Category Archives: Health

Topics such as diseases, healthcare, Medicare, Medicaid, healthcare, and medical research.

Ambient Ozone Pollution and Daily Mortality: A Nationwide Study in 272 Chinese Cities

Author Affiliations open

1National Center for Chronic and Noncommunicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing, China

2Key Laboratory of Public Health Security, School of Public Health, Ministry of Education, Fudan University, Shanghai, China

3Key Laboratory of Health Technique Assessment, School of Public Health, Ministry of Health, Fudan University, Shanghai, China

4Shanghai Key Laboratory of Meteorology and Health, Shanghai, China

5Department of Environmental Health, Rollins School of Public Health, Emory University, Atlanta, Georgia, USA

6Key Laboratory of Reproduction Regulation of National Population and Family Planning Commission, Shanghai Institute of Planned Parenthood Research, Institute of Reproduction and Development, Fudan University, Shanghai, China

PDF icon PDF Version (217 KB)

  • Background:
    Few large multicity studies have been conducted in developing countries to address the acute health effects of atmospheric ozone pollution.
    Objective:
    We explored the associations between ozone and daily cause-specific mortality in China.
    Methods:
    We performed a nationwide time-series analysis in 272 representative Chinese cities between 2013 and 2015. We used distributed lag models and over-dispersed generalized linear models to estimate the cumulative effects of ozone (lagged over 0–3 d) on mortality in each city, and we used hierarchical Bayesian models to combine the city-specific estimates. Regional, seasonal, and demographic heterogeneity were evaluated by meta-regression.
    Results:
    At the national-average level, a 10-μg/m3 increase in 8-h maximum ozone concentration was associated with 0.24% [95% posterior interval (PI): 0.13%, 0.35%], 0.27% (95% PI: 0.10%, 0.44%), 0.60% (95% PI: 0.08%, 1.11%), 0.24% (95% PI: 0.02%, 0.46%), and 0.29% (95% PI: 0.07%, 0.50%) higher daily mortality from all nonaccidental causes, cardiovascular diseases, hypertension, coronary diseases, and stroke, respectively. Associations between ozone and daily mortality due to respiratory and chronic obstructive pulmonary disease specifically were positive but imprecise and nonsignificant. There were no statistically significant differences in associations between ozone and nonaccidental mortality according to region, season, age, sex, or educational attainment.
    Conclusions:
    Our findings provide robust evidence of higher nonaccidental and cardiovascular mortality in association with short-term exposure to ambient ozone in China. https://doi.org/10.1289/EHP1849
  • Received: 4 March 2017
    Revised: 3 October 2017
    Accepted: 20 October 2017
    Published: 21 November 2017

    Address correspondence to H. Kan, Department of Environmental Health, School of Public Health, Fudan University, P.O. Box 249, 130 Dong-An Road, Shanghai 200032, China. Telephone: 86 (21) 5423 7908. Email: kanh@fudan.edu.cn and M. Zhou, National Center for Chronic and Noncommunicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, 27 Nanwei Road, Xicheng District, Beijing 100050 China. Telephone: 86 (10) 6301 5058. Email: maigengzhou@126.com

    Supplemental Material is available online (https://doi.org/10.1289/EHP1849).

    *These authors contributed equally to this work.

    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 (256 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 (138 KB)



Introduction

Ambient air pollution has emerged as a major public health concern worldwide (Cohen et al. 2017). As a key component of the photochemical air pollution mixture, tropospheric ozone is widely considered one of the most important air pollutants (Bell et al. 2004). A large body of epidemiological studies has shown that short-term exposure to air pollution could lead to increased mortality and morbidity, but there is far less evidence on ozone than on particulate matter, and the exisiting evidence is more inconsistent (Atkinson et al. 2014; Bell et al. 2005; Dominici et al. 2006; Pope and Dockery 2006). Interpretation of the epidemiological evidence on ozone was also constrained by a limited range of study sites, by variability in population susceptibility, and by the different statistical methods used to adjust for the confounding effects of concomitant exposures (temperature, particulate matter, etc.). Furthermore, few investigations have provided estimates on cause-specific cardiorespiratory mortality and on the potential effect modifications (Bell and Dominici 2008; Halonen et al. 2010).

Several large multicity studies have evaluated associations between ozone and daily mortality in North America, Europe, and some developed Asian cities (Bell et al. 2004, 2005; Gryparis et al. 2004; Ng et al. 2013; Peng et al. 2013; Wong et al. 2008), but these associations have been investigated in only a small number of cities in developing countries where ozone pollution has become increasingly severe in recent years owing to the rapid growth of urbanization (Anenberg et al. 2010; Bell et al. 2005; Chen et al. 2014; Yan et al. 2013). The lack of evidence in developing countries was mainly driven by the scarce monitoring data on ozone. Ozone was gradually introduced into China’s national air quality monitoring network beginning in January 2013, and its real-time monitoring data were publicly available in ≤366 Chinese cities by the end of 2015.

We performed a nationwide study to evaluate the day-to-day associations between ozone and cause-specific mortality at both national and regional levels. In this study, we further evaluated whether the associations were changed by concomitant exposure to other air pollutants and how the associations varied by regional, seasonal, and demographic characteristics.

Methods

Data Collection

This analysis was based on daily air pollutant concentrations, weather conditions, and cause-specific mortality counts in 272 Chinese cities from 2013 to 2015, which have been described elsewhere (Chen et al. 2017). The 272 cities were selected because we could obtain daily data on ozone, mortality (>3) nonaccidental deaths per day on average), and weather parameters. They included 69 cities with 3-y data, 74 cities with 2-y data, and 129 cities with 1-y data. The cities are dispersed over all 31 provincial administrative regions and include a population of nearly 290 million, accounting for approximately 20% of the total population of mainland China. These cities were classified into four regions according to the common regional divisions in terms of geography, climate, and culture in China (see Figure S1): Northwest (n=21), North (n=107), South (n=140) and Qing-Tibet (n=4). In brief, the Northwest region is an arid area composed of plateaus, basins, deserts, and meadows; the Qing-Tibet region overlaps the Qinghai-Tibet Plateau, the world’s highest plateau; and the North and South regions are populous with apparent differences in climatic and cultural characteristics.

The daily mortality counts for each city were obtained from China’s Disease Surveillance Points System (DSPS), which has been shown to have good national and provincial representativeness (Liu et al. 2016). The DSPS randomly selected several districts or communities within a city. Detailed descriptions of the sampling and development of the DSPS have been published elsewhere (Liu et al. 2016; Yang et al. 2005). According to the International Classification of Diseases, 10th revision (ICD-10, WHO 2016), daily mortality counts in each city were further categorized into total nonaccidental causes (total, codes: A00–R99), cardiovascular diseases (CVD, codes: I00–I99), hypertension (codes: I10–I15), coronary heart disease (CHD, codes: I20–I25), stroke (codes: I60–I69), respiratory diseases (codes: J00–J98), and chronic obstructive pulmonary disease (COPD, codes: J41–J44). Finally, we divided daily deaths into several strata by age ranges (5–64 y, 65–74 y, and ≥75 y), sex, and educational attainment (low: ≤9 years; high: >9 years).

Daily ozone data were derived from China’s National Urban Air Quality Real-time Publishing Platform (http://106.37.208.233:20035/). We calculated maximum 8-h mean concentrations of ozone, which was typically measured from 1000 hours to 1800 hours. To allow adjustment for the simultaneous exposure to copollutants, we also collected daily 24-h average concentrations of particulate matter with an aerodynamic diameter ≤2.5 μm (PM2.5), sulfur dioxide (SO2), nitrogen dioxide (NO2), carbon monoxide (CO), and nitrogen dioxide (NO2). Through 2015, a total of 1,265 monitors were included in the present study; however, the number varied appreciably in each city, with a median of 4 (minimum: 1, maximum: 17). To calculate daily 8-h or 24-h mean concentrations, ≤75% of the 1-h values must have been available on that particular day; otherwise, the entire station was excluded. We then averaged the measurements from all valid monitoring sites in a city. We obtained daily mean temperature and mean relative humidity in each city from the China Meteorological Data Sharing Service System (http://data.cma.cn/). More information about data sources for cause-specific deaths, air pollutants, and weather conditions is available in our previous publication (Chen et al. 2017). Our study protocol was approved by the Institutional Review Board at the School of Public Health, Fudan University (No. 2014-07-0523), with a waiver of informed consent because all data were aggregated at city level, and no subjects were contacted.

Statistical Analysis

Daily mortality counts, ozone concentrations, and weather parameters were linked by date in each city. A two-stage Bayesian hierarchical model was applied to obtain regional- and national-average associations between ozone and cause-specific mortality (Dominici et al. 2006).

In the first stage, the associations in each city were estimated using over-dispersed generalized linear models (GLM). The polynomial distributed lag model (PDLM) was selected a priori for ozone in our main analyses because it can account for collinearity between different lag days and is, in principle, more appropriate than single-day lags and averaging-day lags for exploring the cumulative health risks of an exogenous exposure (Gasparrini 2011). Linear models were used in the present analysis because most previous studies hypothesized linear associations between ozone and adverse health outcomes (Bell et al. 2004; Gryparis et al. 2004; Halonen et al. 2010; Peng et al. 2013). A maximum lag of 3 d in the PDLM (PDLM 0–3) was selected a priori in main analyses because previous studies have reported that associations between adverse health outcomes and short-term air pollutant exposures (including ozone) were limited to exposures on the present day and the previous 3 d (Bell et al. 2004, 2005; Samoli et al. 2009). However, we also considered other lag structures, including single-day lags [same day (lag 0) or 1, 2, or 3 d prior (lag 1–lag 3)], exposures averaged over multiple lag days (lag 0–1, 0–2, and 0–3), and PDLM estimates for cumulative exposures over the same day and 3, 6, or 9 d prior (PDLM 0–3, 0–6, and 0–9, respectively).

Several covariates were incorporated in the main GLM: a) a natural cubic spline smooth function of calendar day with 7 degrees of freedom (df) per year to exclude seasonality in mortality; b) a factor variable for “day of week” to exclude possible variations of mortality within a week; c) a cross-basis function of temperature built by the distributed lag nonlinear model (DLNM) to control for its potentially nonlinear and lagged confounding effects; and d) a natural smooth function with 3 df for the present-day relative humidity. In the DLNM of temperature, a cross-basis function was established by a natural cubic spline for the space of temperature with three internal knots at equally spaced temperature percentiles to account for the potentially nonlinear relationships between temperature and mortality. For the lags, we selected a natural cubic spline with two internal knots at equally spaced log10-values of lags (plus intercept) to allow for more flexible lag effects at short delays (Guo et al. 2014). We selected a cumulative lag over the same day and 13 d prior for temperature in the DLNM (DLNM 0–13) according to previous epidemiological studies on ozone and temperature (Chen et al. 2017, 2014; Qin et al. 2017).

In the second stage, we used Bayesian hierarchical models to combine city-specific estimates for the associations between ozone and cause-specific mortality at national and regional levels. This approach has been widely used to pool risk estimates obtained from multiple locations while accounting for within-city statistical error and between-city variability (heterogeneity) of the true risks (Chen et al. 2012; Dominici et al. 2006; Peng et al. 2013). We reported the posterior mean percentage change and 95% posterior interval (PI) in daily mortality in association with a 10-μg/m3 increase in ozone concentration (Dominici et al. 2006). The 95% PI is the Bayesian formulation of the 95% confidence interval (CI). We also calculated I2 statistics and p-values for the between-city heterogeneity in random-effect models.

Using the model parameters from the main analyses, we further conducted several subgroup analyses on potential effect modification by geographical (four typical regions), seasonal (warm period, from May through October; cool period, from November through April), and demographic (age, sex, and education) factors. First, we used Bayesian hierarchical models to separately estimate the mortality effects of ozone within each subgroup (defined by region, age, sex, or educational attainment). Next, we derived p-values for differences among subgroups based on likelihood ratio tests comparing the fit of a meta-regression model with the potential modifier to the simple meta-analysis model. In addition, for cities with ≥2 years of data (n=143), we evaluated effect modification by season (cool or warm) within each region (Northwest, North, or South) using a meta-regression model with season, region (two indicator terms for North vs. Northwest and South vs. Northwest), and season×region interaction terms. The Qing-Tibet region was excluded from this analysis because of insufficient data. Finally, we evaluated effect modification of the associations between ozone and daily mortality (total nonaccidental, CVD, and respiratory) by annual-mean daily temperatures, PM2.5 concentrations, and ozone concentrations in each city using meta-regression models.

Finally, based on the main models, we performed three sensitivity analyses to assess the robustness of our estimates for the associations between ozone and daily total mortality. First, we fitted two-pollutant models with adjustment for the concomitant exposure to PM2.5, SO2, NO2, and CO, which were introduced by using the same PDLMs as those used for ozone. Second, we controlled for confounding by temperature using alternative lag structures, including parallel lags of temperature on the same day and averaging lags over 1–3 d (abbreviated as “lag 0 and 1–3”), DLNM 0–3, and DLNM 0–6. Third, we changed the degrees of freedom in the smoothness of time from 4 df to 8 df per year.

All analyses were conducted using R version 3.1.1 (R Foundation for Statistical Computing) with the stats package for fitting the GLM, the dlnm package for the PDLM and the DLNM, the tlnise package for the Bayesian hierarchical model, and the metafor package for meta-regression analyses. p-Values<0.05 were considered statistically significant in all analyses, except for a false discovery rate (FDR) of <0.05 in correcting for multiple testing on the between-season differences in each region.

Results

Descriptive Statistics

Table 1 summarizes the environmental and mortality data in 272 cities in China from 2013 to 2015. The annual-mean concentrations of ozone varied considerably, with an average of 77 μg/m3 (ranging from 36 μg/m3 to 113 μg/m3) across all cities. In general, there were appreciable variations in concentrations of ozone and copollutants within each region and between regions (Table 1; see also Table S1). Ozone concentrations were higher during the warm period than during the cool period at both national and regional levels (see Table S2). There were two to three times more deaths in the North and South regions than in the Northwest and Qing-Tibet regions (see Table S1). On average per city, there were 16 deaths from total causes, eight from CVD, one from hypertension, three from CHD, four from stroke, two from respiratory diseases, and two from COPD. The climatic conditions also varied greatly in these cities.

Table 1. Summary statistics of environment and health data in 272 Chinese cities, 2013–2015.
Variable Mean SD Min P25 P50 P75 Max
Ozone (μg/m3)
 Nationwide 77 14 36 68 77 87 113
 Northwest 77 17 44 68 72 93 102
 North 79 13 36 72 79 88 113
 South 75 13 41 67 75 85 104
 Qing-Tibet 76 26 45 59 80 96 99
Daily deaths
 Total 16 16 3 7 12 20 165
 CVD 8 7 1 3 6 10 65
 Hypertension 1 1 0 0 0 1 7
 CHD 3 3 0 1 2 3 28
 Stroke 4 4 0 2 3 5 33
 RD 2 3 0 1 1 3 34
 COPD 2 2 0 0 1 2 29
Weather conditions
 Mean temperature (°C) 15 5 −0.5 12 16 18 25
 Relative humidity (%) 68 10 35 61 71 77 91

Note: CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular diseases; Max, maximum; Min, minimum; P, percentile; RD, respiratory disease; SD, standard deviation.

Regression Results

Figure 1 depicts the lag structures in national-average associations between ozone and daily total mortality. For single-day lags, the association was strongest for ozone exposure on the same day (lag 0) and weakest for exposure three days before death (lag 3). When exposures were lagged over multiple days, associations were stronger for exposures during the 0–2 d and 0–3 d lag periods than for the 0–1 d lag and were similar to the PDLM estimates for all three lag periods evaluated. Notably, the 95% PIs became much broader when the maximum lags were extended from 4 d (as in the main model) to 7 and 10 d. Based on the main model, we estimated that a 10-μg/m3 increase in ozone (PDLM 0–3) was associated with a 0.24% (95% PI: 0.13, 0.35%) increase in daily total mortality.

Plots with posterior mean and posterior intervals plotting percentage difference (y-axis) across lags 0, 1, 2, and 3; lags 01, 02, and 03; and PDLM 0–3, 0–6, and 0–9 (x-axis).

Figure 1. National-average percentage difference (posterior mean and 95% posterior intervals) in daily total mortality per 10-μg/m3 increase in ozone concentration in 272 Chinese cities during single-day lags (lag 0, 1, 2, 3), multiple-day averaging lags (lag 0–1, 0–2, 0–3), and cumulative lags based on a polynomial distributed lag model (PDLM 0–3, 0–6, 0–9). Estimates were generated using over-dispersed generalized linear models adjusted for calendar day [natural cubic spline with 7 degrees of freedom (df)], day of the week, temperature (cross-basis function for temperature lagged for 0–13 d from distributed lag nonlinear model), and humidity (lag 0, natural smooth function, 3 df) to estimate city-specific associations that were combined using hierarchical Bayesian models.

Table 2 presents national average estimates for the associations between ozone concentrations and cause-specific mortality obtained using the main model. At the national level, the association with overall CVD mortality was slightly stronger than the association with total mortality. Associations with CHD and stroke mortality were similar to associations with overall CVD mortality, whereas the association with hypertension was stronger but less precise. Estimated percentage differences in daily mortality per 10-μg/m3 increment in ozone (PDLM 0–3) were 0.27% (95% PI: 0.10%, 0.44%) for CVD, 0.60% (95% PI: 0.08%, 1.11%) for hypertension, 0.24% (95% PI: 0.02%, 0.46%) for CHD, and 0.29% (95% PI: 0.07%, 0.50%) for stroke. Associations with respiratory and COPD mortality were slightly weaker than the association with total mortality but were much less precise and were not statistically significant.

Table 2. National- and regional-average percentage difference (posterior means and 95% posterior intervals) in daily cause-specific mortality per 10-μg/m3 increase in ozone concentration in 272 Chinese cities.
Regions Total CVD Hypertension CHD Stroke RD COPD
Nationwide 0.24 (0.13, 0.35) 0.27 (0.10, 0.44) 0.60 (0.08, 1.11) 0.24 (0.02, 0.46) 0.29 (0.07, 0.50) 0.18 (−0.11, 0.47) 0.20 (−0.13, 0.53)
North 0.28 (0.06, 0.51) 0.26 (0.01, 0.52) 0.15 (−0.72, 1.03) 0.13 (−0.24, 0.50) 0.40 (0.09, 0.70) 0.03 (−0.56, 0.62) 0.15 (−0.51, 0.81)
South 0.24 (0.09, 0.39) 0.31 (0.09, 0.52) 0.66 (0.02, 1.30) 0.30 (0.04, 0.55) 0.25 (0.02, 0.49) 0.29 (−0.05, 0.63) 0.27 (−0.11, 0.65)
Northwest −0.24 (−1.75, 1.28) 0.36 (−1.70, 2.42) 2.11 (−2.68, 6.90) 2.40 (−2.12, 6.92) 0.50 (−1.84, 2.84) −0.02 (−3.94, 3.91) −1.24 (−5.20, 2.72)
Qing-Tibet 0.90 (−2.12, 3.93) 1.47 (−0.81, 3.74) 1.79 (−3.32, 6.90) 1.85 (−3.20, 6.90) 2.23 (−1.64, 6.10) −1.23 (−5.14, 2.68) −1.37 (−5.39, 2.65)

Note: Estimates were generated using over-dispersed generalized linear models and polynomial distributed lag model for cumulative exposures over the same day and 3 days prior, adjusted for calendar day [natural cubic spline with 7 degrees of freedom (df)], day of the week, temperature (cross-basis function for temperature lagged for 0–13 days from distributed lag nonlinear model), and humidity (lag 0, natural smooth function, 3 df) to estimate city-specific associations that were combined using hierarchical Bayesian models. CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; CVD, cardiovascular diseases; RD, respiratory disease.

There was moderate between-city heterogeneity for total mortality (I2=32%) and CVD mortality (I2=32%), and the heterogeneity for respiratory mortality was low (I2=15%). Notably, the associations between ozone and various causes of mortality in the Northwest and Qing-Tibet regions of China were not statistically significant and had much broader 95% PIs than the other two regions (Table 2); however, there were no significant differences among regions according to the results of likelihood ratio tests, with p-values varying from 0.29 to 0.90 by cause of mortality. The meta-regression analysis showed greater effects of daily ozone concentrations on total mortality in cities with lower annual temperatures (p=0.02). For each 10-μg/m3 increment in daily ozone concentration, a city with 10°C lower annual-mean temperature would have an additional increase of 0.14% (95% CI: 0.02%, 0.25%) in daily total mortality. Annual-mean PM2.5 levels did not have a significant influence on associations between ozone and daily total mortality. For daily CVD mortality, only annual-mean PM2.5 had a significant effect of modification (p<0.01), with a 10-μg/m3 decrease leading to an additional increase of 0.11% (95% CI: 0.02%, 0.20%) in mortality per 10-μg/m3 increment in daily ozone concentration. We did not conduct such an analysis for daily respiratory mortality because it was not statistically associated with ozone exposure.

In the meta-regression model with season, region, and the season×region interaction term (Table 3), the associations in the North and South regions were stronger than those in the Northwest region, but differences by region were statistically nonsignificant (p=0.12 and 0.09, respectively); although there were somewhat stronger associations in the cool period at the national-average level, the between-season differences were statistically nonsignificant (p=0.13). In meta-regression models with season for each region, there were stronger associations in the warm season in the Northwest and North regions and weaker associations in the warm season in the South region, but all of these differences remained statistically nonsignificant after correcting for multiple testing (FDR from 0.08 to 0.79).

Table 3. Estimated percentage difference (posterior means and 95% posterior intervals) in daily total mortality per 10-μg/m3 increase in ozone concentration in 143 Chinese cities with ≥2 years of data, according to region and season.
Region All-year p-Valuea Cool season Warm season p-Value FDRb
Nationwide 0.23 (0.11, 0.34) 0.43 (0.21, 0.65) 0.20 (0.08, 0.31) 0.13c
Northwest 0.02 (−1.88, 1.91) −1.65 (−5.81, 2.51) 0.69 (−1.27, 2.64) 0.13b 0.19
North 0.27 (0.03, 0.51) 0.12 0.25 (−0.18, 0.68) 0.39 (0.04, 0.75) 0.79b 0.79
South 0.21 (0.07, 0.35) 0.09 0.51 (0.26, 0.76) 0.13 (−0.06, 0.33) 0.03b 0.08

Note: Analysis excludes the Qing-Tibet region because few cities had ≥2 years of data. Estimates were generated using over-dispersed generalized linear models and polynomial distributed lag model for cumulative exposures over the same day and 3 days prior, adjusted for calendar day [natural cubic spline with 7 degrees of freedom (df)], day of the week, temperature (cross-basis function for temperature lagged for 0–13 days from distributed lag nonlinear model), and humidity (lag 0, natural smooth function, 3 df) to estimate city-specific associations that were combined using hierarchical Bayesian models. —, no comparison or the reference for comparisons; FDR, false discovery rate.

ap-Values comparing effect estimates for the North and South regions to the Northwest (referent) region in meta-regression models with region, season (warm vs. cool), and season×region interaction terms.

bFDR or p-values comparing effect estimates for the warm versus cool seasons from separate meta-regression models stratified by region, with season as the predictor.

cp-Value comparing effect estimates for the warm versus cool seasons over all cities in meta-regression models with region (two indicator terms for North vs. Northwest and South vs. Northwest), season, and season×region interaction terms.

The association between ozone and total mortality also varied with demographic characteristics (Table 4). The effect estimates increased consecutively in the three age groups (5–64 y, 65–74 y, and ≥75 y), but the differences were not significant based on the likelihood ratio test (p=0.12). There were very similar estimates in males and females. The estimated percentage difference in total mortality with a 10-μg/m3 increase in daily ozone was four times as high in less-educated people than in more-educated people, although the difference was not statistically significant (p=0.41). Patterns of association with CVD mortality according to age, sex, and education were similar to those for total mortality, with statistically nonsignificant differences among subgroups (see Table S3).

Table 4. National-average percentage differences (posterior means and 95% posterior intervals) in daily total mortality per 10-μg/m3 increase in ozone concentration in 272 Chinese cities, classified by age, sex and educational attainment.
Characteristic Level Estimates p-Valuea
Age 5–64 y 0.13 (−0.23, 0.48) 0.12
65–74 y 0.19 (0.03, 0.34)
≥75 y 0.42 (0.21, 0.64)
Sex Male 0.26 (0.13, 0.39) 0.75
Female 0.21 (0.05, 0.36)
Education ≤9 y 0.25 (0.14, 0.37) 0.41
>9 y 0.06 (−0.30, 0.43)

Note: Estimates were generated using over-dispersed generalized linear models and polynomial distributed lag model for cumulative exposures over the same day and 3 days prior, adjusted for calendar day [natural cubic spline with 7 degrees of freedom (df)], day of the week, temperature (cross-basis function for temperature lagged for 0–13 days from distributed lag nonlinear model), and humidity (lag 0, natural smooth function, 3 df) to estimate city-specific associations that were combined using hierarchical Bayesian models.

aThe p-values were calculated by performing a likelihood ratio test between the simple meta-analysis model (overall estimates) and a separate meta-regression model with a categorical variable (age, sex, or education).

In sensitivity analyses, the estimates for associations between ozone and daily total mortality were changed little by adjustment for concomitant exposure to PM2.5 or CO in two-pollutant models; they attenuated appreciably but remained statistically significant after controlling for SO2 or NO2 (see Figure S2). The use of alternative lag structures for temperature did not lead to obviously different estimates for the associations (see Figure S3). The estimates for daily total mortality per 10-μg/m3 increase in ozone remained stable when the degrees of freedom per year varied from four to eight (see Figure S4).

Discussion

This multisite analysis in 272 representative Chinese cities had the advantage of analyzing national data using the same protocol and avoiding potential publication bias. It also provided a unique opportunity to explore effect modification by geographical, seasonal, and demographic characteristics. Another key advantage of this analysis was the use of distributed lag models in cumulating the mortality effects of ozone over several days (avoiding underestimates) as well as in controlling for the lagged and nonlinear effects of temperature (a strong confounder). Our estimates were also relatively robust to adjustment of concomitant exposure to copollutants (PM2.5, NO2, SO2, and CO).

This study estimated a 0.24% increase of daily total mortality per 10-μg/m3 increase in ozone concentration, which was generally comparable to previous findings in meta-analyses and multicity studies (Bell et al. 2004, 2005; Peng et al. 2013; Tao et al. 2012; Wong et al. 2008). To allow for comparability when diverse metrics for ozone were adopted, previous estimates were converted based on a metric of per 10-μg/m3 increase in maximum 8-h average concentration with a relationship of 20:15:8 for the 1-h maximum:8-h maximum:daily average. We assumed that 1.96 μg/m3=1 ppb to convert previous estimates into the same metric (Thurston and Ito 2001). For example, a recent multisite study of 21 cities in East Asia (Japan, South Korea, mainland China, Hong Kong, and Taiwan) indicated a 0.30% increase in total mortality (Chen et al. 2014). The Air Pollution and Health: A Combined European and North American Approach project estimated an overall 0.20% increment in total mortality based on a collaborative data set from 125 cities, but the values differed in the United States (0.23%), Canada (0.64%), and Europe (0.14%) (Peng et al. 2013). Investigators reported a 0.38% increase in total mortality in a multicity analysis (Bangkok, Hong Kong, Shanghai, and Wuhan) in the Public Health and Air Pollution in Asia project (Wong et al. 2008). A meta-analysis by Bell et al. (2005) summarized 144 effect estimates from 39 time-series studies worldwide and found that the estimate was 0.33% (95% CI: 0.28%, 0.60%).

We found stronger associations of ozone with CVD mortality than with total mortality, consistent with findings from most previous studies (Bell et al. 2005; Peng et al. 2013; Wong et al. 2008). Few previous studies have evaluated short-term associations between ozone exposure and mortality from hypertension and CHD. A previous meta-analysis estimated a 0.86% increase in stroke mortality per 10-μg/m3 increase in ozone (Shah et al. 2015), which was considerably higher than our corresponding estimate (0.29%). The mechanisms behind the cardiovascular effects of ozone were somewhat biologically plausible. A number of studies have suggested appreciable changes in circulating biomarkers of inflammation, oxidative stress, coagulation, vasoreactivity, lipidology, and glucose metabolism after ozone exposure (Goodman et al. 2015). However, these results lack consistency and are of uncertain clinical relevance; hence, the exact mechanisms remain to be elucidated in further investigations (Goodman et al. 2015). For example, some human studies found that short-term exposure to ozone did not impair vascular function, elevate blood pressure, or affect heart rate variability in either direction (Barath et al. 2013; Hoffmann et al. 2012).

In line with many multicity studies and meta-analyses (Bell et al. 2005; Peng et al. 2013; Wong et al. 2008), we did not observe statistically significant associations between ozone and respiratory mortality. In contrast, some other studies found significant associations with respiratory mortality and morbidity (Ji et al. 2011; Kan et al. 2008; Zmirou et al. 1998). The inconsistent results in the respiratory system might be due to limitations of the time-series approach, different model specifications, location-dependent characteristics, or greater statistical uncertainty in relation to smaller numbers of respiratory deaths than CVD deaths (particularly in warm periods, when ozone concentrations are high).

We did not find significant effect modification by geographical region, but our ability to estimate associations for the Northwest and Qing-Tibet regions was limited by the small numbers of cities (21 and 4, respectively) and daily deaths in these areas. There were no significant differences according to demographic characteristics, although estimates suggested stronger associations in the oldest age group (≥75 y) and among those with less education. In the meta-regression analysis, we found evidence of a stronger association between ozone and daily mortality with a reduction in annual-mean temperature at the city level. Similarly, data from 98 U.S. cities included in the National Morbidity, Mortality, and Air Pollution Study (NMMAPS) indicated that the association between ozone and mortality was strongest in the northeast United States and stronger in association with lower temperatures (Bell and Dominici 2008). Aging and low educational attainment have been the two most commonly reported demographic characteristics to increase susceptibility to short-term ozone exposure in previous studies (Kan et al. 2008; Bell et al. 2014).

We found some evidence of seasonal variation in associations between ozone and daily mortality, although patterns appeared to vary by region, and differences by season within regions did not reach FDR significance. In the Northwest and North regions, the association was stronger during the warm period than during the cool period, consistent with findings from previous multicity studies in North America and western Europe as well as from a recent study in East Asia (Bell et al. 2005; Chen et al. 2014; Peng et al. 2013). Stronger associations during warm seasons might be explained by the following: a) Ozone concentrations were typically higher during the warm period; b) there was widespread use of household heating in northern China during the cool period, which reduced exposure to outdoor ozone because of restricted outdoor activities and reduced natural ventilation; and c) there were more pronounced synergistic effects with high temperatures during the summer (Bell and Dominici 2008; Li et al. 2017). In southern China, however, the association between ozone and mortality was stronger during the cool season than during the warm season, consistent with findings from previous Chinese studies even when extended lags of temperature were adjusted for (Chen et al. 2017; Kan et al. 2008; Qin et al. 2017; Tao et al. 2012). It has been suggested that this pattern might result from the reduced use of air conditioning, and the consequent exposure to more natural ventilation, during cool weather (Kan et al. 2008). A stronger association between ozone and mortality during cool versus warm seasons was also reported in the U.S. NMMAPS study (Chen et al. 2012).

Limitations should be noted when interpreting our results. First, exposure measurement errors would be inevitable because we used the average of measurements across various monitors in a city as the proxy of personal exposure, but such errors have been reported to lead to an underestimate of the effects (Zeger et al. 2000). Second, ecological bias was inherent in the time-series studies because all analyses were conducted on an aggregate level rather than on an individual level. Third, owing to limited availability of such data in China, we were unable to evaluate heterogeneity or effect modification caused by the use of air conditioning, by income, by urbanization, by transportation use, or by time spent outdoors or indoors.

Conclusion

In conclusion, using a large data set covering 272 cities throughout China, we found robust evidence of higher nonaccidental and cardiovascular mortality in association with short-term exposure to ambient ozone. Our results support previous evidence of acute effects of ozone on mortality in developing countries and are generally consistent with findings for populations in North America and Europe.

Acknowledgments

The study was supported by the National Natural Science Foundation of China (91643205), the Public Welfare Research Program of the National Health and Family Planning Commission of China (201502003), the Shanghai 3-Year Public Health Action Plan (GWTD2015S04 and 15GWZK0202), and the China Medical Board Collaborating Program (13-152).

References

Anenberg SC, Horowitz LW, Tong DQ, West JJ. 2010. An estimate of the global burden of anthropogenic ozone and fine particulate matter on premature human mortality using atmospheric modeling. Environ Health Perspect 118(9):1189–1195, PMID: 20382579, 10.1289/ehp.0901220.

Atkinson RW, Kang S, Anderson HR, Mills IC, Walton HA. 2014. Epidemiological time series studies of PM2.5 and daily mortality and hospital admissions: a systematic review and meta-analysis. Thorax 69(7):660–665, PMID: 24706041, 10.1136/thoraxjnl-2013-204492.

Barath S, Langrish JP, Lundbäck M, Bosson JA, Goudie C, Newby DE, et al. 2013. Short-term exposure to ozone does not impair vascular function or affect heart rate variability in healthy young men. Toxicol Sci 135(2):292–299, PMID: 23872581, 10.1093/toxsci/kft157.

Bell ML, Dominici F. 2008. Effect modification by community characteristics on the short-term effects of ozone exposure and mortality in 98 US communities. Am J Epidemiol 167(8):986–997, PMID: 18303005, 10.1093/aje/kwm396.

Bell ML, Dominici F, Samet JM. 2005. A meta-analysis of time-series studies of ozone and mortality with comparison to the national morbidity, mortality, and air pollution study. Epidemiology 16(4):436–445, PMID: 15951661, 10.1097/01.ede.0000165817.40152.85.

Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F. 2004. Ozone and short-term mortality in 95 US urban communities, 1987–2000. JAMA 292(19):2372–2378, PMID: 15547165, 10.1001/jama.292.19.2372.

Bell ML, Zanobetti A, Dominici F. 2014. Who is more affected by ozone pollution? A systematic review and meta-analysis. Am J Epidemiol 180(1):15–28, PMID: 24872350, 10.1093/aje/kwu115.

Chen C, Zhao B, Weschler CJ. 2012. Assessing the influence of indoor exposure to “outdoor ozone” on the relationship between ozone and short-term mortality in U.S. communities. Environ Health Perspect 120(2):235–240, PMID: 22100611, 10.1289/ehp.1103970.

Chen K, Zhou L, Chen X, Bi J, Kinney PL. 2017. Acute effect of ozone exposure on daily mortality in seven cities of Jiangsu Province, China: no clear evidence for threshold. Environ Res 155:235–241, PMID: 28231551, 10.1016/j.envres.2017.02.009.

Chen R, Cai J, Meng X, Kim H, Honda Y, Guo YL, et al. 2014. Ozone and daily mortality rate in 21 cities of East Asia: how does season modify the association? Am J Epidemiol 180(7):729–736, PMID: 25139207, 10.1093/aje/kwu183.

Chen R, Kan H, Chen B, Huang W, Bai Z, Song G, et al. 2012. Association of particulate air pollution with daily mortality: the China Air Pollution and Health Effects Study. Am J Epidemiol 175(11):1173–1181, PMID: 22510278, 10.1093/aje/kwr425.

Chen R, Yin P, Meng X, Liu C, Wang L, Xu X, et al. 2017. Fine particulate air pollution and daily mortality: a nationwide analysis in 272 Chinese cities. Am J Respir Crit Care Med 196(1):73–81, PMID: 28248546, 10.1164/rccm.201609-1862OC.

Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389(10082):1907–1918, PMID: 28408086, 10.1016/S0140-6736(17)30505-6.

Dominici F, Peng RD, Bell ML, Pham L, McDermott A, Zeger SL, et al. 2006. Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases. JAMA 295(10):1127–1134, PMID: 16522832, 10.1001/jama.295.10.1127.

Gasparrini A. 2011. Distributed lag linear and non-linear models in R: the package dlnm. J Stat Softw 43(8):1–20, PMID: 22003319.

Goodman JE, Prueitt RL, Sax SN, Pizzurro DM, Lynch HN, Zu K, et al. 2015. Ozone exposure and systemic biomarkers: Evaluation of evidence for adverse cardiovascular health impacts. Crit Rev Toxicol 45(5):412–452, PMID: 25959700, 10.3109/10408444.2015.1031371.

Gryparis A, Forsberg B, Katsouyanni K, Analitis A, Touloumi G, Schwartz J, et al. 2004. Acute effects of ozone on mortality from the “Air Pollution and Health: A European Approach” project. Am J Respir Crit Care Med 170(10):1080–1087, PMID: 15282198, 10.1164/rccm.200403-333OC.

Guo Y, Gasparrini A, Armstrong B, Li S, Tawatsupa B, Tobias A, et al. 2014. Global variation in the effects of ambient temperature on mortality: A systematic evaluation. Epidemiology 25(6):781–789, PMID: 25166878, 10.1097/EDE.0000000000000165.

Halonen JI, Lanki T, Tiittanen P, Niemi JV, Loh M, Pekkanen J. 2010. Ozone and cause-specific cardiorespiratory morbidity and mortality. J Epidemiol Commun H 64(9):814–820, PMID: 19854743, 10.1136/jech.2009.087106.

Hoffmann B, Luttmann-Gibson H, Cohen A, Zanobetti A, de Souza C, Foley C, et al. 2012. Opposing effects of particle pollution, ozone, and ambient temperature on arterial blood pressure. Environ Health Perspect 120(2):241–246, PMID: 22020729, 10.1289/ehp.1103647.

Ji M, Cohan DS, Bell ML. 2011. Meta-analysis of the association between short-term exposure to ambient ozone and respiratory hospital admissions. Environ Res Lett 6(2):pii: 024006, PMID: 21779304, 10.1088/1748-9326/6/2/024006.

Kan H, London SJ, Chen G, Zhang Y, Song G, Zhao N, et al. 2008. Season, sex, age, and education as modifiers of the effects of outdoor air pollution on daily mortality in Shanghai, China: The Public Health and Air Pollution in Asia (PAPA) Study. Environ Health Perspect 116(9):1183–1188, PMID: 18795161, 10.1289/ehp.10851.

Li J, Woodward A, Hou XY, Zhu T, Zhang J, Brown H, et al. 2017. Modification of the effects of air pollutants on mortality by temperature: a systematic review and meta-analysis. Sci Total Environ 575:1556–1570, PMID: 27780592, 10.1016/j.scitotenv.2016.10.070.

Liu S, Wu X, Lopez AD, Wang L, Cai Y, Page A, et al. 2016. An integrated national mortality surveillance system for death registration and mortality surveillance, China. Bull World Health Organ 94(1):46–57, PMID: 26769996, 10.2471/BLT.15.153148.

Ng CF, Ueda K, Nitta H, Takeuchi A. 2013. Seasonal variation in the acute effects of ozone on premature mortality among elderly Japanese. Environ Monit Assess 185(10):8767–8776, PMID: 23604788, 10.1007/s10661-013-3211-6.

Peng RD, Samoli E, Pham L, Dominici F, Touloumi G, Ramsay T, et al. 2013. Acute effects of ambient ozone on mortality in Europe and North America: results from the APHENA study. Air Qual Atmos Health 6(2):445–453, PMID: 23734168, 10.1007/s11869-012-0180-9.

Pope CA III, Dockery DW. 2006. Health effects of fine particulate air pollution: lines that connect. J Air Waste Manag Assoc 56(6):709–742, PMID: 16805397, 10.1080/10473289.2006.10464485.

Qin LJ, Gu JQ, Liang SJ, Fang F, Bai WM, Liu X, et al. 2017. Seasonal association between ambient ozone and mortality in Zhengzhou, China. Int J Biometeorol 61(6):1003–1010, PMID: 27981338, 10.1007/s00484-016-1279-8.

Samoli E, Zanobetti A, Schwartz J, Atkinson R, LeTertre A, Schindler C, et al. 2009. The temporal pattern of mortality responses to ambient ozone in the APHEA project. J Epidemiol Community Health 63(12):960–966, PMID: 19648130, 10.1136/jech.2008.084012.

Shah AS, Lee KK, McAllister DA, Hunter A, Nair H, Whiteley W, et al. 2015. Short term exposure to air pollution and stroke: systematic review and meta-analysis. BMJ 350:h1295, PMID: 25810496, 10.1136/bmj.h1295.

Tao Y, Huang W, Huang X, Zhong L, Lu SE, Li Y, et al. 2012. Estimated acute effects of ambient ozone and nitrogen dioxide on mortality in the Pearl River Delta of southern China. Environ Health Perspect 120(3):393–398, PMID: 22157208, 10.1289/ehp.1103715.

Thurston GD, Ito K. 2001. Epidemiological studies of acute ozone exposures and mortality. J Expo Anal Environ Epidemiol 11(4):286–294, PMID: 11571608, 10.1038/sj.jea.7500169.

WHO (World Health Organization). 2016. International Statistical Classification of Diseases and Related Health Problems, 10th Revision. http://apps.who.int/classifications/icd10/browse/2016/en [accessed 1 February 2017].

Wong CM, Vichit-Vadakan N, Kan H, Qian Z. 2008. Public health and air pollution in Asia (PAPA): a multicity study of short-term effects of air pollution on mortality. Environ Health Perspect 116(9):1195–1202, PMID: 18795163, 10.1289/ehp.11257.

Yan M, Liu Z, Liu X, Duan H, Li T. 2013. Meta-analysis of the Chinese studies of the association between ambient ozone and mortality. Chemosphere 93(6):899–905, PMID: 23786810, 10.1016/j.chemosphere.2013.05.040.

Yang G, Hu J, Rao KQ, Ma J, Rao C, Lopez AD. 2005. Mortality registration and surveillance in China: history, current situation and challenges. Popul Health Metr 3(1):3, PMID: 15769298, 10.1186/1478-7954-3-3.

Zeger SL, Thomas D, Dominici F, Samet JM, Schwartz J, Dockery D, et al. 2000. Exposure measurement error in time-series studies of air pollution: concepts and consequences. Environ Health Perspect 108(5):419–426, PMID: 10811568, 10.1289/ehp.00108419.

Zmirou D, Schwartz J, Saez M, Zanobetti A, Wojtyniak B, Touloumi G, et al. 1998. Time-series analysis of air pollution and cause-specific mortality. Epidemiology 9(5):495–503, PMID: 9730027, 10.1097/00001648-199809000-00005.

Long-Term Exposure to Transportation Noise in Relation to Development of Obesity—a Cohort Study

Author Affiliations open

1Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden

2Centre for Occupational and Environmental Medicine, Stockholm County Council, Stockholm, Sweden

3Department of Cardiology and Internal Medicine, Belorussian State Medical University, Minsk, Belarus

4Department of Occupational and Environmental Medicine, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

5Department of Molecular Medicine and Surgery, Karolinska Institutet, Stockholm, Sweden

PDF icon PDF Version (666 KB)

  • Background:
    Exposure to transportation noise is widespread and has been associated with obesity in some studies. However, the evidence from longitudinal studies is limited and little is known about effects of combined exposure to different noise sources.
    Objectives:
    The aim of this longitudinal study was to estimate the association between exposure to noise from road traffic, railways, or aircraft and the development of obesity markers.
    Methods:
    We assessed individual long-term exposure to road traffic, railway, and aircraft noise based on residential histories in a cohort of 5,184 men and women from Stockholm County. Noise levels were estimated at the most exposed façade of each dwelling. Waist circumference, weight, and height were measured at recruitment and after an average of 8.9 y of follow-up. Extensive information on potential confounders was available from repeated questionnaires and registers.
    Results:
    Waist circumference increased 0.04 cm/y (95% CI: 0.02, 0.06) and 0.16 cm/y (95% CI: 0.14, 0.17) per 10 dB Lden in relation to road traffic and aircraft noise, respectively. No corresponding association was seen for railway noise. Weight gain was only related to aircraft noise exposure. A similar pattern occurred for incidence rate ratios (IRRs) of central obesity and overweight. The IRR of central obesity increased from 1.22 (95% CI: 1.08, 1.39) in those exposed to only one source of transportation noise to 2.26 (95% CI: 1.55, 3.29) among those exposed to all three sources.
    Conclusion:
    Our results link transportation noise exposure to development of obesity and suggest that combined exposure from different sources may be particularly harmful. https://doi.org/10.1289/EHP1910
  • Received: 17 March 2017
    Revised: 5 October 2017
    Accepted: 9 October 2017
    Published: 20 November 2017

    Address correspondence to A. Pyko, Institute of Environmental Medicine, Karolinska Institutet, SE-171 77 Stockholm, Sweden. Telephone: 46(0) 852487561. Email: Andrei.pyko@ki.se

    Supplemental Material is available online (https://doi.org/10.1289/EHP1910).

    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 (1.3 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 (79 KB)



Introduction

Large parts of the population are exposed to elevated levels of noise, particularly in urban areas. Road traffic is a dominating source but both railway and aircraft noise contribute in certain areas. Exposure to transportation noise may result in annoyance and sleep disturbance (Basner et al. 2014; Miedema and Oudshoorn 2001; Miedema and Vos 2007) as well as in cardiovascular disease (Münzel et al. 2016). Recently, it has been suggested that markers of obesity, such as waist circumference and body mass index (BMI), may be associated with exposure to transportation noise in adults (Christensen et al. 2015a, 2015b; Eriksson et al. 2014; Oftedal et al. 2015; Pyko et al. 2015), but the evidence is not wholly consistent. Only one of the studies focusing on road traffic noise was longitudinal and used self-reported data on weight and waist circumference, which are prone to bias. In addition, there are studies on obesity in relation to residence near major roads (Li et al. 2016) and air pollution exposure (Jerrett et al. 2014), which indicates that it is important to assess both air pollution and noise exposure to elucidate causal associations when exposure from road traffic is focused.

It has been hypothesized that the relation between environmental noise and cardiovascular disease may involve sleep disturbances and psychological stress (Münzel et al. 2016) and the same mechanisms may also be relevant for metabolic diseases such as obesity and type 2 diabetes. Sleep deprivation may lead to dysregulation of appetite-regulating hormones, such as leptin and ghrelin, and contribute to overweight (Chaput et al. 2007; Van Cauter et al. 2008). Furthermore, noise may act as a stressor and lead to the elevation of cortisol levels, thereby promoting central fat deposition and impaired glucose regulation (Björntorp 1997; Rosmond 2003). For example, it has been shown that subjects living near airports have elevated saliva cortisol levels related to noise exposure (Selander et al. 2009a). Combined exposure to several stressors, such as different noise sources or work stress may be particularly harmful (Pyko et al. 2015; Selander et al. 2013; Tétreault et al. 2013). However, data on interactions between exposure to several stressors in relation to development of obesity are limited as well as on the combined effects of noise and air pollution exposure.

In a previous publication we reported results based on a cross-sectional analysis of transportation noise exposure and markers of obesity in a cohort from Stockholm County (Pyko et al. 2015). The present study is based on the same study population, but has a longitudinal design, and uses a newly developed methodology enabling more precise assessment of long-term exposure to transportation noise from different sources as well as objective outcome data. The aim was to estimate the association between exposure to transportation noise and development of obesity markers. As a secondary objective, we assessed the role of combined exposure to multiple sources of transportation noise, including road traffic, railways, and aircraft as well as interactions with air pollution exposure.

Methods

The present study was based on the Stockholm Diabetes Prevention Program cohort, which has been described in detail previously (Eriksson et al. 2008). Briefly, the program was conducted between 1992 and 2006 in Stockholm County with the primary aim to study risk factors for type 2 diabetes as well as to implement and evaluate methods for prevention. Participants were recruited between 1992 and 1998 from five municipalities in Stockholm County (Upplands Bro, Upplands Väsby, Sigtuna, Värmdö, and Tyresö). These municipalities mainly include suburban and semirural areas.

By an enrichment procedure in the original design of the study, approximately half of the participants had a family history of diabetes (52%), that is, at least one first-degree relative (mother, father, or sibling) or two second-degree relatives (grandparent, uncle, or aunt) with diabetes. Those with family history were matched on age and sex with individuals who did not have a family history of diabetes. A total of 3,162 (69.8%) men and 4,946 (70.3%) women accepted the invitation. After the baseline survey and medical examination 34 men and 125 women were excluded due to diabetes diagnosis or other medical reasons. Thus, 7,949 participants 35–56 y of age formed the diabetes-free baseline sample. Some members of the baseline sample died or moved out of Stockholm County during follow-up (n=838), and all remaining 7,111 participants were invited to a new survey 8 to 10 y after the baseline survey (see Figure S1). A total of 5,712 persons (corresponding to 80% of those invited) took part in the follow-up survey and medical examination.

At both baseline and follow-up investigations, participants filled out questionnaires and trained nurses measured weight, height, and waist circumference. The questionnaires covered health status as well as lifestyle habits such as smoking, alcohol intake, and physical activity during leisure time, dietary habits, psychological distress, shift work, insomnia, and job strain. Moreover, the follow-up questionnaire enquired about noise annoyance and noise sensitivity.

We obtained information on residential address history for the participants from the Swedish Population Register through the Swedish Taxation Authority. The residential address history included information on each address where the participants had lived starting from 1990, with precise times of address changes during follow-up.

The study was conducted in accordance with the Helsinki Declaration and approved by the Regional Ethics Review Board at Karolinska Institutet. All participants gave informed consent to the original study which also applied to the present analyses.

Noise Exposure Assessment

To assess long-term individual road transportation noise exposure, a noise database was constructed for Stockholm County to represent the period 1990–2010. The database contains information from several national, regional, and local authorities and includes 3-dimensional terrain data as well as information on ground surface, road net, daily traffic flows, speed limits, and percentage of heavy vehicles. Data were available on the road traffic situation every fifth year (1990, 1995, 2000, 2005, and 2010). To calculate noise levels, we developed and used a modification of the Nordic prediction method for road traffic (Bendtsen 1999; Nielsen 1997). The Nordic method uses traffic flow, distribution of heavy and light vehicles, speed, and pavement type to calculate the noise emission from each road link. The noise level at a receiving point is then calculated by summing the contribution of every road link using a propagation correction based on distance between road and receiver, presence of screening objects such as noise barriers or buildings, terrain shape, and whether the ground is acoustically soft or hard. Meteorological effects are included to some extent, but vegetation is not considered other than as soft ground. We modified the Nordic method for dense urban areas where possible reflection and shielding are taken into account by a ground space index based on building density (Salomons and Pont 2012). Thus, instead of the detailed information on buildings a typical reflection and shielding scenario based on building density was applied. The higher building density and the longer the distance from a source to a receiver, the higher the probability of both reflections and screening by the buildings. This modification led to decreased computational time as well as to avoiding geometrical errors such as the receiver positions being placed inside buildings instead of being exactly on the façade. Our method was validated against the full Nordic prediction method modeled with SoundPLAN (version 6.3; SoundPLAN GmbH) and showed coherent estimates. A more detailed description of the simplified noise modeling methodology instead of using detailed information on buildings is provided elsewhere (Ögren and Barregard 2016). Information on road traffic noise barriers was not included because of insufficient data on year of construction.

Using the residential address history, we estimated noise levels due to road traffic at the most exposed façade for all relevant addresses. The equivalent continuous A-weighted sound pressure level (LAeq) was calculated, and assuming a 24-h distribution of road traffic as 75%/20%/5% for day, evening, and night, respectively, we expressed noise levels as Lden, which corresponds to the equivalent level with a penalty of 3.4 dB considering noise during evening (+5 dB) and night (+10 dB) (Murphy and King 2010). The modeled noise levels were interpolated between the years with road traffic noise estimates. For each participant, we calculated the time-weighted average noise exposure during follow-up taking into account all addresses in Stockholm County where the subject had lived and considering the duration of residence at each address.

For railway noise, we used parts of the same database (3-dimensional terrain data, ground surface, residential history) as for road traffic noise supplemented with relevant information on the railway net, such as speed limits, train counts, and train types, as well as the exact 24-h distribution for different parts of the railway net for the years 1990–2012. Information on railway noise barriers was not taken into account because we generally lacked data on year of construction. As for road traffic, we applied the typical reflection and shielding scenario based on building density instead of detailed information on buildings. Railway noise levels were expressed as Lden with time-weighted exposure during the follow-up computed in a similar way as for road traffic noise.

With assistance from Swedavia, which operates the two main airports (Arlanda and Bromma) in Stockholm County, we obtained noise contour data of annual levels around the airports for the years 1995, 2000, 2005, and 2010. We assumed the same noise level in 1990 as in 1995 because there were no major structural changes in the airports between these years. The noise contour data ranged from 45 dB Lden near Arlanda and from 40 dB Lden near Bromma and were combined with the residential address history data. The annual noise levels at each address point were interpolated between the years with data during the follow-up period. Time-weighted exposure to aircraft noise was computed in relation to residential time at each address.

Outcome Definitions

All measurements of weight and height as well as of waist circumference were performed by trained nurses according to a standard protocol during the baseline and follow-up investigations. Height and weight were measured with the participant standing without shoes and were rounded to the nearest 0.5 cm or 100 g, respectively. Waist circumference was measured with the participant in lying face up, midway between the lower costal margin and iliac crest. Anthropometric markers of obesity were defined according to the WHO criteria for the European population. BMI was calculated as the weight divided by the squared height (kilograms per meter squared) with a cutoff at ≥25 to define overweight. Sex-specific cutoff values for central obesity were applied for waist circumference: ≥88 cm for women and ≥102 cm for men (WHO 2008). We used weight gain in the analyses based on continuous outcomes because it is more easily interpretable than changes in BMI as well as for comparability with previous evidence.

Men and women were investigated during partly different time periods, leading to differences in follow-up time with means of 10.2 and 8.0 y, respectively. Moreover, the follow-up time varied from 6.1 to 11.0 y between individual participants. Therefore, the change in weight and waist circumference was calculated by dividing weight gain and waist circumference increase from baseline to follow-up with the individual follow-up time in years (kilograms per year and centimeters per year, respectively).

Covariates

The covariates evaluated as confounders were identified based on a literature search and by development of a directed acyclic graph (DAG) with DAGitty.net software (see Figure S2) (Textor et al. 2011). We used the DAG to select a set of confounders for assessment of the direct effect of transportation noise on the development of obesity. Information on age, sex, physical activity, dietary habits, psychological distress, family history of diabetes, occupational status, shift work, educational level, marital status, alcohol consumption, smoking status, sleep disturbance, dietary habits, psychological distress, and job strain was obtained from the baseline questionnaire. Data on noise sensitivity and annoyance by transportation noise were obtained from the follow-up questionnaire. Furthermore, information on household mean income in small geographical areas with an average population of 1,000–2,000 subjects was obtained from registers held by Statistics Sweden to adjust for contextual confounding.

Dietary habits were assessed using “recommended” and “non-recommended” food scores (Pyko et al. 2015). In recommended foods we included consumption of low-fat dairy products, whole-meal or hard bread, fruits, vegetables (score +1 if consumed at least two to three times per week), and porridge (+1 if consumed at least one to three times per month). Among the non-recommended foods we included consumption of high-fat dairy products, white bread (score +1 if consumed at least two to three times per week), fast foods (+1 if consumed at least one to three times per month), cakes and sweets (+1 if consumed at least once a week). Summarized, the two scores for recommended and non-recommended foods ranged from 0 to 15 and 0 to 16, respectively. We considered more than 8 of 15 in recommended food score as a healthy diet indicator and more than 8 of 16 in non-recommended food score as an unhealthy diet indicator.

We assessed job strain based on the Swedish version of the Karasek and Theorell demand–decision latitude questionnaire (Agardh et al. 2003). Baseline indices for work-related demands and decision latitude were created and categorized in tertiles, and the highest tertile of demand together with the lowest tertile of decision latitude defined job strain. An index was also created for psychological distress that was assessed from baseline questions on anxiety, apathy, depression, fatigue, and insomnia (Eriksson et al. 2008).

Based on the individual residential history, we calculated time-weighted exposure to nitrogen oxides (NOx) from road traffic for each participant during follow-up using a dispersion modeling methodology developed to assess long-term source-specific exposure in Stockholm County (Bellander et al. 2001; Gruzieva et al. 2012).

Further covariates used in the adjusted models included physical activity during leisure time (sedentary: exercise less than 2 h per week/moderate: exercise at least 2 h per week/regular: exercise at least 30 min one to two times per week/frequent regular: exercise at least 30 min three times or more per week), alcohol consumption (daily/weekly/seldom/never), education (primary school/upper secondary school/university), smoking status (never/former/current), employment status (gainfully employed/unemployed/retired), psychological distress (yes/no), job strain (yes/no), and shift work (yes/no).

Statistical Methods

We tested differences in background characteristics related to road traffic noise exposure with Pearson chi-squared tests for categorical variables and one-way ANOVA for continuous variables. Pearson correlations were used to describe relationships between different transportation noise sources and road traffic–related NOx.

Linear regression models were used to analyze associations between transportation noise exposure and weight and waist circumference changes with the estimation of regression coefficients and 95% confidence intervals (CIs). Homoscedastic variance was checked by residual plots, and normality was assessed by normal probability plots of the residuals. The analyses were performed for continuous transportation noise exposures, and associations are presented for an increment of 10 dB Lden. To examine associations between transportation noise exposure and incidence of central obesity as well as overweight, we used Poisson regression models estimating incidence rate ratios (IRRs) and 95% CIs. In each analysis, those with the outcome at baseline were excluded. We approximated person-time with the length of follow-up and this was specified as an offset parameter in the model.

Additionally, we tested the assumption of linearity between transportation noise and weight as well as waist circumference changes. First, we performed analyses with a categorical exposure variable (<45, 45–49, 50–54, and ≥55 dB Lden) by inserting it in the linear model. Second, we performed restricted cubic splines analyses with 3 knots determined by Harell’s method (knots placed at the 10th, 50, and 90 percentiles). Also, we assessed the effect of combined exposure to multiple noise sources by creating dummy variables, indicating subjects exposed to none, one, two, or three transportation noise sources ≥45 dB Lden. We performed Cuzick nonparametric trend tests for the ranks across exposure groups to estimate p-values for trend.

Results are mainly presented based on two adjustment models. First, a crude model is used with adjustment for only sex and age (35/40/45/50/55 y of age). Second, a fully adjusted model is presented with additional adjustment for dietary habits, physical activity during leisure time, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work.

Effect modification of the association between road traffic noise and development of obesity measures by characteristics from baseline (sex, education, BMI, and waist circumference vs. railway noise and diabetes heredity) as well as follow-up (age, noise annoyance, noise sensitivity, and air pollution) were evaluated by introducing interaction terms into the models and by using F-test statistics.

We performed sensitivity analyses to investigate how the association between road traffic noise exposure and the IRR for central obesity was affected by restriction of the population to those who did not change their residential address, those were exposed to only road traffic noise, or those without diabetes heredity. Moreover, results of additional adjustment for other transportation noise sources, baseline waist circumference, municipality, and contextual confounding (area based mean income) are also presented. In particular, the effects were evaluated of additional adjustments for air pollution from local traffic using NOx as the indicator.

Hypothesis testing for all analyses was based on two-tailed rejection regions and p-values less than 5% were considered statistically significant, except for the interaction terms, where 10% was used as significance level. Statistical analyses were performed using Stata/SE (version 13.1; StataCorp, College Station, TX); and spatial manipulation of data and exposure assessment were performed in QGIS (version 2.10.1; QGIS Development Team).

Results

Of the 5,712 persons participating in the follow-up survey 82 (1.4%) were excluded because of missing exposure data; 94 (1.6%) because of missing data on anthropometric variables at baseline or follow-up; and 352 (6.1%) because of missing data on the covariates included in the main model. Following these exclusions, a study population of 5,184 (91%) individuals remained with complete information and a mean follow-up time of 8.9 y.

Of those included in the study 2,739 (53%) were not exposed to a noise level of ≥45 dB Lden from any transportation noise source; 1,901 (37%) participants were exposed to one of three transportation noise sources at this level or higher; 487 (9%) to two sources of transportation noise, and 57 (1%) to all three sources ≥45 dB Lden (see Figure S3).

Women reported road traffic noise exposure ≥45 dB Lden more often than men because some of them were recruited from one municipality (Upplands-Väsby) that was not included in the recruitment of men (Table 1). Furthermore, participants with higher noise exposure had lower education and socioeconomic status, were less physically active during leisure time, had more job strain and psychological distress, and reported that they were less noise sensitive. Furthermore, those exposed were more annoyed by road traffic noise and more often exposed to railway and aircraft noise.

Table 1. Characteristics of the study cohort from Stockholm County in relation to road traffic noise exposure during follow-up [n (%)].
Individual characteristicsa Time-weighted average road traffic exposure
<45 dB Lden(n=3,457) ≥45 dB Lden(n=1,727) p-Value
Women 1,914 (55) 1,069 (62) 0.007
Age (y) 0.050
 35–39 333 (10) 141 (8)
 40–44 659 (19) 316 (18)
 45–49 1,196 (35) 573 (33)
 50–55 1,269 (37) 697 (40)
Socioeconomic status 0.019
 Low 889 (26) 483 (28)
 Medium 704 (20) 390 (23)
 High 1,609 (47) 746 (43)
 Other 166 (5) 67 (4)
Occupational status 0.185
 Gainfully employed 3,162 (91) 1,573 (91)
 Unemployed 217 (6) 101 (6)
 Retired 78 (2) 53 (3)
Shift work 329 (10) 181 (10) 0.272
Smoking status 0.275
 Current 812 (23) 439 (25)
 Former 1,274 (37) 631 (37)
 Never 1,371 (40) 657 (38)
Physical activity during leisure timeb 0.012
 Sedentary 329 (10) 206 (12)
 Moderate 1,853 (54) 909 (53)
 Regular 986 (29) 497 (29)
 Frequent regular 289 (8) 115 (7)
Alcohol consumption 0.336
 Daily 153 (4) 73 (4)
 Weekly 2,241 (65) 1,131 (65)
 Seldom 949 (27) 451 (26)
 Never 114 (3) 72 (4)
Education level 0.021
 Primary school 1,017 (29) 565 (33)
 Secondary school 1,340 (39) 666 (39)
 University degree or higher 1,100 (32) 496 (29)
Job strainc 381 (11) 221 (13) 0.060
Psychological distressd 696 (20) 410 (24) 0.003
Noise sensitivitye 0.036
 Less sensitive than others 637 (18) 357 (21)
 Same sensitivity as others 2,441 (71) 1,210 (70)
 More sensitive that others 378 (11) 158 (9)
Noise annoyance from road traffice <0.001
 Seldom/never 3,191 (92) 1,263 (73)
 Few times per month 134 (4) 187 (11)
 Few times per week 79 (2) 149 (9)
 Each day 46 (1) 123 (7)
Healthy dietf 318 (9) 175 (10) 0.280
Unhealthy dietf 2,167 (63) 1,030 (60) 0.034
Diabetes heredityg 1,776 (51) 924 (54) 0.148
Railway noise over 45 dB Ldenh 213 (6) 183 (11) <0.001
Aircraft noise over 45 dB Lden h 578 (17) 345 (20) 0.004

aCharacteristics are from the baseline investigation unless stated otherwise. Number of participants in each group, percentages in parenthesis and p-values are reported.

bPhysical activity during leisure time is defined as sedentary (regular exercise less than 2 h per week), moderate (regular exercise at least 2 h per week), regular (regular exercise at least 30 min one to two times per week), frequent regular (at least 30 min three times or more per week).

cJob strain is defined as a combination of the highest tertile of demand together with the lowest tertile of decision latitude at work.

dPsychological distress is assessed as the highest quartile of a summed index based on questions on anxiety, apathy, depression, fatigue, and insomnia.

eFrom follow-up investigation.

fMore than 8 of 15 in recommended food score and 8 of 16 in non-recommended food score (see “Methods” section).

gFamily history of diabetes defined if participants had at least one first-degree relative (parent or sibling) with diabetes or at least two second-degree relatives (grandparents, aunts, uncles) with diabetes.

hTime-weighted average during follow-up period.

The mean follow-up time was 8.0 y in women and 10.2 y in men and mean weight gain was 0.28 kg/y (SD 0.72) among women and 0.32 kg/y (SD 0.56) among men (see Table S1). Mean waist circumference increase differed among women and men and was 0.64 cm/y (SD 0.80) and 0.33 cm/y (SD 0.64), respectively. Excluding those with overweight at baseline, 2,560 (49%), the cumulative incidence of overweight in the remaining 2,624 participants during follow-up was 25% and 36% for women and men, respectively. In contrast, the cumulative incidence of central obesity in those 4,386 without central obesity at baseline was 23% and 16% in women and men, respectively.

Table 2 presents associations between transportation noise exposures from different sources and continuous outcomes. In the fully adjusted model, we observed an association between road traffic noise and waist circumference increase of 0.04 cm/y (95% CI: 0.02, 0.06) per 10 dB Lden. For aircraft noise, the waist circumference increase was 0.16 cm/y (95% CI: 0.14, 0.17) per 10 dB Lden. No clear association was observed between railway noise and waist circumference increase. Weight gain was associated with aircraft noise, and changed 0.03 kg/y (95% CI: 0.01, 0.04) per 10 dB Lden, but not with road or railway traffic noise exposure.

Table 2. Noise exposure from different transportation noise sources during follow-up for the study cohort from Stockholm County in relation to waist circumference increase and weight gain.
Exposurea No. total Waist circumference increase (cm/y) Weight gain (kg/y)
Model 1b β (95% CI) Model 2c β (95% CI) Model 1b β (95% CI) Model 2c β (95% CI)
Road traffic noise
 Continuous per 10 dB Lden 0.04 (0.02, 0.07) 0.04 (0.02, 0.06) 0.01 (−0.006, 0.03) 0.01 (−0.009, 0.03)
Categorical, dB Lden
 <45 3,457 0 Referent 0 Referent 0 Referent 0 Referent
 45–49 958 −0.04 (−0.09, 0.01) −0.05 (−0.10, 0.003) −0.02 (−0.07, 0.03) −0.02 (−0.07, 0.02)
 50–54 565 0.13 (0.06, 0.19) 0.12 (0.06, 0.18) 0.04 (−0.02, 0.10) 0.03 (−0.03, 0.09)
 ≥55 204 0.15 (0.04, 0.25) 0.14 (0.04, 0.25) 0.03 (−0.06, 0.12) 0.03 (−0.07, 0.12)
 Trend p-value <0.001 <0.001 0.337 0.446
Railway noise exposure
 Continuous per 10 dB Lden 0.02 (−0.008, 0.04) 0.01 (−0.01, 0.03) 0.02 (−0.003, 0.04) 0.01 (−0.005, 0.04)
Categorical, dB Lden
 <45 4,788 0 Referent 0 Referent 0 Referent 0 Referent
 45–49 161 0.02 (−0.09, 0.14) 0.01 (−0.10, 0.13) 0.01 (−0.11, 0.10) −0.01 (−0.11, 0.09)
 50–54 125 0.07 (−0.06, 0.20) 0.07 (−0.06, 0.20) 0.10 (−0.02, 0.21) 0.09 (−0.03, 0.20)
 ≥55 110 −0.04 (−0.18, 0.10) −0.05 (−0.19, 0.09) −0.01 (−0.13, 0.12) −0.02 (−0.14, 0.11)
 Trend p-value 0.781 0.967 0.448 0.598
Aircraft noise
 Continuous per 10 dB Lden 0.16 (0.14, 0.17) 0.16 (0.14, 0.17) 0.03 (0.01, 0.04) 0.03 (0.01, 0.04)
Categorical, dB Lden
 <45 4,261 0 Referent 0 Referent 0 Referent 0 Referent
 45–49 126 0.32 (0.19, 0.44) 0.31 (0.18, 0.44) 0.12 (0.00, 0.23) 0.11 (−0.008, 0.22)
 50–54 590 0.45 (0.39, 0.51) 0.44 (0.38, 0.50) 0.06 (0.008, 0.12) 0.06 (0.006, 0.12)
 ≥55 207 0.49 (0.39, 0.59) 0.48 (0.39, 0.58) 0.09 (−0.003, 0.18) 0.09 (−0.005, 0.18)
 Trend p-value <0.001 <0.001 0.003 0.004

aTime-weighted noise exposure expressed as Lden taking into account all addresses where the subject had lived during the follow-up period.

bResults of linear regression model adjusted only for sex and age.

cResults of linear regression model adjusted for sex, age, dietary habits, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work.

Both categorical and restricted cubic splines analyses suggested nonlinearity in the association between road traffic noise exposure and waist circumference increase (Table 2, Figure 1). It is suggested that there might be a threshold in the exposure–response relation at around 45–50 dB Lden. However, departure from linearity was not statistically significant (p-value of departure = 0.099). No corresponding threshold in the exposure–response curve was suggested for aircraft noise.

Figures 1A and 1B are line graphs along with bars plotting waist circumference increase (centimeters per year) (y-axis) across time-weighted exposure to noise (decibels L subscript den) (x-axis) from road traffic and aircraft, respectively.

Figure 1. Waist circumference increase (centimeters per year) in the study cohort from Stockholm County in relation to time-weighted exposure to noise from road traffic (A) and aircraft (B) during follow-up based on restricted cubic spline analyses (n=5,184). Note: Increase of waist circumference (bold central line) and 95% CI (dashed outer bands) in models adjusted for sex, age, dietary habits, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work. Bars indicate number of subjects in different exposure groups.

There were positive trends in incidence of central obesity in relation to aircraft and road traffic noise exposure, with IRRs of 1.19 (95% CI: 1.14, 1.24) and 1.07 (95% CI: 1.00, 1.14) per 10 dB Lden, respectively (Table 3). Aircraft noise was also associated with an increased risk of overweight, showing an IRR of 1.06 (95% CI: 1.01, 1.12) per 10 dB Lden. In sex-specific analyses, clear associations between road traffic or railway noise exposure and central obesity were only seen in women, whereas trends were apparent in both sexes for aircraft noise (see Table S2). For overweight, a statistically significant trend was only seen in women in relation to aircraft noise exposure. Just as for waist circumference, there seemed to be an increased IRR for central obesity primarily above 45–50 dB Lden in relation to road traffic noise exposure, whereas for aircraft noise the trend appears to extend to even lower levels (Table 3, Figure 2).

Figures 2A and 2B are line graphs along with bars plotting annual IRR of central obesity (left y-axis) and subjects (right y-axis) across time-weighted exposure to noise (decibels L subscript den) (x-axis) from road traffic and aircraft, respectively.

Figure 2. Incidence rate ratio for central obesity in the study cohort from Stockholm County in relation to noise exposure from road traffic (A) and aircraft (B) in the fully adjusted model based on restricted cubic spline analyses (n=4,386). Note: IRR of central obesity (bold central line) and 95% CI (dashed outer bands) in model adjusted for sex, age, dietary habits, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work, using 45 dB Lden as reference level. Bars indicate number of subjects in different exposure groups.

Table 3. Risks of central obesity and overweight in a cohort from Stockholm County in relation to transportation noise exposure from different sources.
Exposure Central obesitya Overweightb
No. of subjects/cases Model 1cIRR (95% CI) Model 2dIRR (95% CI) No. of subjects/cases Model 1cIRR (95% CI) Model 2dIRR (95% CI)
Road traffic noisee
 Continuous per 10 dB Lden 4,386/872 1.09 (1.02, 1.16) 1.07 (1.00, 1.14) 2,624/760 1.00 (0.94, 1.07) 0.99 (0.92, 1.05)
Categorical, dB Lden
 <45 2,932/548 1.00 Referent 1.00 Referent 1,784/522 1.00 Referent 1.00 Referent
 45–49 796/154 1.03 (0.88, 1.21) 1.00 (0.85, 1.17) 461/130 0.97 (0.83, 1.14) 0.96 (0.82, 1.13)
 50–54 479/124 1.35 (1.14, 1.60) 1.33 (1.12, 1.58) 276/77 1.02 (0.83, 1.24) 0.99 (0.81, 1.21)
 ≥55 179/46 1.33 (1.02, 1.72) 1.26 (0.96, 1.64) 103/31 1.06 (0.78, 1.43) 1.04 (0.77, 1.39)
 Trend p-value <0.001 0.003 0.837 0.937
Railway noisee
 Continuous per 10 dB Lden 4,386/872 1.07 (1.00, 1.13) 1.05 (0.99, 1.12) 2,624/760 1.04 (0.97, 1.11) 1.03 (0.96, 1.09)
Categorical, dB Lden
 <45 4,057/791 1.00 Referent 1.00 Referent 2,423/702 1.00 Referent 1.00 Referent
 45–49 128/25 0.98 (0.69, 1.40) 0.97 (0.68, 1.38) 77/22 1.08 (0.75, 1.54) 1.04 (0.73, 1.49)
 50–54 110/33 1.48 (1.10, 1.99) 1.43 (1.07, 1.92) 65/17 0.95 (0.63, 1.44) 0.91 (0.61, 1.36)
 ≥55 91/23 1.31 (0.92, 1.87) 1.27 (0.88, 1.81) 59/19 1.09 (0.74, 1.59) 1.04 (0.71, 1.52)
 Trend p-value 0.011 0.025 0.751 0.996
Aircraft traffic noisee
 Continuous per 10 dB Lden 4,386/872 1.20 (1.15, 1.26) 1.19 (1.14, 1.24) 2,624/760 1.06 (1.01, 1.11) 1.06 (1.01, 1.12)
Categorical, dB Lden
 <45 3,590/647 1.00 Referent 1.00 Referent 2,200/620 1.00 Referent 1.00 Referent
 45–49 103/22 1.31 (0.90, 1.92) 1.27 (0.87, 1.85) 55/19 1.17 (0.82, 1.67) 1.10 (0.76, 1.58)
 50–54 508/145 1.69 (1.45, 1.97) 1.62 (1.39, 1.89) 277/90 1.18 (0.99, 1.41) 1.20 (1.00, 1.44)
 ≥55 185/58 2.04 (1.64, 2.56) 1.99 (1.58, 2.50) 92/31 1.16 (0.86, 1.56) 1.15 (0.85, 1.55)
 Trend p-value <0.001 <0.001 0.047 0.045

aGender-specific cutoff values for central obesity were applied for waist circumference: ≥88 cm for women and ≥102 cm for men. Subjects with central obesity at baseline were excluded from analysis.

bCutoff of BMI at ≥25 kg/m2 to define overweight. Subjects with overweight at baseline were excluded from analysis.

cResults of Poisson regression model adjusted only for sex and age.

dResults of Poisson regression model adjusted for sex, age, dietary habits, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work.

eTime-weighted noise exposures expressed as Lden taking into account all addresses where the subject had lived during follow-up period.

We observed an exposure–response relation between the number of transportation noise sources and risk of central obesity (Figure 3). The IRR increased from 1.22 (95% CI: 1.08, 1.39) among those exposed to only one source to 2.26 (95% CI: 1.55, 3.29) among those exposed to all three transportation noise sources (p-value for trend <0.001). No corresponding trend was seen for overweight. For waist circumference a positive trend was observed with a change of 0.14 cm/y (95% CI: 0.09, 0.18) among those exposed to only one source and of 0.48 cm/y (95% CI: 0.29, 0.67) among those exposed to all three transportation noise sources (p-value for trend <0.001) (see Figure S4). On the other hand, no trend was seen in relation to weight gain.

Forest plot shows IRR with 95 percent confidence interval for central obesity and overweight in relation to no traffic noise exposure, one noise source, two noise sources, and three noise sources, all greater than or equal to 45 decibels L subscript den.

Figure 3. Incidence rate ratio (IRR) of central obesity (•) and overweight (⧫) in the study cohort from Stockholm County in relation to noise exposure ≥45 dB Lden from road traffic, railways, and/or aircraft. Note: IRRs are adjusted for sex, age, dietary habits, alcohol consumption, education level, physical activity, smoking status, psychological distress, job strain, and shift work.

The association between exposure to road traffic noise and waist circumference increase was modified by age, with a higher waist circumference increase among those younger than 60 y at follow-up (see Table S3). No other characteristics showed an interaction with road traffic noise exposure. In particular, there was no apparent effect modification by sex as was seen for central obesity (see Table S2). The association between road traffic noise and weight gain also appeared stronger in the younger age group. The effect modification for aircraft noise exposure generally showed the same pattern as for road traffic noise (data not shown).

In sensitivity analyses, we first explored how the results for central obesity related to exposure to road traffic noise were affected by different restrictions and adjustments (see Figure S5). For exposures of ≥45 dB Lden there was an adjusted IRR per 10 dB Lden of 1.22 (95% CI: 1.06, 1.42). The IRR appeared to be little affected by additional adjustment for railway noise, aircraft noise, baseline waist circumference, or contextual confounding (using household mean income in small areas) or following exclusion of those with exposure to railway or aircraft noise ≥45 dB Lden, address change during follow-up or diabetes heredity. However, a lower IRR was suggested after adjustment for municipality. Second, we explored how results were affected by additional adjustment for local air pollution from road traffic using NOx as indicator. No marked influence by adjustment for NOx was seen on the results for different transportation noise sources using waist circumference and weight as continuous outcomes (see Table S4). However, the results on IRR for central obesity related to road traffic noise were affected by adjustment for NOx (see Table S5). On the other hand, results for overweight or for aircraft or railway noise were not influenced. NOx was moderately related to road traffic noise (r=0.56) but not to aircraft or railway noise (r=−0.02 and 0.14, respectively), which contributes to explaining the consequences of NOx adjustment on the associations with obesity markers for different transportation noise sources.

Discussion

This cohort study showed an association between exposure to road traffic noise as well as aircraft noise and waist circumference increase. No corresponding association was observed for railway noise. There were positive trends in risk for central obesity related to each of the sources of transportation noise with a particularly high risk in those exposed to all three sources. The excess risk for central obesity primarily occurred at road traffic noise levels of around 50 dB Lden and higher, whereas the excess risk related to aircraft noise exposure seemed to occur even at lower levels.

There is growing evidence that transportation noise affects adiposity markers. The first study to investigate this relationship focused on aircraft noise and showed an association particularly for waist circumference (Eriksson et al. 2014). A Norwegian cross-sectional study found a positive association between road traffic noise and BMI only among some subgroups (Oftedal et al. 2015). However, based on a longitudinal study a Danish group reported associations between residential exposure to road traffic or railway noise and waist circumference as well as weight changes in adults (Christensen et al. 2015a), confirming earlier cross-sectional evidence in the same cohort (Christensen et al. 2015b). Our observed waist circumference increase related to road traffic (0.02–0.06 cm/y per 10 dB Lden) corresponds well to the findings in the Danish cohort (0.02–0.12 cm/y per 10 dB Lden) (Christensen et al. 2015a). A limitation with the Danish study is that the anthropometric data at follow-up were based on self-reports, which are prone to bias. We already published results on transportation noise exposure and obesity markers based on cross-sectional analyses of the same cohort as in present study (Pyko et al. 2015). This new longitudinal analysis used a much more detailed methodology for assessment of noise exposure and showed associations primarily for waist circumference increase in relation to road traffic or aircraft noise exposure as well as for weight gain in relation to aircraft noise exposure. Overall, the evidence is not fully consistent regarding a role for transportation noise in the development of central obesity (increased waist circumference or waist–hip ratio) or general adiposity (weight gain or BMI). Stress mechanisms as mediated by cortisol excretion would be expected to primarily result in central obesity, although noise-induced sleep disturbances and behavioral changes might also mediate effects on general adiposity (Zaharna and Guilleminault 2010). Elucidation of noise effects on specific adiposity markers may provide evidence on causal mechanisms.

In our study, the waist circumference increase per unit of noise exposure was highest for aircraft noise. The effect appeared lower for road traffic noise, and there was no clear association for railway noise. This pattern is well in line with the effect of noise exposure on annoyance and sleep disturbances (Miedema and Oudshoorn 2001; Miedema and Vos 2007) where aircraft noise causes more pronounced effects than road traffic noise at the same noise levels, and railway noise is less harmful than road traffic noise (WHO 2009). Furthermore, our data suggest a threshold in the effects by road traffic noise on waist circumference and central obesity at around 45–50 dB Lden, which was not apparent for aircraft noise. Aircraft and road traffic noise are qualitatively different, for example, aircraft noise is transient, more intense in a short period, and usually causes a higher arousal level for areas that are directly under the flight paths, which may contribute to the stronger effects. Furthermore, the exposure assessment approaches to both sources of noise were different in our study. There is a great need for further longitudinal evidence on the association between transportation noise from different sources and obesity.

The risk of central obesity related to road traffic noise exposure ≥45 dB Lden was mostly unaffected by different adjustments and restrictions. However, the excess risk was no longer statistically significant following adjustment for air pollution from road traffic, using NOx as marker. On the other hand, no major effect was seen following adjustment for NOx in analyses of road traffic noise and waist circumference increase. We observed a correlation of 0.56 between exposure to noise and air pollution from road traffic in individuals, which is similar to the correlation in another epidemiological study from Stockholm County (Selander et al. 2009b). Other studies on road traffic noise and obesity have generally not found major changes in associations following adjustment for air pollution (Christensen et al. 2015a, 2015b, 2016; Oftedal et al. 2015). It cannot be excluded that air pollution exposure also contributed to the development of obesity in our study population, although we did not find evidence of any interaction between the two exposures. Furthermore, the association between road traffic noise exposure and central obesity appeared somewhat weaker following adjustment for municipality. However, we did not see any marked effect by adjustment for a large number of individual characteristics or contextual confounding. To generally adjust for municipality is not meaningful in our study because aircraft noise primarily affected only two municipalities and due to risks of overadjustment for road traffic noise, it would not have made optimal use of the full range of exposure in our study area given the quite low exposures overall. However, we cannot exclude that residual confounding may have affected some of our results.

We saw clear exposure–response associations related to number of noise sources for both risk of central obesity risks and waist circumference increase. This goes in line with previous studies of sleep effects and annoyance in relation to combined exposure to different noise sources (Griefahn et al. 2006; Miedema 2004). Our findings speak in favor of the multiple environmental stressors theory, where several stressors may enhance the effect of each other (Stansfeld and Matheson 2003). Moreover, interactions have been observed between traffic and occupational noise as well as job strain in relation to the risk of myocardial infarction (Selander et al. 2013). Unfortunately, we did not have any data on stress markers for our study subjects, such as saliva cortisol levels. It is important to further investigate interactions between different environmental stressors including noise for both cardiovascular and metabolic outcomes.

We did not observe significant interactions between exposure to road traffic noise and other risk factors in relation to waist circumference increase, except for age. Our results showed that the association between waist circumference increase and road traffic noise were mainly driven by the age group below 60 y. This is congruent with some noise studies on hypertension (Bodin et al. 2009; De Kluizenaar et al. 2007) but opposite to studies focused on stroke and type 2 diabetes, which indicated stronger effects in those over 60 and 64 y of age, respectively (Sørensen et al. 2011, 2013). Moreover, no age interactions were apparent in other noise studies of obesity (Christensen et al. 2015a, 2015b; Oftedal et al. 2015). We did not see an interaction with BMI or waist circumference at baseline. In contrast, Danish longitudinal results showed stronger effects of noise on adiposity development in those obese or with central obesity at baseline (Christensen et al. 2015a). Furthermore, there was no clear effect modification by noise annoyance or sensitivity, which is opposite to Norwegian results with the strongest effect in noise-sensitive women (Oftedal et al. 2015). Although the literature is limited regarding the influence by these factors on the association between noise exposure and obesity, there is evidence of noise annoyance and sensitivity acting as effect modifiers of the relationship between the noise exposure and cardiovascular outcomes (Babisch et al. 2013; Eriksson et al. 2010). All in all, it is not clear if age or other risk factors modify the association between noise and adiposity.

A limitation of our study is the relatively low road traffic noise levels and the small number of highly exposed participants. Furthermore, in certain aspects, the data on noise exposures are imprecise. For example, we lack information on noise exposure other than from the three transportation sources, such as occupational exposure. Additionally, we do not have information on noise modifiers, such as façade and window insulation as well as bedroom location, open/closed windows, use of earplugs, and so forth. Moreover, by design, the study population was enriched with persons with a family history of diabetes and the results may not be generalizable to the population as a whole. However, the associations were confirmed when we restricted the analysis on road traffic and waist circumference to those without a family history of diabetes.

The strengths of the present study include the prospective design and anthropometric data measured by trained nurses at recruitment as well as at follow-up. Additionally, detailed information was available regarding potential individual confounders (socioeconomic position, diet, alcohol consumption, smoking, physical activity, and so forth) as well as area–base confounders. Nevertheless, residual confounding cannot be excluded. Furthermore, we had a detailed residential history for all participants, allowing exposure assessment for the whole follow-up period. A particular feature of our study is that a sizable proportion of the study participants was exposed to several noise sources, allowing evaluation of health effect following exposure to multiple noise sources.

In conclusion, our study showed associations between exposure to noise from road traffic or aircraft and development of central obesity. The risk appeared particularly high for aircraft noise and in those with concomitant exposure to different sources of transportation noise. These findings support the evidence linking noise to development of obesity, which is an outcome of great public health significance.

Acknowledgments

The authors wish to thank the staff and participants of the Stockholm Diabetes Prevention Program. We thank B. Julin for assistance with the assessment of dietary habits and A. Hilding for valuable comments regarding the manuscript.

The study was supported by grants from the Swedish Research Council for Health, Working Life and Welfare, the Swedish Heart and Lung Foundation, the Stockholm County Council, the Swedish Research Council, the Diabetes Fund of the Swedish Diabetes Association, Novo Nordisk Scandinavia, and GlaxoSmithKline. Both Novo Nordisk Scandinavia and GlaxoSmithKline, Sweden supported the SDPP (Stockholm Diabetes Prevention Program) by unconditional grants, which were used for data base handling and epidemiologic studies. However, none of these grants was used for salary of any author. The authors’freedom to design, conduct, interpret, and publish research was not compromised by any sponsor or funding agency.

References

Agardh EE, Ahlbom A, Andersson T, Efendic S, Grill V, Hallqvist J, et al. 2003. Work stress and low sense of coherence is associated with type 2 diabetes in middle-aged Swedish women. Diabetes Care 26(3):719–724, PMID: 12610028, 10.2337/diacare.26.3.719.

Babisch W, Pershagen G, Selander J, Houthuijs D, Breugelmans O, Cadum E, et al. 2013. Noise annoyance—a modifier of the association between noise level and cardiovascular health? Sci Total Environ 452-453:50–57, PMID: 23500398, 10.1016/j.scitotenv.2013.02.034.

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.

Bellander T, Berglind N, Gustavsson P, Jonson T, Nyberg F, Pershagen G, et al. 2001. Using geographic information systems to assess individual historical exposure to air pollution from traffic and house heating in Stockholm. Environ Health Perspect 109(6):633–639, PMID: 11445519, 10.1289/ehp.01109633.

Bendtsen H. 1999. The Nordic prediction method for road traffic noise. Sci Total Environ 235(1-3):331–338, 10.1016/S0048-9697(99)00216-8.

Björntorp P. 1997. Body fat distribution, insulin resistance, and metabolic diseases. Nutrition 13(9):795–803, PMID: 9290093, 10.1016/S0899-9007(97)00191-3.

Bodin T, Albin M, Ardö J, Stroh E, Östergren PO, Björk J. 2009. Road traffic noise and hypertension: results from a cross-sectional public health survey in southern Sweden. Environ Health 8:38, PMID: 19744313, 10.1186/1476-069X-8-38.

Chaput JP, Després JP, Bouchard C, Tremblay A. 2007. Short sleep duration is associated with reduced leptin levels and increased adiposity: results from the Québec Family Study. Obesity (Silver Spring) 15(1):253–261, PMID: 17228054, 10.1038/oby.2007.512.

Christensen JS, Hjortebjerg D, Raaschou-Nielsen O, Ketzel M, Sørensen TIA, Sørensen M. 2016. Pregnancy and childhood exposure to residential traffic noise and overweight at 7 years of age. Environ Int 94:170–176, PMID: 27258658, 10.1016/j.envint.2016.05.016.

Christensen JS, Raaschou-Nielsen O, Tjønneland A, Nordsborg RB, Jensen SS, Sørensen TIA, et al. 2015a. Long-term exposure to residential traffic noise and changes in body weight and waist circumference: a cohort study. Environ Res 143:154–161, PMID: 26492400, 10.1016/j.envres.2015.10.007.

Christensen JS, Raaschou-Nielsen O, Tjønneland A, Overvad K, Nordsborg RB, Ketzel M, et al. 2015b. Road traffic and railway noise exposures and adiposity in adults: a cross-sectional analysis of the Danish Diet, Cancer, and Health cohort. Environ Health Perspect 124(3):329–335. PMID: 26241990, 10.1289/ehp.1409052.

De Kluizenaar Y, Gansevoort RT, Miedema HM, de Jong PE. 2007. Hypertension and road traffic noise exposure. J Occup Environ Med 49(5):484–492, PMID: 17495691, 10.1097/JOM.0b013e318058a9ff.

Eriksson AK, Ekbom A, Granath F, Hilding A, Efendic S, Östenson CG. 2008. Psychological distress and risk of pre-diabetes and type 2 diabetes in a prospective study of Swedish middle-aged men and women. Diabet Med 25(7):834–842, PMID: 18513304, 10.1111/j.1464-5491.2008.02463.x.

Eriksson C, Bluhm G, Hilding A, Östenson C-G, Pershagen G. 2010. Aircraft noise and incidence of hypertension—gender specific effects. Environ Res 110(8):764–772, PMID: 20880521, 10.1016/j.envres.2010.09.001.

Eriksson C, Hilding A, Pyko A, Bluhm G, Pershagen G, Östenson C-G. 2014. Long-term aircraft noise exposure and body mass index, waist circumference, and type 2 diabetes: a prospective study. Environ Health Perspect 122(7):687–694, PMID: 24800763, 10.1289/ehp.1307115.

Griefahn B, Marks A, Robens S. 2006. Noise emitted from road, rail and air traffic and their effects on sleep. J Sound Vib 295(1-2):129–140, 10.1016/j.jsv.2005.12.052.

Gruzieva O, Bellander T, Eneroth K, Kull I, Melén E, Nordling E, et al. 2012. Traffic-related air pollution and development of allergic sensitization in children during the first 8 years of life. J Allergy Clin Immunol 129(1):240–246, PMID: 22104609, 10.1016/j.jaci.2011.11.001.

Jerrett M, McConnell R, Wolch J, Chang R, Lam C, Dunton G, et al. 2014. Traffic-related air pollution and obesity formation in children: a longitudinal, multilevel analysis. Environ Health 13:49, PMID: 24913018, 10.1186/1476-069X-13-49.

Li W, Dorans KS, Wilker EH, Rice MB, Schwartz J, Coull BA, et al. 2016. Residential proximity to major roadways, fine particulate matter, and adiposity: the Framingham Heart Study. Obesity (Silver Spring) 24(12):2593–2599, PMID: 27804220, 10.1002/oby.21630.

Miedema HM. 2004. Relationship between exposure to multiple noise sources and noise annoyance. J Acoust Soc Am 116(2):949–957, PMID: 15376661.

Miedema HM, Oudshoorn C. 2001. Annoyance from transportation noise: relationships with exposure metrics DNL and DENL and their confidence intervals. Environ Health Perspect 109(4):409–416, PMID: 11335190, 10.1289/ehp.01109409.

Miedema HM, Vos H. 2007. Associations between self-reported sleep disturbance and environmental noise based on reanalyses of pooled data from 24 studies. Behav Sleep Med 5(1):1–20, PMID: 17313321, 10.1207/s15402010bsm0501_1.

Münzel T, Sørensen M, Gori T, Schmidt FP, Rao X, Brook J, et al. 2016. Environmental stressors and cardio-metabolic disease: part I–epidemiologic evidence supporting a role for noise and air pollution and effects of mitigation strategies. Eur Heart J 38(8):550–556. PMID: 27460892,10.1093/eurheartj/ehw269.

Murphy E, King EA. 2010. Strategic environmental noise mapping: methodological issues concerning the implementation of the EU environmental noise directive and their policy implications. Environ Int 36(3):290–298, PMID: 20045195, 10.1016/j.envint.2009.11.006.

Nielsen HL. 1997. Road Traffic Noise: Nordic Prediction Method. Copenhagen, Denmark:Nordic Council of Ministers.

Oftedal B, Krog NH, Pyko A, Eriksson C, Graff-Iversen S, Haugen M, et al. 2015. Road traffic noise and markers of obesity – a population-based study. Environ Res 138:144–153, PMID: 25710788, 10.1016/j.envres.2015.01.011.

Ögren M, Barregard L. 2016. Road traffic noise exposure in Gothenburg 1975–2010. PLoS One 11(5):e0155328, PMID: 27171440, 10.1371/journal.pone.0155328.

Pyko A, Eriksson C, Oftedal B, Hilding A, Östenson CG, Krog NH, et al. 2015. Exposure to traffic noise and markers of obesity. Occup Environ Med 72(8):594–601, PMID: 26009579, 10.1136/oemed-2014-102516.

Rosmond R. 2003. Stress induced disturbances of the HPA axis: a pathway to type 2 diabetes? Med Sci Monit 9(2):RA35–RA39, PMID: 12601304.

Salomons EM, Pont MB. 2012. Urban traffic noise and the relation to urban density, form, and traffic elasticity. Landsc Urban Plan 108(1):2–16, 10.1016/j.landurbplan.2012.06.017.

Selander J, Bluhm G, Nilsson M, Hallqvist J, Theorell T, Willix P, et al. 2013. Joint effects of job strain and road-traffic and occupational noise on myocardial infarction. Scand J Work Environ Health 39(2):195–203, PMID: 23032870, 10.5271/sjweh.3324.

Selander J, Bluhm G, Theorell T, Pershagen G, Babisch W, Seiffert I, et al. 2009a. Saliva cortisol and exposure to aircraft noise in six European countries. Environ Health Perspect 117:1713, PMID: 20049122, 10.1289/ehp.0900933.

Selander J, Nilsson ME, Bluhm G, Rosenlund M, Lindqvist M, Nise G, et al. 2009b. Long-term exposure to road traffic noise and myocardial infarction. Epidemiology 20(2):272–279, PMID: 19116496, 10.1097/EDE.0b013e31819463bd.

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.

Sørensen M, Hvidberg M, Andersen ZJ, Nordsborg RB, Lillelund KG, Jakobsen J, et al. 2011. Road traffic noise and stroke: a prospective cohort study.Eur Heart J 32(6):737–744, PMID: 21266374, 10.1093/eurheartj/ehq466.

Stansfeld SA, Matheson MP. 2003. Noise pollution: non-auditory effects on health. Br Med Bull 68:243–257, PMID: 14757721, 10.1093/bmb/ldg033

Tétreault L-F, Perron S, Smargiassi A. 2013. Cardiovascular health, traffic-related air pollution and noise: are associations mutually confounded? A systematic review. Int J Public Health 58(5):649–666, PMID: 23887610, 10.1007/s00038-013-0489-7.

Textor J, Hardt J, Knüppel S. 2011. DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology 22(5):745, PMID: 21811114, 10.1097/EDE.0b013e318225c2be.

Van Cauter E, Spiegel K, Tasali E, Leproult R. 2008. Metabolic consequences of sleep and sleep loss. Sleep Med 9(suppl 1):S23–S28, PMID: 18929315, 10.1016/S1389-9457(08)70013-3.

WHO (World Health Organization). 2008. Waist Circumference and Waist-Hip Ratio: Report of a WHO Expert Consultation. Geneva, 8–11 December 2008. Geneva, Switzerland:WHO. http://apps.who.int/iris/bitstream/10665/44583/1/9789241501491_eng.pdf [accessed 25 October 2017].

WHO. 2009. Night Noise Guidelines for Europe. Copenhagen, Denmark:WHO Regional Office Europe.

Zaharna M, Guilleminault C. 2010. Sleep, noise and health: review. Noise Health 12(47):64, PMID: 20472951, 10.4103/1463-1741.63205.

Exposure to Perfluoroalkyl Substances and Metabolic Outcomes in Pregnant Women: Evidence from the Spanish INMA Birth Cohorts

Author Affiliations open

1ISGlobal, Centre for Research in Environmental Epidemiology (CREAL), Barcelona, Spain

2Universitat Pompeu Fabra (UPF), Barcelona, Spain

3Spanish Consortium for Research on Epidemiology and Public Health (CIBERESP), Madrid, Spain

4Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA

5Epidemiology and Environmental Health Joint Research Unit, FISABIO–Universitat de València, Valencia, Spain

6Subdirección de Salud Pública y Adicciones de Gipuzkoa, San Sebastián, Spain

7Instituto de Investigación Sanitaria Biodonostia, San Sebastián, Spain

8Institute for Occupational Medicine, Rheinisch-Westfälische Technische Hochschule (RWTH) Aachen University, Aachen, Germany

9Department of Child and Adolescent Psychiatry/Psychology, Erasmus University Medical Centre–Sophia Children’s Hospital, Rotterdam, Netherlands

PDF icon PDF Version (665 KB)

  • Background:
    Exposure to perfluoroalkyl substances (PFASs) may increase risk for metabolic diseases; however, epidemiologic evidence is lacking at the present time. Pregnancy is a period of enhanced tissue plasticity for the fetus and the mother and may be a critical window of PFAS exposure susceptibility.
    Objective:
    We evaluated the associations between PFAS exposures and metabolic outcomes in pregnant women.
    Methods:
    We analyzed 1,240 pregnant women from the Spanish INMA [Environment and Childhood Project (INfancia y Medio Ambiente)] birth cohort study (recruitment period: 2003–2008) with measured first pregnancy trimester plasma concentrations of four PFASs (in nanograms/milliliter). We used logistic regression models to estimate associations of PFASs (log10-transformed and categorized into quartiles) with impaired glucose tolerance (IGT) and gestational diabetes mellitus (GDM), and we used linear regression models to estimate associations with first-trimester serum levels of triglycerides, total cholesterol, and C-reactive protein (CRP).
    Results:
    Perfluorooctane sulfonate (PFOS) and perfluorohexane sulfonate (PFHxS) were positively associated with IGT (137 cases) [OR per log10-unit increase = 1.99 (95% CI: 1.06, 3.78) and OR=1.65 ( 95% CI: 0.99, 2.76), respectively]. PFOS and PFHxS associations with GDM (53 cases) were in a similar direction, but less precise. PFOS and perfluorononanoate (PFNA) were negatively associated with triglyceride levels [percent median change per log10-unit increase = −5.86% (95% CI: −9.91%, −1.63%) and percent median change per log10-unit increase = −4.75% (95% CI: −8.16%, −0.61%, respectively], whereas perfluorooctanoate (PFOA) was positively associated with total cholesterol [percent median change per log10-unit increase = 1.26% (95% CI: 0.01%, 2.54%)]. PFASs were not associated with CRP in the subset of the population with available data (n=640).
    Conclusions:
    Although further confirmation is required, the findings from this study suggest that PFAS exposures during pregnancy may influence lipid metabolism and glucose tolerance and thus may impact the health of the mother and her child. https://doi.org/10.1289/EHP1062
  • Received: 7 September 2016
    Revised: 5 October 2017
    Accepted: 9 October 2017
    Published: 13 November 2017

    Address correspondence to M. Vrijheid, ISGlobal, Centre for Research in Environmental Epidemiology (CREAL), 88 Doctor Aiguader, 08003 Barcelona, Catalonia, Spain. Telephone: 93 214 73 46. Email: martine.vrijheid@isglobal.org

    Supplemental Material is available online (https://doi.org/10.1289/EHP1062).

    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 (243 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 (41 KB)



Introduction

Perfluoroalkyl substances (PFASs) have been used in many applications since the 1950s, including industrial applications and consumer products (Casals-Casas and Desvergne 2011). PFASs bioaccumulate in the food chain and in animal and human tissues, and exposure persists in the environment and living organisms for years (Wang et al. 2014). The routes of human exposure to PFASs include diet (animal and plant-based foods), migration from packaged foods, drinking water, and inhalation of indoor dust (Wang et al. 2014). Detectable blood levels of PFASs have been reported in pregnant women in Spain (Manzano-Salgado et al. 2015) and in other European (Gebbink et al. 2015), North American (Kato et al. 2014), Asian (Okada et al. 2013), and African (Hanssen et al. 2010) regions. PFAS exposure may pass from mother to child through the placenta (Manzano-Salgado et al. 2015) and through breast milk (Mogensen et al. 2015).

Evidence from epidemiological studies suggests that exposures to PFASs and to other endocrine-disrupting chemicals (EDCs) (i.e., synthetic substances that have been shown to alter the function of the endocrine system in intact organisms) may contribute to obesity (de Cock and van de Bor 2014), lipid alterations (Kabir et al. 2015), diabetes (Taylor et al. 2013), and to autoimmune diseases and inflammation (Kuo et al. 2012). A study of female CD-1 mice reported that mice exposed to low doses of perfluorooctanoate (PFOA) in utero had higher serum insulin and leptin concentrations at 21–33 wk of age than controls (Hines et al. 2009). Rats exposed to PFOA or perfluorooctane sulfonate (PFOS) were reported to have lower total serum cholesterol levels and to show evidence of liver toxicity compared with controls (Lau et al. 2007). Experimental evidence further supports that PFOA and PFOS exposures may alter inflammatory responses and the production of cytokines (DeWitt et al. 2012) that play a role in the pathogenesis of metabolic diseases (Caër et al. 2017).

Few epidemiological studies have evaluated associations between PFAS exposures and metabolic outcomes, and findings have been inconclusive. A cross-sectional study of 571 adults in Taiwan reported that PFOS was positively associated with prevalent diabetes, but associations with other PFAS compounds were negative (inverse) (Su et al. 2016). A prospective cohort study of 258 pregnant women in the United States reported a positive association between serum PFOA concentrations and gestational diabetes mellitus (GDM); associations of other PFASs were positive but close to the null (Zhang et al. 2015). However, a prospective study of 1,274 pregnant women in Canada reported limited evidence of a positive association with GDM for plasma perfluorohexane sulfonate (PFHxS) (significant for the second vs. first quartile comparison only) and no evidence of associations for PFOA or PFOS (Shapiro et al. 2016). Similarly, a recent study of 604 pregnant Faroese women that used a multiple-pollutant modeling approach found no association of PFAS exposures with GDM (Valvi et al. 2017). Further, a Norwegian study of PFASs and serum lipid levels in second-trimester samples from 891 pregnant women reported that PFOS was positively associated with total cholesterol, low-density lipoprotein (LDL), and high-density lipoprotein (HDL), and that PFOA, PFHxS, and three other PFASs were also positively associated with HDL (Starling et al. 2014). Finally, a Danish study that examined PFOS, PFOA, and total cholesterol in serum samples collected from 854 women during the 30th week of gestation reported positive associations for both PFASs (Skuladottir et al. 2015).

Exposure during pregnancy may affect the mother and may also affect her child during gestation and in later life (Bach et al. 2015; Casals-Casas and Desvergne 2011). However, until now, only a few studies with relatively small populations have examined the potential role of PFAS exposures on metabolic outcomes at pregnancy (Starling et al. 2014; Skuladottir et al. 2015; Zhang et al. 2015). Therefore, we evaluated associations of exposures in early pregnancy to PFOS, PFOA, and the emerging PFAS compounds PFHxS and PFNA with metabolic outcomes including impaired glucose tolerance (IGT) and GDM, and with serum levels of triglycerides, total cholesterol, and C-reactive protein (CRP), in a cohort of Spanish women.

Methods

Study Population and Data Collection

The population-based birth cohort study INMA [INfancia y Medio Ambiente (Environment and Childhood)] recruited 2,150 pregnant women in the Spanish regions of Valencia (n=855), Sabadell (n=657), and Gipuzkoa (n=638) between 2003 and 2008 (Guxens et al. 2012). Women were enrolled during the first trimester of pregnancy at the primary health care center or hospital of each region and afterwards were followed at the third trimester of pregnancy and at birth. The inclusion criteria were ≥16 years of age, no assisted reproduction, a singleton pregnancy, intention to deliver at the reference hospital, and no communication handicap. Because the main focus of the INMA studies is on child health, plasma PFAS analysis was limited to mother–child pairs who had archived plasma samples from pregnancy and information on child health outcomes at 4 y of age (n=1,243). Of these, 1,240 mothers who also had information about at least one of the metabolic outcomes of interest were included in the present analysis (58% of those initially enrolled) (see Figure S1). Investigators who performed the PFAS analyses did not have information on the women’s metabolic outcomes, and those who assessed the metabolic outcomes did not have information on PFAS concentrations. To assess potential selection bias, we performed a comparison of main characteristics between pregnant women included in and excluded from analysis.

Nonfasting blood samples were collected from the women at the first-trimester prenatal visit, and interview-based questionnaires administered by trained research staff were used to collect information about sociodemographic (including age, education, occupation, and parity history) and lifestyle characteristics (including alcohol consumption and smoking during pregnancy) at first- and third-trimester prenatal visits.

All participants provided written informed consent. The study was approved by the hospital ethics committees of each participating region.

Exposure Assessment of PFAS

Plasma was separated from blood samples collected at approximately 13 wk of gestation (13.1±1.4), aliquoted into 1.5-mL cryotubes, and stored at −80°C until analysis at the Institute for Occupational Medicine, Rheinisch-Westfälische Technische Hochschule (RWTH) Aachen University, Germany, as previously described (Manzano-Salgado et al. 2015). Plasma concentrations of PFASs were determined using column-switching high performance liquid chromatography (HPLC) (Agilent 1100 Series HPLC) coupled to tandem mass spectrometry (Sciex API 3000 LC/MS/MS System). The limit of quantification (LOQ) was 0.20 ng/mL for PFOS, PFOA, and PFHxS, and 0.10 ng/mL for PFNA.

Metabolic Outcomes

In Spain, the routine clinical examination for GDM in pregnant women is performed at 24–28 wk of gestation according to the guidelines recommended by the Spanish Group of Diabetes and Pregnancy. Women whose physicians considered them to be at high risk for GDM [based on age, body mass index (BMI), personal/family history of diabetes, or previous pregnancy complications] were given a 50-g oral glucose challenge test (OGCT), with blood glucose values measured after 1 h. Women with a postchallenge blood glucose ≥140 mg/dL on the OGCT completed a 100-g 3-h oral glucose tolerance test (OGTT) 2–3 wk after the first screening test, with blood glucose concentrations measured at baseline and 1, 2, and 3 h after the glucose challenge. Results of the OGTT are routinely used to classify women as having GDM if two or more of the baseline or postchallenge blood glucose concentrations exceed National Diabetes Data Group (NDDG) reference values (Expert Committee on the Diagnosis and Classification of Diabetes Mellitus 2003) and as having IGT only (without GDM) if one or fewer of the four blood glucose concentrations exceeds NDDG reference values. We extracted information about IGT and GDM diagnosis from the medical records. Information about GDM was available for women from all three INMA subcohorts, but information about IGT only (i.e., IGT in women not diagnosed with GDM) was available in the medical records for women in the Sabadell and Gipuzkoa cohorts. Our a priori hypothesis was that PFAS exposures may interfere with insulin signaling, glucose homeostasis, or both (Yan et al. 2015, Qiu et al. 2016), which could result in elevated blood glucose levels as well as in GDM. Therefore, we used one set of logistic regression models to estimate associations between PFASs and GDM (with noncases defined as all women without a diagnosis of GDM, regardless of whether they had IGT), and a second set of models to estimate associations between PFASs and IGT, where all women diagnosed with GDM or IGT were classified as cases, and all remaining women were classified as noncases. In addition, we performed a sensitivity analysis in which we repeated the logistic regression models of associations with IGT after excluding women from the Valencia subcohort because Valencia women with IGT but not GDM would have been misclassified as noncases owing to missing IGT information.

Total cholesterol and triglycerides were determined in maternal nonfasting serum collected at approximately 13 wk of gestation using colorimetric enzymatic methods. Measurements were performed at the Bizkaia Basque Country Public Health Laboratory for the Gipuzkoa and Sabadell subcohorts and at the General Biochemistry Laboratory of La Fe Hospital (Valencia) for the Valencia subcohort. CRP levels were measured in first-trimester pregnancy serum by immunoturbidimetry at the Consulting Químico Sanitario (CQS) Laboratory for the Gipuzkoa subcohort and at the Echevarne Sabadell Laboratory for the Sabadell subcohort. CRP levels were not measured in the Valencia subcohort.

Other Covariates

We considered a wide range of potential confounders based on previous literature related to PFAS exposures (Berg et al. 2014; Manzano-Salgado et al. 2016; Sagiv et al. 2015) and/or metabolic outcomes in pregnant women (Halldorsson et al. 2012; Perng et al. 2014; Ryckman et al. 2015; Skuladottir et al. 2015; Starling et al. 2014; Taylor et al. 2013; Wild et al. 2015; Zhang et al. 2015). These confounders included subcohort (Valencia, Sabadell, Gipuzkoa), country of birth, age, gestational week at blood extraction, parity, previous breastfeeding duration, marital status (living with the child’s father, other situation), education, social class [based on occupation according to the International Standard Classification of Occupations–88: I and II, professionals and managers; III, other nonmanual and skilled; IV and V, semiskilled and unskilled manual (http://www.ilo.org/public/english/bureau/stat/isco/isco88/index.htm)], occupation at first prenatal visit, smoking during pregnancy, physical activity [in metabolic equivalents of task (METs) per hour per day], and prepregnancy BMI (kilograms/meters squared; based on prepregnancy self-reported weight divided by measured squared height at first prenatal visit). We further considered the potential confounding role of the women’s diet in the past three months before the blood extraction for PFAS analysis; this was assessed at the first prenatal visit using a 101-item food frequency questionnaire (FFQ) previously validated in Spanish adults (Vioque et al. 2013). From the FFQ, we obtained information on total energy (kilocalories/day), alcohol (grams/day), fish and red meat intake (categorized in tertiles by servings/week), and the relative Mediterranean Diet Score (rMED), a proxy measure of diet quality based on the extent of adherence to a Mediterranean diet (Fernández-Barrés et al. 2016).

Statistical Analysis

We substituted maternal PFAS concentration values that were below the LOQ with LOQ/4. We calculated the Spearman correlation coefficients between PFASs. PFAS concentrations, triglycerides, and CRP levels had right-skewed distributions and therefore were log10-transformed.

We used generalized additive models (GAMs) to examine the linearity of the relationships between continuous log10-transformed PFAS concentrations and the metabolic outcomes. Because some of the associations were not linear (i.e., a p-gain in the GAM <0.10; see Table S1), we show effect estimates for both continuous PFASs and per quartile of PFAS exposure. We used logistic regression to estimate associations between log10-transformed PFAS concentrations and dichotomous outcomes (i.e., IGT and GDM), and we used linear regression models to estimate associations with continuous outcomes (serum cholesterol, log10-transformed triglycerides, and log10-transformed CRP concentrations). We estimated all associations using a basic model adjusted for subcohort only and using a fully adjusted model with additional covariates. We tested for heterogeneity among subcohorts by including an interaction term (PFAS×subcohort) in the fully adjusted models (data not shown). All interaction terms were nonsignificant (p>0.10); therefore, we report coefficients from analyses of the overall population only.

The analysis population included 1,240 women with information about GDM/IGT and 1,194 women with information about serum lipid levels (see Figure S1). Analyses of associations with serum CRP levels were limited to 651 women from the Sabadell and Gipuzkoa subcohorts because CRP was not measured in the Valencia subcohort. The fully adjusted models included slightly fewer observations owing to missing data for additional covariates (≤2%). Because of the low proportion of observations with missing covariate data, we performed complete case analyses.

Covariates included in the multivariable-adjusted models were selected using directed acyclic graphs (DAGs, data not shown) including subcohort, women’s country of birth, previous breastfeeding, parity history, prepregnancy BMI, and gestational week at blood collection. Other potential confounders were evaluated and retained if adjustment caused a >10% change in effect estimates for PFASs and the study outcomes; thus, the fully adjusted models also included physical activity and rMED.

We performed sensitivity analyses of associations between PFASs and IGT after excluding women from the Valencia subcohort (because of missing information about their IGT status). In addition, we estimated associations with all metabolic outcomes using multiple-pollutant models adjusted simultaneously for all four PFAS compounds.

To facilitate the interpretation of effect estimates for continuous outcomes, coefficients are expressed as relative percent change in medians (Barrera-Gómez and Basagaña 2015). The level of statistical significance was set to a two-sided p-value<0.05. All analyses were conducted using STATA statistical software (version 12.0; StataCorp LLC).

Results

Study Population Characteristics

Nearly all of the women were born in Spain (93.2%), and most were working during the first pregnancy trimester (73.6%); approximately half were of lower social class (IV–V: 46.9%) and nulliparous (56.1%) (Table 1). Most women had never breastfed (60.7%) and did not report smoking during pregnancy (68.6%), and 18.5% and 7.9% were classified as overweight and obese, respectively, based on their BMI before pregnancy (Table 1). Compared with women excluded from the present analysis, those who were included were slightly older (mean age 31.9 y vs. 31.4 y) and were more likely to be born in Spain (93.2% versus 86.5%), to have a university education (35.1% versus 29.2%), to be of higher social class (Class I–II: 24% vs. 17%), and to have lower average total energy intake (2086 kcal/day vs. 2202 kcal/day) (Table 1).

Table 1. Characteristics of pregnant women included and excluded from the analysis.
Characteristic Included (n=1,240)a Excluded (n=910)a p-Valueb
n Mean±SD or % n Mean±SD or %
Sociodemographic factors
 Subcohort
  Gipuzkoa 322 25.9% 316 34.7% <0.01
  Sabadell 411 33.1% 246 27.1%
  Valencia 507 40.9% 348 38.2%
 Country of birth
  Spanish 1,151 93.2% 761 86.5% <0.01
  Others 84 6.8% 119 13.5%
  Missing 5 0.4% 30 3.3%
 Maternal age (y) 1,237 31.9±4 695 31.4±4.5 0.04
  Missing 3 0.2% 215 24%
 Parity
  Nulliparous 695 56.1% 473 53.7% 0.28
  Multiparous 543 43.9% 407 46.3%
  Missing 2 0.2% 66 7.2%
 Breastfeeding history
  None 751 60.7% 531 60.3% 0.46
  0–6 months 213 17.1% 156 17.8%
  >6 months 274 22.2% 193 21.9%
  Missing 2 0.2% 30 3.2%
 Education
  Primary education or less 282 22.8% 281 31.9% <0.01
  Secondary education 521 42.1% 341 38.8%
  University degree 434 35.1% 257 29.2%
  Missing 3 0.2% 31 3.4%
 Social class
  I–II: Professionals and managers 297 23.9% 150 17% <0.01
  III: Skilled manual/nonmanual 361 29.1% 207 23.5%
  IV–V: Semiskilled/unskilled 582 46.9% 525 59.5%
Lifestyle factors
 Pre-BMI (kg/m2)c
  Underweight 54 4.3% 47 5.4% 0.63
  Normal 858 69.2% 615 69.9%
  Overweight 229 18.5% 152 17.3%
  Obese 99 7.9% 65 7.4%
 Smoking during pregnancy
  Never smoked 841 68.6% 507 65.6% 0.35
  Smoked until 1st trimester 187 15.2% 125 16.2%
  Smoked until 3rd trimester 198 16.1% 141 18.2%
  Missing 14 1.1% 137 15%
 Alcohol intake at 1st trimester [GM (GSD)] 1,234 0.3 (5.2) 900 0.3 (4.8) 0.27
  Missing 6 0.5% 10 1.1%
 Physical activity during 1st trimester (METs/hour/day) 1,235 37.7±3.7 866 38.1±3.7 0.97
  Missing 5 0.4% 44 4.8%
 Relative Mediterranean diet score
  Low 389 31.7% 281 32.9% 0.97
  Middle 551 44.9% 376 44.1%
  High 286 23.3% 197 23.1%
  Missing 14 1.1% 56 6.1%
 Total energy intake (kcal/day) 1,234 2086±514 869 2202±640 <0.01
  Missing 6 0.5% 41 4.5%

Note: BMI, body mass index; GM, geometric mean; GSD, geometric standard deviation; MET, metabolic equivalent of task; SD, standard deviation.

aSome covariates include fewer observations because of missing values.

bp-Value for the comparison between participants included and not included in analysis; Chi-squared test for percentage values comparisons and Student’s t-test for media values comparisons.

cClassification according to World Health Organization (WHO) criteria.

The overall prevalences of GDM and IGT (including GDM cases) were 4.3% and 11%, respectively (Table 2). The INMA-Sabadell subcohort had the highest prevalence of IGT (17%), and the INMA-Valencia subcohort had the highest prevalence of GDM (5.3%). INMA study participants who were excluded from the present analysis were less likely to be diagnosed with GDM (2.6%) or IGT (6.6%) than women who were included. Compared with women who were excluded, women included in the present analysis had similar mean cholesterol (195±33 mg/dL vs. 196±35 mg/dL) and geometric mean CRP concentrations (0.4±2.4 for both groups), but slightly lower geometric mean triglyceride concentrations (98.1±1.5 mg/dL vs. 101.7±1.5 mg/dL, p=0.04). Women from the INMA-Valencia subcohort had higher average cholesterol and triglycerides than women from the other two cohorts (Table 3).

Table 2. Prevalence [n(%)] of gestational diabetes mellitus and impaired glucose tolerance in pregnant women from the INMA birth cohort study.
Prevalence status All INMA subcohorts INMA-Valencia INMA-Sabadell INMA-Gipuzkoa Women excluded from analysis (all INMA subcohorts) p-Valuea
GDM
 No 1,187 (95.7) 480 (94.7) 397 (96.6) 310 (96.3) 886 (97.4) 0.05
 Yes 53 (4.3) 27 (5.3) 14 (3.4) 12 (3.7) 24 (2.6)
IGTb
 No 1,103 (88.9) 480 (94.7)c 323 (83.0) 300 (87.2) 850 (93.4) 0.02
 Yes 137 (11.1) 27 (5.3)c 66 (17.0) 44 (12.8) 60 (6.6)

Note: GDM, gestational diabetes mellitus; IGT, impaired glucose tolerance; INMA, Environment and Childhood Project (INfancia y Medio Ambiente).

aChi-squared test p-value comparing prevalences between participants included and excluded from analysis.

bCases include women diagnosed with GDM (all subcohorts) and women diagnosed with IGT (Sabadell and Gipuzkoa subcohorts only).

cIGT cases and noncases in the Valencia cohort are classified based on GDM only because IGT data were not available for this subcohort.

Table 3. Serum lipids and C-reactive protein in pregnant women from the INMA birth cohort study.
Serum lipid/protein levels n Min–Max Mean (SD) P25 P50 P75
Cholesterol (mg/dL)a
 All subcohorts 1,194 97–324 195 (33) 174 192 214
 INMA-Valencia 479 119–324 201 (33) 180 199 220
 INMA-Sabadell 398 105–283 190 (31) 170 188 208
 INMA-Gipuzkoa 317 97–321 193 (34) 171 188 211
Triglycerides (mg/dL)a
 All subcohorts 1194 35–374 105 (42.8) 76 96 126
 INMA-Valencia 479 38–374 111 (45) 81 103 136
 INMA-Sabadell 398 35–333 107 (44) 77 96 127
 INMA-Gipuzkoa 317 40–250 94 (35) 69 86 112
CRP (mg/dL)b
 All subcohorts 651 0.03–6.5 0.6 (0.7) 0.2 0.4 0.7
 INMA-Valencia NA NA NA NA NA NA
 INMA-Sabadell 329 0.02–6.5 0.7 (0.7) 0.2 0.3 0.6
 INMA-Gipuzkoa 322 0.03–5.4 0.6 (0.7) 0.2 0.3 0.6

Note: CRP, C-reactive protein; INMA, Environment and Childhood Project (INfancia y Medio Ambiente); Max, maximum value; Min, minimum value; NA, not available; P25, percentile 25; P50, percentile 50; P75, percentile 75; SD, standard deviation.

aData for cholesterol and triglycerides were missing for a total of 46 women from Valencia (n=28), Sabadell (n=13), and Gipuzkoa (n=5).

bData for CRP were missing for a total of 589 women from Valencia (n=507) and Sabadell (n=82).

PFOS and PFOA concentrations were quantified in all samples analyzed, and PFHxS and PFNA concentrations were below the LOQ in 3.7% and 0.65% of samples, respectively (Table 4). PFOS had the highest geometric mean concentration, followed by PFOA, PFNA, and PFHxS. PFOA and PFNA were the most highly correlated PFASs (Spearman’s rho=0.68), followed by PFOA and PFHxS (rho=0.55). For other pairs of PFASs, the correlation coefficients ranged from 0.44 to 0.52.

Table 4. Plasma PFAS concentrations (nanograms/milliliter) in pregnant women participating in the INMA birth cohort study (n=1,240).
PFAS n<LOQa(%) GM (GSD) Min Percentiles Max
5th 25th 50th 75th 95th
PFOA 0 2.31 (1.71) 0.28 0.96 1.63 2.35 3.30 5.23 31.64
PFOS 0 5.77 (1.61) 0.28 2.52 4.51 6.05 7.81 11.35 38.58
PFHxS 46 (3.71) 0.55 (1.96) 0.05 0.24 0.40 0.58 0.82 1.39 11.00
PFNA 8 (0.65) 0.64 (1.75) 0.03 0.28 0.49 0.65 0.90 1.49 5.51

Note: GM, geometric mean; GSD, geometric standard deviation; INMA, Environment and Childhood Project (INfancia y Medio Ambiente); LOQ, limit of quantification; Max, maximum value; Min, minimum value; PFAS, perfluoroalkyl substance; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate.

aThe LOQ was 0.20 ng/mL for PFHxS, PFOS, PFOA and 0.10 ng/mL for PFNA.

Associations between PFAS Exposures and Metabolic Outcomes

Association estimates from basic and fully adjusted models were generally consistent, with a nonsignificant positive association between PFOS and GDM [n=53 cases; adjusted OR for a 1-unit increase in log10-PFOS = 2.40 (95% CI: 0.93, 6.18)], without a clear monotonic dose–response trend (Table 5 and Figure 1A). PFHxS was also positively associated with GDM, although ORs were closer to the null and were not significant. ORs for IGT (n=137 cases) were more precise, with a positive association with PFOS [OR for a 1-unit increase in log10-PFOS = 1.99 (95% CI: 1.06, 3.78)] that was similar for all quartiles relative to the lowest quartile (Table 5 and Figure 1B). As for GDM, PFHxS was also positively associated with IGT, although ORs were closer to the null than for PFOS. Associations between PFASs and IGT were generally consistent with the main analyses when women from the Valencia cohort were excluded, although the estimates were less precise (n=719 observations and 108 cases only) (Table 5). All four PFASs were negatively associated with serum triglyceride concentrations, although associations were close to the null for PFOA and somewhat stronger for PFOS [percent change in the median per unit increase in log10-PFOS = −5.86% (95% CI: −9.91, −1.63%)] and PFNA [percent change per unit increase in log10-PFNA = −4.75% (95% CI: −8.16, −0.61%)] than for PFHxS, and estimated associations were similar for all quartiles relative to the lowest quartile (Table 6 and Figure 2A). Associations between PFASs and triglycerides were nonlinear for PFHxS and PFNA (see Table S1). Serum cholesterol levels were positively associated with PFOA [percent change in median per unit increase in log10-PFOA = 1.26% (95% CI: 0.01, 2.54%)], with similar effect estimates for all quartiles (Table 6 and Figure 2B). However, associations between total cholesterol and the other PFASs were essentially null. Associations between PFASs and CRP were based on only 640 observations (because of missing CRP data for Valencia women), and the results were imprecise and inconclusive (Table 6 and Figure 2C).

Figures 1A and 1B respectively plot associations between adjusted ORs for GDM (95 percent confidence interval) and IGT (95 percent confidence interval) (y-axis) and PFAS exposure groups, namely, PFOA, PFOS, PFHxS, and PFNA (x-axis) for exposure quartiles 1, 2, 3 and 4.

Figure 1. Adjusted odds ratios (ORs) [95% confidence intervals (CIs)] for the associations between quartile-specific perfluoroalkyl substances (PFAS)-exposure groups and gestational diabetes mellitus (GDM) (A) and impaired glucose tolerance (IGT) (B). All models are adjusted for subcohort, country of birth, prepregnancy body mass index, previous breastfeeding, parity, gestational week at blood extraction, physical activity, and relative Mediterranean Diet Score (rMED). See Table 4 for corresponding numeric data. PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate.

Figures 2A, 2B, and 2C respectively, plot adjusted percentage change in median of log 10 triglyceride concentrations (95 percent confidence intervals), cholesterol concentrations (95 percent confidence intervals), and log sub 10 C-reactive protein concentrations (95 percent confidence intervals) across PFAS exposure groups, namely, PFOA, PFOS, PFHxS, and PFNA (x-axis) for exposure quartiles 1, 2, 3 and 4.

Figure 2. Adjusted percent changes [95% confidence intervals (CIs)] in median concentrations of log10-triglycerides (A), total cholesterol (B), and log10C-reactive protein (C) per quartile PFAS-exposure group. All models are adjusted for subcohort, country of birth, prepregnancy body mass index, previous breastfeeding, parity, gestational week at blood extraction, physical activity, and relative Mediterranean Diet Score (rMED). See Table 5 for corresponding numeric data. CRP, C-reactive protein; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate.

Table 5. Odds ratios (95% CI) for the associations of PFASs with gestational diabetes mellitus and impaired glucose tolerance.
PFAS (ng/mL) GDM 53 cases, 1,161 noncases IGT (all subcohorts)a 135 cases, 1,079 noncases IGT (Sabadell and Gipzukoa only)b108 cases, 611 noncases
n cases/noncases Basic modelc OR (95% CI) Fully adjustedd OR (95% CI) n cases/noncases Basic modelc OR (95% CI) Fully adjustedd OR (95% CI) n cases/noncases Fully adjustedd OR (95% CI)
PFOA
 Q1: 0.28 to<1.63 10/295 Reference Reference 21/284 Reference Reference 17/169 Reference
 Q2: 1.63 to<2.35 14/292 1.46 (0.63, 3.37) 1.28 (0.55, 3.02) 34/272 1.30 (0.72, 2.35) 1.22 (0.66, 2.25) 27/152 1.17 (0.57, 2.41)
 Q3: 2.35 to<3.30 15/287 1.62 (0.70, 3.78) 1.35 (0.56, 3.23) 41/261 1.37 (0.76, 2.48) 1.28 (0.68, 2.39) 34/145 1.22 (0.58, 2.58)
 Q4: 3.30 to 31.64 14/287 1.53 (0.64, 3.69) 1.25 (0.50, 3.13) 39/262 1.15 (0.63, 2.12) 1.04 (0.54, 2.39) 30/145 0.86 (0.39, 1.89)
 Per log10-unit increase 53/1161 1.34 (0.74, 2.44) 1.20 (0.62, 2.30) 135/1079 1.24 (0.83, 1.86) 1.24 (0.78, 1.94) 108/611 1.19 (0.69, 2.03)
PFOS
 Q1: 0.28 to<4.51 8/298 Reference Reference 18/288 Reference Reference 16/175 Reference
 Q2: 4.51 to<6.05 15/289 1.89 (0.79, 4.53) 1.89 (0.77, 4.64) 35/269 2.26 (1.24, 4.12) 2.11 (1.13, 3.94) 27/146 1.95 (0.97, 3.95)
 Q3: 6.05 to<7.81 13/291 1.64 (0.67, 4.03) 1.54 (0.61, 3.87) 40/264 2.22 (1.22, 4.02) 2.08 (1.12, 3.86) 32/144 1.92 (0.96, 3.85)
 Q4: 7.81 to 38.58 17/283 2.21 (0.93, 5.21) 2.07 (0.85, 5.01) 42/258 2.36 (1.31, 4.25) 2.22 (1.19, 4.13) 33/146 2.02 (1.00, 4.07)
 Per log10-unit increase 53/1161 2.57 (1.05, 6.25) 2.40 (0.93, 6.18) 135/1079 2.06 (1.16, 3.68) 1.99 (1.06, 3.78) 108/611 1.98 (0.96, 4.09)
PFHxS
 Q1: 0.05 to<0.40 9/293 Reference Reference 15/287 Reference Reference 11/168 Reference
 Q2: 0.40 to<0.58 13/292 1.44 (0.61, 3.45) 1.25 (0.51, 3.03) 24/281 1.61 (0.82, 3.16) 1.51 (0.76, 3.02) 18/133 1.85 (0.81, 4.22)
 Q3: 0.58 to<0.82 19/283 2.29 (0.99, 5.29) 1.81 (0.76, 4.28) 42/260 2.28 (1.19, 4.35) 1.99 (1.01, 3.90) 30/133 1.88 (0.82, 4.31)
 Q4: 0.82 to 11.00 12/293 1.63 (0.62, 4.27) 1.15 (0.42, 3.12) 54/251 2.15 (1.11, 4.18) 1.72 (0.85, 3.49) 49/177 1.84 (0.79, 4.28)
 Per log10-unit increase 53/1161 2.06 (0.99, 4.24) 1.58 (0.73, 3.44) 135/1079 1.89 (1.19, 3.02) 1.65 (0.99, 2.76) 108/611 1.51 (0.85, 2.68)
PFNA
 Q1: 0.03 to<0.49 14/289 Reference Reference 20/283 Reference Reference 9/109 Reference
 Q2: 0.49 to<0.65 13/295 0.99 (0.46, 2.17) 1.01 (0.62, 2.23) 29/279 1.08 (0.58, 1.99) 1.16 (0.62, 2.17) 23/151 1.64 (0.69, 3.87)
 Q3: 0.65 to<0.90 17/285 1.40 (0.66, 2.98) 1.27 (0.59, 2.73) 46/256 1.41 (0.58, 1.99) 1.26 (0.68, 2.33) 38/169 1.50 (0.65, 3.48)
 Q4: 0.90 to 5.51 9/292 0.74 (0.31, 1.82) 0.70 (0.28, 1.75) 40/261 0.96 (0.81, 2.62) 0.95 (0.49, 1.80) 38/182 1.36 (0.58, 3.19)
 Per log10-unit increase 53/1161 0.93 (0.46, 1.89) 0.85 (0.40, 1.80) 135/1079 1.08 (0.67, 1.74) 0.95 (0.57, 1.60) 108/611 1.06 (0.57, 1.96)

Note: CI, confidence interval; GDM, gestational diabetes mellitus; IGT, impaired glucose tolerance; OR, odds ratio; PFAS, perfluoroalkyl substance; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; Q1–Q4, quartiles 1 through 4.

aIGT cases include women diagnosed with GDM from all 3 subcohorts, plus women diagnosed with IGT only (without GDM) from the Sabadell and Gipzukoa subcohorts.

bSensitivity analysis of women diagnosed with IGT (without GDM) from the Sabadell and Gipzukoa subcohorts only.

cBasic models are adjusted for subcohort only.

dFully adjusted models include subcohort, country of birth, prepregnancy body mass index, previous breastfeeding, parity, gestational week at blood extraction, physical activity, and relative Mediterranean Diet Score.

Table 6. Percentage of median change (95% CI) for associations of PFASs (categorized into quartiles or continuous log10-transformed) with log10-triglycerides (milligrams/deciliter), total cholesterol (milligrams/deciliter) and log10C-reactive protein (milligrams/deciliter).
PFAS (ng/mL) Log10-triglycerides (mg/dL) n=1,168 Total cholesterol (mg/dL) n=1,168 Log10C-reactive protein (mg/dL) n=640
Basic modela % change (95% CI) Fully adjustedb% change (95% CI) Basic modela% change (95% CI) Fully adjustedb% change (95% CI) Basic modela% change (95% CI) Fully adjustedb% change (95% CI)
PFOA
 Q1: 0.28 to<1.63 Reference Reference Reference Reference Reference Reference
 Q2: 1.63 to<2.35 −4.02 (−11.1, 3.05) −1.98 (−8.61, 5.13) 4.85 (−0.46, 10.1) 2.43 (0.20, 4.81) −12.20 (−29.4, 5.13) −12.19 (−27.3, 6.18)
 Q3: 2.35 to<3.30 −7.69 (−13.9, −1.30) −2.96 (−9.61, 6.13) 3.70 (−1.78, 9.18) 2.33 (−0.01, 4.81) −8.61 (−27.7, 10.5) −3.92 (−22.1, 17.3)
 Q4: 3.30 to 31.64 −7.04 (−14.5, 0.38) −1.98 (−8.61, 6.18) 5.24 (−0.42, 10.9) 3.15 (0.70, 5.76) −7.69 (−25.3, 10.0) 3.05 (−17.3, 28.4)
 Per log10-unit increase −5.49 (−9.03, −2.09) −2.78 (−6.15, 1.42) 0.90 (−0.34, 2.14) 1.26 (0.01, 2.54) −3.47 (−14.2, 7.31) 2.86 (−8.12, 14.3)
PFOS
 Q1: 0.28 to<4.51 Reference Reference Reference Reference Reference Reference
 Q2: 4.51 to<6.05 −8.61 (−14.4, −2.96) −6.76 (−13.1, 1.01) −0.93 (−6.16, 4.30) −0.16 (−2.37, 2.02) 5.13 (−13.4, 23.7) 6.18 (−11.3, 28.4)
 Q3: 0 6.05 to<7.81 −9.52 (−16.3, −2.96) −8.61 (−14.8, −1.00) −2.25 (−5.60, 4.99) −0.49 (−2.66, 1.71) −7.69 (−25.8, 10.5) −6.76 (−22.9, 11.6)
 Q4: 7.81 to 38.58 −11.31 (−17.9, −4.88) −8.61 (−14.8, −1.99) 2.79 (−2.45, 8.04) 1.08 (−1.19, 3.36) −6.41 (−24.0, 11.2) −5.82 (−22.9, 12.7)
 Per log10-unit increase −7.91 (−11.8, −3.97) −5.86 (−9.91, −1.63) 1.17 (−0.26, 2.59) 0.88 (−0.53, 2.37) −12.8 (−22.8, −2.77) −8.41 (−18.4, 3.35)
PFHxS
 Q1: 0.05 to<0.40 Reference Reference Reference Reference Reference Reference
 Q2: 0.40 to<0.58 −7.69 (−14.3, −1.00) −4.88 (−12.2, 2.02) 2.95 (−2.32, 8.22) 1.21 (−1.05, 3.45) −8.61 (−24.4, 7.2) −1.98 (−18.9, 19.7)
 Q3: 0.58 to<0.82 −3.92 (−10.9, 3.05) −4.88 (−11.3, 2.02) 3.06 (−2.41, 8.53) 0.60 (−1.69, 2.94) −10.4 (−28.9, 8.42) −10.4 (−27.4, 9.42)
 Q4: 0.82 to 11.00 −7.69 (−13.9, 1.40) −5.82 (−13.1, 3.05) 3.70 (−2.22, 9.62) 0.70 (−1.86, 3.38) −7.69 (−27.9, 13.0) −4.88 (−23.4, 18.7)
 Per log10-unit increase −4.90 (−9.16, −0.72) −3.53 (−8.2, 1.45) 0.43 (−1.06, 1.91) −0.09 (−8.25, 1.45) −7.59 (−18.2, 3.05) −8.25 (−18.8, 4.40)
PFNA
 Q1: 0.03 to<0.49 Reference Reference Reference Reference Reference Reference
 Q2: 0.49 to<0.65 −8.61 (−15.4, −1.98) −5.82 (−12.2, 1.01) 0.85 (−4.44, 6.15) 0.90 (−1.25, 3.19) 1.01 (−19.3, 21.4) −1.00 (−19.7, 20.9)
 Q3: 0.65 to<0.90 −7.69 (−13.6, −1.88) −7.69 (−13.9, −1.00) 4.12 (−1.32, 9.56) 1.11 (−1.13, 3.47) −12.2 (−32.9, 8.30) −11.3 (−28.1, 8.33)
 Q4: 0.90 to 5.51 −7.69 (−14.8, −0.56) −5.82 (−13.9, 2.02) 3.61 (−2.01, 9.23) 1.81 (−0.57, 4.24) −9.52 (−29.7, 10.6) −3.92 (−22.9, 18.5)
 Per log10-unit increase −5.32 (−9.03, −1.61) −4.75 (−8.16, −0.61) 0.55 (−0.72, 1.82) 0.46 (−0.76, 1.70) −2.40 (−12.8, 7.88) 1.22 (−9.82, 12.3)

Note: Effect estimates from the linear regression models expressed as percent change in medians per log10-unit increase in exposure. CI, confidence interval; OR, odds ratio; PFAS, perfluoroalkyl substance; PFHxS, perfluorohexane sulfonate; PFNA, perfluorononanoate; PFOA, perfluorooctanoate; PFOS, perfluorooctane sulfonate; Q1–Q4, quartiles 1 through 4.

aBasic models are adjusted for subcohort only.

bFully adjusted models include subcohort, country of birth, prepregnancy body mass index, previous breastfeeding, parity, gestational week at blood extraction, physical activity, and relative Mediterranean Diet Score.

Sensitivity Analyses

In the multivariate-adjusted models including all four PFAS compounds, associations with the odds for GDM and IGT were strengthened for PFOS and attenuated for PFHxS (see Table S2). Further, the positive association between PFOA and serum total cholesterol levels increased in magnitude [percent change in median per unit increase in log10-PFOA = 2.04% (95% CI: 0.20, 3.93%)], whereas associations of PFOS and PFNA with triglyceride levels were attenuated (see Table S3).

Discussion

Findings from this large study provide evidence for associations between PFAS exposures and metabolic outcomes in pregnant women. Plasma PFOS concentrations were positively associated with both IGT and GDM, although estimates for GDM were based on fewer cases and were less precise, and ORs were comparable for all quartiles relative to the lowest exposure category. PFHxS was also positively associated with both outcomes, although ORs were closer to the null than those for PFOS. Further, PFOA concentrations were associated with a slight increase in nonfasting serum cholesterol levels, whereas PFOS and PFNA concentrations were associated with small decreases in serum triglyceride levels. In multipollutant models adjusted for all PFAS compounds, the associations between PFOS concentrations and the odds for IGT or GDM and those between PFOA and cholesterol levels were strengthened, whereas associations for PFHxS and PFNA with the study outcomes were attenuated. Moreover, we did not find evidence for associations between PFAS exposures and serum CRP levels, although this analysis was based on only a subset of the study population (n=640), which may in part explain the imprecise coefficients.

To our knowledge, only three previous epidemiologic studies, with median concentrations of PFOS, PFOA, or both that were two to three times as high as those in our population, have evaluated the associations between PFAS concentrations and risk for GDM. In a prospective cohort study of 258 U.S. women, PFOA concentrations were significantly associated with self-reported GDM, whereas ORs for PFOS, PFNA, and four additional PFASs were close to the null (Zhang et al. 2015). In a prospective study of Canadian women, PFHxS was positively associated with IGT (49 cases vs. 1,102 women with normal blood glucose), but the association was nonlinear and was strongest for the second versus first quartile (Shapiro et al. 2016). In addition, there was only a weak positive association between PFHxS and GDM (44 cases), and there was no evidence of positive associations between either outcome and PFOA or PFOS. In a study of pregnant Faroese women (49 GDM cases vs. 555 women without GDM), no clear association with GDM was found for PFOS, PFOA, PFHxS, PFNA, perfluorodecanoic acid (PFDA), or overall PFAS exposure (Valvi et al. 2017). In the present study, associations derived using a multipollutant model that included all four PFASs were substantially attenuated for PFHxS and IGT, whereas associations between PFOS and IGT were comparable to those found using the single-pollutant models. Although we found some evidence of a positive association between PFOS and GDM, our findings for GDM (including null findings for other PFASs) were imprecise owing to the small number of cases (n=53) and should therefore be interpreted with caution. In addition, women were evaluated for IGT and GDM at their physician’s discretion, which may have resulted in some women being misclassified as noncases for both outcomes. Moreover, we did not have information on IGT (without GDM) for women from the Valencia subcohort, although ORs for associations between IGT and PFOS and PFHxS when Valencia women were excluded were generally consistent with those from the primary analysis.

Studies using animal models have indicated that PFOS and PFOA may interfere with the phosphatidylinositol 3-kinase-serine/threonine protein kinase (PI3K-AKT) signaling pathway, which plays an important role in the metabolic actions of insulin (Yan et al. 2015). PFOA exposure decreased expression of the PI3-AKT signaling pathway in rats (Yan et al. 2015), and in vitro experiments using human hepatoma HepG2 cells suggested that PFOS exposure may also inhibit AKT activation, leading to insulin resistance (Qiu et al. 2016). Therefore, the associations of PFOS and IGT observed in our study could be causal and require further confirmation in other populations. Such confirmation is critical because GDM is associated with short- and long-term adverse outcomes in the mother and in her offspring, including an increased risk of fetal macrosomia (i.e., high birth weight) and hyperinsulinemia, as well as with higher risks of cesarean section and hypertensive disorders in the mother (Kampmann et al. 2015). Moreover, women diagnosed with GDM are more likely to develop type 2 diabetes, metabolic syndrome, and cardiovascular disease after pregnancy, and their children are at increased risk of childhood obesity and glucose intolerance (Kampmann et al. 2015).

A previous cross-sectional study of Norwegian women (n=891) reported a positive association between PFOS and total cholesterol and weaker positive associations for all but one of the other six PFASs evaluated (Starling et al. 2014). PFOS and other PFASs were also associated with higher HDL cholesterol, a component of total cholesterol that is generally considered to be beneficial to health. PFOS and PFOA were also positively associated with serum total cholesterol levels in late pregnancy (approximately 30 wk) in a large Danish cohort (n=854) that did not evaluate associations with HDL or other PFASs (Skuladottir et al. 2015). In the present study, we measured PFAS exposures and total cholesterol (but not HDL) at an earlier stage of pregnancy (approximately 13 wk), and we found evidence of a weak positive association between total cholesterol and PFOA, whereas associations with other PFASs were essentially null. Differences in the timing of exposure and outcome measurements and differences in exposure ranges (which were higher for PFOS in both previous studies, and for PFOA in the Danish study, compared with the present study) may have contributed to inconsistencies among the studies. In addition, individual PFAS concentrations were moderately to highly correlated in all three study populations, which makes it difficult to attribute associations to specific individual PFAS compounds. A variety of noncausal mechanisms may also contribute to, or explain, inconsistent findings, including random error, selection bias, confounding, and differences in susceptibility related to coexposures, comorbid conditions, or other factors. We found evidence of negative associations with triglyceride levels, particularly for PFOS and PFNA, although effect estimates were similar for all quartiles above the reference level, and they were attenuated in the multipollutant adjusted model. Starling et al. (2014) examined seven PFASs and triglycerides in Norwegian women, but the results did not support clear dose–response relationships overall [although there were statistically significant associations for the fourth vs. first quartile of perfluoroundecanoic acid (PFUnDA) and the second vs. first quartile of perfluoroheptane sulfonate (PFHpS)]. In an in vitro study, PFASs with 4–12 carbons (including the compounds we studied) activated the human peroxisome proliferator-activated receptor α (PPARα) in transiently transfected COS-1 cells (Wolf et al. 2012). PPARα regulates fatty acid uptake and metabolism, and its activation may reduce plasma triglycerides and increase HDL cholesterol levels (Shah et al. 2010). However, additional observational and experimental studies are needed to confirm associations and to investigate possible PPAR-dependent or independent mechanisms for effects of PFASs on lipid metabolism.

It has been proposed that PFAS exposures may affect health by increasing inflammation (Lau et al. 2007), and serum CRP is a well-established clinical marker of systemic inflammation. However, information on CRP was available for only a subset of the study population, and the overall results were imprecise and inconclusive. Larger studies integrating inflammatory markers beyond CRP are needed to evaluate any potential role of inflammation in PFAS-related metabolic effects.

Compared with INMA participants who were excluded from the present analysis (because of missing information for PFASs or the study outcomes), women who were included were more likely to have been born in Spain, to have higher levels of education and social class, and to be older. Included women also had a higher prevalence of IGT and GDM and somewhat lower triglyceride serum levels and total energy intake. In addition, a previous analysis of the INMA cohort showed that women born outside Spain had lower average PFAS concentrations than women born in Spain (Manzano-Salgado et al. 2016). Therefore, findings for the women included in the present analysis may not be representative of the INMA cohort population overall. Further, cross-sectional associations between PFAS and total cholesterol and triglycerides might be explained by reverse causation if, for example, PFAS pharmacokinetics are affected by lipid levels, or if PFAS and lipid concentrations are both influenced by a common factor. Total cholesterol and triglycerides were measured in nonfasting samples collected from the pregnant women in our study population, which may have resulted in values that were not representative of the usual lipid levels in some women. Although our study had a number of strengths, we cannot rule out the potential influence of random error, bias due to uncontrolled confounding, outcome misclassification, or other methodological issues that might produce misleading results, and it will be important for our findings to be confirmed in additional study populations. Strengths of this study include the assessment of PFAS exposures using plasma biomarkers. The long elimination half-lives of PFOA, PFOS and PFHxS in human blood are estimated to be from 3 to 9 y (Olsen et al. 2007); thus, one single PFAS measurement may reflect chronic exposure. Moreover, we measured PFAS concentrations early in pregnancy, which may reduce the likelihood of confounding due to physiological changes of gestation, such as changes in the glomerular filtration rate (Verner et al. 2015). The large sample size and the broad list of measured confounders are other important strengths of this study.

Conclusion

Our findings suggest that PFAS exposures may influence lipid metabolism and glucose tolerance during pregnancy. To our knowledge, this is the largest study of associations between PFAS exposures and multiple metabolic outcomes in pregnant women to date. Our findings require confirmation but are worthy of further exploration, given their potential implications for the short- and long-term health of mothers and their children.

Acknowledgments

The authors are grateful to all the participants for their generous collaboration. A full roster of the Environment and Childhood Project (INfancia y Medio Ambiente) (INMA) project investigators can be found at http://www.proyectoinma.org/presentacion-inma/listado-investigadores/en listado-investigadores.html.

This study was funded in part by grants from the European Union (FP7-ENV-2011 cod 282957 and HEALTH.2010.2.4.5-1), the Instituto de Salud Carlos III, the Spanish Ministry of Health (Red INMA G03/176; CB06/02/0041; FIS-PI12/01890, FIS-PI041436, FIS- PI081151, FIS-PI06/0867, FIS-PS09/00090; FIS-FEDER: 03/1615, 04/1509, 04/1112, 04/1931, 05/1079, 05/1052, 06/1213, 07/0314, 09/02647, 11/01007, 11/02591, 11/02038, 13/1944, 13/2032, 14/00891, and 14/01687; Miguel Servet-FEDER CP11/0178, MS13/00054, and CPII16/00051; and PFIS-FI14/00099), Generalitat Valenciana (FISABIO: UGP 15-230, UGP-15-244, and UGP-15-249), the Department of Health of the Basque Government (2005111093 and 2009111069), the Provincial Government of Gipuzkoa (DFG06/004 and DFG08/001), the Generalitat de Catalunya-CIRIT (1999SGR 00241), and the National Institutes of Health/National Institute of Environmental Health Sciences (grant number ES021477). This study has been reviewed and approved by the accredited committees of the following institutions: The Municipal Institute of Sanitary Assistance of Barcelona, La Fe University Hospital of Valencia, and The Donostia Hospital.

References

Bach CC, Bech BH, Brix N, Nohr EA, Bonde JPE, Henriksen TB. 2015. Perfluoroalkyl and polyfluoroalkyl substances and human fetal growth: a systematic review. Crit Rev Toxicol 45(1):53–67, PMID: 25372700, 10.3109/10408444.2014.952400.

Barrera-Gómez J, Basagaña X. 2015. Models with transformed variables. Epidemiology 26(2):e16–e17, PMID: 25643111, 10.1097/EDE.0000000000000247.

Berg V, Nøst TH, Huber S, Rylander C, Hansen S, Veyhe AS, et al. 2014. Maternal serum concentrations of per- and polyfluoroalkyl substances and their predictors in years with reduced production and use. Environ Int 69:58–66, PMID: 24815340, 10.1016/j.envint.2014.04.010.

Caër C, Rouault C, Le Roy T, Poitou C, Aron-Wisnewsky J, Torcivia A, et al. 2017. Immune cell-derived cytokines contribute to obesity-related inflammation, fibrogenesis and metabolic deregulation in human adipose tissue. Sci Rep 7(1):3000, PMID: 28592801, 10.1038/s41598-017-02660-w.

Casals-Casas C, Desvergne B. 2011. Endocrine disruptors: from endocrine to metabolic disruption. Annu Rev Physiol 73:135–162, PMID: 21054169, 10.1146/annurev-physiol-012110-142200.

de Cock M, van de Bor M. 2014. Obesogenic effects of endocrine disruptors, what do we know from animal and human studies?. Environ Int 70:15–24, PMID: 24879368, 10.1016/j.envint.2014.04.022.

DeWitt JC, Peden-Adams MM, Keller JM, Germolec DR. 2012. Immunotoxicity of perfluorinated compounds: Recent developments. Toxicol Pathol 40(2):300–311, PMID: 22109712, 10.1177/0192623311428473.

Expert Committee on the Diagnosis and Classification of Diabetes Mellitus. 2003. Report of the expert committee on the diagnosis and classification of diabetes mellitus. Diabetes Care 26(suppl1):S5–S20, PMID: 12502614.

Fernández-Barrés S, Romaguera D, Valvi D, Martínez D, Vioque J, Navarrete-Muñoz EM, et al. 2016. Mediterranean dietary pattern in pregnant women and offspring risk of overweight and abdominal obesity in early childhood: the INMA birth cohort study. Pediatr Obes 11(6):491–499, PMID: 26763767, 10.1111/ijpo.12092.

Gebbink WA, Glynn A, Berger U. 2015. Temporal changes (1997–2012) of perfluoroalkyl acids and selected precursors (including isomers) in Swedish human serum. Environ Pollut 199:166–173, PMID: 25660070, 10.1016/j.envpol.2015.01.024.

Guxens M, Ballester F, Espada M, Fernández MF, Grimalt JO, Ibarluzea J, et al. 2012. Cohort profile: The INMA–INfancia y Medio Ambiente–(Environment and Childhood) Project. Int J Epidemiol 41(4):930–940, PMID: 21471022, 10.1093/ije/dyr054.

Halldorsson TI, Rytter D, Haug LS, Bech BH, Danielsen I, Becher G, et al. 2012. Prenatal exposure to perfluorooctanoate and risk of overweight at 20 years of age: a prospective cohort study. Environ Health Perspect 120(5):668–673, PMID: 22306490, 10.1289/ehp.1104034.

Hanssen L, Röllin H, Odland J, Moe M, Sandanger T. 2010. Perfluorinated compounds in maternal serum and cord blood from selected areas of South Africa: results of a pilot study. J Environ Monit 12(6):1355–1361, PMID: 20424796, 10.1039/b924420d.

Hines EP, White SS, Stanko JP, Gibbs-Flournoy EA, Lau C, Fenton SE. 2009. Phenotypic dichotomy following developmental exposure to perfluorooctanoic acid (PFOA) in female CD-1 mice: low doses induce elevated serum leptin and insulin, and overweight in mid-life. Mol Cell Endocrinol 304(1–2):97–105, PMID: 19433254, 10.1016/j.mce.2009.02.021.

Kabir ER, Rahman MS, Rahman I. 2015. A review on endocrine disruptors and their possible impacts on human health. Environ Toxicol Pharmacol 40(1):241–258, PMID: 26164742, 10.1016/j.etap.2015.06.009.

Kampmann U, Madsen LR, Skajaa GO, Iversen DS, Moeller N, Ovesen P. 2015. Gestational diabetes: a clinical update. World J Diabetes 6(8):1065–1072, PMID: 26240703, 10.4239/wjd.v6.i8.1065.

Kato K, Wong L, Chen A, Dunbar C, Webster GM, Lanphear BP, et al. 2014. Changes in serum concentrations of maternal poly- and perfluoroalkyl substances over the course of pregnancy and predictors of exposure in multiethnic cohort of Cincinnati, Ohio pregnant women during 2003–2006. Environ Sci Technol 48(16):9600–9608, PMID: 25026485, 10.1021/es501811k.

Kuo C-H, Yang S-N, Kuo P-L, Hung C-H. 2012. Immunomodulatory effects of environmental endocrine disrupting chemicals. Kaohsiung J Med Sci 28(7 suppl): S37–S42, PMID: 22871600, 10.1016/j.kjms.2012.05.008.

Lau C, Anitole K, Hodes C, Lai D, Pfahles-Hutchens A, Seed J. 2007. Perfluoroalkyl acids: a review of monitoring and toxicological findings. Toxicol Sci 99(2):366–394, PMID: 17519394, 10.1093/toxsci/kfm128.

Manzano-Salgado CB, Casas M, Lopez-Espinosa MJ, Ballester F, Martinez D, Ibarluzea J, et al. 2016. Variability of perfluoroalkyl substance concentrations in pregnant women by socio-demographic and dietary factors in a Spanish birth cohort. Environ Int 92–93:357–365, PMID: 27132161, 10.1016/j.envint.2016.04.004.

Manzano-Salgado CB, Casas M, Lopez-Espinosa M-J, Ballester F, Basterrechea M, Grimalt JO, et al. 2015. Transfer of perfluoroalkyl substances from mother to fetus in a Spanish birth cohort. Environ Res 142:471–478, PMID: 26257032, 10.1016/j.envres.2015.07.020.

Mogensen UB, Grandjean P, Nielsen F, Weihe P, Budtz-Jørgensen E. 2015. Breastfeeding as an exposure pathway for perfluorinated alkylates. Environ Sci Technol 49(17):10466–10473, PMID: 26291735, 10.1021/acs.est.5b02237.

Okada E, Kashino I, Matsuura H, Sasaki S, Miyashita C, Yamamoto J, Ikeno T, et al. 2013. Temporal trends of perfluoroalkyl acids in plasma samples of pregnant women in Hokkaido, Japan, 2003–2011. Environ Int 60:89–96, PMID: 24013022, 10.1016/j.envint.2013.07.013.

Olsen GW, Burris JM, Ehresman DJ, Froehlich JW, Seacat AM, Butenhoff JL, et al. 2007. Half-life of serum elimination of perfluorooctanesulfonate, perfluorohexanesulfonate, and perfluorooctanoate in retired fluorochemical production workers. Environ Health Perspect 115(9):1298–1305, PMID: 17805419, 10.1289/ehp.10009.

Perng W, Rifas-Shiman SL, Rich-Edwards JW, Stuebe AM, Oken E. 2014. Inflammation and weight gain in reproductive-aged women. Ann Hum Biol 43(1):91–95, PMID: 25510294, 10.3109/03014460.2014.968619.

Qiu T, Chen M, Sun X, Cao J, Feng C, Li D, et al. 2016. Perfluorooctane sulfonate-induced insulin resistance is mediated by protein kinase B pathway. Biochem Biophys Res Commun 477(4):781–785, PMID: 27363333, 10.1016/j.bbrc.2016.06.135.

Ryckman K, Spracklen C, Smith C, Robinson J, Saftlas A. 2015. Maternal lipid levels during pregnancy and gestational diabetes: A systematic review and meta-analysis. BJOG 122(5):643–651, PMID: 25612005, 10.1111/1471-0528.13261.

Sagiv SK, Rifas-Shiman SL, Webster TF, Mora AM, Harris MH, Calafat AM, et al. 2015. Sociodemographic and perinatal predictors of early pregnancy per- and polyfluoroalkyl substance (PFAS) concentrations. Environ Sci Technol 49(19):11849–11858, PMID: 26333069, 10.1021/acs.est.5b02489.

Shah A, Rader DJ, Millar JS. 2010. The effect of PPAR-alpha agonism on apolipoprotein metabolism in humans. Atherosclerosis 210(1):35–40, PMID: 2000551510.1016/j.atherosclerosis.2009.11.010.

Shapiro GD, Dodds L, Arbuckle TE, Ashley-Martin J, Ettinger AS, Fisher M, et al. 2016. Exposure to organophosphorus and organochlorine pesticides, perfluoroalkyl substances, and polychlorinated biphenyls in pregnancy and the association with impaired glucose tolerance and gestational diabetes mellitus: The MIREC Study. Environ Res 147:71–81, PMID: 26852007, 10.1016/j.envres.2016.01.040.

Skuladottir M, Ramel A, Rytter D, Haug LS, Sabaredzovic A, Bech BH, et al. 2015. Examining confounding by diet in the association between perfluoroalkyl acids and serum cholesterol in pregnancy. Environ Res 143(Pt A):33–38, PMID: 26432473, 10.1016/j.envres.2015.09.001.

Starling AP, Engel SM, Whitworth KW, Richardson DB, Stuebe AM, Daniels JL, et al. 2014. Perfluoroalkyl substances and lipid concentrations in plasma during pregnancy among women in the Norwegian Mother and Child Cohort Study. Environ Int 62:104–112, PMID: 24189199, 10.1016/j.envint.2013.10.004.

Su TC, Kuo CC, Hwang JJ, Lien GW, Chen MF, Chen PC. 2016. Serum perfluorinated chemicals, glucose homeostasis and the risk of diabetes in working-aged Taiwanese adults. Environ Int 88:15–22, PMID: 26700417, 10.1016/j.envint.2015.11.016.

Taylor KW, Novak RF, Anderson HA, Birnbaum LS, Blystone C, Devito M, et al. 2013. Evaluation of the association between persistent organic pollutants (POPs) and diabetes in epidemiological studies: a national toxicology program workshop review. Environ Health Perspect 121(7):774–783, PMID: 23651634, 10.1289/ehp.1205502.

Valvi D, Oulhote Y, Weihe P, Dalgård C, Bjerve KS, Steuerwald U, et al. 2017. Gestational diabetes and offspring birth size at elevated environmental pollutant exposures. Environ Int 107:205–215, PMID: 28753482, 10.1016/j.envint.2017.07.016.

Verner M-A, Loccisano AE, Morken N-H, Yoon M, Wu H, McDougall R, et al. 2015. Associations of perfluoroalkyl substances (PFAS) with lower birth weight: an evaluation of potential confounding by glomerular filtration rate using a physiologically based pharmacokinetic model (PBPK). Environ Health Perspect
123(12):1317–1324, PMID: 26008903, 10.1289/ehp.1408837.

Vioque J, Navarrete-Muñoz E-M, Gimenez-Monzó D, García-de-la-Hera M, Granado F, Young IS, et al. 2013. Reproducibility and validity of a food frequency questionnaire among pregnant women in a Mediterranean area. Nutr J 12(1):26, PMID: 23421854, 10.1186/1475-2891-12-26.

Wang Z, Cousins IT, Scheringer M, Buck RC, Hungerbühler K. 2014. Global emission inventories for C4-C14 perfluoroalkyl carboxylic acid (PFCA) homologues from 1951 to 2030, Part I: Production and emissions from quantifiable sources. Environ Int 70:62–75, PMID: 24932785, 10.1016/j.envint.2014.04.013.

Wild R, Weedin EA, Wilson D. 2015. Dyslipidemia in pregnancy. Cardiol Clin 33(2):209–215, 10.1016/j.ccl.2015.01.002.

Wolf CJ, Schmid JE, Lau C, Abbott BD. 2012. Activation of mouse and human peroxisome proliferator-activated receptor-alpha (PPARα) by perfluoroalkyl acids (PFAAs): Further investigation of C4-C12 compounds. Reprod Toxicol 33(4):546–551, PMID: 22107727, 10.1016/j.reprotox.2011.09.009.

Yan S, Zhang H, Zheng F, Sheng N, Guo X, Dai J. 2015. Perfluorooctanoic acid exposure for 28 days affects glucose homeostasis and induces insulin hypersensitivity in mice. Sci Rep 5:11029, PMID: 26066376, 10.1038/srep11029.

Zhang C, Sundaram R, Maisog J, Calafat AM, Barr DB, Buck Louis GM. 2015. A prospective study of prepregnancy serum concentrations of perfluorochemicals and the risk of gestational diabetes. Fertil Steril 103(1):184–189, PMID: 25450302, 10.1016/j.fertnstert.2014.10.001.

Ridge Properties DBA Pain Relief Naturally Issues Voluntary Nationwide Recall of all lots of Naturally HL Bedsore Cream, Pretat by TAT Balm Carbomer Free Gel, & All Naturally HL Hemorrhoid products due to manufacturing concerns at the facility

Ridge Properties DBA Pain Relief Naturally voluntarily recalls all lots within expiry of Naturally HL Bedsore Relief Cream, Extra Strength PreTAT by TAT Balm Carbomer Free Gel and Extra Strength Naturally HL Hemorrhoid Numbing with Lidocaine.

Early-Life Selenium Status and Cognitive Function at 5 and 10 Years of Age in Bangladeshi Children

Author Affiliations open

1Institute of Environmental Medicine, Karolinska Institutet, Stockholm, Sweden

2International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh

PDF icon PDF Version (176 KB)

  • Background:
    In older adults, selenium status has been positively associated with cognitive function. We recently reported a positive association between maternal selenium status in pregnancy and children’s cognitive function at 1.5 y.
    Objective:
    We followed up the children to assess if prenatal and childhood selenium status was associated with cognitive abilities at 5 and 10 y.
    Methods:
    This longitudinal cohort study was nested in Maternal and Infant Nutrition Interventions in Matlab (MINIMat), a population-based, randomized supplementation trial in pregnancy in rural Bangladesh. Selenium in maternal blood [erythrocyte fraction (Ery-Se) at baseline] and in child hair and urine was measured using inductively coupled plasma mass spectrometry. Children’s cognition at 5 and 10 y was assessed using the Wechsler Preschool and Primary Scale of Intelligence™ and the Wechsler Intelligence Scale for Children®, respectively. In total, 1,408 children were included.
    Results:
    Multivariable-adjusted linear regression analyses showed that prenatal selenium status was positively associated with children’s cognitive function at 5 and 10 y. An increase in maternal Ery-Se from the fifth to the 95th percentile [median: 0.44 μg/g hemoglobin (Hb)] was associated with an increase in full developmental score of 3.5 [95% confidence interval (CI): 0.1, 7.0], corresponding to 0.16 standard deviation (SD) at 5 y, and 8.1 (95% CI: 3.8, 13), corresponding to 0.24 SD at 10 y. In addition, urine and hair selenium concentrations at 5 and 10 y of age were positively associated with cognitive function at 10 y, although associations were inverse for concentrations ≥98th percentile. Some associations were slightly stronger for girls than for boys.
    Conclusions:
    Measures of prenatal and childhood (below the 98th percentile) selenium status were associated with higher cognitive function scores at 5 and 10 y of age. https://doi.org/10.1289/EHP1691
  • Received: 30 January 2017
    Revised: 20 September 2017
    Accepted: 21 September 2017
    Published: 07 November 2017

    Address correspondence to M. Kippler, Institute of Environmental Medicine, Karolinska Institutet, Box 210, SE-171 77, Stockholm, Sweden. Telephone: 46 8 524 874 07. Email: maria.kippler@ki.se

    Supplemental Material is available online (https://doi.org/10.1289/EHP1691).

    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 (215 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 (107 KB)

Introduction

Selenium deficiency has been associated with cognitive decline in the elderly (Akbaraly et al. 2007; Gao et al. 2007; Rita Cardoso et al. 2014). Animal studies suggest that selenium supplementation may prevent dopamine loss, degeneration of neurons, and lipid peroxidation in the brain induced by exposure to various pro-oxidative toxicants (al-Deeb et al. 1995; Imam et al. 1999; Zafar et al. 2003). These findings suggest that selenium may help maintain cognitive function. Selenium functions as an essential component of antioxidant systems such as glutathione peroxidases and thioredoxin reductases, and it is also incorporated into iodothyronine deiodinases, which are involved in thyroid hormone metabolism (Roman et al. 2014). All of these systems are important for brain function as well as for neurodevelopment. However, little is known about the potential impact of poor selenium status on neurodevelopment and if there are critical early-life windows for effects of selenium on brain function later in life.

In animal studies, selenium deficiency during gestation affected markers of neurological development in rat offspring (Mitchell et al. 1998), and maternal selenium supplementation was negatively associated with anxiety-like behavior in pups and positively associated with cognitive function in adulthood (Laureano-Melo et al. 2015). In a Chinese study (n=927), cord blood selenium concentrations were positively associated with children’s neonatal behavioral neurological assessments scores (reflexes, passive tone, active tone, behavior, and general assessment) at 3 d of age (Yang et al. 2013). In addition, we recently showed that maternal erythrocyte selenium (Ery-Se) in late pregnancy was positively associated with children’s cognitive function at 1.5 y of age in rural Bangladesh (n=750; Skröder et al. 2015). Subsequently, similar associations were indicated between maternal plasma selenium in pregnancy and Polish children’s psychomotor functions within the first years of life (n=410; Polanska et al. 2016). A study evaluating potential effects of polychlorinated biphenyls, lead, and mercury on neuromotor function in Inuit preschool children (n=110; Després et al. 2005) reported no observed beneficial effect of blood selenium on the outcomes. However, these children all had very high concentrations of blood selenium [whole blood average concentration of 402 μg/L, corresponding to intakes above the upper limit of 150 μg/d for 4–8-y-old children (IOM 2000)]. Thus, there are clear gaps in knowledge about both beneficial intake levels of selenium at different ages and potential effects of early-life selenium on outcomes in older children. In the present study, we followed up Bangladeshi children who were previously assessed at 1.5 y of age to determine whether prenatal and concurrent selenium exposures were associated with cognitive function at 5 and 10 y of age.

Methods

Study Area and Subjects

The study area and participants have been described in detail in our previous publications (Hamadani et al. 2011; Kippler et al. 2012; Skröder et al. 2015; Gustin et al. 2017; Rahman et al. 2016). Briefly, the study area is Matlab, a rural area situated 53 km southeast of Dhaka, in Bangladesh. In this area, the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b), has been collecting information via a health and demographic surveillance system (HDSS) over the past 60 y. Community health workers visit all households in Matlab on a monthly basis and collect information about vital events such as births, deaths, and in- and out-migration; this information is used to continuously update the HDSS. The present cohort was nested in a population-based randomized maternal and infant nutrition intervention study (MINIMat; trial number ISRCTN16581394, http://www.isrctn.com/ISRCTN16581394) that aimed to assess the effects of food and micronutrient supplementation during pregnancy (Persson et al. 2012). Women were recruited in early pregnancy [at approximately gestational week (GW) 8] between November 2001 and October 2003 and were randomly assigned to begin one of three daily micronutrient supplements at the 14-wk visit: 30 mg iron and 400 μg folic acid; 60 mg iron and 400 μg folic acid [the standard supplement for pregnant women recommended by the World Health Organization (WHO 2002)]; or 30 mg iron and 400 μg folic acid plus 13 additional micronutrients (including 65 μg of selenium in the form of sodium selenite). Micronutrient supplementation was combined with food supplementation (608 kcal of energy and 18 g protein per day, 6 d/wk) that women were randomly invited to begin at enrollment (early supplementation, GW9) or at the time of their choosing (usual care, typically initiated at approximately GW20), for a total of six intervention groups (Persson et al. 2012). We have previously reported that maternal selenium status in late pregnancy (GW30) was similar for women in all six supplement groups (n=729; Skröder et al. 2015). Eligibility criteria for enrollment in MINIMat were viable fetus and gestational age <14 wk at the first visit to the icddr,b health care facility (approximately GW8–10; both assessed by ultrasound examination), no severe illness, and written consent for participation (Persson et al. 2012).

For the present study, we invited children who were born between October 2002 and November 2003, who participated in the developmental follow-up at 1.5 and 5 y of age, and who were alive and registered as residents in the study area at 10 y of age (n=1,607). Of these, 1,530 (95%) agreed to participate (Figure 1). The main reasons for not participating were out-migration and parents’ refusal. In total, 1,408 children had data for maternal selenium in erythrocytes at GW14 and hemoglobin (Hb; required for adjustment of selenium concentrations in erythrocytes) and complete outcome and covariate data (described below) at 5 or 10 y. Longitudinal analyses of associations with prenatal selenium included 1,260 children with outcomes measured at 5 y and 1,408 children with outcomes measured at 10 y. Out of these children, cross-sectional analyses included those who had selenium concentrations measured in urine at 5 y (n=1,234) or in urine and hair at 10 y (n=1,330; Figure 1). Reasons for missing selenium measurement at any time point included refusal (to provide urine, blood, or hair), no hemoglobin measurement, or an inadequate blood or hair sample for analysis. Longitudinal analyses of associations between selenium concentrations at 5 y and outcomes at 10 y included 1,167 children with all of the abovementioned measurements. In our previous study, we used selenium concentrations in blood collected at GW30 (Skröder et al. 2015), but because more women had donated a blood sample at GW14, we chose to analyze the GW14 samples for the present study, which resulted in a much larger sample size.

Flow chart.
Figure 1. Flow chart for recruitment into the present study. GW, gestational week; HAZ, height-for-age z-score; HOME, modified version of Home Observation for Measurement of the Environment; SES, socioeconomic status.

The project was approved by the research and ethical review committees at icddr,b as well as by the Regional Ethical Review Board, Stockholm, Sweden, and was conducted in accordance with the Helsinki Declaration. The mothers gave their written consent at recruitment to MINIMat and again before the developmental testing of the children at 5 and 10 y of age (Hamadani et al. 2011; Persson et al. 2012).

Element Analyses

Erythrocyte selenium (Ery-Se) has been suggested to be a suitable biomarker of long-term selenium status (Ashton et al. 2009) and was used as such in both our previous study (follow-up at 1.5 y; Skröder et al. 2015) and the present study. Mothers’ blood samples were collected in 5.5-mL Li-heparin tubes at GW14. Blood collection was performed at health care facilities in 2001–2003, whereupon the samples were transferred to the hospital laboratory for separation of plasma and erythrocytes. Selenium in erythrocytes (stored at −40 to −20°C) was measured during 2014–2015 using inductively coupled plasma mass spectrometry (ICP-MS; Agilent 7700x, Agilent Technologies) with an octopole reaction system operated in hydrogen mode (Lu et al. 2015). Before the analyses, approximately 0.2 mL of erythrocytes were diluted 1:25 in an alkali solution [2% (wt:vol) 1-butanol, 0.05% (wt:vol) ethylenediaminetetraacetic acid (EDTA), 0.05% (wt:vol) Triton X-100, 1% (wt:vol) ammonium hydroxide (NH4OH and 20 μg/L internal standard; Sigma-Aldrich]. The samples were vortex mixed, sonicated for 5 min, and centrifuged at 179×g for 2 min [MSE centrifuge, Super Minor, MSE UK]. The limit of detection (LOD) was <0.002 μg/L, and no samples were below this limit. Quality control was performed by including commercial reference materials for blood (Seronorm™ Trace Elements Whole Blood L-1, lot 1103128, and L-2, lot 1103129), and the obtained results were in good accord with the recommended values (see Table S1). Some of the blood samples included in the present study (n=426) were analyzed in 2007 (after being stored frozen at −20°C or −40°C since collection in 2001–2003) via ICP-MS following acid digestion; details of this method and of the related quality control have been described in detail elsewhere (Kippler et al. 2009). The results from the two analytical methods were highly correlated, although the selenium concentrations from the alkali method were consistently 10% lower than those from the acidic method (Lu et al. 2015). Therefore, to improve the comparability of results across both methods, selenium concentrations derived using the acidic method were multiplied by 0.90. To adjust for differences in hematocrit, Ery-Se was expressed in micrograms per gram Hb (Skröder et al. 2015; Stefanowicz et al. 2013). In addition to selenium, we also measured concentrations of arsenic, cadmium, lead, zinc, and manganese in erythrocytes as described in detail elsewhere (Kippler et al. 2009; Lu et al. 2015).

We recently reported that selenium concentrations in hair correlated positively with concentrations in erythrocytes (Spearman’s rank correlation coefficient of 0.54 for samples representing the time of blood collection) in children from the same study population (Skröder et al. 2017). Therefore, we also used hair selenium concentrations to assess selenium status at the 10-y follow-up.

Hair samples were cut from the occipital part of the children’s heads using 18/8 stainless steel scissors and were transported in high-quality paper envelopes to Karolinska Institutet (Stockholm, Sweden) for analyses (Skröder et al. 2017). A sample of 2 cm of hair (closest to the scalp) was washed in 2% Triton X-100 solution for 1 h, rinsed 10 times with deionized water, and dried for 24 h at room temperature (∼23°C). Samples (approximately 50 mg) were weighed close to an alpha source (placed inside the analytical balance) to decrease static electricity by neutralizing the material. Then, samples were digested in a microwave digestion system (UltraCLAVE, Milestone Inc.) with 2 mL of concentrated nitric acid (Scharlau Trace Analysis Grade; Scharlab) and 3 mL deionized water for 30 min at 250°C and a pressure of 4,000 kPa. After cooling to a temperature below 30°C, the digested solutions were diluted with deionized water to an acid concentration of 20% and were analyzed using ICP-MS (Agilent 7700x, Agilent Technologies). Selenium concentrations measured in a reference hair sample (NCSZC81002b, China National Analysis Center for Iron and Steel) were consistent with the recommended value [means of 593 μg/kg and 590 μg/kg, respectively (Skröder et al. 2017)]. We also repeatedly measured an in-house hair sample (from multiple residents of the Faroe Islands, homogenized in liquid nitrogen in a cryogenic homogenizing system) to assess measurement consistency over the year it took to analyze all of the study hair samples (coefficient of variation=12%). The LOD was 0.022 μg/kg, and no samples had a concentration below this value. In addition, we analyzed the concentration of mercury in hair as described elsewhere (Gustin et al. 2017).

Urinary selenium (U-Se) is a measure of recent intake (within 3 d; Hawkes et al. 2008), in contrast to Ery-Se, which is a marker of average intake over approximately 2–3 months owing to the long life span of erythrocytes (Nève 1995), and hair selenium, which represents average intake over approximately one month per centimeter counting from the scalp (Lemire et al. 2009; Skröder et al. 2017). However, because urine was the only available biological medium at 5 y, we measured selenium in urine samples collected at both 5 and 10 y of age. Children’s urine was collected at 5 and 10 y of age in combination with home interviews (Hamadani et al. 2011). All samples were frozen and were transported to Karolinska Institutet for analysis. The samples were diluted 1:10 in 1% nitric acid, and the selenium concentrations were measured using ICP-MS as described above for erythrocytes and as previously described for urine in younger children from the same study population (Skröder Löveborn et al. 2016). All samples were above the LOD (0.007 μg/L at 5 y and 0.013 μg/L at 10 y). Quality control was performed by including commercial reference materials for urine [Seronorm™ Trace Elements Urine Blank OK4636, Seronorm™ Trace Elements Urine NO2525, Seronorm™ Trace Elements Urine 1011644, Seronorm™ Trace Elements Urine 1011645 and National Institute of Standards and Technology (NIST) Standard Reference Material® 2670a], and the results for these samples were in line with recommended values (see Table S1). We also measured concentrations of arsenic, cadmium, and lead in urine at both 5 and 10 y, and we measured concentrations of manganese in drinking water collected at the 10-y follow-up (Kippler et al. 2012, 2016). All concentrations in urine were adjusted to the average specific gravity (1.012) to compensate for variations in dilution (Nermell et al. 2008).

Outcome Assessment

Children’s cognitive function at 5 y was assessed using the third edition of the Wechsler Preschool and Primary Scale of Intelligence™ (WPPSI-III) at the nearest health care facility (Wechsler 2002). WPPSI-III was slightly modified for use in Bangladeshi children (Hamadani et al. 2011). We used seven subtests of WPPSI-III: information (0–34 points), vocabulary (0–43 points), and comprehension (0–38 points) were summed to form the verbal score. Block design (0–40 points), matrix reasoning (0–29 points), and picture completion (0–32 points) formed the performance score. The verbal and performance scores were then summed together with the processing speed (Coding; 0–65 points), resulting in the full developmental score. Seven testers were trained for conducting the WPPSI-III and were rotated across the four health care facilities to minimize tester-related bias. A supervisor rated 10% of all tests and found adequate interobserver reliability (interobserver reliability kappa>0.92; Kippler et al. 2012).

At 10 y, the Wechsler Intelligence Scale for Children®, fourth edition (WISC-IV), was used as described previously (Rahman et al. 2016). This test was translated to the Bengali language and was culturally adapted to fit the present population with slight changes in the questions. Four female psychologists were trained for six weeks on the included tests before the children’s assessment. Interrater reliabilities were measured between each tester and trainer, and training continued until >85% agreement was achieved (Rahman et al. 2016). The test generates four scales: verbal comprehension (based on vocabulary, information, and comprehension), perceptual reasoning (based on block design, picture concept, and matrix reasoning), working memory (based on digit span and arithmetic), and processing speed (based on coding and symbol search). Higher score for processing speed indicates faster response time. In addition, the full developmental score, which represents the child’s general intellectual ability, was calculated as the sum of the raw scores of the subscales. We used the raw scores from each test to avoid systematic differences between our study population and most populations used as the basis for standardization.

Covariates

Information about the mothers’ background characteristics was obtained at enrollment in MINIMat [maternal age, body mass index (BMI), education, parity; obtained at approximately GW8] or from the HDSS [socioeconomic factors such as assets and housing used to derive the socioeconomic status (SES)]. Family SES was assessed via a wealth index based on information on items such as family ownership of assets, housing structure, and dwelling characteristics (Gwatkin et al. 2000). This information was updated at the follow-up at 10 y. Maternal and paternal education was expressed as the number of years of formal schooling and was updated at the follow-ups at both 5 and 10 y. In addition, Raven’s Coloured Progressive Matrices™ test was performed when mothers brought their children to the health care facilities at the 5-y follow-up and was used to assess maternal nonverbal IQ in terms of abstract logical reasoning (Hamadani et al. 2011;Raven et al. 2003; n=1,530). To assess the quality and quantity of children’s stimulation at home, we used a modified version of Home Observation for Measurement of the Environment (HOME) at 5 and 10 y (Bradley et al. 2003; Caldwell 1967). The HOME consists of questions regarding responsibility, encouragement of maturity, emotional climate, learning materials and opportunities, enrichment, family companionship, family integration, and physical environment. The questions were translated into the Bengali language, and seven questions were dropped because the mothers could not understand them owing to cultural irrelevance. Children’s birth anthropometry was measured by the attending nurse, or by trained paramedics for those having home delivery, using standard methods (Persson et al. 2012). At each developmental testing taking place at the health care facilities, the children’s height was measured using a stadiometer (seca 214, Leicester Height Measure; seca), and weight was measured using a digital scale (Tanita HD–318; Tanita). These measures were converted to age- and gender-standardized z-scores (height-for-age, HAZ; weight-for-age, WAZ; BMI-for-age, BAZ) using the WHO growth references (de Onis et al. 2007; WHO 2010). At the follow-up at 5 y, information on the type of school [none, primary, Madrasa (Islamic), kindergarten, Maktab, or nonformal] was collected, and this information was updated at the follow-up at 10 y [none, public primary, Madrasa, nongovernmental organization (NGO; nonprofit private), or English medium (private)]. Finally, at the 10-y follow-up, we also collected information about the number of years of children’s formal education and the number of children in the family; we also measured children’s Hb in peripheral blood (finger prick) using a HemoCue photometer (Hemocue AB).

Statistical Analysis

Statistical analyses were performed using Stata (version 12, StataCorp). A p-value <0.05 was considered statistically significant. Because U-Se values at 5 and 10 y of age were somewhat right skewed, Spearman correlation was used to assess bivariate associations between continuous variables (Ery-Se, U-Se, hair selenium, maternal IQ, age, BMI, parity, maternal and paternal education, SES, Hb, gestational age and weight at birth, HOME, HAZ, WAZ, and BAZ), whereas associations between categorical variables [gender, tester, type of school, paternal job status, and selenium concentrations above or below cut-offs (see below)] were assessed using the Mann-Whitney U-test or the Kruskal-Wallis, chi-squared, or Fisher’s exact test. The seven testers at the 5-y follow-up were grouped into three categories based on scoring owing to a low number of children scored by some of the testers.

We visually examined all evaluated associations by plotting the outcomes against selenium concentrations measured in the different types of samples. Associations that appeared to be more or less linear were modeled using multivariable-adjusted linear regression: specifically, associations between prenatal selenium concentrations (Ery-Se) and verbal and performance scores at 5 y (see Figure S1); between prenatal selenium concentrations (Ery-Se) and verbal comprehension, working memory, perceptual reasoning, and processing speed at 10 y (see Figure S1); and between prenatal selenium concentrations (Ery-Se) and full developmental score at 5 and 10 y (Figure 2; see also Figure S1). Associations that appeared to be nonlinear were modeled using multivariable-adjusted linear spline models: specifically, associations between 5-y U-Se and all outcomes at 5 and 10 y (Figure 2; see also Figure S2); between 10-y U-Se and all outcomes at 10 y (Figure 2; see also Figure S3); and between 10-y hair selenium and all outcomes at 10 y (Figure 2; see also Figure S2). Spline models included single knots at 34 μg/L for both 5- and 10-y U-Se and at 665 μg/kg for 10-y hair selenium, which were generally consistent with the inflection points for the exposure–outcome curves for all outcomes (see Figures S2, S3, and S4). The choice of linear or linear spline models was also based on the R2 of the models, and differences between models were assessed with the F-test for models including the full developmental score. The linear spline regression models had significantly higher R2 values for the associations between U-Se at 5 y and full developmental score at 5 y (p=0.041) and between hair selenium at 10 y and full developmental score at 10 y (p=0.0062). For U-Se at or 10 y and full developmental score at 10 y, there were no statistical differences between the linear spline regression models and the linear models (p=0.84 and 0.33, respectively); thus, we chose to show the results from the linear spline regression analyses for ease of presentation.

Figure 2A is a scatter plot with a regression line plotting full developmental score at 5 years (y-axis) across erythrocyte Se at GW14 in micrograms per gram Hb (x-axis). Figure 2B is a scatter plot with a regression line plotting full developmental score at 10 years (y-axis) across erythrocyte Se at GW14 in micrograms per gram Hb (x-axis). Figure 2C is a scatter plot with a regression line plotting full developmental score at 5 years (y-axis) across urinary Se at 5 years in micrograms per liter (x-axis). Figure 2D is a scatter plot with a regression line plotting full developmental score at 10 years (y-axis) across urinary Se at 5 years in micrograms per liter (x-axis). Figure 2E is a scatter plot with a regression line plotting full developmental score at 10 years (y-axis) across urinary Se at 10 years in micrograms per liter (x-axis). Figure 2F is a scatter plot with a regression line plotting full developmental score at 10 years (y-axis) across Hair Se at 10 years in micrograms per liter (x-axis).
Figure 2. Scatter plots with smoothed lowess lines for prenatal and childhood selenium (Se) and full developmental score (raw score) at 5 and 10 y. (A) Erythrocyte Se at gestational week (GW) 14 [μg/g hemoglobin (Hb)] and full developmental score at 5 y, (B) erythrocyte Se at GW14 (μg/g Hb) and full developmental score at 10 y, (C) urinary Se at 5 y (μg/L) and full developmental score at 5 y, (D) urinary Se at 5 y (μg/L ) and full developmental score at 10 y, (E) urinary Se at 10 y (μg/L) and full developmental score at 10 y, (F) hair Se at 10 y (μg/kg) and full developmental score at 10 y. The vertical lines in the lower panels represent the turning point used for the knot in the linear spline regression analyses (34 μg/L for urinary Se at 5 and 10 y and 665 μg/kg for hair Se at 10 y).

We used complete subject analyses and built the regression models using backward elimination. In the initial step, we included covariates that are known to influence any of the outcomes (gender and age; a priori) and variables that correlated (Spearman, Mann-Whitney, Kruskal-Wallis, chi-squared, or Fisher’s exact test p<0.05) with selenium concentrations at any time point (GW14, 5 y, or 10 y) and any of the outcomes at 5 or 10 y (eight outcomes in total). The initial covariates selected were gender and child age at testing, maternal age, BMI, parity, maternal and paternal education, paternal job status, and SES (all collected at enrollment); Hb at GW14; gestational age and weight at birth; birth order at 5 y, tester, HOME, type of school attended, HAZ, WAZ, and BAZ (all collected at 5 and 10 y); SES, years of schooling, Hb, and number of children in the family (all collected at 10 y). We then used backward elimination of covariates that did not improve the models and eliminated variables that were not significant (p>0.05), starting with the variable with the highest p-value. The eliminations continued as long as they increased the model R2 for at least one of the outcomes assessed with that model. Two separate multivariable-adjusted models were built for predicting full developmental scores at 5 and 10 y, using Ery-Se at GW14 as the exposure variable. In the model for full developmental score at 5 y, the initial set of potential covariates included all the variables listed above that were collected at enrollment, birth, and at 5 y. For the model at 10 y, variables collected at enrollment, birth, and at 10 y were included. When two potential covariates were highly correlated (rS≥0.60), we selected the covariate that resulted in the greatest improvement of the model R2 to reduce the potential for collinearity. We did not adjust for maternal education because it was highly correlated with SES (rS=0.60) and paternal education (rS=0.60), and adjusting for paternal versus maternal education resulted in a higher R2 value.

Thus, primary models (Model 1) for outcomes assessed at 5 y, including either longitudinal associations with prenatal selenium (Ery-Se at GW14) or cross-sectional associations with U-Se at 5 y, were adjusted for gender and age at testing, parity at enrollment (number of children), family SES at enrollment (scale from −5 to 5), maternal Hb at GW14 (grams/deciliter), birth weight (grams); and, based on information obtained at the 5-y examination, tester (5 categories), HOME score (continuous), HAZ (continuous), school type (none/primary/madrasa/kindergarten/maktab/nonformal), mothers’ cognitive function (Raven’s Coloured and Progressive Matrices™ test, continuous), and paternal education (years of formal schooling).

The longitudinal associations of prenatal selenium and outcomes assessed at 5 y were additionally adjusted for erythrocyte manganese and zinc at GW14 (micrograms/kilogram; Model 2). Because these elements were not measured at 5 y, there was no Model 2 for the cross-sectional analyses at 5 y. In an additional step (Model 3), associations with outcomes at 5 y of age were additionally adjusted for arsenic, cadmium, and lead concentrations in erythrocytes at GW14 (micrograms/kilogram, for the longitudinal analyses) or in urine at 5 y (micrograms per liter, for the cross-sectional analyses) because the antioxidative effects of selenium might be affected by these pro-oxidant neurotoxicants (Bergkvist et al. 2010; Hamadani et al. 2011; Kippler et al. 2012). We did not adjust any model for the nutritional intervention groups (one arm of which included a multiple micronutrient capsule that contained selenium) because we used the selenium concentrations measured at baseline (GW14), that is to say, before the supplementation started.

The primary models for outcomes measured at 10 y (Model 1) and associations with exposures at all three time points (i.e., longitudinal associations with prenatal Ery-Se or with U-Se at 5 y of age and cross-sectional associations with U-Se or hair selenium at 10 y of age) included gender and age at testing; gestational age at birth (in weeks); mothers’ cognitive function at 5 y; and, based on information obtained at the 10-y examination, number of children in the family, tester (3 categories), HOME score, HAZ, school type (none/madrasa/NGO/primary/English medium), SES at 10 y (scale from −10 to 10), years of schooling, and paternal education. Model 2 was additionally adjusted for either erythrocyte manganese and zinc at GW14 (for longitudinal associations with prenatal selenium) or water manganese measured at 10 y (micrograms per liter; for cross-sectional associations with U-Se or hair selenium at 10 y). Model 3 included all Model 1 covariates plus either erythrocyte arsenic, cadmium, and lead concentrations at GW14 (micrograms per kilogram; for longitudinal associations), or urinary arsenic, cadmium, and lead concentrations at 5 y (micrograms per liter; for longitudinal associations with U-Se at 5 y), or hair mercury (micrograms per kilogram) and urinary arsenic, cadmium, and lead concentrations at 10 y (micrograms per liter; for cross-sectional associations). We checked that the residuals of all regression models were normally distributed using quantile-quantile plots and residual versus fitted plots.

In our previous study of prenatal selenium status (measured at GW30, in contrast with GW14 in the present analysis) and measures of cognitive function at 1.5 y of age in children from the same study population (Bayley Scales of Infant Development® and measures of comprehensive and expressive language development based on principles of the MacArthur Communicative Development Inventory; n=750), we noted some evidence of variation in associations between girls and boys (Skröder et al. 2015). Therefore, we also performed a qualitative comparison of associations stratified by gender in the present analysis. In addition, we assessed whether there were any statistical differences between the estimates for girls and boys using the Wald test.

Results

General Characteristics and Selenium Status

The main characteristics of the mothers and children included in the study, by girls and boys (n=1,408) or by hair selenium above and below 665 μg/kg (n=1,330), are shown in Table 1 (see also Table S2 for U-Se above and below (34 μg/L) at 5 (n=1,234) and 10 y (n=1330). The mean ages at developmental testing were 5.4 (5th–95th percentile: 5.3–6.4) and 9.5 (9.4–9.7) y. In general, the children were lean and short, with approximately 33% being stunted [<−2 SD of HAZ, WHO cut-off (WHO 2010)] at 5 y and 28% at 10 y, and approximately 40% were underweight (<−2 SD of WAZ, WHO cut-off) at both 5 and 10 y of age. Girls appeared to be more underweight at both 5 and 10 y (45% vs. 40% underweight girls vs. boys, respectively, at both 5 and 10 y). In addition, girls scored lower than boys in the cognitive subtests at 10 y, but there was no difference in the scores at 5 y. At 5 y, 288 of the 1,408 children (20%; 127 girls and 161 boys) did not attend any school; at 10 y, only 8 children (0.6%; 5 girls and 3 boys) did not go to school.

Table 1. Main characteristics of mothers and children included in the study by girls and boys (n=1,408) and hair selenium below and above 665 μg/kg (n=1,330).
Characteristic n All (n=1,408)
Mean±SD
Girls (n=671) Mean±SD Boys (n=737) Mean±SD p-Valuea Hair Se<665 μg/kg (n=1,296) Mean±SD Hair Se ≥665 μg/kg (n=34) Mean±SD p-Valuea
Mothers
 Parity at GW8b 1,408 1.5±1.4 1.5±1.4 1.4±1.4 0.30 1.5±1.4 1.2±1.6 0.17
 SES at GW8 1,408 −0.14±2.3 −0.17±2.3 −0.12±2.3 0.67 −0.14±2.3 0.50±1.9 0.11
 Ery-Se at GW14 (μg/g Hb) 1,408 0.45±0.11 0.45±0.11 0.45±0.11 0.41 0.45±0.11 0.51±0.15 0.012
 Ery-Zn at GW14 (μg/kg) 1,408 8156±2303 8190±2384 8124±2228 0.70 8145±2302 8389±2036 0.62
 Ery-Mn at GW14 (μg/kg) 1,408 20±7.5 20±7.4 20±7.5 0.66 20±7.5 19±5.1 0.62
 Ery-As at GW14 (μg/kg) 1,408 7.6±8.1 7.4±7.4 7.7±8.7 0.55 7.5±8.1 6.0±5.4 0.43
 Ery-Cd at GW14 (μg/kg) 1,408 1.1±0.69 1.0±0.72 1.1±0.66 0.12 1.1±0.69 0.96±0.49 0.68
 Ery-Pb at GW14 (μg/kg) 1,408 77±45 76±44 78±1.6 0.59 77±45 76±32 0.77
 Raven’s score at 5 y 1,408 25±12 25±12 25±12 0.28 25±12 27±12 0.24
 Education at 5 y (y) 1,408 4.6±4.0 4.6±4.0 4.6±4.0 0.73 4.6±4.0 5.9±3.9 0.049
 Education at 10 y (y) 1,408 5.1±3.7 5.1±3.7 5.2±3.7 0.63 5.1±3.7 6.2±3.4 0.10
 Paternal educationat 5 y 1,399 5.0±4.5 4.9±4.5 5.2±4.4 0.20 5.0±4.5 6.5±3.8 0.045
 Paternal education at 10 y 1,408 5.4±4.3 5.3±4.3 5.5±4.3 0.42 5.4±4.3 6.6±3.8 0.075
Children at birth
 Birth weight (g) 1,327 2690±390 2640±370 2740±400 <0.001 2700±390 2700±460 0.88
 Gestational age at birth (wk) 1,408 39±2.1 39±2.1 39±2.2 0.012 39±2.2 40±1.2 0.027
Children at 5 years
 Age (y) 1,408 5.4±0.13 5.4±0.13 5.4±0.13 0.42 5.4±0.13 5.4±0.16 0.41
 HAZ (z-score) 1,408 −1.6±0.92 −1.6±0.88 −1.6±0.96 0.77 −1.6±0.92 −1.3±1.1 0.17
 HOME 1,344 8.6±4.8 8.7±4.8 8.5±4.9 0.37 8.5±4.8 11±5.1 0.0024
 Tester (%) 1,408 0.45 0.81
  Tester 1 8 9 7 8 6
  Tester 2 78 77 79 78 76
  Tester 3 14 14 14 14 18
 School type (%) 1,408 0.059 0.16
  None 20 18 22 20 29
  Primary 23 23 23 23 26
  Madrasa 5 4 4 5 6
  Kindergarten 6 6 6 6 12
  Maktab 29 27 31 29 21
  Nonformal 17 20 14 17 6
 U-Se (μg/L) 1,357 14±6.6 13±6.3 15±6.7 <0.001 14±6.4 17±9.5 0.033
 U-As (μg/L) 1,357 101±118 100±115 102±120 0.66 102±119 84±85 0.33
 U-Cd (μg/L) 1,357 0.29±0.28 0.31±0.33 0.27±0.23 0.058 0.29±0.28 0.24±0.15 0.46
 U-Pb (μg/L) 1,357 4.6±3.5 4.7±3.7 4.5±3.3 0.31 4.6±3.6 4.4±2.7 0.92
 Full developmental score 1,408 79±22 80±22 79±22 0.29 79±22 82±24 0.46
 Verbal score 1,408 33±11 33±11 33±11 0.57 33±11 33±18 0.54
 Performance score 1,408 34±7.8 34±7.7 34±8.0 0.69 34±7.8 35±8.8 0.39
Children at 10 y
 Age (y) 1,408 9.5±0.095 9.5±0.090 9.5±0.10 0.96 9.5±0.095 9.5±0.072 0.86
 HAZ (z-score) 1,408 −1.4±0.94 −1.5±0.95 −1.4±0.93 0.33 −1.4±0.94 −1.2±1.1 0.24
 HOME 1,408 27±4.9 27±4.8 27±5.1 0.0063 27±4.9 29±6.2 0.15
 Tester (%) 1,408 0.14 0.59
  Tester 1 26 25 28 27 18
  Tester 2 25 27 23 25 32
  Tester 3 24 22 25 23 26
  Tester 4 25 26 24 25 24
 School type (%) 1,408 <0.001 0.93
  None 10 5 15 11 9
  Madrasa 3 3 3 3 0
  NGO 77 84 70 77 79
  Primary 9 7 12 9 12
  English medium 1 1 0 0 0
 Years of schooling 1,408 3.1±1.0 3.2±0.97 2.9±1.1 <0.001 3.1±1.0 3.3±1.0 0.14
 SES 1,408 0.0012±2.6 0.028±2.6 −0.023±2.5 0.98 −0.017±2.6 0.79±3.1 0.055
 U-Se (μg/L) 1,403 15±6.1 14±5.7 15±6.3 <0.001 15±6.1 17±6.9 0.0088
 U-As (μg/L) 1,403 107±123 110±125 104±121 0.96 106±123 113±99 0.47
 U-Cd (μg/L) 1,403 0.29±0.22 0.30±0.23 0.28±0.21 0.078 0.29±0.22 0.24±0.15 0.13
 U-Pb (μg/L) 1,403 1.9±1.5 1.9±1.3 1.9±1.6 0.39 1.9±1.3 2.0±1.2 0.54
 Hair Se (μg/kg) 1,334 487±84 478±85 496±82 <0.001 482±77 701±41 <0.001
 Hair Hg (μg/kg) 1,334 799±684 807±473 791±840 0.083 803±689 661±449 0.037
 Water Mn (μg/L) 1,408 895±1164 942±1247 852±1082 0.90 885±1144 1017±1459 0.84
 Full developmental score 1,408 133±33 131±33 134±34 0.16 132±33 138±44 0.32
 Verbal comprehension 1,408 37±11 36±10 37±11 0.043 37±10 39±14 0.29
 Perceptual reasoning 1,408 32±12 30±11 33±12 <0.001 32±11 34±16 0.53
 Working memory 1,408 30±6.2 29±6.0 30±6.2 <0.001 30±6.1 30±8.5 0.38
 Processing speedc 1,408 34±12 36±12 33±12 <0.001 34±12 34±12 0.63

Note: Ery-As, erythrocyte arsenic; Ery-Cd, erythrocyte cadmium; Ery-Mn, erythrocyte manganese; Ery-Pb, erythrocyte lead; Ery-Se, erythrocyte selenium; Ery-Zn, erythrocyte zinc; GW, gestational week; Hair Hg, hair mercury; Hair Se, hair selenium; HAZ, height-for-age z-score; HOME, quality and quantity of children’s stimulation at home assessed using a modified version of Home Observation for Measurement of the Environment; NGO, nongovernmental organization; SES, socioeconomic status assessed via a wealth index based on information on family ownership of e.g. assets, housing structure, and dwelling characteristics (Gwatkin DR et al. 2000); U-As, urinary arsenic; U-Cd, urinary cadmium; U-Pb, urinary lead; U-Se, urinary selenium; Water Mn, water manganese.

aMann-Whitney U-test, chi-squared, or Fisher’s exact test.

bGW8 corresponds to enrollment into Maternal and Infant Nutrition Interventions in Matlab (MINIMat).

cHigher processing speed indicates faster response time.

The girls had approximately 10% lower mean U-Se concentrations at both 5 and 10 y, and they had slightly lower mean hair selenium at 10 y (Table 1). In bivariate analyses, selenium concentrations at all time points were correlated with each other (Table 2). Ery-Se at GW14 correlated with several of the outcomes at both 5 and 10 y, and U-Se at 5 y correlated with all outcomes measured at both 5 and 10 y. Similarly, both U-Se and hair selenium at 10 y correlated with all outcomes at 10 y (Table 2).

Table 2. Spearman correlations [rS (p-value)] for selenium measurements at gestational week 14, 5 y, and 10 y (n=1167−1408).
Ery-Se GW14 U-Se 5 y U-Se 10 y Hair Se 10 y
Hair Se 10 y 0.13 (<0.001) 0.19 (<0.001) 0.23 (<0.001) 1.0
U-Se 10 y 0.069 (0.012) 0.26 (<0.001 )
U-Se 5 y 0.14 (<0.001)
Full developmental score 5 y 0.049 (0.080) 0.13 (<0.001) NA NA
Verbal score 5 y 0.033 (0.24) 0.11 (<0.001) NA NA
Performance score 5 y 0.056 (0.048) 0.10 (<0.001) NA NA
HAZ 5 y 0.019 (0.50) 0.15 (<0.001) NA NA
WAZ 5 y −0.037 (0.19) 0.10 (<0.001) NA NA
Full developmental score 10 y 0.058 (0.031) 0.17 (<0.001) 0.11 (<0.001) 0.12 (<0.001)
Verbal comprehension 10 y 0.025 (0.35) 0.14 (<0.001) 0.12 (<0.001) 0.13 (<0.001)
Perceptual reasoning 10 y 0.050 (0.063) 0.18 (<0.001) 0.10 (<0.001) 0.10 (<0.001)
Working memory 10 y 0.052 (0.090) 0.12 (<0.001) 0.078 (0.0047) 0.11 (<0.001)
Processing speed 10 ya 0.066 (0.013) 0.11 (<0.001) 0.057 (0.039) 0.053 (0.053)
HAZ 10 y −0.0081 (0.6) 0.13 (<0.001) 0.11 (<0.001) 0.071 (0.0099)
WAZ 10 y −0.0023 (0.93) 0.13 (<0.001) 0.094 (<0.001) 0.087 (0.0015)

Note: Ery-Se, erythrocyte selenium; GW, gestational week; Hair Se, hair selenium; HAZ, height-for-age z-score; NA, not applicable; U-Se, urinary selenium; WAZ, weight-for-age z-score.

aHigher processing speed indicates faster response time.

Prenatal Selenium and Children’s Cognitive Function at 5 and 10 y

In the multivariable-adjusted linear regression analyses, Ery-Se at GW14 was positively associated with all outcomes at 5 and 10 y (Table 3), but associations were stronger for the outcomes measured at 10 y than for those measured at 5 y. None of the associations for the outcomes at 5 y was markedly affected by adjusting for essential elements (manganese and zinc in erythrocytes at GW14; Model 2). The estimates for full developmental score, perceptual reasoning, and processing speed at 10 y increased by 18–42% after adjusting for the essential elements; however, the estimates for working memory and verbal comprehension were very similar after the adjustment (Table 3). When adjusting Model 1 for toxic elements (erythrocyte arsenic, cadmium, and lead at GW14; Model 3), the estimates at both 5 and 10 y were not markedly affected.

Table 3. Multivariable-adjusted regression analyses (linear for Ery-Se, linear spline with knot at 34 μg/L for U-Se) for selenium status in early life and cognitive function in childhood.
Outcomes Ery-Se GW14 (per 0.1 μg/g Hb) U-Se at 5 y (<34 μg/L, per 10 μg/L) U-Se at 5 y (≥34 μg/L, per 10 μg/L)
n B (95% CI) p-Value n B (95% CI) p-Value n B (95% CI) p-Value
5 ya
 Model 1 1,260 1,214 20
  Full developmental score 0.94 (0.027, 1.9) 0.044 1.1 (−0.57, 2.8) 0.19 −8.6 (−17, 0.13) 0.054
  Verbal score 0.47 (−0.0072, 0.95) 0.054 0.51 (−0.38, 1.4) 0.26 −4.4 (−8.9, 0.17) 0.059
  Performance score 0.27 (−0.078, 0.62) 0.13 0.033 (−0.62, 0.69) 0.92 −3.2 (−6.6, 0.15) 0.062
 Model 2 1,260
  Full developmental score 0.97 (−0.16, 2.1) 0.094 NA NA
  Verbal score 0.41 (−0.18, 1.0) 0.17 NA NA
  Performance score 0.32 (−0.11, 0.76) 0.14 NA NA
 Model 3 1,260 1,214 20
  Full developmental score 0.99 (0.0051, 2.0) 0.049 1.2 (−0.51, 3.0) 0.17 −8.2 (−17, 0.72) 0.071
  Verbal score 0.46 (−0.054, 0.98) 0.079 0.67 (−0.24, 1.6) 0.15 −3.8 (−8.4, 0.86) 0.11
  Performance score 0.25 (−0.13, 0.63) 0.19 0.056 (−0.62, 0.73) 0.87 −3.1 (−6.6, 0.29) 0.073
10 yb
 Model 1 1,408 1,149 18
  Full developmental score 2.2 (1.0, 3.3) <0.001 2.5 (−0.033, 5.0) 0.053 1.1 (−12, 14) 0.87
  Verbal comprehension 0.50 (0.12, 0.89) 0.010 0.49 (−0.35, 1.3) 0.25 −1.5 (−5.8, 2.8) 0.50
  Perceptual reasoning 0.53 (0.079, 0.98) 0.021 1.2 (0.18, 2.2) 0.020 3.1 (−20, 8.1) 0.23
  Working memory 0.38 (0.14, 0.62) 0.002 0.34 (−0.18, 0.86) 0.20 −0.13 (−2.8, 2.5) 0.92
  Processing speedc 0.75 (0.29, 1.2) 0.001 0.52 (−0.50, 1.5) 0.32 −0.37 (−5.5, 4.8) 0.89
 Model 2 1,408
  Full developmental score 2.6 (1.2, 3.9) <0.001 NA NA
  Verbal comprehension 0.50 (0.051, 0.95) 0.029 NA NA
  Perceptual reasoning 0.75 (0.22, 1.3) 0.006 NA NA
  Working memory 0.39 (0.11, 0.67) 0.006 NA NA
  Processing speedc 0.94 (0.40, 1.5) 0.001 NA NA
 Model 3 1,408 1,149 18
  Full developmental score 2.2 (1.0, 3.4) <0.001 2.5 (−0.12, 5.1) 0.061 1.0 (−12, 14) 0.88
  Verbal comprehension 0.51 (0.10, 0.92) 0.014 0.52 (−0.34, 1.4) 0.24 −1.4 (−5.8, 3.0) 0.53
  Perceptual reasoning 0.56 (0.081, 1.0) 0.022 1.2 (0.18, 2.2) 0.021 3.2 (−2.0, 8.3) 0.23
  Working memory 0.37 (0.12, 0.62) 0.004 0.25 (−0.28, 0.78) 0.35 −0.28 (−3.0, 2.4) 0.84
  Processing speedc 0.78 (0.30, 1.3) 0.002 0.53 (−0.50, 1.6) 0.32 −0.44 (−5.7, 4.8) 0.87

Note: CI, confidence interval; Ery-Se, erythrocyte selenium; GW, gestational week; Hb, hemoglobin; NA, not applicable; U-Se, urinary selenium.

aAnalyses of outcomes at 5 y are adjusted as follows: Model 1 is adjusted for children’s gender, parity and family socioeconomic status (SES) at enrollment, birth weight, Hb at GW14, age at testing, height-for-age z-score (HAZ), modified Home Observation for Measurement of the Environment score (HOME), testers, school type, mothers’ cognitive function, and paternal education (all assessed at 5-y follow-up). Model 2 is Model 1 further adjusted for erythrocyte zinc and manganese at GW14 (prenatal analyses). There is no Model 2 for the concurrent analyses because there were no exposure markers for zinc or manganese available at 5. Model 3 is Model 1 further adjusted for erythrocyte cadmium, lead, and arsenic at GW14 (prenatal analyses) or urinary arsenic, cadmium, and lead at 5 y (cross-sectional analyses).

bAnalyses of outcomes at 10 y are adjusted as follows: Model 1 is adjusted for children’s gender, gestational age at birth, age at testing, number of children in the family, HAZ, HOME, testers, school type and years of schooling, family SES, and paternal education (all assessed at 10 y), and mothers’ cognitive function assessed at 5 y. Model 2 is Model 1 further adjusted for erythrocyte zinc and manganese at GW14 (prenatal analyses). There is no Model 2 for the longitudinal analyses with U-Se at 5 years because there were no exposure markers for zinc or manganese available at 5 y. Model 3 is Model 1 further adjusted for erythrocyte cadmium, lead, and arsenic at GW14 (prenatal analyses) or urinary arsenic, cadmium, and lead at 5 y (longitudinal analyses with U-Se at 5 y) or urinary arsenic, cadmium, lead, and hair mercury at 10 y (cross-sectional analyses).

cHigher processing speed indicates faster response time.

When stratifying Model 1 by gender, we found positive associations for both girls and boys. The estimates for the outcomes at 5 y were marginally stronger for the girls, although they were not significantly different from the estimates for the boys (p>0.25 for all outcomes; see Figure S5). The estimates were also marginally stronger for the girls for all outcomes at 10 y; however, the 95% confidence intervals were largely overlapping, and there was an indicated statistical difference in the estimates for working memory only (p=0.078; Figure 3).

Plot indicating beta coefficients (left y-axis) and beta coefficients for full score (right y-axis) for a 0.1 μg per gram Hb increase in erythrocyte Se at GW14, for boys and girls across verbal comprehension (p for difference between estimates equals 0.30), perceptual reasoning (p equals 0.84), working memory (p equals 0.078), processing speed (p equals 0.96), and full developmental score (p equals 0.42) (x-axis).
Figure 3. Estimates (B-coefficient) and 95% confidence interval (CI) (straight line) for associations between all outcomes at 10 y and erythrocyte selenium (Se) [per 0.1 μg/g hemoglobin (Hb)] at gestational week (GW) 14, stratified by gender (n=671 girls and 737 boys). p-Value for difference between estimates (Wald test). Adjustments: gestational age at birth, age at testing, mothers’ cognitive function assessed at 5 y, number of children in the family, height-for-age z-score (HAZ), modified version of Home Observation for Measurement of the Environment score (HOME), testers, school type and years of schooling, family socioeconomic status (SES), and paternal education (all assessed at 10 y).

Urinary Selenium at 5 y and Children’s Cognitive Function at 5 and 10 y

In the linear spline regression analyses using U-Se at 5 y as a marker of selenium status, we found positive associations for all outcomes at 5 y up to the spline knot at 34 μg/L, although they were not statistically significant (Table 3). Additionally adjusting for toxic elements (arsenic, cadmium, and lead in urine at 5 y) did not improve the associations between U-Se and the outcomes (Model 3, Table 3). Above the spline knot, the associations with outcomes at 5 y were inverse. However, there were only 20 children with such high U-Se concentrations. When stratifying the analyses (Model 1) by gender, we found no statistical difference between the estimates below the spline knot for girls and boys for full developmental score (p=0.49), performance score (p=0.76), or verbal score (p=0.11), or above the spline knot for any outcome (p>0.13 for all outcomes).

Concerning the outcomes at 10 y of age, we found significant, positive associations between U-Se at 5 y (up to the spline knot at 34 μg/L) and perceptual reasoning (Table 3). This association did not change markedly when adjusting for toxic elements (urinary arsenic, cadmium, and lead at 5 y; Model 3, Table 3). After stratifying by gender, we found essentially no differences between girls and boys below the spline knot (p for difference between estimates: full developmental score, p=0.55; verbal comprehension, p=0.92; processing speed, p=0.42; perceptual reasoning, p=0.63; working memory, p=0.88); or above the spline knot (full developmental score, p=0.61; verbal comprehension, p=0.067; processing speed, p=0.84; perceptual reasoning, p=0.89; working memory, p=0.79).

Selenium Status and Children’s Cognitive Function at 10 y

In the multivariable-adjusted linear spline regression analyses with both selenium status and outcomes at 10 y of age, we found positive associations between hair selenium (below the spline knot at 665 μg/kg) and full developmental score, verbal comprehension, and working memory (Table 4). Additionally adjusting for water manganese did not change the estimates markedly (Table 4, Model 2). Furthermore, adjusting Model 1 for concurrent exposure to toxic elements (urinary arsenic, cadmium, and lead and hair mercury; Model 3), did not affect the estimates for hair selenium (<10% change for all outcomes). Above the spline knot, we found significant inverse associations for full developmental score, working memory, and processing speed, even though there were only 34 children with such high concentrations. When using U-Se instead of hair selenium, we found associations in the same directions as for hair selenium below and above the spline knot at U-Se 34 μg/L, although they were not significant (Table 4).

Table 4. Multivariable-adjusted linear spline regression analyses for concurrent selenium status (urinary and hair selenium) and cognitive function at 10 years (n=1,330).
Outcomes U-Se at 10 y, <34 μg/L (per 10 μg/L, n=1316) U-Se at 10 y, ≥34 μg/L (per 10 μg/L, n=14 ) Hair Se at 10 y, <665 μg/kg (per 100 μg/kg, n=1296) Hair Se at 10 y, ≥665 μg/kg (per 100 μg/kg, n=34)
B (95% CI) p-Value B (95% CI) p-Value B (95% CI) p-Value B (95% CI) p-Value
10 y
 Model 1
  Full developmental score 1.2 (−1.3, 3.7) 0.36 −4.7 (−16, 6.2) 0.40 2.2 (0.46, 3.9) 0.013 −21 (−37, −4.9) 0.011
  Verbal comprehension 0.50 (−0.33, 1.3) 0.24 −2.4 (−6.0, 1.2) 0.19 0.79 (0.23, 1.4) 0.006 −3.6 (−8.9, 1.7) 0.18
  Perceptual reasoning 0.44 (−0.54, 1.4) 0.38 −2.5 (−6.8, 1.8) 0.25 0.60 (−0.072, 1.3) 0.080 −4.6 (−11, 1.6) 0.15
  Working memory 0.10 (−0.42, 0.62) 0.70 −0.58 (−2.9, 1.7) 0.62 0.41 (0.054, 0.76) 0.024 −4.6 (−7.9, −1.3) 0.006
  Processing speeda 0.13 (−0.87, 1.1) 0.81 0.77 (−3.6, 5.1) 0.73 0.38 (−0.31, 1.1) 0.28 −8.0 (−14, −1.6) 0.014
 Model 2
  Full developmental score 1.0 (−1.5, 3.6) 0.42 −4.7 (−16, 6.3) 0.40 2.1 (0.34, 3.8) 0.019 −21 (−37, −4.6) 0.012
  Verbal comprehension 0.47 (−0.36, 1.3) 0.27 −2.4 (−6.0, 1.2) 0.19 0.76 (0.19, 1.3) 0.009 −3.5 (−8.8, 1.7) 0.19
  Perceptual reasoning 0.43 (−0.55, 1.4) 0.39 −2.5 (−6.8, 1.8) 0.25 0.60 (−0.077, 1.3) 0.083 −4.6 (−11, 1.6) 0.15
  Working memory 0.070 (−0.45, 0.59) 0.79 −0.56 (−2.8, 1.7) 0.63 0.38 (0.024, 0.74) 0.036 −4.6 (−7.9, −1.3) 0.007
  Processing speeda 0.074 (−0.92, 1.1) 0.88 0.80 (−2.6, 5.2) 0.72 0.33 (−0.36, 1.0) 0.35 −7.8 (1.4, −1.5) 0.015
 Model 3
  Full developmental score 2.1 (−0.55, 4.8) 0.12 −4.1 (−15, 6.9) 0.47 2.1 (0.41, 3.9) 0.015 −21 (−37, −5.1) 0.010
  Verbal comprehension 0.78 (−0.098, 1.7) 0.082 −2.2 (−5.8, 1.4) 0.23 0.76 (0.19, 1.3) 0.009 −3.8 (−9.0, 1.5) 0.16
  Perceptual reasoning 0.69 (−0.35, 1.7) 0.19 −2.3 (−6.6, 2.0) 0.29 0.59 (−0.085, 1.3) 0.087 −4.7 (−11, 1.5) 0.14
  Working memory 0.33 (−0.22, 0.88) 0.24 −0.48 (−2.8, 1.8) 0.68 0.41 (0.052, 0.76) 0.025 −4.6 (−7.9, −1.3) 0.006
  Processing speeda 0.31 (−0.74, 1.4) 0.56 0.91 (−3.5, 5.3) 0.68 0.38 (−0.30, 1.1) 0.28 −8.0 (−14, −1.7) 0.014

Note: Model 1 has been adjusted as follows: children’s gender, age at testing, number of children in the family, gestational age at birth, height-for-age z-score (HAZ), modified version of Home Observation for Measurement of the Environment score (HOME), testers, school type and years of schooling, family socioeconomic status (SES), and paternal education (all assessed at 10 y), and mothers’ cognitive function assessed at 5 y. Model 2 is Model 1 further adjusted for water manganese at 10 y. Model 3 is Model 1 further adjusted for toxic element (arsenic, cadmium, and lead in children’s urine and mercury in children’s hair at 10 y). CI, confidence interval; Hair Se, hair selenium; U-Se, urinary selenium.

aHigher processing speed indicates faster response time.

When stratifying Model 1 by gender, we found positive associations for both girls and boys (below the spline knot at 665 μg/kg), although the estimates were generally stronger for the girls (Figure 4). In particular, the estimate for the association between hair selenium and working memory was 0.49 [95% confidence interval (CI): 0.11, 1.1] for the girls, compared with 0.18 (95% CI: −0.36, 0.72) for the boys. However, differences in estimates were not significant below the spline knot (Figure 4) or above the spline knot (p>0.39 for all outcomes).

Plot indicating beta coefficients (left y-axis) and beta coefficients per 100 μg increase in hair Se at 10 years, below the spline knot at 665 μg/kg for full score (right y-axis) for boys and girls across verbal comprehension (p equals 0.55), perceptual reasoning (p equals 0.58), working memory (p equals 0.28), processing speed (p equals 0.98), and full developmental score (p equals 0.52) (x-axis) p-Values are for the difference between girls and boys.
Figure 4. Estimates (B-coefficient) and 95% confidence interval (CI) (straight line) for associations between all outcomes at 10 y and hair selenium (Se; below spline knot at 665 μg/kg, per 100 μg/kg) at 10 y, stratified by gender (n=637 girls and 659 boys). p-Value for difference between estimates (Wald test). Adjustments: gestational age at birth, age at testing, mothers’ cognitive function assessed at 5 y, number of children in the family, height-for-age z-score (HAZ), modified version of Home Observation for Measurement of the Environment score (HOME), testers, school type and years of schooling, family socioeconomic status (SES), and paternal education (all assessed at 10 y).

Discussion

To our knowledge, this large longitudinal study is the first to assess the impact of both prenatal and childhood selenium status on child development. Together with our previous findings of a positive association between maternal selenium status in pregnancy (Ery-Se) and language comprehension and psychomotor development at 1.5 y of age in a subset of 750 children from the same cohort (Skröder et al. 2015), the present results provide evidence that adequate selenium status is important for long-term brain development. We found positive associations between maternal Ery-Se in early pregnancy (n=1,408) and the outcomes at 10 y in particular, where an increase from the 5th to the 95th percentile (median 0.44 μg/g Hb, corresponding to 148 μg/kg) was associated with an estimated mean increase in full developmental score of 8.1 points (95% CI: 3.8, 13), corresponding to an increase of 0.24 SD. For the cross-sectional evaluation at 10 y (n=1,330), a linear spline model with a single knot (at 665 μg/kg; 98th percentile) revealed that an increase from the 5th to the 95th percentile in hair selenium was associated with an average increase in full developmental score of 5.9 points (95% CI: 1.3, 11), corresponding to 0.18 SD. The direction of the association changed from positive to negative for children with higher hair selenium concentrations (≥665 μg/kg), although estimates for this group were based on only 34 children. Similar patterns were found for children tested at 5 y of age; however, the short-term exposure biomarker (U-Se) used at 5 y resulted in more imprecise associations. Taken together, the results indicate that both prenatal and childhood selenium status influence brain development. The findings seem robust even after adjustment for multiple potential confounders. The estimated selenium-related mean increases in some outcome scores were higher in girls than in boys, consistent with a greater potential benefit in girls, although the estimates were imprecise and the differences were small.

Very few prospective studies have assessed the impact of prenatal selenium status on child cognition. Following our previous study (Skröder et al. 2015), a Polish study (n=410) reported that maternal plasma selenium during the first trimester (mean±SD, 48.3±10.6 μg/L) was positively associated with motor development at 1 y of age and with language development and cognitive function (borderline significant) at 2 y of age (Polanska et al. 2016). Similarly, a Greek study (n=575) showed a borderline positive association between maternal U-Se (23±8.6 μg/L) and children’s cognition (McCarthy Scales of Children’s Abilities) at 4 y of age (Kippler et al. 2016). However, a recent Spanish study suggested an inverse relationship between serum selenium in the first trimester (79.7±7.9 μg/L) and child neuropsychological development at approximately 1 y of age (Bayley Scales of Infant Development; n=650) above 86 μg/L (Amorós et al. 2017). Moreover, a U.S. study found no association between fairly high Ery-Se concentrations during pregnancy (206 μg/L; n=872) and cognitive function at 7.7 y (Oken et al 2016). Considering the marked variation in selenium levels across studies, there appears to be some consistency concerning the importance of selenium for brain development at lower selenium levels. However, further studies are warranted to evaluate potential susceptibility factors.

There are also a few cross-sectional studies concerning childhood selenium status and cognition. In Bangladesh, a positive association was found between blood selenium and motor function in 9.6-y-old children (n=303; Parvez et al. 2011). A Chinese study of 927 newborns showed a positive association between cord serum selenium and scores in the neonatal behavioral neurological assessment at 3 d of age, but only up to 100 μg/L, after which the association turned inverse (n=80; Yang et al. 2013). This pattern suggests that the range of selenium intake that is beneficial for child development may be narrow, as it is for selenium toxicity in general, although this has not been studied extensively (Jablonska and Vinceti 2015; Vinceti et al. 2014). We also observed indications of inverse associations in the cross-sectional analyses of selenium in hair and urine and the different measures of cognitive function at 5 and 10 y in the children with the highest selenium concentrations. However, there were few children above the turning points of hair selenium at 665 μg/kg (range to 822 μg/kg; n=34) and U-Se of 34 μg/L (range to approximately 60 μg/L; n=20 and 18, at 5 and 10 y, respectively). Despite the generally low selenium levels in the study area (see below), there may well be a risk of excess selenium intake through supplementation, particularly because people may not be aware of the narrow therapeutic index of selenium. However, further studies are needed to determine the potential for toxic levels of different chemical forms of selenium intake through supplementation or even through diet and whether safe levels of intake may vary among different populations.

We assessed prenatal selenium status based on selenium concentrations in maternal erythrocytes, which, like hair selenium, represents average intake over a longer time period (approximately 2–3 months; Skröder et al. 2017) than plasma selenium (a few weeks; Nève 1995) or urine (a few days; Hawkes et al. 2008). For a subsample of the study population (n=98) we previously measured plasma selenium in the beginning of the second trimester (Li et al. 2008), and we found a strong correlation with selenium in the erythrocytes [rS=0.69, p<0.001 (Skröder et al. 2015)]. In that sample, approximately 60% had concentrations <60 μg/L, indicating selenium deficiency (Fairweather-Tait et al. 2011), which is in accord with the reported low selenium content in the agricultural soil in Bangladesh (Ahsan et al. 2009). Consequently, we did not have sufficient numbers of highly exposed women to evaluate potential toxic effects of such selenium levels. The children in the present cohort appeared to have somewhat higher selenium status than their mothers, based on the urine and erythrocyte concentrations measured in another subsample of 488 mother–child (9 y of age) pairs (Skröder Löveborn et al. 2016).

One of the suggested mechanisms for a positive effect of selenium on child cognition is protection against oxidative stress through the antioxidative properties of selenoproteins. Selenium plays an important role in the protection of neurons to a large extent through selenoprotein P, which is a major contributor to selenium content in the brain (Takemoto et al. 2010; Schweizer and Fradejas-Villar 2016). In addition, antioxidative glutathione peroxidases have been shown to be highly active in glial cells (Damier et al. 1993; Power and Blumbergs 2009), and thioredoxin reductases appear important for cerebellar function (Schweizer and Fradejas-Villar 2016). In addition, selenium is essential for thyroid function (Roman et al. 2014), which is important for child development.

Strengths of this study include the longitudinal study design with measurements of a wide range of selenium levels both prenatally and at 5 and 10 y, as well as cognitive assessment at both 5 and 10 y. Essentially no women reported smoking or alcohol consumption. Furthermore, we had a large sample size, and we measured multiple exposures using ICP-MS. A limitation of this study is that we had no plasma selenium measured in the children, which could have simplified comparisons of selenium status with other populations.

Conclusion

Our findings show that adequate selenium status in both fetal life and childhood may be beneficial for children’s cognitive function. We also found evidence of adverse effects in children with the highest selenium levels, based on small numbers of observations. Considering the high prevalence of inadequate selenium intake throughout the world (Fairweather-Tait et al. 2011), the results are likely relevant for other populations. The effect size may be of little clinical relevance on an individual level because the impact of selenium on cognitive abilities was small compared with the effects of factors such as SES, schooling, parental IQ, and stimulation at home. However, on a population level, even small decreases are of importance with respect to the proportion of children falling below the limit for intellectual disability. It is also important to stress that selenium deficiency, unlike some of the other abovementioned factors, is preventable, and that the effects of early-life selenium deficiency may persist into adulthood.

Acknowledgments

We thank all participating women and children and everyone involved in the cognitive testing and collection of samples and data. We also thank M. Levi, M. Grandér, B. Palm, B. Nermell, S. Ahmed, S.M. Rahman, and K. Gustin at Karolinska Institutet for measurements of trace elements. All authors read and approved the final manuscript. This work was supported by grants from the Swedish Research Council (grants 521-2013-2269 and 2015-03206), the Swedish Research Council Formas (grant 210-2013-751), the Swedish International Development Cooperation Agency (grant SWE-186), the European Commission [Public health impact of long-term, low-level mixed element exposure in susceptible population strata (PHIME)], and Karolinska Institutet. The Maternal and Infant Nutrition Interventions in Matlab (MINIMat) supplementation trial in pregnancy was funded by UNICEF, Sida, the U.K. Medical Research Council, the Swedish Research Council, the U.K. Department for International Development, the International Centre for Diarrhoeal Disease Research, Bangladesh, the Global Health Research Fund (Japan), the Child Health and Nutrition Research Initiative, Uppsala University, and the U.S. Agency for International Development.

References

Ahsan DA, DelValls TA, Blasco J. 2009. Distribution of arsenic and trace metals in the floodplain agricultural soil of Bangladesh. Bull Environ Contam Toxicol 82(1):11–15, PMID: 18696001, 10.1007/s00128-008-9502-x.

Akbaraly TN, Hininger-Favier I, Carrière I, Arnaud J, Gourlet V, Roussel AM, et al. 2007. Plasma selenium over time and cognitive decline in the elderly. Epidemiology 18(1):52–58, PMID: 17130689, 10.1097/01.ede.0000248202.83695.4e.

al-Deeb S, al-Moutaery K, Bruyn GW, Tariq M. 1995. Neuroprotective effect of selenium on iminodipropionitrile-induced toxicity. J Psychiatry Neurosci 20(3):189–192, PMID: 7786879.

Amorós R, Murcia M, Ballester F, Broberg K, Iñiguez C, Rebagliato M, et al. 2017. Selenium status during pregnancy: influential factors and effects on neuropsychological development among Spanish infants. Sci Total Environ 1:610–611:741–749, PMID: 28822941, 10.1016/j.scitotenv.2017.08.042.

Ashton K, Hooper L, Harvey LJ, Hurst R, Casgrain A, Fairweather-Tait SJ. 2009. Methods of assessment of selenium status in humans: A systematic review. Am J Clin Nutr 89(6):2025S–2039S, PMID: 19420095, 10.3945/ajcn.2009.27230F.

Bergkvist C, Kippler M, Hamadani JD, Grandér M, Tofail F, Berglund M, et al. 2010. Assessment of early-life lead exposure in rural Bangladesh. Environ Res 110(7):718–724, PMID: 20656285, 10.1016/j.envres.2010.07.004.

Bradley RH, Caldwell BM, Corwyn RF. 2003. The child care home inventories: Assessing the quality of family child care homes. Early Child Res Q 18(3):294–309, 10.1016/S0885-2006(03)00041-3.

Caldwell BM. 1967. Descriptive evaluations of child development and of developmental settings. Pediatrics 40(1):46–54, PMID: 6028898.

Damier P, Hirsch EC, Zhang P, Agid Y, Javoy-Agid F. 1993. Glutathione peroxidase, glial cells and Parkinson’s disease. Neuroscience 52(1):1–6, PMID: 8433802, 10.1016/0306-4522(93)90175-F.

de Onis M, Onyango AW, Borghi E, Siyam A, Nishida C, Siekmann J. 2007. Development of a WHO growth reference for school-aged children and adolescents. Bull World Health Organ 85(9):660–667, PMID: 18026621, 10.1590/S0042-96862007000900010.

Després C, Beuter A, Richer F, Poitras K, Veilleux A, Ayotte P, et al. 2005. Neuromotor functions in Inuit preschool children exposed to Pb, PCBs, and Hg. Neurotoxicol Teratol 27(2):245–257, PMID: 15734276, 10.1016/j.ntt.2004.12.001.

Fairweather-Tait SJ, Bao Y, Broadley MR, Collings R, Ford D, Hesketh JE, et al. 2011. Selenium in human health and disease. Antioxid Redox Signal 14(7):1337–1383, PMID: 20812787, 10.1089/ars.2010.3275.

Gao S, Jin Y, Hall KS, Liang C, Unverzagt FW, Ji R, et al. 2007. Selenium level and cognitive function in rural elderly Chinese. Am J Epidemiol 165(8):955–965, PMID: 17272290, 10.1093/aje/kwk073.

Gustin K, Tofail F, Mehrin F, Levi M, Vahter M, Kippler M. 2017. Methylmercury exposure and cognitive abilities and behavior at 10 years of age. Environ Int 102:97–105, PMID: 28216013, 10.1016/j.envint.2017.02.004.

Gwatkin DR, Rustein S, Johnson K, Pande RP, Wagstaff A. 2000. Socio-economic Differences in Health, Nutrition, and Population in Bangladesh. Washington, DC:World Bank.

Hamadani JD, Tofail F, Nermell B, Gardner R, Shiraji S, Bottai M, et al. 2011. Critical windows of exposure for arsenic-associated impairment of cognitive function in pre-school girls and boys: A population-based cohort study. Int J Epidemiol 40(6):1593–1604, PMID: 22158669, 10.1093/ije/dyr176.

Hawkes WC, Richter BD, Alkan Z, Souza EC, Derricote M, Mackey BE, et al. 2008. Response of selenium status indicators to supplementation of healthy North American men with high-selenium yeast. Biol Trace Elem Res 122(2):107–121, PMID: 18193397, 10.1007/s12011-007-8066-7.

Imam SZ, Newport GD, Islam F, Slikker W Jr, Ali SF. 1999. Selenium, an antioxidant, protects against methamphetamine-induced dopaminergic neurotoxicity. Brain Res 818(2):575–578, PMID: 10082851, 10.1016/S0006-8993(98)01311-0.

Institute of Medicine Panel on Dietary Antioxidants and Related Compounds (IOM). 2000. Dietary Reference Intakes for Vitamin C, Vitamin E, Selenium, and Carotenoids: A Report of the Panel on Dietary Antioxidants and Related Compounds, Subcommittees on Upper Reference Levels of Nutrients and Interpretation and Uses of Dietary Reference Intakes, and the Standing Committee on the Scientific Evaluation of Dietary Reference Intakes, Food and Nutrition Board, Institute of Medicine. Washington, DC:National Academy Press.

Jablonska E, Vinceti M. 2015. Selenium and human health: Witnessing a Copernican revolution? J Environ Sci Health C Environ Carcinog Ecotoxicol Rev 33(3):328–368, PMID: 26074278, 10.1080/10590501.2015.1055163.

Kippler M, Bottai M, Georgiou V, Koutra K, Chalkiadaki G, Kampouri M, et al. 2016. Impact of prenatal exposure to cadmium on cognitive development at preschool age and the importance of selenium and iodine. Eur J Epidemiol 31(11):1123–1134, PMID: 27147065, 10.1007/s10654-016-0151-9.

Kippler M, Goessler W, Nermell B, Ekstrom EC, Lonnerdal B, El Arifeen S, et al. 2009. Factors influencing intestinal cadmium uptake in pregnant Bangladeshi women–a prospective cohort study. Environ Res 109(7):914–921, PMID: 19646688, 10.1016/j.envres.2009.07.006.

Kippler M, Tofail F, Hamadani JD, Gardner RM, Grantham-McGregor SM, Bottai M, et al. 2012. Early-life cadmium exposure and child development in 5-year-old girls and boys: a cohort study in rural Bangladesh. Environ Health Perspect 120(10):1462–1468, PMID: 22759600, 10.1289/ehp.1104431.

Laureano-Melo R, Império GE, da Silva-Almeida C, Kluck GE, Cruz Seara Fde A, da Rocha FF, et al. 2015. Sodium selenite supplementation during pregnancy and lactation promotes anxiolysis and improves mnemonic performance in Wistar rats’ offspring. Pharmacol Biochem Behav 138:123–132, PMID: 26364924, 10.1016/j.pbb.2015.09.007.

Lemire M, Mergler D, Huel G, Passos CJ, Fillion M, Philibert A, et al. 2009. Biomarkers of selenium status in the Amazonian context: Blood, urine and sequential hair segments. J Expo Sci Environ Epidemiol 19(2):213–222, PMID: 18446187, 10.1038/jes.2008.14.

Li L, Ekström EC, Goessler W, Lönnerdal B, Nermell B, Yunus M, et al. 2008. Nutritional status has marginal influence on the metabolism of inorganic arsenic in pregnant Bangladeshi women. Environ Health Perspect 116(3):315–321, PMID: 18335097, 10.1289/ehp.10639.

Lu Y, Kippler M, Harari F, Grandér M, Palm B, Nordqvist H, et al. 2015. Alkali dilution of blood samples for high throughput ICP-MS analysis-comparison with acid digestion. Clin Biochem 48(3):140–147, PMID: 25498303, 10.1016/j.clinbiochem.2014.12.003.

Mitchell JH, Nicol F, Beckett GJ, Arthur JR. 1998. Selenoprotein expression and brain development in preweanling selenium- and iodine-deficient rats. J Mol Endocrinol 20(2):203–210, PMID: 9584835, 10.1677/jme.0.0200203.

Nermell B, Lindberg AL, Rahman M, Berglund M, Persson LA, El Arifeen S, et al. 2008. Urinary arsenic concentration adjustment factors and malnutrition. Environ Res 106(2):212–218, PMID: 17900556, 10.1016/j.envres.2007.08.005.

Nève J. 1995. Human selenium supplementation as assessed by changes in blood selenium concentration and glutathione peroxidase activity. J Trace Elem Med Biol 9(2):65–73, PMID: 8825978, 10.1016/S0946-672X(11)80013-1.

Oken E, Rifas-Shiman SL, Amarasiriwardena C, Jayawardene I, Bellinger DC, Hibbeln JR, et al. 2016. Maternal prenatal fish consumption and cognition in mid childhood: Mercury, fatty acids, and selenium. Neurotoxicol Teratol 57:71–78, PMID: 27381635, 10.1016/j.ntt.2016.07.001.

Parvez F, Wasserman GA, Factor-Litvak P, Liu X, Slavkovich V, Siddique AB, et al. 2011. Arsenic exposure and motor function among children in Bangladesh. Environ Health Perspect 119(11):1665–1670, PMID: 21742576, 10.1289/ehp.1103548.

Persson LA, Arifeen S, Ekström EC, Rasmussen KM, Frongillo EA, Yunus M. 2012. Effects of prenatal micronutrient and early food supplementation on maternal hemoglobin, birth weight, and infant mortality among children in Bangladesh: The MINIMat randomized trial. JAMA 307(19):2050–2059, PMID: 22665104, 10.1001/jama.2012.4061.

Polanska K, Krol A, Sobala W, Gromadzinska J, Brodzka R, Calamandrei G, et al. 2016. Selenium status during pregnancy and child psychomotor development-Polish Mother and Child Cohort study. Pediatr Res 79(6):863–869, PMID: 26885758, 10.1038/pr.2016.32.

Power JH, Blumbergs PC. 2009. Cellular glutathione peroxidase in human brain: Cellular distribution, and its potential role in the degradation of Lewy bodies in Parkinson’s disease and dementia with Lewy bodies. Acta Neuropathol 117(1):63–73, PMID: 18853169, 10.1007/s00401-008-0438-3.

Rahman SM, Kippler M, Tofail F, Bolte S, Hamadani JD, Vahter M. 2016. Manganese in drinking water and cognitive abilities and behavior at 10 years of age: A prospective cohort study. Environ Health Perspect 125(5):057003, PMID: 28564632, 10.1289/EHP631.

Raven JC, Court JH. 2003. Manual for Raven’s Progressive Matrices and Vocabulary Scales. San Antonio, TX:Harcourt Assessment.

Rita Cardoso B, Silva Bandeira V, Jacob-Filho W, Franciscato Cozzolino SM. 2014. Selenium status in elderly: Relation to cognitive decline. J Trace Elem Med Biol 28(4):422–426, PMID: 25220532, 10.1016/j.jtemb.2014.08.009.

Roman M, Jitaru P, Barbante C. 2014. Selenium biochemistry and its role for human health. Metallomics 6(1):25–54, PMID: 24185753, 10.1039/c3mt00185g.

Schweizer U, Fradejas-Villar N. 2016. Why 21? The significance of selenoproteins for human health revealed by inborn errors of metabolism. FASEB J 30(11):3669–3681, PMID: 27473727, 10.1096/fj.201600424.

Skröder Löveborn H, Kippler M, Lu Y, Ahmed S, Kuehnelt D, Raqib R, et al. 2016. Arsenic metabolism in children differs from that in adults. Toxicol Sci 152(1):29–39, PMID: 27056082, 10.1093/toxsci/kfw060.

Skröder HM, Hamadani JD, Tofail F, Persson LÅ, Vahter ME, Kippler MJ. 2015. Selenium status in pregnancy influences children’s cognitive function at 1.5 years of age. Clin Nutr 34(5):923–930, PMID: 25444556, 10.1016/j.clnu.2014.09.020.

Skröder H, Kippler M, Nermell B, Tofail F, Levi M, Rahman S, et al. 2017. Major limitations in using element concentrations in hair as biomarkers of exposure to toxic and essential trace elements in children. Environ Health Perspect 125(6):067021, PMID: 28669939, 10.1289/EHP1239.

Stefanowicz FA, Talwar D, O’Reilly DS, Dickinson N, Atkinson J, Hursthouse AS, et al. 2013. Erythrocyte selenium concentration as a marker of selenium status. Clin Nutr 32(5):837–842, PMID: 23391458, 10.1016/j.clnu.2013.01.005.

Takemoto AS, Berry MJ, Bellinger FP. 2010. Role of selenoprotein P in Alzheimer’s disease. Ethn Dis 20(suppl1):S192–S195, PMID: 20521393.

Vinceti M, Mandrioli J, Borella P, Michalke B, Tsatsakis A, Finkelstein Y. 2014. Selenium neurotoxicity in humans: Bridging laboratory and epidemiologic studies. Toxicol Lett 230(2):295–303, PMID: 24269718, 10.1016/j.toxlet.2013.11.016.

Wechsler D. 2002. Wechsler Preschool and Primary Scale of Intelligence™– Third Edition. San Antonio, TX: The Psychological Corporation.

WHO (World Health Organization). Global Database on Child Growth and Malnutrition. http://www.who.int/nutgrowthdb/about/introduction/en/index5.html [accessed 20 October 2017].

WHO. 2002. Standards for maternal and neonatal care: Iron and folate supplementation. http://www.who.int/reproductivehealth/publications/maternal_perinatal_health/iron_folate_supplementation.pdf [accessed 2 May 2013].

Yang X, Yu X, Fu H, Li L, Ren T. 2013. Different levels of prenatal zinc and selenium had different effects on neonatal neurobehavioral development. Neurotoxicology 37:35–39, PMID: 23570748, 10.1016/j.neuro.2013.04.001.

Zafar KS, Siddiqui A, Sayeed I, Ahmad M, Salim S, Islam F. 2003. Dose-dependent protective effect of selenium in rat model of Parkinson’s disease: Neurobehavioral and neurochemical evidences. J Neurochem 84(3):438–446, PMID: 12558963, 10.1046/j.1471-4159.2003.01531.x.

Long-term Fine Particulate Matter Exposure and Nonaccidental and Cause-specific Mortality in a Large National Cohort of Chinese Men

Author Affiliations open

1National Center for Chronic Noncommunicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, Beijing, China

2School of Population and Public Health, University of British Columbia, Vancouver, British Columbia, Canada

3Health Effects Institute, Boston, Massachusetts, USA

4Health Canada, Ottawa, Ontario, Canada

5Shaanxi Provincial Center for Disease Control and Prevention, Xian, China

PDF icon PDF Version (1.5 MB)

  • Background:
    Cohort studies in North America and western Europe have reported increased risk of mortality associated with long-term exposure to fine particles (PM2.5), but to date, no such studies have been reported in China, where higher levels of exposure are experienced.
    Objectives:
    We estimated the association between long-term exposure to PM2.5 with nonaccidental and cause-specific mortality in a cohort of Chinese men.
    Methods:
    We conducted a prospective cohort study of 189,793 men 40 y old or older during 1990–91 from 45 areas in China. Annual average PM2.5 levels for the years 1990, 1995, 2000, and 2005 were estimated for each cohort location using a combination of satellite-based estimates, chemical transport model simulations, and ground-level measurements developed for the Global Burden of Disease (GBD) 2013 study. A Cox proportional hazards regression model was used to estimate hazard ratios (HR) for nonaccidental cardiovascular disease (CVD), chronic obstructive pulmonary disease (COPD), and lung-cancer mortality. We also assessed the shape of the concentration–response relationship and compared the risk estimates with those predicted by Integrated Exposure-Response (IER) function, which incorporated estimates of mortality risk from previous cohort studies in western Europe and North America.
    Results:
    The mean level of PM2.5 exposure during 2000–2005 was 43.7 μg/m3 (ranging from 4.2 to 83.8 μg/m3). Mortality HRs (95% CI) per 10-μg/m3 increase in PM2.5 were 1.09 (1.08, 1.09) for nonaccidental causes; 1.09 (1.08, 1.10) for CVD, 1.12 (1.10, 1.13) for COPD; and 1.12 (1.07, 1.14) for lung cancer. The HR estimate from our cohort was consistently higher than IER predictions.
    Conclusions:
    Long-term exposure to PM2.5 was associated with nonaccidental, CVD, lung cancer, and COPD mortality in China. The IER estimator may underestimate the excess relative risk of cause-specific mortality due to long-term exposure to PM2.5 over the exposure range experienced in China and other low- and middle-income countries. https://doi.org/10.1289/EHP1673
  • Received: 24 February 2017
    Revised: 01 September 2017
    Accepted: 05 September 2017
    Published: 07 November 2017

    Address correspondence to M. Zhou, National Center for Chronic Noncommunicable Disease Control and Prevention, Chinese Center for Disease Control and Prevention, 27 Nanwei Road, Xicheng District, 100050 Beijing, China. Email: maigengzhou@126.com

    Supplemental Material is available online (https://doi.org/10.1289/EHP1673).

    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 (293 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 (134 KB)

Introduction

Air pollution and especially particulate matter (PM) with aerodynamic diameter less than 2.5 μm (PM2.5) became a focus for both the Chinese government and the Chinese public since the occurrence of heavy smog episodes across most of the areas in China in early 2013 (Chen et al. 2013). In addition, the Global Burden of Disease (GBD) 2010 identified outdoor air pollution as the fourth highest modifiable risk factor for disease burden in China, responsible for more than 1.2 million deaths in 2010. The long-term health effects of ambient air pollution have been studied in many high-income countries with evidence of relatively consistent associations between long-term exposure to ambient air pollution and nonaccidental and cardiovascular (CVD) mortality (Crouse et al. 2012; Gold and Mittleman 2013; Hoek et al. 2013; Beelen et al. 2014a, 2014b).

To date, most studies in China have focused on the effects of short-term exposure and have reported increased rates of daily mortality from natural causes, CVD, and respiratory disease that are consistent with observations in other global regions (Chen et al. 2012, 2013; HEI 2010; Zhou et al. 2015a, 2015b). The effects of long-term PM2.5 exposure have not been adequately investigated. Associations between long-term exposure to ambient air pollution and mortality in China were first reported in the late 1990s. Two aggregate-level (ecologic) studies estimated increased mortality from natural causes, CVD, respiratory disease, and lung cancer associated with long-term average exposure to total suspended particles (TSP) (Jin et al. 1999; Xu et al. 1996). A newer aggregate-level study also reported decreased life-expectancy at birth in northern China associated with coal use and with exposure to TSP (Chen et al. 2013). Recently, several studies have reported increased mortality using individual-level data (Cao et al. 2011; Dong et al. 2012; Wong et al. 2015; Zhang et al. 2011; Zhou et al. 2014). Among them, only a single study assessed the effects of PM2.5, in this case based on high-resolution satellite data and applied to an elderly cohort in Hong Kong (Wong et al. 2015). Several cohort studies, including some with small sample sizes focused on PM with aerodynamic diameter less than 10 μm (PM10) or TSP (Zhou et al. 2014). The other studies were conducted in a single city (Dong et al. 2012; Zhang et al. 2011). More data from large prospective studies are needed to address the association between PM2.5 and mortality among the Chinese population. Furthermore, almost all the studies so far have focused on urban areas, which may have higher pollution levels. Much of the populations in rural areas in China, however, are also exposed to higher levels of PM2.5 compared to urban and rural areas in most high-income countries (Brauer et al. 2012).

Previous estimates of disease burden due to exposure to PM2.5 in China and in other high-pollution locations have used the Integrated Exposure-Response (IER) function (Cohen et al. 2017). The IER incorporated estimates of mortality risk from cohort studies in western Europe and North America, where PM2.5 exposures were generally lower than exposures observed in China. Information from other types of PM2.5 exposure, such as active and passive smoking, were used to extrapolate risk to these higher levels. Direct evidence of the magnitude of mortality risk over the complete global exposure range is therefore lacking. This study was conducted to both obtain direct evidence of the association between exposure to PM2.5 in outdoor air and mortality in China, and to fill the exposure gap between concentrations observed in Western countries and globally.

In the present study, we estimated PM2.5 exposure levels used for the Global Burden of Disease (GBD) 2013 study and assessed the long-term effects on nonaccidental, CVD, COPD, and lung-cancer mortality in a cohort of 222,000 men. We also assessed the shape of the concentration–response relationship in this setting across the range of PM2.5 concentrations in China, including high levels.

Methods

Study Population

The design of the cohort has been described elsewhere (Chen et al. 2012; Yang et al. 2010; Zhou et al. 2008). Briefly, the participants for this study were recruited from 45 districts/counties across China in 1990–1991, randomly selected from China’s 145 Disease Surveillance Points (DSPs). DSPs are nationally representative and cover about 1% of China’s total population. A DSP site was either a district in an urban area or a county in a rural area. In each district/county, two or three residential units were randomly selected, and all resident men older than 40 y were invited to participate in the study, of whom about 80% accepted. As a result, the original study population consisted of 224,064 men. The baseline survey included a standardized questionnaire and physical measurements administered by trained health workers to obtain information on demographics, lifestyle factors (smoking, alcohol drinking, and household solid-fuel use), personal medical history, height, and weight. Vital status was monitored by local DSP staff through death registries, with regular crosschecks from local residential records kept at the Public Security Department and the Social Welfare Department, supplemented by annual active confirmation by local residential committees. The underlying causes of death were ascertained from official death certificates, supplemented with information from medical records and coded according to International Classification of Diseases version 9 (ICD-9). Cohort follow-up was from 1990/1991 to 2006. Deaths from CVD (390–414, 420–459), including ischemic heart disease (IHD) (410–414) and stroke (431–439), COPD (490–496), and lung cancer (162) were specifically examined in the present study. This study received approval from the ethics committee of the Chinese Center for Disease Control and Prevention. All participants gave oral informed consent.

Air Pollution Exposure

Each participant was assigned an area of residence based on DSP site as determined by the baseline survey. Based on the estimates developed for the GBD 2013 study, the PM2.5 exposure level for all 45 district/counties was estimated for 1990, 1995, 2000, and 2005, as described in detail elsewhere (Brauer et al. 2016).

Briefly, PM2.5 exposures were estimated using a combination of satellite-derived and chemical transport model estimates, calibrated to surface measurements. The satellite-based PM2.5 estimates use aerosol optical depth (AOD) retrievals from satellites to estimate near-surface PM2.5 at 0.1°×0.1° (∼11×11 km at the equator) resolution by applying the relationship of PM2.5 to AOD simulated by the GEOS-Chem chemical transport model (van Donkelaar et al. 2015). As satellite retrievals were not available prior to 1998, for 1990 and 1995 estimates were based on the ratio of GEOS-Chem simulations using anthropogenic emissions between 2005 and the respective year of interest, but constant meteorology. These estimates were then combined with simulations from the TM5-FASST (FAst Scenario Screening Tool, a reduced-form version of the TM5 chemical transport model) (van Dingenen and Leitao 2014) based on internally consistent emissions inventories for 1990, 2000, and 2010, and “typical” (year 2001) meteorology, as well as dust and sea salt contributions. TM5-FASST estimates for 1995 and 2005 were generated by fitting a natural cubic spline. Incorporating these separate native estimates (satellite-based and TM5-FASST) helps characterize uncertainty (and how it varies globally) in the final estimate. Finally, we included ground measurements of PM collected from a variety of sources, which were used in a regression calibration approach to combine the mean of the satellite-based estimates and the TM5-FASST simulations with the measurements to produce final estimates at 0.1×0.1° resolution. A total of 4,073 data points from 3,387 unique locations (46% from direct measurements of PM2.5) in 79 countries were included in the global calibration. Among these were 394 measurements from China (90 direct measurements of PM2.5 and 304 where PM2.5 was estimated from PM10 measurements). The measured concentrations in China (during the 2010–2013 period) ranged from 6.6–194.0 μg/m3 annual average PM2.5.

Statistical Analysis

We categorized the participants into four quartiles (Q1–Q4) of PM2.5 exposure and presented the baseline characteristics according to PM2.5 quartiles. We calculated age-adjusted nonaccidental mortality rate for participants of these four categories using the total population in the cohort as reference population. We used Cox proportional hazard models for survival, with the time-scale setting as duration from year of recruitment to the year of death or censored at the year of the follow-up in 2006. We assigned to each subject the average ambient PM2.5 concentration at their DSP site, based on the years 2000 and 2005, and the mean concentrations between 2000 and 2005 were used as the exposure level for the main analysis. This period corresponds to the availability of satellite retrievals. The basic model included age at entry as strata variable by 1 y. In model 1, individual risk factors, such as marital status; educational level (≤6 years or >6 years of education): smoking-related variables; passive smoking; occupational exposure to dust, asbestos dust, coal dust, or stone dust; occupational exposure to coal tar or diesel engine exhaust; alcohol drinking (regular drinkers, nonregular drinkers) and units of alcohol per week (liang/week); body mass index (BMI) (kg/m2); and diet (consumption of fresh fruit and vegetables). In addition, indoor air pollution defined as household solid-fuel use for heating or cooking (no coal or wood) was included. As more than 70% of the participants were smokers, we made comprehensive adjustments for smoking, and the smoking-related variables included smoking status (never, former, and current), current smoker’s years of smoking, current smoker’s years of smoking squared, current smoker’s cigarettes per day, current smoker’s cigarettes per day squared, former smoker’s years of smoking, former smoker’s years of smoking squared, former smoker’s cigarettes per day, and former smoker’s cigarettes per day squared. BMI and BMI squared were included as continuous variables. As hypertension might be mediators of the relationship between particle exposure and mortality, we also examined a separate model with adjustment for hypertension. Hypertension was defined if the participant was diagnosed by a health professional or took any antihypertensive drugs in the last two days before the interview or the measured SBP≥140 mmHg or DBP≥90 mmHg for those who did not take any antihypertensive drugs. A total of 189,793 men were included in the analysis after exclusion of participants with missing data for the covariates in model 1.

In model 2, we added urban/rural, region, and area-level mean years of education as proxy for contextual socioeconomic status (SES). The term urban referred to a district in a city, and rural meant a county in the baseline recruitment, according to criteria by China National Statistics Bureau. Six geographical regions were included, i.e.,Northeast, North, East, South-Central, Southwest, and Northwest (Figure 1), to adjust for broad-scale spatial variation in mortality not accounted for by the risk factors included in the survival model. The DSP-level mean years of education was obtained from the China Census 2000. For each model, the mortality hazard ratio (HR) associated with a 10-μg/m3 increase in mean PM2.5 concentration during 2000–2005 at each participant’s baseline DSP site was estimated. To assess the effect of individual area-level confounders, we performed incremental adjustment by adding the three area-level covariates, individually and in pairs, to the model adjusted for individual-level covariates only. We also calculated the percent changes in −2 Log L values in comparison with the age-adjusted basic model to indicate the effect of adjustment on the model’s degree of fit.

Map of China marking sites with PM sub 2.5 concentration in the air. The PM sub 2.5 concentration for years 2000 to 2005 for urban and rural sites range from 4–10 micrograms per cubic meter to greater than 80 micrograms per cubic meter.
Figure 1. Mean concentrations (2000–2005 mean) of PM2.5 in the 45 urban and rural DSP sites in the cohort. A: North; B: Northeast; C: East; D: Southwest; E: South Central; F: Northwest; G: Hong Kong Special Administrative Region; H: Macao Special Administrative Region. The software used to create the figure is QGIS (version 2.18; QGIS Community, open source).

We performed separate analyses stratified by urban/rural, household solid-fuel use, and smoking status to explore potential effect modification. We also combined the three northern regions and three southern regions and stratified the two large geographical regions. We performed sensitivity analyses with exposures to average PM2.5 for the year 1990 (baseline) and average exposure from 1990 to 2005, and exclusion of participants who died in the first 1–3 y.

To examine the shape of the concentration–response relationship between PM2.5 and mortality, we used an integrated modeling framework in which a class of flexible algebraic concentration–response functions was fit to survival models using standard computer software [Shape Constrained Health Impact Function, (SCHIF)]. We constructed the class by defining transformations of concentration as the product of either a linear or log-linear function of concentration multiplied by a logistic weighting function. This approach was recently illustrated with two large cohort studies: the American Cancer Society Cancer Prevention Study II (ACS) cohort and the Canadian Census Health and Environment Cohort (CanCHEC) (Nasari et al. 2016). The model is monotonically nondecreasing and suitable for health-effect assessment. The model incorporates both sampling and model shape uncertainty. In addition, we estimated the HR for four groupings of PM2.5 exposure based on quintiles relative to the first quintile.

We compared our risk estimates for IHD, stroke, COPD, and lung-cancer mortality to those used by GBD2015 (Cohen et al. 2017) in two ways: First, we used the R routine RMA from the package Metafor to conduct a meta-analysis (R Core Team, 2016). We included in the meta-analysis the logarithm of the HR estimates and their corresponding standard errors per μg/m3 of PM2.5 for each of the four causes of death (IHD, stroke, COPD, and lung cancer) reported in Cohen et al (2017); see Table 1. We specified the random effects model with the REML option to estimate the heterogeneity variance. The meta-analysis estimated a summary logarithm of the HR among the cohorts, examining the association between outdoor concentrations of PM2.5 and cause-specific mortality, in addition to a standard error of this estimate. Second, we calculated the relative risk estimate and confidence interval (CI) for the four causes of death based on IER functions as reported in Cohen et al. (2017) between the 5th (15.5 μg/m3) and 95th (77.1 μg/m3) percentile for the age range of 60–64 y, as this was the average age of follow-up in this cohort. We selected this exposure contrast because it represents the level of exposure experienced by the cohort subjects. We also selected a specific age range for comparison because the IER functions are age specific.A specific exposure contrast is required for this calculation because the IER is nonlinear in PM2.5 concentration, resulting in different relative risk estimates for different values of the exposure contrast. We then calculated the relative risk and corresponding 95% CI for each cause of death separately, using the meta-analysis summary estimate and standard error based on the same exposure contrast as was used for the IER. Here, the relative risk model assumed a linear association between PM2.5 and the logarithm of the HR. All statistical analyses were conducted using the SAS (version 9.3; SAS Institute Inc.) and R software (version 3.2.2; R Core Team).

Table 1. Characteristics of cohort participants at baseline by four categories by quartiles of PM2.5 concentration.
PM2.5 categorya
Variables Total Q1 Q2 Q3 Q4
Participants (n) 189,793 58,034 40,243 42,512 49,004
Individual-level
 Age (years, mean±SD) 54.8±10.7 54.6±10.5 55.2±10.8 54.9±10.9 54.9±10.7
 Marital status (% of married) 91.0 94.7 91.1 91.8 85.9
Smoking (%)
 Never 25.9 22.9 30.0 21.7 29.8
 Former 5.9 6.1 4.0 5.7 7.2
 Current 68.2 71.0 66.0 72.5 63.0
Years of smoking for current smokers (mean±SD) 21.8±17.8 22.9±17.6 21.5±18.3 23.1±17.4 19.6±17.6
Years of smoking for former smokers (mean±SD) 1.7±7.6 1.7±7.7 1.2±6.3 1.6±7.5 2.1±8.5
Cigarettes per day in current smokers 13.5±13.2 13.5±12.7 16.5±15.7 13.4±12.8 11.2±11.4
Cigarettes per day in former smokers 1.1±5.1 1.1±5.4 0.8±4.6 1.0±5.1 1.2±5.2
Passive smoking (%) 79.7 83.2 77.1 83.1 74.9
Education, ≤6 years (%) 72.3 70.8 69.6 74.9 70.7
Indoor air pollution (%)
 Coal 27.6 29.8 17.8 33.6 27.7
 Biomass 35.9 25.0 43.5 45.6 34.1
 No exposure 36.5 45.2 38.7 20.8 38.2
Alcohol drinking (%) 32.6 37.4 37.5 26.4 28.4
Amount of alcohol consumed (drinks/week) 10.7±21.9 12.1±22.5 16.2±28.4 8.0±18.9 7.0±15.2
Occupational exposure (%) 4.5 4.4 3.4 4.0 6.1
Fruit and vegetable consumption (%) 31.8 49.4 20.6 22.6 28.1
BMI (kg/m2, mean±SD) 21.6±2.6 21.3±2.7 21.0±2.4 21.8±2.5 22.1±2.7
Hypertension (%) 26.9 24.9 22.5 31.3 29.2
Area-level factors
 Urban (%) 20.2 13.7 19.3 17.2 31.2
Region
  Northeast (%) 10.2 31.3 3.2 0.0 0.0
  North (%) 8.8 2.1 6.6 24.5 4.6
  Northwest (%) 11.3 5.7 0.0 38.4 3.6
  Southwest (%) 20.7 26.6 22.9 4.2 26.2
  South-central (%) 41.1 20.5 67.4 27.6 55.9
  East (%) 7.9 13.8 0.0 5.4 9.7
Mean years of education (mean±SD)b 7.8±1.2 7.4±1.2 7.6±1.0 8.0±1.0 8.3±1.4
Number of deaths
 Nonaccidental 50,022 15,466 10,348 13,263 13,779
 CVD 18,859 4,797 3,567 5,433 5,062
 COPD 11,989 4,231 2,043 2,754 2,961
 Lung cancer 2,523 797 452 579 695
 Age-adjusted mortality rate (per 100,000 person-years) 2,538 2,355 3,077 2,695

aQuartile cutpoints of PM2.5 concentrations were based on mean exposures during 2000–2005 at each baseline cohort site. Minimum, 4.23 μg/m3; 25th percentile, 28.3 μg/m3; 50th percentile, 38.9 μg/m3; 75th percentile, 60.1 μg/m3; maximum, 83.8 μg/m3.

bArea-level mean years of education.

Results

The geographic location and names of the 45 cohort sites are shown in Figure 1. Descriptive analysis of the participants at baseline is presented in Table 1. The mean age of the participants was 54.8±10.7 y (mean±SD). In addition, 20.2% of the participants lived in urban areas, and 72.3% had an educational level of less than 6 y. Most participants (74.1%) were current or former smokers. The mean BMI was 21.6 kg/m2, and 63.5% reported household solid-fuel use (27.6% with coal and 35.9% with biomass). Participants in the highest exposure category tended to be in urban areas and from areas with higher educational levels. In total, 50,022 nonaccidental deaths occurred during the 15 y of follow-up. These deaths included 18,859 deaths from CVD, 11,989 deaths from COPD, and 2,523 deaths from lung cancer. The age-adjusted mortality rate was 2,538, 2,355, 3,077, and 2,695 per 100,000 person-years, respectively, for participants in the four exposure categories.

The average PM2.5 concentration in the 45 cohort sites increased from 36.4 μg/m3 in 1990 to 46.4 μg/m3 in 2005, with concentrations in urban areas consistently higher than concentrations in rural areas (Table 2). As shown in Table 3, a 10-μg/m3 increase in PM2.5 based on average exposures during 2000–2005 at each participant’s cohort site was associated with increased risk of mortality from nonaccidental causes, CVD, IHD, stroke, and lung cancer in the model adjusted for individual-level covariates. In the fully adjusted model that includes both individual- and area-level covariates, we observed that a 10-μg/m3 increase in PM2.5 was associated with a higher mortality risk for nonaccidental causes (HR=1.09; 95% CI: 1.08, 1.09), CVD (HR=1.09; 95% CI: 1.08, 1.10), IHD (HR=1.09; 95% CI: 1.06, 1.12), stroke (HR=1.14; 95% CI: 1.13, 1.16), COPD (HR=1.12; 95% CI: 1.10, 1.13), and lung cancer (HR=1.12; 95% CI: 1.07, 1.14). Further adjustment for hypertension only slightly reduced the effect size without major changes (Table S1). In the incremental adjustment, the effect estimates increased with inclusion of the three area-level covariates (Table S2). We observed generally larger reductions in the −2LL values when the contextual variables were added to the model in comparison with the individual risk factors, suggesting that there remains considerable spatial variation in mortality that is not accounted for by our individual risk factors (Table S2). We therefore included area/contextual variables to account for some of this residual variation. Furthermore, we included both region and urban/rural specification as stratification variables to adjust for potential confounding effects on the PM2.5–mortality association of these two area factors. In this model specification, PM2.5 exposures are compared among subjects in the same region-urban/rural strata, and not between strata. We included these area-level factors as stratification variables and not as predictors in the survival model, as their baseline mortality rates vary substantially between strata (Table S3). We note that a single PM2.5–mortality effect estimate is obtained from this model, summarizing associations among all strata.

Table 2. Estimated exposure level of PM2.5 (μg/m3) in 45 cohort sites from 1990 to 2005.
PM2.5 category 1990 1995 2000 2005
Total (n=45)
 Mean±SD 36.4±16.3 38.7±17.6 40.7±18.6 46.4±20.2
 Min 3.5 3.8 4.0 4.4
 Q1 23.8 25.0 26.0 31.4
 Q2 31.3 33.6 35.2 42.5
 Q3 49.5 52.9 55.7 63.4
 Max 81.9 81.5 81.5 89.8
Urban (n=23)
 Mean±SD 41.4±17.4 44.3±19.0 46.7±20.2 53.7±22.2
 Min 3.5 3.8 4.0 4.4
 Q1 27.5 27.4 28.5 36.5
 Q2 36.4 38.6 40.7 51.6
 Q3 59.0 65.4 70.6 79.6
 Max 69.5 74.0 77.7 89.8
Rural (n=22)
 Mean±SD 35.1±15.8 37.3±16.9 39.2±17.8 44.5±19.2
 Min 13.2 13.7 14.2 16.4
 Q1 22.6 24.0 25.2 29.1
 Q2 30.2 32.5 34.2 41.1
 Q3 49.5 52.9 55.7 60.7
 Max 81.9 81.5 81.5 79.3
Table 3. Hazard ratios (HRs) (and 95% CI) of mortality associated with 10 μg/m3 increase in PM2.5 levels.
Cause of death No. of deaths Age-adjusted basic model Adjusted model (1)a Adjusted model (2)b
Nonaccidental 50,022 1.03 (1.02–1.03) 1.04 (1.03–1.04) 1.09 (1.08–1.09)
Cardiovascular 18,859 1.07 (1.06–1.07) 1.06 (1.05–1.07) 1.09 (1.08–1.10)
IHD 3,752 1.14 (1.12–1.15) 1.11 (1.09–1.13) 1.09 (1.06–1.12)
Stroke 11,301 1.08 (1.07–1.09) 1.07 (1.06–1.08) 1.14 (1.13–1.16)
COPD 11,989 0.98 (0.97–0.98) 1.00 (0.99–1.01) 1.12 (1.10–1.13)
Lung cancer 2,523 1.03 (1.01–1.05) 1.04 (1.02–1.07) 1.12 (1.07–1.14)

Note: PM2.5 levels were based on mean exposures during 2000–2005 at each participant’s baseline cohort site.

aModel 1: Age-adjusted basic model+individual-level covariates including marital status, educational level, smoking status, years of smoking, cigarettes per day, passive smoking, occupational exposure, alcohol drinking, units of alcohol per week, body mass index (BMI), consumption of fresh fruits and vegetables, and household solid-fuel use.

bModel 2: model 1+area-level covariates including urban/rural, region, and mean years of education.

In stratified analysis (Figure 2), the association was consistently significant for nonaccidental, CVD, IHD, stroke, and COPD mortality in all strata according to urban/rural, north/south, smoking status, or exposure to indoor air pollution. The effect size of nonaccidental causes, CVD, IHD, and COPD in the north was generally higher than the south. The association between PM2.5 and lung cancer was significantly positive for southern regions and negative for the northern regions. Table 4 presents the results of sensitivity analysis. HRs were slightly increased when we use the baseline level of exposure (1990) and mean levels of PM2.5 between 1990 and 2005. When we excluded participants who died 1–3 y after enrollment, the HRs slightly decreased but still remained significant.

Line graph with confidence intervals plotting hazard ratio (95 percent confidence interval) showing mortality for (A: nonaccidental deaths; B: CVD; C: IHD; D: Stroke; E: COPD; F: Lung cancer) associated with 10 micrograms per cubic meter increase in PM sub 2.5 levels, stratified by urban/rural, region, smoking status, and indoor air pollution. PM sub 2.5 distribution was based on mean concentrations during 2000 through 2005 for each participant’s cohort site at baseline.
Line graph with confidence intervals plotting hazard ratio (95 percent confidence interval) showing mortality for (A: nonaccidental deaths; B: CVD; C: IHD; D: Stroke; E: COPD; F: Lung cancer) associated with 10 micrograms per cubic meter increase in PM sub 2.5 levels, stratified by urban/rural, region, smoking status, and indoor air pollution. PM sub 2.5 distribution was based on mean concentrations during 2000 through 2005 for each participant’s cohort site at baseline.
Figure 2. Hazard ratios (HRs) (and 95% CI) of mortality [(A): nonaccidental deaths; (B): CVD; (C): IHD; (D): Stroke; (E): COPD; (F): Lung cancer] associated with a 10-μg/m3 increase in PM2.5 levels, stratified by urban/rural, region, smoking status, and indoor air pollution. PM2.5 distribution was based on mean concentrations during 2000–2005 for each participant’s cohort site at baseline. Note: Stratified models were adjusted for age, individual-level covariates including marital status, educational level, smoking status, years of smoking, cigarettes per day, passive smoking, occupational exposure, alcohol drinking, units of alcohol per week, body mass index (BMI), consumption of fresh fruits and vegetables, and indoor air pollution and area-level covariates, including urban/rural, region, and mean years of education. y-Axis scale was 1.5 for CVD and IHD and 1.3 for other causes of death.
Table 4. Hazard ratio (HR) (95% CI) per 10 μg/m3 increase of PM2.5 for sensitivity analyses for exposure at baseline and for different inclusion criteria.
Cause of death Exposure in 1990 (n=189,793) Average exposure between 1990–2005 (n=189,793) Excluding deaths within 3 years, exposure 2000–2005 (n=179,089)a
Nonaccidental 1.12 (1.11, 1.12) 1.10 (1.10, 1.11) 1.07 (1.06, 1.08)
Cardiovascular 1.10 (1.08, 1.11) 1.11 (1.10, 1.12) 1.08 (1.07, 1.09)
  IHD 1.10 (1.07, 1.14) 1.10 (1.07, 1.13) 1.07 (1.04, 1.10)
  Stroke 1.15 (1.14, 1.18) 1.18 (1.16, 1.19) 1.14 (1.12, 1.16)
COPD 1.19 (1.17, 1.21) 1.13 (1.12, 1.16) 1.12 (1.10, 1.13)
Lung cancer 1.10 (1.06, 1.14) 1.10 (1.06, 1.13) 1.12 (1.07, 1.14)

aExcluded numbers of deaths: 10,083 for nonaccidental causes; 3,489 for cardiovascular diseases; 689 for IHD; 2,158 for stroke; 2,708 for COPD; and 454 for lung cancer.

The HR estimates by quintile of PM2.5 concentrations suggested nonlinear relationships between PM2.5 and mortality from all disease outcomes (Figure S1). Compared with the first quintile, the second quintile had a lower HR during the low concentrations for nonaccidental causes, CVD, stroke, and COPD. The HR estimate associated with the fourth PM2.5 quintile tended to have the highest value, and a lower estimate was observed for the fifth quintile. These plots suggest a sigmodal pattern with little change in the HR for lower concentrations and a diminishing change for the highest exposures.

The SCHIF concentration–response plots for the six causes of deaths display a pattern similar to that of the quintile-based analyses (Figure 3). The estimated curves are monotonic in concentration and display less curvature than the quintile-based analysis, a design feature of the SCHIF. In general, there is a greater change in relative risk for higher concentration in comparison with lower values. We note there is little change in the HR for concentrations below 20 μg/m3 and little uncertainty in these estimates. This finding is due to the slightly negative association in this range, as evidenced by the quintile analyses and the requirement of the SCHIF to yield HR estimates greater than unity and increase with increasing concentration. We also note that the HR estimates increase more rapidly for stroke than for the other causes of death. The much wider CIs for lung cancer suggested larger uncertainty, likely due to the relatively smaller number of lung-cancer deaths in this cohort.

Figures 3A, 3B, 3C, 3D, 3E, and 3F are line graphs plotting hazard ration (95 percent confidence interval) (y-axis) across PM 2.5 concentration (micrograms per cubic meter) (x-axis) for non-accidental causes, CVD, IHD, stroke, COPD, and lung cancer, respectively.
Figure 3. Concentration–response curves and 95% confidence intervals (CIs) for the relationship between PM2.5 and mortality based on the model adjusted for individual-level and area-level covariates: (A) Nonaccidental causes; (B) CVD; (C) IHD; (D) stroke; (E) COPD; (F) lung cancer. PM2.5 concentration was based on mean exposures during 2000–2005 for each cohort site at baseline.

Figure 4 shows the comparison of HR estimates for the four causes of deaths among the present study, IER predictions, and meta-analysis, based on the same exposure contrast between the fifth (15.5 μg/m3) and 95th (77.1 μg/m3) percentile as was used for IER (Cohen et al, 2017) for the age range of 60–64 y. We found the HR estimate from our cohort based on 15.5–77.1  μg/m3 exposure contrast was consistently higher than the corresponding IER estimates for all the four disease outcomes and very similar to the meta-analysis estimates for COPD and lung cancer. In comparison with the meta-analysis estimates, our HR was somewhat lower for IHD and slightly higher for stroke. Both this cohort and the meta-analysis estimates were much larger than the IER predictions for this particular exposure contrast experienced by our cohort subjects.

Line graph with confidence intervals plotting hazard ratio (95 percent confidence interval) (y-axis) across IHD, stroke, COPD, and lung cancer (x-axis) for the present study, IER predictions, and meta-analysis.
Figure 4. Comparison of hazard ratio (HR) estimates and confidence interval (CI) for the four causes of deaths between the present study, IER predictions and meta-analysis for the outdoor air pollution (OAP) study, based on the same exposure contrast between the fifth (15.5 μg/m3) and 95th (77.1 μg/m3) percentile (as was used for IER) for the age range of 60–64 y. IER predictions: estimates based on IER functions as reported in Cohen et al, (2017) between the 5th (15.5 μg/m3) and 95th (77.1 μg/m3) percentile for the age range of 60–64 years. Meta-analysis: HRs (95% CI) presented was calculated from the meta-analysis summary estimate and standard error based on the same exposure contrast as was used for the IER (15.5–77.1  μg/m3), using the cohorts [a list of the cohort was in Table 1 in Cohen et al. (2017)] examining the association between outdoor concentrations of PM2.5 and cause-specific mortality.

Discussion

In this Chinese cohort, we observed an elevated risk for mortality in association with long-term exposure to PM2.5. An increase of 10 μg/m3 of PM2.5 was associated with 9%, 9%, 9%, 14%, 12%, and 12% increases in the risk of mortality from nonaccidental causes, CVD, IHD, stroke, COPD, and lung cancer, respectively, in the fully adjusted model. The concentration response curve suggested a nonlinear relationship between PM2.5 and mortality in China, where the exposure is higher than exposure in developed countries.

To estimate exposure levels, we used a method combining satellite-based estimates, chemical transport models, and ground-level measurements developed for the GBD 2013 study. The use of a continuous spatial surface of PM2.5 estimates allows for assignment of exposure to both urban and rural participants using a common approach, reducing the potential for sampling bias in which only populations in the vicinity of ground measurements are included. The exposure estimates are also internally consistent throughout the period of follow-up. To provide full spatial and temporal coverage, these estimates include chemical transport model simulations and satellite retrievals, but the estimates also incorporate the substantial number of ground measurements from China that have recently become available.

We focus on the 2000–2005 average concentrations because this period included satellite retrievals, whereas the 1990 and 1995 estimates were based on scaling from the year 2000 spatial patterns with chemical transport model simulations, as described previously (Brauer et al. 2016). Although this back-casting does incorporate time-varying spatial variation from the chemical transport model simulations, these are at a lower spatial resolution than the satellite retrievals, and thus there is a likelihood of greater exposure misclassification for the 1990 and 1995 estimates.

We compared our effect estimates of the associations between PM2.5 and mortality with previous large cohort studies, including ACS (Krewski et al. 2009), Six City Study (SCS) (Lepeule et al. 2002), California Teachers Study (CTS) (Lipsett et al. 2011), Dutch Study of Diet and Cancer (DSDC) (Beelen et al. 2014a; Beelen et al. 2014b), Canadian Census Health and Environment Cohort (CanCHEC) (Crouse et al. 2012), and a Hong Kong cohort (Wong et al. 2015) (Figure S2). Previous studies have generally reported HRs in excess of one, although many lacked statistical significance. The HR estimate of nonaccidental and CVD mortality in our study is generally lower than the estimates reported in most of the other major cohort studies, including SCS, CanCHEC, and the Hong Kong Cohort, but higher than the ACS estimate. Our effect estimate of IHD is consistently lower than the estimates in listed cohort studies conducted in high-income countries (Crouse et al. 2012; Krewski et al. 2009; Lepeule et al. 2012; Lipsett et al. 2011; Villeneuve et al. 2015). In line with previous reviews (Hamra et al. 2014; IARC 2016), our study showed a significant effect of PM2.5 exposure on lung-cancer mortality. The observed increased risk of COPD mortality strengthens the current evidence and is consistent with the most recent results of the ACS and other cohort studies (Cohen et al. 2017; Schikowski et al. 2014). The differences of effect size may be explained by various characteristics of study contexts, such as the feature of cohort population (e.g., we included only men in the current study, and CTS included only women), exposure estimate methods, PM2.5 pollution levels, sources and components of air pollutants, and population susceptibility to air pollution (Hedley et al. 2002).

We found a generally higher effect size of nonaccidental causes, CVD, IHD, and COPD in the north than in the south. Different sources and mixtures of PM2.5 from different areas may have differential health impacts. PM2.5 in some areas may be dominated by industrial emissions and power generation, whereas diverse sources’ mixtures contribute to PM2.5 in the other areas across China. Chen et al. (2013), noted above, reported regional differences in life-expectancy associated with long-term exposure to air pollution, and lower life expectancy north of the Huai River in 2000 associated with a national program instituted in the 1950s that provided free coal to northern areas and with elevated TSP levels. The health differences between regions might also be caused by factors other than air pollution, such as health care and access to medical services, which could not be fully adjusted for due to data availability. Further studies are needed to address such differences, especially in the northern provinces, and although we included personal educational level as a proxy for individual-level SES and used mean years of education as area-level SES in the models, accounting for the potential effects of individual and contextual variables related to on mortality remains an important issue for future studies.

The association between PM2.5 air pollution and lung-cancer mortality was positive in the south but negative in the north, which may be due to residual confounding by factors such as occupational exposure, for which we had limited data and therefore was crudely characterized in our study. We also noted that the HR in smokers was higher than among participants who had never smoked, which is not in accordance with the European Study of Cohorts for Air Pollution Effects (ESCAPE), which reported elevated HRs for lung-cancer incidence in association with PM2.5 in individuals who had never smoked (Raaschou-Nielsen et al. 2013).

Trends in air pollution exposure are important in modeling the long-term health effects. Although air pollution levels may vary over time, the changes are usually gradual, and spatial patterns are usually preserved, affecting the regional pollution in a similar way in different time periods (Lipsett et al. 2011), as suggested by our analysis. However, we had limited ability to assess trends and spatial patterns prior to 2000, as estimated exposures for 1990 and 1995 were based on spatial patterns from satellite retrievals in the year 2000, scaled by chemical transport model simulations at a coarser spatial resolution. Accordingly, we used average exposure between 2000 and 2005 as the exposure level for the main analysis, with baseline exposure and average exposure between the years 1990 and 2005 for sensitivity analyses. The effect estimates were similar in all these analyses, which may reflect the similar spatial patterns in pollution over time during our study period.

The accuracy of the specific mortality outcomes is also essential in this study. The mortality outcome in our cohort was ascertained by local DSP staff members through the death surveillance system and confirmed and coded by staff from central DSP office in Beijing, based on official death certificates, medical records, or verbal autopsies when no medical records are available. The good quality and reliability of the DSP have been proved (Liu et al. 2016).

Estimating the exposure–response relationships is critical to assessing the effect of specific regulatory actions to improve air quality on population mortality rates (Cohen et al. 2004; GBD 2013 Risk Factor Collaborators et al. 2015). Previous cohort studies often fit natural, restricted, or smoothing splines with a few degrees of freedom or use simple algebraic nonlinear functions or threshold function to assess the shape of the association between ambient air-pollutant concentrations and mortality (Crouse et al. 2012; Jerrett et al. 2009; Krewski et al. 2009). We used a more complex method to identify the shape of the association between air pollution and mortality in cohort study designs using the Cox proportional hazards model. However, we restricted the shape of the concentration–mortality association such that the HR estimates are always greater than unity and are monotonically nondecreasing, with the amount of curvature that makes them suitable for health-effects assessment. Our SCHIF-estimated functions display a similar pattern to that of using the more unstructured quintiles of exposure. Our study provides much needed direct evidence of the shape of the exposure–response relationship for PM2.5 and mortality over the entire global range. For this reason, it also provides a test of predictions of PM2.5 relative effects based on the IER used in GBD (Burnett et al. 2014; Cohen et al. 2017). Our results suggest that the IER may currently underestimate the excess relative risk over the exposure range experienced in China and other countries with high exposure to PM2.5. Although both the IER and the risk functions fit to the data in this study are constrained to be monotonically increasing, our results corroborate both the nonlinearity in CVD mortality relative risks predicted by the IER and the differences in cause-specific risk function shapes, which was also noted by Pope et al. (2011). They provide important new data points for future updates of the GBD risk functions, but, given the variation in cause-specific effect sizes observed in North American and European studies, it will be important to replicate this study in China and other high-exposure settings.

To our knowledge, this study is the first prospective cohort study to investigate the effect of long-term exposure to PM2.5 on CVD mortality and assess the concentration–response relationship in China. The large sample size and multicity coverage add strength and generalizability to our study. The estimates for PM2.5 exposure across all study areas from 1990 filled an important gap of previously unavailable historical air-pollutant exposure data. Furthermore, no previous cohort studies have examined the long-term health impacts of PM2.5 in rural areas in China. The findings from this study provide important estimates of long-term effects of air pollution for high-exposure settings typical in developing countries, and especially in China, where there is an urgent need to address the challenge of air-pollution exposure reduction for both urban and rural areas. Our estimates can lend support for decision makers to implement cost-effective air-quality improvement strategies and management plans.

Despite its novelty and many strengths, there are also some limitations to our study. First, we included only male participants, because this cohort was originally designed to investigate the effects of smoking (Niu et al. 1998). We therefore cannot assess the effects of long-term exposure to PM2.5 on nonaccidental and CVD mortality in women, although previous studies showed no systematic sex differences in the association between PM2.5 and mortality (Wong et al. 2015). Second, exposure was assessed at the center level and derived from gridded exposure estimates at ∼11×11 km resolution. As such, these estimates do not incorporate finer-scale spatial variation such as that related to topography or proximity to roads and point sources (Beelen et al. 2008). The differences between the proxy estimates and true exposure values are a source of unavoidable measurement error, which may bias the interpretation of air-pollution health effects. In a cross-validation analysis in which 10% of the global measurement sites used for calibration were randomly selected for model evaluation and repeated for a total of three separate sets of 10% testing sites, we estimated a root mean squared error of 9–11  μg/m3 (Brauer et al. 2016). Third, we were not able to adjust for multiple pollutants, such as sulfur dioxide, nitrogen dioxide, and ozone, due to data availability. Fourth, we cannot adjust for some important area-level SES variables, such as health care indicators, as the information was not available in the study period. It will be possible to consider this factor in the future follow-up of the cohort. However, more cohort studies of long-term health effects of PM2.5 exposure should be conducted in China to confirm the magnitude of effects found in the current study and address the limitations.

In conclusion, our study found increased nonaccidental, CVD, COPD, and lung-cancer mortality risks associated with high levels of exposure to ambient PM2.5 in China. The study provides information on PM2.5 exposure–response curves over a much broader range of exposure than previously studied specifically at high PM2.5 concentrations, contributing important new information for burden-of-disease assessments at the national and global levels. Our results suggest that the IER estimator used by the GBD project may underestimate the excess relative risk of cause-specific mortality due to long-term exposure to PM2.5 over the exposure range experienced in China and other low- and middle-income countries, but replication of these results is needed and should be a high priority for future research.

Acknowledgments

P.Y. conducted data analysis and wrote the first draft of the manuscript. Y.L., J.L., R.L., W.W., J.Q., and L.W. participated in data analyses and interpretation. M.B., A.C., and R.T.B. developed and evaluated the exposure models and provided input to analyses and interpretation of the results. M.Z. conceived the original idea and provided input with data analyses and interpretation. All authors were involved in drafting and revising the manuscript.

This study was supported by Gong-Yi Program of China Ministry of Environmental Protection (201509062).

References

Beelen R, Hoek G, van den Brandt PA, Goldbohm RA, Fischer P, Schouten LJ, et al. 2008. Long-term effects of traffic-related air pollution on mortality in a Dutch cohort (NLCS-air study). Environ Health Perspect 116(2):196–202, PMID: 18288318, 10.1289/ehp.10767.

Beelen R, Raaschou-Nielsen O, Stafoggia M, Andersen ZJ, Weinmayr G, Hoffmann B, et al. 2014a. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre escape project. Lancet 383:785–795, PMID: 24332274, 10.1016/S0140-6736(13)62158-3.

Beelen R, Stafoggia M, Raaschou-Nielsen O, Andersen ZJ, Xun WW, Katsouyanni K, et al. 2014b. Long-term exposure to air pollution and cardiovascular mortality: an analysis of 22 European cohorts. Epidemiology 25(3):368–378, PMID: 24589872, 10.1097/EDE.0000000000000076.

Brauer M, Amann M, Burnett RT, Cohen A, Dentener F, Ezzati M, et al. 2012. Exposure assessment for estimation of the global burden of disease attributable to outdoor air pollution. Environ Sci Technol 46(2):652–660, PMID: 22148428, 10.1021/es2025752.

Brauer M, Freedman G, Frostad J, van Donkelaar A, Martin RV, Dentener F, et al. 2016. Ambient air pollution exposure estimation for the Global Burden of Disease 2013. Environ Sci Technol 50(1):79–88, PMID: 26595236, 10.1021/acs.est.5b03709.

Burnett RT, Pope CA, 3rd, Ezzati M, Olives C, Lim SS, Mehta S, et al. 2014. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environ Health Perspect 122(4):397–403, PMID: 24518036, 10.1289/ehp.1307049.

Cao J, Yang C, Li J, Chen R, Chen B, Gu D, et al. 2011. Association between long-term exposure to outdoor air pollution and mortality in china: a cohort study. J Hazard Mater 186(2-3):1594–1600, PMID: 21194838, 10.1016/j.jhazmat.2010.12.036.

Chen R, Kan H, Chen B, Huang W, Bai Z, Song G, et al. 2012. Association of particulate air pollution with daily mortality: the China Air Pollution and Health Effects Study. Am J Epidemiol 175(11):1173–1181, PMID: 22510278, 10.1093/aje/kwr425.

Chen R, Zhang Y, Yang C, Zhao Z, Xu X, Kan H. 2013. Acute effect of ambient air pollution on stroke mortality in the China Air Pollution and Health Effects Study. Stroke 44(4):954–960, PMID: 23391765, 10.1161/STROKEAHA.111.673442.

Chen Y, Ebenstein A, Greenstone M, Li H. 2013. Evidence on the impact of sustained exposure to air pollution on life expectancy from China’s Huai River policy. Proc Natl Acad Sci USA 110(32):12936–12941. PMID: 23836630, 10.1073/pnas.1300018110.

Chen Z, Wang JN, Ma GX, Zhang YS. 2013. China tackles the health effects of air pollution. Lancet 382(9909):1959–1960, PMID: 24332069, 10.1016/S0140-6736(13)62064-4.

Cohen AJ, Anderson HR, Ostro B, Pandey KD, Krzyzanowski M, Künzli N, et al. 2004. Urban air pollution. In: Comparative Quantification of Health Risks: Global and Regional Burden of Disease Attributable to Selected Major Risk Factors. Geneva:World Health Organization, 1353–1433.

Cohen AJ, Brauer M, Burnett R, Anderson HR, Frostad J, Estep K, et al. 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389(10082):1907–1918, PMID: 28408086, 10.1016/S0140-6736(17)30505-6.

Crouse DL, Peters PA, van Donkelaar A, Goldberg MS, Villeneuve PJ, Brion O, et al. 2012. Risk of nonaccidental and cardiovascular mortality in relation to long-term exposure to low concentrations of fine particulate matter: a Canadian national-level cohort study. Environ Health Perspect 120(5):708–714, PMID: 22313724, 10.1289/ehp.1104049.

Dong GH, Zhang P, Sun B, Zhang L, Chen X, Ma N, et al. 2012. Long-term exposure to ambient air pollution and respiratory disease mortality in shenyang, China: a 12-year population-based retrospective cohort study. Respiration 84(5):360–368, PMID: 22116521, 10.1159/000332930.

GBD 2013 Risk Factor Collaborators, Forouzanfar MH, Alexander L, Anderson HR, Bachman VF, Bi, et al. 2015. Global, regional, and national comparative risk assessment of 79 behavioural, environmental and occupational, and metabolic risks or clusters of risks in 188 countries, 1990-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 386(10010):2287–2323, 10.1016/S0140-6736(15)00128-2.

Gold DR, Mittleman MA. 2013. New insights into pollution and the cardiovascular system: 2010 to 2012. Circulation 127(18):1903–1913, PMID: 23648681, 10.1161/CIRCULATIONAHA.111.064337.

Hamra G, Guha N, Cohen A, Laden F, Raaschou-Nielsen O, Samet J, et al. 2014. Outdoor particulate matter exposure and lung cancer: a systematic review and meta-analysis. Environ Health Perspect 122(9):906–911, PMID: 24911630, 10.1289/ehp.1408092.

Hedley AJ, Wong CM, Thach TQ, Ma S, Lam TH, Anderson HR. 2002. Cardiorespiratory and all-cause mortality after restrictions on sulphur content of fuel in Hong Kong: an intervention study. Lancet 360(9346):1646–1652, PMID: 12457788, 10.1016/S0140-6736(02)11612-6.

HEI (Health Effects Institute). 2010. Outdoor air pollution and health in the developing countries of Asia: a comprehensive review. https://www.Healtheffects.Org/publication/outdoor-air-pollution-and-health-developing-countries-asia-comprehensive-review [accessed 17 October 2017].

Hoek G, Krishnan RM, Beelen R, Peters A, Ostro B, Brunekreef B, et al. 2013. Long-term air pollution exposure and cardio-respiratory mortality: a review. Environ Health 12(1):43, PMID: 23714370, 10.1186/1476-069X-12-43.

IARC (International Agency for Research on Cancer). 2016. IARC Monographs on the evaluation of carcinogenic risks to humans: outdoor air pollution. http://monographs.iarc.fr/ENG/Monographs/vol109/index.php [accessed 17 October 2017].

Jerrett M, Burnett RT, Pope CA, 3rd, Ito K, Thurston G, Krewski D, et al. 2009. Long-term ozone exposure and mortality. N Engl J Med 360(11):1085–1095, PMID: 19279340, 10.1056/NEJMoa0803894.

Jin X, QY Xu Z, Chen B. 1999. Association between air pollution and mortality in Benxi [in Chinese]. Chin J Public Health 15(15):211–212.

Krewski D, Jerrett M, Burnett RT, Ma R, Hughes E, Shi Y, et al. 2009. Extended follow-up and spatial analysis of the American Cancer Society study linking particulate air pollution and mortality. Res Rep Health Eff Inst (140):5–114. discussion 115–136, PMID: 19627030.

Lepeule J, Laden F, Dockery D, Schwartz J. 2012. Chronic exposure to fine particles and mortality: an extended follow-up of the Harvard Six Cities study from 1974 to 2009. Environ Health Perspect 120(7):965–970, PMID: 22456598, 10.1289/ehp.1104660.

Lipsett MJ, Ostro BD, Reynolds P, Goldberg D, Hertz A, Jerrett M, et al. 2011. Long-term exposure to air pollution and cardiorespiratory disease in the California teachers study cohort. Am J Respir Crit Care Med 184(7):828–835, PMID: 21700913, 10.1164/rccm.201012-2082OC.

Liu S, Wu X, Lopez AD, Wang L, Cai Y, Page A, et al. 2016. An integrated national mortality surveillance system for death registration and mortality surveillance, China. Bull World Health Organ 94(1):46–57, PMID: 26769996, 10.2471/BLT.15.153148.

Nasari MM, Szyszkowicz M, Chen H, Crouse D, Turner MC, Jerrett M, et al. 2016. A class of non-linear exposure-response models suitable for health impact assessment applicable to large cohort studies of ambient air pollution. Air Qual Atmos Health 9(8):961–972, PMID: 27867428, 10.1007/s11869-016-0398-z.

Niu SR, Yang GH, Chen ZM, Wang JL, Wang GH, He XZ, et al. 1998. Emerging tobacco hazards in China: 2. Early mortality results from a prospective study. BMJ 317(7170):1423–1424, PMID: 9822394, 10.1136/bmj.317.7170.1423.

Pope CA III, Burnett RT, Turner MC, Cohen A, Krewski D, Jerrett M, et al. 2011. Lung cancer and cardiovascular disease mortality associated with ambient air pollution and cigarette smoke: shape of the exposure-response relationships. Environ Health Perspect 119(11):1616–1621, PMID: 21768054, 10.1289/ehp.1103639.

Raaschou-Nielsen O, Andersen Z, Beelen R, Samoli E, Stafoggia M, Weinmayr G, et al. 2013. Air pollution and lung cancer incidence in 17 European cohorts: prospective analyses from the European Study of Cohorts for Air Pollution Effects (ESCAPE). Lancet Oncol 14(9):813–822, PMID: 23849838, 10.1016/S1470-2045(13)70279-1.

Schikowski T, Mills IC, Anderson HR, Cohen A, Hansell A, Kauffmann F, et al. 2014. Ambient air pollution: a cause of COPD?. Eur Respir J 43(1):250–263, PMID: 23471349, 10.1183/09031936.00100112.

Van Dingenen R, Leitao JDF. 2014. A multi-metric global source-receptor model for integrated impact assessment of climate and air quality policy scenarios. EGU Gen Assem In: EGU Abstracts. EGU General Assembly 2014, 27 April – 2 May 2014 Vienna, Austria. id.13949. Geophysical Research Abstracts Vol. 16, EGU2014-13949, http://adsabs.harvard.edu/abs/2014EGUGA..1613949V [accessed 17 October 2017].

van Donkelaar A, Martin RV, Brauer M, Boys BL. 2015. Use of satellite observations for long-term exposure assessment of global concentrations of fine particulate matter. Environ Health Perspect 123(2):135–143, PMID: 25343779, 10.1289/ehp.1408646.

Villeneuve PJ, Weichenthal SA, Crouse D, Miller AB, To T, Martin RV, et al. 2015. Long-term exposure to fine particulate matter air pollution and mortality among Canadian women. Epidemiology 26(4):536–545, PMID: 25894856, 10.1097/EDE.0000000000000294.

Wong CM, Lai HK, Tsang H, Thach TQ, Thomas GN, Lam KB, et al. 2015. Satellite-based estimates of long-term exposure to fine particles and association with mortality in elderly Hong Kong residents. Environ Health Perspect 123(11):1167–1172, PMID: 25910279, 10.1289/ehp.1408264.

Xu Z, Yu D, Chen B, Xu X, Jin L, Yu G, et al. 1996. Effect of air pollution on mortalities in Shenyang City [in Chinese]. Chin J Public Health 15:61–64.

Yang L, Zhou M, Smith M, Yang G, Peto R, Wang J, et al. 2010. Body mass index and chronic obstructive pulmonary disease-related mortality: a nationally representative prospective study of 220,000 men in China. Int J Epidemiol 39(4):1027–1036, PMID: 20400495, 10.1093/ije/dyq051.

Zhang P, Dong G, Sun B, Zhang L, Chen X, Ma N, et al. 2011. Long-term exposure to ambient air pollution and mortality due to cardiovascular disease and cerebrovascular disease in Shenyang, China. PloS one 6(6):e20827, PMID: 21695220, 10.1371/journal.pone.0020827.

Zhou M, He G, Fan M, Wang Z, Liu Y, Ma J, et al. 2015a. Smog episodes, fine particulate pollution and mortality in China. Environ Res 136:396–404, PMID: 25460661, 10.1016/j.envres.2014.09.038.

Zhou M, He G, Liu Y, Yin P, Li Y, Kan H, et al. 2015b. The associations between ambient air pollution and adult respiratory mortality in 32 major Chinese cities, 2006-2010. Environ Research 137:278–286, PMID: 25601729, 10.1016/j.envres.2014.12.016.

Zhou M, Liu Y, Wang L, Kuang X, Xu X, Kan H. 2014. Particulate air pollution and mortality in a cohort of Chinese men. Environ Pollut 186:1–6, PMID: 24333659, 10.1016/j.envpol.2013.11.010.

Zhou M, Offer A, Yang G, Smith M, Hui G, Whitlock G, et al. 2008. Body mass index, blood pressure, and mortality from stroke: a nationally representative prospective study of 212,000 Chinese men. Stroke 39(3):753–759, PMID: 18239175, 10.1161/STROKEAHA.107.495374.

From Intuitive to Evidence Based: Developing the Science of Nature as a Public Health Resource

PDF icon PDF Version (1.1 MB)

  • Published: 6 November 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

Nature Contact and Human Health: A Research Agenda

Howard Frumkin, Gregory N. Bratman, Sara Jo Breslow, Bobby Cochran, Peter H. Kahn Jr, Joshua J. Lawler, Phillip S. Levin, Pooja S. Tandon, Usha Varanasi, Kathleen L. Wolf, and Spencer A. Wood

Doctors nationwide have already begun giving their patients “park prescriptions,” instructions to improve their health by spending more time outdoors.1,2,3 A growing body of evidence suggests that nature, whether the green leaves of a city park or the natural sounds of a back-country wilderness, may help us think better, feel better, and possibly even live longer.4,5,6 But as the authors of a new commentary in Environmental Health Perspectives posit, before nature can truly be tapped as a public health resource, many critical research questions remain to be answered.7

“The notion that nature contact is good for people is very intuitive,” says lead author Howard Frumkin, a professor of environmental and occupational health sciences at the University of Washington. Proof is another matter, however: “There are some basic elements of this science that are just developing,” he says. “There’s an ironic disconnect between how widely held this view is and how early we are in the scientific verification. It may be true that some exposures are very helpful, and others are of little help. Understanding the layers of truth to this is very important.”

Four photographs of people in various settings near nature
What “dose” of nature is enough, and how should it be “administered” to confer potential health benefits? That’s just one of the avenues of inquiry that researchers need to pursue to advance our knowledge of the human–nature relationship. Images, clockwise from top right: © Image Source/Alamy Stock Photo; © Lumi Images/Alamy Stock Photo; © Terry Bruce/Alamy Stock Photo; © Cultura Creative (RF)/Alamy Stock Photo.

What size and type of “dose,” for example, are required to achieve a health benefit? Do we need to be outside, or is peering through a window at a planted courtyard enough? How about watching nature on a screen? What is it about time with nature that helps us: cleaner air, room to roam, quiet contemplation? How, exactly, do the benefits accumulate: physiologically, psychologically, or in some combination? And do people of different cultures experience nature differently?

These are just a few of the questions raised by the commentary, an interdisciplinary effort from the University of Washington that draws upon environmental health, conservation biology, public policy, pediatrics, forestry, and psychology. The authors take a holistic perspective on a subject that is often expressed in simpler terms, such as the relationship between neighborhood green space and risk of type 2 diabetes,8 or living near a park and level of physical activity,9 or views of landscaped spaces and relief from stress and mental fatigue.10 Although such linear, reductionist approaches in research are useful, they should be balanced by more complex systems-based thinking, the authors write.

Even more critically, as the examples above illustrate, the field also must work toward more standardized and reliable definitions of nature “exposure,” the authors note, writing, “Despite the centrality of exposure assessment in epidemiologic research, there is little agreement on how best to define nature contact for research purposes, let alone how to measure it.”7

The implications of getting it right could reach even beyond human health, says study coauthor Josh Lawler, a professor of ecology at the University of Washington. A maturing evidence base could support policies that protect natural landscapes and biodiversity at the same time as human well-being. “Will the aspects of nature that give us these benefits, whether it is stress relief or more rapid healing or other psychological benefits … also provide benefits on the conservation and biodiversity side as well?” he asks.

Rooted in environmental health, the field is indeed growing more interdisciplinary—and for good reason, says Valentine Seymour, a Ph.D. candidate at University College London who authored a 2016 review of research into the relationship between nature and human health.11 “I found quite a lot of existing studies across a broad spectrum of disciplines, and there is a need to bring these together,” she says. “Examining the human–nature relationship from a single disciplinary perspective could lead to partial findings that neglect other important sources.” By contrast, adopting mixed-method approaches and what Seymour calls a “pragmatic outlook” accounting for real-world political, economic, and social forces should support the field’s continued growth.

Payam Dadvand, a senior researcher at the Barcelona Institute for Global Health who was not affiliated with the new commentary, agrees that an important goal going forward will be designing studies whose results can be readily translated into policy. “For example,” he says, “a ten-tree increase around a residential address gives X amount of benefit.”

Local governments in the Pacific Northwest are already clamoring for guidance in designing green infrastructure that can protect both water quality and human health, says coauthor Bobby Cochran, executive director of the Portland, Oregon–based nonprofit Willamette Partnership.12,13 “They are seeing the body of research out there that’s showing that there are benefits, and they are saying, ‘Great, tell me how best to direct my investment.’”


Nate Seltenrich covers science and the environment from Petaluma, California. His work has appeared in High Country News, Sierra, Yale Environment 360, Earth Island Journal, and other regional publications.

References

1. Seltenrich N. 2015. Just what the doctor ordered: using parks to improve children’s health. Environ Health Perspect 123(10):A254–A259, PMID: 26421416, 10.1289/ehp.123-A254.

2. Richards K. 2017. How nature heals: why East Bay doctors are prescribing the outdoors to people of color. East Bay Express, News & Opinion section, online edition. 18 May 2017. https://www.eastbayexpress.com/oakland/how-nature-heals-why-east-bay-doctors-are-prescribing-the-outdoors-to-people-of-color/Content?oid=6788567 [accessed 31 July 2017].

3. Melamed S. 2017. Philly doctors are now prescribing park visits to city kids. Philly.com, Health—Kids & Family section. 6 July 2017. http://www.philly.com/philly/health/kids-families/why-philly-doctors-are-prescribing-park-visits-to-city-kids-20170706.html [accessed 31 July 2017].

4. James P, Hart JE, Banay RF, Laden F. 2016. Exposure to greenness and mortality in a nationwide prospective cohort study of women. Environ Health Perspect 124(9):1344–1352, PMID: 27074702, 10.1289/ehp.1510363.

5. Hartig T, Mitchell R, de Vries S, Frumkin H. 2014. Nature and health. Annu Rev Public Health 35:207–228, PMID: 24387090, 10.1146/annurev-publhealth-032013-182443.

6. James P, Banay RF, Hart JE, Laden F. 2015. A review of the health benefits of greenness. Curr Epidemiol Rep 2(2):131–142, PMID: 26185745, 10.1007/s40471-015-0043-7.

7. Frumkin H, Bratman GN, Breslow SJ, Cochran B, Kahn PH Jr, Lawler JJ, et al. 2017. Nature contact and human health: a research agenda. Environ Health Perspect 125(7):075001, PMID: 28796634, 10.1289/EHP1663.

8. Astell-Burt T, Feng X, Kolt GS. 2014. Is neighborhood green space associated with a lower risk of type 2 diabetes? Evidence from 267,072 Australians. Diabetes Care 37(1):197–201, PMID: 24026544, 10.2337/dc13-1325.

9. 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.

10. Li D, Sullivan WC. 2016. Impact of views to school landscapes on recovery from stress and mental fatigue. Landscape Urban Plan 148:149–158, 10.1016/j.landurbplan.2015.12.015.

11. Seymour V. 2016. The human–nature relationship and its impact on health: a critical review. Front Public Health 4:260, PMID: 27917378, 10.3389/fpubh.2016.00260.

12. Wolf KL. 2014. Water and wellness: green infrastructure for health co-benefits. Stormwater Report, The Report section. 2 April 2014. http://stormwater.wef.org/2014/04/water-wellness/ [accessed 4 October 2017].

13. Oregon Solutions. 2017. Jade Greening Project. http://orsolutions.org/osproject/jade-greening-project [accessed 4 October 2017].

Biological Cleavage of the C–P Bond in Perfluoroalkyl Phosphinic Acids in Male Sprague-Dawley Rats and the Formation of Persistent and Reactive Metabolites

Author Affiliations open

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

PDF icon PDF Version (1.2 MB)

  • Background:
    Perfluoroalkyl phosphinic acids (PFPiAs) have been detected in humans, wildlife, and various environmental matrices. These compounds have been used with perfluoroalkyl phosphonic acids (PFPAs) as surfactants in consumer products and as nonfoaming additives in pesticide formulations. Unlike the structurally related perfluoroalkyl sulfonic and carboxylic acids, little is known about the biological fate of PFPiAs.
    Objectives:
    We determined the biotransformation products of PFPiAs and some pharmacokinetic parameters in a rat model.
    Methods:
    Male Sprague-Dawley rats received an oral gavage dose of either C6/C8PFPiA, C8/C8PFPiA, or C8PFPA. Blood was sampled over time, and livers were harvested upon sacrifice. Analytes were quantified using ultra-high-performance liquid chromatography–tandem mass spectrometry or gas chromatography–mass spectrometry.
    Results:
    PFPiAs were metabolized to the corresponding PFPAs and 1H-perfluoroalkanes (1H-PFAs), with 70% and 75% biotransformation 2 wk after a single bolus dose for C6/C8PFPiA and C8/C8PFPiA, respectively. This is the first reported cleavage of a C-P bond in mammals, and the first attempt, with a single-dose exposure, to characterize the degradation of any perfluoroalkyl acid. Elimination half-lives were 1.9±0.5 and 2.8±0.8 days for C6/C8PFPiA and C8/C8PFPiA, respectively, and 0.95±0.17 days for C8PFPA. Although elimination half-lives were not determined for 1H-PFAs, concentrations were higher than the corresponding PFPAs 48 h after rats were dosed with PFPiAs, suggestive of slower elimination.
    Conclusions:
    PFPiAs were metabolized in Sprague-Dawley rats to form persistent PFPAs as well as 1H-PFAs, which contain a labile hydrogen that may undergo further metabolism. These results in rats produced preliminary findings of the pharmacokinetics and metabolism of PFPiAs, which should be further investigated in humans. If there is a parallel between the disposition of these chemicals in humans and rats, then humans with detectable amounts of PFPiAs in their blood may be undergoing continuous exposure. https://doi.org/10.1289/EHP1841
  • Received: 1 March 2017
    Revised: 20 September 2017
    Accepted: 21 September 2017
    Published: 3 November 2017

    Address correspondence to S. Joudan, Department of Chemistry, University of Toronto, 80 St. George St., Toronto, ON M5S 3H6 Canada. Telephone: (416) 946-7736. Email: shira.joudan@mail.utoronto.ca

    Supplemental Material is available online (https://doi.org/10.1289/EHP1841).

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

    †Current address: Man-Technology-Environment (MTM) Research Centre, Örebro University, Örebro, Sweden.

    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 (145 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 (66 KB)



Introduction

Perfluoroalkyl acids (PFAAs) are ubiquitous environmental contaminants that arise from direct usage as surfactants and as the ultimate breakdown products of many larger fluorinated chemicals (Butt et al. 2014; D’eon and Mabury 2011; Young and Mabury 2010). Strong carbon–fluorine bonds make perfluoroalkyl chains resistant to biological degradation as well as stable in wide temperature, pH, and pressure ranges (Kissa 2001). The persistence, toxicity, and bioaccumulation potential of certain perfluoroalkyl chain lengths of perfluoroalkyl carboxylic and sulfonic acids (PFCAs and PFSAs, respectively) have resulted in numerous phase-outs and bans worldwide (3M 2000; U.S. EPA 2006). Although humans are exposed to an increasing amount of unknown organofluorine-containing molecules (Yeung and Mabury 2016), most pharmacokinetic research has focused on PFCAs and PFSAs.

Although they are structurally similar to PFCAs and PFSAs, there is a paucity of literature on perfluoroalkyl phosphinic and phosphonic acids (PFPiAs and PFPAs, respectively), which have the general structure of [F(CF2)x][F(CF2)y]PO2 and F(CF2)xPO32-, respectively (Figure 1). PFPiAs and PFPAs with chain lengths of x=6, 8, 10, 12 and y=6, 8 (x+y≤18 for PFPiAs) have been identified together in commercial mixtures for use as wetting agents in consumer products and have also been used as antifoaming additives to pesticide formulations (D’eon et al. 2009; Wang et al. 2016). Their current usage patterns are not fully understood, and they have often been incorrectly grouped together with mono- and di-polyfluoroalkyl phosphate esters in regulatory documents. The first environmental observation of any PFPiA or PFPA was when C6, C8, and C10 PFPAs were detected in Canadian surface water and wastewater treatment plant effluent samples collected from 2004–2007 (D’eon et al. 2009). PFPiAs were not targeted analytes in that research.

Three chemical bonding structures of C sub 6 virgule C sub 8 PFPiA, perfluorohexylperfluorooctyl phosphinic acid 802 amu; C sub 8 virgule C sub 8 PFPiA, bis(perfluorooctyl) phosphinic acid 902 amu; and C sub 8 PFPA, perfluorooctyl phosphonic acid 500 amus.

Figure 1. Structures of the PFPiAs and PFPAs investigated in this study. Molar masses are listed as the fully protonated forms of the molecules, but the structures are depicted with charge states at pH 7. amu, atomic mass units; PFPA, perfluoroalkyl phosphonic acid; PFPiA, perfluoroalkyl phosphinic acid.

To our knowledge, there have only been three studies that reported screening for PFPiAs and PFPAs in humans. Although many monitoring programs measure other per- and polyfluorinated substances (PFAS) in the environment, PFPiAs and PFPAs are generally not included as analytes. PFPAs are more challenging to analyze owing to low analytical sensitivity caused by their minus-two charge state. Lee and Mabury (2011) searched for PFPiAs and PFPAs in human sera collected from U.S. residents in 2009; they reported PFPiAs in human sera for the first time but did not detect any PFPAs. The most commonly detected congeners were C6/C6PFPiA and C6/C8PFPiA, which were found in >50% of samples at concentrations from 4–38 pg/mL (Lee and Mabury 2011). Another study reported PFPiAs, and for the first time PFPAs, in human plasma collected as early as 1985 from two German cities (Yeung and Mabury 2016). The authors used an instrument with lower detection limits than that used by Lee and Mabury (2011), which may have allowed them to detect PFPAs. The reported instrumental limit of quantification for all PFPiAs and PFPAs by Yeung and Mabury (2016) was 50 fg on-column in whole blood, compared with 0.5–9.0 pg on-column by Lee and Mabury (2011). PFPiA detections in German plasma samples were less frequent than in the American sera, but the most common congeners were also C6/C6PFPiA and C6/C8PFPiA. German plasma also contained C6 and C8 PFPAs, with all analytes ranging from <10−50 ng/L. Neither PFPiAs nor PFPAs were detected in whole blood collected in 2004 from seven Chinese cities (Yeung and Mabury 2016). These differences highlight the different usage patterns of these compounds worldwide and show that PFPiAs and PFPAs have been used for ≥30 y. A third study did not detect PFPiAs in human whole blood collected in 2010 from the Hong Kong Red Cross; however, it did not target PFPAs (Loi et al. 2013).

Athough there have only been a handful of measurements in humans, there have been additional reports of other environmental PFPiA and PFPA contaminations. Humans may be exposed through household dust (De Silva et al. 2012) and tap water, where only PFPAs were measured (Llorca et al. 2012). Interestingly, PFPiAs have recently been detected in 100% of serum samples from dolphins, cormorants, and pike in North America, with total PFPiA levels reported at 1.87±2.17 ng/g wet weight on average (De Silva et al. 2016). Another recent paper reported PFPiA contamination in sediment from Lake Ontario and two small lakes in the province of Ontario (Guo et al. 2016).

The structure of PFPiAs suggests that they would be persistent against biological transformations because of the perfluoroalkyl chains and the stable carbon–phosphorus bonds that link the hydrophobic and lipophobic perfluoroalkyl groups of the surfactant to the polar head group. A study in juvenile rainbow trout reported the intriguing in vivo formation of PFPAs from PFPiAs, although the rest of the molecule was not accounted for (Lee et al. 2012). Despite their being detected in humans and wildlife, there is no published information on the toxicology or reactivity of these compounds in mammals.

To better understand the fate of PFPiAs in mammals, rats were dosed with either C6/C8PFPiA, C8/C8PFPiA, or C8 PFPA. To determine if PFPiAs degrade to PFPAs in rats and to establish some preliminary pharmacokinetic parameters, rats received a 50-μg/kg single bolus dose via oral gavage (n=3 per treatment). Blood was sampled over time to determine the pharmacokinetics of the individual molecules, and livers were harvested to assess distribution. We hypothesized that 1H-perfluoroalkanes (1H-PFAs) would form as the presumed carbanion leaving group of the PFPiA molecules. To test that hypothesis, a group of rats was redosed with the same molecules at 2 mg/kg (n=3 per treatment and one control). A higher concentration dosage was required because of the volatility and analytical challenges involved in measuring 1H-PFAs.

Methods

Chemicals

All PFPiA, PFPA, and PFCA standards were obtained from Wellington Laboratories. 1H-PFAs were purchased from Synquest Laboratories. For complete details of chemicals used, see Supplemental Material (“Chemicals”).

Animal Treatment and Chemical Administration

All work was performed at the University of Toronto’s Division of Comparative Medicine under an animal use protocol approved by the university’s Animal Care Committee in compliance with the guidelines of the Canadian Council on Animal Care (www.ccac.ca). All animals were treated humanely with regard to alleviation of suffering. Thirty 7-wk-old male Sprague-Dawley rats were obtained from Charles River Laboratories. The animals were housed in triplicate and were exposed to a 12-h light-dark cycle, with food and water available ad libitum. Rats were allowed to adjust to their new environment for 17 d before the first blood collection. Whole blood was collected throughout the experiment into Sarstedt Microvette® CB 300 lithium heparin vials from the tail vein when a single sample was required and from the saphenous veins when multiple samples were required in a day. Blood was collected 3 d before dosing to assess any preexisting contamination. At the time of dosing, rats were 10 wk old and weighed 372±10 g (range: 350–389 g). Three groups of 9 rats were administered either C6/C8 PFPiA, C8/C8 PFPiA, or C8 PFPA at 50 μg/kg via oral gavage at 4mL/kg without prior fasting. This concentration corresponds to 62 mmol/kg C6/C8 PFPiA, 55 mmol/kg C8/C8 PFPiA, or 100 mmol/kg C8 PFPA. Dosing solutions were prepared in a vehicle of 50:50 propylene glycol:water with 0.1% soy lecithin as an emulsifier to assist solubilization (D’eon and Mabury 2010). Three control rats were dosed with the vehicle only. Approximately 150 μL of blood was sampled 0.5 h, 2 h, 4 h, 8 h, 1 d, 3 d, 4 d, 7 d, 10 d, and 14 d postdose. Three rats from each treatment group and one control rat were sacrificed after 3 and 7 d. Blood from time points 0.5 h to 3 d had n=9 per treatment; blood from days 4 and 7 had n=6 per treatment; and blood from days 10 and 14 had n=3 per treatment. Terminal procedures were performed under 2000 mg/kg urethane anesthesia. Urethane was chosen instead of more commonly used halogenated anesthetics to prevent analytical signal overlap, which can occur in our methods. Approximately 5 mL of whole blood was collected via cardiac puncture into BD Vacutainer® tubes with lithium heparin as an anticoagulant. The animals were sacrificed by cervical dislocation, and liver samples were collected and stored in 50-mL Falcon tubes at −20°C until extraction. All blood was stored at 4°C.

To search for potential 1H-PFA metabolites of PFPAs and PFPiAs, the 10 rats sampled up to 14 d were redosed at a higher concentration (2 mg/kg) of the same chemical that they were first administered (n=3 per treatment, plus one control). This concentration corresponds to 2.5 mol/kg C6/C8 PFPiA, 2.2 mol/kg C8/C8 PFPiA, or 4.0 mol/kg C8 PFPA. The rats, then 528±34 g (range: 488–570 g), were sampled for blood after 0.5 h, 2 h, 4 h, 8 h, and 1 d. Rats were sacrificed 48 h after dosing, and blood and liver samples were collected.

Extraction Procedures

Concentrations were measured in whole blood instead of plasma or serum because of evidence showing that PFPAs may enter cellular components (D’eon and Mabury 2010). To extract 100 μL of whole blood, 300 μL of ice-cold acetonitrile (ACN) was added, and the sample was vortexed and then centrifuged at 14,000×g for 10 min at 4°C. The supernatant was transferred to a clean vial for ultra-high-performance liquid chromatography–tandem mass spectrometry (UPLC-MS/MS) analysis. Livers were homogenized using a Tissue Tearor (Biospec Products) with 1 g liver (wet weight) and 0.2 mL 1% potassium chloride solution; the homogenate was then extracted with 2.5 mL ACN acidified to pH 3 with formic acid. After centrifugation for 10 min at 4,000×g, the supernatant was transferred to a clean tube, and the extraction was repeated twice. Combined extracts were evaporated to dryness and reconstituted in 1mL methanol for analysis with UPLC-MS/MS.

Instrumental Analysis

UPLC-MS/MS.

Analysis of PFPiAs, PFPAs and PFCAs was performed using a Waters Acquity UPLC coupled to a Waters Xevo TQ-S triple quadrupole mass spectrometer operating in negative ion mode for electrospray ionization. A Waters Acquity UPLC BEH C18 column (2.1 mm×75 mm, 1.7 μm) was heated to 60°C with a flow rate of 0.5 mL/min. Solvent composition began at 95% A [0.1% ammonium hydroxide (NH4OH)] and 5% B (methanol) and was held at those conditions until 0.5 min, when it was ramped to 30% B. Between 1.0 and 5.5 min, B was increased to 95% and was held there until 6.0 min. At 6.1 min, the composition was returned to 5% B, for a total run time of 8 min. Sample injection volume was 2 μL. Mass spectrometry parameters used for all analytes have been published previously (Yeung and Mabury 2016).

Because no mass-labeled standards were available for PFPAs or PFPiAs, matrix-matched calibration curves were prepared using control rat samples. For the high-concentration experiment, samples had to be diluted 100× in methanol to analyze dosed molecules (not metabolites); therefore, calibration standards were prepared in methanol.

GC-MS with solid phase microextraction.

To probe for the formation of volatile metabolites, approximately 5mL of blood collected from the cardiac puncture of the high-dose experiment was analyzed for 1H-PFAs. Analytes were extracted using headspace solid phase microextraction (SMPE) directly into sealed Vacutainers®, which had been stored at 4°C and never opened to ensure that the volatile molecules would not escape. Blood was heated at 30°C for 20 min to develop an equilibrium between the liquid and gas phases, and the SPME fiber (100 μm polydimethylsiloxane (PDMS) on fused silica; Supelco) was exposed to the headspace for 5 min before GC-MS analysis. To quantify the 1H-perfluorohexane and 1H-perfluorooctane in the blood samples, matrix-matched calibration curves were prepared specifically for each blood sample at the same volume (3.4–6.8 mL). SPME relies on liquid/gas phase partitioning; thus, the volume must be consistent when comparing samples to standards. Owing to the low water solubility of 1H-PFAs, ACN was used as a cosolvent when preparing standards, with a final volume of <3% in the spiked blood.

Samples were analyzed using an Agilent 7890A gas chromatograph with a 5975C MSD mass spectrometer operating in negative chemical ionization mode with methane as a reagent gas. An Agilent GS-GasPro column was used (30m length, inner diameter 0.32mm) with helium as a carrier gas (1.8 mL/min). The GC inlet was in splitless mode and was heated to 200°C to desorb the analytes from the SPME fiber. The GC temperature profile ramped from 140°C to 250°C. The instrument was operated in selected ion monitoring (SIM) mode. For both 1H-perfluorohexane and 1H-perfluorooctane, the [M-HF] peaks of m/z 300 and 400, respectively, were used for quantification. The [M-H] and [M-HF-CF2] peaks were used as confirmation peaks. The retention times were 5.3 min for 1H-perfluorohexane and 8.9 min for 1H-perfluorooctane. Samples were analyzed in triplicate with analytical standard deviations of <3% for all samples.

Quality Control

Limits of detection and quantification.

Limits of detection and quantitation (LODs and LOQs, respectively) were determined using blank blood and liver matrix extracts that were spiked with a mixture of analytes. The limits were empirically determined using concentrations that produced signal-to-noise ratios of 3 for LOD and 10 for LOQ. Values determined for analysis using UPLC-MS/MS are reported as picomoles/gram blood as follows: C6/C8PFPiA, LOD: 0.0050, LOQ: 0.025; C8/C8PFPiA, LOD: 0.0020, LOQ: 0.022; C6PFPA, LOD: 0.050, LOQ: 0.10; C8 PFPA, LOD: 0.080, LOQ: 0.40; perfluorohexanoic acid PFHxA, LOD: 0.64, LOQ: 1.3; perfluoroheptanoic acid (PFHpA), LOD: 0.55, LOQ: 1.1; perfluorooctanoic acid (PFOA), LOD: 0.77, LOQ: 0.97; perfluorononanoic acid (PFNA), LOD: 0.43, LOQ: 0.86.

In liver extracts, the LODs and LOQs are reported as picomoles/gram liver as follows: C6/C8 PFPiA, LOD: 0.0010, LOQ: 0.0050; C8/C8 PFPiA, LOD: 0.0010, LOQ: 0.004; C6PFPA, LOD: 0.015, LOQ: 0.050; C8PFPA, LOD: 0.022, LOQ: 0.072; PFHxA, LOD: 0.19, LOQ: 0.64; PFHpA, LOD: 0.16, LOQ: 0.55; PFOA, LOD: 0.034, LOQ: 0.116; PFNA, LOD: 0.0080, LOQ: 0.028.

For analytes measured using SPME and GC-MS, LODs and LOQs were 8.8 and 29 pmol/g blood, respectively, for 1H-perfluorohexane, and 1.2 and 3.8 pmol/g blood for 1H-perfluorooctane.

Blanks.

Blood collected three days before the first oral gavage was free of PFPiAs, PFPAs, and all PFCAs except for PFOA. Throughout the study, PFOA levels were 0.029±0.025 ng/g [standard deviation (SD); n=17] in the control blood samples. Method LODs and LOQs for PFOA were calculated as the blank levels plus 3 or 10 times the standard deviation to determine the LOD and LOQ of PFOA in blood, respectively, resulting in an LOD of 0.11 ng/g and an LOQ of 0.28 ng/g for PFOA. All blood from control rats was free of PFPiA and PFPA contamination throughout the study, as were the livers from the control rats sacrificed 3 and 7 d after rats were dosed at 50 μg/kg. The control rat sacrificed 48 h after rats were dosed at 2 mg/kg had some contamination in the liver, which may have occurred from cross-contamination during the dissection or from general background contamination depending on the analyte. Levels were 0.088 ng/g PFOA, 0.089 ng/g PFNA, 1.1 ng/g C6PFPA, 12 ng/g C8PFPA, 2.4 ng/g C6/C8PFPiA, and 1.3 ng/g C8/C8PFPiA. All of these concentrations were at least an order of magnitude lower than the levels in the other rats; thus, the concentrations were not corrected. Analytical method blanks were extracted alongside all samples and were always clean.

Extraction recovery.

To determine the extraction recovery from whole blood, 40-ng/mL spike solutions of bovine blood (purchased from BioChemed) were prepared in triplicate. Recovery was 91±5% (SD) for C6/C8PFPiA, 90±8% for C8/C8PFPiA, 63±4% for C6 PFPA, and 67±3% for C8 PFPA. To determine the extraction recovery from liver, 1 ng of each analyte was spiked into liver homogenate in triplicate. Recovery was 90±3% for C8/C8 PFPiA, 78±4% for C8/C8 PFPiA, 65±3% for C6 PFPA, and 52±4% for C8 PFPA. Recovery for PFCAs ranged from 64–81%. Concentrations were not recovery corrected.

Stability tests.

To confirm that the formation of PFPAs and 1H-perfluoroalkanes from PFPiAs was a unique biological process, stability tests were performed using deionized water and bovine blood. Solutions of 40 ng/mL C6/C8 PFPiA were prepared in water and blood and were transferred to small vials to be extracted and analyzed over time. Concentrations of C6/C8 PFPiA were stable over one week, and there was no formation of PFPAs observed. An analogous test was performed in aqueous sodium hydroxide at pH ∼13. Here, we saw the formation of C6 PFPA and C8 PFPA over time, similar to what was observed for the alkaline hydrolysis of C4/C4 PFPiA (Emeléus and Smith 1959).

Statistical Analysis

Data analyses were performed using OriginPro 2017 (Originlab Corporation). For the purpose of calculating means, all values below the limit of detection were treated as one half of the LOD. Absorption and elimination half-lives were calculated using a one-compartment model to obtain clear preliminary estimates of pharmacokinetic values. Assuming first-order absorption and elimination, the following applies for the amount of contaminant in the blood:
start fraction dA sub B over dt end fraction equals k sub abs A sub GI minus k sub elim A sub Bwhere kabs is the first order absorption rate from the gastrointestinal tract (GI), and kelim is the first order elimination rate from the blood (B).

Both the absorption and excretion half-lives (t1/2) are calculated from the rate constants:
t sub 1 virgule 2 equals negative start fraction 0.693 over k end fraction

A plot of ln CB versus time shows a linear relationship when kabs=0, and the slope of that section is −kelim. To calculate t1/2, abs, concentrations during the absorption phase were back-extrapolated to obtain a concentration defined as CB′. (CB′−CB) represents the amount absorbed from the GI tract after exposure. A plot of ln(CB′−CB) versus time results in the slope of −kabs, which allows the t1/2, abs to be calculated. The errors for both half-lives were calculated using the errors from the slope of the line of best fit.

The extent of biotransformation of PFPiAs was calculated using the integration function to obtain the areas under the curve (AUC) of PFPiAs and PFPAs in the plots of molar concentration versus time (Figure 2; Table S2). Using the AUC allowed us to account for the extent of excretion that occurred over time. Because one PFPiA molecule yields one PFPA molecule, the percent biotransformation was calculated after 14 d using the following equation:
percent biotransformation sub 14 equals start fraction summation of AUC open parenthesis mol PFPA close parenthesis sub 14 over AUC open parenthesis mol PFPiA close parenthesis sub 14 end fraction times 100

Three bar graphs with standard errors plotting mean blood concentration in nanomoles per gram (y-axis) across C sub 6 virgule C sub 8 PFPiA, PFPAs, and 1H-PFAs for C sub 6 virgule C sub 8 PFPiA; C sub 8 virgule C sub 8 PFPiA, PFPAs, and 1H-PFAs for C sub 8 virgule C sub 8 PFPiA; and C sub 8 PFPA for the same (x-axis).

Figure 2. Concentrations of PFPiAs and PFPAs over time in rats dosed at 50 μg/kg. Data points represent the means±standard error, with n=9 up to 3 d, then n=6 up to 7 d, and then n=3. Rats dosed with C8 PFPA did not contain any detectable C8 PFPA at the last two time points. The right y-axis shows the mass concentration (nanograms/gram) for the dosed molecules to highlight Cmax values listed in Table 1. PFPA, perfluoroalkyl phosphonic acid; PFPiA, perfluoroalkyl phosphinic acid.

Comparisons between concentration levels of different analytes and biological compartments (Figures 3 and 4) did not have enough samples (n=3) for true statistical power; therefore, observations are discussed in “Results” and “Discussion.”

Three bar graphs with standard errors plotting mean blood concentration in nanomoles per gram (y-axis) across C sub 6 virgule C sub 8 PFPiA, PFPAs, and 1H-PFAs for C sub 6 virgule C sub 8 PFPiA; C sub 8 virgule C sub 8 PFPiA, PFPAs, and 1H-PFAs for C sub 8 virgule C sub 8 PFPiA; and C sub 8 PFPA for the same (x-axis).

Figure 3. Concentrations in whole blood 48 h after dosage at 2 mg/kg. Of all measured analytes, the largest concentration was of the dosed molecules. Between the two classes of metabolites formed, there were greater concentrations of 1H-PFAs than PFPAs. Each bar represents an average of 3 rats, with standard errors plotted as error bars. 1H-PFAs, 1H-perfluoroalkanes; PFPA, perfluoroalkyl phosphonic acid; PFPiA, perfluoroalkyl phosphinic acid.

First part of the figure consists of three bar graphs with standard errors plotting concentrations of PFPiAs and their PFPA metabolites at Day 3 (where 50 micrograms per gram) (y-axis) across blood and liver (x-axis) for C sub 6 virgule C sub 8 PFPiA (for C sub 6 PFPA, C sub 8 PFPA, and C sub 6 virgule C sub 8 PFPiA), C sub 8 virgule C sub 8 PFPiA (for C sub 8 virgule C sub 8PFPiA and C sub 8 PFPA), and C sub 8 PFPA for the same. The second and third parts of the figure consist of three bar graphs with standard errors plotting concentrations of PFPiAs and their PFPA metabolites at Day 7 (50 micrograms per gram) and Day 2 (2 milligrams per kilogram) (y-axis), respectively, across blood and liver (x-axis) for the same perfluoroalkyl phosphinic acids mentioned in the first part of the figure.

Figure 4. Whole blood and liver concentrations of PFPiAs and their PFPA metabolites. All liver-to-blood ratios (LBRs) were >1. At the 50 μg/kg dose, C8 PFPA LBRs were larger in rats dosed with PFPiAs than in rats dosed with C8 PFPA. At the 2 mg/kg dose, all C8 PFPA LBRs were the same regardless of whether the molecule was dosed or formed as a metabolite. Each bar represents an average of 3 rats, with standard errors plotted as error bars. PFPA, perfluoroalkyl phosphonic acid; PFPiA, perfluoroalkyl phosphinic acid.

Results

In order to begin to assess the potential for human exposure, our aim was to understand the elimination kinetics, bio-transformation, and distribution of PFPiAs and their metabolites in rats as a mammalian model.

Pharmacokinetics

After the initial oral gavage, blood concentrations of PFPiAs and PFPAs increased, corresponding to absorption from the gastrointestinal tract into the bloodstream, and then decreased as elimination occurred (Figure 2; Table S1). Whole-blood concentrations were obtained four times within the first day at 0.5, 2, 4, and 8 h postdose, with the maximum concentrations observed at 8 h for C6/C8 PFPiA and C8 PFPA and at 4 h for C8/C8 PFPiA (Table 1). The absorption half-life of C6/C8 PFPiA was the fastest, with a value of 1.3±0.4 h, followed by C8 PFPA and C8/C8 PFPiA, with values of 2.1±1.2 h and 2.7±0.5 h, respectively. The greatest variability in concentrations between the nine replicates was during the absorption phase. Of the three molecules dosed, C8/C8 PFPiA had the highest measured concentration in blood (63±6 ng/g); this value was more than twice the maximum concentration of C8 PFPA and more than three times the maximum concentration of C8/C8 PFPiA. Although the large molecular mass of C8/C8 PFPiA [902 atomic mass units (amu)] may suggest low bioavailability (Lipinski et al. 2001), our results are consistent with the bioavailability of PFPiAs observed in rats for the Masurf® FS-780 commercial mixture containing various congeners of PFPiAs and PFPAs (D’eon and Mabury 2010). The authors suggested that the relatively high bioavailability may be a result of the large mass-to-volume ratio of the fluorine atoms on the molecules, and they also suggested that the PFPiAs may be more bioavailable than PFPAs owing to the difference in their physiological charge states (D’eon and Mabury 2010).

Table 1. Pharmacokinetic parameters in Sprague-Dawley rats after oral gavage at 50 μg/kg.
Dosed molecule t1/2, Absorption, ha t1/2, Elimination, da Tmax, h Cmax, ng/g
(mean±SE)
Percent biotransformation, t=14 db
C6/C8 PFPiA 2.7±0.5 1.9±0.5 8 63±6 70%
C8/C8 PFPiA 1.3±0.4 2.8±0.8 4 17±4 75%
C8 PFPA 2.1±1.2 0.95±0.17 8 24±8

Note: —, not applicable; Cmax, maximum concentration; PFPA, perfluoroalkyl phosphonic acid; PFPiA, perfluoroalkyl phosphinic acid; t1/2 absorption, absorption half-life in blood; t1/2 elimination, elimination half-life in blood; Tmax, time of maximum concentration.

aErrors determined from the error on the slope of the line of best fit.

bPercent biotransformation was approximated for PFPiAs using PFPA metabolite concentrations.

First-order elimination kinetics were determined in blood using data collected after the maximum concentration was reached, up to 14 d postdose. This is the first mammalian study of PFPiAs and PFPAs; therefore, many unknown parameters remain, such as partitioning between compartments, including noncovalent interactions with proteins that have been observed with other PFAAs (Jones et al. 2003). Using a one-compartment model provides the first estimates of pharmacokinetic parameters. Correlation coefficients (r) ranged from 0.67 to 0.82 for elimination data and from 0.53 to 0.87 for absorption data. Elimination was most rapid for C8 PFPA, with a half-life of 0.95±0.17d, followed by C6/C8 PFPiA, with a half-life of 1.9±0.5d, and C8/C8 PFPiA, with a half-life of 2.8±0.8d (Table 1). Elimination processes for PFPiAs included both excretion and biotransformation. Throughout the 14 d after the single bolus dose, PFPAs were formed in rats dosed with PFPiAs, and the concentrations were monitored over time (Figure 2; Table S1). This finding confirmed the preliminary observation of PFPA formation in juvenile rainbow trout dosed with PFPiAs (Lee et al. 2012). All samples were also analyzed for PFCAs as potential breakdown products of PFPiAs and PFPAs. None of these compounds was above the limit of detection, except for PFOA, which could not be confirmed owing to low-level contamination. Therefore, there was no evidence of degradation for PFPAs. C6 and C8 PFPAs were formed in rats dosed with C8/C8 PFPiA, and only C8 PFPA was formed in rats dosed with C6/C8 PFPiA. Detection first occurred after 29 h, and after 14 d, the concentrations of the PFPA metabolites exceeded the concentration of the parent PFPiA molecules. After 14 d, the percent biotransformations of C6/C8 PFPiA and C8/C8 PFPiA to PFPAs were approximately 70% and 75%, respectively, calculated using the AUC for PFPA and PFPiA molar blood concentrations over time (Figure 2; Table S2). Biotransformation appeared to be the dominant elimination process of PFPiAs in rats, which corroborates the low proportion of PFPiAs found in urine and feces of rats dosed with the commercial mixture (D’eon and Mabury 2010).

The whole-blood elimination half-lives we report for C6/C8 PFPiA and C8/C8 PFPiA align with those reported by D’eon and Mabury (2010) after intraperitoneal injection of the Masurf ® FS-780 commercial mixture that contained PFPiAs and PFPAs. However, the elimination half-life of C8 PFPA that we report is 0.95±0.17d, compared with the 1.6±0.1d reported for male rats by D’eon and Mabury (2010). The simultaneous elimination of PFPAs and formation of PFPAs from PFPiAs in the mixture masked the absolute elimination of the dosed PFPA molecules, resulting in an apparent elimination rate two times slower than that reported herein. Thus, PFPA elimination in rats is faster than originally reported.

Formation of 1H-PFAs

To completely characterize the biotransformation of PFPiAs, the entire parent molecule must be accounted for in the products. We hypothesized that cleavage of the C–P bond would yield the corresponding PFPAs and 1H-PFAs, presumably via protonation of a carbanion leaving group. When rats were dosed at 2 mg/kg, 1H-perfluorohexane (1H-PFHx) was detected in the blood of rats dosed with C6/C8 PFPiA, and 1H-perfluorooctane (1H-PFO) was detected in the blood of rats dosed with either C6/C8 PFPiA or C8/C8 PFPiA. 1H-Perfluorooctane was not detected in the blood of rats dosed with C8 PFPA, meaning that PFPAs themselves do not undergo a similar hydrolysis of the C–P bond in vivo. Blood concentrations from the 2 mg/kg dose are reported in Figure S1 and Tables S3 and S4.

Upon sacrifice, the blood from rats dosed with either of the two PFPiA molecules contained more 1H-PFA metabolites than PFPA metabolites (Figure 3). Assuming one PFPiA equivalent is broken down to one PFPA equivalent and one 1H-PFA equivalent, their concentrations would be equal in the blood if they had the same behavior in the body. Rats dosed with C6/C8 PFPiA contained 22 times (±5 SD, n=3) as much 1H-perfluorohexane as C8 PFPA and 23±9 times as much 1H-PFHx as C6 PFPA. Rats dosed with C8/C8 PFPiA contained 7.2±1.3 times as much 1H-PFO as C8 PFPA (Figure 3). A sample size of three rats per treatment was insufficient for meaningful statistical comparisons; therefore, this observation should be further investigated. Either PFPAs were being excreted faster, or they were partitioning out of the blood into other biological compartments. The blood elimination half-life of C8 PFPA (determined to be <1d; Table 1) aligns with the faster disappearance of PFPAs compared with 1H-PFAs in blood. Unlike PFPAs, 1H-PFAs do not contain a polar functional head group and are sparingly soluble in water, which would limit renal excretion and may lead to bioaccumulation. It is important to note that the 1H-PFAs were measured in blood but may be even more concentrated in less-aqueous biological compartments. We chose to measure 1H-PFAs in blood because the use of headspace SMPE allowed us to preclude analyte losses through volatilization during sample preparation. Subsequent studies investigating the disposition of 1H-PFAs will require robust methods in a broader range of sample matrices. At the present time, there is no literature on the elimination route and kinetics of 1H-PFAs.

Partitioning into Liver

The literature indicates that PFPAs and PFPiAs predominantly partition into proteinaceous parts of rats (D’eon and Mabury 2010) and fish (Lee et al. 2012), with the highest concentrations found in the liver. Understanding the partitioning of both parent and metabolite molecules in the body can provide insight into where metabolism is occurring. In our work, all analytes had liver-to-blood ratios (LBRs) >1 (Figure 4; Tables S5 and S6). In rats dosed at 50 μg/kg, the highest LBR was 47±12 (SD; n=3) for C8/C8 PFPiA after 3 d. This value decreased to 4.2±0.5 after 7 d. C8 PFPA measured in the rats dosed with that molecule only had a lower LBR than rats dosed with either C6/C8 PFPiA or C8/C8 PFPiA, which formed C8 PFPA as a metabolite (Figure 4; Table S6). This observation may give insight into the liver’s role in the metabolism: If the PFPA metabolites are being formed in the liver, their concentration will be the highest there before being dispersed throughout the bloodstream. Lee et al. (2012) reported greater LBRs for PFPiAs than for PFPAs in juvenile rainbow trout dosed with those molecules. It is possible that the higher concentration of PFPiAs in the liver allows for higher concentrations of PFPAs to be formed as metabolites rather than if the PFPAs were simply transferred from the blood to the liver. When rats were dosed at 2 mg/kg, the LBRs of C6/C8 PFPiA and C8/C8 PFPiA were lower after 2 d than they were after 3 d at the lower concentration dosage. Additionally, there was no difference in the LBRs of C8 PFPA based on whether it was dosed to the rats or formed as a metabolite in vivo. If metabolism is occurring in the liver, there may be a threshold that is being reached at this higher concentration.

Discussion

PFPiAs have been detected in humans, wildlife, and the environment, whereas PFPA contamination has been reported less frequently. Surprisingly, given the interesting chemistry surrounding PFPiAs and PFPAs, few researchers have included them in their targeted analyte list. Additionally, PFPAs are relatively less sensitive than other PFAS. The aim of this work was to determine if PFPiAs are metabolized to PFPAs in rats, similar to what was observed in juvenile rainbow trout (Lee et al. 2012). After confirming that observation, we then determined that the other half of the PFPiA molecule was forming 1H-PFAs by detecting them in the blood of rats exposed to higher concentrations of PFPiAs. Rats exposed to C8 PFPA did not form the corresponding 1H-PFA, confirming that PFPAs are persistent in a manner similar to all other PFAAs. 1H-PFAs contain a labile hydrogen atom that may undergo further biotransformation. We will discuss possible metabolic end points of 1H-PFAs as they relate to literature information on 1H-perfluoroethane (Harris et al. 1992) and 8:2 fluorotelomer alcohol (Martin et al. 2009; Rand and Mabury 2014).

Cleavage of the C–P Bond

This work detected the first metabolism of any perfluoroalkyl acid (PFAA) identified in a mammalian species, where the C–P bond of PFPiAs was enzymatically cleaved in rats. However, we did not observe a similar cleavage of the C–P bond in PFPAs, which would result in 1H-PFAs and phosphate ions. The phosphorus atoms in PFPAs are relatively more electron-rich than the phosphorus atoms in PFPiAs, which may make them less susceptible to nucleophilic attack. PFPAs have one fewer electron-withdrawing perfluoroalkyl group than PFPiAs, and they also have an additional acidic group, which donates electrons to the phosphorus atom at biological pH. Although both molecules are anions, the single versus double charges of PFPiAs and PFPAs, respectively, may have significant impacts on their reactivity if it is enzymatically controlled.

To our knowledge, this is the first reported cleavage of the C–P bond in mammals. The only reported C–P enzymes are in bacteria such as Escherichia coli (McGrath et al. 2013), but PFPiAs could not undergo the same oxidative mechanisms owing to the location of the fluorine atoms on the molecules. One hypothesis is that enzyme promiscuity is occurring, wherein an enzyme catalyzes a reaction other than that for which it appears to be designed. An example of this phenomenon is paraoxon, the active metabolite of the pesticide parathion, which is metabolized by enzymes with native lactonase activity (Khersonsky and Tawfik 2005).

The observed liver-to-blood ratios suggest that biotransformation occurred in the liver. The continuous formation of PFPA metabolites excludes the gut as a major location of metabolism. After a single bolus dose of either C6/C8 PFPiA or C8/C8 PFPiA, PFPA concentrations increased and then remained approximately stable throughout the 14-d experiment, even with the relatively fast elimination rate of PFPAs. For the asymmetric C6/C8 PFPiA, there was no apparent preference in the location of the C–P bond cleavage. Slightly higher levels of both C6 PFPA and 1H-PFHx were detected over C8 PFPA and 1H-PFO (Figures 2 and 3). If there had been a preference, we would have observed greater amounts of either C6 PFPA and 1H-PFO or C8 PFPA and 1H-PFHx.

Implications for Human Exposure

The PFAAs with the greatest amount of research, regulatory, and public interest have half-lives on the order of years in humans. These molecules do not undergo any biological transformations; thus, the only mode of elimination is by excretion. Perfluorooctane sulfonic acid (PFOS) contains eight perfluorinated carbons and has a half-life of 5.4 y in humans, and PFOA contains seven perfluorinated carbons and has a half-life of <3.8 years (Lau 2012). Within the carboxylic and sulfonic acid subclasses, there is a general trend of half-lives increasing with increasing numbers of perfluorinated carbons in the molecule (Martin et al. 2003). However, there are large differences between toxicokinetics in humans and those in laboratory animals in addition to prominent sex differences within some species. Furthermore, we have shown that in addition to excretion as an elimination process, PFPiAs undergo metabolism to form PFPAs and 1H-perfluoroalkanes in rats, which further complicates extrapolating kinetics to human exposure. The results presented here provide hypotheses about human exposure that should be further investigated.

Half-lives in rats are faster than in humans, and although it is difficult to compare across species, we can gain insight by relating PFPiA data to what is known about PFOS and PFOA. The half-lives of PFOS and PFOA in male rats are 43 d and <6 days, respectively (Lau 2012). Our work revealed half-lives of <3 days for C6/C8PFPiA and C8/C8 PFPiA in male rats, meaning that these molecules are eliminated approximately twice as fast as PFOA, and much faster than PFOS, in rats. Therefore, we would expect the half-lives in humans to be shorter for PFPiAs than for PFOS and PFOA, even though PFPiAs have more perfluorinated carbons. This discrepancy is largely due to metabolism as an additional route of elimination for PFPiAs, compared with only excretion for PFOS and PFOA, suggesting that humans and animals with detectable amounts of PFPiAs may be undergoing continuous environmental exposure to these chemicals. The biological impact of PFPiAs may be potentially greater than that of other PFAAs at a similar exposure level because of the formation of two prominent metabolites, PFPAs, which are persistent, and 1H-PFAs, which may undergo further metabolism. With no information on the toxicology of PFPiAs and their metabolites, it is impossible to predict the physiological effects that may be occurring. Elimination kinetics of PFPiAs and PFPAs have been investigated in juvenile rainbow trout using neat PFPiA or PFPA material separately (Lee et al. 2012) and using a commercial mixture of PFPiAs and PFPAs in rats (D’eon and Mabury 2010) and zebrafish (Chen et al. 2016). The concern with using the commercial mixture is that the elimination rate of PFPAs will appear to be slower because their formation as a metabolite of PFPiAs masks the excretion that occurs simultaneously. Within this small body of literature, there is a notable difference in elimination rates for PFPiAs in rats versus rainbow trout and zebrafish. Elimination half-lives of C6/C8PFPiA and C8/C8 PFPiA in male rats are reported here as 2.2±0.2 and 2.8±0.5 days, respectively, and by D’eon and Mabury (2010) as 1.9±0.5 and 2.8±0.8 days, respectively. However, the half-lives in rainbow trout for C6/C8 PFPiA and C8/C8 PFPiA were 20.4±4.9 and 52.7±15.8 days, respectively (Lee et al. 2012). Chen et al. (2016) found similar values in zebrafish to those reported in rainbow trout. The only data on PFPA elimination that are not affected by the use of a commercial mixture are the half-life in male rats of 0.95±0.17 days reported here and the half-life in rainbow trout of 4.4±0.7 days reported by Lee et al. (2012). Overall, rats have better capabilities for eliminating PFPiAs and PFPAs from their systems.

In this work, we chose to use male Sprague-Dawley rats because other PFAAs have much faster elimination in females than in males. Specifically, PFOA has a half-life of 2–4 h in female rats compared with 4–6 d for male rats (Lau 2012). To ensure that we could monitor metabolite formation, we chose male rats, in case the females eliminated the PFPiAs or their metabolites too quickly. This factor was particularly important because PFOA was a potential metabolite that we did look for but did not detect.

The vast differences between species for PFPiA elimination may be caused by the differences in metabolic capabilities in addition to differences in excretion. In light of these differences and the potential for sex-based differences, more research is needed to assess the exposure to PFPiAs of mammalian and aquatic species, including low-dose chronic exposure studies. In a mouse study of continuous PFOA exposure, a one-compartment kinetic model provided a poor fit for pharmacokinetic data (Lou et al. 2009). At the present time, no PFPiA or PFPA studies have used more sophisticated pharmacokinetic modeling; therefore, this should be pursued in future studies. The observed species-based differences motivate the need for more human-focused research.

Potential Biotransformation of 1H-PFAs

To our knowledge, this is the first observation of the formation of 1H-PFAs in vivo. The shorter hydrofluorocarbon analog 1H-perfluoroethane (i.e., HFC-125) has been shown to be metabolized in rats very slowly following a similar oxidative biotransformation pathway to the anesthetic halothane (Harris et al. 1992). Briefly, 1H-perfluoroethane is oxidized to an alcohol by cytochrome P450s, mainly CYP 2E1 (Dekant 1996), and then forms an acyl fluoride intermediate that can either form trifluoroacetylated proteins or trifluoroacetic acid. Halothane hepatitis is a type of liver necrosis caused by an immune response to trifluoroacetylated hepatic proteins (Ray and Drummond 1991), but in the case of halothane, the metabolism rates are much greater owing to the acyl chloride intermediate (vs. acyl fluoride), leading to more trifluoroacetylated proteins than from 1H-perfluoroethane (Harris et al. 1992). Cytochrome P450s have been proposed to be the enzymes responsible for oxidizing long-chain fluorotelomer molecules. Specifically, CYP 2E1 was reported to oxidize 8:2 fluorotelomer alcohol in rat hepatocytes, which provides evidence that compounds containing longer perfluoroalkyl chains can still undergo these oxidative reactions (Martin et al. 2009). The 8:2 fluorotelomer unsaturated aldehyde reacted to form both PFCAs and to covalently bind to proteins both in vitro (Rand and Mabury 2013) and in vivo (Rand and Mabury 2014). The longer-chained 1H-PFAs would likely have similarly slow reactivity to 1H-perfluoroethane to form PFCAs or other biological nucleophiles such as certain amino acids in proteins. However, if excretion is slow, biotransformation and subsequent reactions may be an important elimination route. In our work, we could not definitively report the formation of PFOA as a metabolite of C6/C8 PFPiA or C8/C8 PFPiA because of low-level background contamination. Future experiments will test this hypothesis using in vitro techniques to minimize PFOA contamination and to improve detection limits.

Conclusion

We have shown that PFPiAs are cleaved in vivo to form persistent PFPAs and labile 1H-PFAs in rats. This is the first reported in vivo formation of 1H-PFAs from any perfluorinated molecule; this is also the first attempt, with a single-dose exposure, to characterize the transformation of a perfluroalkyl acid. Intriguingly, there are no known mammalian enzymes that carry out the biological cleavage of C–P bonds in phosphinic acids. The strong electron-withdrawing effects of two perfluoroalkyl chains apparently increase the lability of the C–P bonds in PFPiAs compared with PFPAs, which do not undergo an analogous transformation. These preliminary results in male Sprague-Dawley rats must be investigated further to determine their relevance to human exposure. PFPiAs have been detected in humans and wildlife; therefore, there may be continuous exposure to these chemicals because these molecules are presumably being degraded and excreted over time. Given the reported environmental detections and the unique biotransformation presented here, more research is required to understand the toxicokinetics and metabolism of human PFPiA exposure, including further biotransformation of 1H-PFAs.

Acknowledgments

We thank the staff and veterinarians at the University of Toronto, Division of Comparative Medicine, particularly R. Duke and L. Peters. We thank D. Wang, of the University of Toronto, for assistance with sample preparation and J. D’eon, of the University of Toronto, and D. Jackson (York University, Toronto, Ontario, Canada) for useful discussion.

This work was funded by a National Science and Engineering Research Council Discovery grant to S.A.M.

References

3M. 2000. Phase-out Plan for POSF-based Products. U.S. EPA Public Docket OPPT-2002-0043; 3M Specialty Materials Markets Group, St. Paul, MN.

Butt CM, Muir DCG, Mabury SA. 2014. Biotransformation pathways of fluorotelomer-based polyfluoroalkyl substances: a review. Environ Toxicol Chem 33(2):243–267, PMID: 24114778, 10.1002/etc.2407.

Chen F, Gong Z, Kelly BC. 2016. Bioavailability and bioconcentration potential of perfluoroalkyl-phosphinic and -phosphonic acids in zebrafish (Danio rerio): comparison to perfluorocarboxylates and perfluorosulfonates. Sci Total Environ 568:33–41, PMID: 27285794, 10.1016/j.scitotenv.2016.05.215.

D’eon JC, Crozier PW, Furdui VI, Reiner EJ, Libelo EL, Mabury SA. 2009. Perfluorinated phosphonic acids in Canadian surface waters and wastewater treatment plant effluent: discovery of a new class of perfluorinated acids. Environ Toxicol Chem 28(10):2101–2107, PMID: 19463027, 10.1897/09-048.1.

D’eon JC, Mabury SA. 2010. Uptake and elimination of perfluorinated phosphonic acids in the rat. Environ Toxicol Chem 29(6):1319–1329, PMID: 20821575, 10.1002/etc.167.

D’eon JC, Mabury SA. 2011. Is indirect exposure a significant contributor to the burden of perfluorinated acids observed in humans? Environ Sci Technol 45(19):7974–7984, PMID: 21630688, 10.1021/es200171y.

De Silva AO, Allard CN, Spencer C, Webster GM, Shoeib M. 2012. Phosphorus-containing fluorinated organics: polyfluoroalkyl phosphoric acid diesters (diPAPs), perfluorophosphonates (PFPAs), and perfluorophosphinates (PFPIAs) in residential indoor dust. Environ Sci Technol 46(22):12575–12582, PMID: 23102111, 10.1021/es303172p.

De Silva AO, Spencer C, Ho KCD, Al Tarhuni M, Go C, Houde M. 2016. Perfluoroalkylphosphinic acids in Northern pike (Esox lucius), double-crested cormorants (Phalacrocorax auritus), and bottlenose dolphins (Tursiops truncatus) in relation to other perfluoroalkyl acids. Environ Sci Technol 50(20):10903–10913, PMID: 27677975, 10.1021/acs.est.6b03515.

Dekant W. 1996. Toxicology of chlorofluorocarbon replacements. Environ Health Perspect 104(suppl1):75–83, PMID: 8722112, 10.2307/3432698.

Emeléus HJ, Smith JD. 1959. The heptafluoropropyliodophosphines and their derivatives. J Chem Soc 375–381, 10.1039/JR9590000375.

Guo R, Megson D, Myers AL, Helm PA, Marvin C, Crozier P, et al. 2016. Application of a comprehensive extraction technique for the determination of poly- and perfluoroalkyl substances (PFASs) in Great Lakes Region sediments. Chemosphere 164:535–546, PMID: 27619064, 10.1016/j.chemosphere.2016.08.126.

Harris JW, Jones JP, Martin JL, LaRosa AC, Olson MJ, Pohl LR, et al. 1992. Pentahaloethane-based chlorofluorocarbon substitutes and halothane: Correlation of in vivo hepatic protein trifluoroacetylation and urinary trifluoroacetic acid excretion with calculated enthalpies of activation. Chem Res Toxicol 5(5):720–725, PMID: 1446014, 10.1021/tx00029a020.

Jones PD, Hu W, Coen WD, Newsted JL, Giesy JP. 2003. Binding of perfluorinated fatty acids to serum proteins. Environ Toxicol Chem 22(11):2639–2649, PMID: 14587903.

Khersonsky O, Tawfik DS. 2005. Structure-reactivity studies of serum paraoxonase PON1 suggest that its native activity is lactonase. Biochemistry 44(16):6371–6382, PMID: 15835926, 10.1021/bi047440d.

Kissa E. 2001. Fluorinated Surfactants and Repellents. 2nd ed. New York, NY:Mercel Dekker.

Lau C. 2012. Perfluoroalkyl acids: Recent research highlights. Reprod Toxicol 33(4):405–409, PMID: 22429996, 10.1016/j.reprotox.2012.03.002.

Lee H, De Silva AO, Mabury SA. 2012. Dietary bioaccumulation of perfluorophosphonates and perfluorophosphinates in juvenile rainbow trout: Evidence of metabolism of perfluorophosphinates. Environ Sci Technol 46(6):3489–3497, PMID: 22335432, 10.1021/es204533m.

Lee H, Mabury SA. 2011. A pilot survey of legacy and current commercial fluorinated chemicals in human sera from United States donors in 2009. Environ Sci Technol 45(19):8067–8074, PMID: 21486041, 10.1021/es200167q.

Lipinski CA, Lombardo F, Dominy BW, Feeney PJ. 2001. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development setting. Adv Drug Deliv Rev 46(1–3):3–26, PMID: 11259830.

Llorca M, Farré M, Picó Y, Müller J, Knepper TP, Barceló D. 2012. Analysis of perfluoroalkyl substances in waters from Germany and Spain. Sci Total Environ 431:139–150, PMID: 22683491, 10.1016/j.scitotenv.2012.05.011.

Loi EIH, Yeung LWY, Mabury SA, Lam PKS. 2013. Detections of commercial fluorosurfactants in Hong Kong marine environment and human blood: a pilot study. Environ Sci Technol 47(9):4677–4685, PMID: 23521376, 10.1021/es303805k.

Lou I, Wambaugh JF, Lau C, Hanson RG, Lindstrom AB, Strynar MJ, et al. 2009. Modeling single and repeated dose pharmacokinetics of PFOA in mice. Toxicol Sci 107(2):331–341, PMID: 19005225, 10.1093/toxsci/kfn234.

Martin JW, Chan K, Mabury SA, O’Brien PJ. 2009. Bioactivation of fluorotelomer alcohols in isolated rat hepatocytes. Chem Biol Interact 177(3):196–203, PMID: 19041856, 10.1016/j.cbi.2008.11.001.

Martin JW, Mabury SA, Solomon KR, Muir DCG. 2003. Dietary accumulation of perfluorinated acids in juvenile rainbow trout (Oncorhynchus mykiss). Environ Toxicol Chem 22(1):189–195, PMID: 12503764, 10.1002/etc.5620220125.

McGrath JW, Chin JP, Quinn JP. 2013. Organophosphonates revealed: new insights into the microbial metabolism of ancient molecules. Nat Rev Microbiol 11(6):412–419, PMID: 23624813, 10.1038/nrmicro3011.

Rand AA, Mabury SA. 2013. Covalent binding of fluorotelomer unsaturated aldehydes (FTUALs) and carboxylic acids (FTUCAs) to proteins. Environ Sci Technol 47(3):1655–1663, PMID: 23256684, 10.1021/es303760u.

Rand AA, Mabury SA. 2014. Protein binding associated with exposure to fluorotelomer alcohols (FTOHs) and polyfluoroalkyl phosphate esters (PAPs) in rats. Environ Sci Technol 48(4):2421–2429, PMID: 24460105, 10.1021/es404390x.

Ray DC, Drummond GB. 1991. Halothane hepatitis. Br J Anaesth 67(1):84–99, PMID: 1859766.

U.S. Environmental Protection Agency (EPA). 2006. PFOA Stewardship Program. U.S. EPA Public Docket OPPT-2006-0621. Washington, DC:U.S. EPA.

Wang Z, Cousins IT, Berger U, Hungerbühler K, Scheringer M. 2016. Comparative assessment of the environmental hazards of and exposure to perfluoroalkyl phosphonic and phosphinic acids (PFPAs and PFPiAs): current knowledge, gaps, challenges and research needs. Environ Int 89–90:235–247, PMID: 26922149, 10.1016/j.envint.2016.01.023.

Yeung LWY, Mabury SA. 2016. Are humans exposed to increasing amounts of unidentified organofluorine? Environ Chem 13(1):102–110, 10.1071/EN15041.

Young CJ, Mabury SA. 2010. Atmospheric perfluorinated acid precursors: chemistry, occurence, and impacts. In: Reviews of Environmental Contamination and Toxicology, Volume 208: Perfluorinated alkyl substances. de Voogt P ed. New York, NY:Springer, 1–109.

The NIH Collaboratory Launches a New Resource on Methods and Best Practices for Pragmatic Clinical Trials

The NIH Health Care Systems Research Collaboratory, an NIH Common Fund project, has been supporting nine large-scale pragmatic clinical trials in partnership with health care systems around the United States.  NCCIH and NIA have been leading the Collaboratory program, and NCCIH staff have previously blogged about its progress.

New Blood: The Promise of Environmental Health Citizen Science Projects

PDF icon PDF Version (4.5 MB)

  • Published: 2 November 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.



Mónica Ramírez-Andreotta sat on a folding chair holding a notebook and pen in the Humboldt Elementary School gym. It was August 2008, and Leah Butler, a project manager with the U.S. Environmental Protection Agency (EPA), was leading a public meeting about the recent designation of a Superfund site within the small community of Dewey-Humboldt, Arizona. The agency had determined that more than 4 million m3 of mine tailings left behind by the Iron King Mine and the Humboldt Smelter posed a health risk to the residents.1

Butler explained to the audience that wind and water erosion could carry the hazardous waste, which contained arsenic, lead, and other contaminants, from the former industrial properties into Dewey-Humboldt neighborhoods, potentially contaminating residents’ water, soil, and air. She then outlined what the agency’s cleanup efforts would entail and how the community would be involved.2

At the conclusion of the presentation, residents’ hands shot up, and Ramírez-Andreotta scribbled down their questions. Then a doctoral student at the University of Arizona, she was also employed as the research translation coordinator for the University of Arizona Superfund Research Program. The program, which is funded by the National Institute of Environmental Health Sciences, promotes multidisciplinary research on human health and environmental issues related to hazardous substances. Ramírez-Andreotta’s job that night was to listen to community members’ concerns and let them know about the program.

Photograph of students assembling kits.
Graduate students and researchers from Virginia Tech assembled and shipped 300 water testing kits for Flint citizen scientists. The kits included labeled bottles and an instruction sheet for sampling tap water. Image: © Marc Edwards/Virginia Tech.

One question she heard more than once from residents struck her as particularly interesting: Was it safe to eat the vegetables they grew in their gardens?

After the meeting, Ramírez-Andreotta approached some of the audience members, introduced herself, and made a proposal about their gardening concerns. “I cannot give you a specific answer at this point in time,” she said. “But are you interested in working together to do the research and come up with the answer?”

There is a long history of laypeople participating in scientific research, from Charles Wilson Peale’s efforts in the 18th century using public donations to create the first U.S. natural history museum3 to 20th-century homemaker Lois Gibbs canvassing her neighbors in the Love Canal community to record their children’s birth defects.4

Still, it has only been in the past few decades, since the term “citizen science” was coined in the mid-1990s,5 that the practice has been recognized formally by the scientific community.6 Although some investigators remain wary of citizen science6 (and some citizens in environmentally contested areas are skeptical that scientists will investigate their true concerns), an increasing number of studies have reaped benefits from community involvement.7

Citizen science projects also have the potential to democratize science by exposing a greater and more diverse section of the population to the scientific process. This is especially true in the environmental health research realm because a disproportionately high number of people of color and of low income—who also are underrepresented in the research community—live near environmentally compromised spaces.8

“I like to stress that now we live in a time where your ZIP code may be more important than your genetic code,” Ramírez-Andreotta says. “Where you live, your proximity to waste, your socioeconomic status, and your physical environment can, in some cases, mean more in terms of your health than the genetic makeup that you were born with.”

Citizen science projects involving environmental health research can be challenging for both investigators and residents in affected communities however. Some scientists, for instance, may worry they will be labeled environmental activists and their scientific neutrality compromised if they help a group of concerned citizens to investigate potential environmental contamination. Others may wonder if they can obtain reliable data from people who have no formal scientific training. On the flipside, residents may become frustrated if their health concerns are not taken seriously and if they are not included equitably in decision making and data sharing while the investigation is ongoing.9

Screen shot from the instructional video that is labeled as follows: Sampling Water for Lead in Flint MI – Instruction Video.
The Virginia Tech students also produced an instructional video (https://www.youtube.com/watch?v=dEQDaPws2xk) to show how to correctly collect tap water samples using the kits. Correct sampling is critical to accurately capture how much lead residents may be ingesting through drinking water. Image: Courtesy of YouTube.

Concerns and Confidence

Marc Edwards, a professor at Virginia Polytechnic Institute and State University, stood on the lawn of City Hall at Flint, Michigan, at a press conference. He held two small plastic bottles, one filled with orange water. Next to him was LeeAnne Walters, mother of four, who several months earlier had sent Edwards 30 samples of her tap water. Tests showed the water to have high levels of lead—one sample had more than 1,300 times the World Health Organization’s maximum acceptable limit of 10 ppb for lead in water. “It was the worst I’ve seen in 25 years,” Edwards says. He estimated at the press conference that 5,000 Flint homes had tap water lead levels that exceeded 10 ppb.

After he measured the lead in Walters’s water, Edwards and his graduate students distributed water test kits to Flint residents. His students also made a video10 explaining how to take samples of drinking water, and local community groups and representatives of the American Civil Liberties Union held training sessions in the basement of a local church.

Edwards says he was confident that the Flint residents could collect the water samples properly. His assurance stemmed from several factors. For one, local water utilities already routinely rely on citizens to send in their own samples to determine compliance with the Lead and Copper Rule, which is enforced by the U.S. Environmental Protection Agency.11 In addition, his previous experience of working with residents of Washington, DC, whose water had been found to have high lead levels years earlier, showed him that most laypeople who suspect pollution in their water are, as he puts it, “very careful and more concerned than most scientists about making a sampling mistake.” And finally, he points out that the U.S. EPA protocol allows for some mistakes and that his team looks for sampling anomalies.

Under the Lead and Copper Rule, the agency throws out the 9 samples with the highest concentrations of lead from every 100 collected. That is because, although there is no maximum contaminant level (MCL) for lead, utilities must take corrective action if more than 10% of households sampled have concentrations of the metal above the action level set by the U.S. EPA.11

Despite Edwards’s confidence, sampling error in citizen science projects is something that concerns many scientists. In a 2014 study, investigators Hauke Riesch and Clive Potter conducted qualitative interviews with scientists participating with the Open Air Laboratories network, a citizen science network in the United Kingdom. They found that data quality was “a clear area of worry for the majority of interviewees.”6

To combat that concern, the scientists interviewed by Riesch and Potter employed a range of methods, from extensive training and supervision, to cross-checking the data with their own observations and/or data from previously published studies, to simplifying research questions and the data collection protocols as much as possible. “Needless to say,” they wrote, “the issues about data quality were in the end solved to the satisfaction of the participating scientists and therefore represented no stumbling block for the enterprise as a whole.”6

Ramírez-Andreotta followed a similar protocol in Dewey-Humboldt. She held extensive training sessions with the residents and conducted a controlled greenhouse study growing similar vegetables in three different soil types.12 She also randomly selected households where she collected her own samples to compare against the residents’ samples. She was exceedingly careful, she says, “because I knew if we observed a contaminant of concern at an elevated concentration, that one of the first things others might want to challenge is the methodology; they’d say, ‘Oh, they did not know how to collect the samples.’”13

Photographs of the interior and exterior of the bucket sampler.
An early bucket sampler used by the Louisiana Bucket Brigade employed a hand-held vacuum cleaner to pump the air out of a five-gallon plastic container, then draw ambient air into through a stainless steel inlet to a nonreactive plastic bag. Image: © Gwen Ottinger/Drexel University.

Finding Common Ground

So what does it take for a regulatory agency to take the work of citizen scientists seriously enough to influence research agendas, affect policy, or change scientific standards of proof? Gwen Ottinger, an associate professor in Drexel University’s Center for Science, Technology and Society, has investigated that very question. Beginning in 2001 she looked at whether citizen science projects conducted by “fenceline” communities—those located adjacent to oil refineries and chemical manufacturing plants—were able to influence state and federal regulatory agencies. The focus of her study was the community of New Sarpy, Louisiana, which adjoins a Shell Chemical plant.14

Residents of New Sarpy worried that their health was being harmed by pollution spikes during events such as gas flaring and venting, plant start-ups, malfunctions, and accidents.15 They used “bucket sampling,” an inexpensive method for grabbing air samples at discrete points in time, to provide evidence that their industrial neighbors occasionally released high levels of dangerous chemicals into the air. As the name suggests, air is drawn into a nonreactive plastic bag inside a bucket, then the bag’s valve is closed, and the bag is shipped overnight to a laboratory for analysis.

Regulatory clean air standards are designed to reflect pollution measurements averaged over longer periods of time. For instance, regulators use sampling instruments that may sample continuously over 24 hours, every sixth day.14 This mixes air collected during pollution peaks with the relatively clean air present during the rest of the sampling period. The results, Ottinger wrote in a 2010 article, render “pollution spikes invisible in the process of comparing air quality measurements to air quality standards.”14

Because bucket sampling does not align with federal monitoring standards, Ottinger says the validity of the citizens’ results was questioned by federal and state scientists. This presented an impasse of sorts, she says, with the scientists claiming the data were not legitimate and the citizens claiming the standards do not address the “right questions.”14 She believes this difference of views toward citizen science—one that is “scientific authority-driven” citizen science and the other “social movement-based” citizen science, needs to be addressed if the true potential of citizen science is ever to be fulfilled.

“For people in the environmental justice movement who are trying to make a change, science is a mixed bag,” Ottinger says, “because a lot of versions of science tell them you’re not experiencing what you’re experiencing, whereas the environmental justice citizen scientists want to ask a research question that makes visible what they know from experience… . It’s probably more of a spectrum than two poles, but they are important distinctions.”

Photograph of a man using a bucket sampler.
A citizen scientist collects air samples near the Norco Shell Chemical Plant. The samples provided evidence that the plant occasionally released high levels of pollutants that were not reflected in regulatory measurements averaged over longer periods of time. Image: © Louisiana Bucket Brigade.

What is the solution? In the case of the Louisiana bucket brigade, some consensus was reached because the laboratory that analyzed the citizens’ samples used the same method that the U.S. EPA uses to analyze its own samples.16 That made the two sets of results directly comparable. In addition, the activists had used a U.S. EPA laboratory in California to conduct quality assurance testing on the use of the bucket sampler. Thus, they could argue that they were using a “U.S. EPA-approved” monitoring method. This allowed the bucket data to be received with some credibility among the research scientists.17

Another solution is for activists to work to change the air quality standards. “Standards are meant to be ‘invisible’ so we do not have to think about them,” Ottinger says. It takes time and effort to lobby for change, and so it is not necessarily a place where social movement groups would want to spend their limited resources.14 Nevertheless, she says, citizen scientists may have to get deeper into the weeds and figure out how National Ambient Air Quality Standards are set and how they can make their voices heard when the standards are periodically reviewed. Anytime such a review is set to take place, there is a science policy workshop to gather input from both the scientific community and the public regarding policy-relevant issues and questions that can help inform the review.18

Overall, Ottinger believes bucket testing and other citizen science pollution monitoring projects are pushing the U.S. EPA to be more proactive in finding methods to help citizens and the agency work together. She points to the U.S. EPA’s new website, the Air Sensor Toolbox for Citizen Scientists, Researchers, and Developers,19 which was designed to help members of the public choose from the plethora of low-cost sensors now available to conduct air quality monitoring. The site also provides information on funding, training, and pilot studies for citizen scientists.19

At a 2015 workshop where community scientists received training in the use of the U.S. EPA’s Air Sensor Toolbox,19 agency scientists told the attendees that using such tools would make it easier for the agency to act on the data they gather. “They said if you want [regulators] to take your data seriously, you need to have them involved from the very beginning, consult with them, tell them what is coming, how you did it,” Ottinger recalls.

Although a tool that helps citizens and regulatory scientists align their goals makes sense, it is still unlikely that even this will erase all the debate between lay and regulatory scientists, Ottinger says. “The rejoinder from some in the audience was, ‘Well, wait a minute, we did all that, and it still did not work for us, because you did not like what you were seeing in the data.’”

Agency scientists may fear that citizen scientists have undertaken data collection with a biased mindset—for instance, convinced that a pollutant is causing health problems—which could result in skewed sampling. Ottinger, however, argues that citizens understand that any data they collect will be carefully scrutinized by agency scientists for evidence of problems. So the citizens have a strong motivation to collect high-quality data in a nonbiased way, in addition to the intrinsic motivation of protecting their health and families, says Ramírez-Andreotta.

Image of a page from the instructional manual.
As part of the Gardenroots project, researchers developed an instruction manual (https://superfund.arizona.edu/sites/superfund.arizona.edu/files/photofiles/gardenroots_instructional_manual.pdf) to help participants properly collect soil, water, and vegetable samples from their homes. Image: © University of Arizona.

Despite concerns, the U.S. EPA is investigating the possibilities of citizen science. “Data quality is always a priority when measuring air pollution, and EPA has well-defined data collection methods and guidelines to ensure that the data the agency uses in regulatory decisions is of the highest quality,” says an agency spokesperson who requested anonymity per agency policy. “Engaging local, state, or federal environmental agencies in the early planning of a citizen science project will help in determining the data quality required for intended applications.”

At the end of 2016 an advisory council established by the U.S. EPA urged the agency to “embrace citizen science” and recommended several actions the agency could take to maximize citizen science and integrate it into its work. These included embracing citizen science as a core tenet of environmental protection; dedicating funding for citizen science for communities, partners, and the agency; enabling the use of citizen science data at the agency; and integrating citizen science into the full range of work at the U.S. EPA.20

Co-creating Citizen Science

In 2004, Rose Eitemiller, a resident of Dewey-Humboldt, bought a newly constructed house on Sweet Pea Lane, near the Humboldt smelter. Eitemiller’s real estate agent had assured her the house could not have been built if the land were contaminated. A few years later, though, after she and her husband found debris in their yard connected to the smelter, Eitemiller called the U.S. EPA project manager who was analyzing the mine tailing site.

“I said, ‘I am wondering if we’re contaminated, too,’ and she said, ‘We haven’t even thought about looking over there,’ and I said, ‘You are kidding!’” recalls Eitemiller. She later had her young son tested, with the results showing he had elevated levels of arsenic in his urine that exceeding the 50th percentile for his age range. The U.S. EPA eventually included the smelter in its Superfund listing, and removed and replaced about 61 cm (2 ft) of contaminated soil in most of the yards on Eitemiller’s street.21

Eitemiller’s experiences align with Ramírez-Andreotta’s notion that residents who live near a hazardous waste site are knowledgeable about the site and intrinsically motivated to learn about the potential negative ecological and health outcomes posed by the site. This makes them experts in their own right, she says, and a group worth listening to and working alongside.

“When you’re working with families living near contamination, they have a great deal of insight and important viewpoints,” she says. “They’re the ones taking the pictures of the mine tailing waste blowing off-site on windy days. They’re the ones who are perhaps distressed and experiencing illness.”

After attending that first U.S. EPA meeting in August 2008, Ramírez-Andreotti continued to return to Dewey-Humboldt and talk to residents. Eventually she obtained a U.S. EPA grant to conduct a citizen science project to characterize the uptake of arsenic by vegetables commonly grown in the Dewey-Humboldt community. She named the project Gardenroots: The Dewey-Humboldt Arizona Garden Project.13

Photograph of vegetables awaiting testing.
Each Gardenroots participant received the results of the arsenic testing on his or her homegrown vegetables. Afterward, many indicated they planned to take new precautions to avoid arsenic exposure through gardening. Image: © Mónica Ramírez-Andreotta/University of Arizona.

From the start, Ramírez-Andreotta was determined to make Gardenroots as equitable as possible by having the residents decide the research question, design how the experiment would be conducted on their properties, take the samples, and learn how to interpret the results. This approach is called “co-created citizen science.”22

Community members, in addition to providing samples of irrigation water and soil, collected vegetables that they grew. The garden crops spanned the major plant families12 and ultimately provided an invaluable data set, Ramírez-Andreotta says. By taking a public participatory approach, she says, the study was more applicable to the community, based upon what they actually grew and ate, and was thus fundamentally aligned with their research questions.

After analyzing residents’ vegetable, soil, and water samples for arsenic, Ramírez-Andreotta faced the challenge of communicating the probability of developing cancer from exposure to the arsenic levels within the various media. Late one night, she sat in front of her computer, designing customized booklets that she planned to give each resident listing the contamination levels in their soil, water, and vegetables. But how to explain their risk?

The U.S. EPA uses a risk assessment threshold of 1 cancer case in 10,000 as a basis for determining whether to clean up a Superfund site.23 If risk posed by a given environmental pollutant falls between 1 in 10,000 and 1 in 1 million cases, this is deemed an acceptable risk.23 Ramírez-Andreotta initially decided to frame the residents’ results in terms of risk that would fall right in the middle of that range; she would give residents a chart listing how much of each vegetable they could eat from their gardens before hitting a risk level of 1 in 100,000.

But then she had an “aha” moment. “I realized I had chosen their risk level for them!” she says. She went back to her computer and changed each booklet, making charts with three risk categories: 1 in 10,000, 1 in 100,000, and 1 in 1 million so residents could decide on their own how much they felt comfortable eating. She also gave them the raw data for their individual samples of vegetables, soil, and water.

Afterward, Ramírez-Andreotta said she was pleased that residents continued to garden but changed some of their practices. For instance, incidental soil ingestion and drinking water were estimated to pose a greater risk of arsenic exposure than eating the vegetables themselves, and a survey of the residents after the data-sharing event indicated they planned to take precautions to avoid these exposures. These included avoiding gardening on windy days, washing their hands after gardening, and storing gardening tools outdoors.24 Some also recalculated their risk exposure because they knew there were certain vegetables they would not eat as often.

The project had also uncovered an unforeseen problem. A number of residents had water samples whose arsenic concentrations tested above the U.S. EPA’s MCL of 10 μg/L (or ppb). Among these were residents on town water, whose exceedances triggered a state notice of violation and a fine to the utility. Ramírez-Andreotta reported these findings to the U.S. EPA and the Arizona Department of Environmental Quality. However, the fact that the arsenic occurred naturally from the region’s geology made the contamination a state drinking water matter, not a Superfund matter. The U.S. EPA therefore had to delegate the issue to the Arizona department.

Ramírez-Andreotta says the U.S. EPA’s inability to get involved was frustrating to residents, who did not want to be exposed to arsenic through their drinking water whether it was there naturally or not. Still, she believes that as a result of Gardenroots the residents had increased their capacity for advocating for their best interests. Having their residential site-specific data was critical in addressing the MCL exceedance, and it provided the evidence needed to move the water company into compliance.

Citizen Concerns Change over Time

Ramírez-Andreotta and colleagues documented a shift in the concerns that Dewey-Humboldt residents expressed through the initial 5 years (2008–2013) of being listed as a Superfund site. Using U.S. EPA public meeting records, town council minutes, Gardenroots meeting accounts, newspaper stories, and U.S. EPA factsheets mailed to residents, she and her coauthors observed that community members moved from a passive position of absorbing information to an action-oriented position of applying scientific knowledge to protect themselves.25 The shift in residents’ outlooks from passive to active, Ramírez-Andreotta believes, occurred in part as a result of the community’s engagement in research with the University of Arizona Superfund Research Program, which provided a platform for free-choice learning.

The U.S. EPA spokesperson says the agency believes it is critical to involve communities in the entire Superfund process. “Communities play a key role in informing EPA of how they want to be engaged in the process,” says the spokesperson. “In turn, EPA tailors its outreach to meet community needs, offering a wide range of opportunities for communities to learn about the science and health issues related to cleaning up Superfund sites.”

Ramírez-Andreotta recommends that the U.S. EPA work more to involve residents in the Superfund process beyond an initial typically one-time survey. She says residents can be involved during the remedial investigation that the agency conducts when a site is placed on the National Priorities List for cleanup. “Involving the affected communities via community-engaged research and participation in environmental projects during the USEPA’s Superfund management is critical,” Ramírez-Andreotta wrote in a 2016 article.25 “It can lead to improvements in one’s knowledge and awareness, sense of control and ability to make informed decisions and take measures to mitigate exposures.”


Nancy Averett writes about science and the environment from Cincinnati, Ohio. Her work has been published in Pacific Standard, Audubon, Discover, E/The Environmental Magazine, and a variety of other publications.

References

1. U.S. EPA (U.S. Environmental Protection Agency). 2017. Iron King Mine and Humboldt Smelter Description and History [website]. Washington, DC:U.S. Environmental Protection Agency. https://yosemite.epa.gov/r9/sfund/r9sfdocw.nsf/ViewByEPAID/az0000309013 [accessed 5 September 2017].

2. Tone S. 2008. Two chances to hear EPA’s update on Iron King. Prescott Valley Tribune, online edition, News section, 14 August 2008. http://www.pvtrib.com/news/2008/aug/14/two-chances-to-hear-epas-update-on-iron-king-site/ [accessed 5 September 2017].

3. Diethorn K. 2017. Peale’s Philadelphia Museum [website]. http://philadelphiaencyclopedia.org/archive/peales-philadelphia-museum/ [accessed 5 September 2017].

4. Yerman MG. 2014. Lois Gibbs: ‘The Government Wouldn’t Help Me, So I Decided to Do It Myself.’ Huffington Post, The Blog section, 22 April 2014. http://www.huffingtonpost.com/marcia-g-yerman/the-government-wouldnt-he_b_5188348.html [accessed 5 September 2017].

5. Irwin A. 1995. Citizen Science: A Study of People, Expertise and Sustainable Development. New York, NY:Psychology Press.

6. Riesch H, Potter C. 2013. Citizen science as seen by scientists: methodological, epistemological and ethical dimensions. Public Underst Sci 23(1):107–120, PMID: 23982281,10.1177/0963662513497324.

7. Nature. 2015. Rise of the citizen scientist [editorial]. Nature 524(7565):265, PMID: 26289171, 10.1038/524265a

8. Soleri D, Long JW, Ramirez-Andreotta MD, Eitemiller R, Pandya R. 2016. Finding pathways to more equitable and meaningful public-scientist partnerships. Citiz Sci Theory Pract 1(1):9, 10.5334/cstp.46.

9. Ramirez-Andreotta MD, Brusseau ML, Artiola JF, Maier RM, Gandolfi AJ. 2014. Environmental Research Translation: enhancing interactions with communities at contaminated sites. Sci Total Environ 497-498:651–664, PMID: 25173762, 10.1016/j.scitotenv.2014.08.021.

10. American Civil Liberties Union of Michigan. 2016. Here’s to Flint . Detroit, MI:American Civil Liberties Union of Michigan. http://www.aclumich.org/herestoflint [accessed 5 September 2017].

11. U.S. EPA. 2017. Drinking Water Requirements for States and Public Water Systems. Lead and Copper Rule [website]. https://www.epa.gov/dwreginfo/lead-and-copper-rule [accessed 5 September 2017].

12. Ramirez-Andreotta MD, Brusseau ML, Artiola JF, Maier RM. 2013. A greenhouse and field-based study to determine the accumulation of arsenic in common homegrown vegetables grown in mining-affected soils. Sci Total Environ 443:299–306, PMID: 23201696, 10.1016/j.scitotenv.2012.10.095.

13. Ramírez-Andreotta M. 2016. Cultivating Citizen Science to Reduce Environmental Risks [presentation, 31 October 2016. Presented at ScienceWriters 2016, 28 October–1 November 2016, San Antonio, TX:National Association of Science Writers.

14. Ottinger G. 2010. Buckets of resistance: standards and the effectiveness of citizen science. Sci Technol Human Values 35(2):244–270, 10.1177/0162243909337121.

15. Louisiana Bucket Brigade. 2015. Louisiana Bucket Brigade [website]. http://www.labucketbrigade.org/content/bucket [accessed 5 September 2017].

16. U.S. EPA. 1999. Air Method, Toxic Organics-15 (TO-15): Compendium of Methods for the Determination of Toxic Organic Compounds in Ambient Air, Second Edition: Determination of Volatile Organic Compounds (VOCs) in Air Collected in Specially-Prepared Canisters and Analyzed by Gas Chromatography/Mass Spectrometry (GC/MS). EPA 625/R-96/010b. Washington, DC:U.S. Environmental Protection Agency. https://www.epa.gov/homeland-security-research/epa-air-method-toxic-organics-15-15-determination-volatile-organic [accessed 5 September 2017].

17. Frickel S, Gibbon S, Howard J, Kempner J, Ottinger G, Hess DJ. 2010. Undone science: charting social movement and civil society challenges to research agenda setting. Sci Technol Human Values 35(4):444–473, 10.1177/0162243909345836.

18. U.S. EPA. 2016. Process of Reviewing the National Ambient Air Quality Standards [website]. https://www.epa.gov/criteria-air-pollutants/process-reviewing-national-ambient-air-quality-standards [accessed 5 September 2017].

19. U.S. EPA. 2017. Air Sensor Toolbox for Citizen Scientists, Researchers and Developers [website]. https://www.epa.gov/air-sensor-toolbox [accessed 5 September 2017].

20. National Advisory Council for Environmental Policy and Technology. 2016. Environmental Protection Belongs to the Public: A Vision for Citizen Science at EPA. EPA 219-R-16-001. https://www.epa.gov/faca/nacept-2016-report-environmental-protection-belongs-public-vision-citizen-science-epa [accessed 5 September 2017].

21. Tone S. 2015. Dewey-Humboldt’s Superfund site still poses risks and triggers frustration. Prescott Valley Tribune, online edition, 29 April 2015. https://www.pvtrib.com/news/2015/apr/29/dewey-humboldts-superfund-site-still-poses-risks-/ [accessed 5 September 2017].

22. Shirk JL, Ballard HL, Wilderman CC, Phillips T, Wiggins A, Jordan R, et al. 2012. Public participation in scientific research: a framework for deliberate design. Ecol Soc 17(2):29, 10.5751/ES-04705-170229.

23. U.S. EPA. 1992. Understanding Superfund Risk Assessment. 9285.7-06FS. Washington, DC:U.S. Environmental Protection Agency.

24. University of Arizona. 2017. Simple Steps to Start Collecting: Tips [website]. https://gardenroots.arizona.edu/get-started#tips [accessed 5 September 2017].

25. Ramirez-Andreotta M, Lothrop N, Wilkinson ST, Root RA, Artiola JF, Klimecki W, et al. 2016. Analyzing patterns of community interest at a legacy mining waste site to assess and inform environmental health literacy efforts. J Environ Stud Sci 6(3):543–555, PMID: 27595054, 10.1007/s13412-015-0297-x.

Erratum: “Urinary Concentrations of Organophosphate Flame Retardant Metabolites and Pregnancy Outcomes among Women Undergoing in Vitro Fertilization”

PDF icon PDF Version (233 KB)

  • Received: 1 September 2017
    Accepted: 9 September 2017
    Published: 1 November 2017

    Address correspondence to C. Carignan, Department of Environmental Health, Building 1, 14th Floor, 665 Huntington Avenue, Boston, MA 02115 USA. Phone: (617) 432-4572. Email: carignan@hsph.harvard.edu

    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.



Carignan et al. have noted an error in Figure 2 of their paper, “Urinary Concentrations of Organophosphate Flame Retardant Metabolites and Pregnancy Outcomes among Women Undergoing in Vitro Fertilization.” The correct figure is included in this erratum and was part of the original submission, peer review, and accepted article. The error occurred when the authors responded to a proof request for a formatting edit, and a mistake was made in recreating the bottom row of the figure. The authors regret the error.

First set of four plots showing changes in the mean proportion of cycles resulting in implantation (y-axis) across gamma PFR, BDCIPP, DPHP, and ip-PPP in Q1, Q2, Q3, and Q4 (x-axis; respective p trend values are 0.02, 0.06, 0.02, and 0.05). Second set of four plots showing changes in mean proportion of cycles resulting in clinical pregnancy (y-axis) across the above-mentioned organophosphate flame retardant metabolite concentrations in the four quartiles (x-axis; respective p trend values are 0.004, 0.07, 0.01, and 0.19). Third set of four plots showing changes in mean proportion of cycles resulting in live birth (y-axis) across the above-mentioned organophosphate flame retardant metabolite concentrations in the four quartiles (x-axis; respective p trend values are 0.05, 0.16, 0.08, and 0.04).

Figure 2. Adjusted mean [95% confidence interval (CI)] proportion of cycles resulting in implantation, live birth, and clinical pregnancy by quartile of urinary organophosphate flame retardant (PFR) metabolite concentrations among 211 women undergoing 297 in vitro fertilization (IVF) cycles. Gray shading indicates change in means from the first and fourth quartile. Adjusted models control for maternal age (continuous), body mass index (BMI) (continuous), race/ethnicity (black/Asian/other, white/Caucasian), year of IVF treatment cycle (continuous), and primary Society for Assisted Reproductive Technology (SART) infertility diagnosis at study entry (female, male, unknown), with continuous variables at their mean level and categorical variables weighted by their frequency in the study population. *Significantly different from the lowest quartile (Q1) at the α = 0.05 level.

Unwell: The Public Health Implications of Unregulated Drinking Water

PDF icon PDF Version (1.5 MB)

  • Published: 1 November 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

Strategies to Improve Private-Well Water Quality: A North Carolina Perspective

Jacqueline MacDonald Gibson, and Kelsey J. Pieper

Roughly one in seven U.S. residents relies on a private well for drinking water.1 Unlike the rest of the population served by the nation’s many public water systems,2 these 44.5 million Americans are not protected by the federal Safe Drinking Water Act,3 which regulates 87 biological and chemical contaminants.4 This has significant implications for public health, according to the authors of a new review in Environmental Health Perspectives,5 and although solutions exist for ensuring that well water is safe to drink, it is unclear how and whether they can be implemented.

At best, private wells receive minimal oversight from local and state authorities, such as limited testing upon installation and, in some states, when properties change hands, write authors Jacqueline MacDonald Gibson of the University of North Carolina at Chapel Hill and Kelsey Pieper of Virginia Polytechnic Institute and State University. Many wells escape regulation altogether, leaving the onus entirely on individuals to screen for pollutants and to mitigate them when they are discovered.6

Although precise statistics are not available, most well owners do not perform or pay for recommended tests at regular intervals, MacDonald Gibson says. This neglect is partly due to a simple lack of knowledge: “A lot of people assume that if their water looks and tastes okay, that the water is safe,” she says. But common contaminants such as bacteria, arsenic, and nitrate are colorless, odorless, and tasteless, and they can easily go undetected.7,8,9

Cost can be another barrier to keeping well water safe. Testing for volatile organic compounds, pesticides, metals, nitrate, bacteria, and radioactive contaminants could exceed $500, says Lynda Knobeloch, a former toxicologist with the Wisconsin Department of Health Services who was not involved with the paper. Effective in-home filters can cost hundreds to thousands of dollars, and well replacement can cost $10,000 or more, Knobeloch says. Furthermore, in some states, including Wisconsin, well contamination must be disclosed upon the sale of the home, providing an additional disincentive for some homeowners to ever test at all.

Photograph of a rural well
This poorly constructed and largely unprotected well tested high for fecal coliforms. But even a state-of-the-art well can produce unsafe water without routine maintenance, testing, and treatment when necessary. Having a septic system nearby only increases the risk. Image: © Bryan Swistock/Pennsylvania State University.

Many people who use private wells are left potentially exposed to harmful pollutants. In a 2013 study of nearly 4,000 private wells in rural Wisconsin, for example, Knobeloch found that 47% exceeded at least one health-based water quality standard.10 And a recent study in North Carolina led by MacDonald Gibson found that between 2007 and 2013, 99% of emergency department visits for acute gastrointestinal illness caused by microbial contamination of drinking water were associated with private wells.11

The new review focused on North Carolina, where 35% of the population relies on private wells (the third most of any state),3 and 3.2 million people live in rural areas (the second most of any state).12 Furthermore, a history of racial segregation and what is known as “municipal underbounding”—in which expanding cities engulf poor and minority communities without affording them city services—has left many black residents of North Carolina’s periurban areas without access to public water systems, even if adjacent newer neighborhoods are fully connected.13

Because underbounded communities often also lack sewer service, the authors suggest that residents run the risk of contaminating their own water with fecal organisms, particularly in higher-density areas. “There are little pockets of people with old and failing septic systems in their backyard and a well nearby, and there may be a water line across the street,” says MacDonald Gibson.

To help address these challenges in North Carolina and beyond, the Research Triangle Environmental Health Collaborative, an alliance of government, academic, industry, and public-interest groups, held a two-day summit in October 2015. The new review outlines findings and recommendations from this summit, which include developing a state database of private well locations, funding studies to identify areas underserved by municipal water and sewer systems, and providing additional support and resources to individuals for monitoring and maintenance.

Knobeloch says the recommendations are sensible, if not necessarily innovative. A number of these recommendations have already been embraced by states including New York and Minnesota, whereas others, such as fully annexing underserved communities to extend city services, are unlikely to work for political and economic reasons.

One additional improvement she believes should have been included is discouraging the development of new communities without central sewer and water systems. “You can prevent more of the problems you already have,” she says.

The review highlights important and in some cases universal challenges with private wells, says Bryan Swistock of Pennsylvania State University, who has studied water resources and well water quality extensively in his home state and was not involved in the review. “It agrees so much with what we’ve seen here over thirty years,” he says. “There needs to be a better recognition that private water wells are a really critical part of the nation’s infrastructure. But they are mostly ignored by the government because they are viewed as a private property issue.”

References

1. U.S. Geological Survey. 2017. Domestic Water Use. https://water.usgs.gov/edu/wudo.html [accessed 29 September 2017].

2. U.S. Environmental Protection Agency (EPA). 2015. “Providing Safe Drinking Water in America: 2013 National Public Water Systems Compliance Report.” EPA Document 305R15001. Washington, DC: U.S. Environmental Protection Agency. https://www.epa.gov/sites/production/files/2015-06/documents/sdwacom2013.pdf [accessed 29 September 2017].

3. Maupin M, Kenny J, Hutson S, Lovelace J, Barber N, Linsey K. 2014. “Estimated Use of Water in the United States in 2010.” U.S. Geological Survey Circular 1405. Washington, DC: U.S. Department of the Interior; U.S. Geological Survey. https://pubs.usgs.gov/circ/1405/pdf/circ1405.pdf [accessed 29 September 2017].

4. U.S. EPA. 2017. Ground Water and Drinking Water: National Primary Drinking Water Regulations. https://www.epa.gov/ground-water-and-drinking-water/table-regulated-drinking-water-contaminants [accessed 29 September 2017].

5. MacDonald Gibson J, Pieper KJ. 2017. Strategies to improve private-well water quality: a North Carolina perspective. Environ Health Perspect 125(7):076001, PMID: 28728142, 10.1289/EHP890.

6. U.S. EPA. 2017. Private Drinking Water Wells. https://www.epa.gov/privatewells [accessed 29 September 2017].

7. PennState Extension. 2017. Lead in Drinking Water. http://extension.psu.edu/natural-resources/water/drinking-water/water-testing/pollutants/lead-in-drinking-water [accessed 29 September 2017].

8. U.S. EPA. 2016. Drinking Water Requirements for States and Public Water Systems. Chemical contaminant rules. https://www.epa.gov/dwreginfo/chemical-contaminant-rules [accessed 29 September 2017].

9. Minnesota Department of Health. 2017. Nitrate in Well Water: Well Management Program. http://www.health.state.mn.us/divs/eh/wells/waterquality/nitrate.html [accessed 29 September 2017].

10. Knobeloch L, Gorski P, Christenson M, Anderson H. 2013. Private drinking water quality in rural Wisconsin. J Environ Health 75(7):16–20, PMID: 23505770.

11. 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 124(10):1583–1591. (2016), PMID: 27203131, 10.1289/EHP160.

12. U.S. Census Bureau. 2012. 2010 Census Urban and Rural Classification and Urban Area Criteria. https://www.census.gov/geo/reference/ua/urban-rural-2010.html [accessed 29 September 2017].

13. 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.

An Investigation of the Single and Combined Phthalate Metabolite Effects on Human Chorionic Gonadotropin Expression in Placental Cells

Author Affiliations open

1Department of Epidemiology, University of Pittsburgh Graduate School of Public Health, Pittsburgh, Pennsylvania, USA

2Department of Obstetrics, Gynecology and Reproductive Sciences, University of Pittsburgh, Pittsburgh, Pennsylvania, USA

3Department of Obstetrics, Gynecology and Reproductive Sciences, University of California, San Francisco, San Francisco, California, USA

4Department of Clinical Chemistry, Biomedicum, Helsinki University and Helsinki University Central Hospital, Helsinki, Finland

5Institute of Reproductive and Developmental Biology, Imperial College of London, London, United Kingdom

PDF icon PDF Version (1.6 MB)

  • Background:
    Observational studies have reported associations between maternal phthalate levels and adverse outcomes at birth and in the health of the child. Effects on placental function have been suggested as a biologic basis for these findings.
    Objective:
    We evaluated the effects of phthalates on placental function in vitro by measuring relevant candidate genes and proteins.
    Materials and Methods:
    Human trophoblast progenitor cells were isolated at 7–14 wk of pregnancy (two female and three male concepti), and villous cytotrophoblast cells (vCTBs) were isolated at 15–20 wk (three female and four male concepti). Cells were cultured in vitro with four phthalate metabolites and their combination at concentrations based on levels found previously in the urine of pregnant women: mono-n-butyl (MnBP, 200 nM), monobenzyl (MBzP, 3 μM), mono-2-ethylhexyl (MEHP, 700 nM), and monoethyl (MEP, 1.5 μM) phthalates. mRNA levels of CGA, CGB, PPARG, CYP19A1, CYP11A1, PTGS2, EREG, and the intracellular β subunit of human chorionic gonadotropin (hCGβ) and peroxisome proliferator activated receptor γ (PPARγ) were measured in the cellular extracts, and protein levels for four forms of secreted hCG were measured in the conditioned media.
    Results:
    Previously reported associations between maternal phthalates and placental gene expression were reproduced experimentally: MnBP with CGA, MBzP with CYP11A1, and MEHP with PTGS2. CGB and hCGβ were up-regulated by MBzP. In some cases, there were marked, even opposite, differences in response by sex of the cells. There was evidence of agonism in female cells and antagonism in male cells of PPARγ by simultaneous exposure to multiple phthalates.
    Conclusions:
    Concentrations of MnBP, MBzP and MEHP similar to those found in the urine of pregnant women consistently altered hCG and PPARγ expression in primary placental cells. These findings provide evidence for the molecular basis by which phthalates may alter placental function, and they provide a preliminary mechanistic hypothesis for opposite responses by sex. https://doi.org/10.1289/EHP1539
  • Received: 23 December 2016
    Revised: 6 September 2017
    Accepted: 18 September 2017
    Published: 31 October 2017

    Address correspondence to J.J. Adibi, 130 Desoto Street, Parran Hall 5132, Pittsburgh, PA 15261 USA. Telephone: 412-624-1913. Email: adibij@pitt.edu

    Supplemental Material is available online (https://doi.org/10.1289/EHP1539).

    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 (123 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

Phthalates are a class of synthetic, endocrine-disrupting chemicals that are detected in all pregnant women in the United States (Woodruff et al. 2011) Prenatal maternal exposure to phthalates has been associated with short-term outcomes in pregnancy, such as the duration of labor (Adibi et al. 2009; Ferguson et al. 2014; Latini et al. 2003), the risk of preeclampsia (Cantonwine et al. 2016), and fetal sex differentiation as determined at birth (Swan et al. 2015). Long-term outcomes in the children may include negative trends in behavior, IQ, attention, and social communication [11 studies of children 0 to 12 y old reviewed by Ejaredar et al. (2015)].

The role of the placenta in these associations has been partially identified. Placental prostaglandin production, which might be a target of phthalates, is highest at the end of pregnancy and is required for the induction of labor (Tetz et al. 2015). Phthalates may alter circulating levels of antiangiogenic molecules that are produced by the placenta and are indicators of preeclampsia risk (Ferguson et al. 2015). We have identified a potential role of sex-specific placental hormone production in contributing to the effect of phthalate exposures on a neonatal marker predicting the future reproductive health of the child (Adibi et al. 2015a). Clarifying the role of the placenta in these associations will improve temporal and spatial precision in estimating the short- and long-term health consequences of prenatal exposure.

It is not possible to directly observe the critical time points in early pregnancy when mRNA and protein expression in the placenta and in the embryo/fetus are vulnerable to perturbation by tissue and circulating levels of maternal phthalates. Owing to genetic, anatomic, and physiologic differences in placental–fetal biology between species (Maltepe et al. 2010; Rawn and Cross 2008), animal models alone are insufficient to identify causal relationships in human pregnancy. As such, we believe a combination of experimental and observational models that are human-specific and informative of early pregnancy relationships is best suited to deliver these necessary insights.

Another motivation to work simultaneously in experimental and observational systems is the need to overcome a source of intractable confounding: between-person differences in placental metabolism. This is a special problem in phthalate epidemiology because the urinary biomarkers used to assess exposure are also partially products of placental metabolism (Hakkola et al. 1998). At the present time, we have no methods to assess or control for this type of bias. Experimental methods (and statistical methods applied to observational data) may drastically reduce this type of bias and give greater confidence and reproducibility regarding the unconfounded effects of the chemicals on gene and protein expression.

We selected a list of candidate genes to determine if the types of phthalate effects on gene expression that have been identified in other species and in other tissue types were also targets in the human placenta (Adibi et al. 2010). In a companion to this manuscript, we have reported sex-specific associations of eight of these candidate mRNAs measured in term placental tissue biopsies with concentrations of six maternal urinary phthalates (Adibi et al. 2017). Here, we report the experimental follow-up study to those associations using models of undifferentiated and differentiated trophoblasts (Tbs) isolated from first- and second-trimester placentas. In the placental tissue, chorionic gonadotropin α (CGA) was the most strongly associated with the highest number of phthalate metabolites (Adibi et al. 2017). CGA encodes an α subunit of glycoprotein hormones including placenta-specific gonadotropin, called human chorionic gonadotropin (hCG), which is a primary end point in the present study. Here, we follow up on this finding by reporting phthalate effects on mRNAs of both α- and β-subunits of hCG (encoded by CGA and by CGB and its homologues, respectively), on protein levels of intracellular and secreted hCG (dimer of α- and β-subunits), and on the mRNA and intracellular levels of a regulatory factor of hCG, peroxisome proliferation activated receptor γ (PPARG/PPARγ). In this analysis, we compared the molecular response of human Tbs to phthalate exposure according to fetal sex differences, application of single phthalate metabolites versus mixtures, cell type, and consistency between mRNA and protein effects.

Materials and Methods

Study Subjects

Women undergoing elective pregnancy terminations between 7 and 20 wk gestation at the Women’s Options Center at the University of California, San Francisco between 2012 and 2013 donated tissue anonymously. Donation was restricted to women without fetal anomalies. Written informed consents were obtained from all subjects. The University of California, San Francisco Committee on Human Research approved the tissue collection.

Cell Culture

The trophoblast progenitor cell (TBPC) represents an undifferentiated, multipotent population derived from the mesoderm of the chorion at gestational weeks 7.0–14.9. Derivation and maintenance conditions are described elsewhere (Genbacev et al. 2011; Genbacev et al. 2016). TBPCs were cultured in six-well plates that had been incubated with 0.5% gelatin for 30 min. They were plated at 250,000 cells/well. Details on the cell-culture medium are described elsewhere (Genbacev et al. 2016). The villous cytotrophoblasts (vCTBs) were isolated and purified from placentas obtained at 15 wk 4 d to 20 wk 1 d of gestation using microdissection, enzymatic digestion, and cell culture techniques described elsewhere (Hunkapiller and Fisher 2008). Cells were plated on Matrigel® in 24-well plates at 500,000 cells/well, dosed in duplicate, and cultured in serum-free medium (Hunkapiller and Fisher 2008). The vCTBs were sampled from placentas obtained from three female and four male placentas, and the TBPCs were sampled from two female and three male placentas (these cells are referred to hereafter as female and male cells, respectively). Details on experiments and replicates are in the Supplemental Material (see Tables S5 and S6). Single phthalate metabolite doses were designed to mimic maternal urinary concentrations measured in a birth cohort study (Adibi et al. 2017): 200 nM mono-n-butyl phthalate (MnBP; TCI America), 3 μM monobenzyl phthalate (MBzP; TCI), 700 nM mono-2-ethylhexyl phthalate (MEHP; Wako), and 1.5 μM monoethyl phthalate (MEP; Sigma-Aldrich). Specifically, we chose urinary concentrations that in our previous study were associated with CGA mRNA levels and where the association differed in male versus female placentas [Figures 1A, B, D in Adibi et al. (2017)]. In addition, we combined the four abovementioned concentrations of single phthalates into a mixture dose. Vehicle control cells for the single metabolite doses received 0.1% dimethyl sulfoxide (DMSO), and controls for the phthalate mixture received 0.4% DMSO; both of these doses are below the DMSO concentration that affects Tb differentiation (Thirkill and Douglas 1997). The self-renewing TBPC cultures were grown for 72 h, which is the point at which they reach 90% confluency. The vCTB cultures were maintained for 40 h from the time of dosing. vCTBs are nonrenewing; therefore, 40 h is optimal to evaluate functional changes in hCG secretion while cells are metabolically active. In all experiments, cell morphology and cell density were assessed and documented by phase contrast microscopy.

Gene Expression in Cultured Cells

The genes selected for this study were measured previously in placental tissue biopsies and were found to be correlated with maternal urinary phthalates (Adibi et al. 2010, 2017), with the exception of CGB and EREG. CGB was selected because it encodes a β subunit of human chorionic gonadotropin (hCG), a placental hormone that we hypothesize to be a target of phthalate exposure. The CGB primer set amplifies CGB, CGB3, CGB5, CGB7, and CGB8 (TaqMan™, ThermoFisher Scientific). EREG is an ovarian target of luteinizing hormone (LH), a gonadotropin that binds to the same receptor as hCG (Park et al. 2004). In vivo, hCG has been found to stimulate EREG, which is why we also selected EREG for this study (Huber et al. 2007). RNA was isolated using an RNeasy Plus Mini Kit (Qiagen) and measured on a NanoDrop™ spectrophotometer (ThermoFisher Scientific). Reverse transcription was carried out using an iScript™ cDNA Synthesis Kit (BioRad). A TaqMan™ assay for RPS4Y1 was used to assign sex to all control cells. The Ct (amplification cycle at which the mRNA concentration was detectable) for female cells was either >35 cycles, or amplification was nondetectable. For males, amplification of RPS4Y1 was detected at ≤25 cycles. RPS4Y1, a Y-linked gene, was selected for this purpose after reviewing three independently generated transcriptome data sets of placental cells in culture and placental biopsies to determine which of the Y-linked genes showed the greatest discrimination in males and females (J. Adibi, unpublished data, 2013). TaqMan™ gene expression assays (see Table S4) were used for mRNA quantitation, and 10 ng of cDNA was loaded in a 10-μL reaction volume and run in triplicate on a 7900HT quantitative polymerase chain reaction (PCR) instrument (Applied Biosystems). RN18S1 RNA was used as an internal control based on the validated convention of the laboratory where the experiments were designed and conducted (Winn et al. 2007).

hCG in Conditioned Media

Conditioned media from each well from both cell types were aliquoted and stored at −20°C. Aliquots were shipped to the University of Helsinki (Finland), where they were analyzed for a panel of hCG subunits and variants including intact hCG, hCGα, hCGβ, and the hyperglycosylated form of hCG (hCG-h). Intact hCG was measured using time-resolved immunofluorometric assays (IFMAs) (DELFIA®, Perkin-Elmer Wallace), and the other forms were measured using in-house assays with monoclonal antibodies (Alfthan et al. 1992; Lee et al. 2013). The lower limit of quantitation was 0.5 pmol for intact hCG, 5 pmol for hCGα, 1 pmol for hCGβ, and 10.0 pmol for hCG-h. The intact hCG assay measures only the hCG αβ dimer, irrespective of whether it is hyperglycosylated or not. The hCGβ assay measures only free hCGβ subunit, including hyperglycosylated hCGβ. The hCGα assay measures the free α subunit. The hyperglycosylated hCG assay (hCG-h) measures hyperglycosylated intact hCG and hyperglycosylated hCGβ. The hCGα subunit is a gonadotropin subunit that is shared by all glycoprotein hormones, that is to say, hCG, TSH, LH, and FSH. hCGβ lacks hCG activity but may play a role in Tb invasion (Lee et al. 2013).

Quantitative Western Blots

Protein was purified from the same vCTB lysate used for RNA isolation according to the manufacturer’s instructions (AllPrep® RNA/Protein Kit and RNeasy Plus, Qiagen) and was additionally cleaned up by acetone precipitation. Protein lysate was mixed with four volumes of ice-cold acetone and incubated at −20°C for 30 min. After centrifuging at 12,000×g for 10 min, the supernatant was discarded, and the pellet was air-dried. The pellet was then resuspended in 5% sodium dodecyl sulfate (SDS). The protein concentration was measured using a Direct Detect® Spectrometer (EMD Millipore). The protein was denatured at 95°C for 5 min in 4×protein loading buffer (Li-Cor) supplemented with 5% β-mercaptoethanol (Bio-Rad) and separated using Any kD™ Precast Protein Gels (Bio-Rad). The proteins were transferred to nitrocellulose membrane using a Trans-Blot® Turbo™ Transfer System (Bio-Rad). After blocking in phosphate-buffered saline (PBS) blocking buffer (Li-Cor) at room temperature (20–22°C) for 1 h, the membrane was incubated with primary antibodies in blocking buffer at 4°C overnight. The membrane was then washed and incubated with secondary antibodies at room temperature (20–22°C) for 50 min. The reactive proteins were detected using an Odyssey CLx (Li-Cor) imaging system, a method that has been optimized and validated for quantitation (Wang et al. 2007). The Western blot bands were quantified using Image Studio (version 5.0; LI-COR). The primary antibodies were anti-hCG (1:2,000, Dako, A0231), anti-PPAR-γ (1:1,000, Cell Signaling, 24,355), and anti-β-Actin (1:5,000, Santa Cruz Biotechnology, sc4778). The secondary antibodies were IRDye 800CW donkey anti-rabbit IgG (H+L) (Li-Cor) and IRDye 680LT donkey anti-mouse IgG (H+L) (Li-Cor). The hCGβ antibody yielded four bands in the Western blot, of which we quantified three: 42 kDa, 23 kDa, and 21 kDa. The 42-kDa band gave the strongest signal, and the 23- and 21-kDa bands were closest in size to the predicted band. The goal was to determine which band had the highest correlation with CGB mRNA and secreted hCG protein as validation of the Dako antibody and as further insight into possible sex differences in hCG synthetic machinery. The 42-kDa band was 4 log units higher on average than the 23-kDa band and strongly correlated with β-actin in the same sample (r=0.61, p<0.0001). The 21-kDa band was only detected in 57% of female cells and 43% of male cells, and it was not analyzed further (χ2 p=0.02). Results for the 23-kDa band are presented (Figure 3F). The 23-kDa band has been used for hCG quantitation in previous studies (Racca et al. 2011). The PPARγ antibody gave a single band at 51 kDa, consistent with previous studies (Senol-Cosar et al. 2016) (Figure 3G).

Statistical Analysis

For the RNA analysis, mean Ct log2 values (amplification cycle at which the mRNA in the sample was detected; i.e., higher Ct values indicate lower mRNA concentration in the sample) of the technical replicates (n=2) were calculated for each sample. Because of the significant variability in baseline gene and protein expression between biologic replicates and because it is established that hCG also differs significantly by day of gestation (Wald et al. 2003), we chose to analyze the data using multivariate mixed effects models. This method allows us to quantitate differences in gene and protein expression between the treated and control cells after adjustment for the clustering of values within each biologic replicate (random effect) and adjustment for fixed effects (sex, gestational age, etc.). In this case, the random effect or experiment number is a proxy for unmeasured biologic variability at baseline (i.e., genetic and epigenetic variability, maternal health, preprocedure exposures). We included as model covariates gestational age at the time of pregnancy termination, RN18S mRNA levels in the sample, sex of the cells, phthalate dose, and institution at which the experiment was conducted. We report the population marginal means and their 95% confidence intervals (CIs) for all treatment groups by sex. For plotting the changes, we used the β coefficient (equivalent to the ΔΔCt) and its 95% CIs after transformation to the linear scale (2ΔΔct). Sex-specific parameters were calculated, using the Estimate statement in SAS (SAS Institute Inc.), from a model that included a term for phthalate dose by sex. We used two models to estimate all dose effects: the first included the 0.1% DMSO vehicle control samples and the single-metabolite dose groups (also with 0.1% DMSO), and the second included 0.4% DMSO and the phthalate-mixture dose group (also with 0.4% DMSO). Dose group was treated as a categorical variable and interacted on sex.

We applied the same strategy in the analysis of intracellular protein and secreted hCG levels. We report all results for protein on a log scale; a one loge unit difference is equivalent to a 2.7-fold change on a linear scale. For ease of comparison and interpretation, we report log unit differences as fold changes (the difference between the 2 log values transformed to the linear scale). In the analysis of the Western blot data, β-actin was a covariate to control for total protein in the sample. Intracellular hCGβ and PPARγ were modeled as loge-transformed intensity values, and secreted hCG variants were modeled as log-transformed concentrations. To estimate subunit-specific effects, we calculated ratios of the α, β, and hyperglycosylated hCG to intact hCG. This calculation was performed to normalize for overall hCG production, which serves as an indicator of quantity and viability of the cells. We used Spearman rank correlations to quantitate the relationship between mRNAs and intracellular and secreted proteins that were normalized for internal controls. In all mixed effects models, we estimated empirical standard errors that are more robust when the assumption of equal covariance across experiments cannot be confirmed (Verbeke and Molenberghs 2000). Analyses were performed using SAS (version 9.3; SAS Institute Inc.).

Results

The intracellular and secreted hCG protein levels from the TBPCs were primarily below the level of detection and were not analyzed (data not shown). All mRNA and protein end points were well above the detection limit in the vCTBs. Differences in cellular morphology and cell density were not observed in treated versus control cells. We present the sex-specific means (Ct values, protein intensities, and secreted protein concentrations) of cells treated with single metabolites and with the combined metabolites.

mRNA

MnBP stimulated the CGA mRNA expression in female TBPCs (2.1-fold; 95% CI: 1.4, 3.1) and vCTBs (1.3-fold; 95% CI: −0.9, 1.7) (Figures 1A, 2A; see also Tables S1 and S2), whereas an opposite and weaker effect was observed in male cells (TBPCs −0.9-fold; 95% CI: −0.6, 1.5; vCTBs −0.9-fold; 95% CI: −0.7, 1.1). This was the only sex-specific effect on mRNA that was common to both cell types (Figures 1, 2). In the TBPCs, the phthalate mixture increased the CGB levels in female cells 1.4-fold (95% CI: 1.0, 1.9) and decreased the levels in male cells 0.7-fold (95% CI: −0.5, −0.9). In the female cells, the mixture’s effect on CGB resembled the average effect of the single metabolites (1.4-fold vs. 1.5-fold in the mixture-dosed cells). In the male cells, the effect of the mixture was similar to that of MEHP, which significantly reduced CGB (−0.8-fold by MEHP alone vs. −0.7-fold in the mixture-dosed cells). Conversely, in the female vCTBs, the mixture lowered CGB mRNA −0.9-fold (95% CI: −0.8, −1.0) and increased CGB in the male vCTBs 1.7-fold (95% CI: 0.9, 3.2) relative to the controls. There was inadequate biologic replication for the MEP-dosed TBPCs (see Table S5, results not shown).

Figures 1A to 1D are four whisker plots with confidence intervals plotting fold change in TBPC mRNA relative to control cells (mean, 95 percent CI) (y-axis) across phthalates CGA, CGB, and PPARG (x-axis) for metabolites MnBP, MBzP, MEHP, and their mixture.

Figure 1. Phthalate effects on mRNAs. The effects are expressed as mean relative fold change (2ΔΔct) and 95% confidence intervals (CIs) compared with DMSO-treated control cells in undifferentiated trophoblasts (TBPCs). (A) 200 nM MnBP; (B) 3 μM MBzP; (C) 700 nM MEHP; (D) mixture of all four metabolites (MnBP, MBzP, MEHP, MEP). The black lines indicate female-specific effects, and the gray lines indicate male-specific effects. Overall effects that were significant (p≤0.05) are indicated by a line and marked “sex-specific” if the phthalate effect differed in male and female cells. CG, chorionic gonadotropin; DMSO, dimethyl sulfoxide; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate; PPARG, peroxisome proliferator activated receptor gamma.

Figures 2A to 2E are five whisker plots with confidence intervals plotting fold change in vCTB mRNA relative to control cells (mean, 95 percent CI) (y-axis) across phthalates CGA, CGB, and PPARG (x-axis) for metabolites MnBP, MBzP, MEHP, MEP, and their mixture.

Figure 2. Phthalate effects on mRNAs. The effects are expressed as mean relative fold change (2ΔΔct) and 95% confidence intervals (CIs) compared with DMSO-treated control cells in differentiated cytotrophoblasts (vCTBs). (A) 200 nM MnBP; (B) 3 μM MBzP; (C) 700 nM MEHP; (D) 1.5 μM MEP; (E) mixture of all four metabolites. The black lines indicate female-specific effects, and the gray lines indicate male-specific effects. Overall effects that were significant (p≤0.05) are indicated by a line and marked “sex-specific” if the phthalate effect differed in male and female cells. CG, chorionic gonadotropin; DMSO, dimethyl sulfoxide; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate; PPARG, peroxisome proliferator activated receptor gamma.

In the vCTBs, we analyzed three genes that are targets of phthalates in other cell types (Howdeshell et al. 2007; Lovekamp and Davis 2001; Schlezinger et al. 2004) and that are important to placental function: CYP19A1 and CYP11A1 are involved in steroidogenesis and in xenobiotic metabolism, respectively (Hakkola et al. 1998; Miller 1998), and PTGS2 (COX-2) is involved in prostaglandin production by the placenta (Challis et al. 2005). CYP19A1 and CYP11A1 were up-regulated by phthalates in both sexes. MBzP increased CYP11A1 levels in the female cells 1.4-fold (95% CI: 1.2, 1.6) compared with the control cells (see Figure S1). PTGS2 levels were increased in female cells by MBzP 1.2-fold (95% CI: −0.9, 1.6) and decreased in male cells −0.7-fold (95% CI: −0.5, 1.0). In male and female cells combined, MEHP lowered PTGS2 −0.7-fold (95% CI −0.5, −1.0, p=0.02).

Intracellular Proteins

MnBP and MBzP increased hCGβ (2.5-fold and 2.2-fold, respectively) and PPARγ (1.5-fold and 1.8-fold, respectively) (Table 1, Figure 3). The mixture dose had a stronger sex-specific effect on PPARγ than the single metabolites. PPARγ was increased in female cells [1.7-fold (95% CI: 1.1, 2.4)] and was decreased in male vCTBs [−0.6-fold (95% CI: −0.4, 1.1)].

Figures 3A to 3E are five whisker plots with confidence intervals plotting log difference in protein intensity, relative to control cells (mean, 95 percent CI) (y-axis) across phthalates hCG beta and PPAR gamma (x-axis) for metabolites MnBP, MBzP, MEHP, MEP, and their mixture. Figure 3F and 3G are Western blots for hCG beta and PPAR gamma.

Figure 3. Phthalate effects on intracellular hCGβ and PPARγ levels in differentiated cytotrophoblasts (vCTBs). The effects are expressed as the mean natural log difference in protein intensity and 95% confidence intervals (CIs) compared with DMSO-treated control cells. (A) 200 nM MnBP; (B) 3 μM MBzP; (C) 700 nM MEHP; (D) 1.5 μM MEP; (E) mixture of all four metabolites. The black lines indicate female-specific effects, and the gray lines indicate male-specific effects. Overall effects that were significant (p≤0.05) are indicated by a line and marked “sex-specific” if the phthalate effect differed in male and female cells. Examples of (F) hCGβ and (G) PPARγ Western blots. Each dose group was assayed in duplicate (two lanes). This represents a single experiment conducted on cells isolated from a female placenta at 15.6 wk gestation. DMSO, dimethyl sulfoxide; hCGβ, human chorionic gonadotropin β; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate; PPARγ, peroxisome proliferator activated receptor gamma.

Table 1. Mean log intensities of protein (95% CI) measurements of intracellular hCG-β and PPARγ protein expression in vCTBs treated with phthalate metabolites compared with DMSO-treated control cells.
Protein Metabolite Females Males Overall p-Value Sex-specific p-value
hCGβ 0.1% DMSO 3.43 (1.40, 5.47) 4.51 (3.08, 5.94) 3.91 (2.83, 5.43) Reference Reference
hCGβ MnBP 4.63 (4.12, 5.13) 5.00 (3.90, 6.11) 4.81 (4.11, 5.52) 0.08 0.38
hCGβ MBzP 3.94 (2.64, 5.23) 5.43 (4.16, 6.71) 4.68 (3.57, 5.79) 0.04* 0.46
hCGβ MEHP 4.50 (3.84, 5.16) 4.39 (3.14, 5.64) 4.44 (3.57, 5.32) 0.35 0.19
hCGβ MEP 4.18 (2.88, 5.48) 4.72 (3.45, 5.99) 4.44 (3.23, 5.65) 0.35 0.59
hCGβ 0.4% DMSO 4.00 (2.41, 5.60) 4.93 (3.28, 6.57) 4.47 (3.45, 5.48) Reference Reference
hCGβ Mixture 4.41 (3.81, 5.00) 5.21 (3.90, 6.52) 4.81 (4.11, 5.50) 0.34 0.85
PPARγ 0.1% DMSO 5.80 (4.50, 7.10) 5.39 (3.11, 7.68) 5.60 (4.32, 6.87) Reference Reference
PPARγ MnBP 6.05 (4.76, 7.34) 5.99 (4.31, 7.67) 6.02 (4.95, 7.08) 0.01* 0.29
PPARγ MBzP 6.34 (5.57, 7.10) 6.03 (3.95, 8.10) 6.18 (5.12, 7.24) 0.01* 0.81
PPARγ MEHP 6.16 (5.01, 7.31) 5.53 (3.71, 7.34) 5.84 (4.81, 6.87) 0.37 0.65
PPARγ MEP 5.79 (5.14, 6.45) 5.27 (4.15, 6.39) 5.59 (4.83, 6.34) 0.98 0.87
PPARγ 0.4% DMSO 6.36 (5.36, 7.37) 5.94 (4.82, 7.07) 6.11 (5.39, 6.83) Reference Reference
PPARγ Mixture 6.87 (5.95, 7.79) 5.51 (4.67, 6.34) 6.23 (5.45, 7.01) 0.38 0.01*

Note: p-Values are reported for the overall and sex-specific effects of the phthalate dose on secreted hCG. *p≤0.05. Means were estimated by using a mixed effects model with a random intercept for experiment, allowing for control for between- versus within-placenta variability in protein expression. The final sample included three female and three male biologic replicates. Dose groups: MnBP, 200 nM; MBzP, 3 μM; MEHP, 700 nM; MEP, 1.5 μM. The mixture includes all 4 concentrations. CI, confidence interval; DMSO, dimethyl sulfoxide; hCG, human chorionic gonadotropin; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate; PPARγ, peroxisome proliferator activated receptor gamma; vCTB, villous cytotrophoblast cells.

Secreted hCG Isoforms and Subunits

Phthalate effects on secreted hCG levels in the conditioned media differed by metabolite (Table 2, Figure 4). MEP had the strongest effect, with a −0.66-fold decrease (95% CI: −0.59, −0.74) in hCG-h secreted from female cells. In male cells, MEHP increased hCGα secretion 1.21-fold (95% CI: 1.11, 1.33). The mixture significantly suppressed hCGβ (−0.93-fold; 95% CI: −0.88, −0.99) and hCG-h (−0.86-fold 95% CI: −0.75, −0.98) levels in cells of both sexes.

Figures 4A to 4E are five whisker plots with confidence intervals plotting difference in log hCG concentration, relative to control cells (mean difference, 95% confidence intervals) (y-axis) srm of hCG and intact hCG (x-axis) for metabolites MnBP, MBzP, MEHP, MEP, and their mixture.

Figure 4. Phthalate effects on secreted hCG forms in the conditioned media of differentiated trophoblasts (villous cytotrphoblasts, vCTBs). The effects are expressed as the difference in mean natural log concentration and 95% confidence intervals (CIs) compared with DMSO-treated control cells. (A) 200 nM MnBP; (B) 3 μM MBzP; (C) 700 nM MEHP; (D) 1.5 μM MEP; (E) mixture of all four metabolites. The black lines indicate female-specific effects, and the gray lines indicate male-specific effects. Overall effects that were significant (p≤0.05) are indicated by a line and marked “sex-specific” if the phthalate effect differed in male and female cells. DMSO, dimethyl sulfoxide; hCG human chorionic gonadotropin; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate.

Table 2. Mean log concentrations (95% CI) of hCG in the conditioned media of vCTBs treated with four phthalate metabolites, a mixture of the metabolites, and with DMSO.
Protein Metabolite Females Males Overall p-Value Sex-specific p-value
hCGα 0.1% DMSO 7.06 (6.77, 7.35) 6.74 (6.54, 6.94) 6.90 (6.72, 7.07) Reference Reference
hCGα MnBP 7.04 (6.81, 7.28) 6.73 (6.55, 6.91) 6.88 (6.73, 7.04) 0.71 0.95
hCGα MBzP 7.10 (6.75, 7.45) 6.83 (6.65, 7.02) 6.97 (6.77, 7.16) 0.05* 0.49
hCGα MEHP 7.16 (6.83, 7.48) 6.94 (6.72, 7.15) 7.04 (6.84, 7.24) 0.001* 0.18
hCGα MEP 7.05 (6.78, 7.33) 6.84 (6.63, 7.04) 6.94 (6.77, 7.12) 0.29 0.26
hCGα 0.4% DMSO 7.05 (6.73, 7.37) 6.87 (6.59, 7.15) 6.96 (6.76, 7.15) Reference Reference
hCGα Mixture 7.03 (6.64, 7.42) 6.88 (6.57, 7.19) 6.95 (6.73, 7.18) 0.94 0.58
hCGβ 0.1% DMSO 5.14 (4.00, 6.29) 5.62 (4.88, 6.36) 5.38 (4.73, 6.03) Reference Reference
hCGβ MnBP 5.14 (4.10, 6.17) 5.61 (4.82, 6.40) 5.37 (4.73, 6.00) 0.79 0.98
hCGβ MBzP 5.19 (4.10, 6.27) 5.72 (4.96, 6.49) 5.45 (4.82, 6.09) 0.01* 0.26
hCGβ MEHP 5.03 (3.86, 6.21) 5.58 (4.82, 6.34) 5.30 (4.63, 5.97) 0.08 0.33
hCGβ MEP 5.06 (3.90, 6.22) 5.68 (4.88, 6.47) 5.37 (4.69, 6.04) 0.88 0.27
hCGβ 0.4% DMSO 5.19 (3.50, 6.88) 5.94 (4.84, 7.04) 5.56 (4.64, 6.48) Reference Reference
hCGβ Mixture 5.11 (3.44, 6.78) 5.87 (4.77, 6.97) 5.49 (4.57, 6.41) 0.03 0.87
hCG-h 0.1% DMSO 4.47 (3.68, 5.27) 4.65 (4.06, 5.23) 4.55 (4.08, 5.02) Reference Reference
hCG-h MnBP 4.33 (3.44, 5.22) 4.53 (3.91, 5.15) 4.42 (3.91, 4.94) 0.02* 0.79
hCG-h MBzP 4.40 (3.46, 5.33) 4.66 (4.04, 5.28) 4.53 (3.99, 5.07) 0.69 0.37
hCG-h MEHP 4.40 (3.36, 5.44) 4.53 (3.92, 5.15) 4.47 (3.87, 5.06) 0.34 0.78
hCG-h MEP 4.06 (3.19, 4.93) 4.52 (3.92, 5.11) 4.28 (3.78, 4.79) 0.001* 0.01*
hCG-h 0.4% DMSO 4.47 (2.97, 5.98) 4.84 (4.04, 5.64) 4.66 (3.88, 5.44) Reference Reference
hCG-h Mixture 4.36 (2.93, 5.79) 4.69 (3.88, 5.49) 4.52 (3.76, 5.28) 0.01* 0.55
Intact hCG 0.1% DMSO 5.15 (4.48, 5.83) 5.22 (4.76, 5.68) 5.18 (4.78, 5.57) Reference Reference
Intact hCG MnBP 5.05 (4.48, 5.62) 5.16 (4.72, 5.61) 5.10 (4.75, 5.46) 0.21 0.74
Intact hCG MBzP 5.15 (4.44, 5.86) 5.30 (4.87, 5.72) 5.23 (4.83, 5.62) 0.13 0.25
Intact hCG MEHP 5.20 (4.52, 5.87) 5.25 (4.80, 5.70) 5.22 (4.81, 5.63) 0.11 0.66
Intact hCG MEP 4.85 (4.31, 5.38) 5.19 (4.72, 5.67) 5.01 (4.67, 5.36) 0.04* 0.02*
Intact hCG 0.4% DMSO 5.20 (4.27, 6.12) 5.38 (4.78, 5.97) 5.29 (4.77, 5.80) Reference Reference
Intact hCG 0.1% DMSO 5.14 (4.15, 6.13) 5.33 (4.76, 5.91) 5.24 (4.71, 5.76) 0.21 0.79

Note: p-Values are reported for the overall and sex-specific effects of the phthalate dose on secreted hCG. *p≤0.05. Means were estimated by using a mixed effects model with a random intercept for experiment, allowing for control for between- versus within-placenta variability in hCG secretion. The final sample included three female and four male biologic replicates. Dose groups: MnBP, 200 nM; MBzP, 3 μM; MEHP, 700 nM; MEP, 1.5 μM. The mixture includes all 4 concentrations. CI, confidence interval; DMSO, dimethyl sulfoxide; hCG, human chorionic gonadotropin; MBzP, monobenzyl phthalate; MEHP, mono-2-ethylhexyl phthalate; MEP, monoethyl phthalate; MnBP, mono-n-butyl phthalate; PPARγ, peroxisome proliferator activated receptor gamma; vCTB, villous cytotrophoblast cells.

To evaluate specific effects of phthalates on the levels of hCG subunits, we used ratios of the subunits to intact hCG as our end points (see Figure S2). Here, intact hCG serves as an indicator of overall hCG production. We did not detect sex-specific effects. MnBP significantly up-regulated %hCGβ (1.07-fold; 95% CI: 1.00, 1.15), and MEHP significantly down-regulated %hCGβ (−0.89-fold; 95% CI: −0.82, −0.96). MEP up-regulated %hCGα (1.22-fold; 95% CI: 1.06, 1.42) and %hCGβ (1.17-fold; 95% CI: 1.07, 1.28). The mixture did not significantly alter the hCG subunits.

Correlations between mRNA and Intracellular and Secreted Protein

Even though it is only a snapshot, we can evaluate these correlations as indicators of potential sex differences in the underlying in vitro mechanisms of hCG synthesis and secretion. A direct and positive correlation between CGB mRNA and intracellular hCGβ was only observed in females (23 kDa r=0.58, p<0.0001; Figure 5; see also Table S3). This finding is similar to measurements obtained by others who used this antibody, but without consideration of the sex of the placental cells (Uhlén et al. 2015). In males, there was no correlation between levels of mRNAs for hCG and the intracellular protein levels. Correlations that were common to males and females were the negative correlation between CGA and secreted hCGα and the positive correlation between CGB and secreted hCGβ. PPARG mRNA was positively correlated with its encoded protein only in the male (r=0.49, p=0.002) but not the female vCTBs (r=−0.05, p=0.74). These are univariate correlations and do not take into account clustering within placentas or sources of variability in the correlations other than the sex of the cells.

Figures 5A to 5C are flowcharts showing Spearman rank correlations.

Figure 5. Spearman rank correlations between levels of mRNAs, intracellular proteins, and secreted proteins in female and male differentiated trophoblasts (villous cytotrphoblasts, vCTBs). (A) Correlations that are common to male and female vCTBs; (B) Correlations detected only in female and in male cells (p≤0.05). Positive correlations are drawn as solid black lines, and negative correlations are drawn as dotted lines. hCG, human chorionic gonadotropin; PPAR, peroxisome proliferator activated receptor.

Discussion

Using experimental methods, we generated data that reproduced, in cell models, observed relationships between prenatal exposure to MnBP, MBzP, and MEHP and genes essential to placental gonadotropin synthesis (CGA), placental progesterone synthesis (CYP11A1), and placental prostaglandin production (PTGS2). To further evaluate the biological relevance of these changes in mRNA, we measured two corresponding proteins: hCG and PPARγ. MnBP and MBzP changed hCGβ, and MEHP changed hCGα, in ways that were consistent with mRNA effects. To maximize the two-way translational value of our findings to human pregnancy, in the present study, we used primary human placental cells and dosed them with phthalate metabolites at concentrations found in the urine of pregnant women exposed to environmental levels of phthalates. Compared with working with homogenous or immortalized cell lines or high doses of phthalates, this approach presents unique experimental and statistical challenges, yet it produces results with greater translational significance to pregnancy.

In two birth cohort studies, we previously reported that maternal urinary phthalates were associated with higher levels of mRNA and protein in female placentas/fetuses and with lower levels in male placentas/fetuses (Adibi et al. 2015a, 2017). In the present study, we explored this further by studying mRNA and protein effects in tandem and by including a transcription factor that regulates hCG in the placenta that is also activated by phthalates—PPARγ (Fournier et al. 2011; Handschuh et al. 2007; Hurst and Waxman 2003). It has been hypothesized that PPARγ may be the mechanism by which phthalates can exert endocrine-disrupting effects (Desvergne et al. 2009; Lovekamp-Swan and Davis 2003).

Unexpectedly, we observed sex-specific relationships between PPARγ and hCG. PPARG mRNA and PPARγ protein were positively correlated with hCGβ in male but not female cells. hCG synthesis also differed by sex. In female cells, CGB mRNA was positively correlated with intracellular and secreted hCGβ, as expected. In male cells, CGB mRNA was not correlated with intracellular hCGβ. hCGα and hCGβ are subunits of intact hCG but may also have unique functions independent of classical LH/hCG-receptor (Blithe and Nisula 1987; Blithe et al. 1991; Hussa 1980, 1982; Lee et al. 2013). hCGα and hCGβ were highly positively correlated at the mRNA level in both sexes, but not at the protein level. In female cells, the subunits were inversely correlated, and they were not correlated in male cells. hCG subunit variation may be relevant to sex-specific hCG regulation or to other types of posttranscriptional regulation of hCG.

PPARγ may be a key intermediary between phthalate exposure and placental hCG levels, explaining why hormonal effects are opposite in direction for males and females. The effects of the mixture dose on PPARγ were opposite in males and females. The sex difference in the correlation of PPARG with hCGβ and hCGα could explain the opposite effects of phthalates on hCG. These are important and novel insights that give rise to testable hypotheses that can be further studied using biomarkers in human pregnancy and in vitro by using experimental techniques. Additional levels of complexity in this relationship should be considered, such as the epigenetic regulation of PPARG (Lendvai et al. 2016), mitochondrial expression of PPARG in the placenta (Calabuig-Navarro et al. 2016), and sex-specific mitochondrial dysfunction in response to maternal exposures (Muralimanoharan et al. 2015).

We based our experimental doses on phthalate concentrations in maternal urine that were correlated with placental tissue CGA mRNA to compare the two sets of results and to evaluate reproducibility (Adibi et al. 2017). In three cases, the sex-specific associations between phthalates and placental mRNA expression were supported by the in vitro replication. For CYP11A1, the association with MBzP was stronger in magnitude than the in vitro effect. PTGS2 was down-regulated by MEHP in male placentas in both studies. Discrepancy in results between the two study designs may indicate that isolated trophoblast cells, cultured in the absence of fetal tissue and signals from the fetal pituitary/adrenal/gonadal cells, exhibit weaker or even a reversal of hCG sex differences measured in vivo. We observed heterogeneity in the direction and magnitude of the hCG effects by phthalate metabolite, by sex, by hCG subunit, and by differentiated versus undifferentiated Tbs. We offer these as testable hypotheses to be pursued in future studies in vitro and in human populations. hCG is an essential hormone for pregnancy maintenance and is correlated with many obstetric outcomes (Filicori et al. 2005; Yaron et al. 2002b), yet it has not been considered in studies of fetal endocrine disruption to the same degree as androgens, estrogens, and progesterone. In both sexes, there is evidence that hCG can act as a potent gonadotropin at different points in development. In females, hCG is used to stimulate ovulation for the purpose of in vitro fertilization (Yen et al. 2014). In males, hCG has been used to induce virilization and penile growth in prepubertal males with hypogonadotrophic hypogonadism (Bistritzer et al. 1989) and to induce spermatogenesis in adult life. In normal pregnancy, hCG binds to the luteinizing hormone/chorionic gonadotropin (LH/CGR) receptor in the male fetus during the first trimester, stimulating testicular steroidogenesis and thereby indirectly guiding genital differentiation (Huhtaniemi et al. 1977). If the LH/CGR receptor is inactive because of mutation of its gene, males are born with defective genital masculinization (XY, disorder of sexual differentiation) (Kremer et al. 1995). Women with 20% lower circulating hCG had an increased chance of giving birth to cryptorchid boys (Chedane et al. 2014). Taken together, these findings support the idea that disruption of hCG production and function by phthalates or by other endocrine-disrupting chemicals during pregnancy may have effects on fetal sex differentiation.

There are not likely to be real-life situations where a person would only be exposed to one phthalate metabolite at a time; therefore, we evaluated and compared the effects of phthalate mixtures and of single phthalates. We used a nonbalanced approach (i.e., nonequivalent doses) in the design of our doses to accurately reflect real-life exposures during pregnancy (Evans et al. 2012). We detected three cases of a significant effect of the mixture. In the case of CGB mRNA in undifferentiated trophoblasts, the mixture effect was analogous to the single-metabolite effects. We interpret this to mean that there was a common mechanism that was not overwhelmed by the phthalate concentrations used. In the case of intracellular PPARγ in the differentiated trophoblasts, the mixture effect was stronger than the single-metabolite effects, sex-specific, and opposite in direction. The mechanism of PPARγ activation may differ in the case of multiple versus single phthalates. In the case of the female cells, the different metabolites may have synergized to increase the strength of the positive effect on PPARγ (i.e., agonism). In the case of the male cells, the metabolites may have competed for or antagonized (or both) a common mechanism to cause the down-regulation of PPARγ. This latter scenario may also apply to secreted hCGβ, where the negative effect of the mixture was a reversal of the positive effect of MBzP. Comparisons between the effects of single metabolites versus mixtures are critical in establishing which metabolites are more biologically potent and should be prioritized in efforts to reduce risks to the placenta and fetus.

Sex and gestational variation in hCG have been previously established at the population level in analyses of hCG biomarker data (Adibi et al. 2015b; Bremme and Eneroth 1983; Buckberry et al. 2014; Clements et al. 1976; Cowans et al. 2009; Nagy et al. 1994a, 1994b; Steier et al. 1999; Yaron et al. 2002a). This type of variation is generally not considered when analyzing in vitro experimental data. We observed in our primary tissue cultures that the sex and gestational-age variation in mRNA and secreted protein levels that were present at baseline persisted and rendered our biologic replicates less comparable. For this reason, we controlled for these variables in the data analysis using basic multivariate statistical techniques.

There are noteworthy caveats in making comparisons between observational and experimental findings. In the in vitro study, our controls received no phthalates, whereas in the observational studies, we compared pregnancies with lower but not zero exposure because all subjects were environmentally exposed. Similarly, all of the pregnant women were exposed to a phthalate mixture even though we estimated associations with single metabolites to which we compared the results of the present study (Adibi et al. 2017). Phthalate concentrations in placental tissue are most likely lower than the urine concentrations modeled here [phthalates are not measured in blood owing to a short half-life and to a high risk of phthalate contamination in the sampling process (Calafat et al. 2015)]. In a small pilot study, we estimated that MnBP, MBzP, and MEP were higher, but within an order of magnitude, in urine than in placental tissue by 14-, 28-, and 8-fold respectively. MEHP was 4-fold higher in placental tissue; therefore, we may have slightly underestimated the true exposure to the placenta in this study (J. Adibi and N. Snyder, unpublished data, 2017); this may increase the translational value over previously published studies that dosed with concentrations of MEHP that are 1–3 orders of magnitude higher than urinary levels (Meruvu et al. 2016; Tetz et al. 2013; Wang et al. 2016).

These relationships and the relatively small effect sizes are supported by studies conducted by other investigators. In a study of immature rat Leydig cells, CYP11A1 mRNA was 30–40% higher than in controls at 50–500 nM MnBP (Li et al. 2016), similar to the effect we found in male placental cells (33%) in the present study. In an immortalized first-trimester cell line (sex not specified), PTGS, the gene that encodes the COX-2 protein, increased approximately 2- to 3-fold at a 90- μM MEHP dose (129-fold as high as the dose used in our study) (Tetz et al. 2013). In another study, the effects of MEHP on the COX-2 protein were not detected within the dose range that we used, but only at higher doses (Wang et al. 2016). We measured a significant reduction in PTGS2 by MEHP. Results cannot be easily compared owing to the large differences in dose and to a high likelihood of different mechanisms at low versus high doses. In a review of 35 published studies that reported associations of prenatal phthalate exposures with obstetric outcomes, the authors indicated that knowledge of a mechanism and of ways to measure specific biologic intermediaries in human pregnancy are lacking (Marie et al. 2015). Our findings address this gap by offering biologic insight into correlations of prenatal phthalate exposure with placental end points.

Conclusion

In conclusion, we moved one step beyond an observational association by showing sex-specific relationships between MnBP and placental CGA, and between MBzP and placental CYP11A1 using experimental methods with primary cells. The finding that MnBP can alter chorionic gonadotropin α (CGA) was extended to other phthalate metabolites (MBzP, MEHP, and MEP) and to intracellular and secreted hCG and its subunits. The sex-specific effects of phthalates may be challenging to reproduce in vitro in the absence of the fetus. However, our results support the hypothesis that hCG is altered by low concentrations of single and combined phthalates, which is relevant for environmental exposure to phthalates. We are the first to report sex differences in hCG transcription and translation, which we believe to be partially regulated by PPARγ. This finding provides a testable hypothesis to better understand why the hormonal effects of phthalates are opposite in direction between males and females. In future studies, it will be important to quantify the functional significance (for cells, for organs, and for the future child) of small perturbations in hCG by ubiquitous phthalate exposures.

Acknowledgments

We acknowledge S. Fisher and O. Genbacev for their mentoring in placental biology and to Y. Zhou, M. Gormley, N. Hunkapiller, B. Hromatka, and members of the Fisher Lab who all provided valuable mentoring, technical training resources, and feedback on this project. We thank the D. Lewis and S. Gollin laboratories in the Department of Human Genetics and the University of Pittsburgh Genomics Core for support in completing experiments at our new university; we also thank T. Grönholm for technical assistance with hCG immunoassays. Funding was received from the National Institute of Envirnomental Health Sciences/National Institutes of Health (NIEHS/NIH) (grant nos. 1K99 ES017780-01, 5R00ES017780 – 06) and from the Science Innovation Fund of the Passport Foundation (J.J.A.). Funds were provided by the Department of Epidemiology at the University of Pittsburgh to complete this project. Contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.

References

Adibi JJ, Hauser R, Williams PL, Whyatt RM, Calafat AM, Nelson H, et al. 2009. Maternal urinary metabolites of di-(2-ethylhexyl) phthalate in relation to the timing of labor in a US multicenter pregnancy cohort study. Am J Epidemiol 169(8):1015–1024, PMID: 19251754, 10.1093/aje/kwp001.

Adibi JJ, Whyatt RM, Hauser R, Bhat HK, Davis BJ, Calafat AM, et al. 2010. Transcriptional biomarkers of steroidogenesis and trophoblast differentiation in the placenta in relation to prenatal phthalate exposure. Environ Health Perspect 118(2):291–296, PMID: 20123604, 10.1289/ehp.0900788.

Adibi JJ, Lee MK, Naimi AI, Barrett E, Nguyen RH, Sathyanarayana S, et al. 2015a. Human chorionic gonadotropin partially mediates phthalate association with male and female anogenital distance. J Clin Endocrinol Metab 100(9):E1216–E1224, PMID: 26200238, 10.1210/jc.2015-2370.

Adibi JJ, Lee MK, Saha S, Boscardin WJ, Apfel A, Currier RJ. 2015b. Fetal sex differences in human chorionic gonadotropin fluctuate by maternal race, age, weight and by gestational age. J Dev Orig Health Disease 6(6):493–500, PMID: 26242396, 10.1017/S2040174415001336.

Adibi JJ, Buckley JP, Lee MK, Williams PL, Just AC, Zhao Y, et al. 2017. Maternal urinary phthalates and sex-specific placental mRNA levels in an urban birth cohort. Environ Health 16(1):35, PMID: 28381288, 10.1186/s12940-017-0241-5.

Alfthan H, Haglund C, Dabek J, Stenman UH. 1992. Concentrations of human choriogonadotropin, its beta-subunit, and the core fragment of the beta-subunit in serum and urine of men and nonpregnant women. Clin Chem 38(10):1981–1987, PMID: 1382894.

Bistritzer T, Lunenfeld B, Passwell JH, Theodor R. 1989. Hormonal therapy and pubertal development in boys with selective hypogonadotropic hypogonadism. Fertil Steril 52(2):302–306, PMID: 2753178, 10.1016/S0015-0282(16)60859-2.

Blithe DL, Nisula BC. 1987. Similarity of the clearance rates of free alpha-subunit and alpha-subunit dissociated from intact human chorionic gonadotropin, despite differences in sialic acid contents. Endocrinology 121(4):1215–1220, PMID: 3653025, 10.1210/endo-121-4-1215.

Blithe DL, Richards RG, Skarulis MC. 1991. Free alpha molecules from pregnancy stimulate secretion of prolactin from human decidual cells: a novel function for free alpha in pregnancy. Endocrinology 129(4):2257–2259, PMID: 1717245, 10.1210/endo-129-4-2257.

Bremme K, Eneroth P. 1983. Fetal sex dependent hormone levels in early pregnant women with elevated maternal serum alpha-fetoprotein. Int J Gynaecol Obstet 21(6):451–457, PMID: 6198221, 10.1016/0020-7292(83)90034-6.

Buckberry S, Bianco-Miotto T, Bent SJ, Dekker GA, Roberts CT. 2014. Integrative transcriptome meta-analysis reveals widespread sex-biased gene expression at the human fetal-maternal interface. Mol Hum Reprod 20(8):810–819, PMID: 24867328, 10.1093/molehr/gau035.

Calabuig-Navarro V, Puchowicz M, Glazebrook P, Haghiac M, Minium J, Catalano P, et al. 2016. Effect of ω-3 supplementation on placental lipid metabolism in overweight and obese women. Am J Clin Nutr 103(4):1064–1072, PMID: 26961929, 10.3945/ajcn.115.124651.

Calafat AM, Longnecker MP, Koch HM, Swan SH, Hauser R, Goldman LR, et al. 2015. Optimal exposure biomarkers for nonpersistent chemicals in environmental epidemiology. Environ Health Perspect 123(7):A166–A168, PMID: 26132373, 10.1289/ehp.1510041.

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.

Challis JR, Bloomfield FH, Bocking AD, Casciani V, Chisaka H, Connor K, et al. 2005. Fetal signals and parturition. J Obstet Gynaecol Res 31(6):492–499, PMID: 16343248, 10.1111/j.1447-0756.2005.00342.x.

Chedane C, Puissant H, Weil D, Rouleau S, Coutant R. 2014. Association between altered placental human chorionic gonadotrophin (hCG) production and the occurrence of cryptorchidism: a retrospective study. BMC Pediatr 14:191, PMID: 25064170, 10.1186/1471-2431-14-191.

Clements JA, Reyes FI, Winter JS, Faiman C. 1976. Studies on human sexual development. Iii. Fetal pituitary and serum, and amniotic fluid concentrations of LH, CG, and FSH. J Clin Endocrinol Metab 42(1):9–19, PMID: 1249196, 10.1210/jcem-42-1-9.

Cowans NJ, Stamatopoulou A, Maiz N, Spencer K, Nicolaides KH. 2009. The impact of fetal gender on first trimester nuchal translucency and maternal serum free beta-hCG and PAPP-A MoM in normal and trisomy 21 pregnancies. Prenat Diagn 29(6):578–581, PMID: 19288535, 10.1002/pd.2246.

Desvergne B, Feige JN, Casals-Casas C. 2009. PPAR-mediated activity of phthalates: A link to the obesity epidemic?. Mol Cell Endocrinol 304(1–2):43–48, PMID: 19433246, 10.1016/j.mce.2009.02.017.

Ejaredar M, Nyanza EC, Ten Eycke K, Dewey D. 2015. Phthalate exposure and childrens neurodevelopment: a systematic review. Environ Res 142:51–60, PMID: 26101203, 10.1016/j.envres.2015.06.014.

Evans RM, Scholze M, Kortenkamp A. 2012. Additive mixture effects of estrogenic chemicals in human cell-based assays can be influenced by inclusion of chemicals with differing effect profiles. PloS One 7(8):e43606, PMID: 22912892, 10.1371/journal.pone.0043606.

Ferguson KK, McElrath TF, Meeker JD. 2014. Environmental phthalate exposure and preterm birth. JAMA Pediatr 168(1):61–67, PMID: 24247736, 10.1001/jamapediatrics.2013.3699.

Ferguson KK, McElrath TF, Cantonwine DE, Mukherjee B, Meeker JD. 2015. Phthalate metabolites and bisphenol-A in association with circulating angiogenic biomarkers across pregnancy. Placenta 36(6):699–703, PMID: 25913709, 10.1016/j.placenta.2015.04.002.

Filicori M, Fazleabas AT, Huhtaniemi I, Licht P, Rao ChV, Tesarik J, et al. 2005. Novel concepts of human chorionic gonadotropin: reproductive system interactions and potential in the management of infertility. Fertil Steril 84(2):275–284, PMID: 16084861, 10.1016/j.fertnstert.2005.02.033.

Fournier T, Guibourdenche J, Handschuh K, Tsatsaris V, Rauwel B, Davrinche C, et al. 2011. PPARγ and human trophoblast differentiation. J Reprod Immunol 90(1):41–49, 10.1016/j.jri.2011.05.003.

Genbacev O, Donne M, Kapidzic M, Gormley M, Lamb J, Gilmore J, et al. 2011. Establishment of human trophoblast progenitor cell lines from the chorion. Stem Cells 29(9):1427–1436, PMID: 21755573, 10.1002/stem.686.

Genbacev O, Larocque N, Ona K, Prakobphol A, Garrido-Gomez T, Kapidzic M, et al. 2016. Integrin α4-positive human trophoblast progenitors: functional characterization and transcriptional regulation. Hum Reprod 31(6):1300–1314, PMID: 27083540, 10.1093/humrep/dew077.

Hakkola J, Pelkonen O, Pasanen M, Raunio H. 1998. Xenobiotic-metabolizing cytochrome P450 enzymes in the human feto-placental unit: role in intrauterine toxicity. Crit Rev Toxicol 28(1):35–72, PMID: 9493761, 10.1080/10408449891344173.

Handschuh K, Guibourdenche J, Tsatsaris V, Guesnon M, Laurendeau I, Evain-Brion D, et al. 2007. Human chorionic gonadotropin produced by the invasive trophoblast but not the villous trophoblast promotes cell invasion and is down-regulated by peroxisome proliferator-activated receptor-γ. Endocrinology 148(10):5011–5019, PMID: 17628005, 10.1210/en.2007-0286.

Howdeshell KL, Furr J, Lambright CR, Rider CV, Wilson VS, Gray LE. Jr
2007. Cumulative effects of dibutyl phthalate and diethylhexyl phthalate on male rat reproductive tract development: altered fetal steroid hormones and genes. Toxicol Sci 99(1):190–202, PMID: 17400582, 10.1093/toxsci/kfm069.

Huber A, Hudelist G, Knöfler M, Saleh L, Huber JC, Singer CF. 2007. Effect of highly purified human chorionic gonadotropin preparations on the gene expression signature of stromal cells derived from endometriotic lesions: potential mechanisms for the therapeutic effect of human chorionic gonadotropin in vivo. Fertil Steril 88(4 Suppl):1232–1239, PMID: 17561010, 10.1016/j.fertnstert.2007.02.006.

Huhtaniemi IT, Korenbrot CC, Jaffe RB. 1977. HCG binding and stimulation of testosterone biosynthesis in the human fetal testis. J Clin Endocrinol Metab 44(5):963–967, PMID: 192755, 10.1210/jcem-44-5-963.

Hunkapiller NM, Fisher SJ. 2008. Chapter 12. Placental remodeling of the uterine vasculature. Methods Enzymol 445:281–302, PMID: 19022064, 10.1016/S0076-6879(08)03012-7.

Hurst CH, Waxman DJ. 2003. Activation of PPARα and PPARγ by environmental phthalate monoesters. Toxicol Sci 74(2):297–308, PMID: 12805656, 10.1093/toxsci/kfg145.

Hussa RO. 1980. Biosynthesis of human chorionic gonadotropin. Endocr Rev 1(3):268–294, PMID: 6262070, 10.1210/edrv-1-3-268.

Hussa RO. 1982. Clinical utility of human chorionic gonadotropin and alpha-subunit measurements. Obstet Gynecol 60(1):1–12, PMID: 6283446.

Kremer H, Kraaij R, Toledo SP, Post M, Fridman JB, Hayashida CY, et al. 1995. Male pseudohermaphroditism due to a homozygous missense mutation of the luteinizing hormone receptor gene. Nat Genet 9(2):160–164, PMID: 7719343, 10.1038/ng0295-160.

Latini G, De Felice C, Presta G, Del Vecchio A, Paris I, Ruggieri F, et al. 2003. Exposure to di(2-ethylhexyl)phthalate in humans during pregnancy. A preliminary report. Biol Neonate 83(1):22–24, PMID: 12566679.

Lee CL, Chiu PC, Hautala L, Salo T, Yeung WS, Stenman UH, et al. 2013. Human chorionic gonadotropin and its free β-subunit stimulate trophoblast invasion independent of LH/hCG receptor. Mol Cell Endocrinol 375(1–2):43–52, PMID: 23684886, 10.1016/j.mce.2013.05.009.

Lendvai A, Deutsch MJ, Plösch T, Ensenauer R. 2016. The peroxisome proliferator-activated receptors under epigenetic control in placental metabolism and fetal development. Am J Physiol Endocrinol Metab 310(10):E797–E810, PMID: 26860983, 10.1152/ajpendo.00372.2015.

Li L, Chen X, Hu G, Wang S, Xu R, Zhu Q, et al. 2016. Comparison of the effects of dibutyl and monobutyl phthalates on the steroidogenesis of rat immature Leydig cells. Biomed Res Int 2016:1376526, PMID: 27148549, 10.1155/2016/1376526.

Lovekamp TN, Davis BJ. 2001. Mono-(2-ethylhexyl) phthalate suppresses aromatase transcript levels and estradiol production in cultured rat granulosa cells. Toxicol Appl Pharmacol 172(3):217–224, PMID: 11312650, 10.1006/taap.2001.9156.

Lovekamp-Swan T, Davis BJ. 2003. Mechanisms of phthalate ester toxicity in the female reproductive system. Environ Health Perspect 111(2):139–145, PMID: 12573895.

Maltepe E, Bakardjiev AI, Fisher SJ. 2010. The placenta: transcriptional, epigenetic, and physiological integration during development. J Clin Invest 120(4):1016–1025, PMID: 20364099, 10.1172/JCI41211.

Marie C, Vendittelli F, Sauvant-Rochat MP. 2015. Obstetrical outcomes and biomarkers to assess exposure to phthalates: a review. Environ Int 83:116–136, PMID: 26118330, 10.1016/j.envint.2015.06.003.

Meruvu S, Zhang J, Bedi YS, Choudhury M. 2016. Mono-(2-ethylhexyl) phthalate induces apoptosis through miR-16 in human first trimester placental cell line HTR-8/SVneo. Toxicol In Vitro 31:35–42, PMID: 26597031, 10.1016/j.tiv.2015.11.010.

Miller WL. 1998. Steroid hormone biosynthesis and actions in the materno-feto-placental unit. Clin Perinatol 25(4):799–817, PMID: 9891616.

Muralimanoharan S, Guo C, Myatt L, Maloyan A. 2015. Sexual dimorphism in miR-210 expression and mitochondrial dysfunction in the placenta with maternal obesity. Int J Obes (Lond) 39(8):1274–1281, PMID: 25833255, 10.1038/ijo.2015.45.

Nagy AM, Glinoer D, Picelli G, Delogne-Desnoeck J, Fleury B, Courte C, et al. 1994a. Total amounts of circulating human chorionic gonadotrophin α and β subunits can be assessed throughout human pregnancy using immunoradiometric assays calibrated with the unaltered and thermally dissociated heterodimer. J Endocrinol 140(3):513–520, PMID: 7514205.

Nagy AM, Jauniaux E, Jurkovic D, Meuris S. 1994b. Placental production of human chorionic gonadotrophin α and β subunits in early pregnancy as evidenced in fluid from the exocoelomic cavity. J Endocrinol 142(3):511–516, PMID: 7525825.

Park JY, Su YQ, Ariga M, Law E, Jin SL, Conti M. 2004. EGF-like growth factors as mediators of LH action in the ovulatory follicle. Science 303(5658):682–684, PMID: 14726596, 10.1126/science.1092463.

Racca AC, Camolotto SA, Ridano ME, Bocco JL, Genti-Raimondi S, Panzetta-Dutari GM. 2011. Krüppel-like factor 6 expression changes during trophoblast syncytialization and transactivates βhCG and PSG placental genes. PloS One 6(7):e22438, PMID: 21799854, 10.1371/journal.pone.0022438.

Rawn SM, Cross JC. 2008. The evolution, regulation, and function of placenta-specific genes. Annu Rev Cell Dev Biol 24:159–181, PMID: 18616428, 10.1146/annurev.cellbio.24.110707.175418.

Schlezinger JJ, Howard GJ, Hurst CH, Emberley JK, Waxman DJ, Webster T, et al. 2004. Environmental and endogenous peroxisome proliferator-activated receptor γ agonists induce bone marrow B cell growth arrest and apoptosis: interactions between mono(2-ethylhexyl)phthalate, 9-cis-retinoic acid, and 15-deoxy-Δ12,14-prostaglandin J2. J Immunol 173(5):3165–3177, PMID: 15322177, 10.4049/jimmunol.173.5.3165.

Senol-Cosar O, Flach RJ, DiStefano M, Chawla A, Nicoloro S, Straubhaar J, et al. 2016. Tenomodulin promotes human adipocyte differentiation and beneficial visceral adipose tissue expansion. Nat Commun 7:10686, PMID: 26880110, 10.1038/ncomms10686.

Steier JA, Myking OL, Bergsjø PB. 1999. Correlation between fetal sex and human chorionic gonadotropin in peripheral maternal blood and amniotic fluid in second and third trimester normal pregnancies. Acta Obstet Gynecol Scand 78(5):367–371, PMID: 10326878.

Swan SH, Sathyanarayana S, Barrett ES, Janssen S, Liu F, Nguyen RH, et al. 2015. First trimester phthalate exposure and anogenital distance in newborns. Hum Reprod 30(4):963–972, PMID: 25697839, 10.1093/humrep/deu363.

Tetz LM, Cheng AA, Korte CS, Giese RW, Wang P, Harris C, et al. 2013. Mono-2-ethylhexyl phthalate induces oxidative stress responses in human placental cells in vitro. Toxicol Appl Pharmacol 268(1):47–54, PMID: 23360888, 10.1016/j.taap.2013.01.020.

Tetz LM, Aronoff DM, Loch-Caruso R. 2015. Mono-ethylhexyl phthalate stimulates prostaglandin secretion in human placental macrophages and THP-1 cells. Reprod Biol Endocrinol 13:56, PMID: 26036283, 10.1186/s12958-015-0046-8.

Thirkill TL, Douglas GC. 1997. Differentiation of human trophoblast cells in vitro is inhibited by dimethylsulfoxide. J Cell Biochem 65(4):460–468, PMID: 9178096, 10.1002/(SICI)1097-4644(19970615)65:4<460::AID-JCB2>3.0.CO;2-M.

Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, Mardinoglu A, et al. 2015. Tissue-based map of the human proteome. Science 347(6220):1260419, PMID: 25613900, 10.1126/science.1260419.

Verbeke G, Molenberghs G. 2000. Linear Mixed Models for Longitudinal Data. New York:Springer.

Wald NJ, Rodeck C, Hackshaw AK, Walters J, Chitty L, Mackinson AM. 2003. First and second trimester antenatal screening for Down’s syndrome: the results of the Serum, Urine and Ultrasound Screening Study (SURUSS). J Med Screen 10(2):56–104, PMID: 14746340, 10.1258/096914103321824133.

Wang XK, Agarwal M, Parobchak N, Rosen A, Vetrano AM, Srinivasan A, et al. 2016. Mono-(2-ethylhexyl) phthalate promotes pro-labor gene expression in the human placenta. PloS One 11(1):e0147013, PMID: 26751383, 10.1371/journal.pone.0147013.

Wang YV, Wade M, Wong E, Li YC, Rodewald LW, Wahl GM. 2007. Quantitative analyses reveal the importance of regulated Hdmx degradation for p53 activation. Proc Natl Acad Sci USA 104(30):12365–12370, PMID: 17640893, 10.1073/pnas.0701497104.

Winn VD, Haimov-Kochman R, Paquet AC, Yang YJ, Madhusudhan MS, Gormley M, et al. 2007. Gene expression profiling of the human maternal–fetal interface reveals dramatic changes between midgestation and term. Endocrinology 148(3):1059–1079, PMID: 17170095, 10.1210/en.2006-0683.

Woodruff TJ, Zota AR, Schwartz JM. 2011. Environmental chemicals in pregnant women in the united states: NHANES 2003–2004. Environ Health Perspect 119(6):878–885, PMID: 21233055, 10.1289/ehp.1002727.

Yaron Y, Lehavi O, Orr-Urtreger A, Gull I, Lessing JB, Amit A, et al. 2002a. Maternal serum HCG is higher in the presence of a female fetus as early as week 3 post-fertilization. Hum Reprod 17(2):485–489, PMID: 11821300, 10.1093/humrep/17.2.485.

Yaron Y, Ochshorn Y, Heifetz S, Lehavi O, Sapir Y, Orr-Urtreger A. 2002b. First trimester maternal serum free human chorionic gonadotropin as a predictor of adverse pregnancy outcome. Fetal Diagn Ther 17(6):352–356, PMID: 12393965, 10.1159/000065384.

Yen SSC, Strauss JF, Barbieri RL. 2014. Yen and Jaffe’s Reproductive Endocrinology: Physiology, Pathophysiology, and Clinical Management. 7th ed. Philadelphia, PA:Elsevier/Saunders, 243–271.

Air Conditioning Use and Heat-Related Deaths: How a Natural Disaster Presented a Unique Research Opportunity

PDF icon PDF Version (1.1 MB)

  • Published: 30 October 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.



The aftermath of the Great East Japan Earthquake and resulting tsunami of 11 March 2011 provided researchers an unexpected opportunity to explore the relationship between air conditioning and risk of heat-related death.1 Many earlier studies reported associations between access to and/or use of air conditioning and lower mortality risk.2,3,4,5,6 However, the new results, published in Environmental Health Perspectives, suggest that limiting the use of air conditioning during summer heat may not necessarily have adverse consequences for human health.1

The 2011 earthquake and tsunami severely damaged the Fukushima Daiichi Nuclear Power Plant and caused severe electrical shortages in the Tokyo metropolitan area. To conserve energy, the government strongly encouraged residents served by Tokyo Electric Power Company and Tohoku Electric Power Company to reduce electricity consumption by 15% from July to September of 2011.7

Photograph of Tokyo residents shopping for fans
Campaigns promoting energy conservation have long been a summertime fixture in Japan.14 But the sudden loss of the Fukushima Daiichi Nuclear Power Plant in 2011 gave a new urgency to reducing energy consumption. Among other measures, many residents swapped air conditioning for fans. Image: © Associated Press.

In the new study, a team led by professor Masahiro Hashizume of Nagasaki University compared the number of heat-related deaths in Japan in 2011 and 2012 to the number reported between 2008 and 2010. In 13 of the 15 prefectures with the greatest drop in electrical consumption, the researchers estimated that heat-related deaths did not increase but instead decreased by 5–9%. The prefectures with less change in electricity consumption also saw little change in risk of heat-related death.1

In Japan, 90% of households have air conditioners.8 Before the Fukushima earthquake, an estimated 53% of the electricity consumed during the peak hours in summer went toward running air conditioners.7 The restrictions on electrical use after the disaster are estimated to have reduced overall household summer electricity consumption by up to 18%.7

A mass media and online public information campaign9 urged people to set their air conditioners to 28°C (82°F), to run them less, and to switch to electric fans. People also were advised to drink more water, dress in cool clothing, wear hats outdoors, and use blinds and curtains to block sunlight. Businesses allowed far more casual work attire than usual and shifted hours of operation to cooler times. Lights were dimmed in public areas, workplaces, and stores.10

About two-thirds of households changed their energy use habits, whereas only 4% of households bought and installed newer energy-efficient air conditioners.11 “This suggests that behavioral changes, rather than technological measures, were a primary reason for the reductions in household electricity consumption,” says Hashizume. “Public information campaigns seem quite effective to raise people’s awareness about how to prevent heat-related illness.”

By one estimate,12 air conditioning around the world consumes 1 trillion kilowatt-hours of electricity yearly, and the use of air conditioning could increase 10 times by 2050 in response to rising temperatures. This anticipated surge in power production likely will add to greenhouse gas emissions and promote global-scale climate change, while the “waste heat” put off by air conditioning units will additionally warm urban areas.13

More research is needed to determine whether the Japanese experience could help guide prevention of heat-related deaths of people living elsewhere. “It is unknown whether similar public service campaigns, without the backdrop of a natural crisis, would change behaviors and reduce energy use as effectively,”cautions Hashizume.

David Hondula, an assistant professor at Arizona State University’s Urban Climate Research Center, who was not part of the study, says the insights into human adaptive capacity and behavior are encouraging. “Understanding how people cope with the coupled hazards of extreme heat and power failure is important in an era of increasing temperatures and demands on urban infrastructure,” Hondula says. “The new finding that heat-related mortality did not increase, despite electricity limitations, suggests that we need to re-evaluate the importance of air conditioning versus other adaptive strategies as determinants of heat-related health risks.”


Carol Potera, based in Montana, also writes for Microbe, Genetic Engineering News, and the American Journal of Nursing

References

1. Kim Y, Gasparrini A, Hashizume M, Honda Y, Ng CFS, Armstrong B. 2017. Heat-related mortality in Japan after the 2011 Fukushima disaster: an analysis of potential influence of reduced electricity consumption. Environ Health Perspect 125(7):077005, PMID: 28686555, 10.1289/EHP493.

2. Zhang Y, Nitschke M, Krackowizer A, Dear K, Pisaniello D, Weinstein P, et al. 2017. Risk factors for deaths during the 2009 heat wave in Adelaide, Australia: a matched case-control study. Int J Biometeorol 61(1):35–47, PMID: 27221967, 10.1007/s00484-016-1189-9.

3. Ostro B, Rauch S, Green R, Malig B, Basu R. 2010. The effects of temperature and use of air conditioning on hospitalizations. Am J Epidemiol 172(9):1053–1061, PMID: 20829270, 10.1093/aje/kwq231.

4. Medina-Ramón M, Schwartz J. 2007. Temperature, temperature extremes, and mortality: a study of acclimatisation and effect modification in 50 US cities. Occup Environ Med 64(12):827–833, PMID: 17600037, 10.1136/oem.2007.033175.

5. O’Neill MS, Zanobetti A, Schwartz J. 2005. Disparities by race in heat-related mortality in four US cities: the role of air conditioning prevalence. J Urban Health 82(2):191–197, PMID: 15888640, 10.1093/jurban/jti043.

6. Rogot E, Sorlie PD, Backlund E. 1992. Air-conditioning and mortality in hot weather. Am J Epidemiol 136(1):106–116, PMID: 1415127.

7. Agency for Natural Resources and Energy. 2012. “Annual Report on Heisei 22 Energy.” Energy White Paper 2011. Tokyo, Japan:Agency for Natural Resources and Energy, Japanese Ministry of Economy, Trade and Industry. http://www.enecho.meti.go.jp/about/whitepaper/2011html [accessed 25 September 2017].

8. Statistics Bureau. 2014. National Survey of Family Income and Expenditure 2014 [website]. http://www.stat.go.jp/english/data/zensho/index.htm [accessed 25 September 2017].

9. Nishiyama H. 2013. Japan’s Policy on Energy Conservation. Presented at: Energy Management Action Network 4th Workshop, 31 January 2013, Tokyo, Japan. https://www.iea.org/workshops/energy-management-action-network-emak-workshop-4.html [accessed 25 September 2017].

10. Murakoshi C, Nakagami H, Hirayama S. 2012. Electricity crisis and behavior change in the residential sector: Tokyo before and after the Great East Japan Earthquake. In: Proceedings from the 2012 ACEEE Summer Study on Energy Efficiency in Buildings, Part 7: Building Efficiency, Human Behavior, and Social Dynamics, 12–17 August 2012, Pacific Grove, CA. Washington, DC:Omnipress, 198–211.

11. NISHIO K-i, OFUJI K. 2012. Differences in electricity conservation rates by households and effects of conservation measures. Nihon Kenchiku Gakkai Kankyokei Ronbunshu 77(679):753–759, 10.3130/aije.77.753.

12. Cox S. 2012. Cooling a warming planet: a global air conditioning surge. Yale Environment 360, Features section. 10 July 2012. http://e360.yale.edu/features/cooling_a_warming_planet_a_global_air_conditioning_surge [accessed 25 September 2017].

13. Lundgren K, Kjellstrom T. 2013. Sustainability challenges from climate change and air conditioning use in urban areas. Sustainability 5(7):3116–3128, 10.3390/su5073116.

14. Smart R. 2015. Ditch the tie and reduce the AC—Japan’s Cool Biz gets summer hell just about right. Quartz. 29 July 2015. https://qz.com/465327/ditch-the-tie-and-reduce-the-ac-japans-cool-biz-gets-summer-hell-just-about-right [accessed 25 September 2017].

Longer-Term Impact of High and Low Temperature on Mortality: An International Study to Clarify Length of Mortality Displacement

Author Affiliations open

1Department of Social and Environmental Health Research, London School of Hygiene and Tropical Medicine, London, UK

2School of Forestry and Environmental Studies, Yale University, New Haven, Connecticut, USA

3Institute of Advanced Studies at the University of São Paulo, São Paulo, Brazil

4Department of Environmental and Occupational Medicine, National Taiwan University, Taipei, Taiwan

5Department of Epidemiology and Preventive Medicine, School of Public Health and Preventive Medicine, Monash University, Melbourne, Australia

6Environmental Health Sciences Institute, Dublin Institute of Technology, Dublin, Ireland

7Department of Pediatric Infectious Diseases, Institute of Tropical Medicine, Nagasaki University, Nagasaki, Japan

8Faculty of Health and Sport Sciences, University of Tsukuba, Tsukuba, Japan

9Graduate School of Public Health, Seoul National University, Seoul, Republic of Korea

10School of Epidemiology, Public Health and Preventive Medicine, University of Ottawa, Ottawa, Canada

11Department of Epidemiology, Lazio Regional Health Service, Rome, Italy

12Department of Environmental Health, Harvard T.H. Chan School of Public Health, Boston, Massachusetts, USA

13Institute of Environmental Assessment and Water Research (IDAEA), Spanish Council for Scientific Research (CSIC), Barcelona, Spain

14Shanghai Children’s Medical Center, Shanghai Jiao Tong University, Shanghai, China

15School of Public Health, Anhui Medical University, Hefei, China

16School of Public Health and Social Work, Queensland University of Technology, Brisbane, Australia

17Department of Public Health, National Taiwan University, Taipei, Taiwan

18Institute of Environment, Health and Societies, Brunel University London, London, UK

PDF icon PDF Version (739 KB)

  • Background:
    In many places, daily mortality has been shown to increase after days with particularly high or low temperatures, but such daily time-series studies cannot identify whether such increases reflect substantial life shortening or short-term displacement of deaths (harvesting).
    Objectives:
    To clarify this issue, we estimated the association between annual mortality and annual summaries of heat and cold in 278 locations from 12 countries.
    Methods:
    Indices of annual heat and cold were used as predictors in regressions of annual mortality in each location, allowing for trends over time and clustering of annual count anomalies by country and pooling estimates using meta-regression. We used two indices of annual heat and cold based on preliminary standard daily analyses: a) mean annual degrees above/below minimum mortality temperature (MMT), and b) estimated fractions of deaths attributed to heat and cold. The first index was simpler and matched previous related research; the second was added because it allowed the interpretation that coefficients equal to 0 and 1 are consistent with none (0) or all (1) of the deaths attributable in daily analyses being displaced by at least 1 y.
    Results:
    On average, regression coefficients of annual mortality on heat and cold mean degrees were 1.7% [95% confidence interval (CI): 0.3, 3.1] and 1.1% (95% CI: 0.6, 1.6) per degree, respectively, and daily attributable fractions were 0.8 (95% CI: 0.2, 1.3) and 1.1 (95% CI: 0.9, 1.4). The proximity of the latter coefficients to 1.0 provides evidence that most deaths found attributable to heat and cold in daily analyses were brought forward by at least 1 y. Estimates were broadly robust to alternative model assumptions.
    Conclusions:
    These results provide strong evidence that most deaths associated in daily analyses with heat and cold are displaced by at least 1 y. https://doi.org/10.1289/EHP1756
  • Received: 10 February 2017
    Revised: 25 September 2017
    Accepted: 26 September 2017
    Published: 27 October 2017

    Address correspondence to B. Armstrong, London School of Hygiene and Tropical Medicine, Keppel St., London WC1E 7HT, UK. Telephone: 0044 (0)20 79272232. Email: ben.armstrong@lshtm.ac.uk

    Supplemental Material is available online (https://doi.org/10.1289/EHP1756).

    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 (276 KB)
    Zip icon Supplemental Code and Data Zip File (9 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 (102 KB)

Introduction

In many locations worldwide, daily time-series analyses have identified adverse associations between daily mortality and temperatures that are toward the hot and cold ends of the temperature range for each location (Gasparrini et al. 2015; Guo et al. 2014). Many of these analyses employed distributed lag models (Armstrong 2006; Schwartz 2000) that discount mortality displacement or harvesting (Zanobetti et al. 2000) within the lag interval, typically 2–3 wk in recent studies. However, they cannot identify the extent of displacement of mortality beyond this period. Such displacement may be quite short, for example, just a few months. The uncertainty about this makes it hard to estimate the actual life lost to weather and thereby judge the priority that should be given to public health policies to reduce those excesses.

One way to clarify the extent of mortality displacement due to the acute effects of heat and cold is to estimate associations between annual mortality and annual summaries of heat and cold. If such associations were null, it would indicate that excess deaths due to acute effects were wholly compensated by deficits later in the same year, as would be the case if all reflected mortality displacement. Positive associations of annual series would indicate that all or some of the excess deaths were displaced at least into the next period, suggesting that mortality was advanced by at least 1 y.

Two published single-city studies have used this method. In both London between 1947 and 2006 (Rehill et al. 2015) and Hong Kong from 1976 to 2012 (Goggins et al. 2015), years with cold winters and those with hot summers were associated with high mortality, though in temperate London, the association with heat was not significant. Those studies were not linked to daily studies, so it was not possible to directly estimate from them whether a proportion of heat and cold deaths found in daily studies were nevertheless displaced by less than 1 y.

In this multicountry study, we estimated the association between annual mortality and annual summaries of heat and cold in 278 locations from 12 countries. We also develop the method further to give insight, for positive associations, into whether the magnitude of association indicates partial or total absence of short-term displacement. This is easily the largest study to date addressing this question, indeed the only study including more than one location.

Methods

The data were assembled from daily counts of deaths and mean temperatures from 278 locations from 12 countries of total durations of 10–40 y from years between 1972 and 2012. These data are those that have been previously described (Gasparrini et al. 2015), with the addition of six locations from Ireland (all the island of Ireland: two regions in Northern Ireland and four in the Republic). We also used a longer series from some countries, Japan in particular, and excluded locations with less than 10 y of qualifying data (see below). Descriptive data on included locations are summarized in Table 1 and given in more detail in Supplementary Material (Table S1).

Table 1. Summary descriptive statistics by country.
Country n Period MMT (percentile) Mean (SD) AF% Mean (SD) °C
Heat Cold Heat Cold
Australia 3 1988–2008 22.8 (82%) 0.5 (0.2) 5.9 (0.5) 0.4 (0.1) 4.8 (0.3)
Brazil 15 1997–2010 25.1 (62%) 0.7 (0.3) 2.5 (0.5) 0.5 (0.1) 1.7 (0.2)
Canada 21 1986–2010 17.2 (82%) 0.5 (0.2) 5.1 (0.3) 0.5 (0.2) 11.1 (0.8)
Ireland 6 1984–2006 17.5 (97%) 0.0 (0.0) 11.3 (1.1) 0.0 (0.0) 7.8 (0.5)
Italy 2 1988–2009 21.8 (78%) 1.7 (1.4) 9.8 (0.9) 0.7 (0.3) 7.5 (0.5)
Japan 47 1972–2011 24.8 (84%) 0.4 (0.2) 10.5 (0.8) 0.4 (0.2) 10.1 (0.6)
South Korea 7 1992–2009 25.6 (90%) 0.3 (0.3) 6.6 (0.2) 0.2 (0.1) 12.0 (0.5)
Spain 50 1990–2009 21.4 (78%) 1.1 (0.5) 5.6 (0.7) 0.6 (0.2) 6.6 (0.5)
Taiwan 3 1994–2006 25.8 (55%) 0.9 (0.2) 3.8 (0.6) 1.1 (0.1) 2.9 (0.3)
United Kingdom 10 1993–2005 17.2 (90%) 0.3 (0.2) 8.4 (0.8) 0.2 (0.1) 6.9 (0.4)
United States 114 1985–2005 24.3 (83%) 0.3 (0.2) 5.4 (0.3) 0.4 (0.1) 9.8 (0.6)

Note: The table shows number n of locations, period of study (start years of first to last complete Nov–Oct year), and the mean over locations of MMT from daily analyses (in degrees and as percentile), mean, and SD over included years of annual attributable fraction of deaths estimated due to heat and cold using methods identical those for a previously published daily analysis (Gasparrini 2015), and mean (SD) of annual mean degrees above and below the MMT. AF, attributable fraction; MMT, minimum mortality temperature; SD, standard deviation.

Broadly, statistical methods follow those of the previous single-location annual studies (Goggins et al. 2015; Rehill et al. 2015). Mortality and temperature daily series were collapsed to an annual series using two different year dividers: a) starting years at the beginning of summer (the month before that including the summer solstice, May, in the northern hemisphere) to allow maximum time for shorter-term displacement of heat-related deaths to be discounted; b) similarly, starting years at the beginning of winter (November) for optimal cold analysis. Location years with more than 1% of days missing key variables were excluded.

In these annual series, (logged) annual mortality was regressed on annual indices of heat and cold. We used two indices described below; the first (“Basic Model”) is simpler and we hope easier to follow, and the second (“Modified Model”) more informative at the expense of some additional complexity.

Basic Model

In a preliminary step, conventional daily analysis was used to identify minimum mortality temperature (MMT), the daily mean temperature at which mortality was predicted to be minimum, at each location. Specifically, we used exactly the same model specification as Gasparrini et al. (2015), a distributed lag nonlinear model (lags to 21 d) with spline control for seasonality and trend [8 degrees of freedom (df)/year] and day of week indicators.

Then, after collapsing daily to annual data as described above, in each country, location i and year t, mean daily deaths Yit were regressed on mean of degrees above (below) these MMTis in that year and location:
log open parenthesis Y sub i t close parenthesis equals alpha sub i plus beta sub heat, heat underscore degrees sub i, t plus beta sub cold, cold underscore degrees sub i,t plus s sub i open parenthesis t,df sub i close parenthesis plus omega sub t plus epsilon sub i t

Where

  • heat_degreesi,t=∑d=1,365{max[(tempi,t,d−MMTi),0]}/365, with tempi,t,d being mean temperature on day d, year t, location i.
  • cold_degreesi,t=∑d=1,365{max[(MMTi−tempi,t,d),0]}/365.
  • si(t,dfi) is a natural cubic spline of year t specific to location i with two df per decade (rounded).
  • ωt and εit are Gaussian independent and identically distributed noise at year and location–year levels, respectively.

In model (1), a Gaussian distribution was adopted in preference to the quasi-Poisson models used in daily analysis for modeling simplicity and because the large size of annual death counts made the Gaussian approximation very good (Goggins et al. 2015; Rehill et al. 2015). For presentation of results, we transform the coefficients βheat and βcold to percent excess relative risks (ERR%) per degree of annual high and low temperatures: ERR%=100[exp(β)−1]

In these models, R code and data for one country are given in Supplemental Material, R code and data.

That model (1) is at country rather than location level is unusual, and indeed, we initially planned to fit a regression model of annual mortality model separately for each location, as did the previous single-location studies, and then combine results in a second-stage meta-analysis of location coefficients. However, exploratory analyses showed appreciable correlations of mortality and temperature anomalies across locations in the same country; years with anomalous mortality residuals in one location tended to be similarly anomalous in other locations also, even after allowing for temperature effects. This falsified the assumption of independence of estimates over locations necessary for conventional two-stage analysis with location-specific analyses at first level, leading to spurious precision in coefficients. We therefore reformulated the model for our analysis to have first-level regressions using all data from each country combined and a second-level meta-analysis of country (heat and cold) coefficients. The country-level regressions allowed separate intercepts and time spline curves for each location and an additional random effect of calendar year to allow for the shared year anomalies in mortality noted above.

Modified Model to Relate Results to Those from Daily Analyses

Model (1) allows concluding whether some deaths were displaced by more than 1 y by heat or cold (if the coefficients of the annual heat_degrees and cold_degrees are positive), but does not easily provide evidence whether such longer-displaced deaths comprised all or just a fraction of the deaths found attributable to heat and cold in daily analyses. To provide this more quantitative evidence, we replaced the heat degrees and cold degrees variables in model (1) with annual fraction of deaths attributable to heat and cold, heat_AFdailyi,t and cold_AFdailyi,t, where AF stands for attributable fraction. These were estimated in the preliminary daily analyses by aggregating to annual level daily fractions of deaths attributable to heat and cold obtained as described in Gasparrini (2015), using the forward method described in Gasparrini and Leone (2014). The model was in other respects the same as model (1):
log open parenthesis Y sub i t close parenthesis equals alpha sub i plus beta sub heat, heat underscore AF daily sub i,t plus beta sub cold, cold underscore AF daily sub i,t plus s sub i open parenthesis t,df sub i close parenthesis plus omega sub t plus epsilon sub i t

Under model (2), if in each year t, the excess fraction of deaths due to heat (cold) found in annual regression analyses are the same as the fraction found by aggregating the daily attributable deaths, then the coefficients βheat and βcold would, to a close approximation, be 1. The approximation is addressed with a complexity cost in a sensitivity analysis, but we motivate the argument with a simplified example: Suppose in one location in just 1 y, daily analyses indicated that 2% of deaths were due to heat (none to cold), so heat_AFdailyi,t=0.02. Then, if all of those deaths were displaced by 1 y or more, mortality would be raised by a factor (rate ratio) of close to 1.02, and log mortality be 0.02 higher. To fit this, the coefficient βheat=1.

Thus, we can interpret the actual coefficients estimated from model (2) as follows: A coefficient of 0 is consistent with all deaths attributed to heat by daily analyses being displaced by less than 1 y. A coefficient of 1 is consistent with all the daily attributable deaths being displaced more than 1 y. Intermediate values are consistent with some but not all the daily attributable deaths being displaced by more than 1 y. Effects of heat and cold at longer lags than those considered in daily analyses may also play a part. If all the daily attributable deaths were displaced by more than 1 y, such additional longer lags would lead to coefficients >1. If there was some short-term displacement, this might be compensated by longer lag effects, so, for example, a coefficient of 1 might reflect the two things balancing exactly. However, public health importance seems independent of whether a coefficient of 1, say, is due to absence of short-term displacement or its compensation by longer lag effects.

Sensitivity Analyses

To identify sensitivity to specific features of the model (2) it was modified as follows:

  1. Time spline changed from 2 to 1 and 3 df/decade.
  2. The benchmark for excluding years dues to missing values changed from 1% to 0% (excluded 9 locations entirely) and 10% (added 18 locations).
  3. Influenza data were available only for locations in Canada, Ireland, Japan, Spain, and the United Kingdom (137 locations). For these countries, a term λ(influenza%)it was added, where influenza% is the proportion of deaths coded as due to influenza (International Classification of Diseases Eighth Revision (ICD-8) 470-474, Ninth Revision (ICD-9) code 487, and Tenth Revision (ICD-10) (WHO 1966, 1978, 2016)]. Unadjusted results restricted to the same five countries are also given for comparison.
  4. Step changes in death rates at calendar-year changes (1 January) were included wherever significant at a) p<0.01, and b) p<0.001. These corresponded to false discovery rates of included steps of about 50% and 20%, respectively (estimated by Storey’s q-value). Steps were selected first in calendar year analyses, modeled as indicators step(T)it=1 if t<T, 0 otherwise. In the main models with years starting in May and November, the indicators were modified to step(T)it=1 if t<T, step(T)it=4/12(May) or 10/12 (November), 0 otherwise.
  5. Three alternative approaches to clustering by year were undertaken: a) ignoring clustering by applying a conventional two-stage analysis with each location analyzed separately; b) using the country-level model as for the main model, but without random year effect with standard error of heat and cold index coefficients estimated by jackknife clustering on year; and c) the main model with addition of allowance for random variation in heat and cold coefficients (slopes) across locations.
  6. To allow interpretation of coefficients with respect to proportion of deaths attributable to heat and cold in daily analyses without approximation, the explanatory variables in model (2) were modified to heat_indexi,t=−log(1–heat_AFdailyi,t), and equivalently for cold. This we deduced from standard theory (Steenland and Armstrong 2006), by which for an explanatory variable value x for 1 y, the AF is: AF(x)=1−exp(−βx). Changing the subject of this expression, the value x=−log[1−AF(x)] gives the x variable that would give a coefficient of 1 if AF(x) were known. Thus, if heat_AFdailyi,t indeed were the same as the AF from annual analysis, heat_AFannuali,t, then the coefficient of x=−log(1–heat_AFdailyi,t) would be 1. We did not anticipate replacing AF by −log(1−AF) would make much difference to results, since at the low values for AF in these data (maximum 0.08 for heat, 0.2 for cold), AF≈−log(1−AF) to a good approximation.
  7. An approximate correction was made for estimation uncertainty in the annual AFdaily used as explanatory variables in our main analysis. Because of the complexity of the structure of this error in variables (highly correlated across years within a location), we chose the SIMEX method (Carroll et al. 2006) in which additional error is simulated at various magnitudes and the pattern of mean ERRs and their standard errors estimated from (100) simulations at each magnitude extrapolated to estimate what would have been observed at zero error if the pattern persisted. Specifically, we took values of additional error variance multiplier λ=0.5, 1, 1.5, and 2, and used a quadratic polynomial to extrapolate the association between mean of estimated β values (βλ) with λ back to λ=−1 to predict a value βSIMEX. To obtain SE(βSIMEX), we similarly extrapolated variance differences: mean[Varjλj)]−Var(βλj) to λ=−1, where j is the simulation number.

Results

A summary of distributions of key variables in the 278 included locations is shown by country in Table 1, and the values of those variables for each location are in Table S1. The United States, Japan, and Spain stand out as having the most locations, and Japan had easily the longest series (40 y). In some countries, the average location-specific variation over years of our two heat indices (fraction of deaths attributable to heat and degrees above MMT) is rather small, which limits study precision and power in those places. Ireland in particular, with mean MMT at the 97th percentile of temperature, showed an average heat-attributable fraction of just 0.03%, with standard deviation of 0.05%.

The association of mortality with mean annual degrees of heat and cold was positive on average and in most countries (Figure 1), with overall estimates of ERR per degree of heat above MMT 1.7% [(95% confidence interval (CI): 0.3, 3.1] and of cold below MMT 1.1% (95% CI: 0.6, 1.6), though variation between countries in ERRs was appreciable (I2=67% and 72%). This indicates that some deaths associated with heat and cold were indeed brought forward by at least 1 y.

Two forest plots, one for heat and one for cold, show excess relative risks in percentage along with their 95 percent confidence intervals for countries, namely, Australia, Brazil, Canada, Ireland, Italy, Japan, South Korea, Spain, Taiwan, UK, and USA. For heat, the pooled mean excess relative risk per degree was 1.7%, with I squared equal to 67 percent and p heterogeneity equals 0.00081; and for cold, the pooled mean excess relative risk per degree was 1.1%, with I squared equal to 72 percent and p heterogeneity equals 0.00011. The country-specific excess relative risks ranged from negative 1.1 to 16.5 for both heat and cold.

Figure 1. Association of annual mortality with mean annual degrees above and below minimum mortality temperature (MMT) by daily analysis (model 1): (A) Heat, (B) Cold. Percent excess relative risks ERR% are increments in RRs (%) per 1°C increase in annual mean degrees above/below MMT.

An ERR% of 1.0% has no special significance in the results in Figure 1 [model (1)], which cannot elucidate what proportion the excesses found comprise of all deaths found attributable to heat and cold in daily analyses. The results in Figure 2 from model (2) do this; if all the deaths found attributable to heat and cold in daily analyses were displaced by at least 1 y, then the regression coefficient β would be 1. For both heat and cold, our pooled annual analysis coefficients were indeed close to 1, at 0.8 (95% CI: 0.2, 1.3) and 1.1 (95% CI: 0.9, 1.4), respectively. This indicates that about the same numbers of deaths were found associated with these weather conditions in annual as in daily analyses, evidence that most of the latter were brought forward (displaced) by at least 1 y (not just harvested).

Two forest plots, one for heat and one for cold, show beta values with 95 percent confidence intervals for countries, namely, Australia, Brazil, Canada, Ireland, Italy, Japan, South Korea, Spain, Taiwan, UK, and USA. For heat, the pooled mean beta was 0.8, with I squared equal to 60 percent and p heterogeneity equals 0.0052; and for cold, the pooled mean beta was 1.1, with I squared equal to 28 percent and p heterogeneity equals 0.17. The country-specific betas ranged from negative 0.6 to 2.6 for both heat and cold.

Figure 2. Association of annual mortality with mean annual deaths attributed to heat and cold by daily analysis (model 2). (A) Heat, (B) Cold. Betas are regression coefficients for log(mortality) on fraction of deaths attributable to heat and cold in daily analyses; a value of 1.0 indicates exactly the deaths expected from daily analyses if all such deaths were displaced beyond the year end.

For cold, the country-specific associations in Figure 2 varied no more than expected by chance. There was, however, appreciable heterogeneity of heat coefficients across countries (I-sq=60%), despite these being less precise than for cold. Most of the results that were not clearly positive were very imprecise, but for the United States and Canada, the incremental risks, 0 (95% CI: −0.5, 0.5) and −0.6 (95% CI: −1.8, 0.6), had CIs excluding 1.0, thus providing some evidence that in these countries, some deaths shown in daily analyses to be associated with heat were displaced by less than 1 y.

Few of the alternative models explored in sensitivity analyses changed the overall estimates appreciably (Figures 3 and 4). Models ignoring the year clustering across locations within countries (“Alt. app. to year clustering” first row) had much narrower confidence intervals for country-specific coefficients, which is reflected in the high heterogeneity I-sq values. Conversely, those using a jackknife to allow for clustering (second row) had wider country-level intervals leading to low I-sq values. Allowing for error in the AFdaily values by SIMEX (“Correction for AF est. error”) moved estimates away from the null and slightly widened CIs. Here, overall estimates increased by multiples of 1.24 and 1.13 for heat and cold, respectively, and patterns by country remained the same (Figure S1). Investigation of sensitivity to the choice of month marking the start of years for annual aggregation (May and November in main analyses) showed little such variation for cold and somewhat more for heat, with November and February start months showing higher estimates, though all CIs overlapped considerably (Figure S2; not on Figures 3 and 4).

Figure 3A is a forest plot for the heat-mortality association showing beta values along with 95 percent confidence intervals for various models. The pooled mean betas range from 0.58 to 1.03. Figure 3B is a forest plot for the cold-mortality association indicating beta values along with 95 percent confidence intervals for various models.

Figure 3. Sensitivity of overall heat–mortality association (baseline model 2) to changing model features. Alt. time spline: Degrees of freedom (df)/decade changed from 2 df/decade (baseline) to 1 or 3 df/decade. Alt. max. prop. missing: criterion for excluding years with missing values changed from 1% (baseline) to 0% (9 locations excluded) or 10% (18 locations added.) Flu adjustment: adjusted for the proportion of deaths with influenza as the cause of death; data available only for locations in Canada, Ireland, Japan, Spain, and the United Kingdom (137 locations); unadjusted estimate restricted to the same countries for comparison. Allow for steps: Steps (breaks) were allowed for by including any single-step indicator variables significant at each of the two stated levels, which were estimated to result in false discovery rates of 0.5 (p<0.01; 109 steps) and 0.2 (p<0.001; 24 steps). Alt. app. to year clustering: The estimate ignoring clustering (“ignore”) was a conventional analysis with each location analyzed separately; the jackknife estimate was from the country-level model without random year effect, but standard error estimated by jackknife clustering on year; the “randslopes” estimate (baseline model) allowed for random variation in coefficients for heat and cold across locations. Alt. AFdaily x–var: Explanatory variables modified to heat_indexi,t=−log(1−heat_AFdailyi,t)and cold_indexi,t=−log(1−cold_AFdailyi,t), respectively, to allow interpretation of coefficients with respect to proportion of deaths attributable to heat and cold in daily analyses without approximation. Correction for AF estimation error: SIMEX correction made for error in AFdailyi,t values.

Figure 4 is a forest plot for the cold-mortality association showing beta values along with 95 percent confidence intervals for various models. The pooled mean betas range from 0.94 to 1.30. Figure 3B is a forest plot for the cold-mortality association indicating beta values along with 95 percent confidence intervals for various models.

Figure 4. Sensitivity of overall cold–mortality association (baseline model 2) to changing model features. Alt. time spline: Degrees of freedom (df)/decade changed from 2 df/decade (baseline) to 1 or 3 df/decade. Alt. max. prop. missing: criterion for excluding years with missing values changed from 1% (baseline) to 0% (9 locations excluded) or 10% (18 locations added.) Flu adjustment: adjusted for the proportion of deaths with influenza as the cause of death; data available only for locations in Canada, Ireland, Japan, Spain, and the United Kingdom (137 locations); unadjusted estimate restricted to the same countries for comparison. Allow for steps: Steps (breaks) were allowed for by including any single-step indicator variables significant at each of the two stated levels, which were estimated to result in false discovery rates of 0.5 (p<0.01; 109 steps) and 0.2 (p<0.001; 24 steps). Alt. app. to year clustering: The estimate ignoring clustering (“ignore”) was a conventional analysis with each location analyzed separately; the jackknife estimate was from the country-level model without random year effect, but standard error estimated by jackknife clustering on year; the “randslopes” estimate (baseline model) allowed for random variation in coefficients for heat and cold across locations. Alt. AFdaily x–var: Explanatory variables modified to heat_indexi,t=−log(1−heat_AFdailyi,t) and cold_indexi,t=−log(1−cold_AFdailyi,t), respectively, to allow interpretation of coefficients with respect to proportion of deaths attributable to heat and cold in daily analyses without approximation. Correction for AF estimation error: SIMEX correction made for error in AFdailyi,t values.

Discussion

This study found strong evidence that, on average, over all country studies, annual mortality was associated with deviations of temperatures from normal, in particular rising in years that experienced long or severe high or low temperatures. The strength of the association was consistent with all of the deaths found associated acutely with cold in daily analyses being brought forward by at least 1 y. These deaths thus represent significant life shortening rather than short-term mortality displacement. The associations with heat in the United States and Canada were exceptions, however. These suggested that in those countries, many of the deaths attributable in daily analyses to heat (though not cold) were advanced by less than 1 y.

Our overall results are broadly in line with the two previously published studies using a broadly similar approach, each with data from a single location but much longer periods (37 and 57 y) (Goggins et al. 2015; Rehill et al. 2015). Direct quantitative comparison is difficult. Many other studies have considered the mortality-displacement issue for heat-related deaths using distributed lag models with daily data, some finding evidence of significant proportions of heat-related deaths being displaced by only a few weeks (Basu and Malig 2011; Hajat et al. 2005; Kyselý 2004), but others found no such evidence (Kysely and Kim 2009; Le Tertre et al. 2006). In recent years, many studies have considered distributed lag models with maximum lag of several weeks (Gasparrini et al. 2015), so that temperature-related excesses of deaths with very short-term displacement will be matched by deficits later on in the total lag interval, so the overall cumulative risk for that interval will not include the deaths displaced within it. However, it is difficult for such studies to exclude displacement, which, though short, is more than a month or two.

Some studies have considered patterns of deaths after heat waves to elucidate displacement. From considering mortality for the 50 d following the 1995 Chicago heat wave, it was estimated that 26% of excess deaths during the heat wave were displaced by less than 50 d (Kaiser et al. 2007). From the experience in the year following the 2003 heat wave in Paris, it was concluded that deficits of mortality in next year could explain no more than 5,000 of the 15,000 excess due to heat in 2003 (Toulemon and Barbieri 2008). It would be interesting to explore whether duration of displacement differed for such extreme temperatures and the more moderate heat and cold dominating the temperature-related excesses found by Gasparrini (Gasparrini et al. 2015).

The patterns of reduced vulnerability to heat following winters of high mortality found in several studies (Ha et al. 2011; Qiao et al. 2015; Rocklöv et al. 2009; Stafoggia et al. 2009) provide some indirect evidence that some cold-related deaths are among those who would otherwise be vulnerable to heat, so they would likely have died within the next several months if they had survived the cold. However, it seems impossible from these studies to quantify the proportion of cold-related deaths that this would affect. Our results indicated that it is small.

The robustness of our results to quite extensive sensitivity analyses adds credibility to our conclusions. A further strength of this study is the wide geographic coverage and large number of locations. However, the gain in precision we hoped to achieve by multiple replication was somewhat reduced by the correlation of annual anomalies of mortality (year clustering) across cities in the same country. This meant that most country-specific results lack the precision for detailed comparison. A notable exception is Japan, where precise estimates illustrate the value of long (40 y) series for analyses of annual data.

There were some other limitations. Because we did not have annual population data for many locations, we could not allow for population changes, as did Goggins et al. (2015), relying rather on the spline curve to allow for this and other gradual temporal changes. We also did not consistently have cause- or age-specific death counts, precluding an investigation of these potential modifiers of the associations. Although our geographic coverage was wide, it is not globally representative, in particular overwhelmingly being comprised of more developed countries.

Conclusion

This study provides new evidence that over a wide range of locations, for the deaths associated acutely with extremes of heat and cold, most lives were shortened by at least 1 y. Adverse health effects of high and low temperatures are thus confirmed as significant public health concerns and not merely short-term displacement of times of death. Future research could usefully use similar methods, preferably with longer series, to clarify how much longer than 1 y temperature-attributable deaths are displaced, and whether extent of displacement varies by factors such as age, poverty, and preexisting conditions, and how extreme were the heat and cold.

Acknowledgments

B.A. was supported for this work by the U.K. National Institute for Health Research Health Protection Research Unit in Environmental Change and Health; A.G. and F.S. were supported by a grant from the U.K. Medical Research Council (grant ID:MR/M022625/1); M.L.B. was supported by Assistance Agreement No. 83587101 awarded by the U.S. Environmental Protection Agency. Y.G. was supported by the Career Development Fellowship of Australian National Health and Medical Research Council (APP1107107). A.T. was supported by Ministry of Education (Spain) grant PRX17/00705. H.K. was supported by the Global Research Lab (#K21004000001-10A0500-00710) through the National Research Foundation of Korea. Y.L.G. is supported by NHRI-105-EMSP09 from National Health Research Institutes, Taiwan. Y.H. and M.H. were supported by the Environment Research and Technology Development Fund (S-14) of the Environmental Restoration and Conservation Agency.

References

Armstrong B. 2006. Models for the relationship between ambient temperature and daily mortality. Epidemiology 17(6):624–631, PMID: 17028505, 10.1097/01.ede.0000239732.50999.8f.

Basu R, Malig B. 2011. High ambient temperature and mortality in California: Exploring the roles of age, disease, and mortality displacement. Environ Res 111(8):1286–1292, PMID: 21981982, 10.1016/j.envres.2011.09.006.

Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu CM. 2006. Measurement Error in Nonlinear Models: A Modern Perspective. 2nd edition. Boca Raton, FL:Chapman & Hall/CRC.

Gasparrini A, Guo Y, Hashizume M, Lavigne E, Zanobetti A, Schwartz J, et al. 2015. Mortality risk attributable to high and low ambient temperature: a multicountry observational study. Lancet 386(9991):369–375, PMID: 26003380, 10.1016/S0140-6736(14)62114-0.

Gasparrini A, Leone M. 2014. Attributable risk from distributed lag models. BMC Med Res Methodol 14:55, PMID: 24758509, 10.1186/1471-2288-14-55.

Goggins WB, Yang C, Hokama T, Law LS, Chan EY. 2015. Using annual data to estimate the public health impact of extreme temperatures. Am J Epidemiol 182(1):80–87, PMID: 26009315, 10.1093/aje/kwv013.

Guo Y, Gasparrini A, Armstrong B, Li S, Tawatsupa B, Tobias A, et al. 2014. Global variation in the effects of ambient temperature on mortality: a systematic evaluation. Epidemiology 25(6):781–789, PMID: 25166878, 10.1097/EDE.0000000000000165.

Ha J, Kim H, Hajat S. 2011. Effect of previous-winter mortality on the association between summer temperature and mortality in South Korea. Environ Health Perspect 119(4):542–546, PMID: 21233056, 10.1289/ehp.1002080.

Hajat S, Armstrong BG, Gouveia N, Wilkinson P. 2005. Mortality displacement of heat-related deaths: a comparison of Delhi, São Paulo, and London. Epidemiology 16(5):613–620, PMID: 16135936, 10.1097/01.ede.0000164559.41092.2a.

Kaiser R, Le Tertre A, Schwartz J, Gotway CA, Daley WR, Rubin CH. 2007. The effect of the 1995 heat wave in chicago on all-cause and cause-specific mortality. Am J Public Health 97(suppl 1):S158–S162, PMID: 17413056, 10.2105/AJPH.2006.100081.

Kysely J, Kim J. 2009. Mortality during heat waves in South Korea, 1991 to 2005: How exceptional was the 1994 heat wave?. Clim Res 38:105–116, 10.3354/cr00775.

Kyselý J. 2004. Mortality and displaced mortality during heat waves in the Czech Republic. Int J Biometeorol 49(2):91–97, PMID: 15549422, 10.1007/s00484-004-0218-2.

Le Tertre A, Lefranc A, Eilstein D, Declercq C, Medina S, Blanchard, et al. 2006. Impact of the 2003 heatwave on all-cause mortality in 9 French cities. Epidemiology 17(1):75–79, PMID: 16357598, 10.1097/01.ede.0000187650.36636.1f.

Qiao Z, Guo Y, Yu W, Tong S. 2015. Assessment of short-and long-term mortality displacement in heat-related deaths in Brisbane, Australia, 1996–2004. Environ Health Perspect 123(8):766–772, PMID: 25794410, 10.1289/ehp.1307606.

Rehill N, Armstrong B, Wilkinson P. 2015. Clarifying life lost due to cold and heat: a new approach using annual time series. BMJ Open 5(4):e005640, PMID: 25877269, 10.1136/bmjopen-2014-005640.

Rocklöv J, Forsberg B, Meister K. 2009. Winter mortality modifies the heat-mortality association the following summer. Eur Respir J 33(2):245–251, PMID: 18799511, 10.1183/09031936.00037808.

Schwartz J. 2000. The distributed lag between air pollution and daily deaths. Epidemiology 11(3):320–326, PMID: 10784251, 10.1097/00001648-200005000-00016.

Stafoggia M, Forastiere F, Michelozzi P, Perucci CA. 2009. Summer temperature-related mortality: effect modification by previous winter mortality. Epidemiology 20(4):575–583, PMID: 19295435, 10.1097/EDE.0b013e31819ecdf0.

Steenland K, Armstrong B. 2006. An overview of methods for calculating the burden of disease due to specific risk factors. Epidemiology 17(5):512–519, PMID: 16804473, 10.1097/01.ede.0000229155.05644.43.

Toulemon L, Barbieri M. 2008. The mortality impact of the august 2003 heat wave in France: investigating the “harvesting” effect and other long-term consequences. Popul Stud (Camb) 62(1):39–53, PMID: 18278672, 10.1080/00324720701804249.

WHO (World Health Organization). 1966. International Classification of Diseases, Eighth Revision (ICD-8) 470–474, http://apps.who.int/iris/handle/10665/89197 [accessed 1 October 2016].

WHO. 1978. International Classification of Diseases, Ninth Revision (ICD-9) code 487, http://apps.who.int/iris/handle/10665/39473 [accessed 1 October 2016].

WHO. 2016. International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) WHO version for 2016. Chapter X; codes J9-J11, http://apps.who.int/classifications/icd10/browse/2016/en#/X [accessed 1 October 2016].

Zanobetti A, Wand M, Schwartz J, Ryan LM. 2000. Generalized additive distributed lag models: quantifying mortality displacement. Biostatistics 1(3):279–292, PMID: 12933509, 10.1093/biostatistics/1.3.279.

Examining Lead Exposures in California through State-Issued Health Alerts for Food Contamination and an Exposure-Based Candy Testing Program

Author Affiliations open

1Division of Infectious Disease Epidemiology, Department of Epidemiology and Biostatistics, University of California, San Francisco, San Francisco, California, USA

2UCSF Center for Vulnerable Populations at Zuckerberg San Francisco General Hospital, Division of General Internal Medicine, Department of Medicine, San Francisco General Hospital–UCSF School of Medicine, San Francisco, California, USA

3Family Medicine Residency Program, Natividad Medical Center, University of California, San Francisco, San Francisco, California, USA

4Food Safety Section, Food and Drug Branch, California Department of Public Health, Sacramento, California, USA

PDF icon PDF Version (292 KB)

  • Summary:
    In California, the annual number of children under age 6 y of age with blood lead levels (BLL) ≥10 μg/dL is estimated at over 1,000 cases, and up to 10,000 cases when BLL between 4.5 and 9.5 μg/dL are included. State-issued health alerts for food contamination provide one strategy for tracking sources of food-related lead exposures. As well, California passed legislation in 2006 for the Food and Drug Branch (FDB) of the state health department to test and identify lead in candy. This report presents health alert data from California over a 14-y period, compares data before and after the candy testing program began, and examines country of origin, ZIP code data, and time from candy testing to release of health alerts for lead-contaminated candies for 2011–2012. After 2007, health alerts issued for lead in candy and food increased significantly. Analysis of candy-testing data indicated that multiple counties and ZIP codes were affected. Seventeen candies with high lead concentrations were identified, resulting in rapid dissemination (<2 wk) of health alerts to local health departments and community clinicians and to the public. Surveillance of lead exposures from state-based food and candy testing programs provides an opportunity to identify and immediately act to remove nonpaint sources of lead affecting children. https://doi.org/10.1289/EHP2582
  • Received: 28 July 2017
    Revised: 6 September 2017
    Accepted: 7 September 2017
    Published: 26 October 2017

    Address correspondence to M.A. Handley, Center for Vulnerable Populations, Department of Medicine, Division of General Internal Medicine, 1001 Potrero Ave., Building 13, San Francisco General Hospital–University of California, San Francisco, San Francisco, CA 94110 USA. Telephone: (415) 206-7366. Email: margaret.handley@ucsf.edu

    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

In 2011, California ranked fifth in the United States for the number of children under 5 y of age with blood lead levels (BLLs) ≥10 μg/dL, with approximately 950 cases a year between 2007 and 2011 (Raymond and Brown 2015). In 2012, more than 11,000 cases of elevated lead levels (ELLs), defined as levels >4.5 μg/dL, were reported for children under 6 y of age in California (CLPPB 2017).

Lead poisoning prevention in the United States has focused on the removal of lead from industrial sources, including paint and gasoline. Other sources of lead, such as tainted tap water and contaminated foods, warrant more thorough assessment. Case reports from local and state lead poisoning prevention programs suggest that up to 30% of pediatric lead poisoning cases investigated (with cases defined as BLLs ≥10 μg/dL), did not identify an immediate lead paint hazard (Brown and Margolis 2012). Many childhood lead poisoning case investigations have identified nonpaint related exposures, including tap water, food, candy, home remedy products, take-home contamination from workplace exposures, or hobby-related contamination (Brown and Margolis 2012; FDA 2006).

Field investigation data also suggest that nonpaint lead exposures are often insufficiently characterized and their importance thus underestimated. For example, a 2012 report from the Centers for Disease Control and Prevention (CDC) indicated that lead program inspections primarily focus on looking for lead paint hazards in the physical structures where children with BLLs ≥10 μg/dL spend time, and that nonpaint sources are sought only when no paint hazards are found (Brown and Margolis 2012).

In this Brief Communication we report the benefits of active surveillance of health alerts in identifying nonpaint sources of lead and preventing future lead exposure in children. In particular, we focused on health alerts stemming from a program of the Food and Drug Branch (FDB) of the California Department of Public Health (CDPH), in which a specific food type, candy, is tested for lead contamination.

Discussion

Background

The current permissible tolerable lead level in food likely to be consumed by small children (such as candy) was lowered by the U.S. Food and Drug Administration (FDA) to 0.10 parts per million (ppm)following several case reports of lead-contaminated candy resulting from field investigations across the United States (FDA 2006). In California alone, several reports have been published about lead poisoning cases associated with candy and food. Among these were a case-series investigation into childhood lead poisoning associated with imported candy (CDC 2002); an outbreak investigation that identified food-related sources of lead exposure among pregnant women and children in Seaside, Monterey County (Handley et al. 2007); and a series of investigative reports focusing on candy imported from other countries, especially Mexico (Orange County Register 2004). In a case investigated in Seaside, some food and candy samples tested had lead levels as high as 2,300 ppm (Handley et al. 2007; CDHS 2003), and case investigations in Orange and Stanislaus counties identified candy with over 21,000 ppm (CDHS 2001).

In a 2002 report from the California Childhood Lead Poisoning and Prevention program, the authors wrote that candy produced in Mexico was identified as a possible exposure source in approximately 15% of about 1,000 cases of (which was at the time defined as ≥10 μg/dL) that were reported to the California Department of Health Services over a period of 9 mo (CDC 2002). The number and severity of many of these reports in California resulted in legislation being passed in 2006 requiring increased surveillance of lead in candy and public reporting. Currently, the state FDB is responsible for collection of candy samples, and the Food and Drug Lab Branch conducts testing. Extensive candy testing began in 2007.

As in many other states, health officials in the CDPH and the FDB prepare and disseminate health alerts to regional and county public health programs, practicing community clinicians, and the general public. In the context of food, health alerts warn of potential toxic food exposures identified by public health and food safety program, including recalls, seafood-related quarantines, and episodes of contamination that have potential widespread public health impact. California also disseminates health alerts across the state to warn health care providers and the public health community about potential lead hazards. Health alerts related to pediatric lead poisoning cases can provide information about nonindustrial lead exposures such as candy and other foods. The CDPH issues health alerts through its newsroom to notify the public about emerging health threats and to encourage voluntary recalls of contaminated products.

Data Sources

We obtained CDPH health alerts from the department’s website for the years 2007 to 2014, and from the California Department of Health Services newsroom for alerts issued between January 2001 and June 2007. Health alerts issued between 2001 and 2014 were reviewed by four independent reviewers (M.A.H., K.N., C.C., E.S.) to determine whether a) health alerts were considered warnings, which focus on higher-level concerns, compared with less serious advisories and announcements also included in news releases, and b) health alerts involved food contamination. For the purposes of this report, we excluded warnings that focused on fishing restrictions—including shellfish quarantines—and on harvesting wild mushrooms, as well as warnings that focused on products identified as home remedies or alternative medicines.

Warnings about food contamination were then data-abstracted into a spreadsheet and coded by the biological or chemical name of the contaminant; the year; the food product contaminated; country of origin for the food; and, for alerts involving lead, the level reported in parts per million or micrograms per serving (ppm or μg/serving). Using criteria established by the CDC (Gould et al. 2013), foods were assigned to one of 17 commodity groups: fish, crustaceans, mollusks, dairy, eggs, beef, game, pork, poultry, grains/beans, oils/sugars, fruits/nuts, fungi, leafy vegetables, root vegetables, sprouts, and vegetables from a vine or stalk. Additionally, semi-prepared foods, such as hummus, dips, spreads, and cookie dough, were classified as prepared foods. Herbal teas or supplements, spices, and drinks were each given a separate category.

We obtained data on candies tested in 2011 and in 2012 from the CDPH FDB, Food Safety Section (P. Kennelly, written communication, May 2014). We reviewed lab slip requisition forms, abstracted data on country of origin, lead levels detected, sample submission dates, and ZIP codes for sites where samples were collected. We focused on two years, 2011 and 2012, the most recent years for which there were available candy data at the time of the analysis. For candies that were tested multiple times, only one observation was selected and the highest tested lead level included.

The sampling strategy for candies attempted to include a wide range of retail stores and candy distributors across the state. The data were reviewed to a) determine how many of the candies tested had high levels of lead exceeding the established limit of 0.10 ppm; b) determine how many individual candies that were tested subsequently resulted in a health alert being issued, as well as the time frame from testing to health alert release; c) describe the levels of lead among those candies with amounts exceeding the FDA limit; d) identify the country of origin of lead-contaminated candies; and e) determine which ZIP codes and counties the contaminated candies included in the sample represented.

Findings

A total of 164 health alerts were issued for food contamination in California between 2001 and 2014 [California Department of Health Services, written communication, June–July 2012; California Department of Health Services online archive (archive taken off line 6 October 2017; accessed 15 June 2015)]. Of these, the largest percentage (36.6%) was issued for lead contamination in foods (Figure 1).

Bar graph showing number of health alerts (y-axis) across following contaminants (x-axis): lead, salmonella, E. Coli, Botulism, Listeria, prescription drugs or antibiotics, bacteria, inorganic, shigella, cryptosporidium, Bacillus Cereus, melamine, and others.

Figure 1. Food-related health alerts issued in California by contaminant type, 2001–2014 (n=164).

Of the 60 lead-related health alerts issued over this period, 55 (91.6%) were for imported foods, and the remaining 5 were for food products manufactured in the United States. Almost all of the health alerts for lead-contaminated imported foods (96.3%) were for candy products. Of the two noncandy related health alerts focusing on imported food products, one was for chapuline (a toasted grasshopper snack food), and one was for spices/herbs. Lead-contaminated imported candies came primarily from Mexico (34%), China (24%), and India (20%).

There was a substantial increase in the total number of health alerts issued after 2006, primarily for lead in candy. Between 2001 and 2007, before the candy testing program was widespread, 22% of the 48 total health alerts issued related to lead contamination. In the 7 y after the candy testing program was in effect, between 2008 and 2014, lead-related health alerts made up approximately 42% of the total of 116 issued (p<0.05).

Over the time period 2011–2012, 1,346 candies were tested for lead, and 65 unique products were identified as having lead levels >0.05 ppm (4.8%). Close to two-thirds (n=40) had lead levels ≥0.10 ppm; the highest level of lead was 2.4 ppm. These test results prompted 17 health alerts for lead-contaminated candies.

The median length of time between the date of the test results and the date of health alert issuance was 6 d. All but one was issued within 2 wk of the time that the test results were reported. The most frequent countries of origin for the 65 candies identified over the 2-y period were India (35%), Taiwan (12%), China (11%), the United States (11%), Mexico (9%), Pakistan (6%), Hong Kong (4%), United Kingdom (3%), with 1 candy each for Germany, Indonesia, Thailand, Turkey, and Spain. ZIP codes were not identified for 6 of the 65 candies (9.2%). The ZIP codes for the 17 candies for which health alerts were issued represented 11 counties across the state.

Conclusions

Based on these observations, lead-contaminated candies represent an important contribution to lead exposures in California. The fact that a large number of unique products were identified among the contaminated candies presents an ongoing challenge for exposure-based testing programs, as does the large number of candies tested for which no lead was detected. After legislation was passed requiring more widespread testing of candies for lead, many imported candies were identified as containing lead, suggesting that imported candy might be considered a public health risk for lead poisoning in California. However, this approach does not convey the magnitude of that risk, because the sampling methods are not currently population-based. More work is needed to determine the best approaches to sampling in order to determine the magnitude of the problem.

We do not know whether other food sources (besides candy) that are not tested through the candy testing program, or candies not included in the sampling, also contribute to lead exposures in California. Furthermore, these data do not represent prevalence estimates of lead risks at the county and ZIP code level. Nevertheless, the candy testing program does provide a means of proactively detecting sources of lead, and the health alerts may help prevent lead poisoning cases in vulnerable populations, particularly children.

We know that consumption of contaminated foods, such as lead-contaminated candies, can immediately result in ELLs, especially in children. A recent study of lead in candy in Mexico identified a significant association between the previous week’s lead intake through the consumption of candy and the proportion of children with BLLs above the CDC action level of 4.5 μg/dL (Tamayo y Ortiz et al. 2016). In that study, lead levels in candy consumed the previous week ranged between 0.13 and 0.70 ppm—which are similar levels to those found in California candies, as reported above—and exceeded 0.10 ppm, the FDA reporting threshold.

Although there have been few exposure-based programs to track lead exposures in foods, we believe that the California program to test candies and rapidly translate these results into health alerts and recall programs is a useful model. The availability of new technologies to screen foods for lead, such as XRF (X-ray fluorescence) screening tools (Reames and Charlton 2013), can allow for local programs to increase exposure-based surveillance activities. These screening results can, in turn, help prioritize which food and candy samples should be sent to local and state laboratory testing programs. Multiple strategies, including the one presented in this report, along with case-based investigations, are needed to assist health departments and clinicians in practice to translate health alert data into implementation strategies and clinician practice.

Acknowledgments

The authors acknowledge funding from the National Institute on Minority Health and Health Disparities, National Institutes of Health (grant P60MD006902).

References

Brown MJ, Margolis S. 2012. Lead in drinking water and human blood lead Levels in the United States. MMWR Suppl 61(4):1–9, PMID: 22874873.

CDC (Centers for Disease Control and Prevention). 2002. Childhood lead poisoning associated with tamarind candy and folk remedies—California, 1999–2000. MMWR Morb Mortal Wkly Rep 51(31):684–686, PMID: 12233910.

CDHS (California Department of Health Services). 2001. State health department issues warning on lead-contaminated candy from Mexico. No. 30-01, Sacramento, CA:California Department of Health Services.

CDHS. 2003. State health department issues health warning on lead-contaminated chapulines (grasshoppers). No. 03-92, Sacramento, CA:California Department of Health Services.

CLPPB (Childhood Lead Poisoning Prevention Branch). 2017. California Blood Lead Data, 2012. https://www.cdph.ca.gov/Programs/CCDPHP/DEODC/CLPPB/CDPH%20Document%20Library/BLL%20Counts%202012%20by%20LHD%20final.pdf [accessed 17 January 2017].

FDA (U.S. Food and Drug Administration). 2006. “Supporting Document for Recommended Maximum Level for Lead in Candy Likely To Be Consumed Frequently by Small Children.” Docket No. 2005D-0481. https://www.fda.gov/food/foodborneillnesscontaminants/metals/ucm172050.htm [accessed 17 January 2017].

Gould LH, Walsh KA, Vieira AR, Herman K, Williams IT, Hall AJ, et al. 2013. Surveillance for foodborne disease outbreaks – United States, 1998–2008. MMWR Surveill Summ 62(2):1–34, PMID: 23804024.

Handley MA, Hall AC, Sanford ED, Diaz E, Gonzalez-Mendez E, Drace K, et al. 2007. Globalization, binational communities and imported food risks: results of an outbreak investigation of lead poisoning in Monterey County, California. Am J Public Health 97(5):900–906, PMID: 17395841, 10.2105/AJPH.2005.074138.

Orange County Register. 2004. Toxic Treats. http://www.ocregister.com/2004/04/26/toxic-treats-part-1-extras/ [accessed 11 February 2017].

Raymond J, Brown MJ. 2015. Summary of notifiable noninfectious conditions and disease 275 outbreaks: childhood blood lead levels – United States, 2007–2012. MMWR Morb Mortal Wkly Rep 62(54):76–80, PMID: 26505220, 10.15585/mmwr.mm6254a5.

Reames G, Charlton V. 2013. Lead detection in food, medicinal, and ceremonial items using a portable X-ray fluorescence (XRF) instrument. J Environ Health 75(6):16–20, PMID: 23397645.

Tamayo y Ortiz M, Téllez-Rojo MM, Hu H, Hernández-Ávila M, Wright R, Amarasiriwardena C, et al. 2016. Lead in candy consumed and blood lead levels of children living in Mexico City. Environ Res 147(2016):497–502, PMID: 26974363, 10.1016/j.envres.2016.03.007.