Pesticide residue intake from fruits and vegetables and alterations in the serum metabolome of women undergoing infertility treatment

Abstract

BACKGROUND.

Pesticide exposure is linked to a myriad of negative health effects; however, the mechanisms underlying these associations are less clear. We utilized metabolomics to describe the alterations in the serum metabolome associated with high and low pesticide residue intake from fruits and vegetables (FVs), the most common route of exposure in humans.

METHODS.

This analysis included 171 women undergoing in vitro fertilization who completed a validated food frequency questionnaire and provided a serum sample during controlled ovarian stimulation (2007-2015). FVs were categorized as high or low-to-moderate pesticide residue using a validated method based on pesticide surveillance data from the USDA. We conducted untargeted metabolic profiling using liquid chromatography with high-resolution mass spectrometry and two chromatography columns. We used multivariable generalized linear models to identified metabolic features (p<0.005) associated with high and low-to-moderate pesticide residue FV intake, followed by enriched pathway analysis.

RESULTS.

We identified 50 and 109 significant features associated with high pesticide residue FV intake in the C18 negative and HILIC positive columns, respectively. Additionally, we identified 90 and 62 significant features associated with low-to-moderate pesticide residue FV intake in the two columns, respectively. Four metabolomic pathways were associated with intake of high pesticide residue FVs including those involved in energy, vitamin, and enzyme metabolism. 12 pathways were associated with intake of low-to-moderate pesticide residue FVs including cellular receptor, energy, intercellular signaling, lipid, vitamin, and xenobiotic metabolism. One energy pathway was associated with both high and low-to-moderate pesticide residue FVs.

CONCLUSIONS.

We identified limited overlap in the pathways associated with intake of high and low-to-moderate pesticide residue FVs, which supports findings of disparate health effects associated with these two exposures. The identified pathways suggest there is a balance between the dietary antioxidant intake associated with FVs intake and heightened oxidative stress as a result of dietary pesticide exposure.

Graphical Abstract

1. INTRODUCTION

Pesticides have a wide range of uses including but not limited to: reducing crop loss from pests, reducing pests in homes, and reducing diseases by killing disease vectors (i.e. malaria, dengue, yellow fever, Lyme’s disease etc.). Because of the wide range of uses, exposure to pesticides is ubiquitous.¹ While pesticide exposure occurs through a variety of routes, one of the main routes of exposure in the general population is ingestion of pesticide residue on conventionally grown fruits and vegetables (FVs).¹ In 2016, 85% of FVs in U.S. markets had detectable pesticide residues, and nearly half of FVs had detectable levels of three or more pesticides.² Both dietary and occupational exposures to pesticides have been linked to health effects including increased risk of neurological disorders, digestive and liver diseases, respiratory issues, and adverse reproductive and pregnancy outcomes.^1,3–8 While many in vivo and ex vivo animal experiments have examined specific biologic mechanisms underlying the health effects of high exposure to pesticides, the effects of chronic low-level exposure to pesticide residues through diet requires further study.

Metabolomics is a relatively new field that studies small molecules in biological tissues and fluids, which can be used to advance our understanding of the biological mechanisms underlying associations between exposure and disease. In recent years there has been an interest in using metabolomics to measure contaminants in food as well as to study the mediating biologic pathways that might underlie why exposure to pesticide may lead to higher risk of human diseases.^7,9–12 Several applications of metabolomics methods have already been used to identify contaminants, including pesticides, in foods^13–15 and describe the health effects of pesticide exposure in humans.^7,9–12,16 These previous human studies have observed that higher pesticide levels in serum, saliva, and urine were associated with changes in lipid metabolism, amino acid metabolism, polyamine metabolism and energy metabolism as well as an increased need for antioxidants.^7,9–11,16 While these studies have provided some evidence for potential biological effects of pesticide exposure, these studies only describe pesticide exposure from residential or occupational sources (or do not identify a source), rather than from food sources, and their effects on human health with the exception of one study examining pesticide levels in human breast milk and infant health.¹²

We previously developed and validated a questionnaire-based method for assessment of pesticide residues in FVs: the Pesticide Residue Burden Score (PRBS).^17,18 The PRBS previously has been found to be associated with a variety of reproductive and fertility outcomes including poorer semen quality and lower probability of clinical pregnancy and live birth.^19,20 Here, we used the PRBS to evaluate the association of fruit and vegetable intake considering their pesticide residue status with alterations in the serum metabolome among a prospective cohort of women undergoing infertility treatment. Identifying potential pathways and metabolites associated with higher dietary intake of pesticides could lead to identification of biomarkers of exposure as well as provide information about potential biological mechanisms between dietary pesticide intake and disease states. Based on previous studies of pesticide exposure in occupational and residential settings, we hypothesized that we would observe alterations in lipid metabolism, amino acid metabolism, and energy metabolism among the serum of individuals exposed to higher levels of high pesticide residue FV intake.

2. MATERIALS & METHODS

2.1. Study population.

Women in our study were participants in the Environment and Reproductive Health (EARTH) study.²¹ Briefly, this study was a prospective cohort that sought to evaluate dietary and environmental factors associated with fertility. The EARTH study enrolled couples seeking infertility evaluation and treatment at the Massachusetts General Hospital Fertility Center (2005-2019) and followed them through each of their treatment cycles. The EARTH study was approved by the Institutional Review Boards at the MGH and Harvard T.H. Chan School of Public Health. All study participants provided written informed consent.

Since this sub-study was originally designed to evaluate air pollution exposure, 345 women with complete air pollution data who underwent a fresh, autologous assisted reproductive technology (ART) cycle between 2005 and 2015 were initially eligible (Supplemental Figure 1). Of these we randomly selected 200 women for the EARTH metabolomics sub-study. As described previously,²² there was little difference in terms of demographic or ART cycle characteristics between the 345 women who were eligible and the 200 included in the sub-study. Samples from 345 women could not be included due to budget constraints. Of the 200 women with metabolomics data, 22 joined the EARTH Study prior to the introduction of the food frequency questionnaire (FFQ), and 7 women did not complete an FFQ. This left 171 women with complete dietary and metabolomics information to be included in our analysis.

2.2. Pesticide measures.

To quantify dietary pesticide intake from fruits and vegetables (FVs), we used data collected from a validated 131-food item FFQ,²³ which quantified each women’s average intake of fruits and vegetables over the previous year. The US Department of Agriculture’s pesticide data program reports (2006-2015)² were used to determine the pesticide residue burden score (PRBS) of FVs. The PRBS has been used and described elsewhere.^17,19,24 In brief, we classified FVs into those with high and low-to-moderate pesticide residue burden based on three factors: (1) percentage of samples with any detectable pesticides, (2) percentage of samples tested with pesticide exceeding the tolerance level, and (3) percentage of samples with 3 or more individual detectable pesticides. We classified FVs into tertiles for each of these three measures and assigned a score of 0 for the bottom tertile, 1 for the middle tertile, and 2 for the top tertile. We then summed these scores across the three measures. Any fruit or vegetable with a summed score 4 or higher was considered a high pesticide residue FV while the rest were considered a low-to-moderate pesticide residue FV. Fourteen FVs were classified as high pesticide residue burden while 22 were classified as low-to-moderate pesticide residue burden. We then summed intakes of high and low-to-moderate pesticide residue FVs, separately, to create an estimated servings per day for each woman.

2.3. Covariate assessment.

Upon enrollment into the EARTH study, women completed a self-administered baseline questionnaire that covered demographics, lifestyle factors, medical history, and select environmental exposures. For example, women self-reported whether their home or apartment had been treated with pesticides for ants, roaches, insects, pests, or rodents in the past five years with a response of yes or no. Two data-derived dietary pattern scores, excluding fruits and vegetable intake, the prudent and Western pattern, were also calculated to summarize overall food choices based the FFQ food group data. Fruits and vegetables were excluded from these dietary patterns since our exposure was tied to FV intake and we did not want to overadjust our associations. Women’s height and weight were measured at entry into the study by a trained staff member to calculate body mass index (BMI; kg/m²).

2.4. High-resolution metabolomics.

All women provided a non-fasting blood sample, during controlled ovulation stimulation, to facilitate metabolomic profiling. Approximately 6-ml of blood was collected via venipuncture during a routine morning appointment (7 AM to 10 AM). Serum was centrifuged, aliquoted, and stored at −20 °C initially before being transferred to Harvard for storage at −80 °C. We used established protocols to analyze the stored samples.^25–27 Serum samples were treated with two volumes of acetonitrile and were analyzed in triplicate. Samples were analyzed with liquid chromatography with high-resolution mass spectrometry (LC-HRMS) (Dionex Ultimate 3000 RSLCnano; Thermo Orbitrap Fusion). To maximize the metabolomic coverage, we used two chromatography columns: (1) C18 hydrophobic reversed-phase chromatography (C18 negative) with negative electrospray ionization (ESI) and (2) hydrophobic interaction liquid chromatography (HILIC) with positive ESI. For quality control purposes, at the beginning and end of each batch, we included two standards, NIST 1950²⁸ and pooled human plasma (Equitech Bio). Raw data files were converted with ProteoWizard to .mzML files using apLCMS and xMSanalyzer.^29–31 Unique metabolic features were identified based on their mass-to-charge ratio, retention time, and ion density. Features detected in less than 10% of samples were removed. Additionally, samples with a median coefficient of variation among technical replicates greater than 30% and a Pearson correlation less than 0.7 were not included in the analysis. We then log transformed the average intensity of the remaining features for analysis.

2.5. Statistical analysis.

We used an untargeted metabolomics approach and identified metabolomic features associated with high and low-to-moderate pesticide residue FV intake. We used generalized linear models to identify significant associations use the following model:

$$\begin{gathered} {\text{Y}}_{\text{ji}}=\text{α}+{\text{β}}_{\text{1j}}{\text{High PRBS}}_{\text{i}}+{\text{β}}_{\text{2j}}{\text{Low PRBS}}_{\text{i}}+{\text{β}}_{\text{3j}}{\text{Age}}_{\text{i}}+{\text{β}}_{\text{4j}}{\text{BMI}}_{\text{i}}+{\text{β}}_{\text{5j}}{\text{Smoking}}_{\text{i}}+{\text{β}}_{\text{6j}}{\text{Folate}}_{\text{i}} \\ +{\text{β}}_{\text{7j}}{\text{Residential Pesticide}}_{\text{i}}+{\text{β}}_{\text{8j}}{\text{Prudent Diet}}_{\text{i}}+{\text{β}}_{\text{9j}}{\text{Western Diet}}_{\text{i}} \\ +{\text{β}}_{\text{10j}}{\text{Caloric Intake}}_{\text{i}}+{\text{β}}_{\text{11j}}{\text{Infertility}}_{\text{i}}+{\text{E}}_{\text{ij}} \end{gathered}$$

In these models, Y_ji was the natural log of the intensity for feature j and participant i. High PRBS_i was woman i’s intake of high pesticide residue FVs (in servings/day). Low PRBS_i was woman i’s intake of low-to-moderate pesticide residue FVs (in servings/day). Finally, these models also included the woman’s age (Age_i) (continuous), body mass index (BMI_i) (continuous), smoking status (Smoking,) (ever, never), supplemental folate intake (Folate_i) (continuous), residential pesticide usage (Residential Pesticide_i) (yes, no), prudent diet score (Prudent Diet_i), western diet score (Western Diet_i) (continuous), total caloric intake (Caloric Intake_i) (continuous), and initial infertility diagnosis (Infertility_i) (male, female, unexplained). E_ij denotes the residual normal error. We only had one model for high-pesticide residue FV intake and low-to-moderate pesticide FV intake because they may confound each other. We used two separate models for the two technical columns (C18 negative and HILIC positive) to maximize our coverage of metabolomic features. We identified features at increasing levels of significance (p<0.05, p<0.005, p<0.0005). To address multiple comparisons, we also corrected these raw p-values using the Benjamini-Hochberg false discovery rate (FDR) at two thresholds (q-value<0.20 and q-value<0.05).

2.5. Pathway analysis and metabolite confirmation.

We used mummichog (v. 1.0.10) to complete the pathway analysis. Briefly, mummichog is a validated innovative bioinformatics tool that identifies pathways from a reference list of significant metabolomic features without needing to first identify the metabolites.³² Mummichog uses the features mass-to-charge ratio and retention time to accomplish pathway identification and it provides an adjusted p-value for each pathway by resampling the reference input file using a gamma distribution.³² We used two references (one for each technical column) that included significant features at the p<0.005 level. We used an uncorrected p-value because very few features were significant after FDR correction. We used a more stringent p-value than the typical 0.05 level to be more conservative. We used mummichog to identify pathways associated with high and low-to-moderate pesticide residue FV intake, separately. Pathways with less than 10% overlapping features identified from the overall pathway size were excluded to further prevent false positives. Heat maps were used to compare pathways associated with the two different exposures.

For metabolite annotation, we selected the features that were significantly associated with either high or low-to-moderate pesticide residue FV intake (p<0.005). We examined extracted ion chromatography for retention time, isotope patterns, and peak quality Significant features with high quality peaks were then compared to authentic standards from our laboratory that were analyzed with the same methods (level-1 evidence).³³ Significant features were matched to authentic standards by comparing the mass-to-charge ratio, retention time, and ion dissociation.

3. RESULTS

3.1. Sample characteristics.

The average age of women in this study was 34.8 years (standard deviation [SD]: 3.9 years) and the average BMI was 23.8 (SD: 4.4) (Table 1). The average total caloric intake was 1811.5 calories (SD: 620.6). On average women consumed 1.8 (SD: 1.1) servings per day of high pesticide residue FVs and 2.9 (SD: 1.5) servings per day of low-to-moderate pesticide residue FVs. High pesticide FVs (in ascending order of pesticide residue score) included tomatoes, apple sauce, blueberries, kale/mustard/chard greens, winter squash, fresh apple/pear, string beans, grape/raisin, potatoes, spinach (cooked), peach/plum, strawberries, spinach (raw), green/yellow/red peppers, and celery. Low-to-moderate pesticide FVs (in ascending order of pesticide residue score) included peas/lima beans, dried plums/prunes, onions, beans/lentils, avocado, corn, cabbage/coleslaw, orange juice, tomato sauce, apple juice/cider, cauliflower, grapefruit, cantaloupe, tofu, bananas, eggplant/summer squash/zucchini, yam/sweet potatoes, oranges, broccoli, carrots, and head lettuce/leaf lettuce.¹⁹ The Spearman correlation between intake of high and low-to-moderate pesticide residue FVs was moderate (r=0.60).

Table 1.

Descriptive characteristics of the women enrolled in the EARTH Study who were included in the metabolomics sub-study with diet information (n=171).

	Overall	High Pesticide Reside FV		Low-to-moderate Pesticide Residue FV
	(n=171)	Q1 (n=43)	Q4 (n=43)	Q1 (n=43)	Q4 (n=43)
Pesticide residue, mean (SD)		0.72 (0.19)	3.32 (1.03)	1.36 (0.28)	5.04 (1.25)
DEMOGRAPHICS
Age, years, mean (SD)	34.8 (3.9)	35.5 (3.8)	34.5 (3.3)	35 (4.3)	34 (3.9)
Body Mass Index, kg/m2, mean (SD)	23.8 (4.4)	23.1 (2.9)	23.2 (4.6)	23.7 (3.7)	23.3 (3.6)
White, n (%)	144 (82.4)	36 (83.7)	33 (76.7)	36 (83.7)	34 (79.1)
Education, n (%)
<College	15 (8.8)	4 (9.3)	2 (4.7)	5 (11.6)	6 (14.0)
College	50 (29.2)	11 (25.6)	10 (23.3)	11 (25.6)	12 (27.9)
Graduate degree	106 (62.0)	28 (65.1)	31 (72.1)	27 (62.8)	25 (58.1)
Never Smoked, n (%)	127 (74.3)	31 (72.1)	30 (69.8)	32 (74.4)	23 (53.5)
Infertility Diagnosis, n (%)
Male Factor	47 (27.5)	12 (27.9)	8 (18.6)	10 (23.3)	8 (18.6)
Female Factor	53 (31.0)	11 (25.6)	17 (39.5)	16 (37.2)	15 (34.9)
Unexplained	71 (41.5)	20 (46.5)	18 (41.9)	17 (39.5)	20 (46.5)
Residential Pesticide Use, n (%)
Yes	40 (23.4)	12 (27.9)	12 (27.9)	8 (18.6)	10 (23.3)
No	131 (76.6)	31 (72.1)	31 (72.1)	35 (81.4)	33 (76.7)
DIET
Total Caloric Intake, kcal/day, mean (SD)	1881.5 (620.6)	1418.8 (472.6)	2115.2 (667.9)	1388.7 (431.3)	2213 (701.3)
Prudent Diet, mean (SD)	−0.03 (1.00)	−0.23 (0.70)	−0.17 (1.00)	−0.24 (0.82)	0.01 (1.25)
Western Diet, mean (SD)	0.06 (1.04)	−0.82 (0.43)	1.19 (1.19)	−0.73 (0.57)	1.21 (1.18)
Supplemental Folate, μg/day, mean (SD)	622.4 (360.8)	607.3 (305.9)	639.8 (459.5)	648.5 (305.3)	634.1 (405.0)

3.2. Significant features.

Overall, 50 and 109 metabolic features were significantly associated (p<0.005) with intake of high pesticide residue FVs in the C18 negative and HILIC positive columns, respectively. Similarly, 99 and 65 metabolic features were significantly associated with intake of low-to-moderate pesticide residue FVs from the two analytical columns (Table 2). Almost none of these features remained statistically significant after FDR adjustment.

Table 2.

Number of significant features associated with intake of high and low-to-moderate pesticide residue fruits and vegetables.

	C18 Negative (n=10,803)					HILIC Positive (n=12,968)
	Raw P-values			Corrected Q-values		Raw P-values			Corrected Q-values
Exposure	<0.05	<0.005	<0.0005	<0.20	<0.05	<0.05	<0.005	<0.0005	<0.20	<0.05
High (Total) A	497	50	2	0	0	909	109	10	1	1
Low-to-moderate (Total) A	945	99	11	4	0	572	65	11	2	0
High (Total) A,B	529	50	1	0	0	914	119	9	1	1
Low-to-moderate (Total) A,B	975	108	11	5	0	577	65	10	3	0

^AGeneralized linear models were adjusted for age, BMI, smoking status, folate intake, organic food intake, residential pesticide usage, prudent diet score, western diet score, total caloric intake, and infertility diagnosis. Models were also adjusted for high or low-to-moderate pesticide residue FV intake.

^BModels to obtain number of significant features did not co-adjust the other pesticide burden residue score.

Visual inspection of the volcano plots demonstrated that in the C18 column, a roughly equal number of metabolites were negatively and positively associated with the intake of high pesticide residue FVs when examining the beta coefficients (Figure 1). In contrast, in the HILIC column, more metabolites were positively associated with the intake of high pesticide residue FVs than were negatively associated when examining the beta coefficients. More metabolites were negatively associated than were positively associated in the C18 column while in the HILIC column a roughly equal number of metabolites were positively associated and negatively associated with the intake of low-to-moderate pesticide residue FVs when examining the beta coefficients (Figure 2).

Figure 1.

Volcano Plot of Metabolic Features Associated with High Pesticide Residue Fruit and Vegetable Intake (p<0.005).^A

^A Generalized linear models were adjusted for age, BMI, smoking status, folate intake, organic food intake, residential pesticide usage, prudent diet score, western diet score, total caloric intake, and infertility diagnosis. Models were co-adjusted for high and low-to-moderate pesticide residue FV intakes.

Figure 2.

Volcano Plot of Metabolic Features Associated with Low-to-Moderate Pesticide Residue Fruit and Vegetable Intake (p<0.005).^A

^A Generalized linear models were adjusted for age, BMI, smoking status, folate intake, organic food intake, residential pesticide usage, prudent diet score, western diet score, total caloric intake, and infertility diagnosis. Models were also adjusted for both pesticide residue.

3.3. Pathway analysis.

We observed four pathways significantly and solely associated with intake of high pesticide residue FVs (Table 3). Two of these pathways were vitamin metabolism pathways (vitamin A and vitamin B5), 1 was an energy metabolism pathway (starch and sucrose metabolism), and 1 was an enzyme metabolism pathway (CoA catabolism).

Table 3.

Pathway analysis of significant features (<0.005) associated with high and low-to-moderate pesticide residue fruit and vegetable intake.

^AESI: Electrospray ionization for: (−) C18 Negative column and (+) HILIC positive column

^BHigh pesticide residue fruits and vegetables

^CLow-to-moderate pesticide residue fruits and vegetables

We observed twelve pathways solely associated with intake of low-to-moderate pesticide residue FVs (Table 3). These pathways included four energy metabolism pathways, 3 cellular receptor metabolism pathways, 2 xenobiotic metabolism pathways, 1 intercellular signaling pathway, 1 lipid metabolism pathway, and 1 vitamin metabolism pathway. The energy metabolism pathways included: galactose metabolism, hexose phosphorylation, fructose and mannose metabolism, and pentose phosphate pathway. The cellular receptor metabolism pathways included: chondroitin sulfate degradation, heparan sulfate degradation, and N-glycan degradation. The xenobiotic metabolism pathways included: caffeine metabolism and drug metabolism. The intercellular signaling metabolism pathway was phosphatidylinositol phosphate metabolism. Glycosphingolipid metabolism was the lipid metabolism pathway and vitamin D3 (cholecalciferol) metabolism was the vitamin metabolism pathway. Lastly, one energy metabolism pathway (pentose and glucoronate interconversion) was associated with both high and low-to-moderate pesticide residue FVs.

3.4. Metabolite identification.

Using level 1 evidence, we confirmed 4 significant metabolites (p<0.005) associated with high or low-to-moderate FV intake (Table 4). Mannitol (beta: −0.737) was negatively associated with intake of high pesticide residue FVs while cortisol (beta: 0.170) was positively associated with intake of high pesticide residue FVs. Gluconic acid (beta: −0.125) and methyl-histidine (beta: −0.249) were negatively associated with intake of low-to-moderate pesticide residue FVs.

Table 4.

Significant metabolites (<0.005) associated with high and low-to-moderate pesticide residue fruit and vegetable intake confirmed using Level 1 evidence.

Exposure	ESIA	Chemical Name	Beta B	M/Z	Time	Super Class	Class
High C	−	Mannitol E	−0.737	217.0482	21.3	Organic oxygen compounds	Organooxygen compounds
High C	+	Cortisol	0.170	363.2161	26.0	Lipids and lipid-like molecules	Steroids and steroid derivatives
Low-to-moderate D	−	Gluconic acid	−0.125	195.0514	20.4	Organic oxygen compounds	Organooxygen compounds
Low-to-moderate D	+	Methyl-histidine	−0.249	170.0924	78.9	Organic acids and derivatives	Carboxylic acids and derivatives

^AESI: Electrospray ionization for: (−) C18 Negative column and (+) HILIC positive column

^BBeta coefficient for the association between the metabolite and either the intake of high pesticide residue FVs or the intake of low-to-moderate pesticide residue FVs.

^CHigh pesticide residue fruits and vegetables

^DLow-to-moderate pesticide residue fruits and vegetables

^EMannitol is indistinguishable from other sugar alcohols; however, mannitol was assigned because it is the most common sugar alcohol.

3.5. Sensitivity analysis.

When we reran our analysis of intake of high and low-to-moderate pesticide residue FVs without co-adjustment for each other we observed similar results. In most instances, we observed slightly more significant metabolites associated with FV intake from both high and low-to-moderate sources (Table 2; Supplemental Figure 2; Supplemental Figure 3). Among the pathways associated with intake of high pesticide residue FVs, we observed one new pathway (linoleate metabolism) while CoA catabolism and vitamin B5 metabolism were no longer significant when compared to the co-adjusted results (Supplemental Table 1). All other pathways observed for the intake of high pesticide residue FVs were the same as the main results. In the sensitivity analysis, we observed all the same pathways associated with intake of the low-to-moderate pesticide residue FVs when compared to the main analysis. In the sensitivity analysis, pentose and glucoronate interconversion was still associated with both exposures. Without co-adjustment for the two exposures, we confirmed 2 additional metabolites associated with intake of low-to-moderate pesticide residue FVs (Supplemental Table 2) (in addition to the four metabolites that were confirmed in the main analysis). The two new metabolites, 4-hydroxyl-phenylglycine (beta: −0.216) and pyridoxal (beta: −0.216) were negatively associated with the intake of low-to-moderate pesticide residue FVs.

4. DISCUSSION

4.1. Key findings.

We identified several different biologic pathways associated with intake of high and low-to-moderate pesticide residue FVs. Four metabolomic pathways were significantly altered with higher intake of high pesticide residue FVs including energy metabolism, enzyme metabolism, and vitamin metabolism while 12 pathways were altered with higher intake of low-to-moderate pesticide residue FVs including cellular receptor metabolism, energy metabolism, intercellular signaling, lipid metabolism, vitamin metabolism, and xenobiotic metabolism. We also observed one energy metabolism pathway associated with both intake of high and low-to-moderate pesticide residue FVs. In addition to the metabolomic pathways identified, using level-1 evidence we confirmed the identity of 4 metabolites, 2 associated with intake of high pesticide residue FVs and 2 associated with low-to-moderate pesticide residue FVs. The alterations we observed in the serum metabolome from pesticide residue from FV intake should be considered biologically plausible. Previous literature exploring the PRBS have observed poorer semen quality and lower probability of clinical pregnancy and live birth with increasing consumption of high pesticide residue FVs.^19,20 Although not directly addressed in this manuscript, it is possible that some of the specific metabolites and pathways identified may offer insight into the biological mechanisms underlying these associations..

We identified similarities in the broad categories of pathways associated with intake of high and low-to-moderate pesticide residue FVs, but not for the individual pathways. For example, energy metabolism and vitamin metabolism pathways were associated with intake of FVs, regardless of pesticide residue status. However, the energy metabolism pathways differed; intake of high pesticide residue FVs was associated with starch and sucrose metabolism while intake of low-to-moderate pesticide residue FVs was associated with pentose phosphate pathway, fructose and mannose metabolism, hexose phosphorylation, and galactose metabolism. Pentose and glucuronate interconversion was the lone pathway observed to be associated with both high and low-to-moderate pesticide residue FV intake. Alterations in vitamin metabolism were also common with higher intake of both high and low-to-moderate pesticide residue FV intake. With high pesticide residue FV intake, metabolic features associated with vitamin A and vitamin B5 metabolism were commonly altered while with low-to-moderate pesticide residue FV intake metabolic features associated with vitamin D3 metabolism was commonly altered. On a whole, intake of low-to-moderate pesticide residue FVs was associated with many more biologic pathways including cellular receptor metabolism, intercellular signaling, lipid metabolism, and xenobiotic metabolism.

Comparing our results to previous literature is difficult given that many studies of metabolomics and pesticide exposure tend to focus on pesticides measured in biological fluids rather than noting the specific source of pesticides, and those studies that do measure pesticide exposure from diet focus on breast milk.^12,34 However, we did observe some similarities and differences that are worth highlighting. For example, our finding that several energy pathways were associated with higher dietary pesticide intake is not unexpected. In an untargeted metabolomics study of 51 adults occupationally exposed to several pesticides in India, Ch et al observed several energy metabolism pathways that were altered in the worker’s urine and saliva as compared to pesticide applicators who were not working.¹¹ Increased energy metabolism may be needed after exposure to pesticides because pesticides can induce oxidative stress and damage cells. A review of the toxicity mechanisms for persistent organochlorinated pesticides in humans and animals confirms that these pesticides can damage cells from oxidative stress³⁵ offering credence to our results.

In addition to energy metabolism, several human studies have identified alterations in lipid metabolism with higher levels of pesticides in various biological fluids,^{7,10,16,36–38} which is not entirely surprising given the lipophilic nature of many pesticides. In a study of male Wistar rats that were subcutaneously exposed to either chlorpyrifos, diisopropylphosphorofluoridate, or parathion, lipid peroxidation products increased and were associated with oxidative injury.³⁶ In humans, an untargeted metabolomics study of 34 healthy adults from a larger study investigating markers of meat consumption in Northern Ireland, higher levels of persistent organic pollutants (POPs) in the serum, including several pesticides, were associated with alterations in sphingolipids metabolism.¹⁰ We similarly found that higher dietary intake of pesticides from FVs was associated with altered glycosphingolipids metabolism, which is a subtype of sphingolipids. Additionally, in another study in Sweden where the primary chemicals measured were dichlorodiphenyldichloroethylene (p,p’-DDE) and hexachlorobenzene (HCB) concentrations in the serum of 1,016 adults from a population registry, metabolites in sphingolipids pathway was found to be altered with higher exposure to these chemicals.¹⁶ These consistent findings suggesting dysregulation of the sphingolipid metabolism pathway across numerous populations and settings strongly suggests that future studies should specifically explore how this pathway may ultimately influence health. In addition to sphingolipid metabolism, several studies^7,10,16 including Yang et al which evaluated serum levels of 37 pesticide metabolites among 104 pregnant women in China,⁷ also observed some alterations in other lipid pathways including glycerophospholipids and glycerolipids metabolism, which we did not observe in our analysis.

In addition to energy and lipid metabolism pathways that have been commonly observed in previous studies, we also found several pathways that could be unique to pesticide intake from FVs or possibly not identified by previous researchers. For example, we observed that several vitamin metabolism pathways were altered with higher intake of high pesticide residue FVs. One potential explanation for this finding could be the increased need for antioxidants to combat inflammation and oxidative stress induced by pesticide exposure.^35–38 For example, exposure to permethrin (a specific type of pesticide) has been shown to increase oxidative stress across several studies.³⁸ Additionally, in a cluster randomized crossover trial of organic diets, children exposed to organic diets with lower levels of pyrethroid and neonicotinoid had decreases in markers of oxidative stress and inflammation.³⁷ A review of persistent organochlorinated pesticides in animals and humans found that these pesticides can damage cells through oxidative stress among other mechanisms,³⁵ which may require cells to increase vitamin metabolism to combat this damage. However, since FVs are also key dietary sources of vitamins, these findings could also reflect residual confounding due to higher levels of these specific nutrients. As such, additional studies – ideally randomized controlled trials of conventional versus organic FV intake – are needed to confirm these pathways.

Several studies have observed alterations in amino acid metabolism,^9,11 which we did not observe after applying our minimum 10% rule for overlapping features. Again, in the study by Ch et al, the authors observed alterations in glycine, serine, and threonine metabolism, tyrosine metabolism, arginine and proline metabolism, lysine metabolism, phenylalanine metabolism, tryptophan metabolism, aminoacyl-tRNA metabolism, and cyano amino acid metabolism. There are many potential explanations for the divergent results including differences in pesticides present in the US versus India, differences in the primary route (e.g. ingestion vs inhalation/dermal contact) and magnitude of exposure (e.g. occupational versus solely through diet), differences in the biological fluids collected and analyzed (e.g. urine and saliva vs serum), and differences in the bioinformatic tools used (Metaboanalyst vs Mummichog). In another study of 83 pregnant women in France, women were classified into 3 exposure groups based on the percentage of land that was dedicated to agricultural crops in their town of residence. Nuclear magnetic resonance-based metabolomics analyses were then performed on urine samples provided in early pregnancy. These authors found that women with higher estimated exposure to a complex mixture of pesticides had altered glycine and threonine metabolism.⁹ Differences between the results from the study from France and our study may be due in part to differences in the pesticides used in the US versus France and how pesticide exposure was defined between the two studies. For example, many pesticides commonly detected on fruits and vegetables in the US such as atrazine, dacthal, and chlorpyrifos are banned or not approved for use in France (and the rest of the European Union).³⁹ Furthermore, in the study from France, cereal crops were the main agricultural product being grown in the region of study while in our study FVs were the focus; these two products have different pesticide profiles, which may have resulted in divergent findings.³⁹ An additional difference worth noting is the analytical methods used between the two studies because NMR tends to be less sensitive than LC-MS.

In addition to the pathway analysis, we identified and confirmed 4 metabolites associated with high and low-to-moderate pesticide FV intake using level-1 evidence including mannitol, cortisol, gluconic acid, and methyl-histidine. Mannitol is a common sugar than is often found in fruits and vegetables so the negative association with high pesticide residue FV intake warrants further investigation. In one study of intestinal epithelial cell lines exposed in vitro to glyphosate, a common herbicide, increased cell permeability was observed and mannitol permeability increased.⁴⁰ It is possible therefore, that high pesticide residue FV intake may be affecting cell permeability. However, if this was the case we would expect mannitol to have a positive association with high pesticide residue FVs. Instead, it is possible that the body is expending more energy to reduce the negative impact of pesticide residue and is therefore utilizing more mannitol as an energy source which would explain its negative association. In either case, understanding why high pesticide residue FVs are negatively associated with mannitol requires further study. Cortisol, unlike the previous metabolite, was positively associated with high pesticide residue FV intake indicating that as consumption of high pesticide residue FVs increases cortisol also increases. Cortisol is a stress hormone, a marker of inflammation, and has been previously linked to pesticide exposure in both humans and animals.^41,42 This may indicate that pesticide residue is stressing the body and may also explain the heightened demand for energy and vitamin metabolism (as observed in the pathway analysis and with the negative association with mannitol). Additionally, cortisol has been linked to early pregnancy loss⁴³ which may indicate that cortisol is a potential mediator of the association between high pesticide residue FVs and pregnancy loss which was previously observed in this sample.²⁰ Energy and vitamins in this instance may be necessary for the body to mitigate and repair damage due to oxidative stress created by higher intake of pesticide residues.

The identification of gluconic acid as being negatively associated with low-to-moderate pesticide FV intake is interesting and validates some of our findings given that energy metabolism pathways were commonly associated with low-to-moderate pesticide FV intake. Additionally, gluconic acid has previously been linked to low chronic levels of cadmium and chlorpyrifos in the brains of rat.⁴⁴ Finding gluconic acid in multiple matrices across multiple species may indicate its utility as a biomarker for low levels of pesticide exposure but this will need to be confirmed with additional biomarker studies. Methyl-histidine, was negatively associated with low-to-moderate pesticide residue FV intake. Methyl-histidine has been tentatively linked to several health effects as well as a marker of muscle protein turnover.^45–47 It could be possible therefore, that individuals consuming more low-to-moderate pesticide FV intake have lower meat consumption thus decreasing this metabolite. Therefore, this may be a marker of higher FV intake and lower meat consumption but again further biomarker studies are needed to confirm such an association.

4.3. Strengths & Limitations.

We recognize that our study has several limitations. First, our sample only included women undergoing infertility treatment at a fertility clinic in the Northeastern US, which likely affected the generalizability of our study. The majority of the women included were White and highly educated, which is typical of studies focusing on infertility clinic populations but may limit the generalizability of our findings to the broader population. However, we do not expect that our sample selection influenced either our exposure or outcome in this study. On average women in our sample had a higher intake of fruits and vegetables than the average American woman,⁴⁸ but women in our sample had a wide range of fruit and vegetable intake and the higher average likely reflects higher socioeconomic status. Additionally, most women (70%) had their blood samples collected prior to receiving fertility treatment (e.g. controlled ovarian stimulation). Thus, it is highly unlikely the serum metabolome results were materially altered by this selection criteria. Additional studies that focus on populations outside of an infertility clinic setting should be undertaken to confirm these results. Of note, we did observe many of the same pathways, including energy¹¹ and lipid metabolism pathways,^7,10,16 as previous pesticide studies which offers some consistency for our results. Second, we did not directly measure pesticide intake but rather relied on a proxy measure, the PRBS which is estimated based on national pesticide surveillance data and individual diet intake data rather than actual measured pesticide residue in food consumed by the participants. In addition, questions on fruit and vegetable intake did not separate raw from cook food consumption, thus increasing the likelihood of exposure misclassification. Indirectly measuring pesticide levels from diet may not fully capture actual pesticide levels consumed and this may distort the relationship between pesticide exposure from diet and metabolomic changes. However, previous studies have validated the PRBS and found good reliability when compared to urinary biomarkers of pesticide intake.¹⁷ Future studies that utilize more direct quantification of pesticide exposure from diet in various populations and settings will be needed to confirm the results of this study. In addition, we only measured pesticide residue intake and the serum metabolome at a singular timepoint which means reverse causation is a possibility but unlikely as metabolomic pathways are unlikely to influence individuals to consume more or less high and low-to-moderate pesticide residue FVs. Additionally, because we were only able to measure the serum metabolome at a single timepoint, we were unable to assess variation within an individual and this especially concerning when assessing non-persistent pesticides. However, we did analyze samples in triplicate and removed features that had low correlation or high coefficient of variation. Future studies should consider multiple samples to better examine intra-individual variability. Third, while we were only interested in examining pesticide residue burden from FVs in our analysis, other dietary sources of pesticide exposure were likely present and could have contributed to the measurement error of our exposure. However, the Food and Drug Administration has observed that domestic fruits and vegetables are most often the food items with detectable pesticide residue compared to other food groups.⁴⁹ Therefore, we are confident that by focusing on fruits and vegetables we are capturing the highest sources of pesticide exposure in subjects’ diet. In addition, we were unable to account for other chemical exposures from food packaging (e.g. bisphenol A) or other sources. However, these are unlikely to be highly correlated with high or low-to-moderate FV intake and therefore unlikely to be a confounder. Future studies examining the effects of these environmental chemicals on the serum metabolome will enable us to further evaluate the specificity of these metabolites and pathways. Fourth, residual confounding is possible due to unknown or unmeasured confounders. However, we did adjust for many known confounders including dietary patterns, supplement intake, smoking status, BMI, and age. Fifth, we did not require that the women fast prior to blood draw, which may have had an impact on the metabolomics results. To minimize this impact, we applied a comprehensive metabolomics workflow, which we and others have successfully applied to the analysis of non-fasting samples, using pooled standards and internal references. Sixth, we were able to identify only 4 metabolites using level-1 evidence (based on authentic standards from our laboratory) which means that 90-95% of the significant features associated with pesticide residue FV intake were not identified. Metabolomics is a newer field with less features being positively identified when compared to other omics fields. To address this limitation, we provided the ten most significant features associated with pesticide residue FVs that may help future identification of these metabolites (Supplemental Table 3–4). Lastly, we were unable to use the FDR corrected p-values due to there being too few significant metabolites after adjustment which means our results should be interpreted with caution. It is likely that limited sample size and non-differential measurement error of our exposure and outcome measures reduced our power and precluded the use of the FDR correction. Instead, we attempted to safeguard against false positives by using a more conservative p-value for significant metabolites (p-value<0.005). Our study does have several strengths. First, we utilized both a validated measure of pesticide residue burden in food¹⁷ and a validated measure of food intake.²³ Second, we used standard protocols to extract metabolomic features from serum samples. Finally, we used validated standards to confirm, using level-1 evidence, the identity of several metabolites in the serum metabolite associated with high and low-to-moderate pesticide burden fruits and vegetables.

5. CONCLUSIONS

In this study of women from a fertility clinic, we identified several unique metabolomic pathways associated with dietary intake of high pesticide residue FVs and these pathways had limited overlap with intake of low-to-moderate FVs. These results may provide valuable insight into the potential mechanisms underlying associations between dietary intake of pesticides and increased risk of human disease. Additionally, our findings could be used to identify potential biomarkers of high pesticide residue FV intake; however, this will require additional studies to confirm and validate our results.

HIGHLIGHTS.

• Untargeted metabolomics was used to describe alterations in the serum metabolome.

• Pathways associated high residue FVs include energy, vitamin, and enzyme metabolism.

• Pathways associated low residue FVs include energy, cellular receptor, and xenobiotic metabolism.

• Four unique metabolites were identified with level-1 evidence.

ABBREVIATIONS:

PRBS

Pesticide residue burden score

FFQ

Food frequency questionnaire

FV

Fruits and vegetables

DATA AVAILABILITY STATEMENT

Data are not available.