Concurrent associations of glyphosate, dithiocarbamate fungicides, and PFAS with body fat composition and BMI in adolescents from agricultural communities

Abstract

Background:

Dithiocarbamate fungicides, glyphosate, and per- and polyfluoroalkyl substances (PFAS) are chemicals found in agricultural settings. These chemicals have been linked to anthropometric indicators, but findings are contradictory. This study evaluated associations of PFAS, glyphosate, and dithiocarbamate metabolites (ethylene thiourea [ETU] and propylene thiourea [PTU]) with anthropometric indicators among adolescents in agricultural settings.

Methods:

We examined 535 adolescents in agricultural communities in Ecuador (ESPINA study) in 2016. Concentrations of urinary pesticides, 6 serum PFAS, and all co-variates of interest were measured in 507 participants. Associations between chemicals and anthropometric measures were calculated using linear regression, and chemical mixtures were assessed using weighted quantile sum regression (WQS). Models adjusted for sex, race, parental education, and number of smokers at home.

Results:

Glyphosate had a significant negative association with body mass index (BMI)-for-age z score (difference per a 2-fold increase of chemical concentration (β) = −0.06 SD [95 % CI: −0.12, −0.02]) and % body fat (β = −0.31 % [95 % CI: −0.58, −0.04]). ETU also had a significant negative association with BMI-for-age z score (β = −0.07 SD [−0.12, −0.01]). The mixture of chemicals was also negatively associated with BMI-for-age z scores (difference per a decile increase in the mixture (β_WQS): −0.10 SD [−0.14, −0.05]) and % body fat (β_WQS: −0.44 % [−0.72, −0.17]).

Conclusions:

Glyphosate, and to an extent ETU, had negative associations with BMI and % body fat in adolescents in these cross-sectional analyses. The mixture of all 4 chemicals had stronger associations with these outcomes compared to any individual chemical. Longitudinal analyses are warranted. These exploratory findings may inform future hypothesis-driven research on pesticide exposures and adolescent body composition.

1. Introduction

Many types of chemicals, including herbicides, insecticides, fungicides and per- and poly-fluoroalkyl substances (PFAS), are used in agriculture. Many of these pesticides, including fungicides like dithiocarbamates and herbicides like glyphosate, are frequently used, but their associations with health outcomes, including child development are understudied. Furthermore, PFAS are nearly ubiquitous in humans due to their persistence in the environment and their various uses, including in non-stick coatings, water-resistant textiles, firefighting foams, personal care products and more. PFAS are also often used in agriculture as pesticides, such as sulfuramid, or to prolong the effectiveness of other pesticides (Kemi, 2015; Wang et al., 2017). There is a growing body of literature showing how these pesticides and persistent pollutants can act as endocrine disrupters (Coperchini et al., 2017; Muñoz et al., 2021). However, limited data are available on the anthropometric effects of their exposure among children and adolescents in agricultural settings who may have chronically elevated exposure.

The current evidence on the associations of PFAS with anthropometric measures is contradictory. A review by Frangione et al. revealed no effect of prenatal PFAS exposure on pediatric obesity, while postnatal exposure was inversely associated with obesity (Frangione et al., 2024). Braun et al. reported greater adiposity in 8-year-old children born to mothers with higher PFOA levels and greater body mass index (BMI) gains in children aged 2-to-8 years with higher serum PFOA (Braun et al., 2016). However, other studies have found no associations between prenatal PFAS and anthropometric indicators (Andersen et al., 2013; Bloom et al., 2022). Many studies on postnatal PFAS exposure have reported inverse associations with anthropometric measurements (Canova et al., 2021; Fassler et al., 2019; Lee et al., 2021; Papadopoulou et al., 2021; Pinney et al., 2019). Conversely, a study of a large representative sample of US adolescents examined between 1999 and 2012 revealed positive associations between serum PFAS concentrations and BMI (Geiger et al., 2021). Finally, two studies found no association between early childhood PFAS exposure and adult obesity (Barry et al., 2014; Domazet et al., 2016).

Few studies have measured glyphosate in relation to anthropometric measures. In a representative sample of U.S. adults, higher urinary concentrations of glyphosate were associated with a greater likelihood of obesity (Li et al., 2023). In children living in agricultural settings in California, USA, urinary glyphosate residues were positively associated with a large waist circumference at 18 years (Eskenazi et al., 2023). Waist circumference is an independent anthropometric measure that is a marker of visceral adiposity (Brambilla et al., 2006). However, findings from the same Indiana-based cohort of high-risk pregnancies have been mixed: Parvez et al. (2018) reported no association between prenatal urinary glyphosate and fetal growth indicators, whereas Gerona et al. (2022) later observed an inverse association when glyphosate was measured earlier in pregnancy (Gerona et al., 2022; Parvez et al., 2018).

Ethylene thiourea (ETU) is an impurity, metabolite, and environmental decomposition product of ethylene bis-dithiocarbamate (EBDC) fungicides (i.e., mancozeb, maneb, metiram, nabam, zineb) and is widely measured in urine as a marker of EBDC exposure. Propylene thiourea (PTU) is the metabolite and degradation product of propylene bis-dithiocarbamate (PBDC) fungicides (i.e., propineb). Currently, limited data have been published on the associations of bis-dithiocarbamate fungicides with anthropometric measures. The only existing study on this topic, to our knowledge, found that prenatal urinary concentration of ETU in pregnant women in Costa Rica was not associated with birth weight or body length in their offspring (van Wendel de Joode et al., 2024). To our knowledge, no studies have evaluated the associations between propylene thiourea (PTU) and anthropometric indices. Studies characterizing these associations in other age groups are needed.

This study aimed to examine the independent and combined effects of exposure to glyphosate, ETU, PTU, and PFAS on anthropometric indicators among adolescents living in a floricultural and agricultural region of Ecuador. Prior research has reported limited and inconsistent associations between these chemical exposures and anthropometric outcomes, underscoring the need for further investigation, particularly among children and adolescents in agricultural communities, where pesticide exposures are substantially higher than in the general population. In this context, we evaluated the potential health effects of co-exposure to PFAS, an herbicide (glyphosate), and a fungicide (ETU), given their concurrent use and presence in agricultural settings. We specifically included PFAS (PFOA and PFOS) in our analyses because they have been documented as co-formulants in pesticide products and have been detected in agricultural soils and water (Donley et al., 2024; Libenson et al., 2024; Mandal et al., 2025). For example, approximately 14 % of pesticide active ingredients in the U.S. are PFAS (Donley et al., 2024), and growing evidence of PFAS contamination in treated fields has raised concerns about their role as adjuvants in pesticide formulations. Reflecting these concerns, the U.S. EPA recently removed multiple PFAS compounds from its list of approved pesticide inert ingredients (United States Environmental Protection Agency, 2021). Together, these findings support the plausibility and importance of assessing the health impacts of co-exposure to PFAS and commonly used pesticides in farming communities.

2. Methods

2.1. Study description and participants

The present study analyzes cross-sectional data of the follow-up year 8b examination conducted in July–October 2016 of participants of the study of Secondary Exposures to Pesticides among Children, and Adolescents (ESPINA). ESPINA is a prospective cohort of participants living in the agricultural county of Pedro Moncayo, Pichincha, Ecuador, that started in 2008. During initial recruitment in 2008, 313 children aged 4–9 years were examined and recruited through their participation in the “2004 Survey of Access and Demand of Health Services in Pedro Moncayo County” (n = 228, 73 %), a representative survey of Pedro Moncayo County collected by Fundación Cimas de Ecuador in conjunction with local communities. Participant recruitment for ESPINA has been previously described (Suarez-Lopez et al., 2012).

In 2016, two examinations were conducted—330 participants in April and 535 in July–October —for a total of 554 participants examined in 2016 (311 in both time periods). Participants in 2016 were 12–17 years of age and included 238 participants examined in 2008 and 316 new volunteers. New participants in 2016 were invited to participate using the System of Local and Community Information (SILC) developed by Fundación Cimas del Ecuador. In 2008 and 2016, none of the participants reported working in agriculture. Of the 535 participants in the July–October 2016 examination, 507 had complete sets of anthropometric outcomes, covariates, serum PFAS and urinary pesticide data, and were included in the current analysis. Fig. S1 shows a flowchart of the participants.

2.2. Data collection overview

During the examination period from July to October 2016, children were assessed at seven schools in Pedro Moncayo County while the schools were not in session. The examiners were unaware of the participants’ potential exposures. Home interviews were conducted with parents and adult residents to collect socio-economic and demographic data. The 2016 examination of the ESPINA study was approved by the Institutional Review Boards of the University of California San Diego (UCSD), Universidad San Francisco de Quito and the Ministry of Public Health of Ecuador and endorsed by the Commonwealth of Rural Parishes of Pedro Moncayo County. We collected informed consent from adult participants (aged 18 years or older) and parents, as well as parental permission of participation and assent of child participants.

2.3. Anthropometric measures

Participants’ height was measured using a height board according to the recommended guidelines (World Health Organization, 2008), and their weight and % body fat was recorded with a digital scale (Tanita BF-689; Tanita Corporation of America, Arlington Heights, IL, USA). We computed z scores for height-for-age and BMI-for-age based on the World Health Organization growth standards (World Health Organization, 2006). All anthropometric measurements were taken by trained personnel using standardized protocols, and equipment was calibrated daily. Duplicate measurements were taken when anomalies or outliers were observed to minimize misclassification. BMI was calculated as weight (kg) divided by height squared (m²). We computed z scores for height-for-age and BMI-for-age based on the World Health Organization growth standards (World Health Organization, 2006). Z scores were calculated using WHO AnthroPlus software, which normalizes for age and sex based on WHO reference populations for children and adolescents aged 5–19 years (Onis et al., 2007; World Health Organization, 2006).

2.4. Collection of biospecimens

Urine samples were collected for each participant on the same day as their examination. Urine samples were collected by the participants upon awakening and were brought to the examination site, where they were aliquoted and frozen at −20 °C.

Blood was obtained through venipuncture of the median cubital or cephalic veins in the arm. On-site processing followed the collection, with the extracted serum being aliquoted and frozen on dry ice in coolers. Urine and blood samples were subsequently transported at the end of the day to Quito, Ecuador and stored at −70 °C.

Using a specialized courier, the frozen urine and serum samples were then shipped from Quito to UCSD, where they were stored long-term at −80 °C. Throughout transportation, the samples remained frozen. Serum samples were subsequently shipped frozen from UCSD to the Minnesota CHEAR Lab for PFAS quantification. Urine samples were shipped frozen to the Laboratory for Exposure Assessment and Development in Environmental Research at Emory University (Atlanta, GA) for the quantification of creatine, glyphosate, PTU and ETU.

2.5. . PFAS quantification in serum

The analysis of PFAS in serum was performed through protein precipitation followed by centrifugation, concentration, and analysis using liquid chromatography-tandem mass spectrometry (LC/MS/MS). In brief, a stable isotope-labeled internal standard solution and acetonitrile were added to a 400 μl aliquot of serum in microcentrifuge tubes. The tubes were mixed and centrifuged, and the supernatant was transferred to 96-well plates for concentration. After concentration, the sample was analyzed by LC/MS/MS (Shimadzu Prominence coupled to AB Sciex 5500Q) using electrospray ionization with multiple reaction monitoring (MRM) and retention time windows. Bovine calf serum was used to prepare matrix-matched calibration curves, and quantitation was performed via isotope dilution. A minimum of 2 MRM transitions are monitored for each analyte (when possible), and unknowns are verified through retention time and ion ratio matching with the calibration curve. A minimum of 3 quality control (QC) samples were run with each batch of 20 unknowns, a method blank, and low and high concentration levels. The quality control materials were prepared in bovine calf serum. An additional QC is run if an unknown sample requires dilution, either due to a low volume of sample or an initial concentration higher than the calibration curve allows. Batches with a QC recovery outside of 70–130 % were reanalyzed; in cases where a QC failure could not be resolved through reinjection/reanalysis, the resulting unknown data were qualified to reflect this result. The measured PFAS chemicals included perfluorohexane-1-sulfonic acid (PFHXS), perfluorononanoic acid (PFNA), perfluorooctanesulfonamide (PFOSA), N-ethylperfluoro-1-octanesulfonamide (ETFOSA), perfluorooctanoic acid (PFOA), and perfluorooctanesulfonic acid (PFOS). The limits of quantification (LOQs) are reported in Table 2.

Table 2.

Distribution of serum PFAS and urinary pesticide metabolite concentrations for participants included in the present analyses (N = 507).

Analyte	N above LOQ or LOD (percent)	N above LOD including machine- derived valuesa (percent)	LOD	LOQ	Geometric Mean (SD)	Min	Percentiles					Max
							10th	25th	50th	75th	90th
ETFOSA, ng/mL	40 (7.9)	NA	NA	0.100	NC	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	0.430
PFHXS, ng/mL	14 (2.8)	NA	NA	0.100	NC	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	0.170
PFNA, ng/mL	155 (30.6)	NA	NA	0.100	NC	<LOQ	<LOQ	<LOQ	<LOQ	0.110	0.140	2.10
PFOA, ng/mL	507 (100)	NA	NA	0.100	0.32 (1.45)	0.11	0.20	0.25	0.32	0.40	0.51	1.20
PFOS, ng/mL	507 (100)	NA	NA	0.100	0.54 (1.66)	0.12	0.29	0.39	0.54	0.72	1.00	3.70
PFOSA, ng/mL	0 (0)	NA	NA	0.100	NC	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ	<LOQ
ETU, ng/g of creatinine	413 (81.5)	458 (90.3)	0.625	NA	1.65 (2.39)	<LOD	<LOD	0.88	1.72	2.95	4.57	29.49
PTU, ng/g of creatinine	51 (10.1)	368 (72.6)	1.625	NA	0.92 (2.35)	<LOD	<LOD	<LOD	<LOD	<LOD	2.67	104.01
GLY, ng/g of creatinine	441 (87.0)	507 (100)	0.250	NA	0.82 (5.38)	<LOD	0.25	0.53	0.98	1.90	3.38	21.94

LOD = Limit of Detection; LOQ = Limit of Quantification. Pesticide biomarkers (ETU, PTU, Glyphosate) were measured using LOD thresholds, while PFAS biomarkers (PFOA, PFOS, PFNA) were measured using LOQthresholds; NC: not calculated; GLY: glyphosate; ng/mL: nanograms per milliliter.

^aValues below the LOD were imputed using the machine-derived concentration (where available) or LOD/√2.

2.6. Glyphosate quantification in urine

Aliquots of 250 μL of urine were spiked with isotopically labeled glyphosate. These aliquots were then diluted to 1 mL using doubly deionized water and extracted through C18 solid-phase extraction (SPE). Glyphosate was derivatized to form its heptafluorobutyl analogue, which was then concentrated for subsequent analysis. Prior to assessing urinary glyphosate levels, all aliquoted urine samples were randomized using a Fisher-Yates shuffling algorithm with the aim of minimizing potential batch effects (Fisher and Yates, 1948; Knuth, 1960). Analysis of the concentrated extracts of the heptafluorobutyl analogue was performed with gas chromatography–mass spectrometry with electron impact ionization in multiple ion monitoring mode. The limit of detection (LOD) was 0.25 ng/mL, with a relative standard deviation (RSD) of 3 %.

2.7. ETU and PTU quantification in urine

All samples were randomized using a Fisher-Yates shuffling algorithm prior to analysis to reduce any potential batch effects (Fisher and Yates, 1948; Knuth, 1960). An 800 μL aliquot of urine was spiked with isotopically labeled internal standards and 50 μL of 2.2 N hydrochloric acid. The samples were then extracted using an Isolute solid–liquid extraction (SLE) cartridge. All eluates were concentrated to dryness and reconstituted with 100 μL of 0.1 % formic acid. Extracts were analyzed using liquid chromatography–mass spectrometry (LC/MS) with electrospray ionization. ETU and PTU and their labeled analogs were measured using multiple ion monitoring mode (Montesano et al., 2007). The concentrations of ETU and PTU were determined using isotope dilution calibration. The LOD for ETU was 0.625 ng/mL, for PTU it was 1.625 ng/mL, and had an RSD of 7 %.

2.8. Creatinine quantification in urine

Urinary creatinine concentrations were measured and used to correct urinary metabolite concentrations to account for urinary dilution (Carrieri et al., 2000). To quantify urinary creatinine, a 10-μL aliquot of the sample was diluted and analyzed via high-performance liquid chromatography coupled with tandem mass spectrometry with electrospray ionization (Kwon et al., 2012). The LOD was 5 mg/dL, with an RSD of 7 %.

2.9. Imputation for values below the LOD

For the PFASs, the limits of quantification (LOQs) reported were the average of the daily LOQs for the whole study. The daily LOQs were determined across different analytical runs, as slight batch-to-batch variations occurred in instrumental sensitivity during the study period. The daily LOQs were used to identify the valid values for each sample. A surrogate value, LOQ/√2, was reported for measures below the LOQ since machine-read values were not available for measures below the LOQ. There was a single participant with a PFOA measurement below the LOQ and no participants with PFOS measurements below the LOQ. For pesticide analytes, the LODs were reported in the same manner. Trace concentrations below the LOD that had machine-read values were considered for analysis of pesticides since machine-read values were available for all measures, including the values that resulted in zero. There were 310 (84.2 %) observations for PTU, 45 (9.8 %) observations for ETU, and 66 (13.0 %) observations for glyphosate below the LOD. To account for the right skew of these environmental data, a log2 transformation was applied to all exposure concentrations. Due to the low detectability (<50 %) of many of these environmental chemical concentrations, we focused our assessments on PFOS, PFOA, glyphosate, and ETU (Table 2), each of which had a detectability above 80 %.

2.10. Statistical analysis

Participant characteristics.

This study utilized cross-sectional data from the ESPINA examination, which was conducted between July and October 2016. We computed medians and interquartile ranges (IQRs) for various participant characteristics (Table 1). Categorical variables were analyzed using column percentages to establish model covariates. Associations between covariates and log2-transformed glyphosate, ETU, and PFOA concentrations were tested using t-test, ANOVA or linear regression, as appropriate.

Table 1.

Descriptive statistics of the study population (N = 507).

Covariates	N (% of total), or Mean (SD)	Glyphosate (μg/g creatinine)		ETU (μg/g creatinine)		PFOA (ng/mL)
		Median (Q1, Q3), or Pearson Correlation Coefficient	p valuea	Median (Q1, Q3), or Pearson Correlation Coefficient	p valuea	Median (Q1, Q3), or Pearson Correlation Coefficient	p valuea
Age, years	14.5 (1.8)	−0.16	<0.001	−0.03	0.592	−0.13	0.003
Gender			0.142		0.540		<0.001
Female	259 (51 %)	0.84 (0.43, 1.80)		1.87 (0.99, 3.07)		0.29 (0.23–0.37)
Male	248 (49 %)	1.05 (0.60, 2.04)		1.86 (1.04, 3.19)		0.34 (0.27–0.41)
Ethnicity			0.055b		0.999		<0.001
Indigenous	112 (22 %)	1.40 (0.76, 2.42)		2.08 (1.19, 3.15)		0.26 (0.22–0.33)
Mestizo	394 (78 %)	0.87 (0.47, 1.72)		1.79 (0.94, 3.08)		0.33 (0.26–0.40)
White	1 (0.2 %)	0.51 (0.51, 0.51)		2.72 (2.72, 2.72)		0.28 (0.28–0.28)
Number of smokers at home			0.804			0.276	0.875
0	397 (77 %)			1.86 (1.00–3.01)		0.32 (0.25–0.40)
1	115 (23 %)			1.93 (1.07–3.56)		0.31 (0.25–0.40)
Parental education, years	8.1 (3.5)	−0.12	0.009	−0.05	0.307	0.27	<0.001
Body fat, %	22.3 (7.3)	−0.10	0.027	−0.01	0.895	−0.14	0.002
Height-for-age z score, SD	−1.5 (0.9)	−0.05	0.233	−0.04	0.391	0.12	0.005
BMI-for-age z score, SD	0.4 (0.8)	−0.06	0.200	−0.06	0.212	−0.11	0.012

Abbreviations: SD: standard deviation, Q1: 25th percentile cutoff, Q3: 75th percentile cutoff, BMI: body mass index.

^aAssociations between covariates and glyphosate, ETU, and PFOA concentration were tested using t tests, ANOVA, or linear regression, as appropriate.

^bStatistical comparison of glyphosate across ethnicity categories excludes ‘White’ due to the small sample size.

Outcomes.

The main anthropometric variables included BMI-for-age z scores, % body fat, and height-for-age z scores. Each chemical exposure-outcome hypothesis was assessed using appropriate linear contrasts. The distribution of each outcome was assessed graphically, and symmetry was maintained, thereby supporting assumptions of normality.

Covariates.

All models were adjusted for covariates selected a priori, which included sex (male vs female) [Ref. = Female], race (indigenous vs mestizo vs white) [Ref. = Indigenous], parental education (categorized to ≤6 years, 7–12 years, and ≥12 years) [Ref. = parental education ≤6 years] and number of smokers in the household [Ref. = nonsmokers]. These covariates were selected based on prior evidence of their associations with both environmental exposures and anthropometric outcomes (Fig. S2) in adolescents (Han et al., 2024; Zhang et al., 2023). ETU, PTU and glyphosate were corrected for creatinine by dividing the wet weight concentration for urinary creatinine * 100 (μg/g of creatinine).

Single chemical analysis.

Single chemical multivariable linear regression models were run for each chemical to estimate the associations with linear outcomes, including BMI-for-age z scores, % body fat, and height-for-age z scores. False discovery rate (FDR) correction was applied to p values for each analysis to account for multiple comparisons (Glickman et al., 2014). Specifically, for each outcome, raw p-values from the five single-chemical models (glyphosate, ETU, PTU, PFOA, PFOS) were ranked, and q-values were computed using the Benjamini–Hochberg procedure. A q-value <0.05 was considered statistically significant. This approach controls the expected proportion of false positives and is recommended in environmental health studies involving multiple comparisons (Bai et al., 2022).

Chemical mixture analyses.

Repeated holdout validation for the weighted quantile sum regression (WQS_rh) approach was utilized to characterize the variability, improve the stability of WQS estimates, and assess the relationship between the outcome and WQS index estimated from quantiled exposure concentrations. Furthermore, this method combined cross-validation using repeated holdouts and bootstrap resampling, accounting for the complex correlation pattern between chemical exposures and further reduced dimensionality of the exposure (Carrico et al., 2015) (Fig. S3). This validation technique is an important tool used in predictive modeling to assess the generalizability and evaluate the replicability of results. In this case, the exposed chemicals were deciled based on the best fit from visual inspection of the data using different quantiles (i.e., quartiles, deciles, and percentiles). This strategy randomly selects 40 % of the participants into the training set to estimate weights and uses the remaining 60 % as a validation set to test the association of the weighted index on the outcome of interest. Specifically, the repeated holdout version of WQS repeats this process of randomly splitting the data 100 times (100 bootstraps per repetition for robustness), resulting in a distribution of effect estimates and weight estimates (Tanner et al., 2019). The models were run to assess associations constrained to either the positive or negative directions to assess mixtures that may increase or decrease the overall outcome. When more than 97.5 % of the estimates are above or below the null, the association is significant. Using the Busgang criteria (Busgang et al., 2022) for interpreting the weights, when 90 % of weight contributions are above the threshold of concern (defined as 1/c, where c is the total number of chemicals), the specified chemical is defined as a probable contributor to the mixture effect. When 50–90 % of the weight contributions are greater than the threshold of concern, the chemical is defined as a possible contributor. When 90 % of the weight contribution is below the threshold of concern, the chemical is defined as a probable noncontributor, and if 50–90 % of the weight contribution is less than the threshold, it is considered a possible noncontributor (Bennett et al., 2022). Given the likelihood of co-exposure in agricultural settings, PFAS and EBDC fungicides were included together in the mixture analysis. Both chemical classes have been implicated in endocrine and metabolic disruption (DeWitt, 2015; Mora et al., 2018), with EBDCs affecting thyroid hormone synthesis (Maranghi et al., 2013) and PFAS interfering with lipid metabolism and adiposity regulation (Liu et al., 2018). These shared biological pathways justify their joint evaluation in relation to BMI during adolescence. Statistical significance was defined as a p-value <0.05, or a FDR–adjusted q-value <0.05, where applicable.

3. Results

3.1. Participant characteristics

Participant characteristics are shown in Table 1. The mean age was 14.5 years; 78 % of the participants were mestizo, 22 % were indigenous, 51 % were female, and the mean body fat composition was 22.3 % (SD: 7.3). Seventy-seven percent of the children lived in households with no smokers. The mean parental education level was 8.1 years (SD: 3.5). The mean BMI-for-age z score was 0.4 (SD: 0.08), and the mean height-for-age z score was −1.5 (SD: 0.9). Fewer than 0.5 % of the participants (i.e., only 1 individual) exhibited thinness, as indicated by a BMI-for-age z score below −2. In contrast, 2.2 % (i.e., 11 participants) were classified as overweight, with BMI-for-age z scores exceeding 2. Additionally, stunting—defined by a height-for-age z score below −2—was observed in more than one quarter of the participants (143 individuals, or 28.20 %), and no one had a height-for-age z score above 2. Overall, indigenous participants had higher urinary glyphosate concentrations than mestizo race participants. In these unadjusted models, all the anthropometric indicators were negatively correlated with urinary glyphosate. In addition to glyphosate, ETU and PFOA concentrations showed similar directional trends—primarily inverse correlations—with several anthropometric measures, although not all associations were statistically significant (Table 1).

Table 2 shows the detection rates of each chemical measured. Only chemicals with detection rates exceeding 80 % were included in the subsequent analysis. These chemicals included PFOA and PFOS, with 100 % detection rates, and glyphosate and ETU, with 87 % and 82 % detection rates, respectively. The distributions of PFAS, glyphosate, and ETU are presented in Figure S4 (A–E) as histograms with a log2 transformation to account for the right-skewed data.

3.2. Single chemical associations with BMI-for-age, % body fat, and height-for-age z scores

All four chemicals were found to have negative associations with BMI-for-age and % body fat. Glyphosate was statistically significantly negatively associated with BMI-for-age z scores and % body fat after FDR correction for multiple comparisons (difference per 2-fold increase in chemical concentration β = −0.06 SD [95 % CI: −0.10, −0.02]; corrected p value = 0.02 for BMI-for-age z scores; and β = −0.31 SD [95 % CI: −0.58, −0.04]; corrected p value = 0.04 for % body fat; Table 3). ETU also had significant negative associations with BMI-for-age z scores (β = −0.07 [−0.12, −0.01]; corrected p value = 0.04) and a borderline nonsignificant negative association with body fat composition. PFOA had a borderline nonsignificant negative association with BMI-for-age z scores. None of the chemicals had statistically significant associations with height-for-age z scores. Sensitivity analyses using uncorrected pesticide concentrations with urinary creatinine included as a covariate yielded similar effect estimates and levels of statistical significance. For example, glyphosate remained inversely associated with BMI-for-age z-score (β = −0.06, p = 0.01 with creatinine correction vs. β = −0.06, p = 0.009 with creatinine as a covariate). Associations for ETU and other metabolites also remained consistent across both approaches, indicating the robustness of our findings to the method of urinary dilution adjustment. To evaluate potential effect modification by sex, we included interaction terms between each pesticide biomarker and sex in fully adjusted models. No statistically significant interactions were observed after FDR correction (all p-values >0.05), indicating that associations were similar in males and females. In exploratory analyses, PFNA was dichotomized at the limit of detection (LOD), but no significant associations with BMI, percent body fat, or other anthropometric outcomes were observed in any model specification (data not shown).

Table 3.

Differences in anthropometric measures associated with a 2-fold increase of a given pesticide (n = 507).

	Height-for-age z score, SD			BMI-for-age z score, SD			Body fat composition, %
	Difference (95 % CI)	p value	FDR-corrected p value	Difference (95 % CI)	p value	FDR-corrected p value	Difference (95 % CI)	p value	FDR-corrected p value
PFOA	0.10 (−0.06, 0.25)	0.23	0.32	−0.14 (−0.28, 0.00)	0.05	0.09	−0.33 (−1.21, 0.55)	0.47	0.53
PFOS	0.05 (−0.06, 0.16)	0.40	0.46	−0.04 (−0.14, 0.06)	0.45	0.53	−0.05 (−0.67, 0.57)	0.88	0.88
ETU	−0.05 (−0.11, 0.01)	0.12	0.17	—0.07 (−0.12, −0.01)	0.02	0.04	−0.32 (−0.67, 0.04)	0.08	0.11
PTU	−0.01 (−0.07, 0.06)	0.79	0.79	−0.02 (−0.08, 0.04)	0.55	0.55	−0.09 (−0.45, 0.27)	0.63	0.68
GLY	−0.01 (−0.05, 0.04)	0.80	0.80	—0.06 (−0.10, −0.02)	0.01	0.02	—0.31 (−0.58, −0.04)	0.03	0.04

Models adjusted for gender, race, parental education, and number of smokers in the household. ETU and GLY were creatinine corrected. FDR correction was applied per outcome across five exposures.

3.3. Chemical mixture associations with BMI-for-age, % body fat, and height-for-age z scores

For all the outcomes, WQS_rh regression was conducted to constrain the inference in each direction. There was no significant positive effect of the exposure mixture with BMI-for-age z score (Table 4, Fig. S5). However, there was a significant negative association (100 % for the negative constraint) between the exposure mixture and BMI-for-age z score (difference per decile increase in the mixture [β_WQS]: −0.10 SD [95 % CI: −0.14, −0.05]; Fig. 1 [I-III]). ETU and glyphosate contributed the most to the negatively constrained chemical mixture, with median weights of 31 % and 29 %, respectively (Fig. 1 [I - III]). Both ETU and glyphosate were determined to be possible contributors based on the Busgang criteria (Busgang et al., 2022) detailed previously. Additionally, PFOA was determined to be a probable noncontributor, and PFOS was determined to be a possible noncontributor (Fig. 1 [II]). Similarly, there was a significant inverse association (100 % for the negative constraint) between the exposure mixture and % body fat. A one-decile increase in the exposure mixture was associated with a 0.44 % (−0.72, −0.17) decrease in % body fat, and this effect was driven by glyphosate and ETU, with median weights of 35 % and 31 %, respectively (Fig. 1 [II]). Both glyphosate and ETU were determined to be possible contributors (Busgang et al., 2022). Additionally, PFOA and PFOS were determined to be possible noncontributors (Fig. 1 [II]).

Table 4.

Differences in anthropometric measures in relation to a 1-decile increase in the pesticide mixture using WQS regression. Analyses using positively and negatively constrained β coefficients to estimate weights are presented.

Anthropometric differences per decile increase in the chemical mixture (95 % CI)
BMI-for-age z score, SD		Body fat composition, %		Height-for-age z score, SD
Positive constraint	Negative constraint	Positive constraint	Negative constraint	Positive constraint	Negative constraint
−0.05 (−0.10, 0.00)	−0.10 (−0.14, −0.05)	−0.25 (−0.51, 0.01)	−0.44 (−0.72, −0.17)	−0.01 (−0.05, 0.03)	0.00 (−0.04, 0.04)

Models adjusted for gender, race, parental education, and number of smokers in the household. The mixture used urinary creatinine-corrected metabolite concentrations of glyphosate, PTU and ETU.

Fig. 1.

Results from 100 repeated holdouts of the WQS_rh, of the pesticide mixture on anthropometric measures, using negatively constrained β coefficients to estimate weights. N = 507.

Data points represent the estimate for each of the 100 holdouts, and the closed diamond shows the mean effect estimate. The box plot shows the 25th and 75th percentiles, and the line within represents the 50th percentile. Whiskers show the 10th and 90th percentiles of weights for the 100 holdouts. * Median weight % (95 % CI). Data points depict the covariate-adjusted WQS_rh index in relation to the BMI-for-age z score, % body fat and height-for-age z score observations. The blue line is the LOESS fitted line with its 95 % CI (gray areas). Models adjusted for gender, race, parental education, and number of smokers in the household. The mixture used creatinine-corrected metabolite concentrations. WQS_rh β is the difference in anthropometric measures in relation to a 1-decile increase in the mixture.

There was no significant positive or negative association of the exposure mixture on height-for-age z score (Fig. 1 [I-III], Fig. S4 ([I-III] and Table 4) using WQS_rh regression.

4. Discussion

This is one of the first studies to characterize the anthropometric associations of pesticide and persistent pollutant exposures. As an exploratory study, our findings provide preliminary evidence of potential associations. We found that glyphosate concentrations, and to a lesser extent the ETU concentration, were inversely associated with BMI-for-age z scores and % body fat among adolescents living in agricultural settings, based on independent chemical modeling. In contrast, no associations were observed for PFAS concentrations. This was corroborated by chemical mixture modeling, which also showed an inverse relationship with BMI-for-age z scores and % body fat. In this analysis, glyphosate and ETU contributed the most to the association and were identified as potential contributors, whereas PFOS and PFOA were probable or possible non-contributors, respectively.

The concentrations of PFAS in ESPINA were nearly six times lower than among 12–19-year-old adolescents in the U.S. examined in 2011–2018 for PFOS (3.35 μg/L vs 0.54 μg/L) and PFNA (0.54 μg/L vs 0.09 μg/L), and about five times lower for PFOA (1.5 μg/L vs 0.3 μg/L) (Center for Disease Control, 2024). In that same population of U.S. adolescents, PFAS including PFOA and PFOS, were found to be significantly positively associated with BMI (Geiger et al., 2021). It is plausible that the relatively lower concentrations of PFAS in this Ecuadorian population may explain the lack of associations observed. However, multiple other studies assessing concurrent PFAS and anthropometric measures have also reported null or inverse associations (Canova et al., 2021; Fassler et al., 2019; Papadopoulou et al., 2021; Pinney et al., 2019), making it difficult to attribute causality to these findings. Notably, Lee et al. (2021), in a recent systematic review, concluded that current human evidence remains limited and inconsistent regarding PFAS-induced effects on physical development during childhood and adolescence (Lee et al., 2021). These inconsistencies may reflect differences in study populations, age ranges, timing of exposure and outcome assessments, and exposure levels—factors that are especially important given the relatively low PFAS concentrations observed in our study population. Our findings contribute to this nuanced body of literature by presenting data from an underrepresented adolescent population with comparatively low exposure levels.

ESPINA study participants in 2016 had 91 % higher concentrations of urinary glyphosate than did a representative sample of adolescents in the U.S. from 2013 to 2014 (geometric means [GM]: ESPINA: 0.84 μg/g creatinine [0.74 μg/L [wet weight (ww)]] vs 0.44 μg/g creatinine [0.41 μg/L (ww)], detected in 87 % of participants in both studies) (Ospina et al., 2022). Similarly, ESPINA participants in 2016 had 175 %–381 % higher concentration [0.77 μg/L specific gravity-corrected (sp. gr)] than adolescents in an agricultural region in California (16 μg/L (sp. gr) for 18 years, detected in 55 % of participants and 0.28 μg/L (sp. gr) for 14 years, detected in 79 % of participants) (Eskenazi et al., 2023), and 833 % higher concentrations than adolescents in Germany (GM: 0.09 μg/g of creatinine, detected in 52 % of participants) (Lemke et al., 2021). Participants in our study were not reported to be agricultural workers; hence much of their pesticide exposures are likely from para-occupational sources such as from the introduction of pesticides into the home by agricultural workers living with participants (Suarez-Lopez et al., 2012), or residential proximity to agricultural crops (Suárez-López et al., 2020).

To our knowledge, this is the first study to characterize an inverse association of concurrent glyphosate exposure with BMI-for-age z scores and % body fat, but further confirmation in prospective analyses is needed. Importantly, the absence of an association with height-for-age z scores, despite the negative associations with BMI and % body fat, suggests that pesticide exposure may influence body composition, particularly fat accumulation, without affecting stature. This pattern suggests that the effects observed may be metabolic or affecting hunger/satiety regulation rather than growth-restrictive in nature. Although the relationship between glyphosate exposure and BMI has not been well established, several animal studies suggest obesogenic potential. Notably, Kubsad et al. (2019) demonstrated that gestational glyphosate exposure resulted in lower weaning weight in F1 offspring of Sprague Dawley rats, followed by increased obesity and metabolic dysfunction in unexposed F2 and F3 generations via presumed epigenetic mechanisms (Kubsad et al., 2019). Similar effects have been reported in earlier toxicology studies (Bukowska et al., 2022; Cox, 1993) and a few epidemiological studies have also reported associations with anthropometric outcomes (Andreotti et al., 2010; Li et al., 2023; Noppakun and Juntarawijit, 2021). In the few epidemiological studies that have evaluated the associations between glyphosate exposure and different anthropometric measures at different ages, none of the studies evaluated urinary glyphosate in relation to BMI in childhood or adolescence. In the prenatal context, two studies from the same Indiana-based cohort of high-risk pregnancies reported differing findings depending on timing of exposure and outcome. In an earlier analysis, Parvez et al. (2018) found no association between maternal urinary glyphosate levels and birth weight or head circumference (Parvez et al., 2018). However, in a subsequent study from the same group, Gerona et al. (2022) reported that glyphosate exposure measured in early pregnancy was significantly associated with reduced fetal growth percentiles (Gerona et al., 2022). Despite similar mean glyphosate levels, the null findings in Parvez et al. may reflect limited power due to a smaller sample size (n = 71 vs. 187) and slightly lower detection rate (93 % vs. 99 %). In addition, consistent with our findings, prenatal exposures to other persistent organic pollutants—such as organochlorine pesticides (OCPs), polybrominated diphenyl ether (PBDE) flame retardants, and polychlorinated biphenyls (PCBs)—have also been associated with reduced fetal growth measures, even at low concentrations, as shown in a racially diverse U.S. cohort (Ouidir et al., 2020). Among a large study of adults of the Agricultural Health Study in the U.S., the total number of days of exposure to glyphosate was not associated with BMI, but rather, days of exposure to triazine herbicides and the herbicide atrazine were positively associated with BMI (LaVerda et al., 2015).

Glyphosate exposure may influence BMI through gut microbiota disruption (Walsh et al., 2023), affecting energy metabolism, nutrient absorption (Scotti et al., 2017), reduction in short-chain fatty acid (SCFA) production, and greater inflammation (Barnett et al., 2022; Lehman et al., 2023; Leino et al., 2021; Pappolla et al., 2021). Glyphosate also acts as an endocrine disruptor, affecting the HPG axis and hormones like estradiol and testosterone, which modulate BMI (Geier and Geier, 2023; Prasad et al., 2022; Wrobel, 2018; Xia et al., 2020; Watts et al., 2017). Further research on glyphosate’s effects on inflammation and endocrine disruption is needed.

Urinary ETU concentrations vary significantly across populations. In the INMA (Environment and Childhood) cohort study conducted in Spain, urine samples were collected from 1539 children aged 7–11 years residing in urban and rural areas. This study reported a detection rate of 51.8 % and a median concentration of 0.081 μg/L (Castiello et al., 2023). Similarly, a longitudinal cohort study in Sweden analyzed urine samples from 1060 adolescents aged 11–18 years from the general population between 2000 and 2017, finding a detection rate of approximately 50 % and a median concentration below 0.4 μg/L (Norén et al., 2020). In contrast, the ESPINA study, focusing on adolescents from agricultural communities in Ecuador, observed a higher detection rate of 82 % and a median concentration of 1.47 μg/L (1.72 μg/g creatinine). Additionally, a study involving 13 adult vineyard workers in Italy, who were occupationally exposed to the fungicide mancozeb, reported the highest ETU concentrations, with a median of 2.5 μg/g creatinine (Corsini et al., 2005). These variations in ETU levels may be attributed to differences in age, community settings (general vs. agricultural populations), and occupational exposures across the studied cohorts.

To our knowledge, there have been no published studies directly assessing the relationship between ETU and BMI in adolescents. EBDC fungicides, including ETU, have been associated with endocrine disruption, including anti-androgenic effects in vitro (Manabe et al., 2006; Yu et al., 2015), and with reduced thyroid hormone levels in animals (Axelstad et al., 2011) and humans (Goldner et al., 2010; Medda et al., 2017). These hormonal disruptions are often linked to weight gain (Sanyal and Raychaudhuri, 2016); thus, the inverse association we observed warrants further exploration. Notably, Castiello et al. (2023) reported that higher urinary ETU concentrations were associated with earlier pubertal onset in Spanish adolescents from INMA-Granada cohort in Spain, and that this relationship appeared to be modified by BMI (Castiello et al., 2023). Although BMI was not examined as a primary outcome in their study, the observed effect modification supports the plausibility that ETU may influence growth and metabolic pathways, particularly among leaner adolescents.

It is unclear why ETU may be associated with lower BMI. Although epidemiologic evidence remains limited, several mechanisms may underlie the observed inverse association between ETU and BMI. ETU is a metabolite of EBDC fungicides and has been associated with thyroid disruption, including reduced circulating thyroid hormone levels in both animal models (Axelstad et al., 2011) and human studies (Goldner et al., 2010; Medda et al., 2017). Given the essential role of thyroid hormones in growth, metabolism, and energy balance, such disruption could impair weight gain or fat accumulation in adolescents. Furthermore, in vitro studies suggest anti-androgenic and estrogenic activity of EBDC compounds (Manabe et al., 2006; Yu et al., 2015), which could potentially alter hormonal pathways influencing body composition. Animal studies provide supporting evidence; for example, ETU exposure in rats led to dose-dependent reductions in thyroid hormone levels and body weight gain, effects that reversed upon ETU withdrawal—suggesting a thyroid-mediated mechanism of growth impairment (Arnold et al., 1983). These endocrine-disrupting effects may be particularly relevant during adolescence, a period of dynamic hormonal regulation and somatic growth. However, more research is needed to establish causality and clarify dose-response relationships. These findings may offer indirect insight into our results and emphasize the need for further investigation into the role of ETU in adolescent development and body composition. This biological plausibility supports the inclusion of ETU as an exploratory exposure in relation to BMI and body fat. While glyphosate and ETU emerged as possible contributors in our mixture analysis based on the Busgang criterion, these findings should be interpreted cautiously and considered hypothesis-generating rather than confirmatory. The independent models showed consistent inverse associations; however, the Busgang weights indicated only modest contribution, and further research is needed to validate these associations.

This study has several strengths. This is one of the largest studies assessing the health effects of herbicides and PFAS exposures and adolescent health. Furthermore, the unique exposure profile, characterized by comparatively high glyphosate and ETU concentrations and low PFAS levels, offers a valuable contrast to prior adolescent studies from high-income countries, enabling evaluation of underexplored exposure–outcome associations. Standardized data collection protocols and comprehensive biomarker assessments enhance internal validity. While this analysis is cross-sectional, it draws from a well-established cohort with detailed contextual information, improving the precision and interpretation of exposure–health relationships in a real-world agricultural setting. Additionally, considering the population-based study design of ESPINA, we believe the findings of this study population may be generalizable to other populations with similar exposure patterns (high pesticide and low PFAS exposures) such as many low- and middle-income countries.

The present study has certain limitations that must be acknowledged. Its cross-sectional design limits causal inference and precludes the assessment of critical windows of susceptibility—particularly important during adolescence, a period marked by rapid hormonal and metabolic changes. Additionally, the possibility of reverse causation cannot be ruled out; for instance, leaner adolescents may have distinct behavioral or physiological characteristics that influence their pesticide exposure or metabolism. Beyond the clear benefits of conducting longitudinal analyses, studies involving pesticide biomarker measures with multiple and sequential samples can improve exposure accuracy when the half-lives of such pesticides are short (Attfield et al., 2014; LaKind et al., 2014), such as EBDC fungicides like maneb or mancozeb or their metabolite ETU, which have half-lives of 17–23 h (Lindh et al., 2008). Pesticides with short half-lives, such as organophosphates and pyrethroids, exhibit substantial variability in exposure assessment based on urinary biomarkers, with within-person variability often exceeding between-person variability by several-fold(Attfield et al., 2014). This high within-person variability can lead to exposure misclassification, particularly when relying on a single urine sample to represent an individual’s typical exposure. Such misclassification is generally non-differential with respect to the outcome and therefore tends to bias effect estimates towards the null, potentially underestimating true associations. Glyphosate has a relatively long half-life in humans, ranging from approximately 3 to 14 days, and is primarily excreted unchanged in urine without significant accumulation in tissues (Bradberry et al., 2004). Consequently, urinary glyphosate concentrations are more reflective of subacute exposures than those of ETU. In contrast, PFAS are highly persistent chemicals with exceptionally long biological half-lives—about 3 years for PFOA and 2.9 years for PFOS—and are best measured in serum as indicators of chronic exposures (Li et al., 2019). While PFAS and herbicides differ in their environmental persistence and mechanisms of action, their inclusion in joint mixture models reflects real-world co-exposures. Our study advances the understanding of cumulative risks in populations facing complex exposure profiles. Additionally, we did not collect dietary intake data, which is an important potential confounder of both pesticide exposure, especially for biomarkers like glyphosate and ETU, and body composition. The absence of this information may contribute to residual confounding. Although creatinine-normalized urinary pesticide biomarker provides a convenient index of maternal exposure, it remains an indirect measure of the dose reaching target tissues (O’Brien et al., 2017) and thus an imperfect surrogate for actual fetal exposure (Boeniger et al., 1993). Variability in maternal hydration status, renal function, and creatinine generation (e.g., differences in muscle mass or metabolism) can skew this biomarker, potentially leading to over- or underestimation of true pesticide dose (Boeniger et al., 1993; O’Brien et al., 2017). Furthermore, individuals with diabetes have been shown to exhibit significantly different concentrations of urinary arsenic when values are adjusted for creatinine, compared to non-diabetic individuals; a difference not observed when specific gravity is used for adjustment (Yassine et al., 2012). Creatinine excretion has been found to depend more heavily than specific gravity on factors such as body size, age, sex (Kuiper et al., 2021), and season (Nermell et al., 2008). As a result, creatinine adjustment may substantially overestimate urinary arsenic concentrations in children and/or malnourished individuals, such as those in this study population, relative to older or better-nourished individuals (Nermell et al., 2008). Overall, the relationship between glyphosate exposure with BMI and adiposity in humans is multifaceted and influenced by various biological factors that warrant careful consideration. Future research endeavors are warranted to better understand the complex relationship between glyphosate and ETU exposure with BMI in adolescents. Replication studies using longitudinal approaches with large sample sizes, and ideally with pesticide quantification at multiple time points are warranted.

5. Conclusion

Our study found that concentrations of glyphosate, and to a lesser extent ETU, were inversely related to BMI-for-age z scores and % body fat in adolescents living in agricultural areas. PFAS and PFOA showed no association with BMI-for-age z scores and adiposity after FDR correction. Both independent chemical modeling and chemical mixture modeling supported these findings, with glyphosate and ETU contributing the most to the inverse relationship. Height-for-age z score was not associated with any of the chemicals measured. The lack of an association with height further supports the idea that glyphosate and ETU exposures may influence adiposity or metabolic processes specifically, rather than impairing overall linear growth. However, given the cross-sectional nature of the study, these findings should be interpreted with caution. Replication of these findings in longitudinal studies with measures of pesticide biomarkers at multiple time periods are warranted.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.envres.2025.122378.

Data availability

Data will be made available on request.