Urinary concentrations of Dialkylphosphate Metabolites of Organophosphate pesticides in the study of Asian Women and their Offspring’s Development and Environmental Exposures (SAWASDEE)

Abstract

Background:

Measurements of urinary dialkyl phosphate (DAP) metabolites are often used to characterize exposures to organophosphate (OP) insecticides; however, some challenges to using urinary DAP metabolites as an exposure measure. OP insecticides have short biological half-lives with measurement in a single urine sample typically only reflecting recent exposure within the last few days. Because of the field staff and participant burden of longitudinal sample collection and the high cost of multiple measurements, typically only one or two urine samples have been used to evaluate OP insecticide exposure during pregnancy, which is unlikely to capture an accurate picture of prenatal exposure.

Methods:

We recruited pregnant farmworker women in Chom Thong and Fang, two districts of Chiang Mai province in northern Thailand (N=330) into the Study of Asian Women and their Offspring’s Development and Environmental Exposures (SAWASDEE) from 2017-2019. We collected up to 6 serial urine samples per participant during gestation and composited the samples to represent early, mid, and late pregnancy. We measured concentrations of urinary DAP metabolites in the composited urine samples and evaluated the within- and between-participant variability of these levels. We also investigated predictors of OP insecticide exposure.

Results:

DAP metabolite concentrations in serial composite samples were weakly to moderately correlated. Spearman correlations indicated that composite urine samples were more highly correlated in Fang participants than in Chom Thong participants. The within-person variances (0.064-0.65) exceeded the between-person variances for DETP, DEP, ΣDEAP, DMP, DMTP, ΣDMAP, ΣDAP. The intraclass correlations (ICCs) for the volume-based individual metabolite levels (ng/mL) ranged from 0.10-0.66. For ΣDEAP, ΣDMAP, and ΣDAP the ICCs were, 0.47, 0.17, 0.45 respectively. We observed significant differences between participants from Fang compared to those from Chom Thong both in demographic and exposure characteristics. Spearman correlations of composite samples from Fang participants ranged from 0.55 to 0.66 for the ΣDEAP metabolite concentrations in Fang indicating moderate correlation between pregnancy periods. The ICCs were higher for samples from Fang participants from Fang participants, which drove the overall ICCs.

Conclusions:

Collecting multiple (~6) urine samples during pregnancy rather than just 1 or 2 improved our ability to accurately assess exposure during the prenatal period. By compositing the samples, we were able to still obtain trimester-specific information on exposure while keeping the analytic costs and laboratory burden low. This analysis also helped to inform how to best conduct future analyses within the SAWASDEE study. We observed two different exposure profiles in participants in which the concentrations and variability in data were highly linked to the residential location of the participants.

Introduction

Organophosphate (OP) insecticides are a class of pesticides that have been widely used globally in agricultural, residential, and public health applications ^1–4. Their broad spectrum of applications and potent toxicity to insects make OP insecticides such as chlorpyrifos and diazinon desirable in both agricultural and residential settings ⁵. Although exposure to OP insecticides is widespread, agricultural workers are likely to have a greater exposure than the general population ^1,2. Direct dermal and inhalational occupational exposure to OP insecticides are typical routes of occupational exposure, however, insecticide residues from spraying may be carried away from the field via wind dispersion, or appearing on field equipment or clothing. As a result, OP insecticides are also likely to be present in soil, surface waters, ⁶ on the surface of crops, ⁷ and within residential settings.⁸

OP insecticides are non-persistent with biologic half-lives of hours to days ⁹. After exposure occurs, the body quickly metabolizes OP insecticides into biologically active oxons, “specific” metabolites that make the insecticide unique, and/or common diakylphosphate (DAP) metabolites ^10,11. Most OP pesticides are O,O-dimethyl or O,O-diethyl substituted and are quickly metabolized after intake into one or more of six common DAP metabolites: dimethylphosphate (DMP), dimethylthiophosphate (DMTP), dimethyldithiophosphate (DMDTP), diethylphosphate (DEP), diethylthiophosphate (DETP), diethyldithiophospahte (DEDTP) ^12,13. After a single exposure, OP insecticide metabolites are excreted in urine within a day or two of exposure ¹², hence a single urinary metabolite measurement will only represent recent exposure increasing the potential for exposure misclassification of used to predict long-term exposures. To estimate OP insecticide exposure accurately over an extended period, such as through pregnancy, multiple serial samples are needed ¹⁴.

Exposure to OP insecticides has been linked to several adverse health outcomes including neurobehavior problems, attention deficit hyperactivity disorder (ADHD), and neurodegenerative disease ^15,16. During pregnancy, OP exposure has been associated with adverse neurobehavioral outcomes in infants with the timing of exposure of particular importance ^15,17,18. Thus, understanding exposure during critical windows of development is pivotal in trying to decipher potential mechanistic pathways of toxicity and for developing intervention strategies ¹⁹.

OP insecticides are among the most commonly used insecticides in Thailand ²⁰. Previous studies indicated that people in Thailand have some of the highest concentrations of urinary DAP metabolites ^21,22 measured worldwide, which is linked to the large fraction of the workforce employed in the agricultural sector; the agricultural industry employs at least 60% of the workforce in Thailand ^4,23. As the agricultural industry continues to grow, the amount of pesticides used annually increases ²⁴ suggesting exposures are likely to increase in the future.

In this paper we have measured the distribution of urinary OP insecticide metabolites and their determinants in the Study of Asian Women and their Offspring’s Development and Environmental Exposures (SAWASDEE). SAWASDEE is a longitudinal birth cohort of pregnant farm workers and their children in northern Thailand in which prenatal insecticide exposure and child neurodevelopmental trajectories are evaluated. Our focus in this analysis is on longitudinal variability in OP exposures with an eye toward establishing a benchmark on the number of samples needed to estimate exposures throughout pregnancy and during specific windows of susceptibility for the developing fetus. We hypothesized that having more frequently collected urine samples during pregnancy, we would observe better congruence between farming activities and pesticide metabolite levels.

Methods

Study population

The SAWASDEE study is a longitudinal occupational birth cohort in Chiang Mai province, Thailand. Chang Mai province was selected for this investigation, because of its robust agricultural sector and its generalizability to other low/middle income countries (LMIC) that are similarly reliant on agriculture²⁵. We recruited study participants from Fang and Chom Thong, two districts in Chiang Mai Province (Figure 1) in which the crops grown differ in their pesticide application pattern. Participants were recruited between July 2017 and June 2019 during their antenatal care visit at district hospitals or community health clinics. Enrolled women (1) were agricultural workers or lived within 50 meters of an agricultural field; (2) had a Thai identification card permitting hospital and antenatal clinic access; (3) resided in their regional district for ≥ 6 months and planned residence at least three years after delivery, (4) spoke Thai language at home, (5) were in good general health (i.e., no major medical conditions such as hypertension, diabetes, thyroid disease, HIV), (6) consumed fewer than two alcoholic beverages (beer, wine, liquor) per day and did not use illegal drugs, (7) were less than 16 weeks of gestation, and (8) were 18-40 years of age. Expectant mothers with non-singleton pregnancies or major pregnancy complications that could affect fetal growth and development were excluded from further participation at the time of diagnosis. Additional details of the cohort are described elsewhere (Baumert et al., Submitted, 2021). The study sample size was based upon published means and standard deviations of neurological endpoints (not discussed here; memory, attention, and processing) that would enable us to have 80% power to differences with a two- tailed, two-sided t-test at the 0.05 significance level. Gestational age was determined using physician-determined gestational age which included use of the last menstrual period (LMP) with ultrasound correction as necessary, in accordance with the recommendations of the American Association of Obstetricians and Gynecologists²⁶.

Figure 1.

This map of Thailand has highlighted in yellow the Chiang Mai Province in the North. The two areas highlighted in bright green are our sampling sites. Chom Thong is near the center of Chiang Mai Province and Fang is further north, at to the Myanmar border.

Urine Collection and Analysis of DAP Metabolites

Urine samples were collected from each participant up to six times during pregnancy at each antenatal care visit. Samples were collected in 100-ml polypropylene urine collection containers and were aliquoted and stored in a −20°C freezer until analysis. To retain trimester-specific average measurements and characterize exposure adequately across pregnancy while keeping the analytic burden and cost low, urine samples were composited using equal volumes of each sample fitting the specific criterion to create early- (0-14 weeks gestation), mid- (14 weeks 1 day-27 weeks gestation) and late- (>27 weeks gestation) pregnancy samples that roughly corresponded to trimester. These composites represent the average of the analyte concentrations in each individual sample collected.

Urine composites were analyzed for DAP metabolites of OP insecticides using a previously validated method that was also cross-validated by gas chromatography-mass spectrometry ²⁷. All samples were randomized using a Fisher-Yates shuffling algorithm prior to analysis to reduce any potential batch effects ^28,29. Briefly, 5 mL of urine was spiked with dibutylphosphate (DBP) as a surrogate internal standard, then acidified with HCl (PMID 24280209). The acidified urine was extracted with ethyl acetate and acetone. DAPs were derivatized to their pentafluorobenzyl phosphate esters, which were isolated from the reaction mixture using a hexane extraction. The concentrated extract was analyzed by gas chromatography-flame photoionization detection (PMID 24280209). Data were quantified using a calibration curve normalizing on the internal standard area. For each analytical run of 34 unknown samples, two blank samples (negative control) and four positive quality control samples at two different levels were analyzed concurrently. The quality control pools consisted of unspiked pooled urine (low pool) and spiked pooled urine where DAP metabolites were spiked at various levels (high pool). The quality control information is provided in Supplemental Table 1. Successful participation in the German External Quality Assessment Scheme (GEQUAS) served as an additional quality assurance parameter of the method. The limits of detection (LOD) were 5 ng/mL (DMP), 1 ng/mL (DMTP), 0.5 ng/mL (DMDTP), 1 ng/mL (DEP), 0.125 ng/mL (DETP), and 0.25 ng/mL (DEDTP) and the relative recoveries ranged from 94–119%. For statistical analysis, the LOD divided by the square root 2 was imputed for all values below the LOD. Three summary variables, (ΣDEAP, ΣDMAP, and ΣDAP), were also created by summing using the equations below. The final units of these summary variables are μmol/L or μM.

$$\Sigma DMAP=\left(\frac{DMP}{126}\right)+\left(\frac{DMTP}{142}\right)+(\frac{DMDTP}{158})$$

$$\Sigma DEAP=\left(\frac{DEP}{154}\right)+\left(\frac{DETP}{170}\right)+(\frac{DEDTP}{186})$$

$$\Sigma DAP=\Sigma DMAP+\Sigma DEAP$$

Creatinine Measurement

Creatinine was measured by diluting urine samples 1000-fold with water after spiking with its isotopically labeled analogue. Diluted samples were analyzed by liquid chromatography electrospray ionization coupled with tandem mass spectrometry (LC-MS). For creatinine, two ion transition were monitored (m/z 113.9 → m/z 44.2 and m/z 113.9 →86) and only one ion transition was monitored for labeled creatinine (m/z 116.9 → m/z 47.2)³⁰. Quantification was achieved using an isotope dilution calibration. Quality control and assurance included the concurrent measurement of calibrants, blanks and quality control materials and semi-annual certification by the GEQUAS program. The LOD was 5 mg/dL with a relative standard deviation of 5

Questionnaire Information

Seven questionnaires were administered during pregnancy, each of which was completed at routine antenatal appointments, and detailed elsewhere (Baumert et al., 2021 submitted). Of the questionnaires administered during pregnancy, three were specific to exposure assessment and were administered during early, middle, and late pregnancy. The exposure questionnaires included questions on work-related tasks, household and occupational use of pesticides, personal habits of parents (e.g., alcohol consumption, smoking), household characteristics and cleanliness, medical histories, and demographics.

Statistical Analysis

Descriptive statistics (i.e., geometric mean (GM), geometric standard deviation (GSD), median, distribution percentiles, and range) were determined for each individual and summed DAP variable. Possible differences between the mean metabolites levels for different periods were tested with an analysis of variance. Spearman correlation coefficients were calculated between urinary DAP concentrations in serial composite samples across pregnancy. Between-person and within-person (between days) variance in DAP concentrations were determined by fitting mixed effect models in which an identifier for the mother was introduced as a random factor to account for the possibility of correlation between the serial measures in the same participant. To estimate reliability or how well the DAP concentrations in the serial composite urine samples correlated, the intraclass correlation coefficient (ICC) was calculated as (between-person variance) / (between-person variance + within-person variance).

Bivariate mixed effect models were used to analyze the associations between maternal characteristics and metabolite concentrations by introducing the characteristics as the fixed effect. The analyses were performed using the natural log-transformed molar summed DAP variables. The variables from the bivariate models that were found to be predictive of DAP levels from the bivariate models (p-values for one of the categories < 0.1) were then added to a linear mixed model that included the combined predictors. Because the mixed-effects regression model includes several variance components including both fixed and random effects we used the Nakagawa marginal and conditional R^{2 31}. Nakagawa and Schielzeth define R² for the mixed effects model as (1) marginal R² is the variance explained by only the fixed effects and (2) conditional R² is the variance explained by both fixed and random effects. The marginal R² is similar in interpretation to the traditional R² statistic from the ordinary least squares method. All analyses were completed in SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) or R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Study Characteristics

We enrolled 330 pregnant women from Chom Thong and Fang districts of Chiang Mai province in northern Thailand (Table 1). Participants’ ages ranged from 18 to 39 years with a mean age at enrollment of 25.1 years (STD=5.2) (Table 1). The average gestational age of the participants at enrollment was 9.9 weeks (Table 1) and did not differ between Chom Thong and Fang.

Table 1.

Descriptive characteristics of the 330 mother-child pairs in the SAWASDEE cohort

Variable	SAWASDEE (N = 330)	CHOM THONG (n = 222)	FANG (n = 108)
Age of mother at enrollment, (years, mean ± SD)	25.1 ± 5.2	25.6 ± 5.3	24.0 ± 5.0
Gestational age at enrollment, (weeks, mean ± SD)	9.9 ± 2.6	9.7 ± 2.6	10.2 ± 2.5
Missing	2	0	2
Parity, (n (%))
Primiparous	141 (42.7)	83 (37.4)	58 (53.7)
Multiparous	189 (57.3)	139 (62.6)	50 (46.3)
Household income, (Thai baht, mean ± SD)	10612.2 (8352.4)	11612.6 (9493.6)	8601.9 (4816.0)
Missing	11	9	2
Ethnicity of the mother, (n (%))
Thai	210 (63.6)	202 (91)	8 (7.4)
Hmong	10 (3)	10 (4.5)	0 (0.0)
Thai Yai	25 (7.6)	1 (0.5)	24 (22.2)
Karen (Pagayo)	6 (1.8)	6 (2.7)	0 (0.0)
Burmese	5 (1.5)	2 (0.9)	3 (2.8)
Akha	5 (1.5)	0 (0.0)	5 (4.6)
Pa-Long (Dara-ang)	35 (10.6)	0 (0.0)	35 (32.4)
Lahu	33 (10)	0 (0.0)	33 (30.6)
Lawa	1 (0.3)	1 (0.5)	0 (0.0)
Mother, highest education level (n (%))
Never attended school	58 (17.6)	12 (5.4)	46 (42.6)
P. 1-6 (primary)	50 (15.2)	20 (9.0)	30 (27.8)
M. 1-3 (junior high/high school)	107 (32.4)	87 (39.2)	20 (18.5)
M. 4-6 (high school/did not graduate)	71 (21.5)	64 (28.8)	7 (6.5)
Diploma/technical school/equivalent	29 (8.8)	25 (11.3)	4 (3.7)
Attended college but did not graduate	1 (0.3)	1 (0.5)	0 (0.0)
College graduate or more	14 (4.2)	13 (5.9)	1 (0.9)
Maternal smoking (ever/never) during pregnancy, (n (%))
Yes	4 (1.2)	2 (0.9)	2 (1.9)
No	326 (98.8)	220 (99.1)	106 (98.1)
Marital status, (n (%))
Legally married	49 (14.8)	48 (21.6)	1 (0.9)
Living as married	274 (83)	170 (76.6)	104 (96.3)
Widowed	2 (0.6)	2 (0.9)	0 (0.0)
Divorced	2 (0.6)	2 (0.9)	0 (0.0)
Separated	3 (0.9)	0 (0.0)	3 (2.8)
Proximity of home from nearest agricultural field or orchard, (n (%))
Within 50 meters	98 (31.6)	59 (28.4)	39 (38.2)
More than 50 meters	212 (68.4)	149 (71.6)	63 (61.8)
Missing	20	14	6

Both sites are rural areas in which agricultural-related work is the predominant occupation. Of the 330 women included in this analysis 327 reported at recruitment that they were working in an agriculture-related job at least 3 days a week prior to becoming pregnant (Table 2). We observed that many women stopped working as pregnancy progressed. By the third trimester of pregnancy, about 40% of the participants who reported working at enrollment were still working in agriculture.

Table 2.

Farming characteristics of the 330 mothers in the SAWASDEE cohort

Variable	SAWASDEE (N = 330)	CHOM THONG (n = 222)	FANG (n = 108)
Did agriculture related work during pregnancy: (n (%))
First trimester
Yes	257 (81.3)	178 (84.8)	79 (74.5)
No	59 (18.7)	32 (15.2)	27 (25.5)
Missing	14	12	2
Second trimester
Yes	191 (59.5)	137 (63.1)	54 (51.9)
No	130 (40.5)	80 (36.9)	50 (48.1)
Missing	9	5	4
Third trimester
Yes	137 (43.5)	99 (46.3)	38 (37.6)
No	178 (56.5)	115 (53.7)	63 (62.4)
Missing	15	8	7
Did farm work for more than 3 days per week before pregnant (n (%))
Yes	327 (100.0)	219 (100.0)	108 (100.0)
No	0 (0.0)	0 (0.0)	0 (0.0)
Missing	3	3	0
Handle pesticide containers during pregnancy, (n (%))+
Yes	73 (25.9)	58 (29.2)	15 (18.1)
No	209 (74.1)	141 (70.8)	68 (81.9)
Apply pesticides during pregnancy, (n (%))+
Yes	34 (12.1)	27 (13.6)	7 (8.4)
No	248 (87.9)	172 (86.4)	26 (91.6)
Farming task during pregnancy: (n (%)) +
Packed fruits, vegetables or flowers for storage/sale
Yes	150 (53.2)	115 (57.8)	35 (42.2)
No	132 (46.8)	84 (42.2)	48 (57.8)
Worked in nursery or greenhouse
Yes	24 (8.5)	22 (11.1)	2 (2.4)
No	258 (91.5)	177 (88.9)	81 (97.6)
Waxed fruit*
Yes	40 (14.9)	36 (18.2)	4 (5.6)
No	229 (85.1)	162 (81.8)	67 (94.4)
Worked with the following crops during pregnancy: (n (%)) +
Rice
Yes	54 (19.1)	41 (20.6)	13 (15.7)
No	228 (80.9)	158 (79.4)	70 (84.3)
Tangerines
Yes	44 (15.6)	0 (0.0)	44 (53.0)
No	238 (84.4)	199 (100.0)	39 (47.0)
Longons
Yes	60 (21.3)	58 (29.2)	2 (2.4)
No	222 (78.7)	141 (70.8)	81 (97.6)
Cabbage
Yes	47 (16.7)	46 (23.1)	1 (1.2)
No	235 (83.3)	153 (76.9)	82 (98.8)
Garlic
Yes	12 (4.3)	9 (4.5)	3 (3.6)
No	270 (95.7)	190 (95.5)	80 (96.4)

⁺n=282 (Chom Thong n=199, Fang n=83), number of women reported ever doing agricultural work during pregnancy,

^*n=281, 1 missing

We noted differences in characteristics of the participants between the two sites. The majority of women in Chom Thong received at least a high school education (Table 1); only 5.4% of women in Chom Thong reported never attending school, compared to nearly 43% of women in Fang. There was also a greater percentage of women who reported completing education beyond high school in Chom Thong than in Fang. Only about 5% of women received more than a high school degree in Fang compared to nearly 14% in Chom Thong. Most study participants (n=202, 91%) in Chom Thong were of Thai ethnicity (Table 1) compared to only 7.4% (n=8) in Fang. Fang district is quite diverse and is predominately comprised of tribal communities. There are three predominant hill tribes among the study participants from Fang: Thai Yai (n=24, 22.2%), Palaung (De’ang).” (n=35, 32.4%), and Lahu (n=33, 30.6%) people. Participants from Chom Thong were more likely to be legally married than women in Fang (Table 1), although, the greatest percentage of women in both sites were living as married without a legal marriage.

DAP Metabolite Levels in Urine

There were 299 early pregnancy urine measures of DAP concentrations (Table 3). Most early pregnancy measures were comprised of a single sample (n=278) although 21 women provided two urine samples in early pregnancy. For the mid-pregnancy (n=316) period, 13 women provided one urine sample, 154 women provided two urine samples, 140 women provided three samples, and 9 women provided four samples. Most study participants provided two samples (n=177) for the late-pregnancy period. There were 137 women who provided three urine samples for late pregnancy and 1 participant provided a single late-pregnancy sample.

Table 3.

DAP Metabolite Levels for pregnant women of the SAWASDEE study measured during three periods (nmol/L)

	SAWASDEE (N=330)				CHOM THONG (N=222)			FANG (N=108)
	N	AM	GM	GSD	N	GM	GSD	N	GM	GSD
Early Pregnancy ΣDEAP	299	107.55	44.99	3.25	196	32.33	2.62	103	84.39	3.65
Mid Pregnancy ΣDEAP	316	113.81	48.82	2.98	213	35.40	2.31	103	94.90	3.50
Late Pregnancy ΣDEAP	315	77.23	40.97	2.86	213	31.05	2.41	102	73.07	3.15
Early Pregnancy ΣDMAP	299	50.14	41.99	1.57	196	40.24	1.43	103	45.53	1.80
Mid Pregnancy ΣDMAP	316	45.40	39.30	1.46	213	37.89	1.29	103	42.41	1.71
Late Pregnancy ΣDMAP	315	68.36	39.83	1.55	213	39.75	1.60	102	39.99	1.42
Early Pregnancy ΣDAP	299	157.69	100.00	2.18	196	80.14	1.77	103	152.39	2.55
Mid Pregnancy ΣDAP	316	159.21	99.07	2.12	213	79.57	1.63	103	155.91	2.64
Late Pregnancy ΣDAP	315	145.59	91.23	2.05	213	77.24	1.86	102	129.13	2.19

The urinary DAP metabolites concentrations are shown in Table 3 and Supplemental Table 2 as the uncorrected metabolites (nmol/L). Geometric means for the ΣDAP metabolite concentrations for early-, mid- and late-pregnancy were 100.00 (GSD 2.18), 99.07 (GSD 2.12), 91.23 (GSD 2.05) nmol/g (Table 3), respectively. The creatinine-corrected levels are reported in Supplementary Table 2. The DEAP metabolites were observed in higher concentration than the DMAP metabolites and represented more than 60% of the total DAP concentration. Specifically, DETP concentrations dominated the DEAP metabolites in all pregnancy periods.

We observed differences between Chom Thong and Fang participants in DAP metabolite concentrations (Table 3). Most of the participants had metabolite concentrations <LOD for the DMAP metabolites in both Chom Thong and Fang. In Chom Thong the percentage <LOD for the DAP metabolites ranged from about 2-99% compared to 0-91% in Fang (Supplemental Table 2). The DEAP metabolite concentrations were much higher than the DMAP concentrations in both Chom Thong and Fang; however, most of the concentrations were greatest in Fang. Specifically, DETP and DEP were found in in the highest concentrations. In Fang, DETP was detected in all participants during the mid (ranging from 0.66-327.6 ng/mL) and late (ranging from 0.58-68.2 ng/mL) pregnancy periods. Most participants in Chom Thong also had measurable DETP concentrations >LOD; however, the levels were lower than in Fang (Table 3). DAP metabolite concentrations were weakly to moderately correlated³² during pregnancy. Spearman correlations indicated that serial composite urine samples were more correlated in Fang than in Chom Thong (Table 5). This is likely because most of the metabolite concentrations in Chom Thong were <LOD. The Spearman analyses for Chom Thong DAP metabolite levels do not indicate strong correlation between the pregnancy periods. For Fang, the Spearman correlations ranged from 0.15 to 0.66 (Table 4). The DMAP metabolites were weakly correlated³² throughout pregnancy. Again, this is because many participants had DMAP concentrations >LOD. Spearman correlations ranged from 0.32 to 0.66 for the DEAP metabolites in Fang and the ΣDEAP metabolite concentrations ranged from 0.55 to 0.66 indicating moderate correlation between pregnancy periods. Creatinine levels (0.23-0.40) were weakly correlated across pregnancy (Table 4).

Table 5.

Bivariate mixed effect models based on lognormal-transformed DAP metabolite values in which maternal characteristics are fixed effects controlling for creatinine levels. Note: Only variables found statistically significant are shown.

	ΣDAP		ΣDEAP		ΣDMAP
	Estimate	p-Value	Estimate	p-value	Estimate	p-Value
Location
Fang	0.68	< 0.0001	1.1	< 0.0001	0.1	0.002
Chom Thong	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Ethnicity of mother
Thai	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Thai Yai	1.03	< 0.0001	1.61	< 0.0001	0.1	0.1
Burmese	1.03	< 0.0001	1.4	< 0.0001	0.084	0.5
Pa-Long (Dara-ang)	1.00	< 0.0001	1.47	< 0.0001	0.28	< 0.0001
Education of mother
Never attended school	0.80	< 0.0001	1.1	< 0.0001	0.18	0.001
Graduated high school and/or higher education	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Proximity of family home from nearest agricultural field or orchard
Within 50 meters	0.27	0.0002	0.39	0.0003	0.04	0.2
More than 50 meters	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Work with Rice
Yes	−0.28	0.002	−0.49	0.0004	−0.079	0.08
No	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Work with longons
Yes	−0.29	0.001	−0.32	0.01	−0.092	0.03
No	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Work with cabbage
Yes	−0.39	< 0.0001	−0.66	< 0.0001	−0.082	0.09
No	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.
Work with tangerines
Yes	1.1	< 0.0001	1.6	< 0.0001	0.18	0.0002
No	Ref.	Ref.	Ref.	Ref.	Ref.	Ref.

Table 4.

Intraclass Correlation Coefficients (ICC) and pairwise spearman correlation coefficients for DAP metabolite levels of 330 pregnant women from the SAWASDEE study (metabolite sums expressed as nmol/L and individual metabolites expressed as ng/mL). Presented for entire study and separated by site.

SAWASDEE (N=330)
Metabolite	ICC	Period 1-2 Corr (p-value)	Period 2-3 Corr (p-value)	Period 1-3 Corr (p-value)
DEDTP	0.66	0.62 (<0.0001)	0.62 (<0.0001)	0.60 (<0.0001)
DETP	0.38	0.37 (<0.0001)	0.53 (<0.0001)	0.41 (<0.0001)
DEP	0.47	0.32 (<0.0001)	0.43 (<0.0001)	0.32 (<0.0001)
ΣDEAP	0.47	0.37 (<0.0001)	0.53 (<0.0001)	0.39 (<0.0001)
DMDTP	0.17	0.36 (<0.0001)	0.14 (0.01)	0.18 (0.05)
DMP	0.10	0.060 (0.31)	0.21 (0.0002)	0.17 (0.004)
DMTP	0.25	0.17 (0.0042)	0.24 (<0.0001)	0.22 (0.0002)
ΣDEMP	0.17	0.11 (0.05)	0.21 (0.0002)	0.12 (0.03)
ΣDAP	0.45	0.34 (<0.0001)	0.51 (<0.0001)	0.39 (<0.0001)
FANG (N=108)
Metabolite	ICC	Period 1-2 Corr (p-value)	Period 2-3 Corr (p-value)	Period 1-3 Corr (p-value)
DEDTP	0.56	0.32 (<0.0001)	0.43 (<0.0001)	0.32 (<0.0001)
DETP	0.49	0.60 (<0.0001)	0.66 (<0.0001)	0.63 (<0.0001)
DEP	0.61	0.43 (<0.0001)	0.59 (<0.0001)	0.52 (<0.0001)
ΣDEAP	0.59	0.55 (<0.0001)	0.66 (<0.0001)	0.61 (<0.0001)
DMDTP	0.19	0.42 (<0.0001)	0.16 (0.11)	0.18 (0.082)
DMP	0.09	0.15 (0.13)	0.28 (0.006)	0.37 (0.0002)
DMTP	0.32	0.30 (0.0026)	0.47 (<0.0001)	0.29 (0.0035)
ΣDEMP	0.22	0.26 (0.0080)	0.34 (0.0006)	0.33 (0.0009)
ΣDAP	0.51	0.53 (<0.0001)	0.59 (<0.0001)	0.59 (<0.0001)
Chom Thong (N=220)
Metabolite	ICC	Period 1-2 Corr (p-value)	Period 2-3 Corr (p-value)	Period 1-3 Corr (p-value)
DEDTP	-	0.038(0.61)	0.00050 (0.94)	−0.054 (0.46)
DETP	0.11	0.14 (0.055)	0.39 (<0.0001)	0.18 (0.012)
DEP	0.22	0.035 (0.63)	0.24 (0.0004)	0.024 (0.74)
ΣDEAP	0.17	0.099 (0.18)	0.35 (<0.0001)	0.11 (0.14)
DMDTP	0.24	−0.0094 (0.99)	−0.011 (0.87)	−0.024 (0.74)
DMP	0.11	0.0060 (0.94)	0.17 (0.012)	0.049 (0.50)
DMTP	0.10	0.035 (0.63)	0.068 (0.33)	0.17 (0.016)
ΣDEMP	0.10	0.022 (0.77)	0.15 (0.032)	0.010 (0.89)
ΣDAP	0.13	0.065 (0.38)	0.36 (<0.0001)	0.12 (0.099)

The within-person variances (0.06-0.65) exceeded the between-person variances for DETP, DEP, ΣDEAP, DMP, DMTP, ΣDMAP, ΣDAP (Supplemental Table 4). For DEDTP and DMDTP, between-person variance was greater than within-person variance, but this is largely because both metabolites were predominately below the LOD. The resulting ICCs for the volume-based individual metabolite levels (ng/mL) ranged from 0.10-0.66 (Supplemental Table 4). For ΣDEAP, ΣDMAP, and ΣDAP the ICCs are, 0.47, 0.17, 0.45 respectively (Supplemental Table 4).

Predictive factors of DAP concentrations

We performed several bivariate analyses (Table 5). The following variables were found to have statistically significant associations with ΣDAP, ΣDEAP, and ΣDMAP levels: location, ethnicity, education, proximity of home to nearest agricultural fields, and working with tangerines. The bivariate analyses revealed that lower ΣDEAP levels were observed in those living more than 50 meters from an agricultural field. Education was also found to be a predictive factor of DAP levels. Those with lower education, particularly participants who reported never attending school or only completing primary school, had significantly higher levels of ΣDAP than those with higher education. Because ethnicity and study location (Fang and Chom Thong) were highly correlated (Pearson’s r = 0.81) we elected to include only location in the model.

Table 6 shows the linear mixed effects model in which the parameters included location (ref=Chom Thong), education (ref=graduated high school or more), proximity of home to agricultural field (ref=more than 50 m) and working with tangerines during pregnancy (ref=no). Higher ΣDEAP concentrations were observed in those who work with tangerines and have never attended school. The geometric mean levels of those who work with tangerines (N=44) was 442.65 (GSD=2.2), 316.61 (GSD=3.1), and 81.52 (GSD=2.3) nmol/g creatinine for ΣDAP (data not shown), ΣDEAP, and ΣDMAP, respectively. Specifically, for ΣDEAP levels this is more than four times the cohort geometric mean. We also observed differences in ΣDAP, ΣDEAP, ΣDMAP concentrations in participants between the two study regions. Those in Fang had higher concentrations of all three summed metabolites values. Based on the mixed effects model the marginal R² values for ΣDAP, ΣDEAP, and ΣDMAP were 0.37, 0.38, and 0.35. When we stratified the analysis to evaluate participants from Chom Thong and Fang separately, we observed considerably different R² marginal and conditional values (Table 6). When stratifying the analysis, the marginal R² value for ΣDEAP in Fang was 0.48 compared to 0.16 in Chom Thong. These results suggest that the predictors are essentially predictors of ΣDEAP within Fang.

Table 6.

Linear mixed effect model including predictors from bivariate linear mixed effects models (n=267). A → ΣDAP, B → ΣDEP, C → ΣDMP

A	ΣDAP
	Crude model		Full model
Fixed effects	β (95% CI)	P-value	β (95% CI)	p-value
Intercept	4.57 (4.5, 4.6)	<0.0001	3.89 (3.7, 4.0)	<0.0001
Creatinine	-	-	0.0044 (0.0036, 0.005)	<0.0001
Study location
Fang	-	-	0.19 (0.02, 0.36)	0.03
Chom Thong	-	-	ref	ref
Mother, highest education level
Never attended school	-	-	0.28 (0.07, 0.49)	0.009
Graduated high school or more			ref	ref
Proximity of home from nearest agricultural field or orchard	-	-
Within 50 meters	-	-	0.13 (0.015, 0.24)	0.03
More than 50 meters			ref	ref
Worked with tangerines during pregnancy	-	-
Yes	-	-	0.83 (0.63, 1.02)	<0.0001
No	-	-	ref	ref
		-		-
Random Effects	Component		Component
Study participants	0.25	-	0.09	-
Residuals	0.31	-	0.29	-
Fixed factors	0.00	-	0.22	-
R2 GLMM (marginal)	-	-	0.37	-
R2 GLMM (conditional)	-	-	0.52	-
B	ΣDEP
	Crude model		Full model
Fixed effects	β (95% CI)	p-value	β (95% CI)	p-value
Intercept	3.8 (3.70, 3.90)	<0.0001	2.7 (2.4, 2.9)	<0.0001
Creatinine		-	0.0073 (0.006, 0.008)	<0.0001
Study location
Fang		-	0.38 (0.13, 0.64)	0.003
Chom Thong		-	ref	ref
Mother, highest education level
Never attended school		-	0.33 (0.018, 0.64)	0.04
Graduated high school or more		-	ref	ref
Proximity of home from nearest agricultural field or orchard
Within 50 meters		-	0.19 (0.02, 0.4)	0.03
More than 50 meters		-	ref	ref
Worked with tangerines during pregnancy
Yes		-	1.19 (0.9, 1.5)	<0.0001
No		-	ref	ref
Random Effects	Component		Component	-
Study participants	0.58	-	0.23	-
Residuals	0.65	-	0.57	-
Fixed factors	0.00	-	0.48	-
R2 GLMM (marginal)		-	0.38	-
R2 GLMM (conditional)		-	0.55	-
C	ΣDMP
	Crude model		Full model
Fixed effects	β (95% CI)	p-value	β (95% CI)	p-value
Intercept	3.7 (3.67, 3.73)	<0.0001	3.6 (3.4, 3.7)	<0.0001
Creatinine		-	0.001 (0.0005, 0.002)	<0.0001
Study location
Fang		-	0.03 (−0.09, 0.1)	0.6
Chom Thong		-	ref	ref
Mother, highest education level
Never attended school		-	0.09 (−0.05, 0.24)	0.2
Graduated high school or more		-	ref	ref
Proximity of home from nearest agricultural field or orchard
Within 50 meters		-	−0.005 (−0.08, 0.07)	0.9
More than 50 meters		-	ref	ref
Worked with tangerines during pregnancy
Yes		-	0.12 (−0.01, 0.3)	<0.0001
No		-		ref
Random Effects	Component		Component	-
Study participants	0.03	-	0.03	-
Residuals	0.15	-	0.16	-
Fixed factors	0.00	-		-
R2 GLMM (marginal)		-	0.35	-
R2 GLMM (conditional)		-	0.19	-

Discussion

Most birth cohort studies evaluating pesticide exposures and health outcomes have based prenatal exposure assessments on one or two urine samples collected in the 2^nd and 3^rd trimesters of pregnancy. Because pesticide exposure usually consists of a lower level dietary exposure with episodic periods of higher exposure^33,34, exposure misclassification is a problematic issue. The SAWASDEE study was designed with a large focus on collecting refined exposure assessment data to enable us to evaluate critical windows of susceptibility between pesticide exposure and health outcomes. Our goal was to collect many serial samples during pregnancy to provide confidence in our prenatal exposure assessment. We collected urine samples from pregnant women at up to six antenatal appointments. Most of our study participants provided 2-3 samples for the mid and late pregnancy periods, but we typically were only able to collect 1 urine sample representing early pregnancy despite our success in early recruitment (mean gestational age at enrollment, 9.9 ± 2.6 weeks). While measurement of pesticide metabolites in each discrete samples collected serially, the analytical cost and burden are extensive. To provide a compromise, we composited samples to be reflective of average exposure during a given trimester period which still provided similar information but at about ½ the cost. We intentionally recruited participants from Chom Thong and Fang, two districts in Chiang Mai province, because they represent two potential exposure scenarios (i.e., high application, short duration and low application, longer duration). Additionally, different types of pesticides are often used with different crops, which varied across the sites. For example, fruits trees such as tangerine trees require intensive application of insecticides when the fruit blossom petals fall off and again within the next few weeks resulting in a larger application amount in a shorter application period. Conversely, rice crops require insecticide application more regularly during the growing season but at lower quantities. These exposure scenarios were clearly observed in the urinary metabolite concentrations in our cohort. Rice, longans, and cabbage were the predominant crops of Chom Thong participants whereas tangerines were the predominant crop of Fang participants. We found much higher levels of DEAP metabolites compared to DMAP metabolites. We also saw that the highest DAP levels were among Fang participants where higher application rates were experienced. While we did see that some of the participants had high DMAP metabolite concentrations most people had concentrations that were below the LOD. Total DAP levels were driven by the concentrations of DEAP metabolites—accounting for about 60% of the total. The participants who’s urinary DMAP levels were above the LOD were considerably higher than those found in other studies; but high DMAP metabolite levels were not universally observed in our cohort. Our study is best suited to evaluate the associations with ΣDEAP as most people, especially those living in Fang, had DEAP concentrations above the LODs.

Urinary DAP metabolites represent exposure to parent OP insecticides and their preformed metabolites from multiple sources. While they are non-specific markers of OP insecticide exposure and we are not able to connect DAPs to a specific pesticide, tangerine farmers have documented high use of chlorpyrifos, a DEAP-producing insecticide, which is likely the driver of our observed DEAP levels in Fang as all participants reported working with tangerines. Location was highly correlated with crop type and farming activities so our location-specific observations were likely related to the crop grown which drives the pesticide use pattern.

Total DEAP concentrations in our study were higher than those reported in previous US birth cohorts with measures of DAP metabolite levels. The CHAMACOS (N=377) study reported the highest geometric mean DEAP level (15.9 nmol/g creatinine) of the US birth cohorts³⁵; however, it is considerably lower than the ΣDEAP concentrations in the SAWASDEE cohort. This is likely because a high number of women in our cohort actively worked in agriculture during pregnancy and the use of personal protective equipment is typically insufficient or non-existent. Total DMAP levels, however, were higher in the CHAMACOS study (Figure 1) likely because of the extensive use of oxydemton-methyl in Salinas Valley agriculture, while methyl OPs are not used extensively in the regions monitored in this study. Co-exposures to other environmental toxicants can often alter or saturate metabolic pathways resulting in different outcomes³⁶. For example, nicotine exposure is known to influence the metabolism and toxicity of OP insecticides. However, in Thailand, smoking is uncommon in women. In fact, fewer than 1% of the women in our cohort were active smokers. Given the very small number of women smoking, it is not surprising that, our sensitivity analysis conducted excluding smokers did not change our effect estimates.

Demographics factors such as ethnicity and education differed between the two study sites. Site location was a surrogate for ethnicity with the hill tribe ethnicities dominating the Fang participants. “Cultural” marriages were more common in Fang hill tribes so fewer legal marriages existed (1% Fang vs. 22% Chom Thong). Our study was not designed to look at the impact of statelessness on participants but likely the effect was greater on Fang participants because many hill tribes are considered stateless. Marital status or legal marriage may reflect this impact as stateless persons lack legal rights.

A large within-person variability presents challenges for designing epidemiological studies with sufficient statistical power. Analysis of more than a single measure per study participant reduces the sample size needed to detect associations. In our study, the within-person variability of creatinine-corrected ΣDAP levels were greater than the between-person variability. For the ΣDEAP metabolites, the ICCs were moderate to relatively strong compared to previous studies³⁷. CHAMACOS, a California based study (USA), observed ICCs ranging from 0.30-0.39 for ΣDMAP metabolites, 0.11-0.16 for ΣDEAP metabolites, and 0.27-0.35 for ΣDAP concentrations ³⁸. It is likely that we observe higher ICCs for the ΣDEAP metabolites because we have measured metabolites levels in samples composited from up to 4 serially collected samples that better represent average exposure during pregnancy periods than a single or two samples.

Location specific variability was important in our study as the exposure scenarios were quite different between Chom Thong and Fang. In Chom Thong, where exposure was lower, the ICCs were more moderate. The ICCs were stronger for the DMAP metabolites in Chom Thong than the DEAP metabolites. Because most samples were below the LOD for DAP metabolites in Chom Thong, little variance was observed, thus the ICCs were derived from the few participants with high DAP levels. In Fang, the ICCs were much higher, especially for the DEAP metabolite levels (range 0.49-0.61). For DEDTP and ΣDEAP the between-person variability was greater than the within-person variability. This is unlike other birth cohorts that have measured DAPs during the pregnancy period in which within-person variability is often greater than between-person indicating the need for additional samples during the sampling window. The stronger ICCs are likely a result of having multiple urinary samples per pregnancy period or because of having an occupational exposure pathway.

Our ICCs in Fang are stronger than those in previous studies of urinary DAPs during pregnancy which is also reflected in the Spearman correlations (ΣDAP in Fang, ρ=0.53-0.59). Focusing on the DEAP metabolites in Fang the correlation coefficients ranged, from 0.32 to 0.66 and are consistent within the specific periods for a given metabolite. In Chom Thong the data are again weakly reliable (ΣDAP in Chom Thong, ρ=0.06-0.36) and not consistent among periods. The HOME study, a Cincinnati (USA) based cohort, found weakly correlated creatinine-corrected DAP measures (r=0.22) ³⁹. The CHAMACOS study also found urinary DAP concentrations to be weakly correlated (r=0.14, P=0.005) ³⁸. Previous studies have been limited to one or two maternal urine specimens during pregnancy. Weak correlation between the serial samples for the same participant suggests the need for increased samples to estimate true exposure. By increasing the number of samples per participant during gestation there is a reduction in the potential for exposure misclassification also increasing power.

Our study is not without limitations. Because of cost restraints, we composited samples, but we would have had better refinement of exposure during pregnancy if each discrete sample was analyzed separately. Regardless, having up to six serial samples, even if composited, provides a more robust estimation of exposure during pregnancy than only one or two samples might provide given the short biological half-life of OP metabolites. It is also possible that women who had more samples per pregnancy period differed from participants who only provided a single sample per pregnancy period. Additionally, urinary DAPs will include a contribution from preformed environmental metabolites in the diet. Our study could also be improved if we had dietary data. However, given our population experienced occupational exposures that occurred after known pesticide application events, we expect the fractional contribution from dietary or other environmental sources to be smaller. Another limitation is that we did not measure all pesticides to which participants may have been exposed.

Additionally, because we composited samples, we are not able to look at the effects of season on DAP concentrations because a single composite sample often spanned seasons.

Specifically, among the Fang study participants we observed relatively strong ICCs of DAP metabolite concentrations during pregnancy. The correlations for specific DAP metabolites between trimester were also consistent. This is likely a result of the increased sampling compared to previous studies. Because so many individuals in Chom Thong had DAP concentrations below the LOD, identifying predictors of DAP levels for the entire cohort or specifically for Chom Thong was challenging. However, within Fang we see that proximity of home to agricultural fields, never attending school, and working with tangerines are significant predictors of urinary DAP levels especially for the DE metabolites. Our results also indicate the importance of stratifying future analyses by study location site as the exposure scenarios are quite different and the populations are not homogenous.

Pesticide policies and risk assessment in Thailand have been lacking^23,40. The overall lack of policy enforcement has resulted in the widespread use of legal and illegal pesticides at whim. Because pesticides are typically inexpensive, greatly increase crop yields, and loosely regulated, farm owners are heavily reliant upon them^23,40. In 2019, Thailand announced plans to ban three pesticides including chlorpyrifos; however, after political and manufacture pressure to ease up on these restrictions, Thailand decided to delay the ban despite the continued evidence of human harm they can cause. Furthermore, they expect farmworkers to comply with their pesticide-use patterns or risk losing their jobs. Use of adequate PPE is not common because of both expense and traditional practices. As farmworkers are often immigrants or indigenous Thai, they tend to be poorer and disenfranchised and know little of the harmful effects that pesticides may have on a developing fetus.

Our study shows that farm workers in Thailand, specifically tangerine workers are highly exposed to OP insecticides. We hope that these results encourage conversations around and supply evidence to support policy changes as it relates to the use of OP insecticides, especially involving vulnerable populations like children and women.

Conclusions

Our study in northern Thailand has demonstrated the high level of exposure that pregnant Thai farmworkers have during pregnancy. By collecting 6 serial samples over pregnancy, we were able to better understand their overall exposure and link that exposure to crop and pesticide use patterns. With the data we are gathering, we plan to determine what birth outcomes and early indicators of neurological deficits may be linked to these exposures. Ultimately, we hope this study will provide farmworkers with practical and affordable exposure mitigation strategies and will help the Thai government to adopt and enforce health-protective pesticide policy.

Figure 2.

A comparison of the sum of urinary dimethyl phosphates, diethyl phosphates, and dialkyl phosphates measured in the SAWASDEE Study overall and for the two sub-sites with concentration in California, Cincinnati, and New York City. All measures are in moles/g creatinine. Sample sizes for each study are noted.