Category Archives: Environmental Health

Location Is Everything: The Pollutants in Yellowfin Tuna Depend on Where It’s Caught

PDF icon PDF Version (312 KB)

  • Published: 25 July 2017

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Related EHP Article

Geographic Differences in Persistent Organic Pollutant Levels of Yellowfin Tuna

Sascha C.T. Nicklisch, Lindsay T. Bonito, Stuart Sandin, and Amro Hamdoun

Fish is a highly nutritious food, but it can also be a dietary source of persistent organic pollutants (POPs).1 In a new study in EHP, researchers investigated the extent to which contaminant levels within a single commercially important fish species varied depending on where the fish was caught.2 Their results suggest that capture location may be an important yet overlooked variable when assessing the risk of exposure to POPs from eating wild fish.

Governmental fish consumption advisories suggest limiting intake of certain fish species to reduce human exposures to POPs.3 These potentially harmful chemicals accumulate in body fat, and larger animals farther up the food chain tend to have higher levels. That is why characteristics, such as a fish’s species, fat content, body size, and trophic level, are used as predictors of pollutant load.4

Photo of a yellowfin tuna on a dock.
A fisherman unloads yellowfin tuna at a market in Semporna, Malaysia. Because these large predators do not migrate widely, their body burdens of persistent organic pollutants (POPs) are a good indicator of how polluted their home waters might be. © shahreen/Shutterstock.

However, pollutant levels are not uniform across the world’s oceans. Instead, POPs form a patchy distribution. Many of the fish species widely consumed in the United States, such as tuna, are harvested throughout the world’s oceans, says the study’s senior author Amro Hamdoun, a professor of biology at the Scripps Institution of Oceanography, University of California, San Diego. Geographic variation in pollutant levels in ocean waters could therefore affect the safety of fish that people eat.

Researchers from the Scripps Institute of Oceanography, led by Sascha Nicklisch, analyzed POPs in the dorsal muscle fillets of 117 wild-caught yellowfin tuna from 12 capture locations in the Atlantic, Eastern Pacific, Western Pacific, and Indian oceans. These commonly eaten fish are found around the world, but they do not travel far,5 making them a likely sentinel of geographic variation in POP levels.2

The authors found that levels of pollutants in tuna varied between sites by a factor of 36. Fish caught in the offshore waters of North America and Europe had pollutant levels that were, on average, more than an order of magnitude higher than fish caught in the waters of Asia and Oceania. The 10 most contaminated fish, which were found in the Northeast Pacific Ocean, the Gulf of Mexico, and the Northeast Atlantic Ocean, had POP levels of 20–29 ng/g wet weight of fish. The 10 least contaminated fish, taken from the Northwest and Southwest Pacific Ocean, the South China Sea, and the Indian Ocean, had pollutant levels of 0.2–0.4 ng/g wet weight.2

The researchers also measured specific POPs belonging to a group of chemicals they have dubbed transporter interfering compounds (TICs). In earlier work, they identified TICs as compounds that may interfere with transporter proteins, hampering their ability to eliminate foreign substances from the body.6 In the current study, the researchers found that the concentrations and distribution of TICs reflected those of total POPs. Hamdoun says, “TICs could help explain why certain chemicals are so persistent in body tissue,” although future research is needed to better understand the connection.

To assess the impact of geographic variation on risk-based consumption advice, the researchers calculated monthly meal recommendations based on POP levels in individual fish. Based on their estimates, most of the yellowfin tuna they caught would have been safe to eat in unlimited quantities, says Hamdoun. However, 9 of the 10 fish captured in the Northeast Atlantic Ocean and 5 of the 8 fish caught in the Gulf of Mexico contained POP levels that would trigger health advisories even for people who eat less fish than currently recommended by the American Heart Association (AHA) and U.S. Department of Agriculture (USDA). USDA dietary guidelines suggest that people eat 12 ounces of cooked fish per week,7 while the AHA recommends two 3.5-ounce servings of cooked fish each week.8

The new findings expand on earlier studies that pointed out that capture location might be an important factor in the contaminant load in fish. In one study, researchers used skipjack tuna to assess the distribution of POPs, including polychlorinated biphenyls and organochlorine pesticides in different parts of the Pacific Ocean.9 In another, farm-raised salmon from Europe had higher levels of organochlorine pesticides than those raised in North or South America.10 However, these studies were limited geographically, compared with the current study.

Unlike previous studies of farmed and wild-caught fish, Hamdoun and colleagues did not always find that POP levels correlated with the size or body fat percentage of the fish; in many cases, he says, fatty fish from a clean site were found to have lower levels of POPs than lean fish from a more polluted site. Based on these findings, Hamdoun concludes that lipid content alone is not enough to predict pollutant load when comparing fish from different sites.

According to David Carpenter, a public health physician at State University of New York at Albany who has studied POPs in farmed and wild-caught salmon and Great Lakes fish, these results are unexpected, since POPs accumulate in lipids. Carpenter, who was not involved in the current study, says it is possible that too few fish were captured at each location to suss out statistically significant differences between bigger and smaller fish at each site.

Overall, says Carpenter, the new study adds to mounting evidence that “where a fish comes from is an important factor to consider in determining likely contaminant loads and fish consumption advisories.” Unfortunately, he adds, the new findings are likely to frustrate consumers who seldom receive information about capture location when purchasing seafood at supermarkets and restaurants.


Lindsey Konkel is a New Jersey–based journalist who reports on science, health, and the environment.

References

1. Schecter A, Haffner D, Colacino J, Patel K, Päpke O, Opel M, et al. 2010. Polybrominated diphenyl ethers (PBDEs) and hexabromocyclodecane (HBCD) in composite U.S. food samples. Environ Health Perspect 118(3):357–362, PMID: 20064778, 10.1289/ehp.0901345.

2. Nicklisch SCT, Bonito LT, Sandin S, Hamdoun A. 2017. Geographic differences in persistent organic pollutant levels of yellowfin tuna. Environ Health Perspect 125(6):067014, PMID: 28686554, 10.1289/EHP518.

3. U.S. Environmental Protection Agency. 2000. Guidance for assessing chemical contaminant data for use in fish advisories. In: Risk Assessment and Fish Consumption Limits, Vol 2. 3rd edition. Washington, DC:Office of Science and Technology, Office of Water, U.S. Environmental Protection Agency. https://www.epa.gov/sites/production/files/2015-06/documents/volume2.pdf [accessed 21 February 2017].

4. Elskus A, Collier TK, Monosson E. 2005. Interactions between lipids and persistent organic pollutants in fish. In: Biochemistry and Molecular Biology of Fishes, Vol. 6. Mommsen TP, Moon TW, eds. San Diego, CA:Elsevier, 119–152.

5. Block BA, Jonsen ID, Jorgensen SJ, Winship AJ, Shaffer SA, Bograd SJ, et al. 2011. Tracking apex marine predator movements in a dynamic ocean. Nature 475(7354):86–90, PMID: 21697831, 10.1038/nature10082.

6. Nicklisch SCT, Rees SD, McGrath AP, Gökirmak T, Bonito LT, Vermeer LM, et al. 2016. Global marine pollutants inhibit P-glycoprotein: environmental levels, inhibitory effects, and cocrystal structure. Sci Adv 2(4):e1600001, PMID: 27152359, 10.1126/sciadv.1600001.

7. U.S. Department of Agriculture. 2015. Dietary Guidelines for Americans 2015–2020. 8th edition. Washington, DC:U.S. Department of Agriculture. https://health.gov/dietaryguidelines/2015/resources/2015-2020_Dietary_Guidelines.pdf [accessed 21 February 2017].

8. American Heart Association. 2016. Fish and Omega-3 Fatty Acids. Updated 6 October 2016. Dallas:TX:American Heart Association. http://www.heart.org/HEARTORG/HealthyLiving/HealthyEating/HealthyDietGoals/Fish-and-Omega-3-Fatty-Acids_UCM_303248_Article.jsp#.WKxzwBIrLJw [accessed 21 February 2017].

9. Ueno D, Takahashi S, Tanaka H, Subramanian AN, Fillmann G, Nakata H, et al. 2003. Global pollution monitoring of PCBs and organochlorine pesticides using skipjack tuna as a bioindicator. Arch Environ Contam Toxicol 45(3):378–389, PMID: 14674591, 10.1007/s00244-002-0131-9.

10. Hites RA, Foran JA, Carpenter DO, Hamilton MC, Knuth BA, Schwager SJ. 2004. Global assessment of organic contaminants in farmed salmon. Science 303(5655):226–229, PMID: 14716013, 10.1126/science.1091447.

Race/Ethnicity, Socioeconomic Status, Residential Segregation, and Spatial Variation in Noise Exposure in the Contiguous United States

Author Affiliations open

1Robert Wood Johnson Foundation Health & Society Scholars Program, University of California, San Francisco and University of California, Berkeley, California, USA

2Department of Environmental Science, Policy, and Management, and the School of Public Health; University of California, Berkeley, California, USA

3Department of Electrical & Computer Engineering, Colorado State University, Fort Collins, Colorado, USA

4Natural Sounds and Night Skies Division, Natural Resource Stewardship and Science Directorate, National Park Service, Fort Collins, Colorado, USA

5Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, Maryland, USA

6Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Healthcare Institute; Departments of Epidemiology and Environmental Health, Harvard TH Chan School of Public Health; Channing Division of Network Medicine, Department of Medicine, Brigham and Women’s Hospital & Harvard Medical School, Boston, Massachusetts, USA

PDF icon PDF Version (3.4 MB)

  • Background:
    Prior research has reported disparities in environmental exposures in the United States, but, to our knowledge, no nationwide studies have assessed inequality in noise pollution.
    Objectives:
    We aimed to a) assess racial/ethnic and socioeconomic inequalities in noise pollution in the contiguous United States; and b) consider the modifying role of metropolitan level racial residential segregation.
    Methods:
    We used a geospatial sound model to estimate census block group–level median (L50) nighttime and daytime noise exposure and 90th percentile (L10) daytime noise exposure. Block group variables from the 2006–2010 American Community Survey (ACS) included race/ethnicity, education, income, poverty, unemployment, homeownership, and linguistic isolation. We estimated associations using polynomial terms in spatial error models adjusted for total population and population density. We also evaluated the relationship between race/ethnicity and noise, stratified by levels of metropolitan area racial residential segregation, classified using a multigroup dissimilarity index.
    Results:
    Generally, estimated nighttime and daytime noise levels were higher for census block groups with higher proportions of nonwhite and lower-socioeconomic status (SES) residents. For example, estimated nighttime noise levels in urban block groups with 75% vs. 0% black residents were 46.3 A-weighted decibels (dBA) [interquartile range (IQR): 44.3–47.8 dBA] and 42.3 dBA (IQR: 40.4–45.5 dBA), respectively. In urban block groups with 50% vs. 0% of residents living below poverty, estimated nighttime noise levels were 46.9 dBA (IQR: 44.7–48.5 dBA) and 44.0 dBA (IQR: 42.2–45.5 dBA), respectively. Block groups with the highest metropolitan area segregation had the highest estimated noise exposures, regardless of racial composition. Results were generally consistent between urban and suburban/rural census block groups, and for daytime and nighttime noise and robust to different spatial weight and neighbor definitions.
    Conclusions:
    We found evidence of racial/ethnic and socioeconomic differences in model-based estimates of noise exposure throughout the United States. Additional research is needed to determine if differences in noise exposure may contribute to health disparities in the United States. https://doi.org/10.1289/EHP898
  • Received: 01 August 2016
    Revised: 06 February 2017
    Accepted: 06 February 2017
    Published: 25 July 2017

    Address correspondence to J.A. Casey, University of California, Berkeley, 13B University Hall, Berkeley, CA 94610 USA. Telephone: 541-760-8477; Email: joanacasey@berkeley.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP898).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (908 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (99 KB)

Introduction

A growing body of evidence links environmental noise—a biologic stressor usually generated by mechanized sources: transportation, industry, power generation, power tools, and air-conditioning—to hearing loss and other health outcomes (Basner et al. 2014). The human body initially reacts to noise with activation of the central nervous system, even while asleep. This can result in release of stress hormones and increased blood pressure, heart rate, and cardiac output (Evans et al. 1995; Lercher 1996). While individual noise sensitivities differ, the World Health Organization (WHO) estimated a “no observed effect level” for average outdoor nighttime noise of 30 A-weighted decibels (dBA) based on evidence that sleep is not disturbed by noise below 30 (dBA) (WHO 2009). The Federal Highway Administration noise abatement criteria near hospitals and schools is 70 dBA, a recommendation that balances health, communication, and economic interests (U.S. DOT 2015). Exposure to these noise levels has been associated with impaired cognitive performance (Clark et al. 2012) and behavioral problems in children (Hjortebjerg et al. 2016), as well as hypertension (van Kempen and Babisch 2012), type 2 diabetes (Sørensen et al. 2013), cardiovascular disease (Gan et al. 2012), and reduced birth weight (Gehring et al. 2014). The WHO (2011) has estimated >1 million disability adjusted life years are lost annually in Western Europe due to environmental noise, attributable primarily to annoyance and sleep disturbance. The WHO calculation was based on estimated noise exposures and previous research on associations between noise and health outcomes.

Environmental noise is typically measured as sound pressure level, a logarithmic quantity expressed in decibels (dB); for example, an increase of 3 dB is a doubling of sound energy. With every 5.5-dB increase, the proportion of individuals highly annoyed by residential noise exposure appears to double (ANSI 2003). Measurements of sound pressure level are commonly adjusted by A-weighting to reflect how humans perceive sound across frequency, denoted as dBA (Murphy and King 2014). Because sound levels vary over time, metrics describing the statistical behavior of the variation are utilized. The energy average, or equivalent, indicator is abbreviated Leq. Multiple exceedance levels are used to characterize magnitude, rate of occurrence, and duration of environmental noise. The L50 is the noise level exceeded half of the time, whereas the L10 is the level exceeded 10% of the time.

Like other exposures, the impact of noise varies by intensity, duration, and frequency. Noise sensitivity or degree of reactivity to the same level of noise can differ from person to person and by source of noise (Janssen et al. 2011; van Kamp et al. 2004). Time of day may also play a role, such that associations between noise and health outcomes appear to be stronger for noise exposure during the night vs. day (Basner et al. 2014). Despite evidence of noise-related adverse health effects, the most recent nationwide noise pollution estimates were made by the U.S. Environmental Protection Agency (U.S. EPA) in 1981 (Simpson and Bruce 1981). By extrapolating the U.S. EPA’s 1981 estimate of the prevalence of noise exposure to the current U.S. population, Hammer et al. (2014) estimated that 145.5 million Americans experience annual noise levels that exceed those recommended to protect public health with an adequate margin of safety. Moreover, the distribution of noise is not uniform across communities, and some groups may have heightened vulnerability to noise (van Kamp and Davies 2013). The spatial distribution of noise exposure may contribute to health disparities seen in the United States and elsewhere.

A body of environmental justice literature from the United States suggests that air pollution and exposure to hazardous waste often follows a social gradient such that racial/ethnic minorities, the poor, and the undereducated endure greater exposure (Bell and Ebisu 2012; Hajat et al. 2015; Mohai and Saha 2007). A more limited body of scholarship from Europe frames environmental injustices by social categories, but not usually race/ethnicity, finding, for example, that those in the 10% most deprived areas in England are the most exposed to chemical, metal, and waste facilities (Laurent 2011; Walker et al. 2005). Researchers theorize that in the United States, communities of color and the poor are disproportionately exposed to environmental hazards due to a variety of factors, including weak regulatory enforcement in marginalized neighborhoods and lack of capacity to engage in land use decision-making, which may contribute to the concentration of potentially hazardous mobile and stationary emission sources in these communities (Morello-Frosch 2002; Pulido 2000).

Only a few studies have evaluated community-level inequality in exposure to estimated noise pollution. Studies in Minneapolis and St. Paul, Minnesota (Nega et al. 2013) and Montreal, Quebec, Canada (Carrier et al. 2016; Dale et al. 2015) found that lower neighborhood socioeconomic status (SES) or a higher proportion of minority race/ethnicities was associated with higher noise levels. Outside the United States and Canada, results have been mixed and more focused on SES than race/ethnicity as an explanatory variable. In Hong Kong, Lam and Chan (2008) reported a weak, but statistically significant, association between lower income and educational attainment and higher noise exposure at the street block level. Haines et al. (2002) estimated noise exposure at 123 schools near the Heathrow Airport in the United Kingdom, and found in a subanalysis that a higher proportion of students eligible for free lunch was associated with higher noise exposure. In nearby Birmingham, a higher proportion of black residents at the enumeration district level was weakly associated with estimated daytime noise levels (Brainard et al. 2004). In Marseilles, France, census blocks with intermediate levels of deprivation had the highest estimated exposure to road noise, whereas in Berlin, Germany, there was no straightforward association between SES and noise exposure at the planning unit level (Lakes et al. 2014).

At the individual level, one study in Wales, United Kingdom (Poortinga et al. 2008) and another in Germany (Kohlhuber et al. 2006) found that lower SES participants reported more neighborhood noise. However, in Paris, France, individuals living in neighborhoods with the highest housing values and highest levels of educational attainment also had the highest estimated noise exposures (Havard et al. 2011).

To our knowledge, no prior studies have evaluated demographic disparities in noise pollution across the United States or considered how racial segregation (an indicator of metrowide social inequality) is associated with overall noise levels. Prior U.S.–based studies have found increased racial segregation associated with more air pollution (Bravo et al. 2016; Jones et al. 2014), ambient air toxins (Morello-Frosch and Jesdale 2006; Rice et al. 2014), and less tree canopy cover (Jesdale et al. 2013). In highly segregated metropolitan areas in the United States, political power is asymmetrical along racial, ethnic, and economic lines. Further, segregation spatially binds communities of color and working class residents through the concentration of poverty, lack of economic opportunity, and exclusionary housing development and lending policies (Massey and Denton 1993). These power differences may lead to disparities in environmental hazard exposures, including noise, as more powerful residents influence decisions about the siting of undesirable land uses in ways that are beneficial to them (Cushing et al. 2015; Morello-Frosch and Lopez 2006). Because segregation can make it easier for more powerful communities to displace hazardous land uses onto disadvantaged communities where regulations may not be consistently enforced, this scenario can lead to higher pollution overall (Ash et al. 2013). Segregation may also lead to spatial segmentation between neighborhoods, workplaces, and basic services, resulting in more driving, longer commute times, and higher levels of mobile source pollution (Ash et al. 2013; Morello-Frosch and Jesdale 2006).

In the present study, we utilized a nationwide noise model (Mennitt and Fristrup 2016) to a) estimate differences in noise exposure along racial/ethnic and socioeconomic lines; and to b) examine whether segregation modifies the association between race/ethnicity and noise exposure across the contiguous United States.

Methods

We conducted a cross-sectional analysis to investigate the spatial distribution of demographic characteristics at the census block group level in relation to noise exposure across the contiguous United States. Prior work identified U.S. block group–level socioeconomic measures as a relevant spatial scale for measuring socioeconomic inequality (Krieger et al. 2003). In 2010, the study area (i.e., the contiguous United States) contained 216,331 block groups—statistical divisions of census tracts generally containing 600 to 3,000 people—that we assigned to 933 Core Based Statistical Areas (CBSAs) based on the location of their centroid, using 2010 TIGER/Line shapefiles (U.S. Census Bureau 2010a, 2010b). CBSAs are counties grouped by common commuting patterns. We excluded 1,669 block groups without residents or that were missing data on any socioeconomic variable (0.8%) and 557 block groups missing noise exposure estimates (0.3%). We then designated block groups located in CBSAs containing >100,000 people as urban (n=175,373) and those located elsewhere as suburban/rural (n=38,732).

Dependent Variables

Exposure to noise was based on a previously published geospatial model of environmental sound levels (Mennitt and Fristrup 2016). Similar to land use regression models of air quality, expected noise exposure was modeled using empirical acoustical data and geospatial features such as topography, climate, hydrology, and anthropogenic activity. The acoustical data included 1.5 million h of long-term (durations of ≥25 d at natural sites and in 14 U.S. cities, and ≥30 d near all U.S. airports) measurements from 492 urban and rural sites located across the contiguous United States during 2000–2014. The explanatory variables fell into seven groups (location, climatic, landcover, hydrological, anthropogenic, temporal, and equipment) and are described in detail elsewhere (Sherrill 2012). Cross-validation procedures were used to evaluate model performance and identify variables with predictive power. The method utilized random forest, a tree-based machine-learning algorithm, to perform the regression. A cross-validation procedure was used to evaluate the accuracy of national scale projections (see Table S1; Mennitt and Fristrup 2016). The resulting geospatial sound model enabled mapping of ambient sound levels at 270-m resolution. We then used the zonal statistics function in ArcGIS (version 10.4; Esri) to estimate the mean sound level in each block group across the contiguous United States.

We examined three metrics of sound pressure from anthropogenic sources to assess the robustness of findings at different times of day and different levels of noise: a) A-weighted L10 (representing the loudest transient events or proximate sources) during the daytime; b) A-weighted L50 (median of the data, representing typical sound levels) during the daytime; and c) A-weighted L50 during the nighttime. All levels were projected for the summer season in order to maintain temporal consistency across noise estimates. Daytime was defined as 0700 to 1900 hours and nighttime as 1900 to 0700 hours.

Independent Variables

We used block group-level data from the 5-y 2006–2010 American Community Survey (ACS) to characterize area-level race/ethnicity and socioeconomic conditions (NHGIS Database). We used self-identified race/ethnicity to generate variables for the proportion of the population in each block group that fell in five race/ethnicity categories: Non-Hispanic American Indian, Non-Hispanic Asian, Non-Hispanic black, and Non-Hispanic white, and Hispanic (any race), referred to as American Indian, Asian, black, white, and Hispanic for the duration. Other block group–level variables included total population, population density (defined as number of people per km2), age (defined as percent of population <5 y old), and those selected to describe multiple dimensions of neighborhood socioeconomic context (Kawachi and Berkman 2003). We characterized neighborhood socioeconomic context at the block group level by: low educational attainment (defined as percent of adults ≥25 y of age with <high school education), median household income (defined as block group median income in dollars in the past 12 mo), poverty (defined as percent of individuals with income below the Census Bureau poverty threshold based on family size and number of children), civilian family unemployment (defined as percent of families with ≥1 family member unemployed), housing tenure (defined as percent of households comprised of renters or owners), and linguistic isolation (defined as percent of households where no one >14 y speaks English “very well”). Housing tenure may reflect residential instability as well as area-level income and wealth. Linguistically isolated households may face racial discrimination and reduced access to public services and ability to engage with regulatory processes (Gee and Ponce 2010). Some variables, like housing tenure, may have differential meaning in urban vs. rural settings, and, therefore, we conducted analyses stratified by urban/rural status (Bertin et al. 2014). To test the hypothesis that segregation was associated with higher levels of noise for all race/ethnic groups, we calculated a CBSA-level segregation measure (Sakoda 1981) for urban block groups only, using 5-y 2006–2010 ACS data (Jesdale et al. 2013). Our focus on urban areas was aimed at facilitating comparisons with prior studies on segregation and environmental hazards (Bravo et al. 2016; Jones et al. 2014; Morello-Frosch and Jesdale 2006; Rice et al. 2014). The multigroup dissimilarity index (Dm) characterizes the residential distribution of Non-Hispanics: Asians, blacks, and whites; and Hispanics (any race) among block groups located in CBSAs. Dm ranges from 0 to 1 and represents the proportion of the racial/ethnic minority population that would need to change block groups within a metro area to achieve an even distribution (Massey and Denton 1989). Block groups located in the same CBSA received the same Dm value. We selected Dm, a metro-level indicator of social inequality, because prior research indicated that land use decision-making tends to be regionally rooted (Morello-Frosch 2002; Pastor et al. 2000).

Statistical Analysis

We used weights to extrapolate block group–level noise estimates to the individual, family, and household level across the United States. For each characteristic of interest (e.g., non-Hispanic blacks), we used the wquantile function in R (version 3.2.3; R Development Core Team) to compute the weighted 25th, 50th, and 75th percentiles of noise.

We evaluated the association between 12 socioeconomic variables and log–transformed L50 nighttime, L50 daytime, and L10 daytime noise exposure by specifying 12 separate regression models for each noise metric, controlled for population size and population density. We checked the residuals from ordinary least square models and found evidence of spatial autocorrelation using Moran’s I statistic (p<0.001 indicating clustering; data not shown). Therefore, we implemented a spatial econometric approach in R using the spdep package (Bivand and Piras 2015). To discriminate between spatial autocorrelation in the error terms vs. the noise values themselves, we used a Lagrange multiple diagnostic test (implemented with lm.LMtests in spdep). p-values from this test suggested that a spatial error model was preferable to a spatial lag model for our data. A spatial error model specifies a linear relationship between the independent variable and the dependent variable, but unlike a traditional ordinary least squares model, errors are not assumed to be independent and identically distributed; rather, they are distributed according to a spatial autoregressive process:

where y is an n×1 vector of log L50 or L10 daytime or nighttime sound pressure, and n is the total number of block groups; X is an n×j matrix of independent variables, and j is the number of independent variables; W is an n×n spatial weights matrix; e∼N(0, σ2I); and λ is a spatial autoregressive coefficient (Anselin 2002). In addition to adjustment for block group population (continuous) and population density (continuous, individuals/km2), all models included polynomial terms for the independent variable of interest to allow for nonlinearity. We hypothesized that nonwhite race/ethnicities would experience higher overall levels of noise and steeper slopes as the percent minority increased in more segregated CBSAs. Therefore, we conducted stratified analyses [i.e., Dm<0.4 (low to moderate), 0.4 to <0.5 (high), 0.5 to <0.6 (very high), and ≥0.6 (extreme)] (Jesdale et al. 2013; Morello-Frosch and Jesdale 2006) from which American Indians were excluded due to small numbers in urban areas. In all analyses, we selected the number of polynomial terms (up to 10) using likelihood ratio testing by adding polynomial terms until the improvement in model fit was no longer significant at α=0.05. This procedure was completed separately for urban/rural and for L50 nighttime/L10 and daytime/L50 daytime models. In the results section, we refer to an association as nonlinear and statistically significant when likelihood ratio testing indicated that the final model with polynomial terms was a significantly better fit at the α=0.05 level than the model without the predictor. When the final model only included a first-degree predictor, we refer to it as a linear association; statistical significance (at the α=0.05 level) was determined using a t-test. We present results as scatterplots of fitted values with locally weighted smoothing functions (LOESS lines) to aid in interpretability. We did not predict noise exposure at specific values of the independent variables because predicting at specified populations and population densities would ignore model-derived weights applied to each census block to account for spatial correlations. Instead, we report median and interquartile ranges to summarize the distribution of predicted values for each noise metric according to block group race/ethnicity and socioeconomic characteristics.

In order to calculate W, the spatial weight matrix, in the main analysis, we defined queen-based neighbors (i.e., only block groups that share a common border or a single common point were considered neighbors) and S-coding scheme weights (i.e., variance-stabilizing weights) (Bivand et al. 2008; Tiefelsdorf et al. 1999). Because the choice of neighbors and weights can influence model fit and coefficient estimates, we performed several sensitivity analyses. First, we used distance-based neighbors, where block groups within 5 km of a block group centroid were classified as neighbors. Second, we applied a W-coding scheme for weights (i.e., row-standardized). Third, we combined the first two sensitivity analyses.

In additional sensitivity analyses, we simultaneously adjusted models for poverty, education, and housing tenure to assess whether these three dimensions of neighborhood socioeconomic context were confounders. All statistical analysis was performed in R (version 3.2.3; R Development Core Team) using Amazon Web Services and mapping was conducted in QGIS (version 2.12.0; QGIS, http://qgis.osgeo.org).

Results

The analysis spanned the contiguous United States and included 933 CBSAs and 214,105 block groups—175,373 urban and 38,732 suburban/rural—after excluding 2,226 block groups missing census or noise variables (Figure 1). The urban block groups contained 257,192,214 individuals, and the rural block groups 45,999,315 individuals. There was a concentration of poverty, nonwhite individuals, and low educational attainment in the South and Southwest. Urban block groups, compared to suburban/rural areas, had, on average, more racial/ethnic minorities (38.0% vs. 19.3%), more renter-occupied homes (34.6% vs. 26.9%), and slightly lower levels of poverty (13.3% vs. 16.8%) [Table 1 (urban) and Table S2 (suburban/rural)]. We observed moderate correlations (Spearman’s ρ=0.2–0.4) between many of the independent variables, for example, ρ=0.20, 0.25, and −0.43 between proportion in poverty and proportions unemployed, linguistically isolated, and of white race, respectively, in urban areas (see Figure S1). Individuals, households, and families in urban areas with lower SES had, on average, higher nighttime and daytime noise (Table 1). For example, in urban areas, the median L10 daytime noise estimated for households in the lowest quartile of median income (≤$39,224) was 54.5 dBA [interquartile range (IQR): 52.4–56.5 dBA] compared to 52.6 dBA (IQR: 50.4–54.5) estimated for households with median income >$39,224. Racial residential segregation was common; 83.4% of the urban study population resided in segregated CBSAs (Dm>0.4) (Table 1), with the most segregated regions in the Northeast and Midwest (Figure 1). Individuals living in more segregated CBSAs, compared to those living in CBSAs with Dm≤0.4, had higher L50 nighttime [median=44.5 dBA (IQR: 42.5–46.7 dBA) vs. 42.9 (IQR: 39.2–45.2 dBA)] and L50 daytime [median=48.2 dBA (IQR: 45.7–50.5 dBA) vs. 46.7 dBA (IQR: 41.5–49.4 dBA)] noise levels (Table 1).

Six spatial distribution maps of USA
Figure 1. Spatial distribution of (A) anthropogenic L50 nighttime noise; (B) population density; (C) racial residential segregation (urban CBSAs only); (D) non-Hispanic, nonwhite race/ethnicity; (E) poverty; and (F) <high school education (deciles) at the block group-level in the contiguous United States estimated from 2006–2010 American Community Survey data; 2010 shapefiles used to generate these maps downloaded from the NHGIS site: http://www.nhgis.org.
Table 1. Distribution of anthropogenic L50 nighttime, L50 daytime, and L10 daytime noise among urban residents by race/ethnicity and socioeconomic characteristics at the block group level from the 2006–2010 American Community Survey.
Median (IQR) anthropogenic noise, dBAa
Characteristic Total, n (%) L50 nighttime L50 daytime L10 daytime
Total population 254,328,850 (100) 44.3 (42.1–46.5) 48.0 (45.1–50.3) 52.9 (50.7–55.0)
 Population<5 y 17,112,446 (6.7) 44.5 (42.3–46.7) 48.1 (45.4–50.5) 53.0 (50.9–55.0)
 Population≥5 y 237,216,404 (93.3) 44.3 (42.1–46.5) 48.0 (45.0–50.3) 52.9 (50.7–55.0)
Race/ethnicityb
 Hispanic 44,095,827 (17.3) 45.6 (43.3–47.5) 49.5 (47.5–52.3) 54.1 (52.3–56.0)
 Non-Hispanic
 American Indian 1,209,132 (0.5) 42.9 (37.9–45.7) 46.1 (37.8–49.7) 51.5 (44.8–54.4)
 Asian 13,081,414 (5.1) 45.4 (43.9–47.1) 49.1 (47.4–51.1) 54.0 (52.4–55.7)
 Black 32,935,749 (13.0) 45.6 (43.8–47.6) 49.7 (47.6–52.6) 54.2 (52.4–56.3)
 White 157,730,767 (62.0) 43.6 (41.3–45.7) 47.1 (43.3–49.2) 52.3 (49.6–54.2)
 Income≤poverty threshold 33,194,588 (13.3) 45.2 (42.8–47.5) 49.2 (46.6–52.2) 54.0 (51.7–56.1)
 Income>poverty thresholdc 216,181,346 (86.7) 44.2 (42.0–46.3) 47.9 (44.9–50.0) 52.8 (50.6–54.8)
CBSA-level segregation
 0.14≤Dm<0.40 42,124,233 (16.6) 42.9 (39.2–45.2) 46.7 (41.5–49.4) 51.9 (48.0–54.3)
 0.40≤Dm<0.70 212,204,617 (83.4) 44.5 (42.5–46.7) 48.2 (45.7–50.5) 53.1 (51.1–55.1)
Total population≥25 y 35,298,009 (100) 44.6 (42.4–46.8) 48.5 (45.7–50.9) 53.3 (51.2–55.4)
 <High school education 5,837,943 (16.5) 45.4 (43.0–47.6) 49.4 (46.7–52.3) 54.1 (51.8–56.1)
 ≥High school education 29,460,066 (83.5) 44.4 (42.3–46.6) 48.3 (45.6–50.7) 53.2 (51.1–55.3)
Total households 95,455,047 (100) 44.3 (42.2–46.5) 48.1 (45.2–50.4) 52.0 (50.9–55.1)
 Median household income (USD)
 Quartile 1 ($2,868–$39,229) 23,863,693 (25.0) 45.6 (43.2–47.7) 49.8 (47.3–52.8) 54.5 (52.4–56.5)
 Quartiles 2–4 ($39,230–$249,896) 71,591,354 (75.0) 44.0 (41.8–46.0) 47.6 (44.5–49.6) 52.6 (50.4–54.5)
 Linguistically isolated households 5,140,332 (5.4) 45.9 (43.9–47.9) 50.4 (48.2–53.3) 54.8 (53.0–56.8)
 Nonlinguistically isolated households 90,314,715 (94.6) 44.2 (42.1–46.4) 48.0 (45.0–50.2) 52.9 (50.7–54.9)
 Housing tenure
 Renter-occupied homes 32,996,266 (34.6) 45.3 (43.3–47.4) 49.5 (47.2–52.3) 54.2 (52.3–56.3)
 Owner-occupied homes 62,458,781 (65.4) 43.8 (41.6–45.9) 47.4 (43.9–49.4) 52.5 (50.0–54.3)
Total families 63,521,803 44.1 (41.9–46.3) 47.7 (44.6–49.9) 52.7 (50.4–54.7)
Unemployed families 3,343,134 (5.3) 44.5 (42.3–46.8) 48.2 (45.4–50.6) 53.1 (51.0–55.2)
Employed families 60,180,669 (94.7) 44.1 (41.9–46.2) 47.7 (44.5–49.9) 52.7 (50.4–54.7)
Note: CBSA, Core Based Statistical Area; dBA, A-weighted decibels; IQR, interquartile range.

aPopulation-weighted by block group population (population<5 y, and race/ethnicity), by number of families (unemployment), by households (household income and linguistic isolation, and renters/owners), by population for whom poverty status was determined (poverty), and by population≥25 y (<high school education).

bRace/ethnicity does not sum to total; 5,275,961 individuals were of mixed or other race/ethnicity.

c4,952,916 people did not have poverty status determined and thus are not included in the poverty summary.

Median population-weighted anthropogenic L50 nighttime, L50 daytime, and L10 daytime noise levels in urban block groups were 44.3 dBA (IQR: 42.1–46.5 dBA), 48.0 dBA (IQR: 45.1–50.3 dBA), and 52.9 dBA (IQR: 50.7–55.0 dBA) (Table 1) compared to 38.8 dBA (IQR: 33.7–42.2 dBA), 37.0 dBA (IQR: 32.8–43.2 dBA), and 43.7 dBA (IQR: 40.2–49.6) in suburban/rural block groups (Table S1), respectively. While the noise metrics differed in magnitude, the spatial distribution was similar with noise levels highest in major metropolitan areas and lowest in the western United States (Figure 1, Figure S2). Overall, estimated noise levels were highest for Asians, blacks, Hispanics, and those of lower SES (Table 1, Table S2).

Figure 2 presents the results from adjusted spatial error models of the association between our 12 racial/ethnic and sociodemographic variables and nighttime L50 noise in urban and suburban/rural block groups. Daytime L50 and L10 results are available in Figures S3 and S4. Patterns of associations between each predictor and each noise metric were generally consistent in both urban and rural census blocks; as the proportion of nonwhite and low-SES individuals, families, and households increased, estimated nighttime and daytime noise increased. Table S4 summarizes the polynomial terms used for predictors in spatial error models.

Figure 2A comprises 12 plots indicating L sub 50 nighttime noise in decibels (y-axis) for urban block groups across percentage less than 5 y (log scale), percentage less than high school, income ($1000, log scale), percentage of poverty, percentage of unemployed, percentage of American Indians, percentage of Asians, percentage of Black, percentage of Hispanic, percentage of Whites, percentage of linguistically isolated, and percentage of renter-occupied (x-axis). Likewise, Figure 2B comprises plots of similar data for suburban/rural block groups.
Figure 2. Race/ethnicity and socioeconomic characteristics and anthropogenic L50 nighttime noise in (A) urban block groups (n=175,373); and (B) suburban/rural block groups (n=38,732). The figure displays the fitted values (points) showing the relationship between noise and each of 12 demographic characteristics adjusted for block group population and population density and using a queen neighbor definition and variance-stabilizing weights. Four of the plots [%<5 y, median household income (in thousands)], % unemployed, and % linguistically isolated) use a log scale x-axis as noted on the figure. The LOESS line was only estimated when there were >100 observations.

We observed significant associations between increased proportions of Asian, black, and Hispanic individuals and higher levels of noise in urban and suburban/rural areas, with the exception of Hispanic ethnicity in rural areas, which was not associated with noise. In urban block groups that contained 75% black residents, the median (interquartile range) nighttime noise level was 46.3 dBA (IQR: 44.3–47.8), while in block groups with 0% black residents, the level was 42.3 dBA (IQR: 40.4–45.5) (Table S3). Conversely, in urban block groups that contained 75% white, the median noise level was 44.8 dBA (IQR: 43.0–46.5), which rose to 47.0 dBA (IQR: 45.4–48.7) in block groups with 0% white residents. For race/ethnicity, the models with best fit all included polynomial terms (Table S4), except for Hispanic race/ethnicity, where we estimated a linear association between nighttime noise in both urban and suburban/rural block groups. Estimated urban nighttime noise increased from 43.7 dBA (IQR: 40.1–45.9) in block groups with 0% Hispanics to 46.5 dBA (IQR: 43.7–48.3) within block groups with 75% Hispanics (Table S3). For American Indian populations, we generally observed reduced noise as the percent of American Indians increased. Particularly for nighttime noise, the best model fit suggested shape of the association was fairly flat until 7–8% of the population was American Indian, and then there were more rapid reductions in noise levels. In contrast, the best model fit suggested a steeper slope (i.e., more rapid increases in estimated noise) for Asian and black populations at the lower tail of the distribution. For example, nighttime noise was estimated to be about 1.3 dBA higher in block groups containing 0% compared to 10% black individuals, but only 0.2 dBA higher in block groups containing 50% compared to 60% black individuals.

Block groups with higher proportions of individuals with less than a high school education, living in poverty, linguistically isolated, renting, and with a higher proportion of children <5 y were generally associated with higher nighttime and daytime noise levels (Figure 2, Figure S3–S4). For example, urban block groups with 50% of residents in poverty had nighttime noise levels, on average, 3 dBA higher than block groups with 0% (Table S3). The highest levels of noise were estimated in the block groups with the lowest median income; from there, noise levels declined until the median income, where noise levels plateaued.

Figure 3 shows the associations between race/ethnicity and L50 nighttime noise across segregation categories (see Table S5 for summary of polynomial terms). Three patterns emerged: first, that across all CBSAs and race/ethnicities, increasing segregation was associated with increased nighttime noise; second, that across all levels of CBSA-level segregation, block groups with higher proportions of Asian, Hispanic, and black residents generally had higher levels of exposure to nighttime L50 noise than those with higher proportions of white residents; and third, that the estimated curve shapes (i.e., the LOESS line of the fitted values) remained similar across levels of segregation, with the exception of Hispanics in the least segregated CBSAs, where there was no estimated increase in nighttime noise as the proportion of Hispanics increased above 25%. Results for L50 and L10 daytime noise were similar (Figures S6 and S7), but differences were less pronounced by level of segregation.

Four line graphs plotting L sub 50 nighttime noise in decibels (y-axis) across percentage of Asians, Black, Hispanic, and Whites (x-axis), where the D sub m is less than 0.4, 0.4 to less than 0.5, 0.5 to less than 0.6, and greater than 0.6.
Figure 3. Race/ethnicity and anthropogenic L50 nighttime noise in urban block groups (n=175,373), stratified by multigroup racial/ethnic segregation (Dm) for (A) Asians; (B) blacks; (C) Hispanics; and (D) whites. American Indians were excluded due to small numbers in urban areas. The figure displays the LOESS line of the fitted values from 16 (i.e., four categories of segregation and four race/ethnicities) separate spatial error models adjusted for block group population and population density and using a queen neighbor definition and variance-stabilizing weights. The LOESS line was only estimated when there were >100 observations.

Simultaneous adjustment for poverty, education, and housing tenure in our main analysis had little effect on race/ethnicity and SES estimates in suburban/rural block groups, except for <5 y and education, which, after adjustment, were no longer statistically significant (data not shown). In urban block groups, after adjustment for poverty, education, and housing tenure, the associations for <5 y, Hispanic race/ethnicity, education, and unemployment became nonsignificant (data not shown).

In the primary analysis, we defined queen neighbors and used variance-stabilizing weights. We also conducted sensitivity analyses using three additional neighbor/weight combinations (i.e., queen/W-coding, distance/S-coding, and distance/W-coding). Estimated associations were similar for all neighbor/weight combinations across both urban and suburban/rural block groups (Figure S5).

Discussion

Our findings suggest inequality in the spatial distribution of noise pollution along racial/ethnic and socioeconomic lines across the contiguous United States. Multiple indicators of neighborhood socioeconomic context were associated with increased night and daytime noise, including poverty, unemployment, linguistic isolation, and a high proportion of renters and those not completing high school. Block groups with higher proportions of Asians, blacks, and Hispanics had higher levels of noise, but relationships were rarely linear. The magnitude of these differences may be relevant for health outcomes (Basner et al. 2014); for example, we estimated that census block groups containing 25% black residents were exposed to a median 3.0 dBA higher nighttime noise than those with 0% black residents. In general, for all race/ethnicity groups evaluated, estimated noise exposures were higher for CBSAs with higher levels of racial segregation. However, disparities persisted between block groups with relatively high proportions of white (>50%) compared to relatively low proportions, except for block groups with high proportions of Hispanics in the least segregated CBSAs. As an example, the median nighttime noise level in block groups containing 75% of each race/ethnicity in the most segregated CBSAs were estimated to be 48.9 dBA, 46.9 dBA, 47.0 dBA, and 45.3 dBA for Asian, black, Hispanic, and white race/ethnicity, respectively.

Early indications of inequality in noise pollution in the United States came from the U.S. EPA in the 1970s. Survey respondents of higher SES tended to live in quieter neighborhoods and reported hearing fewer airplanes, traffic, and people’s voices, but more motorcycles, garden power tools, and sports cars (U.S. EPA 1977). More recently, in nearly 2000 block groups in the Twin Cities in Minnesota, Nega et al. (2013) modeled 24-h average traffic noise using data on roadways, traffic volume, building height, airplane flight paths, and other information. They reported significantly increased traffic noise as block group median household income and housing value fell and the proportion of non-white residents and persons aged >18 y increased, results from a spatial error model that simultaneously adjusted for all four independent variables. Nega et al. (2013) joined a handful of others to account for spatial dependence in their models when assessing inequality in noise pollution, with heterogeneous results (Bocquier et al. 2013; Carrier et al. 2016; Havard et al. 2011). Carrier et al. (2016) modeled mean 24-h traffic noise levels in 7,456 city blocks in Montreal, Canada, and used spatial error models to estimate associations with race/ethnicity and SES at the city block level. Like us, they observed increasing noise levels with an increasing proportion of low-income and nonwhite individuals. In Marseilles, France, Bocquier et al. (2013) reported that census blocks of intermediate SES (defined by a deprivation index constructed from 17 variables) had the highest modeled noise levels. In Paris, France, Havard et al. (2011) found an inverse relationship where modeled noise levels were higher in a 250-m buffer surrounding the residences of individuals with more education and higher valued homes. Desire of individuals to live near transportation networks may explain the inverse relationship. Indeed, among nearly 2 million individuals in Rome, Cesaroni et al. (2010) found that higher area SES and individual education were associated with increased traffic in a 150-m buffer around participants’ homes, except in the city center, where traffic density was highest, and less affluent neighborhoods and individuals were closer to roads. We observed indications of the same phenomenon; the relationship between median household income and noise was U-shaped in both urban and rural areas.

There is a broad literature concerning variation in air pollution in relation to social factors (Bell and Ebisu 2012; Hajat et al. 2015; Miranda et al. 2011). Our results are consistent with and may partially overlap this literature, due to co-occurrence of noise and air pollution. However, despite co-occurrence, noise and air pollution have only been moderately correlated (Spearman’s correlation coefficients 0.3–0.6) in New York City, London, and Vancouver (Davies et al. 2009; Fecht et al. 2016; Ross et al. 2011), and correlations did not differ by deprivation in London (Fecht et al. 2016). Furthermore, epidemiologic studies have reported associations between noise and multiple health outcomes after adjusting for air pollution, including associations with cognition and behavioral problems in children (Clark et al. 2012; Hjortebjerg et al. 2016), birth weight (Gehring et al. 2014), cardiovascular mortality (Gan et al. 2012), and diabetes (Sørensen et al. 2013).

To our knowledge, no prior studies have reported a positive association between noise levels and a community-level measure of social inequality, in this case, racial segregation. This observation is consistent with Boyce’s power-weighted decision theory (Boyce 1994) that social inequalities are associated with the distribution of environmental pollution, perhaps due to political power imbalances between the wealthy and the poor. It is also consistent with recent U.S. literature reporting positive associations between modeled air pollution and community segregation (Bravo et al. 2016; Jones et al. 2014; Rice et al. 2014). In the case of environmental noise, spatial segmentation of neighborhoods, workplaces, and basic service locations due to CBSA-level racial segregation may increase vehicle miles traveled (Morello-Frosch and Jesdale 2006), which could, in turn, contribute to noise pollution. This situation could create a feedback loop in which worse noise pollution catalyzes segregation (Bjørnskau 2005).

Our study suggests racial disparities in noise exposure, and noise has previously been linked to a number of negative health outcomes, including hypertension and sleep difficulties (Haralabidis et al. 2008; Münzel et al. 2014; Muzet 2007). Future work is needed to estimate how much differences in noise exposure may explain racial disparities in noise-related health outcomes. Disadvantaged populations may also have increased susceptibility to noise. For example, a recent German study reported that among 3,300 participants free from depressive symptoms at baseline, annual average noise exposure >55 dBA was associated with depressive symptoms at the 5-y follow-up, but only among those with <13 y of education (Orban et al. 2016). Worse quality housing, increased exposure to indoor noise, and comorbid conditions may help explain this disparity (Evans and Marcynyszyn 2004). Wealthier individuals may have greater ability to invest in noise abatement technologies (e.g., triple-paned windows, air-conditioning), and thus may have lower actual exposures to noise than less affluent individuals living in census blocks with the same estimated level of noise exposure. While living near urban centers may provide benefits like access to public transit and cultural assets, in some cities, the accompanying road and rail traffic may also increase the level of outdoor noise (Fyhri and Klaeboe 2006).

For the past 40 y, many noise studies have relied on a simple relationship between population density and community noise exposure to estimate ambient noise in the absence of in situ measurements (Schomer et al. 2011). In addition to population density, our geospatial sound model incorporated multiple explanatory variables describing the type and intensity of human activity, but did not include demographic descriptors. Unfortunately, we were only able to assess the distribution of noise at the block group level in this cross-sectional study. We could not identify the processes and procedures—for example, regulatory policies, neighborhood economic sorting, or land use decisions—that might explain inequality in noise levels, nor could we explain why some groups appeared more exposed to noise than others. We were unable to examine individual modifying characteristics, such as housing quality, work environment and location, prevalent illness, or propensity to noise sensitivity (Stansfeld and Shipley 2015). Additionally, our noise model did not differentiate between various anthropogenic sources of sound, which may have differential health effects (Basner et al. 2011). Although the noise model performed well (R2≥0.8), our prediction of outdoor noise is likely to contain measurement error (Mennitt and Fristrup 2016). Finally, our analysis did not include Leq estimates (i.e., a measure of equivalent continuous sound often used for noise standards). Therefore, we are unable to make comparisons to established WHO or U.S. EPA noise guidelines or to the majority of epidemiologic studies (Basner et al. 2014).

Despite these limitations, our study made several contributions. We characterized noise pollution across the contiguous United States for the first time since the early 1980s. The use of polynomial terms in this large sample allowed us to characterize nonlinear associations without assuming a constant slope over all values of each predictor. We accounted for spatial autocorrelation and implemented a spatial error model to avoid violating modeling assumptions. We defined neighbors and weights in multiple ways, given prior evidence that changing these definitions can impact inference (Bivand et al. 2008). Our sensitivity analyses demonstrated the stability of our results. Regardless of neighbor/weight definitions, we estimated similar relationships between race/ethnicity, SES, and noise almost universally, with the highest estimated noise levels in block groups with higher percentages of minorities and lower SES individuals.

Conclusion

Our analysis of estimated outdoor noise exposures in census block groups throughout the contiguous United States found evidence of higher noise exposures in census block groups characterized by lower SES and higher proportions of American Indian, Asian, black, and Hispanic residents. These associations were stronger in more racially segregated communities. Differences in noise exposure may have implications for more fully understanding drivers of environmental health disparities in the United States.

Acknowledgments

We thank B. Jesdale for assembling the segregation data and the University of California, Berkeley Statistical Consulting Service for statistical support. This work was supported by the Robert Wood Johnson Foundation Health & Society Scholars program (J.A.C.), the National Cancer Institute Award K99CA201542 (P.J.), and the Hewlett and Kellogg and Foundations (R. M.-F.).

References

American National Standard Institute (ANSI). 2003. ANSI Standard S12.9. Quantities and procedures for description and measurement of environmental sound – Part 4: Noise assessment and prediction of long term community response, London, UK:IHS Markit.

Anselin L. 2002. Under the hood issues in the specification and interpretation of spatial regression models. Agr Econ 27(3):247–267, 10.1111/j.1574-0862.2002.tb00120.x.

Ash M, Boyce JK, Chang G, Scharber H. 2013. Is environmental justice good for white folks? Industrial air toxics exposure in urban America. Soc Sci Q 94(3):616–636, 10.1111/j.1540-6237.2012.00874.x.

Basner M, Babisch W, Davis A, Brink M, Clark C, Janssen S, et al. 2014. Auditory and non-auditory effects of noise on health. Lancet 383(9925):1325–1332, PMID: 24183105, 10.1016/S0140-6736(13)61613-X.

Basner M, Muller U, Elmenhorst EM. 2011. Single and combined effects of air, road, and rail traffic noise on sleep and recuperation. Sleep 34(1):11–23, PMID: 21203365.

Bell ML, Ebisu K. 2012. Environmental inequality in exposures to airborne particulate matter components in the United States. Environ Health Perspect 120(2):1699–1704, PMID: 22889745, 10.1289/ehp.1205201.

Bertin M, Chevrier C, Pelé F, Serrano-Chavez T, Cordier S, Viel JF. 2014. Can a deprivation index be used legitimately over both urban and rural areas?. Int J Health Geogr 13:22, 10.1186/1476-072X-13-22.

Bivand R, Piras G. 2015. Comparing implementations of estimation methods for spatial econometrics. J Stat Soft 63(18):1–36, 10.18637/jss.v063.i18.

Bivand RS, Pebesma, EJ., Gomez-Rubio, V. 2008. Applied Spatial Data Analysis with R. New York, NY:Springer, 239–287.

Bjørnskau T. 2005. Road traffic change: A catalyst for segregation? Urban Studies 42(1):69–89, 10.1080/0042098042000309702.

Bocquier A, Cortaredona S, Boutin C, David A, Bigot A, Chaix B, et al. 2013. Small-area analysis of social inequalities in residential exposure to road traffic noise in Marseilles, France. Eur J Public Health 23(4):540–546, PMID: 22645238, 10.1093/eurpub/cks059.

Boyce JK. 1994. Inequality as a cause of environmental degradation. Ecol Econ 11(3):169–178, 10.1016/0921-8009(94)90198-8.

Brainard JS, Jones AP, Bateman IJ, Lovett AA. 2004. Exposure to environmental urban noise pollution in Birmingham, UK. Urban Studies 41(13):2581–2600, 10.1080/0042098042000294574.

Bravo MA, Anthopolos R, Bell ML, Miranda ML. 2016. Racial isolation and exposure to airborne particulate matter and ozone in understudied US populations: Environmental justice applications of downscaled numerical model output. Environ Int 92-93:247–255, PMID: 27115915, 10.1016/j.envint.2016.04.008.

Carrier M, Apparicio P, Séguin A-M. 2016. Road traffic noise in Montreal and environmental equity: What is the situation for the most vulnerable population groups?. J Transp Geogr 51:1–8, 10.1016/j.jtrangeo.2015.10.020.

Cesaroni G, Badaloni C, Romano V, Donato E, Perucci CA, Forastiere F. 2010. Socioeconomic position and health status of people who live near busy roads: the Rome Longitudinal Study (RoLS). Environ Health 9:41, PMID: 20663144, 10.1186/1476-069X-9-41.

Clark C, Crombie R, Head J, van Kamp I, van Kempen E, Stansfeld SA. 2012. Does traffic-related air pollution explain associations of aircraft and road traffic noise exposure on children’s health and cognition? A secondary analysis of the United Kingdom sample from the RANCH project. Am J Epidemiol 176(4):327–337, PMID: 22842719, 10.1093/aje/kws012.

Cushing L, Morello-Frosch R, Wander M, Pastor M. 2015. The haves, the have-nots, and the health of everyone: the relationship between social inequality and environmental quality. Annu Rev Public Health 36:193–209, PMID: 25785890, 10.1146/annurev-publhealth-031914-122646.

Dale LM, Goudreau S, Perron S, Ragettli MS, Hatzopoulou M, Smargiassi A. 2015. Socioeconomic status and environmental noise exposure in Montreal, Canada. BMC Public Health 15:205, PMID: 25885357, 10.1186/s12889-015-1571-2.

Davies HW, Vlaanderen JJ, Henderson SB, Brauer M. 2009. Correlation between co-exposures to noise and air pollution from traffic sources. Occup Environ Med 66(5):347–350, PMID: 19017692, 10.1136/oem.2008.041764.

Evans GW, Hygge S, Bullinger M. 1995. Chronic noise and psychological stress. Psychol Sci: 6(6):333–338, 10.1111/j.1467-9280.1995.tb00522.x.

Evans GW, Marcynyszyn LA. 2004. Environmental justice, cumulative environmental risk, and health among low- and middle-income children in upstate New York. Am J Public Health 94(11):1942–1944, PMID: 15514234, 10.2105/AJPH.94.11.1942.

Fecht D, Hansell AL, Morley D, Dajnak D, Vienneau D, Beevers S, et al. 2016. Spatial and temporal associations of road traffic noise and air pollution in London: Implications for epidemiological studies. Environ Int 88:235–242, PMID: 26773394, 10.1016/j.envint.2015.12.001.

Fyhri A, Klaeboe R. 2006. Direct, indirect influences of income on road traffic noise annoyance. J Environ Psychol 26(1):27–37, 10.1016/j.jenvp.2006.04.001.

Gan WQ, Davies HW, Koehoorn M, Brauer M. 2012. Association of long-term exposure to community noise and traffic-related air pollution with coronary heart disease mortality. Am J Epidemiol 175(9):898–906, PMID: 22491084, 10.1093/aje/kwr424.

Gee GC, Ponce N. 2010. Associations between racial discrimination, limited English proficiency, and health-related quality of life among 6 Asian ethnic groups in California. Am J Public Health 100(5):888–895, PMID: 20299644, 10.2105/AJPH.2009.178012.

Gehring U, Tamburic L, Sbihi H, Davies HW, Brauer M. 2014. Impact of noise and air pollution on pregnancy outcomes. Epidemiology 25(3):351–358, PMID: 24595395, 10.1097/EDE.0000000000000073.

Haines MM, Stansfeld SA, Head J, Job RF. 2002. Multilevel modelling of aircraft noise on performance tests in schools around Heathrow Airport London. J Epidemiol Community Health 56(2):139–144, 10.1136/jech.56.2.139.

Hajat A, Hsia C, O’Neill MS. 2015. Socioeconomic disparities and air pollution exposure: a global review. Curr Environ Health Rep 2(4):440–450, PMID: 26381684, 10.1007/s40572-015-0069-5.

Hammer MS, Swinburn TK, Neitzel RL. 2014. Environmental noise pollution in the United States: developing an effective public health response. Environ Health Perspect 122(2):115–119, PMID: 24311120, 10.1289/ehp.1307272.

Haralabidis AS, Dimakopoulou K, Vigna-Taglianti F, Giampaolo M, Borgini A, Dudley ML, et al. 2008. Acute effects of night-time noise exposure on blood pressure in populations living near airports. Eur Heart J 29(5):658–664, PMID: 18270210, 10.1093/eurheartj/ehn013.

Havard S, Reich BJ, Bean K, Chaix B. 2011. Social inequalities in residential exposure to road traffic noise: an environmental justice analysis based on the RECORD Cohort Study. Occup Environ Med 68(5):366–374, PMID: 21273211, 10.1136/oem.2010.060640.

Hjortebjerg D, Andersen AM, Christensen JS, Ketzel M, Raaschou-Nielsen O, Sunyer J, et al. 2016. Exposure to road traffic noise and behavioral problems in 7-year-old children: A cohort study. Environ Health Perspect 124(2):228–234, PMID: 26126294, 10.1289/ehp.1409430.

Janssen SA, Vos H, Eisses AR, Pedersen E. 2011. A comparison between exposure-response relationships for wind turbine annoyance and annoyance due to other noise sources. J Acoust Soc Am 130(6):3746–3753, PMID: 22225031, 10.1121/1.3653984.

Jesdale BM, Morello-Frosch R, Cushing L. 2013. The racial/ethnic distribution of heat risk-related land cover in relation to residential segregation. Environ Health Perspect 121(7):811–817, PMID: 23694846, 10.1289/ehp.1205919.

Jones MR, Diez-Roux AV, Hajat A, Kershaw KN, O’Neill MS, Guallar E, et al. 2014. Race/ethnicity, residential segregation, and exposure to ambient air pollution: the Multi-Ethnic Study of Atherosclerosis (MESA). Am J Public Health 104(11):2130–2137, 10.2105/AJPH.2014.302135.

Kawachi I, Berkman LF. 2003. Neighborhoods and Health. New York, NY:Oxford University Press, 46–61.

Kohlhuber M, Mielck A, Weiland SK, Bolte G. 2006. Social inequality in perceived environmental exposures in relation to housing conditions in Germany. Environ Res 101(2):246–255, PMID: 16290820, 10.1016/j.envres.2005.09.008.

Krieger N, Chen JT, Waterman PD, Soobader MJ, Subramanian SV, Carson R. 2003. Choosing area based socioeconomic measures to monitor social inequalities in low birth weight and childhood lead poisoning: The Public Health Disparities Geocoding Project (US). J Epidemiol Community Health 57(3):186–199, PMID: 12594195, 10.1136/jech.57.3.186.

Lakes T, Brückner M, Krämer A. 2014. Development of an environmental justice index to determine socio-economic disparities of noise pollution and green space in residential areas in Berlin. J Environ Plann Man 57(4):538–556, 10.1080/09640568.2012.755461.

Lam K-C, Chan P-K. 2008. Socio-economic status and inequalities in exposure to transportation noise in Hong Kong. TOENVIRSJ 2(1):107–113, 10.2174/1876325100802010107.

Laurent É. 2011. Issues in environmental justice within the European Union. Ecol Econ 70(11):1846–1853, 10.1016/j.ecolecon.2011.06.025.

Lercher P. 1996. Environmental noise and health: An integrated research perspective. Environ Int 22(1):117–129, 10.1016/0160-4120(95)00109-3.

Massey DS, Denton NA. 1989. Hypersegregation in U.S. metropolitan areas: black and Hispanic segregation along five dimensions. Demography 26(3):373–391, PMID: 2792476.

Massey DS, Denton NA. 1993. American Apartheid: Segregation and the Making of the Underclass. Cambridge, MA:Harvard University Press, 118–142.

Mennitt DJ, Fristrup KM. 2016. Influence factors and spatiotemporal patterns of environmental sound levels in the contiguous United States. Noise Cont Engng J 64(3):342–353, 10.3397/1/376384.

Miranda ML, Edwards SE, Keating MH, Paul CJ. 2011. Making the environmental justice grade: the relative burden of air pollution exposure in the United States. Int J Environ Res Public Health 8(6):1755–1771, PMID: 21776200, 10.3390/ijerph8061755.

Mohai P, Saha R. 2007. Racial inequality in the distribution of hazardous waste: A national-level reassessment. Social Problems 54(3):343–370, 10.1525/sp.2007.54.3.343.

Morello-Frosch R, Jesdale BM. 2006. Separate and unequal: residential segregation and estimated cancer risks associated with ambient air toxics in U.S. metropolitan areas. Environ Health Perspect 114(3):386–393, PMID: 16507462.

Morello-Frosch R, Lopez R. 2006. The riskscape and the color line: examining the role of segregation in environmental health disparities. Environ Res 102(2):181–196, PMID: 16828737, 10.1016/j.envres.2006.05.007.

Morello-Frosch R. 2002. Discrimination and the political economy of environmental inequality. Environ Plann C Gov Policy 20(4):477–496, 10.1068/c03r.

Münzel T, Gori T, Babisch W, Basner M. 2014. Cardiovascular effects of environmental noise exposure. Eur Heart J 35(13):829–836, PMID: 24616334, 10.1093/eurheartj/ehu030.

Murphy E, King EA. 2014. Chapter 2 – Principles of environmental noise. In: Environmental Noise Pollution. Boston, MA:Elsevier, 9–49. http://dx.doi.org/10.1016/B978-0-12-411595-8.00002-1 [accessed 8 January 2017].

Muzet A. 2007. Environmental noise, sleep and health. Sleep Med Rev 11(2):135–142, PMID: 17317241, 10.1016/j.smrv.2006.09.001.

Nega TH, Chihara L, Smith K, Jayaraman M. 2013. Traffic noise and inequality in the Twin Cities, Minnesota. Hum Ecol Risk Assess 19(3):601–619, 10.1080/10807039.2012.691409.

NHGIS (National Historical Geographic Information System Database: Version 11.0). NHGIS Database: 2010 Census state and block group boundary shapefiles (released 30 November 2010) and 2006–2010 American Community Survey (released 8 December 2011). Minneapolis:University of Minnesota. http://doi.org/10.18128/D050.V11.0 [accessed 20 December 2016].

Orban E, McDonald K, Sutcliffe R, Hoffmann B, Fuks KB, Dragano N, et al. 2016. Residential road traffic noise and high depressive symptoms after five years of follow-up: results from the Heinz Nixdorf Recall Study. Environ Health Perspect 124(5):578–585, PMID: 26606640, 10.1289/ehp.1409400.

Pastor M, Dreier P, Grigsby E, Garza J, Lopez-Garza M. 2000. Regions That Work: How Cities and Suburbs Can Grow Together. Minneapolis, MN:University of Minnesota Press, 103–111.

Poortinga W, Dunstan FD, Fone DL. 2008. Neighbourhood deprivation and self-rated health: the role of perceptions of the neighbourhood and of housing problems. Health Place 14(3):562–575, PMID: 17997343, 10.1016/j.healthplace.2007.10.003.

Pulido L. 2000. Rethinking environmental racism: White privilege and urban development in Southern California. Ann Assoc Am Geogr 90(1):12–40, 10.1111/0004-5608.00182.

Rice LJ, Jiang C, Wilson SM, Burwell-Naney K, Samantapudi A, Zhang H. 2014. Use of segregation indices, Townsend Index, and air toxics data to assess lifetime cancer risk disparities in metropolitan Charleston, South Carolina, USA. Int J Environ Res Public Health 11(5):5510–5526, PMID: 24852759, 10.3390/ijerph110505510.

Ross Z, Kheirbek I, Clougherty JE, Ito K, Matte T, Markowitz S, et al. 2011. Noise, air pollutants and traffic: continuous measurement and correlation at a high-traffic location in New York City. Environ Res 111(8):1054–1063, PMID: 21958559, 10.1016/j.envres.2011.09.004.

Sakoda JM. 1981. A generalized index of dissimilarity. Demography 18(2):245–250, PMID: 7227588.

Schomer P, Freytag J, Machesky A, Luo C, Dossin C, Nookala N, et al. 2011. A re-analysis of Day-Night Sound Level (DNL) as a function of population density in the United States. Noise Control Eng J 59(3):290–301, 10.3397/1.3560910.

Sherrill K. 2012. GIS Metrics – Soundscape Modeling, Standard Operating Procedure. Natural Resource Report NPS/NRSS/IMD/NRR—2012/596. Fort Collins, CO:National Park Service. http://irmafiles.nps.gov/reference/holding/460450 [accessed 9 January 2017].

Simpson M, Bruce R. 1981. Noise in America: Extent of the Noise Problem. (Report 550/9-81-101). Washington, DC:U.S. Environmental Protection Agency.

Sørensen M, Andersen ZJ, Nordsborg RB, Becker T, Tjønneland A, Overvad K, et al. 2013. Long-term exposure to road traffic noise and incident diabetes: a cohort study. Environ Health Perspect 121(2):217–222, PMID: 23229017, 10.1289/ehp.1205503.

Stansfeld SA, Shipley M. 2015. Noise sensitivity and future risk of illness and mortality. Sci Total Environ 520:114–119, PMID: 25804878, 10.1016/j.scitotenv.2015.03.053.

Tiefelsdorf M, Griffith DA, Boots B. 1999. A variance-stabilizing coding scheme for spatial link matrices. Environment and Planning A 31(1):165–180, 10.1068/a310165.

U.S. Census Bureau. 2010a. “TIGER/Line Shapefiles.” https://www.census.gov/geo/maps-data/data/tiger-line.html [accessed 10 June 2016].

U.S. Census Bureau. 2010b. “2010 TIGER/Line Shapefiles Technical Documentation.” http://www2.census.gov/geo/pdfs/maps-data/data/tiger/tgrshp2010/TGRSHP10SF1.pdf [accessed 9 January 2017].

U.S. DOT (U.S. Department of Transportation). 2015. “Highway Traffic Noise.” https://www.fhwa.dot.gov/environment/noise/regulations_and_guidance/analysis_and_abatement_guidance/polguide02.cfm [accessed 4 April 2016].

U.S. EPA (U.S. Environmental Protection Agency). 1977. “The Urban Noise Survey.” 550/9-77-100. Washington, DC:U.S. EPA.

van Kamp I, Davies H. 2013. Noise and health in vulnerable groups: a review. Noise Health 15(64):153–159, PMID: 23689296, 10.4103/1463-1741.112361.

van Kamp I, Job RF, Hatfield J, Haines M, Stellato RK, Stansfeld SA. 2004. The role of noise sensitivity in the noise-response relation: a comparison of three international airport studies. J Acoust Soc Am 116(6):3471–3479, PMID: 15658698.

van Kempen E, Babisch W. 2012. The quantitative relationship between road traffic noise and hypertension: a meta-analysis. J Hypertens 30(6):1075–1086, PMID: 22473017, 10.1097/HJH.0b013e328352ac54.

Walker G, Mitchell G, Fairburn J, Smith G. 2005. Industrial pollution and social deprivation: evidence and complexity in evaluating and responding to environmental inequality. Local environment 10(4):361–377, 10.1080/13549830500160842.

WHO (World Health Organization). 2009. “Night Noise Guidelines for Europe.” http://www.euro.who.int/__data/assets/pdf_file/0017/43316/E92845.pdf [accessed 5 January 2016].

WHO. 2011. “Burden of Disease from Environmental Noise.” http://www.euro.who.int/__data/assets/pdf_file/0008/136466/e94888.pdf [accessed 5 January 2016].

Maiden Voyage of the Collaborative Cross Mouse: Exploring Variability in Animals’ Response to Perchloroethylene

PDF icon PDF Version (742 KB)

  • Published: 24 July 2017

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Related EHP Article

Characterization of Variability in Toxicokinetics and Toxicodynamics of Tetrachloroethylene Using the Collaborative Cross Mouse Population

Joseph A. Cichocki, Shinji Furuya, Abhishek Venkatratnam, Thomas J. McDonald, Anthony H. Knap, Terry Wade, Stephen Sweet, Weihsueh A. Chiu, David W. Threadgill, and Ivan Rusyn

Researchers have spent more than a decade perfecting an alternative laboratory mouse model—the Collaborative Cross (CC)—that is designed to mimic the genetic diversity of the human population.1 This issue of Environmental Health Perspectives contains the first published toxicology study to be performed with CC mice: an exploration of how the toxic effects of perchloroethylene vary from animal to animal.2

Most inbred mouse strains are genetically identical, which means that individuals from the same strain respond similarly to the same chemical exposure. This limits researchers’ ability to study how differences in metabolism and other factors influence the effects of exposure and to understand how effects might vary in genetically diverse human populations.

To address that shortcoming, researchers began to develop the CC model in 2002. First, they identified eight “founder” strains from the three major laboratory and wild subspecies of Mus musculus, otherwise known as the house mouse. They crossbred these strains and then crossbred their offspring.

This unique breeding process generated inbred CC strains that each have a random sampling of the genetic diversity contained in the founder mice. Although individuals from a given CC strain are genetically identical to one another, the strains themselves are genetically distinct. Therefore, using a mouse population comprising multiple CC strains produces a facsimile of a genetically diverse human population. Several dozen genetically unique CC strains have been commercialized since the breeding effort began.3

The eight founder strains

Collaborative Cross mouse strains were developed by crossbreeding eight “founder” strains of mice. Using a mouse population comprising multiple Collaborative Cross strains produces a facsimile of a genetically diverse human population. © Jennifer Torrance.

For the present study, males from 45 CC strains were exposed to perchloroethylene, or perc, an industrial solvent that is also a common environmental contaminant. Perc is metabolized primarily to trichloroacetate (TCA), which has been shown to cause liver cancer in certain strains of rodents.4 The goal of the study was to look for strain-specific differences in perc’s toxicokinetics (i.e., how much perc and how much TCA accumulate in various organs) and its toxicodynamics (i.e., the effects caused by perc and TCA at target sites within the body).

Investigators from Texas A&M University gave the animals a single high oral dose of 1,000 mg/kg body weight. After 1, 2, 4, 12, or 24 hours, they euthanized one animal from each strain and collected serum and tissue samples for analysis. According to their results, the toxicokinetics of perc and TCA varied significantly from one strain to the next. The amounts in liver, for instance, spanned a roughly 8-fold difference from the strain with the highest level to the strain with the lowest level.

The toxicodynamics results, meanwhile, were less straightforward. The study’s first author, Joseph Cichocki, a postdoctoral research fellow at Texas A&M’s Department of Veterinary Integrative Biosciences, says that liver effects from perc were previously attributed to the formation of TCA, which in turn triggers peroxisome proliferator–activated receptors (PPARs), including PPARα, in exposed cells.5 PPARs are known to be involved in cell proliferation and in the onset of liver cancer in mice.6

Cichocki and his colleagues homed in on two PPARα-induced genes—Acox1 and Cyp4a10—and found that their induction was highly variable in the different strains. According to Cichocki, it had previously been assumed that Acox1 and Cyp4a10 expression levels would rise proportionately with increasing doses of perc. However, their results showed that this was not the case: Some mice had low Acox1 and Cyp4a10 expression levels despite being exposed to high doses of the chemical; in other mice, the opposite was true. The implication, Cichocki says, is that perc doses may not reliably predict the magnitude of their toxic effects. He asserts that additional factors, some that are related to toxicokinetics and others that are not, are responsible for interindividual variability in susceptibility to perc-induced toxicity.

“Our study was not without limitations,” Cichoki says. “But the takeaway message is that by using CC mice, we could characterize and quantify interindividual variability across the population.”

Steven Munger, an assistant professor at Jackson Laboratory, where development of the CC mouse first began, says that traditional single-strain toxicology studies provide only a limited view of the potential responses to a given chemical. “With Cichocki’s study, we’ve seen a distribution that looks a lot more like what you’d expect to see in exposed humans,” he says. “Some strains are outliers for particular phenotypes, just as a minority of people might exhibit a response that many others do not. And we still need to account for these people in our risk assessments.”

Cichocki explains that risk assessors ordinarily use default “uncertainty factors” to estimate interindividual variation in chemical effects. By providing a glimpse into the biological nature of that variation, experts anticipate that experiments with CC mice could enable risk assessors to replace those default values with real data. “This is what we’re concluding with our paper,” Cichocki says. “We’re trying to generate the experimental data for population-level variability that regulators need.”


Charles W. Schmidt, MS, an award-winning science writer from Portland, ME, writes for Scientific American, Science, various Nature publications, and many other magazines, research journals, and websites.

References

1. Threadgill DW, Churchill GA. 2012. Ten years of the Collaborative Cross. G3 (Bethesda) 2(2):153–156, PMID: 22384393, 10.1534/g3.111.001891.

2. Cichocki JA, Furuya S, Venkatratnam A, McDonald TJ, Knap AH, Wade T, et al. 2017. Characterization of variability in toxicokinetics and toxicodynamics of tetrachloroethylene using the Collaborative Cross mouse population. Environ Health Perspect 125(5):057006, PMID: 28572074, 10.1289/EHP788.

3. Schmidt CW. 2015. Diversity Outbred: a new generation of mouse model. Environ Health Perspect 123(3):A64–A67, PMID: 25730842, 10.1289/ehp.123-A64.

4. IARC (International Agency for Research on Cancer). 2014. Trichloroethylene, tetrachloroethylene, and some other chlorinated agents. Monogr Eval Carcinog Risk Hum 106:1–525, http://monographs.iarc.fr/ENG/Monographs/vol106/mono106.pdf [accessed 13 July 2017].

5. Maloney EK, Waxman DJ. 1999. trans-Activation of PPARα and PPARγ by structurally diverse environmental chemicals. Toxicol Appl Pharmacol 161(2):209–218, PMID: 10581215, 10.1006/taap.1999.8809.

6. Bakiri L, Wagner EF. 2013. Mouse models for liver cancer. Mol Oncol 7(2):206–223, PMID: 23428636, 10.1016/j.molonc.2013.01.005.

Northern Trek: The Spread of Ixodes scapularis into Canada

PDF icon PDF Version (384 KB)

  • Published: 24 July 2017

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Related EHP Article

Expansion of the Lyme Disease Vector Ixodes scapularis in Canada Inferred from CMIP5 Climate Projections

Michelle McPherson, Almudena García-García, Francisco José Cuesta-Valero, Hugo Beltrami, Patti Hansen-Ketchum, Donna MacDougall, and Nicholas Hume Ogden

For a decade Nicholas Ogden, a researcher at the National Microbiology Laboratory of the Public Health Agency of Canada, has tracked the northern expansion of the deer tick (Ixodes scapularis), the vector for Lyme disease. He has found a strong correlation between rising winter temperatures and the spread of the tick population.1,2 Now Ogden has collaborated with Hugo Beltrami, Canada Research Chair in Climate Dynamics at St. Francis Xavier University, and other researchers to forecast the range expansion of I. scapularis under a greater number of possible climate scenarios.3

Lyme disease was first identified in coastal Connecticut in 1976, and the bacterium that causes it, Borrelia burgdorferi, was isolated in 1982.4 Eastern Canada’s first infected ticks were found on the Ontario shore of Lake Erie in the early 1990s.5 The ticks have since expanded their range farther north into Ontario and parts of Manitoba, Quebec, New Brunswick, and Nova Scotia, Ogden says. The number of reported Lyme disease cases in Canada is rising steadily, from 144 in 2009 to 917 in 2015.6

The inference that temperature thresholds have a strong impact on tick survival fits with a growing body of evidence showing that the ranges of Ixodes ticks in Europe are limited by cold temperatures.7 In a 2014 study, Ogden and his colleagues used a single climate model to forecast the spread of the tick into Canada.1 They concluded that climatic conditions suitable for I. scapularis populations to expand steadily northward would likely occur during the coming century.

Photo of an Ixodes scapularis tick on a blade of grass.

Ticks do not jump, fly, or drop onto passersby. Instead, they wait on vegetation with their front legs raised in a “questing” pose. When an appropriate host brushes past, the tick hitches a ride and attaches itself for a blood meal. © Juniors Bildarchiv GmbH/Alamy Stock Photo.

The basic needs of I. scapularis include woodland habitat and an assortment of vertebrate hosts to bite. After hatching, ticks pass through three life stages and require a blood meal to fuel their development from one to the next. As larvae and nymphs, the ticks most often obtain these meals from white-footed mice or other small rodents, although they occasionally latch onto other creatures—a raccoon, a bird, or an unfortunate human. Adult ticks feed primarily on white-tailed deer.8

Deer are growing in numbers and expanding their range to the north, as are white-footed mice.9 In addition, Ogden says, recent warming has occurred in southern Canadian regions with new influxes of ticks, which are moving in a geographic pattern consistent with temperature being an important factor in their becoming established.

Under even the most optimistic scenario, in which the increase in global average temperature is limited to 1.5°C above preindustrial temperatures, the authors’ models showed Lyme disease continuing to spread in Canada. They conclude that people in Nova Scotia and in southern Ontario—home to more than 85% of the provincial population—will need to be aware of and adapt to the risk of bites from infected ticks. Under the worst-case scenario modeled, in which global greenhouse gas emissions are not curtailed, the authors estimate I. scapularis will spread into northern Ontario, a region not yet colonized by deer ticks.

“This study is an extension of previous work published in 20141 showing the predicted expansion of the distribution of the Lyme disease tick vector into Canada,” says Maria Diuk-Wasser, a professor at Columbia University who focuses on the emergence of vector-borne diseases. “Although the results are not qualitatively different, it represents an improvement on the previous study by incorporating the full range of and most up-to-date climate models and emission scenarios.” Importantly, she says, the new study accounts for the inherent uncertainty in such models and scenarios but also indicates that an increased risk can be expected in any event.

Canada’s public health officials track the leading edge of the tick’s range expansion in several ways: through laboratory identification of ticks found on patients by doctors and veterinarians; by conducting surveys in which a large cloth is dragged across a woodland floor, picking up any ticks that are questing for a host to bite; and by compiling data on reported cases of human infection. The Canadian government has provided information on how to avoid tick bites and identify the symptoms of Lyme disease.10 “We’ll all need to participate in adapting to the tick’s arrival,” says Ogden.


Sharon Levy based in Humboldt County, CA, has covered ecology, evolution, and environmental science since 1993. She is at work on the book The Marsh Builders: Wetlands in the Fight for Clean Water.

References

1. Ogden N, Radojević M, Wu X, Duvvuri VR, Leighton PA, Wu J. 2014. Estimated effects of projected climate change on the basic reproductive number of the Lyme disease vector Ixodes scapularis. Environ Health Perspect 122(6):631–638, PMID: 24627295, 10.1289/ehp.1307799.

2. Gabriele-Rivet V, Arsenault J, Badcock J, Cheng A, Edsall J, Goltz J, et al. 2015. Different ecological niches for ticks of public health significance in Canada. PLoS One 10(7):e0131282, PMID: 26131550, 10.1371/journal.pone.0131282.

3. McPherson M, García-García A, Cuesta-Valero FJ, Beltrami H, Hansen-Ketchum P, MacDougall D, et al. 2017. Expansion of the Lyme disease vector Ixodes scapularis in Canada inferred from CMIP5 climate projections. Environ Health Perspect 125(5):057008, PMID: 28599266, 10.1289/EHP57.

4. Steere A, Coburn J, Glickstein L. 2004. The emergence of Lyme disease. J Clin Invest 113(8):1093–1101, PMID: 15085185, 10.1172/JCI200421681.

5. Barker IK, Surgeoner GA, Artsob H, McEwen SA, Elliott LA, Campbell GD, et al. 1992. Distribution of the Lyme disease vector, Ixodes dammini (Acari: Ixodidae) and isolation of Borrelia burgdorferi in Ontario, Canada. J Med Entomol 29(6):1011–1022, PMID: 1460617, 10.1093/jmedent/29.6.1011.

6. Government of Canada. Surveillance of Lyme Disease [website]. Updated 20 September 2016. https://www.canada.ca/en/public-health/services/diseases/lyme-disease/surveillance-lyme-disease.html [accessed 6 March 2017].

7. Ostfeld RS, Brunner JL. 2015. Climate change and Ixodes tick-borne diseases of humans. Philos Trans R Soc Lond B Biol Sci 370(1665):20140051, PMID: 25688022, 10.1098/rstb.2014.0051.

8. U.S. Centers for Disease Control and Prevention. Lifecycle of Blacklegged Ticks [website]. Updated 15 November 2011. https://www.cdc.gov/lyme/transmission/blacklegged.html [accessed 6 March 2017].

9. Simon JA, Marrotte RR, Desrosiers N, Fiset J, Gaitan J, Gonzalez A, et al. 2014. Climate change and habitat fragmentation drive the occurrence of Borrelia burgdorferi, the agent of Lyme disease, at the northeastern limit of its distribution. Evol Appl 7(7):750–764, PMID: 25469157, 10.1111/eva.12165.

10. Government of Canada. Lyme Disease [website]. Updated 17 February 2017. https://www.canada.ca/en/public-health/services/diseases/lyme-disease.html [accessed 6 March 2017].

Weekly Personal Ozone Exposure and Respiratory Health in a Panel of Greek Schoolchildren

Author Affiliations open

12nd Pulmonary Department, ATTIKON University Hospital, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece

2Department of Hygiene, Epidemiology and Medical Statistics, School of Medicine, National and Kapodistrian University of Athens, Athens, Greece

3Pulmonary Department, G. Papanikolaou Hospital, Medical School, Aristotle University of Thessaloniki, Thessaloniki, Greece

4School of Chemical Engineering, National Technical University of Athens, Athens, Greece

5Department of Primary Care & Public Health Sciences and Environmental Research Group, King’s College, London, UK

PDF icon PDF Version (312 KB)

  • Background:
    The association of ozone exposure with respiratory outcomes has been investigated in epidemiologic studies mainly including asthmatic children. The findings reported had methodological gaps and inconsistencies.
    Objectives:
    We aimed to investigate effects of personal ozone exposure on various respiratory outcomes in school-age children generally representative of the population during their normal activities.
    Methods:
    We conducted a panel study in a representative sample of school-age children in the two major cities of Greece, Athens and Thessaloniki. We followed 188, 10- to 11-y-old, elementary school students for 5 wk spread throughout the 2013–2014 academic year, during which ozone was measured using personal samplers. At the end of each study week, spirometry was performed by trained physicians, and the fractional concentration of nitric oxide in exhaled air (FeNO) was measured. Students kept a daily time–activity–symptom diary and measured PEF (peak expiratory flow) using peak flow meters. Mixed models accounting for repeated measurements were applied.
    Results:
    An increase of 10 μg/m3 in weekly ozone concentration was associated with a decrease in FVC (forced vital capacity) and FEV1 (forced expiratory volume in 1 s) of 0.03 L [95% confidence interval (CI): −0.05, −0.01] and 0.01 L (95% CI: −0.03, 0.003) respectively. The same increase in exposure was associated with a 11.10% (95% CI: 4.23, 18.43) increase in FeNO and 19% (95% CI: −0.53, 42.75) increase in days with any symptom. The effect estimates were robust to PM10 adjustment. No inverse association was found between ozone exposure and PEF.
    Conclusions:
    The study provides evidence that airway inflammation and the frequency of respiratory symptoms increase, whereas lung function decreases with increased ozone exposure in schoolchildren. https://doi.org/10.1289/EHP635
  • Received: 09 June 2016
    Revised: 08 March 2017
    Accepted: 13 March 2017
    Published: 21 July 2017

    Please address correspondence to A. Karakatsani, 2nd Pulmonary Department, ATTIKON University Hospital, School of Medicine, National and Kapodistrian University of Athens, 1, Rimini St., 124 62 Haidari, Greece. Telephone: 30-210-5831184. Email: annakara@otenet.gr, akarakats@med.uoa.gr

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Introduction

Ozone (O3), a very reactive gas and strong oxidant, is found as a secondary pollutant in the troposphere. Although, its presence in the stratosphere is essential for life because it filters harmful ultraviolet radiation, increased concentrations of O3 in the air we breathe have been linked to adverse health effects mainly concerning the respiratory system (WHO 2013). A number of controlled human exposure studies have demonstrated changes in FEV1 and respiratory symptoms (Adams 2006; Horstman et al. 1990; Kim et al. 2011) as well as lung inflammation (Devlin et al. 1991; Kim et al. 2011), but those included healthy adults, not children. In an attempt to investigate the acute effects of ozone under real-world exposures several camp studies have been conducted involving sequential measurements of lung function and ambient ozone measurements on children attending summer camps (Burnett et al.1990; Higgins et al. 1990; Raizenne et al.1989; Spektor et al. 1988; Spektor et al. 1991). However, despite the evidence they provide of an association between daily outdoor O3 concentrations and decreased lung function (Kinney et al. 1996), the lack of personal measurements, variations in data reporting and analysis as well as concerns about potential confounding by other pollutants that may co-vary with O3, limit their use in risk assessment. Therefore, so far there is lack of studies conducted in children under real living conditions, at current concentrations of ozone, using personal exposure assessment.

Children’s lungs are considered to be more sensitive and vulnerable to air pollution as they have, even at rest, a larger surface area for body size compared with adults and are still growing. Children breathe more air per unit of body weight and have different breathing patterns than adults that alter the deposition and toxicity of inhaled gases and particles (Arcus-Arth and Blaisdell 2007; Bateson and Schwartz 2008; Warren et al. 1990). Moreover, because they spend more time outdoors and are more active, they have higher ventilation rates leading to increased intake and deposition of pollutants (Bateson and Schwartz 2008). So far, epidemiological studies on the effects of ozone on children’s lung function and respiratory symptoms have mainly focused on asthmatics and have reported inconsistent results (Castro et al. 2009; Declercq and Macquet 2000; Jalaludin et al. 2000; Just et al. 2002; Li et al. 2012; Mortimer et al. 2002; Scarlett et al. 1996). However, response to ozone exposure displays individual variability that cannot be explained by underlying morbidity (Schelegle et al. 2012; WHO 2013), and thus, from a public health point of view, it is critical to determine to what extent ambient O3 exposure is responsible for adverse respiratory effects in the general population of children.

We conducted a panel study (Respiratory Effects of Ozone Exposure in children; RESPOZE) in a representative sample of the general population of schoolchildren in the two major cities of Greece, Athens (state capital) and Thessaloniki. Greece provides a unique opportunity to study O3 health effects because of its sunny, warm weather as well as the high concentrations of precursor pollutant emissions encountered in some of its cities.

In this paper we examine the association of weekly personal ozone measurements with lung function parameters, airway inflammation, and respiratory symptoms.

Materials and Methods

Study Design: Subjects

A panel study was conducted, in the two largest and most polluted cities of Greece, Athens and Thessaloniki (total population of about 4 million). The research was approved by the Ethics Committee of the National and Kapodistrian University of Athens, and it complied with all relevant national and local regulations. The study population consisted of all children in the fifth grade (age 10–11 y) in state elementary schools representative of the general population of students in that age group. State schools were chosen because all attending students are required to live in the school neighborhood. A two-stage sampling design was applied. First, state elementary schools located near a fixed monitoring site (within 2 km), were identified, and permission to implement the project was obtained from the Ministry of Education. In the second stage, the research team visited all schools and informed fifth grade school children in class about the project. Informed consent was obtained from the parents of children willing to participate and the final sample was formed. We had estimated that a sample of 200 school children (100 from Athens and 100 from Thessaloniki) would provide power >90% to detect a decrease in FEV1 (forced expiratory volume in 1 s) of 0.14% per 10 μg/m3 in O3 exposure, at the 5% significance level, considering that the overall proportion of variance explained would be 41%. In each city 60% of the sample was drawn from high-O3 areas (suburbs), to ensure that we had enough relatively highly exposed children, and 40% from low-O3 areas (city centers, where ozone is scavenged by primary pollutants). All field work was implemented during the academic year 2013–2014 and consisted of five weeks of intensive follow-up per study participant: two in the fall period, one in winter, two in spring/summer. During each period, we used a staged entry of the participants for practical reasons. As a result, the period of data collection was longer, and thus, the likelihood for uncontrolled factors or unexpected events to influence the association between air pollution and health decreased (Roemer et al. 2000). Before the start of the field work, trained interviewers visited the schoolchildren’s families at home to fill in an extensive questionnaire with information on demographic, life style, and residential characteristics and the medical history of the child. During this visit they also provided a peak flow meter (Mini-Wright, Clement Clarke International Ltd.) and trained the student on how to use it. A team of three field workers, including one pediatrician or pulmonologist, visited each school twice for each field work week. At the first visit an O3 personal sampler (Ogawa USA) and a time–activity diary (TAD) were given to the participating students. At the second visit, a week later (same weekday), the O3 personal sampler and completed TADS were collected, spirometry was performed, the fractional concentration of nitric oxide in exhaled air (FeNO) was measured and a 24-h dietary recall questionnaire was filled in by interview. [See “TAD-PEF (Peak Expiratory Flow): Symptoms”; “Spirometry”; “FeNO”; “Ozone Exposure Assessment”].

TAD-PEF (Peak Expiratory Flow): Symptoms

Participating schoolchildren were instructed to complete a TAD collecting information on his or her location (at home, outdoors, indoors, but not at home and in transport) at 15-min intervals for each day of the field work period. In the TAD, they additionally recorded a total of nine PEF recordings per day, performed with the Mini-Wright peak flow meter three times every morning, afternoon, and night before the use of any medication. In addition, at the end of each day they were asked to record all their daily symptoms, namely cough, wheezing, breathing difficulties (dyspnea), fever, and stuffy nose; school absenteeism; and daily medication use (yes vs. no, as well as identification of medicine).

Spirometry

Spirometry was performed at school at the end of each of the five intensive field work weeks. Thus, a total of five spirometric assessments were available per schoolchild. All spirometry maneuvers were performed by the same pulmonologists or by the specially trained pediatrician in a separate office provided by each school that was quiet, well ventilated and had a standard temperature independent of season. Children were tested in a seated position and with nose clips in place. Portable spirometers (Spiropalm, Cosmed Srl, Italy) compatible with ATS (American Thoracic Society) and ERS (European Respiratory Society) requirements were used (Miller et al. 2005). At least two technically satisfactory maneuvers had to be performed with a difference between the two highest values of FVC (forced vital capacity) and FEV1 of <5% of each other. The best FVC and FEV1 were recorded independently of the order of the maneuver, whereas the values for the other parameters were chosen from the maneuver having the largest sum of FVC and FEV1 (Miller et al. 2005). PEF and FEF25–75% were also recorded.

FeNO

A measurement of FeNO was performed at school at the end of each field study week just before spirometry in the same room where spirometry was performed (see above), thus, a total of five measurements were available per schoolchild. The NIOX MINO (Aerocrine, Solna, Sweden), an instrument that has also a scrubber to eliminate ambient NO from the subject’s sample, was used for all measurements according to American Thoracic Society (ATS) and European Respiratory Society (ERS) recommendations (ATS/ERS 2005).

Ozone Exposure Assessment

To assess O3 weekly exposure, each student was provided with a personal Ogawa O3 sampler (Ogawa & Co. USA Inc., Pompano Beach, FL) at the first visit of each field work week and was instructed on how to wear and handle it. The sampler was collected during the second visit at the end of the week. One sampler was placed in the outdoor space of each participating school also providing a weekly measurement. Additionally, concentrations of daily ambient O3 as well as PM10 (particulate matter with aerodynamic diameter <10 μm) and NO2 (nitrogen dioxide) were obtained by the nearest fixed monitoring site from the state network (http://www.ypeka.gr/).

Confounder Data

A possible learning or fatigue effect as well as a growth factor throughout the academic year was taken into account by including a time trend as a variable taking values 1–5 for each consecutive week (week ID). Additionally, all models were adjusted for sex, city (i.e., Thessaloniki vs. Athens) and area (high- vs. low-ozone area, as used in the sampling frame), father’s education (years) as socioeconomic index, outdoor ambient temperature (°C, weekly average and alternatively same day temperature), consumption of antioxidant foods (yes/no, in the corresponding 24-h recall), time spent outdoors (hours/day, from data recorded in the corresponding week TAD), and medication use (yes/no, during the corresponding week). Height (cm) and weight (kg) were adjusted in models with lung function outcomes. Models were additionally adjusted for PM10 levels from the nearest fixed monitoring site.

Quality Assurance and Quality Control

Standard operating procedures (SOPs) were applied for air pollution and health measurements. A training workshop for all field workers was organized before the start of the fieldwork. It should be noted that especially for the TAD, which are demanding on the subject’s time and accuracy, scrupulous data quality procedures were applied as well as controls with manual inspections of the data and cross references from the field workers’ progress diaries.

Statistical Analyses

Random effects models, incorporating a random intercept, Stata Statistical Software (Release 12; StataCorp LP), were used to account for the repeated measurements, for each child. We applied multiple linear regression models when we investigated O3 effects on continuous variables: PEF measurements from the Mini-Wright (taken as the average of seven daily values calculated as the mean of the three maximum values from the morning, afternoon, and evening measurements), spirometry indices FVC, FEV1, PEF, and FEF25–75% (forced expiratory flow at 25–75%) as well as FeNO measurements, which were log-transformed because the distribution was skewed. Poisson models were used for counts: the number of days per week when any symptom occurred and the number of days when the child was absent from school. In all models, we adjusted for all the potential confounders mentioned above. Because ozone concentrations are higher in spring–summer, we repeated the analysis for this period only. In addition, because asthmatic children may present special characteristics, we repeated the analysis excluding this subgroup using two alternative definitions. The first definition was based on parents’ reporting a doctor-diagnosed asthma. The second definition further included those students having FEV1/FVC and/or FEF25−75%<LLN (lower limit of normal) at least once during the study period (provided that all spirometry maneuvers were performed according to the project’s protocol), which provides evidence of obstructive ventilatory defect (Lougheed et al. 2012; Lum and Stocks 2010) and may denote (undiagnosed) asthma. We also conducted additional sensitivity analyses in healthy children only, excluding alternatively the two groups (asthmatic and asthmatic plus those presenting obstructive ventilatory defect) according to the definitions above.

Results are presented per 10-μg/m3 increase in O3 levels, representing approximately an interquartile range of the weekly measurements and thus, allowing comparability with previously published results based on measurements from fixed sites monitors. In sensitivity analyses we tested for heterogeneity between children, which included adding random slopes to our models.

Results

The final sample included 188 school-age children (97 in Athens and 91 in Thessaloniki). Table 1 shows personal, medical, socioeconomic characteristics, and respiratory health indices of the study population, by city and ozone concentration area (as used for the sampling scheme), during the academic year 2013–2014. Based on demographic data children residing in the two cities were similar. Parents reported a doctor-diagnosed asthma for 21 schoolchildren (7 in Athens and 14 in Thessaloniki). Of the remaining 167 children without a reported history of asthma, 22 (10 in Athens and 12 in Thessaloniki) had FEV1/FVC and/or FEF25–75%<LLN (lower limit of normal) for at least once during the study period.

Table 1. Personal, medical, socioeconomic characteristics, and respiratory health indices (presented as 5-wk averages) of the study population, by city and ozone concentration area.
Athens Thessaloniki
Ozone concentration areaa
Characteristic/respiratory health outcome Low (n=37) High (n=60) Low (n=33) High (n=58)
Boys [n (%)] 22 (59.5) 28 (46.7) 14 (42.4) 29 (50.0)
Age (y) 10.3±0.3 10.3±0.3 10.4±0.4 10.4±0.3
Height (cm) 147.2±6.7 143.5±7.7 145.9±9.5 144.3±7.3
BMI (kg/m2) 18.2±2.8 18.5±3.6 18.0±3.4 18.0±2.8
Father’s education (y) 14.0±2.8 15.2±3.7 15.3±3.4 14.1±3.3
Working father [yes; n (%)] 29 (78.4) 55 (91.7) 27 (81.8) 52 (89.7)
Asthma [n (%)] 2 (5.4) 5 (8.3) 4 (12.1) 10 (17.2)
FVC (L) 2.5±0.4 2.4±0.3 2.6±0.4 2.5±0.4
FEV1 (L) 2.2±0.4 2.1±0.3 2.2±0.4 2.2±0.4
PEF (L/s) 4.7±0.8 4.8±0.8 4.9±0.9 4.8±0.9
FEF25–75% (L/s) 2.6±0.7 2.5±0.5 2.6±0.8 2.5±0.6
FeNO (ppb) 17.7±17.8 15.6±12.3 15.7±13.8 16.5±15.5
PEF-Mini-Wrightb (L/min) 293.5±53.9 287.8±45.3 300.8±55.8 297.5±58.3
Students with any symptom during the study periodc [n (%)] 29 (78.4) 40 (66.7) 27 (84.4) 45 (79.0)
Students with at least 1 d absence during the study period [n (%)] 18 (48.7) 31 (51.7) 12 (37.5) 22 (38.6)

Note: Data are presented as mean±SD or n (%), unless otherwise stated.

aThis definition of high- and low-concentration areas was based on previous years and used as a basis for the sampling procedure.

bWeekly averages from daily peak expiratory flow values (measured with the personal Mini-Wright peak flow meter) were used.

cAt least one symptom during the study period.

In Table 2 descriptive characteristics of the air pollutants and temperature by city and high- or low- O3 area (as used in the sampling scheme) are presented. O3 personal measurements were much lower than outdoor measurements at schools or fixed sites, reflecting the amount of time spent indoors, where ozone concentrations are lower. Furthermore, O3 as well as PM10 concentrations and temperature were higher in Athens than in Thessaloniki. Personal ozone and outdoor school ozone concentrations were in a range of 25–40% higher in high-versus low-ozone areas in both cities. In Athens, PM10 concentrations were lower by 20% in high-compared with low-ozone areas but, in contrast, in Thessaloniki they were 5% higher in high-ozone compared with low-ozone areas. Therefore, in Athens there was a larger contrast in both O3 and PM10 concentrations between high- and low-ozone areas. In additional, contrasts were higher in outdoor school measurements compared with personal measurements.

Table 2. Weekly air pollution and temperature levels by city.
Athens Thessaloniki
Ozone concentration area
Environmental indicator Low (n=37) High (n=60) Low (n=32) High (n=57)
O3 personal measurements (μg/m3) 8.2±6.7 10.8±7.8 4.7±4.8 5.9±6.6
O3 outdoor at schools (μg/m3) 45.9±14.7 64.3±20.1 35.2±20.7 45.6±19.4
O3 at fixed sites (μg/m3) 24.6±13.8 63.8±16.4 36.3±16.7 41.3±18.5
PM10 at fixed sitesa (μg/m3) 28.9±7.4 23.1±7.1 18.9±3.8 21.0±9.3
Temperature (°C) 17.6±2.5 17.6±2.5 14.6±3.9 13.7±3.8

Note: Data are presented as mean±SD. The high- or low-O3 concentration areas were used for the sampling scheme.

aAverage of 24-h values.

Concerning the assessment of health effects, a negative association between weekly O3 exposure and FVC was observed (both in the one and two pollutants models including PM10). An increase of 10 μg/m3 in weekly O3 concentration was associated with a decrease in FVC of 0.03 L or 30 mL (95% CI: −0.05 L, −0.01 L). When we restricted the analyses to the spring–early summer period the negative effect of O3 on FVC persisted. The results concerning the weekly O3 effect on FEV1 were similar. A 10-μg/m3 increase in weekly O3 concentration was associated with a decrease in FEV1 of 0.01 L or 10 mL (95% CI: −0.03 L, 0.003 L). Peak expiratory flow, as assessed by spirometry, and FEF25–75%, were not associated with weekly O3 exposure in any period of analyses (Table 3). Weekly averages from daily values of PEF, taken with the Mini-Right flow meter, were positively associated with O3 and this association remained when we restricted the analyses to the spring–early summer period (Table 3). Inclusion of PM10 did not change the above-mentioned results.

Table 3. Mean change in lung function indices associated with weekly increase of 10 μg/m3 in O3 exposure measured by personal monitors in 178 school children.
O3 effect per 10 μg/m3
All study periods Spring–early summer period
Health outcome Models β-coefficient (95% CI) Wald statistic β-coefficient (95% CI) Wald statistic
Spirometry indices
 FVC (L) Ozone only −0.03 (−0.05, −0.01)* 7.25* −0.02 (−0.04, 0.003) 2.95
Ozone+PM10a −0.03 (−0.05, −0.004)* 5.55* −0.02 (−0.04, 0.01) 2.43
 FEV1 (L) Ozone only −0.01 (−0.03, 0.003) 2.65 −0.02 (−0.04, 0.01) 1.70
Ozone+PM10a −0.01 (−0.03, 0.01) 0.70 −0.01 (−0.04, 0.01) 1.26
 PEF (L/s) Ozone only 0.004 (−0.07, 0.08) 0.01 −0.05 (−0.14, 0.04) 0.71
Ozone+PM10a 0.02 (−0.06, 0.10) 0.23 −0.05 (−0.14, 0.04) 0.62
 FEF25–75% (L/s) Ozone only 0.01 (−0.04, 0.05) 0.07 0.03 (−0.04, 0.09) 0.39
Ozone+PM10a 0.02 (−0.02, 0.07) 1.05 0.03 (−0.04, 0.10) 0.41
Self measured peak expiratory flow (Mini-Wright)
 PEFb (L/min) Ozone only 4.80 (1.17, 8.43)* 6.73* 3.45 (−1.14, 8.05) 2.17
Ozone+PM10a 5.34 (1.62, 9.06)* 8.08* 3.37 (−1.39, 8.13) 2.03
Exhaled nitric oxide fraction (FeNO) log-transformed
FeNO (ppb)c Ozone only 11.10 (4.23, 18.43)* 10.43* 11.79 (2.61, 21.80)* 6.49*
Ozone+PM10a 9.48 (2.46, 16.98)* 7.17* 11.77 (2.54, 21.82)* 6.41*

Note: Results of mixed models adjusting for sex, height, weight, exposure area (low/high), study area (Athens/Thessaloniki), years of father’s education, air temperature (7-d average), mean time spent outdoors daily, citrus fruits consumption (yes/no), and week ID.

a7-d average, measurements from nearest fixed site.

bWeekly averages from daily values (daily value is the average of three maximum values of morning, noon, and night measurements).

cIn this model we did not adjust for height, weight, and week ID.

*p<0.05%.

Analysis of the association of FeNO with O3 showed that a 10-μg/m3 increase in weekly personal O3 concentration was associated with an increase in FeNO values by 11.10% (95% CI: 4.23%, 18.43%). The direction of the effect and the magnitude of the association was practically the same after adjusting for PM10 [9.5% (95% CI: 2.46%, 16.98%)] (Table 3). Associations with FeNO remained significant when we restricted the analysis in the spring–early summer study period 11.79% (95% CI: 2.54%, 21.82%).

An increase of 19% (95% CI: −0.53%, 42.75%), (21%, 95% CI: 0.42%, 45.66%, after adjusting for PM10) in the number of days with at least one symptom within each of the five weeks was associated with a 10-μg/m3 increase in weekly ozone exposure (Table 4). The association was smaller and nonsignificant when only the spring–summer period was considered (data not shown). No association between weekly O3 personal exposure with “any absenteeism” (i.e., the number of days the student was absent from school during each of the five weeks) was observed (Table 4). In sensitivity analyses, the direction of the effects remained unchanged when the groups of asthmatic, asthmatic plus those having obstructive ventilatory defect, or healthy children were analyzed separately. We observed no significant heterogeneity in the effects between children when we allowed for random slopes in our models (data not shown). Results remained unchanged when the same day temperature was included in the models. In addition, no effect modification by area (high or low ozone) or city (Athens or Thessaloniki) was detected.

Table 4. Association between the number of days with any respiratory symptom or absenteeism with weekly O3 exposure measured with personal monitors in the panel of 178 school children.
Health outcome Model All study periods % change per 10 μg/m3 O3 (95% CI) Wald statistic
Number of days with any symptoma Ozone only 19.16 (−0.53, 42.75) 3.62
Ozone+PM10b 20.94 (0.42, 45.66)* 4.02*
Number of days of absenteeismc Ozone only −28.95 (−55.25, 12.79) 2.10
Ozone+PM10b −31.32 (−56.88, 9.39) 2.50

Note: Results of Poisson models adjusting for sex, exposure area (low/high), study area (Athens/Thessaloniki), years of father’s education, air temperature (7-d average), mean time spent outdoors daily, citrus fruits consumption (yes/no), and week ID.

aNumber of days within the week that any symptom occurred.

b7-d average, measurements from nearest fixed site.

cNumber of days within the week that the student was absent from school.

*p<0.05%.

Discussion

In this panel study, we found consistent associations between increased weekly O3 personal exposure and a decrease mainly in FVC and also FEV1, in 10-y-old school children following their normal daily schedules not modified by their participation in the study. We also observed corresponding increases in the number of days that any respiratory symptom occurred and in FeNO, a marker of airway inflammation. Our sample included asthmatic children (about 7%), but separate analyses by asthmatic and nonasthmatic children did not show a modification in the effect estimates. The students were sampled from low- and high-O3 areas and from two major cities, but no effect modification by area or city was detected.

Previous longitudinal studies investigating the effect of short-term exposure to O3 on children’s respiratory health involved mainly asthmatic children and relied on fixed monitoring sites rather than on personal measurements (Li et al. 2012). Most studies used PEF as lung function parameter. Moreover, among the few panel studies concerning healthy children no one, to the best of our knowledge, combined inflammatory response and change in lung function with personal measurements of O3 exposure (Barraza-Villarreal et al. 2008; Chen et al. 2015).

One very consistent and robust finding in our study is the increase in FeNO that persisted after controlling for PM10, probably reflecting the capacity of O3 to induce airway inflammation (Mudway and Kelly 2000) even in healthy children. Very few studies investigated exposure to O3 and FeNO levels. A panel study, involving school children living in Mexico City, demonstrated an association of FeNO with acute exposure to traffic-related air pollutants in both asthmatics and nonasthmatics (Barraza-Villarreal et al. 2008). However, the 8-h moving average outdoor concentration of O3 was associated with FeNO in asthmatic children only. Another study investigated the effects of exposure to several pollutants on FeNO in asthmatic children in the Mexican-U.S. border and found respiratory effects attributed to PM (several size fractions) and NO2 but not of O3 (Sarnat et al. 2012). Exposure assessment in these studies was based on fixed monitoring sites and passive samplers located in schools, resulting perhaps in less accurate assessment of O3 exposure.

Furthermore, we found significant negative associations, robust after adjusting for potential confounders and more strongly during warmer days, between O3 weekly personal exposure and FVC as well as FEV1 (but of borderline significance).

Based on our results, the mean decrease found in FVC (0.0 3L, per 10-μg/m3 increase in personal O3 exposure) may be of clinical significance if repeated exposures lead to a more permanent adverse effect. Moreover, the range of personal exposure measurements is about 40 μg/m3, depending not only on ambient concentrations but also on the child’s time spent outdoors, and in the highest exposure range the weekly decrease in FVC may be clinically relevant. Concerning our findings on FeNO, it follows that a 20% increase in FeNO is associated with 20-μg/m3 increase in O3 personal exposure (an exposure contrast well within the range of measured exposures from 1 to 42 μg/m3) considered to indicate an effect that may have clinical importance (Dweik et al. 2011).

An unexpected finding of our study was the significant positive associations of O3 with the weekly averages of daily PEF self-assessed values using Mini-Wright peak flow meters. PEF is the most usual measurement in panel studies evaluating the effect of air pollution on children’s lung function but results have been inconsistent across studies. Some panel studies reported PEF decrements in healthy (Declercq and Macquet 2000; Höppe et al. 2003; Li et al. 2012) and asthmatic school children (Just et al. 2002; Li et al. 2012; Mortimer et al. 2002; Romieu et al.1997) as well as enhanced daily PEF variability in the asthmatics (Just et al. 2002; Li et al. 2012), whereas others found no effect (Peacock et al. 2003; Scarlett et al. 1996). Likewise, in the older studies of children attending summer camps in the United States and Canada, inconsistent results associating ambient O3 exposure with PEF were observed, with the largest study (at Pine Springs Ranch, east of Los Angeles) reporting a positive and statistically significant association (Kinney et al. 1996). In a Brazilian study, Castro et al. (2009) observed a protective effect of O3 to school children’s PEF. Recently, Altuǧ et al. (2014) showed that PEF levels were negatively associated with weekly average O3 levels only in children without upper respiratory tract complaints.

Two panel studies examined the effect of O3 exposure on FEV1 in asthmatic children and found no effect (Dales et al. 2009; Delfino et al. 2004). Both of them relied on self-measured FEV1 by handheld electronic devices. More consistent findings have been observed in the few studies using spirometry performed by trained personnel. In a reanalysis of six summer camp studies, a consistent decrease in FEV1 has been found associated with increased ambient exposure to O3 (Kinney et al, 1996). In the Mexican study (Barraza-Villarreal et al. 2008), using fixed site outdoor measurements, the investigators report negative associations with FVC and FEV1, although none reached the nominal level of significance. In a study in Taiwan (Chang et al. 2012), based also on fixed site measurements for the assessment of O3 exposure, a significant negative association was observed with FVC and a nonsignificant negative association with FEV1. The decrements, we observed mainly in FVC and also in FEV1, are also consistent with previous experimental studies in young adults. The main reason for this O3-induced decrease is believed to be impaired inspiration as a result of stimulation of airway receptors causing a reduction in the level of inflation achieved at full inspiration (Blomberg et al. 1999; Hazucha et al. 1989; Kjærgaard et al. 2004). Based on all the above, our results support that ambient O3 exposure affects inspiratory capacity and is related to airway inflammation.

In the present study, we also found a significant increase in respiratory symptoms after increased exposure to O3. This is a consistent finding among studies that investigate O3 effects at different time scales, mostly in asthmatic children. Thus, Declerq and Macquet (2000), Delfino et al. (2002), Escamilla-Nuñez et al. (2008), Just et al. (2002), Mortimer et al. (2002), and Schlink et al. (2002) found an association of O3 exposure and increase in respiratory symptoms, but few others found no effect (Jalaludin et al. 2004; Ostro et al. 2001). In our study asthmatic children were not found to be more susceptible than nonasthmatic children and this has been reported by other studies as well (Ward et al. 2002).

Our study has certain limitations. Our sample included children who agreed to participate from randomly selected schools; however, they did not form a random sample from the students of their class. This may have introduced some selection bias because children from more educated families or families more sensitive to environmental problems may have been preferentially included. However, we think that this did not bias our results away from the null given that the students attending a specific school are socioeconomically comparable (they live in the neighborhood) and the percentage of asthmatic children in our sample is comparable to that reported for the population (Papadopoulou et al. 2011). Another limitation of our project is the fact that assessment of exposure to pollutants other than O3 was not based on personal measurements but was estimated using the nearest fixed site monitors. This might have led to residual confounding, especially taking into account exposure to PM. We believe that the probability of residual confounding is limited because the correlations of O3 with PM10 and NO2 was not the same across the high- and low-O3 areas and between the two cities. Furthermore, PM is more homogeneously distributed and penetrates indoors making the fixed site and personal measurements less different from those of O3. Absenteeism might be a source of bias to the null for the estimated effects. During the field work, on each of the days our researchers visited the schools for spirometry, only a few children were absent from school (0–4 children each time, except during the last summer week in Thessaloniki when 17 children could not be found because specific schools had end-of-year activities), and it may be hypothesized that some of these children might have had a respiratory problem because the reason for the absence was not reported by the school. Another limitation of our study is the inability to study children during the hottest months of the year, July and August, because of school holidays. These months are also characterized by high-O3 concentrations. We were not able to reach schoolchildren during the holidays, but it should be mentioned that most families in Greece take their holidays during this time and the population of children in the cities is greatly reduced.

An advantage of the present study is the use of personal O3 monitoring. The higher contrast observed in outdoor school measurements compared with personal ones denotes the importance of relying on personal measurements when assessing O3 exposure health effects on school children. Another advantage is the spirometry performed by physicians, which provides more accurate and valid results compared with self-monitoring. This may explain our findings associating ozone exposure with the weekly FVC and FEV1, provided through spirometry, whereas an analysis of daily PEF measured by peak flow meters, handled by the same sample of students, did not provide significant results (Samoli et al. 2017).

Among the strengths of our study is the study design, as panel studies allow great flexibility in investigating multiple exposures and multiple outcomes; the location, as Greece is a warm, sunny country with inhabitants exposed to high-O3 concentrations; the large number of students and the relatively long period of follow up; the limited number of participants lost to follow-up or with incomplete data; the personal O3 monitoring; and the wealth of data collected over a variety of respiratory morbidity indices using standardized procedures. Moreover, our results were robust after controlling for particles and under several sensitivity/subgroup (e.g., after considering only the asthmatic schoolchildren) analyses.

Conclusion

The study provides evidence that airway inflammation and the frequency of respiratory symptoms increase while lung function decreases with increased O3 exposure in school children generally representative of the population of children 10- to 11-y-old in two large Southern European cities, where high- ozone concentrations are observed. Schoolchildren should be considered at a moderate risk of suffering respiratory health adverse effects following short-term exposure to the current levels of ambient O3, in line with what has been found in controlled human exposure studies at ambient concentrations. Focusing on asthmatic children may lead to an underestimation of the real burden of O3 induced respiratory adverse events.

Acknowledgments

The work was cofunded by the European Commission and the Greek government through the National Strategic Reference Framework 2007–2013 contract ref RESPOZE-children/2248.

References

Adams WC. 2006. Comparison of chamber 6.6-h exposure to 0.04–0.08 ppm ozone via square-wave and triangular profiles on pulmonary responses. Inhal Toxicol 18(2):127–136, PMID: 16393927, 10.1080/08958370500306107.

Altuǧ H, Gaga EO, Döǧeroǧlu T, Brunekreef B, Hoek G, Doorn WV. 2014. Effects of ambient air pollution on respiratory tract complaints and airway inflammation in primary school children. Sci Total Environ 479-480:201–209, PMID: 24561926, 10.1016/j.scitotenv.2014.01.127.

Arcus-Arth A, Blaisdell RJ. 2007. Statistical distributions of daily breathing rates for narrow age groups of infants and children. Risk Anal 27(1):97–110, PMID: 17362403, 10.1111/j.1539-6924.2006.00862.x.

ATS/ERS (American Thoracic Society and the European Thoracic Society). 2005. ATS/ERS recommendations for standardized procedures for the online and offline measurement of exhaled lower respiratory nitric oxide and nasal nitric oxide, 2005. Am J Respir Crit Care Med 171(8):912–930, 10.1164/rccm.200406-710ST.

Barraza-Villarreal A, Sunyer J, Hernandez-Cadena L, Escamilla-Nuñez MC, Sienra-Monge JJ, Ramírez-Aguilar M, et al. 2008. Air pollution, airway inflammation, and lung function in a cohort study of Mexico City Schoolchildren. Environ Health Perspect 116(6):832–838, PMID: 18560490, 10.1289/ehp.10926.

Bateson TF, Schwartz J. 2008. Children’s response to air pollutants. J Toxicol Environ Health Part A 71(3):238–243, PMID: 18097949, 10.1080/15287390701598234.

Blomberg A, Mudway IS, Nordenhäll C, Hedenström H, Kelly FJ, Frew AJ, et al. 1999. Ozone-induced lung function decrements do not correlate with early airway inflammatory or antioxidant responses. Eur Respir J 13(6):1418–1428, PMID: 10445622.

Burnett R, Raizenne ME, Krewski D. 1990. Acute health effects of transported air pollution: a study of children attending a residential summer camp. Can J Statistics 18(4):367–373, 10.2307/3315843.

Castro HA, Cunha MF, Mendonca GA, Junger WL, Cunha-Cruz J, Leon AP. 2009. Effect of air pollution on lung function in schoolchildren in Rio de Janeiro, Brazil. Rev Saude Publica 43(1):26–34, PMID: 19169573.

Chang YK, Wu CC, Lee LT, Lin RS, Yu YH, Chen YC. 2012. The short-term effects of air pollution on adolescent lung function in Taiwan. Chemosphere 87(1):26–30, PMID: 22189374, 10.1016/j.chemosphere.2011.11.048.

Chen CH, Chan CC, Chen BY, Cheng TJ, Guo YL. 2015. Effects of particulate air pollution and ozone on lung function in non-asthmatic children. Environ Res 137:40–48, PMID: 25486544, 10.1016/j.envres.2014.11.021.

Dales R, Chen L, Frescura AM, Liu L, Villeneuve PJ. 2009. Acute effects of outdoor air pollution on forced expiratory volume in 1 s: a panel study of schoolchildren with asthma. Eur Respir J 34:316–323, PMID: 19251781, 10.1183/09031936.00138908.

Declercq C, Macquet V. 2000. Short-term effects of ozone on respiratory health of children in Armentieres, North of France [in French]. Rev Epidemiol Sante Publique 48(suppl 2):2S37–2S43.

Delfino RJ, Quintana PJ, Floro J, Castanaga VM, Samini BS, Kleinman MT. 2004. Association of FEV1 in asthmatic children with personal and microenvironmental exposure to airborne particulate matter. Environ Health Perspect 112(8):932–941, PMID: 15175185, 10.1289/ehp.6815.

Delfino RJ, Zeiger RS, Seltzer JM, Street DH, McLaren CE. 2002. Association of asthma symptoms with peak particulate air pollution and effect modification by anti-inflammatory medication use. Environ Health Perspect 119(10):A607–A617, PMID: 12361942, 10.1289/ehp.021100607.

Devlin RB, McDonnell WF, Mann R, Becker S, House DE, Koren HS. 1991. Exposure of humans to ambient levels of ozone for 6.6 hours causes cellular and biochemical changes in the lung. Am J Respir Cell Mol Biol 4(1):72–81, PMID: 1846079, 10.1165/ajrcmb/4.1.72.

Dweik RA, Boggs PB, Erzurum SC, Irvin CG, Leigh MW, Lundberg JO, et al. 2011. An official ATS clinical practice guideline: interpretation of exhaled nitric oxide levels (FeNO) for clinical applications. Am J Respir Crit Care Med 184(5):602–615, PMID: 21885636, 10.1164/rccm.9120-11ST.

Escamilla-Nuñez NMC, Barraza-Villarreal A, Hernandez-Cadena L, Moreno-Macias H, Ramirez-Aguilar M, Sienra-Monge JJ, et al. 2008. Traffic-related air pollution and respiratory symptoms among asthmatic children, resident in Mexico City: the EVA cohort study. Respir Res 9:74, PMID: 19014608, 10.1186/1465-9921-9-74.

Hazucha MJ, Bates DV, Bromberg PA. 1989. Mechanism of action of ozone on the human lung. J Appl Physiol 67(4):1535–1541, PMID: 2793755.

Higgins ITT, D’Arcy JB, Gibbons DI, Avol EL, Gross KB. 1990. Effect of exposure to ambient ozone on ventilatory lung function in children. Am Rev Respir Dis 141(5 part 1):1136–1146, PMID: 2339836, 10.1164/ajrccm/141.5_Pt_1.1136.

Höppe P, Peters A, Rabe G, Praml G, Lindner J, Jakobi G, et al. 2003. Environmental ozone effects in different population subgroups. Int J Hyg Environ Health 206(6):505–516, PMID: 14626898, 10.1078/1438-4639-00250.

Horstman DH, Folinsbee LJ, Ives PJ, Abdul-Salaam S, McDonnell WF. 1990. Ozone concentration and pulmonary response relationships for 6.6-hour exposure with five hours of moderate exercise to 0.08, 0.10 and 0.12 ppm. Am Rev Respir Dis 142(5):1158–1163, PMID: 2240838, 10.1164/ajrccm/142.5.1158.

Jalaludin BB, Chey T, O’Toole BI, Smith WT, Capon AG, Leeder SR. 2000. Acute effects of low levels of ambient ozone on peak expiratory flow rate in a cohort of Australian children. Int J Epidemiol 29:549–557, 10.1093/intjepid/29.3.549.

Jalaludin BB, O’Toole BI, Leeder SR. 2004. Acute effects of urban ambient air pollution on respiratory symptoms, asthma medication use, and doctor visits for asthma in a cohort of Australian children. Environ Res 95(1):32–42, PMID: 15068928, 10.1016/S0013-9351(03)00038-0.

Just J, Segala C, Sahraoui F, Priol G, Grimfeld A, Neukirch F. 2002. Short-term health effects of particulate and photochemical air pollution in asthmatic children. Eur Respir J 20(4):899–906, PMID: 12412681.

Kim CS, Alexis NE, Rappold AG, Kehrl H, Hazucha MJ, Lay JC, et al. 2011. Lung function and inflammatory response in healthy young adults exposed to 0.06 ppm ozone for 6.6 hours. Am J Respir Crit Care Med 183(9):1215–1221, PMID: 21216881, 10.1164/rccm.201011-1813OC.

Kinney PL, Thurston GD, Raizenne M. 1996. The effects of ambient ozone on lung function in children: a reanalysis of six summer camp studies. Environ Health Perspect 104(2):170–174, PMID: 8820584.

Kjærgaard SK, Pedersen OF, Miller MR, Rasmussen TR, Hansen JC, Mølhave L. 2004. Ozone exposure decreases the effect of a deep inhalation on forced expiratory flow in normal subjects. J Appl Physiol 96(5):1651–1657, PMID: 14688031, 10.1152/japplphysiol.00507.2003.

Li S, Williams G, Jalaludin B, Baker P. 2012. Panel studies of air pollution on children’s lung function and respiratory symptoms: a literature review. J Asthma 49(9):895–910, PMID: 23016510, 10.3109/02770903.2012.724129.

Lougheed MD, Leniere C, Ducharme FM, Licskai C, Dell SD, Rowe BH, et al. 2012. Canadian Thoracic Society 2012 guideline update: diagnosis and management of asthma in preschoolers, children and adults: executive summary. Can Respir J 19(6):e81–e88, 10.1155/2012/214129.

Lum S, Stocks J. 2010. Forced expiratory manoeuvres. In: Paediatric Lung Function. Frey U, Merkus PJFM, eds. Lausanne, Switzerland:European Respiratory Society, 46–65.

Miller MR, Hankinson J, Brusasco V, Burgos F, Casaburi R, Coates A, et al. 2005. Standardisation of spirometry. Eur Respir J 26(2):319–338, PMID: 16055882, 10.1183/09031936.05.00034805.

Mortimer KM, Neas LM, Dockery DW, Redline S, Tager IB. 2002. The effect of air pollution on inner-city children with asthma. Eur Respir J 19(4):699–705, PMID: 11999000.

Mudway IS, Kelly FJ. 2000. Ozone and the lung: a sensitive issue. Mol Aspects Med 21:1–48, PMID: 10804262.

Ostro B, Lipsett M, Mann J, Braxton-Owens H, White M. 2001. Air pollution and exacerbation of asthma in African-American children in Los Angeles. Epidemiology 12(1–2):200–208, PMID: 11246581.

Papadopoulou A, Hatziagorou E, Matziou VN, Grigoropoulou DD, Panagiotakos DB, Tsanakas JN, et al. 2011. Comparison in asthma and allergy prevalence in the two major cities in Greece: the ISAAC phase II survey. Allergol Immunopathol (Madr) 39(6):347–355, PMID: 21342745, 10.1016/j.aller.2010.10.003.

Peacock JL, Symonds P, Jackson P, Bremner SA, Scarlett JF, Strachan DP, et al. 2003. Acute effects of winter air pollution on respiratory function in schoolchildren in southern England. Occup Environ Med 60(2):82–89, PMID: 12554833.

Raizenne ME, Burnett RT, Stern B, Franklin CA, Spengler JD. 1989. Acute lung function responses to ambient acid aerosol exposures in children. Environ Health Perspect 79:179–185, PMID: 2707197.

Roemer W, Hoek G, Brunekreef B. 2000. Pollution effects on asthmatic children in Europe, the PEACE study. Clin Exp Allergy 30(8):1067–1075, PMID: 10931113.

Romieu I, Meneses F, Ruiz S, Huerta J, Sienra JJ, White M, et al. 1997. Effects of intermittent ozone exposure on peak expiratory flow and respiratory symptoms among asthmatic children in Mexico City. Arch Environ Health 52(5):368–376, PMID: 9546760, 10.1080/00039899709602213.

Samoli E, Dimakopoulou K, Evangelopoulos D, Rodopoulou S, Karakatsani A, Veneti L, et al. 2017. Is daily exposure to ozone associated with respiratory morbidity and lung function in a representative sample of schoolchildren? Results from a panel study in Greece. J Expo Sci Environ Epidemiol 27(3):346–351.

Sarnat SE, Raysoni AU, Li WW, Holguin F, Johnson BA, Flores Luevano S, et al. 2012. Air pollution and acute respiratory response in a panel of asthmatic children along the U.S.-Mexico border. Environ Health Perspect 120(3):437–444, PMID: 21896396, 10.1289/ehp.1003169.

Scarlett JF, Abbott KJ, Peacock JL, Strachan DP, Anderson HR. 1996. Acute effects of summer air pollution on respiratory function in primary school children in southern England. Thorax 51(1):1109–1114, PMID: 8958894.

Schelegle ES, Adams WC, Walby WF, Marion MS. 2012. Modelling of individual subject ozone exposure response kinetics. Inhal Toxicol 24(7):401–415, PMID: 22642289, 10.3109/08958378.2012.683891.

Schlink U, Fritz GJ, Herbarth O, Richter M. 2002. Longitudinal modelling of respiratory symptoms in children. Int J Biometeorol 47(1):35–48, 10.1007/s00484-002-0142-2.

Spektor DM, Lippmann M, Lioy PJ, Thurston GD, Citak K, James DJ, et al. 1988. Effects of ambient ozone on respiratory function in active, normal children. Am Rev Respir Dis 137(2):313–320, PMID: 3341625, 10.1164/ajrccm/137.2.313.

Spektor DM, Thurston GD, Mao J, He D, Hayes C, Lippman M. 1991. Effects of single- and multiday ozone exposures on respiratory function in active normal children. Environ Res 55(2):107–122, 10.1016/S0013-9351(05)80167-7.

Ward DJ, Roberts KT, Jones N, Harrison RM, Ayres JG, Hussain S, et al. 2002. Effects of daily variation in outdoor particulates and ambient acid species in normal and asthmatic children. Thorax 57(6):489–502, PMID: 12037223.

Warren DW, Hairfield WM, Dalston ET. 1990. Effect of age on nasal cross-sectional area and respiratory mode in children. Laryngoscope 100(1):89–93, PMID: 2293706, 10.1288/00005537-199001000-00018.

WHO (World Health Organization). 2013. Review of Evidence on Health Aspects of Air Pollution – REVIHAAP Project: Final Technical Report. http://www.euro.who.int/en/health-topics/environment-and-health/air-quality/publications/2013/review-of-evidence-on-health-aspects-of-air-pollution-revihaap-project-final-technical-report [accessed 22 May 2016].

The Prevalence of Extended-Spectrum Beta-Lactamase-Producing Multidrug-Resistant Escherichia Coli in Poultry Chickens and Variation According to Farming Practices in Punjab, India

Author Affiliations open

1Center for Disease Dynamics, Economics & Policy, Washington, DC, USA

2Public Health Foundation of India, New Delhi, Delhi, India

3Dept. of Veterinary Population Medicine, University of Minnesota, St. Paul, Minnesota, USA

4School of Public Health and Zoonoses, Guru Angad Dev Veterinary and Animal Sciences University, Ludhiana, Punjab, India

5Dept. of Ecology and Evolutionary Biology, Princeton University, Princeton, New Jersey, USA

6SRL Diagnostics, FORTIS Hospital, Noida, Uttar Pradesh, India

7SRL Diagnostics, Mumbai, Maharashtra, India

8Veterinary & Animal Husbandry Extension, Guru Angad Dev Veterinary and Animal Sciences University, Ludhiana, Punjab, India

9Princeton Environmental Institute, Princeton, New Jersey, USA

PDF icon PDF Version (144 KB)

  • Background:
    Agricultural use of antimicrobials in subtherapeutic concentrations is increasing in response to the rising demand for food animal products worldwide. In India, the use of antimicrobials in food animal production is unregulated. Research suggests that many clinically important antimicrobials are used indiscriminately. This is the largest study to date in India that surveys poultry production to test for antimicrobial resistance and the occurrence of extended-spectrum β-lactamases (ESBLs) modulated by farming and managerial practices.
    Objectives:
    Our goal was to survey poultry production for resistance to eleven clinically relevant antimicrobials and phenotypic occurrence of ESBLs as modulated by farming and managerial practices.
    Methods:
    Eighteen poultry farms from Punjab were surveyed, and 1,556 Escherichia coli isolates from 530 birds were tested for susceptibility to 11 antimicrobials using the disk diffusion method and validated using VITEK 2 (bioMérieux, Marcy-L’Étoile, France). Samples from 510 of these birds were phenotypically tested for ESBL production using the combination disk method and confirmed using VITEK 2. Generalized linear mixed models were used to infer differences in resistance profiles associated with different farming practices and facility types.
    Results:
    Resistance profiles were significantly different between broiler and layer farms. Broiler farms were 2.2 [ampicillin (AMP), p=0.017] to 23 [nalidixic acid (NX), p<0.001] times more likely to harbor resistant E. coli strains than layer farms. Adjusting for farm type (broiler vs. layer), the odds of resistance (although not statistically significant) to all antimicrobials except nitrofurantoin (NIT) were higher in independent facilities (IUs) as compared to contracted facilities (CFs). Increased prevalence of multidrug resistance (MDR; 94% compared to 60% in layers), including prevalence of ESBL-producing strains (87% compared to 42% in layers), was observed in broiler farms.
    Conclusions:
    Our findings suggest that unregulated use of clinically relevant antimicrobials in Indian broiler and layer farms may contribute to the emergence of resistance and support the need to curb the nontherapeutic use of medically important antimicrobials in food animal production. https://doi.org/10.1289/EHP292
  • Received: 31 March 2016
    Revised: 16 December 2016
    Accepted: 17 December 2016
    Published: 20 July 2017

    Address correspondence to R. Laxminarayan and C. Brower, Center for Disease Dynamics, Economics & Policy, 1400 Eye St. NW, #500, Washington, DC 20005 USA. Telephone: 202-939-3300. Email: ramanan@cddep.org; charlie.brower@gmail.com

    Supplemental Material is available online (https://doi.org/10.1289/EHP292).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (333 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (164 KB)

Introduction

Medically important antimicrobials are used extensively in food animal production for disease prevention (e.g., prophylaxis and metaphylaxis), treatment, and growth promotion. It is estimated that two-thirds of antimicrobials produced globally are consumed in the livestock sector (CDDEP 2015). Numerous studies suggest that the widespread use of agricultural antimicrobials contributes to increased clinical resistance to antimicrobials (Chang et al. 2015; Marshall and Levy 2011; Silbergeld et al. 2008). Since antimicrobials are routinely added to animal feeds, bacterial populations are repeatedly exposed to subtherapeutic doses ideal for the emergence and spread of antimicrobial resistance (Chang et al. 2015; Marshall and Levy 2011; Silbergeld et al. 2008; You and Silbergeld 2014). Nearly every class of antimicrobial is used in agriculture, including many closely related to clinically relevant antimicrobials, such as penicillins, cephalosporins, fluoroquinolones, tetracyclines, sulfonamides, and aminoglycosides (Marshall and Levy 2011; Schwarz et al. 2001; Silbergeld et al. 2008). The extensive use of antimicrobials in agriculture results in human exposure to antimicrobial-resistant bacteria via direct and indirect pathways. These include exposure via direct contact with livestock or contaminated food products, indirect gene transfer across bacterial species, and the widespread release of antimicrobial-resistant pathogens into the environment (Silbergeld et al. 2008). This raises serious human health concerns, since the occurrence of cross-resistance between antimicrobials of the same class is highly likely.

The use of antimicrobials in subtherapeutic concentrations is increasing in response to heightened demand for food animal products worldwide, particularly in South and Southeast Asia, due to rising incomes (Teillant et al. 2015; Van Boeckel et al. 2015). The global average annual consumption of antimicrobials in food animals was conservatively estimated in 2010 to be 63,151 (±1,560) tons. In chickens, this corresponds to a consumption of antimicrobials per kilogram of animal produced of approximately 148 Mg·kg−1 (Van Boeckel et al. 2015). Global antimicrobial consumption in food animal production is expected to rise by 67% by 2030, driven primarily by BRICS (Brazil, Russia, India, China, and South Africa) countries shifting to large-scale, intensive farming operations where antimicrobials are used routinely in subtherapeutic doses. Antimicrobial consumption in food animal production in India is projected to grow by 312%, a concerning development further compounded by India’s status as the largest consumer of antimicrobials of humans (Van Boeckel et al. 2014, 2015).

A recent study of antimicrobial residues in chicken meat sold for human consumption in New Delhi, India, found that of the 70 chicken meat samples tested, 40% contained antimicrobial residues (Sahu and Saxena 2014). The most common antimicrobials detected were enrofloxacin (20%), ciprofloxacin (14.3%), doxycycline (14.3%), oxytetracycline (11.4%), and chlortetracycline (1.4%). The high use of fluoroquinolones detected is particularly concerning, given the importance of this broad-spectrum agent in human clinical medicine. Fluoroquinolones were banned for veterinary use in poultry in the United States in 2005 in response to numerous studies linking its use in treating respiratory diseases in poultry to the emergence and spread of fluoroquinolone resistance in humans and animals (Endtz et al. 1991; Nelson et al. 2007). In 2006, the European Union banned the use of antimicrobials for growth promotion in food animals in response to evidence suggesting growth promoters drove the emergence and spread of resistance (European Commission 2005).

In light of evidence of antimicrobial use in Indian poultry production and the projected 312% increase in agricultural antimicrobial consumption over the next 15 years, this study seeks to understand the prevalence of resistance [and, in particular, extended-spectrum β-lactamase (ESBL) production] in Escherichia coli and other Enterobacteriaceae in poultry and how farm operational and managerial practices in poultry production influence resistance emergence. Previous studies reported a high prevalence of ESBL-producing Enterobacteriaceae in poultry production (Dierikx et al. 2013b, 2013a; Kar et al. 2015; Laube et al. 2013). Moreover, ESBL-producing pathogens are resistant to numerous antimicrobials and are associated with longer hospital stays and other negative clinical outcomes in humans (Lautenbach et al. 2001). Specifically, this study investigates how resistance profiles vary between broiler and layer farms, between independent (IUs) and contracted facilities (CFs), and among farms reporting antimicrobial use (including for growth promotion) and those that do not.

This study compares various farm (i.e., broiler vs. layer) and facility types (i.e., contracted vs. independent), as these are the primary types of poultry operations in Punjab. Additionally, past research has reported that broiler farms tend to use more antimicrobials and harbor a higher level of resistance than layer farms (van den Bogaard et al. 2002). Higher antimicrobial usage in broilers is reasonable, considering the need to sustain rapid growth of chickens over short periods of time (∼35–50 days), whereas layer farms typically use fewer antimicrobials to sustain consistent egg production over longer periods of time (∼52–56 weeks). Moreover, unlike independent facilities, contracted farms are owned by large-scale poultry producers and follow a strict production process established by the producer, wherein all input materials (e.g., day-old chicks, feed, antimicrobials, etc.) are supplied by the contracting firm. Contracted farms are obligated to adhere to all instructions and protocols from the contracting firm. Given the differences in farming practices among these types of operations, it is necessary to understand how resistance prevalence varies to appropriately target policy interventions where they would have the greatest impact.

Materials and Methods

Sampling Protocol

Eighteen poultry farms (nine layers and nine broilers) were randomly selected from a list of farms provided by Guru Angad Dev Veterinary and Animal Sciences University (GADVASU), Ludhiana, Punjab, India. Layer and broiler farming operations were categorized as CFs or IUs, based on the nature of their contractual agreements with large-scale poultry producers. In this cross-sectional design, all samples were collected over a 5-month period from July to November 2014. All farms were located within six districts across the state of Punjab, India.

Sixty cloacal swabs were collected from 30 birds selected at random (two swabs per bird) from each farm and transported to the laboratory for isolation of E. coli (three isolates per bird) and other Enterobacteriaceae (one isolate per bird) for phenotypic detection of ESBLs. Only 20 birds were sampled from the first layer farm visited due to extreme heat causing danger to the health of the birds and only a single swab could be collected from each bird. Putative ESBL-producing Enterobacteriaceae were not isolated from these samples. Thus, 530 birds (n=530) were sampled for general susceptibility testing of E. coli, and 510 birds (n=510) were sampled for ESBL-producing Enterobacteriaceae. Since three E. coli isolates were sampled per bird, a total of 1,556 viable E. coli isolates (n=1,556) were tested for susceptibility against 11 antimicrobials covering a range of clinically relevant antimicrobial classes (tetracyclines, cephalosporins, quinolones, fluoroquinolones, penicillins, carbapenems, aminoglycosides, sulfonamides, and so on) (Figure 1).

Flow diagram.
Figure 1. Sampling framework depicting differences between on-farm and in-laboratory sampling protocols.

After sampling was complete, a 30-min structured interview covering sanitation, disease control, and antimicrobial use practices was conducted with each farm manager (see Supplemental Material, Survey). All protocols used to sample isolates from animals in this study were humane and approved by the Institutional Animal Care and Use Committee at Princeton University. Informed consent was obtained from all farmers participating in interviews, and the survey protocol was approved by the Institutional Ethics Committee at the Public Health Foundation of India.

Microbiological Methodology

Cloacal swabs were transported in a Cary-Blair transport medium (HiMedia, Mumbai, India) on ice to the laboratory for bacterial isolation. One swab from each bird was cultured on selective MacConkey Lactose agar (HiMedia, Mumbai, India), while the other was inoculated in a selective pre-enrichment broth [trypticase soy broth supplemented with 1 mg/L cefotaxime (CTX)] (HiMedia, Mumbai, India) [Diederen et al. 2012; Dierikx et al. 2013b, 2013a; EFSA Panel on Biological Hazards (BIOHAZ) 2011]. Both were incubated at 37°C under aerobic conditions for 20–24 hours.

Isolation of Escherichia coli for susceptibility testing.

Three well-isolated E. coli-like colonies (pink, doughnut-shaped) were selected at random from the MacConkey Lactose agar plate and subcultured on three separate Eosin Methylene Blue (HiMedia, Mumbai, India) agar plates at 37°C under aerobic conditions for 20–24 hours (Corry et al. 2011). Selecting bacterial isolates without selection for a specific resistance profile will detect only the most prevalent flora; thus, resistance at low levels may not be detected (Aarestrup et al. 2001). Subsequently, a single, well-isolated colony with the metallic-green sheen characteristic of E. coli on Eosin Methylene Blue from each plate was subcultured onto three separate trypticase soy agar (HiMedia, Mumbai, India) plates and incubated for 20–24 hours at 37°C under aerobic conditions for the isolation of pure cultures.

A single, well-isolated colony from each trypticase soy agar plate was then suspended in 5 mL of trypticase soy broth and incubated at 37°C until the inoculum achieved 0.5 McFarland standard (≥0.10D at 620 nm). Isolates were confirmed as E. coli by using HiIMViC biochemical test kits (HiMedia, Mumbai, India). Susceptibility to 11 drugs was tested using the Kirby-Bauer disk diffusion methodology (Clinical and Laboratory Standards Institute 2013a, 2013b) (see Table S1 for zone diameter breakpoints). All 11 antimicrobial disks used for susceptibility testing: ampicillin (AMP), gentamicin (GEN), ciprofloxacin (CIP), nitrofurantoin (NIT), co-trimoxazole (COT), tetracycline (TE), cefuroxime (CXM), imipenem (IPM), nalidixic acid (NX), chloramphenicol (C), and ceftriaxone (CTR) (HiMedia, Mumbai, India), were subjected to weekly quality control tests using the standard strain E. coli ATCC 25922 (HiMedia, Mumbai, India).

Isolation of ESBL-producing Enterobacteriaceae for phenotypic confirmation.

Following incubation, the second cloacal sample, which was inoculated in trypticase soy broth supplemented with 1 mg/L CTX (HiMedia, Mumbai, India), was vortexed and subcultured onto MacConkey agar supplemented with 1 mg/L CTX [Diederen et al. 2012; Dierikx et al. 2013b, 2013a; EFSA Panel on Biological Hazards (BIOHAZ) 2011]. After incubation for 20–24 hours at 37°C under aerobic conditions, a single, lactose-fermenting (pink-red) colony was then subcultured onto trypticase soy agar and incubated for 20–24 hours at 37°C under aerobic conditions for the isolation of pure colonies.

A single, well-isolated colony was then suspended in 5 mL of trypticase soy broth and incubated at 37°C until the inoculum achieved 0.5 McFarland standard (≥0.10D at 620 nm). Species identity was confirmed using HiIMViC biochemical test kits. Phenotypic ESBL production was tested using the combination disk method using CTX, cefotaxime–clavulanic acid (CEC), ceftazidime (CAZ), and ceftazidime–clavulanic acid (CAC) (Clinical and Laboratory Standards Institute 2013b; Dierikx et al. 2013b, 2013a). All antimicrobial disks were subjected to weekly quality control tests using the standard strain E. coli ATCC 25922, as well as an ESBL-positive Klebsiella pneumoniae isolate.

Isolate storage and quality assurance testing.

All isolates were inoculated in trypticase soy broth supplemented with 30% glycerol (v/v) (HiMedia, Mumbai, India) in cryovials and stored at −80ºC for further analysis and quality assurance testing. All 347 putative ESBL-producing Enterobacteriaceae isolates (i.e., isolates that exhibited growth on MacConkey agar supplemented with CTX) and 540 E. coli isolates for general susceptibility testing were sent to SRL Diagnostics at Fortis Hospital, Noida, Uttar Pradesh, India, for quality assurance by a separate team of researchers at an independent laboratory accredited by the National Accreditation Board for Testing and Calibration Laboratories, Government of India.

Of these isolates, all putative ESBL-producing and 395 (∼25% of total sample size, selected at random) E. coli isolates underwent species identification and susceptibility testing using VITEK 2 (bioMérieux, Marcy-L’Étoile, France). Species identification was confirmed using GN ID cards (bioMérieux, Marcy-L’Étoile, France) capable of identifying more than 150 fermentative and nonfermentative gram-negative bacilli. Antimicrobial susceptibility and phenotypic detection of ESBL production was confirmed using AST N280 cards (bioMérieux, Marcy-L’Étoile, France), which test susceptibility to the following drugs: amikacin, amoxicillin–clavulanic acid, ampicillin, cefepime, cefoperazone, sulbactam, ceftriaxone, cefuroxime, ciprofloxacin, colistin, ertapenem, gentamicin, imipenem, meropenem, nalidix acid, nitrofurantoin, piperacillin tazobactam, tigecycline, and trimethoprim/sulfamethoxazole (i.e., COT). Thus, this procedure provided quality assurance for 9 of the 11 drugs tested in this study (excluding chloramphenicol and tetracycline) as well as phenotypic detection of ESBL activity.

Statistical Analysis

Generalized linear mixed models with a logit link function were used to model the resistance profiles against farm and facility type for each antimicrobial, where the outcome for each isolate was classified as either resistant or susceptible. Isolates classified as intermediate were considered resistant for these analyses. Logistic regression models using the state of each isolate as a binary outcome (combining intermediate and resistant as nonsusceptible) with farm and facility type as the explanatory variables were used to model resistance profiles modulated by farming practices (i.e., farm or facility type). For each antimicrobial, the odds of an isolate being nonsusceptible were computed against farm and facility type. Since birds within a farm were likely to be treated similarly, random effects were incorporated in each model to account for similar resistance profiles within farms. Using the model parameters estimated from the logistic regression models, we derived prediction estimates (along with 95% prediction intervals using diagonal elements of the variance–covariance matrix of the predicted means and the estimated variance of the random intercept) for the population-based probability of resistant strains in isolates from randomly sampled broiler or layer farms, given the type of facility (contracted or independent).

Chi-square tests of independence were employed to test the difference in prevalence of ESBL-producing Enterobacteriaceae strains among farm types. Of the 11 antimicrobials tested in the study, the number of antimicrobials to which an isolate was classified as resistant was computed for every isolate. Since the number of antimicrobials did not follow a normal distribution, a categorical variable describing multidrug resistance was constructed with four levels [0=susceptible, 1=singly-resistant, 2−4=moderately multidrug resistant (MDR), >4=extremely MDR]. This multidrug resistance categorical variable was analyzed against farm (broiler vs. layer) and facility (independent vs. contracted) types using a proportional odds logistic regression model in order to account for the severity of resistance. Predicted probabilities for multidrug resistance were computed using the estimated parameters from the ordinal logistic regression model.

The resistance prevalence on farms that reported use of antimicrobials for growth promotion compared to those that did not was analyzed using logistic regression models with a random intercept for each farm. Adjusting for farm type (broiler vs. layer), ordinal logistic regressions were employed to understand the association of antimicrobial growth promoter (AGP) usage with multidrug resistance. Finally, a random effects logistic regression, adjusted for farm type, was used to analyze the association of AGP usage with the presence of ESBL-positive strains.

Sensitivity analyses were implemented to understand the effects of treating intermediate isolates as susceptible instead of resistant. Logistic regression models were employed to estimate the odds of resistance prevalence in broiler farms as compared to layer farms for each antimicrobial. Additionally, logistic regression models, stratified by farm type, were used to estimate the effect of farm size (in terms of number of birds) on resistance prevalence to each antimicrobial. All statistical analyses were carried out in R (version 3.2.1; R Foundation for Statistical Computing).

Results

Summary of Survey Results

Table S2 summarizes survey responses from 16 of the 18 farms sampled; two farms did not provide responses to detailed survey questions. In total, we sampled nine broiler farms (three contracted, six independent) and nine layer farms (three contracted, six independent). The average number of birds on each farm was 57,324 (interquartile range=12,000–40,000). Of the farms that elected to participate in the survey, all 16 reported using antimicrobials for disease treatment and prevention, while 12 (67%) reported using antimicrobials for growth promotion. Table S3 reports purpose of antimicrobial use disaggregated by farm and facility type. Tetracyclines and fluoroquinolones were the most commonly reported antimicrobials used, with nine (56%) farms reporting their use.

Overall prevalence of resistant Escherichia coli in Poultry Farms

The overall prevalence of resistance in the 1,556 E. coli isolates was analyzed across all farm (broiler vs. layer) and facility (contracted vs. independent) types. Percentages of isolates classified as susceptible, intermediate, or resistant were used to summarize resistance prevalence overall and disaggregated by farm and facility type (Table S4). A high prevalence of E. coli resistant to NX (86.1%), TE (47.0%), AMP (43.8%), COT (42.2%), and CIP (39.4%) was observed. The degree of resistance to TE, NX, AMP, and COT was consistently high across all farm and facility types. However, resistance to CIP was detected at high levels only in broiler farms. No resistance to IPM was detected among any of the farms.

Quality Assurance Validation Results

Resistance profiles of isolates against antimicrobials common to both datasets were compared. Validation analysis reveals that the original analysis conservatively underestimated the prevalence of CIP resistance by ∼15–20% across various farm and facility types (Figure 2). For all other antimicrobials, the validation analysis reveals either no differences or differences only in the classification of isolates being susceptible vs. intermediate.

Five ternary diagrams indicating broiler and layer farm types, contracted and independent facility types, and overall. The prevalence of following antibiotics is reported: ampicillin, ceftriaxone, cefuroxime, ciprofloxacin, cotrimoxazole, gentamicin, imipenem, nalidixic acid, and nitrofurantoin.
Figure 2. Ternary diagrams showing differences in resistance prevalence between original and validation data for 395 isolates against 9 common antimicrobials. Prevalence in the original data is shown as solid triangles, and those in the validation data are shown as solid dots. Each point represents a three-component vector showing the prevalence of susceptible, intermediate, and resistant isolates that sum to 100%. A point closer to a vertex (for instance, R) represents a high prevalence of the “resistant” state, also indicated by the arrows along each edge.

Resistance Profiles Modulated by Farming Practices

Figure 3A shows the odds ratio (OR) of resistance against each antimicrobial in broiler farms as compared to layer farms, adjusted for type of facility (contracted or independent). It is evident that for all antimicrobials, except NIT, the odds of finding a resistant isolate in broiler farms was at least two times greater than in layer farms (See also Table S6). Specifically, odds of E. coli resistant to antimicrobials, such as NX and CIP, was more than 10 times higher in broiler farms. Independent facilities had a significantly higher risk of E. coli resistant to C and NX as compared to contracted facilities, adjusted for farm type (Figure 3B). The results suggest that the odds of resistance (although not significant) to all antimicrobials tested, except NIT were higher in independent facilities.

Graphical representation of Odd’s ratio with confidence intervals.
Figure 3. Results from logistic regression modeling the risk of resistance prevalence against farm and facility type, with random intercepts for each farm. Left panel presents the risk (in terms of odds ratios) of Escherichia coli resistance in broiler farms relative to layer farms, adjusted for facility type. Right panel presents the same risk in independent facilities as compared to contracted facilities, adjusted for farm type. For all of these analyses, intermediate isolates are treated as resistant. The x-axis represents odds ratios in a log scale. The horizontal lines represent 95% confidence intervals for the estimated odds ratios.

This model was further used to estimate the population average probabilities of the occurrence of E. coli strains resistant or susceptible to each antimicrobial across all farm and facility types (Figure 4). A high probability of resistance to NX, CIP, and TE was observed, with higher probabilities in broiler farms, and specifically, independent broiler facilities.

Predicted probabilities are plotted (y-axis) across 10 antimicrobials by contracted and layer type; independent and layer type; contracted and broiler type; and independent and broiler type, respectively.
Figure 4. Predicted probabilities of resistance (solid blue dots) and corresponding 95% confidence intervals (blue error-bars) against 10 antimicrobials [imipenem (IPM) was not included, as no resistant isolates were detected], by farm and facility type, based on a logistic regression model with random intercepts for farm. Each isolate is assumed to be either resistant or susceptible in this analysis.

Prevalence of ESBL-Positive Strains and Multidrug Resistance

Understanding the prevalence of ESBL-positive bacteria is particularly relevant, given that infections caused by these organisms are more difficult to treat in humans. Of 510 cloacal samples collected for ESBL detection, 305 E. coli, 13 K. pneumoniae, 8 Escherichia fergusonii, 3 Proteus mirabilis, and 1 Escherichia hermannii isolate were phenotypically confirmed as ESBL-producing Enterobacteriaceae. Additionally, seven non-Enterobacteriaceae gram-negative isolates were confirmed as ESBL-positive: three Pseudomonas aeruginosa, two Bordetella trematum, and two Acinetobacter spp., and are included in our overall analysis, as they still indicate that a particular bird carried ESBL-positive flora (Table S5). The prevalence of ESBL-positive strains was significantly higher [OR=9.55, confidence interval (CI): 6.14, 14.85, p<0.0001] in broiler farms (87%) than layer farms (42%) (Figure 5A). There was no statistically significant difference in the prevalence of ESBL-positive strains (∼87%) between contracted and independent broiler farms. Contracted layer farms had a higher prevalence (49%) than independent layer farms (38%), although the difference was not statistically significant.

Figure A is a bar graph plotting percentage ESBL (y-axis) across facility types (x-axis) for the broiler and layer farm types. Figure B is a box-and-whisker plot, plotting number of antibiotics (y-axis) across facility types (x-axis) for the broiler and layer farm types. Figure C shows four panels of bar graphs plotting predicted probability for independent and contracted facilities for broiler and layer.
Figure 5. (A) Distribution of ESBL-producing status of the 510 cloacal samples, disaggregated by farm and facility type. ESBL status was tested for all Enterobacteriaceae isolated, including seven non-Enterobacteriaceae, gram-negative isolates. The figure shows proportions and associated 95% confidence intervals of ESBL-positive strains within each group. (B) Distribution of 1,556 multidrug-resistant Escherichia coli isolates by farm type and facility type. The vertical axis shows the number of antimicrobials (maximum 10) to which an isolate was resistant. The horizontal lines within the boxes indicate the median number of antibiotics to which an isolate was resistant, while the length of the box represents the interquartile range (IQR). The lower and upper limits of the whiskers represent Q1−1.5 IQR and Q3+1.5 IQR, where Q1 and Q3 are the first and third quartiles, respectively. (C) Predicted probabilities and associated 95% confidence intervals for multidrug resistance according to a proportional odds logistic regression of categories of multidrug resistance against farm and facility type.

Given that all isolates were tested against multiple antimicrobials, the prevalence of MDR E. coli strains was analyzed against farming practices (Table S6). A higher prevalence of MDR E. coli was detected on broiler farms (94%) as compared to layer farms (60%) (Figure 5B). Isolates on independent broiler farms were 15 times more likely (CI: 10.3, 21.9, p<0.001) to be multidrug resistant than those from an independent layer farm. The odds were significant yet lower in the case of contracted farming facilities (OR=5.4, CI: 4.1, 7.0, p<0.001). Predicted probabilities (Figure 5C) show that 90% of isolates were likely to be moderately or extremely resistant in broiler farms, while contracted layer farms had a lower probability (36%) of being moderately or extremely resistant.

Impact of Antimicrobial Use for Growth Promotion on Resistance Profiles

The prevalence patterns of antimicrobial resistance within certain subgroups were analyzed to better understand the impact of farming practices on resistance. Prevalence percentages within farms that reported using antimicrobials for growth promotion in poultry chickens were analyzed compared to farms that did not. Of the 16 farms that elected to participate in the survey, 75% reported using antimicrobials for growth promotion. Irrespective of the type of farming operation, our results indicate increased resistance prevalence in farms using AGPs. AGP use is correlated with significantly increased odds of resistance for all antimicrobials except CTR and NIT. Within broiler farms that reported use of AGPs, a significantly higher prevalence of resistance to CIP, C, NIT, and COT was detected. Within layer farms, significantly higher odds of resistance were observed for all antimicrobials except C, NIT, CXM, and CTR (Table 1).

Table 1 lists antimicrobials in the first column; the corresponding odds ratios stratified by farm type for overall, broilers, and layers are listed in the other columns.

Table 1. Odds ratios of increased resistance prevalence to all antibiotics for farms reporting antimicrobial use for growth promotion as compared to farms that did not report antimicrobial use for growth promotion.
Odds ratios stratified by farm type
Antimicrobial Overall Broilers Layers
Ampicillin (AMP) 1.413* (1.088, 1.839) 0.784* (0.485, 1.256) 1.460* (1.031, 2.079)
Chloramphenicol (C) 1.698* (1.179, 2.489) 2.721* (1.361, 6.040) 0.916 (0.562, 1.513)
Ciprofloxacin (CIP) 4.980** (3.792, 6.567) 2.258* (1.168, 4.189) 4.297** (2.899, 6.473)
Cotrimoxazole (COT) 4.242** (3.131, 5.809) 1.669* (1.042, 2.693) 6.638** (4.221, 10.764)
Ceftriaxone (CTR) 1.680 (0.885, 3.464) 1.284 (0.529, 3.769) 1.061 (0.363, 3.492)
Cefuroxime (CXM) 2.032* (1.130, 3.923) 1.888 (0.792, 5.479) 0.993 (0.388, 2.747)
Gentamicin (GEN) 2.559** (1.734, 3.875) 1.151 (0.694, 1.954) 3.279* (1.498, 8.190)
Nalidixic Acid (NX) 5.572** (3.761, 8.298) N/A 3.977** (2.590, 6.158)
Nitrofurantoin (NIT) 1.290 (0.984, 1.698) 2.942** (1.669, 5.457) 0.879 (0.624, 1.240)
Tetracycline (TE) 1.969** (1.490, 2.598) 1.101 (0.563, 2.036) 1.633* (1.162, 2.295)

Note: Fisher’s exact test was employed to compute the odds ratios and associated 95% confidence intervals (given in parentheses below) with the outcome as the presence or absence of resistant isolates. Odds ratios are presented for all farms (“overall”) and disaggregated by farm type (“broilers” vs. “layers”). Note that an OR could not be calculated for NX in broiler farms, as the cell frequency was zero. *p<0.05; **p<0.001.

After accounting for farm type (broiler vs. layer), our results indicate that reported AGP use had no significant association with the prevalence of ESBL-positive strains (OR=0.55, CI: 0.01, 30.58, p=0.773). In contrast, isolates from farms reporting AGP use were 2.92 times more likely (CI: 2.24, 3.81, p<0.001) to be multidrug resistant than those from farms not reporting AGP use. Moreover, in this analysis, the effect of farm type on the odds of multidrug resistance was maintained as broiler farms were still 6.17 times (CI: 4.80, 8.00, p<0.001) more likely to harbor MDR strains compared to layer farms.

Sensitivity Analysis

We analyzed the impact of treating intermediate isolates as susceptible instead of resistant on the associations described earlier, and we provide these results in the Supplemental Material (Figure S1). We did not observe any significant changes in the direction and statistical significance of associations between farming practices and patterns of antimicrobial resistance. However, in some cases (e.g., GEN, CXM, C), the associations were stronger, indicating a higher prevalence of intermediate isolates in layer farms.

We also conducted a sensitivity analysis to examine associations between farm size (number of birds) and the prevalence of resistance among different types of farms (broilers and layers). The median farm size in terms of number of birds was 15,000 in broiler farms compared to 40,000 in layer farms. With an increase in farm size by 10,000 birds, we observed that changes in the odds of resistance to all antimicrobials were not statistically significant within broiler farms. In the case of layer farms, we observed marginally lower odds of resistance to NX, CIP, and NIT with increases in farm size. These results are summarized in the Supplemental Materials (Figure S2).

Discussion

Research on antimicrobial use and resistance in food animal production in India remains a relatively new field. However, research is urgently needed given the projected large-scale increase in poultry production and antimicrobial use in the poultry sector (Brahmachari et al. 2013; Van Boeckel et al. 2015). Sarma et al. (1981) isolated E. coli from healthy and diseased fowl in Ludhiana, Punjab, India, and discovered that approximately 80% of isolates were resistant to chlortetracycline, tetracycline, oxytetracycline, and triple sulfas. Shivachandra et al. (2004) found 100% resistance to sulfadiazine and widespread resistance to amikacin, carbenicillin, erythromycin, and penicillin in Pasteurella multocida isolates from chickens and other birds from 11 separate states in India. More recently, Dhanarani et al. (2009) found extensive resistance to streptomycin (75%), erythromycin (57%), tobramycin (54%), ampicillin (50%), rifampicin (46%), and kanamycin (40%) in Staphylococcus and other bacterial isolates from poultry litter in Tamil Nadu, India.

Recent surveys suggest that 70–90% of Enterobacteriaceae in India are ESBL producers and that colonization of humans with such bacteria is widespread(Hawkey 2008; Kumarasamy et al. 2010; Mathai et al. 2002). However, researchers in India have only recently begun to investigate ESBL-producing bacteria of food animal origin. Kar et al. (2015) conducted the first systematic study on multidrug resistant ESBL-producing E. coli in food producing animals from India in which 316 E. coli isolates were collected from poultry and dairy cattle in Odisha with 18 (6%) isolates confirmed as ESBL-positive by combination disc method and ESBL E-test. A study in Hyderabad, India, isolated E. coli from 150 food samples (vegetable salad, raw egg surface, raw chicken, unpasteurized milk, and raw meat) and detected 6 (4%) ESBL producers, two of which were isolated from raw chicken samples (Rasheed et al. 2014). Another study in West Bengal, India, of 360 healthy layers and their environment did not detect ESBL production by PCR, but did record high levels of phenotypic resistance to several drugs: erythromycin (95.83%), chloramphenicol (87.52%), and cotrimoxazole (78.%) (Samanta et al. 2014).

We found a high degree of antimicrobial resistance and ESBL production in poultry facilities, which varied according to farming practices (i.e., farm and facility type). Moreover, we found a high prevalence of E. coli resistant to antimicrobials, such as CIP (fluoroquinolone), AMP (β-lactam), and tetracycline, which are commonly used in clinical settings. Broiler farms, and especially independent broiler farms, were associated with a higher prevalence of resistant E. coli strains than layer farms, corroborating past research (van den Bogaard et al. 2002). Additionally, the prevalence of ESBL-producing Entero-bacteriaceae was higher for broiler farms (87%) than layer farms (42%), and for contracted layer farms (49%) than independent layer farms (38%). These results may indicate that independent layer farms are using more drugs such as tetracyclines and fluoroquinolones, while contracted layer and broiler farms are beginning to shift to more recently developed drugs, such as third-generation cephalosporins, accounting for the higher prevalence of ESBLs. Additionally, higher odds of resistance among independent facilities may suggest that contracted facilities are employing better hygiene practices and veterinary care, which seems reasonable, considering these protocols and services are supplied by the contracting firm. In contrast, independent farmers do not have a comparable support system and could be misusing antimicrobials to a greater degree. However, it is critical that future studies identify the specific farming practices that are driving increased prevalence of resistant strains in order to mitigate the risk of spreading antimicrobial resistance.

Most of the farms surveyed in this study were large (average number of birds greater than 50,000), and all farms (that participated in the survey) reported using antimicrobials. Large flocks in small, confined areas, a lack of proper sanitation measures, and the unregulated application of broad-spectrum antimicrobials drive the spread of resistance at the farm level. The questionnaire-based surveys employed in this study revealed disturbing trends regarding the indiscriminate use of antimicrobials; all poultry farms included in this study used antimicrobials, and over half of them used antimicrobials for growth promotion rather than solely for disease prevention or treatment. Antimicrobials are often employed when broilers are being transported or held prior to slaughter to help them tolerate stress. Anecdotally, one farmer noted that antimicrobials were more effective than hygiene or sanitation measures because labor on poultry farms is unskilled, making it difficult to ensure that all hygiene procedures are followed. A majority of the farmers surveyed reported being unaware of the presence of AGPs premixed in chicken feed purchased from feed mills. Given the size and reach of these poultry farms in the retail market, the risk of exposure to resistant bacteria and antimicrobial compounds to humans is a significant concern.

This study reports a high prevalence of ESBL-producing Enterobacteriaceae in poultry farms (87% and 42% in broilers and layers, respectively). These results corroborate similar studies in broiler production facilities, but the prevalence reported here is similar or higher (Dierikx et al. 2013b, 2013a; Kar et al. 2015; Laube et al. 2013). ESBL-producing Enterobacteriaceae are highly resistant to multiple drugs, can contribute to acquired resistance through horizontal gene transfer across a wide range of bacterial species, and are associated with longer hospital stays and negative clinical outcomes in humans (Lautenbach et al. 2001; Marshall and Levy 2011; Rawat and Nair 2010; Silbergeld et al. 2008). Emerging resistance to widely used antimicrobials, such as fluoroquinolones and cephalosporins, reduces the efficacy of treating enteric, urinary tract, and skin infections, resulting in prolonged and more serious courses of illness. Enterobacteriaceae resistant to third-generation cephalosporins as a result of cephalosporin overuse in poultry production have been associated with increased human deaths in Europe (Collignon et al. 2013). In the present study, the prevalence of ESBL-producing E. coli and other Enterobacteriaceae was higher in broiler facilities, where antimicrobials are more commonly administered for growth promotion and disease prevention. This highlights the need to regulate the use of antimicrobials in these intensive farming operations, especially since researchers predict that this region will experience a major shift towards this type of farming operation over the next 15 years (Van Boeckel et al. 2015).

Direct contact with livestock colonized with resistant bacteria is the most documented route of resistance transmission from the agricultural reservoir into human populations (Bergeron et al. 2012; Jakobsen et al. 2010a, 2010b; Marshall and Levy 2011; Schmithausen et al. 2015; van den Bogaard et al. 2002; Zhang et al. 2009). These high-risk individuals provide a conduit of entry for resistant bacteria and resistance genes into the community and hospitals, where further person-to-person transmission is possible (Marshall and Levy 2011; Silbergeld et al. 2008). Of particular concern in this study was the lack of sanitation measures to prevent the transfer of resistant bacteria from animals to farm workers. Among survey respondents, 67% indicated they take no precautions when entering poultry sheds (and farm workers often do not wear closed-toe shoes); thus, the risk of colonization of farm workers is likely much higher than in other countries where strict disease control practices are implemented.

Despite extensive evidence linking the use of antimicrobials in food animal production to resistance in human populations, little has been done to address the problem in the majority of developing and developed countries worldwide. Recent publications have highlighted the rise of antimicrobial resistance and the emergence of new mechanisms of resistance in the Indian subcontinent (Kumarasamy et al. 2010). Policy makers and researchers have focused their attention on the clinical overuse and misuse of antimicrobials (to treat colds and other viral infections, for example) as drivers of resistance emergence. Furthermore, a recent review of the effectiveness of AGPs in food animal production suggests that the effects of AGPs on improving production and decreasing mortality in the poultry industry are minimal and do not offset the costs of the AGPs themselves (Cogliani et al. 2011; Engster et al. 2002; Graham et al. 2007; Sneeringer et al. 2015; Teillant et al. 2015). Moreover, it is possible to reduce the prevalence of antimicrobial resistance by placing restrictions on the use of antimicrobials in food animal production without negative impacts on productivity, as evidenced by the experience of both the poultry and pork industries in Denmark (Aarestrup et al. 2001; Levy 2014).

Notwithstanding the growing body of evidence, India has no regulatory provisions for the use of antimicrobials in cattle, chicken, or pigs raised for domestic consumption (Ganguly et al. 2011; Van Boeckel et al. 2015). The only laws on antimicrobial use in food animal production for domestic consumption mandate withdrawal of antimicrobials before processing of food animal products (Brahmachari et al. 2013). The lack of uniform regulations by the various agencies involved in poultry farming (and other food animal production industries) hinders enforcement of the appropriate use of antimicrobials. Policy actions should be implemented immediately in order to safeguard the effectiveness of antimicrobials, since antimicrobial effectiveness is a globally shared resource and responsibility (Ganguly et al. 2011).

This study analyzes a large sample size of isolates for resistance patterns and ESBL production, utilizing samples from farms following a variety of different operational and managerial practices, but only covers 18 farms located in one particular region of India. Future surveys of farms across multiple locations will be needed to obtain estimates at the district, state, and national levels. Furthermore, resistance prevalence at the farm level should be tracked longitudinally over longer periods of time to account for seasonality and to ascertain how resistance profiles are fluctuating over time. Supply of, and demand for, poultry meat and the economics of poultry farming are other critical factors that determine the proliferation of broiler farms and associated farming practices. Augmenting this study with antimicrobial sales and consumption data would help in formulating strategies to curb the rise of resistance in this growing reservoir.

This study only examines the prevalence of resistance on farms that reported using antimicrobials. Future studies should attempt to include antimicrobial-free or organic farms in their sampling frame. However, this may be difficult to accomplish in this region, as we did not encounter any such farms. Additionally, this study relies on survey data to create analytical variables used in our statistical models; all variables used were selected for analysis based on their likelihood of impacting resistance profiles and whether survey responses were reliable across all farms. For instance, we were unable to obtain detailed information on the types and amounts of antimicrobials used for all farms, since some farmers were unaware of this information or unwilling to report it. In light of these limitations, our study primarily compares resistance profiles among different farm (broiler vs. layer) and facility (independent vs. contracted) types instead of investigating specific practices influencing resistance development. Future studies should augment this survey data with observational periods on farms to better understand specific managerial and antimicrobial use practices.

Finally, given the high rates of background resistance in both human populations and the environment in India, this study cannot definitively link increased farm use of antimicrobials to increased resistance. A more detailed genetic investigation of the isolates to identify specific resistance genes could shed light on the mechanisms of resistance propagation within and among high-risk farms. Moreover, genetic analysis also would enable tracking of resistance genes in poultry birds to farmers and laborers in order to better quantify the risk of transmission from animals to humans, as well as help us understand the sources of resistance (e.g., humans, animals, environment, etc.).

Conclusion

This study presents evidence that antimicrobial resistance in E. coli and other pathogenic bacteria is correlated with particular farming practices. In particular, the results of this study revealed that broiler farms were associated with a higher prevalence of resistance, including ESBL-producing Enterobacteriaceae and multidrug resistance, than layer farms. Moreover, our findings suggest that antimicrobial use for growth promotion promoted the development of reservoirs of highly resistant bacteria on the studied farms, with potentially serious implications for human health. The prevalence of resistance to multiple antimicrobials was higher in both broiler and layer farms that used antimicrobials for growth promotion. ESBL-positive and multidrug-resistant strains are equipped with an arsenal of mechanisms that enable them to survive against last-resort treatments in human clinical settings. Furthermore, these highly resistant strains contribute to acquired resistance through horizontal gene transfer of resistance determinants to other microbial strains and species (including commensal microbes) further propagating antimicrobial resistance across various reservoirs of resistance, a threat that is both real and imminent (Marshall and Levy 2011; Silbergeld et al. 2008).

Until recently, resistance to polymyxin (a drug of last resort when other modern antimicrobials are ineffective) had only been reported to evolve via chromosomal mutations. A recent study of commensal E. coli in Chinese food animal production has identified a plasmid-mediated polymyxin resistance mechanism, MCR-1, in Enterobacteriaceae that has spread from animals to humans (detected in 1% of inpatients with infection) (Liu et al. 2015). Although we did not test for MCR-1 in this study, the emergence of such a highly mobile resistance determinant to such an important class of antimicrobials, with risk of global dissemination similar to NDM-1 (New Delhi metallo-β-lactamase), further emphasizes the need to regulate and curb antimicrobial overuse in food animal production (Liu et al. 2015). We conclude that withdrawal of nontherapeutic use of agricultural antimicrobials in India would be prudent to protect public health.

Acknowledgments

C.H.B. and R.L.’s time, as well as the cost of the study, was funded by the Global Antibiotic Resistance Partnership (funded by the Bill and Melinda Gates Foundation). C.H.B.’s time was also partially funded by the Fulbright-Nehru Fellowship. S.P. was funded by Princeton University’s Grand Challenges in Health program. We thank the poultry farmers for providing access to their farms to sample bacteria from their birds, SRL and Guru Angad Dev Veterinary and Animal Sciences University (GADVASU) for providing laboratory facilities, bioMérieux for providing us VITEK supplies, Suraj Pant for helping us create the figures presented in this article, and Manish Kakkar and Elizabeth Rogawski for helping develop the survey we administered.

References

Aarestrup FM, Seyfarth AM, Emborg HD, Pedersen K, Hendriksen RS, Bager F. 2001. Effect of abolishment of the use of antimicrobial agents for growth promotion on occurrence of antimicrobial resistance in fecal Enterococci from food animals in Denmark. Antimicrob Agents Chemother 45(7):2054–2059, PMID: 11408222, 10.1128/AAC.45.7.2054-2059.2001.

Bergeron CR, Prussing C, Boerlin P, Daignault D, Dutil L, Reid-Smith RJ, et al. 2012. Chicken as reservoir for extraintestinal pathogenic Escherichia coli in humans, Canada. Emerging Infect Dis 18(3):415–421, PMID: 22377351, 10.3201/eid1803.111099.

Brahmachari SK, Misri J, Vijayakumar K, George VI, Chakraborty P, Bhalla S, et al. 2013. Report of the Committee on Antibiotic Use and Resistance in Food Animals. New Delhi, India:
The Public Health Foundation of India.

Center for Disease Dynamics, Economics & Policy (CDDEP). 2015. State of the World’s Antibiotics, 2015. Washington, DC:CDDEP. http://www.cddep.org/publications/state_worlds_antibiotics_2015#sthash.XhiMCKG6.dpbs [accessed 14 July 2016].

Chang Q, Wang W, Regev-Yochay G, Lipsitch M, Hanage WP. 2015. Antibiotics in agriculture and the risk to human health: how worried should we be?. Evol Appl 8(3):240–247, PMID: 25861382, 10.1111/eva.12185.

Clinical and Laboratory Standards Institute (CLSI). 2013a. “Performance Standards for Antimicrobial Disk and Dilution Susceptibility Tests for Bacteria Isolated from Animals; Approved Standard––Fourth Edition; CLSI Document VET01-A4.” Wayne, PA:Clinical and Laboratory Standards Institute.

Clinical and Laboratory Standards Institute (CLSI). 2013b. “Performance Standards for Antimicrobial Susceptibility Testing; Twenty-Third Informational Supplement; CLSI Document M100-S23.” Wayne, PA:Clinical and Laboratory Standards Institute.

Cogliani C, Goossens H, Greko C. 2011. Restricting antimicrobial use in food animals: lessons from Europe. Microbe 6(6):274–279, 10.1128/microbe.6.274.1.

Collignon P, Aarestrup FM, Irwin R, McEwen S. 2013. Human deaths and third-generation cephalosporin use in poultry, Europe. Emerging Infect Dis 19(8):1339–1340, PMID: 23876416, 10.3201/eid.1908.120681.

Corry JEL, Curtis GDW, Baird RM, eds. 2011. Culture media for detection and enumeration of “total” Enterobacteriaceae, coliforms and Escherichia coli from foods. In: Handbook of Culture Media for Food and Water Microbiology. Cambridge, UK:Royal Society of Chemistry, 233–260.

Dhanarani TS, Shankar C, Park J, Dexilin M, Kumar RR, Thamaraiselvi K. 2009. Study on acquisition of bacterial antibiotic resistance determinants in poultry litter. Poult Sci 88(7):1381–1387, PMID: 19531707, 10.3382/ps.2008-00327.

Diederen B, Chang C, Euser S, Cohen Stuart J. 2012. Evaluation of four screening protocols for detection of extended-spectrum β-lactamase-producing members of the Enterobacteriaceae. J. Med Microbiol 61:452–453, 10.1099/jmm.0.036467-0.

Dierikx CM, van der Goot JA, Smith HE, Kant A, Mevius DJ. 2013a. Presence of ESBL/AmpC -producing Escherichia coli in the broiler production pyramid: A descriptive study. A. Cloeckaerted. PLoS ONE 8:e79005, 10.1371/journal.pone.0079005.

Dierikx C, van der Goot J, Fabri T, van Essen-Zandbergen A, Smith H, Mevius D. 2013b. Extended-spectrum-β-lactamase- and AmpC-β-lactamase-producing Escherichia coli in Dutch broilers and broiler farmers. J. Antimicrob Chemother 68(1):60–67, 10.1093/jac/dks349.

EFSA Panel on Biological Hazards (BIOHAZ). 2011. Scientific Opinion on the public health risks of bacterial strains producing extended-spectrum β-lactamases and/or AmpC β-lactamases in food and food-producing animals. EFSA J 9(8):1–95, 10.2903/j.efsa.2011.2322.

Endtz HP, Ruijs GJ, van Klingeren B, Jansen WH, van der Reyden T, Mouton RP. 1991. Quinolone resistance in campylobacter isolated from man and poultry following the introduction of fluoroquinolones in veterinary medicine. J Antimicrob Chemother 27(2):199–208, PMID: 2055811.

Engster HM, Marvil D, Stewart-Brown B. 2002. The effect of withdrawing growth promoting antibiotics from broiler chickens: a long-term commercial industry study. J Appl Poult Res 11(4):431–436, 10.1093/japr/11.4.431.

European Commission. 2005. Ban on Antibiotics as Growth Promoters in Animal Feed Enters into Effect. http://europa.eu/rapid/press-release_IP-05-1687_en.htm [accessed 4 March 2016].

Ganguly NK, Arora NK, Chandy SJ, Fairoze MN, Gill JP, Gupta U, et al. 2011. Rationalizing antibiotic use to limit antibiotic resistance in India. Indian J Med Res 134:281–294, PMID: 21985810.

Graham JP, Boland JJ, Silbergeld E. 2007. Growth promoting antibiotics in food animal production: an economic analysis. Public Health Rep 122(1):79–87, PMID: 17236612, 10.1177/003335490712200111.

Hawkey PM. 2008. Prevalence and clonality of extended-spectrum beta-lactamases in Asia. Clin. Microbiol. Infect 14(suppl 1):159–165, PMID: 18154540, 10.1111/j.1469-0691.2007.01855.x.

Jakobsen L, Kurbasic A, Skjøt-Rasmussen L, Ejrnaes K, Porsbo LJ, Pedersen K, et al. 2010a. Escherichia coli isolates from broiler chicken meat, broiler chickens, pork, and pigs share phylogroups and antimicrobial resistance with community-dwelling humans and patients with urinary tract infection. Foodborne Pathog. Dis 7(5):537–547, 10.1089/fpd.2009.0409.

Jakobsen L, Spangholm DJ, Pedersen K, Jensen LB, Emborg HD, Agersø Y, et al. 2010b. Broiler chickens, broiler chicken meat, pigs and pork as sources of ExPEC related virulence genes and resistance in Escherichia coli isolates from community-dwelling humans and UTI patients. Int J Food Microbiol 142(1-2):264–272, 10.1016/j.ijfoodmicro.2010.06.025.

Kar D, Bandyopadhyay S, Bhattacharyya D, Samanta I, Mahanti A, Nanda PK, et al. 2015. Molecular and phylogenetic characterization of multidrug resistant extended spectrum beta-lactamase producing Escherichia coli isolated from poultry and cattle in Odisha, India. Infect Genet Evol 29:82–90, 10.1016/j.meegid.2014.11.003.

Kumarasamy KK, Toleman MA, Walsh TR, Bagaria J, Butt F, Balakrishnan R, et al. 2010. Emergence of a new antibiotic resistance mechanism in India, Pakistan, and the UK: a molecular, biological, and epidemiological study. Lancet Infect Dis 10(9):597–602, PMID: 20705517, 10.1016/S1473-3099(10)70143-2.

Laube H, Friese A, von Salviati C, Guerra B, Käsbohrer A, Kreienbrock L, et al. 2013. Longitudinal monitoring of extended-spectrum-beta-lactamase/AmpC-producing Escherichia coli at German broiler chicken fattening farms. Appl Environ Microbiol 79:4815–4820, PMID: 23747697, 10.1128/AEM.00856-13.

Lautenbach E, Patel JB, Bilker WB, Edelstein PH, Fishman NO. 2001. Extended-spectrum β-lactamase-producing Escherichia coli and Klebsiella pneumoniae: risk factors for infection and impact of resistance on outcomes. Clin Infect Dis 32(8):1162–1171, 10.1086/319757.

Levy S. 2014. Reduced antibiotic use in livestock: How Denmark tackled resistance. Environ Health Perspect 122(6):A160–A165, PMID: 24892505, 10.1289/ehp.122-A160.

Liu YY, Wang Y, Walsh TR, Yi LX, Zhang R, Spencer J, et al. 2015. Emergence of plasmid-mediated colistin resistance mechanism MCR-1 in animals and human beings in China: a microbiological and molecular biological study. Lancet Infect. Dis 16(2):161–168.

Marshall BM, Levy SB. 2011. Food animals and antimicrobials: Impacts on human health. Clin Microbiol Rev 24(4):718–733, PMID: 21976606, 10.1128/CMR.00002-11.

Mathai D, Rhomberg PR, Biedenbach DJ, Jones RN, India Antimicrobial Resistance Study Group, 2002. Evaluation of the in vitro activity of six broad-spectrum β-lactam antimicrobial agents tested against recent clinical isolates from India: a survey of ten medical center laboratories. Diagn Microbiol. Infect Dis 44(4):367–377, 10.1016/S0732-8893(02)00466-2.

Nelson JM, Chiller TM, Powers JH, Angulo FJ. 2007. Fluoroquinolone-resistant Campylobacter species and the withdrawal of fluoroquinolones from use in poultry: a public health success story. Clin Infect Dis 44(7):977–980, PMID: 17342653, 10.1086/512369.

Rasheed MU, Thajuddin N, Ahamed P, Teklemariam Z, Jamil K. 2014. Antimicrobial drug resistance in strains of Escherichia coli isolated from food sources. Rev Inst Med Trop Sao Paulo 56(4):341–346, PMID: 25076436.

Rawat D, Nair D. 2010. Extended-spectrum β-lactamases in Gram Negative Bacteria. J Glob Infect Dis 2(3):263–274, PMID: 20927289, 10.4103/0974-777X.68531.

Sahu R, Saxena P. 2014. Antibiotics in Chicken Meat. New Delhi, India:Centre for Science and Environment. http://www.cseindia.org/userfiles/Antibiotics%20in%20Chicken_Lab%20Report_Final%2029%20July.pdf [accessed 3 August 2014].

Samanta I, Joardar SN, Das PK, Das P, Sar TK, Dutta TK, et al. 2014. Virulence repertoire, characterization, and antibiotic resistance pattern analysis of Escherichia coli isolated from backyard layers and their environment in India. Avian Dis 58(1):39–45, PMID: 24758111, 10.1637/10586-052913-Reg.1.

Sarma DK, Sambyal DS, Baxi KK. 1981. Drug susceptibility of Escherichia coli from domestic fowl. Zentralbl Veterinarmed B 28(4):333–335, PMID: 7027684.

Schmithausen RM, Schulze-Geisthoevel SV, Stemmer F, El-Jade M, Reif M, Hack S, et al. 2015. Analysis of transmission of MRSA and ESBL-E among pigs and farm personnel. PloS One 10(9):e0138173, PMID: 26422606, 10.1371/journal.pone.0138173.

Schwarz S, Kehrenberg C, Walsh TR. 2001. Use of antimicrobial agents in veterinary medicine and food animal production. Int J Antimicrob Agents 17(6):431–437, PMID: 11397611.

Shivachandra SB, Kumar AA, Biswas A, Ramakrishnan MA, Singh VP, Srivastava SK. 2004. Antibiotic sensitivity patterns among Indian strains of avian Pasteurella multocida. Trop Anim Health Prod 36(8):743–750, PMID: 15643810.

Silbergeld EK, Graham J, Price LB. 2008. Industrial food animal production, antimicrobial resistance, and human health. Annu Rev Public Health 29:151–169, PMID: 18348709, 10.1146/annurev.publhealth.29.020907.090904.

Sneeringer S, MacDonald JM, Key N, McBride W, Mathews K. 2015. Economics of Antibiotic Use in U.S. Livestock Production, ERR-200. Washington, DC:U.S. Department of Agriculture, Economic Research Service. https://www.ers.usda.gov/publications/pub-details/?pubid=45488 [accessed 14 July 2016].

Teillant A, Brower CH, Laxminarayan R. 2015. Economics of antibiotic growth promoters in livestock. Annu Rev Resour Econ 7:349–374, 10.1146/annurev-resource-100814-125015.

Van Boeckel TP, Brower C, Gilbert M, Grenfell BT, Levin SA, Robinson TP, et al. 2015. Global trends in antimicrobial use in food animals. Proc Natl Acad Sci USA 112(18):5649–5654, PMID: 25792457, 10.1073/pnas.1503141112.

Van Boeckel TP, Gandra S, Ashok A, Caudron Q, Grenfell BT, Levin SA, et al. 2014. Global antibiotic consumption 2000 to 2010: an analysis of national pharmaceutical sales data. Lancet Infect Dis 14(8):742–750, PMID: 25022435, 10.1016/S1473-3099(14)70780-7.

van den Bogaard AE, Willems R, London N, Top J, Stobberingh EE. 2002. Antibiotic resistance of faecal enterococci in poultry, poultry farmers and poultry slaughterers. J Antimicrob Chemother 49(3):497–505, PMID: 11864950.

You Y, Silbergeld EK. 2014. Learning from agriculture: understanding low-dose dantimicrobials as drivers of resistome expansion. Front Microbiol 5:284, 10.3389/fmicb.2014.00284.

Zhang XY, Ding LJ, Yue J. 2009. Occurrence and characteristics of class 1 and class 2 integrons in resistant Escherichia coli isolates from animals and farm workers in northeastern China. Microb Drug Resist 15(4):323–328, 10.1089/mdr.2009.0020.

The Saliva Exposome for Monitoring of Individuals’ Health Trajectories

Author Affiliations open

1Department of Chemistry, University of Waterloo, Waterloo, Ontario, Canada

2Center for Exposure Biology, School of Public Health, University of California, Berkeley, Berkeley, California, USA

PDF icon PDF Version (2 MB)

  • Background:
    There is increasing evidence that environmental, rather than genetic, factors are the major causes of most chronic diseases. By measuring entire classes of chemicals in archived biospecimens, exposome-wide association studies (EWAS) are being conducted to investigate associations between a myriad of exposures received during life and chronic diseases.
    Objectives:
    Because the intraindividual variability in biomarker levels, arising from changes in environmental exposures from conception onwards, leads to attenuation of exposure–disease associations, we posit that saliva can be collected repeatedly in longitudinal studies to reduce exposure–measurement errors in EWAS.
    Methods:
    From the literature and an open-source saliva–metabolome database, we obtained concentrations of 1,233 chemicals that had been detected in saliva. We connected salivary metabolites with human metabolic pathways and PubMed Medical Subject Heading (MeSH) terms, and performed pathway enrichment and pathway topology analyses.
    Results:
    One hundred ninety-six salivary metabolites were mapped into 49 metabolic pathways and connected with human metabolic diseases, central nervous system diseases, and neoplasms. We found that the saliva exposome represents at least 14 metabolic pathways, including amino acid metabolism, TCA cycle, gluconeogenesis, glutathione metabolism, pantothenate and CoA biosynthesis, and butanoate metabolism.
    Conclusions:
    Saliva contains molecular information worthy of interrogation via EWAS. The simplicity of specimen collection suggests that saliva offers a practical alternative to blood for measurements that can be used to characterize individual exposomes. https://doi.org/10.1289/EHP1011
  • Received: 23 August 2016
    Revised: 08 November 2016
    Accepted: 18 November 2016
    Published: 20 July 2017

    Address correspondence to V. Bessonneau, Silent Spring Institute, 320 Nevada St., Suite 302, Newton, Massachusetts 02460 USA. Telephone: (617) 332-4288. Email: vincent.bessonneau@gmail.com

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.


  • Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

Introduction

Because genetic factors typically account for only about 18% of chronic disease risks, it is reasonable to infer that nongenetic factors (i.e., exposures) are major causes of chronic diseases (Rappaport 2016). Given the myriad exposures from both exogenous and endogenous sources that an individual experiences during life [the “exposome” (Wild 2005)], investigators are performing exposome-wide association studies (EWAS) that interrogate levels of chemicals in biospecimens to discover causes of chronic diseases (Patel et al. 2010; Rappaport 2012; Wild et al. 2013). By measuring entire classes of chemicals (e.g., small molecules, protein modifications, antigens) in archived biospecimens from incident disease cases and matched controls, EWAS can pinpoint discriminating features that then generate hypotheses for targeted follow-up studies (Rappaport 2011; Rappaport et al. 2014). For example, Hazen and coworkers employed this avenue to implicate joint microbial/human metabolism of the nutrient choline as a potentially major cause of coronary heart disease (Wang et al. 2011; Tang et al. 2013; Koeth et al. 2013).

An important challenge to designing EWAS is the intraindividual variability in levels of circulating molecules arising from changes in diet, lifestyle factors, and sources of pollutants during decades of life that precede disease onset. This within-person variability in biomarker levels leads to exposure measurement errors that attenuate causal signals and obscure disease associations (Lin et al. 2005; Sampson et al. 2013). One way to circumvent such measurement errors is to perform longitudinal studies with repeated biospecimens, collected from subjects during critical stages of life (Rappaport 2011; Robinson and Vrijeheid 2015). The most logical approach for doing this relies on prospective cohorts that archived blood or other biospecimens repeatedly from the same subjects. However, such cohorts are rare and repeated collection of blood, the main archival specimen, is difficult to perform (Hansen et al. 2007; Randell et al. 2016).

Saliva (also referred as oral fluid) is a natural filtrate of blood that contains omic features (small molecules, metals, proteins, and DNA) worthy of interrogation via EWAS. Because collection is “stress-free,” repeated specimens of saliva are routinely obtained for determination of steroid hormones in psychobiological studies (Hjortskov et al. 2004; Kajantie and Phillips 2006; Hunter et al. 2011). Sampling of saliva is straightforward and protocols are available that allow subjects to collect their own samples and ship them to a laboratory or repository.

Metabolomics is recognized as a powerful top-down approach for detecting small molecules in biological matrices (Nicholson and Wilson 2003; German et al. 2005). These small molecules can be either substrates or end products of cellular metabolism and can originate from exogenous sources via inhalation, ingestion and dermal absorption, or from endogenous processes including human and microbial metabolism. Adductomics is another top-down technique that employs modifications of blood proteins like hemoglobin or human serum albumin (HSA) to characterize exposures to reactive electrophiles that are inherently toxic but cannot be measured directly in biospecimens (Rubino et al. 2009; Li et al. 2011; Carlsson et al. 2014; Grigoryan et al. 2016; Rappaport 2012). Because blood is in equilibrium with the tissues and saliva is in equilibrium with blood, both blood and saliva represent dynamic reservoirs of small molecules that are present in the body at a given time. Given the potential utility of saliva as a biospecimen for EWAS, we will evaluate the linkages between salivary metabolites and human metabolic pathways, as well as those between these pathways and chronic diseases. We will also consider methods for collection and analysis of saliva via untargeted metabolomics and adductomics.

Methods

Saliva Metabolome

Salivary metabolites (n=1,233) were obtained from the saliva metabolome database (http://www.salivametabolome.ca/) that was recently integrated into the Human Metabolome Database (HMDB) (Wishart et al. 2013; Dame et al. 2015). This database compiles important physical, chemical and biological information of metabolites derived from peer-reviewed articles, including chemical structure, chemical class, origin, biological function, and cellular location. Normal and abnormal concentrations are also reported for metabolites previously quantified in humans as well as standard deviations of the measurements that can be used in some cases to estimate the intra- and interindividual variability. It is important to note that this database was built from a recent study by Dame et al. who combined contemporary measurements of salivary metabolites in healthy subjects with literature data derived from the healthy and/or diseased subjects (Dame et al. 2015). Those investigators measured 308 salivary metabolites in 16 healthy adult subjects sampled once and two subjects sampled three times in a day (morning, midday, and afternoon/evening). They also reported levels of an additional 708 metabolites from peer-reviewed articles. All metabolites included in the saliva metabolome database were used for analyses.

Visualization of Human Metabolic Pathways

Salivary metabolites were connected to their human metabolic pathways using the Metscape 3.1 App (Karnovsky et al. 2012) for Cytoscape 3.2.1. (Shannon et al. 2003). The network of metabolites and reactions was built using the internal Metscape database that integrates metabolomics and gene-expression data derived from all available biospecimens, and compiled in the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the Edinburgh Human Metabolic Network (EHMN) (Karnovsky et al. 2012). Nodes (i.e., metabolites) were colored according to their source (e.g., host, microbial). Edges connecting nodes represent KEGG and EHMN biochemical reactions. Only salivary metabolites with a KEGG IDs (196 metabolites) were retained with “human” as the model organism. The network type “compound” was used to map metabolic connections between small molecules.

Pathway Enrichment and Pathway Topology Analysis

Pathway enrichment and topology analyses of the saliva metabolome were performed using MetaboAnalyst 3.0 (Xia et al. 2015) with “Homo sapiens” as the model organism. Pathway enrichment analysis was conducted using the hypergeometric test to determine whether saliva metabolites were represented in a particular KEGG metabolic pathway more than expected by chance. p-Values≤0.05 indicate statistically significant representations of metabolic pathways. Pathway topology analysis was performed using the “relative-betweenness” centrality measure to estimate the importance of saliva metabolites relative to the biological pathway structure. Metabolite centrality is reported on a scale from 0 (isolated metabolite) to 1 (key metabolite).

Connections between the Saliva Metabolome and Human Chronic Diseases

Salivary metabolites were connected with human chronic diseases using the MetDisease App for Cytoscape 3.2.1 (Duren et al. 2014). MetDisease is a text-mining App that queried metabolites previously reported in any biospecimen as associated with PubMed Medical Subject Heading (MeSH) terms in peer-reviewed human and/or animal studies. For each disease, the total number of salivary metabolites, associated with a particular MeSH term, was determined. Salivary metabolites were queried using their KEGG IDs as identifiers and their shared names as select attributes. MeSH terms related to both chronic and nonchronic diseases (e.g., bacterial infections) were included in the network. The network of MeSH terms was built with Cytoscape 3.2.1 (Shannon et al. 2003) using the organic layer. The number of salivary metabolites associated with MeSH terms was used as the node attributes. Edges connecting nodes represented interconnections between MeSH terms.

Results and Discussion

The Saliva Metabolome

Saliva is a mixture of fluids originating mainly from the parotid, submandibular, sublingual, and minor salivary glands, and to a lesser extent from oral and nasal mucosa (de Almeida et al. 2008). Saliva consists of approximately 99% water with the remaining 1% comprised of electrolytes, mucus, cellular debris, proteins, and small molecules (Humphrey and Williamson 2001; de Almeida et al. 2008). Transfer of small molecules from blood to saliva relies on passive diffusion of lipophilic compounds and ultrafiltration of hydrophilic compounds of low molecular weight (<300–1,900 Da) (Gallardo and Queiroz 2008; de Almeida et al. 2008; Thieme 2012; Higashi 2012). The pKa of a molecule plays an important role in transfer from blood to saliva. For example, basic molecules accumulate in saliva due to ion-trapping phenomena associated with their transfer from neutral blood (pH 7.4) to acidic saliva (pH 6) (Thieme 2012). Also, several factors including disease states, diet, drug consumption, and physical activity can significantly affect saliva excretion and saliva pH can be influenced by variation in bicarbonate concentrations (Aps and Martens, 2005).

Several studies have evaluated the correlation between saliva and blood concentrations for compounds with diverse physical and chemical properties, as summarized in Table 1. The median value of the Pearson correlation coefficients was 0.92 with a range of 0.30≤r≤0.98). High correlations were observed between blood and saliva concentrations for neutral molecules that are not affected by changes in saliva pH (Hill et al. 2001; Sakaguchi and Hasegawa 2005; Juniarto et al. 2011; Gunnala et al. 2015). Lower correlations for acidic and basic compounds have been attributed to intraindividual variation in saliva pH as well as buccal contamination (for nicotine) (Fisher et al. 2013). However, these studies have been conducted only on a small subset of compounds (mainly drugs and pollutants). Because the blood–saliva transfer of small molecules relies on physical and chemical processes, more in-depth studies are needed to determine effects of pKa, polarity, physical activity, diet, and disease state.

Table 1. Blood:saliva concentration ratios and correlations for various classes of small molecules.
Compound Log Pa pKa Blood: saliva ratio (mean) Correlation coefficient (r) Reference
Therapeutic drugs
 Clozapine 3.7 15.9 3.6b 0.728 Fisher et al. 2013
 Norclozapine 3.2 15.9 3.6b 0.806 Fisher et al. 2013
 Quetiapine 2.9 15.1 3.0b 0.843 Fisher et al. 2013
 Risperidone 3.4 8.8 2.6b 0.954 Fisher et al. 2013
 9-hydroxyrisperidone 2.3 13.7 2.5b 0.640 Fisher et al. 2013
 Alprazolam 2.2 18.3 2.3 ND Gjerde et al. 2014
 Clonazepam 2.8 11.9 7.1 ND Gjerde et al. 2014
 Diazepam 2.6 2.9 27.0 ND Gjerde et al. 2014
 Nordiazepam 2.8 12.3 22.0 ND Gjerde et al. 2014
 Busulfan −0.9 NA 0.92b 0.980 Rauh et al. 2006
 Methylphenidate 1.5 8.9 0.5b ND Seçilir et al. 2013
 Mycophenolic acid 2.4 3.6 NRb 0.922 Shen et al. 2009
 Voriconazole 1.6 12.7 2.04b 0.943 Vanstraelen et al. 2015
Illicit drugs
 Cocaine 1.9 8.8 0.4b ND Moolchan et al. 2000
 Benzoylecgonine 1.7 3.1 5b ND Moolchan et al. 2000
 Ecgonine methyl ester 0.1 14.6 1b ND Moolchan et al. 2000
Pollutants
 Nicotine 0.9 8.9 NRb 0.300 Shin et al. 2002
 Cotinine 0.4 4.8 2.3b 0.980 Shin et al. 2002
 Perchlorate −0.9 −6.9 0.07c 0.927 Oldi et al. 2009
Endogenous
 Cortisol 1.8 12.6 0.87 0.933 Gunnala et al. 2015
 7-HydroxyDHEA 1.8 18.2 4.5c 0.975 Hill et al. 2001
 Androstenedione 2.9 19.0 NRb 0.781 Juniarto et al. 2011
 17-Hydroxyprogesterone 2.9 12.7 NRb 0.964 Juniarto et al. 2011
 Testosterone 2.9 19.1 NRc 0.843 Sakaguchi and Hasegawa 2005

Note: ND, not determined.

aFrom in silico prediction using ALOGPS 2.1(http://www.vcclab.org/lab/alogps/).

bPlasma:saliva ratio.

cSerum:saliva ratio.

Dame et al. (2015) recently combined results from metabolomic analyses of human saliva with a literature review of salivary metabolites. This characterization of the saliva metabolome included 1,233 small molecules, metals and ions, which is about one-fourth of the 4,549 metabolites that have been reported in human blood (Wishart et al. 2013). The difference in coverage of the two metabolomes reflects the lower metabolite concentrations observed in saliva (nM to low μM) compared with blood (μM to mM). We classified the sources of these salivary molecules and metals as follows: a) host endogenous (879), b) microbial endogenous (52), c) food (225), d) drugs (15), e) pollutants (25), and f) metals (37) (Figure 1).

Pie chart of the distribution of following sources of salivary metabolites: host endogenous: 72 percent; food: 18 percent; microbial endogenous: 4 percent; metals: 3 percent; pollutants: 2 percent; drugs: 1 percent.
Figure 1. Source category of salivary metabolites compiled from the saliva metabolome database (Dame et al. 2015).

Using Metscape, we mapped the 1,233 salivary metabolites into biological pathways (Figure 2). Metscape uses an internal database that integrates metabolomic and gene-expression data from KEGG and EHMN (Karnovsky et al. 2012). Because only 196 of the 1,233 (16%) metabolites detected in saliva were included in these databases, an additional 529 metabolites were added to the network through linkages (i.e., direct neighbor in metabolic pathways) with the 196 salivary metabolites through 49 recognized pathways. Most of these additional 529 metabolites were either phase-II conjugates or compounds tightly bound to proteins. In saliva, metabolites are mainly present in the free form (i.e., unbound fraction) due to blood-saliva transfer processes. Phase-II conjugation increases the molecular weight, acidity, and hydrophilicity of each parent compound (Figure 2A), whereas protein binding increases the size of the compound. Both of these processes limit the efficiency of blood–saliva transfer. For compounds tightly bound to proteins, such as eicosanoids (markers of inflammation) (Figure 2A), the free form represents only about 1% of the total concentration in the systemic circulation (Brodersen et al. 1990; Fujiwara and Amisaki 2013; Bessonneau et al. 2015b). Salivary levels of these compounds are very low (nM) and, therefore, difficult to detect with untargeted analyses.

Linkage pathways of salivary metabolites classified by their source, namely, host endogenous, microbial endogenous, metals, food, drugs, and pollutants. Insets A and B are provided as magnified images in the same figure.
Figure 2. Network of metabolic pathways reconstructed from metabolites detected in human saliva. Gray nodes represent metabolites that had not previously been detected in saliva but have direct neighbors in metabolic pathways. Edges represent biochemical connections between metabolites.

Next, we performed pathway enrichment and pathway topology analyses to identify the most significant pathways in the saliva exposome. Table 2 summarizes the 14 most significant metabolic pathways (p≤0.05), which prominently includes amino acid metabolism, TCA cycle, gluconeogenesis, glutathione metabolism, pantothenate and CoA biosynthesis, and butanoate metabolism.

Table 2. Metabolic pathways represented by the saliva metabolome (metaboanalyst – homosapiens) for pathways having p-values<0.05 from pathway enrichment analysis. Impact values are derived from pathway topology analysis.
Pathway Total metabolitesa Hitsb p-valuec Impactd
Alanine, aspartate, and glutamate metabolism 24 13 1.4e−4 0.88
beta-Alanine metabolism 28 13 9.8e−4 0.38
Phenylalanine metabolism 45 18 9.9e−4 0.45
Arginine and proline metabolism 77 26 1.7e−3 0.73
Nitrogen metabolism 39 15 4.0e−3 0.22
Citrate cycle (TCA cycle) 20 9 7.7e−3 0.43
Glycolysis or gluconeogenesis 31 12 9.3e−3 0.32
D-Glutamine and D-glutamate metabolism 11 6 9.7e−3 0.64
Glycine, serine, and threonine metabolism 48 16 1.4e−2 0.56
Glutathione metabolism 38 13 2.1e−2 0.35
Pantothenate and CoA biosynthesis 27 10 2.4e−2 0.35
Valine, leucine, and isoleucine biosynthesis 27 10 2.4e−2 0.36
Butanoate metabolism 40 13 3.3e−2 0.21
D-Arginine and D-ornithine metabolism 8 4 5.0e−2 0.50

aTotal number of metabolites involved in the pathway.

bNumber of salivary metabolites involved in the pathway.

cFisher’s exact test was used to measure the association between input metabolites (i.e., salivary metabolites) and pathways. p-Values≤5.0e−2 indicate that associations are less likely due to random chance.

dImpact represents, on a scale from 0 to 1, the centrality of salivary metabolites in their respective pathways.

Connections between the Saliva Metabolome and Human Diseases

In order to provide additional biological context for the saliva metabolome, we evaluated associations between salivary metabolites and MeSH terms using MetDisease. The resulting network (Figures 3 and 4) indicates that salivary metabolites were associated with most human diseases (Table 3). For example, substantial evidence from breast cancer epidemiology shows the influence of hormones at different stages of women’s development. A nested case–control study found that premenopausal concentrations of testosterone were associated with breast cancer risk (Zeleniuch-Jacquotte et al. 2012). Another case–control study reported lower salivary testosterone levels in women with breast cancer, suggesting a protective effect of testosterone (Dimitrakakis et al. 2010). The carboxylic acid ester phenylacetate and its analogs have also been linked to breast neoplasms (Sawatsri et al. 2001) due to their antiproliferative activities on human breast cancer cells. Recently, targeted metabolomics of salivary levels of polyamines such as spermidine has discriminated breast cancer patients from healthy subjects (Takayama et al. 2016). Polyamines are essential molecules for normal cell growth and regulate gene expression by modulating ligand–receptor interactions, including estradiol binding to estrogens receptors (Cervelli et al. 2014). Previous studies have attributed the proliferation of highly invasive breast cancer tumor cells to the upregulation of polyamine metabolism (Cervelli et al. 2014).

Linkage pathways indicating the MeSH terms for most human chronic diseases associated with salivary metabolites.
Figure 3. Network of PubMed Medical Subject Headings (MeSH) terms reported associated with salivary metabolites. The size of a node and text reflects the number of metabolites associated with MeSH terms. Edges represent links between MeSH terms.
Linkage pathways indicating the MeSH terms for neoplasms associated with salivary metabolites.
Figure 4. Subnetwork of neoplasm-related PubMed Medical Subject Headings (MeSH) terms reported associated with salivary metabolites. The size of a node and text reflects the number of metabolites associated with MeSH terms. Edges represent links between MeSH terms.
Table 3. Summary of the most important human chronic diseases associated with salivary metabolites.
Human chronic disease Number of salivary metabolites
Congenital, hereditary, and neonatal diseases and abnormalities 145
Nutritional and metabolic diseases 144
Central nervous system diseases 138
Neoplasms 118
Digestive system diseases 105
Urogenital diseases 103
Cardiovascular diseases 101
Immune system diseases 76
Respiratory tract diseases 69
Endocrine system diseases 64

There is also growing evidence that the human gut microbiome and its complex interactions with exogenous exposures play roles in disease processes (Jiménez et al. 2008; Orešič et al. 2008; Nicholson et al. 2012). Of the 14 metabolites of known microbial origin, 12 were associated with congenital, hereditary, and neonatal diseases, 11 with nervous system diseases, 10 with nutritional and metabolic diseases, eight with digestive system diseases, and seven each with neoplasms and urogenital and pregnancy complications.

The above analyses suggest that the saliva metabolome captures a biologically meaningful fraction of exposures that are associated with human diseases. Thus, any discriminating salivary molecules detected by EWAS should be well worth following up in targeted studies to examine sources, causality, disease mechanisms, and interventions (Rappaport 2012). We envision that creating exposure terms, similar to MeSH terms indexed to the peer-reviewed literature, would allow annotation of metabolites based on their origins and facilitate investigation of exposure sources and temporality of exposure–disease associations. For example, Scalbert et al. (2014) performed a comprehensive review of metabolites associated with specific food groups (e.g., red meat, fish, vegetables) from human dietary interventions and cross-sectional studies. They reported that many metabolites were significantly correlated (Pearson correlation r>0.3, p<0.01) with food consumption; e.g., apple consumption was associated with kaempferol, isorhamnetin, m-coumaric acid and phloretin, carrot consumption with α-carotene, and soy consumption with daidzein, genistein, isoflavones, and O-desmethylangolensin. Likewise, Rappaport et al. (2014) compiled literature values for 94 environmental chemicals and nutrients that have been measured in the general population.

Saliva as a Convenient Biospecimen for Longitudinal Studies

Saliva offers several advantages over traditional biospecimens archived in population-based studies. As with blood, saliva provides a snapshot of the internal exposome at the time of collection. However, whereas blood sampling usually requires a trained phlebotomist, saliva can be collected by the subjects themselves, using commercial kits (Shirtcliff et al. 2000; Abraham et al. 2012). This advantage should increase participation rates in cohort studies, particularly when repeated sampling is desired. In a recent study of inflammatory bowel disease, Randell et al. (2016) reported that the participation rate was significantly higher across 591 participants when they were asked to contribute a saliva sample (38% participation), compared with blood sample (23% participation). Similarly, Hansen et al. (2007) found a participation rate of 72% when individuals were asked to deliver saliva samples for DNA analysis, compared with 31% for venous blood samples collected at a medical facility. Another investigation revealed that the participation rates for collection of saliva samples and dried blood spots (DBS) were equal at 71.0% for 4,600 women (Sakhi et al. 2015). Although DBS specimens are also minimally invasive and amenable to self-collection by participants, the volume of blood is small (approximately 50 μL per drop) and the analytical matrix consists of both serum and hemolysate, thereby complicating untargeted analysis (Vuckovic 2012). Also, it is unclear whether subjects would be amenable to repeated collection of DBS, which does involve some discomfort. Thus, saliva specimens offer an attractive alternative to blood sampling for investigating individual exposomes with the high temporal resolution (i.e., repeated measurements) needed to explore health trajectories. Several studies have found low day-to-day variability in salivary levels of tightly regulated metabolites such as steroid hormones (Gann et al. 2001; Viardot et al. 2005). For example, Gann et al. (2001) observed that the intraclass correlation for peak progesterone between two consecutive menstrual cycles in healthy premenopausal women ranged from 0.72 to 0.76, indicating that cycle-to-cycle variability within women was smaller than variability across women.

Analytical Considerations

Saliva can easily be collected with a variety of commercial devices such as Salivette® or Drugwipe®. The most common approach is to collect saliva with an absorbent pad or swab, from which saliva is then recovered in the laboratory by centrifugation (Drummer 2008). Prior to sampling, saliva production can be stimulated with citric acid and by chewing of gum or sterile pads (Gröschl et al. 2008; Lund et al. 2011). This approach is often used to increase the volume of saliva collected and to control the pH so as to standardize the transfer of acidic and basic molecules. Unstimulated saliva can be collected passively with a collection tube or an oral swab (Drummer 2008). Collection devices should be selected carefully because some media, including cotton and polypropylene pads, can nonspecifically bind small molecules (Gallardo and Queiroz 2008; de Almeida et al. 2008; Gröschl et al. 2008; Higashi 2012). Given that the buccal cavity can be contaminated by components originating from previous oral ingestion (Shin et al. 2002), it is useful for subjects to refrain from eating and drinking for a few hours prior to saliva collection. Once collected, saliva samples can be easily transported by mail to the laboratory for analysis. Because a large number of salivary metabolites such as amino acids, steroids, or fatty acids are under circadian control (i.e., high diurnal variation) (Dallmann et al. 2012; Dame et al. 2015), it is important to collect repeated samples at similar times (e.g., all samples collected during early morning). Saliva is also an attractive alternative matrix for pediatric populations, due to the noninvasiveness of the sample collection. Although several collection devices have been specifically developed for children 6 mo–6 y old, more studies are required to address the safety of saliva sampling devices and procedures in young children.

Several studies have demonstrated that various classes of compounds, including illicit drugs (Lund et al. 2011), therapeutic drugs (Mendonza et al. 2006; Moore et al. 2007; Ogawa et al. 2014), pollutants (Bentley et al. 1999; Wang and Lu 2009; Gherardi et al. 2010), and endogenous steroids (Higashi et al. 2011; Alvi et al. 2013) are stable in saliva for a few days at room temperature and for at least 1 year when stored at low temperature (≤−20°C).

Sample Preparation

The strategy for preparing samples plays an important role in the quality of metabolomics data. Issues related to the quenching of metabolism prior to analysis, ion suppression during mass spectrometry and metabolite instability can adversely impact the interpretation of data from untargeted analysis. The most common techniques for preparing saliva samples for mass spectrometry are liquid–liquid extraction (LLE), protein precipitation (PP), and solid-phase extraction (SPE) (Higashi 2012). Although LLE and PP provide exhaustive extraction of small molecules, such procedures are labor intensive, difficult to automate, require multistep sample handling, and are prone to suppression or enhancement of ionization in mass spectrometric analysis. Most of metabolites in the saliva metabolome database have been obtained through LLE or PP of saliva samples. These analytical techniques often used an internal standard in order to correct for possible loss and/or degradation during sample preparation and for matrix effects during analysis.

In contrast to LLE and PP, solid-phase microextraction (SPME) offers a simple, fast, and sensitive technique for preparing saliva and other biological fluids for metabolomics analysis (Bojko et al. 2014; Bessonneau et al. 2013, 2015a). In fact, when coupled to the Concept-96 system (employing 96-well plates), thin-film SPME (TF-SPME) motivates automation of all preconditioning, extraction, rinsing, and desorption steps (Jiang and Pawliszyn 2012). For example, preparation for saliva via TF-SPME can be achieved in less than 2 min per sample and prevents matrix effects and adsorption of macromolecules compared with LLE (Bessonneau et al. 2015a).

The small size and biocompatibility of SPME materials also permits in vivo sampling of saliva (Bessonneau et al. 2015a). By placing the SPME probe in the mouth for a short period of time, this technique can be used for rapid extraction and stabilization of metabolites for analysis (Bessonneau et al. 2015b). For example, SPME coatings were able to rapidly stabilize highly reactive metabolites (i.e., eicosanoids) formed in vivo and prevented their autooxidation during storage (Bessonneau et al. 2015b). Although promising, in vivo sampling of saliva with SPME is a new technique that will require substantial validation to address variability arising from salivary flow rate, pH, sample volume, and agitation.

The Saliva Adductome

Biotransformation of dietary chemicals and pollutants leads to the formation of reactive electrophiles, including reactive oxygen and nitrogen species, aldehydes, oxiranes, and quinones that can modify DNA, proteins, and lipids and can lead to dysregulation of homeostasis (Farmer and Davoine 2007; Fritz and Petersen 2013). Because reactive species are unstable, it is difficult to measure them in their free-circulating form. However, one can detect adducts formed from reactions between reactive electrophiles and nucleophilic residues on DNA and proteins. Several targeted studies have reported correlations between adducts of reactive electrophiles from endogenous and exogenous sources with DNA (Sturla 2007,) and prominent blood proteins, that is, hemoglobin (Hb) and human serum albumin (HSA) [reviewed by Törnqvist et al. 2002; Rubino et al. 2009].

Both DNA and proteins are also present in saliva, albeit at lower concentrations than in blood. For example, HSA is found in saliva at concentrations of about 0.5 mg/mL compared with 35–50 mg/mL in serum (Wang et al. 2012; Shaila et al. 2013; Metgud and Patel 2014; Nam et al. 2015); whereas Hb is present at a 50,000-fold lower concentration in saliva (3.3 μg/mL) compared with blood (150 mg/mL) (Nomura et al. 2012). For DNA, mean salivary levels (12 μg/mL) are 2-fold lower than those in blood (26 μg/mL) (Abraham et al. 2012). Although the analysis of salivary adducts is still in its infancy, targeted studies have reported adducts of salivary DNA with electrophiles from smoking and dietary sources (Bessette et al. 2010; Chen and Lee 2014; Chen and Lin 2014), and of HSA adducts in nasal lavage fluid from workers exposed to reactive electrophiles (Kristiansson et al. 2004; Jeppsson et al. 2009).

Untargeted analysis of adducts of DNA, HAS, and Hb has motivated adductomics for untargeted characterization of exposures to reactive electrophiles (Rappaport et al. 2012). That is, an adductome represents the totality of adducts generated from reactions between reactive electrophiles and a particular nucleophilic locus on one of these molecules. By characterizing a complete adductome, it is possible to map systemic exposures that occurred over the in vivo residence time of the nucleophile, which can range from hours to months. Particular attention has been paid to mass spectrometric characterization of DNA adductomes in human tissue samples (reviewed by Balbo et al. 2014) and the blood proteins, HSA (Li et al. 2011; Grigoryan et al. 2016) and Hb (Carlsson et al. 2014).

Here, we posit that salivary HSA and Hb residence times would be similar to those in blood (i.e., 30 days for HSA and 60 days for Hb), which are much longer than those for DNA (a few days). Consequently, by performing untargeted mass spectrometry of HSA and Hb adducts in saliva samples, it should be possible to characterize systemic exposures to reactive electrophiles during 1–2 mo prior to specimen collection. With repeated saliva sampling every few years, one can construct an individual exposure history of relevance to a person’s health trajectory.

Conclusions

Saliva contains a rich set of molecular information for circulating chemicals that can be interrogated via EWAS, with metabolomics and adductomics, to discover exposure–risk factors for chronic diseases. Although the number of metabolites detected in saliva is smaller than that in blood, we can anticipate that analytical improvements will uncover many additional salivary metabolites, present at low concentrations.

Given the simplicity and noninvasiveness of specimen collection, saliva offers a practical alternative to blood for longitudinal measurements of individual’s exposomes. Several studies have shown that participants are more willing to donate saliva specimens than venous blood, and saliva can be collected by the subjects themselves. This would eventually improve participation rates in cohort studies and, therefore, generate larger sets of biospecimens.

Data-driven studies that utilize repeated omic measurements from individuals at different stages of life should reduce exposure measurement errors and thereby increase power to discover unknown causes of chronic diseases. Given the ease of collection, saliva could well be the specimen of choice for obtaining repeated samples to profile small molecules, DNA, and proteins.

Acknowledgments

Support for this work was provided by the Natural Sciences and Engineering Research Council of Canada Industrial Research program and the Canada Research Chairs program (V.B. and J.P.) and by the U.S. National Institutes of Health through grants P42ES04705 and R33CA191159 (S.M.R.).

References

Abraham JE, Maranian MJ, Spiteri I, Russell R, Ingle S, Luccarini C, et al. 2012. Saliva samples are a viable alternative to blood samples as a source of DNA for high throughput genotyping. BMC Med Genomics 5:19, 10.1186/1755-8794-5-19.

Alvi SN, Al Dgither S, Hammami MM. 2013. Development and validation of LC-MS/MS method for determination of testosterone level in human saliva using lovastatin as internal standard. J Bioequiv Availab 5:228–232.

Aps JK, Martens LC. 2005. Review: the physiology of saliva and transfer of drugs into saliva. Forensic Sci Int 150:119–131, PMID: 15944052, 10.1016/j.forsciint.2004.10.026.

Balbo S, Turesky RJ, Villalta PW. 2014. DNA adductomics. Chem Res Toxicol 27:356–366, PMID: 24437709, 10.1021/tx4004352.

Bentley MC, Abrar M, Kelk M, Cook J, Phillip K. 1999. Validation of an assay for the determination of cotinine and 3-hydroxycotinine in human saliva using automated solid-phase extraction and liquid-chromatography with tandem mass spectrometric detection. J Chromatogr B Biomed Sci Appl 723:185–194, 10.1016/S0378-4347(98)00494-0.

Bessette EE, Spivack SD, Goodenough AK, Wang T, Pinto S, Kadlubar FF, et al. 2010. Identification of carcinogen DNA adducts in human saliva by linear quadrupole ion trap/multistage tandem mass spectrometry. Chem Res Toxicol 23:1234–1244, PMID: 20443584, 10.1021/tx100098f.

Bessonneau V, Bojko B, Pawliszyn J. 2013. Analysis of human saliva metabolome by direct immersion solid-phase microextraction LC and benchtop orbitrap MS. Bioanalysis 5:783–792, PMID: 23534423, 10.4155/bio.13.35.

Bessonneau V, Boyaci E, Maciazek-Jurczyk M, Pawliszyn J. 2015a. In vivo solid-phase microextraction sampling of human saliva for non-invasive and on-site monitoring. Anal Chim Acta 856:35–45.

Bessonneau V, Zhan Y, Lannoy IAM, Saldivia V, Pawliszyn J. 2015b. In vivo solid-phase microextraction liquid-chromatography-tandem mass spectrometry for monitoring blood eicosanoids time profile after lipopolysaccharide-induced inflammation in Sprague-Dawley rats. J Chromatogr A 1424:134–138.

Bojko B, Reyes-Garcés N, Bessonneau V, Goryński K, Mousavi F, Souza Silva EA, et al. 2014. Solid-phase microextraction in metabolomics. Trends Anal Chem 61:168–180, 10.1016/j.trac.2014.07.005.

Brodersen R, Andersen S, Vorum H, Nielsen SU, Pedersen O. 1990. Multiple fatty acid binding to albumin in human blood plasma. Eur J Biochem 189:343–349, PMID: 2338079.

Carlsson H, von Stedingk H, Nilsson U, Törnqvist M. 2014. LC-MS/MS screening strategy for unknown adducts to N-terminal valine in hemoglobin applied to smokers and nonsmokers. Chem Res Toxicol 27:2062–2070, PMID: 25350717, 10.1021/tx5002749.

Cervelli M, Pietropaoli S, Signore F, Amendola R, Mariottini P. 2014. Polyamines metabolism and breast cancer: state of the art and perspectives. Breast Cancer Res Treat 148:233–248, PMID: 25292420, 10.1007/s10549-014-3156-7.

Chen HJ, Lee CR. 2014a. Detection and simultaneous quantification of three smoking-related ethylthymidine adducts in human salivary DNA by liquid chromatography tandem mass spectrometry. Toxicol Lett 224:101–107.

Chen HJ, Lin CR. 2014b. Noninvasive measurement of smoking-associated N(3)-ethyladenine and N(7)-ethylguanine in human salivary DNA by stable isotope dilution nanoflow liquid chromatography-nanospray ionization tandem mass spectrometry. Toxicol Lett 225:27–33.

Dallmann R, Viola AU, Tarokh L, Cajochen C, Brown SA. 2012. The human circadian metabolome. Proc Natl Acad Sci U S A 109:2625–2629, PMID: 22308371, 10.1073/pnas.1114410109.

Dame ZT, Aziat F, Mandal R, Krishnamurthy R, Bouatra S, Borzouie S, et al. 2015. The human saliva metabolome. Metabolomics 11:1864–1883, 10.1007/s11306-015-0840-5.

de Almeida Pdel V, Grégio AM, Machado MA, de Lima AA, Azevedo LR. 2008. Saliva composition and functions: a comprehensive review. J Contemp Dent Pract 9:72–80, PMID: 18335122.

Dimitrakakis C, Zava D, Marinopoulos S, Tsigginou A, Antsaklis A, Glaser R. 2010. Low salivary testosterone in patients with breast cancer. BMC Cancer 10:547–554, PMID: 20937135, 10.1186/1471-2407-10-547.

Drummer OH. 2008. Introduction and review of collection techniques and applications of drug testing of oral fluid. Ther Drug Monit 30:203–206, PMID: 18367981, 10.1097/FTD.0b013e3181679015.

Duren W, Weymouth T, Hull T, Omenn GS, Athey B, Burant C, et al. 2014. MetDisease—connecting metabolites to diseases via literature. Bioinformatics 30:2239–2241, PMID: 24713438, 10.1093/bioinformatics/btu179.

Farmer EE, Davoine C. 2007. Reactive electrophile species. Curr Opin Plant Biol 10:380–386, PMID: 17646124, 10.1016/j.pbi.2007.04.019.

Fisher DS, van Schalkwyk GI, Seedat S, Curran SR, Flanagan RJ. 2013. Plasma, oral fluid, and whole-blood distribution of antipsychotics and metabolites in clinical samples. Ther Drug Monit 35:345–351, PMID: 23666566, 10.1097/FTD.0b013e318283eaf2.

Fritz KS, Petersen DR. 2013. An overview of the chemistry and biology of reactive aldehydes. Free Radic Biol Med 59:85–91, PMID: 22750507, 10.1016/j.freeradbiomed.2012.06.025.

Fujiwara S, Amisaki T. 2013. Fatty acid binding to serum albumin: molecular simulation approaches. Biochim Biophys Acta Gen Subj 1830:5427–5434, 10.1016/j.bbagen.2013.03.032.

Gann PH, Giovanazzi S, Van Horn L, Branning A, Chatterton RT Jr. 2001. Saliva as a medium for investigating intra- and interindividual differences in sex hormone levels in premenopausal women. Cancer Epidemiol Biomarkers Prev 10:59–64, PMID: 11205490.

Gallardo E, Queiroz JA. 2008. The role of alternative specimens in toxicological analysis. Biomed Chromatogr 22:795–821, PMID: 18506679, 10.1002/bmc.1009.

German JB, Hammock BD, Watkins SM. 2005. Metabolomics: building on a century of biochemistry to guide human health. Metabolomics 1:3–9, PMID: 16680201, 10.1007/s11306-005-1102-8.

Gherardi M, Gordiani A, Gatto M. 2010. Development and validation of method for analysis of some ototoxic solvents in saliva matrix by headspace gas chromatography/mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 878:2391–2396, PMID: 20727840, 10.1016/j.jchromb.2010.07.007.

Gjerde H, Langel K, Favretto D, Verstraete AG. 2014. Detection of 4 benzodiazepines in oral fluid as biomarker for presence in blood. Ther Drug Monit 36:252–256, PMID: 24061449, 10.1097/FTD.0b013e3182a3ab42.

Gröschl M, Köhler H, Topf HG, Rupprecht T, Rauh M. 2008. Evaluation of saliva collection devices for the analysis of steroids, peptides and therapeutic drugs. J Pharm Biomed Anal 47:478–486, PMID: 18325706, 10.1016/j.jpba.2008.01.033.

Grigoryan H, Edmands W, Lu SS, Yano Y, Regazzoni L, Iavarone AT, et al. 2016. An adductomics pipeline for untargeted analysis of modifications to Cys34 of human serum albumin. Anal Chem , 10.1021/acs.analchem.6b02553.

Gunnala V, Guo R, Minutti C, Durazo-Arvizu R, Laporte C, Mathews H, et al. 2015. Measurement of salivary cortisol level for the diagnosis of critical illness-related corticosteroid insufficiency in children. Pediatr Crit Care Med 6(4):e101–e106, 10.1097/PCC.0000000000000361.

Hansen TO, Simonsen MK, Nielsen FC, Hundrup YA. 2007. Collection of blood, saliva, and buccal cell samples in a pilot study on the Danish nurse cohort: comparison of the response rate and quality of genomic DNA. Cancer Epidemiol Biomarkers Prev 16:2072–2076, 10.1158/1055-9965.EPI-07-0611.

Higashi T. 2012. Salivary hormone measurement using LC/MS/MS: specific and patient-friendly for assessment of endocrine function. Bio Pharm Bull 35:1401–1408, 10.1248/bpb.b212009.

Higashi T, Ito K, Narushima M, Sugiura T, Inagaki S, Min JZ, et al. 2011. Development and validation of stable-isotope dilution liquid chromatography–tandem mass spectrometric method for determination of salivary progesterone. Biomed Chromatogr 25:1175–1180, PMID: 21294140, 10.1002/bmc.1586.

Hill M, Lapcík O, Havlíková H, Morfin R, Hampl R. 2001. 7-Hydroxydehydroepiandrosterone epimers in human serum and saliva. Comparison of gas chromatography-mass spectrometry and radioimmunoassay. J Chromatogr A 935:297–307, PMID: 11762781.

Hjortskov N, Garde AH, Ørbæk P, Hansen AM. 2004. Evaluation of salivary cortisol as a biomarker of self-reported mental stress in field studies. Stress Health 20:91–98, 10.1002/smi.1000.

Humphrey SP, Williamson RT. 2001. A review of saliva: normal composition, flow and function. J Prosthet Dent 85:162–169, PMID: 11208206, 10.1067/mpr.2001.113778.

Hunter AL, Minnis H, Wilson P. 2011. Altered stress responses in children exposed to early adversity: a systematic review of salivary cortisol studies. Stress 14:614–626, PMID: 21675865, 10.3109/10253890.2011.577848.

Jeppsson MC, Lindh CH, Kristiansson MH, Nielsen J, Jönsson BA. 2009. Methylhexahydrophthalic anhydride adducted albumin tryptic peptides in nasal lavage fluid. Inhal Toxicol 21:1013–1020, PMID: 19772480, 10.1080/08958370802715997.

Jiang R, Pawliszyn J. 2012. Thin-film microextraction offers another geometry for solid-phase microextraction. Trends Anal Chem 39:245–253, 10.1016/j.trac.2012.07.005.

Jiménez E, Marín ML, Martín R, Odriozola JM, Olivares M, Xaus J, et al. 2008. Is meconium from healthy newborns actually sterile? Res Microbiol 159:187–193, PMID: 18281199, 10.1016/j.resmic.2007.12.007.

Juniarto AZ, Goosens K, Setyawati BA, Drop SL, de Jong FH, Faradz SM. 2011. Correlation between androstenedione and 17-hydroxyprogesterone levels in saliva and plasma of patients with congenital adrenal hyperplasia. Singapore Med J 52:810–813.

Kajantie E, Phillips DIW. 2006. The effects of sex and hormonal status on the physiological response to acute psychosocial stress. Psychoneuroendocrinology 31:151–178, 10.1016/j.psyneuen.2005.07.002.

Karnovsky A, Weymouth T, Hull T, Tarcea VG, Scardoni G, Laudanna C, et al. 2012. Metscape 2 bioinformatics tool for the analysis and visualization of metabolomics and gene expression data. Bioinformatics 28:373–380, PMID: 22135418, 10.1093/bioinformatics/btr661.

Koeth RA, Wang Z, Levison BS, Buffa JA, Org E, Sheehy BT, et al. 2013. Intestinal microbiota metabolism of L-carnitine, a nutrient in red meat, promotes atherosclerosis. Nat Med 19:576–585, 10.1038/nm.3145.

Kristiansson MH, Lindh CH, Jönsson BA. 2004. Correlations between air levels of hexahydrophthalic anhydride (HHPA) and HHPA-adducted albumin tryptic peptides in nasal lavage fluid from experimentally exposed volunteers. Rapid Commun Mass Spectrom 18:1592–1598, PMID: 15282784, 10.1002/rcm.1527.

Li H, Grigoryan H, Funk WE, Lu SS, Rose S, Williams ER, et al. 2011. Profiling Cys34 adducts of human serum albumin by fixed-step selected reaction monitoring. Mol Cell Proteomics 10:M110.004606, 10.1074/mcp.M110.004606.

Lin YS, Kupper LL, Rappaport SM. 2005. Air samples versus biomarkers for epidemiology. Occup Environ Med 62:750–760, PMID: 16234400, 10.1136/oem.2004.013102.

Lund HME, Øiestad EL, Gjerde H, Christophersen AS. 2011. Drugs of abuse in oral fluid collected by two different sample kits – stability testing and validation using ultra performance tandem mass spectrometry analysis. J Chromatogr B 879:3367–3377, 10.1016/j.jchromb.2011.09.002.

Mendonza AE, Gohh RY, Akhlaghi F. 2006. Analysis of mycophenolic acid in saliva using liquid chromatography tandem mass spectrometry. Ther Drug Monit 28:402–406, PMID: 16778726, 10.1097/01.ftd.0000211826.65607.05.

Metgud R, Patel S. 2014. Serum and salivary levels of albumin as diagnostic tools for oral pre-malignancy and oral malignancy. Biotech Histochem 89:8–13, PMID: 23738795, 10.3109/10520295.2013.793394.

Moolchan ET, Cone EJ, Wstadik A, Huestis MA, Preston KL, et al. 2000. Cocaine and metabolite elimination patterns in chronic cocaine users during cessation: plasma and saliva analysis. J Anal Toxicol 24:458–466, PMID: 11043647.

Moore C, Rana S, Coulter C. 2007. Determination of meperidine, tramadol and oxycodone in human oral fluid using solid phase extraction and gas chromatography-mass spectrometry. J Chromatogr B Anlyt Technol Biomed Life Sci 850:370–375, 10.1016/j.jchromb.2006.12.008.

Nam SH, Jung HI, Kang SM, Inaba D, Kwon HK, Kim BI. 2015. Validity of screening methods for periodontitis using salivary hemoglobin level and self-report questionnaires in people with disabilities. J Periodontol 86:536–545, PMID: 25569125, 10.1902/jop.2015.140457.

Nicholson JK, Holmes E, Kinross J, Burcelin R, Gibson G, Jia W, et al. 2012. Host-gut microbiota metabolic interactions. Science 336:1262–1267, PMID: 22674330, 10.1126/science.1223813.

Nicholson JK, Wilson ID. 2003. Understanding ‘global’ systems biology: metabonomics and the continuum of metabolism. Nat Rev Drug Discov 2:668–676, PMID: 12904817, 10.1038/nrd1157.

Nomura Y, Tamaki Y, Eto A, Kakuta E, Ogino D, Nakamura Y, et al. 2012. Screening for periodontal diseases using salivary lactate dehydrogenase, hemoglobin level, and statistical modeling. J Dent Sci 7:379–383, 10.1016/j.jds.2012.09.005.

Ogawa S, Tadokoro H, Sato M, Higashi T. 2014. Enantioselective determination of ibuprofen in saliva by liquid chromatography/tandem mass spectrometry with chiral electrospray ionization-enhancing and stable isotope-coded derivatization. J Pharm Biomed Anal 98:387–392, PMID: 24999866, 10.1016/j.jpba.2014.06.024.

Oldi JF, Kannan K. 2009. Perchlorate in human blood serum and plasma: Relationship to concentrations in saliva. Chemosphere 77(1):43–47, PMID: 19564037, 10.1016/j.chemosphere.2009.05.047.

Orešič M, Simell S, Sysi-Aho M, Näntö-Salonen K, Seppänen-Laakso T, Parikka V, et al. 2008. Dysregulation of lipid and amino acid metabolism precedes islet autoimmunity in children who later progress to type 1 diabetes. J Exp Med 205:2975–2984, PMID: 19075291, 10.1084/jem.20081800.

Patel CJ, Bhattacharya J, Butte AJ. 2010. An Environment-Wide Association Study (EWAS) on type 2 diabetes mellitus. PLoS One 5(5):e10746, PMID: 20505766, 10.1371/journal.pone.0010746.

Randell RL, Gulati AS, Cook SF, Martin CF, Chen W, Jaeger EL, et al. 2016. Collecting biospecimens from an internet-based prospective cohort study of inflammatory bowel disease (CCFA Partners): a feasibility study. JMIR Res Protoc 5(1):e3, 10.2196/resprot.5171.

Rappaport SM. 2011. Implications of the exposome for exposure science. J Expo Sci Environ Epidemiol 21:5–9, PMID: 21081972, 10.1038/jes.2010.50.

Rappaport SM. 2012. Biomarkers intersect with the exposome. Biomarkers 17:483–489, PMID: 22672124, 10.3109/1354750X.2012.691553.

Rappaport SM. 2016. Genetic factors are not the major causes of chronic diseases. PLoS One 11(4):e0154387, PMID: 27105432, 10.1371/journal.pone.0154387.

Rappaport SM, Barupal DK, Wishart D, Vineis P, Scalbert A. 2014. The blood exposome and its role in discovering causes of disease. Environ Health Perspect 122:769–774, PMID: 24659601, 10.1289/ehp.1308015.

Rappaport SM, Li H, Grigoryan H, Funk W, Williams ER. 2012. Adductomics: characterizing exposures to reactive electrophiles. Toxicol Lett 213:83–90, PMID: 21501670, 10.1016/j.toxlet.2011.04.002.

Rauh M, Stachel D, Kuhlen M, Groschl M, Holter W, Rascher W. 2006. Quantification of busulfan in saliva and plasma in haematopoietic stem cell transplantation in children: validation of liquid chromatography tandem mass spectrometry method. Clin Pharmacokinet 45(3):305–316, PMID: 16509762, 10.2165/00003088-200645030-00006.

Robinson O, Vrijeheid M. 2015. The pregnancy exposome. Curr Environ Health Rep 2:204–213, PMID: 26231368, 10.1007/s40572-015-0043-2.

Rubino FM, Pitton M, Di Fabio D, Colombi A. 2009. Toward an “omic” physiopathology of reactive chemicals: thirty years of mass spectrometric study of the protein adducts with endogenous and xenobiotic compounds. Mass Spectrom Rev 28:725–784, PMID: 19127566, 10.1002/mas.20207.

Sakaguchi K, Hasegawa T. 2005. Analysis of salivary testosterone by liquid chromatography-tandem mass spectrometry: correlation with serum bioavailable testosterone and aging. Rinsho Byori 53:388–394.

Sakhi AK, Bastani NE, Ellingjord-Dale M, Gundersen TE, Blomhoff R, Ursin G. 2015. Feasibility of self-sampled dried blood spot and saliva samples sent by mail in population-based study. BMC Cancer 15:265–273, PMID: 25886002, 10.1186/s12885-015-1275-0.

Sampson JN, Boca SM, Shu XO, Stolzenberg-Solomon RZ, Matthews CE, Hsing AW, et al. 2013. Metabolomics in epidemiology: sources of variability in metabolite measurements and implications. Cancer Epidemiol Biomarkers Prev 22:631–640, PMID: 23396963, 10.1158/1055-9965.EPI-12-1109.

Sawatsri S, Samid D, Malkapuram S, Sidell N. 2001. Inhibition of estrogen-dependent breast cell responses with phenylacetate. Int J Cancer 93:687–692, PMID: 11477579.

Scalbert A, Brennan L, Manach C, Andres-Lacueva C, Dragsted LO, Rappaport SM, et al. 2014. The food metabolome: a window over dietary exposure. Am J Clin Nutr 99:1286–1308, PMID: 24760973, 10.3945/ajcn.113.076133.

Seçilir A, Schrier L, Bijleveld YA, Toersche JH, Jorjani S, Burggraaf J, et al.. 2013. Determination of methylphenidate in plasma and saliva by liquid chromatography/tandem mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 923–924:22–28, PMID: 23454305, 10.1016/j.jchromb.2013.01.027.

Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. 2003. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 13:2498–2504, PMID: 14597658, 10.1101/gr.1239303.

Shaila M, Pai GP, Shetty P. 2013. Salivary protein concentration, flow rate, buffer capacity and pH estimation: a comparative study among young and elderly subjects, both normal and with gingivitis and periodontitis. J Indian Soc Periodontol 17:42–46, 10.4103/0972-124X.107473.

Shen B, Li S, Zhang Y, Yuan X, Fan Y, Liu Z, et al. 2009. Determination of total, free and saliva mycophenolic acid with a LC-MS/MS method: application to pharmacokinetic study in healthy volunteers and renal transplant patients. J Pharm Biomed Anal 50(3):515–521, PMID: 19574013, 10.1016/j.jpba.2009.05.030.

Shin HS, Kim JG, Shin YJ, Jee SH. 2002. Sensitive and simple method for the determination of nicotine and cotinine in human urine, plasma and saliva by gas chromatography-mass spectrometry. J Chromatogr B Analyt Technol Biomed Life Sci 769:177–183, PMID: 11936690.

Shirtcliff EA, Granger DA, Schwartz EB, Curran MJ, Booth A, Overman WH. 2000. Assessing estradiol in biobehavioral studies using saliva and blood spots: simple radioimmunoassay protocols, reliability, and comparative validity. Horm Behav 38:137–147, PMID: 10964528, 10.1006/hbeh.2000.1614.

Sturla SJ. 2007. DNA adduct profiles: chemical approaches to addressing the biological impact of DNA damage from small molecules. Curr Opin Chem Biol 11:293–299, PMID: 17574899, 10.1016/j.cbpa.2007.05.021.

Tang WHW, Wang Z, Levison BS, Koeth RA, Britt EB, Fu X, et al. 2013. Intestinal microbial metabolism of phosphatidylcholine and cardiovascular risk. N Engl J Med 368:1575–1584, PMID: 23614584, 10.1056/NEJMoa1109400.

Takayama T, Tsutsui H, Shimizu I, Toyama T, Yoshimoto N, Endo Y, et al. 2016. Diagnostic approach to breast cancer patients based on target metabolomics in saliva by liquid chromatography with tandem mass spectrometry. Clin Chim Acta 452:18–26, PMID: 26523874, 10.1016/j.cca.2015.10.032.

Thieme D. 2012. Potential and limitations of alternative specimens in doping control. Bioanalysis 4:1613–1622, PMID: 22831477, 10.4155/bio.12.150.

Törnqvist M, Fred C, Haglund J, Helleberg H, Paulsson B, Rydberg P. 2002. Protein adducts: quantitative and qualitative aspects of their formation, analysis and applications. J Chromatogr B Analyt Technol Biomed Life Sci 778:279–308, PMID: 12376136.

Vanstraelen K, Maertens J, Augustijns P, Lagrou K, de Loor H, Mols R, et al. 2015. Investigation of saliva as an alternative to plasma monitoring of voriconazole. Clin Pharmacokinet 54(11):1151–1160, PMID: 25910879, 10.1007/s40262-015-0269-z.

Viardot A, Huber P, Puder JJ, Zulewski H, Keller U, Müller B. 2005. Reproducibility of nighttime salivary cortisol and its use in the diagnosis of hypercortisolism compared with urinary free cortisol and overnight dexamethasone suppression test. J Clin Endocrinol Metab 90:5730–5736, 10.1210/jc.2004-2264.

Vuckovic D. 2012. Current trends and challenges in sample preparation for global metabolomics using liquid chromatography-mass spectrometry. Anal Bioanal Chem 403:1523–1548, PMID: 22576654, 10.1007/s00216-012-6039-y.

Wang RE, Tian L, Chang YH. 2012. A homogeneous fluorescent sensor for human serum albumin. J Pharm Biomed Anal 63:165–169, PMID: 22326845, 10.1016/j.jpba.2011.12.035.

Wang VS, Lu MY. 2009. Application of solid-phase microextraction and gas chromatography-mass spectrometry for measuring chemical in saliva of synthetic leather workers. J Chromatogr B Analyt Technol Biomed Life Sci 877:24–32, PMID: 19036646, 10.1016/j.jchromb.2008.11.006.

Wang Z, Klipfell E, Bennett BJ, Koeth R, Levison BS, Dugar B, et al. 2011. Gut flora metabolism of phosphatidylcholine promotes cardiovascular disease. Nature 472:57–63, PMID: 21475195, 10.1038/nature09922.

Wild CP. 2005. Complementing the genome with an “exposome”: the outstanding challenge of environmental exposure measurement in molecular epidemiology. Cancer Epidemiol Biomarkers Prev 14:1847–1850, PMID: 16103423, 10.1158/1055-9965.EPI-05-0456.

Wild CP, Scalbert A, Herceg Z. 2013. Measuring the exposome: a powerful basis for evaluating environmental exposures and cancer risk. Environ Mol Mutagen 54:480–499, PMID: 23681765, 10.1002/em.21777.

Wishart DS, Jewison T, Guo AC, Wilson M, Knox C, Liu Y, et al. 2013. HMDB 3.0—the Human Metabolome Database in 2013. Nucleic Acids Res 41:D801–D807, PMID: 23161693, 10.1093/nar/gks1065.

Xia J, Sinelnikov I, Han B, Wishart DS. 2015. MetaboAnalyst 3.0—making metabolomics more meaningful. Nucl Acids Res 43(W1):W251–W257, , 10.1093/nar/gkv380.

Zeleniuch-Jacquotte A, Afanasyeva Y, Kaaks R, Rinaldi S, Scarmo S, Liu M, et al. 2012. Premenopausal serum androgens and breast cancer risk: a nested case-control study. Breast Cancer Res 14:R32, 10.1186/bcr3117.

Pesticide Use and Age-Related Macular Degeneration in the Agricultural Health Study

Author Affiliations open

1Epidemiology Branch, National Institute of Environmental Health Sciences (NIEHS), National Institutes of Health (NIH), Department of Health and Human Services (DHHS), Research Triangle Park, North Carolina, USA

2Department of Ophthalmology, Duke University Medical School, Durham, North Carolina, USA

3Biostatistics and Computational Biology Branch, NIEHS, NIH, DHHS, Research Triangle Park, North Carolina, USA

4Westat, Durham, North Carolina, USA

5Social & Scientific Systems, Inc., Durham, North Carolina, USA

6Occupational and Environmental Epidemiology Branch, National Cancer Institute, NIH, DHHS, Rockville, Maryland, USA

7Department of Medicine, Duke University Medical School, Durham, North Carolina, USA

PDF icon PDF Version (312 KB)

  • Background:
    Age-related macular degeneration (AMD) is a leading cause of blindness in developed countries. Few studies have investigated its relationship to environmental neurotoxicants. In previous cross-sectional studies, we found an association between pesticide use and self-reported retinal degeneration.
    Objective:
    We evaluated the association of pesticide use with physician-confirmed incident AMD.
    Methods:
    The Agricultural Health Study (AHS) is a prospective cohort of pesticide applicators and their spouses enrolled from 1993–1997 in Iowa and North Carolina. Cohort members reported lifetime use of 50 specific pesticides at enrollment. Self-reports of incident AMD during follow-up through 2007 were confirmed by reports from participants’ physicians and by independent evaluation of retinal photographs provided by the physicians. Confirmed cases ( n=161) were compared with AHS cohort members without AMD ( n=39,108). We estimated odds ratios (ORs) and 95% confidence intervals (CIs) by logistic regression with adjustment for age, gender, and smoking.
    Results:
    AMD was associated with ever use of organochlorine [OR=2.7 (95% CI: 1.8, 4.0)] and organophosphate [OR=2.0 (95% CI: 1.3, 3.0)] insecticides and phenoxyacetate herbicides [OR=1.9 (95% CI: 1.2, 2.8)]. Specific pesticides consistently associated with AMD included chlordane, dichlorodiphenyltrichloroethane (DDT), malathion, and captan; others with notable but slightly less consistent associations were heptachlor, diazinon, phorate, 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), and 2,4-dichlorophenoxyacetic acid (2,4-D). Results were similar for men and women. Some specific pesticides were associated with both early- and late-stage AMD, but others were associated with only one stage.
    Conclusions:
    Exposures to specific pesticides may be modifiable risk factors for AMD. https://doi.org/10.1289/EHP793
  • Received: 12 July 2016
    Revised: 17 March 2017
    Accepted: 18 March 2017
    Published: 19 July 2017

    Address correspondence to F. Kamel, Epidemiology Branch, National Institute of Environmental Health Sciences, 111 TW Alexander Dr., Bldg. 101, Room 327, PO Box 12233, MD A3-05, Research Triangle Park, NC 27709 USA. Telephone: 919-541-1581; Email: kamel@niehs.nih.gov

    Supplemental Material is available online (https://doi.org/10.1289/EHP793).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (782 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (17 KB)

Introduction

Age-related macular degeneration (AMD) is a degenerative condition of the central portion of the retina, the macula (Velez-Montoya et al. 2014). AMD is the leading cause of blindness in older individuals in developed countries, affecting >8 million U.S. residents. The early stage of the disease is often asymptomatic, but late AMD, either geographic atrophy (“dry” AMD) or the neovascular form (“wet” AMD), results in the loss of central, high-acuity vision. Factors affecting risk of early AMD may differ from those affecting progression to late-stage disease (Evans and Lawrenson 2012a; Evans and Lawrenson 2012b).

Both genetic and environmental factors play a role in the etiology of AMD (Sobrin and Seddon 2014). AMD is associated with polymorphisms in approximately 20 genes, most notably complement factor H (CFH) (Sofat et al. 2012) and the age-related maculopathy susceptibility 2/HtrA serine peptidase (ARMS2/HTRA1) locus at chromosome 10q26 (Tong et al. 2010). Smoking is associated with increased risk of AMD, and adiposity may also be important (Chakravarthy et al. 2010). However, these factors do not explain all cases of AMD.

Limited evidence suggests an association of pesticide exposure with retinal dysfunction. Several case series reported signs of macular degeneration in pesticide workers (Dementi 1994; Misra et al. 1985), and experimental studies of rodents have shown biochemical, morphological, and functional changes in the retina after systemic (Imai et al. 1983) or intraocular (Zhang et al. 2006) treatment with pesticides. Nevertheless, few epidemiologic studies have addressed this issue. The Agricultural Health Study (AHS) is a study of licensed pesticide applicators and their spouses who have been followed since enrollment in the mid-1990s. In a cross-sectional analysis of AHS data collected at enrollment, we found that self-reported prevalent retinal or macular degeneration was associated with use of fungicides and organochlorine and organophosphate insecticides in pesticide applicators (Kamel et al. 2000) and with use of fungicides in AHS spouses (Kirrane et al. 2005).

The present study extends these findings. We exploited the prospective design of the AHS to evaluate the association of pesticide use with medically confirmed incident cases of AMD, thus overcoming some limitations of our previous studies.

Methods

Population

The AHS cohort includes 52,394 private pesticide applicators (mostly farmers) and 32,345 of their spouses enrolled between 1993 and 1997 in Iowa and North Carolina. Most applicators were men (97%), most spouses were women (99%), and the race/ethnicity of most cohort members was non-Hispanic white (97%). At enrollment, participants completed self-administered questionnaires that collected information on demographics, lifestyle characteristics, medical history, lifetime pesticide use, and other farming practices. Follow-up telephone interviews were conducted in 1999–2003 and 2005–2010.

To investigate the relationship of pesticide use to AMD incidence, we conducted a case–control study nested within the AHS cohort. We used information from the two follow-up interviews to identify potential incident cases through 1 September 2007. Among 84,739 AHS cohort members, 6 had requested no further contact, 26,002 had not completed either follow-up interview, and 2,554 had died. We also excluded 13,975 persons who were <50 y old on 1 September 2007 because AMD is rare before that age, 324 who had reported retinal or macular degeneration at enrollment, and 15 for other reasons. Thus, 41,863 cohort members were eligible to participate.

Medical histories collected in the follow-up interviews included self-report of physician-diagnosed retinal or macular degeneration. We verified self-reports using information from participants’ eye-care physicians (Figure 1). We screened 552 of 886 participants who reported AMD at either follow-up; of these, 315 affirmed their diagnosis and provided permission for retrieval of medical records. In addition, we screened 257 of 442 AHS cohort members who, although not reporting AMD, did report using an Amsler grid, a self-test of vision loss sometimes recommended for patients with early signs of AMD; of these, 30 affirmed a diagnosis of AMD and provided permission for retrieval of medical records. We contacted one or more physicians for 345 potential AMD cases and obtained diagnostic information for 311 study participants. Physicians either completed a short questionnaire on AMD diagnosis, retinal pathology, and treatment or provided relevant medical records. In addition, some physicians provided retinal photographs for one or both eyes for 101 participants. An optometrist abstracted medical records, and the study ophthalmologist (E.P.) evaluated retinal photographs.

Flowchart.
Figure 1. Recruitment and validation of AMD cases.

Case Definition

Cases were participants for whom the treating physician confirmed the diagnosis of AMD with supporting pathology or for whom the study ophthalmologist diagnosed AMD from retinal photographs. Early AMD was defined by the presence of large, soft, or confluent drusen, with or without pigmentary changes. Individuals with physician reports indicating small, hard drusen as the only sign of AMD were not included as cases unless the retinal photograph provided additional evidence to support the diagnosis. Late AMD was defined by the presence of at least one of the following pathological signs: geographic atrophy, disciform scar, retinal pigment epithelial detachment, and subretinal hemorrhage. AMD unknown stage was assigned when both the physician report and the retinal photograph indicated AMD but disagreed regarding stage. Overall, there was 76% agreement between diagnoses of AMD assigned with or without the photographs; the major difference was the identification of 19 additional early cases using supporting pathology from photographs. We used the more severe diagnosis from either eye to assign a diagnosis to the participant. The final assignments were unclear diagnosis (6), no AMD (133), AMD early stage (58), AMD late stage (79), and AMD unknown stage (35), with a total of 172 AMD cases of any stage.

The controls were AHS cohort members remaining after excluding both the 172 AMD cases and 1,156 individuals identified as possible cases but not confirmed. We also excluded from analysis 10 cases whose physician-reported diagnoses occurred before enrollment in the AHS and 1 case and 1,427 controls with incomplete smoking data. The final sample included 161 incident AMD cases diagnosed between 1994 and 2007 and 39,108 controls.

Institutional review boards of the National Institutes of Health and its contractors approved the study, and all participants signified consent by completing questionnaires; written consent was obtained for release of medical records.

Pesticide Exposure

We used information on ever use of 50 specific pesticides provided by AHS cohort members at enrollment. We also combined information on specific pesticides to create four categories based on function (insecticides, herbicides, fungicides, and fumigants), three chemical classes of insecticides (organochlorines, organophosphates, and carbamates), and two chemical classes of herbicides (phenoxyacetate and triazine herbicides).

Pesticide applicators, but not spouses, provided additional information on the duration (years) and frequency (days per year) of use for 22 individual pesticides. Some applicators (53% of controls and 72% of cases included in this analysis) also completed a second questionnaire, providing information on the frequency and duration of use for 28 additional pesticides. We multiplied frequency and duration to calculate lifetime days of use for each specific pesticide and categorized the results (0, 0.01–10, 10.01–100, and >100 lifetime days), combining categories for some pesticides with sparse data. Because of potential overlap in periods of use, we did not calculate lifetime days of use for pesticide groups.

Data Analysis

We examined the association of AMD with pesticide exposure using multivariable logistic regression to estimate odds ratios (ORs) and 95% confidence intervals (CIs). We included age on 1 September 2007 (50–69, 70–79, ≥80 y), gender, and smoking at enrollment (ever/never) in models because all are important risk factors for AMD and are associated with pesticide use. Adjustment for age using a six-level categorical variable, continuous age, or continuous age plus age squared gave estimates virtually identical to those using the three-level categorical variable. We also evaluated associations in models stratified by age (<70 vs.≥70); these models were further adjusted for age with a continuous variable. Adjusting for smoking status (never, former, or current smoking) gave results similar to those using smoking as a binary variable. We considered body mass index, education, and study site (North Carolina or Iowa) as potential confounders; none of these factors substantially altered effect estimates (<15%), so our final models included only age, gender, and smoking. Because the results were generally similar for men and women when analyzed separately, we present the results for men and women together (adjusted for gender) as our main analysis. In additional analyses, we restricted cases to individuals diagnosed with either early- or late-stage AMD (excluding those of unknown stage) and used multinomial logistic regression to compare each case group with controls. Because sun exposure may increase AMD risk, we evaluated AMD–pesticide associations in models including a variable for hours per day of sun exposure (≥3 vs.≤2).

We evaluated the association of AMD with ever use of 47 pesticides for which at least five cases reported use. We also examined associations with lifetime days of use for 38 pesticides for which users could be categorized into at least two groups with at least five cases in each; these analyses were restricted to men because few women were applicators and because spouses were not asked to provide data on frequency or duration of pesticide use. We evaluated correlations of ever-use variables between pairs of pesticides. Whenever the correlation coefficient for a pair was ≥0.25 and at least one member of the pair was associated with AMD risk, we ran an additional model including both pesticides.

To address concerns regarding possible selection bias, we conducted a quantitative bias analysis (Lash et al 2009) (for details, see Supplemental Material, “Quantitative Bias Analysis”). Briefly, individuals who were not screened were allocated to AMD case or control status based on covariate distribution and pesticide use. We then estimated ORs associated with pesticide exposure among the “complete” set of cases and controls. Separate analyses were performed for each pesticide.

We used SAS (version 9.2; SAS Institute Inc.) and data from AHS data releases REL201004.00, P2REL0506.03, P1REL0506.01, P2REL0506.03, and P3REL0707.01 (https://aghealth.nih.gov/) in our analyses.

Results

We attempted to screen 1,328 potential cases, individuals identified from the AHS cohort because they reported a diagnosis of AMD or use of an Amsler grid (Figure 1). Among those screened, 43% (237/552) of those initially reporting AMD and 88% (227/257) reporting Amsler grid use denied AMD. If all potential cases had been screened and similar proportions had denied AMD, we project that a total of 557 would have affirmed AMD. We enrolled 345 cases (62% of the projected 557), and approximately half of these were confirmed after evaluation of medical records. Comparing potential cases ( n=1,328) with those included in the analysis ( n=161; AMD confirmed, diagnosis after enrollment, smoking data available), we found that the latter were older and more likely to be ever smokers (data not shown); they were also considerably more likely to have used pesticides ≥25 lifetime days (66% and 74%, respectively) and to have ever used organochlorines (50% and 62%, respectively).

AMD risk was positively associated with age and smoking and was slightly elevated among women, those with more than a high school education, and those who consumed alcohol more frequently; AMD was not related to race/ethnicity, state, or body mass index (Table 1). Both early AMD (57 cases) and late AMD (72 cases) were associated with age and smoking; late AMD was also associated with residence in North Carolina and having more than a high school education (see Table S1). Comparing case groups with one another, late AMD cases were slightly older (χ2=3.52, p=0.17) and more likely to be from North Carolina (χ2=5.58, p=0.018) than early cases; other characteristics were similar.

Table 1. Characteristics of incident AMD cases and controls among pesticide applicators and their spouses, AHS 1993-2007.
Case Control
Characteristic n % n % ORa 95% CI
Age at enrollment in AMD study, y
 50–69 36 22 28,801 74 1.0 Reference
 70–79 82 51 8,286 21 7.9 5.3 11.7
 80+ 43 27 2,021 5 17.5 11.2 27.3
Gender
 Men 95 59 22,658 58 1.0 Reference
 Women 66 41 16,450 42 1.3 0.9 1.8
Race/ethnicity
 White, non-Hispanic 158 98 37,984 97 1.0 Reference
 Other 3 2 1,124 3 0.5 0.2 1.6
State
 Iowa 95 59 25,612 65 1.0 Reference
 North Carolina 66 41 13,496 35 0.9 0.7 1.3
Education
 ≤High school 92 59 19,877 54 1.0 Reference
 >High school 64 41 17,046 46 1.3 0.9 1.8
Ever smoker
 No 75 47 23,292 60 1.0 Reference
 Yes 86 53 15,816 40 1.8 1.3 2.5
Alcohol consumption (frequency)
 Never 77 49 16,100 42 1.0 Reference
 <1 to 3 times per mo 41 26 13,471 35 1.0 0.7 1.4
 Once a week or more 38 24 8,530 22 1.3 0.9 2.0
BMI (kg/m2)
 <25 54 34 12,123 32 1.0 Reference
 25–30 78 50 17,139 45 1.0 0.7 1.5
 >30 25 16 87,98 23 0.7 0.5 1.2

Note: AMD, age-related macular degeneration; BMI, body mass index; CI, confidence interval; OR, odds ratio.

aAll models include age, gender, and smoking.

AMD risk was elevated among ever users of insecticides and fungicides as classes but not among ever users of herbicides or fumigants (Table 2). Among chemical classes, AMD was associated with organochlorine and organophosphate insecticides and with phenoxyacetate herbicides. Specific organochlorines significantly associated with AMD were aldrin, chlordane, dichlorodiphenyltrichloroethane (DDT), dieldrin, heptachlor, and lindane; specific organophosphates significantly associated with AMD were diazinon, dichlorvos, malathion, parathion, and phorate; specific phenoxyacetate herbicides significantly associated with AMD were 2,4-dichlorophenoxyacetic acid (2,4-D), 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), and 2-propionic acid (fenoprop; 2,4,5-TP); specific chemicals in other classes significantly associated with AMD were the insecticide permethrin used on crops; the herbicide glyphosate; the fungicides benomyl and captan; and the fumigant ethylene dibromide. Adjustment for sun exposure (≥3 vs.≤2 h/d) had no effect on the AMD–pesticide association.

Table 2. Incident AMD and ever use of specific pesticides in pesticide applicators and their spouses, AHS 1993-2007.
Case Control
n % n % ORa 95% CI
Insecticides (any) 126 78 28558 73 1.6 1.1 2.5
 Organochlorines (any) 98 62 14870 38 2.7 1.8 4.0
  Aldrin 37 26 5433 15 1.5 0.95 2.3
  Chlordane 61 42 7696 21 2.4 1.7 3.6
  DDT 70 47 7728 21 2.1 1.4 3.1
  Dieldrin 21 15 1960 5 1.9 1.1 3.2
  Heptachlor 36 26 4490 12 1.9 1.2 3.0
  Lindane 32 22 5148 14 1.9 1.2 3.0
  Toxaphene 27 19 3957 11 1.5 0.9 2.3
 Organophosphates (any) 117 73 25157 64 2.0 1.3 3.0
  Chlorpyrifos 45 29 10522 27 1.3 0.9 1.9
  Coumaphos 10 7 2302 6 1.1 0.6 2.2
  Diazinon 58 40 9404 26 2.0 1.4 2.9
  Dichlorvos 18 12 3121 9 1.8 1.1 3.0
  Fonofos 20 14 5463 15 1.0 0.6 1.7
  Malathion 103 68 19889 53 2.2 1.5 3.3
  Parathion 28 20 3877 11 1.9 1.2 3.0
  Phorate 42 30 8070 22 1.7 1.1 2.6
  Terbufos 35 24 9210 25 1.1 0.7 1.7
 Other insecticides
  Aldicarb 5 4 2511 7 0.5 0.2 1.3
  Carbaryl 91 61 18890 51 1.4 0.99 2.0
  Carbofuran 31 21 7231 20 1.1 0.7 1.7
  Permethrin (crops) 16 11 3197 9 1.8 1.03 3.0
  Permethrin (animals) 11 8 3541 10 1.3 0.7 2.4
Herbicides (any) 119 74 28973 74 1.2 0.8 1.9
 Phenoxyacetate (any) 101 64 21222 55 1.9 1.2 2.8
  2,4,5-T 46 32 5841 16 2.0 1.3 3.0
  2,4,5-TP 18 13 2390 7 1.7 1.03 2.9
  2,4-D 98 62 20689 54 1.8 1.2 2.7
 Triazine (any) 79 50 19414 50 1.2 0.7 1.8
  Atrazine 74 47 17889 47 1.2 0.8 1.8
  Cyanazine 41 28 10175 28 1.3 0.9 2.0
  Metribuzin 40 28 10756 30 1.2 0.8 1.9
 Other herbicides
  Alachlor 59 39 13154 36 1.4 0.9 2.1
  Butylate 30 21 7886 22 1.2 0.7 1.8
  Chlorimuron ethyl 28 20 7887 22 1.2 0.7 1.9
  Dicamba 44 30 12012 33 1.1 0.7 1.7
  EPTC 17 12 4618 13 1.2 0.7 2.0
  Glyphosate 103 64 23493 61 1.4 0.99 2.0
  Imazethapyr 32 23 9503 26 1.1 0.7 1.8
  Metolachlor 41 28 10728 29 1.2 0.8 1.8
  Paraquat 30 20 5542 15 1.5 0.9 2.3
  Pendimethalin 31 22 9912 27 0.9 0.6 1.4
  Petroleum oil 39 27 11165 31 0.9 0.6 1.4
  Trifluralin 51 36 12775 35 1.2 0.8 1.9
Fungicides (any) 51 32 9337 24 1.5 1.04 2.1
  Benomyl 18 12 2548 7 1.7 0.99 2.8
  Captan 20 14 3009 8 2.0 1.2 3.3
  Chlorothalonil 11 7 1965 5 1.2 0.7 2.3
  Maneb 13 9 2627 7 1.1 0.6 2.0
  Metalaxyl 24 17 5459 15 1.1 0.7 1.8
Fumigants (any) 28 18 6032 16 1.0 0.6 1.5
  Carbon tetrachloride 11 8 1542 4 1.4 0.7 2.6
  Ethylene dibromide 10 7 884 2 2.8 1.5 5.6
  Methyl bromide 17 11 4106 11 0.8 0.5 1.4

Note: AHS, Agricultrual Health Study; AMD, age-related macular degeneration; BMI, body mass index; CI, confidence interval; DDT, dichlorodiphenyltrichloroethane; EPTC, S-ethyl dipropylthiocarbamate; OR, odds ratio. 2,4-D, 2,4-dichlorophenoxyacetic acid; 2,4,5-T, 2,4,5-trichlorophenoxyacetic acid; 2,4,5,-TP, 2-propionic acid (fenoprop).

aAdjusted for age, gender, and smoking.

In gender-stratified analyses, the results were generally similar for men (95 cases) and women (66 cases), although many chemicals were used by too few women to permit analysis (see Table S2). Only organochlorines as a class and the specific chemicals chlordane and malathion were significantly associated with AMD risk in both early and late AMD (see Table S3). Several other exposures—insecticides and organophosphates as classes, DDT, 2,4,5-T, and ethylene dibromide—were associated with both subtypes (OR>1.5) but significantly so only in one subtype. Associations were weaker for late than for early AMD for organochlorines and phenoxyacetate and triazine herbicides as classes and for aldrin, dieldrin, 2,4-D, cyanazine, butylate, and metolachlor (p<0.05). Associations were stronger for late than for early AMD for paraquat, petroleum oil, and benomyl.

In models stratified by age (<70 vs.≥70 y old), ORs for most pesticides were similarly elevated in both age groups, and most age-by-pesticide interactions were unimportant (p>0.8; data not shown). The exception was phenoxyacetate herbicides: as a group, these pesticides were associated with AMD in those ≥70 y old [OR=2.1 (95% CI: 1.0, 4.4)] but not in those <70 y old [OR=1.3 (95% CI: 0.7, 2.7)]; p-interaction=0.24.

Most pesticides were not strongly correlated with one another. We constructed models including each correlated pair (r>0.25), one pair per model (see Table S4). The results suggested that some pesticides (chlordane, DDT, heptachlor, diazinon, malathion, parathion, 2,4,5-T) had strong independent effects: their associations with AMD persisted in models including other pesticides. Others (aldrin, toxaphene, carbaryl, 2,4,5-TP, glyphosate, paraquat, benomyl) did not have independent effects: when modeled with other pesticides, their associations with AMD became weaker and nonsignificant. Some (dieldrin, lindane, phorate, 2,4-D) were intermediate, affected by modeling with some pesticides but not with others. The remainder were not correlated with other pesticides; therefore, their effects were presumed to be independent.

Information on lifetime days of use was available for 38 pesticides for applicators but not spouses (Table 3). Because the data were sparse, we considered a trend to be notable if ptrend<0.10. Such trends were observed for the insecticides chlordane, DDT, lindane, malathion, parathion, and phorate; the herbicides 2,4-D, alachlor, and glyphosate; and the fungicides captan and maneb/mancozeb.

Table 3. Dose–response trends for pesticide use and risk of incident AMD among male pesticide applicators, AHS 1993-2007.
Case Control
Cumulative days of use n % n % ORa 95% CI p-Value for trend
Organochlorine insecticides
 Aldrinb
  0 43 67 8,950 79 1.0 Reference
  >0–10 8 13 1,107 10 1.0 0.5 2.2
  >10–100 8 13 1,060 9 1.0 0.5 2.1
  >100 5 8 211 2 3.0 1.1 7.6 0.233
 Chlordaneb
  0 43 65 8,771 77 1.0 Reference
  >0–10 9 14 1,734 15 0.8 0.4 1.7
  >10 14 21 834 7 2.4 1.3 4.5 0.025
 DDTb
  0 30 45 8,308 73 1.0 Reference
  >0–10 9 14 1,396 12 0.9 0.4 1.8
  >10–100 19 29 1,033 9 2.3 1.3 4.2
  >100 8 12 575 5 1.9 0.8 4.2 0.011
 Heptachlorb
  0 54 81 9,708 85 1.0 Reference
  >0–10 4 6 885 8 0.6 0.2 1.6
  >10 9 13 819 7 1.4 0.7 2.8 0.670
 Lindaneb
  0 49 74 9,535 84 1.0 Reference
  >0–10 5 8 817 7 1.2 0.5 3.0
  >10–100 7 11 683 6 1.9 0.9 4.3
  >100 5 8 290 3 3.5 1.4 9.0 0.005
 Toxapheneb
  0 54 81 9,876 87 1.0 Reference
  >0–10 6 9 786 7 1.1 0.5 2.6
  >10 7 10 750 7 1.3 0.6 2.8 0.527
Organophosphate insecticides
 Chlorpyrifos
  0 52 57 12,679 57 1.0 Reference
  >0–10 15 16 3,137 14 1.3 0.7 2.3
  >10–100 15 16 4,287 19 1.1 0.6 1.9
  >100 10 11 2,081 9 1.7 0.8 3.3 0.224
 Diazinonb
  0 46 71 8,782 78 1.0 Reference
  >0–10 7 11 1,196 11 1.1 0.5 2.5
  >10 12 18 1,326 12 1.6 0.8 3.0 0.174
 Dichlorvos
  0 70 86 17,938 88 1.0 Reference
  >0–10 5 6 736 4 1.9 0.8 4.8
  >10 6 7 1,779 9 1.1 0.5 2.5 0.558
 Fonofos
  0 63 77 15,648 76 1.0 Reference
  >0–10 4 5 1,500 7 0.8 0.3 2.2
  >10 15 18 3,451 17 1.2 0.7 2.2 0.522
 Malathionb
  0 21 32 3,720 33 1.0 Reference
  >0–10 12 18 2,961 26 0.8 0.4 1.5
  >10–100 15 23 3,247 29 0.9 0.4 1.7
  >100 17 26 1,352 12 2.0 1.1 3.9 0.093
 Parathionb
  0 52 81 10,353 92 1.0 Reference
  >0–10 8 13 403 4 3.3 1.6 7.1
  >10 4 6 548 5 1.3 0.5 3.8 0.087
 Phorateb
  0 39 61 7,570 67 1.0 Reference
  >0–10 6 9 1,552 14 0.8 0.3 1.9
  >10–100 9 14 1,664 15 1.0 0.5 2.2
  >100 10 16 557 5 3.5 1.7 7.2 0.020
 Terbufos
  0 49 61 12,154 59 1.0 Reference
  >0–10 4 5 2,047 10 0.6 0.2 1.6
  >10 27 34 6,371 31 1.3 0.8 2.0 0.394
Other insecticides
 Carbofuran
  0 54 67 13,797 68 1.0 Reference
  >0–10 8 10 2,614 13 0.7 0.3 1.5
  >10–100 13 16 2,857 14 1.1 0.6 2.1
  >100 6 7 1,121 5 1.5 0.6 3.5 0.463
 Carbarylb
  0 27 40 6,178 55 1.0 Reference
  >0–10 15 22 2,042 18 1.5 0.8 2.9
  >10–100 11 16 1,728 15 1.1 0.5 2.3
  >100 15 22 1,355 12 1.9 1.0 3.6 0.101
 Permethrin (crops)
  0 67 84 17,630 87 1.0 Reference
  >0–10 8 10 1,394 7 2.1 0.99 4.4
  >10 5 6 1,287 6 1.4 0.6 3.5 0.150
Herbicides
 2,4-D
  0 13 14 4,580 21 1.0 Reference
  >0–10 10 11 2,502 11 1.4 0.6 3.2
  >10–100 26 28 6,877 31 1.4 0.7 2.8
  >100 44 47 8,116 37 2.2 1.2 4.1 0.011
 2,4,5,Tb
  0 44 69 8,833 78 1.0 Reference
  >0–10 10 16 1,430 13 1.0 0.5 2.0
  >10 10 16 1,116 10 1.2 0.6 2.4 0.635
 Alachlor
  0 30 36 8,622 42 1.0 Reference
  >0–10 12 14 2,633 13 1.4 0.7 2.8
  >10–100 19 23 5,025 25 1.2 0.7 2.2
  >100 22 27 4,176 20 1.9 1.1 3.3 0.046
 Atrazine
  0 24 26 5,566 25 1.0 Reference
  >0–10 8 9 2,686 12 0.8 0.4 1.9
  >10–100 28 30 6,561 30 1.2 0.7 2.1
  >100 34 36 7,387 33 1.5 0.9 2.5 0.111
 Butylateb
  0 49 74 7,968 70 1.0 Reference
  >0–10 5 8 1,177 10 0.8 0.3 2.0
  >10 12 18 2,246 20 1.2 0.6 2.3 0.668
 Chlorimuron ethylb
  0 47 71 7,907 69 1.0 Reference
  >0–10 12 18 2,274 20 1.2 0.6 2.2
  >10 7 11 1,240 11 1.2 0.5 2.6 0.633
 Cyanazine
  0 41 52 11,217 55 1.0 Reference
  >0–10 11 14 2,816 14 1.3 0.6 2.5
  >10–100 17 22 3,986 19 1.4 0.8 2.5
  >100 10 13 2,556 12 1.5 0.8 3.1 0.141
 Dicamba
  0 39 49 9,454 46 1.0 Reference
  >0–10 10 13 3,089 15 1.0 0.5 2.0
  >10–100 16 20 4,934 24 1.1 0.6 1.9
  >100 15 19 2,887 14 1.9 1.03 3.5 0.112
 EPTC
  0 65 82 16,124 79 1.0 Reference
  >0–10 7 9 1,902 9 1.1 0.5 2.5
  >10 7 9 2,284 11 1.0 0.5 2.2 0.917
 Glyphosate
  0 15 16 5,104 23 1.0 Reference
  >0–10 18 19 4,929 22 1.3 0.6 2.5
  >10–100 33 35 7,403 33 1.7 0.9 3.1
  >100 28 30 4,783 22 2.6 1.4 4.9 0.002
 Imazethapyr
  0 45 59 11,670 57 1.0 Reference
  >0–10 13 17 3,957 19 1.1 0.6 2.1
  >10 18 24 4,748 23 1.5 0.9 2.6 0.167
 Metolachlor
  0 41 53 10,720 52 1.0 Reference
  >0–10 10 13 2,435 12 1.2 0.6 2.5
  >10–100 13 17 4,275 21 1.0 0.5 1.8
  >100 13 17 3,078 15 1.5 0.8 2.9 0.325
 Metribuzinb
  0 40 60 6,746 59 1.0 Reference
  >0–10 13 19 2,253 20 1.2 0.6 2.3
  >10 14 21 2,389 21 1.4 0.7 2.6 0.282
 Paraquatb
  0 53 80 9,560 84 1.0 Reference
  >0–10 7 11 1,032 9 1.2 0.5 2.6
  >10 6 9 829 7 1.4 0.6 3.2 0.413
 Pendimethalinb
  0 42 66 7,250 64 1.0 Reference
  >0–10 14 22 1,924 17 1.4 0.8 2.6
  >10 8 13 2,239 20 0.8 0.4 1.8 0.980
 Petrolium oilb
  0 48 79 8,894 78 1.0 Reference
  >0–100 7 11 1,771 16 0.8 0.4 1.8
  >100 6 10 679 6 1.9 0.8 4.5 0.370
 Trifluralin
  0 35 45 9,185 45 1.0 Reference
  >0–10 9 12 1,813 9 1.3 0.6 2.7
  >10–100 15 19 4,862 24 0.9 0.5 1.7
  >100 19 24 4,657 23 1.4 0.8 2.5 0.386
Fungicides
 Captan
  0 66 81 17,971 89 1.0 Reference
  >0–10 7 9 1,436 7 1.8 0.8 3.9
  >10 8 10 677 3 2.9 1.4 6.2 0.002
 Manebb
  0 56 84 10,447 92 1.0 Reference
  >0–100 7 10 694 6 1.7 0.8 3.7
  2)100+ 4 6 239 2 2.5 0.9 6.9 0.039
 Metalaxylb
  0 54 82 9,216 81 1.0 Reference
  >0–10 5 8 723 6 1.2 0.5 3.0
  >10 7 11 1,394 12 0.9 0.4 1.9 0.822
Fumigants
 Methyl bromide
  0 77 82 18,556 84 1.0 Reference
  >0–10 6 6 923 4 1.2 0.5 2.9
  >10 11 12 2,728 12 0.8 0.4 1.5 0.582

Note: AHS, Agricultrual Health Study; AMD, age-related macular degeneration; BMI, body mass index; CI, confidence interval; DDT, dichlorodiphenyltrichloroethane; EPTC, S-ethyl dipropylthiocarbamate; OR, odds ratio. 2,4-D, 2,4-dichlorophenoxyacetic acid; 2,4,5-T, 2,4,5-trichlorophenoxyacetic acid; 2,4,5,-TP, 2-propionic acid (fenoprop).

aAdjusted for age and smoking.

bData available only for subset of applicators who completed the take-home questionnaire.

Table 4 summarizes results of the various analyses. We defined a consistent association of a pesticide with AMD as one that was evident in all five of the following: a) the ever use analysis (Table 2); b) the analysis with adjustment for correlated pesticides (see Table S4); c) the lifetime use analysis (Table 3); d) either men or women (see Table S2); and e) either early or late AMD (see Table S3). Four pesticides met these criteria fully: chlordane, DDT, malathion, and captan. Five additional pesticides met most of the criteria for consistency: phorate and 2,4-D were each significantly associated with AMD in one of two analyses with correlated pesticides, and each had evidence of dose–response; heptachlor, diazinon, and 2,4,5-T were each associated with AMD in analyses of correlated pesticides but did not have evidence of dose–response, possibly because small numbers of individuals were exposed.

Table 4. Summary of analyses of AMD and pesticide use.
Ever Use By Gender By AMD stage
Ever use Adjusted for correlated pesticides Cumulative use Men Women Early Late
(Table 2) (Table S4) (Table 3) (Table S2) (Table S3)
Insecticides (any) + ND ND + + +
 Organochlorines (any) + ND ND + + + +
  Aldrin + + +
  Chlordane + + + + + + +
  DDT + + + + + +
  Dieldrin + + +
  Heptachlor + + + +
  Lixane + + + +
  Toxaphene +
 Organophosphates (any) + ND ND + + +
  Chlorpyrifos
  Coumaphos
  Diazinon + + + + +
  Dichlorvos + + +
  Fonofos +
  Malathion + + + + + + +
  Parathion + +
  Phorate + + + +
  Terbufos
 Other insecticides
  Aldicarb
  Carbaryl + +
  Carbofuran
  Permethrin (crops) + +
  Permethrin (animals) +
Herbicides (any) ND ND
 Phenoxyacetate (any) + ND ND +
  2,4,5-T + + + +
  2,4,5-TP + +
  2,4-D + + + + +
 Other herbicides ND ND
  Alachlor +
  Atrazine +
  Butylate
  Chlorimuron ethyl
  Cyanazine
  Dicamba
  EPTC
  Glyphosate + + +
  Imazethapyr
  Metolachlor +
  Metribuzin +
  Paraquat +
  Peximethalin
  Petroleum oil
  Trifluralin
Fungicides (any) + ND ND +
  Benomyl + +
  Captan + + + + +
  Chlorothalonil
  Maneb +
  Metalaxyl
Fumigants (any) ND ND
  Carbon tetrachloride
  Ethylene dibromide + + +
  Methyl bromide

Note: AMD, age-related macular degeneration; DDT, dichlorodiphenyltrichloroethane; EPTC, S-ethyl dipropylthiocarbamate; ND, not done. Blank cells indicate that no association was present. ND indicates that analyses of correlated pesticides and cumulative use could not be tested for grouped pesticides. +, AMD was associated with the pesticide in the indicated analysis; 2,4-D, 2,4-dichlorophenoxyacetic acid ; 2,4,5-T, 2,4,5-trichlorophenoxyacetic acid; 2,4,5,-TP, 2-propionic acid (fenoprop).

We performed a quantitative bias analysis in which individuals who were not screened were allocated to AMD case or control status based on covariate distribution and pesticide use (see Table S5). The ORs were somewhat attenuated but remained elevated, and our interpretation was not qualitatively affected.

Discussion

To our knowledge, this is the first epidemiologic study to examine the relationship between specific pesticides and physician-confirmed AMD. We found associations of incident AMD with specific pesticides in several functional and chemical groups. The results were strongest and most consistent for organochlorine and organophosphate insecticides and phenoxyacetate herbicides together with specific pesticides from these classes and the fungicide captan (Table 4). Results were qualitatively similar for men and women, but some differences were apparent between early and late AMD. The present findings are consistent with results from two previous cross-sectional analyses in the AHS, which evaluated cases prevalent at enrollment (not included in the present study). These earlier studies implicated organochlorine and organophosphate insecticides and fungicides as risk factors for AMD (Kamel et al. 2000; Kirrane et al. 2005).

The etiology of AMD likely involves both genetic susceptibility and environmental exposures. AMD has been associated with polymorphisms in approximately 20 genes (Fritsche et al. 2014; Sobrin and Seddon 2014) including CFH (Sofat et al. 2012), other genes in complement pathways (Schramm et al. 2014), and genes involved in inflammation and immune regulation, lipid metabolism and transport, maintenance of the extracellular matrix, and angiogenesis (Fritsche et al. 2014; Sobrin and Seddon 2014). Smoking is positively associated with AMD (Chakravarthy et al. 2010; Sobrin and Seddon 2014), and certain dietary factors, including vitamins, minerals, and omega-3 fatty acids, are inversely associated with AMD (Sobrin and Seddon 2014; Zampatti et al. 2014). A meta-analysis of 24 studies found consistent associations with adiposity, hypertension, and cardiovascular disease (Chakravarthy et al. 2010). Genetic variation may modify associations of environmental factors such as smoking with AMD (Seddon et al. 2006; Wang et al. 2008).

Two fundamental mechanisms critical to AMD pathogenesis are inflammation and oxidative stress. The importance of the former is supported by the genetic evidence cited above and by associations of AMD with changes in inflammation biomarkers (Hong et al. 2011). High levels of oxidative stress are normally present in the retina and may be further increased by aging or environmental factors such as smoking (Handa 2012). Biomarkers of oxidative stress are elevated in patients with AMD (Zafrilla et al. 2013), and dietary antioxidants may retard AMD progression (Evans and Lawrenson 2012b; Zampatti et al. 2014). Oxidative stress may provoke the innate immune system and further increase inflammation, perhaps particularly in the presence of genetic variation in complement factors (Handa 2012).

Many specific pesticides associated with AMD are polychlorinated cyclic hydrocarbons. Of 47 specific pesticides evaluated in the ever-use analysis, 11 of 14 (79%) polychlorinated cyclic hydrocarbons were associated with AMD compared with 10 of 33 (30%) pesticides having other structures (χ2=9.3, p=0.002). Many polychlorinated cyclic hydrocarbons are persistent, perhaps because they are lipophilic, which might account both for the greater toxicity of these chemicals and for the inconsistent association of adiposity with AMD: perhaps the latter can be detected primarily in populations exposed to persistent lipophilic toxicants.

Polychlorinated cyclic hydrocarbons may activate mechanisms involved in AMD. Organochlorine insecticides and polychlorinated biphenyls (PCBs) increase both oxidative stress (Bagchi et al. 1995; Lee and Opanashuk 2004) and inflammation (Kim KS et al. 2012; Hayley et al. 2011). Pesticides from other chemical classes, including organophosphate and pyrethroid insecticides and the bipyridyl herbicide paraquat, increase oxidative stress in the retina (Cingolani et al. 2006; Rotstein et al. 2003; Yu et al. 2008). Further, many of the polycyclic pesticides associated with AMD are aromatic, and polycyclic aromatic hydrocarbons are prone to phototoxic reactions (Wielgus and Roberts 2012). These persistent lipophilic compounds could accumulate in the retina and increase oxidative stress in response to light exposure. Potentially of great importance to AMD are observations that complement pathways can be activated by some pesticides, including DDT (Dutta et al. 2008), other organochlorine insecticides (Kumar et al. 2014), malathion (Ayub et al. 2001), the pyrethroid fenvalerate (Dutta and Das 2011), and paraquat (Kim YS et al. 2012).

Risk of late AMD is greater in women than in men (Rudnicka et al. 2015), and risk factors may differ between the genders (Adams et al. 2011; Erke et al. 2014; Klein et al. 1998). We found, however, that the pesticide–AMD association was qualitatively similar in men and women, although fewer women used pesticides. Risk factors for early and late AMD may differ. For example, vitamin supplements do not appear to reduce the risk of early AMD but may delay progression to the late forms (Evans and Lawrenson 2012a; Evans and Lawrenson, 2012b). Similarly, smoking has a stronger association with late than with early AMD (McKay et al. 2011), as does adiposity (Adams et al. 2011; Klein et al. 2007). Differences in pesticide–AMD associations for early and late AMD might be attributable to chance or to confounding: for example, residual confounding by age-related factors. One would, however, expect the latter to produce greater discrepancies for pesticides with greater secular trends, such as organochlorine insecticides (which were banned in the United States beginning in the 1970s), but organophosphate insecticides and herbicides accounted for more of the differences between early and late AMD. Further, neither adjustment for age using finer-grained variables nor stratification by age affected the pesticide–AMD associations.

Because pesticides from several classes were associated with AMD, general farming-related exposures might confound pesticide–AMD associations. A possible confounder is sun exposure, which is a risk factor for AMD that is potentially related to pesticide use (Sui et al 2013). However, adjustment for hours per day of sun exposure did not alter pesticide–AMD associations.

That many potential cases were not included in the final analysis raises a concern of selection bias. Some nonparticipation may have resulted from individuals misreporting AMD on the original questionnaires because they misunderstood the question or for other reasons; this could explain why 43% of those initially reporting AMD subsequently denied the diagnosis on screening. However, we expected that most Amsler grid users would deny the diagnosis. We enrolled 62% of potential cases, but only approximately half of these were confirmed by evaluation of medical records. Overall, these results suggest primarily that self-reports are unreliable and underscore the importance of our effort to validate cases. Quantitative bias analysis (Lash et al 2009) indicated that the results were qualitatively similar after accounting for the possible case–control status of unscreened individuals, although the ORs were slightly weaker. Thus, our findings do not appear to be the result of selection bias.

A further concern is that 25% of individuals otherwise eligible for the study did not complete either follow-up interview and so could not be considered for screening. In a previous study, we found that reporting either a health condition or greater pesticide use at enrollment in the AHS was associated with slightly greater odds of participation at follow-up (Montgomery et al 2010). Thus, we are unlikely to have lost exposed cases from the study population, which might have biased our estimates to the null. Further, our previous study showed that associations of pesticide use with health conditions either in the full cohort or in those participating in the follow-up gave very similar results (Montgomery et al 2010), and in another study, we found that loss to follow-up created notable bias only in analyses of strongly related exposures and outcomes (e.g., smoking and lung cancer) and was not a general problem (Rinsky et al. 2016).

Differences in age and smoking between potential AMD cases and those included in the analysis are not surprising because these factors are known to be related to AMD. Differences in pesticide use are a greater concern; however, they undoubtedly reflect the greater age of AMD cases. Further, it is difficult to understand why greater exposure to pesticides would select for substantially greater participation. Although our previous study found a small association of ever use of pesticides with greater participation at follow-up, increasing frequency of use was associated with nonparticipation, and neither duration of use nor ever use of insecticides was related to participation, either positively or inversely (Montgomery et al 2010). Thus, greater use of pesticides among confirmed AMD cases may represent a real difference and not selection bias.

Based on the age structure and race/ethnicity of the AHS cohort, we expected to find ∼84 incident late AMD cases compared with the 72 we identified (Rudnicka et al. 2015). There are few estimates of incidence for early AMD, but we are likely to have missed more of these cases. Early cases may be less aware of their disease or less likely to report it and so would not have been contacted for screening. Our inability to identify asymptomatic early AMD cases is unlikely to have produced spurious associations unless underascertainment of early cases was related to pesticide use: an unlikely event. Further, pesticide associations with early AMD were often stronger than those with late AMD, suggesting that underascertainment of early cases may have resulted in underestimation of pesticide–AMD associations.

Strengths of the present study include its prospective design and its uniquely detailed information on the use of specific pesticides. Reporting of pesticide use by AHS applicators is generally reliable (Blair et al. 2002), and remaining misclassification would likely be nondifferential and bias associations toward the null (Blair et al. 2011). Although virtually all applicators and approximately half of spouses had used some pesticides, the use of specific chemicals varied considerably, enabling us to compare exposed and unexposed individuals within the cohort. Information was available to control for confounding by many risk factors for AMD, and the use of cases and controls from the same cohort would tend to further minimize confounding. The study was sufficiently large to evaluate exposure–response trends for many pesticides and to compare findings for men and women and for early and late AMD.

Conclusion

We found associations of AMD with use of organochlorine and organophosphate insecticides and phenoxyacetate herbicides as classes as well as with individual pesticides. Specifically, there were consistent associations with chlordane, DDT, malathion, and captan. Additional pesticides with slightly less consistent but nevertheless notable associations were heptachlor, diazinon, phorate, 2,4,5-T and 2,4-D. Overall, these results are consistent with experimental studies of mechanisms underlying AMD, including oxidative stress, inflammation, and complement activation. Our study involved a relatively small number of cases, and its novel results require replication. Nevertheless, it suggests that use of specific pesticides may be a modifiable risk factor for AMD.

Acknowledgments

B. Wujciak evaluated medical records provided by physicians. This work was supported in part by the Intramural Research Program of the National Institutes of Health/National Institute of Environmental Health Sciences (Z01-ES049030) and the National Cancer Institute (Z01-CP-1-119).

References

Adams MK, Simpson JA, Aung KZ, Makeyeva GA, Giles GG, English DR, et al. 2011. Abdominal obesity and age-related macular degeneration. Am J Epidemiol 173(11):1246–1255, PMID: 21422060, 10.1093/aje/kwr005.

Ayub S, Deb K, Sharma M, Das N. 2001. Activation of the human serum complement cascade by insecticides. Curr Sci 80(12):1592–1595.

Bagchi D, Bagchi M, Hassoun EA, Stohs SJ. 1995. In vitro and in vivo generation of reactive oxygen species, DNA damage and lactate dehydrogenase leakage by selected pesticides. Toxicology 104(1–3):129–140, PMID: 8560491.

Blair A, Tarone R, Sandler D, Lynch CF, Rowland A, Wintersteen W, et al. 2002. Reliability of reporting on life-style and agricultural factors by a sample of participants in the Agricultural Health Study from Iowa. Epidemiology 13(1):94–99, PMID: 11805592.

Blair A, Thomas K, Coble J, Sandler DP, Hines CJ, Lynch CF, et al. 2011. Impact of pesticide exposure misclassification on estimates of relative risks in the Agricultural Health Study. Occup Environ Med 68(7):537–541, PMID: 21257983, 10.1136/oem.2010.059469.

Chakravarthy U, Wong TY, Fletcher A, Piault E, Evans C, Zlateva G, et al. 2010. Clinical risk factors for age-related macular degeneration: A systematic review and meta-analysis. BMC Ophthalmol 10:31, PMID: 21144031, 10.1186/1471-2415-10-31.

Cingolani C, Rogers B, Lu L, Kachi S, Shen J, Campochiaro PA. 2006. Retinal degeneration from oxidative damage. Free Radic Biol Med 40(4):660–669, PMID: 16458197, 10.1016/j.freeradbiomed.2005.09.032.

Dementi B. 1994. Ocular effects of organophosphates – a historical perspective of Saku disease. J Appl Toxicol 14(2):119–129, PMID: 8027507.

Dutta R, Das N. 2011. Immunomodulation of serum complement (C3) and macrophages by synthetic pyrethroid fenvalerate: In vitro study. Toxicology 285(3):126–132, PMID: 21557984, 10.1016/j.tox.2011.04.012.

Dutta R, Mondal AM, Arora V, Nag TC, Das N. 2008. Immunomodulatory effect of DDT bis[4-chlorophenyl]-1,1,1-trichloroethane on complement system and macrophages. Toxicology 252(1–3):78–85, PMID: 18755234, 10.1016/j.tox.2008.07.063.

Erke MG, Bertelsen G, Peto T, Sjolie AK, Lindekleiv H, Njolstad I. 2014. Cardiovascular risk factors associated with age-related macular degeneration: the Tromsø Study. Acta Ophthalmol 92(7):662–669, PMID: 24460653, 10.1111/aos.12346.

Evans JR, Lawrenson JG 2012a. Antioxidant vitamin and mineral supplements for preventing age-related macular degeneration. Cochrane Database Syst Rev 6: CD000253.

Evans JR, Lawrenson JG 2012b. Antioxidant vitamin and mineral supplements for slowing the progression of age-related macular degeneration. Cochrane Database Syst Rev 11:CD000254.

Fritsche LG, Fariss RN, Stambolian D, Abecasis GR, Curcio CA, Swaroop A. 2014. A
ge-related macular degeneration: Genetics and biology coming together. Annu Rev Genomics Hum Genet 15:151–171, PMID: 24773320, 10.1146/annurev-genom-090413-025610.

Handa JT. 2012. How does the macula protect itself from oxidative stress?. Mol Aspects Med 33(4):418–435, PMID: 22503691, 10.1016/j.mam.2012.03.006.

Hayley S, Mangano E, Crowe G, Li N, Bowers WJ. 2011. An in vivo animal study assessing long-term changes in hypothalamic cytokines following perinatal exposure to a chemical mixture based on Arctic maternal body burden. Environ Health 10:65, PMID: 21745392, 10.1186/1476-069X-10-65.

Hong T, Tan AG, Mitchell P, Wang JJ. 2011. A review and meta-analysis of the association between C-reactive protein and age-related macular degeneration. Surv Ophthalmol 56(3):184–194, PMID: 21420705, 10.1016/j.survophthal.2010.08.007.

Imai H, Miyata M, Uga S, Ishikawa S. 1983. Retinal degeneration in rats exposed to an organophosphate pesticide fenthion. Environ Res 30(2):453–465, PMID: 6832126, 10.1016/0013-9351(83)90231-1.

Kamel F, Boyes WK, Gladen BC, Rowland AS, Alavanja MC, Blair A, et al. 2000. Retinal degeneration in licensed pesticide applicators. Am J Ind Med 37(6):618–628, PMID: 10797505.

Kim KS, Hong NS, Jacobs DR, Jr, Lee DH. 2012. Interaction between persistent organic pollutants and C-reactive protein in estimating insulin resistance among non-diabetic adults. J Prev Med Public Health 45(2):62–69, 10.3961/jpmph.2012.45.2.62.

Kim YS, Jung H, Gil HW, Hong SY, Song HY. 2012. Proteomic analysis of changes in protein expression in serum from animals exposed to paraquat. Int J Mol Med 30(6):1521–1527, PMID: 23023206, 10.3892/ijmm.2012.1143.

Kirrane EF, Hoppin JA, Kamel F, Umbach DM, Boyes WK, Deroos AJ, Alavanja M, Sandler DP. 2005. Retinal degeneration and other eye disorders in wives of farmer pesticide applicators enrolled in the Agricultural Health Study. Am J Epidemiol 161(11):1020–1029, PMID: 15901622, 10.1093/aje/kwi140.

Klein R, Deng Y, Klein BEK, Hyman L, Seddon J, Frank RN, et al. 2007. Cardiovascular disease, its risk factors and treatment, and age-related macular degeneration: women’s Health Initiative Sight Exam ancillary study. Am J Ophthalmol 143(3):473–483, PMID: 17317391, 10.1016/j.ajo.2006.11.058.

Klein R, Klein BE, Moss SE. 1998. Relation of smoking to the incidence of age-related maculopathy. The Beaver Dam Eye Study. Am J Epidemiol 147(2):103–110, PMID: 9456998.

Kumar J, Lind PM, Salihovic S, van Bavel B, Ekdahl KN, Nilsson B, et al. 2014. Influence of persistent organic pollutants on the complement system in a population-based human sample. Environ Int 71:94–100, PMID: 24996157, 10.1016/j.envint.2014.06.009.

Lash TL, Fox M, Fink AK. 2009. Applying Quantitative Bias Analysis to Epidemiologic Data. New York, NY:Springer, 43–58.

Lee DW, Opanashuk LA. 2004. Polychlorinated biphenyl mixture aroclor 1254-induced oxidative stress plays a role in dopaminergic cell injury. Neurotoxicology 25(6):925–939, PMID: 15474611, 10.1016/j.neuro.2004.05.005.

McKay GJ, Patterson CC, Chakravarthy U, Dasari S, Klaver CC, Vingerling JR, et al. 2011. Evidence of association of APOE with age-related macular degeneration: A pooled analysis of 15 studies. Hum Mutat 32(12):1407–1416, PMID: 21882290, 10.1002/humu.21577.

Misra UK, Nag D, Misra NK, Mehr MK, Ray PK. 1985. Some observations on the macula of pesticide workers. Hum Toxicol 4(2):135–145, PMID: 4007878.

Montgomery MP, Kamel F, Hoppin JA, Beane Freeman LE, Alavanja MC, Sandler DP. 2010. Effects of self-reported health conditions and pesticide exposures on probability of follow-up in a prospective cohort study. Am J Ind Med 53(5):486–496, PMID: 20017198, 10.1002/ajim.20789.

Rinsky JL, Richardson DB, Wing S, Beard JD, Alavanja M, Beane Freeman LE, et al. 2016. Assessing the potential for bias from non-response to a study follow-up interview: An example from the Agricultural Health Study. Am J Epidemiol, PMID: 28486574, 10.1093/aje/kwx098.

Rotstein NP, Politi LE, German OL, Girotti R. 2003. Protective effect of docosahexaenoic acid on oxidative stress-induced apoptosis of retina photoreceptors. Invest Ophthalmol Vis Sci 44(5):2252–2259, PMID: 12714668.

Rudnicka AR, Kapetanakis VV, Jarrar Z, Wathern AK, Wormald R, Fletcher AE, et al. 2015. Incidence of late-stage age-related macular degeneration in American whites: systematic review and meta-analysis. Am J Ophthalmol 160(1):85–93, PMID: 25857680, 10.1016/j.ajo.2015.04.003.

Schramm EC, Clark SJ, Triebwasser MP, Raychaudhuri S, Seddon JM, Atkinson JP. 2014. Genetic variants in the complement system predisposing to age-related macular degeneration: a review. Mol Immunol 61(2):118–125, PMID: 25034031, 10.1016/j.molimm.2014.06.032.

Seddon JM, George S, Rosner B, Klein ML. 2006. CFH gene variant, Y402H, and smoking, body mass index, environmental associations with advanced age-related macular degeneration. Hum Hered 61(3):157–165, PMID: 16816528, 10.1159/000094141.

Sobrin L, Seddon JM. 2014. Nature and nurture- genes and environment- predict onset and progression of macular degeneration. Prog Retin Eye Res 40:1–15, PMID: 24374240, 10.1016/j.preteyeres.2013.12.004.

Sofat R, Casas JP, Webster AR, Bird AC, Mann SS, Yates JR, et al. 2012. Complement factor H genetic variant and age-related macular degeneration: effect size, modifiers and relationship to disease subtype. Int J Epidemiol 41(1):250–262, 10.1093/ije/dyr204.

Sui GY, Liu GC, Liu GY, Gao YY, Deng Y, Wang WY, et al. 2013. Is sunlight exposure a risk factor for age-related macular degeneration? A systematic review and meta-analysis. Br J Ophthalmol 97(4):389–394, PMID: 23143904, 10.1136/bjophthalmol-2012-302281.

Tong Y, Liao J, Zhang Y, Zhou J, Zhang H, Mao M. 2010. LOC387715/HTRA1 gene polymorphisms and susceptibility to age-related macular degeneration: A HuGE review, meta-analysis. Mol Vis 16:1958–1981, PMID: 21031019.

Velez-Montoya R, Oliver SC, Olson JL, Fine SL, Quiroz-Mercado H, Mandava N. 2014. Current knowledge and trends in age-related macular degeneration: genetics, epidemiology, and prevention. Retina 34(3):423–441, PMID: 24285245, 10.1097/IAE.0000000000000036.

Wang JJ, Ross RJ, Tuo J, Burlutsky G, Tan AG, Chan CC, et al. 2008. The LOC387715 polymorphism, inflammatory markers, smoking, and age-related macular degeneration. A population-based case-control study. Ophthalmology 115(4):693–699, PMID: 17675241, 10.1016/j.ophtha.2007.05.038.

Wielgus AR, Roberts JE. 2012. Retinal photodamage by endogenous and xenobiotic agents. Photochem Photobiol 88(6):1320–1345, PMID: 22582903, 10.1111/j.1751-1097.2012.01174.x.

Yu F, Wang Z, Ju B, Wang Y, Wang J, Bai D. 2008. Apoptotic effect of organophosphorus insecticide chlorpyrifos on mouse retina in vivo via oxidative stress and protection of combination of vitamins C and E. Exp Toxicol Pathol 59(6):415–423, 10.1016/j.etp.2007.11.007.

Zafrilla P, Losada M, Perez A, Caravaca G, Mulero J. 2013. Biomarkers of oxidative stress in patients with wet age related macular degeneration. J Nutr Health Aging 17(3):219–222, PMID: 23459973, 10.1007/s12603-012-0095-z.

Zampatti S, Ricci F, Cusumano A, Marsella LT, Novelli G, Giardina E. 2014. Review of nutrient actions on age-related macular degeneration. Nutr Res 34(2):95–105, PMID: 24461310, 10.1016/j.nutres.2013.10.011.

Zhang X, Jones D, Gonzalez-Lima F. 2006. Neurodegeneration produced by rotenone in the mouse retina: A potential model to investigate environmental pesticide contributions to neurodegenerative diseases. J Toxicol Environ Health A 69(18):1681–1697, PMID: 16864419, 10.1080/15287390600630203.

High-Throughput Analysis of Ovarian Cycle Disruption by Mixtures of Aromatase Inhibitors

Author Affiliations open

1Models for Ecotoxicology and Toxicology Unit (DRC/VIVA/METO), Institut National de l’Environnement Industriel et des Risques (INERIS), Verneuil en Halatte, France

2S-IN Soluzioni Informatiche Srl (S-IN), Vicenza, Italy

3School of Information Sciences, University of Illinois at Urbana-Champaign, Champaign, Illinois, USA

PDF icon PDF Version (2 MB)

  • Background:
    Combining computational toxicology with ExpoCast exposure estimates and ToxCast™ assay data gives us access to predictions of human health risks stemming from exposures to chemical mixtures.
    Objectives:
    We explored, through mathematical modeling and simulations, the size of potential effects of random mixtures of aromatase inhibitors on the dynamics of women’s menstrual cycles.
    Methods:
    We simulated random exposures to millions of potential mixtures of 86 aromatase inhibitors. A pharmacokinetic model of intake and disposition of the chemicals predicted their internal concentration as a function of time (up to 2 y). A ToxCast™ aromatase assay provided concentration–inhibition relationships for each chemical. The resulting total aromatase inhibition was input to a mathematical model of the hormonal hypothalamus–pituitary–ovarian control of ovulation in women.
    Results:
    Above 10% inhibition of estradiol synthesis by aromatase inhibitors, noticeable (eventually reversible) effects on ovulation were predicted. Exposures to individual chemicals never led to such effects. In our best estimate, ∼10% of the combined exposures simulated had mild to catastrophic impacts on ovulation. A lower bound on that figure, obtained using an optimistic exposure scenario, was 0.3%.
    Conclusions:
    These results demonstrate the possibility to predict large-scale mixture effects for endocrine disrupters with a predictive toxicology approach that is suitable for high-throughput ranking and risk assessment. The size of the effects predicted is consistent with an increased risk of infertility in women from everyday exposures to our chemical environment. https://doi.org/10.1289/EHP742
  • Received: 30 June 2016
    Revised: 07 December 2016
    Accepted: 24 February 2017
    Published: 19 July 2017

    Address correspondence to F.Y. Bois, INERIS, DRC/VIVA, Parc ALATA, BP 2, 60550 Verneuil en Halatte, France. Telephone: 33-344-234-385; Email: frederic.bois@ineris.fr

    Supplemental Material is available online (https://doi.org/10.1289/EHP742).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (2 MB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (120 KB)

Introduction

Concern is growing worldwide over the negative human health and environmental impacts of chemical pollutants that can interfere with the production, metabolism, and action of natural hormones, the so-called endocrine-disrupting chemicals (EDCs). In humans, EDCs have been linked to reproductive disorders (Sweeney et al. 2015), abnormal or delayed development in children (Schug et al. 2015), changes in immune function (Rogers et al. 2013), and cancer (Birnbaum and Fenton 2003). Exposure to mixtures of EDCs may result in effects that can depart from mere summation (Kortenkamp 2007), and human subgroups (e.g., women) may not be sufficiently protected against mixtures of EDCs by current regulatory limits (Kortenkamp 2014).

Each menstrual cycle in women involves hormonal regulation of follicular growth and maturation resulting in ovulation of a single oocyte (Falcone and Hurd 2013). The cycle is controlled by coordinated stimulations and inhibitions along the hypothalamus–pituitary–ovarian axis. Gonadotropin-releasing hormone (GnRH), secreted by the hypothalamus, stimulates the secretion of gonadotropins [follicle-stimulating hormone (FSH) and luteinizing hormone (LH)] by the anterior pituitary gland. Those hormones, in turn, regulate the secretion of ovarian hormones, such as estradiol (E2) or progesterone (P4). Exposures to EDCs that interfere directly or indirectly with any of these hormones can eventually induce infertility or other pathological outcomes. Aromatase is critical because it irreversibly converts testosterone to E2 and androstenedione to estrone, maintaining the dynamic balance between androgens and estrogens.

The objective of the present work was to explore predictively the effects of exposure to large-scale (i.e., potentially real-life) mixtures of aromatase inhibitors on the dynamics of menstrual cycling in women. We input exposure estimates from ExpoCast (Wambaugh et al. 2013) and biological effect data from ToxCast™ (Dix et al. 2007) to coupled pharmacokinetic (PK) and ovarian cycle models; this provided a quantitative mechanistic link between exposure to mixtures of EDCs and their potential adverse effects on the menstrual cycle in women. We compared the expected effects of exposures to single EDCs, as is usually considered by risk assessment provisions in different regulations, with estimated effects of cumulative and concurrent exposures.

Methods

Workflow Overview

The overall computational workflow is pictured in Figure S1. Briefly, after selecting the chemicals of interest, we sampled millions of random mixtures of chemicals using the exposure estimates provided by ExpoCast (Wambaugh et al. 2013). Both constant and time-varying exposure scenarios of an adult woman were considered. A pharmacokinetic model of intake and disposition was then used to estimate the blood concentration (over 2 y) for each chemical present in each mixture. The resulting aromatase activity inhibition was estimated using the Hill’s dose–response model parameters provided by ToxCast™ (Dix et al. 2007). A mathematical model of the hypothalamus–pituitary–ovarian hormonal events [based on a study by Chen and Ward (2014)] was used to predict the levels of E2, P4, and other quantities characterizing the ovarian cycle, for a reference cycle and following exposure to the mixtures generated. Monte Carlo sampling (Bois et al. 2010) was used to propagate uncertainties in exposure, kinetics, and dose–response relationships up to ovarian cycle perturbation.

Databases, Chemical Selection, and Mixture Sampling

ExpoCast (Wambaugh et al. 2013) provides exposure estimates, with measures of uncertainty, for 1,936 chemicals. Those exposure estimates were obtained using far-field, mass-balance human exposure models (USEtox® and RAIDAR).

In ToxCast™ (11 December 2013 release), the Tox21-aromatase-inhibition assay is a cell-based assay measuring CYP19A1 (aromatase) gene activity via a fluorescent protein reporter gene. Chemicals acting on aromatase mRNA synthesis, degradation, or translation, or on aromatase itself, should give positive results in this assay (Chen et al. 2015). For each chemical x assayed, ToxCast™ provides the geometric mean and standard error for the parameter values AC50,x (half-maximal response), Wx (exponent), Bx (baseline value) and Tx (maximum value) of a Hill function fitted to the concentration–inhibition data (scaled using the positive and negative controls’ data):

In ToxCast™, 1,102 chemicals are identified as aromatase inhibitors by the Tox21-aromatase-inhibition assay (on MCF-7 human breast cells) with a fitting that “mate basic requirements of Hill model with some minimal confidence in T and B.” Among those, 256 chemicals (matching either by CAS number or by chemical name) also had exposure estimates in ExpoCast. However, cytotoxicity has been shown to induce many false positive results in ToxCast™ (Judson et al. 2016). Of the 256 chemicals mentioned above, we kept only the 86 that had an AC90 (90% of maximal response) for aromatase inhibition lower than their cytotoxicity AC10 (10% of maximal response) (as measured by the ToxCast™ proliferation decrease assay on T47D human breast cells). The virtual mixtures generated included all of those 86 chemicals.

Exposure Modeling

ExpoCast provided the molecular mass, geometric mean, and lower and upper 95% confidence limits of the exposure rate (milligrams per kilogram per day) for each chemical present in the randomly generated mixtures. For constant exposure modeling over 10 mo, we sampled a rate for each chemical from the corresponding log-normal distribution, but with a standard deviation (SD) scaled by the square root of the number of days of exposure simulated (because a constant exposure rate should be a time average level in that case) and converted it to micromoles per kilogram per minute.

More realistic time-varying exposures over a 2-y period were also modeled [similarly to the report of Bertail et al. (2010)] using exposure windows of random length and intensity. We first sampled the number n of exposure windows for each chemical in the mixture from a scaled exponential distribution (with a rate parameter equal to 5). That yielded on average 145 exposure events over 2 y (median: 100 events, first quartile: 40 events, third quartile: 200 events). The n exposures’ start and end times were sampled uniformly over the 2-y period. The exposure rate during each of the n exposure windows was randomly sampled from the log-normal distribution given by ExpoCast, with an SD scaled by the number of days of the exposure window considered (or unscaled if the exposure lasted less than a day).

To obtain a lower bound on the effect of mixtures, we similarly simulated random nonoverlapping exposures to the 86 EDCs selected (i.e., each person was exposed to the 86 chemicals, in random order, at random times, but to only one chemical at a time).

Pharmacokinetics Modeling

For each chemical in each mixture, a one-compartment PK model was used to estimate its internal concentration (in micromoles) at steady state in the case of constant oral exposure or at any point in time in the case of varying oral exposures. Steady-state internal concentrations for chemical x were calculated as:
where Fx is the bioavailability of x (unitless), Ex is its exposure rate (in micromoles per kilogram per minute, sampled as indicated above), and Ke,x is its total body clearance rate constant (per minute). A body density of 1 was assumed.

For time-varying exposures, internal concentrations were obtained as a function of time by numerical integration of the following differential over a 2-y period (with an initial value set to zero):

We used quantitative structure-activity relationships (QSAR) to obtain central estimates of Fx and Ke,x for each of the 86 aromatase inhibitors considered. The robustness of the prediction was evaluated by examining compounds from the training set similar to the target substances, together with literature data and references.

Fx central estimates were obtained at several oral dose levels and were linearly interpolated between dose levels as needed. Beyond the dose rates of 0.001 to 10 mg/d, the Fx value at the closest bound was used.

To take into account the uncertainty affecting the QSAR-estimated PK parameters, we randomly sampled F values from beta distributions (naturally bounded between 0 and 1), with parameters calculated such that the distribution modes corresponded to the interpolated value of Fx with a coefficient of variation (CV) of 20% (for null Fx modes, a and b were set to 1 and 50, yielding a median at 0.01, a first quartile at 0.006, and a third quartile at 0.03, approximately). Ke,x values were log-normally sampled with a geometric mean equal to the central estimates obtained by QSAR and a geometric SD corresponding to a factor of 3.

Ovarian Cycle Model

We adapted the menstrual cycle model presented by Chen and Ward (2014). The model describes the inhibitory and stimulatory effects of hormones E2 and P4 on the hypothalamus–pituitary axis in women (Figure 1). The equations and definitions of all parameters used in the model are given in the Supplemental Material (see “Menstrual cycle model equations”; see also Tables S3 and S4). In the original model, E2 and P4 were assumed to be instantaneously in equilibrium between blood and the ovaries. Instead, we described the kinetics of E2 and P4 using differential equations (Equations 15–19 in the Supplemental Material, “Menstrual cycle model equations”). That modification had practically no impact on the time course of the model variables during a normal cycle [equilibrium between blood and ovaries is fast, as assumed by Chen and Ward (2014)], but it allowed us to coherently integrate the dynamic aspect of estradiol synthesis inhibition by EDCs. The three additional parameters (blood and ovarian volumes, ovarian blood flow) were obtained from the literature (see Table S3).

Conceptual diagram of the ovarian cycle model, showing the hormonal controls between the brain and the ovaries.
Figure 1. Regulatory pathways of the human menstrual cycle as implemented in the model. During the follicular phase (1: germ cells; 2: developing follicle; 3: mature follicle; 4: ovulation; 5: corpus luteum formation; 6: corpus luteum degradation), negative feedback by estradiol (E2) reduces follicle-stimulating hormone (FSH) secretion, leading to the selection of one follicle for ovulation. Gonadotropin-releasing hormone (GnRH) secretion is promoted by E2 and inhibited by progesterone (P4), inducing a luteinizing hormone (LH) peak and consecutive ovulation. E2 is mainly produced by follicles and corpus luteum and P4 by corpus luteum.

An additional variable, the ratio of disrupted over basal E2 synthesis rate constants (EDCYP19), was introduced to link the internal doses of chemicals in mixtures to aromatase inhibition. EDCYP19 was calculated using Hill’s model, parameterized with the chemical-specific values provided by ToxCast™, as a cumulative product of remaining activity for each of the m chemicals of the mixture considered:

For constant exposures, Cx was set to the steady-state internal concentrations Cx,ss. For time-varying exposures, Cx was computed by integration as explained above. The parameter Bx was set to zero because a positive inhibition with no dosage would not make sense. AC50,x, Wx, and Tx values were randomly sampled using the mean and standard error provided by ToxCast™. AC50,x was sampled from a log-normal distribution (its logarithm is actually the ToxCast™ fitted value). Wx and Tx were sampled from truncated normal distributions. Truncation was from 0 to 10 for Wx (values beyond 10 would be found for some chemicals for which Wx is poorly identified, but have no biological meaning). Truncation was from 0 to 100 for letrozole’s Tx (the positive control). For the other chemicals, truncation was from 0 to 10,000 over Tx for letrozole to properly rescale the ToxCast™ Tx values between 0 and 100.

EDCYP19 was entered as an input to the ovarian cycle model, which was then solved to obtain the time profile of its output variables over 2 y of simulated time. For constant exposures, we computed the square root of the sum of the squared Euclidean distances between a reference E2 concentration (EDCYP19 set at zero) and the perturbed concentrations (at a fixed set of times) as a summary measure of disruption.

Software Used

The ACD/Labs Percepta platform modules ACD/Oral Bioavailability and ACD/PK Explorer were used for the prediction of oral bioavailability (F) and the total body clearance rate constant (Ke), respectively (see Supplemental Material, “PK modules of ACD/Labs”; see also Table S1 and Figures S2 and S3). GNU MCSim v5.6.5 (www.gnu.org/software/mcsim) (Bois 2009) was used to build the ovarian cycle model. R v3.1.1 (R Development Core Team) with the parallel, deSolve, and EnvStats packages was used for database processing, numerical integration of the models, and graphics.

Results

Estimates of Internal Dose

The relationship between constant exposure rates and steady-state internal concentrations for the 86 EDCs considered indicates that exposures ranged from 10−8μmole/kg/d to 10−3μmole/kg/d and that the resulting steady-state internal concentrations ranged from 10−13μM to 10−3μM (see Figure S4). The exposure rates and pharmacokinetic parameters were Monte Carlo sampled as described above. For any single EDC, uncertainty is approximately a factor of 10 for exposures and approximately a factor of 1,000 for the resulting internal concentrations. For time-varying exposures, Figure S5 shows an example of a simulated random 2-y time course of internal concentration for lindane. Such profiles were obtained for each chemical in each simulated mixture.

Cycle Model Behavior

Our implementation of the ovarian cycle model proposed by Chen and Ward (2014) correctly reproduces their results. Human data from McLachlan et al. (1990) and Welt et al. (1999) on LH, FSH, E2, and P4 normal cycles are correctly simulated except for McLachlan FSH data, for which the baseline levels are not well matched. There is a large intra- and inter-subject variability in hormonal levels across women in those data sets (see Figure S6). We took the model-simulated normal cycling of E2 as the “reference cycle” in the following. Constant exposure scenarios result, at steady state, in a constant level of aromatase inhibition. In that case, perturbation depends only on that parameter (according to the model assumptions), so the distance between the perturbed and reference cycles is a useful measure of effect (see Figure 2). As aromatase inhibition increases, cycles become increasingly perturbed and exhibit chaotic features (hence the misalignment of the points in Figure 2). At 5% inhibition (95% of normal aromatase activity), cycles are shortened, baseline levels change little, and peak levels either increase or decrease less than proportionally except for LH. Simply put, the regulations dampen the effect of perturbation. At ∼10% inhibition, LH peaks disappear after approximately five cycles, and a major bifurcation in cycle patterns occurs: cycles are further shortened, baseline levels are much increased (doubled for E2 and P4, for example, even though E2 synthesis by aromatase is decreased), and peak levels mostly decreased; E2 distance to normal increases up to a maximum (Figure 2). At higher inhibition levels, the cycles increasingly dampen and disappear completely between 30% and 40% inhibition (see Figures S7–S10). Overall, according to this model, having ∼10% constant inhibition of aromatase activity in vivo leads to perturbations of the cycle, which is still under control and should be compatible with normal reproductive function. Beyond 10% inhibition, an actual disruption of the system seems to occur.

Scatter plot with a regression line plotting Euclidian distance to reference E2 (y-axis) across percentage of aromatase inhibition (x-axis).
Figure 2. Euclidean distance between an estradiol (E2) normal cycle and a perturbed cycle as a function of aromatase inhibition at steady state. Distance is computed over 2 y based on 7,301 time points (one every 144 min).

Effects of Single Chemicals

We first simulated 1,000,000 constant exposures to each of the 86 chemicals considered, taken individually. In that case, despite accounting for uncertainty in exposure levels and dose–response parameters, none induced >1% aromatase inhibition. Hence, none of those chemicals alone was able to induce a significant disruption of the ovarian cycle. Figure 3 places those chemicals on a map with the slope W of the Hill dose–response curve at AC50 and the log-margin of exposure as coordinates. The margin of exposure was defined as the ratio of the 97.5th percentile of internal concentrations over AC50. The log10-margins of the chemicals studied ranged from −10 to −1.8, indicating that for all chemicals, the high end of internal exposure concentrations was at most 1% of AC50. In that case, Equation 1 shows that the logarithm of aromatase inhibition is approximately equal to the product of W times the log-margin of exposure. The color background of the map codes for the resulting risk index (i.e., the log10-inhibitions) and ranges from <−2 (1%) to approximately −80(10−78 %), much too low to elicit changes in ovarian cycles such as those in Figure 2. Therefore, no effects can be expected from typical exposures to those chemicals when considered alone. Table 1 gives the list of the 10 chemicals for which individual risk is the highest. Note that letrozole is the reference chemical for the Tox21 aromatase inhibition assay, which is consistent with its high rank. The others are found in therapeutic drugs, agrochemicals, food contaminants, consumer products, and other materials.

Inhibition map plotting log sub 10 margin of exposure (y-axis) across slope of W at AC sub 50 (x-axis).
Figure 3. Map of the aromatase inhibitors studied over a plane defined by dose–response slope W and log-margin of exposure (see text). Colors vary linearly with powers of 10 of aromatase inhibition resulting from W and margin of exposure combinations, from red (10−2) to blue (10−80). For the chemicals–numbers correspondence, see Table S1.
Table 1. Top-ranking chemicals according to their individual risk indices. Those chemicals are in the top left corner of Figure 3.
ExpoCast name CAS number Risk indexa
Letrozole 112809-51-5 −2.99
Estrone 53-16-7 −3.41
Fulvestrant 129453-61-8 −4.30
Triflumizole 68694-11-1 −4.81
2,4,7,9-Tetramethyl-5-decyne-4,7-diol 126-86-3 −5.11
n-Methyl-2-pyrrolidone 872-50-4 −5.49
Rhodamine 6G 989-38-8 −5.51
Anastrozole 120511-73-1 −6.04
Fenvalerate 51630-58-1 −6.05
Imazalil 35554-44-0 −6.31

aRisk index=W×log10(E95/AC50), where E95 is the 95th percentile of the exposure values sampled, W is the Hill exponent, and AC50 the half-maximal response.

Effects of Mixtures of Chemicals

We generated 1,000,000 hypothetical mixtures of the 86 aromatase inhibitors studied and evaluated their global effect on E2 synthesis and the resulting ovarian cycle disruption. Figure 4 shows a histogram of the resulting inhibition levels. Depending on the (random) composition of the mixtures, and given the uncertainties on exposures and effect parameters, responses ranged from 0% to 100% inhibition, but on average were very high. Such inhibition levels may lead to perturbation of the ovarian cycle, according to the model.

Histogram of the average percent inhibition of E2 synthesis by random mixtures at constant exposure levels.
Figure 4. Histogram of the average percent inhibition of estradiol (E2) synthesis by random mixtures (n=1,000,000) of 86 aromatase inhibitors at constant exposure levels.

Real-life exposures to chemicals do not usually occur at constant levels. We simulated time-varying exposures (see Figure S5) to investigate the resulting effects on aromatase inhibition and ovarian cycle disruption. To define a lower bound on mixture effects, we first simulated fluctuating levels of exposure to EDCs, but without concomitant exposures to them. Interactions can still occur in that case because of storage in the body or because of persistent effects on the ovarian cycle. Figure 5 shows that only 0.3% of the simulated exposures caused >10% average aromatase inhibition. The maximum inhibition found was close to 50%. More realistic exposure scenarios do not prohibit concomitant exposures. In that case (Figure 6), the distribution of simulated time-averaged inhibitions is shifted toward greater effects, and average inhibitions >20% are not uncommon (yet they do not reach the extreme levels observed in Figure 4). Because inhibition changes with time, the distances between normal and perturbed cycles do not follow the pattern shown in Figure 2 (distances can be much larger), and the link between estradiol inhibition and cycle disruption is harder to establish. Examination of the time course of the dominant follicle mass (F) for 1,000 random simulations of mixtures of the 86 chemicals shows that perturbed cycles typically have a baseline shifted up (which may or may not return to normal) and an irregular succession of peaks (corresponding to ovulation) (Figure 7). The 1,000 simulations examined can be classified into four groups. In group 1 (17% of the samples), the cycles are practically normal with no baseline shifts and at most one or two missing ovulations. In group 2 (73% of cases), baseline shifts are always present but without major irregularities in ovulation. Group 3 (7% of cases) has systematic baseline shifts and frequent or prolonged anovulations. Such cycling would clearly impair fertility. In group 4 (3% of cases), disruption is catastrophic or total. Figure S11 shows the corresponding plots for E2 time courses. Judging by these plots, E2 profiles can have a shifted baseline even in normal ovulation profiles, but otherwise, the patterns are rather similar.

Histogram of the average percent inhibition of E2 synthesis by random mixtures at nonconcomitant exposure levels.
Figure 5. Histogram of the average percent inhibition of estradiol (E2) synthesis by random mixtures (n=1,000,000) of 86 aromatase inhibitors with time-varying nonconcomitant exposures. For visibility, the first histogram bar has been truncated (it represents ∼90,000 simulations). Note: E2, estradiol.
Histogram of the average percent inhibition of E2 synthesis by random mixtures at variable exposure levels.
Figure 6. Histogram of the average percent inhibition of estradiol (E2) synthesis by random mixtures (n=1,000,000) of 86 aromatase inhibitors with time-varying exposures.
Simulated time profiles plotting follicular mass in arbitrary units (y-axis) across time (x-axis).Figure 7. Typical simulated time profiles of dominant follicle mass during time-varying exposures to random mixtures of 86 aromatase inhibitors. Four classes (rows) of increasing disruption are illustrated (see text). Left column: least disrupted profile in its class; Right column: most disrupted. Responses range from regular ovulation (1A) to complete disruption (4B).

Discussion

We linked ToxCast™ data and ExpoCast estimates of exposures to (mixtures of) aromatase inhibitors and estimated their effects on the ovarian cycle in women. To our knowledge, this is the first application of computational toxicology and high-throughput testing to assessment of the combined effects of exposures to a large number of EDCs.

Our approach is predictive, and we had to make many assumptions and simplifications. ToxCast™ and ExpoCast are incomplete, and a full inventory of all the EDCs to which women are exposed is not yet available. Therefore, we were only able to look at a subset of the potential EDCs. We used human exposure estimates reported by Wambaugh et al. (2013), who gave summary statistics for the distribution of exposures to individual chemicals. This information allowed us to take the correspondingly large uncertainty into account via Monte Carlo simulations. However, we cannot differentiate between uncertainty and variability in those exposure estimates, and we cannot identify subgroups of sensitive individuals. We cannot even focus on women, our target population. More sophisticated exposure models (Isaacs et al. 2014) could help in that respect, but they still deal only with single chemicals and provide no data or estimates on coexposures. Depending on age, occupation, socioeconomic status, ethnicity, and health condition, we are exposed to different cocktails of chemicals in our diet, workplace, environment, and so forth (Tornero-Velez et al. 2012). We modeled coexposures by random sampling, either at a constant level or, more realistically, with time-varying exposure profiles and hence with time-varying mixture complexity. That approach is still imperfect, and we had to guess about the distribution of the number of exposure windows, for example. We respected the distribution of population exposure levels documented by ExpoCast, but lacking coexposure information, our estimates might be lower or higher than in reality. Efforts are ongoing to collect relevant data in, for example, the European Total Diet Study (Vin et al. 2014). An analysis of such data (Traoré et al. 2016) shows that among 153 synthetic chemicals studied in seven typical French diets (food associations), three are aromatase inhibitors according to ToxCast™: zearalenone, triadimenol, and lindane. In this regard, mixtures of ≥86 aromatase inhibitors may seem unrealistic, but only food contaminants were studied by Traoré et al. (2016).

We searched for the 86 selected aromatase inhibitors in a database of consumer products marketed in the United States (Gabb and Blake 2016). Briefly, this database was constructed by scraping product information from online retailers and currently contains 53,743 products. Twelve of the 86 aromatase inhibitors were detected in 5,701 products (representing 11% of the products in the database). [It is worth noting that none of the 86 aromatase inhibitors is among the volatile fragrance chemicals detected in consumer products (Steinemann 2015; Steinemann et al. 2011), so aromatase inhibitors are unlikely to be hidden in generic “fragrance” or “flavor” designations on consumer product labels.] Two-way combinations of these chemicals were found in 220 products, and the three-way combination of carminic acid, FD&C blue no. 1, and retinol was found in 3 products. These findings may not seem to indicate a large problem, but it is an incomplete view of combinatorial exposure. Consider that 3,660 of 4,501 makeup products (81%) in the database contain at least one of the aromatase inhibitors evaluated (carminic acid, retinol, and artificial colors are common ingredients in makeup) and that a typical consumer uses several products each day, possibly even several makeup products. This increases the likelihood of combined exposures. In addition, no readily available data address associations for all near- and far-field exposures for the 86 aromatase inhibitors, which are used in industrial or agricultural processes (18), consumer product formulations (12), biocidal applications (38), and pharmaceutical drugs (18). These usage categories are likely to be independent, so focusing on a few known associations would only give partial answers and would underestimate global risk. We are striving for a more extensive picture. To address the potential overestimation of mixture effects when generating purely random associations, we present the results of a very optimistic exposure scenario (with no coexposures at all). This scenario gives a lower bound estimate: 0.3% of exposures would lead to >10% average aromatase inhibition in women.

ToxCast™ aromatase inhibition data were obtained by exposing cells in vitro. We had no easy way to assess the in vitro kinetics of the substances assayed, and reconstruction methods (Armitage et al. 2014) require input data that we did not have. We assumed that the nominal assay concentrations were those actually experienced by the cells and that equivalent extracellular concentrations in vivo would lead to the same levels of aromatase inhibition. That is a typical assumption, but it is not necessarily correct (Coecke et al. 2013). To obtain extracellular concentrations in vivo, we estimated bioavailability and total clearance with QSAR methods and input them in a simple one-compartment PK model with oral exposure only (even though inhalation or dermal exposures might be more relevant). Again, more sophisticated [physiologically based pharmacokinetic (PBPK)] models and additional PK data would give more precise and more accurate predictions (El-Masri et al. 2016; Wambaugh et al. 2015).

On the effect side, we only considered chemicals for which the ToxCast™ dose–response parameters were estimated with reasonable confidence. This was a conservative choice, and additional chemicals would have been included if more relaxed criteria had been chosen (at the cost of lower confidence in the results). A concern with large databases such as ToxCast™ is the quality assurance for the data provided. Aromatase inhibition may not be the most sensitive toxicity end point, for example, or the action may be due to a burst effect of cytotoxicity (Judson et al. 2016). From the original set of 256 aromatase inhibitors for which we had exposure estimates and ToxCast™ data, 170 were excluded based on cytotoxicity. Those were mostly weak inhibitors (data not shown). For the screen, we kept only the 86 substances that had an aromatase inhibition AC90 lower than their cytotoxicity AC10 as measured by the ToxCast™ proliferation decrease assay on T47D human breast cells, a cell type similar to the MCF-7 used by the aromatase inhibition assays (Aka and Lin 2012). We preferred that screen to the omnibus z-score criterion, which aggregates cytotoxicity results from different cell types and species. For the remaining 86 substances, we may still have downplayed other types of toxicity. Cytotoxicity, it should be noted, is not a negligible end point, and an evaluation of the cytotoxicity of mixtures would be interesting in its own right.

The evidence provided by ToxCast™ is also not perfectly predictive of in vivo outcomes in humans. For example, the Tox21 aromatase inhibition assay uses the MCF-7 breast cancer cell line, which might not respond as normal ovary cells would. In addition, our model of the ovarian cycle (Chen and Ward 2014) describes only approximately the complex dynamic interactions between ovarian follicular growth and hormonal homeostasis. The hypothalamus and pituitary gland are treated as a single compartment, and the description of the central hormonal controls is simplified. More complex models have been proposed (Hendrix et al. 2014), but they still make many assumptions and do not seem to offer dramatically better performance. In addition, we note that we treated the parameters of the ovarian cycle as constant, when in fact they are affected by both uncertainty and variability in response to EDCs, adding to the tails of the distributions of our results. However, we did not have sufficient information to define statistical distributions for those parameters.

In terms of results, an obvious question is that of the “bad” actors, that is, the chemicals responsible for the predicted effects. The answer is provided by Figure 3 (and partially by the top-ten Table 1) because here, the impact of individual chemicals on aromatase is only conditioned by internal dose and inhibition potency. Figure 3 is a useful prioritization tool. It is rather simple to construct and does not require running the ovarian cycle (the PK model is needed). However, it does not give an answer in terms of magnitude of effect at the subject level. For that, we need the whole-body ovarian cycle model.

One of the consistent features of E2 cycle perturbation that we found is that the baseline (interovulation) levels of E2 tend to increase (to approximately twice the normal level) in response to aromatase inhibition (which implies a lower rate of E2 synthesis by aromatase). That counterintuitive feature of the complex cycle dynamic is induced by central nervous system (CNS) feedback. Beyond a certain level of inhibition, the control of E2 remains in effect (peaks are still observed) but moves the baseline to a higher value. We do not have confirmation that this is the case in women exposed to EDCs, but that would be interesting information and a potential biomarker of effects. We also note that we lack good measures of perturbation for such complex systems. We used visual inspection to classify cycles for 1,000 time-varying exposures (Figure 7). Perturbation analysis of more cycles would require more sophisticated tools. Finally, many other perturbation pathways exist for ovarian cycle disruption that were not accounted for (e.g., actions mediated by the androgen or estrogen receptor). The simplicity of our model also precludes investigation of synergistic or antagonistic effects that could result from metabolic or toxicodynamic interactions (Cheng and Bois 2011).

Conclusion

High-throughput data collection requires high-throughput analysis, extrapolation, and decision tools if we want to avoid a bottleneck and accumulation of unused data. We developed such a tool, making use of our increasing understanding of toxicity mechanisms.

This exercise in prediction suggests large data gaps. Our knowledge of exposure to actual mixtures is minimal except in a few cases (e.g., tar, tobacco smoke). For the chemicals studied here, quantitative knowledge of their routes of exposure and PK parameters is also lacking. Our knowledge of endocrine disruption mechanisms is still in its infancy. We should check, for example, the inhibition potential of the 86 substances studied here with better tests and with better characterization of the in vitro fate of the chemicals. Nevertheless, we now have a prioritized list and some reason to deepen our investigations. We should also research relevant human biomarkers of exposure and effects for validation of the results and for better actions.

The basic assumption of regulatory practice and most risk assessments is that keeping individual chemicals under control with reasonable safety factors will keep their joint effects at bay. That may not be the case. Some women have various levels of ovarian dysfunction that can be caused by various internal and external factors. Environmental chemical exposures add on to that background (National Research Council 2009; Zeise et al. 2013). We found that even though individual chemicals are likely “safe” as used now, their joint effects, when they exceed a few dozen in number, can lead to severe disruption of the ovarian cycle in women in a sizable number of cases. Obviously, this study does not provide definite proof that some fertility problems can be caused by real-life exposures to EDCs. Nevertheless, risk assessment practice and regulations should start thinking of the problem of mixtures not as an unsolvable one, but as needing a clearly laid out research agenda. In the case in point, the simple graphical map of internal dose and potency we propose could already be used for prioritization. Given the magnitude of the range of the predicted effects on ovarian cycling, while waiting for confirmation of our results, a cautionary attitude should be adopted.

Acknowledgments

The authors thank the reviewers for their helpful comments.

We thank the U.S. Environmental Protectin Agency Computational Toxicology Research team for its help in using ToxCast™ and ExpoCast. The research leading to these results received funding from the European Union’s 7th Framework Program and Cosmetics Europe [grant agreement 266835 (COSMOS)], the European Union’s Horizon 2020 research and innovation program [grant agreement 633172 (Euromix)], and the French Ministry for the Environment (PRG190).

References

Aka JA, Lin S-X. 2012. Comparison of functional proteomic analyses of human breast cancer cell lines T47D and MCF7. PLoS ONE 7(2):e31532, PMID: 22384035, 10.1371/journal.pone.0031532.

Armitage JM, Wania F, Arnot JA. 2014. Application of mass balance models and the chemical activity concept to facilitate the use of in vitro toxicity data for risk assessment. Environ Sci Technol 48(16):9770–9779, PMID: 25014875, 10.1021/es501955g.

Bertail P, Clémençon S, Tressou J. 2010. Statistical analysis of a dynamic model for dietary contaminant exposure. J Biol Dyn 4(2):212–234, PMID: 22876987, 10.1080/17513750903222960.

Birnbaum LS, Fenton SE. 2003. Cancer and developmental exposure to endocrine disruptors. Environ Health Perspect 111(4):389–394, PMID: 12676588, 10.1289/ehp.5686.

Bois FY. 2009. GNU MCSim: Bayesian statistical inference for SBML-coded systems biology models. Bioinformatics 25(11):1453–1454, PMID: 19304877, 10.1093/bioinformatics/btp162.

Bois FY, Jamei M, Clewell HJ. 2010. PBPK modelling of inter-individual variability in the pharmacokinetics of environmental chemicals. Toxicology 278(3):256–267, PMID: 20600548, 10.1016/j.tox.2010.06.007.

Chen CY, Ward JP. 2014. A mathematical model for the human menstrual cycle. Math Med Biol 31(1):65–86, PMID: 23329626, 10.1093/imammb/dqs048.

Chen S, Hsieh JH, Huang R, Sakamuru S, Hsin LY, Xia M, et al. 2015. Cell-based high-throughput screening for aromatase inhibitors in the Tox21 10K library. Toxicol Sci 147(2):446–457, PMID: 26141389, 10.1093/toxsci/kfv141.

Cheng S, Bois FY. 2011. A mechanistic modeling framework for predicting metabolic interactions in complex mixtures. Environ Health Perspect 119(12):1712–1718, PMID: 21835728, 10.1289/ehp.1103510.

Coecke S, Pelkonen O, Leite SB, Bernauer U, Bessems JG, Bois FY, et al. 2013. Toxicokinetics as a key to the integrated toxicity risk assessment based primarily on non-animal approaches. Toxicol. In Vitro 27(5):1570–1577, PMID: 22771339, 10.1016/j.tiv.2012.06.012.

Dix DJ, Houck KA, Martin MT, Richard AM, Setzer RW, Kavlock RJ. 2007. The ToxCast program for prioritizing toxicity testing of environmental chemicals. Toxicol Sci 95(1):5–12, PMID: 16963515, 10.1093/toxsci/kfl103.

El-Masri H, Kleinstreuer N, Hines RN, Adams L, Tal T, Isaacs K, et al. 2016. Integration of life-stage physiologically based pharmacokinetic models with adverse outcome pathways and environmental exposure models to screen for environmental hazards. Toxicol Sci 152(1):230–243, PMID: 27208077, 10.1093/toxsci/kfw082.

Falcone T, Hurd WW, eds. 2013. Clinical Reproductive Medicine and Surgery. New York, NY:Springer.

Gabb HA, Blake C. 2016. An informatics approach to evaluating combined chemical exposures from consumer products: a case study of asthma-associated chemicals and potential endocrine disruptors. Environ Health Perspect 124:1155–1165, PMID: 26955064, 10.1289/ehp.1510529.

Hendrix AO, Hughes CL, Selgrade JF. 2014. Modeling endocrine control of the pituitary–ovarian axis: androgenic influence and chaotic dynamics. Bull Math Biol 76(1):136–156, PMID: 24272388, 10.1007/s11538-013-9913-7.

Isaacs KK, Glen WG, Egeghy P, Goldsmith MR, Smith L, Vallero D, et al. 2014. SHEDS-HT: an integrated probabilistic exposure model for prioritizing exposures to chemicals with near-field and dietary sources. Environ Sci Technol 48(21):12750–12759, PMID: 25222184, 10.1021/es502513w.

Judson R, Houck K, Martin M, Richard AM, Knudsen TB, Shah I, et al. 2016. Editor’s Highlight: analysis of the effects of cell stress and cytotoxicity on in vitro assay activity across a diverse chemical and assay space. Toxicol Sci 152(2):323–339, PMID: 27208079, 10.1093/toxsci/kfw092.

Kortenkamp A. 2014. Low dose mixture effects of endocrine disrupters and their implications for regulatory thresholds in chemical risk assessment. Curr Opin Pharmacol 19:105–111, PMID: 25244397, 10.1016/j.coph.2014.08.006.

Kortenkamp A. 2007. Ten years of mixing cocktails: A review of combination effects of endocrine-disrupting chemicals. Environ Health Perspect 115(suppl 1):98–105, PMID: 18174957, 10.1289/ehp.9357.

McLachlan RI, Cohen NL, Dahl KD, Bremner WJ, Soules MR. 1990. Serum inhibin levels during the periovulatory interval in normal women: relationships with sex steroid and gonadotrophin levels. Clin Endocrinol (Oxf) 32(1):39–48, PMID: 2110047.

National Research Council. 2009. Science and Decisions: Advancing Risk Assessment. Washington, D.C:National Academies Press.

Rogers JA, Metz L, Yong VW. 2013. Review: Endocrine disrupting chemicals and immune responses: A focus on bisphenol-A and its potential mechanisms. Mol Immunol 53(4):421–430, PMID: 23123408, 10.1016/j.molimm.2012.09.013.

Schug TT, Blawas AM, Gray K, Heindel JJ, Lawler CP. 2015. Elucidating the links between endocrine disruptors and neurodevelopment. Endocrinology 156(6):1941–1951, PMID: 25714811, 10.1210/en.2014-1734.

Steinemann A. 2015. Volatile emissions from common consumer products. Air Qual Atmos Health 8:273–281, 10.1007/s11869-015-0327-6.

Steinemann AC, MacGregor IC, Gordon SM, Gallagher LG, Davis AL, Ribeiroc DS, et al. 2011. Fragranced consumer products: chemicals emitted, ingredients unlisted. Environ. Impact Assess. Rev 31(3):328–333, 10.1016/j.eiar.2010.08.002.

Sweeney MF, Hasan N, Soto AM, Sonnenschein C. 2015. Environmental endocrine disruptors: Effects on the human male reproductive system. Rev Endocr Metab Disord 16(4):341–357, PMID: 26847433, 10.1007/s11154-016-9337-4.

Tornero-Velez R, Egeghy PP, Cohen Hubal EA. 2012. Biogeographical analysis of chemical co-occurrence data to identify priorities for mixtures research: biogeographical analysis of chemical co-occurrence data. Risk Anal 32(2):224–236, PMID: 21801190, 10.1111/j.1539-6924.2011.01658.x.

Traoré T, Béchaux C, Sirot V, Crépet A. 2016. To which chemical mixtures is the French population exposed? Mixture identification from the second French Total Diet Study. Food Chem. Toxicol 98(part B):179–188, 10.1016/j.fct.2016.10.028.

Vin K, Papadopoulos A, Cubadda F, Aureli F, Oktay Basegmez HI, D’Amato M, et al. 2014. TDS exposure project: relevance of the Total Diet Study approach for different groups of substances. Food Chem. Toxicol 73:21–34, PMID: 25106751, 10.1016/j.fct.2014.07.035.

Wambaugh JF, Setzer RW, Reif DM, Gangwal S, Mitchell-Blackwood J, Arnot JA, et al. 2013. High-throughput models for exposure-based chemical prioritization in the ExpoCast project. Environ Sci Technol 47(15):8479–8488, PMID: 23758710, 10.1021/es400482g.

Wambaugh JF, Wetmore BA, Pearce R, Strope C, Goldsmith R, Sluka JP, et al. 2015. Toxicokinetic triage for environmental chemicals. Toxicol Sci 147(1):55–67, PMID: 26085347, 10.1093/toxsci/kfv118.

Welt CK, McNicholl DJ, Taylor AE, Hall JE. 1999. Female reproductive aging is marked by decreased secretion of dimeric inhibin. J Clin Endocrinol Metab 84(1):105–111, PMID: 9920069, 10.1210/jcem.84.1.5381.

Zeise L, Bois FY, Chiu WA, Hattis D, Rusyn I, Guyton KZ. 2013. Addressing human variability in next-generation human health risk assessments of environmental chemicals. Environ Health Perspect 121(1):23–31, PMID: 23086705, 10.1289/ehp.1205687.

Indoor Wood-Burning Stove and Fireplace Use and Breast Cancer in a Prospective Cohort Study

Author Affiliations open

Epidemiology Branch, National Institute of Environmental Health Sciences, NIH, Research Triangle Park, North Carolina, USA

PDF icon PDF Version (144 KB)

  • Background:
    Indoor burning of fuel for heating or cooking releases carcinogens. Little is known about the impact of indoor air pollution from wood-burning stoves or fireplaces on breast cancer risk.
    Objectives:
    In a large prospective cohort study, we evaluated the risk of breast cancer in relation to indoor heating and cooking practices.
    Methods:
    Sister Study participants (n=50,884) were recruited from 2003–2009. Breast cancer–free women in the United States or Puerto Rico, 35–74 y old, with a sister with breast cancer were eligible. Participants completed questionnaires on indoor heating and cooking practices for both their enrollment and their longest adult residence. Cox regression was used to estimate adjusted hazard ratios (HRs) and 95% confidence intervals (95% CIs) for the association between indoor heating/cooking and breast cancer.
    Results:
    A total of 2,416 breast cancer cases were diagnosed during follow-up (mean=6.4 y). Having an indoor wood-burning stove/fireplace in the longest adult residence was associated with a higher breast cancer risk [HR=1.11 (95% CI: 1.01, 1.22)]; the risk increased with average frequency of use [≥once/week, HR=1.17 (95% CI: 1.02, 1.34)] (p for trend=0.01). An elevated HR was seen for women burning wood [HR=1.09 (95% CI: 0.98, 1.21)] or natural gas/propane [HR=1.15 (95% CI: 1.00, 1.32)]. No association was observed for burning artificial fire-logs [HR=0.98 (95% CI: 0.85, 1.12)] except among women from western states [HR=1.36 (95% CI: 1.02, 1.81)].
    Conclusions:
    In this prospective study, using an indoor wood-burning stove/fireplace in the longest adult residence at least once a week and burning either wood or natural gas/propane was associated with a modestly higher risk of breast cancer. https://doi.org/10.1289/EHP827
  • Received: 18 July 2016
    Revised: 16 November 2016
    Accepted: 09 December 2016
    Published: 18 July 2017

    Address correspondence to A.J. White, Epidemiology Branch, National Institute of Environmental Health Sciences, National Institutes of Health, Research Triangle Park, NC 27709-2233 USA. Telephone: (919) 316-4867. Email: alexandra.white@nih.gov

    Supplemental Material is available online (https://doi.org/10.1289/EHP827).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (184 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (96 KB)

Introduction

Indoor air pollution, also referred to as household air pollution, is a public health issue of concern throughout the world. It was estimated that approximately 2.8 billion people across the globe were exposed to indoor air pollution from burning biomass indoors in 2010 (Bonjour et al. 2013). The International Agency for Research on Cancer (IARC) has classified indoor air pollution from the combustion of biomass as a Group 2A, or probable, carcinogen (IARC 2010b). The majority of the previous research on the carcinogenic potential of indoor air pollution has focused on cancers of the lung (Hosgood et al. 2010; Lissowska et al. 2005; Sapkota et al. 2008) or aerodigestive tract (Sapkota et al. 2008). Understanding the impact of indoor air pollution on other types of cancer, such as breast cancer, has recently been highlighted as a research priority (Reid et al. 2012).

Burning biomass, such as wood, in the home can result in exposure to carcinogens that are similar to those found in tobacco smoke (IARC 2010b); for example, wood burning in the home is highly correlated with levels of polycyclic aromatic hydrocarbons (PAHs), benzene, and 1,3-butadiene, as well as other compounds (Gustafson et al. 2007; Gustafson et al. 2008). Wood burning in an open fireplace produces PAHs at levels that are comparable to those in ambient urban air (Alfheim and Ramdahl 1984). Only one prior report, a retrospective case–control study, evaluated indoor wood-burning stove/fireplace use as a breast cancer risk factor (White et al. 2014). The study observed a positive association with breast cancer for women who reported burning artificial or synthetic fire-logs (White et al. 2014).

In this large prospective cohort study, we considered measures of residential indoor heating and cooking in association with breast cancer risk. We hypothesized that using an indoor wood-burning stove/fireplace would be associated with a higher breast cancer risk and that the risk may vary with material burned and frequency of use. Indoor wood-burning stove/fireplace use is a potentially modifiable breast cancer risk factor, and better understanding of the role played by indoor air pollution in breast carcinogenesis may inform public health strategies to reduce breast cancer risk.

Methods

Study Population

The Sister Study is a prospective volunteer-based cohort study that was designed to evaluate environmental risk factors for breast cancer. Women were recruited for the study from 2003–2009 using a multimedia campaign and a network of breast cancer professionals and advocates. Eligibility criteria included having no personal history of breast cancer, living in the United States or Puerto Rico, being between 35–74 y of age, and having a sister who had been previously diagnosed with breast cancer. At baseline, Sister Study participants completed an extensive telephone questionnaire covering demographics, lifestyle factors, medical and family history, and residential history, including questions on indoor heating and cooking at both baseline and their longest adult residence.

This research was approved by the Institutional Review Boards of the National Institute of Environmental Health Sciences (NIEHS), the National Institutes of Health (NIH), and the Copernicus Group. Written informed consent was obtained from all participants. This study includes breast cancer cases that were diagnosed before July 1, 2014 (Sister Study Data Release 4.1).

Outcome Assessment

Study participants complete annual health updates to notify the study about any changes in health as well as additional comprehensive questionnaires every 2–3 y. Participation rates have been >90% throughout the follow-up period (NIEHS 2016).

Medical records are used to confirm the breast cancer diagnosis and have been successfully obtained for ∼80% of cases. The agreement between self-reported tumor characteristics and information abstracted from the medical record is high, and therefore, self-reported data are used when medical record data are not available (D’Aloisio et al., unpublished data, 2016).

Exposure and Covariate Assessment

As part of the residential history questionnaire given at baseline, study participants were asked for details on their indoor heating and cooking sources for both their baseline residence and their longest adult residence (since the age of 20). For both the baseline and the longest adult residence, information on the age starting and stopping living in the home was collected. Women were asked whether there was an indoor wood-burning stove or a fireplace in their home (yes, no) and the frequency of use (average number of times per year). Frequency of indoor stove/fireplace use was collapsed to less than once per month, 1–4 times per month, and at least once per week. Women who reported having an indoor wood-burning stove/fireplace in their residence but estimated zero uses per year (11%) were excluded from the frequency-of-use analysis. The questionnaire also asked about the type of material burned in the indoor stove/fireplace (wood, coal, natural gas or propane, or artificial logs). For material burned, participants could select multiple fuel types. Study participants also answered questions on the main source of heating in the home (electricity, natural gas, fuel oil, propane, wood, or other) and the energy source for the cooking stove top or top range (electricity, gas or natural gas, or other). See Supplemental Material, “Sister Study Baseline Residential Questionnaire,” for specific wording of the questions.

Covariates of potential interest, including demographics, reproductive history, cigarette smoking, and use of exogenous hormones, were obtained from the baseline interview. A trained examiner measured height and weight at the baseline home visit; measurements were used to determine body mass index (BMI; kilograms/square meter).

Statistical Analysis

To evaluate the association between indoor heating and cooking and breast cancer risk, multivariable Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Follow-up was accrued from age at baseline to age at breast cancer diagnosis or censoring (defined as the age of last follow-up).

Overall breast cancer (n=2,416), invasive breast cancer (n=1,843), and ductal carcinoma in situ (DCIS) (n=533) were considered as outcomes of interest. We also evaluated the association between breast cancer and indoor wood-burning stove and fireplace use when breast cancer was defined using estrogen receptor status (ER+,ER-) and menopausal status at diagnosis (premenopausal, postmenopausal) as secondary outcomes. When evaluating by ER subtype or invasive/DCIS status, cases without the breast cancer subtype of interest were censored at the time of diagnosis. ER analyses were restricted to women with invasive breast cancer because ER status is less frequently reported for in situ disease. A case–case analysis was used to test whether effect estimates differed by subtype or invasive/DCIS status (Begg and Zhang 1994). When evaluating breast cancer risk by menopausal status at diagnosis, women who became postmenopausal during the follow-up period were censored for premenopausal breast cancer at age of menopause and then became at risk for postmenopausal breast cancer. The assumption of proportional hazards for the Cox model was evaluated visually using log–log survival plots and via an interaction term in the model between each covariate and survival time (using an α=0.05). There was no evidence of time-variant associations.

Effect measure modification of the relationship between breast cancer and indoor wood-burning stove/fireplace use by geographic region (northeast, west, south, midwest), smoking status (never smoker, ever smoker), education (≤high school graduate, some college, 4-y college degree, graduate degree), and number of first-degree relatives with breast cancer was tested. Confounders were identified using a directed acyclic graph (Greenland et al. 1999). Multivariable-adjusted models included race (non-Hispanic white, other), education (<high school degree, completed high school or General Education diploma (GED), some college but no degree, associate’s or technical degree, bachelor’s degree, Master’s degree, doctoral degree), marital status (never married, legally married or living as married, widowed or divorced or separated), annual household income (<20,000 USD, 20,000–49,999 USD, 50,000–99,999 USD, 100,000–199,999 USD, ≥200,000 USD), parity (nulliparous, parous), use of oral contraceptives (ever, never), hormone replacement therapy use at enrollment (none, estrogen only, estrogen and progesterone combined or both estrogen and estrogen and progesterone combined), age at menopause (premenopausal, <40 y, 40–50 y, 51–55 y, >55 y based on enrollment information), and BMI (<18.5 kg/m2, 18.5–24.9 kg/m2, 25.0–29.9 kg/m2, ≥30 kg/m2). All confounders were modeled as categorical variables using the characterizations described above.

For sensitivity analyses, we restricted to women whose longest adult residence was also their baseline residence. Two-sided tests were used with a p-value of 0.05 to evaluate statistical significance. All analyses were performed using SAS software (version 9.3; SAS Institute Inc.).

Results

Having an indoor wood-burning stove or fireplace was common in our study population, with over half of the participants reporting having one at their longest adult residence. Women in the Sister Study who had a wood-burning stove/fireplace in their home were more likely to be non-Hispanic white, to have higher educational attainment, and to have higher annual household income (Table 1). Indoor wood-burning stove/fireplace users were also more likely to use oral contraceptives and postmenopausal hormones.

Table 1. Study population characteristics at baseline by presence of an indoor wood-burning stove/fireplace at longest adult residence, National Institute of Environmental Health Sciences (NIEHS) Sister Study.
Characteristic No indoor wood-burning stove/fireplace n=18,017 Indoor wood-burning stove/fireplace n=29,495
Continuous, mean (standard deviation)
 Age at baseline 54.6 (9.2) 55.7 (8.7)
 Age at menopausea 47.5 (6.7) 48.6 (6.2)
 Body mass index (kg/m2) 28.8 (6.8) 27.2 (5.9)
Categorical, n (%)
Parity
 Parous 14,220 (79.0) 24,567 (83.3)
Smoking status
 Ever 8,244 (45.8) 12,551 (42.6)
Race
 Non-Hispanic white 13,543 (75.2) 26,274 (89.1)
Education
 Less than high school degree 412 (2.3) 167 (0.6)
 High school degree or equivalent 3,136 (17.4) 3,440 (11.7)
 Some college, no degree 3,949 (21.9) 5,308 (18.0)
 Associate’s degree 2,827 (15.7) 3,903 (13.2)
 4-y degree 4,248 (23.6) 8,667 (29.4)
 Master’s degree 2,884 (16.0) 6,592 (22.4)
 Doctoral degree 558 (3.1) 1,414 (4.8)
Marital status
 Never married 1,611 (8.9) 972 (3.3)
 Legally married or living as married 11,800 (65.5) 23,765 (80.6)
 Separated, divorced or widowed 4,603 (25.6) 4,752 (16.1)
Annual household income
 <20,000 USD 1,443 (8.3) 666 (2.4)
 20,000–49,999 USD 5,060 (29.0) 4,372 (15.5)
 50,000–99,999 USD 7,299 (41.8) 11,343 (40.2)
 100,000–199,999 USD 3,112 (17.8) 9,031 (32.0)
 ≥200,000 USD 535 (3.1) 2,833 (10.0)
Geographic location
 Northeast 3,605 (21.3) 5,242 (17.8)
 Midwest 5,470 (32.4) 7,864 (26.7)
 South 5,260 (31.1) 8,874 (30.2)
 West 2,572 (15.2) 7,427 (25.3)
Use of oral contraceptives
 Ever 14,726 (81.8) 25,342 (86.0)
Postmenopausal hormone usea
 None 4,636 (42.9) 6,897 (36.4)
 Estrogen only (E) 3,083 (28.5) 5,285 (27.9)
 Estrogen and Progesterone (E+P) or E and E+P 3,093 (28.6) 6,747 (35.6)

Note: Chi-squared p-value for all study characteristics p<0.0001. USD, U.S. dollar.

aLimited to those who were postmenopausal at baseline (n=32,457).

There were 2,416 breast cancers diagnosed during the follow-up period (mean=6.4 y), 1,843 of which were determined to be invasive. Having a wood-burning stove/fireplace in the longest adult residence was associated with slightly increased breast cancer risk [Number of exposed cases (n)=1,500; HR=1.11 (95% CI: 1.01, 1.22)] (Table 2). There was a trend of higher breast cancer risk with increased frequency of use (p=0.01); those who used an indoor wood-burning stove/fireplace at least once a week had the highest HR [n=327; HR=1.17 (95% CI: 1.02, 1.34)] relative to those who did not have a wood-burning stove/fireplace in their longest adult residence. Although the HR for invasive cases did not differ significantly from in situ cases in a direct comparison, the estimate for weekly wood-burning stove/fireplace use was apparent for invasive breast cancer [n=265; HR=1.25 (95% CI: 1.07, 1.46)] but not for DCIS [n=59; HR=0.96 (95% CI: 0.70, 1.31)] (invasive vs. DCIS, p=0.7).

Table 2. Indoor heating/cooking at longest adult residence and breast cancer, National Institute of Environmental Health Sciences (NIEHS) Sister Study.
Multivariable-adjusteda
Indoor air pollution at longest adult residence Person-years (n=362,242) All breast cancer (n=2,416) Invasive breast cancer (n=1,843) DCIS (n=533) Age-adjusted Overall HR (95% CI) Overall HR (95% CI) Invasive HR (95% CI) DCIS HR (95% CI)
Indoor wood-burning stove/fireplace
 No stove/fireplace 113,647 754 567 170 1.00 (Referent) 1.00 (Referent) 1.00 (Referent) 1.00 (Referent)
 Yes 192,317 1500 1151 329 1.15 (1.05, 1.25)* 1.11 (1.01, 1.22)* 1.13 (1.02, 1.26)* 1.07 (0.87, 1.31)
Indoor wood-burning stove/fireplace frequency of use
 No stove/fireplace 113,647 754 567 170 1.00 (Referent) 1.00 (Referent) 1.00 (Referent) 1.00 (Referent)
 Less than once/month 59,782 449 340 102 1.11 (0.99, 1.25) 1.06 (0.93, 1.20) 1.07 (0.93, 1.24) 1.01 (0.77, 1.32)
 1–4 times/mo 53,920 418 307 103 1.15 (1.02, 1.30)* 1.13 (0.99, 1.28) 1.11 (0.95, 1.28) 1.22 (0.94, 1.58)
 at least once a week 40,856 327 265 59 1.18 (1.03, 1.34)* 1.17 (1.02, 1.34)* 1.25 (1.07, 1.46)* 0.96 (0.70, 1.31)
Indoor wood-burning stove/fireplace fuelb
 No stove/fireplace 113,647 754 567 170 1.00 (Referent) 1.00 (Referent) 1.00 (Referent) 1.00 (Referent)
 Wood 120,549 924 718 190 1.12 (1.02, 1.24)* 1.09 (0.98, 1.21) 1.13 (1.00, 1.27) 0.97 (0.77, 1.21)
 Gas 45,578 353 271 79 1.16 (1.03, 1.32)* 1.15 (1.00, 1.32)* 1.19 (1.02, 1.39)* 1.07 (0.81, 1.43)
 Artificial logs 49,737 339 245 87 1.02 (0.90, 1.16) 0.98 (0.85, 1.12) 0.94 (0.80, 1.10) 1.08 (0.82, 1.42)
Energy source for the cooking stove top
 Electricity 185,435 1397 1061 316 1.00 (Referent) 1.00 (Referent) 1.00 (Referent) 1.00 (Referent)
 Gas 107,850 794 605 174 1.00 (0.91, 1.09) 0.99 (0.90, 1.08) 1.00 (0.90, 1.10) 0.94 (0.78, 1.14)
 Other 32,956 225 177 43 0.86 (0.74, 0.99)* 0.86 (0.75, 1.00)* 0.90 (0.76, 1.06) 0.73 (0.52, 1.02)
Main source of heating
 Electricity 75,395 523 395 121 1.00 (Referent) 1.00 (Referent) 1.00 (Referent) 1.00 (Referent)
 Gas 170,459 1302 987 295 1.07 (0.97, 1.19) 1.09 (0.98, 1.21) 1.09 (0.97, 1.23) 1.07 (0.86, 1.33)
 Fuel oil 30,457 249 193 50 1.11 (0.96, 1.29) 1.13 (0.97, 1.32) 1.14 (0.95, 1.37) 1.04 (0.74, 1.46)
 Propane 12,293 66 53 12 0.75 (0.58, 0.97)* 0.83 (0.64, 1.07) 0.89 (0.66, 1.18) 0.62 (0.33, 1.15)
 Wood 7,457 54 47 6 1.03 (0.78, 1.36) 1.09 (0.82, 1.45) 1.27 (0.93, 1.72) 0.55 (0.24, 1.26)
 Other 5,952 37 26 9 0.90 (0.64, 1.25) 0.90 (0.63, 1.27) 0.87 (0.58, 1.31) 0.90 (0.44, 1.85)

Note: CI, confidence interval; DCIS, ductal carcinoma in situ; HR, hazard ratio

aAdjusted for age, race, education, income, marital status, parity, use of hormonal birth control, use of postmenopausal hormones, age at menopause and menopausal status, and body mass index (BMI).

bFuel types are not mutually exclusive.

*p<0.05.

An increased risk of breast cancer was observed for women who burned wood (n=924 HR=1.09 (95% CI: 0.98, 1.21)] or natural gas/propane [n=353 HR=1.15 (95% CI 1.00, 1.32)] in their wood-burning stove/fireplace compared with those without a wood-burning stove/fireplace in their home. There was no evidence of an increase in risk for burning artificial fire-logs [n=339; HR=0.98 (95% CI: 0.85, 1.12)]. Although there was not a significant interaction for any wood-burning stove/fireplace by geographic region (p for interaction=0.1), there was some suggestion of variability in the association of any wood-burning stove/fireplace use and breast cancer risk by geographic region (see Table S1). For example, the association with having an indoor wood-burning stove/fireplace was most pronounced in women residing in western states [n=404; HR=1.32 (95% CI: 1.05, 1.66)], and this elevated association was observed for all materials burned including artificial fire logs [wood, n=247; HR=1.30 (95% CI: 1.02, 1.67); gas, n=91; HR=1.30 (95% CI: 0.96, 1.76); artificial logs, n=111; HR=1.36 (95% CI: 1.02, 1.81)]. Frequency of fireplace use was highest in women living in the west (mean=40 times/y) or northeast (mean=41 times/y) compared with those living in the south (mean=24.9 times/y) or midwest (mean=34.7 times/y).

There were slight increases in breast cancer risk observed for gas [n=1,302; HR=1.09 (95% CI: 0.98, 1.21)] or fuel oil [n=249; HR=1.13 (95% CI: 0.97, 1.32)] as the main heating sources in the home compared with electricity (Table 2). Although few women reported wood as the primary heating source of the home, there was suggestive evidence of elevated invasive breast cancer risk for wood heating relative to electricity use [n=47; HR=1.27 (95% CI: 0.93, 1.73)]. Estimates were similar when restricting to women whose longest residence was also their baseline residence (see Table S2). Estimates were attenuated toward the null when we considered indoor wood-burning stove/fireplace and heating/cooking exposure information from the baseline residence compared with estimates for exposure at the longest adult residence (see Table S3).

There was little evidence that the association between indoor wood-burning stove/fireplace use and breast cancer differed with increasing years lived in the longest residence [<10 y, n=255; HR=1.09 (95% CI: 0.89, 1.35); ≥10 y, n=1,245; HR=1.12 (95%CI: 1.00, 1.24)] (Table 3), although the estimate was statistically significant only among those who reported living ≥10 y in the longest adult residence. Having an indoor wood-burning stove/fireplace in the longest adult residence was associated with both invasive ER+ [n=891 ; HR=1.16 (95%CI 1.02, 1.31)] and invasive ER– [n=165; HR=1.30 (95% CI 0.96, 1.74)] tumor status (Table 4). The HRs did not vary by menopausal status at diagnosis.

Table 3. Indoor wood-burning stove/fireplace at longest adult residence and overall breast cancer risk by years lived at residence, National Institute of Environmental Health Sciences (NIEHS) Sister Study.
Residence duration
<10 y ≥10 y
Indoor wood-burning stove/fireplace in longest adult residence n HR (95% CI)a n HR (95% CI)a
Indoor wood-burning stove/fireplace
 No stove/fireplace 164 1.00 (Referent) 589 1.00 (Referent)
 Yes 255 1.09 (0.89, 1.35) 1,245 1.12 (1.00, 1.24)*
Indoor wood-burning stove/fireplace fuel
 No indoor stove/fireplace 164 1.00 (Referent) 589 1.00 (Referent)
 Wood 154 1.13 (0.89, 1.42) 770 1.08 (0.96, 1.22)
 Gas 79 1.16 (0.86, 1.55) 274 1.15 (0.98, 1.34)
 Artificial logs 62 0.86 (0.63, 1.17) 277 1.01 (0.86, 1.17)

Note: CI, confidence interval; HR, hazard ratio.

aAdjusted for age, race, education, income, marital status, parity, use of hormonal birth control, use of postmenopausal hormones, age at menopause and menopausal status, and body mass index (BMI).

*p<0.05.

Table 4. Indoor wood-burning stove/fireplace at longest adult residence and breast cancer tumor characteristics, National Institute of Environmental Health Sciences (NIEHS) Sister Study.
No indoor wood-burning stove/fireplace Indoor wood-burning stove/fireplace
Tumor characteristics n HR (95% CI)a n HR (95% CI)a
Estrogen receptor (ER) statusb
 ER+ 416 1.00 (Referent) 891 1.16 (1.02, 1.31)*
 ER− 75 1.00 (Referent) 165 1.30 (0.96, 1.74)
Menopausal status at diagnosis
 Premenopausal 180 1.00 (Referent) 315 1.09 (0.90, 1.33)
 Postmenopausal 569 1.00 (Referent) 1,180 1.10 (0.99, 1.23)

Note: CI, confidence interval; HR, hazard ratio.

aEstrogen receptor analyses and postmenopausal models adjusted for age, race, education, income, marital status, parity, use of hormonal birth control, use of postmenopausal hormones, age at menopause and menopausal status, and body mass index (BMI). Premenopausal models adjusted for age, race, education, income, marital status, parity, use of hormonal birth control, use of postmenopausal hormones, and BMI.

bEstrogen receptor status analyses limited to invasive cases.

*p<0.05.

There was no evidence to suggest that the association between indoor wood-burning stove/fireplace use varied by number of first-degree relatives with breast cancer (see Table S4) or by smoking history (see Table S5). Similarly, there was no evidence of a significant interaction by education (see Table S6).

Discussion

In this large, prospective cohort of women with a family history of breast cancer, those who used an indoor wood-burning stove or fireplace at their longest adult residence were at a higher risk of developing breast cancer. The risk of breast cancer increased with more frequent use, and the association varied based on the material burned, with both wood and natural gas/propane being associated with an elevated risk. There were also modest associations observed for reporting that gas, fuel oil, or wood was the main source of heating at the longest adult residence relative to electricity. This is the first prospective study to consider measures of indoor heating/cooking in association with breast cancer risk.

The association between indoor air pollution and breast cancer is biologically plausible. Use of an open fireplace has been associated with higher DNA adduct levels (Pedersen et al. 2009), which have been related to breast carcinogenesis (Gammon et al. 2004). Although gas fireplaces are thought to produce less air pollution (U.S. EPA 2016), gas fireplaces produce PAHs, nitrogen dioxide, and carbon monoxide (Dutton et al. 2001). Burning wood releases numerous compounds, including PAHs, 1,3-butadiene, polychlorinated dibenzodioxins and dibenzofurans (PCDDs/Fs), polychlorinated biphenyls (PCBs), hexachlorobenze (HxCBz), and particulate matter, among others (Gullett et al. 2003; Gustafson et al. 2007; McDonald et al. 2000; Rogge et al. 1998). Once PAHs are inhaled, they can be rapidly absorbed and can eventually accumulate in the breast (IARC 2010a). PAHs are established carcinogens and can bind to DNA to form bulky adducts in the breast tissue that, if not sufficiently repaired, can lead to somatic mutations (IARC 2010a). Another potentially relevant biologic mechanism is that of epigenetic modification. PAH exposure sources have been associated with aberrant DNA methylation of breast cancer–related genes, which can result in altered expression patterns that promote carcinogenesis (White et al. 2016).

The pollutants released from indoor heating and cooking methods are estimated to be at a concentration higher than that of environmental tobacco smoke but lower than active smoking (Smith and Peel 2010). Previous studies have reported modest, but positive, associations between breast cancer and both active smoking and environmental tobacco smoke (DHHS 2014). Similarly, the growing body of literature on the association between outdoor air pollution measures and breast cancer supports the findings observed in this study. Most studies have reported an increased risk of breast cancer in association with markers of higher traffic-related air pollution, including nitrogen dioxide and PAHs (Crouse et al. 2010; Mordukhovich et al. 2016; Reding et al. 2015). Two studies of early-life exposure to total suspended particles, a proxy for PAH exposure, also reported positive associations with later breast cancer risk (Bonner et al. 2005; Nie et al. 2007). Outdoor air pollution is a source of indoor air pollution and thus could potentially interact with indoor air pollution from indoor wood-burning stoves/fireplaces to influence breast cancer risk. Thus, any geographic variability in the constituents of outdoor air pollution could potentially have contributed to any geographic differences in the estimates reported in this study.

The findings reported here conflict somewhat with those of the Long Island Breast Cancer Study Project (LIBCSP), a retrospective case–control study of women on Long Island, New York (n=1,508 cases and n=1,556 controls). In the LIBCSP, there was an increase in odds of breast cancer observed in women who reported burning synthetic logs, whereas no increase in risk was observed for women who reported burning wood or gas (White et al. 2014). The discrepancies between these two studies may be due to a number of factors. First, women in the LIBCSP were geographically constrained to Long Island, NY based on enrollment criteria, whereas women in the Sister Study population could have lived anywhere in the United States or Puerto Rico. To address this specific difference in these study populations, we conducted an analysis stratifying by geographic region. When restricted to Sister Study participants living in the northeast, burning wood or gas in wood-burning stoves/fireplaces was not associated with breast cancer risk, similar to the LIBCSP findings. However, we did not see a similar association with use of artificial fire-logs in this subgroup.

The women in the LIBCSP were diagnosed between 1996 and 1997 (Gammon et al. 2002), whereas the women in our study population were diagnosed after their enrollment in the study between 2003 and 2009. Changes in fireplace construction and/or indoor ventilation may have led to variation in indoor air emissions from indoor wood-burning stoves/fireplaces over time (Houck et al. 1998), and construction may vary geographically along with frequency of wood-burning stove/fireplace use. The composition of synthetic fire-logs has also changed over time (Li and Rosenthal 2006), which may partially explain some of the differences in the findings. Finally, the LIBCSP study did not consider frequency of indoor wood-burning stove/fireplace use and thus was not able to identify those with higher levels of exposure and could not rule out the potential for recall bias.

In addition to observing an increase in risk for burning wood or natural gas in an indoor wood-burning stove/fireplace, we found similar associations for wood, gas, and fuel oil as the main heating sources relative to electricity. The association observed with wood as fuel for the stove/fireplace and as the main heating source was only evident for invasive tumors. Invasive breast cancer and in situ disease may have distinct etiologies and thus different risk factors (Kerlikowske et al. 1997).

We saw little evidence that associations with wood-burning stove/fireplace use were stronger with increasing years lived at the longest adult residence. However, we did observe that associations were attenuated for indoor heating/cooking measures at the baseline residence compared with the longest adult residence. This finding suggests that there is some importance of increasing duration of exposure to indoor air pollution with respect to breast cancer. There was an inverse association observed with other energy sources for a cooking-stove top; it is unclear why this would be the case, but estimates were imprecise.

Although the test of effect measure modification was not statistically significant, there was some evidence of variability in the association between wood-burning stove/fireplace use and breast cancer by geographic region. In particular, estimates were most notable in women who were living in the western United States. It is unclear why this would be the case; the frequency of wood-burning stove/fireplace use in women who lived in western states was similar to the frequency of use in women who lived in the northeast, where no increase in risk was observed. A limitation of this study is that we did not capture differences in frequency of use by seasons, which is highly likely to vary by geographic region. A report from the U.S. EPA suggested that fireplaces in the west were notably larger and thus would produce more emissions (Houck et al. 1998), which may explain, possibly in conjunction with differences in age of home and ventilation, the increase in risk observed for women in the west, if this effect is real.

The information on indoor heating and cooking is self-reported. Our questionnaire was limited to the participant’s residence at study baseline as well as their longest adult residence, which may have helped to maximize recall. Any recall error would not have been differential by case status because information on fireplace use was collected before breast cancer diagnosis. However, because we did not collect information on residences other than longest adult and baseline residence, we were unable to consider exposure information for some women who may have used wood-burning indoor stoves/fireplaces at other residences, and we did not capture information on ventilation or other factors that may affect level of exposure from fireplaces and indoor wood-burning stoves. In our study, although the HR for invasive cancer was greater for those who used fireplaces at least once a week, we did not have the sample size to distinguish very heavy users (e.g., those who used them every day) from more occasional users.

The prospective design of this study is an important strength. Additionally, the inclusion of both frequency of use and material burned are strengths of this analysis. However, the questionnaire combined indoor wood-burning stoves and fireplaces into a single question; thus, we were unable to differentiate between the two. This is a limitation because sealed indoor stoves may release reduced emissions compared with open fireplaces (Gullett et al. 2003; McDonald et al. 2000), and use of wood-burning stoves versus fireplaces may vary geographically as well as by other participant characteristics such as socioeconomic status. Indoor wood-burning stove/fireplace use was more common in women who had a higher income or educational attainment. Higher socioeconomic status is an established risk factor for breast cancer (Yost et al. 2001). Although we were able to adjust for detailed education and income information, as well as for marital status, we cannot rule out the possibility that there was some residual confounding by socioeconomic status that we were unable to resolve.

The women in this study population have a family history of breast cancer. To evaluate whether family history modified the association between wood-burning stove/fireplace use and breast cancer, we considered degree of family history as an effect measure modifier. There was no evidence that the association varied based on the number of first-degree family members with breast cancer. Although we cannot be sure that these results are generalizable to all women, we do expect women in this study population to have similar risk factor distributions to the general population (Weinberg et al. 2007).

Conclusions

In conclusion, this is the first prospective study to examine the association between indoor stove/fireplace use and breast cancer risk. We found that using an indoor wood-burning stove/fireplace was associated with a higher risk of breast cancer in a study population of women with a family history of breast cancer. The association with indoor wood-burning stove/fireplace use increased with both frequency and years of use. Indoor air pollution is of worldwide concern, particularly in areas where burning of wood and other materials for both heating and cooking is common; in our study population, over half of the women reported using an indoor wood-burning stove or fireplace. Despite the modest 10–15% increase in risk observed in this study, the high prevalence of indoor wood-burning stove and fireplace use as well as the continued high incidence of breast cancer suggest that these findings could have substantial public health impact. Indoor air pollution from indoor wood-burning stoves/fireplaces is a widespread and potentially modifiable breast cancer risk factor.

Acknowledgments

This research was supported by the Intramural Research Program of the NIH, National Institute of Environmental Health Sciences (Z01-ES044005).

References

Alfheim I, Ramdahl T. 1984. Contribution of wood combustion to indoor air pollution as measured by mutagenicity in Salmonella and polycyclic aromatic hydrocarbon concentration. Environ Mutagen 6(2):121–130, PMID: 6368216.

Begg CB, Zhang ZF. 1994. Statistical analysis of molecular epidemiology studies employing case-series. Cancer Epidemiol Biomarkers Prev 3(2):173–175, PMID: 8049640.

Bonjour S, Adair-Rohani H, Wolf J, Bruce NG, Mehta S, Prüss-Ustün A, et al. 2013. Solid fuel use for household cooking: Country and regional estimates for 1980–2010. Environ Health Perspect 121(7):784–790, 10.1289/ehp.1205987.

Bonner MR, Han D, Nie J, Rogerson P, Vena JE, Muti P, et al. 2005. Breast cancer risk and exposure in early life to polycyclic aromatic hydrocarbons using total suspended particulates as a proxy measure. Cancer Epidemiol Biomarkers Prev 14(1):53–60, PMID: 15668476.

Crouse DL, Goldberg MS, Ross NA, Chen H, Labrèche F. 2010. Postmenopausal breast cancer is associated with exposure to traffic-related air pollution in Montreal, Canada: a case-control study. Environ Health Perspect 118(11):1578–1583, PMID: 20923746, 10.1289/ehp.1002221.

DHHS (Department of Health & Human Services). 2014. The health consequences of smoking—50 years of progress: A report of the Surgeon General. Atlanta, GA: US Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health. https://www.surgeongeneral.gov/library/reports/50-years-of-progress/full-report.pdf [accessed 4 December 2016].

Dutton SJ, Hannigan MP, Miller SL. 2001. Indoor pollutant levels from the use of unvented natural gas fireplaces in Boulder, Colorado. J Air Waste Manag Assoc 51(12):1654–1661, PMID: 15666470.

Gammon MD, Neugut AI, Santella RM, Teitelbaum SL, Britton JA, Terry MB, et al. 2002. The Long Island Breast Cancer Study Project: Description of a multi-institutional collaboration to identify environmental risk factors for breast cancer. Breast Cancer Res Treat 74(3):235–254, 10.1023/A:1016387020854.

Gammon MD, Sagiv SK, Eng SM, Shantakumar S, Gaudet MM, Teitelbaum SL, et al. 2004. Polycyclic aromatic hydrocarbon-DNA adducts and breast cancer: a pooled analysis. Arch Environ Health 59(12):640–649, PMID: 16789472, 10.1080/00039890409602948.

Greenland S, Pearl J, Robins JM. 1999. Causal diagrams for epidemiologic research. Epidemiology: 37–48, 10.1097/00001648-199901000-00008.

Gullett BK, Touati A, Hays MD. 2003. PCDD/F, PCB, HxCBz, PAH, and PM emission factors for fireplace and woodstove combustion in the San Francisco Bay region. Environ Sci Technol 37(9):1758–1765, PMID: 12775046.

Gustafson P, Barregard L, Strandberg B, Sällsten G. 2007. The impact of domestic wood burning on personal, indoor and outdoor levels of 1, 3-butadiene, benzene, formaldehyde and acetaldehyde. J Environ Monit 9(1):23–32, PMID: 17213939, 10.1039/b614142k.

Gustafson P, Ostman C, Sällsten G. 2008. Indoor levels of polycyclic aromatic hydrocarbons in homes with or without wood burning for heating. Environ Sci Technol 42(14):5074–5080, PMID: 18754350.

Hosgood HD 3rd, Boffetta P, Greenland S, Lee YC, McLaughlin J, Seow A, et al. 2010. In-home coal and wood use and lung cancer risk: A pooled analysis of the International Lung Cancer Consortium. Environ Health Perspect 118(12):1743, 10.1289/ehp.1002217.

Houck JE, Tiegs PE, McCrillis RC. 1998. Residential Wood Combustion Technology Review Volume 1. Technical Report. Prepared for the United States Environmental Protection Agency, Office of Research and Development.

IARC (International Agency for Research on Cancer). 2010a. Household use of solid fuels and high-temperature frying. IARC. 2010. Monog Eval Carcinog Risk Hum 95:7–430. http://www.who.int/indoorair/guidelines/hhfc/IARC_monographs_Vol.95.pdf [accessed 4 December 2016].

IARC. 2010b. Some non-heterocyclic polycyclic aromatic hydrocarbons and some related exposures. IARC. 2010. Monog Eval Carcinog Risk Hum 92:35–818. https://monographs.iarc.fr/ENG/Monographs/vol92/mono92.pdf [accessed 4 December 2016].

Kerlikowske K, Barclay J, Grady D, Sickles EA, Ernster V. 1997. Comparison of risk factors for ductal carcinoma in situ and invasive breast cancer. J Natl Cancer Inst 89(1):76–82, PMID: 8978410.

Li VS, Rosenthal S. 2006. Content and emission characteristics of Artificial Wax Firelogs. https://www3.epa.gov/ttnchie1/conference/ei15/poster/li.pdf [accessed 4 December 2016].

Lissowska J, Bardin-Mikolajczak A, Fletcher T, Zaridze D, Szeszenia-Dabrowska N, Rudnai P, et al. 2005. Lung cancer and indoor pollution from heating and cooking with solid fuels: The IARC International Multicentre Case–Control Study in Eastern/Central Europe and the United Kingdom. Am J Epidemiol 162(4):326–333, PMID: 16014775, 10.1093/aje/kwi204.

McDonald JD, Zielinska B, Fujita EM, Sagebiel JC, Chow JC, Watson JG. 2000. Fine particle and gaseous emission rates from residential wood combustion. Environ Sci Technol 34(11):2080–2091, 10.1021/es9909632.

Mordukhovich I, Beyea J, Herring AH, Hatch M, Stellman SD, Teitelbaum SL, et al. 2016. Vehicular traffic-related polycyclic aromatic hydrocarbon exposure and breast cancer incidence: The Long Island Breast Cancer Study Project (LIBCSP). Environ Health Perspect 124(1):30–38.

Nie J, Beyea J, Bonner MR, Han D, Vena JE, Rogerson P, et al. 2007. Exposure to traffic emissions throughout life and risk of breast cancer: the Western New York Exposures and Breast Cancer (WEB) study. Cancer Causes Control 18(9):947–955, PMID: 17632764, 10.1007/s10552-007-9036-2.

NIEHS (National Institute of Environmental Health Sciences.) 2016. “Sister Study Response Rates for Annual and Detailed Follow-Up.” https://sisterstudy.niehs.nih.gov/English/images/SIS-RespRatesFollowUps-website-20160208-508.pdf [accessed 4 December 2016].

Pedersen M, Wichmann J, Autrup H, Dang DA, Decordier I, Hvidberg M, et al. 2009. Increased micronuclei and bulky DNA adducts in cord blood after maternal exposures to traffic-related air pollution. Environ Res 109(8):1012–1020, PMID: 19783246, 10.1016/j.envres.2009.08.011.

Reding KW, Young MT, Szpiro AA, Han CJ, DeRoo LA, Weinberg C, et al. 2015. Breast cancer risk in relation to ambient air pollution exposure at residences in the Sister Study cohort. Cancer Epidemiol Biomarkers Prev 24(12):1907–1909, PMID: 26464427, 10.1158/1055-9965.EPI-15-0787.

Reid BC, Ghazarian AA, DeMarini DM, Sapkota A, Jack D, Lan Q, et al. 2012. Research opportunities for cancer associated with indoor air pollution from solid-fuel combustion. Environ Health Perspect 120(11):1495–1498, PMID: 22846419, 10.1289/ehp.1204962.

Rogge WF, Hildemann LM, Mazurek MA, Cass GR, Simoneit BR. 1998. Sources of fine organic aerosol. 9. Pine, oak, and synthetic log combustion in residential fireplaces. Environ Sci Technol 32(1):13–22, 10.1021/es960930b.

Sapkota A, Gajalakshmi V, Jetly DH, Roychowdhury S, Dikshit RP, Brennan P, et al. 2008. Indoor air pollution from solid fuels and risk of hypopharyngeal/laryngeal and lung cancers: a multicentric case–control study from India. Int J Epidemiol 37(2):321–328, PMID: 18234740, 10.1093/ije/dym261.

Smith KR, Peel JL. 2010. Mind the gap. Environ Health Perspect 118(12):1643, PMID: 20729177, 10.1289/ehp.1002517.

U.S. EPA (U.S. Environmental Protection Agency). 2016. Burn Wise Best Burn Practices. www.epa.gov/burnwise/burn-wise-best-burn-practices [accessed 4 December 2016].

Weinberg CR, Shore DL, Umbach DM, Sandler DP. 2007. Using risk-based sampling to enrich cohorts for endpoints, genes, and exposures. Am J Epidemiol 166(4):447–455, PMID: 17556763, 10.1093/aje/kwm097.

White AJ, Chen J, Teitelbaum SL, McCullough LE, Xu X, Hee Cho Y, et al. 2016. Sources of polycyclic aromatic hydrocarbons are associated with gene-specific promoter methylation in women with breast cancer. Environ Res 145:93–100, PMID: 26671626, 10.1016/j.envres.2015.11.033.

White AJ, Teitelbaum SL, Stellman SD, Beyea J, Steck SE, Mordukhovich I, et al. 2014. Indoor air pollution exposure from use of indoor stoves and fireplaces in association with breast cancer: a case-control study. Environ Health 13:108, PMID: 25495350, 10.1186/1476-069X-13-108.

Yost K, Perkins C, Cohen R, Morris C, Wright W. 2001. Socioeconomic status and breast cancer incidence in California for different race/ethnic groups. Cancer Causes Control 12(8):703–711, PMID: 11562110.

Pesticide Exposure and Risk of Rheumatoid Arthritis among Licensed Male Pesticide Applicators in the Agricultural Health Study

Author Affiliations open

1Occupational and Environmental Health Branch, Public Health Institute, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil

2Epidemiology Branch, National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, North Carolina, USA

3Occupational and Environmental Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA

PDF icon PDF Version (161 KB)

  • Background:
    The occupation of farming has been associated with rheumatoid arthritis (RA); pesticides may account for this association, but there are few studies.
    Objectives:
    We investigated associations between RA and use of pesticides in the Agricultural Health Study.
    Methods:
    The study sample was drawn from male pesticide applicators enrolled in 1993–1997 who provided questionnaire data at baseline and at least once during follow-up (over a median 18 y; interquartile range 16–19). Incident RA cases (n=220), confirmed by physicians or by self-reported use of disease-modifying antirheumatic drugs, were compared with noncases (n=26,134) who did not report RA. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using logistic regression, adjusting for enrollment age, state, smoking pack-years, and education. We evaluated the association of RA with the use of 46 pesticides and across 4 levels (never use and tertiles) of lifetime days of use for 16 pesticides with OR≥1.2 for ever use.
    Results:
    Incident RA was associated with ever use of fonofos (OR = 1.70; 95% CI: 1.22, 2.37), carbaryl (OR = 1.51; 95% CI: 1.03, 2.23), and chlorimuron ethyl (OR = 1.45; 95% CI: 1.01, 2.07) compared with never use. Statistically significant exposure–response trends in association with RA were observed for lifetime days of use of atrazine [ORtertile3= 1.62 (95% CI: 1.09, 2.40); ptrend=0.01] and toxaphene [ORtertile3= 2.42 (95% CI: 1.03, 5.68); ptrend=0.02]. Exposure–response was nonlinear for fonofos [ORtertile1= 2.27 (95% CI: 1.44, 3.57); ORtertile2= 0.98 (95% CI: 0.54, 1.80); ORtertile3= 2.10 (95% CI: 1.32, 3.36); ptrend=0.005] and suggestive for carbaryl (ptrend=0.053).
    Conclusions:
    Our results provide novel evidence of associations between exposure to some pesticides and RA in male farmers. https://doi.org/10.1289/EHP1013
  • Received: 24 August 2016
    Revised: 19 January 2017
    Accepted: 13 February 2017
    Published: 14 July 2017

    Address correspondence to C.G. Parks, Epidemiology Branch, A3-05, National Institite of Environmental Health Sciences, National Institutes of Health, PO Box 12233, Research Triangle Park, NC 27599 USA; Telephone: (919) 541-2577; Email: Parks1@mail.nih.gov

    Supplemental Material is available online (https://doi.org/10.1289/EHP1013).

    The authors declare that they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (210 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (88 KB)

Introduction

Rheumatoid arthritis (RA) is a systemic autoimmune inflammatory disease (McInnes and Schett 2011) that affects 0.5–1% of the world’s adult population (Alamanos and Drosos 2005), with higher rates in North America and northern Europe (Helmick et al. 2008; Tobón et al. 2010; Widdifield et al. 2014). Because of its chronic nature and elevated mortality, RA is a disease of great personal, socioeconomic, and public health concern (Cross et al. 2014; Firestein 2003).

RA is a multifactorial disease that is influenced by genetic and environmental risk factors (Silman and Pearson 2002). Established environmental risk factors for RA include cigarette smoking and silica exposure (Alamanos and Drosos 2005; Mostafalou and Abdollahi 2013; Tobón et al. 2010). Several studies have suggested farming, pesticides, or both as a risk factor for RA (Gold et al. 2007; Li et al. 2008; Lundberg et al. 1994; Milham 1988; Olsson et al. 2004). Although pesticides have not been studied in animal models of RA, diverse immunotoxic effects seen across different types of pesticides support a range of plausible mechanisms (Corsini et al. 2013; Mokarizadeh et al. 2015) consistent with the hypothesis that pesticides may contribute to RA. However, only a few studies have examined types of pesticides in relation to RA.

Parks et al. (2011) observed a higher risk of RA and a related disease, systemic lupus erythematosus, among women who self-reported use of insecticides in the Women’s Health Initiative Observational Study, with a greater risk in women reporting a farming background. In a sample of the U.S. population in the National Health and Nutrition Examination Survey, higher serum levels of organochlorine insecticides were associated with self-reported arthritis, including RA (Lee et al. 2007). In the Agricultural Health Study (AHS), a prospective study of licensed pesticide applicators from Iowa and North Carolina, an early analysis did not show any statistically significant associations between specific pesticides and RA among female spouses (cases: n=135; controls: n=675) of licensed pesticide farmer-applicators (De Roos et al. 2005). However, a more recent analysis of the spouses including a larger number of incident cases (n=275) found a statistically significant association of incident RA with the fungicides maneb/mancozeb and the herbicide glyphosate in addition to elevated association with dichlorodiphenyltrichloroethane (DDT) (Parks et al. 2016).

Because of the higher prevalence of RA in females, with few exceptions (Gold et al. 2007; Lee et al. 2007; Olsson et al. 2004), studies of farming or pesticides and RA have focused on women. However, in the AHS, most farming activities, including pesticide handling and application, are performed by licensed pesticide applicators, of whom 97% are men (Blair et al. 2005). The present study investigated the risk of RA associated with pesticide exposures among male licensed pesticide applicators enrolled in the AHS. Given the limited prior research in humans (and none in animal models of RA), no prior hypotheses were specified, and the full range of pesticides was considered in our analyses. To our knowledge, this is the first time that the association of RA with specific pesticides has been investigated in males.

Methods

Study Population

The AHS has been previously described (Alavanja et al. 1996; Alavanja et al. 2003). Between 1993 and 1997 (phase 1), 52,394 private pesticide applicators (most of whom were male farmers) enrolled when they completed a questionnaire (“enrollment questionnaire”) at the licensing site where they sought to obtain or renew their licenses for pesticide application in North Carolina and Iowa. Enrolled farmers were given another questionnaire to complete at home (“take-home questionnaire”). Over 80% of eligible applicators completed the enrollment questionnaire, and 44% of those who completed the enrollment questionnaire returned the take-home questionnaire. The questionnaires (http://aghealth.nih.gov/collaboration/questionnaires.html) collected information on demographics, lifetime pesticide use, and medical history. Participants were asked to complete follow-up questionnaires at three time points [phase 2 (P2: 1999–2003), phase 3 (P3: 2005–2010), and phase 4 (P4: 2013–2015)] to update their exposures and health status, including RA. This study was approved by the relevant institutional review boards.

Case Ascertainment

Individuals were considered eligible to be incident cases if they answered yes to the P3 and/or P4 questionnaire question “Have you ever been diagnosed specifically with rheumatoid arthritis (an autoimmune disease)?” RA was not assessed for male applicators at P2. Prevalent cases reported RA diagnosed at or before the year of enrollment. Because self-report has proven to overestimate the prevalence of RA (Cross et al. 2014; Walitt et al. 2008), we sought additional information to confirm diagnoses. Eligible cases identified through P3 were screened by telephone to confirm their diagnosis and to provide additional information on the disease, including the use of disease-modifying antirheumatic drugs (DMARDs), as previously described (Parks et al. 2016). In the P4 questionnaire, all self-reported cases were asked about current or past use of DMARDs. Previous validation studies showed that self-reported RA together with DMARD use dramatically increases positive and negative predictive values (Formica et al. 2010).

Inclusion Criteria for Cases and Noncases

Of 51,036 enrolled male pesticide applicators, 20,733 (40.5%) did not return questionnaires in P3 or P4 (see Figure S1). Among the remaining 30,303 applicators, we excluded 503 with missing data on RA or age at diagnosis, 33 with childhood RA, 537 who reported RA at enrollment or at P3 that they refuted in a later questionnaire, and 580 prevalent RA cases (self-reported, including 152 confirmed by DMARD use or physician validation). Of the remaining 28,650 participants eligible for the study sample, 906 individuals were eligible for further RA screening questions (99 were eligible owing to a related autoimmune disease, lupus): of these, 84 could not be reached or had missing data on RA status on the screening interview, 211 reported not having RA, and 200 were considered unlikely to have RA because they did not report use of DMARDs or steroids for RA or being diagnosed or seen by a rheumatologist, leaving 411 identified as potential RA cases. Excluding those missing data on potential confounders, 220 cases were classified as “probable RA” for primary analyses of cases confirmed by DMARD use, and 160 were classified as “possible RA” for sensitivity analyses if they took steroids for RA or were diagnosed or seen by a rheumatologist. Of 27,744 potential noncases (no reported RA), 26,134 had complete covariate data.

Pesticide Exposure Assessment

In the enrollment questionnaire, applicators provided information on ever use of 50 pesticides and detailed data on duration (years) and frequency (days per year) of use for 22 of those pesticides. Duration and frequency of use for the remaining 28 pesticides was collected in the take-home questionnaire. Cumulative lifetime days of use was calculated by multiplying the midpoint of each duration and frequency stratum. This product was multiplied by an intensity score to estimate intensity-weighted lifetime days of pesticide use, a measure that takes into account information on repair of pesticide application equipment, use of personal protective equipment, and application methods (Coble et al. 2011). Based on the distribution of lifetime days (LD) of use (or intensity-weighted LD) in cases, each pesticide was grouped by tertiles and compared with never users. We also evaluated lifetime number of pesticides used (0–5, 6–9, 10–13, ≥14).

Statistical Analysis

Odds ratios (ORs) and 95% confidence intervals (95% CIs) calculated by logistic regression were used to estimate the association of RA with pesticide exposure variables. Covariates were identified based on observed or hypothesized associations with RA and overall pesticide use, and confirmed through selection by stepwise regression. The final model included age at enrollment (continuous), state of enrollment (NC or IA), number of cigarette packs-years (none, <5, 5–18, >18), and education (≤high school degree and >high school). Current body mass index (BMI) and alcohol use at enrollment were considered as potential confounders but were not independently associated with RA and so were not included in final model. Other characterizations of age (e.g., age squared) did not materially affect results.

Adjusted ORs estimated the association of incident RA with ever use of each of 46 pesticides (with at least five exposed cases) and the lifetime number of pesticides used. Exposure–response analyses examined associations across tertiles of lifetime days of pesticide use and intensity-weighted lifetime days compared with no use for those pesticides with ≥20 exposed cases and an OR≥1.2 for ever use. Statistical tests for trend were performed on the categorical (4 levels) exposure variables, comparing tertiles with no use.

In exploratory analyses, models for ever use were stratified by smoking and by age (<50 and ≥50 y), which may be related to the number and types of pesticides used. Tests for interaction and analyses of lifetime days within strata were not performed owing to small numbers and lack of prior hypotheses. We also explored the impact of case definition on results for ever use and lifetime days by including possible incident cases who reported steroid use or being seen or treated by a rheumatologist for RA (160 possible+220 probable cases; total n=380).

Results

RA cases tended to be older at enrollment (Table 1), and after adjusting for age, incident RA was associated with NC residence [OR = 1.60 ( 95% CI: 1.22, 2.10)], current smoking [OR = 1.67 (95% CI: 1.01, 2.76)], and pack-years [e.g., OR = 1.95 (95% CI: 1.39, 2.75), >18 pack-years vs. never].

Table 1. Main characteristic of rheumatoid arthritis (RA) cases and non-cases among male licensed private applicators in the Agricultural Health Study.
Noncases (n=26,134) Incident RA cases (n=220)
Characteristic n (%) n (%) ORa (95% CI)
Age at enrollment (years)
 <40 7,858 (30) 39 (18) Reference
 40–49 7,956 (30) 72 (33) 1.82 (1.23, 2.70)
 50–59 6,035 (23) 75 (34) 2.50 (1.70, 3.69)
 ≥60 4,285 (16) 34 (15) 1.60 (1.01, 2.54)
Body mass index (kg/m2)
 <25 6,157 (26) 41 (20) Reference
 25 to <30 12,353 (51) 114 (55) 1.36 (0.95, 1.94)
 ≥30 5,551 (23) 51 (25) 1.36 (0.90, 2.06)
State of enrollment
 Iowa 17,972 (69) 125 (57) Reference
 North Carolina 8,162 (31) 95 (43) 1.60 (1.22, 2.10)
Race
 White 25,454 (97) 212 (96) Reference
 Nonwhite 659 (3) 8 (4) 1.39 (0.69, 2.84)
Education
 High school or less 14,117 (54) 137 (62) Reference
 More than high school 12,017 (46) 83 (38) 0.77 (0.58, 1.02)
Smoking status at enrollment
 Never smoker 15,046 (58) 100 (45) Reference
 Past smoker 7,637 (29) 74 (34) 1.30 (0.96, 1.78)
 Current smoker 3,392 (13) 46 (21) 2.04 (1.44, 2.90)
Pack-years of cigarette smoked
 None 15,046 (58) 100 (45) Reference
 <5 3,460 (13) 23 (10) 0.98 (0.62, 1.54)
 5–18 3,760 (15) 42 (19) 1.60 (1.11, 2.30)
 >18 3,868 (15) 55 (25) 1.95 (1.39, 2.75)
Alcohol consumption (times/wk)
 None 8,143 (32) 82 (38) Reference
 ≤1 11,778 (47) 84 (39) 0.78 (0.57, 1.07)
 >1 5,331 (21) 49 (23) 1.01 (0.70, 1.44)

Note: CI, confidence interval; OR, odds ratio.

aAdjusted for continuous age.

Incident RA was associated with ever use of the organophosphate insecticide fonofos [OR = 1.70 (95% CI: 1.22, 2.37)], the carbamate insecticide carbaryl [OR = 1.51 (95% CI: 1.03, 2.23)], and the sulfonylurea herbicide chlorimuron ethyl [OR = 1.45 (95% CI: 1.01, 2.07)] compared with never use (Table 2). Although confidence limits included the null, elevated odds ratios (≥1.40) were also observed for the organochlorine insecticides dieldrin [OR = 1.63 (95% CI: 0.77, 3.43)] and toxaphene [OR = 1.44 (95% CI: 0.90, 2.14)], the organophosphate insecticide dichlorvos [OR = 1.40 (95% CI: 0.91, 2.14)], and the fumigant methyl bromide [OR = 1.42 (95% CI: 0.97, 2.08)]. RA was also associated with more pesticides reported [e.g., OR = 1.52 (95% CI: 1.02, 2.32), for ≥14 vs.≤5].

Table 2. Ever use of specific pesticides and incident rheumatoid arthritis (RA) among male licensed pesticide applicators in the Agricultural Health Study.
Noncases (n=26,134) Incident RA cases (n=220)
Pesticide n (%) n (%) ORa (95%CI)
Insecticides
 Organochlorines
  Aldrin 2,370 (17) 29 (23) 1.30 (0.82, 2.05)
  Chlordane 2,590 (19) 36 (28) 1.32 (0.88, 1.98)
  DDT 2,920 (21) 40 (31) 1.15 (0.75, 1.75)
  Dieldrin 491 (4) 8 (6) 1.63 (0.77, 3.43)
  Heptachlor 1,723 (13) 15 (12) 0.88 (0.49, 1.55)
  Lindane 2,006 (15) 18 (14) 0.96 (0.58, 1.59)
  Toxaphene 1,487 (11) 23 (18) 1.44 (0.90, 2.29)
 Organophosphates
  Chlorpyrifos 11,261 (43) 104 (48) 1.30 (0.99, 1.70)
  Coumaphos 2,259 (10) 13 (7) 0.70 (0.40, 1.23)
  Diazinon 2,791 (21) 34 (27) 1.16 (0.77, 1.75)
  Dichlorvos 2,855 (12) 26 (13) 1.40 (0.91, 2.14)
  Fonofos 5,739 (24) 58 (29) 1.70 (1.22, 2.37)
  Malathion 8,983 (66) 87 (67) 1.05 (0.73, 1.53)
  Parathion 1,022 (8) 11 (8) 0.85 (0.45, 1.60)
  Phorate 4,345 (32) 40 (31) 1.14 (0.76, 1.70)
  Terbufos 10,169 (42) 86 (43) 1.24 (0.93, 1.66)
 Carbamates
  Aldicarb 950 (7) 13 (10) 1.08 (0.58, 2.01)
  Carbaryl 5,708 (42) 74 (57) 1.51 (1.03, 2.23)
  Carbofuran 7,066 (29) 63 (32) 1.08 (0.80, 1.46)
 Pyrethroids
  Permethrin Cb 3,352 (14) 31 (16) 1.17 (0.79, 1.73)
  Permethrin Pb 3,637 (15) 25 (13) 1.05 (0.68, 1.62)
 Herbicides
  2,4-D 20,436 (79) 169 (79) 1.16 (0.83, 1.64)
  2,4,5-T 2,612 (19) 29 (22) 1.11 (0.72, 1.71)
  2,4,5-TP 675 (5) 7 (5) 1.00 (0.46, 2.15)
  Alachlor 13,695 (56) 122 (60) 1.26 (0.95, 1.68)
  Atrazine 19,320 (74) 166 (76) 1.29 (0.94, 1.79)
  Butylate 3,912 (29) 35 (27) 1.04 (0.70, 1.54)
  Chlorimuron ethyl 4,470 (33) 51 (39) 1.45 (1.01, 2.07)
  Cyanazine 10,960 (45) 76 (38) 0.96 (0.69, 1.31)
  Dicamba 13,402 (55) 92 (46) 0.90 (0.65, 1.25)
  EPTC 5,181 (22) 24 (12) 0.62 (0.40, 0.96)
  Glyphosate 20,060 (77) 166 (76) 0.90 (0.65, 1.24)
  Imazethapyr 11,292 (45) 86 (44) 1.32 (0.94, 1.86)
  Metolachlor 11,909 (49) 85 (43) 0.90 (0.67,1.20)
  Metribuzin 5,506 (40) 48 (37) 1.08 (0.73, 1.59)
  Paraquat 2,131 (16) 24 (18) 0.92 (0.57, 1.49)
  Pendimethalin 5,183 (38) 49 (37) 0.98 (0.69, 1.40)
  Trifluralin 13,590 (56) 112 (56) 1.23 (0.92, 1.66)
 Fungicides
  Benomyl 979 (7) 9 (7) 0.64 (0.32, 1.31)
  Captan 2,933 (12) 21 (11) 0.90 (0.57, 1.43)
  Chlorothalonil 1,861 (7) 24 (11) 1.27 (0.81, 2.01)
  Maneb 1,031 (8) 14 (11) 0.97 (0.53, 1.78)
  Metalaxyl 2,511 (18) 35 (27) 1.20 (0.77, 1.88)
 Others
  80/20 mix 595 (4) 5 (4) 0.76 (0.31, 1.89)
  Methylbromide 3,825 (15) 53 (24) 1.42 (0.97, 2.08)
  Petroleum oil 3,028 (22) 19 (14) 0.65 (0.40, 1.06)
 Total use reported
  0 – 5 pesticides 5,741 (22) 44 (20) Reference
  6 – 9 7,520 (29) 68 (31) 1.41 (0.96, 2.09)
  10 – 13 6,274 (24) 47 (21) 1.26 (0.82, 1.93)
  ≥14 6,494 (25) 60 (27) 1.54 (1.02, 2.32)

Note: 2,4-D, 2,4-dichlorophenoxyacetic acid; 2,4,5-T, 2,4,5-Trichlorophenoxyacetic acid; 2,4,5-TP, 2- propionic acid (fenoprop); 80/20 mix, carbon tetrachloride/carbon disulfide; CI, confidence interval; DDT, dichlorodiphenyltrichloroethane; EPTC, S-ethyl dipropylthiocarbamate; OR, odds ratio.

aAdjusted for age, state of enrollment, pack-years smoking, and education.

bPermethrin C: used on crops; permethrin P: used on poultry and livestock.

For lifetime days of pesticide use (Table 3), positive exposure–response relationships with incident RA were observed for toxaphene [ORtertile3=2.42 (95% CI: 1.03, 5.68); ptrend=0.02] and for the herbicide atrazine [ORtertile3=1.62 (95% CI: 1.09, 2.40); ptrend=0.01], but not for fonofos [ORtertile1=2.10 (95% CI: 1.32, 3.36), ORtertile3=2.77 (95% CI: 1.44, 3.57)] or chlorimuron ethyl (ptrend=0.10). Other positive associations were observed for which confidence limits excluded the null: the lowest tertile of the herbicide imazethapyr [ORtertile1=1.70 (95% CI: 1.18, 2.55)] and the top tertile of the fungicide chlorothalonil [ORtertile3=2.35 (95% CI: 1.07, 5.14)] compared with never use. Exposure–response patterns were similar for intensity-weighted lifetime days (see Table S1), but we also noted a new trend for the herbicide alachlor [ORtertile3=1.44 (95% CI: 1.00, 2.08); ptrend=0.04] and an association for the top tertile of the fungicide metalaxyl [ORtertile3=1.98 (95% CI: 1.08, 3.64)].

Table 3. Lifetime days use of specific pesticides in relation to incident rheumatoid arthritis (RA) among male licensed pesticide applicators in the Agricultural Health Study.
Noncases (n=26,134) RA cases (n=220)
Insecticides n (%) n (%) ORa (95%CI)
Organochlorines
 Aldrin
  Never 11,257 (83) 99 (78) Reference
  <20 1,095 (8) 14 (11) 1.38 (0.77, 2.50)
  ≥20 to <24.5 116 (1) 2 (2) NC
  ≥24.5 1,084 (8) 12 (10) 1.18 (0.63, 2.22)
  p-trend 0.45
 Chlordane
  Never 11,035 (81) 93 (72) Reference
  <8.75 639 (5) 8 (6) 1.14 (0.54, 2.38)
  ≥8.75 to <20 1,051 (8) 15 (12) 1.41 (0.81, 2.48)
  ≥20 825 (6) 13 (10) 1.48 (0.81, 2.69)
  p-trend 0.11
 Toxaphene
  Never 12,143 (89) 104 (83) Reference
  <24.5 859 (6) 7 (6) 0.82 (0.38, 1.78)
  ≥24.5 to <62.5 392 (3) 9 (7) 2.00 (0.99, 4.04)
  ≥62.5 202 (1) 6 (5) 2.42 (1.03, 5.68)
  p-trend 0.02
Organophosphates
 Chlorpyrifos
  Never 14,678 (57) 111 (52) Reference
  <20 3,756 (15) 34 (16) 1.27 (0.86, 1.88)
  ≥20 to <56 3,727 (15) 37 (17) 1.42 (0.97, 2.06)
  ≥56 3,602 (14) 32 (15) 1.25 (0.84, 1.86)
  p-trend 0.09
 Dichlorvos
  Never 21,220 (88) 170 (87) Reference
  <24.5 1,052 (4) 9 (5) 1.33 (0.67, 2.63)
  ≥24.5 to <173.25 936 (4) 7 (4) 1.14 (0.53, 2.44)
  ≥173.25 796 (3) 9 (5) 1.76 (0.87, 3.48)
  p-trend 0.11
 Fonofos
  Never 18,544 (77) 141 (71) Reference
  <20 1,825 (8) 24 (12) 2.27 (1.44, 3.57)
  ≥20 to <56 2,074 (9) 12 (6) 0.98 (0.54, 1.80)
  ≥56 1,725 (7) 22 (11) 2.10 (1.32, 3.36)
  p-trend 0.005
 Terbufos
  Never 14,178 (59) 116 (58) Reference
  <24.5 3,548 (15) 27 (14) 1.14 (0.74, 1.75)
  ≥24.5 to <63.75 3,464 (14) 29 (15) 1.25 (0.82, 1.89)
  ≥63.75 2954 (12) 28 (14) 1.33 (0.88, 2.03)
  p-trend 0.12
Carbamates
 Carbaryl
  Never 7,966 (59) 55 (44) Reference
  <20 2,339 (17) 23 (18) 1.33 (0.81, 2.18)
  ≥20 to <62.5 1,775 (13) 25 (20) 1.62 (0.96, 2.71)
  ≥62.5 1,434 (11) 23 (18) 1.61 (0.90, 2.86)
  p-trend 0.05
Herbicides
 Alachlor
  Never 10,698 (45) 80 (40) Reference
  <38.75 5,931 (25) 50 (25) 1.25 (0.87, 1.79)
  ≥38.75 to <108.5 3,001 (13) 24 (12) 1.17 (0.74, 1.86)
  ≥108.5 4,362 (18) 44 (22) 1.37 (0.94, 1.98)
  p-trend 0.11
 Atrazine
  Never 6,685 (26) 52 (24) Reference
  <50.75 8,199 (32) 60 (28) 1.10 (0.75, 1.62)
  ≥50.75 to <178.5 6,002 (23) 49 (23) 1.28 (0.85, 1.91)
  ≥178.5 4,914 (19) 54 (25) 1.62 (1.09, 2.40)
  p-trend 0.01
 Chlorimuron ethyl
  Never 9,217 (68) 81 (63) Reference
  <8.75 1,116 (8) 12 (9) 1.47 (0.79, 2.72)
  ≥8.75 to <24.5 1,881 (14) 20 (16) 1.34 (0.82, 2.21)
  ≥24.5 1,397 (10) 16 (12) 1.42 (0.83, 2.45)
  p-trend 0.10
 Imazethapyr
  Never 12,842 (54) 111 (57) Reference
  <20 4,910 (21) 50 (26) 1.74 (1.18, 2.55)
  ≥20 to <24.5 599 (3) 5 (3) 1.42 (0.56, 3.59)
  ≥24.5 5,582 (23) 30 (15) 0.97 (0.62, 1.52)
  p-trend 0.82
 Trifuralin
  Never 10,781 (45) 89 (45) Reference
  <38.75 4,520 (19) 35 (18) 1.11 (0.75, 1.66)
  ≥38.75 to <108.5 3,572 (15) 31 (16) 1.34 (0.87, 2.06)
  ≥108.5 5,204 (22) 43 (22) 1.25 (0.85, 1.84)
  p-trend 0.18
Fungicides
 Chlorothalonil
  Never 24,052 (93) 194 (91) Reference
  <54.25 862 (3) 7 (3) 0.82 (0.38, 1.77)
  ≥54.25 to <200 572 (2) 6 (3) 1.02 (0.44, 2.35)
  ≥200 304 (1) 7 (3) 2.35 (1.07, 5.14)
  p-trend 0.18
 Metalaxyl
  Never 11,164 (83) 96 (74) Reference
  <8 714 (5) 7 (5) 1.05 (0.48, 2.28)
  ≥8 to <50.75 991 (7) 14 (11) 1.20 (0.64, 2.23)
  ≥50.75 616 (5) 13 (10) 1.76 (0.91, 3.41)
  p-trend 0.13
Others
 Methyl bromide
  Never 22,116 (86) 166 (76) Reference
  <12.25 1,004 (4) 16 (7) 1.74 (1.00, 3.02)
  ≥12.25 to <54.25 1,469 (6) 17 (8) 1.18 (0.68, 2.05)
  ≥54.25 1,203 (5) 19 (9) 1.54 (0.90, 2.63)
  p-trend 0.11

Note: CI, confidence interval; OR, odds ratio; NC, not calculated.

aAdjusted for age, state of enrollment, pack-years smoking, and education.

Most pesticide and RA associations observed in the overall sample did not substantially differ by smoking status (see Table S2), but some new statistically positive associations were seen in never smokers, including chlordane, DDT, dieldrin, the herbicide 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), and the fungicide chlorothalonil. No differences were found for total number of pesticides used, but we did not examine exposure–response for individual pesticides owing to the smaller sample size in these stratified analyses.

Most associations between RA and ever use of specific pesticides were similar by age (<50 y and ≥50 y) or were stronger in older participants (see Table S3) except for two new associations observed only in older participants: dichlorvos [OR = 1.82 (95% CI: 1.04, 3.20) and imazethapyr [OR = 1.85 (95% CI: 1.11, 3.05). A higher total number of pesticides (≥14 vs.≤5) was associated with RA in older [≥50 y, OR=2.16 (95% CI: 1.19, 3.90)] but not younger [<50 y, OR = 0.99 (95% CI: 0.55, 1.78)] participants. When limited to chemicals with suggestive associations (OR≥1.2 for ever use), less of an age difference was seen [e.g., ≥7 vs.≤2 pesticides: ≥50 y, OR=2.57 (95% CI: 1.45, 4.56); <50 y, OR = 1.97 (95% CI: 1.10, 3.54)].

In a sensitivity analysis including an additional 160 possible incident cases (total n=380 probable+possible cases; see Tables S4–S6), ORs with RA were increased with age [OR = 2.92 (95% CI: 2.07, 4.11), ≥ 60 y vs. <40 y] and for nonwhites [OR=2.07 (95% CI: 1.27, 3.17)]. Compared with the primary analysis of probable cases, the association was confirmed for fonofos, whereas the OR was reduced for carbaryl and increased for chlorothalonil). Dose–response results were similar for atrazine, but a new exposure–response trend was noted for phorate, and our findings confirmed trends for alachlor and chlorothalonil as well as associations with the highest levels of toxaphene and metalaxyl.

Discussion

In this prospective cohort of male farmers, we observed increased associations of incident RA, and exposure–response trends, for several pesticides. To the best of our knowledge, this is the first epidemiologic study to investigate the development of RA in men in relation to many types of pesticides. Given the limited prior research, our results provide novel evidence on the role of pesticides as environmental risk factors for RA, including robust findings of dose–response associations of RA with the commonly used herbicide atrazine and an organochlorine insecticide, toxaphene. Along with other associations for lifetime use of the insecticides fonofos and carbaryl and the herbicide chlorimuron ethyl, these results warrant replication in other samples.

The AHS offers a unique opportunity to examine the development of RA in a large cohort of farmers with a detailed lifetime history of pesticide use, which is infeasible for population-based case–control studies. However, statistical power is limited for analyses of uncommon outcomes such as RA, particularly when considering exposure–response data on chemicals no longer on the market in the mid-1990s that were covered in the take-home questionnaire completed by only 44% of the sample (e.g., toxaphene).

We observed expected associations of RA with older age and smoking (Karlson and Deane 2012; Linos et al. 1980; Symmons et al. 1994; Uhlig et al. 1998). Smoking-stratified analyses showed no substantial differences in primary results for ever use and total number used (except for some new associations for organochlorines). However, most associations were stronger or only apparent in older participants (e.g., dichlorvos, imazehapyr, chlorimuron ethyl). Explanations for this outcome could include differences in types of pesticide exposures in older cohorts, age at exposure, and opportunities over time for higher doses or multiple exposures. Genetic predisposition may have a lesser impact on older-onset RA (Scott et al. 2013); therefore, it seems plausible that pesticide associations with RA might simply be more apparent in the absence of strong risk factors such as smoking or genotypes associated with early-onset disease.

In these analyses of incident disease, exposures were reported before RA diagnosis, reducing the likelihood of recall bias. Nondifferential exposure misclassification is a possible limitation. However, self-reported pesticide use is considered accurate in the AHS (Hoppin et al. 2002), and detailed data were collected on application frequency, duration, and practices that might influence exposure levels. Indeed, exposure–response trends for lifetime days were observed for cumulative days of use for some pesticides, and intensity-weighted results were not notably different. However, if short-term, intense exposure were more relevant to immune effects, these metrics may not have adequately captured the relevant dose. Future analyses could explore potential determinants of higher-intensity exposures based on average days used per year and on other types of data. Although farmers may experience similar background exposures to the general population from residential and other sources, agricultural pesticides are typically more concentrated chemicals, and most farmers used multiple types. For each comparison, the “unexposed” referent group included those with background exposures and exposures to other pesticides, which could potentially influence he appearance or strength of individual effect estimates. Furthermore, our analyses focused on lifetime exposures reported at baseline but did not include those more proximal to RA diagnoses. The timing of and required dose for pesticide effects on RA pathogenesis is not known, and the relevant window could extend from early-life immune development through processes involved in the generation of preclinical autoimmunity and the development of symptomatic disease.

Early-life pesticide exposures are likely in the cohort but may not be captured by self-reported lifetime exposure questions. Moreover, most participants experienced exposures to several different pesticides over the course of adult use; this is part of a broader picture of multiple pesticide exposures in the AHS, in addition to other potentially relevant exposures, such as organic and inorganic dusts, solvents, and heavy metals. Methods are being developed to help identify critical components of exposure mixtures or potential synergy across correlated exposures (Sun et al, 2013), but they are not robust when the number of cases is small and the number of potential exposures is great. In our study, RA was associated with increasing lifetime number of pesticides used, but the small number of RA cases using individual combinations of chemicals limits our ability to evaluate multiple exposures. Moreover, because we did not identify strong correlations between pesticides, these were not considered as potential confounders. Other occupational exposures, both on and off the farm, may also be risk factors for RA but were not considered as confounders or modifiers for pesticide associations in this analysis. For example, silica dust is an established risk factor for RA (Parks et al. 2003; Stolt et al. 2005); soil silica levels vary across the study region (Stopford and Stopford 1995). Future analyses are planned to explore these questions in the cohort.

Self-reported RA is nonspecific, and many AHS participants who initially reported RA later refuted it in subsequent follow-up and case confirmation. For those with available data (not shown), a refuted RA diagnosis from phase 3 showed high reliability, that is to say, 155 of 180 (86%) repeated their negative response for RA in phase 4. Along with unconfirmed RA cases, these were excluded from the analysis sample, although their potential influence as false negatives would have been unsubstantial compared with their influence as false positives. In contrast, most self-reported RA cases at phase 3 who used DMARDs confirmed their self-report at phase 4, that is to say, 91 of 98 cases (93%). Of 906 potential RA cases eligible for screening questions about their diagnosis and medications, fewer than half (n=411) were considered as possible cases, and only half of these were classified as probable cases confirmed by DMARDs use. Although identifying RA cases based on self-report and use of DMARDs has high specificity (Formica et al. 2010; Walitt et al. 2008), this definition may lack sensitivity for males and for older individuals, who may be less likely to use DMARDs (Schmajuk et al. 2011; Solomon et al. 2012); therefore, it could have limited generalizability and missed associations if use of specialized medical treatment is related to pesticide use. Analyses adding in the possible RA cases, who reported use of steroids for RA or being diagnosed or treated by a rheumatologist, yielded similar results for many, but not all, pesticides. We found increased associations for RA with older age and with nonwhite race for the more inclusive group of RA cases (probable+possible). This finding confirms the potential influence of biased case ascertainment; therefore, differences in pesticide associations may warrant further consideration.

Loss to follow-up could also have biased the results. Smoking is related to loss to follow-up in AHS participants, but recent analyses show little impact on cancer associations (and a lack of compensation by inverse-probability weighting) when the associations with smoking were not extreme (Rinsky et al. 2017). Smoking was a moderate risk factor for RA in this sample, suggesting limited potential for bias.

Although the immune effects of pesticides are variable, several associations observed in the study suggest plausible biologic mechanisms related to RA pathogenesis. For example, fonofos altered the methylation levels of genes involved in the regulation of the immune response (Zhang et al. 2012). Immunotoxic effects of other organophosphates have been described (Corsini et al. 2013), and a more detailed assessment of updated AHS organophosphate exposures is justified given their widespread use and the changes in types of organophosphates used over time in the cohort and in the general population. Interestingly, carbamates (e.g., carbaryl) and organophosphates share the same main mechanism of toxicity: inhibition of acetylcholinesterase enzyme activity in neuronal and neuromuscular synapses (Fukuto 1990). This ability to inhibit serine hydrolase and protease enzymes, which play an essential role in the immune system, may hold a common explanation for some alterations in the immune function induced by pesticides of both classes (Casale et al. 1992; Galloway and Handy 2003; Guo et al. 2014; Long and Cravatt 2011).

We previously noted an association of DDT use with incident RA in female AHS spouses (Parks et al. 2016). Although incident RA was not significantly associated with DDT in the present study overall, associations with DDT (and other organochlorines) were apparent in never smokers. Other evidence of organochlorine pesticide associations with RA includes elevated serum organochlorines in female RA cases in the National Health and Nutrition Examination Survey (1999–2002) (Lee et al. 2007) and experimental studies that have suggested a role for DDT in inflammatory processes and immune disruption. (Cardenas-Gonzalez et al. 2013; Kim et al. 2004; Dutta et al. 2008). Toxaphene was associated with RA in our analyses, although the estimates were imprecise because of small numbers. Banned in 1990, toxaphene was often used on cotton in the southeastern United States, along with other organochlorines.

We observed an exposure–response trend in the association of incident RA with atrazine, a triazine herbicide for which immunotoxic effects have been observed in experimental animals. In mice, atrazine has been shown to induce apoptosis of splenocytes, whereas leukocytes increased in a dose-dependent manner (Zhang et al. 2011); to elicit a broad inhibitory effect on cell-mediated, humoral, and nonspecific immunity function (Chen et al. 2013); and to significantly decrease natural killer (NK) cell lysis activity in vitro (Rowe et al. 2007; Whalen et al. 2003). The sulfonylurea herbicide chlorimuron ethyl was also associated with RA risk. Sulfonylurea herbicides have been considered to be relatively nontoxic to humans (Mushak and Piver 1992). Another common herbicide, glyphosate, was previously associated with incident RA in AHS female spouses (Parks et al. 2016), but we saw no glyphosate association with incident RA in the male applicators. Glyphosate use in U.S. agriculture has dramatically increased over the past two decades; therefore, updated glyphosate exposure data are needed to properly examine associations with incident RA.

Conclusions

The results from our study provide evidence that exposure to some pesticides may play a role in the development of RA among male farmers regardless of age, smoking, and educational level. Because this is the first study to observe increased RA risk associated with these pesticides among men, our findings need confirmation in other populations.

Acknowledgments

The authors thank S. Long (Westat) for programming support and for helpful review of manuscript drafts.

This work was supported by the Intramural Research Program of the National Institutes of Health/National Institute of Environmental Health Sciences (Z01-ES049030) and by the National Cancer Institute (Z01-CP010119). A.M. was funded by Coordination for the Improvement of Higher Education Personnel (CAPES), Brazil (grant no. 2478/2015-03).

References

Alamanos Y, Drosos AA. 2005. Epidemiology of adult rheumatoid arthritis. Autoimmun Rev 4(3):130–136, PMID: 15823498, 10.1016/j.autrev.2004.09.002.

Alavanja MCR, Samanic C, Dosemeci M, Lubin J, Tarone R, Lynch CF, et al. 2003. Use of agricultural pesticides and prostate cancer risk in the Agricultural Health Study cohort. Am J Epidemiol 157(9):800–814, 10.1093/aje/kwg040.

Alavanja MC, Sandler DP, McMaster SB, Zahm SH, McDonnell CJ, Lynch CF, et al. 1996. The Agricultural Health Study. Environ Health Perspect 104(4):362–369, PMID: 8732939.

Blair A, Sandler DP, Tarone R, Lubin J, Thomas K, Hoppin JA, et al. 2005. Mortality among participants in the Agricultural Health Study. Ann Epidemiol 5(4):279–285, PMID: 15780775, 10.1016/j.annepidem.2004.08.008.

Cardenas-Gonzalez M, Gaspar-Ramirez O, Perez-Vazquez FJ, Alegria-Torres JA, Gonzalez-Amaro R, Perez-Maldonado IN. 2013. p,p′-DDE, a DDT metabolite, induces proinflammatory molecules in human peripheral blood mononuclear cells “in vitro”. Exp Toxicol Pathol 65(5):661–665, 10.1016/j.etp.2012.07.006.

Casale GP, Bavari S, Gold RE, Vitzthum EF. 1992. Inhibition of interleukin-2-stimulated enhancement of human natural killer (NK) cell activity by carbaryl, an anticholinesterase insecticide. Toxicol Lett 63(3):299–311, PMID: 1488778.

Chen JY, Song Y, Zhang LS. 2013. Immunotoxicity of atrazine in Balb/c mice. J Environ Sci Health B 48(8):637–645, PMID: 23638890, 10.1080/03601234.2013.777308.

Coble J, Thomas KW, Hines CJ, Hoppin JA, Dosemeci M, Curwin B, et al. 2011. An updated algorithm for estimation of pesticide exposure intensity in the Agricultural Health Study. Int J Environ Res public health 8(12):4608–4622, PMID: 22408592, 10.3390/ijerph8124608.

Corsini E, Sokooti M, Galli CL, Moretto A, Colosio C. 2013. Pesticide induced immunotoxicity in humans: a comprehensive review of the existing evidence. Toxicol 307:123–135, 10.1016/j.tox.2012.10.009.

Cross M, Smith E, Hoy D, Carmona L, Wolfe F, Vos T, et al. 2014. The global burden of rheumatoid arthritis: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis 73(7):1316–1322, PMID: 24550173, 10.1136/annrheumdis-2013-204627.

De Roos AJ, Cooper GS, Alavanja MC, Sandler DP. 2005. Rheumatoid arthritis among women in the Agricultural Health Study: risk associated with farming activities and exposures. Ann Epidemiol 15(10):762–770, PMID: 16257361, 10.1016/j.annepidem.2005.08.001.

Dutta R, Mondal AM, Arora V, Nag TC, Das N. 2008. Immunomodulatory effect of DDT (bis[4-chlorophenyl]-1,1,1-trichloroethane) on complement system and macrophages. Toxicol 252(1):78–85, 10.1016/j.tox.2008.07.063.

Firestein GS. 2003. Evolving concepts of rheumatoid arthritis. Nature 423:356–361, PMID: 12748655, 10.1038/nature01661.

Formica MK, McAlindon TE, Lash TL, Demissie S, Rosenberg L. 2010. Validity of self-reported rheumatoid arthritis in a large cohort: results from the Black Women’s Health Study. Arthritis Care Res (Hoboken) 62:235–241.

Fukuto TR. 1990. Mechanism of action of organophosphorus and carbamate insecticides. Environ Health Perspect 87:245–254, PMID: 2176588.

Galloway T, Handy R. 2003. Immunotoxicity of organophosphorous pesticides. Ecotoxicology 12(1–4):345–363, PMID: 12739880.

Gold LS, Ward MH, Dosemeci M, De Roos AJ. 2007. Systemic autoimmune disease mortality and occupational exposures. Arthritis Rheum 56(10):3189–3201, PMID: 17907164, 10.1002/art.22880.

Guo Y, Wu Q, Ni B, Mou Z, Jiang Q, Cao Y, et al. 2014. Tryptase is a candidate autoantigen in rheumatoid arthritis. Immunol 142(1):67–77, PMID: 24433328, 10.1111/imm.12197.

Helmick CG, Felson DT, Lawrence RC, Gabriel S, Hirsch R, Kwoh CK, et al. 2008. Estimates of the prevalence of arthritis and other rheumatic conditions in the United States: part I. Arthritis Rheum 58(1):15–25, PMID: 18163481, 10.1002/art.23177.

Hoppin JA, Yucel F, Dosemeci M, Sandler DP. 2002. Accuracy of self-reported pesticide use duration information from licensed pesticide applicators in the Agricultural Health Study. J Expo Anal Environ Epidemiol 12(5):313–318, PMID: 12198579, 10.1038/sj.jea.7500232.

Karlson EW, Deane K. 2012. Environmental and gene-environment interactions and risk of rheumatoid arthritis. Rheum Dis Clin North Am 38(2):405–426, PMID: 22819092, 10.1016/j.rdc.2012.04.002.

Kim JY, Choi CY, Lee KJ, Shin DW, Jung KS, Chung YC, et al. 2004. Induction of inducible nitric oxide synthase and proinflammatory cytokines expression by o,p′-DDT in macrophages. Toxicol Lett 147(3):261–269, PMID: 15104118.

Lee DH, Steffes M, Jacobs DR. 2007. Positive associations of serum concentration of polychlorinated biphenyls or organochlorine pesticides with self-reported arthritis, especially rheumatoid type, in women. Environ Health Perspect 115(6):883–888, PMID: 17589595, 10.1289/ehp.9887.

Li X, Sundquist J, Sundquist K. 2008. Socioeconomic and occupational risk factors for rheumatoid arthritis: a nationwide study based on hospitalizations in Sweden. J Rheumatol 36(6):986–991, PMID: 18464310.

Linos A, Worthington JW, O’Fallon M, Kurland LT. 1980. The epidemiology of rheumatoid arthritis in Rochester, Minnesota: a study of incidence, prevalence, and mortality. Am J Epidemiol 111(1):87–98, 10.1093/oxfordjournals.aje.a112878.

Long JZ, Cravatt BF. 2011. The metabolic serine hydrolases and their functions in mammalian physiology and disease. Chem Rev 111(10):6022–6063, PMID: 21696217, 10.1021/cr200075y.

Lundberg I, Alfredsson L, Plato N, Sverdrup B, Klareskog L, Kleinau S. 1994. Occupation, occupational exposure to chemicals and rheumatological disease. A register based cohort study. Scand J Rheumatol 23(6):305–310, PMID: 7801054.

McInnes IB, Schett G. 2011. The pathogenesis of rheumatoid arthritis. N Engl J Med 365(23):2205–2219, PMID: 22150039, 10.1056/NEJMra1004965.

Milham S. Jr
1988. Using multiple cause of death coding in occupational mortality studies. Am J Ind Med 14(3):341–344, 10.1002/ajim.4700140311.

Mokarizadeh A, Faryabi MR, Rezvanfar MA, Abdollahi M. 2015. A comprehensive review of pesticides and the immune dysregulation: mechanisms, evidence and consequences. Toxicol Mech Methods 25(4):258–278, PMID: 25757504, 10.3109/15376516.2015.1020182.

Mostafalou S, Abdollahi M. 2013. Pesticides and human chronic diseases: evidences, mechanisms, and perspectives. Toxicol Appl Pharmacol 268(2):157–177, PMID: 23402800, 10.1016/j.taap.2013.01.025.

Mushak EW, Piver WT. 1992. Agricultural chemical utilization and human health. Environ Health Perspect 97:269–274, PMID: 1396466.

Olsson AR, Skogh T, Axelson O, Wingren G. 2004. Occupations and exposures in the work environment as determinants for rheumatoid arthritis. Occup Environ Med 61(3):233–238, PMID: 14985518.

Parks CG, Cooper GS, Nylander-French LA, Storm JF, Archer JD. 2003. Assessing exposure to crystalline silica from farm work: a population-based study in the Southeastern United States. Ann Epidemiol 13(5):385–392, PMID: 12821278.

Parks CG, Hoppin JA, De Roos AJ, Costenbader KH, Alavanja MC, Sandler DP. 2016. Rheumatoid arthritis in Agricultural Health Study spouses: associations with pesticides and other farm exposures. Environ Health Perspect 124(11):1728–1734, PMID: 27285288, 10.1289/EHP129.

Parks CG, Walitt BT, Pettinger M, Chen JC, de Roos AJ, Hunt J, et al. 2011. Insecticide use and risk of rheumatoid arthritis and systemic lupus erythematosus in the Women’s Health Initiative Observational Study. Arthritis Care Res (Hoboken) 63(2):184–194, PMID: 20740609, 10.1002/acr.20335.

Rinsky JL, Richardson DB, Wing S, Beard JD, Alavanja M, Beane Freeman LE, et al. 2017. Assessing the potential for bias from non-response to a study follow-up interview: an example from the Agricultural Health Study. Am J Epidemiol. 10.1093/aje/kwx098.

Rowe AM, Brundage KM, Barnett JB. 2007. In vitro atrazine-exposure inhibits human natural killer cell lytic granule release. Toxicol Appl Pharmacol 221(1):179–188, PMID: 17475299, 10.1016/j.taap.2007.01.012.

Schmajuk G, Trivedi AN, Solomon DH, et al. 2011. Receipt of disease-modifying antirheumatic drugs among patients with rheumatoid arthritis in medicare managed care plans. JAMA 305(5):480–486, PMID: 21285425, 10.1001/jama.2011.67.

Scott IC, Seegobin SD, Steer S, et al. 2013. Predicting the risk of rheumatoid arthritis and its age of onset through modelling genetic risk variants with smoking. PLoS Genet 9(9):e1003808 PMID: 24068971, 10.1371/journal.pgen.1003808.

Silman AJ, Pearson JE. 2002. Epidemiology and genetics of rheumatoid arthritis. Arthritis Res 4:S265–S272.

Solomon DH, Ayanian JZ, Yelin E, Shaykevich T, Brookhart MA, Katz JN. 2012. Use of disease-modifying medications for rheumatoid arthritis by race and ethnicity in the National Ambulatory Medical Care Survey. Arthritis Care Res (Hoboken) 64(2):184–189, PMID: 22012868, 10.1002/acr.20674.

Stolt P, Källberg H, Lundberg I, Sjögren B, Klareskog L, Alfredsson L. 2005. Silica exposure is associated with increased risk of developing rheumatoid arthritis: results from the Swedish EIRA study. Ann Rheum Dis 64(4):582–586, PMID: 15319232, 10.1136/ard.2004.022053.

Stopford CM, Stopford W. 1995. Potential for respirable quartz exposure from North Carolina farm soils. Scand J Work Environ Health 21(Suppl 2):44–46, PMID: 8929688.

Sun Z, Tao Y, Li S, Ferguson KK, Meeker JD, Park SK, et al. 2013. Statistical strategies for constructing health risk models with multiple pollutants and their interactions: possible choices and comparisons. Environ Health 12(1):85, PMID: 24093917, 10.1186/1476-069X-12-85.

Symmons DP, Barrett EM, Bankhead CR, Scott DG, Silman AJ. 1994. The incidence of rheumatoid arthritis in the United Kingdom: results from the Norfolk Arthritis Register. Br J Rheumatol 33(8):735–739, PMID: 8055200.

Tobón GJ, Youinou P, Saraux A. 2010. The environment, geo-epidemiology, and autoimmune disease: rheumatoid arthritis. J Autoimmun 35(1):10–14, PMID: 20080387, 10.1016/j.jaut.2009.12.009.

Uhlig T, Kvien TK, Glennas A, Smedstad LM, Forre O. 1998. The incidence and severity of rheumatoid arthritis, results from a county register in Oslo, Norway. J Rheumatol 25(6):1078–1084, PMID: 9632067.

Walitt BT, Constantinescu F, Katz JD, Weinstein A, Wang H, Hernandez RK, et al. 2008. Validation of self-report of rheumatoid arthritis and systemic lupus erythematosus: The Women’s Health Initiative. J Rheumatol 35(5):811–818, PMID: 18398940.

Whalen MM, Loganathan BG, Yamashita N, Saito T. 2003. Immunomodulation of human natural killer cell cytotoxic function by triazine and carbamate pesticides. Chem Biol Interact 145(3):311–319, PMID: 12732457.

Widdifield J, Paterson JM, Bernatsky S, Tu K, Tomlinson G, Kuriya B, et al. 2014. The epidemiology of rheumatoid arthritis in Ontario, Canada. Arthritis Rheumatol 66(4):786–793, 10.1002/art.38306.

Zhang X, Wallace AD, Du P, Kibbe WA, Jafari N, Xie H, et al. 2012. DNA methylation alterations in response to pesticide exposure in vitro. Environ Mol Mutagen 53(7):542–549, PMID: 22847954, 10.1002/em.21718.

Zhang X, Wang M, Gao S, Ren R, Zheng J, Zhang Y. 2011. Atrazine-induced apoptosis of splenocytes in balb/c mice. BMC Med 9:117, PMID: 22032520, 10.1186/1741-7015-9-117.

A Longitudinal Analysis of the Influence of the Neighborhood Environment on Recreational Walking within the Neighborhood: Results from RESIDE

Author Affiliations open

1School of Population Health, The University of Western Australia, Perth, Western Australia, Australia

2Centre for the Built Environment and Health, School of Earth and Environment & School of Sports Science, Exercise and Health, The University of Western Australia, Perth, Western Australia, Australia

3Telethon Kids Institute, The University of Western Australia, Perth, Western Australia, Australia

4McCaughey VicHealth Community Wellbeing Unit, School of Population and Global Health, The University of Melbourne, Melbourne, Victoria, Australia

PDF icon PDF Version (408 KB)

  • Background:
    There is limited longitudinal evidence confirming the role of neighborhood environment attributes in encouraging people to walk more or if active people simply choose to live in activity-friendly neighborhoods. Natural experiments of policy changes to create more walkable communities provide stronger evidence for a causal effect of neighborhood environments on residents’ walking.
    Objectives:
    We aimed to investigate longitudinal associations between objective and perceived neighborhood environment measures and neighborhood recreational walking.
    Methods:
    We analyzed longitudinal data collected over 8 yr (four surveys) from the RESIDential Environments (RESIDE) Study (Perth, Australia, 2003–2012). At each time point, participants reported the frequency and total minutes of recreational walking/week within their neighborhood and neighborhood environment perceptions. Objective measures of the neighborhood environment were generated using a Geographic Information System (GIS).
    Results:
    Local recreational walking was influenced by objectively measured access to a medium-/large-size park, beach access, and higher street connectivity, which was reduced when adjusted for neighborhood perceptions. In adjusted models, positive perceptions of access to a park and beach, higher street connectivity, neighborhood esthetics, and safety from crime were independent determinants of increased neighborhood recreational walking. Local recreational walking increased by 9 min/wk (12% increase in frequency) for each additional perceived neighborhood attribute present.
    Conclusions:
    Our findings provide urban planners and policy makers with stronger causal evidence of the positive impact of well-connected neighborhoods and access to local parks of varying sizes on local residents’ recreational walking and health. https://doi.org/10.1289/EHP823
  • Received: 19 July 2016
    Revised: 29 January 2017
    Accepted: 10 February 2017
    Published: 12 July 2017

    Address correspondence to H. Christian, School of Population Health (M707), The University of Western Australia, 35 Stirling Highway, Crawley WA 6009, Australia. Telephone: 61-8-6488 8501. Email: hayley.christian@uwa.edu.au

    Supplemental Material is available online (https://doi.org/10.1289/EHP823).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (96 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (89 KB)

Introduction

Increasingly, there are calls to “rethink” approaches to the prevention of disease in the face of global rises in noncommunicable diseases and obesity (Das and Horton 2012; Giles-Corti et al. 2016; Kleinert and Horton 2015). Working with sectors outside of health to create more supportive and sustainable built environments is recognized as an important strategy with a range of cobenefits, such as improving health and the environment, reducing traffic congestion and heat island effects, and mitigating the negative impacts of climate change (Cheng and Berry 2013; Sallis et al. 2016; Watts et al. 2015).

Designing cities that promote health is now a multisector global priority, building on the World Health Organization (WHO) decade-old Healthy Cities agenda (Duhl 1996). For example, the 2015 United Nations Sustainable Development Goals 2030 targets both making cities more inclusive, safe, resilient, and sustainable, and ensuring healthy lives and promoting well-being (United Nations 2015). The 2016 United Nations Conference on Housing and Sustainable Urban Development (HABITAT III) set the stage for the New Urban Agenda, including standards for achieving sustainable urban development worldwide (United Nations 2016). Moreover, in late 2016, the WHO Shanghai Declaration reaffirmed its commitment to planning cities to promote health (World Health Organization 2016). Effective translation of findings from transdisciplinary international built environment and health research to urban planning policy and practice will help guide implementation of the sustainable urban development agenda while also promoting health and well-being (Giles-Corti et al. 2016; Sallis et al. 2016; Stevenson et al. 2016).

There is a growing body of evidence showing associations between neighborhood attributes and physical activity, particularly transport walking (Ding and Gebel 2012; McCormack and Shiell 2012; Saelens and Handy 2008; Sallis et al. 2009, 2016). However, to date, much of this evidence is cross-sectional, which limits causal inferences being drawn. Only a handful of recent cohort studies with longitudinal data (Hirsch et al. 2014; Panter et al. 2013) and natural experiments of changes to the built environment (Goodman et al. 2013) provide evidence supporting a causal effect of the built environment on residents’ physical activity (Giles-Corti et al. 2013; Halonen et al. 2015; Knuiman et al. 2014; Ranchod et al. 2014; Turrell et al. 2014).

Residential preferences (i.e., self-selection factors) are an important consideration when examining causal relationships between the neighborhood environment and local walking (Boone-Heinonen et al. 2010; Giles-Corti et al. 2008). Longitudinal studies collect repeated measures over time on built environment attributes, walking, and potential confounders. There are now modeling approaches available that utilize all available data on each individual, adjust for measured confounders, and which isolate and compare the between-person (cross-sectional) and within-person (longitudinal) effects of built environment attributes on walking (Allison 2005; Fitzmaurice et al. 2012; Knuiman et al. 2014). This is important because, unlike the between-person effect, the within-person effect is not subject to confounding by unmeasured (time constant) self-selection factors and other confounders (Allison 2005; Fitzmaurice et al. 2012; Knuiman et al. 2014).

Another important consideration in studies of the effect of the neighborhood environment on local walking is the effect of different neighborhood environment attributes on different types of behavior. We have previously argued (Giles-Corti et al. 2013) and shown that transport and recreational walking are distinct behaviors and impacted by different neighborhood features. Transport walking involves walking specifically to get to or from somewhere, such as walking to a shop, work, or public transport, while recreational walking is undertaken for recreation, health, or fitness purposes (Giles-Corti et al. 2006). As such, they should be examined using different models using context-specific measures of the behavior (e.g., recreational walking in neighborhood) and behavior-specific determinants (e.g., access to park). To date, relatively few studies have examined the impact of the built environment on recreation-based walking using a longitudinal design (Giles-Corti et al. 2013; Halonen et al. 2015; Ranchod et al. 2014). After a 3-yr follow-up of adults (n=6,814) in the U.S. Multi-Ethnic Study of Atherosclerosis, an increase in access to objectively measured neighborhood recreational facility density over time was associated with a smaller decline in recreational physical activity, particularly in older adults (Ranchod et al. 2014). This study was limited because recreational physical activity undertaken in the local neighborhood environment was not measured, and hence the behavioral measure was not matched to the exposure variable, which assessed neighborhood recreational facility density. Moreover, other neighborhood attributes (e.g., parks) previously associated with recreational physical activity, particularly walking (Bancroft et al. 2015), were not investigated.

To date, it appears that only one study, the RESIDential Environments Project (RESIDE), has examined the influence of the neighborhood environment on local recreational walking using a longitudinal design and context-specific measures of both exposure and behavior (Giles-Corti et al. 2013). RESIDE was a natural experiment of an urban policy intervention, and our previous analysis investigated the effect of gaining access to neighborhood destinations on local walking 1 yr after relocating to a new home (Giles-Corti et al. 2006, 2008). Among participants who gained access to recreational destinations (i.e., the beach, a park, or sports fields), their recreational walking increased by almost 18 min/wk for each type (range: 0–3) of recreational destination gained after relocation (Giles-Corti et al. 2013). However, notably, participants had only lived in their new neighborhood for about a year. Additionally, while key infrastructure, such as public parks, are typically delivered early in the land development process, there are often delays in the on-ground delivery of other community infrastructure until there are sufficient people in place to support shops, service, and public transport.

Longer-term follow-up is required to account for the evolving nature of new neighborhoods and the impact of these changes on residents’ recreational walking levels over time (Giles-Corti et al. 2013). Hence, this paper advances our previous RESIDE analyses by using data collected over 8 yr (four surveys) to investigate longitudinal associations between objective and perceived neighborhood environment measures and neighborhood recreational walking.

Methods

Sample and Data Collection

RESIDE commenced in 2003 and is a longitudinal natural experiment of 1,813 people building homes in 73 new housing developments across metropolitan Perth, Western Australia. Details of participant recruitment procedures have been reported elsewhere (Giles-Corti et al. 2008). Briefly, participants moving to each development were invited to participate by the state water authority following the land transfer transaction. The following eligibility criteria were applied: English proficiency, ≥18 yr, intention to relocate by December 2005, and willingness to complete surveys four times over 8 yr. Participants were recruited by telephone and one person from each household randomly selected. Participants were surveyed four times: baseline (T1: n=1,813), then approximately 1 (T2: n=1,467), 3 (T3: n=1,230), and 7 yr (T4: n=565) later. Almost all participants (99%) relocated between T1 and T2, and 10% moved again after T2.

Sociodemographic Factors

Gender, age, marital status (married/de facto, separated/divorced/widowed/single), education (secondary or less, trade/apprenticeship/certificate, bachelor degree or higher), occupation (manager/administrator, blue collar, clerical/sales/service/other, not in workforce), hours of work per week (≤19, 20–38, 39–59, ≥60; not in workforce), minutes/day of work travel (work from home/≤30, 31–60, ≥60; work multiple locations; not in workforce), level of physical activity at work (physically inactive, regular walking, moderately active, vigorously active, not in workforce), children at home, and dog ownership status was collected.

Local Recreational Walking

Participants reported the frequency and total minutes of recreational walking within their neighborhood (defined as a 15-min walk from their home) over a usual week using the Neighborhood Physical Activity Questionnaire (Giles-Corti et al. 2006). The neighborhood was defined as a 1600-m road network buffer around each participant’s home (Giles-Corti et al. 2008).

Objective Measures of the Neighborhood Environment

At each time point, temporally relevant objective measures of the neighborhood environment were generated using a Geographic Information System (GIS), and included street connectivity and residential density (Christian et al. 2011; Frank et al. 2005). The presence of pocket/small (≤0.5 ha) and medium/large (neighborhood) size (>0.5 to ≤5 ha) parks within a 400-m neighborhood service area and the presence of a district- or regional-size (district >5 to ≥15 ha, regional >15 ha) park and beach access within 1,600-m network-based neighborhood service area of each participant’s home were calculated (Bull et al. 2013; Christian et al. 2015; Western Australian Department of Sport and Recreation 2012). Neighborhood service areas were selected to maximize variation between and within participants over time. An overall “objective” neighborhood environment index (range: 0 to 6) was calculated and included the presence or absence (yes=1, no=0) of pocket/small park, medium/large park, district/regional park, beach access, as well as street connectivity and residential density. Street connectivity and residential density values were converted to z-scores using the mean and standard deviation at baseline (T1). Dichotomized scores (yes=1, no=0) were based on the median split in z-score where an above median z-score=1 and a below median z-score=0.

Perceptions of the Neighborhood Environment

Subscales from the Neighborhood Environment and Walking Scale (NEWS) questionnaire (Cerin et al. 2006) were used to measure participants’ neighborhood environment perceptions. Scales included street connectivity, infrastructure and safety for walking, neighborhood esthetics, traffic safety, and crime safety (Cerin et al. 2006). To complement our objective measures, the presence of a park within 5 min perceived walking distance from home and the presence of a sports field and beach within 15 min were calculated from responses to the NEWS destinations subscale (Cerin et al. 2006). A perceived neighborhood environment index (range: 0 to 8) was calculated and included the perceived presence (yes=1, no=0) of a park, sports field, or beach, and perceived street connectivity, infrastructure and safety for walking, neighborhood aesthetics, traffic safety, and crime safety (z-scores above baseline median score=1, z-scores below baseline median score=0).

Statistical Analysis

Associations between neighborhood environment measures and frequency of recreational walking were assessed using negative binomial log-linear repeated measures regression models with generalized estimating equations (GEE) estimation and robust standard errors (Fitzmaurice et al. 2012). Associations with total minutes of recreational walking were assessed using a marginal linear repeated measures (unrestricted covariance) regression model. Three multivariate models were fitted. Model 1 included the objectively measured neighborhood environment variables. Model 2 included comparable perceived neighborhood environment variables. Model 3 included both the objectively measured and perceived neighborhood environment variables. All models adjusted for sociodemographic factors (age, gender, marital status, education level, occupation, weekly hours of work, daily minutes of work travel, physical activity level at work, children at home, and dog ownership) as time-varying covariates. For the log-linear models, the exponentiated coefficient estimates are reported [with 95% confidence intervals (CI)] and represent the relative change in frequency of recreational walking for a 1-unit change in the neighborhood environment measure. For the linear models, the coefficient estimates are reported and represent the absolute change in minutes/week of recreational walking in the neighborhood.

In addition to estimating the overall effect of each neighborhood environment measure, we fitted additional models that simultaneously included both the mean of the measure (averaged over time) and the mean-centered deviation in order to separately estimate and compare the between-person (cross-sectional) and within-person (longitudinal) effects of the measure (Allison 2005). The between-person and within-person effects were found to be similar for all measures; thus, only the overall combined between-person and within-person effect estimates have been reported.

Furthermore, we conducted a mediation analysis to determine whether a particular objectively measured neighborhood environment attribute (e.g., GIS-derived street connectivity z-score) was mediated by the corresponding perceived neighborhood environment measure (e.g., NEWS street connectivity scale) (Figure 1). We tested whether: a) the objective measure was related to recreational walking without adjustment for the perceived measure (Model 1); b) the objective measure was related to the perceived measure (Table S1); c) the perceived measure was related to recreational walking without adjustment for the objective measure (Model 2); and d) the objective measure has a smaller effect and the perceived measure a significant effect on recreational walking when both are included together in a model (Model 3).

Mediation relationship between objective and perceived measures of the neighborhood environment and recreational walking.
Figure 1. Mediation relationship between objective and perceived measures of the neighborhood environment and recreational walking.

Results

Baseline Sociodemographic Characteristics and Recreational Walking in Cohort

Table 1 shows the baseline sociodemographic characteristics of the study cohort (n=1,771). The mean age of participants was 40 yr, 59% were female, 82% were married or in a de facto relationship, 23% had a bachelor degree or higher level of education, 49% had children at home, and 44% had a dog.

Table 1. Baseline sociodemographic characteristics of cohort (n=1,771).
Variable %
Female 59.3
Mean age (SD) 40.0 (11.9)
Marital status
 Married/de facto 81.6
 Separated/divorced/widowed/single 18.4
Education level
 Secondary or less 39.7
 Trade/apprenticeship/certificate 37.5
 Bachelor degree or higher 22.8
Occupation
 Manager/administrator 15.1
 Professional 27.4
 Blue collar 17.1
 Clerical/sales/service/other 23.2
 Not in workforce 17.2
Hours of work per week
 ≤19 10.5
 20–38 26.7
 39–59 41.1
 ≥60 4.5
 Not in workforce 17.2
Minutes of work travel per day
 Work from home/≤30 20.7
 31–60 23.0
 ≥61 17.8
 Work multiple locations 21.2
 Not in workforce 17.2
Level of physical activity at work
 Physically inactive 32.1
 Regular walking 27.7
 Moderately active 15.6
 Vigorously active 7.3
 Not in workforce 17.2
Children at home 49.0
Dog owner 44.1

At baseline, participants walked for recreation in the neighborhood an average of twice per week for an average duration of 69 min. Local recreational walking increased following relocation and then stabilized, with the average frequency/duration per week at 1-, 3-, and 8-yr follow-up being: 2.6/89 min, 2.4/91 min, and 2.4/87 min, respectively (Table 2). There was considerable (cross-sectional) between-person variation, with the standard deviation in frequency of walking/week being 2.6 to 3.0 at each time point. There was also considerable (longitudinal) within-person variation with the standard deviation in the frequency of walking/week for the baseline to 1-yr follow-up change being 2.6, and for 1- to 3-yr follow-up and 3- to 8-yr follow-up being 2.7 and 3.0, respectively.

Table 2. Neighborhood recreational walking and attributes of the objectively measured and perceived neighborhood environment at each time point [mean±SD or n (%)].
T1 T2 T3 T4
n=1,771 n=1,383 n=1,176 n=541
Neighborhood recreational walking
 Frequency per week 2.04 (2.64) 2.57 (2.82) 2.35 (2.87) 2.40 (3.03)
 Minutes per week 69.0 (98.4) 89.3 (113.5) 90.7 (128.2) 86.7 (121.2)
Objectively measured
 Connectivity z-score 0.00 (1.00) 0.68 (1.42) 0.95 (1.44) 1.15 (1.52)
 Residential density z-score −0.00 (0.99) −0.30 (0.69) −0.09 (0.66) −0.10 (0.51)
 Presence of a pocket/small park within 400 m (%)a 767 (43.3) 810 (58.6) 656 (55.8) 356 (65.8)
 Presence of a medium/large park within 400 m (%)b 781 (44.1) 725 (52.4) 618 (52.6) 294 (54.3)
 Presence of a district/regional park within 1,600 m (%)c 1,307 (73.8) 604 (43.7) 557 (47.4) 337 (62.3)
 Presence of a beach access point within 1,600 m (%) 197 (11.1) 90 (6.5) 75 (6.4) 54 (10.0)
Perceived measures
 Street connectivityd 3.40 (0.65) 3.54 (0.63) 3.56 (0.60) 3.62 (0.60)
 Infrastructure and safety for walkingd 3.09 (0.57) 3.14 (0.53) 3.12 (0.52) 3.17 (0.54)
 Traffic safetyd 3.61 (0.76) 3.90 (0.64) 3.81 (0.59) 3.85 (0.60)
 Neighborhood aestheticsd 3.48 (0.76) 3.83 (0.69) 3.61 (0.67) 3.59 (0.71)
 Crime safetyd 3.52 (0.78) 3.98 (0.60) 3.84 (0.61) 3.80 (0.67)
 Presence of a park within 5-min walk (%) 890 (50.3) 978 (70.7) 802 (68.2) 368 (68.0)
 Presence of a sports field within 15-min walk (%) 990 (55.9) 570 (41.2) 599 (50.9) 294 (54.3)
 Presence of a beach within 15-min walk (%) 211 (11.9) 116 (8.4) 94 (8.0) 51 (9.4)

Note: T1=baseline; T2=1-yr follow-up; T3=3-yr follow-up; T4=8-y follow-up.

aPocket/small park=≤0.5 ha.

bMedium/large park=≤5 ha.

cDistrict/regional park=≤15 ha.

dNEWS 5-point scale: Strongly disagree to Strongly agree.

Perceived and Objectively Measured Neighborhood Environment Attributes over Time

Three objectively measured neighborhood environment attributes (street connectivity, presence of pocket/small park, and presence of medium/large park within 400 m from home) improved after relocation and remained higher than baseline throughout follow-up (Table 2). Values for all other neighborhood environment attributes (e.g., shops and services, larger public open spaces) reduced immediately after relocation, but by 8-yr follow-up, had almost returned to, although did not exceed, baseline levels.

Participant’s perceptions of street connectivity, infrastructure and safety for walking, traffic safety, neighborhood esthetics, crime safety, and perceived access to a park within a 5-min walk from home all increased after relocation and, in most instances, continued to improve with each follow-up (Table 2). Perceived access to a sports field and the beach within a 15-min walk from home reduced after relocation, but had also almost returned to baseline levels at 8-yr follow-up.

Associations between Frequency of Recreational Walking and Objective and Perceived Measures of the Neighborhood Environment

Objective neighborhood environment measures.

Objectively measured neighborhood environment determinants of increased frequency of neighborhood recreational walking included access to a medium-/large-size park within 400 m from home and beach access within 1,600 m from home (both p<0.05; Table 3, Model 1). On average, participants with a medium-/large-size park within 400 m from home had a 10% higher frequency of recreational walking than participants who did not have a medium-/large-size park within 400 m from home (relative change: 1.10, 95% CI=1.03−1.18). Participants with a beach access point within 1,600 m from home had 11% higher frequency of recreational walking than participants without a beach access point within the same distance (relative change: 1.11, 95% CI=1.00−1.24). However, inexplicably, the presence of a district-/regional-size park within 1,600 m from home appeared to reduce the frequency of recreational walking (relative change: 0.92; 95% CI: 0.86, 0.99), compared with not having a district/regional park at all. Notably, the presence of a pocket/small park within 400 m was not a significant independent determinant of local recreational walking (all p>0.05), nor were street connectivity or residential density.

Table 3. Associations between frequency of walking for recreation per week and perceived and objective measures of the neighborhood environment.
Model 1ab Model 2ac Model 3ad
Relative change (95% CI) Relative change (95% CI) Relative change (95% CI)
Objectively measured
 Connectivity z-score 1.01 (0.99, 1.04) 0.98 (0.95, 1.01)
 Residential density z-score 1.00 (0.96, 1.04) 1.02 (0.98, 1.06)
 Presence of a pocket/small park within 400 me 1.00 (0.93, 1.07) 0.97 (0.91, 1.04)
 Presence of a medium/large park within 400 mf 1.10 (1.03, 1.18) 1.03 (0.96, 1.11)
 Presence of a district/regional park within 1,600 mg 0.92 (0.86, 0.99) 0.96 (0.90, 1.04)
 Presence of a beach access point within 1,600 m 1.11 (1.00, 1.24) 0.95 (0.84, 1.09)
Perceived measures
 Street connectivityh 1.12 (1.06, 1.19) 1.12 (1.06, 1.19)
 Infrastructure and safety for walkingh 1.02 (0.96, 1.09) 1.03 (0.96, 1.10)
 Traffic safetyh 1.03 (0.98, 1.08) 1.03 (0.98, 1.09)
 Neighborhood aestheticsh 1.20 (1.14, 1.27) 1.20 (1.14, 1.26)
 Crime safetyh 1.08 (1.02, 1.14) 1.08 (1.02, 1.14)
 Presence of a park within 5-min walk 1.12 (1.05, 1.20) 1.12 (1.05, 1.20)
 Presence of a sports field within 15-min walk 1.04 (0.97, 1.11) 1.04 (0.97, 1.11)
 Presence of a beach within 15-min walk 1.10 (1.00, 1.21) 1.14 (1.01, 1.28)
 Objective neighborhood environment indexi 1.01 (0.98, 1.03) 0.98 (0.95, 1.01)
 Perceived neighborhood environment indexj 1.12 (1.10, 1.14) 1.12 (1.10, 1.14)

Note: T1=baseline; T2=1-yr follow-up; T3=3-yr follow-up; T4=8-yr follow-up.

aAll models adjusted for baseline age, gender, marital status, level of education, occupation, hours of work/week, minutes/day of work travel, level of physical activity at work, children at home, and dog ownership.

bModel 1: Objective measures of the neighborhood environment only.

cModel 2: Perceived measures of the neighborhood environment only.

dModel 3: Objective and perceived measures of the neighborhood environment.

ePocket/small park=≤0.5 ha.

fMedium/large park=≤5 ha.

gDistrict/regional park=≤15 ha.

hNEWS 5-point scale: Strongly disagree to Strongly agree.

iPresence or absence (yes=1, no=0) of pocket/small park, medium/large park, district/regional park, beach access, and street connectivity and residential density (z-scores above baseline median score=1, z-scores below baseline median score=0); range: 0–6.

jPerceived presence (yes=1, no=0) of park, sports field, or beach, and perceived street connectivity, infrastructure, and safety for walking, neighborhood esthetics, traffic safety, and crime safety (z-scores above baseline median score=1, z-scores below baseline median score=0); range: 0–8.

Neighborhood perceptions.

The frequency of recreational walking was 12% higher in participants who perceived they had a park within a 5-min walk from home compared with participants who did not (relative change: 1.12; 95% CI: 1.05, 1.20) (Table 3, Model 2). Higher frequencies of recreational walking were also reported by participants who perceived they had beach access within a 15-min walk from home and higher levels street connectivity, neighborhood esthetics, and safety from crime (all p<0.05; relative change: 1.08 to 1.20). Perceived infrastructure and safety for walking, traffic safety, and presence of a sports field within a 15-min walk from home were not significant independent determinants of local recreational walking.

Objective and perceived neighborhood environments.

In models combining both objective and perceived neighborhood environment measures (Table 3, Model 3), all significant perceived measures from Model 2 remained significant, but associations with objective neighborhood measures were no longer significant. The frequency of local recreational walking increased by 12% for each additional neighborhood environment attribute perceived to be present (p<0.01; relative change=1.12). In contrast, after adjustment for neighborhood perceptions, no individual objective neighborhood environment measures, nor the cumulative index of exposures, remained significant determinants of local walking frequency in the fully adjusted model. None of the significant neighborhood environment variables in the three models had significantly different between-person (cross-sectional) and within-person (longitudinal) effect estimates (results not shown), suggesting that there was little confounding by uncontrolled time-constant factors, including time-constant self-selection factors.

Associations between Minutes of Recreational Walking and Objective and Perceived Measures of the Neighborhood Environment

The results for minutes of local recreational walking were generally similar to those for the frequency of local recreational walking with one exception; objectively measured connectivity remained statistically significant even after full adjustment (Tables 3 and 4, Model 1). Consistent with the frequency analyses, in Model 3 (adjusted for both objective and perceived measures), all perceived measures from Model 2 remained significant, while all but one objective measure (i.e., the presence of pocket/small parks within 400 m of home) remained significant determinants of the minutes of local recreational walking (Table 4, Model 3). Notably, participants with small/pocket parks within 400 m of their home undertook 6 min less recreational walking weekly compared with others. Conversely, participants who perceived they had access to a local park and access to the beach, and higher levels of street connectivity, neighborhood esthetics, and safety from crime reported significantly higher total minutes/week of recreational walking (absolute increases between 7 and 22 min; all p<0.05). Minutes of local recreational walking increased by almost 9 min/wk for each additional perceived neighborhood environment present (p<0.01).

Table 4. Associations between minutes of walking for recreation per week and perceived and objective measures of the neighborhood environment.
Model 1ab Model 2ac Model 3ad
Estimate (95% CI) Estimate (95% CI) Estimate (95% CI)
Objectively measured
 Connectivity z-score 3.29 (0.76, 5.82) 0.44 (–2.16, 3.04)
 Residential density z-score 0.48 (–3.38, 4.34) 1.59 (–2.28, 5.46)
 Presence of a pocket/small park within 400 me −3.87 (−9.79, 2.05) −6.24 (−12.11, −0.37)
 Presence of a medium/large park within 400 mf 7.50 (1.55, 13.45) 2.90 (−3.10, 8.91)
 Presence of a district/regional park within 1,600 mg −8.38 (−14.62, −2.13) −4.95 (−11.31, 1.42)
 Presence of a beach access point within 1,600 m 15.10 (4.31, 25.89) −4.43 (−18.20, 9.32)
Perceived measures
 Street connectivityh 8.57 (3.95, 13.18) 8.65 (3.99, 13.30)
 Infrastructure and safety for walkingh 4.75 (−0.78, 10.29) 4.70 (−0.98, 10.38)
 Traffic safetyh −0.02 (−4.41, 4.37) −0.23 (−4.64, 4.17)
 Neighborhood estheticsh 10.43 (6.03, 14.83) 10.44 (6.00, 14.87)
 Crime safetyh 7.85 (3.35, 12.35) 7.49 (2.91, 12.08)
 Presence of a park within 5-min walk 10.03 (4.07, 15.99) 9.52 (3.37, 15.67)
 Presence of a sports field within 15-min walk 2.44 (−3.27, 8.15) 2.72 (−3.09, 8.53)
 Presence of a beach within 15-min walk 19.37 (9.42, 29.33) 21.51 (8.59, 34.43)
 Objective neighborhood environment indexi 1.50 (−0.82, 3.82) −0.40  (−2.72, 1.92)
 Perceived neighborhood environment indexj 8.76 (7.09, 10.44) 8.81 (7.12, 10.50)

Note: T1=baseline; T2=1-yr follow-up; T3=3-yr follow-up; T4=8-yr follow-up.

aAll models adjusted for baseline age, gender, marital status, level of education, occupation, hours of work/week, minutes/day of work travel, level of physical activity at work, children at home, and dog ownership.

bModel 1: Objective measures of the neighborhood environment only.

cModel 2: Perceived measures of the neighborhood environment only.

dModel 3: Objective and perceived measures of the neighborhood environment.

ePocket/small park=≤0.5 ha.

fMedium/large park=≤5 ha.

gDistrict/regional park=≤15 ha.

hNEWS 5-point scale: Strongly disagree to Strongly agree.

iPresence or absence (yes=1, no=0) of pocket/small park, medium/large park, district/regional park, beach access, and street connectivity and residential density (z-scores above baseline median score=1, z-scores below baseline median score=0); range: 0–6.

jPerceived presence (yes=1, no=0) of park, sports field, or beach, and perceived street connectivity, infrastructure, and safety for walking, neighborhood esthetics, traffic safety, and crime safety (z-scores above baseline median score=1, z-scores below baseline median score=0); range: 0–8.

Mediation Relationship between Objective and Perceived Measures of the Neighborhood Environment and Recreational Walking

As the effects of objectively measured neighborhood environment exposures decreased when adjusted for perceptions, it was plausible that neighborhood perceptions mediated the relationship between objective measures and recreational walking. With the exception of small/pocket park access at time point 2, all objective measures were significantly (although mostly modestly) correlated with their corresponding perceived neighborhood environment measure (Table S1). Correlations between perceived and objectively measured beach access were substantially higher (range: 0.58–0.73) than for any other neighborhood attribute. The mediation analyses showed that the relationship between objective measures of the neighborhood environment and recreational walking was mediated by the perceived measures. Almost all estimates of objective measures of the neighborhood environment decreased and became nonsignificant in Model 3 when adjusted for neighborhood perceptions. Finally, objective measures of the neighborhood environment were associated with recreational walking when not adjusting for the perceived measure (Model 1), and perceived measures were related to recreational walking when not adjusted for the objective measure (Model 2), thus confirming that the perceived neighborhood environment mediates the relationship between the objectively measured neighborhood environment and local recreational walking.

Discussion

Consistent with numerous previous cross-sectional studies conducted in Australia and North America, we found that the frequency and duration of recreational walking was determined by objectively measured access to a medium-/large-size park (but not small pocket parks), beach access, and street connectivity (duration but not frequency of walking). However, these effects attenuated when adjusted for neighborhood perceptions. In fully adjusted models, positive perceptions of access to a park and beach and street connectivity, neighborhood esthetics, and safety from crime remained independent determinants of increased neighborhood recreational walking. Notably, local recreational walking increased by 9 min/wk (12% increase in frequency) for each additional perceived neighborhood attribute present. These latter findings are consistent with the International Physical Activity and the Environment Network 11 country study, which showed a linear gradient in the association between neighborhood environment perceptions and achieving recommended levels of physical activity (Sallis et al. 2009). With adjustment for unmeasured time-constant self-selection factors, we found between-person cross-sectional and within-person longitudinal effect estimates to be similar. This suggests that the longitudinal estimates observed of changes in the neighborhood environment on change in recreational walking can be interpreted in the same way as cross-sectional between-person effects.

Longitudinal analysis of RESIDE data 12 mo after relocating to a new neighborhood found that participants who favorably changed their neighborhood environment perceptions reported 2 mins/wk of additional recreational walking for each (range: 0–14) positive neighborhood perception (Giles-Corti et al. 2013). The 8-yr follow-up findings further highlight how local recreational walking is determined by resident perceptions about access to recreational opportunities (parks and the beach), street connectivity, safety from crime, and esthetics. Furthermore, recreational walking may increase over time as these positive perceptions grow. A previous cross-sectional study of RESIDE participants highlighted the importance of the quality of neighborhood environments (Sugiyama et al. 2015). This study found that walking to local public open space was associated with public open spaces being attractive with gardens, grassed areas, walking paths, water features, wildlife, amenities, dog-related facilities, and off-leash areas for dogs (Sugiyama et al. 2015). Notably, in the current study, we found that lower levels of walking were determined by the presence of smaller pocket parks, which is important, given that many subdivision design codes encourage the creational of smaller local public open spaces. Hence, providing residents with larger attractive public open space (Giles-Corti et al. 2005a; Sugiyama et al. 2010) and activating these spaces with social, cultural, and physical activity opportunities may positively impact on local recreational walking (de Blasio 2016). Nevertheless, the effects observed in this study are relatively modest, and greater efforts to encourage more residents to increase their physical activity are warranted. This is underscored by findings from Sugiyama and colleagues (Sugiyama et al. 2013), who found that while access to a park may assist active residents to maintain levels of walking, other strategies were required to encourage inactive participants to initiate physical activity. Encouraging local residents to initiate park use through park activation strategies (e.g., walking groups, permitting private service providers to use parks for physical activity programs, community music, or cultural events) may encourage residents to commence walking as well as increase their walking levels.

Participants who perceived they lived in an attractive neighborhood with recreational destinations, such as a park within a 5-min walk or a beach within a 15-min walk, did an additional 10 and 22 min/wk of local recreational walking, respectively (see Table 4). We observed similar unadjusted effect sizes for objectively measured park and beach access; however, these results were reduced when the perceptions were added to the model. These findings support the idea of the “coastal effect” reported by Bauman and colleagues, who found that those living in coastal suburbs were 27% more likely to meet recommended levels of physical activity (Bauman et al. 1999; White et al. 2014). The “coastal effect” might imply that those drawn to live by the coast may be those predisposed to be active; however, our longitudinal study suggests that the attractiveness of the ocean may also encourage those less predisposed to be more recreationally active. Although we did not investigate, similar findings may be found for access to walking paths along rivers and lakes, which would provide more residents with convivial recreational opportunities. From an equity perspective, there is a need for policies that facilitate access to pleasant recreational opportunities. This includes affordable housing located within close proximity to the beach, as well as developing walkways along other waterways, such as rivers and lakes. However, inequities in access to recreational destinations, such as the beach, could be reduced by ensuring people have access to high-quality larger parks. Although the effects were smaller, we found that access to a local medium-size park, but not smaller pocket parks, determines levels of recreational walking. Thus, in areas without proximate access to the beach or other waterways, additional resources are required for the design, management, and maintenance of higher-quality neighborhood parks. In disadvantaged areas, this may produce important cobenefits (de Blasio 2016). For example, a Scottish study found that increased access to green space decreased perceived and objectively measured stress levels in residents of disadvantaged areas (Roe et al. 2013). However, importantly, these parks must be well maintained, particularly in lower income areas. Another RESIDE cross-sectional study found that parks with disorder (e.g., graffiti, rubbish, and vandalism) discouraged recreational walking (Sugiyama et al. 2010), while an English study concluded that poorly maintained public open space may produce poorer mental health outcomes in lower income areas (Mitchell and Popham 2007). A number of pathways could explain these findings: areas with poorly maintained parks may become stigmatized, which may be internalized by residents, but may also increase levels of fear in residents. Both of these outcomes would be detrimental to the health and well-being of residents.

We have previously shown the potential walking benefits from gaining access to recreational destinations-and reported that this relationship is partly mediated by its effect on residents’ enjoyment of walking (Giles-Corti et al. 2013). Neighborhood esthetics are repeatedly associated with recreational walking: green, leafy suburbs with high-quality local open space provide attractive destinations and pleasant routes, and together can be powerful determinants of physical activity by encouraging more recreational walking (Sugiyama et al. 2012). There is already considerable evidence that park quality is a critical factor in encouraging more walking, while the presence of disorder (Foster et al. 2012) and high-speed traffic (Kaczynski et al. 2014) are deterrents (Sugiyama et al. 2010). Given the importance of a variety of public open spaces for increasing physical activity (Giles-Corti et al. 2009), a systems view is required to avoid any unintended consequences from only focusing on the needs of adults and ignoring the impact on the health and well-being of children and adolescents (Giles-Corti and King 2009).

This study appears to be one of the first longitudinal studies to show an association between street connectivity and local recreational walking. A 1-unit increase in the street connectivity z-score was associated with 3 additional min/wk of local recreational walking. This also supports cross-sectional findings showing that living in neighborhoods with high levels of street connectivity encourages more park-based activity (Kaczynski et al. 2014). Although we found access to smaller parks decreased recreational walking, others have found an interaction between park presence (even small parks) and connected street networks (Hooper et al. 2015b). People who reside in neighborhoods with more connected streets are able to access local destinations such as parks more easily because more connected street networks offer a more direct route to local destinations (Hooper et al. 2015b).

Our findings are important for achieving the UN’s Sustainable Development Goals particularly related to increasing access to green space to create resilient cities that will promote health. Our longitudinal analysis suggested that the presence of smaller parks was negatively associated with duration (but not frequency) of local recreational walking. This is not surprising, given that larger parks with more amenities are more interesting and offer users more opportunities and variety. Compared with smaller parks, larger parks may attract residents to walk further to visit them and to visit for longer once at the park, and they will have more amenities present that encourage users to be more active while there. It is possible that very small parks do not attract local residents to be physically active, although they may be important for some population subgroups (e.g., young children with parents, older adults, inner city residents). However, in established areas where there are only smaller local parks, particularly in areas with connected street networks that foster a range of alternative routes, it may be possible to generate a network of smaller local parks to create attractive recreational walking routes. Such a network may foster positive neighborhood perceptions, which our study shows determine recreational walking. Interventions of this type, particularly in lower socioeconomic areas where residents tend to undertake less recreational walking, are worth exploring, given the challenge of retrofitting established neighborhoods. However, more evidence is required to explore what role small parks play (if any) and for whom, and how the quality of smaller parks affects both physical activity and mental health outcomes (Francis et al. 2012). In the meantime, given the exponential growth of cities around the globe and global targets to increase access to public open space, both our longitudinal and cross-sectional findings suggest that fewer larger open spaces may be preferable than providing many small public open spaces (Sugiyama et al. 2010, 2015).

Our findings highlight that each individual attribute of the objectively measured neighborhood environment is associated with a relatively small increase in recreational walking per week. However, they are potentially more significant when their combined effects are considered. We found that local recreational walking increased by 9 min/wk (12% increase in frequency) for each additional perceived neighborhood attribute present. This suggests that improving just one attribute of the neighborhood environment may have little effect on residents’ overall walking levels (Hooper et al. 2015a, 2015b). Rather, it is likely that recreational walking is determined by a combination of factors, that is, having access to safe attractive neighborhoods with highly connected street networks and good access to larger local public open space. In addition, individual attitudes and abilities and the social environment cannot be ignored. Previously, we have found that individual, social, and built environment factors are equally important (Giles-Corti and Donovan 2003). Hence, a combination of intervention strategies are required to maximize local walking levels, and simply having one attractive park alone may be insufficient to change the behavior of residents (Sugiyama et al. 2013). It may be that interventions designed to encourage more park use and encourage inactive residents to use local facilities are warranted. Future natural experiments might investigate both the individual and combined effects of changes to the neighborhood environment on recreational walking and the added benefit of implementing strategies to encourage residents to make better use of their local recreational destinations. Moreover, examining the effect of changes to the neighborhood environment for other domains of physical activity is warranted to better understand the potential tradeoff (if any) of neighborhood environment features for recreational compared with transport-related physical activities and whether local neighborhood attributes encourage other types of physical activity in adults, children, and adolescents.

Our mediation analysis results show that participants’ perceptions of the neighborhood environment mediate the relationship between objectively measured attributes of the neighborhood environment and local recreational walking. However, in this study, not all neighborhood environment perceptions (e.g., neighborhood esthetics and safety from crime) had a corresponding objectively measured neighborhood environment variable. Thus, future longitudinal studies should consider including both perceived and objective measures of crime and neighborhood attractiveness to fully explore mediating relationships. Notably, our findings do not suggest that objective measures of the neighborhood environment are not important determinants of local recreational walking. Rather, they highlight that individuals’ perceptions of their environment are more proximal determinants of behavior change than the environment. Thus, they exert a greater influence on walking behavior. Yet, the actual neighborhood environment likely determines these neighborhood perceptions (Giles-Corti and Donovan 2002; Giles-Corti et al. 2005b). Future studies may wish to further explore the importance of this relationship, and interventions need to focus on people and places to maximize physical activity outcomes (Giles-Corti 2006).

Study Limitations

A potential source of bias was the lower participant retention rate at the fourth survey, which may have been due to a considerably longer survey than in previous waves. Analysis of the factors associated with participant dropout was associated with various demographic variables (being younger and male, having children at home), but not with prior walking behavior. Dropout patterns such as this (conditionally on the covariates) are called missing completely at random. Estimates from our longitudinal repeated measures models are unbiased under this pattern of dropout, provided the covariates related to dropout are included in the models as time-varying covariates and there are no further unmeasured covariates related to dropout (Fitzmaurice et al. 2012).

This study was not able to examine the impact of other relevant objectively measured built environment attributes, as data was not available across all time points. For example, future studies should consider evaluating the impact of footpath length, traffic exposure, quality of public open space (which in this study was only measured at two but not four time points), neighborhood greenness, and terrain on local recreational walking. This study was also limited by the potential confounding from unmeasured time-varying factors, such as general health status. Finally, the study site, Perth, Western Australia, represents an above-average quality of life and a low-density city with a Mediterranean climate. Thus, findings may not be representative of other lower-income medium- to high-density cities with different climatic conditions. For instance, patterns of walking tend to differ by neighborhood disadvantage, whereby residents in relatively disadvantaged areas tend to walk less for recreation, but more for transport than those in advantaged areas (Miles et al. 2008; Turrell et al. 2010, 2013). Further, residents in low-income settings face an increased exposure to many potential barriers to recreational walking, such as crime and disorder, poorer upkeep and esthetics (Foster et al. 2011, 2015), and increased traffic (Giles-Corti and Donovan 2002). Disorder, in particular, tends to cluster near nonresidential land uses (e.g., shops and parks) (Perkins et al. 1992), so even when access is equitable, the quality of these spaces may not be, thereby potentially rendering these sites less appealing as walking settings/destinations.

Conclusions

Longitudinal repeated measures studies allow a stronger assessment of the effect of the built environment on recreational walking than cross-sectional studies. Perceived neighborhood attractiveness and safety influences recreational walking, and this influence grows with more positive perceptions. The provision of larger (but not small) public open space and street connectivity supports more recreational walking. In areas with smaller parks and connected street networks, it may be possible to encourage recreational walking by creating a network of smaller parks. These findings provide urban planners and policy makers with stronger evidence of a causal relationship between the neighborhood environment and recreational walking, and highlight specific interventions required to increase recreational walking and meet residents’ health needs when planning new or retrofitting established neighborhoods.

Acknowledgments

The RESIDE Study was funded by grants from the Western Australian Health Promotion Foundation (Healthway) (#11828), the Australian Research Council (LP0455453), and an Australian National Health and Medical Research Council (NHMRC) Capacity Building Grant (#458688). H.C. is supported by an NHMRC/National Heart Foundation Early Career Fellowship (#1036350) and National Heart Foundation Future Leader Fellowship (#100794). B.G.C. was supported by an NHMRC Senior Principal Research Fellow Award (#1107672). S.F. is supported by an ARC Discovery Early Career Researcher Award (DE160100140). P.H. was supported by National Health and Medical Research Council (NHMRC) Capacity Building Grant (#458688) and the Centre for Research Excellence in Healthy Liveable Communities (#1061404). The authors gratefully acknowledge the GIS team (N. Middleton, S. Hickey and B. Beasley) for their assistance in the development of the GIS measures used in this study.

References

Allison PD. 2005. Fixed Effects Regression Methods for Longitudinal Data Using SAS. Cary, NC:SAS Institute Inc.

Bancroft C, Joshi S, Rundle A, Hutson M, Chong C, Weiss CC, et al. 2015. Association of proximity and density of parks and objectively measured physical activity in the United States: A systematic review. Soc Sci Med 138:22–30, PMID: 26043433, 10.1016/j.socscimed.2015.05.034.

Bauman A, Smith B, Stoker L, Bellew B, Booth M. 1999. Geographical influences upon physical activity participation: Evidence of a ‘coastal effect.’ Aust N Z J Public Health 23(3):322–324, PMID: 10388181, 10.1111/j.1467-842X.1999.tb01265.x.

Boone-Heinonen J, Guilkey DK, Evenson KR, Gordon-Larsen P. 2010. Residential self-selection bias in the estimation of built environment effects on physical activity between adolescence and young adulthood. Int J Behav Nutr Phys Act 7:70, PMID: 20920341, 10.1186/1479-5868-7-70.

Bull F, Beesley B, Hooper P, Boruff BB, Wood L. 2013. Public Open Space (POS) Tool. Perth, Australia:
Centre for Built Environment and Health, School of Population Health, The University of Western Australia.

Cerin E, Saelens BE, Sallis JF, Frank LD. 2006. Neighborhood Environment Walkability Scale: Validity and development of a short form. Med Sci Sports Exerc 38(9):1682–1691, PMID: 16960531, 10.1249/01.mss.0000227639.83607.4d.

Cheng JJ, Berry P. 2013. Health co-benefits and risks of public health adaptation strategies to climate change: A review of current literature. Int J Public Health 58(2):305–311, PMID: 23111371, 10.1007/s00038-012-0422-5.

Christian H, Bull FC, Middleton NJ, Knuiman MW, Divitini ML, Hooper P, et al. 2011. How important is the land use mix measure in understanding walking behaviour? Results from the RESIDE study. Int J Behav Nutr Phys Act 8:55, PMID: 21631958, 10.1186/1479-5868-8-55.

Christian H, Klinker CD, Villanueva K, Knuiman MW, Foster SA, Zubrick SR, et al. 2015. The effect of the social and physical environment on children’s independent mobility to neighborhood destinations. J Phys Act Health 12(suppl 1):S84–S93, PMID: 25599389, 10.1123/jpah.2014-0271.

Das P, Horton R. 2012. Rethinking our approach to physical activity. Lancet 380(9838):189–190, PMID: 22818931, 10.1016/S0140-6736(12)61024-1.

de Blasio B. 2016. Healthier neighbourhoods through healthier parks. Lancet 388(10062):2850–2851, PMID: 27671665, 10.1016/S0140-6736(16)31579-3.

Ding D, Gebel K. 2012. Built environment, physical activity, and obesity: What have we learned from reviewing the literature? Health Place 18(1):100–105, PMID: 21983062, 10.1016/j.healthplace.2011.08.021.

Duhl LJ. 1996. An ecohistory of health: The role of ‘Healthy Cities.’ Am J Health Promot 10(4):258–261, PMID: 10159706, 10.4278/0890-1171-10.4.258.

Fitzmaurice G, Laird N, Ware J. 2012. Applied Longitudinal Analysis. 2nd edition.Hoboken, NJ:John Wiley & Sons.

Foster S, Giles-Corti B, Knuiman M. 2011. Creating safe walkable streetscapes: Does house design and upkeep discourage incivilities in suburban neighbourhoods?. J Environ Psychol 31(1):79–88, 10.1016/j.jenvp.2010.03.005.

Foster S, Giles-Corti B, Knuiman M. 2012. Does fear of crime discourage walkers? A social-ecological exploration of fear as a deterrent to walking. Environ Behav 46(6):698–717, 10.1177/0013916512465176.

Foster S, Wood L, Francis J, Knuiman M, Villanueva K, Giles-Corti B. 2015. Suspicious minds: Can features of the local neighbourhood ease parents’ fears about stranger danger? J Environ Psychol 42:48–56, 10.1016/j.jenvp.2015.02.001.

Francis J, Wood LJ, Knuiman M, Giles-Corti B. 2012. Quality or quantity? Exploring the relationship between Public Open Space attributes and mental health in Perth, Western Australia. Soc Sci Med 74(10):1570–1577, PMID: 22464220, 10.1016/j.socscimed.2012.01.032.

Frank LD, Schmid TL, Sallis JF, Chapman J, Saelens BE. 2005. Linking objectively measured physical activity with objectively measured urban form: Findings from SMARTRAQ. Am J Prev Med 28(2 suppl 2):117–125, PMID: 15694519, 10.1016/j.amepre.2004.11.001.

Giles-Corti B. 2006. People or places: What should be the target? J Sci Med Sport 9(5):357–366, PMID: 16931155, 10.1016/j.jsams.2006.06.021.

Giles-Corti B, Broomhall MH, Knuiman M, Collins C, Douglas K, Ng K, et al. 2005a. Increasing walking: How important is distance to, attractiveness, and size of public open space? Am J Prev Med 28(2 suppl 2):169–176, PMID: 15694525, 10.1016/j.amepre.2004.10.018.

Giles-Corti B, Bull F, Knuiman M, McCormack G, Van Niel K, Timperio A, et al. 2013. The influence of urban design on neighbourhood walking following residential relocation: Longitudinal results from the RESIDE study. Soc Sci Med 77:20–30, PMID: 23206559, 10.1016/j.socscimed.2012.10.016.

Giles-Corti B, Donovan RJ. 2002. Socioeconomic status differences in recreational physical activity levels and real and perceived access to a supportive physical environment. Prev Med 35(6):601–611, PMID: 12460528, 10.1006/pmed.2002.1115.

Giles-Corti B, Donovan RJ. 2003. Relative influences of individual, social environmental, and physical environmental correlates of walking. Am J Public Health 93(9):1583–1589, PMID: 12948984, 10.2105/AJPH.93.9.1583.

Giles-Corti B, Kelty SF, Zubrick SR, Villanueva KP. 2009. Encouraging walking for transport and physical activity in children and adolescents: How important is the built environment?. Sports Med 39(12):995–1009, PMID: 19902982, 10.2165/11319620-000000000-00000.

Giles-Corti B, King AC, 2009. Creating active environments across the life course: “Thinking outside the square”. Br J Sports Med 43(2):109–113, PMID: 19136501, 10.1136/bjsm.2008.054700.

Giles-Corti B, Knuiman M, Timperio A, Van Niel K, Pikora TJ, Bull FC, et al. 2008. Evaluation of the implementation of a state government community design policy aimed at increasing local walking: Design issues and baseline results from RESIDE, Perth Western Australia. Prev Med 46(1):46–54, PMID: 17881044, 10.1016/j.ypmed.2007.08.002.

Giles-Corti B, Timperio A, Bull F, Pikora T. 2005b. Understanding physical activity environmental correlates: Increased specificity for ecological models. Exerc Sports Sci Rev 33(4):175–181, PMID: 16239834, http://journals.lww.com/acsm-essr/Fulltext/2005/10000/Understanding_Physical_Activity_Environmental.5.aspx.

Giles-Corti B, Timperio A, Cutt H, Pikora TJ, Bull FC, Knuiman M, et al. 2006. Development of a reliable measure of walking within and outside the local neighborhood: RESIDE’s Neighborhood Physical Activity Questionnaire. Prev Med 42(6):455–459, PMID: 16574208, 10.1016/j.ypmed.2006.01.019.

Giles-Corti B, Vernez-Moudon A, Reis R, Turrell G, Dannenberg AL, Badland H, et al. 2016. City planning and population health: A global challenge. Lancet 388(10062):2912–2924, PMID: 27671668, 10.1016/S0140-6736(16)30066-6.

Goodman A, Sahlqvist S, Ogilvie D, iConnect Consortium. 2013. Who uses new walking and cycling infrastructure and how? Longitudinal results from the UK iConnect study. Prev Med 57(5):518–524, PMID: 23859933, 10.1016/j.ypmed.2013.07.007.

Halonen JI, Stenholm S, Kivimäki M, Pentti J, Subramanian SV, Kawachi I, et al. 2015. Is change in availability of sports facilities associated with change in physical activity? A prospective cohort study. Prev Med 73:10–14, PMID: 25602907, 10.1016/j.ypmed.2015.01.012.

Hirsch JA, Diez Roux AV, Moore KA, Evenson KR, Rodriguez DA. 2014. Change in walking and body mass index following residential relocation: The multi-ethnic study of atherosclerosis. Am J Public Health 104(3):49–56, PMID: 24432935, 10.2105/AJPH.2013.301773.

Hooper P, Knuiman M, Bull F, Jones E, Giles-Corti B. 2015a. Are we developing walkable suburbs through urban planning policy? Identifying the mix of design requirements to optimise walking outcomes from the ‘Liveable Neighbourhoods’ planning policy in Perth, Western Australia. Int J Behav Nutr Phys Act 12(1):63, PMID: 25981916, 10.1186/s12966-015-0225-1.

Hooper P, Knuiman M, Foster S, Giles-Corti B. 2015b. The building blocks of a ‘Liveable Neighbourhood’: Identifying the key performance indicators for walking of an operational planning policy in Perth, Western Australia. Health Place 36:173–183, PMID: 26606456, 10.1016/j.healthplace.2015.10.005.

Kaczynski AT, Koohsari MJ, Stanis SA, Bergstrom R, Sugiyama T. 2014. Association of street connectivity and road traffic speed with park usage and park-based physical activity. Am J Health Promot 28(3):197–203, PMID: 23875985, 10.4278/ajhp.120711-QUAN-339.

Kleinert S, Horton R. 2015. Rethinking and reframing obesity. Lancet 385(9985):2326–2328, PMID: 25703115, 10.1016/S0140-6736(15)60163-5.

Knuiman M, Christian HE, Divitini ML, Foster SA, Bull FC, Badland HM, et al. 2014. A longitudinal analysis of the influence of the neighborhood built environment on transport walking: The RESIDE Study. Am J Epi 180(5):453–461, PMID: 25117660, 10.1093/aje/kwu171.

McCormack GR, Shiell A. 2012. In search of causality: A systematic review of the relationship between the built environment and physical activity among adults. Int J Behav Nutr Phys Act 8:125, PMID: 22077952, 10.1186/1479-5868-8-125.

Miles R, Panton LB, Jang M, Haymes EM. 2008. Residential context, walking and obesity: Two African-American neighborhoods compared. Health Place 14(2):275–286, PMID: 17822941, 10.1016/j.healthplace.2007.07.002.

Mitchell R, Popham F. 2007. Greenspace, urbanity and health: Relationships in England. J Epidemiol Community Health 61(8):681–683, PMID: 17630365, 10.1136/jech.2006.053553.

Panter J, Griffin S, Dalton AM, Ogilvie D. 2013. Patterns and predictors of changes in active commuting over 12 months. Prev Med 57(6):776–784, PMID: 23938464, 10.1016/j.ypmed.2013.07.020.

Perkins DD, Meeks JW, Taylor RB. 1992. The physical environment of street blocks and resident perceptions of crime and disorder: Implications for theory and measurement. J Environ Psychol 12(1):21–34, 10.1016/S0272-4944(05)80294-4.

Ranchod YK, Diez Roux AVD, Evenson KR, Sánchez BN, Moore K. 2014. Longitudinal associations between neighborhood recreational facilities and change in recreational physical activity in the multi-ethnic study of atherosclerosis, 2000–2007. Am J Epi 179(3):335–343, PMID: 24227016, 10.1093/aje/kwt263.

Roe JJ, Thompson CW, Aspinall PA, Brewer MJ, Duff EI, Miller D, et al. 2013. Green space and stress: Evidence from cortisol measures in deprived urban communities. Int J Environ Res Public Health 10(9):4086–4103, PMID: 24002726, 10.3390/ijerph10094086.

Saelens B, Handy S. 2008. Built environment correlates of walking: A review. Med Sci Sports Exerc 40(7 suppl):S550–S566, PMID: 18562973, 10.1249/MSS.0b013e31817c67a4.

Sallis JF, Bowles HR, Bauman A, Ainsworth BE, Bull FC, Craig CL, et al. 2009. Neighborhood environments and physical activity among adults in 11 countries. Am J Prev Med 36(6):484–490, PMID: 19460656, 10.1016/j.amepre.2009.01.031.

Sallis JF, Bull F, Burdett R, Frank LD, Griffiths P, Giles-Corti B, et al. 2016. Use of science to guide city planning policy and practice: How to achieve healthy and sustainable future cities. Lancet 388(10062):2936–2947, PMID: 27671670, 10.1016/S0140-6736(16)30068-X.

Stevenson M, Thompson J, de Sá TH, Ewing R, Mohan D, McClure R, et al. 2016. Land use, transport, and population health: Estimating the health benefits of compact cities. Lancet 388(10062):2925–2935, PMID: 27671671, 10.1016/S0140-6736(16)30067-8.

Sugiyama T, Francis J, Middleton NJ, Owen N, Giles-Corti B. 2010. Associations between recreational walking and attractiveness, size, and proximity of neighborhood open spaces. Am J Public Health 100(9):1752–1757, PMID: 20634455, 10.2105/AJPH.2009.182006.

Sugiyama T, Giles-Corti B, Summers J, du Toit L, Leslie E, Owen N. 2013. Initiating and maintaining recreational walking: A longitudinal study on the influence of neighborhood green space. Prev Med 57(3):178–182, PMID: 23732245, 10.1016/j.ypmed.2013.05.015.

Sugiyama T, Gunn LD, Christian H, Francis J, Foster S, Hooper P, et al. 2015. Quality of public open spaces and recreational walking. Am J Public Health 105(12):2490–2495, PMID: 26469676, 10.2105/AJPH.2015.302890.

Sugiyama T, Neuhaus M, Cole R, Giles-Corti B, Owen N. 2012. Destination and route attributes associated with adults’ walking: A review. Med Sci Sports Exerc 44(7):1275–1286, PMID: 22217568, 10.1249/MSS.0b013e318247d286.

Turrell G, Haynes M, Burton NW, Giles-Corti B, Oldenburg B, Wilson LA, et al. 2010. Neighborhood disadvantage and physical activity: Baseline results from the HABITAT multilevel longitudinal study. Ann Epidemiol 20(3):171–181, PMID: 20159488, 10.1016/j.annepidem.2009.11.004.

Turrell G, Haynes M, Wilson L-A, Giles-Corti B. 2013. Can the built environment reduce health inequalities? A study of neighbourhood socioeconomic disadvantage and walking for transport. Health Place 19:89–98, PMID: 23207291, 10.1016/j.healthplace.2012.10.008.

Turrell G, Hewitt B, Haynes M, Nathan A, Giles-Corti B. 2014. Change in walking for transport: A longitudinal study of the influence of neighbourhood disadvantage and individual-level socioeconomic position in mid-aged adults. Int J Behav Nutr Phys Act 11:151–157, PMID: 25492854, 10.1186/s12966-014-0151-7.

United Nations. 2015. Sustainable Development Goals: 17 Goals to Transform our World. http://www.un.org/sustainabledevelopment/sustainable-development-goals/ [accessed 9 December 2016].

United Nations. 2016. Habitat III Conference – UN-Habitat. http://unhabitat.org/habitat-iii-conference/ [accessed 9 December 2016].

Watts N, Adger WN, Agnolucci P, Blackstock J, Byass P, Cai W, et al. 2015. Health and climate change: Policy responses to protect public health. Lancet 386(10006):1861–1914, PMID: 26111439, 10.1016/S0140-6736(15)60854-6.

Western Australian Department of Sport and Recreation. 2012. Classification Framework for Public Open Space. Perth, Australia:WA Department of Sport and Recreation.

White MP, Wheeler BW, Herbert S, Alcock I, Depledge MH. 2014. Coastal proximity and physical activity: Is the coast an under-appreciated public health resource? Prev Med 69:135–140, PMID: 25284259, 10.1016/j.ypmed.2014.09.016.

World Health Organization. 2016. Shanghai Declaration on promoting health in the 2030 Agenda for Sustainable Development. In: Proceedings of the 9th Global Conference on Health Promotion, 2016. Shanghai, 21-24 November 2016. World Health Organization [accessed 13 December 2016].

First-Trimester Urinary Bisphenol A Concentration in Relation to Anogenital Distance, an Androgen-Sensitive Measure of Reproductive Development, in Infant Girls

Author Affiliations open

1Department of Epidemiology, Environmental and Occupational Health Sciences Institute, Rutgers University School of Public Health, Piscataway, New Jersey, USA

2Department of Obstetrics and Gynecology, University of Rochester School of Medicine and Dentistry, Rochester, New York, USA

3Department of Pediatrics, University of Washington, Seattle, Washington, USA

4Seattle Children’s Research Institute, Seattle, Washington, USA

5Department of Biostatistics and Computational Biology, University of Rochester School of Medicine and Dentistry, Rochester, New York, USA

6Department of Medicine, University of Minnesota, Minneapolis, Minnesota, USA

7Department of Epidemiology and Community Health, University of Minnesota, Minneapolis, Minnesota, USA

8Department of Environmental Medicine and Public Health, Icahn School of Medicine at Mount Sinai, New York, New York, USA

PDF icon PDF Version (281 KB)

  • Introduction:
    Evidence from animal models suggests that prenatal exposure to bisphenol A (BPA), a ubiquitous endocrine-disrupting chemical, is associated with adverse reproductive outcomes in females. Exposure during early gestation, a critical period for reproductive development, is of particular concern. Anogenital distance (AGD) is a sensitive biomarker of the fetal hormonal milieu and a measure of reproductive toxicity in animal models. In some studies, the daughters of BPA-exposed dams have shorter AGD than controls. Here, we investigate this relationship in humans.
    Methods:
    BPA was assayed in first-trimester urine samples from 385 participants who delivered infant girls in a multicenter pregnancy cohort study. After birth, daughters underwent exams that included two measures of AGD (AGD-AC: distance from center of anus to clitoris; AGD-AF: distance from center of anus to fourchette). We fit linear regression models to examine the association between specific gravity–adjusted (SPG-adj) maternal BPA concentrations and infant AGD, adjusting for covariates.
    Results:
    BPA was detectable in 94% of women. In covariate-adjusted models fit on 381 eligible subjects, the natural logarithm of SpG-adj maternal BPA concentration was inversely associated with infant AGD-AC [β=−0.56, 95% confidence interval (CI): −0.97, −0.15]. We observed no association between maternal BPA and infant AGD-AF.
    Conclusion:
    BPA may have toxic effects on the female reproductive system in humans, as it does in animal models. Higher first-trimester BPA exposure was associated with significantly shorter AGD in daughters, suggesting that BPA may alter the hormonal environment of the female fetus. https://doi.org/10.1289/EHP875
  • Received: 27 July 2016
    Revised: 18 January 2017
    Accepted: 19 January 2017
    Published: 11 July 2017

    Address correspondence to E. S. Barrett, Dept. of Epidemiology, Environmental and Occupational Health Sciences Institute, Rutgers University School of Public Health, 170 Frelinghuysen Road, Piscataway, NJ 08854 USA, Telephone: 848-445-0197, Fax: 732-445-0784 USA, Email: Emily.barrett@eohsi.rutgers.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP875).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (719 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (113 KB)

Introduction

Bisphenol A (BPA) is a synthetic chemical widely used in consumer products, including food and drink containers, thermal receipts, medical equipment, and other plastic products (CDC 2013). BPA is detectable in over 90% of the population in the United States (Calafat et al. 2008), and may act on the endocrine system in numerous ways, including binding to and activating numerous nuclear and membrane endocrine receptors, and stimulating changes in estrogen, androgen, progesterone, and thyroid hormone activity (Gentilcore et al. 2013; Jones et al. 2016; Rehan et al. 2015; Sohoni and Sumpter 1998; Teng et al. 2013; Vandenberg et al. 2009). Dozens of studies in humans have examined BPA exposure in relation to a wide range of health end points, including reproductive, perinatal, and pediatric outcomes. That epidemiological research is complemented by findings from animal models and in vitro studies indicating that many tissues and organ systems (including the mammary gland, prostate gland, adipose tissue, reproductive system, and brain) are sensitive to BPA (Ariemma et al. 2016; Berger et al. 2016; de Lima et al. 2015; Vandenberg et al. 2007; Wolstenholme et al. 2011).

In animal models and humans, BPA can cross the placenta to enter fetal circulation (Balakrishnan et al. 2010; Gerona et al. 2013; Ikezuki et al. 2002; Nishikawa et al. 2010; Takahashi and Oishi 2000). Because fetal development is a period of rapid cell proliferation and differentiation, tissue development, and organ growth, prenatal exposure to environmental chemicals such as BPA may be of particular concern. In humans, linking prenatal BPA exposure to postnatal outcomes can be challenging for many reasons, including the long lag between exposure and the outcomes of interest. This is particularly true of reproductive end points because decades may lapse between prenatal exposure and outcomes such as pubertal development, sex steroid production, and, ultimately, fertility.

However, even in the case of these reproductive end points, there are early indicators that may reflect signal altered development. For example, anogenital distance (AGD), the distance from the anus to the genitals, can be measured from birth and is important because a) it is a well-documented index of prenatal androgen exposure in animal models; and b) in humans, it has been linked to clinically relevant measures of adult reproductive health in both sexes (Castaño-Vinyals et al. 2012; Eisenberg et al. 2012; Mendiola et al. 2011; Mendiola et al. 2016). Across numerous mammalian species (including humans), AGD is 50–100% longer in males than in females, and in animal models, it is responsive to experimental manipulation of the prenatal endocrine environment (Dean and Sharpe 2013). Here, we focus on females, acknowledging there is an even larger literature on AGD in males. When androgens are administered to a pregnant dam, her female offspring have longer, more masculine AGD than controls (Hotchkiss et al. 2007). In humans, an early study observed that three infant girls with congenital adrenal hyperplasia (a condition characterized by supranormal adrenal androgen exposures in utero) all had AGD ratios above the 95% confidence limit, as established based on measurements in 115 healthy infant girls (Callegari et al. 1987). We have reported, furthermore, that infant girls born to mothers with polycystic ovary syndrome (PCOS) (n=24), another hyperandrogenic condition, have longer AGD at birth than infant girls born to a PCOS-free comparison group (n=232) (Barrett et al. 2016). In a study of 100 young adult women, longer AGD was associated with higher testosterone levels (Mira-Escolano et al. 2014), as well as multifollicular ovaries (Mendiola et al. 2012). Most recently, a case–control study demonstrated that women with endometriosis (n=114) had shorter AGD than controls (n=105) (Mendiola et al. 2016).

Thus, AGD may be an important marker of prenatal endocrine-disrupting exposures and potentially a predictor of reproductive sequelae in females. Several studies in rodents have examined the effect of low-dose maternal BPA exposure on AGD in female offspring, but the results have been inconsistent. In a rat study, decreases in female AGD at birth were observed following prenatal and lactational BPA exposure at 0.025 mg/kg body weight (BW) per d, an amount well below the no observed adverse effect level of 5 mg/kg BW per d (Christiansen et al. 2014), but still considerably higher than the likely human daily exposure, estimated at 0.04–1.5 μg/kg BW per d (EFSA 2015; Lakind and Naiman 2011; WHO 2011). In another rat study, at age 1 mo, AGD was shorter in the female offspring of dams exposed to BPA (at doses of 0.17 and 1.7 mg/kg BW per d) compared with controls; however, the differences in AGD were attenuated by 3 mo of age (Kobayashi et al. 2012). At the same time, a handful of other studies in mice and rats have observed no significant changes in AGD in female offspring following gestational BPA exposure at similar or lower concentrations (2–200 μg/kg BW per d) (Honma et al. 2002; Howdeshell et al. 2008; Ryan et al. 2010). Differences in timing of exposure, dose, strain, sample size, and age at AGD measurement all may contribute to the inconsistencies across studies.

To date, very few studies have examined the relationship between prenatal BPA exposure and AGD in humans, and thus far, all of them have focused on male offspring. In a Chinese study, the sons of workers with occupational BPA exposure during pregnancy (n=56) had shorter AGD than the sons of controls who worked in related industries (n=97) (Miao et al. 2011). However, maternal urinary BPA (a more precise exposure metric) was not measured, and AGD was measured in the sons at a wide range of ages (0–17 yr). By contrast, in a second study, maternal urinary BPA concentration in late pregnancy was not associated with AGD in male infants at birth (n=137), but it was inversely associated with testosterone concentrations and the testosterone-to-estradiol ratio in cord blood (Liu et al. 2016). In this case, the timing of urine and blood collection was outside of the period of greatest relevance, the reproductive programming window (estimated to be approximately 8–14 weeks gestation), during which AGD appears to be most responsive to exposures (Welsh et al. 2008). The objective of the current study was to expand upon this small literature by using data from a large pregnancy cohort study to examine the relationship between maternal BPA concentrations in the early pregnancy reproductive programming window and AGD in the resulting daughters at birth.

Methods

Study Population and Overview

The Infant Development and the Environment Study (TIDES) was a pregnancy cohort study designed to examine exposure to endocrine disrupting chemicals in relation to infant reproductive development. Women in their first trimester of pregnancy were recruited in 2010–2012 at four academic medical centers: University of California, San Francisco (UCSF), University of Minnesota (UMN), University of Rochester Medical Center (New York) (URMC), and University of Washington (UW). Eligibility criteria included: age 18 or older, able to read and write English, no major medical complications, <13 weeks pregnant, and planning to deliver in a participating study hospital. In each trimester, subjects gave urine samples and completed questionnaires, which included items on demographics, health, lifestyle, and reproductive history. All study activities were approved by the relevant institutional review boards prior to study implementation, and all subjects signed informed consent. The current analysis includes TIDES subjects who gave a first-trimester urine sample and went on to deliver a daughter who underwent a TIDES physical examination shortly after birth. Gestational age at birth was determined based on the first ultrasound in the medical record. When that was not available, the physician’s estimate of gestational age at birth was used instead.

Bisphenol A Measurement and Analysis

Urine samples were collected in BPA-free containers and frozen at −80°C until they were shipped on dry ice to the Division of Laboratory Sciences, National Center for Environmental Health at the Centers for Disease Control and Prevention (CDC). Due to funding constraints, BPA was only analyzed in first trimester urine samples from mothers who gave birth to girls. At the CDC, total urinary BPA (free plus conjugated species) was measured using online solid phase extraction–high-performance liquid chromatography–isotope dilution mass spectrometry (Ye et al. 2005). For quality control, each batch also included field blanks, reagent blanks, analytical standards, and matrix-based quality control materials.

Samples with BPA concentrations below the limit of detection (0.07 μg/L) were assigned a value of LOD divided by the square root of 2, following convention (Hornung and Reed 1990). We adjusted for urine dilution using a standard formula: BPAadj=BPA[(1.014−1)/(SpG−1)], where BPAadj is the specific gravity–adjusted (SpG-adj) BPA concentration, BPA is the ssconcentration measured in the individual sample, 1.014 is the mean SpG for all TIDES samples, and SpG is the SpG of the individual urine sample (Boeniger et al. 1993). SpG-adj BPA concentrations were then natural log–transformed.

Infant Physical Examinations and Anogenital Distance

Prior to hospital discharge (typically at 1–2d of age), the study team visited the TIDES mother and child to conduct the infant physical examination. For infants born preterm or fragile, exams were delayed until the clinical team felt the infant was ready. TIDES exams consisted of weight and length measurements (following standardized protocols), as well as comprehensive genital exams conducted by experienced examiners who had undergone 2-d, face-to-face, multicenter intensive standardized trainings at the beginning and middle of the study (Sathyanarayana et al. 2015). AGD was measured following protocols developed based on our previous work (Swan et al. 2005; Swan 2008). The infant was placed flat on her back with her legs held in a “frog-leg” position. Trained study coordinators obtained two measurements of AGD (in mm) using dial calipers, shown in Figure 1. The shorter measurement, the anus–fourchette distance (AGD-AF), is from the posterior end of the fourchette to the center of the anus (1). The longer measurement, the anus–clitoris distance (AGD-AC), is from the anterior surface of the clitoral hood to the center of the anus (Figure 1). Each measurement was repeated three times, and the mean of the three measurements was used in data analysis. At each study center, measurements were independently repeated by a different examiner in at least 10% of infants to assess quality control. The repeated measurements allowed us to quantify inter- and intraobserver (intraclass correlations) variation (as further described by Sathyanarayana et al. 2015). On a monthly basis, the TIDES coordinating center assessed interexaminer and intercenter variation, and any issues were immediately addressed to ensure data quality.

Conceptual diagram
Figure 1. Measurement of anogenital distance in female newborns [adapted from Sathyanarayana et al. (2010) and reprinted with permission from John Wiley and Sons].

Statistical Methods

We first calculated univariate statistics for all variables of interest, including counts and percentages for categorical variables, means, and summary statistics (minimum, 25th percentile, median, 75th percentile, maximum) for continuous variables (Table 1). Variables of interest included our main exposure (log–transformed, SpG-adj BPA) and outcome variables (AGD-AF and AGD-AC), all of which were continuous. We selected a set of covariates a priori based on our prior analyses and the previous literature (Swan et al. 2015). Those covariates were postconception age at the time of exam (calculated as gestational age at birth plus age at exam, continuous), infant size at exam [weight-for-length z-score (ZWL), continuous], mother’s age (continuous), child’s race (non-Hispanic white vs. other; categorical indicator) time of urine collection (time since midnight in hours), and study center (categorical indicator). We elected to use ZWL based on World Health Organization (WHO) standard curves (WHO Multicentre Growth Reference Study Group 2009) to adjust for body size in our models because our previous work in this cohort suggests that among possible infant size metrics to consider (including weight, weight for age, and length-for-age z-scores), ZWL is the strongest predictor of genital measurements (Swan et al. 2015).

Table 1. Characteristics of the study population (n=381).
Percentiles
Continuous variables Mean±SD Min 25th 50th 75th Max Association with log(SpG-adj BPA) (r)c
Maternal age (years) 31.1±5.6 18.3 27.4 31.8 35.3 45.2 −0.16
Gestational age at birth (weeks) 39.4±1.6 32.9 38.9 39.6 40.6 42.3 0.03
Age at exam (days) 4.4±10.4 0 1.0 1.0 2.0 65.0 0.11
z-score weight for age −0.4±1.3 −5.3 −1.0 −0.3 0.4 8.6 0.06
Postconception age at exam (weeks) 40.1±2.0 35.0 39.1 40.0 41.0 49.0 0.11
AGD-AC (mm)a 36.6±3.8 16.5 34.1 36.8 39.1 53.3 −0.10
AGD-AF (mm) 16.0±3.2 8.0 13.9 15.7 18.2 28.5 0.03
Time of urine collection (hours since midnight) 12.6±2.5 7.5 10.5 12.5 14.8 19.8 0.16
Gestational age at urine collection (weeks) 10.8±2.0 5.1 9.4 11.0 12.3 15.7 0.01
SpG-adj BPA(μg/L) 1.8±2.5 0.04 0.6 1.0 1.9 27.1
Categorical variables n (%)b
Center
UCSF 98 (25.7)
UMN 94 (24.7)
URMC 109 (28.6)
UW 80 (21.0)
Race/ethnicity
White/non-Hispanic 225 (60.5)
Other 147 (39.5)
Education
High school or less 55 (14.7)
Some college or more 320 (85.3)

Note: AGD-AC, anogenital distance from the anus to the clitoris; AGD-AF, anogenital distance from the anus to the fourchette; BPA, Bisphenol A; Max, maximum; Min, minimum; SD, standard deviation; SpG-adj, specific gravity–adjusted; UCSF, University of California, San Francisco; UMN, University of Minnesota; URMC, University of Rochester Medical Center; UW, University of Washington.

aOne infant did not have an AGD-AC measurement, so n=380.

bPercentages may not total exactly 100% due to rounding.

cLog-transformed specific gravity–adjusted concentrations. Pearson’s correlation was used to examine associations with continuous variables.

We used scatterplots to examine the relationships between our outcome variables (AGD-AF and AGD-AC) and all continuous covariates. Four unusual observations were detected in bivariate analyses. Three babies were measured at ages much older than the median of 1 d (110, 132, and 153 d). An additional mother had a urinary SpG value that was biologically implausible (1.062) (Boeniger et al. 1993). Those four mother–infant dyads were removed from all subsequent analyses.

To examine the relationship between maternal BPA concentrations and infant AGD, we fit unadjusted and adjusted linear regression models. In the unadjusted models, only SpG-adj BPA was used as a covariate, while in the adjusted models, all covariates specified above were included. We conducted several sets of sensitivity analyses. First, we refit models stratifying based on whether the infants were examined by the study team within 2 d after birth. Second, we refit models replacing the composite variable “postconception age at exam” with its constituent variables, gestational age at birth and age at exam. Model diagnostics were conducted to investigate any possible violations of the linear model assumptions of normality (Q-Q plots), homogeneity of variance (residuals as a function of predicted values), independence of the observations (Durbin-Watson test), and linearity of the relationship between the outcome variables and the covariates (plots of residual vs. predicted values as well as component plus residual plots). Outliers were identified using the studentized residuals (>3 or <−3), influential observations were identified using Cook’s distance (>0.5), and leverage points were identified using the diagonal elements of the hat matrix (for our sample, h>0.0417). Adjusted models were also checked for collinearity using the variance inflation factors (VIFs) of each variable (VIF>10 indicating collinearity issues). In adjusted models, several violations were noted, including violation of the constant variance assumption (AGD-AC models only), and nonlinear relationships between some covariates (postconception age, ZWL, and BPA) and outcome variables in some models. Although several potential outliers were noted across the various models and some observations had slightly high leverage (but were not influential), lacking further justification for excluding them, they were retained. Given the model violations noted above, we log–transformed the outcome variables and refit models. This did not lead to significant improvement in satisfying the linear assumptions; therefore, we also explored generalized additive models (GAMs). Ultimately, because the intent of this analysis is inference, not prediction, we present the linear models as primary, with the nonlinear models presented secondarily for reference. All analyses were done using R (version 3.23; R Development Core Team), and p-values<0.05 were considered significant.

Results

Demographic characteristics of TIDES participants have been previously described (Barrett et al. 2014; Swan et al. 2015), and are summarized here briefly. A total of 385 mothers (and their infant daughters) had data on first-trimester maternal BPA concentrations, infant AGD, and relevant covariates. After removing from the analysis the four excluded mother–child pairs, 381 dyads were included in the current analyses (Table 1). On average, mothers in the study were 31.1±5.6 yr of age. Participants were predominantly white (65.3%). The remaining women self-identified as black (14.8%), Asian (6.7%), and other/unknown (13.2%). Most women were non-Hispanic (87.5%), and 85.3% had at least a high school education. There was roughly equal representation across the four study centers (California: 25.7%; Minnesota: 24.7%; New York: 28.6%; Washington: 21.0%) (see Table S1). Eight women (2.1%) gave a urine sample outside of the first trimester (range: 14–15 weeks ); however, they were retained in analyses given that the reproductive programming window is believed to extend into that approximate gestational age range.

The infants in the current analyses were mostly born at term (91.9% at ≥37 weeks gestation). Median infant weight for length z-score at the time of examination was slightly negative (median: −0.32), presumably due to the water weight loss that typically occurs in the days immediately following birth (Mulder and Gardner 2015). Of the 381 babies included in this analysis, exams were conducted on 68 babies (17.8%) at greater than 2 d old. Of these, sixteen exams were delayed due to neonatal intensive care unit (NICU) admission (following preterm birth), while the remaining 52 exams were delayed due to logistical issues, where mothers were discharged from the hospital before they could be reached by the study team.

Across all girls, mean AGD-AC was 36.6±3.8 mm, and mean AGD-AF was 16.0±3.2 mm. The intraexaminer intraclass correlations (ICCs) (looking at consistency of measurements within a single examiner) were 0.92 for both AGD measures. Fifty-four infants in this analysis underwent repeated measurements by two examiners, and the interexaminer ICCs were 0.73 and 0.79 for AGD-AF and AGD-AC, respectively. BPA concentration was below the limit of detection in 6.3% of samples, and the median BPA concentration was 0.90 μg/L (0.99 μg/L after SpG adjustment). A check of model assumptions showed evidence of nonhomogeneous residual variance with untransformed AGD, so models were also fit using the natural logarithm of AGD. Conclusions were similar and results from the transformed models are presented in Table S2. Scatterplots of AGD in relation to covariates are shown in Figures S1 and S2.

In unadjusted models, log(SpG-adj BPA) showed a nonsignificant, inverse association with AGD-AC [β=−0.36, 95% confidence interval (CI): −0.78, 0.06] and a nonsignificant, weakly positive relationship with AGD-AF (β=0.16, 95% CI: −0.19, 0.52) (Table 2). In multivariable models adjusting for mother’s age, infant’s postconception age, weight-for-length z-score, time of urine collection, infant’s race, and study center, log(SpG-adj BPA) was significantly associated with AGD-AC (β=−0.56, 95% CI: −0.97, −0.15), but not with AGD-AF (β=0.03, 95% CI: −0.30, 0.37). Since the third quartile of SpG-adj BPA is 317% that of the first quartile, multiplying the slopes for log(SpG-adj BPA) by log (3)=1.1 estimates the change in AGD for a change in BPA from the first to the third quartile. The transformed slopes in the adjusted model are −0.62 mm for AGD-AC and 0.03 mm for AGD-AF. The infant’s postconception age and ZWL were both strongly and positively associated with AGD-AC (Table 2). AGD-AF varied by study center and was inversely associated with maternal age, but positively associated with infant’s postconception age (Table 2).

Table 2. Linear regression models examining the relationship between log(SpG-adj BPA) and covariates and anogenital distance measures in newborn daughters (n=381).
AGD-AC AGD-AF
Characteristic Unadjusted β (95% CI) Adjusted β (95% CI) Unadjusted β (95% CI) Adjusted β (95% CI)
Log(SpG-adj BPA) −0.36 (−0.78, 0.06) −0.56 (−0.97, −0.15) 0.16 (−0.19, 0.52) 0.03 (−0.30, 0.37)
Maternal age −0.05 (−0.13, 0.03) −0.08 (−0.14, −0.01)
Infant’s postconception age 0.54 (0.35, 0.73) 0.17 (0.02, 0.33)
Weight-for-length z-score 0.66 (0.35, 0.96) 0.22 (−0.02, 0.47)
Race 0.30 (−0.54, 1.13) −0.06 (−0.73, 0.62)
Urine collection time 0.04 (−0.11, 0.20) 0.04 (−0.08, 0.17)
UCSF center 0.61 (−0.54, 1.75) −1.53 (−2.46, −0.61)
URMC center −0.47 (−1.63, 0.68) −1.09 (−2.02, −0.16)
UW center 0.08 (−1.06, 1.22) 1.99 (1.07, 2.90)

Note: AGD-AC, anogenital distance from the anus to the clitoris; AGD-AF, anogenital distance from the anus to the fourchette; BPA, Bisphenol A; mm, millimeters; CI, confidence interval; SpG-adj, specific gravity–adjusted; UCSF, University of California, San Francisco; URMC, University of Rochester Medical Center; UW, University of Washington. For race, the referent is non-Hispanic, white. For center, the referent is UMN. All other variables are continuous.

In secondary GAM models that allowed for nonlinearity between covariates and AGD, AGD-AC showed nonlinear associations with both log(SpG-adj BPA) and ZWL, while AGD-AF showed nonlinear associations with log(SpG-adj BPA) and postconception age. The association between log(SpG-adj BPA) on AGD-AC was relatively constant for small values of BPA, but AGD-AC decreased more rapidly in association with higher BPA concentrations (Figure S3). The nature of this relationship was similar when stratified by postconception age greater or less than 2 d (not shown). GAM models for AGD-AF showed nonmonotonic relationships with log(SpG-adj BPA) (Figure S4).

When we refit models stratifying by age at examination, in the subset of infants measured within 2 d of birth (n=313), the adjusted slope for BPA was −0.40 (95% CI: −0.82, 0.01), whereas the relationship was stronger among infants measured after 2 d of age (n=68; adjusted slope=−0.84; 95% CI: −2.26, 0.59). In sensitivity models replacing postconception age at exam with gestational age at birth and age at exam, results were virtually identical to our primary models (AGD-AC: β=−0.55, 95% CI: −0.97, −0.14; AGD-AF: β=0.04, 95%CI: −0.30, 0.37).

Discussion

In this pregnancy cohort study, first-trimester maternal urinary BPA concentration was inversely associated with one of two measures of daughters’ AGD at birth. To our knowledge, this is the first study to examine BPA exposure during early pregnancy in relation to AGD. It is also the first study to consider prenatal BPA exposure and AGD in females. Our findings are consistent with several (but not all) rodent studies that found BPA administration during gestation to be associated with shortened AGD in newborn female offspring (Christiansen et al. 2014; Kobayashi et al. 2012), and it has been hypothesized that shorter female AGD may represent hyperfeminization, possibly resulting from estrogen receptor agonism (Christiansen et al. 2014). Given the body of evidence that AGD is, in fact, a sensitive measure of endocrine activity during early fetal development, our results provide further support for the hypothesis that BPA exposure during pregnancy alters typical gestational endocrine signaling pathways. The extent to which early endocrine changes may program long-term reproductive development and trajectories is uncertain, but merits additional research.

Research across a number of model species has linked prenatal BPA exposure to outcomes including changes in oogenesis and ovarian steroidogenesis (Fernández et al. 2010; Hunt et al. 2012; Susiarjo et al. 2007; Xi et al. 2011), a polycystic ovarian syndrome-like phenotype (Adewale et al. 2009; Fernández et al. 2010), uterine and endometrial defects (Newbold et al. 2009; Signorile et al. 2010), and morphological changes in the mammary gland (Paulose et al. 2015). Although there has been considerable epidemiological research on adolescent and adult BPA exposure in relation to reproductive outcomes (reviewed by Peretz et al. 2014), to our knowledge, there is currently only one study that measured prenatal BPA levels and followed the resulting children to reproductive maturity, when reproductive outcomes are more easily measured and clinically relevant. In that Mexican cohort, BPA concentration in third-trimester urine was not associated with steroid hormone levels or sexual maturation in girls at age 8–13 y (n=115) (Watkins et al. 2014), but was inversely associated with odds of having reached adrenarche or pubarche in boys (n=107) (Ferguson et al. 2014). Additional longitudinal research is needed to replicate these findings in other populations to examine exposures earlier in pregnancy and to investigate reproductive outcomes beyond the peripubertal period in humans.

Based on the body of evidence from numerous in vitro, animal model, and human studies, AGD is usually considered an androgen-sensitive measure (reviewed in Dean and Sharpe 2013; Thankamony et al. 2016). By contrast, BPA is best known for its estrogenic activity (Alonso-Magdalena et al. 2012; Vom Saal et al. 2012), and the mechanisms underlying the relationship between prenatal BPA exposure and AGD are uncertain. In animal models, there is evidence that BPA may act as an androgen receptor antagonist (vom Saal and Hughes 2005; Wetherill et al. 2007), a mechanism that is consistent with findings from a Chinese study demonstrating that maternal occupational exposure to BPA (as estimated by personal air sampling during pregnancy) was associated with shorter age- and weight-adjusted AGD in sons during childhood (n=153) (Miao et al. 2011). In that study, daughters were not examined. However previous work in animal models and humans suggests that prenatal exposure to antiandrogens [such as di(2-ethylhexyl) phthalate (DEHP)] is not associated with alterations in AGD in females (Christiansen et al. 2009; Hass et al. 2007; Swan et al. 2015). Thus, if BPA exposure in early gestation affects AGD in females, it may be through an alternative mechanism. For instance, BPA may increase activity of the enzyme aromatase, which converts testosterone to estradiol (Castro et al. 2013; Séralini and Moslemi 2001). This possibility is supported by a cross-sectional study that found that BPA concentration in third-trimester maternal urine was inversely associated with the testosterone to estradiol ratio in cord blood of 137 newborn boys (Liu et al. 2016). It is also possible that BPA exposure may affect AGD through estrogenic pathways. Some studies have found that gestational exposure to known estrogenic compounds, like diethylstilbestrol (DES), ethinyl estradiol (EE2), or genistein, decreases AGD in female rodents (Delclos et al. 2009; Levy et al. 1995), whereas other studies have found increases in AGD, depending on the compound used, dose, species, and the study design (Casanova et al. 1999; Mandrup et al. 2013). Finally, it is worth considering other potential mechanisms; in large-scale toxicity testing (ToxCast program), BPA had effects in 101 of 467 in vitro screening assays. Notably, of 309 environmental chemicals studied, BPA had the second-highest toxic potential score, suggesting its ability to act through numerous endocrine pathways (Reif et al. 2010).

A strength of our study is the longitudinal cohort design with maternal samples collected from early pregnancy, arguably the period of most relevance for fetal reproductive system development. In addition, we recruited healthy pregnant women broadly from four U.S. cities, giving us a relatively diverse sample (in terms of age, race, education, and socioeconomic status) compared to some other notable pregnancy cohorts, which have focused on a very specific subject population, such as migrant farm workers (Harley et al. 2013b) or low-income inner-city families (Braun et al. 2011; Harley et al. 2013a; Wolff et al. 2008). Notably, BPA concentrations varied quite considerably across study centers. For instance, women at the UCSF study center, who tended to be very well educated and had high income, had much lower BPA levels (median, 0.81 μg/L) than women at the other TIDES study centers (URMC: 1.18 μg/L; UMN: 1.11 μg/L, UW: 0.91 μg/L). Their BPA levels were also lower than those reported in a number of other recent studies of pregnant women in Boston (geometric mean: 1.34 μg/L), Denmark (median 1.52 μg/L), and at National Children’s Study Vanguard sites (geometric mean: 1.4 μg/L) (Cantonwine et al. 2016; Frederiksen et al. 2014; Mortensen et al. 2014). Given the widespread concern about BPA exposure, better understanding the specific factors (including consumer choices) linked to low BPA levels (such as those seen in our UCSF population) may shed light on potential ways to reduce exposure.

Only one of the two infant female AGD measurements, AGD-AC, was significantly associated with maternal BPA concentration. Overall, the high intra- and interobserver ICCs indicate that our infant AGD measurements were consistent and reproducible. All examiners were highly trained, and our protocols were carefully designed to be quick, minimally invasive, and acceptable to families. Nevertheless, as previously reported, the AGF-AF measurement tends to be relatively difficult to replicate because the fourchette landmark can be difficult to visualize (Sathyanarayana et al. 2015), and there was significant intercenter variation in that measurement. Based on these data, we believe that AGD-AC is the more reliable measurement in the current study; however, it is also worth considering the possibility that AGD-AF was not associated with first-trimester maternal BPA exposure because that distance reflects different developmental processes that may be affected by different exposures and/or during a different critical period. It is worth noting, however, that even for AGD-AC, the magnitude of the observed differences was small. Holding all other covariates constant, the daughter of a mother in the 75th percentile of BPA concentration would be expected to have 0.63 mm shorter AGD than the daughter of a mother in the 25th percentile. This difference corresponds to only a 2% shorter AGD for the average girl. While this difference is small, it is on par with the magnitude of observed associations between prenatal phthalate exposure and AGD in male infants. Our previous work suggests that depending on the AGD measurement and the metabolite, a maternal increase from the 10th to the 90th percentile of first-trimester DEHP exposure corresponds to a 2–5% decrease in boys’ AGD at birth (Swan et al. 2015). The clinical relevance of these subtle variations in AGD is an important question that requires continued longitudinal follow-up.

Our study has several limitations of note. First, women who agree to participate in intensive, longitudinal research studies may not be representative of pregnant women as a whole; however, we cannot directly address this issue, as we do not have data on women who declined to participate or could not be approached. Another concern is the use of single urine samples to assess maternal BPA exposure. BPA levels can vary considerably over time; two studies have estimated the interclass correlation for serial BPA samples collected across pregnancy to be roughly 0.2 to 0.3 (Meeker et al. 2013; Teitelbaum et al. 2008). Moreover, we did not measure BPA in maternal samples collected later in pregnancy, nor in mothers of boys, which would be informative as far as further understanding critical windows and potential sex differences. Finally, although AGD was measured by highly trained examiners and quality control measures were implemented throughout the study, there was significant intercenter variation in AGD-AF, making that measurement less informative. As with any multicenter study, there is the possibility of uncontrolled confounding by unidentified factors that may vary across the study sites. Finally, we do not have serial measures of AGD at multiple ages in girls, so we are unable to address whether associations observed between prenatal BPA concentrations and AGD are stable across childhood and into adulthood. Regardless of whether the relationship between early BPA exposure and AGD persists throughout life, our findings provide further evidence suggesting that BPA may disrupt the early endocrine environment.

Conclusion

Our study provides further evidence that prenatal exposure to BPA may impact reproductive development in females. More research is needed to confirm the current findings and investigate additional reproductive end points in human females, including reproductive hormones and ovarian reserve in infancy, as well as reproductive outcomes later in life. Whether the changes we observed persist across the lifespan and contribute to clinically relevant outcomes should be investigated further. These results may further inform the ongoing controversy over the widespread use of BPA in consumer products and provide evidence regarding policy to limit its use in manufactured goods in the United States as well as abroad (Heindel et al. 2015; Metz 2016).

Acknowledgments

We thank the TIDES Study Team for their contributions. Coordinating Center: F. Liu, E. Scher, S. Evans; UCSF: M. Stasenko, E. Ayash, M. Schirmer, J. Farrell, M.-P. Thiet, L. Baskin; UMN: H. L. Gray, C. Georgesen, B. J. Rody, C. A. Terrell, K. Kaur; URMC: E. Brantley, H. Fiore, L. Kochman, J. Marino, W. Hulbert, R. Mevorach, E. Pressman; UW/SCH: R. Grady, K. Ivicek, B. Salveson, G. Alcedo; and the families who participated in the study. In addition, we thank A. Calafat (Centers for Disease Control and Prevention) for BPA analysis, the TIDES families for their participation, and the residents at URMC and UCSF who assisted in birth exams. Funding for TIDES was provided by the following grants from the National Institute of Environmental Health Sciences: R01ES016863-04 and R01ES016863-02S4. Support for the current analysis was provided by T32ES007271, P30ES001247, and P30ES005022.

References

Adewale HB, Jefferson WN, Newbold RR, Patisaul HB. 2009. Neonatal bisphenol-A exposure alters rat reproductive development and ovarian morphology without impairing activation of gonadotropin-releasing hormone neurons. Biol Reprod 81(4):690–699, PMID: 19535786, 10.1095/biolreprod.109.078261.

Alonso-Magdalena P, Ropero AB, Soriano S, García-Arévalo M, Ripoll C, Fuentes E, et al. 2012. Bisphenol-A acts as a potent estrogen via non-classical estrogen triggered pathways. Mol Cell Endocrinol 355(2):201–207, PMID: 22227557, 10.1016/j.mce.2011.12.012.

Ariemma F, D’Esposito V, Liguoro D, Oriente F, Cabaro S, Liotti A, et al. 2016. Low-dose Bisphenol-A impairs adipogenesis and generates dysfunctional 3T3-L1 adipocytes. PLoS One 11(3):e0150762, PMID: 26942597, 10.1371/journal.pone.0150762.

Balakrishnan B, Henare K, Thorstensen EB, Ponnampalam AP, Mitchell MD. 2010. Transfer of bisphenol A across the human placenta. Am J Obstet Gynecol 202(4):393.e1–7, PMID: 20350650, 10.1016/j.ajog.2010.01.025.

Barrett E, Hoeger K, Sathyanarayana S, Redmon JB, Nguyen RH, Swan SH. 2016. Anogenital distance, a biomarker of prenatal androgen exposure, is longer among newborn daughters of women with polycystic ovary syndrome (PCOS). In: Endocrine Society’s 98th Annual Meeting and Expo, April 1–4, 2016 – Boston. Boston, MA:Endocrine Society (ENDO).

Barrett ES, Sathyanarayana S, Janssen S, Redmon JB, Nguyen RH, Kobrosly R, et al. 2014. Environmental health attitudes and behaviors: Findings from a large pregnancy cohort study. Eur J Obstet Gynecol Reprod Biol 176:119–125, PMID: 24647207, 10.1016/j.ejogrb.2014.02.029.

Berger A, Ziv-Gal A, Cudiamat J, Wang W, Zhou C, Flaws JA. 2016. The effects of in utero bisphenol A exposure on the ovaries in multiple generations of mice. Reprod Toxicol 60:39–52, PMID: 26746108, 10.1016/j.reprotox.2015.12.004.

Boeniger MF, Lowry LK, Rosenberg J. 1993. Interpretation of urine results used to assess chemical exposure with emphasis on creatinine adjustments: A review. Am Ind Hyg Assoc J 54(10):615–627, PMID: 8237794, 10.1080/15298669391355134.

Braun JM, Kalkbrenner AE, Calafat AM, Yolton K, Ye X, Dietrich KN, et al. 2011. Impact of early-life bisphenol A exposure on behavior and executive function in children. Pediatrics 128(5):873–882, PMID: 22025598, 10.1542/peds.2011-1335.

Calafat AM, Ye X, Wong LY, Reidy JA, Needham LL. 2008. Exposure of the U.S. Population to bisphenol A and 4-tertiary-octylphenol: 2003-2004. Environ Health Perspect 116(1):39–44, PMID: 18197297, 10.1289/ehp.10753.

Callegari C, Everett S, Ross M, Brasel JA. 1987. Anogenital ratio: Measure of fetal virilization in premature and full-term newborn infants. J Pediatr 111(2):240–243, PMID: 3612396, 10.1016/S0022-3476(87)80075-6.

Cantonwine DE, Meeker JD, Ferguson KK, Mukherjee B, Hauser R, McElrath TF. 2016. Urinary concentrations of bisphenol A and phthalate metabolites measured during pregnancy and risk of preeclampsia. Environ Health Perspect 124(10):1651–1655, PMID: 27177253, 10.1289/EHP188.

Casanova M, You L, Gaido KW, Archibeque-Engle S, Janszen DB, Heck HA. 1999. Developmental effects of dietary phytoestrogens in Sprague-Dawley rats and interactions of genistein and daidzein with rat estrogen receptors alpha and beta in vitro. Toxicol Sci 51(2):236–244, PMID: 10543025, 10.1093/toxsci/51.2.236.

Castaño-Vinyals G, Carrasco E, Lorente JA, Sabaté Y, Cirac-Claveras J, Pollán M, et al. 2012. Anogenital distance and the risk of prostate cancer. BJU Int 110(11 Pt B):E707–E710, PMID: 22984847, 10.1111/j.1464-410X.2012.11516.x.

Castro B, Sanchez P, Torres JM, Preda O, del Moral RG, Ortega E. 2013. Bisphenol a exposure during adulthood alters expression of aromatase and 5α-reductase isozymes in rat prostate. PLoS One 8(2):e55905, PMID: 23405234, 10.1371/journal.pone.0055905.

CDC. 2013. Bisphenol A (BPA) Factsheet. Atlanta, GA:Centers for Disease Control National Biomonitoring Program.

Christiansen S, Axelstad M, Boberg J, Vinggaard AM, Pedersen GA, Hass U. 2014. Low-dose effects of bisphenol A on early sexual development in male and female rats. Reproduction 147(4):477–487, PMID: 24298045, 10.1530/REP-13-0377.

Christiansen S, Scholze M, Dalgaard M, Vinggaard AM, Axelstad M, Kortenkamp A, et al. 2009. Synergistic disruption of external male sex organ development by a mixture of four antiandrogens. Environ Health Perspect 117(12):1839–1846, PMID: 20049201, 10.1289/ehp.0900689.

de Lima RF, Rodriguez DA, Campos MS, Biancardi MF, dos Santos IF, de Oliveira WD, et al. 2015. Bisphenol-A promotes antiproliferative effects during neonatal prostate development in male and female gerbils. Reprod Toxicol 58:238–245, PMID: 26529182, 10.1016/j.reprotox.2015.10.016.

Dean A, Sharpe RM. 2013. Anogenital distance or digit length ratio as measures of fetal androgen exposure: Relationship to male reproductive development and its disorders. J Clin Endocrinol Metab 98(6):2230–2238, PMID: 23569219, 10.1210/jc.2012-4057.

Delclos KB, Weis CC, Bucci TJ, Olson G, Mellick P, Sadovova N, et al. 2009. Overlapping but distinct effects of genistein and ethinyl estradiol (EE2) in female Sprague-Dawley rats in multigenerational reproductive and chronic toxicity studies. Reprod Toxicol 27(2):117–132, PMID: 19159674, 10.1016/j.reprotox.2008.12.005.

EFSA. 2015. Scientific opinion on the risks to public health related to the presence of bisphenol A (BPA) in foodstuffs. EFSA Journal 13(1):3978, 10.2903/j.efsa.2015.3978.

Eisenberg ML, Shy M, Walters RC, Lipshultz LI. 2012. The relationship between anogenital distance and azoospermia in adult men. Int J Androl 35(5):726–730, PMID: 22519659, 10.1111/j.1365-2605.2012.01275.x.

Ferguson KK, Peterson KE, Lee JM, Mercado-García A, Blank-Goldenberg C, Téllez-Rojo MM, et al. 2014. Prenatal and peripubertal phthalates and bisphenol A in relation to sex hormones and puberty in boys. Reprod Toxicol 47:70–76, PMID: 24945889, 10.1016/j.reprotox.2014.06.002.

Fernández M, Bourguignon N, Lux-Lantos V, Libertun C. 2010. Neonatal exposure to bisphenol a and reproductive and endocrine alterations resembling the polycystic ovarian syndrome in adult rats. Environ Health Perspect 118(9):1217–1222, PMID: 20413367, 10.1289/ehp.0901257.

Frederiksen H, Jensen TK, Jørgensen N, Kyhl HB, Husby S, Skakkebæk NE, et al. 2014. Human urinary excretion of non-persistent environmental chemicals: An overview of Danish data collected between 2006 and 2012. Reproduction 147(4):555–565, PMID: 24395915, 10.1530/REP-13-0522.

Gentilcore D, Porreca I, Rizzo F, Ganbaatar E, Carchia E, Mallardo M, et al. 2013. Bisphenol A interferes with thyroid specific gene expression. Toxicology 304:21–31, PMID: 23238275, 10.1016/j.tox.2012.12.001.

Gerona RR, Woodruff TJ, Dickenson CA, Pan J, Schwartz JM, Sen S, et al. 2013. Bisphenol-A (BPA), BPA glucuronide, and BPA sulfate in midgestation umbilical cord serum in a northern and central California population. Environ Sci Technol 47(21):12477–12485, PMID: 23941471, 10.1021/es402764d.

Harley KG, Aguilar Schall R, Chevrier J, Tyler K, Aguirre H, Bradman A, et al. 2013a. Prenatal and postnatal bisphenol A exposure and body mass index in childhood in the CHAMACOS cohort. Environ Health Perspect 121(4):514–520, PMID: 23416456, 10.1289/ehp.1205548.

Harley KG, Gunier RB, Kogut K, Johnson C, Bradman A, Calafat AM, et al. 2013b. Prenatal and early childhood bisphenol A concentrations and behavior in school-aged children. Environ Res 126:43–50, PMID: 23870093, 10.1016/j.envres.2013.06.004.

Hass U, Scholze M, Christiansen S, Dalgaard M, Vinggaard AM, Axelstad M, et al. 2007. Combined exposure to anti-androgens exacerbates disruption of sexual differentiation in the rat. Environ Health Perspect 115 (Suppl 1):122–128, PMID: 18174960, 10.1289/ehp.9360.

Heindel JJ, Newbold RR, Bucher JR, Camacho L, Delclos KB, Lewis SM, et al. 2015. NIEHS/FDA CLARITY-BPA research program update. Reprod Toxicol 58:33–44, PMID: 26232693, 10.1016/j.reprotox.2015.07.075.

Honma S, Suzuki A, Buchanan DL, Katsu Y, Watanabe H, Iguchi T. 2002. Low dose effect of in utero exposure to bisphenol A and diethylstilbestrol on female mouse reproduction. Reprod Toxicol 16(2):117–122, PMID: 11955942, 10.1016/S0890-6238(02)00006-0.

Hornung RW, Reed LD. 1990. Estimation of average concentration in the presence of nondetectable values. Applied Occupational and Environmental Hygiene 5(1):46–51, 10.1080/1047322X.1990.10389587.

Hotchkiss AK, Lambright CS, Ostby JS, Parks-Saldutti L, Vandenbergh JG, Gray LE Jr. 2007. Prenatal testosterone exposure permanently masculinizes anogenital distance, nipple development, and reproductive tract morphology in female Sprague-Dawley rats. Toxicol Sci 96(2):335–345, PMID: 17218470, 10.1093/toxsci/kfm002.

Howdeshell HL, Furr J, Lambright CR, Wilson VS, Ryan BC, Gray LE Jr. 2008. Gestational and lactational exposure to ethinyl estradiol, but not bisphenol A, decreases androgen-dependent reproductive organ weights and epididymal sperm abundance in the male long evans hooded rat. Toxicol Sci 102(2):371–382, PMID: 18096570, 10.1093/toxsci/kfm306.

Hunt PA, Lawson C, Gieske M, Murdoch B, Smith H, Marre A, et al. 2012. Bisphenol A alters early oogenesis and follicle formation in the fetal ovary of the rhesus monkey. Proc Natl Acad Sci USA 109(43):17525–17530, PMID: 23012422, 10.1073/pnas.1207854109.

Ikezuki Y, Tsutsumi O, Takai Y, Kamei Y, Taketani Y. 2002. Determination of bisphenol A concentrations in human biological fluids reveals significant early prenatal exposure. Hum Reprod 17(11):2839–2841, PMID: 12407035, 10.1093/humrep/17.11.2839.

Jones BA, Wagner LS, Watson NV. 2016. The effects of bisphenol A exposure at different developmental time points in an androgen-sensitive neuromuscular system in male rats. Endocrinology 157(8):2972–2977, PMID: 27022676, 10.1210/en.2015-1574.

Kobayashi K, Kubota H, Ohtani K, Hojo R, Miyagawa M. 2012. Lack of effects for dietary exposure of bisphenol A during in utero and lactational periods on reproductive development in rat offspring. J Toxicol Sci 37(3):565–573, PMID: 22687996, 10.2131/jts.37.565.

Lakind JS, Naiman DQ. 2011. Daily intake of bisphenol A and potential sources of exposure: 2005-2006 national health and nutrition examination survey. J Expos Sci Environ Epidemiol 21(3):272–279, PMID: 20237498, 10.1038/jes.2010.9.

Levy JR, Faber KA, Ayyash L, Hughes CL Jr. 1995. The effect of prenatal exposure to the phytoestrogen genistein on sexual differentiation in rats. Proc Soc Exp Biol Med 208(1):60–66, PMID: 7892297, 10.3181/00379727-208-43832.

Liu C, Xu X, Zhang Y, Li W, Huo X. 2016. Associations between maternal phenolic exposure and cord sex hormones in male newborns. Hum Reprod 31(3):648–656, PMID: 26724800, 10.1093/humrep/dev327.

Mandrup KR, Jacobsen PR, Isling LK, Axelstad M, Dreisig K, Hadrup N, et al. 2013. Effects of perinatal ethinyl estradiol exposure in male and female Wistar rats. Reprod Toxicol 42:180–191, PMID: 24036065, 10.1016/j.reprotox.2013.09.001.

Meeker JD, Cantonwine DE, Rivera-González LO, Ferguson KK, Mukherjee B, Calafat AM, et al. 2013. Distribution, variability, and predictors of urinary concentrations of phenols and parabens among pregnant women in Puerto Rico. Environ Sci Technol 47(7):3439–3447, PMID: 23469879, 10.1021/es400510g.

Mendiola J, Roca M, Mínguez-Alarcón L, Mira-Escolano MP, López-Espín JJ, Barrett ES, et al. 2012. Anogenital distance is related to ovarian follicular number in young Spanish women: A cross-sectional study. Environ Health 11:90, PMID: 23217457, 10.1186/1476-069X-11-90.

Mendiola J, Sánchez-Ferrer ML, Jiménez-Velázquez R, Cánovas-López L, Hernández-Peñalver AI, Corbalán-Biyang S, et al. 2016. Endometriomas and deep infiltrating endometriosis in adulthood are strongly associated with anogenital distance, a biomarker for prenatal hormonal environment. Hum Reprod 31(10):2377–2383, PMID: 27357299, 10.1093/humrep/dew163.

Mendiola J, Stahlhut RW, Jørgensen N, Liu F, Swan SH. 2011. Shorter anogenital distance predicts poorer semen quality in young men in Rochester, New York. Environ Health Perspect 119(7):958–963, PMID: 21377950, 10.1289/ehp.1103421.

Metz CM. 2016. Bisphenol A: Understanding the controversy. Workplace Health Saf 64(1):28–36; quiz 37, PMID: 26800896, 10.1177/2165079915623790.

Miao M, Yuan W, He Y, Zhou Z, Wang J, Gao E, et al. 2011. In utero exposure to bisphenol-A and anogenital distance of male offspring. Birth Defects Res Part A Clin Mol Teratol 91(10):867–872, PMID: 21987463, 10.1002/bdra.22845.

Mira-Escolano MP, Mendiola J, Mínguez-Alarcón L, Melgarejo M, Cutillas-Tolín A, Roca M, et al. 2014. Longer anogenital distance is associated with higher testosterone levels in women: A cross-sectional study. BJOG 121(11):1359–1364, 10.1111/1471-0528.12627.

Mortensen ME, Calafat AM, Ye X, Wong LY, Wright DJ, Pirkle JL, et al. 2014. Urinary concentrations of environmental phenols in pregnant women in a pilot study of the National Children’s Study. Environ Res 129:32–38, PMID: 24529000, 10.1016/j.envres.2013.12.004.

Mulder PJ, Gardner SE. 2015. The healthy newborn hydration model: A new model for understanding newborn hydration immediately after birth. Biol Res Nurs 17(1):94–99, PMID: 25504955, 10.1177/1099800414529362.

Newbold RR, Jefferson WN, Padilla-Banks E. 2009. Prenatal exposure to bisphenol A at environmentally relevant doses adversely affects the murine female reproductive tract later in life. Environ Health Perspect 117(6):879–885, PMID: 19590677, 10.1289/ehp.0800045.

Nishikawa M, Iwano H, Yanagisawa R, Koike N, Inoue H, Yokota H. 2010. Placental transfer of conjugated bisphenol A and subsequent reactivation in the rat fetus. Environ Health Perspect 118(9):1196–1203, PMID: 20382578, 10.1289/ehp.0901575.

Paulose T, Speroni L, Sonnenschein C, Soto AM. 2015. Estrogens in the wrong place at the wrong time: Fetal BPA exposure and mammary cancer. Reprod Toxicol 54:58–65, PMID: 25277313, 10.1016/j.reprotox.2014.09.012.

Peretz J, Vrooman L, Ricke WA, Hunt PA, Ehrlich S, Hauser R, et al. 2014. Bisphenol A and reproductive health: Update of experimental and human evidence, 2007-2013. Environ Health Perspect 122(8):775–786, PMID: 24896072, 10.1289/ehp.1307728.

Rehan M, Ahmad E, Sheikh IA, Abuzenadah AM, Damanhouri GA, Bajouh OS, et al. 2015. Androgen and progesterone receptors are targets for bisphenol A (BPA), 4-methyl-2,4-bis-(p-hydroxyphenyl)pent-1-ene–a potent metabolite of BPA, and 4-tert-octylphenol: A computational insight. PLoS One 10(9):e0138438, PMID: 26379041, 10.1371/journal.pone.0138438.

Reif DM, Martin MT, Tan SW, Houck KA, Judson RS, Richard AM, et al. 2010. Endocrine profiling and prioritization of environmental chemicals using ToxCast data. Environ Health Perspect 118(12):1714–1720, PMID: 20826373, 10.1289/ehp.1002180.

Ryan BC, Hotchkiss AK, Crofton KM, Gray LE Jr. 2010. In utero and lactational exposure to bisphenol A, in contrast to ethinyl estradiol, does not alter sexually dimorphic behavior, puberty, fertility, and anatomy of female LE rats. Toxicol Sci 114(1):133–148, PMID: 19864446, 10.1093/toxsci/kfp266.

Sathyanarayana S, Beard L, Zhou C, Grady R. 2010. Measurement and correlates of ano-genital distance in healthy, newborn infants. Int J Androl 33(2):317–323, PMID: 20132349, 10.1111/j.1365-2605.2009.01044.x.

Sathyanarayana S, Grady R, Redmon JB, Ivicek K, Barrett E, Janssen S, et al. 2015. Anogenital distance and penile width measurements in The Infant Development and the Environment Study (TIDES): Methods and predictors. J Pediatr Urol 11(2):76.e1–e6, PMID: 25824881, 10.1016/j.jpurol.2014.11.018.

Séralini G, Moslemi S. 2001. Aromatase inhibitors: Past, present and future. Mol Cell Endocrinol 178(1-2):117–131, PMID: 11403901, 10.1016/S0303-7207(01)00433-6.

Signorile PG, Spugnini EP, Mita L, Mellone P, D’Avino A, Bianco M, et al. 2010. Pre-natal exposure of mice to bisphenol A elicits an endometriosis-like phenotype in female offspring. Gen Comp Endocrinol 168(3):318–325, PMID: 20350546, 10.1016/j.ygcen.2010.03.030.

Sohoni P, Sumpter JP. 1998. Several environmental oestrogens are also anti-androgens. J Endocrinol 158(3):327–339, PMID: 9846162, 10.1677/joe.0.1580327.

Susiarjo M, Hassold TJ, Freeman E, Hunt PA. 2007. Bisphenol A exposure in utero disrupts early oogenesis in the mouse. PLoS Genet 3(1):e5, PMID: 17222059, 10.1371/journal.pgen.0030005.

Swan SH. 2008. Environmental phthalate exposure in relation to reproductive outcomes and other health endpoints in humans. Environ Res 108(2):177–184, PMID: 18949837, 10.1016/j.envres.2008.08.007.

Swan SH, Main KM, Liu F, Stewart SL, Kruse RL, Calafat AM, et al. 2005. Decrease in anogenital distance among male infants with prenatal phthalate exposure. Environ Health Perspect 113(8):1056–1061, PMID: 16079079.

Swan SH, Sathyanarayana S, Barrett ES, Janssen S, Liu F, Nguyen RH, Redmon JB, et al. 2015. First trimester phthalate exposure and anogenital distance in newborns. Hum Reprod 30(4):963–972, PMID: 25697839, 10.1093/humrep/deu363.

Takahashi O, Oishi S. 2000. Disposition of orally administered 2,2-Bis(4-hydroxyphenyl)propane (Bisphenol A) in pregnant rats and the placental transfer to fetuses. Environ Health Perspect 108(10):931–935, PMID: 11049811.

Teitelbaum SL, Britton JA, Calafat AM, Ye X, Silva MJ, Reidy JA, et al. 2008. Temporal variability in urinary concentrations of phthalate metabolites, phytoestrogens and phenols among minority children in the United States. Environ Res 106(2):257–269, PMID: 17976571, 10.1016/j.envres.2007.09.010.

Teng C, Goodwin B, Shockley K, Xia M, Huang R, Norris J, et al. 2013. Bisphenol A affects androgen receptor function via multiple mechanisms. Chem Biol Interact 203(3):556–564, PMID: 23562765, 10.1016/j.cbi.2013.03.013.

Thankamony A, Pasterski V, Ong KK, Acerini CL, Hughes IA. 2016. Anogenital distance as a marker of androgen exposure in humans. Andrology 4(4):616–625, PMID: 26846869, 10.1111/andr.12156.

Vandenberg LN, Maffini MV, Sonnenschein C, Rubin BS, Soto AM. 2009. Bisphenol-A and the great divide: A review of controversies in the field of endocrine disruption. Endocr Rev 30(1):75–95, PMID: 19074586, 10.1210/er.2008-0021.

Vandenberg LN, Maffini MV, Wadia PR, Sonnenschein C, Rubin BS, Soto AM. 2007. Exposure to environmentally relevant doses of the xenoestrogen bisphenol-A alters development of the fetal mouse mammary gland. Endocrinology 148(1):116–127, PMID: 17023525, 10.1210/en.2006-0561.

vom Saal FS, Hughes C. 2005. An extensive new literature concerning low-dose effects of bisphenol A shows the need for a new risk assessment. Environ Health Perspect 113(8):926–933, PMID: 16079060.

Vom Saal FS, Nagel SC, Coe BL, Angle BM, Taylor JA. 2012. The estrogenic endocrine disrupting chemical bisphenol A (BPA) and obesity. Mol Cell Endocrinol 354(1-2):74–84, PMID: 22249005, 10.1016/j.mce.2012.01.001.

Watkins DJ, Téllez-Rojo MM, Ferguson KK, Lee JM, Solano-Gonzalez M, Blank-Goldenberg C, et al. 2014. In utero and peripubertal exposure to phthalates and BPA in relation to female sexual maturation. Environ Res 134:233–241, PMID: 25173057, 10.1016/j.envres.2014.08.010.

Welsh M, Saunders PT, Fisken M, Scott HM, Hutchison GR, Smith LB, et al. 2008. Identification in rats of a programming window for reproductive tract masculinization, disruption of which leads to hypospadias and cryptorchidism. J Clin Invest 118(4):1479–1490, PMID: 18340380, 10.1172/JCI34241.

Wetherill YB, Akingbemi BT, Kanno J, McLachlan JA, Nadal A, Sonnenschein C, et al. 2007. In vitro molecular mechanisms of bisphenol A action. Reprod Toxicol 24(2):178–198, PMID: 17628395, 10.1016/j.reprotox.2007.05.010.

WHO (World Health Organization). 2011. Toxicological and health aspects of bisphenol A. In: Joint FAO/WHO Expert Meeting to Review Toxicological and Health Aspects of Bisphenol A: Final Report, Including Report of Stakeholder Meeting on Bisphenol A, 1-5 November 2010, Ottawa, Canada. Geneva, Switzerland: World Health Organization.

WHO Multicentre Growth Reference Study Group. 2009. WHO Child Growth Standards: Growth Velocity Based on Weight, Length, and Head Circumference. Methods and Development. Geneva, Switzerland:World Health Organization.

Wolff MS, Engel SM, Berkowitz GS, Ye X, Silva MJ, Zhu C, et al. 2008. Prenatal phenol and phthalate exposures and birth outcomes. Environ Health Perspect 116(8):1092–1097, PMID: 18709157, 10.1289/ehp.11007.

Wolstenholme JT, Rissman EF, Connelly JJ. 2011. The role of bisphenol A in shaping the brain, epigenome and behavior. Horm Behav 59(3):296–305, PMID: 21029734, 10.1016/j.yhbeh.2010.10.001.

Xi W, Lee CK, Yeung WS, Giesy JP, Wong MH, Zhang X, et al. 2011. Effect of perinatal and postnatal bisphenol a exposure to the regulatory circuits at the hypothalamus-pituitary-gonadal axis of CD-1 mice. Reprod Toxicol 31(4):409–417, PMID: 21182934, 10.1016/j.reprotox.2010.12.002.

Ye X, Kuklenyik Z, Needham LL, Calafat AM. 2005. Quantification of urinary conjugates of bisphenol A, 2,5-dichlorophenol, and 2-hydroxy-4-methoxybenzophenone in humans by online solid phase extraction-high performance liquid chromatography-tandem mass spectrometry. Anal Bioanal Chem 383(4):638–644, PMID: 16132150, 10.1007/s00216-005-0019-4.

Prostate Cancer Risk and DNA Methylation Signatures in Aging Rats following Developmental BPA Exposure: A Dose–Response Analysis

Author Affiliations open

1Department of Urology, College of Medicine, University of Illinois at Chicago, Chicago, Illinois, USA

2University of Illinois Cancer Center, Chicago, Illinois, USA

3Department of Environmental Health and Center for Environmental Genetics, University of Cincinnati College of Medicine, Cincinnati, Ohio, USA

4Cincinnati Cancer Center, Cincinnati, Ohio, USA

5Cincinnati Veteran Affairs Hospital Medical Center, Cincinnati, Ohio, USA

6Department of Medicinal Chemistry & Pharmacognosy, College of Pharmacy, University of Illinois at Chicago, Chicago, Illinois, USA

PDF icon PDF Version (3 MB)

  • Background:
    Previous studies have uncovered heightened prostatic susceptibility to hormone-induced neoplasia from early-life exposure to low-dose bisphenol A (BPA). However, significant data gaps remain that are essential to address for biological relevance and necessary risk assessment.
    Objectives:
    A complete BPA dose–response analysis of prostate lesions across multiple prostatic lobes was conducted that included internal BPA dosimetry, progression to adenocarcinoma with aging and mechanistic connections to epigenetically reprogramed genes.
    Methods:
    Male neonatal Sprague-Dawley rats were briefly exposed to 0.1 to 5,000 μg BPA/kg BW on postnatal days (PND) 1, 3, and 5. Individual prostate lobes plus periurethral prostatic ducts were evaluated at 7 mo or 1 y of age without or with adult testosterone plus estradiol (T+E) to promote carcinogenesis. DNA methylation of five genes was quantified by bisulfite genomic sequencing in d-200 dorsal prostates across BPA doses. Serum free-BPA and BPA-glucuronide were quantitated in sera of individual PND 3 pups collected 1 hr postexposure utilizing ultra-high-pressure tandem mass spectrometry (UHPLC-MS-MS).
    Results:
    The lowest BPA dose initiated maximal hormonal carcinogenesis in lateral prostates despite undetectable free BPA 1 hr postexposure. Further, prostatic intraepithelial neoplasia (PIN) progressed to carcinoma in rats given neonatal low-dose BPA with adult T+E but not in rats given adult T+E alone. The dorsal and ventral lobes and periurethral prostatic ducts exhibited a nonmonotonic dose response with peak PIN, proliferation and apoptotic values at 10–100 μg/kg BW. This was paralleled by nonmonotonic and dose-specific DNA hypomethylation of genes that confer carcinogenic risk, with greatest hypomethylation at the lowest BPA doses.
    Conclusions:
    Developmental BPA exposures heighten prostate cancer susceptibility in a complex dose- and lobe-specific manner. Importantly, elevated carcinogenic risk is found at doses that yield undetectable serum free BPA. Dose-specific epigenetic modifications of selected genes provide a mechanistic framework that may connect early-life BPA to later-life predisposition to prostate carcinogenesis. https://doi.org/10.1289/EHP1050
  • Received: 1 September 2016
    Revised: 13 January 2017
    Accepted: 19 January 2017
    Published: 11 July 2017

    Address correspondence to G.S. Prins, Dept. of Urology, University of Illinois at Chicago, 820 South Wood St., M/C 955, Chicago, IL 60612 USA. Telephone: (312) 413-9766. Email: gprins@uic.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP1050).

    The authors declare they have no actual or potential competing interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (935 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (105 KB)

Introduction

Bisphenol A (BPA), a high-volume chemical and widely used synthetic plasticizer, has known estrogenic activity and is a recognized endocrine-disrupting chemical (EDC). Extensive studies over the past two decades have evaluated its potential effects in multiple organs and biological systems including its capacity for augmenting carcinogenesis (Chapin et al. 2008; Gore et al. 2015; Rochester 2013; Seachrist et al. 2016). Although research consensus has not been reached on adverse outcomes from BPA doses relevant to human exposures, it is widely appreciated that the developmental period is particularly sensitive to BPA exposures that may lead to long-lasting effects over the life span.

The prostate gland is a hormone-dependent reproductive organ that possesses a high rate of disease with aging. Currently, prostate cancer is the most common noncutaneous cancer and the second leading cause of cancer-related deaths in U.S. men (Siegel et al. 2016). Whereas androgens are essential for prostate growth and function, substantial evidence indicates that estrogens play key roles in prostate homeostasis and disease (Nelles et al. 2011). Importantly, inappropriate estrogen exposures during prostate development, in terms of timing, type and dose, can reprogram the gland, drive differentiation defects and predispose to an increased risk of prostate cancer (Prins et al. 2001; Prins and Ho 2010). Work from our laboratory (Ho et al. 2006; Prins et al. 2011), recently confirmed in an independent study (Wong et al. 2015), determined that brief early-life exposure to low-dose BPA (10–50 μg/kg BW), although not sufficient to induce prostate lesions on its own, increased susceptibility to estrogen-driven prostatic intraepithelial neoplasia (PIN) in adulthood. This is germane because relative estradiol levels rise in the aging male (Vermeulen et al. 2002), estrogens can transform adult prostate epithelium (Bosland et al. 1995; Hu et al. 2011), accelerate cancer progression (Chakravarty et al. 2014; Setlur et al. 2008; Takizawa et al. 2015) and estrogen activity is amplified in advanced disease (Montgomery et al. 2008). Thus we posit that BPA reprograms prostate cells early in life resulting in a cellular memory that augments adult hormonal sensitivity. The molecular underpinnings of reprogramed prostatic memory appear to lie in epigenetic modifications that have been identified in prostates exposed perinatally to low-dose BPA that poise the cells for amplified responses to later estrogenic exposures. These include hypo- or hypermethylation of DNA that directly modifies gene expression (Cheong et al. 2016; Ho et al. 2006; Tang et al. 2012), alterations in histone methylation marks that directly change gene transcription or prime the gene for elevated response to transcriptional signals in later life (Wang et al. 2016), as well as changes in the expression of noncoding RNAs (Ho et al. 2015). These epigenetic modifications initiated by BPA are mediated, in part, by changes in the activity of DNA methyltransferases (DMNTs), methyl-CpG binding domain proteins (Mbd2/4) and histone methyltransferases (HMTs) (Tang et al. 2012; Wang et al. 2016). Recent work from our laboratory has further identified that human prostate stem and progenitor cells are direct targets of BPA exposures leading to epigenomic modifications and, due to their long-lived nature, increased carcinogenic susceptibility (Calderon-Gierszal and Prins 2015; Ho et al. 2015; Prins et al. 2014; Prins et al. 2015).

Despite inroads that have been made in identifying BPA’s potential role in prostate carcinogenesis, significant knowledge gaps remain that have hindered full utilization of this work in risk assessment analysis (Chapin et al. 2008). First, a dose–response study of the prostatic carcinogenic response to BPA has not been undertaken and is necessary across a range of exposures that include average human exposure levels and occupational risks. Another critical factor is accurate determination of internal free-BPA levels soon after exposure, irrespective of exposure route, to provide environmental applicability of the responses noted as well as knowledge of the precise BPA quantities to which animals and tissues are exposed. An essential element that remains unresolved is whether the precancerous PIN lesions found in prior studies that implicate BPA-driven carcinogenic susceptibility can in fact progress to prostate cancer. Finally, connection of epigenetic reprogramed genes to carcinogenesis and documentation of altered DNA methylation in a dose-responsive manner is necessary to determine the pathological relevance of these molecular modifications.

The present study sought to address these critical elements by a multi-pronged approach. We first undertook a large dose–response analysis of rat prostatic lesions at 7 mo of age as a function of early-life BPA exposures, both without and with a 2-fold increase in adult estradiol levels to promote carcinogenesis. Endpoint analysis included detailed histopathology, apoptosis/proliferation assessments and separate examination of the lateral, dorsal, ventral prostate lobes and periurethral prostatic ducts—all with known differential sensitivities to hormonal carcinogenesis. This is especially important because some past BPA evaluations have only examined the larger ventral lobe, which has no homology in the human prostate gland (Price 1963). Periurethral prostatic ducts are particularly susceptible to estrogen-driven cancers (Bosland et al. 1995) and provide added value to the present dataset. Next, the progression of high-grade PIN lesions to microinvasion and prostate adenocarcinoma was evaluated by undertaking a properly powered study to 1 y of age. Internal dosimetry for all BPA doses in individual rat pups was addressed through development of low volume quantitative capacity for both free-BPA and BPA-glucuronide (BPA-G) using UHPLC-MS-MS. Finally, we extended our ongoing DNA methylation analysis of identified reprogramed genes (Cheong et al. 2016) across a dose-range and found that permanent methylation changes were often most robust at the lowest BPA exposure levels. Together, these results firmly document that developmental exposures to low, environmentally relevant levels of BPA modify the prostatic DNA methylome in a dose-responsive manner and drive a significant increase in rat prostate cancer incidence that is lobe-specific and BPA dose-dependent.

Materials and Methods

Animal Housing and Treatments

All animals were treated humanely and with regard for alleviation of suffering, using protocols approved by the Animal Use Committee at UIC.Timed pregnant Sprague-Dawley rats between 3 and 6 mo of age (Zivic-Miller Laboratories, Pittsburgh, PA) were shipped on gestation d 12 and housed under strict conditions as described (Prins et al. 2011). Rooms were maintained at 21°C with 50% relative humidity and a 14-hr:10-hr light:dark schedule. All rats were housed in polysulfone solid-bottom cages with steel covers and double deionized water was supplied from glass bottles. Animals were fed ad libitum a soy-free, phytoestrogen-reduced diet (Zeigler Reduced Rodent Diet 2, Ziegler Bros, Inc., Gardners, PA). Pregnant dams were monitored and the day of birth was designated postnatal d 0 (PND 0). Litter size was culled to 10 pups on PND 0 by removing or adding female pups.

Newborn male pups were assigned to one of eight neonatal treatment groups with 22–32 pups/group (see Figure S1A). To control for litter effects, male pups in each litter were randomly assigned to different treatment groups and tattooed for permanent identification. The 8 neonatal groups were a) tocopherol stripped corn oil vehicle as controls, b) high dose β17-estradiol 3-benzoate (E2), 2,500 μg E2/kg BW, c) low-dose E2, 0.1 μg E2/kg BW, or 4–8) 0.1, 1.0, 10, 100, or 5,000 μg BPA/kg BW. The highest BPA dose is the current LOAEL for BPA established by the U.S. National Toxicology Program (Chapin et al. 2008). All dosing was administered by subcutaneous (s.c.) depot injection in the nape of the neck on PND 1, 3 and 5 as previously described (Ho et al. 2006; Prins et al. 2011) thus allowing direct comparison of findings with our prior results. The timing of exposures coincides with the d 1–6 critical window characterized for the rodent prostate gland (Pylkkänen et al. 1991). On PND 3, tail vein blood was collected at 60 min postinjection from male rats in control and BPA dose groups (n=6–8/group) using Microvette® CB300 capillary tubes (Sarstedt, Newton, NC). Sera was separated and frozen for internal dosimetry measurement of free BPA and BPA-G. All products used for collection and storage of samples were confirmed to be free of BPA contamination. The pups were weaned at PND 21 and housed two/cage.

At PND 90, approximately half of the rats from each neonatal treatment group were given implants of Silastic capsules (Dow Corning, Midland, MI; i.d. 1.02 mm, o.d. 2.16 mm) packed with estradiol (one 1-cm tube) and testosterone (two 2-cm tubes) (T+E) to drive prostate carcinogenesis as described (Ho et al. 2006). The T capsules maintain physiologic testosterone levels and are needed to maintain prostate homeostasis because estrogen treatment alone results in feedback inhibition of endogenous testosterone secretion with resultant prostatic involution. The estrogen capsules double the circulating estradiol levels, which is sufficient to promote prostate cancer in a rat model (Bosland et al. 1995). Fresh tubes were replaced every 8 wk to ensure consistent steroid levels over time. At 7 mo of age (d 200), the animals were sacrificed by decapitation, blood collected and the prostatic-urethral complex quickly removed. This included the bilateral ventral, lateral and dorsal prostate lobes and the periurethral collecting ducts from each lobe that drain into the urethra (see Figure S1B). Using a dissection microscope, a single ventral, lateral and dorsal lobe from one side of each complex was removed, snap frozen, and stored in liquid nitrogen for molecular analysis. The remaining prostate lobes and full urethral complex were fixed en masse in methacarn (BBC Biochemical, Mt Vernon, WA) for 48–72 hr, rinsed and stored in 70% EtOH until histological processing. Serum was separated and frozen for hormone assays.

To examine potential progression of prostatic lesions to carcinoma, 60 control or T+E treated rats were sacrificed at 1 y of age (D365) with a noted loss of ∼20% of rats due to bladder outlet obstruction prior to 1 y. Five treatment groups (n=12/group) included a) neonatal vehicle controls with empty implants at PND90, b) neonatal vehicle controls with T+E implants at PND 90, c) neonatal low-dose E2 (0.1 μg E2/kg BW) with T+E implants at PND90, d) neonatal high-dose E2 (2,500 μg E2/kg BW) with T+E implants at PND90, and e) neonatal 10 μg BPA/kg BW with T+E implants at PND 90. The T+E implants were replaced every 8 wk. At 1 y, the rats were killed by decapitation, blood was collected, and the prostatic complex was dissected and processed for histopathologic diagnosis to assess carcinoma rates. This initial analysis of cancer rates was used for power analysis that determined that a doubling of the animal number would be required for proper calculation of carcinoma incidence. The entire study was then duplicated using identical vendors, diets, conditions, and rat strains for an additional 60 rats. When the histology data was decoded for treatment group, the results were similar to the original cohort and use of Bartlett’s test for homogeneity of variance permitted the pooling of data between the two cohorts.

Histopathology

The fixed prostatic tissues were dehydrated and paraffin embedded as described (Prins et al. 2011) with the ventral, lateral, and dorsal prostate lobes mounted along one plane surrounding the urethral region. Coronal sections of this complex permitted viewing of prostate structures en masse (see Figure S1). Three to four serial sections (4 μm) were made at four levels of the block 150 μm apart to permit pathologic analysis along the tissue depth and 12–16 sections were analyzed for each tissue. The sections were coded to prevent reader bias and stained with hematoxylin and eosin. Each prostatic region was read in a blinded fashion and scored for presence, severity and extent of lesions that include prostatic epithelial hyperplasia, inflammatory cell infiltration, and PIN. The PIN lesions were characterized by the presence of nuclear atypia (enlarged and elongated nuclei, hyperchromasia, prominent nucleoli) with or without aberrant cellular piling. Regions of cribiform pattern and aberrant cell piling without nuclear atypia were scored as atypical hyperplasia. PIN lesions were graded on a 0–3 scale with 0=no atypia, 1=low-grade PIN, 2=focal high-grade PIN (HGPIN) and 3=extensive HGPIN. Other pathology included adenoma and squamous metaplasia and, at 1 y, basement membrane breakdown with local epithelial microinvasion and adenocarcinoma (Shappell et al. 2004). In instances with uncertain diagnosis, a second pathology opinion was obtained and final classification reached by consensus. Once all histopathology diagnosis were completed, data were entered into a secured Excel database by a third party followed by decoding and data analysis. Incidences of the separate prostate lesions in each lobe were analyzed by chi-square and scores were analyzed by ANOVA followed by Fischer’s exact test with significance accepted at p<0.05.

Immunohistochemistry and in Situ Apoptosis Labeling

To assess epithelial proliferation rates, d-200 specimens (n=7−9/group) were immunostained for Ki–67 using a polyclonal Ki–67 primary antibody (1:2,500, Novacastra, Newcastle, UK) with adjacent sections incubated in normal rabbit IgG (0.5 μm/mL) as negative controls. To determine apoptosis rates by TUNEL staining, d-200 sections were reacted with an ApopTag Peroxidase In Situ Apoptosis Detection Kit according to manufacturer’s instructions (Chemicon International, Temecula, CA). To calculate proliferation and apoptotic indices, multiple areas of each lobe were captured with a color digital AxioCam camera on an Axioskop microscope (Carl Zeiss, Inc., Thornwood, NY). Positive and negative Ki–67-stained or TUNEL-labeled epithelial cells were counted using Zeiss Image Version 3.0 (Carl Zeiss) with ∼1,000 cells counted per slide. Data was analyzed by ANOVA and Bonferroni post-tests with p<0.05 considered significant.

Bisphenol A Quantitation

Serum BPA was quantitated using a previously described HPLC-MS-MS methodology that permitted direct and simultaneous measures of free BPA and BPA-G in 25 μl sera (Prins et al. 2014). As a participant in the NIEHS-coordinated round-robin BPA analysis that rigorously examined criteria required for accurate blood BPA measurements (Vandenberg et al. 2014), quality control procedures were incorporated that included procedural blanks (LC-MS grade water; Burdick & Jackson, Honeywell, Muskegon, MI) and experimental blanks (charcoal-dextran stripped serum) analyzed with each experimental run. All equipment and supplies used in sera collection and storage and BPA measurements were confirmed free of BPA contamination. [d6]-BPA (Cambridge Isotope Laboratories, Andover, MA) [rings-13C12]-BPA-G, BPA (Sigma-Aldrich) and bisphenol A mono-β-D-glucuronide (BPA-G; Midwest Research Institute, Kansas City, MO) were used as standards. Spiked BPA samples were analyzed for measurement accuracy.

Stock and working solutions of BPA and BPA-G were prepared in HPLC-grade methanol. Calibration standards were prepared by mixing 1 μL of each working solution with 24 μL blank rat serum. Sera from the 5,000 μg BPA/kg BW rats were initially run undiluted and repeated with 1:50 dilutions to ensure accuracy. All other samples were measured undiluted and deidentified for treatment groups. Each serum sample (25 μL) or calibration standard was mixed with 100 μL HPLC-grade acetonitrile containing the surrogate standards 5 ng/mL[d6]-BPA and 5 ng/mL[13C12]-BPA-G, centrifuged at 13,000×g at 4°C for 15 min and the supernatant was removed and evaporated to dryness. The residue was reconstituted in 25 μL of 50% aqueous methanol and a 5-μL aliquot was injected onto the UHPLC-MS-MS system for analysis.

Chromatographic separations were carried out using a Shimadzu (Kyoto, Japan) LCMS-8050 triple quadrupole mass spectrometer equipped with a Shimadzu Nexera UHPLC system. Free BPA and BPA-G were separated on a Waters (Milford, MA) Acquity UPLC BEH (2.1×50 mm, 1.7-μm) C18 column. A 1.5-min linear gradient was used from 10% to 100% acetonitrile in water followed by a hold at 100% for 0.3 min at a flow rate of 0.4 mL/min. Negative ion electrospray mass spectrometry with selected reaction monitoring (SRM) was used for measurement of each analyte using previously detailed SRM transitions (Prins et al. 2014). Data acquisition was carried out using Shimadzu Labsolution software for external calibration curve construction from standards run in each assay. The lower limits of detection (LOD) for free BPA and BPA-G in rat sera were 0.02 and 0.01 ng/mL, respectively, whereas the lower limits of quantitation (LLOQ) for free BPA and BPA-G were 0.2 and 0.1 ng/mL, respectively.

Steroid radioimmunoassays (RIA).

Frozen serum samples for testosterone (T) and estradiol-17β (E2) analysis were shipped to the Ligand Assay and Analysis Core Laboratory (University of Virginia, Charlottesville, VA). Hormone levels were measured using murine RIA kits (TKTT2 for T and TKE21 for E2; Siemens Medical Solutions Diagnostics). Sensitivity for T was 0.1 ng/mL and the intra- and interassay coefficients of variance were 4.0% and 7.1%, respectively. For E2, sensitivity was 10 pg/mL and the intra- and interassay coefficients of variance were 7.1% and 11.6%, respectively.

Bisulfite PCR sequencing analysis.

Genome-wide DNA methylation analysis using Roche-NimbleGen Rat ChIP 385 Promoter methylation arrays previously identified 20 genes with significant differentially methylated promoter regions in d-90 dorsal prostates of rats neonatally exposed to high-dose E2 or 10 μg BPA/kg BW as compared to oil controls (Cheong et al. 2016). Bisulfite sequencing validated the differential methylation patterns in 15 of the 20 genes and seven were confirmed to have gene expression status inversely correlated with promoter methylation status. In the present study, the CpG methylation status of five of these seven confirmed genes—Sox2, Creb314, Paqr4, Pitx3 and Tpd52—was examined across the five neonatal BPA doses and high-dose E2 exposure of d-200 dorsal prostates from rats without exogenous T+E exposures to examine their persistence with aging and whether methylation status exhibited a dosage-dependent response.

Genomic DNA from d-200 dorsal prostates was extracted using DNeasy Blood & Tissue kit (Qiagen, Valencia, CA) with RNase A. Bisulfite sequencing of the gene promoters was conducted using 500-ng genomic DNA and EZ DNA Methylation kits (Zymo Research, Irvine, CA) as described (Cheong et al. 2016). Following 40 cycles of PCR amplification, the amplicons were gel-purified and TA-cloned into pGEM T Easy Vector (Promega, Madison, WI). Plasmids from a single E. coli colony were directly amplified by TempliPhi DNA amplification kits (GE Healthcare, Buckinghamshire, UK) and sequenced (Macrogen USA, Rockville, MD). The methylation status at each CpG site was analyzed using BiQ Analyzer. Five day-200 dorsal prostate samples were used in each group. Promoter methylation at each CpG site was expressed as mean±SEM from four to five samples/group with 4–6 clones/sample. The exception was Creb3/4 of which bisulfite sequencing analysis was performed on pooled samples (n=5) with six clones/sample. For statistical analysis, area under the curve for promoter CpG methylation sites at each dose was calculated and significance determined by two-way ANOVA and Tukey test when compared to vehicle control prostates, with p<0.05 considered significant.

Results

Quantitation of Free and Glucuronidated BPA in D-3 Rat Serum

An essential element for evaluating the biologic relevance of BPA administration across a dose–response range is internal dosimetry measurements that accurately quantitate circulating free-BPA and BPA-G levels shortly after exposure, irrespective of mode of BPA administration. Because BPA was administered on PND 1, 3, and 5, serum samples were collected from individual PND 3 rats via tail vein sampling 60 min after s.c. injection to determine the median free-BPA and BPA-G levels to which the developing tissues were exposed. As shown in Table 1, all vehicle-treated control rats had undetectable (<LOD) levels of free BPA and BPA-G, documenting a contamination-free system. Total BPA rose linearly in rat sera with increasing doses of BPA up to 5,000 μg/kg BW. At the two lowest BPA doses of 0.1 and 1.0 μg BPA/kg BW, free BPA was below the LOD and LLOQ, respectively, whereas BPA-G was <LLOQ and 2.32 μg BPA/kg BW, respectively. Free BPA was first quantifiable in rats treated with 10 μg BPA/kg BW, appearing at 0.76 ng/mL, which is similar to our previous study (Prins et al. 2011) and within the range of fetal and newborn human exposures (Gerona et al. 2013; Padmanabhan et al. 2008), thus implicating direct biologic relevance of this dosage. A nonlinear increase in free BPA was observed in the 100 and 5,000 μg BPA/kg BW treated rats, reaching ∼20 and 1,492 ng/mL, respectively. Direct measurement of BPA-G revealed 33.55 ng/mL in the 10 μg BPA/kg BW treated rats, which is markedly higher than our previous measures using indirect enzymatic treatment and extrapolation for BPA-G (Prins et al. 2011), emphasizing the improved accuracy and importance of direct BPA-G measurements. With this direct quantitative method, % free BPA in PND 3 rats was 2.2%, 7.7%, and 15% of total BPA for 10, 100, and 5,000 μg BPA/kg BW dosing levels, respectively, at 60 min postexposure.

Table 1. Serum BPA levels in individual d 3 rats quantitated by UHPLC-MS-MS.
BPA Concentration (ng/mL)
Treatment (μg BPA/kg BW) Free BPA G-BPA Total BPAa
Vehicle <LOD <LOD <LOD
0.1 <LOD <LLOQ <LLOQ
1 <LLOQ 2.318±0.873 2.318±0.873
10 0.762±0.548 33.552±8.064 34.312±8.538
100 20.32±1.22 244.01±36.98 264.33±36.15
5000 1492.19±652.94 8354.60±1268.70 9846.85±1654.00
Note: BPA in oil administered by subcutaneous injection. Serum collected at 1 h via tail vein. Mean±SEM; n=12–14 pups per group; G-BPA, glucuronidated BPA; LOD, limit of detection; LLOQ, lower limit of quantitation.

aTotal=Free BPA+G-BPA.

Neonatal BPA Exposures Increase Prostatic PIN and Hyperplasia at D 200 in a Lobe- and Dose-Specific Manner

Rats exposed to neonatal hormones without or with adult T+E treatment were aged to d 200 and their prostates assessed in a blinded manner for the presence of PIN, atypical hyperplasia, epithelial hyperplasia and inflammatory cell infiltration. Neonatal exposure to increasing doses of BPA without adult hormone exposure produced minimal prostatic lesions at d 200 with no significant differences compared to aging controls (data not shown). The only group with significant pathologic alterations was neonatal high-dose E2 as previously described by our laboratory (Ho et al. 2006; Prins 1992).

In contrast, marked prostatic pathology was noted in a lobe-specific and dose-responsive manner in all neonatal E2 and BPA treated rats when exposed to T+E for 4 mo during adulthood (Figure 1, Table 2; see also Table S1). It is important to note that the T+E capsules produced identical serum testosterone levels and only 2-fold higher estradiol-17β levels as compared to control rats given empty capsules (see Figure S2), which is sufficient to initiate moderate prostate carcinogenesis with aging in this rat model (Bosland et al. 1995). As expected, the lateral prostate lobe was most sensitive to adult hormone-induced carcinogenesis. PIN incidence (%) and severity score (scale 0–3) increased from 20% and 0.30, respectively, in the neonatal vehicle controls-empty adult capsule group to 67% and 1.46 in neonatal controls given T+E during adulthood (Table 2). Importantly, PIN incidence and scores in the lateral lobe were further increased to near maximal values when exposed early in life to high-E2 and all BPA doses plus adult T+E, with an incidence and score of 100% and 2.54, respectively, at the lowest dose of 0.1&μg BPA/kg BW (Figure 1, Table 2). It is noteworthy that serum free BPA and BPA-G were <LOD and LLOQ, respectively, at 1 hr postexposure in this dosing group, documenting that lateral prostatic effects occur at exceedingly low BPA exposures. Lateral lobe epithelial hyperplasia was prevalent in all aged prostates including neonatal controls without additional adult hormone exposures (empty) (Table 2) and this was not further exacerbated with neonatal E2 or BPA exposures or adult T+E. Of note, the lateral lobes contained marked inflammatory cell infiltration upon adult T+E treatment that was significantly augmented by neonatal high-dose E2 and BPA exposures (Table 2).

Three panels of bar graphs plotting PIN scores (y-axis) across vehicle-treated, low-dose E<sub>2</sub>, high-dose E<sub>2</sub>, and 10 micrograms BPA and 100 micrograms BPA (x-axis).
Figure 1. Prostatic intraepithelial neoplasia (PIN) scores (0–3 scale), % proliferation (Ki67+cells) and % apoptosis (TUNEL labeling) in the lateral, dorsal and ventral prostate lobes and periurethral prostate ducts at d 200. Rats were treated neonatally with vehicle, low-dose E2, high-dose E2 and increasing doses of BPA on d 1, 3, and 5 of life. All rats were given T+E implants at d 90 to drive hormonal carcinogenesis. n=8–11/treatment group. *=p<0.05 and **=p<0.01 vs, neonatal vehicle within each lobe.
Table 2. Lateral prostate (T&E) pathology in d-200 rats.
PIN Hyperplasia Inflammation
Score Incidence Score Incidence Score Incidence
μg/kg BW All HGPIN
Vehicle+Empty 0.3 20% 0% 0.90 40% 0.00 0%
Vehicle+T E 1.46 67% 33% 1.10 60% 0.89 67%
Low E2 1.18 62% 25% 0.38 25% 1.38 75%
High E2 2.46a,c 91% 73%c 0.73 64% 2.00b 100%
0.1 BPA 2.54a,c 100% 70% 0.80 60% 1.70a 90%
1.0 BPA 2.27c 89% 56% 0.56 67% 1.33 89%
10 BPA 2.54a,c 100% 67% 0.67 56% 2.22b 100%
100 BPA 2.18 75% 62% 0.62 38% 1.25 50%
5K BPA 2.75b,c 90% 80%a,c 0.80 80%c 1.90a 100%
(a) vs. Veh+T E p<0.05 NS p<0.05 NS NS p<0.05 NS
(b) vs. Veh+T E p<0.01 p<0.01
(c) vs. Low E2 p<0.05 p<0.05 p<0.05
Note: Scores were analyzed by ANOVA with post hoc Dunnett multiple comparisons.

Incidence was analyzed by chi-square and Fischer’s exact test.

All PIN includes tissues with 1, 2, or 3 PIN scores.

HGPIN includes only 2 and 3 PIN scores. n=8–11/group; NS, nonsignificant.

Prostatic lesions in the dorsal and ventral lobes and periurethral prostatic ducts presented at reduced incidence and severity upon neonatal E2/BPA plus adult hormone exposures when compared to the lateral prostate. This provided an opportunity to assess the dose-responsiveness of neonatal BPA exposures on altering carcinogenic susceptibility. PIN lesions in the neonatal vehicle-adult T+E group were low in these three regions, not affected by neonatal low-dose E2 and increased by high-dose E2 exposure (Figure 1; see also Table S1). In contrast to the lateral lobe, PIN scores exhibited an inverted U-shaped dose response to rising neonatal BPA doses. Although nonsignificant increases were noted at 1 and 10 μg BPA/kg BW as compared to neonatal vehicle controls, a significant increase in PIN scores was observed in dorsal and ventral lobes at 100 μg BPA/kg BW. Notably, this dropped at the high dose of 5,000 μg BPA/kgBW to levels seen with lower-dose BPA. In the periurethral prostatic ducts, PIN scores increased at all BPA doses compared to neonatal vehicle controls with significance at 1, 10, 100, and 5,000 μg BPA/kg BW and peak values at the 10- and 100-μg doses. The decrease PIN score noted at the highest BPA dose suggests that increased carcinogenic response in these regions may be limited to low-dose BPA exposures. Although dorsal lobe hyperplasia was not affected by neonatal E2/BPA (see Table S1), the ventral lobe and periurethral prostatic ducts exhibited significant augmentation in hyperplastic incidence and severity in the high E2 and all BPA doses as compared to adult T+E only (see Table S1). Additionally, inflammation was low in these prostatic regions with only modest and nonsignificant increases from neonatal BPA exposures.

Neonatal BPA Augments Prostatic Epithelial Proliferation and Apoptosis

Perturbations in epithelial proliferation and apoptosis are accepted biomarkers of precancerous prostatic lesions when combined with histopathology (Shappell et al. 2004). To assess the pathological significance of high-grade PIN lesions, tissues from neonatal E2 and selected BPA dose-groups given adult T+E through d 200 were histologically assessed for epithelial proliferation and apoptosis using Ki67 and TUNEL assays. Although not affected by neonatal low-dose E2, the proliferative and apoptotic indices were markedly increased in all prostate regions with neonatal high-dose E2 exposure plus adult T+E (Figure 1), matching the known capacity for carcinoma progression with combined hormonal exposures (Leav et al. 1988; Yuen et al. 2005). Importantly, neonatal exposures to 10 or 100 μg BPA/kg BW increased proliferation in the lateral, dorsal and ventral prostate epithelium to levels observed with high-E2 exposure. Although a marked response to high-E2/adult T+E was seen in the periurethral prostatic ducts, the effect of neonatal BPA was less pronounced with a proliferative increase only noted at 100 μg BPA/kg BW. Matching the proliferative changes, a significant increase in epithelial apoptosis was noted in all prostatic regions at 10 μg BPA/kg BW and a further augmentation was noted at the 100-μg dose (Figure 1). Together, this imbalance in proliferation/apoptosis rates indicates pre-carcinogenic activity and provides objective support for the pathologic relevance of the heightened PIN susceptibility upon neonatal BPA exposures.

Progression to Local Invasion and Adenocarcinoma upon Aging to 1 Y

To directly determine whether the d-200 high-grade PIN lesions progressed to cancer, a cohort of control rats (neonatal vehicle without or with adult T+E) and rats neonatally exposed to low/high E2 or 10 μg BPA/kg BW with adult T+E were aged to 1 y for detailed histopathologic analysis. Although serum testosterone and estradiol were lower at d-365 vs. d-200 rats given empty capsules, the T+E capsules maintained equivalent testosterone and a 2-fold elevation in circulating estradiol-17β in rats at 1 y (see Figure S2). Control rats without adult hormones showed no lesion progression whereas controls with T+E exhibited limited cancerous progression with a 17% incidence of microinvasion, the earliest recognizable form of prostate cancer as evidenced by basement membrane breakdown with epithelial cells invading the stroma (Bostwick 1996; Bostwick and Cheng 2012). However, there were no newly formed malignant acini/glands, used for grading adenocarcinoma (Shappell et al. 2004), in any of the prostatic regions in the vehicle empty or T+E control rats (Figure 2, Table 3; see also Table S2). In contrast, a progressive increase in microinvasion and formation of focal well-differentiated adenocarcinoma glands (carcinoma – glandular) were observed in prostates of rats treated neonatally with E2 or BPA plus adult hormones. The response to low- and high-dose E2 was dose-related in the lateral lobe with maximal incidence of microinvasion (58%; p=0.006) and newly formed adenocarcinoma (carcinoma – glandular incidence of 23%; p=0.032) seen with neonatal high-dose E2 (Table 3). Notably, neonatal 10 μg BPA/kg BW produced an equivalent incidence of microinvasive carcinoma (62%; p=0.004) as high-dose E2 in the lateral lobe, whereas the incidence of newly formed cancerous glands was somewhat lower (carcinoma: glandular incidence of 14%; p=0.079) in the 10-μg BPA-treated rats. Similar progression to microinvasive cancer without or with discreet, newly formed cancerous glands was found in the other prostatic regions for the low-dose BPA group, although this did not reach significance (see Table S2). Together, these findings demonstrate for the first time that neonatal exposure to low-dose BPA combines with rising adult estradiol levels to drive prostate cancer with aging.

Photomicrographs
Figure 2. Prostatic histopathology lesions at d 365 in the lateral lobe (A–G) and periurethral prostate ducts (H) of rats treated neonatally with 10 μg BPA/kg BW and T+E implants at d 90. Examples of HG-PIN and in situ carcinoma (A,C) and local microinvasion of epithelium into stroma (B,D, arrows) were observed in the majority of BPA/T+E treated lateral prostates. Focal regions of well-differentiated adenocarcinoma (arrowheads in E, 10 × and F, 20 × of same region and G from a different lateral lobe) were seen in several lateral prostates as evidenced by irregular small glandular structures within stroma with abortive glandular lumens, back-to-back lumens and loss of basement membranes (arrowheads). Regions of adenocarcinoma were also noted in periurethral prostatic ducts (H, arrowheads). A–D, F–H, same magnification as G with bar=50 μm; E bar=100 μm.
Table 3. Lateral prostate (T&E) pathology in aged (d 365) rats.
PIN Carcinoma-Microinvasion Carcinoma- Glandular Inflammation Hyperplasia
μg/kg BW Score Incidence Incidence Incidence Score Incidence Score Incidence
Vehicle Empty 0.12 1/17 6% 0/17 0% 0/17 0% 0.00 0% 0.76 47%
Vehicle T&E 1.50a 15/18 83%a 3/18 17% 0/18 0% 1.28a 78%a 0.50 39%
Lo E2 T&E 2.00a 12/14 86%a 6/14 43%a 1/14 7% 1.14a 71%a 0.86 57%
Hi E2 T&E 2.50a,b 24/26 92%a 15/26 58%a,b 6/26 23%a,b 1.77a 77%a 1.81a,b 88%a,b
BPA 10 T&E 2.48a,b 20/21 95%a 13/21 62%a,b 3/21 14% 1.43a 95%a 0.67 52%
(a) vs. Veh empty p<0.001 p<0.0001 p<0.01 p<0.01 p<0.001 p<0.0001 p<0.001 p<0.001
(b) vs. Veh T&E p<0.01 NS p<0.01 p<0.05 NS NS p<0.001 p<0.01
Note: Scores were analyzed by ANOVA with post hoc Dunnett multiple comparison.

Incidence was analyzed by chi-square and Fischer’s exact Test. n=14–21/group. NS, nonsignificant.

DNA Methylome reprograming across BPA Doses at D 200

We recently identified multiple genes in d-90 rat dorsal prostates whose promoters bore reprogramed DNA methylation patterns as a result of neonatal E2 and BPA exposures, leading to altered gene transcription (Cheong et al. 2016). In the present study, five of these genes were selected for a dose–response analysis of DNA methylation status using bisulfite PCR sequencing in d-200 dorsal prostates of rats without adult T+E treatment. The five genes (Creb3l4, Tpd52, Pitx3, Paqr4, and Sox2) were selected because they showed the greatest changes in promoter methylation in PND90 dorsal prostates as a function of neonatal E2/BPA exposures as compared to vehicle-treated controls. Although the lateral lobes exhibited the highest PIN and carcinoma rates upon T+E treatment, the dorsal lobes were selected given that a dose–response PIN phenotype incidence was observed thus pathologically validating other dose–response measures. Importantly, the dorsal lobe exhibited minimal inflammatory cells with hormone treatments whereas lateral prostates had a high incidence of inflammatory cell infiltration thus precluding a clean separation of prostatic cells from immune cells in lateral lobes. As shown in Figure 3, Creb314, Tpd52, Pitx3, Paqr4, and Sox2 were hypomethylated at promoter CpG sites by neonatal high-dose E2 and BPA in a dose-dependent manner when compared to vehicle controls. Three dose-dependent patterns emerged. For three genes—Creb314, Tpd52 and Pitx3—significant hypomethylation was observed at the lower BPA doses of 0.1, 1.0, and/or 10 BPA μg/kg BW, with a return towards vehicle control methylation levels at the higher BPA exposures. Of note, the hypomethylation changes in Creb314 and Tpd52 induced by neonatal 0.1 or 1.0 BPA μg/kg BW, respectively, were greater than that observed with high-dose E2. A second methylation pattern was noted in Paqr4 in which high-dose E2 and all BPA doses significantly reduced promoter gene methylation to the same extent as compared to the vehicle control group. The third pattern observed was seen with Sox 2 where hypomethylation first appeared at 1.0 BPA μg/kg BW and gradually decreased further with increasing BPA doses, reaching a nadir at the highest BPA dose of 5,000 μg/kg BW (p<0.01).

Panel A is a graphical representation plotting methylation in percentage (y-axis) across incidence (x-axis). Panel B is a panel of bar graphs plotting methylation in percentage (y-axis) across the dosage (x-axis) for the treatment groups, namely, vehicle, EB, BPA 0.1, BPA 1, BPA 10, BPA 100, BPA 5,000.
Figure 3. Dose–response analysis of promoter DNA methylation patterns of five previously identified E2/BPA reprogramed rat prostate genes (Cheong et al. 2016). A) Promoter methylation at each CpG site is expressed as mean±SEM from four to five samples/group with 4–6 clones/sample for each BPA dose, E2 treatment and vehicle control. The exception is Creb3/4 of which bisulfite sequencing analysis was performed on pooled samples (n=5) with six clones/sample. B) Mean % methylation of all promoter-region CpG sites combined across the separate doses. The dotted line represents the vehicle control total promoter % methylation for comparison.

Discussion

The present study fills several critical data gaps necessary to thoroughly assess the influence of developmental BPA, at levels relevant to daily human exposures, on prostate cancer risk with aging. Most notably, it provides a logarithmic dose–response analysis across separate rat prostate lobes, direct measurement of internal free BPA and BPA-G across doses in individual rat pups, documentation of PIN lesion progression to locally invasive carcinoma, and mechanistic connections to epigenetically reprogramed genes across multiple doses. At all doses tested, early-life BPA exposure alone was insufficient to drive prostate precancerous lesions with aging, which substantiates previous findings (Ho et al. 2006; Milman et al. 2002; Yoshino et al. 2002) and indicates that up to the current LOAEL (5 mg/kgBW), BPA is not a complete carcinogen in the prostate. Nevertheless, we found that BPA exposures across multiple doses, prostatic lobes and endpoints act in conjunction with elevated estradiol levels, as seen with aging, to heighten prostate carcinogenesis and progression. Combined with previously published studies (Cheong et al. 2016; Ho et al. 2006; Ho et al. 2015; Seachrist et al. 2016; Tang et al. 2012; Wang et al. 2016; Wong et al. 2015), the present work further supports the postulation that early-life BPA exposure acts on estrogen-sensitive prostate cells to epigenetically reprogram and prime selective genes for enhanced responses to later-life estrogenic triggers. To that extent, BPA may be considered an epigenetic initiator of tumorigenesis during early development that results in increased cancer risk to rising estrogens that act as a promoter (Sharma et al. 2010).

Specific BPA dose–response patterns were observed across the separate prostatic regions, which aligns with known differential sensitivities to multiple hormones, including estrogens, in the prostate gland (Bosland et al. 1995; Prins 1987; Prins 1989; Prins 1992). Of particular note, the lateral prostate lobe exhibited maximal carcinogenic susceptibility to adult estradiol when neonatally exposed to 0.1 μg BPA/kg BW, that is, at a 100-fold lower dose than previously reported (Ho et al. 2006; Prins et al. 2011), despite no detectable free BPA or BPA-G 1 hr after exposure. This indicates that markedly lower BPA doses will be required to reach a LOAEL for the rat prostate. Remarkably, this low-dose BPA response was equivalent to that found with neonatal high-dose E2. These results are similar to mammary gland lesions induced by perinatal exposure to the very low dose of 25 ng BPA/kg BW (Durando et al. 2007) and emphasize that BPA levels within the range of human gestational exposures can predispose to adverse effects with aging. That similar responses were observed with BPA and high-dose E2 treatment, but not the low-dose E2 treatment suggests that BPA may act through additional pathways beyond estrogen receptors as has been documented in many studies including our recent findings with human prostate progenitor cells (Delfosse et al. 2014b; Ho et al. 2015).

Relative to the lateral prostate, the less estrogen-sensitive dorsal and ventral lobes and periurethral prostatic ducts exhibited a nonmonotonic dose response for T+E-induced PIN incidence and scores. That a rising carcinogenic response was observed across these three regions with rising neonatal BPA levels through the 100-μg/kgBW dose, supports the biologic relevance of the response. Of note, a significant increase in PIN scores was found in periurethral ducts at 1 μg BPA/kg BW, when serum free BPA was <LLOQ, and rose to peak values at the 10 μg dose when mean serum free BPA was 0.76 ng/mL, a value reported in some pregnant women (Gerona et al. 2013). That peak neoplastic responses were observed between 10 and 100 μg BPA/kgBW and dropped to lower responses at the 5,000-μg/kg dose in these three prostatic regions is likely explained by the multiple receptors and signaling pathways that can be engaged by different BPA levels, leading to a complex response pattern over the wide range of exposures tested herein (Delfosse et al. 2014a; Viñas and Watson 2013). Similar shaped BPA dose responses have been reported for other organs and benign prostatic growth (Vandenberg et al. 2012; vom Saal et al. 1997) and the current data extend this to carcinogenic susceptibility.

A complete prostate lobe-specific analysis was essential for multiple reasons: a) the rat dorsal and lateral lobes have homology in humans, both embryologically and histologically, whereas there is no human homolog for the ventral prostate (Price 1963), b) the individual lobes express specific genes, particularly the high-expression genes that encode lobe-specific secretory proteins (Gerhardt et al. 1983), c) the ventral prostate is larger than the dorsal and lateral lobes combined and analysis of genes in entire prostatic complex will mask significant changes in the smaller lateral and dorsal lobes, and d) the ventral lobe undergoes branching morphogenesis between PND 1 and 6, whereas the lateral and dorsal lobes are delayed by 4–5 d, thus studies prior to 1 wk of age will be primarily ventral in nature (Hayashi et al. 1991). This may explain why a recent study that examined global genomic DNA methylation % and gene expression changes in the d-4 and -90 full prostate complex was not able to identify a dose–response effect in the prostate (Camacho et al. 2015). This further emphasizes the necessity for lateral and dorsal prostate evaluations given that they are considered the human homolog in rodent studies and will have the greatest applicability to prostate diseases in aging men.

The pathological relevance of the PIN lesions to prostate cancer was confirmed in the present studies in two ways. First, significantly accelerated rates of epithelial proliferation and apoptosis were found in all prostate regions at the 10–100 μg BPA/kg BW doses when elevated and maximal PIN scores were noted, respectively. These aberrant cell turn-over rates are considered key evidence of similarity to human high-grade PIN, the precursor to prostate cancer (Shappell et al. 2004). Second, and most importantly, the prostate PIN lesions progressed to microinvasion and adenocarcinoma in the lateral lobes of rats treated with neonatal BPA (10 μg/kg BW) or E2 plus adult T+E when aged to 1 y as compared to neonatal vehicle-controls. Although not significant, similar trends were noted in the dorsal lobe and periurethral ducts aged to 1 y. That adult T+E exposures alone had limited invasion (17%) and no glandular carcinoma in the lateral prostate, whereas the addition of neonatal BPA produced a 62% incidence of microinvasive carcinoma and 14% glandular carcinoma incidence clearly confirms, for the first time, that early-life BPA exposures heighten prostate cancer risk.

The route of BPA exposure in the present study deserves discussion. A prior study from our laboratory directly compared the BPA pharmacokinetics in PND 3 rat pups after exposure to 10 μg BPA/kg BW by s.c. injection or orally and further assessed prostatic lesions in adult T+E treated rats on d 200 (Prins et al. 2011). Whereas free BPA was significantly higher at 30 min after s.c. injection compared with oral exposure, this rapidly declined by 1–2 hours to equivalent levels in the two exposure groups with serum values matching the present study. Importantly, the prostate exhibited nearly identical heightened susceptibility to PIN lesions with either exposure route suggesting that the early metabolic differences in free BPA did not influence the pathologic outcomes. A recent independent study comparing exposure routes in PND 3 rats reported equivalent serum free-BPA levels 1 hr postexposure in pups given 10 μg BPA/kg BW by s.c. injection and oral administration of 50 μg BPA/kg BW, the current BPA NOAEL, with similar carcinogenic risk at both doses (Wong et al. 2015). Although the present experiments utilized the s.c. injection route to permit direct comparisons with previous datasets, we directly measured the internal BPA dosimetry in the neonatal pups, and most importantly, find significant prostatic lesions when free BPA was below detection limits. It is now appreciated that humans are exposed to BPA through multiple routes including ingestion, skin absorption, inhalation, and intravenous medical devices with which newborns are in increasing contact (Duty et al. 2013; Hines et al. 2017; Vandenberg et al. 2007). As such, studies that utilize nonoral routes should not be discounted a priori and may in fact encompass the variety of human exposures to this chemical. Thus internal dosimetry measures utilizing a contamination-free system as performed herein should be the ultimate determinant of applicability of study results for risk assessment and relevance.

The measurement of free BPA and BPA-G in 25 μL of sera permitted direct monitoring of BPA levels in individual exposed rats in this study, markedly increasing the robustness of the present results. Prior work on rat neonates required pooling sera from 10 pups for BPA quantitation (Prins et al. 2011; Wong et al. 2015), extrapolation from higher-dose exposures or utilization of isotopically labeled tracers (Doerge et al. 2010). Further, until recently, direct BPA-G measures were not possible due to lack of an available standard, relying instead on enzymatic digestion for indirect BPA-G calculations (Doerge et al. 2010; Prins et al. 2011). Using a direct assay, the current study found that the BPA-G levels were considerably higher than we previously observed in PND3 rats and result in a 2–15% level of free BPA that closely aligns with studies by Doerge for PND-4 rats (Doerge et al. 2010). The higher levels of free BPA with increasing BPA doses is expected due to the immaturity of UDP glucuronosyltransferase 2 family, polypeptide B1 (Ugt2b1) expression and thus limited capacity for glucuronidation in the neonatal liver (Matsumoto et al. 2002). As a result, the 100-μg BPA/kg BW dose, which produced peak PIN lesions in the dorsal and ventral prostates, resulted in 20 ng/mL free BPA 1 hr postexposure that, while far higher than daily exposures reported in the adult human population, has been reported in mid-gestational sera of pregnant women using a documented contamination-free sera collection and assay (Gerona et al. 2013).

Complementing the nonmonotonic dose response in carcinogenic risk, we herein identified nonmonotonic dose responses for DNA hypomethylation in several gene promoters of d-200 dorsal prostates, with three of five examined genes (Creb3L4, Tpd52, Pitx3) showing marked changes only at the lowest BPA doses. This is supported by recent reports of nonmonotonic DNA methylation responses to low- and high-dose BPA in the mouse and human fetal liver (Faulk et al. 2015; Kim et al. 2014). In addition to the nonmonotonic low-dose methylation marks, other patterns emerged including consistent Paqr4 promoter hypomethylation at all doses and increasing Sox2 promoter hypomethylation with increasing BPA levels. The five prostate genes examined at d 200 in the present study were selected from a previous genome-wide DNA methylation analysis of the same rat cohort that identified 86 genes differentially methylated by neonatal BPA exposure in d-90 dorsal prostates as compared to neonatal vehicle controls (Cheong et al. 2016). These genes were specifically chosen as potential BPA- reprogramed gene candidates because they exhibited an inverse relationship between DNA methylation and gene expression at d 90, long after neonatal BPA exposure. That differential methylation modifications persist through d 200 indicates that these genes are permanently reprogramed throughout life by developmental BPA exposures. Our previous work found that upon neonatal low-dose BPA treatment, the prostatic de novo DNA methylation transferases (Dmnt3a/b) and methyl-CpG binding domain proteins (Mbd2/4) were elevated throughout life and modified by adult estrogens which may underpin the dynamic alterations in DNA methylation marks across genes, doses and with aging (Tang et al. 2012). It is important to note that the hypomethylation marks in the present study were quantified in the d-200 dorsal prostate in the absence of adult hormone treatment. Given that the onset of dorsal lobe PIN lesions and microinvasion was only observed with the addition of adult T+E, the present data suggest that the altered gene methylomes may contribute to increased carcinogenic susceptibility to adult estrogen exposure. Utilizing the same rat prostate model, a parallel mechanism of neonatal BPA- reprogramed genes in the KEGG prostate cancer pathway was recently identified (Wang et al. 2016). In those studies, persistent H3K4me3 marks were elevated at several cancer-associated gene promoters, a function of increased histone methyltransferase mixed-lineage leukemia 1 (MLL1) activation by BPA, which primed them for enhanced sensitivity to T+E induction in the adult prostate. Undoubtedly several epigenomic modifications initiated by early-life BPA, including altered noncoding RNAs such as SNORDs as recently shown in prostate progenitor cells (Cheong et al. 2016; Faulk et al. 2015), act in parallel to poise the prostate gland for increased carcinogenic risk as a function of later-life events, supporting the proposal that BPA acts as an epigenetic initiator in early life as epigenetic marks are established in developing tissues.

Given that the five BPA- reprogramed genes examined herein were hypomethylated with aging, their expression may be aberrantly elevated in the aging prostate. Paqr4 (progestin and adipoQ receptor family member 4) was hypomethylated at equivalent levels by all BPA doses and matched high-dose E2, suggesting maximal and sustained expression changes at low-to-high BPA exposures. Although studies on Paqr4 are limited, AdipoQ, its ligand, plays an important role in cellular metabolism and abnormal expression has been associated with prostate cancer risk (Kaklamani et al. 2011). Pitx3 is a homeobox transcription factor with established roles in development and stem cells, although its function in the prostate has not been examined. Interestingly, although Tpd52 and Creb3L4 were hypermethylated in d-90 prostates by neonatal BPA/E2 exposures with resultant down regulation of gene expression (Cheong et al. 2016), the profiles of these two genes was reversed with aging, becoming significantly hypomethylated by d 200. This is noteworthy because Tpd52 and Creb3L4, along with Sox 2, are upregulated in prostate cancer (Jia et al. 2011; Labrie et al. 2008; Tennstedt et al. 2014) and, along with Pitx3, have established roles in stem cells. Because epigenetic mechanisms are central to maintaining stem cell identity, disruption of their epigenome may give rise to high-risk progenitor populations that readily become neoplastic upon gain of additional insults throughout life (Sharma et al. 2010). Taken together with our recent findings that BPA targets human embryonic and adult prostate stem cells and modifies their epigenome (Calderon-Gierszal and Prins 2015; Ho et al. 2015; Prins et al. 2014; Prins et al. 2015) and that chronic low-dose BPA treatment alters rat prostate stem cell homeostasis (Hu et al. 2015), a picture emerges whereby early-life BPA exposure heightens later-life carcinogenic risk by permanently modifying prostate stem and progenitor cells for increased sensitivity to carcinogenic hits. Although stem and progenitor cells were not studied in the present experiments, the epigenetic modifications of certain prostate genes related to stemness found herein leads us to postulate that BPA exposures might epigenetically poise prostate epithelium towards a cancer stem cell phenotype following additional hormonal stimuli. Of particular note, these genes were found to be associated with recurrence-free survival in human prostate cancer when the Cancer Genome Atlas (TCGA) data set was interrogated, indicating potential clinical relevance of these genes to human prostate cancer (Cheong et al. 2016).

In summary, the present study extends our knowledge on the effects of developmental BPA on adult prostate health, documenting a dose-dependent effect, providing internal BPA dosimetry for direct comparisons to humans and showing that low-dose exposures heighten the risk for developing lesions that progress to prostate cancer. That PIN progresses to adenocarcinoma provides the required biologic and pathologic relevance of these earlier lesions and raises the bar for adverse outcomes due to early-life BPA exposures. Further, dose-specific modifications in the DNA methylome of genes connected to human prostate cancer provide a mechanistic framework for connecting early-life exposures to later-life disease risk. Together, this data reinforces the assertion that low-dose BPA at levels comparable to human exposures negatively affects prostatic health. This may bear particular relevance to at-risk populations for developing prostate cancer due to race, metabolic polymorphisms, hormonal therapeutics and/or genetics.

Acknowledgments

This study was supported by grants from the National Institute of Environmental Health Sciences/National Institutes of Health (NIH): R01ES015584 (G.S.P., S.-M.H.), RC2ES018758 (G.S.P., S.-M.H.), P30ES006096 (S.-M.H.) and the Michael Reese Research and Education Foundation (G.S.P.). The authors gratefully thank G. Shi, D. Hu (Department of Urology, UIC) for their technical assistance; M. Bosland (Department of Pathology, UIC) for pathologic consultations; J. Ying (UC) for biostatistics consultation; M. Medvedovich (UC) for Bioinformatics assistance; and the Genomics, Epigenomics and Sequencing Core (UC) for the array service. The University of Virginia’s Center for Research in Reproduction Ligand Assay and Analysis Core performed steroid RIA services supported by Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD)/NIH grant U54-HD28934.

References

Bosland MC, Ford H, Horton L. 1995. Induction at high incidence of ductal prostate adenocarcinomas in NBL/Cr and Sprague-Dawley Hsd:SD rats treated with a combination of testosterone and estradiol-17β or diethylstilbestrol. Carcinogenesis 16(6):1311–1317, PMID: 7788848.

Bostwick DG. 1996. Progression of prostatic intraepithelial neoplasia to early invasive adenocarcinoma. Eur Urol 30(2):145–152, PMID: 8875195.

Bostwick DG, Cheng L. 2012. Precursors of prostate cancer. Histopathology 60(1):4–27, PMID: 22212075, 10.1111/j.1365-2559.2011.04007.x.

Calderon-Gierszal E, Prins GS. 2015. Directed differentiation of human embryonic stem cells into prostate organoids in vitro and its perturbation by low-dose bisphenol A exposure. PloS One 10(7):e0133238, PMID: 26222054, 10.1371/journal.pone.0133238.

Camacho L, Basavarajappa MS, Chang CW, Han T, Kobets T, Koturbash I, et al. 2015. Effects of oral exposure to bisphenol A on gene expression and global genomic DNA methylation in the prostate, female mammary gland, and uterus of NCTR Sprague-Dawley rats. Food Chem Toxicol 81:92–103, PMID: 25862956, 10.1016/j.fct.2015.04.009.

Chakravarty D, Sboner A, Nair SS, Giannopoulou E, Li R, Hennig S, et al. 2014. The oestrogen receptor alpha-regulated lncRNA NEAT1 is a critical modulator of prostate cancer. Nat Commun 5:5383, PMID: 25415230, 10.1038/ncomms6383.

Chapin RE, Adams J, Boekelheide K, Gray LE Jr, Hayward SW, Lees PS, et al. 2008. NTP-CERHR expert panel report on the reproductive and developmental toxicity of bisphenol A. Birth Defects Res B Dev Reprod Toxicol 83(3):157–395, PMID: 18613034, 10.1002/bdrb.20147.

Cheong A, Zhang X, Cheung YY, Tang WY, Chen J, Ye SH, et al. 2016. DNA methylome changes by estradiol benzoate and bisphenol A links early-life environmental exposures to prostate cancer risk. Epigenetics 11(9):674–689, PMID: 27415467, 10.1080/15592294.2016.1208891.

Delfosse V, Grimaldi M, Cavaillès V, Balaguer P, Bourguet W. 2014a. Structural and functional profiling of environmental ligands for estrogen receptors. Environ Health Perspect 122(12):1306–1313, PMID: 25260197, 10.1289/ehp.1408453.

Delfosse V, Grimaldi M, le Maire A, Bourguet W, Balaguer P. 2014b. Nuclear receptor profiling of bisphenol-A and its halogenated analogues. Vitam Horm 94:229–251, PMID: 24388193, 10.1016/B978-0-12-800095-3.00009-2.

Doerge D, Twaddle NC, Vanlaningham M, Fischer JW. 2010. Pharmacokinetics of bisphenol-A in neonatal and adult Sprague-Dawley rats. Toxicol Appl Pharmacol 247(2):158–165, PMID: 20600215, 10.1016/j.taap.2010.06.008.

Durando M, Kass L, Piva J, Sonnenschein C, Soto AM, Luque EH, et al. 2007. Prenatal bisphenol A exposure induces preneoplastic lesions in the mammary gland in Wistar rats. Environ Health Perspect 115(1):80–86, PMID: 17366824, 10.1289/ehp.9282.

Duty SM, Mendonca K, Hauser R, Calafat AM, Ye X, Meeker JD, et al. 2013. Potential sources of bisphenol A in the neonatal intensive care unit. Pediatrics 131(3):483–489, PMID: 23420909, 10.1542/peds.2012-1380.

Faulk C, Kim JH, Jones TR, McEachin RC, Nahar MS, Dolinoy DC, et al. 2015. Bisphenol A-associated alterations in genome-wide DNA methylation and gene expression patterns reveal sequence-dependent and non-monotonic effects in human fetal liver. Environ Epigenet 1(1):dvv006, PMID: 27358748, 10.1093/eep/dvv006.

Gerhardt PG, Mevåg B, Tveter KJ, Purvis K. 1983. A systematic study of biochemical differences between the lobes of the rat prostate. Int J Androl 6(6):553–562, PMID: 6668082, 10.1111/j.1365-2605.1983.tb00346.x.

Gerona RR, Woodruff TJ, Dickenson CA, Pan J, Schwartz JM, Sen S, et al. 2013. Bisphenol-A (BPA), BPA glucuronide, and BPA sulfate in midgestation umbilical cord serum in a northern and central California population. Environ Sci Technol 47(21):12477–12485, PMID: 23941471, 10.1021/es402764d.

Gore AC, Chappell VA, Fenton SE, Flaws JA, Nadal A, Prins GS, et al. 2015. EDC-2: the Endocrine Society’s Second Scientific Statement on Endocrine-Disrupting Chemicals. Endocr Rev 36:E1–E150, PMID: 26544531, 10.1210/er.2015-1010.

Hayashi N, Sugimura Y, Kawamura J, Donjacour AA, Cunha GR. 1991. Morphological and functional heterogeneity in the rat prostatic gland. Biol Reprod 45(2):308–321, PMID: 1786296.

Hines CJ, Jackson MV, Deddens JA, Clark JC, Ye X, Christianson AL, et al. 2017. Urinary bisphenol A (BPA) concentrations among workers in industries that manufacture and use BPA in the USA. Ann Work Expo Health 61(2):164–182, PMID: 28395354, 10.1093/annweh/wxw021.

Ho SM, Cheong A, Lam HM, Hu WY, Shi GB, Zhu X, et al. 2015. Exposure of human prostaspheres to bisphenol A epigenetically regulates SNORD family non-coding RNAs via histone modification. Endocrinology 156(11): 3984–3995, PMID: 26248216, 10.1210/en.2015-1067.

Ho SM, Tang WY, Belmonte de Frausto J, Prins GS. 2006. Developmental exposure to estradiol and bisphenol A increases susceptibility to prostate carcinogenesis and epigenetically regulates phosphodiesterase type 4 variant 4. Cancer Res 66(11):5624–5632, PMID: 16740699, 10.1158/0008-5472.CAN-06-0516.

Hu WY, Shi GB, Lam HM, Hu DP, Ho SM, Madueke IC, et al. 2011. Estrogen-initiated transformation of prostate epithelium derived from normal human prostate stem-progenitor cells. Endocrinology 152(6):2150–2163, PMID: 21427218, 10.1210/en.2010-1377.

Hu WY, Shi GB, Xie L, Birch L, Hu DP, Majumdar S, et al. 2015. Chronic low-dose bisphenol A (BPA) exposure alters rat prostate stem cell homeostasis. In: Proceedings of the Society of Toxicology 54th Annual Meeting, 22-26 March 2015. San Diego, CA, Vol. 144. Toxicologist: Late-Breaking Supplement to Toxicological Sciences 144(1). Abstract No. 2835.

Jia X, Li X, Xu Y, Zhang S, Mou W, Liu Y, et al. 2011. SOX2 promotes tumorigenesis and increases the anti-apoptotic property of human prostate cancer cell. J Mol Cell Biol 3(4):230–238, PMID: 21415100, 10.1093/jmcb/mjr002.

Kaklamani V, Yi N, Zhang K, Sadim M, Offit K, Oddoux C, et al. 2011. Polymorphisms of ADIPOQ and ADIPOR1 and prostate cancer risk. Metab Clin Exp 60(9):1234–1243, PMID: 21397927, 10.1016/j.metabol.2011.01.005.

Kim JH, Sartor MA, Rozek LS, Faulk C, Anderson OS, Jones TR, et al. 2014. Perinatal bisphenol A exposure promotes dose-dependent alterations of the mouse methylome. BMC genomics 15:30, PMID: 24433282, 10.1186/1471-2164-15-30.

Labrie C, Lessard J, Ben Aicha S, Savard MP, Pelletier M, Fournier A, et al. 2008. Androgen-regulated transcription factor AibZIP in prostate cancer. J Steroid Biochem Mol Biol 108(3–5):237–244, PMID: 17933519, 10.1016/j.jsbmb.2007.09.008.

Leav I, Ho S, Ofner P, Merk F, Kwan P, Damassa D. 1988. Biochemical alterations in sex hormone-induced hyperplasia and dysplasia of the dorsolateral prostates of Noble rats. J Natl Cancer Inst 80(13):1045–1053, PMID: 2457709.

Matsumoto J, Yokota H, Yuasa A. 2002. Developmental increases in rat hepatic microsomal UDP-glucuronyltransferase activities toward xenoestrogens and decreases during pregnancy. Environ Health Perspect 110(2):193–196, PMID: 11836149, 10.1289/ehp.02110193.

Milman HA, Bosland MC, Walden PD, Heinze JE. 2002. Evaluation of the adequacy of published studies of low-dose bisphenol A on rodent prostate for use in human risk assessment. Regul Toxicol Pharmacol 35(3):338–346, PMID: 12202049.

Montgomery RB, Mostaghel EA, Vessella R, Hess DL, Kalhorn TF, Higano CS, et al. 2008. Maintenance of intratumoral androgens in metastatic prostate cancer: a mechanism for castration-resistant tumor growth. Cancer Res 68(11):4447–4454, PMID: 18519708, 10.1158/0008-5472.CAN-08-0249.

Nelles JL, Hu WY, Prins GS. 2011. Estrogen action and prostate cancer. Expert Rev Endocrinol Metab 6(3):437–451, PMID: 21765856, 10.1586/eem.11.20.

Padmanabhan V, Siefert K, Ransom S, Johnson T, Pinkerton J, Anderson L, et al. 2008. Maternal bisphenol-A levels at delivery: a looming problem? J Perinatol 28(4):258–263, PMID: 18273031, 10.1038/sj.jp.7211913.

Price D. 1963. Comparative aspects of development and structure in the prostate. In: Biology of the Prostate and Related Tissues, Vol. 12. Vollmer EP, ed. Washington, DC:National Cancer Institute, 1–27.

Prins GS. 1987. Prolactin influence on cytosol and nuclear androgen receptors in the ventral, dorsal, and lateral lobes of the rat prostate. Endocrinology 120(4):1457–1464, PMID: 3493896, 10.1210/endo-120-4-1457.

Prins GS. 1989. Differential regulation of androgen receptors in the separate rat prostate lobes: androgen independent expression in the lateral lobe. J Steroid Biochem 33(3):319–326, PMID: 2779222.

Prins GS. 1992. Neonatal estrogen exposure induces lobe-specific alterations in adult rat prostate androgen receptor expression. Endocrinology 130(6):3703–3714, PMID: 1597166, 10.1210/endo.130.6.1597166.

Prins GS. 1992. Neonatal estrogen exposure induces lobe-specific alterations in adult rat prostate androgen receptor expression. Endocrinology 130(4):2401–2412, PMID: 1547747, 10.1210/endo.130.4.1547747.

Prins GS, Birch L, Habermann H, Chang WY, Tebeau C, Putz O, et al. 2001. Influence of neonatal estrogens on rat prostate development. Reprod Fertil Dev. 13(4):241–252, 10.1071/RD00107.

Prins GS, Calderon-Gierszal EL, Hu WY. 2015. Stem cells as hormone targets that lead to increased cancer susceptibility. Endocrinology 156(10):3451–3457, PMID: 26241068, 10.1210/en.2015-1357.

Prins GS, Ho SM. 2010. Early life estrogens and prostate cancer in an animal model. J Dev Orig Health Dis 1(6):365–370, PMID: 24795802, 10.1017/S2040174410000577.

Prins GS, Hu WY, Shi GB, Hu DP, Majumdar S, Li G, et al. 2014. Bisphenol A promotes human prostate stem-progenitor cell self-renewal and increases in vivo carcinogenesis in human prostate epithelium. Endocrinology 155(3):805–817, PMID: 24424067, 10.1210/en.2013-1955.

Prins GS, Ye SH, Birch L, Ho SM, Kannan K. 2011. Serum bisphenol A pharmacokinetics and prostate neoplastic responses following oral and subcutaneous exposures in neonatal Sprague-Dawley rats. Repro Toxicology 31(1):1–9, PMID: 20887781, 10.1016/j.reprotox.2010.09.009.

Pylkkänen L, Santti R, Newbold R, McLachlan J. 1991. Regional differences in the prostate of the neonatally estrogenized mouse. Prostate 18(2):117–129, PMID: 2006118.

Rochester JR. 2013. Bisphenol A and human health: a review of the literature. Reprod Toxicol 42:132–155, PMID: 23994667, 10.1016/j.reprotox.2013.08.008.

Seachrist DD, Bonk KW, Ho SM, Prins GS, Soto AM, Keri RA. 2016. A review of the carcinogenic potential of bisphenol A. Reprod Toxicol 59:167–182, PMID: 26493093, 10.1016/j.reprotox.2015.09.006.

Setlur SR, Mertz KD, Hoshida Y, Demichelis F, Lupien M, Perner S, et al. 2008. Estrogen-dependent signaling in a molecularly distinct subclass of aggressive prostate cancer. J Natl Cancer Inst 100(11):815–825, 10.1093/jnci/djn150.

Shappell S, Thomas GV, Roberts RL, Herbert R, Ittmann MM, Rubin MA, et al. 2004. Prostate pathology of genetically engineered mice: definitions and classification. The consensus report from the Bar Harbor meeting of the Mouse Models of Human Cancer Consortium Prostate Biology Committee. Cancer Res 64(6):2270–2305, PMID: 15026373, 10.1158/0008-5472.CAN-03-0946.

Sharma S, Kelly TK, Jones PA. 2010. Epigenetics in cancer. Carcinogenesis 31(1):27–36, PMID: 19752007, 10.1093/carcin/bgp220.

Siegel RL, Miller KD, Jemal A. 2016. Cancer statistics, 2016. CA: A Cancer Journal for Clinicians 66(1):7–30, 10.3322/caac.21332.

Takizawa I, Lawrence MG, Balanathan P, Rebello R, Pearson HB, Garg E, et al. 2015. Estrogen receptor alpha drives proliferation in PTEN-deficient prostate carcinoma by stimulating survival signaling, MYC expression and altering glucose sensitivity. Oncotarget 6(2):604–616, PMID: 25436982, 10.18632/oncotarget.2820.

Tang WY, Morey LM, Cheung YY, Birch L, Prins GS, Ho SM. 2012. Neonatal exposure to estradiol/bisphenol A alters promoter methylation and expression of Nsbp1 and Hpcal1 genes and transcriptional programs of Dnmt3a/b and Mbd2/4 in the rat prostate gland throughout life. Endocrinology 153(1):42–55, PMID: 22109888, 10.1210/en.2011-1308.

Tennstedt P, Bölch C, Strobel G, Minner S, Burkhardt L, Grob T, et al. 2014. Patterns of TPD52 overexpression in multiple human solid tumor types analyzed by quantitative PCR. Int J Oncol 44(2):609–615, PMID: 24317684, 10.3892/ijo.2013.2200.

Vandenberg LN, Colborn T, Hayes TB, Heindel JJ, Jacobs DR Jr, Lee DH, et al. 2012. Hormones and endocrine-disrupting chemicals: low-dose effects and nonmonotonic dose responses. Endocr Rev 33(3):378–455, PMID: 22419778, 10.1210/er.2011-1050.

Vandenberg LN, Gerona RR, Kannan K, Taylor JA, van Breemen RB, Dickenson CA, et al. 2014. A round robin approach to the analysis of bisphenol A (BPA) in human blood samples. Environ Health 13:25, 10.1186/1476-069X-13-25.

Vandenberg LN, Hauser R, Marcus M, Olea N, Welshons WV. 2007. Human exposure to bisphenol A (BPA). Repro Toxicology 24(2):139–177, 10.1016/j.reprotox.2007.07.010.

Vermeulen A, Kaufman JM, Goemaere S, van Pottelberg I. 2002. Estradiol in elderly men. Aging Male 5(2):98–102, PMID: 12198740.

Viñas R, Watson CS. 2013. Mixtures of xenoestrogens disrupt estradiol-induced non-genomic signaling and downstream functions in pituitary cells. Environ Health 12:26, 10.1186/1476-069X-12-26.

vom Saal FS, Timms BG, Montano MM, Palanza P, Thayer KA, Nagel SC, et al. 1997. Prostate enlargement in mice due to fetal exposure to low doses of estradiol or diethylstilbestrol and opposite effects at high doses. Proc Natl Acad Sci U S A 94(5):2056–2061, PMID: 9050904.

Wang Q, Trevino LS, Wong RL, Medvedovic M, Chen J, Ho SM, et al. 2016. Reprogramming of the epigenome by MLL1 links early-life environmental exposures to prostate cancer risk. Mol Endocrinol 30(8):856–871, 10.1210/me.2015-1310.

Wong RLY, Wang Q, Treviño LS, Bosland MC, Chen J, Medvedovic M, et al. 2015. Identification of secretaglobin Scgb2a1 as a target for developmental reprogramming by BPA in the rat prostate. Epigenetics 10(2):127–134, PMID: 25612011, 10.1080/15592294.2015.1009768.

Yoshino H, Ichihara T, Kawabe M, Imai N, Hagiwara A, Asamoto M, et al. 2002. Lack of significant alteration in the prostate or testis of F344 rat offspring after transplacental and lactational exposure to bisphenol A. J Toxicol Sci 27(5):433–439, PMID: 12533913.

Yuen M, Leung L, Wang J, Wong YC, Chan FL. 2005. Enhanced induction of prostatic dysplasia and carcinoma in the Noble rat model by combination of neonatal estrogen exposure and hormonal treatments in adulthood. Int J Oncology 27(6):1685–1695, PMID: 16273225.

Strategies to Improve Private-Well Water Quality: A North Carolina Perspective

Author Affiliations open

1Department of Environmental Sciences and Engineering, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, Chapel Hill, North Carolina, USA

2Department of Civil and Environmental Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virginia, USA

PDF icon PDF Version (759 KB)

  • Background:
    Evidence suggests that the 44.5 million U.S. residents drawing their drinking water from private wells face higher risks of waterborne contaminant exposure than those served by regulated community water supplies. Among U.S. states, North Carolina (N.C.) has the second-largest population relying on private wells, making it a useful microcosm to study challenges to maintaining private-well water quality.
    Objectives:
    This paper summarizes recommendations from a two-day summit to identify options to improve drinking-water quality for N.C. residents served by private wells.
    Methods:
    The Research Triangle Environmental Health Collaborative invited 111 participants with knowledge of private-well water challenges to attend the Summit. Participants worked in small groups that focused on specific aspects and reconvened in plenary sessions to formulate consensus recommendations.
    Discussion:
    Summit participants highlighted four main barriers to ensuring safe water for residents currently relying on private wells: (1) a database of private well locations is unavailable; (2) racial disparities have perpetuated reliance on private wells in some urbanized areas; (3) many private-well users lack information or resources to monitor and maintain their wells; and (4) private-well support programs are fragmented and lack sufficient resources. The Summit produced 10 consensus recommendations for ways to overcome these barriers.
    Conclusions:
    The Summit recommendations, if undertaken, could improve the health of North Carolinians facing elevated risks of exposure to waterborne contaminants because of their reliance on inadequately monitored and maintained private wells. Because many of the challenges in N.C. are common nationwide, these recommendations could serve as models for other states. https://doi.org/10.1289/EHP890
  • Received: 29 July 2016
    Revised: 22 December 2016
    Accepted: 15 March 2017
    Published: 07 July 2017

    Address correspondence to J. M. Gibson, Dept. of Environmental Sciences and Engineering, Gillings School of Global Public Health, 135 Dauer Dr., University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7431, USA. Telephone: (919) 969-1594. Email: jackie.macdonald@unc.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP890).

    J.M.G. is employed by the University of North Carolina at Chapel Hill, which partially sponsored N.C. Research Triangle Environmental Health Collaborative’s 2015 Summit, “Safe Water from Every Tap.” The other authors declare they have no actual or competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (172 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (90 KB)

Introduction

The introduction of municipal water-treatment systems was one of the greatest public health advances in the U.S. during the twentieth century. A 2005 study of historical public health data in 13 major U.S. cities attributed nearly half of overall mortality reduction, two-thirds of child mortality reduction, and three-quarters of infant mortality reduction between 1900 and 1936 to the installation of municipal water chlorination and filtration systems (Cutler and Miller 2005). The same study estimated that the public health benefits of these investments exceeded total construction, operation, and maintenance costs by a factor of 23 over that period.

Today, communities served by municipal water systems are protected by the Safe Drinking Water Act (SDWA) of 1974 and its subsequent amendments (Tiemann 2014). Under the SDWA, public water systems must ensure that the water they deliver meets health-based water-quality requirements, known as maximum contaminant levels (MCLs), for 91 contaminants. To ensure compliance, utilities must monitor water quality on a monthly or more frequent basis, depending on the size of the system. When monitoring detects contaminants at concentrations exceeding one or more MCLs, the water utility must notify its consumers and take corrective action, such as upgrading its treatment processes. The SDWA provides funds that states can use to assist public water systems with the costs of improvements, and the U.S. Environmental Protection Agency (EPA) offers technical assistance to plan such improvements.

Although most U.S. residents benefit from municipally treated water regulated by the SDWA, the 14% of the population (44.5 million people) who obtain their drinking water from private wells are excluded from these protections (Maupin et al. 2014). An analysis of waterborne disease outbreak data from 1971 to 2006 showed that although outbreaks associated with public water supplies decreased during that period, the number of outbreaks from private well contamination increased (Craun et al. 2010). Evidence suggests that although many private wells deliver water of high quality, a substantial fraction may be contaminated. Past state-level surveys report that 40–58% of private wells exceed at least one SDWA health-based standard, most commonly for bacterial contamination (Swistock et al. 2012; Knobeloch et al. 2013; Pieper et al. 2015). Concerns about children’s exposures led the American Academy of Pediatrics in 2009 to issue a policy statement recommending that pediatricians ask families if they obtain their water from private wells to determine whether water contamination could be a source of illnesses (such as gastrointestinal illness and lead poisoning) (American Academy of Pediatrics Committee on Environmental Health and Committee on Infectious Diseases, 2009). The policy statement encourages pediatricians to recommend that parents test and maintain their wells at least annually for coliform bacteria and nitrates along with lead if a child has an elevated blood lead level.

Among U.S. states, North Carolina (N.C.) has the second-highest total population (3.3 million) after Pennsylvania (Figure 1) and the third-highest population percentage (35%, after Maine and Alaska) of residents relying on private wells for drinking water (Maupin et al. 2014). Although this distinction in part reflects that N.C. has the second-largest rural population in the nation (3.2 million), 939,000 private well users (28.4% of the total) are in the six counties classified as urban by the N.C. Rural Economic Development Center due to their high population densities (above 750 people per square mile) (U.S. Geological Survey 2016; The Rural Center 2016). Recent research has revealed associations between N.C. private-well water quality and health risks. For example, our recent research demonstrated that 99% of N.C. emergency-department hospital visits for acute gastrointestinal illness associated with exposure to waterborne microbial contaminants are attributable to private-well contamination (DeFelice et al. 2016). Other recent research has linked elevated concentrations of metals in N.C. private wells to the risk of birth defects (Sanders et al. 2014). With such high incidence of reliance on private wells and with recent evidence of significant health risks for some private-well users, N.C. offers a useful microcosm in which to analyze challenges and solutions to ensuring that private wells deliver safe drinking water.

Horizontal bar graph plotting 50 US states (y-axis) across population obtaining drinking water from private wells in thousands (x-axis).
Figure 1. Among U.S. states, North Carolina has the second-largest number of people relying on private wells for their drinking water. Data source: Maupin et al. 2014.

Recognizing the need for new solutions to N.C.’s private-well water quality challenges and the potential broader national relevance of these challenges, the N.C. Research Triangle Environmental Health Collaborative (Collaborative) convened a summit, “Safe Water from Every Tap,” on 26–27 October 2015, to identify and discuss strategies to reduce health risks from private-well water contamination. This paper provides context to understand obstacles to maintaining private-well water quality in N.C. and summarizes major recommendations from the Summit. Although the Summit focused on N.C., many challenges to private-well stewardship are common nationwide. In the discussion, we highlight the broader relevance of the Summit recommendations to other states.

Methods

The Summit was organized by the Collaborative’s 14-member executive committee, which is chaired by former N.C. Public Health Director Leah Devlin and Department of Environmental Quality (DEQ) Secretary William Ross (Table S1). To represent a variety of perspectives, the committee invited representatives of county, state, and federal agencies, public health practitioners, academic scientists, nonprofit community-based organizations, elected officials and policy makers, industry and water utility representatives, and environmental and public health consultants with knowledge of private-well water quality challenges. In all, 111 participants attended (Figure S1; Table S2). Several participants (e.g., from the Southeast Rural Community Assistance Project and community-based organizations) represented constituencies of private-well owners. In addition, other attendees served a dual role representing their organization (such as the N.C. Division of Public Health) and as a current or former private-well owner.

To facilitate interactive discussions and cross-disciplinary exchanges, the Summit alternated between expert plenaries and small work-group breakout sessions. Participants registered for one of four work groups on the themes of (1) private-well owner education, (2) governance and policy for private wells, (3) groundwater pollution prevention, and (4) user-friendly technology for private-well monitoring and maintenance. Participants remained in the same group throughout multiple breakout sessions to facilitate the development of within-group recommendations.

The plenary on the Summit’s first day featured presentations on federal and state programs related to private-well water quality and a panel discussion including communities affected by private-well contamination. During the afternoon, work groups met separately to define private-well water quality challenges related to their group’s theme. All participants then reconvened to discuss and debate challenges highlighted by each group and to identify common themes. The second day followed a similar structure, with a morning plenary presenting innovative technologies for managing decentralized water supplies, followed by work-group breakout sessions to identify and prioritize solutions to the challenges identified on the first day. During the last plenary, participants integrated and prioritized common themes and recommendations from the four groups. The executive committee further refined recommendations through e-mail feedback after the Summit.

Results

Participants in the “Safe Water from Every Tap” Summit engaged in a consensus process to identify challenges to and recommendations to improve N.C. private-well water quality. Participants decided to focus their recommendations on four critical challenges:

  1. The N.C. population relying on private wells is poorly characterized. Consequently, the N.C. Division of Public Health and county health departments lack the information they need to target households for risk communication, technical assistance, or other interventions.
  2. Racial discrimination in the establishment of municipal boundaries excluded some peri-urban N.C. communities from public water service. As a result, these communities rely on private wells despite their proximity to municipal water lines and are at risk of exposure to well-water contaminants brought about by high population densities.
  3. Many N.C. private-well users lack the knowledge and/or resources needed to routinely monitor and maintain their well water. These well users are therefore at risk of exposure to contamination that could be detected and removed if the households were part of a well-managed and regulated community water system or if household treatment were installed and properly maintained by the user.
  4. Programs to protect private-well water quality and to support homeowners in managing their wells are fragmented across state and county agencies. These programs lack the resources to help private-well owners ensure that their drinking water meets recommended health-based standards, such as the standards that community water supplies are required to meet under the U.S. SWDA.

Participants proposed 10 high-priority recommendations to address these challenges. Table 1 summarizes the recommendations and proposes a time sequence for implementation. The following sections provide context to understand the need for these recommendations.

Table 1. Ten recommendations for protecting the health of households relying on private wells.
Recommendation Responsible organization(s) Timing
Challenge 1: Private Well Population Is Poorly Characterized
 1. The Division of Public Health (DPH) or Department of Environmental Quality (DEQ) should collect and coordinate all available state and county data relevant to characterizing private well locations. North Carolina (NC) DPH or DEQ Immediate
Challenge 2: Racial Disparities Have Perpetuated Reliance on Private Wells in Some Communities
 2.1. The General Assembly should authorize and fund a study to identify areas underserved by community water and sewer service that could be connected to existing municipal water lines. NC General Assembly After recommendation 1 (since information on well locations is required)
 2.2. The DPH, DEQ, or a private foundation should fund a preliminary state-wide analysis of the capital costs of extending municipal water service to underbounded neighborhoods, and areas in need of service extension should be prioritized. NC DPH, DEQ, and/or private foundation After recommendation 2.1 (since data on locations of underserved neighborhoods are required)
Challenge 3: Many Private Well Users Lack Knowledge and Resources to Routinely Test and Maintain Their Wells
 3.1. The DPH should develop targeted marketing campaigns to promote private well testing and maintenance. NC DPH Immediate
 3.2. An appropriate NC state agency or foundation should fund a study to analyze options for providing financial assistance to low-income private well users to afford the costs of well monitoring and maintenance. NC DPH, DEQ, and/or private foundation After recommendation 3.1 (since marketing campaign results can reveal needs for assistance)
 3.3. DPH, DEQ, or a private foundation should support a study of options for promoting the development of affordable private well contract maintenance services, in which private system users pay subscription fees for routine well maintenance and testing and for assistance in installing and maintaining household water treatment systems where necessary. NC DPH, DEQ, and/or private foundation After recommendations 1 and 3.2 (since well location data can assist in planning and marketing data can reveal needs)
 3.4. The General Assembly should allocate resources to DEQ to build an interactive mapping tool that well owners and health departments can use to identify wells at risk of contamination. NC General Assembly After recommendation 1(since well location data are required)
 3.5. The DPH should update and upgrade its existing web sites to assist homeowners in finding state-accredited water testing labs, selecting contaminants for monitoring, collecting samples, interpreting test results, and selecting water treatment technologies. NC DPH Immediate
 3.6. The DPH or DEQ should create a state-wide network of professionals that provides information and training on private well issues. NC DPH or DEQ Immediate
Challenge 4: Private Well Programs Are Fragmented and Insufficiently Resourced
 4. The NC General Assembly should commission a study of the adequacy of existing private well regulations and programs. NC General Assembly Immediate

Challenge 1: Private-Well Population is Poorly Characterized

Ideally, county health departments would send regular communications to private-well owners reminding them to test their wells each year and providing information about what to do if tests revealed water-quality problems. However, neither the individual counties nor the state has a complete database with addresses of private-well owners. As a result, delivering these messages or other interventions to help well owners poses a major challenge.

The most comprehensive nationwide private-well population inventory (the basis for Figure 1) was compiled by the U.S. Geological Survey (USGS) (Maupin et al. 2014). Although the USGS data are useful in identifying counties with high incidence of reliance on private wells, these data lack geographic locations of private wells. The USGS data were developed by obtaining data regarding the size of the population served by each public water-supply system in each county and then subtracting the total from the county population.

Several other data sources could help to identify private-well locations. Potential sources include utility service area maps, water pipeline maps, N.C. private-well permitting and testing databases, and the U.S. Census. However, our investigations have shown that none of these sources provides complete and accurate spatial coverage (Leker 2015). GIS files containing utility service-area boundaries are often overly inclusive, for example, showing the entire county as being included in the water utility’s service area even in counties where a large population relies on private wells. Detailed water-utility maps that show water-main locations have been difficult to obtain due to post-9/11 security concerns. To our knowledge, the most recent map of water-main locations, compiled by the N.C. Rural Economic Development Center, includes only 75 of the state’s 100 counties and has not been updated since 1997 (Leker 2015). Even when utility pipeline maps are publicly available, they indicate water-main locations but not addresses of households served. We are aware of neighborhoods that are bisected by water-service lines but that contain households unconnected to those lines (Heaney et al. 2013; MacDonald Gibson et al. 2014).

Statewide well permitting and testing databases are another potential source of private-well-owner addresses. Since 1967, N.C. has required permits for all new drinking-water wells. However, rather than being issued and tracked by the state, permits are issued and filed by counties in paper copies. Some counties (for example, Chatham http://www.chathamnc.org/index.aspx?page=1887) have digitally scanned some of its permits and are beginning to construct databases, but a complete, statewide inventory is unavailable. Furthermore, wells constructed before 1967 are exempt. Since 1 January 2009, N.C. has required that all new wells must be tested for selected contaminants; county health departments collect water samples and send them to the N.C. State Laboratory of Public Health. The locations of tested wells could be used to map locations of households that rely on private wells, but only for wells constructed since 2009. In addition, because test results are typically submitted as paper copies, adequate resources would need to be provided to the N.C. State Laboratory of Public Health to convert all of the paper forms to a searchable database.

A final potential information source on N.C. private well locations is the U.S. Census. Through 1990, the Census collected household-level data on drinking-water sources. However, this question was classified as nonmandatory after 1990 as part of a process of streamlining the Census questionnaire (U.S. Census Bureau 2009). The streamlined 2000 Census, unlike its 1990 predecessor, excluded all questions that were not required by federal law. Local, county, or state governments can request that the U.S. Census Bureau conduct a special census that could include questions about water and sewer access (U.S. Census Bureau 2015). However, the requesting governments must pay for this service, and the U.S. Office of Management and Budget must approve the questions.

In summary, the Summit participants identified lack of address-level and demographic data on N.C. households relying on private wells as a critical barrier to developing outreach to residents and other programs to ensure adequate water quality. To address this barrier, participants recommended the following:

Summit recommendation 1.

The N.C. Division of Public Health (DPH) or DEQ should collect and coordinate all available state and county data relevant to characterizing the locations of private wells. Data sources may include well-construction and well-abandonment forms, N.C. State Laboratory of Public Health well-sampling records, and municipal water-pipe maps. Gaps in the inventory should be identified, and a program initiated (possibly including door-to-door efforts in targeted communities) to fill the gaps. Although this effort should be coordinated with local health departments, these departments are insufficiently resourced to bear the burden of data collection; therefore, state funding and coordination are essential to success. If completed, such a database could serve as a model for other states that also lack comprehensive, centralized data on private-well locations and characteristics.

Challenge 2: Racial Disparities Have Perpetuated Reliance on Private Wells in Some Communities

Some peri-urban neighborhoods in N.C. are excluded from nearby water service despite their proximity to municipal water lines. Due to high population and septic-system densities, wells in such neighborhoods may be at increased risk of contamination (Borchardt et al. 2003; Stillo and MacDonald Gibson 2017).

Several N.C. case studies have documented African American communities on the borders of or surrounded by towns and cities that are excluded from nearby municipal services. For example, a 2004 case study documented that the Mebane City Council over its history had systematically drawn discontiguous municipal boundaries in order to exclude four black communities (Johnson et al. 2004). Although one of the communities neighbored the municipal sewage-treatment plant, neither this community nor the other three had access to municipal water or sewer service. Exclusionary zoning practices continued in Mebane through at least the 1990s as the town annexed satellite parcels slated for high-income residential development and continued to exclude the historically black communities. In 2005, the New York Times reported on a similar predicament affecting African American communities around Pinehurst, site of the U.S. Open Golf Tournament that year (Dewan 2005). In 2013, Heaney et al. documented exclusion from municipal services in the Rogers Road community neighboring Chapel Hill and Carrboro (Heaney et al. 2013). In 2014, researchers found statistical evidence of racial exclusion from water service in extraterritorial jurisdiction (ETJ) areas of Wake County (MacDonald Gibson et al. 2014). These ETJ areas border or are surrounded by municipalities, and the municipalities are allowed to control zoning decisions there, but they are not required to provide municipal services (although they may elect to do so). We found that in Wake County ETJ census blocks, every 10% increase in the black population proportion increased the odds of exclusion from municipal water service by 3.8% (MacDonald Gibson et al. 2014). In addition, we found a high prevalence of bacterial contaminants (29% of samples tested positive for total coliform bacteria and 6.4% were positive for Escherichia coli) in household drinking water in these excluded communities, and showed an increased risk of visiting an emergency department for acute gastrointestinal illness (Stillo and MacDonald Gibson 2017).

Summit participants agreed that improving drinking-water quality in peri-urban households still relying on private wells should be a high priority due to their proximity to regulated municipal water supplies and the increased risks to water quality from the relatively high population densities. Our prior research interviewing 25 key informants and 18 private-well owners in ETJ areas in four N.C. counties (Wake, New Hanover, Hoke, and Transylvania) explored the major barriers to connecting to municipal services (Naman and MacDonald Gibson 2015; Fizer 2016). Cost was the most prominent concern for officials, who would need to authorize service extensions, and for well owners. For example, a town mayor told us, “We’ve got a section in town here that does want to be annexed. The [city] will not do it. We did a study on it… . The payback was like 115 years” (Naman and MacDonald Gibson 2015). Homeowners doubted their ability to pay for connections to municipal water systems and to afford monthly water bills. For example, one private-well owner said, “But I know I am going to have to [deal with well water the rest of my life] because I cannot afford to have the city tapped in … . It is like five houses on this street that we all have well water and we would like to have city water” (Fizer 2016). Some homeowners also communicated that they preferred their well water because of its flavor and odor, even in households where we detected bacterial contaminants (Fizer 2016; Stillo and MacDonald Gibson 2017). In other communities, homeowners may desire access to municipal supplies but fear advocating for such services due to fear that public health officials could condemn their land if their septic system is failing (Naman and MacDonald Gibson 2015). Some communities (such as in Mebane) have advocated for municipal service extensions for decades, but their requests have been continually denied or only partially fulfilled (Johnson et al. 2004; Wilson et al. 2008).

As a start toward improving the safety of drinking water in peri-urban areas historically excluded from nearby municipal water service, Summit participants recommended that the state undertake two studies:

Summit recommendation 2.1.

The N.C. General Assembly should authorize and fund a study to identify ETJ areas underserved by community water and sewer service that could be connected to existing municipal water lines. The study should evaluate the benefits and drawbacks of extending community water services and should consider other potential mechanisms (for example, designating a responsible environmental or public health management entity) to ensure drinking water quality.

Summit recommendation 2.2.

A preliminary statewide analysis of the capital costs of water-service extensions to underserved ETJ communities should be completed, and areas in need of service extension should be prioritized. Existing and potential innovative options for financing capital costs of service extensions should be identified. This analysis should also examine the feasibility of establishing third-party options to administer the funds and mechanisms to help low-income communities afford monthly water and sewer bills. In addition, this effort should evaluate the legislative changes to annexation necessary for municipalities to extend services and to assess secondary impacts of infrastructure extension (e.g., changes in impervious surfaces, economic development, and public health).

Challenge 3: Many Private Well Users Lack Knowledge and Resources to Test and Maintain Their Wells

Statewide private-well testing data indicate that few N.C. well owners monitor their water quality on a routine basis. Although N.C. has required testing of new wells since January 1, 2009, the number of wells tested illustrate this program’s limited reach. Over the five-year period 1 January 2009–31 December 2013, the N.C. State Center for Health Statistics reported that 16,138 well-water samples statewide had been tested. Assuming each sample represents a distinct well, only 1.2% of all 1.3 million self-supplied domestic water wells were tested. Our interviews with 18 private-well owners in Wake County found that one (5.5%) tested at the recommended annual frequency, another two (11%) tested every 2–3 y, two (11%) tested every 4–5 y, and the rest (72%) tested less than every 5 y or not at all (Fizer 2016). Among this group, only eight (44%) reported ever taking any action to maintain their wells. All 18 interviewees thought they could detect contaminants through taste, odor, and appearance. All also mentioned costs as a barrier to testing. In Wake County, bacteriological analysis costs $25 per sample, $20–105 per sample with additional sample collection fees of $50 for some contaminants.

Even if well owners do monitor their water quality, the costs of home-treatment systems may pose a barrier to taking action when contaminants are detected. Purchase and installation of a whole-house water filter to remove contaminants typically costs hundreds to thousands of dollars, depending on the type of unit. Point-of-use treatment devices, which treat water for only a single tap, are less costly (with purchase costs as low as $25), but they do not provide complete coverage of all faucets. Some Wake County well owners we interviewed mentioned cost as a barrier to the purchase of a treatment system (Fizer 2016). One homeowner reported that her water had been tested and contaminants found, but that she could not afford a treatment unit, so she stopped drinking the water and only uses it for nonpotable purposes.

To promote monitoring and maintenance of private wells, Summit participants recommended the following new initiatives:

Summit recommendation 3.1.

The DPH should develop marketing campaigns to promote private well testing and maintenance. Social marketing campaigns could target new parents (for example, through prenatal classes and medical practices); child care centers; K–12 schools; recipients of Women, Infants, and Children program benefits; health care providers; mobile health clinics; Medicare recipients; and faith-based groups and homes in areas known to be at risk of contamination (especially those with wells pre-dating mandatory new-well testing). The campaigns could include distribution of drinking-water test kits with instructions about where to send the kits for analysis and links to a website to help homeowners interpret test results (see recommendation 3.5). The campaigns also could include information about contamination prevention, such as proper septic-system maintenance.

Summit recommendation 3.2.

A state agency or foundation should fund a study to analyze options for providing financial assistance to low-income private well users to afford the costs of private-well monitoring and maintenance. Existing programs, supported by federal and state governments and private organizations, that help low-income households pay their energy bills could serve as models for similar programs for private-well owners. For example, the N.C. Department of Health and Human Services operates a federally funded Low-Income Energy Assistance Program to provide one-time payments for households unable to pay their winter heating or summer cooling bills (North Carolina Department of Health and Human Services 2016); many counties and energy utilities offer similar programs (Duke Energy, 2017).

Summit recommendation 3.3.

DPH, DEQ, or a private foundation should support a study of options for promoting the development of affordable private-well contract maintenance services, in which private-system users pay subscription fees for routine well maintenance and testing and for assistance in installing and maintaining household water-treatment systems where contamination is identified. These services could also include septic-system maintenance in areas where septic systems threaten private-well water quality.

Summit recommendation 3.4.

The N.C. General Assembly should allocate resources to the DEQ to build an interactive mapping tool for use by well owners and county health departments in identifying wells at risk of contamination. Such a tool could be modeled after the U.S. EPA’s Drinking Water Mapping Application for Protecting Source Waters. Other state agencies should be required to contribute relevant data. Adequate resources should be provided to enable the DEQ to fill gaps in data necessary to delineate areas where private wells are at highest risk of contamination and to monitor contaminant trends.

Summit recommendation 3.5.

The DPH should update and upgrade its existing websites to assist homeowners in finding state-accredited water testing labs, selecting contaminants for monitoring, collecting samples, interpreting test results, and selecting water-treatment technologies. The website should be linked to the interactive mapping tool to be developed by the DEQ. It should include a comprehensive catalog of currently available, easily used point-of-use and whole-house treatment technologies and an interactive decision tool to help homeowners select an appropriate technology. The website could be modeled on New Hampshire’s Be Well Informed and Pennsylvania’s Drinking Water Interpretation Tool. One feature of these sites is that they enable users to enter their measured water-quality parameters. The sites then generate a customized interpretation, including information about whether the parameters were above or below the SDWA MCLs and links to more information. The New Hampshire tool also provides information about possible treatment options and discusses the adverse health effects of each parameter.

Summit recommendation 3.6.

The DPH or DEQ should create a statewide network of professionals who provide information and training on private-well issues. This network could host workshops and presentations on such issues. Professionals could be encouraged to participate by offering continuing-education credits. Use of a webinar format would facilitate communication among agencies and universities and minimize travel and financial burdens. This network could be linked to the U.S. Centers for Disease Control (CDC) and Prevention’s Private Well Community of Practice, which hosts a bimonthly webinar highlighting cutting-edge research addressing water and health problems in private wells throughout the United States.

Challenge 4: Private-Well Programs are Fragmented and Insufficiently Resourced

As previously mentioned, the SDWA does not regulate private wells. The N.C. legislature has enacted several laws governing various aspects of well construction, along with programs to protect groundwater sources tapped by private wells (Table 2). In addition, county health departments provide well testing services upon request, and some offer reduced-cost testing for low-income households. However, as Table 2 illustrates, administration of these programs is fragmented across agencies. No single organization is in charge of helping private-well owners ensure the safety of their drinking water.

Table 2. Existing private well protection programs in North Carolina (NC).
Program Description Implementing agencies
Permitting, inspection, and testing of new wells Since July 1, 2008, every new private drinking water well must be permitted, inspected, and tested by the local health department. Testing includes analysis for arsenic, barium, cadmium, chromium, copper, fluoride, lead, iron, magnesium, manganese, mercury, nitrates, nitrites, selenium, silver, sodium, zinc, pH, and bacterial indicators. Follow-up testing after construction is not required. Local health departments, with oversight of Division of Public Health (DPH)
Well construction standards Every well must be constructed to meet statewide minimum standards for location, casing, grouting, and screening. Some counties have enacted more stringent standards. Local health departments in conjunction with DPH
Well contractor certification Any person engaged in well construction, installation, repair, or abandonment must be certified by the NC Well Contractor Certification Commission. Certification is based on a written exam, work experience, and field observation. NC Well Contractor Certification Commission (staffed by DPH)
Voluntary well testing Local health departments offer low-cost well testing upon request. Sampling by local health departments. Analysis by the State Laboratory for Public Health, certified private lab, or local health department.
Health risk evaluations DPH provides recommendations for well water use based on results of the mandatory sampling of new wells or voluntary sampling. DPH and local health departments
Groundwater classifications and quality standards NC law has established drinking water as the best intended use for groundwater, and the NC Department of Environmental Quality (DEQ) has developed standards to protect the resource for that use. Violations trigger corrective action, with restoration to potable standards as the goal (though alternative standards are possible). DEQ
Bernard Allen Emergency Drinking Water Fund This fund pays for notification of well owners, water sampling, and alternative water sources near known contamination for qualifying individuals when no responsible party or other fund is available. DEQ-Division of Waste Management

Note: Table developed by E. Kane, hydrogeologist, Wake County Environmental Services Department.

To improve coordination across state and county programs and evaluate the adequacy of existing programs, Summit participants recommended the following:

Summit recommendation 4.

The N.C. General Assembly should commission a study of the adequacy of existing private-well regulations and programs. The study should evaluate well-construction standards, consider requiring operating permits for private wells on rental properties, assess the value and feasibility of requiring well testing at the time of property resale, and evaluate the potential for support programs for low-income well owners. It should also evaluate the adequacy of DPH and DEQ staffing to track contamination and maintenance issues and provide technical assistance to well owners and local health departments.

Discussion

The 2016 NC Environmental Health Collaborative Summit highlighted four principal barriers to ensuring that N.C. residents who rely on private wells receive drinking water of sufficient quality to protect their health. These barriers are (1) lack of a comprehensive database of private-well locations, (2) exclusion of some peri-urban minority communities from municipal services, (3) lack of well owner compliance with recommendations for monitoring and maintaining their wells, and (4) a fragmented system of regulations and programs for supporting private-well owners.

Other states face similar challenges. Elsewhere, as in N.C., databases with locations of private-well owners are incomplete. For example, in Oregon, private-well testing data are collected on paper forms that are not routinely digitized or collected by a central state agency (Hoppe et al. 2011). Texas began digitizing private-well construction reports in 2003, but records of wells built before then are not included in the state’s database, although scanned, portable document files of prior hand-written well reports are available (Texas Water Development Board 2016). The U.S. CDC has recognized the lack of data on private-well locations and has established a Private Well Initiative, the first goal of which is to answer the question, “Where are the unregulated drinking water systems in the U.S.?” (Backer and Tosta 2011). However, to date, a national database of well locations does not exist.

Racial disparities in access to nearby community water services also have been documented elsewhere. Aiken was the first to describe the use of selective annexation to exclude black communities from municipal boundaries and the services offered to municipal residents (Aiken 1987). In a 1987 study in the Mississippi Yazoo Delta, he found that towns incorporated white neighborhoods at the urban fringe and excluded similar African-American neighborhoods to dilute the voting strength of African-American citizens. He referred to such practices as “municipal underbounding.” In a 2007 analysis of U.S. Census data from 1,992 towns and communities in eight southern states, Lichter found that towns with high white population percentages were significantly less likely to annex and provide water and other municipal services to African-American municipal fringe areas than they were to annex predominantly white fringe neighborhoods (Lichter et al. 2007). Beyond the southern plantation crescent, recent research has documented racial underbounding in the Texas Lower Rio Grande Valley (Durst 2014) and California’s Central Valley (Ranganathan and Balazs 2015). In their article on the Central Valley, Ranganathan and Balazs referred to underbounded communities as “a piece of the Third World in the First World.”

Lack of routine monitoring of water quality in private wells is a nationwide problem as well. For example, a 1998 survey of 244 upstate New York well owners found that 47% had never tested their water (Schwartz et al. 1998). A 2009 Wisconsin survey of 2,600 well owners found that although 67% reported having ever tested their water, only 24% had done so in the past year (Knobeloch 2009). A 2012 study of 622 Pennsylvania well owners found that 30% had never tested their water, and 44% had tested it just once, usually only for coliform bacteria (Swistock et al. 2012). All of these studies found that well owners were largely unaware of testing recommendations (Schwartz et al. 1998; Swistock et al. 2012). For example, in Wisconsin, 45% of 2,600 survey respondents said they did not know what to test for, and 42% said they did not know where to send samples for analysis (Knobeloch 2009). Just as in N.C., the Wisconsin study revealed the misperception that drinking water contaminants can be detected through sensory perception: 82% of those who had not tested their water said their reason for not testing was that the water “tastes and looks fine.” Financial barriers are an impediment to testing elsewhere, too. The New York study found that survey respondents living in low-income/low-education counties reported lower testing prevalence (41%) than respondents living in high-income/high-education counties (64%) (Schwartz et al. 1998). Similarly, the Wisconsin study reported that 33% of families earning less than $20,000 per year had ever tested their well water, and 71% of families earning more than $75,000 per year had done so (Knobeloch 2009). A New Hampshire study found that among those who reported never having tested their well water for arsenic (widespread in New Hampshire’s geologic formations), 25% said cost was the major barrier to testing (Borsuk et al. 2014).

Fragmentation of programs to support private-well owners also occurs nationwide. Some states, such as Texas (Texas Commission on Environmental Quality 2014) and Pennsylvania (Swistock et al. 2012), do not regulate private wells at all. Among states with regulations, programs and responsible agencies vary substantially, with the latter ranging from state or local health departments to water resources agencies, natural resources departments, environmental protection agencies, and various land grant university extension services (Rogan and Brady 2009). A few states have stringent testing requirements. For example, New Jersey, Rhode Island, and Oregon require well-water testing at time of resale of residences (Fox et al. 2016), although the testing requirements and enforcement of the laws vary substantially in these three state programs. Like N.C., many states have well-construction standards and contractor-certification or licensing requirements, but these requirements can vary substantially by state. For example, in California, local water agencies or water districts determine required minimum depths for grouting between the upper portions of the borehole and the well casing, whereas other states establish mandatory state-wide minimum depths (California State Water Resources Control Board 2015). Some states require the use of pitless adapters (which allow the pipe delivering water from the well to the home to be buried below the frost line), but other states do not.

One consequence of the lack of federal regulations and the fragmentation across state programs is that private-well owners lack access to the financial support available to regulated public water systems. The SDWA established a low-cost loan program for water utility infrastructure improvements that provided $13.1 billion in federal investments through 2009 (Copeland and Tiemann 2010). Prior to the SDWA, communities with municipal systems also were eligible for grants and loans under the 1972 Clean Water Act and its predecessors through programs that Copeland and Tiemann called “the largest nonmilitary public works programs since the Interstate Highway System” (Copeland and Tiemann 2010). During the Reagan Administration, the amount of federal funding for water and sewer infrastructure dropped substantially and shifted from grants to loans (Copeland and Tiemann 2010). Notably, much of the period of active federal investment in clean water infrastructure coincided with the period of systematic exclusion of southern African-American communities from municipal services, denying these communities access to federal benefits afforded to similar white communities.

Although programs to support private-well users, study the demographics of well users, and record water quality characteristics vary nationwide, prior research indicates that the challenges to private-well stewardship identified for N.C. are common nationwide. As a result, the Summit recommendations may be useful in other states that are grappling with challenges to supporting private-well owners in maintaining safe drinking water. Indeed, a report from a recent CDC workshop on private wells emphasized the need for “building an infrastructure for stewardship” of private wells nationwide (Fox et al. 2016).

One challenge not discussed by Summit participants is the effects of drought on private wells. N.C. has a relative abundance of water, with average rainfalls ranging from around 45–60 in per year, depending on region (mountain, piedmont, or coastal plain) (National Oceanic and Atmospheric Administration 2016). In contrast, rainfall in arid regions is a fraction of this amount, leaving private-well users at risk of running out of water. For example, case studies of Latino communities around Fresno, California, where average annual rainfall is 11.5 in, have reported instances of private wells running dry, forcing residents to purchase water from local grocery stores (Pannu 2012). In our discussions with N.C. private-well owners, we have encountered households that were forced to haul water from grocery stores or gas stations because of well components freezing during winter. However, water quantity did not emerge as a major concern at the Summit, likely due to the general abundance of water in N.C. on average.

Conclusions

Because N.C. is an economically and racially diverse state, with land uses ranging from highly developed urban and industrial areas to rural crop and livestock production, the challenges to maintaining private-well water quality in this state are likely to reflect those in other states as well. The Collaborative Summit recommendations presented in this article should apply not only in N.C., but also in other states struggling with similar issues related to protecting the health of populations that draw their water from private wells. Indeed, N.C. could serve as a test bed for innovative private-well protection programs, such as contract maintenance, financial support, data system improvement, and outreach initiatives that could ensure private-well water quality and better health for all residents.

Acknowledgments

The Summit described in this article was convened by the N.C. Research Triangle Environmental Health Collaborative, which connects organizations and institutions; links research and policy; and joins government, academia, industry, and public interest groups to mutually consider, discuss, and debate the future of environmental health on a regional, national, and international level. Funding for the Summit was provided by the U.S. Environmental Protection Agency, National Institute of Environmental Health Sciences, RTI International, Duke University Nicholas School of the Environment, North Carolina State University, the University of North Carolina at Chapel Hill Gillings School of Global Public Health, MDB Inc., Social & Scientific Systems Inc., Visionpoint Marketing, the N.C. Biotechnology Center, ILS, and the N.C. Department of Environmental Quality.

References

Aiken CS. 1987. Race as a factor in municipal underbounding. Ann Assoc Am Geogr 77(4):564–579, 10.1111/j.1467-8306.1987.tb00181.x.

American Academy of Pediatrics. Committee on Environmental Health and Committee on Infectious Diseases, 2009. Drinking water from private wells and risks to children. Pediatrics 123(6):1599–1605, PMID: 19482772, 10.1542/peds.2009-0751.

Backer LC, Tosta N. 2011. Unregulated drinking water initiative for environmental surveillance and public health. J Environ Health 73(7):31–32, PMID: 21413560, http://www.ncbi.nlm.nih.gov/pubmed/21413560.

Borchardt MA, Chyou PH, DeVries EO, Belongia EA. 2003. Septic system density and infectious diarrhea in a defined population of children. Environ Health Perspect 111(5):742–748, PMID: 12727604, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1241485/.

Borsuk M, Rardin L, Paul M, Hampton T. 2014. Arsenic in private wells in NH. Hanover, NH:Thayer School of Engineering at Dartmouth. http://www.dartmouth.edu/~toxmetal/assets/pdf/Wellreport.pdf [accessed 21 June 2017].

California State Water Resources Control Board. 2015. “A Guide for Private Domestic Well Owners,” http://www.waterboards.ca.gov/gama/docs/wellowner_guide.pdf.

Copeland C, Tiemann M. 2010. “Water Infrastructure Needs and Investment: Review and Analysis of Key Issues.” Congressional Research Service Report RL31116. Washington, DC. https://www.fas.org/sgp/crs/homesec/RL31116.pdf.

Craun GF, Brunkard JM, Yoder JS, Roberts VA, Carpenter J, Wade T, et al. 2010. Causes of outbreaks associated with drinking water in the United States from 1971 to 2006. Clin Microbiol Rev 23(3):507–528, PMID: 20610821, 10.1128/CMR.00077-09.

Cutler D, Miller G. 2005. The role of public health improvements in health advances: The twentieth-century United States. Demography 42(1):1–22, PMID: 15782893, 10.1353/dem.2005.0002.

DeFelice NB, Johnston JE, Gibson JM. 2016. Reducing emergency department visits for acute gastrointestinal illnesses in North Carolina (USA) by extending community water service. Environ Health Perspect 24(10):1583–1591, PMID: 27203131, 10.1289/EHP160.

Dewan S. 2005. In county made rich by golf, some enclaves are left behind. The New York Times 7 June. http://www.nytimes.com/2005/06/07/national/07pinehurst.html?pagewanted=all&_r=0#.

Duke Energy. 2017. “Customer Assistance Programs.” https://www.duke-energy.com/community/customer-assistance-programs [accessed 21 June 2017].

Durst NJ. 2014. Municipal annexation and the selective underbounding of colonies in Texas’ Lower Rio Grande Valley. Environ Plann A 46(7):1699–1715, 10.1068/a130038p.

Fizer, Chelsea. 2016. “Barriers to Private Well and Septic Management: An Analysis of Homeowner Decision-Making.” [Master Thesis] Chapel Hill, NC:University of North Carolina.

Fox MA, Nachman KE, Anderson B, Lam J, Resnick B. 2016. Meeting the public health challenge of protecting private wells: Proceedings and recommendations from an expert panel workshop. Sci Total Environ 554–555:113–118, PMID: 26950625, 10.1016/j.scitotenv.2016.02.128.

Heaney CD, Steve W, Wilson SM, Campbell RL, O’Shea S. 2013. Public infrastructure disparities and the microbiological and chemical safety of drinking and surface water supplies in a community bordering a landfill. J Environ Health 75(10):24–36, PMID: 23858663, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4514614/.

Hoppe BO, Harding AK, Staab J, Counter M. 2011. Private well testing in Oregon from real estate transactions: An innovative approach toward a state-based surveillance system. Public Health Rep 126(1):107–115, PMID: 21337936, 10.1177/003335491112600115.

Johnson JH, Parnell A, Joyner AM, Christman CJ, Marsh B. 2004. Racial apartheid in a small North Carolina town. Rev Black Polit Econ 31(4):89–107, 10.1007/s12114-004-1012-4.

Knobeloch, L. 2009. “2008-9 Survey of Private Well Owners.” Wisconsin Department of Health Services.

Knobeloch L, Patrick G, Megan C, Henry A. 2013. Private drinking water quality in rural Wisconsin. J Environ Health 75(7):16–20, PMID: 23505770, http://www.ncbi.nlm.nih.gov/pubmed/23505770.

Leker, H. 2015. “Relationship Between Race and Community Water and Sewer Service in North Carolina.” Master’s Technical Report. Chapel Hill, NC:University of North Carolina. https://cdr.lib.unc.edu/record/uuid:1dac3da8-752b-416e-80b7-0044de2db0ef.

Lichter DT, Parisi D, Grice SM, Taquino M. 2007. Municipal underbounding: annexation and racial exclusion in small southern towns. Rural Sociol 72(1):47–68, 10.1526/003601107781147437.

MacDonald Gibson J, DeFelice N, Sebastian D, Leker H. 2014. Racial disparities in access to community water supply service in Wake County, North Carolina. Frontiers in Public Health Services and Systems Research 3(3):6, http://uknowledge.uky.edu/frontiersinphssr/vol3/iss3/6/.

Maupin, MA, Kenny JF, Hutson SS, Lovelace JK, Barber NL, Linsey KS. 2014. “Estimated Use of Water in the United States in 2010.” U.S. Geological Survey Circular, no. 1405. US Geological Survey, Washington, DC. http://www.mendeley.com/catalog/estimated-water-united-states-2010-8.

Naman JM, MacDonald Gibson J. 2015. Disparities in water and sewer services in North Carolina: An analysis of the decision-making process. Am J Public Health 105(12):320–326, PMID: 26270307, 10.2105/AJPH.2015.302731.

National Oceanic and Atmospheric Administration. 2016. “National Centers for Environmental Information,” https://www.ncdc.noaa.gov/ [accessed 21 June 2017].

North Carolina Department of Health and Human Services. 2016. “Low Income Energy Assistance.” http://www.ncdhhs.gov/assistance/low-income-services/low-income-energy-assistance [accessed 14 December 2016].

Pannu C. 2012. Drinking water and exclusion: A case study from California’s Central Valley. California Law Rev 100(1):223–268, http://www.californialawreview.org/5drinking-water-and-exclusion-a-case-study-from-californias-central-valley/.

Pieper KJ, Krometis LA, Gallagher DL, Benham BL, Edwards M, et al. 2015. Incidence of waterborne lead in private drinking water systems in Virginia. J Water Health 13(3):897–908, PMID: 26322775, 10.2166/wh.2015.275.

Ranganathan M, Balazs C. 2015. Water marginalization at the urban fringe: Environmental justice and urban political ecology across the North–South divide. Urban Geography 36(3):403–423, 10.1080/02723638.2015.1005414.

Rogan WJ, Brady MT. 2009. Drinking water from private wells and risks to children. Pediatrics 123(6):e1123–e1137, PMID: 19482745, 10.1542/peds.2009-0752.

Rural Center. 2016. “Rural County Map: Rural Counites in North Carolina.” http://www.ncruralcenter.org/rural-county-ma.

Sanders AP, Desrosiers TA, Warren JL, Herring AH, Enright D, Olshan AF, et al 2014. Association between arsenic, cadmium, manganese, and lead levels in private wells and birth defects prevalence in North Carolina: A semi-ecologic study. BMC Public Health 14(1):955, PMID: 25224535, 10.1186/1471-2458-14-955.

Schwartz JJ, Waterman AB, Lemley AT, Wagenet LR, Landre P, Allee DJ. 1998. Homeowner perceptions and management of private water supplies and wastewater treatment systems. J Soil Water Conserv 53(4):315–319, http://www.jswconline.org/content/53/4/315.abstract.

Stillo F, MacDonald Gibson J. 2017. Exposure to contaminated drinking water and health disparities in North Carolina. Am J Public Health 107(1):180–185, PMID: 27854523, 10.2105/AJPH.2016.303482.

Swistock BR, Clemens S, Sharpe WE, Rummel S. 2012. Water quality and management of private drinking water wells in Pennsylvania. J Environ Health 75(6):60–66, PMID: 23397651.

Texas Commission on Environmental Quality. 2014. “Groundwater Regulation for Private Well Owners.” https://www.tceq.texas.gov/response/drought/groundwater_regulation.html.

Texas Water Development Board. 2016. “Submitted Drillers Reports Database.” http://www.twdb.texas.gov/groundwater/data/drillersdb.asp.

Tiemann, M. 2017. “Safe Drinking Water Act (SDWA): A Summary of the Act and Its Major Requirements.” CRS Report No. 7-5700. Washington, DC:Congressional Research Service, https://fas.org/sgp/crs/misc/RL31243.pdf [accessed 21 June 2017].

U.S. Census Bureau. 2009. “History: 2000 Census of Population and Housing.” Washington DC. http://www.census.gov/history/pdf/Census2000v1.pdf.

U.S. Census Bureau. 2015. “Special Census Program: Frequently Asked Questions. http://www.census.gov/programs-surveys/specialcensus/faqs.html#addquestions.

U.S. Geological Survey. 2016. ”Water Use Data for North Carolina.“ https://waterdata.usgs.gov/nc/nwis/wu.

Wilson SM, Heaney CD, Cooper J, Wilson O. 2008. Built environment issues in unserved and underserved African-American neighborhoods in North Carolina. Environ Justice 1(2):63–72, 10.1089/env.2008.0509.

Levels and Determinants of DDT and DDE Exposure in the VHEMBE Cohort

Author Affiliations open

1Center for Environmental Research and Children’s Health (CERCH), School of Public Health, University of California at Berkeley, Berkeley, California, USA

2Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montréal, QC, Canada

3Maryland Institute for Applied Environmental Health, School of Public Health, University of Maryland, College Park, Maryland, USA

4Department of Environmental Health and Rollins School of Public Health, Emory University, Atlanta, Georgia, USA

5Department of Urology, University of Pretoria, Pretoria, South Africa

6University of Pretoria Centre for Sustainable Malaria Control and School of Health Systems and Public Health, University of Pretoria, Pretoria, South Africa

PDF icon PDF Version (1.2 MB)

  • Background:
    Although indoor residual spraying (IRS) is an effective tool for malaria control, its use contributes to high insecticide exposure in sprayed communities and raises concerns about possible unintended health effects.
    Objective:
    The Venda Health Examination of Mothers, Babies and their Environment (VHEMBE) is a birth cohort study initiated in 2012 to characterize prenatal exposure to IRS insecticides and exposures’ impacts on child health and development in rural South Africa.
    Methods:
    In this report, we describe the VHEMBE cohort and dichlorodiphenyltrichloroethane (DDT) and dichlorodiphenyldichloroethylene (DDE) serum concentrations measured in VHEMBE mothers when they presented for delivery. In addition, we applied a causal inference framework to estimate the potential reduction in population-level p,p′-DDT and p,p′-DDE serum concentrations under five hypothetical interventions. A total of 751 mothers were enrolled.
    Results:
    Serum concentrations of p,p′ isomers of DDT and DDE were above the limit of detection (LOD) in ≥98% of the samples, whereas the o,p′ isomers were above the LOD in at least 80% of the samples. Median (interquartile range) p,p′-DDT and p,p′-DDE serum concentrations for VHEMBE cohort participants were 55.3 (19.0–259.3) and 242.2 (91.8–878.7) ng/g-lipid, respectively. Mothers reporting to have lived in a home sprayed with DDT for malaria control had ∼5–7 times higher p,p′-DDT and p,p′-DDE serum concentrations than those who never lived in a home sprayed with DDT. Of the five potential interventions tested, we found increasing access to water significantly reduced p,p′-DDT exposure and increasing the frequency of household wet mopping significantly reduced p,p′-DDT and p,p′-DDE exposure.
    Conclusion:
    Our findings suggest that several intervention approaches may reduce DDT/DDE exposure in pregnant women living in IRS communities. https://doi.org/10.1289/EHP353
  • Received: 01 May 2016
    Revised: 19 December 2016
    Accepted: 19 January 2017
    Published: 07 July 2017

    Address correspondence to B. Eskenazi, Center for Environmental Research and Children’s Health. 1995 University Ave., Suite 265. Berkeley, CA 94704, USA. Email: eskenazi@berkeley.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP353).

    The authors declare they have no actual or potential competing financial interests.

    Note to readers with disabilities: EHP strives to ensure that all journal content is accessible to all readers. However, some figures and Supplemental Material published in EHP articles may not conform to 508 standards due to the complexity of the information being presented. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.

  • PDF icon Supplemental Material PDF (805 KB)


    Note to readers with disabilities: EHP has provided a 508-conformant table of contents summarizing the Supplemental Material for this article (see below) so readers with disabilities may determine whether they wish to access the full, nonconformant Supplemental Material. If you need assistance accessing journal content, please contact ehponline@niehs.nih.gov. Our staff will work with you to assess and meet your accessibility needs within 3 working days.
    PDF icon Supplemental Table of Contents PDF (114 KB)

Introduction

In 2015, malaria infected approximately 214 million people worldwide and resulted in nearly 438,000 deaths (World Health Organization 2015). Indoor residual spraying (IRS), the application of insecticides to interior walls, ceilings, and eaves, is a malaria-vector control policy adopted by 88 countries (World Health Organization 2014b), protecting approximately 116 million people worldwide (World Health Organization 2015). The World Health Organization’s (WHO) Pesticide Evaluation Scheme recommends 12 insecticides from four chemical classes for IRS that include organochlorine (OC), organophosphate, carbamate, and pyrethroid insecticides (World Health Organization 2014a). Although banned in most countries, at least 10 countries, including Botswana, Democratic Republic of Congo, Gambia, India, Mozambique, Namibia, South Africa, Swaziland, Zambia, and Zimbabwe, used the OC insecticide dichlorodiphenyltrichloroethane (DDT) for IRS in 2014 (World Health Organization 2014b). The comparative advantages of DDT to other insecticides used for IRS include its longer residual efficacy (>6 months) (World Health Organization 2014a) and noncontact spatial repellent properties (Grieco et al. 2007). In some areas of South Africa, such as the Limpopo Province, DDT had been continuously used since the 1940s (Mabaso et al. 2004).

Although the benefits of decreased malaria infection are clear (Kim et al. 2012; Mabaso et al. 2004), the use of DDT for malaria control has contributed to uniquely high DDT exposure in sprayed communities (Aneck-Hahn et al. 2007; Bouwman et al. 2006; Channa et al. 2012; Ortiz-Pérez et al. 2005; Sereda et al. 2009; Van Dyk et al. 2010; Whitworth et al. 2014). Of particular concern is exposure to pregnant women, as DDT can cross the placental barrier and expose the developing fetus (Waliszewski et al. 2000). Biomonitoring studies of pregnant women living in IRS areas are sparse, but Channa et al. (2012) reported median p,p′-DDT and p,p′-dichlorodiphenyldichloroethylene (DDE) plasma concentrations in women delivering in a high-risk malaria area of KwaZulu-Natal Province in South Africa (n=91) to be 2,788 and 4,092 ng/g-lipid, respectively. These levels are substantially higher than median p,p′-DDT and p,p′-DDE plasma concentrations in South African women giving birth in areas of low malaria risk (n=47; p,p′-DDT = 27 and p,p′-DDT = 184   ng/g-lipid) and in nonmalarial areas (n=117 p,p′-DDT = 7 and p,p′-DDE = 26 ng/g-lipid) (Channa et al. 2012).

The few studies that have investigated determinants of DDT exposure in populations living in IRS areas have found that living in either a home or village sprayed for malaria control was associated with higher DDT body burden (Aneck-Hahn et al. 2007; Bouwman et al. 2006; Channa et al. 2012; Herrera-Portugal et al. 2005; Manaca et al. 2011; Sereda et al. 2009; Van Dyk et al. 2010; Whitworth et al. 2014), but few additional determinants have been examined. Whitworth et al. (2014) recently reported that, in a subset of Limpopo women living in unsprayed villages (n=175), women with water piped into their yards had 73% and 61% lower DDT and DDE levels, respectively, than had women whose water source was a public tap. In addition, Limpopo women living in DDT-sprayed homes (n=100) who performed more than six preventative measures to prepare their home for IRS (e.g., covering food/water, taking furniture out of the house) had 40% lower DDT serum levels than women who performed fewer than four preventative measures (Whitworth et al. 2014). In the only study investigating the determinants of prenatal exposure to DDT in an IRS population (n=255), Channa et al. (2012) found that length of breastfeeding, age, parity, level of education, and permanent employment of the mothers were inversely associated with p,p′-DDT/E plasma concentrations.

Most studies have quantified determinants of xenobiotic exposures by fitting a single regression model with all covariates and interpreting the coefficients as the association of each variable with serum levels. This approach may not give valid estimates of effects and inference if the model inputs are not a priori specified or improper assumptions about the relationship between the exposure and outcome are made (e.g., linear relationship) (Ritter et al. 2014). Further, covariates along the causal pathway between exposure and outcome (mediators) are often improperly included within a single model, biasing the results (Schisterman et al. 2009). Public health researchers are ultimately interested in the marginal (population-level) effect of specific interventions on exposure. Under a causal inference framework, one can test the population change in insecticide levels that would be observed if specified interventions were implemented (e.g., increasing access to water, cleaning floors at a given frequency) (Pearl 2000). Therefore, the results of a causal inference analysis are tailored to the ultimate purpose of the study — instituting public health policies that reduce insecticide exposure in IRS communities.

In our study, we investigated p,p′/o,p′-DDT and p,p′/o,p′-DDE serum concentrations of pregnant women enrolled in the Venda Health Examination of Mothers, Babies and their Environment (VHEMBE) study. We examined bivariate determinants of DDT and DDE exposure and evaluated five hypothetical interventions aiming to reduce exposure using a causal inference framework.

Methods

Study Population

VHEMBE is a birth cohort study based in the rural Vhembe district of South Africa’s Limpopo Province. The study aims to investigate the potential effects of IRS insecticide exposure on child growth and development. Between August 2012 and December 2013, we enrolled mother–newborn dyads at the time of maternal presentation for delivery at Tshilidzini Hospital. Eligible women were ≥18 years old, spoke Tshivenda at home, lived within 20 km of the hospital, planned to remain in the area, had not been diagnosed with malaria during pregnancy, had contractions >5 minutes apart, and gave birth to a viable singleton. We obtained informed consent by verbally explaining the study procedures prior to the collection of study data. All human subject protocols were approved by institutional review boards at the University of California, Berkeley; McGill University; the University of Pretoria; the Limpopo Department of Health and Social Development; and Tshilidzini Hospital.

Out of the 1,649 women approached to participate in the VHEMBE study, 920 were eligible (∼57%). Of those eligible, 152 refused enrollment (∼16%), 14 did not complete a baseline questionnaire (∼1%), and three did not provide a sufficient blood sample for DDT analysis (<1%). In total, 751 mothers completed a baseline questionnaire and provided a blood sample, and 722 were visited at their homes by our staff one week after delivery (∼96%). On average, mothers enrolled in the VHEMBE study were 1.6 years younger and had given birth to 0.2 fewer children prior to the index child than had eligible mothers who refused enrollment (p-values<0.05).

Maternal and Home Characteristics

Tshivenda-speaking study staff administered a baseline questionnaire before hospital discharge to collect data on demographic characteristics (e.g., maternal age, primary language, marital status, education, and household income), parity, length of cumulative breastfeeding, hygiene/cleaning habits, and housing and IRS-use history. Household income was compared with the food poverty line determined by Statistics South Africa (W. Ruch, written communication, May 2014; Statistics South Africa 2014). We also assessed nutrient intake by administering a quantitative food frequency questionnaire (FFQ) validated in the Limpopo population (MacIntyre et al. 2001a,b; MacIntyre et al. 2001c). FFQ parameters were generated using the Food Finder 3 program (Nutritional Intervention Research Unit and Biomedical Research Division). Maternal height was measured using a wall-mounted stadiometer (Charder HM210D; Taichung City, Taiwan), and weight was measured with a digital scale (Beurer PS06; Ulm, Germany). All measurements were performed in triplicate, with the mean values used to calculate body mass index (BMI). At the one-week visit, we performed home inspections to collect information on household water source, building type, and homes’ latitude and longitude coordinates.

Generating Spatial Variables

We used 2009 Spot 5 satellite imagery to create spatial variables to test for the association between location and p,p′-DDT/E serum concentrations. We calculated the minimum distance from each participant’s home to the nearest body of water using ArcGIS’s ‘Near’ tool. Water bodies were defined based on publicly accessible national datasets but were supplemented with manual additions, drawn using ArcScan based on the Spot 5 imagery. The distance-to-body-of-water variable was created based on the hypothesis that participant homes located near bodies of water (potential mosquito habitats) would be more likely to undergo IRS applications and proximity would result in higher exposure to participants. In addition, we used the kernel density ArcTool to calculate the number of structures per hectare within 250 and 1,000 m buffers from the participant’s home. Our structure density variable was created based on the hypothesis that density of IRS use in the area (spraying is done by structure) would influence the exposure of the participant within that area. This variable was generated using ArcScan to extract imagery pixels with a radiometric resolution of 220 or higher. The resulting extracted pixel layer was cleaned by hand (to minimize misclassification of other features such as roads and structures). The pixels were then converted to points, and the kernel density was completed.

Measurement of p,p′ and o,p′ Isomers of DDT/E

Maternal blood was collected into red-top vacutainer tubes by study nurses prior to delivery (n=595) or immediately after delivery (n=156). Samples were immediately processed and stored at −80°C. Serum aliquots were sent on dry ice to Emory University’s Rollins School of Public Health for the measurement of p,p′ and o,p′ DDT/E using gas chromatography-tandem mass spectrometry (GC-MS) with isotope dilution quantification (Barr et al. 2003). The limit of detection (LOD) and limit of quantification (LOQ) for p,p′-DDT, o,p′-DDT, and o,p′-DDE were 0.01 and 0.05 ng/mL, respectively. For p,p′-DDE, the LOD and LOQ were 0.03 and 0.15 ng/mL, respectively. Total lipid concentrations were estimated based on triglycerides and total cholesterol concentrations (Phillips et al. 1989), measured using standard enzymatic methods (Roche Chemicals, Indianapolis, IN). Quality-control procedures included field spikes, field blanks, matrix-matched calibrants, and laboratory-prepared serum and reagent blanks analyzed concurrently with participants’ samples. The Supplemental Information (SI) describes the laboratory method used to quantify the p,p′ and o,p′ isomers of DDT/E and provides detailed quality control information.

Data Analysis

We used Spearman’s correlation or Kruskal-Wallis tests to examine the bivariate relationships of participant characteristics and potential determinants of exposure with DDT/DDE serum concentrations. Only the p,p′ isomers of DDT and DDE were considered for these analyses due to lower detection frequencies in the o,p′ isomers of DDT and DDE. For the p,p′-DDT/E serum concentrations below the laboratory’s LOQ, but above the LOD, we assigned those values the GC-MS machine-read value (n=27 for p,p′-DDT and n=12 for p,p′-DDE). For p,p′-DDT serum concentrations below the laboratory’s LOD (n=15), we imputed those values from maximum likelihood estimates of the lognormal distribution of the detected serum values (Lubin et al. 2004). Spearman’s correlation tests were also used to examine the correlation between p,p′-DDT and p,p′-DDE serum concentrations. Associations were considered statistically significant if p-values were <0.05.

For the causal inference analysis, we aimed to estimate the marginal geometric mean difference in p,p′-DDT/E serum levels (Y) if, contrary to fact, all VHEMBE mothers were given an intervention (A=1) relative to a scenario in which none of the mothers were given that intervention (A=0): E[E(Y|A=1,W)−E(Y|A=0,W)], where W is a matrix of covariates. In addition, we tested the effect of potential interventions by whether the mother reported ever living in a home sprayed with DDT to explore exposure reduction effect modification by spray status. Covariates used in the TMLE analysis included the following: if the mother ever lived in a home sprayed with DDT for malaria control (W1, categorical); if the mother lived in a home sprayed with DDT for malaria control during pregnancy (W2, categorical); the frequency of IRS in the home where the mother lived during pregnancy (W3, ordinal); if the mother lived in a village sprayed for malaria control during pregnancy (W4, categorical); the frequency of IRS in the village where the mother lived during pregnancy (W5, ordinal); the time spent in an IRS home (W6, no. of years); the mother’s age at delivery (W7, years); the education level of mother at delivery (W8, ordinal); household income (W9, Rands per household member per month); whether the pregnancy home was a rondavel with earthen walls and thatched roof (W10, categorical); parity of mother at delivery (W11, no. of previous births); breastfeeding history (W12, no. of months); presence of a rondavel on homestead (W13, categorical); if the household owned livestock (W14, categorical); proximity of mother’s home to the nearest body of water (W15, meters); structure density within 250-m radius of the mother’s home (W16, no./hectare); and maternal BMI after delivery (W17, kg/m2).

The potential interventions that we evaluated included: 1) living in a home with piped water [A1=A1(A4,5,W1−17)];)]; 2) living in a home in which floors were mopped more than seven times weekly (median frequency reported by mothers) [A2=A2(A1,A3−5,W1−17)], 3) washing bed sheets more than two times per month (median frequency reported by mothers) [A3=A3(A1,2,A4,5W1−17)]; 4) avoiding a high-fat diet (<the 75th percentile of fat intake among VHEMBE women) [A4=A4(A1−3,A5,W1−16)]; and 5) avoiding local dairy/meat/poultry fish products during pregnancy [A5=A5(A1−4,W1−17)]. The potential interventions for this analysis were selected because they may be modifiable characteristics that were hypothesized to reduce DDT/DDE exposure, while maintaining effective malaria control.

The marginal geometric mean difference of p,p′-DDT/DDE serum concentrations for each intervention was evaluated in separate models using targeted maximum likelihood estimations (TMLE), a doubly robust substitution estimator that generates unbiased estimates if either models for the estimation of the exposure [E(Y|A,W)] or determinant mechanisms [E(A|W)] are correct (Rose and van der Laan 2011; van der Laan 2006; van der Laan and Rubin 2006). A directed acyclic graph (DAG) was generated to conceptualize the estimation of serum levels and interventions and to identify potential confounders (Figure S1) (Textor et al. 2011). Missing covariate values (<5%) were imputed at random based on their observed probability distributions.

To estimate [E(Y|A,W)] and E(A|W), we used the Super Learner algorithm, an ensemble machine learning algorithm that uses a weighted combination of algorithms to return a prediction function that minimizes cross-validated mean squared error (van der Laan et al. 2007). We assessed positivity using the propensity score for each intervention and found that our positivity assumption holds for all interventions as the lowest propensity score was 0.07, and the median propensity scores across all interventions ranged from 0.53 to 0.75 (Table S1). To estimate E(Y|A,W) and E(A|W), we used the Super Learner algorithm with the following candidate algorithms: generalized linear models, generalized additive models, Bayesian linear model, support vector machine, recursive partitioning and regression trees, elastic net, neural network, local polynomial regression, and random forest. The associated weights used by Super Learner for estimating E(Y|A,W) and E(A|W) are presented in the Supplemental Material (Tables S2 and S3). We used bootstrapping (n=1,000) to estimate 95% confidence intervals (CI) based on the percentile method (Efron 1979). Data analyses were performed using the statistical programs R (version 3.1.3; R Development Core Team) and ArcGIS (version 10.3; ESRI Corporation).

Exposure Levels Comparison with Other Populations

We compared VHEMBE lipid-adjusted p,p′-DDT and p,p′-DDE serum concentrations to serum/plasma levels previously reported in 1) adults living in IRS communities and 2) pregnant women from non-IRS communities in the United States. The median and inter-quartile ranges (IQR) were used to compare serum/plasma concentrations across studies, as those descriptive statistics were the most commonly reported. Because Aneck-Hahn et al. (2007) reported only the arithmetic mean and standard deviation (SD) of men living and not living in DDT-sprayed homes, the geometric mean (GM) and geometric standard deviation (GSD) were estimated according to equations presented in Jean and Helms (1983). We sampled 1,000 values from a log-normal distribution using the estimated GM and GSD to estimate the median and IQR of population from Aneck-Hahn et al. (2007). As only wet-weight concentrations (ng/mL) were presented by Whitworth et al. (2014), the Study of Women and Babies (SOWB) researchers graciously provided the lipid-adjusted distributions for comparison (K.W. Whitworth, written communication, October 2014). We compare only the Van Dyk et al. (2010) results for p,p′-DDE in adults living in home sprayed ∼60 days prior to blood collection because the detection frequency for p,p′-DDT in sprayed homes (5%) and p,p′-DDT/E in unsprayed communities were low (0 and 33%, respectively). Only the lipid-adjusted values from the control group (n=283) were used from the case-control study of Bhatia et al. (2004). The p,p′-DDT/E serum concentrations from pregnant women who participated in the 1999–2000, 2001–2002, and 2003–2004 National Health and Nutrition Examination Study (NHANES) (DDT n=263, DDE n=277) were combined (Center for Disease Control 2000, 2002, 2004). In the three NHANES surveys, p,p′-DDE was detected in 100% of the samples, and p,p′-DDT was detected in 37% of the samples (LOD∼5.1 ng/g-lipid).

Results

Study Participants

All mothers were born in South Africa and were black Africans. They had a mean (SD) age of 24.9 (6.3) years at delivery (Table 1). Most of the mothers had less than a 12th-grade education (54.9%), lived below the South African food poverty line of $25 per person per month (58.3%), and were multiparous (56.8%). Almost a third (31.4%) of the mothers reported living in a village sprayed for malaria control during pregnancy, 3.1% reported living in a home sprayed with DDT during pregnancy, and 33.8% reported living in a home sprayed with DDT for malaria control in their lifetime. Of those mothers reporting that their home was sprayed for malaria control during pregnancy (n=40), the majority reported that they were inside the home during IRS (80%) and did not move household items outside prior to IRS (58%). However, 68% did report that they covered household items prior to IRS.

Table 1. Demographic characteristics of participants in the VHEMBE study, Limpopo, South Africa (n=751).
Characteristic n (%)a
Maternal age (years)
 18–24 377 (50.2)
 25–30 172 (22.9)
 30–35 111 (14.8)
 >35 91 (12.1)
Primary language
 Tshivenda 734 (97.7)
 Tshitsonga (Xitsonga) 14 (1.9)
 Tshipedi (Sepedi) 3 (0.4)
Married or living as married
 No 392 (52.2)
 Yes 359 (47.8)
Education
 <Grade 12 412 (54.9)
 Completed grade 12 229 (30.5)
 Further studies started 50 (6.7)
 Diploma or further degree 60 (8.0)
Povertyb
 Above food poverty line 310 (41.3)
 Below food poverty line 438 (58.3)
 Don’t know 3 (0.4)
Parity
 0 325 (43.3)
 1 201 (26.8)
 ≥2 225 (30.0)
Mother ever had malaria
 No 727 (96.8)
 Yes 24 (3.2)

aPercentages may not add to 100% due to rounding.

bFood poverty line based on Statistics South Africa (370 Rands or about $25 monthly income per household member) (W. Ruch, written communication, May 2014; Statistics South Africa 2014).

Serum Concentrations of DDT and DDE

Serum concentrations were typically above the LOD for all DDT/DDE isomers (>82%), with p,p′-DDT and p,p′-DDE above the LOD in 98% and 100% of the samples, respectively. Furthermore, p,p′-DDT/E concentrations were above the LOQ in >90% of the samples and o,p′ DDT/E were above the LOQ in <67% of the samples (Table 2). Median (IQR) p,p′-DDT and p,p′-DDE concentrations were 55.3 (19.0−259.3) and 242.2 (91.8–878.7) ng/g-lipid, respectively. Extreme serum concentration outliers were observed for both p,p′-DDT (90th%ile=946.2 ng/g-lipid) and p,p′-DDE (90th%ile=2,577.7 ng/g-lipid). VHEMBE mothers’ p,p′-DDT and p,p′-DDE serum concentrations were strongly correlated (rho=0.86, p-value<0.01). Exposure to p,p′-DDT and p,p′-DDE was elevated among women living south and west of Tshilidzini Hospital (Figures 1 and S2).

Spatial distribution map of South Africa indicating sites with exposure to p,p’-DDT by quintile.
Figure 1. Spatial distribution of p,p‘-DDT concentrations in relation to Tshilidzini Hospital.
Table 2. DDT and DDE serum concentrations in VHEMBE participants, Limpopo, South Africa (ng/g-lipid).
Isomer %>LODa %>LOQb GMc GSDc Min 10th% 25th% Median 75th% 90th% Max
p,p′-DDT 98.0 90.7 69.6 6.7 <LOQ 8.1 19.0 55.3 259.3 946.2 15027.6
o,p′-DDT 90.5 66.6 <LOQ <LOQ <LOQ 7.1 22.6 72.0 2029.3
p,p′-DDE 100.0 97.2 287.9 4.8 <LOQ 44.7 91.8 242.2 878.7 2577.7 26301.3
o,p′-DDE 82.7 48.2 <LOQ <LOQ <LOQ <LOQ 6.9 13.0 117.5

aLOD=limit of detection. LOD for p,p′-DDT, o,p′-DDT, and o,p′-DDE was 0.01 ng/mL and the LOD for p,p′-DDE was 0.03 0.03 ng/mL.

bLOQ=limit of quantification. LOQ for p,p′-DDT, o,p′-DDT, and o,p′-DDE was 0.03 ng/mL and the LOQ for p,p′-DDE was 0.15 ng/mL.

cGM=geometric mean, GSD=geometric standard deviation. GM and GSD not calculated for o,p′ isomers due to lower detection frequencies. For p,p′-DDT/E, GM and GSD calculations include values below the LOD using imputed values from maximum likelihood estimates of the lognormal distribution and values below the LOQ, but above the LOD, using GC/MS machine-read values.

Bivariate Determinants Analysis

Mothers reporting that the villages in which they lived during pregnancy were sprayed for malaria control every year (n=63) had significantly higher (p<0.01) p,p′-DDT and p,p′-DDE serum concentrations than mothers who lived in an unsprayed village had (n=516) (median p,p′-DDT: 562.7 vs. 38.1 ng/g-lipid; median p,p′-DDE: 1,431.1 vs. 178.1 ng/g-lipid, respectively) (Table 3 and S4). Mothers who lived in a home that was sprayed with DDT during their pregnancy (n=23) had median p,p′-DDT and p,p′-DDE concentrations an order of magnitude higher than those who did not (n=720) (p,p′-DDT: 736.9 vs. 50.0 ng/g-lipid; p,p′-DDE: 2,129.0 vs. 230.9 ng/g-lipid, respectively).

Tab