Association of Polychlorinated Biphenyls and Organochlorine Pesticides with Autism Spectrum Disorder in Jamaican Children

Abstract

Background:

Polychlorinated biphenyls (PCBs) and organochlorine (OC) pesticides are suspected to play a role in autism spectrum disorder (ASD).

Objectives:

To investigate associations of PCBs and OC pesticides with ASD in Jamaican children and explore possible interaction between PCBs or OC pesticides with glutathione S-transferase (GST) genes (GSTT1, GSTM1, GSTP1) in relation to ASD.

Methods:

Participants included n=169 age- and sex-matched case-control pairs of Jamaican children 2–8 years old. Socioeconomic status and food frequency data were self-reported by the parents/guardians. Blood from each participant was analyzed for 100 PCB congeners and 17 OC pesticides and genotyped for three GST genes. PCBs and OC pesticides concentrations below the limit of detection (LoD) were replaced with (LoD/√2). We used conditional logistic regression (CLR) models to assess associations of PCBs and OC pesticides with ASD, individually or interactively with GST genes (GSTT1, GSTM1, GSTP1).

Results:

We found inverse associations of PCB-153 [adjusted MOR (95% CI) = 0.44 (0.23–0.86)] and PCB-180 [adjusted MOR (95% CI) = 0.52 (0.28–0.95)] with ASD. When adjusted for covariates in a CLR the interaction between GSTM1 and PCB-153 became significant (P < 0.01).

Discussion:

Differences in diet between ASD and typically developing control groups may play a role in the observed findings of lower concentrations of PCB-153 and PCB-180 in individuals with ASD than in controls. Considering the limited sample size and high proportion of concentrations below the LoD, these results should be interpreted with caution but warrant further investigation into associations of PCBs and OC pesticides with ASD.

1. Introduction

Polychlorinated biphenyls (PCBs) and organochlorine (OC) pesticides are persistent organic pollutants (POPs) of increasing public health concern. They are suspected to play a role in many adverse human health effects including developmental abnormalities in infants and children (ATSDR, 2000; ATSDR, 2011; ATSDR, 2002a; ATSDR, 1994; ATSDR, 2002b). Many countries have banned the use of PCBs and certain OC pesticides, however their long half-lives and lipophilic properties have allowed them to persist in the environment (ATSDR, 2000; ATSDR, 2002a; ATSDR, 1994; ATSDR, 2002b). The most common routes of exposure to PCBs are through consumption of contaminated fish, meat, and dairy products (ATSDR, 2000). Likewise, the main routes of exposure to OC pesticides are through meat, poultry, dairy products, fish, and leafy vegetables (ATSDR, 2002a; ATSDR, 1994; ATSDR, 2002b). On the other hand, POPs have potential for bioaccumulation in the food chain (Mackay & Fraser, 2000). The dietary patterns also have an effect on levels of PCBs and OC pesticides in human body. For example, in the US seafood contains higher levels of PCBs compared to beef (Environmental Working Group, 2003), but since the consumption of beef is more than that of fish (Zeng et al., 2019), beef accounts for a higher percentage of PCBs in the total diet.

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by restricted or repetitive behavioral patterns and deficits in social interaction and communication (APA, 2013). The most recent estimate of ASD prevalence increased to 1 in 59 children (Baio et al., 2018), but ASD etiology remains poorly understood. There is a clear genetic component, (Tick, Bolton, Happe, Rutter, & Rijsdijk, 2016) though ASD is generally believed to be a multifactorial condition with many potential genetic and environmental exposures involved (Lyall et al., 2017a).

Much of the research on ASD in recent years has been devoted to investigating environmental exposures, particularly endocrine disrupting chemicals. These chemicals, including PCBs and OC pesticides, are hypothesized to play a role in ASD by interfering with neurodevelopment (Kalkbrenner, Schmidt, & Penlesky, 2014; Hertz-Picciotto et al., 2008). Despite this concern, epidemiologic studies on the association between PCBs or OC pesticides and ASD remain sparse and the findings are inconsistent (Lyall et al., 2017b). Among those reporting positive associations, there is some inconsistency in which specific PCB congener or OC pesticide is associated with ASD (e.g. PCB-153 or PCB-138/158). However, most studies thus far have reported null associations with most POPs analyzed. Evidence from developing countries is lacking as prior studies have been focused on countries such as the United States (Braun et al., 2014; Lyall et al., 2017b; Roberts et al., 2007; Granillo et al., 2019), Canada (Bernardo et al., 2019), and Finland (Brown et al., 2018; Cheslack-Postava et al., 2013). Furthermore, to our knowledge, none of the prior studies have accounted for potential gene-environment interactions between these POPs and genes that encode the enzymes responsible for metabolizing them, such as the glutathione S-transferase (GST) family genes (Bohannon, Porter, Lavoie, & Ottinger, 2018).

In general, there is limited literature on postnatal and early childhood exposures in ASD (Rossignol, Genuis, & Frye, 2014). Neurodevelopment is a process that begins early in gestation, but structural and functional brain development continues after birth, through childhood and adolescence, and into early adulthood (Jernigan, Baare, Stiles, & Madsen, 2011). For example, synapse formation and pruning occur simultaneously (Rakic, Bourgeois, Eckenhoff, Zecevic, & Goldman-Rakic, 1986), however rapid synapse formation continues in postnatal periods before being outpaced by pruning during childhood (Penzes, Cahill, Jones, VanLeeuwen, & Woolfrey, 2011; Tang et al., 2014). Between ages 2 and 10 years, synaptic pruning results in elimination of more than 50% of extra synapses (Navlakha, Barth, & Bar-Joseph, 2015). One suspected mechanism behind ASD is a deficit in pruning (Tang et al., 2014; Martinez-Cerdeno, 2017). Complex, genetically determined processes such as synapse formation interact with the environment, possibly contributing to behavioral differences in children (Jernigan et al., 2011). A study has reported that PCBs could have an effect on synaptogenesis (Howd & Fan, 2007). Therefore, it is possible that environmental exposure to PCBs during childhood may interact with dynamic brain development processes and contribute to or exacerbate ASD.

The population in Jamaica may be at high risk of exposure to PCBs and OC pesticides. Historically, both PCBs and OC pesticides were widely used in Jamaica (Fernandez, Singh, & Jaffe, 2007; Mansingh, Robinson, Henry, & Lawrence, 2000). Certain pesticides such as dichlorodiphenyltrichloroethane (DDT), chlordane, and hexachlorobenzene (HCB) were banned in Jamaica in the 1970s (Robinson & Mansingh, 1999). However, in the 1990s, many were found to still be in use (Fernandez et al., 2007) and one report indicated that PCBs and OC pesticides were some of the most common organic pollutants in the Wider Caribbean Region (Farrington & Tripp, 1995). Since then, PCBs have been detected around highly populated areas such as Montego Bay, the second largest city in Jamaica (Jaffe et al., 2003). OC pesticides including DDT, dichlorodiphenyldichloroethylene (DDE), chlordane, and HCB have also been detected in water, sediment, fauna, and shrimp samples in various regions across Jamaica (Mansingh et al., 2000; Robinson & Mansingh, 1999; Fernandez et al., 2007; Jaffe et al., 2003).

Although not focused on OC pesticides, two recent studies have found that pregnant women in Jamaica had higher urine concentrations of other classes of pesticides compared to pregnant women from other Caribbean countries (Forde et al., 2015), the US, and Canada (Dewailly et al., 2014). We have previously reported concentrations of certain PCB congeners, including PCB-153, PCB-180, PCB-206, and total-PCB, as well as 4,4’-DDE (hexane fraction) and total-DDE in umbilical cord blood serum to serve as a reference for Jamaican newborns (Rahbar et al., 2016a). We have also previously reported that use of household pesticides (mainly pyrethroids) during pregnancy and breastfeeding was associated with ASD in Jamaican children. This association may be modified by other chemical exposures (Christian et al., 2018). However, to our knowledge, there have been no studies on the associations of PCBs and OC pesticides with ASD in Jamaica. The objective of this paper is to investigate the relationship between ASD and childhood serum concentrations of PCBs and OC pesticides, including potential effect modification by GST genes, in Jamaican children. We will compare our results to those previously published using Jamaican cord blood serum, as well as results from studies conducted in developed countries.

2. Methods

2.1. Study Population

Investigators at the University of Texas Health Science Center (UTHealth) and the University of the West Indies (UWI), Mona Campus in Kingston, Jamaica have collaborated to conduct a case-control study, in which we enrolled Jamaican children ages 2–8. Methods regarding recruitment and analysis of biospecimens have been previously reported (Rahbar et al., 2012b; Rahbar et al., 2012a; Rahbar et al., 2016a). In brief, cases were identified from an existing Jamaican Autism Database, developed by Professor Samms-Vaughan over the past two decades. Cases were reassessed and classified as having ASD based on both the Autism Diagnostic Observation Schedule, second edition (ADOS-2) (Lord et al., 2012) and the Autism Diagnostic Interview-Revised (ADI-R) (Rutter, Le, & Lord, 2003). Typically developing (TD) controls were recruited from local schools and well-child clinics and matched 1:1 to ASD cases based on sex and age (± 6 months). TD controls were assessed using the Social Communication Questionnaire (SCQ) (Rutter, Bailey, & Lord, 2003) and had to have a score ≤ 6 to rule out the possibility of ASD or other developmental delays. At the time of enrollment, the parents/guardians completed socio-economic status (SES) and food frequency questionnaires. The SES questionnaire obtained data such as demographics of the child and the parents, education level and occupational history of the parents, and whether the family owned a car (used as a proxy for SES in Jamaica). The food frequency questionnaire obtained data on how frequently the child consumed foods such as meat, fish, dairy, fruits, and vegetables. Both SES and food questionnaires have been described in detail previously (Rahbar et al., 2012b; Rahbar et al., 2012a). The data included in the present study represent 169 age- and sex-matched case-control pairs. Study data were compiled using Research Electronic Data Capture (REDCap) (Harris et al., 2009). The study protocol was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Boards (IRBs) of UTHealth (HSC-SPH-09–0059), UWI, and Michigan Department of Health and Human Services (MDHHS). In this study, we obtained written consent from parents/guardians and child’s assent (if the child was 7–8 years old).

2.2. Analysis of PCBs and OC pesticides

As previously described (Rahbar et al., 2016a), children had 8–10 mL of whole blood drawn in 10 mL red top vacutainers, which were further processed by the Caribbean Genetics (CARIGEN) lab at UWI to obtain 4–5 mL of serum. After serum was obtained, samples were frozen at −10 to −20 °C and stored in Jamaica until ready to be shipped to the MDHHS in Lansing, Michigan using Cryoport® dry vapor shippers (Irvine, CA, USA). Serum samples were analyzed for 94 individual PCB congeners and three congener pairs (066/095, 138/163, and 196/203), and 17 OC pesticides, including DDT (2,4′-DDT (o,p′-DDT) and 4,4′-DDT (p,p′-DDT)) and its metabolites DDE (4,4′-DDE (p,p′-DDE) hexane fraction and benzene fraction), and DDD (2,4′-DDD (o,p′-DDD) and 4,4′-DDD (p,p′-DDD)). Other OC pesticides included chlordane and chlordane-related isomers and metabolites (γ-chlordane, α-chlordane, trans-nonachlor, cis-nonachlor, oxychlordane, heptachlor, and heptachlor epoxide), hexachlorocyclohexane (HCCH) isomers (γ-HCCH and β-HCCH), HCB, and Mirex. A detailed description of the laboratory analysis and quality control is provided in the Supplementary Materials of our previous manuscript (Rahbar et al., 2016a). The limit of detection (LoD) for total PCB is the algebraic sum of the LoDs of the three most common PCB congeners found in biomonitoring samples (153, 170 and 180). For total 4,4′-DDE, we used the LoD from the lowest fraction (i.e. hexane) (Rahbar et al., 2016a).

2.3. Genetic Analysis

Whole blood was processed by the CARIGEN lab at UWI and then shipped to the UTHealth School of Public Health (UTSPH) Human Genetics Center (HGC) Laboratory in Houston, Texas. Genomic DNA was isolated from buffy coat using the Gentra PUREGENE Blood Kit (Qiagen, N.V., Venlo, The Netherlands). Regions of the GSTM1 and GSTT1 genes were amplified in two independent TaqMan Copy Number Assay reactions; GSTM1 Assay ID: Hs02575461_cn and GSTT1 Assay ID: Hs00010004_cn (www.thermofisher.com). Each quantitative PCR reaction contained 10 ng of genomic DNA, TaqMan Genotyping Master Mix, TaqMan Copy Number Assay, and TaqMan Copy Number Reference Assay (RNaseP) in a 5 μL reaction in accordance with instructions provided by the manufacturer. Thermal cycling conditions were 2 min at 50°C, 10 min at 95°C, followed by 40 cycles of 15 sec at 95°C and 1 min at 60°C. The real-time QuantStudio and 12K Flex Software v1.2.2 were used to assay each sample in quadruplicate. Samples were normalized to RNaseP and averaged to obtain a single ΔCt value (FAM dye Ct - VIC dye Ct) which was imported into CopyCaller Software v2.1 (www.thermofisher.com) to determine the copy number of each genomic DNA target. GSTM1 and GSTT1 homozygous deletions were coded as DD and the presence of an insertion was coded as I*. Methods for genetic analysis of the GSTP1 Ile105Val polymorphism (rs1695) using the TaqMan Drug Metabolism SNP genotyping assay C_3217198_20 have been described in detail previously (Rahbar et al., 2014; Rahbar et al., 2016b).

2.4. Statistical Analysis

Differences in demographic characteristics and other covariates between ASD cases and TD controls were assessed using univariable conditional logistic regression (CLR). We confirmed that the GSTP1 polymorphism in the TD control group met Hardy-Weinberg equilibrium expectations using Pearson’s Chi-square test (P = 0.58). For POPs with ≥ 30% of samples above LoD (i.e. PCB-153, PCB-180, total PCB, and 4,4′-DDE (hexane fraction)), we reported the mean and standard deviation (SD) of lipid-adjusted serum concentrations for ASD cases and TD controls. For descriptive purposes, we have reported both the arithmetic and geometric means, where arithmetic means are untransformed and geometric means have undergone a log transformation due to the distribution being skewed to the right. Observations below LoD were replaced by (LoD/√2) (Hornung & Reed, 1990; Meeker, Sathyanarayana, & Swan, 2009). The geometric mean concentrations of POPs were compared between ASD cases and TD controls using univariable CLR.

For our main analyses, we used separate CLR models where ASD is the independent variable and the exposure variables of interest include PCB-153, PCB-180, total PCB, and 4,4′-DDE (p,p′-DDE) (hexane fraction), which were dichotomized using two different methods as follows. First, to be consistent with prior studies (Brown et al., 2018; Cheslack-Postava et al., 2013), POPs exposure variables were dichotomized as greater than/equal to vs. below the 75^th percentile among TD controls. We then dichotomized POPs exposure variables based on whether concentrations were above or below their respective LoDs. Case-control pairs with missing data were dropped from CLR models. Since a sizeable portion of data were below LoD, we performed sensitivity analyses using subsets of case-control pairs in which both the case and control had concentrations above LoD (van Buuren, 2012) for PCB-153, PCB-180, total PCB, and 4,4′-DDE (hexane fraction).

Potential confounders included age of the mother and father, education level of the mother and father, SES, parish of child’s birth, and food variables. Due to sparse data in food frequency categories, the food variables were dichotomized based on whether the parents/guardians reported the child ever/never ate each food item. For additive models (i.e. without interactions), food variables (e.g., fish, meat, poultry, eggs and dairy products) that were considered as potential confounders varied depending on the POP and its common routes of exposure, which we obtained from Agency for Toxic Substances and Disease Registry (ATSDR) reports (Division of Environmental Health, 2009; Mackay & Fraser, 2000). Potential confounders were included in final models if they were associated with ASD and POP exposure variables at P ≤ 0.2 to be conservative, and if inclusion of the variable in the models resulted in a ≥ 10% change in the matched odds ratio (MOR). In addition, we considered some other relevant covariates suggested by the CHARGE study that included seafood consumption, maternal age, maternal education, and SES (Hertz-Picciotto et al., 2011). Each final model included its own unique set of potential confounders or covariates, which are listed in the footnotes of their respective tables in the Results section.

We reported MORs and corresponding 95% confidence intervals (CIs), and results were considered statistically significant if P ≤ 0.05. Finally, we also explored the possibility that GST genotypes may modify the association between POPs exposures and ASD using interactive CLR models. The CONTRAST statement in SAS 9.4 (SAS Institute Inc., 2013) was used to report MORs and 95% CIs specific to each level of the potential effect modifier. All analyses for this study were conducted using SAS 9.4 (SAS Institute Inc., 2013).

3. Results

A majority of the children in this study were male (78.7%) and > 95% of children and their parents were Afro-Caribbean. The mean age of ASD cases and TD controls was 60.4 months and 60.6 months, respectively. Compared to TD controls, mothers (P = 0.04) and fathers (P < 0.01) of ASD cases were significantly older. There were no significant differences between cases and controls regarding education of the mothers and fathers and SES. A greater proportion of TD controls were from Kingston parish compared to ASD cases (P < 0.01). Fewer TD controls were homozygous for the GSTM1 deletion (P = 0.05), but there were no differences between ASD cases and TD controls regarding genotype distributions for GSTP1 or GSTT1. Descriptive characteristics of cases and controls are reported in Table 1.

Table 1.

Characteristics of the study population and their parents (169 matched pairs or 338 children).

Population Characteristics	ASD Cases N (%)	TD Controls N (%)	P-valuea
Child’s sex
Male	133 (78.7)	133 (78.7)	1.00
Child’s age (months)	60.4 ± 18.5b	60.6 ± 17.5b	0.52
Child’s race
Afro-Caribbean	162 (95.9)	163 (96.5)	0.78
Maternal age at birth of the child (years)c	28.4 ± 6.7b	26.9 ± 6.7b	0.04
Paternal age at birth of the child (years)d	34.1 ± 8.4b	31.3 ± 8.8b	<0.01
Maternal race
Afro-Caribbean	164 (97.0)	163 (96.5)	0.76
Paternal racee
Afro-Caribbean	162 (95.9)	161 (96.4)	0.78
Maternal education at birthf
Up to high schoolg	91 (53.9)	88 (52.4)	0.84
Beyond high schoolh	78 (46.2)	80 (47.6)
Paternal education at birth of the childi
Up to high schoolg	102 (63.0)	113 (71.5)	0.15
Beyond high schoolh	60 (37.0)	45 (28.5)
Socioeconomic status
Family owns a car	82 (48.5)	74 (43.8)	0.35
Parish of child’s birth
Kingston	60 (35.5)	101 (59.8)	<0.01
Otherj	109 (64.5)	68 (40.2)
GSTP1k
Ile/Ile	39 (23.6)	41 (25.2)	0.72
Ile/Val	96 (58.2)	85 (52.2)
Val/Val	30 (18.2)	37 (22.7)
GSTM1l
D/Dm	52 (31.5)	36 (22.5)	0.05
I/I or I/Dn	113 (68.5)	124 (77.5)
GSTT1o
D/Dm	41 (24.9)	43 (27.0)	0.53
I/I or I/Dn	124 (75.2)	116 (73.0)

Note: ASD, autism spectrum disorder; TD, typically developing.

^aP-values from Wald’s test in conditional logistic regression models comparing cases and controls.

^bMean ± SD for continuous variables.

^cData missing for 1 case.

^dData missing for 2 cases and 3 controls.

^eData missing for 2 controls.

^fData missing for 1 control.

^gIncludes Primary/ Jr. Secondary, and Secondary/High/Technical schools.

^hIncludes Vocational, Tertiary College, or University.

ⁱData missing for 7 cases and 11 controls.

^jOther parishes include Trelawny, Westmoreland, Clarendon, St. Andrew, St. Mary, St. James, St. Elizabeth, St. Catherine, St. Thomas, St. Ann, and Manchester.

^kData missing for 4 cases and 6 controls.

^lData missing for 4 cases and 9 controls.

^mIndicates the null alleles for GSTM1 and GSTT1.

ⁿIndicates the homozygote (I/I) or heterozygote (I/D) for GSTM1 and GSTT1;

^oData missing for 4 cases and 10 controls.

Of the PCB congeners and OC pesticides analyzed in this study, only PCB-153, PCB-180, total PCB, and 4,4′-DDE (hexane fraction) had ≥ 30% of observations above their respective LoDs, as shown in Table 2. The remaining POPs analyzed in this study are reported in Table S1, along with the percent below LoD and their respective LoDs. Although not statistically significant, ASD cases had slightly lower geometric mean concentrations of PCB-153, PCB-180, and total PCB compared to TD controls. ASD cases did have significantly lower geometric mean serum concentrations of 4,4′-DDE (hexane fraction) compared to TD controls (geometric mean = 57.9 ng/g-lipid vs. 75.3 ng/g-lipid, P = 0.02). There was one TD control that had an extremely high concentration of 4,4′-DDE (hexane fraction) (2,436.1 ng/g-lipid), however there remained a significant difference in concentrations between ASD cases and TD controls when we excluded this observation (P = 0.04, data not shown). This particular observation was confirmed to be accurate by MDHHS, therefore it was retained in all analyses. In the sensitivity analysis of pairs above LoD, there was no significant difference between ASD cases and TD controls with respect to geometric mean concentrations of 4,4’DDE (hexane fraction) or the PCB congeners (Table 2).

Table 2.

Distributions of lipid-adjusted serum concentrations (ng/g-lipid) of persistent organic pollutants for cases and controls separately, along with sensitivity analysis results by fitting the final model with pairs of concentrations above LoD.

Persistent organic pollutants	n = 338 children (169 pairs)							Sensitivity Analysis n = children in pairs with both concentrations above LoD
	ASD Cases (N=169)			TD Controls (N=169)				ASD Cases			TD Controls
	% above LoDa	A-mean ± SDb	G-mean (SD)c	% above LoDa	A-mean ± SDb	G-mean (SD)c	P-valued	n	A-mean ± SDb	G-mean (SD)c	n	A-mean ± SDb	G-mean (SD)c	P-valued
Polychlorinated biphenyls
PCB-153	47.9	18.6 ± 23.5	14.4 (1.9)	47.9	20.0 ± 17.8	16.0 (1.9)	0.14	39	26.4 ± 26.2	22.3 (16)	39	29.0 ± 23.5	24.3 (1.7)	0.45
PCB-180	40.5e	8.2 ± 8.4	6.6 (1.8)	37.9	8.9 ± 8.3	7.1 (1.8)	0.19	27	13.7 ± 15.0	10.6 (18)	27	17.1 ± 14.4	14.1 (1.7)	0.12
Total PCB	33.7	37.2 ± 59.3	26.2 (2.0)	32.5	41.7 ± 71.0	28.4 (2.0)	0.28	20	57.6 ± 41.7	48.6 (1.7)	20	106.6 ± 170.7	60.8 (2.5)	0.35
Organochlorine pesticides
4,4’-DDEf	74.6	112.5 ± 201.2	57.9 (2.8)	85.8	139.9 ± 272.9	75.3 (2.6)	0.02	107	152.3 ± 239.0	88.1 (2.6)	107	142.6 ± 226.1	92.2 (2.2)	0.71

Note: ASD, autism spectrum disorder; DDE, dichlorodiphenyldichloroethylene; LoD, limit of detection; PCB, polychlorinated biphenyl; SD, standard deviation; TD, typically developing.

^aObservations below LoD replaced with LoD/√2.

^bA-mean = Arithmetic mean = Mean (serum concentrations).

^cG-mean = Geometric mean = Exp [ Mean (ln serum concentrations)].

^dP-value from Wald statistic in conditional logistic regression (CLR) models comparing geometric means between the cases and controls. (child age and sex were study matching factors and accounted for in CLR models)

^eData not available for 1 case.

^fHexane fraction.

ASD cases and TD controls differed in their consumption of cheese, yogurt, eggs, goat, pork, salt water fish, canned fish (e.g. sardine or mackerel), salt fish or pickled mackerel, shellfish (e.g. lobster and crab), shrimp, lettuce, callaloo/broccoli/pak choi, and cabbage (all P < 0.05), as reported in Table 3.

Table 3.

Univariable comparisons of those who reported eating food items between cases and controls (169 matched pairs or 338 children).

Food Items	ASD Cases N (%)	TD Controls N (%)	P-valuea
Dairy Products/Eggs
Milk	75 (44.4)	91 (53.9)	0.09
Cheese	112 (66.3)	132 (78.1)	0.02
Yogurt	32 (18.9)	58 (34.3)	<0.01
Eggs	122 (72.2)	142 (84.0)	<0.01
Meat/Poultry
Beef	53 (31.4)	67 (39.6)	0.11
Lamb/Mutton	18 (10.7)	28 (16.6)	0.10
Goat	58 (34.3)	82 (48.5)	<0.01
Pork	68 (40.2)	96 (56.8)	<0.01
Liver/Kidney	82 (48.5)	99 (58.6)	0.07
Chicken	157 (92.9)	164 (97.0)	0.07
Fish & Seafood
Salt water fish	87 (51.5)	106 (62.7)	0.03
Fresh water fish	43 (25.4)	40 (23.7)	0.71
Canned Sardine/Mackerel	119 (70.4)	145 (85.8)	<0.01
Canned Tuna	43 (25.4)	53 (31.4)	0.21
Salt fish/Pickled mackerel	94 (55.6)	135 (79.9)	<0.01
Lobster/Crab	4 (2.4)	26 (15.4)	<0.01
Shrimp	6 (3.6)	28 (16.6)	<0.01
Packaged fish	43 (25.4)	30 (17.75)	0.08
Leafy Vegetables
Lettuce	41 (24.3)	89 (52.7)	<0.01
Callaloo, broccoli, or pak choi	99 (58.6)	132 (78.1)	<0.01
Cabbage	91 (53.9)	127 (75.2)	<0.01

Note: ASD, autism spectrum disorder; TD, typically developing.

^aP-value from Wald statistic in conditional logistic regression (CLR) models comparing cases and controls. (child age and sex were study matching factors and accounted for in CLR models)

Among TD controls, the 75^th percentiles of PCB-153, PCB-180, total PCB, and 4,4′-DDE (hexane fraction) were 24.1 ng/g-lipid, 10.6 ng/g-lipid, 34.0 ng/g-lipid, and 123.4 ng/g-lipid, respectively (Table S2).

Those with PCB-153 ≥ 75^th percentile significantly differed from those with PCB-153 < 75^th percentile in consumption of milk, beef, and goat (all P < 0.05) (Table S3). There were no statistically significant differences in food consumption between those with PCB-180 ≥ 75^th vs. PCB-180 < 75^th (Table S4). Those with total PCB ≥ 75^th percentile significantly differed from those with total PCB < 75^th percentile only on consumption of beef (P = 0.01) (Table S5). There were no statistically significant differences in food consumption for those with 4,4′-DDE (hexane fraction) ≥ 75^th vs. 4,4′-DDE (hexane fraction) < 75^th, however there was a marginally significant difference in cabbage consumption (P = 0.06) (Table S6).

Table 4 reports the analyses in which the POPs are dichotomized as greater than/equal to vs. below the 75^th percentile among TD controls. After adjusting for age of the father and consumption of milk, yogurt, beef, and goat, ASD cases had 56% lower odds of having serum PCB-153 concentration ≥ 75^th percentile compared to TD controls [adjusted MOR (95% CI) = 0.44 (0.23–0.86)]. In a sensitivity analysis of pairs in which concentrations in both members were above LoD, this association was in the same direction but was no longer statistically significant [adjusted MOR (95% CI) = 0.33 (0.08, 1.32)]. There was also an inverse association between PCB-180 and ASD. After adjusting for consumption of milk, yogurt, and shrimp, ASD cases had 48% lower odds of having serum PCB-180 concentration ≥ 75^th percentile compared to TD controls [adjusted MOR (95% CI) = 0.52 (0.28–0.95)]. The sensitivity analysis resulted in an even more significant inverse association [adjusted MOR (95% CI) = 0.12 (0.02, 0.65)]. In contrast, there was no significant difference in the odds of having total PCB [adjusted MOR (95% CI) = 0.85 (0.51–1.41)] or 4,4′-DDE (hexane fraction) [adjusted MOR (95% CI) = 1.10 (0.63–1.92)] concentrations ≥ 75^th percentile between ASD cases and TD controls in adjusted analyses. These results remained non-significant in sensitivity analyses.

Table 4.

Associations of select dichotomized persistent organic pollutants (concentrations above or below the 75^th percentile) with autism spectrum disorder, along with sensitivity analysis results by fitting the final model with pairs of concentrations above LoD.

Dichotomized persistent organic pollutants	n = 338 children (169 pairs)				Sensitivity Analysis n = children in pairs with both concentrations above LoD
	Unadjusted		Adjusted		Unadjusted			Adjusted
	MOR (95% CI)	P-valuea	MOR (95% CI)	P-valuea	n	MOR (95% CI)	P-valuea	n	MOR (95% CI)	P-valuea
Polychlorinated biphenyls
PCB-153	0.79 (0.47, 1.32)	0.36	0.44 (0.23, 0.86)b	0.02b	78	0.70 (0.27, 1.84)	0.46	76	0.33 (0.08, 1.32)b	0.12b
PCB-180	0.57 (0.33, 0.99)c	0.05	0.52 (0.28, 0.95)d	0.03d	54	0.15 (0.04, 0.68)	0.01	54	0.12 (0.02, 0.65)d	0.01d
Total PCB	0.94 (0.58, 1.53)	0.80	0.85 (0.51, 1.41)e	0.52e	40	1.00 (0.29, 3.45)	1.00	40	1.00 (0.28, 3.57)e	1.00e
Organochlorine pesticides
4,4’-DDEf	0.80 (0.49, 1.32)	0.38	1.10 (0.63, 1.92)g	0.73g	214	1.19 (0.67, 2.13)	0.56	214	1.70 (0.87, 3.31)g	0.12g

Note: CI, confidence interval; DDE, dichlorodiphenyldichloroethylene; MOR, matched odds ratio; PCB, polychlorinated biphenyl.

^aP-value from Wald statistic in conditional logistic regression (CLR) models that relates to the comparison of the cases and controls while adjusted for potential confounders. (child age and sex were study matching factors and accounted for in CLR models)

^bAdjusted for consumption of milk, yogurt, beef and goat, and age of the father at child’s birth.

^cData not available for 1 case.

^dAdjusted for consumption of milk, yogurt, and shrimp.

^eAdjusted for consumption of yogurt.

^fHexane fraction

^gAdjusted for consumption of cabbage and parish of child’s birth.

Table 5 reports the analyses in which the POPs are dichotomized based on whether concentrations were above or below their respective LoDs. Although ASD cases had lower, but not statistically different, concentrations of PCB-153, PCB-180, and total PCB compared to TD controls, slightly more ASD cases than TD controls had concentrations above the LoD for PCB-180 (40.5% vs. 37.9%) and total PCB (33.7% vs. 32.5%). However, these proportions were very similar. The proportions of both ASD cases and TD controls with concentrations above the LoD for PCB-153 were equivalent (47.9% each), as shown in Table 2. Therefore, it follows naturally that there was no significant difference between ASD cases and TD controls in the odds of having concentrations above LoD for PCB-153 [adjusted MOR (95% CI) = 0.83 (0.50–1.36)], PCB-180 [adjusted MOR (95% CI) = 1.03 (0.64–1.67)], and total PCB [adjusted MOR (95% CI) = 0.92 (0.56–1.49)] (Table 5). In contrast, ASD cases had both lower concentrations and lower percent above LoD compared to TD controls for 4,4′-DDE (hexane fraction) (Table 2), and therefore had 42% lower odds of having 4,4′-DDE (hexane fraction) concentrations above LoD compared to TD controls, although this did not reach significance at the 5% level [adjusted MOR (95% CI) = 0.58 (0.31–1.08)].

Table 5.

Associations of select dichotomized persistent organic pollutants (concentrations above or below limit of detection) and autism spectrum disorder (169 matched pairs or 338 children).

Dichotomized persistent organic pollutants	Unadjusted		Adjusted
	MOR (95% CI)	P-valuea	MOR (95% CI)	P-valuea
Polychlorinated biphenyls
PCB-153	1.00 (0.65, 1.53)	1.00	0.83 (0.50, 1.36)b	0.45b
PCB-180c	1.11 (0.71, 1.73)	0.65	1.03 (0.64, 1.67)d	0.90d
Total PCB	1.06 (0.67, 1.68)	0.81	0.92 (0.56, 1.49)e	0.73e
Organochlorine pesticides
4,4’-DDEf	0.50 (0.29, 0.87)	0.01	0.58 (0.31, 1.08)g	0.09g

Note: CI, confidence interval; DDE, dichlorodiphenyldichloroethylene; MOR, matched odds ratio; PCB, polychlorinated biphenyl.

^aP-value from Wald statistic in conditional logistic regression (CLR) models that relates to the comparison of the cases and controls while adjusted for potential confounders. (child age and sex were study matching factors and accounted for in CLR models)

^bAdjusted for consumption of milk, yogurt, beef and goat, and age of the father at child’s birth.

^cData not available for 1 case.

^dAdjusted for consumption of milk, yogurt, and shrimp.

^eAdjusted for consumption of yogurt.

^fHexane fraction

^gAdjusted for consumption of cabbage and parish of child’s birth.

We found evidence of a possible interaction between genotypes of GSTM1 and category of PCB-153 concentration (≥ 75^th percentile vs. < 75^th percentile) in relation to ASD status of the children (overall interaction P = 0.08), as shown in Table 6. Specifically, among those with the null (D/D) genotype for GSTM1, there was a positive association between PCB-153 concentration and ASD [MOR (95% CI) = 1.82 (0.59–5.59)]. In contrast, among those with a homozygous (I/I) or heterozygous (I/D) genotype for GSTM1, there was an inverse association between PCB-153 concentration and ASD [MOR (95% CI) = 0.60 (0.33–1.10)]. Neither of these associations reached statistical significance at the 5% level. However, we when adjusted for three covariates (eating yogurt, consumption of salt fish/pickled mackerel, and age of father at child’s birth) the overall interaction between GSTM1 and PCB-153 was significant (P < 0.01), though among those with a homozygous (I/I) or heterozygous (I/D) genotype for GSTM1, there was a significant (P < 0.01) inverse association between PCB-153 concentrations and ASD [adjusted MOR (95% CI) = 0.25 (0.11–0.59)].

Table 6.

Interactive association between dichotomized exposure to PCB-153 (concentrations above or below the 75^th percentile) and ASD status modified by GSTM1 genotypes, along with sensitivity analysis results by fitting the final model with pairs of concentrations above LoD.

GST M1 Genotype	n = 338 children (169 pairs) (Model with GSTM1 interaction)							Sensitivity Analysis n = children in pairs with both concentrations above LoD (Model with GSTM1 interaction)
	Category of PCB-153	ASD Cases (n)	TD Controls (n)	Unadjustedc		Adjustedd		ASD Cases (n)	TD Controls (n)	Unadjustede		Adjustedf
				MOR (95% CI)a	P-valueb	MOR (95% CI)a	P-valueb			MOR (95% CI)a	P-valueb	MOR (95% CI)a	P-valueb
D/Dg	≥ 75th percentile	13	5	1.82 (0.59, 5.59)	0.29	2.55 (0.68, 9.62)	0.17	7	3	2.10 (0.36, 12.12)	0.41	2.99 (0.40, 22.39)	0.29
	< 75th percentile	39	31					4	5
I/I or I/Dh	≥ 75th percentile	21	35	0.60 (0.33, 1.10)	0.10	0.25 (0.11, 0.59)	<0.01	9	15	0.48 (0.15, 1.49)	0.20	0.19 (0.04, 0.92)	0.04
	< 75th percentile	92	89					19	14

Note: ASD, autism spectrum disorder; CI, confidence interval; MOR, matched odds ratio; PCB, polychlorinated biphenyl; TD, typically developing.

^aMOR represents those in Q4 (≥75%) compared to those in Q1, Q2, and Q3 (< 75%) of PCB-153 concentration.

^bP-value from Wald statistic in conditional logistic regression (CLR) model. (child age and sex were study matching factors and accounted for in CLR models)

^cP-value for the interaction term in the unadjusted CLR model based on the full data set is 0.08.

^dP-value for the interaction term in the CLR model adjusted for consumption of yogurt and salt fish/pickled mackerel, and age of the father based on the full data set is <0.01.

^cP-value for the interaction term in the unadjusted CLR model based on the subset of pairs with both concentrations above LoD is 0.14.

^fP-value for the interaction term in the CLR model adjusted for consumption of yogurt and salt fish/pickled mackerel, and age of the father at child’s birth based on the subset (concentrations above LoD) is 0.09.

^gIndicates the null alleles for GSTM1.

^hIndicates the homozygote (I/I) or heterozygote (I/D) for GSTM1.

The sensitivity analysis of interactive CLR models with pairs with both concentrations above LoD revealed similar findings as in interactive CLR models based on the full dataset. There was a tendency towards positive association among those with the null genotype [MOR (95% CI) = 2.10 (0.36, 12.12)] and an inverse association among those with the homozygous or heterozygous genotypes [MOR (95% CI) = 0.48 (0.15, 1.49)]. However, the overall interaction between GSTM1 and PCB-153 was not significant, P = 0.14). When we adjusted for the aforementioned three covariates, there was a significant (P = 0.04) inverse association between PCB-153 concentration and ASD [adjusted MOR (95% CI) = 0.19 (0.04–0.92)], though the overall interaction was not significant (P = 0.09). We did not find any other interactions between combinations of the GST genes and POPs analyzed in this study (data not shown).

4. Discussion and Implications

In the present study, we measured serum concentrations of PCBs and OC pesticides among 2–8 year old ASD cases and TD controls in Jamaica. We report inverse associations of the highest quartiles of PCB-153 and PCB-180 with ASD among Jamaican children. Furthermore, the association between PCB-153 and ASD may be modified by GSTM1 genotype, though the interaction between GSTM1 and PCB-153 was not statistically significant. We found no associations between the highest quartiles of total PCB and 4,4’-DDE (hexane fraction) and ASD. The remaining PCB congeners and OC pesticides assessed as part of the study had nearly 100% of concentrations below LoD for both ASD cases and the TD control group, hence the associations with ASD could not be assessed.

A high proportion of the samples had concentrations below LoD and thus were imputed. Therefore, we tested whether imputing these concentrations drastically changed our results by performing sensitivity analyses using only pairs in which both the ASD case and TD control had a detectable concentration of the POP. Our sensitivity analyses had the same overall results as discussed above, with the exception of the CLR with PCB-153. In the sensitivity analysis, there was no longer a significant association between PCB-153 and ASD, though the MOR was in the same direction and of similar magnitude compared to the full model. For all sensitivity analyses, since a small subset of the study population was used, the confidence intervals were wider, indicating less precise effect estimates. However, the overall similarity between MORs in the main analyses and sensitivity analyses is promising, as it indicates that imputing values below LoD did not have a substantial effect on our conclusions.

Only one prior epidemiologic study by Braun et al. (Braun et al., 2014) has reported an inverse association between PCB-153 and ASD similar to our own results, however this study differed from ours in a few important ways: 1) timing of exposure assessment (i.e. gestational vs. childhood), 2) autism assessment (i.e. autistic behavior vs. autism diagnosis), and 3) statistical analysis (i.e. semi-Bayesian model adjusting for co-pollutants and traditional confounders vs. CLR models adjusting for traditional confounders). Braun et al. also reported that PCB-138/158 was positively associated, PCB-178 was (marginally) inversely associated, and the rest of the tested congeners had no association. However, these specific associations were imprecise (Braun et al., 2014). In contrast, analyses from the Early Markers for Autism (EMA) study which also assessed prenatal exposures, found increased risk of ASD for those in the highest quartile of PCB-138/158 (Lyall et al., 2017b) and PCB-153 (Traglia et al., 2017; Lyall et al., 2017b). Using Bayesian methods, Bernardo et al. found similar trends as the EMA study (Bernardo et al., 2019). A pilot study found ASD cases to have higher prenatal concentrations of PCBs compared to controls (Cheslack-Postava et al., 2013), however investigators were unable to replicate this finding with a larger sample size (Brown et al., 2018). Granillo et al. grouped PCB congeners (measured during pregnancy) by biological mechanism of action instead of investigating individual associations for their main analysis. Their overall results were not statistically significant, but the authors called for further study in larger populations (Granillo et al., 2019).

Among studies using biomarkers to measure exposure to OC pesticides, one found an association between gestational concentrations of trans-nonalchlor and autistic behavior (Braun et al., 2014), and a more recent study reported an association between the highest quartile of maternal serum concentrations of p,p′-DDE (i.e. 4,4′-DDE) (hexane fraction) and ASD in children in Finland (Brown et al., 2018). The EMA study reported no association (Traglia et al., 2017; Lyall et al., 2017b). This was in line with our findings, though the EMA study measured gestational concentrations. Another study found an increasing dose-response relationship between prenatal residential proximity to agricultural pesticide application sites and risk of ASD in their children (Roberts et al., 2007). The Childhood Autism Risks from Genetics and the Environment (CHARGE) study also used prenatal proximity to agricultural application of pesticides and found low prevalence of OC pesticide exposure. However, they did find a positive association between organophosphate pesticide exposure and ASD (Shelton et al., 2014). The CHARGE study later went on to report that the associations of several classes of pesticides and ASD may be attenuated by folic acid supplementation, though they were unable to analyze OC pesticides (Schmidt et al., 2017). Many of these studies have limitations, including failing to account for co-exposures to other POPs, and the possibility that dietary exposures to other nutrients (e.g. folate) counterbalance the harmful effects of exposure to endocrine disrupting chemicals (Kalkbrenner et al., 2014) as was demonstrated in the CHARGE study (Schmidt et al., 2017).

It is possible that the differences in timing of exposure measurement in prior studies compared to our study may explain some of the discrepancy in results. While we only measured PCBs and OC pesticides exposure during the postnatal period in this case-control study, we have previously measured concentrations of these POPs in cord blood in a sub-study of the second Jamaican birth cohort (JA Kids study) (Rahbar et al., 2016a). Compared to these previously published results from Jamaica, we found that children 2–8 years old have higher proportions of detectable POPs and higher lipid-adjusted mean concentrations than newborns. For example, 98.5% of cord blood serum samples had PCB-153 concentrations below LoD, compared to only 52.1% of childhood serum samples among TD controls. Similar comparisons can be made for PCB-180 (98.5% vs. 62.1%), total PCB (100% vs. 67.5%), and 4,4′-DDE (hexane fraction) (62.7% vs. 14.2%). The arithmetic lipid-adjusted mean (SD) concentration of total 4,4′-DDE (hexane fraction) in cord blood was 61.60 (70.76) ng/g-lipid, which was much lower than the concentration among TD children in the present study (139.9 (272.9) mg/g-lipid). However, it is important to note that the cord blood study used a different method of imputing values below LoD to calculate mean (SD) concentrations than we used in the present study, which makes our results slightly less comparable. Regardless, these comparisons suggest that POPs, particularly 4,4′-DDE (hexane fraction), are still persistent in Jamaica and accumulating in the serum of children as they age. Since the Jamaican population has continuous exposure to these POPs and neurodevelopment continues into adolescence, it is of interest to explore associations during later windows of neurodevelopment.

To our knowledge, we are the first to investigate childhood concentrations of these POPs and their associations with ASD. However, the CHARGE study reported on a structurally similar class of chemicals, polybrominated diphenyl ethers (PBDEs), measured in serum sampled during childhood and found no associations of prenatal exposures with ASD (Hertz-Picciotto et al., 2011). In this study, the authors discussed key differences in sources of prenatal and postnatal exposures and that, despite PBDEs having a long half-life, childhood exposures to PBDEs may not always accurately reflect prenatal exposures (Hertz-Picciotto et al., 2011). Other studies have investigated childhood concentrations of POPs with disorders and disabilities that are frequently associated with ASD. Among these studies, some reported associations of postnatal/childhood concentrations of POPs with cognitive function (Cartier et al., 2016), ADHD, and learning disabilities (Lee, Jacobs, & Porta, 2007), while others reported no associations with ADHD (Forns et al., 2018; Newman, Behforooz, Khuzwayo, Gallo, & Schell, 2014). It is possible that exposure to these POPs during later windows of neurodevelopment can contribute to or exacerbate existing ASD symptoms. Further study of this exposure window is required to examine differences compared to prenatal exposures.

It is also possible that the differences in dietary habits between ASD cases and TD controls, as well as differences in dietary habits between Jamaicans and populations from other countries contribute to the overall conflicting results in the literature. Many children with ASD have narrow food preferences due to a variety of factors (Bandini et al., 2017), which may contribute to lower exposures to these POPs if children with ASD are avoiding certain foods. In our study, ASD cases reported lower consumption of nearly every food variable we ascertained compared to TD controls. However, the dietary differences between ASD cases and TD controls may also partly be explained by food availability dependent on the parish of residence. Kingston parish is the most urban parish in Jamaica, and more of our TD controls were from Kingston parish compared to ASD cases. While the literature on food availability in Jamaica is very limited, prior studies have indicated that about 26% of Jamaican children experience food insecurity (Dubois et al., 2011), and children living in urban areas may have better access to nutritious foods than those in rural areas, especially among poor families (Walker, Powell, Hutchinson, Chang, & Grantham-McGregor, 1998). Furthermore, high food costs have also been reported in parishes with relatively high poverty rates (Henry, Caines, & Eyre, 2015). We have attempted to account for these issues in our analyses by including food variables and parish of child’s birth as potential confounders, when relevant. However, there may be complex relationships between food preferences, food insecurity, SES, and parish of residence at play that warrant further investigation into dietary differences between ASD cases and TD controls and the effects on blood concentrations of POPs in Jamaica.

The National Health and Nutrition Examination Survey (NHANES) has also reported the lipid-adjusted serum concentrations of these POPs that are representative of the US population (CDC, 2019). According to the 2003–2004 survey, which is the most recent survey with data available, US children and adolescents 12–19 years old had lower geometric mean PCB-153 (5.86 ng/g-lipid vs. 16.0 ng/g-lipid) and PCB-180 (3.06 ng/g-lipid vs. 7.1 ng/g-lipid) concentrations compared to TD Jamaican children 2–8 years old. However, US children and adolescents had higher geometric mean 4,4′-DDE (hexane fraction) concentrations compared to TD Jamaican children (105 ng/g-lipid vs. 75.3 ng/g-lipid). Since some of the main routes of exposure to PCBs and OC pesticides are via consumption of contaminated fish and meat, it is possible that Jamaicans have different exposure levels compared to the US population due to a seafood-heavy diet (Institute of Medicine of National Academy, 2003; Environmental Working Group, 2003). Due to accumulation of POPs in the food chain, a population’s overall exposure level depends on both the types and amounts of meat and fish being consumed. As discussed in our previous paper on PCB and OC pesticide concentrations in Jamaican newborns (Rahbar et al., 2016a), compared to the US and Canada, the Jamaican population consumes about twice as much fish and about half as much meat per capita.

To our knowledge, we are the first to report on possible gene-environment interactions that potentially modify the associations of childhood concentrations of these POPs with ASD. Individuals who are homozygous for the GSTM1 deletion (i.e. D/D) have complete loss of function of the glutathione S-transferase mu 1 enzyme, which can potentially increase susceptibility to xenobiotics including POPs. Our results are in line with this. We found a positive association of PCB-153 with ASD among those with loss of enzyme function (i.e. D/D) while there was an inverse association among those with enzyme function (i.e. I/*). These results suggest that having glutathione S-transferase mu 1 enzyme function may attenuate the positive association of PCB-153 and ASD reported in previous studies. However, the demographic distribution and exposure measurement timing in our study is very different compared to previous studies, so we cannot draw any conclusions at this time. It is important to note that neither of these associations were statistically significant. The overall p-value for interaction between GSTM1 and PCB-153 (GSTM1*PCB-153) was P = 0.08 and did not reach statistical significance at the 5% level. However, when adjusted for the child’s consumption of yogurt and salt fish/pickled mackerel, and age of father at child’s birth, the overall interaction between GSTM1 and PCB-153 became significant (P < 0.01). In this study, we may not have adequate power to detect this interaction, so these results should be considered exploratory. However, as ASD is a multifactorial disorder, it is important to consider these types of interactions (Hertz-Picciotto, Schmidt, & Krakowiak, 2018). Our findings warrant further study before any conclusions can be drawn on the role of GSTM1 deletions in altering susceptibility to POPs as they may relate to ASD.

We acknowledge that his study has several limitations. First, due to the timing of our exposure being in early childhood after ASD had already been diagnosed, we are unable to draw conclusions on the possible role of these POPs in ASD etiology. We had relatively high proportions of PCBs and OC pesticides with observations below LoD, though we used a similar cut-off for inclusion in analysis (≥ 30% above LoD) and method for imputing values below LoD (LoD/√2) as other prior studies have used, and we conducted sensitivity analyses using pairs in which both the case and control had concentrations above LoD. It is possible that our results are affected by residual confounding due to imperfect adjustment for food variables. However, known risk factors for ASD, such as maternal age, that were measured in this study were assessed as potential confounders and found not to affect the MORs. The findings from the CLR models with interactions should be interpreted with caution, as we may not have adequate power to detect significant interactions for both the overall interaction and significant genotype specific 95% CIs for MORs that do not include 1.0. The TD controls enrolled in this study were more likely to be from the Kingston area compared to the ASD cases. Since extent of exposure to PCBs and OC pesticides may depend on parish, this may have introduced selection bias. It also suggests that results from TD controls may not be generalizable to the entire Jamaican population.

In this paper, we reported that among Jamaican children, ASD cases had lower geometric lipid-adjusted serum concentration of 4,4′-DDE (hexane fraction) compared to TD controls (57.9 mg/g-lipid vs. 75.3 ng/g-lipid). Compared to TD controls, ASD cases also had lower odds of having lipid-adjusted concentrations ≥ 75^th percentile for PCB-153 [adjusted MOR (95% CI) = 0.44 (0.23–0.86)] and PCB-180 [adjusted MOR (95% CI) = 0.52 (0.28–0.95)]. Furthermore, the association between PCB-153 and ASD may be modified by genotype of GSTM1, though this interaction was not statistically significant at the 5% level. Specifically, there was a tendency towards a positive association among those with the null (D/D) genotype [MOR (95% CI) = 1.82 (0.59–5.59)] and a tendency towards an inverse association among those with the homozygous (I/I) or heterozygous (I/D) genotypes [MOR (95% CI) = 0.60 (0.33–1.10)]. However, when adjusted for the child’s consumption of yogurt and salt fish/pickled mackerel, and age of father at child’s birth, the overall interaction between GSTM1 and PCB-153 became significant (P < 0.01) and the inverse association among those with the homozygous (I/I) or heterozygous (I/D) genotypes became significant, [MOR (95% CI) = 0.25 (0.11–0.59)]. The analysis of interactions was exploratory and the results should be interpreted with caution. These results are among the first to be reported from Jamaica and warrant further investigation into the relationship between these POPs and ASD.

Highlights.

• We report inverse association of PCB-153 & PCB-180 concentrations with ASD in Jamaica.

• Differences in diet between ASD and TD controls may play a role in these findings.

• The association of PCB-153 levels with ASD status may be modified by GSTM1 genotype.

Informed consent:

In this study, informed consent was obtained from parents/guardians and child’s assent was sought, if the child was 7–8 years old.