Healthy dietary patterns are associated with exposure to environmental chemicals in a pregnancy cohort

Healthy dietary patterns, such as the alternate Mediterranean diet and alternate Healthy Eating Index, benefit cardiometabolic health. However, several food components of these dietary patterns are primary sources of environmental chemicals. Here, using data from a racially and ethnically diverse US cohort, we show that healthy dietary pattern scores were positively associated with plasma chemical exposure in pregnancy, particularly for the alternate Mediterranean diet and alternate Healthy Eating Index with polychlorinated biphenyls and per- and poly-fluoroalkyl substances. The associations appeared stronger among Asian and Pacific Islanders. These findings suggest that optimizing the benefits of a healthy diet requires concerted regulatory efforts aimed at lowering environmental chemical exposure.

Both aHEI and aMED represent the guidelines that are more updated in time and have been shown to be consistently associated with lower risks of chronic diseases 4 .A higher value of each score indicates greater adherence to the healthy dietary pattern.aMED, adapted from the Mediterranean diet score originally proposed by Trichopoulou et al., is comprised of 8 components [vegetable, legume, fruit, nut, whole grain, red & processed meat, fish, the ratio of monounsaturated fatty acids to saturated fatty acids (MUFA: SFA ratio)] 5 .aHEI was subsequently developed based on the original Healthy Eating Index, with the use of more specific food items and the inclusion of alcohol and multivitamin use 6 .It is comprised of 10 food groups and nutrients [vegetable, whole fruit, whole grain, sugar-sweetened beverage (SSB, including juice), nut and legume, red & processed meat, trans-fat, eicosapentaenoic acid and docosahexaenoic acid (EPA+DHA), polyunsaturated fatty acids excluding eicosapentaenoic acid and docosahexaenoic acid (PUFA excluding EPA & DHA), and sodium].Fatty acids, including EPA+DHA and PUFA excluding EPA & DHA, were mainly derived from food items of FFQ without supplement use included.The aMED score is calculated by totaling the scores for each dietary component, assigning a score of 1 when the consumption amount of a food group/nutrient exceeds its corresponding median, and a score of 0 when it falls below the median.The aHEI score is calculated by assigning points to specific dietary components and then summing the score of each component.DASH was derived from the sum of 8 food group scores [fruit, vegetable, nut and legume, whole grain, low fatty dairy, sodium, red & processed meat, and SSB (excluding juice)] and each of the calculated components were scaled from 1 (least healthy) to 5 (most healthy) according to the quintiles of food consumption amounts.Several common food groups, such as vegetables, whole grain and red & processed meat, were used in the scoring of different dietary patterns.
Alcohol consumption was removed from aMED and aHEI as it remains controversial whether it should be part of a healthy diet for pregnant women 2 .
Briefly, for the analysis of PCBs, PBDEs, and OCPs, 1 ml of plasma was spiked with a 13Clabeled internal standard mixture (250 pg each for OCPs, PBDEs, and PCBs), vortexed, and refrigerated overnight.Subsequently, 1 ml of 88% formic acid was added, and the mixture was sonicated for 15 minutes, followed by the addition of 2 ml of Milli-Q water.The samples were then subjected to solid-phase extraction (SPE) using cartridges packed with 1.3 g of Sepra C18-E (Rapid Trace SPE Workstation) and eluted with dichloromethane, which was then concentrated to 1 ml.The extracts were further purified using SPE cartridges packed with 0.2 g of silica gel/1.1 g of sulfuric acid silica gel and eluted with 30% dichloromethane in hexane, then concentrated to a final volume of 50 μl under a gentle stream of nitrogen.
Quantification of PBDEs, OCPs, and PCBs was based on the isotope dilution method with 13C-labeled internal standards.Two procedural blanks and SRM1958 were analyzed for every set of 27 samples.
For PFAS quantification, briefly, 200 μl of plasma was transferred into a polypropylene (PP) tube and spiked with 13C-labeled internal standards.To this mixture, 100 μl of 10% ammonia solution (v/v) was added.After 30 minutes, 780 μl of 1% ammonium formate in methanol (w/v) was added and vortexed.The sample was then centrifuged, and the supernatant was loaded onto a Hybrid-SPE cartridge (Supleco, Bellefonte, PA).The eluent was concentrated three times under a gentle stream of nitrogen.The target analytes in the eluate were quantified using an ultra-performance liquid chromatography system (Acquity I Class; Waters, Milford, MA, US) coupled with an electrospray triple quadrupole tandem mass spectrometer (API 5500; AB SCIEX, Framingham, MA, US).Analyte separation was achieved using an Acquity UPLC BEH C18 column (1.7 μm, 50×2.1 mm, Waters).Serum cotinine was measured using ultra-performance liquid chromatography coupled with electrospray triple quadrupole tandem mass spectrometry.
Concentrations of metals were measured in blood plasma samples collected during the late first trimester of pregnancy (median: 12 weeks' gestation) and stored at −70°C pending analysis.Plasma specimens were shipped to the Wadsworth Center, New York State Department of Health, for trace element analysis using inductively coupled plasma-mass spectrometry (ICP-MS), a multi-element method optimized for serum/plasma samples and validated for use in biomonitoring studies.The ICP-MS instrument was calibrated with matrix-matched standards traceable to the National Institute of Standards and Technology (NIST).Levels of serum internal quality control materials were included in each analytical run, and 2% of all samples were analyzed in duplicate.Method validation was established using NIST standard reference materials, and method performance was assessed by successful participation in external proficiency testing programs for serum/plasma trace elements operated by the Center de Toxicologie du Québec, UK NEQAS for Trace Elements, German EQUALM, and the New York State proficiency testing program for trace elements.
All metal(loid) measurements were above the limit of detection, calculated according to the International Standards Organization / International Union of Pure and Applied Chemistry harmonized guidelines.
Concentrations of POPs and metals in all samples were expressed in ng/mL.The level of quantifications (LOQs) was adopted in the formal analysis to determine the chemical detection rate and varied by analytes between 0.0025-0.05ng/mL for OCPs, 0.0025-0.01ng/mL for PBDEs, 0.005 ng/mL for PCBs, 0.007-0.01ng/mL for PFASs and 0.01-340 ng/mL for metals.Chemicals with concentrations below LOQ were replaced with LOQ/sqrt (2).Total OCPs, total PBDEs, total PCBs, and total PFASs were calculated by summing over the detected chemical concentrations within each chemical class due to their similar food sources and environmental fate 8,10 .Metals, regardless of the route and sources of exposure and health effects, were totalled as well to reflect the general changes introduced by adherence to dietary patterns 11 .Some POPs such as OCPs, PBDEs and PCBs are lipophilic and the plasma concentrations of these chemicals might be influenced by lipids 12 .Thus, we further measured lipid concentrations using commercially available enzymatic methods 13 and calculated total lipids by the equation as follows: plasma total lipids (mg/dL) = plasma total cholesterol×2.27+plasmatriglycerides+62.3 14 .Plasma cotinine level was measured by LC-MS to confirm the status of active or passive smoking.

Covariates
Data on maternal characteristics and socio-economic status such as age, weight, height, education level, tobacco and alcohol use, physical activity level, and household income were collected using standardized questionnaires at recruitment.Parity and disease history were extracted from medical records.Pre-pregnancy body mass index (BMI, kg/m 2 ) was calculated as weight in kilograms divided by the square of height in meters.Maternal race/ethnicity was self-identified by the participants, including Hispanic, non-Hispanic White, non-Hispanic Black and Asian & Pacific Islander.Maternal education level was grouped into high school or below, some college and undergraduate, graduate and postgraduate.Parity was classified into primiparous and multiparous.Maternal income was grouped into three levels based on the income during the past year.We classified an individual's status of tobacco exposure as yes or no based on whether the plasma cotinine concentration was above the limit of detection.Alcohol consumption was excluded due to a low proportion of drinkers (0.75%).

Physical activity was measured using the validated Pregnancy Physical Activity
Questionnaire and reported in weekly physical activity level (MET-hour/w).Total energy intake (kcal) was estimated from the same FFQ administered at recruitment.

Data analysis
Distributions of women's characteristics were reported according to quartiles of the different dietary pattern scores and tested using the Chi-square test for categorical characteristics or ANOVA for continuous ones.Chemicals (except for PFASs and metals) were corrected by total lipids at recruitment to account for variation in the concentrations due to lipid solubility.
To estimate the percent difference in chemicals per SD change in dietary pattern scores, we categorized the dietary pattern scores into binary low and high groups by the median of each dietary pattern score, and then estimated the group-specific chemical concentrations.Group differences were examined by non-parametric test.
Multivariable linear regression models were used to assess the associations of individual dietary pattern scores and food components with each of the chemicals.Models were adjusted for maternal age, physical activity level, pre-pregnancy BMI, education, income, parity, total energy intake, and tobacco exposure.Both dietary pattern scores and consumption of each food group were modelled as continuous variables in the main models.Covariates included in the models were selected based on a priori evidence depicted in a causal diagram using a directed acyclic graph (DAG) (Figure S2) 9 .To aid in the interpretation of the results, beta coefficients were converted into percent difference using the following formula: (e β − 1) × 100, which represents the percent difference in plasma chemical concentrations by 1-unit SD increase in each dietary pattern score or contributing food group/nutrient.Stratified analyses for different covariates (race/ethnicity, parity, pre-pregnancy BMI) were conducted to explore potential effect modifications.To account for the possible multiple comparisons, significant levels of the P-values were adjusted by the Benjamini-Hochberg procedure in all the association analyses.
To examine the association between the dietary pattern and plasma concentrations of chemicals at a finer scale, we performed reduced rank regression analysis (RRR) to assess the contribution of each constituent food group/nutrient to the variations of individual chemicals and chemical classes, which has been generally adopted to identify key food groups/nutrients and obtain factor scores which would describe the degree of a participant's adherence to each identified dietary pattern.The workflow of the RRR analysis is summarized in the conceptual diagram (Fig 1a) 15 .For the RRR analysis, natural log-transformed chemical plasma concentrations were included and the food group/nutrient consumption was standardized by default in the model set of SAS procedure pls.As RRR does not allow the incorporation of confounding variables directly, we used the residual method of adjustment to minimize the correlation between the dietary patterns and confounding variables 16 .To calculate the residuals, we regressed each of the chemicals on confounding covariates.The covariateadjusted chemicals and all food groups/nutrients were included and modelled as dependent and independent variables, respectively.We further derived the model loadings of each constitution food group/nutrient, which represent the strength and direction of contributions to variation in chemicals.
Several additional analyses were performed to assess the robustness of our results.1) To account for the baseline difference between included and excluded participants, we calculated the population weight by inverse probability weighting and did additional analysis while incorporating weight to represent the total cohort population.2) We additionally adjusted for total lipids as previous studies showed that incorporating both covariate-adjusted standardization and the inclusion of lipids as a covariate in the regression model might have low bias and perform well 17 .3) The associations of dietary pattern scores with chemicals might be attenuated due to the relatively low concentrations of exposure in the study population.Thus, each of the chemicals was dichotomized according to the 80 th percentile (high level: ≥ 80th, common level< 80th) to examine whether the associations would remain significant.4) Since chemical concentration and dietary pattern may vary by geographical regions, we additionally adjusted for clinical centers as a random effect intercept using generalized linear mixed models.5) Given that certain nutrients, such as EPA+DHA, PUFA excluding EPA & DHA, Trans-fat, and MUFA:SFA ratio, were derived from other food groups, their inclusion in the analysis may introduce collinearity concerns.To address this, we conducted RRR analyses by excluding these nutrients.Additionally, we employed elastic network regression (ENR) models to identify key food groups/nutrients and assess the reliability and robustness of our findings.Both inclusion and exclusion of nutrients were considered in the ENR models.6) Multivariate Imputation by Chained Equations (MICE) was applied to impute chemical values below the limit of detection (LOD) by specifying the skewed distribution and setting an upper limit according to the LOQ of each chemical 18 .

Fig
Fig S1| Flowchart of the study analytical population.

Fig
Fig S4| Associations of different dietary patterns with chemicals, stratified by race/ethnicity among the NICHD Fetal Growth Study-Singletons cohort.a, Change% [ (exp(beta) − 1) × 100] was reported with different colors to benefit interpretation: the redder the grid cell, the higher the change%, while the bluer, the lower the change%.When change% is associated with raw p-value< 0.05, the numbers are shown in black.To account for multiple comparisons, change% with adjusted p-value< 0.05, < 0.01, < 0.001 are marked as * , ** , *** , respectively.b, Interaction test was examined between Asian & Pacific Islander and other races/ethnicities, with p-value< 0.05 shown in black number and adjusted p-value< 0.05, < 0.01, < 0.001 marked as * , ** , *** ,

Fig
Fig S7| Elastic net regression models for the associations between food groups and different chemical classes (β) among the NICHD Fetal Growth Study-Singletons cohort.Coefficients (β) of each food group were visualized to indicate the strength and direction of the relationship between food groups and the chemical classes.All models were adjusted for maternal age, physical activity, maternal race/ethnicity, maternal BMI, maternal educational level, household income level, parity, tobacco exposure, and total energy intake.a, all food groups (including nutrients) were included.b, food groups excluding EPA+DHA, PUFA excluding EPA & DHA, Trans-fat, and MUFA: SFA ratio were included.SSB: sugar-sweetened beverage.MUFA: monounsaturated fatty acids.SFA: saturated fatty acids.PUFA: polyunsaturated fatty acid.

Table S1 . Categories and distribution of the consumption amounts of different food groups among the NICHD Fetal Growth Study-Singletons cohort. No
supplements were included in the calculations of dietary pattern scores.aMED score was derived from 8 food groups: vegetable, legume, fruit, nut, whole grain, red & processed meat, fish, the ratio of monounsaturated fatty acids to saturated fatty acids (MUFA: SFA ratio); the

Table S2 . Characteristics of study participants at enrollment (gestational weeks 8-13, n = 1,618) according to quartiles of the aMED, aHEI, and DASH scores among the NICHD Fetal Growth Study- Singletons cohort. The
chi-square test and analysis of variance were used to examine the group differences.

Table S3 . Plasma concentration differences of chemicals according to aMED, aHEI, and DASH scores among the NICHD Fetal Growth Study-Singletons cohort.
Group differences of chemicals (High vs. Low, dichotomized by the median of each dietary pattern score) were examined by a non-parametric test.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.Chemicals (ng/g lipid, except for PFASs and metals, ng/mL) were standardized by total lipids.aHEI: alternate Healthy Eating Index; aMED: alternate Mediterranean diet; DASH: Dietary Approaches to Stop Hypertension.Chemicals were reported with median± interquartile range (IQR).DR: detection rate.LOQ: limit of quantification (ng/mL).High vs. Low, dichotomized by the median of each dietary pattern score) were examined by a non-parametric test.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.Chemicals (ng/g lipid, except for PFASs and metals, ng/mL) were standardized by total lipids.aHEI: alternate Healthy Eating Index; aMED: alternate Mediterranean diet; DASH: Dietary Approaches to Stop Hypertension.Chemicals were reported with median± interquartile range (IQR).DR: detection rate.LOQ: limit of quantification (ng/mL).High vs. Low, dichotomized by the median of each dietary pattern score) were examined by a non-parametric test.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.Chemicals (ng/g lipid, except for PFASs and metals, ng/mL) were standardized by total lipids.aHEI: alternate Healthy Eating Index; aMED: alternate Mediterranean diet; DASH: Dietary Approaches to Stop Hypertension.Chemicals were reported with median± interquartile range (IQR).DR: detection rate.LOQ: limit of quantification (ng/mL).

Table S4 . Percent difference in grouped and individual plasma chemical concentrations per 1 SD increase in dietary pattern indices of aHEI, aMED, and DASH among the NICHD Fetal Growth Study-Singletons cohort.
Multiple linear regression model was used.a, no covariate was adjusted; b, adjusted for maternal race/ethnicity, age, physical activity level, pre-pregnancy BMI, education level, income, parity, tobacco exposure, and total energy intake.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.Chemicals (except for PFASs and metals) were standardized by total lipids, and all chemicals were further log-transformed and scaled.All chemicals were log-transformed.Change% [ (exp(beta) − 1) × 100] was reported to benefit interpretation.

Table S5 . Associations of dietary pattern scores with chemicals with accounting for non-response of FFQ by inverse probability weighting among the NICHD Fetal Growth Study-Singletons cohort.
Multiplelinear regression model was applied with weight calculated and included in the model to represent the total population.Statistically significant associations were bolded with the significance of two-sided adjusted p-

Continued Table S5. Associations of dietary pattern scores with chemicals with accounting for non- response of FFQ by inverse probability weighting among the NICHD Fetal Growth Study-Singletons cohort.
Multiple linear regression model was applied with weight calculated and included in the model to represent the total population.Statistically significant associations were bolded with the significance of twosided adjusted p-values indicated by the asterisk: * : p-value<0.05;** : p-value<0.01;*** : p-value<0.001,respectively.All models were adjusted for maternal age, physical activity level, pre-pregnancy BMI, education level, income, parity, total energy intake, and tobacco exposure; All chemicals were log-transformed.Chemicals (except for PFASs and metals) were standardized by total lipids, and all chemicals were further log-transformed and scaled.Change% [ (exp(beta) − 1) × 100] was reported to benefit interpretation.

5) Continued Table S5. Associations of dietary pattern scores with chemicals with accounting for non- response of FFQ by inverse probability weighting among the NICHD Fetal Growth Study-Singletons cohort.
Multiple linear regression model was applied with weight calculated and included in the model to represent the total population.Statistically significant associations were bolded with the significance of twosided adjusted p-values indicated by the asterisk: * : p-value<0.05;** : p-value<0.01;*** : p-value<0.001,respectively.All models were adjusted for maternal age, physical activity level, pre-pregnancy BMI, education level, income, parity, total energy intake, and tobacco exposure; All chemicals were log-transformed.Chemicals (except for PFASs and metals) were standardized by total lipids, and all chemicals were further log-transformed and scaled.Change% [ (exp(beta) − 1) × 100] was reported to benefit interpretation.

Table S6 . Associations of dietary patterns with each lipophilic chemical with additional adjustment for total lipids among the NICHD Fetal Growth Study-Singletons cohort.
Multiple linear regression model was used.Two-sided p< 0.05 was bolded, with Benjamini-Hochberg (BH) adjusted p values calculated and marked as " * ", " ** ", " *** ", if adjusted p value< 0.05, < 0.01, < 0.001, respectively.All models were adjusted for total lipids (except for PFASs and metals), maternal race/ethnicity, age, physical activity level, pre-pregnancy BMI,

Continued Table S6. Associations of dietary patterns with each lipophilic chemical with additional adjustment for total lipids among the NICHD Fetal Growth Study-Singletons cohort.
Multiple linear regression model was used.Two-sided p< 0.05 was bolded, with Benjamini-Hochberg (BH) adjusted p values calculated and marked as " * ", " ** ", " *** ", if adjusted p value< 0.05, < 0.01, < 0.001, respectively.All models were adjusted for total lipids (except for PFASs and metals), maternal race/ethnicity, age, physical activity level,

Table S7 . Associations of dietary patterns with each chemical: high exposure level VS common exposure level among the NICHD Fetal Growth Study-Singletons cohort.
Binary logistics regression model was used.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.All models were adjusted for maternal race/ethnicity, age, physical activity level, prepregnancy BMI, education level, income, parity, tobacco exposure, and total energy intake; Chemicals (except for PFASs and metals) were standardized by total lipids, and all chemicals were further log-transformed and scaled.Each of the chemical was dichotomized according to the 80 th percentage (high level: ≥ 80 th , common level< 80 th ).

Table S8 . Associations of dietary patterns with each chemical and chemical class while including clinical centers as random effect intercept among the NICHD Fetal Growth Study-Singletons cohort.
GeneralizedLinear Mixed Models (GLMM) was used with clinical centers included as a random effect intercept.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All

Continued Table S8. Associations of dietary patterns with each chemical and chemical class while including clinical centers as random effect intercept among the NICHD Fetal Growth Study-Singletons cohort.
Generalized Linear Mixed Models (GLMM) was used with clinical centers included as a random effect intercept.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-

Continued Table S8. Associations of dietary patterns with each chemical and chemical class while including clinical centers as random effect intercept among the NICHD Fetal Growth Study-Singletons cohort.
Generalized Linear Mixed Models (GLMM) was used with clinical centers included as a random effect intercept.Estimations with raw p-value< 0.05 were bolded.To account for multiple comparisons, Benjamini-Hochberg (BH) adjusted p-values were calculated with p< 0.001, <0.01 and <0.05 marked as *** , ** , and * respectively.All p-values were two-sided.All models were adjusted for total lipids (except for PFASs and metals), maternal race/ethnicity, age, physical activity level, pre-pregnancy BMI, education level, income, parity, tobacco exposure, and total energy intake; All chemicals were log-transformed and scaled.change% [ (exp(beta) − 1) × 100] was reported to benefit interpretation.