key: cord-0430132-h82sygle
authors: Middleton, L. Y.; Nguyen, V. K.; Dou, J.; Park, S. K.; Colacino, J. A.; Bakulski, K. M.
title: Environmental chemical-wide associations with immune biomarkers in the US: A cross-sectional analysis
date: 2022-03-23
journal: nan
DOI: 10.1101/2022.03.22.22272789
sha: ac48ebc035d67db3940ec3e53b23e2c3d243c2c0
doc_id: 430132
cord_uid: h82sygle

Exposure to environmental chemicals influence immune system functions, and humans are exposed to a wide range of chemicals, termed the chemical exposome. Thus, a comprehensive analysis of the effects across multiple chemical families with immune biomarkers is needed. In this study, we tested the associations between environmental chemicals and immune biomarkers. We analyzed the United States cross-sectional National Health and Nutrition Examination Survey (NHANES 1999-2018). Chemicals were measured in blood or urine (198 chemicals, 17 families). Immune biomarkers included percentages of lymphocytes, neutrophils, monocytes, basophils, and eosinophils, and counts of red blood cells, white blood cells, and mean corpuscular volume. We conducted survey-weighted, multivariable linear regressions of log2-transformed chemicals on immune measures, adjusted for age, sex, race/ethnicity, poverty-income ratio, waist circumference, cotinine concentration, creatinine for urinary chemicals, and survey cycle. We accounted for multiple comparisons using a false discovery rate (FDR). Among 45,528 adult participants, using survey weights, the mean age was 45.7 years, 51.4% were female, and 69.3% were Non-Hispanic White. There were 65 chemicals associated with white blood cell count. For example, a doubling in the concentration of blood lead was associated with a decrease of 61 white blood cells per L (95% CI: 23-99; FDR=0.005). 122 (61.6%) chemicals were associated with at least one of the eight immune biomarkers. Chemicals in the Metals family were associated with all eight immune measures. Concentrations of a wide variety of biomarkers of exposure to chemicals such as metals and smoking-related compounds, were highly associated with immune system biomarkers, with implications for immune function and toxicology. This environmental chemical-wide association study identified chemicals from multiple families for further toxicological and epidemiological investigation.

The immune system is dynamic and complex, consisting of diverse cell types with varying forms and function. Proper function of immune cells in the context of a system is essentialas dysregulation has been linked to poor vaccine efficacy, increased susceptibility to infection, autoimmune diseases, and cancer. [1] [2] [3] [4] Clinically, immune system function is screened through measurement of blood cell populations, including quantification of lymphocytes, neutrophils, monocytes, basophils, eosinophils, red blood cells (RBCs), white blood cells (WBCs), and measurement of mean corpuscular volume (MCVa measure of the average volume of RBCs). These immune measures are influenced by several individual-level factors including age, 5,6 sex, 5, 7, 8 and genetics. 5, 9, 10 Heritability studies of the variation in these measures estimate that genetics only contribute 25-50% of the population variation. 5, [9] [10] [11] With a projected 5.08 million deaths due to top infectious diseases in 2030 12 and the current six million (March 2022) global deaths due to COVID-19, 13 it is critical to identify the modifiable environmental risk factors that contribute to altered immune measures.

Environmental chemical exposures can have significant impact on immune system function. Humans are exposed to environmental chemicals through air, water, food, personal care product use, industrial byproducts, or occupation. Chemicals often distribute in the blood stream, thus blood cells and the immune system become a major target for toxicological effects.

For example, exposure to heavy metals, pesticides, and per-and polyfluoroalkyl substances (PFAS) are associated with immune cell proportion changes, immunosuppression, and decreased vaccine efficacy. 1, [14] [15] [16] [17] [18] [19] A significant challenge in understanding how chemicals impact the immune system in humans, however, is the breadth of the possible exposures -there are more than 86,000 chemicals in commercial use in the US alone with poorly characterized immunotoxicity data. 20 The exposome refers to considering the totality of environmental factors a person experiences with their health. 21 While substantial progress has been made in epidemiological studies documenting associations between a few chemicals or single chemical families and immune measure outcomes, 22 a comprehensive analysis of the effects of chemical exposures across a broad range of chemical families on immune system measures is needed.

To investigate a broad range of chemical exposures with immune biomarkers, our study tested the hypothesis that chemical exposure biomarkers would be associated with altered immune system measures, and more specifically, that chemicals from different families would have distinct patterns of immune marker dysregulation. To accomplish this, we used data from the United States cross-sectional National Health and Nutrition Examination Survey (NHANES) to test the associations between 198 chemical exposure biomarkers from 17 families of chemicals (groups of chemicals that share similar features and/or uses), with 8 immune measures across 45,528 study participants. In an exposomic analysis of the immune system, we 1) quantified associations between individual chemical exposures and specific immune measure alterations, 2) described patterns of chemical co-exposure, and 3) identified the individual chemicals and chemical families with the greatest impact on immune measures.

NHANES is a cross-sectional study conducted annually by the United States Centers for Disease Control and Prevention. 23 NHANES is nationally representative of the noninstitutionalized, civilian United States population. Data are collected by questionnaire, clinical measurements, as well as health-related biomarker measurements, including those reflecting immune system function and chemical exposures. Data on independent samples of participants are released in cycles every two years. We included ten sets of continuous cycles (1999 -2018) . Data are publicly available and supported by the National Center for Health Statistics (https://www.cdc.gov/nchs/nhanes/index.htm). Participants provided written informed consent at . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ;  https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint the time of participation. The University of Michigan Institutional Review Board (IRB) approved these secondary data analyses (HUM00116291).

At the time of the study visit, participants provided urine samples and trained phlebotomists collected participant venous blood in the NHANES Mobile Examination Centers.

Complete laboratory methods used to measure the chemical concentrations in blood and urine and immune measures in blood are available. 24 A total of 495 chemical biomarkers were available for initial analysis. Concentrations of all of the chemical biomarkers were not however, measured in the same cycles or in the same participants. Chemical biomarker measures were cleaned with the same process as previously established (Figure 1) . 25, 26 Specifically, we ensured that there is one unique codename for each chemical biomarker to resolve differences in chemical codenames for the same chemical biomarker. Next, we ensured that units for concentrations were consistent over the study period for each chemical. For lipophilic chemicals measured in blood, we used the lipid-adjusted chemical measures (n=79). We excluded one measurement of urinary 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanol due to no corresponding creatinine measurement for that participant.

We included chemical measures after considering the lower limit of detection (LOD) for each measure based on instrument specifications. NHANES coded measurements below the LOD as the LOD value divided by the square root of two. 27 For chemicals with multiple comment codenames, we ensure that there is one unique comment codename per chemical biomarker.

Next, we excluded chemicals with greater than 50% of their measurements below the LOD (n=209). During the study period, advances in laboratory technology influenced detection frequencies for some chemicals. To minimize such influence on our analysis, we excluded measurements that showed large changes in LOD over the study period as previously . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint described, and used the cycles with the lowest LODs available. 25 For example, the LODs for PCB 196 were 10.50 ng/g for cycle 2 and 0.40 ng/g for cycle 3, thus we excluded measurements from cycle 2. Our analytic sample included 196 chemicals from 17 chemical families. All chemical measures were log2 transformed for regression analysis. Information about each chemical, including NHANES chemical codenames, full chemical names, survey cycles in which the chemical was measured, chemical family, and limit of detection were provided in Supplemental Table 1 .

Complete blood counts with 5-part differentials including MCV measurements were measured in the Mobile Examination Centers for all NHANES cycles with the Beckman Coulter method. 28 The immune measures we included in this study were percentages of basophils, eosinophils, lymphocytes, monocytes, and neutrophils as well as MCV, RBC count, and WBC count. The participants included in our study sample had complete data on all immune measures.

The participant covariates age (years), sex (male/female), race/ethnicity (Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, Other Race/Multiracial), and poverty-income ratio were collected by interview. We considered the following as reference groups: male, Non-Hispanic White, and the first survey cycle with measurements for each chemical. To protect anonymity, participants who reported an age greater than 85 years until 2006 and greater than 80 years from 2007-2018 were recorded in NHANES as 85 years or 80 years. We used the poverty-income ratio as a proxy for socio-economic status. The povertyincome ratio was calculated as household income divided by the poverty level for that year, adjusted for family size and inflation. Values ranged from 0 -5 where less than 1 indicated . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint below the poverty level, and values above 5 were rounded down to 5. Waist circumference (cm), urinary creatinine concentration (mg/dL), and cotinine concentration (ng/mL) of the participants was also collected in the Mobile Examination Centers. Creatinine and cotinine were log2 transformed.

All analyses were performed using R version 4.0.0. Code to reproduce data compilation, cleaning, and analyses are available (https://github.com/bakulskilab). To account for NHANES complex sampling design, we produced statistics representative of the non-institutionalized, civilian US population by applying survey weights to our statistical models. 26 We selected survey weights corresponding to the smallest analysis subpopulation, and this varied by chemical measure. 29 For the initial 101,316 participants, we compiled data on covariates (age, sex, race/ethnicity, poverty-income ratio, waist circumference, creatinine, cotinine, and survey cycle), immune system biomarkers (basophil, eosinophil, lymphocyte, monocyte, neutrophil, mean corpuscular volume, red blood cell count, and white blood cell count), and chemical biomarkers (n=486 chemicals). Participants were excluded for missing covariates and white blood cell counts above the maximum LOD. Urinary chemical measurements were excluded when missing creatinine measurements. We excluded creatinine measurements if the value was zero. We also excluded urinary cadmium measurements that were contaminated with molybdenum. We compared descriptive statistics for the demographic and immune measures on included and excluded study samples (Supplemental Table 2 ).

For each chemical, we also calculated descriptive statistics, including participant count, minimum concentration, maximum concentration, mean, standard deviation, median, interquartile range, first and 99 th percentiles, and percent of measurements above the LOD. To . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. To quantify the association between chemical exposures (log2-transformed) and immune measure outcomes, we used survey-weighted generalized linear models that were adjusted for age (continuous), sex (categorical), race/ethnicity (categorical), poverty-income ratio (continuous), waist circumference (continuous), log2 urinary creatinine (continuous), log2 cotinine (continuous), and survey cycle (categorical) (Equation 1). Separate models were used for each pairwise combination of chemical and immune measure. We only included urinary creatinine in the urinary chemical models to adjust for urinary dilution. 25, 30 For chemicals that were only measured in only one survey cycle, survey cycle was not included as a covariate in the regression model. Cotinine was used as a covariate to adjust for smoking exposures.

Cotinine was also included as one of the chemical exposures, but was not used as an adjustment variable for itself. The beta coefficients from these models were reported in the text and would be interpreted as: a doubling in the chemical concentration was associated with an increase or decrease of x units of the immune measure. To account for multiple comparisons across chemicals and immune biomarkers, we calculated the false discovery rate (FDR), 31 and we considered statistical significance as FDR<0.05. To compare the beta coefficients across chemicals with different concentration scales in the figures, we z-score standardized the log2transformed chemical concentrations and conducted the same generalized linear models using . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) To identify patterns of association with immune measures within chemical families from the generalized linear models, we created a heatmap of the percent of chemicals significantly associated with each immune measure, separated by chemical family. We visualized the chemical beta coefficients by stratifying the results by immune measure, ordering the chemicals by chemical family, and constructing a forest plot. To visualize the strength of associations between chemical exposures and differences in immune measurements, we created a volcano plot (effect estimate versus -log10 FDR) for each immune measure.

Several of the chemicals in this analysis were components of cigarette smoke. As a sensitivity analysis to assess associations unadjusted for the smoking biomarker (cotinine), we ran a third set of log2-transformed, z-score standardized, survey-weighted, generalized linear models that were adjusted for age, sex, race/ethnicity, poverty-income ratio, waist circumference, log2 urinary creatinine, and survey cycle, but not cotinine. We compared the results from the cotinine adjusted models to those of the non-cotinine adjusted models in a series of Pearson correlation plots for all eight immune measures to quantify the impact of tobacco smoking on our findings.

As a sensitivity analysis, we quantified the associations between age and immune measure using a series of generalized linear models adjusted for age, sex, race/ethnicity, poverty-income ratio, waist circumference, log2 urinary creatinine, smoking, and survey cycle.

To understand the likelihood of a participant having a high white blood cell count compared to a normal count with exposure to blood cadmium, cotinine, or copper, we ran three separate survey-weighted logistic regressions. The chemical concentrations were log2-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint transformed, and the included covariates were age, sex, race/ethnicity, poverty-income ratio, waist circumference, log2 cotinine for blood cadmium and copper, and survey cycle.

Our analytic sample size was 45,528 participants (Figure 1 ). Among participants, the survey-weighted mean age was 45.7 years (standard deviation: 17.1), 51.4% were female, and 69.3% were Non-Hispanic White ( Table 1 ). The median survey-weighted values for most immune measures were within the normal adult range ( Table 2 ). The percentage of monocytes, however, was slightly above the normal adult range. Compared to the included participants, excluded participants were younger (mean: 21.6 years, p<0.001), a similar percentage were female (50.8%, p=0.15), and a lower percentage were Non-Hispanic White (56.7%, p<0.001) (Supplemental Table 2 ).

We calculated descriptive statistics for the 198 included chemicals (Supplemental Table 3 ). Three chemicals (cotinine, blood lead, and blood cadmium,) were measured in more than 40,000 participants (Supplemental Figure 1 ). On average, smoking-related chemicals were measured in the most participants (mean=17,532) and PCBs were measured in the fewest (mean=1,365) (Supplemental Figure 2) . Within the participants for each chemical, 16 chemicals were detected in 100% of participants and 91 chemicals were measured in ≥95% of participants.

We calculated the Spearman correlations between the chemical concentrations ( Figure   2 , Supplemental Table 4 ). Smoking-related compounds (n=4) were highly correlated with themselves (mean r=0.88). PAHs (n=12) were also highly correlated with themselves (mean . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint r=0.75). Conversely, concentrations of metals (n=26) had a low correlation with other metals (mean r=0.18). The average pairwise correlation across all chemicals was r=0.20.

We also calculated the Spearman correlations between the immune measures (Supplemental Figure 3 , Supplemental Table 5 ). Percentages of lymphocytes and neutrophils were highly inversely correlated (r=-0.94). The other immune measures were weakly correlated (absolute value minimum r=0.002, absolute value maximum r=0.37). Table 6 ). Cotinine and chloroform were associated with six of the eight immune measures.

In the adjusted analysis, we observed 31 (25.0%) chemicals associated with lymphocyte percentage (FDR<0.05) (Figure 3 ). Of these, 22 (70.9%) of chemicals were associated with higher levels of lymphocyte percentage. Serum copper and formaldehyde were associated with the lowest and highest percentages of lymphocytes, respectively, with a doubling in the concentration of serum copper associated with 2.6% lower lymphocytes (95% CI: 1.5-3.8%;

FDR=3.7x10 -4 ) and a doubling in the concentration of formaldehyde associated with 2.4% higher lymphocytes (95% CI: 0.8-4.1%; FDR=0.02) (Figure 4 , Supplemental Table 6 ).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Chemicals in the families of acrylamides (n=1, 50.0%), metals (n=9, 34.6%), pesticides (n=6, 30 .0%), smoking-related compounds (n=2, 50.0%), PFAS (n=5, 55.6%), aldehydes (n=1, 14 .3%), and other (n=2, 50.05%) were associated with lymphocytes (Figure 3) . Chemicals with the smallest FDRs were the PFAS chemicals: perfluorooctanoic acid (FDR=3.9x10 -7 ), perfluorooctane sulfonic acid (FDR=3.3x10 -6 ), and perfluorohexane sulfonic acid (FDR=6.9x10 -6 ) (Supplemental Figure 4) .

In the adjusted analysis, we observed 33 (16.7%) chemicals were associated with neutrophil percentage (FDR<0.05) (Figure 3 ). Of these chemicals, 10 (30.3%) were associated with higher levels of neutrophil percentage. Serum zinc and serum copper were associated with the lowest and highest percentages of neutrophils, respectively, with a doubling in the concentration of serum zinc associated with 2.9% lower neutrophils (95% CI: 1.6-4.2%;

FDR=7.4x10 -4 ) and a doubling in the concentration of serum copper associated with 3.3% higher neutrophils (95% CI: 2.0-4.6%; FDR=8.4x10 -5 ) (Figure 4 , Supplemental Table 6 ). In the adjusted analysis, we observed 33 (16.7%) chemicals were associated with monocyte percentages (FDR<0.05) (Figure 3 ). Of these chemicals, 12 (36.4%) were associated with higher levels of monocyte percentage. Serum copper and cobalt were associated with the lowest and highest percentages of monocytes, respectively, with a doubling in the concentration of serum copper associated with 0.4% lower monocytes (95% CI: 0.1-0.6%; FDR=0.005), and a doubling in the concentration of cobalt associated with 0.2% higher monocytes (95% CI: 0.07-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Table 6 ). Chemicals in the families of acrylamides (n=1, 50.0%), PCBs (n=1, 2.9%), metals (5, 19 .2%), pesticides (1, 5.0%), PAHs (n=4, 33.3%), VOCs (9, 36.0%), smoking-related compounds (n=2, 50.0%) and other (1, 25.0%) were associated with monocytes (Figure 3) . Chemicals with the smallest FDRs included cotinine (FDR=2.9x10 -14 ), glycideamide (FDR=1.1x10 -6 ), and urinary thiocyanate (FDR=5.4x10 -6 ) (Supplemental Figure 4) .

Basophils (mean=0.7%, sd=0.5%) and eosinophils (mean=2.8, sd=2.0%) were the smallest percentages of total white blood cells. There were 6 (3.0%) chemicals associated with basophil percentage (FDR<0.05). There were 13 (6.6%) chemicals associated with eosinophil percentage (FDR<0.05).

In the adjusted analysis, we observed 65 (32.8%) chemicals were associated with WBC count (FDR<0.05). Of these chemicals, 45 (69.2%) were associated with higher WBC count In the adjusted analysis, we observed 68 (34.3%) chemicals were associated with RBC count (FDR<0.05). Of these chemicals, 18 (26.5%) were associated with higher RBC count Table 6 ). No chemicals in the BFRs, furans, and aldehydes families were associated with RBC count. Chemicals with the smallest FDRs included blood lead (FDR=1.3x10 -32 ), blood cadmium (FDR=1.2x10 -17 ), and blood manganese (FDR=1.7x10 -11 ) (Supplemental Figure 4) .

In the adjusted analysis, we observed 55 (27.8%) chemicals were associated with MCV (FDR<0.05). Of these chemicals, 50 (90.9%) were associated with higher MCV (Figure 3) . (Supplemental Figure 4) .

As a sensitivity analysis to assess confounding by cigarette smoke, we compared the zscore standardized chemical beta coefficients from the regressions that included cotinine as a covariate and the regressions that were not adjusted for cotinine (Supplemental Figure 5) . The correlations for each of the immune measure categories ranged from 0.74 to 0.98 (mean=0.90, p<1.1x10 -35 ).

As a sensitivity analysis to compare the effects of a 10-year increase in age compared to increased chemical exposures, we ran generalized linear models for each of the immune measures that did not include chemical exposures as a covariate. A 10-year increase in age was associated with a 0.3 decrease in percent lymphocytes (95% CI: 0.3-0.3; FDR<1.0x10 -316 ).

A 10-year increase in age was associated with a 0.06 increase in percent neutrophils (95% CI:

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 10-year increase in age was associated with a 0.08 increase in percent eosinophils (95% CI: 0.08-0.08; FDR<1.0x10 -316 ). A 10-year increase in age was associated with associated with 12 fewer WBCs per µL (95% CI: 11-12; FDR<1.0x10 -316 ). A 10-year increase in age was associated with associated with 5,408 fewer RBCs per µL (95% CI: 5,368-5,449; FDR<1.0x10 -316 ). A 10-year increase in age was associated with associated with 0.07fL larger MCV (95% CI: 0.07-0.07; FDR<1.0x10 -316 ).

As another sensitivity analysis, we conducted three logistic regressions with exposure to blood cadmium, cotinine, or copper. We observed that for a doubling of the concentration of blood cadmium, there was 1. 16 

Chemical exposure-linked immune system dysregulation includes poor vaccine efficacy, increased susceptibility to infection, autoimmune diseases, and cancer, but few of the total chemicals that Americans are regularly exposed to have been studied. [1] [2] [3] [4] We tested associations between 198 chemical exposures, across 17 chemical families, with eight immune system measures using nationally representative data from 45,528 NHANES participants. We observed 122 (61.6%) chemicals were associated (FDR<0.05) with at least one immune measure. Only percent of lymphocytes and neutrophils were highly correlated (r=-0.93), which indicated that each additional immune measure provided new information. Our results showed . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. MCV is a biomarker used to differentiate types of anemia based on the average volume of RBCs. 32 We observed 55 chemicals (27.8%) associated with MCV (FDR<0.05). Prior research on MCV and environmental chemicals have noted conflicting results. In particular, associations between blood lead and MCV have varied depending on the sample size and population. 33 A small study of car battery workers in Iran and study of pregnant women in Mexico both observed no association between blood lead and MCV, 33,34 but another small study of pregnant women in Iran found higher blood levels were associated with lower MCV. 35 Higher blood lead levels of lead recycling workers in Taiwan also were associated with lower MCV. 36 These findings were supported by toxicological research where dosing rats with a mixture of heavy metals including lead decreased MCV after 90 days. 37 In our large, generalizable US sample, we found that a doubling of blood lead concentration was not associated with MCV (FDR<0.05). Based on these observations, MCV may decrease in response to longer-term lead exposure. MCV is measured as part of complete blood panels, and values that are smaller than 80 fL or larger than 100fL can indicate diseases including iron deficiency anemia and copper deficiencies. 32, 38 Future studies may investigate the association between environmental chemicals and MCV, adjusting for intake of iron, folate, vitamin B12, and other dietary vitamins and minerals. 39, 40 Associations between MCV and environmental chemicals outside of heavy metals have been under-studied and could be a new and interesting direction for research.

A low RBC count is an indicator of anemia while a high RBC count can indicate conditions including chronically low oxygen levels or a hormonal imbalance. 41, 42 We observed 68 (34.3%) chemicals were associated with RBC count (FDR<0.05), and 73.5% of those chemicals were associated with lower RBC counts. The directions of associations in the literature were variable. In a sample of lead-exposed workers in Japan, higher blood lead was associated with lower RBC counts, 43 while a study of pregnant women in Iran found no . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint association between blood lead and RBC counts. 35 Blood lead levels of lead recycling workers in Taiwan were associated with higher RBC counts. 36 Another study using data from NHANES found that, on average, RBC counts were lowest for people with the highest PCB concentrations. 19 The differences in study populations and lead concentrations may explain the variation in the results across these studies. Given our finding of 68 chemicals associated with RBC count, a wider array of chemicals should be tested.

WBC counts are indicative of conditions including infections and inflammation. 42 In our study, the chemicals with the largest absolute value effect estimates for WBCs were copper (increase of 952 WBCs/µL; FDR=1.5x10 -7 ) and zinc (decrease of 1,088 WBCs/µL; FDR=5.8x10 -5 ) (Figure 4) . Copper and zinc are essential minerals that are important for immune system functions and can be ingested through food sources or exposed from occupational sources. [44] [45] [46] One prior study of 251 adolescents observed that higher copper and zinc concentrations were associated with higher WBC counts. 47 One rat study found no association between zinc and copper deficiency and WBC count, 48 while another found that as zinc intake increased, WBC count decreased. 49 In vitro and in vivo studies in humans have found that zinc deficiency and high levels of zinc supplementation inhibit immune system functions. 50 Copper deficiency and excessive levels can also result in organ system dysfunctions. 51, 52 The NHANES participants had a range of 24.7-306.6µg/dL copper (median: 113.8µg/dL) and 40.9-232.5µg/dL zinc (median: 80.4µg/L). The reference range for serum zinc was 60 -120µg/dL and the recommended plasma copper range was 70-140µg/dL. 51 In prior research, zinc and copper concentrations were inversely correlated, 51 although we did not find evidence of that in our nonsurvey-weighted sample (r=-0.02, p=0.2). NHANES captured both deficiencies and excessive levels of copper and zinc, representing a wide range of nutritional statuses that are associated with WBC counts. Additionally, we found that increased cotinine concentrations, a biomarker of tobacco smoke, were associated with increased WBC counts (FDR=3.9x10 -64 ). Following smoking cessation, prior studies observed WBC counts decreased, compared to continued . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint smokers. 53, 54 In addition to being a biomarker of infection and inflammation, WBC counts may also be indicative of nutritional and chemical exposures.

Neutrophils are the largest component of white blood cells and function to phagocytize bacteria and cellular debris. 42, 55, 56 Although in our NHANES dataset, concentrations of BFRs in blood were not shown to be associated with percentages and counts of immune cells, molecular toxicology studies showed that BFRs affect granulocyte functions. 57, 58 Smoking was previously associated with increased neutrophil counts 59 which matches our findings that cotinine was associated with an increase in neutrophil percentage. We found that 7 (35.0%) pesticides were associated with a lower percentage of neutrophils. Previous literature on the associations between pesticide exposures and neutrophil counts have conflicting findings. One study of agricultural workers found decreased neutrophil counts associated with pesticide exposure. 60 Another found no difference in neutrophil count in exposed and unexposed participants. 61 Studies of agricultural workers and hospitalized patients with pesticide poisoning found an increase in neutrophil counts compared to unexposed participants. 62 Monocytes are important sources for inflammatory cytokines, and there are rarely disorders that result in abnormalities in only the monocytes. 64 In our study, we observed two (50.0%) smoking-related compounds, 9 (36.0%) VOCs, and 4 (33.3%) PAHs were associated with lower percentages of monocytes. A study of cigarette smokers (exposed to smoking-related compounds, VOCs, and PAHs) found that compared to non-smokers, smoking was associated with lower monocyte counts. 65 In contrast, two other studies observed that smoking or number of cigarettes per day was associated with an increase in monocyte counts. 59, 66 It would be of interest to consider mixtures of these chemical families rather than the association of each individual chemical to determine overall effects.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Our study had some important limitations. NHANES has a cross-sectional survey design, thus chemical and immune biomarkers were only measured in each person one time.

We cannot determine temporality or longitudinal trends within each participant. We also cannot determine causality of the relationship between chemical concentrations and immune measure differences. For example, elevated blood lead could increase the RBC count through a molecular mechanism, or if lead is stored in RBCs, an increased RBC count might result in a higher blood lead concentration. In addition, NHANES does not capture long-term exposure for all chemicals. Some chemicals, like phthalates in urine, have a half-life of under an hour, while others such as cadmium in urine and lead in bones have half-lives of almost 30 years. 25 It is possible that this study is affected by a survivorship bias. Potential participants who had the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint highest levels of chemical exposures may not have survived to be included in NHANES. Future studies could examine these chemicals in a longitudinal study to measure changes over time.

We excluded chemicals with less than 50% of their measurements beyond the LOD, but it is likely that some of these chemicals would be associated with the immune measures as well.

A future analysis could test additional chemicals with immune measures. We used linear regressions to analyze the relationships between chemical exposures and immune measure outcomes, but there is evidence that chemicals have a non-linear association with other outcomes. [70] [71] [72] When we examined plots of chemical concentrations and immune measures (data not shown), we observed non-linear associations for a few of the chemicals. Future analyses could test additional exposures and incorporate non-linear analyses into their methods.

We focused on immune biomarkers in our discovery study, and based on these findings, future work may consider additional immune health outcomes. For example, a future environment-wide association study could investigate the associations between chemicals and persistent infections in NHANES, expanding the work of a similar study on four PFAS chemicals. 2 Prior research showed exposure to PCB and PFAS chemicals were associated with decreased antibodies against diphtheria and tetanus vaccines in children, potentially reducing vaccine efficacy. 1, 73, 74 Our adult analysis had measures of all six PCBs and all five PFAS examined in the childhood and adolescent vaccine efficacy studies. We observed associations between six PCBs and monocytes, RBCs, and WBCs (FDR<0.05). We were not able to differentiate lymphocyte subtypes, so it is possible that B cell proportions decreased and the other subtype proportions increased more, or all subtypes increased in cell count, but these chemicals disrupted antibody production. Future studies could directly measure or estimate the proportions of lymphocyte subtypes, including B cells as well as antibody levels in response to . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022. ; https://doi.org/10.1101/2022.03.22.22272789 doi: medRxiv preprint PCB and PFAS chemical exposures. Other chemical exposures may also be associated with reduced humoral responses to vaccines and could be analyzed using the NHANES dataset. 75 Several studies have documented associations in humans between a few chemicals or single chemical families and immune measure outcomes 40, 59, [76] [77] [78] [79] [80] 

Oulhote Y, Shamim Z, Kielsen . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022 

ߙ . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted March 23, 2022 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) 6 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) Effect Estimate

Serum Vaccine Antibody Concentrations in Adolescents Exposed to Perfluorinated Compounds

Associations of exposure to perfluoroalkyl substances individually and in mixtures with persistent infections: Recent findings from NHANES 1999-2016

Cellular and molecular mechanisms of immune dysregulation and autoimmunity

Paradoxical roles of the immune system during cancer development

The cellular composition of the human immune system is shaped by age and cohabitation

The immune system and aging: a review

Blood Groups and Red Cell Antigens

Age-related changes in peripheral blood counts in humans

Variation in the human immune system is largely driven by non-heritable influences

ImmVar project: Insights and design considerations for future studies of "healthy" immune variation

Genetic variants regulating immune cell levels in health and disease

The International Bank for Reconstruction and Development / The World Bank

An interactive web-based dashboard to track COVID-19 in real time

Approaches to detecting immunotoxic effects of environmental contaminants in humans

Some Immunological Effects of Lead, Cadmium and Methylmercury

Lead exposure and its impact on immune system: A review

Immunotoxicology of cadmium: Cells of the immune system as targets and effectors of cadmium toxicity

National Health and Nutrition Examination Survey

Characterization of age-based trends to identify chemical biomarkers of higher levels in children

A comprehensive analysis of racial disparities in chemical biomarker concentrations in United States women

Laboratory Procedure Manual -Complete Blood Count

Weighting. CDC/National Center for Health Statistics

Urinary Creatinine Concentrations in the U.S. Population: Implications for Urinary Biologic Monitoring Measurements

Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing

Mean Corpuscular Volume

Relationship Between Blood Lead Levels and Hematological Indices in Pregnant Women

Correlation Between Iron Deficiency and Lead Intoxication in the Workers of a Car Battery Plant

Study of the Relationship between Blood Lead Level and Complete Blood Counts in Pregnant Women

Interactive Effects between Chronic Lead Exposure and the Homeostatic Iron Regulator Transport HFE Polymorphism on the Human Red Blood Cell Mean Corpuscular Volume (MCV)

Immunosuppressive Effect of Subchronic Exposure to a Mixture of Eight Heavy Metals, Found as Groundwater Contaminants in Different Areas of India, Through Drinking Water in Male Rats

Laboratory Tests and Diagnostic Procedures

Relationship between mean corpuscular volume and cognitive performance in older adults

The association of cadmium and lead exposures with red cell distribution width

Henry's Clinical Diagnosis and Management by Laboratory Methods

Understanding the complete blood count with differential

Benchmark Dose of Lead Inducing Anemia at the Workplace

Modulatory effects of selenium and zinc on the immune system

Effects of Copper Deficiency on the Immune System

Biofortification of crops with seven mineral elements often lacking in human diets -iron, zinc, copper, calcium, magnesium, selenium and iodine

Effect of dietary zinc and copper interrelationships on blood parameters of the rat

Studies of Zinc Metabolism in the Rat

Zinc-Altered Immune Function

Chapter 17 -Disorders of trace metals

Chapter 35 -Copper

Changes in hemorheological and biochemical parameters following short-term and long-term smoking cessation induced by nicotine replacement therapy (NRT)

Effects of Biochemically Confirmed Smoking Cessation on White Blood Cell Count

Disorders of Neutrophil Function

Molecular Mechanisms Involved in the Toxic Effects of Polychlorinated Biphenyls (PCBs) and Brominated Flame Retardants (BFRs)

A Commercial Mixture of the Brominated Flame Retardant Pentabrominated Diphenyl Ether (DE-71) Induces Respiratory Burst in Human Neutrophil Granulocytes In Vitro

Smoking and Increased White and Red Blood Cells

Adverse Respiratory Health and Hematological Alterations among Agricultural Workers Occupationally Exposed to Organophosphate Pesticides: A Cross-Sectional Study in North India

Assessment of Pesticide Residues in Human Blood and Effects of Occupational Exposure on Hematological and Hormonal Qualities

Occupational pesticide exposure and adverse health effects at the clinical, hematological and biochemical level

Neutrophil-lymphocyte ratio in patients with pesticide poisoning

Classification and Clinical Manifestations of Disorders of Monocytes and Macrophages

Blood eosinophil and monocyte counts are related to smoking and lung function

Cigarette smoking and peripheral blood leukocyte differentials

Association between exposure to pesticides and disorder on hematological parameters and kidney function in male agricultural workers

Polychlorinated biphenyls, lead, and mercury are associated with liver disease in American adults: NHANES

Insecticide and metal exposures are associated with a surrogate biomarker for non-alcoholic fatty liver disease in the National Health and Nutrition Examination Survey

Nonmonotonic Dose-Response Curves Occur in Dose Ranges That Are Relevant to Regulatory Decision-Making

Serum Selenium Levels and All-Cause, Cancer, and Cardiovascular Mortality Among US Adults

A Nonlinear Relation Between Maternal Red Blood Cell Manganese Concentrations and Child Blood Pressure at Age 6-12 y: A Prospective Birth Cohort Study

Serum Concentrations of Antibodies Against Vaccine Toxoids in Children Exposed Perinatally to Immunotoxicants

Serum Vaccine Antibody Concentrations in Children Exposed to Perfluorinated Compounds

Vaccination efficacy and environmental pollution

Immunologic effects of background exposure to polychlorinated biphenyls and dioxins in Dutch preschool children

Associations between blood BTEXS concentrations and hematologic parameters among adult residents of the U.S. Gulf States

Effect of Cigarette Smoking on Haematological Parameters in Healthy Population