Combined Exposure to Folate and Lead during Pregnancy and Autistic-Like Behaviors among Canadian Children from the MIREC Pregnancy and Birth Cohort

Abstract

Background:

Folic acid (FA) supplementation may attenuate the associations between gestational exposure to certain chemicals and autism or autistic-like behaviors, but to our knowledge, this has not been assessed for lead.

Objectives:

We examined whether the relationship between gestational blood-lead levels (BLLs) and autistic-like behaviors was modified by gestational plasma total folate concentrations, FA supplementation, and maternal methylenetetrahydrofolate reductase (MTHFR) 677C>T genotype.

Methods:

We used data from the Maternal–Infant Research on Environmental Chemicals study (2008–2011), a Canadian pregnancy and birth cohort study. Childhood autistic-like behaviors were documented in 601 children 3–4 y of age with the Social Responsiveness Scale-2 (SRS-2), where higher scores denote more autistic-like behaviors. We measured BLLs and plasma total folate concentrations during the first and third trimesters of pregnancy. We also estimated gestational FA supplementation via surveys and genotyped the maternal MTHFR 677C>T single nucleotide polymorphism (SNP). We estimated the confounder-adjusted associations between ${\log }_2$-transformed BLLs and SRS-2 scores by two indicators of folate exposure and maternal MTHFR 677C>T genotype using linear regression.

Results:

Third-trimester BLLs were associated with increased SRS-2 scores [${\mathrm{\beta }}_{adj}=3.3$; 95% confidence interval (CI): 1.1, 5.5] among participants with low ($<10\text{th}$ percentile), third-trimester, plasma total folate concentrations, but BLL-SRS-2 associations were null (${\mathrm{\beta }}_{adj}=-0.3$; 95% CI: $-1.2$, 0.5) among those in the middle category ($\ge 10\text{th}$ and $<80\text{th}$ percentiles) (p-interaction $<0.001$). FA supplementation also attenuated these associations. Both folate indicators modified first-trimester BLL-SRS-2 associations, but to a lesser extent. Third-trimester BLL-SRS-2 associations were slightly stronger among participants who were homozygous for the T (minor) allele of the MTHFR 677C>T SNP (${\mathrm{\beta }}_{adj}=0.9$; 95% CI: $-1.2$, 3.1) than those without the T allele (${\mathrm{\beta }}_{adj}=-0.3$; 95% CI: $-1.3$, 0.7), but the difference was not statistically significant ($p\text{-interaction}=0.28$).

Discussion:

Folate may modify the associations between gestational lead exposure and childhood autistic-like behaviors, suggesting that it mitigates the neurotoxic effects of prenatal lead exposure. https://doi.org/10.1289/EHP14479

Introduction

Lead exposure has a detrimental effect on child neurodevelopment.^1–3 Although exposure is decreasing, no safe level or threshold has been identified.³ Moreover, lead exposure is higher among people in marginalized communities and low- to middle-income countries.^2–4 Some observational studies suggest that childhood and gestational lead exposure is associated with autism diagnosis or autistic-like behaviors.^5–12

Conversely, a growing body of research suggests that folate may attenuate the relationship between gestational exposure to toxic chemicals and child neurodevelopmental outcomes.¹³ Previous epidemiological evidence suggests that prenatal folic acid (FA) supplementation attenuates the associations of certain phthalate metabolites,^14,15 air pollutants,¹⁶ and pesticides,¹⁷ with autism or autistic-like behaviors. This attenuation was not observed for arsenic.¹⁸ In addition, the association of childhood blood-lead levels (BLLs) and intelligence quotient (IQ) deficits may be larger in children with lower erythrocyte folate concentrations.¹⁹ This previous research suggests that folate plays a role in preventing the toxic effects of chemicals in addition to its many well-established benefits.¹³ We are unaware, however, of studies examining whether folate can attenuate the associations between BLLs and autism or autistic-like behaviors.

We used data from a pan-Canadian pregnancy cohort to assess whether the associations between gestational BLLs and childhood autistic-like behaviors are modified by gestational plasma total folate concentrations or FA supplementation. In addition, we assessed whether these associations were modified by maternal MTHFR 677C>T genotype, a common single nucleotide polymorphism (SNP) that impacts the metabolic pathway that folate is involved in²⁰ and is associated with lower folate status.²¹

Methods

Study Participants

We used data from the Maternal–Infant Research on Environmental Chemicals (MIREC) Study. MIREC is a pregnancy and birth cohort conducted in 10 Canadian cities from 2008 to 2011, as previously described.²² Inclusion criteria were as follows: $\ge 18$ y of age, $<14$ wk pregnant, able to communicate in French or English, planning on delivering at a local hospital, and consenting to cord blood collection. Participants were excluded for the following: carrying a fetus with a known abnormality, having a history of medical complications, using illicit drugs, or threatened abortion. A total of 8,716 individuals were approached, 1,983 met the eligibility criteria, and 1,861 had a singleton live birth. Sociodemographic and nutritional information (intake of supplements at study enrollment and the 16th wk of gestation) was recorded via questionnaires completed during pregnancy. We subsequently assessed neurodevelopment in a convenience subsample of 610 children 3–4 y of age from 6 of the 10 cities.²³ Our analysis was restricted to the 601 mother–child dyads for which autistic-like behaviors were documented (Figure S1).

Health Canada, Simon Fraser University, and all study sites [Izaak Walton Killam Health Center (Halifax), McMaster University (Hamilton), Kingston General Hospital (Kingston), Center Hospitalier Universitaire Sainte-Justine, Jewish General Hospital (Montreal), Mount Sinai Hospital, Princess Margaret Cancer Center, Sunnybrook Health Sciences Center (Toronto), British Columbia Children’s and Women’s Hospital (Vancouver)] approved the MIREC Study. All participants gave their informed consent to participate in the study.

Autistic-like Behaviors

The preschool-age version of the Social Responsiveness Scale-2 (SRS-2) 65-item questionnaire was completed by one parent during an in-home visit.²⁴ The SRS-2 is a sensitive, valid, and reliable tool for documenting autistic-like behaviors in several cultures and settings.^25–29 Higher SRS-2 scores denote a higher degree of communication problems and repetitive/nonreciprocal behaviors. The SRS-2 recognizes the dimensional nature of autistic traits,³⁰ which fits into the inclusivity and neurodiversity framework (autism is a result of inherent diversity in human behaviors and cognition), instead of designating autism as a disease.^27,30,31 This recognition also makes it well-suited for population-based cohort studies where the number of autistic children would be low.

Measurement of BLLs

Gestational BLLs were measured in whole blood samples collected during the first (6–13 wk gestation) and third (32–34 wk gestation) trimesters of pregnancy. All specimens were maintained at $-20°\mathrm{C}$ before being analyzed with inductively coupled plasma–mass spectrometry (ICP-MS) at the Centre de Toxicologie du Québec, Institut National de Santé Publique du Québec, Quebec, Canada, as previously described.³² No left-censoring was needed because all BLLs were above the limit of detection ($1\mathrm{\mu g/L}$).

Potential Modifying Variables

We evaluated whether four variables modified the associations between gestational BLL and child SRS-2 scores (referred to as “BLL-SRS-2 associations”): child sex, plasma total folate concentrations (also referred to as “folate status”), FA supplementation, and maternal MTHFR 677C>T (rs1801133) genotype (measured with the Sequenom MassARRAY system). We calculated plasma total folate concentrations by summing five folate vitamers: tetrahydrofolate (THF), 5-formyl-THF, 5,10-methenyl-THF, unmetabolized folic acid, and 5-methyl-THF. The limits of detection of these vitamers are 0.018, 0.02, 0.019, 0.029, and $0.019\mathrm{nmol}/\mathrm{L}$, respectively. Folate vitamers were measured using liquid chromatography–tandem mass spectrometry, as previously described.³³ These samples were collected at the first and third trimesters of pregnancy, the same times that BLLs were assessed. Most samples were not drawn after fasting, which may decrease precision.^32–34

FA supplementation was primarily measured using a structured survey conducted at 16 wk gestation, where participants were asked to list all “vitamins, minerals, homeopathic medicines, and/or natural products” they took in the prior 30 d. They reported the product name/description, the drug identification number (DIN) or natural product number (NPN),³⁵ the amount taken each time, and the frequency of intake. The FA dose for each product was assessed by searching the DINs and NPNs on their respective databases (drug product database and Health Canada licensed natural health products database). If no valid DIN or NPN was provided, we searched the provided product name in these databases to extract its FA dosage per tablet. If more than one product was active at the time, then we recorded their average dose. For instance, there were two “Jamieson Folic Acid” products active at the time, with dosages of $400\mathrm{\mu }\mathrm{g}/\text{tablet}$ and $\mathrm{1,000}\mathrm{\mu }\mathrm{g}/\text{tablet}$. Therefore, if participants simply reported taking this product with no further details, they were assigned a dosage of $700\mathrm{\mu }\mathrm{g}/\text{tablet}$. If no valid DIN/NPN was provided and the product name was incomplete, then the median dosage across all supplements in that “group” was assigned. Groups included: adults’ multivitamins, prenatal/postpartum multivitamins, women’s multivitamins, children’s multivitamins, and FA. Finally, we multiplied the dosage per tablet by the number taken per day for all the products each participant reported taking to estimate their total daily FA supplementation.

In addition to the 30-d recall form described above, we also estimated daily FA supplementation using two additional survey tools: the 24-h recall form (also completed at 16 wk gestation) and the baseline questionnaire (complete on study enrollment, 6–13 wk gestation). Although the 24-h recall form was otherwise identical to the 30-d recall form, the baseline questionnaire queried “vitamin/mineral supplementation” in the past 3 months. If a participant completed and returned the 30-d recall form and indicated FA supplementation, we estimated FA supplementation as described above with the 30-d recall form alone ($n=492$). If a participant indicated no FA supplementation on the 30-d recall form, then we confirmed this with the 24-h recall form ($n=5$). If a participant reported FA supplementation on the 24-h but not the 30-d recall form, then we estimated FA supplementation using only the 24-h recall form ($n=4$) (the 24-h recall form appeared first and was sometimes filled in with more detail). If the 30-d recall form was not filled in and returned, then we estimated FA supplementation with the 24-h recall form ($n=9$). In some instances, we cross-referenced the baseline questionnaire to fill in missing details on the 24-h recall form ($n=17$). Finally, if neither the 30-d nor the 24-h recall forms were returned, we estimated FA supplementation with the baseline questionnaire ($n=74$). Our analysis assumed that FA supplementation was constant throughout pregnancy. In addition, we performed an exploratory analysis to check whether FA supplementation was constant across the three surveys we used.

Finally, we categorized participants’ FA intake from supplements as being $<400$, 400–1,000, or $>\mathrm{1,000}\mathrm{\mu }\mathrm{g}/\mathrm{d}$. These cutoffs correspond to the recommended supplementation for individuals who could become pregnant or are pregnant or lactating^36,37 and the upper limit for over-the-counter prenatal supplements in Canada.^35,37 Nearly all participants ($>99\%$) had plasma total folate concentrations $>25.5\text{\hspace{0.17em}nmol}/\mathrm{L}$, a level associated with maximal neural tube defect risk reduction among vitamin B12–adequate individuals, as is the case in the MIREC cohort.³⁸ As such, no biologically plausible and feasible cutoffs for plasma total folate concentrations could be identified. Because approximately 10% and 20% of our sample reported a daily FA supplementation of $<400$ and $>\mathrm{1,000}\mathrm{\mu }\mathrm{g}$, respectively, we selected the 10th and 80th percentiles of plasma total folate concentrations as cutoffs. We anticipated that making both variables (FA supplementation and plasma total folate concentrations) equally stringent in classifying participants would make them easier to compare. Otherwise, the lowest category of plasma total folate concentrations may be diluted with participants who actually have relatively high plasma total folate concentrations, and this may obscure the modifying effect of being in the lowest category.

Covariates

We used a directed acyclic graph based on a literature review to identify variables that may confound the relationship between gestational BLLs and child SRS-2 scores (Figure S2) a priori.^39,40 We identified the following confounding variables: maternal age at enrollment (in years), household income (in CAD: $\le \$\mathrm{40,000}$, $40,001–$80,000, $80,001–$100,000, $>\$\mathrm{100,000}$), maternal education (high school or less, college or trade school, undergraduate university degree, graduate university degree), maternal relationship status (did they live with their partner or not), parity (0, 1, or $\ge 2$ previous pregnancies), city of residence, year of enrollment (2008, 2009, 2010, or 2011), maternal cigarette smoking (did they smoke during pregnancy or not), and maternal race (do they exclusively self-identify as being “White” or not).^41–44 Maternal race was included as a confounder because it is a proxy for environmental injustices caused by systematic racism, which may impact lead exposure and childhood neurodevelopment.^4,42,45,46 The “non-White” racial groups were combined because the number of participants who identified as being “Black” ($n=20$), “Asian” ($n=30$), “Indigenous” ($n=7$), “Latin American” ($n=16$), or “Other” ($n=20$) was small. The number of participants in each of these racial groups does not sum to 62 because participants could select multiple racial groups. We did not consider maternal race as an effect modifier because it did not fit within the scope of our study objectives. We also decided a priori to control for variables that are plausibly associated with SRS-2 scores to improve the precision of our linear regression models.⁴⁷ This included: child sex, child age at SRS-2 assessment (in months), and Home Observation for Measurement of the Environment (HOME) score (continuous).^48–50 We collected covariate information via a questionnaire administered on study enrollment. HOME scores were determined by in-home observations and interviews with the child’s caregiver(s) by trained interviewers at the same time as the SRS-2 assessment and reflect the quality of the stimulation provided to the child.⁵⁰ As is the case for almost all observational studies, we cannot rule out the presence of unmeasured or residual confounding.⁵¹

Statistical Analyses

We first created linear regression models to examine the associations between ${\log }_2$-transformed gestational BLLs and SRS-2 scores in the entire sample:

$$Y_i=\mathrm{\alpha }+{\mathrm{\beta }}_x{x}_i+{\mathrm{\beta }}_C{C}_i+{ϵ}_i,$$ (1)

where $Y_i$ represents a child’s estimated SRS-2 score, $x_i$ represents an individual mother’s ${\log }_2$-transformed BLL, $C_i$ represents a vector of the covariates, and ${ϵ}_i$ represents model errors. Next, we created separate linear regression models with product interaction terms to estimate BLL-SRS-2 associations by three modifying variables: plasma total folate concentrations, FA supplementation, and maternal MTHFR 677C>T genotype (all of which are categorical variables with $k=3$ levels):

$$Y_i=\mathrm{\alpha }+{\mathrm{\beta }}_x{x}_i+{\mathrm{\beta }}_{z_2}{z}_{2,i}+{\mathrm{\beta }}_{z_3}{z}_{3,i}+{\mathrm{\beta }}_{x\text{:}{z}_2}{x}_i:z_{2,i}+{\mathrm{\beta }}_{x\text{:}{z}_3}{x}_i:z_{3,i}+{\mathrm{\beta }}_C{C}_i+{ϵ}_i,$$ (2)

where $z_i$ represents a modifying variable, and $z_1$ is the references level. The association between gestational BLLs and child SRS-2 when $z_k=1$, 2, and 3 is denoted by ${\mathrm{\beta }}_x$, ${\mathrm{\beta }}_x+{\mathrm{\beta }}_{x:z_2}$, and ${\mathrm{\beta }}_x+{\mathrm{\beta }}_{x:z_3}$, respectively. We used a modified version of this model to assess modification by child sex (see Supplementary Materials, Web appendix). CIs were constructed using the formulas in Figueiras et al.⁵²; see the web appendix for further details. All analyses were carried out separately for the first and third trimesters of pregnancy.

We recorded the two-sided $p$-values (with a type I error rate of 5%) of interaction terms to assess whether differences in associations between strata were statistically significant. Still, we primarily relied on comparing point estimates (and their 95% CIs) and the overall pattern of results over null hypothesis significance testing when interpreting our results. Missing data (including BLLs, plasma total folate concentrations, MTHFR genotype, household income, maternal education, and HOME scores) were addressed using multiple imputation by chained equations (MICE) with 10 multiply imputed datasets.^53,54 We assumed that these data were missing at random because the missingness can plausibly be explained by the other variables used in our analysis.⁵³ The following auxiliary variables were also used to aid with the imputation: maternal alcohol consumption during pregnancy (yes, no), maternal birthplace (Canada, United States, Mexico, China, or Other), prepregnancy body mass index (continuous), gestational (first and third trimester) concentrations of arsenic, cadmium, mercury (see Alampi et al.¹⁰ for further details), and gestational (first and third trimester) concentrations of unmetabolized folic acid, 5-mTHF, and the sum of (THF, 5-formyl-THF, 5,10-methenyl-THF). All analyses were carried out using R (version 4.3.3; R Development Core Team).

Exploratory Analyses

We also assessed the modifying effect of continuous plasma total folate concentrations because categorizing concentrations causes information to be lost.⁵⁵ To do this, we assessed BLL-SRS-2 associations while fixing plasma total folate concentrations to a range of values.⁵⁶ This approach was not taken for FA supplementation because it has discrete values. Because folate may modify BLL-SRS-2 associations in a nonlinear fashion, we fit plasma total folate concentrations with a restricted cubic spline using the interactionRCS package.⁵⁷ We conducted a complete case analysis here due to software incompatibilities between MICE and interactionRCS.

The T allele of the MTHFR 677C>T SNP is associated with a thermolabile MTHFR enzyme that requires higher folate concentrations to be stable.^58,59 Consequently, we created linear regression models with three-way interaction terms to simultaneously assess whether plasma total folate concentrations and maternal MTHFR 677C>T genotype modified BLL-SRS-2 associations:

$$\begin{array}{c}{Y}_i=\mathrm{\alpha }+{\mathrm{\beta }}_x{x}_i+{\mathrm{\beta }}_z{z}_i+{\mathrm{\beta }}_{w_2}{w}_{2,i}+{\mathrm{\beta }}_{w_3}{w}_{2,i}+{\mathrm{\beta }}_{x\text{:}z}{x}_i:z_i+{\mathrm{\beta }}_{x\text{:}{w}_2}{x}_i:w_{2,i}+{\mathrm{\beta }}_{x\text{:}{w}_3}{x}_i:w_{3,i}+{\mathrm{\beta }}_{z\text{:}{w}_2}{z}_i:w_{2,i}+\\ {\mathrm{\beta }}_{z\text{:}{w}_3}{z}_i:w_{3,i}+{\mathrm{\beta }}_{x\text{:}z\text{:}{w}_2}{x}_i:z_i:w_{2,i}+{\mathrm{\beta }}_{x\text{:}z\text{:}{w}_3}{x}_i:z_i:w_{3,i}+{\mathrm{\beta }}_C{C}_i+{ϵ}_i,\end{array}$$ (3)

where $z_i$ is now a binary variable that denotes plasma total folate concentrations and that equals 1 when concentrations are $<20\text{th}$ percentile and equals 0 otherwise, and $w_i$ is a categorical variable with $k=3$ levels, where $w_1$ is the reference level, which denotes MTHFR 677C>T genotype. See the web appendix for further details.⁵² For this analysis, we decided a priori to simplify the cutoffs for plasma total folate concentration to ensure that enough people were in the lowest category and had two copies of the minor (T) MTHFR 677C>T allele.

In our primary analyses, we assessed modification by finding BLL-SRS-2 associations among participants with a $\mathrm{C}|\mathrm{C}$, $\mathrm{C}|\mathrm{T}$, or $\mathrm{T}|\mathrm{T}$ MTHFR 677C>T genotype separately, because each genotype is likely associated with a MTHFR enzyme with different levels of stability.⁵⁸ We also considered dominant (grouping participants with a $\mathrm{C}|\mathrm{T}$ and $\mathrm{T}|\mathrm{T}$ genotype together) and recessive (participants with a $\mathrm{C}|\mathrm{C}$ and $\mathrm{C}|\mathrm{T}$ genotype together) models.

Results

Descriptive Statistics

Most participating mothers lived with their spouse (97%), self-identified as being White (90%), did not report smoking cigarettes during pregnancy (92%), and had an undergraduate degree or higher (67%) (Table 1). The average ($\pm \text{standard deviation}$) maternal age at enrollment was $33\pm 4.6$ y of age. Eleven children (2%) had a SRS-2 score $\ge 60$, indicating at least mild autistic-like behaviors.^24,60 SRS-2 scores tended to be higher in male children, children from socioeconomically disadvantaged families (lower household income, lower maternal education, parents did not live together), and grew up in less-stimulating environments, as indicated by HOME score (Table 1).

Table 1.

Sociodemographic characteristics, SRS-2 scores, and BLLs ($\mathrm{\mu }\mathrm{g}/\mathrm{L}$) of participants, MIREC study, Canada, 2008–2011 ($n=601$).

Variable	n (%)	Mean SRS-2 $\text{score}\pm \text{SD}$	GM BLL (GSD)
			First trimester ($n=590$)	Third trimester ($n=550$)
All	601 (100.0)	$45.3\pm 6.1$	6.21 (1.62)	5.91 (1.67)
Child sex
Male	290 (48.3)	$46.6\pm 6.5$	6.33 (1.64)	6.15 (1.67)
Female	311 (51.7)	$44.2\pm 5.3$	6.11 (1.60)	5.69 (1.66)
Maternal age at enrollment
18–29	132 (22.0)	$46.7\pm 5.6$	5.52 (1.61)	5.26 (1.57)
30–35	292 (48.6)	$45.3\pm 6.4$	6.06 (1.60)	5.74 (1.68)
$\ge 36$	177 (29.5)	$44.4\pm 5.8$	7.05 (1.62)	6.75 (1.68)
Living with spouse
Yes	580 (96.5)	$45.2\pm 6.0$	6.21 (1.62)	5.87 (1.67)
No	21 (3.5)	$48.8\pm 7.3$	6.36 (1.54)	7.13 (1.50)
Maternal race
Self-identifies as White	539 (89.7)	$45.1\pm 6.1$	6.09 (1.61)	5.80 (1.66)
Does not self-identify as White	62 (10.3)	$47.1\pm 5.7$	7.38 (1.63)	6.96 (1.66)
Maternal education
High school or less	30 (5.0)	$47.8\pm 7.1$	6.36 (1.50)	6.25 (1.49)
College or trade school	167 (27.8)	$46.2\pm 5.8$	5.58 (1.62)	5.35 (1.63)
Undergraduate university degree	239 (39.8)	$45.4\pm 6.6$	6.36 (1.65)	6.13 (1.73)
Graduate university degree	163 (27.1)	$44.0\pm 5.1$	6.70 (1.57)	6.14 (1.64)
Missing	2 (0.3)	—	—	—
Annual household income ($CAD)
$\le \$\mathrm{40,000}$	61 (10.1)	$47.9\pm 6.1$	6.74 (1.62)	7.36 (1.68)
$40,001–$80,000	173 (28.8)	$46.3\pm 6.3$	6.05 (1.64)	5.75 (1.57)
$80,001–$100,000	116 (19.3)	$45.1\pm 6.1$	5.98 (1.58)	5.37 (1.69)
$>\$\mathrm{100,000}$	231 (38.4)	$44.1\pm 5.6$	6.40 (1.61)	6.06 (1.69)
Missing	20 (3.3)	—	—	—
Parity
Nulliparous	261 (43.4)	$46.1\pm 6.1$	6.24 (1.65)	6.06 (1.65)
Uniparous	251 (41.8)	$44.6\pm 5.7$	6.26 (1.60)	5.74 (1.68)
Multiparous	89 (14.8)	$45.1\pm 6.9$	6.01 (1.58)	5.93 (1.69)
HOME score
48–55 ($\ge$ median)	323 (53.7)	$44.3\pm 5.3$	6.13 (1.60)	5.70 (1.64)
27–47 ($<$ median)	260 (43.3)	$46.7\pm 6.7$	6.34 (1.65)	6.12 (1.70)
Missing	18 (3.0)	—	—	—
Maternal cigarette smoking during pregnancya
Smoker	47 (7.8)	$48.2\pm 6.8$	6.51 (1.68)	6.69 (1.65)
Nonsmoker	554 (92.2)	$45.1\pm 6.0$	6.19 (1.61)	5.84 (1.67)
Year of enrollment
2008	10 (1.7)	$48.0\pm 5.0$	9.38 (1.55)	9.15 (1.44)
2009	187 (31.1)	$45.2\pm 6.1$	6.55 (1.61)	6.18 (1.62)
2010	381 (63.4)	$45.3\pm 6.1$	6.03 (1.61)	5.77 (1.69)
2011	23 (3.8)	$46.0\pm 6.2$	5.54 (1.67)	4.84 (1.53)
Child age at SRS-2 assessment
40–48 months ($\ge$ median)	322 (53.6)	$45.0\pm 6.5$	6.47 (1.64)	6.13 (1.66)
36–39 months ($<$ median)	279 (46.4)	$45.8\pm 5.5$	5.94 (1.60)	5.66 (1.67)
Folic acid supplementationb
$<400\mathrm{\mu g}$ per day	34 (5.7)	$45.4\pm 5.5$	6.68 (1.58)	6.48 (1.66)
400–1,000 µg per day	423 (70.4)	$45.3\pm 6.2$	5.98 (1.62)	5.71 (1.68)
$>\mathrm{1,000}\mathrm{\mu }\mathrm{g}$ per day	144 (24.0)	$45.4\pm 6.0$	6.85 (1.60)	6.39 (1.62)
Plasma total folate concentrations (first trimester, 6–13 wk gestation)c
$<10\text{th}$ percentile ($65.6\mathrm{nmol}/\mathrm{L}$)	58 (9.7)	$45.3\pm 6.2$	6.29 (1.57)	5.49 (1.52)
$\ge 10\text{th}$ and $<80\text{th}$ percentiles	400 (66.6)	$45.3\pm 6.3$	6.15 (1.59)	5.93 (1.67)
$\ge 80\text{th}$ percentile ($125\text{\hspace{0.17em}nmol}/\mathrm{L}$)	115 (19.1)	$45.5\pm 5.6$	6.27 (1.77)	6.11 (1.77)
Missing	28 (4.7)	—	—	—
Plasma total folate concentrations (third trimester, 32–34 wk gestation)c
$<10\text{th}$ percentile ($49.3\text{\hspace{0.17em}nmol}/\mathrm{L}$)	54 (9.0)	$46.7\pm 7.3$	5.98 (1.64)	6.18 (1.76)
$\ge 10\text{th}$ and $<80\text{th}$ percentiles	372 (61.9)	$45.2\pm 6.0$	6.22 (1.60)	5.87 (1.65)
$\ge 80\text{th}$ percentile ($158\text{\hspace{0.17em}nmol}/\mathrm{L}$)	107 (17.8)	$44.7\pm 5.4$	6.18 (1.70)	5.88 (1.70)
Missing	68 (11.3)	—	—	—
Maternal MTHFR 677 C>T genotype
$\mathrm{C}|\mathrm{C}$	264 (43.9)	$45.2\pm 5.9$	6.25 (1.59)	6.05 (1.68)
$\mathrm{C}|\mathrm{T}$	264 (43.9)	$45.7\pm 6.3$	6.34 (1.63)	5.93 (1.65)
$\mathrm{T}|\mathrm{T}$	58 (9.7)	$44.7\pm 6.0$	5.63 (1.68)	5.37 (1.69)
Missing	15 (2.5)	—	—	—

Note: —, no data; BLL, blood-lead level; CAD, Canadian dollar; GM, geometric mean; GSD, geometric standard deviation; HOME, Home Observation for Measurement of the Environment; MIREC, Maternal–Infant Research on Environmental Chemicals Study; MTHFR, methylenetetrahydrofolate reductase, SD, standard deviation; SRS-2, Social Responsiveness Scale-2; THF, tetrahydrofolate; UMFA, unmetabolized folic acid.

^a “Smoker” includes current smokers and individuals who quit during pregnancy. “Nonsmoker” includes participants who did not smoke and former smokers who quit before pregnancy.

^b Folic acid supplementation was primarily measured via a survey conducted at 16 wk gestation, which queried intake in the prior 30 d. We also used data from the 24-h recall version of this survey (administered at 16 wk gestation) and a questionnaire completed at study enrollment (administered at 6–13 wk gestation).

^c Sum of 5-formyl-THF, 5-10-methylene-THF, THF, UMFA, 5-methyl-THF.

All participants had detectable concentrations of lead in their blood (Table 2). Maternal BLLs were slightly higher in the first trimester of pregnancy than in the third trimester [geometric mean (geometric standard deviation): 6.21 (1.62) vs. 5.91 (1.67) $\mathrm{\mu g}/\mathrm{L}$; Pearson $\text{correlation}=0.75$]. BLLs tended to be higher in participants who were socioeconomically disadvantaged, younger, and cigarette smokers (Table 1).

Table 2.

Distribution of maternal lead and folate biomarkers, MIREC study, Canada, 2008–2011 ($n=601$).

Trimester (weeks of gestation)	$n$	$\%>\text{LOD}$ a	GM (GSD)	Percentile
				25th	50th	75th	95th
Lead (whole blood concentrations; $\mathrm{\mu }\mathrm{g}/\mathrm{L}$)
First (6–13)	590	100	6.21 (1.62)	4.40	6.01	8.29	14.00
Third (32–34)	550	100	5.91 (1.67)	4.14	5.80	8.24	13.28
Total folate (plasma concentrations, $\mathrm{nmol}/\mathrm{L}$)b
First (6–13)	573	NAc	98.91 (1.49)	79.44	97.63	117.81	188.93
Third (32–34)	533	NAc	105.44 (1.81)	76.76	101.10	140.73	356.03

Note: —, no data; BLL, blood-lead level; GM, geometric mean; GSD, geometric standard deviation; LOD, limit of detection; MIREC, Maternal–Infant Research on Environmental Chemicals Study; MTHFR, methylenetetrahydrofolate reductase; NA, not available; SD, standard deviation; SRS-2, Social Responsiveness Scale-2; THF, tetrahydrofolate; UMFA, unmetabolized folic acid.

^a Limit of detection for BLLs is $1\mathrm{\mu g/L}$.

^b Sum of 5-formyl-THF, 5-10-methylene-THF, THF, UMFA, 5-methyl-THF. The limits of detection of these vitamers are 0.02, 0.019, 0.018, 0.029, and $0.019\text{\hspace{0.17em}nmol}/\mathrm{L}$, respectively.

^c Limit of detection not reported because plasma total folate concentrations are calculated as the sum of five individual folate vitamers, and each vitamer has a different detection rate.

Median (interquartile range) plasma total folate concentrations were similar (Pearson $\text{correlation}=0.32$) for the first [97.63 (38.37) $\text{nmol}/\mathrm{L}$] and third trimesters [101.1 (63.97) $\text{nmol}/\mathrm{L}$] but third-trimester concentrations were more right-skewed (Table 2; Figure S3). Seventy percent of participants’ FA supplementation was within the recommended range (400–1,000 $\mathrm{\mu }\mathrm{g}/\mathrm{d}$) (Table 1; Figure S4).^35–37 FA supplementation was most frequently estimated with the 30-d recall form (83%), whereas the baseline questionnaire (12%) and especially the 24-h recall form (5%) were less frequently used (Table S1). Although only 6% of participants had a below-recommended ($<400\mathrm{\mu }\mathrm{g}/\mathrm{d}$) FA supplementation,^36,37 16% of participants whose FA supplementation was estimated using the baseline questionnaire had below-recommended FA supplementation (Table S1). Finally, 44%, 44%, and 10% of mothers had zero, one, and two copies of a minor MTHFR 677C>T allele, respectively (Table 1).

Associations between SRS-2 Scores and Gestational BLLs

After controlling for confounders, each two-fold difference in first-trimester BLLs was associated with a 0.4-point increase in child SRS-2 scores (95% CI: $-0.3$, 1.1). Third-trimester BLL-SRS-2 associations were essentially null (${\mathrm{\beta }}_{adj}=0.1$; 95% CI: $-0.7$, 0.8). We found no evidence that child sex modified BLL-SRS-2 associations (Table 3).

Table 3.

Adjusted associations between blood-lead levels (2-fold difference) and SRS-2 scores by child sex, gestational plasma total folate concentrations, gestational folic acid supplementation, and maternal MTHFR 677C>T genotype MIREC study, Canada, 2008–2011 ($n=601$).

Time of BLL measurement	Modifying variable	β (95% CI)a	$p$-Interaction
First trimester (6–13 wk gestation)	Full sample ($n=601$)	0.4 ($-0.3$, 1.1)	NA
	Child sex	—	—
	Male ($n=290$)	0.5 ($-0.5$, 1.5)	Ref
	Female ($n=311$)	0.3 ($-0.7$, 1.3)	0.77
	First trimester plasma total folate concentrationsb,c	—	—
	$\ge 10\text{th}$ and $<80\text{th}$ percentiles	0.0 ($-0.9$, 0.9)	Ref
	$<10\text{th}$ percentile ($65.6\mathrm{nmol}/\mathrm{L}$)	1.7 ($-0.6$, 4.0)	0.16
	$\ge 80\text{th}$ percentile ($125\mathrm{nmol}/\mathrm{L}$)	1.0 ($-0.3$, 2.3)	0.19
	Folic acid supplementationd	—	—
	400–1,000 $\mathrm{\mu }\mathrm{g}/\mathrm{d}$ ($n=423$)	0.2 ($-0.6$, 1.1)	Ref
	$\mathrm{<400\hspace{0.17em}\mu }\mathrm{g}/\text{day}$ ($n=34$)	0.8 ($-2.2$, 3.7)	0.73
	$>\mathrm{1,000}\mathrm{\mu }\mathrm{g}/\mathrm{d}$ ($n=144$)	0.9 ($-0.5$, 2.4)	0.40
	MTHFR 677 C>T genotypec	—	—
	$\mathrm{C}|\mathrm{C}$	$-0.1$ ($-1.2$, 1.0)	Ref
	$\mathrm{C}|\mathrm{T}$	0.8 ($-0.2$, 1.9)	0.21
	$\mathrm{T}|\mathrm{T}$	0.4 ($-1.6$, 2.5)	0.64
Third trimester (32–34 wk gestation)	Full sample ($n=601$)	0.1 ($-0.7$, 0.8)	NA
	Child sex	—	—
	Boys ($n=290$)	0.1 ($-0.8$, 1.1)	Ref
	Girls ($n=311$)	0.0 ($-1.0$, 1.0)	0.79
	Third trimester plasma total folate concentrationsb,c	—	—
	$\ge 10\text{th}$ and $<80\text{th}$ percentiles	$-0.3$ ($-1.2$, 0.5)	Ref
	$<10\text{th}$ percentile ($49.3\mathrm{nmol}/\mathrm{L}$)	3.3 (1.1, 5.5)	$p<.001$
	$\ge 80\text{th}$ percentile ($158\mathrm{nmol}/\mathrm{L}$)	$-0.9$ ($-2.4$, 0.5)	0.47
	Folic acid supplementationd	—	—
	400–1,000 $\mathrm{\mu }\mathrm{g}/\mathrm{d}$ ($n=423$)	$-0.1$ ($-0.9$, 0.7)	Ref
	$\mathrm{<400\hspace{0.17em}\mu }\mathrm{g}/\mathrm{d}$ ($n=34$)	2.1 ($-0.9$, 5.1)	0.13
	$\mathrm{>}\mathrm{1,000}\mathrm{\mu }\mathrm{g}/\mathrm{d}$ ($n=144$)	0.1 ($-1.4$, 1.6)	0.77
	MTHFR 677 C>T genotypec	—	—
	$\mathrm{C}|\mathrm{C}$	$-0.3$ ($-1.3$, 0.7)	Ref
	$\mathrm{C}|\mathrm{T}$	0.2 ($-0.9$, 1.3)	0.44
	$\mathrm{T}|\mathrm{T}$	0.9 ($-1.2$, 3.1)	0.28

Note: Linear regression models were used. “Interaction $p$-value” refer to the two-sided $p$-value of the interaction term. —, no data; BLL, blood-lead level; CI, confidence interval; MIREC, Maternal–Infant Research on Environmental Chemicals; MTHFR, methylenetetrahydrofolate reductase; NA, not available; Ref, reference; SRS-2, Social Responsiveness Scale-2; THF, tetrahydrofolate; UMFA, unmetabolized folic acid.

^a Controlled for the following variables: child sex, child age at SRS-2 assessment, HOME score, household income, relationship status, maternal education, maternal race, maternal age, parity, smoking status, city of residence, and year of enrollment.

^b The sum of 5-formyl-THF, 5-10-methylene-THF, THF, UMFA, and 5-methyl-THF.

^c The number of participants in each category cannot be reported because the plasma total folate and maternal MTFHR 677C>T genotype variables were imputed with MICE. We created 10 imputed datasets, each with a different number of participants in each imputed dataset.

^d Folic acid supplementation was primarily measured via a survey conducted at 16 wk gestation, which queried intake in the past 30 d. We also used data from the 24-h recall version of this survey (administered at 16 wk gestation) and a questionnaire completed at study enrollment (administered at 6–13 wk gestation).

Next, we assessed whether folate status could modify BLL-SRS-2 associations. First-trimester BLL-SRS-2 associations were slightly stronger among participants in the lowest category of plasma total folate concentrations ($<10\text{th}$ percentile) (${\mathrm{\beta }}_{adj}=1.7$; 95% CI: $-0.6$, 4.0) in comparison with those in the middle category ($\ge 10\text{th}$ and $<80\text{th}$ percentiles) (${\mathrm{\beta }}_{adj}=0$; 95% CI: $-0.9$, 0.9) ($p\text{-interaction}=0.16$). Third-trimester BLL-SRS-2 associations were also stronger among participants in the lowest category of third-trimester plasma total folate concentrations (${\mathrm{\beta }}_{adj}=3.3$; 95% CI: 1.1, 5.5) vs. the middle category (${\mathrm{\beta }}_{adj}=-0.3$; 95% CI: $-1.2$, 0.5) ($p\text{-interaction}<0.001$) (Table 3).

Being in the highest category plasma total folate concentrations ($\ge 80\text{th}$ percentile) did not clearly modify BLL-SRS-2 associations. First-trimester BLL-SRS-2 associations were slightly stronger among participants in the highest (${\mathrm{\beta }}_{adj}=1.0$; 95% CI: $-0.3$, 2.3) vs. the middle category (${\mathrm{\beta }}_{adj}=0$; 95% CI: $-0.9$, 0.9) of first-trimester plasma total folate concentrations ($p\text{-interaction}=0.19$). However, this was not observed using third-trimester biomarkers (Table 3).

Low FA supplementation may have also modified BLL-SRS-2 associations, but the differences were not apparent in our data at all time points. First-trimester BLL-SRS-2 associations were slightly stronger among participants with low ($<400\mathrm{\mu }\mathrm{g}/\mathrm{d}$) (${\mathrm{\beta }}_{adj}=0.8$; 95% CI: $-2.2$, 3.7) vs. adequate (400–1,000 $\mathrm{\mu }\mathrm{g}/\mathrm{d}$) (${\mathrm{\beta }}_{adj}=0.2$; 95% CI: $-0.6$, 1.1) FA supplementation, but the difference was imprecise ($p\text{-interaction}=0.73$). Among participants with low FA supplementation, third-trimester BLLs were more strongly associated with SRS-2 scores (${\mathrm{\beta }}_{adj}=2.1$; 95% CI: $-0.9$, 5.1) compared with those with adequate FA supplementation (${\mathrm{\beta }}_{adj}=-0.1$; 95% CI: $-0.9$, 0.7) ($p\text{-interaction}=0.13$) (Table 3).

Again, first-trimester BLL-SRS-2 associations were slightly stronger among participants with high ($>\mathrm{1,000}\mathrm{\mu }\mathrm{g}/\mathrm{d}$) (${\mathrm{\beta }}_{adj}=0.9$; 95% CI: $-0.5$, 2.4) vs. adequate (${\mathrm{\beta }}_{adj}=0.2$; 95% CI: $-0.6$, 1.1) FA supplementation ($p\text{-interaction}=0.40$). However, third-trimester BLL-SRS-2 associations were similar among participants with high and adequate FA supplementation (Table 3).

Next, first-trimester BLL-SRS-2 associations were null (${\mathrm{\beta }}_{adj}=-0.1$; 95% CI: $-1.2$, 1.0) among participants with no minor alleles ($\mathrm{C}|\mathrm{C}$) but were slightly stronger (${\mathrm{\beta }}_{adj}=0.8$; 95% CI: $-0.2$, 1.9) among participants with one copy ($\mathrm{C}|\mathrm{T}$) of the minor allele ($p\text{-interaction}=0.21$). This trend reversed among participants with two copies ($\mathrm{T}|\mathrm{T}$) of the minor allele (${\mathrm{\beta }}_{adj}=0.4$; 95% CI: $-1.6$, 2.5) (p-interaction vs. $\mathrm{C}|\mathrm{C}=0.64$). Third-trimester BLL-SRS-2 associations were null among participants with no minor alleles ($\mathrm{C}|\mathrm{C}$) (${\mathrm{\beta }}_{adj}=-0.3$; 95% CI: $-1.3$, 0.7), but associations became progressively stronger in participants with one ($\mathrm{C}|\mathrm{T}$) (${\mathrm{\beta }}_{adj}=0.2$; 95% CI: $-0.9$, 1.3) and two ($\mathrm{T}|\mathrm{T}$) (${\mathrm{\beta }}_{adj}=0.9$; 95% CI: $-1.2$, 3.1) minor alleles (p-interaction for $\mathrm{C}|\mathrm{T}$ vs. $\mathrm{C}|\mathrm{C}=0.44$; for $\mathrm{T}|\mathrm{T}$ vs. $\mathrm{C}|\mathrm{C}=0.28$) (Table 3).

Exploratory Analyses

Next, we flexibly modeled BLL-SRS-2 associations across plasma total folate concentrations (Figures 1A and 1B; Excel Tables S1 and S2). First-trimester BLL-SRS-2 associations were positive when first-trimester plasma total folate concentrations were lower and higher (U-shaped) (Figure 1A; Excel Table S1). Third-trimester BLL-SRS-2 associations were positive when plasma total folate concentrations were lower and negative otherwise (Figure 1B; Excel Table S2). These results are consistent with our primary analysis (Table 3).

Figure 1.

Adjusted associations (solid black line; controlled for the following variables: child sex, child age at SRS-2 assessment, HOME score, household income, relationship status, maternal education, maternal race, maternal age, parity, smoking status, city of residence, and year of enrollment) and 95% CIs (dashed gray lines) between gestation BLLs and SRS-2 scores while fixing plasma total folate concentrations (the sum of 5-formyl-THF, 5-10-methylene-THF, THF, UMFA, and 5-methyl-THF) to the specified value on the x-axis. (A) First trimester BLLs and plasma total folate concentrations. (See Supplementary Excel Tables S1 for numeric results). (B) Third trimester BLLs and plasma total folate concentrations. (See Supplementary Excel Tables S2 for numeric results). Note: Plasma total folate concentrations were fitted with a restricted cubic spline. Fifth to 95th percentiles of plasma total folate concentrations are displayed; full range not shown to avoid extrapolation. MICE was not implemented due to software incompatibilities. Note: BLL, blood-lead level; CI, confidence interval; MICE, multiple imputation by chained equations; MIREC, Maternal–Infant Research on Environmental Chemicals; SRS-2, Social Responsiveness Scale-2; THF, tetrahydrofolate; UMFA, unmetabolized folic acid.

We also fit a three-way interaction model to simultaneously assess whether plasma total folate concentrations and maternal MTHFR 677C>T genotype can modify BLL-SRS-2 associations (Table 4). It was unclear whether this SNP could modify BLL-SRS-2 associations among participants with lower ($<20\text{th}$ percentile) plasma total folate concentrations, because CIs were very wide. Among participants with higher ($\ge 20\text{th}$ percentile) third-trimester plasma total folate concentrations, third-trimester BLL-SRS-2 associations were stronger among participants with two minor alleles ($\mathrm{T}|\mathrm{T}$) (${\mathrm{\beta }}_{adj}=1.6$; 95% CI: $-0.6$, 3.8) vs. zero minor alleles ($\mathrm{C}|\mathrm{C}$) (${\mathrm{\beta }}_{adj}=-1.3$; 95% CI: $-2.5$, $-0.1$) ($p\text{-interaction}=0.02$) (Table 4). This trend was also observed in our primary analysis, although the differences in associations were less stark (Table 3). In addition, among participants with a $\mathrm{C}|\mathrm{C}$ genotype, third-trimester BLL-SRS-2 associations were stronger among participants with lower ($<20\text{th}$ percentile) third-trimester plasma total folate concentrations (${\mathrm{\beta }}_{adj}=2.2$; 95% CI: 0.1, 4.3) vs. those with higher ($>20\text{th}$ percentile) concentrations (${\mathrm{\beta }}_{adj}=-1.3$; 95% CI: $-2.5$, $-0.1$) ($p\text{-interaction}=0.003$) (Table 4). Again, this finding is consistent with our main analysis (Table 3).

Table 4.

Adjusted associations between blood-lead levels (2-fold difference) and SRS-2 scores with a three-way interaction model including gestational plasma total folate concentrations and maternal MTHFR 677 C>T genotype, MIREC study, Canada, 2008–2011 ($n=601$).

Time of BLL and plasma total folate concentration measurement	Plasma total folate concentrationb	MTHFR genotype	β (95% CI)a	$p$-Interaction
First trimester (6–13 wk gestation)	$\ge 20\text{th}$ percentile ($76.2\mathrm{nmol}/\mathrm{L}$)	$\mathrm{C}|\mathrm{C}$	$-0.4$ ($-1.6$, 0.7)	Ref
	—	$\mathrm{C}|\mathrm{T}$	0.2 ($-0.9$, 1.4)	0.42
	—	$\mathrm{T}|\mathrm{T}$	0.7 ($-1.8$, 3.1)	0.43
	$<20\text{th}$ percentile	$\mathrm{C}|\mathrm{C}$	1.6 ($-1.4$, 4.6)	0.20
	—	$\mathrm{C}|\mathrm{T}$	3.2 (0.1, 6.3)	0.66
	—	$\mathrm{T}|\mathrm{T}$	$-1.1$ ($-5.7$, 3.5)	0.19
Third trimester (32–34 wk gestation)	$\ge 20\text{th}$ percentile ($69.0\mathrm{nmol}/\mathrm{L}$)	$\mathrm{C}|\mathrm{C}$	$-1.3$ ($-2.5$, $-0.1$)	Ref
	—	$\mathrm{C}|\mathrm{T}$	$-0.1$ ($-1.5$, 1.3)	0.15
	—	$\mathrm{T}|\mathrm{T}$	1.6 ($-0.6$, 3.8)	0.02
	$<20\text{th}$ percentile	$\mathrm{C}|\mathrm{C}$	2.2 (0.1, 4.3)	0.003
	—	$\mathrm{C}|\mathrm{T}$	1.9 ($-0.6$, 4.4)	0.37
	—	$\mathrm{T}|\mathrm{T}$	$-0.8$ ($-17.1$, 15.5)	0.47

Note: Linear regression models were used. “Interaction $p$-value” refers to the two-sided $p$-value of the interaction term. —, no data; BLL, blood-lead level; CI, confidence interval; MIREC, Maternal–Infant Research on Environmental Chemicals; MTHFR, methylenetetrahydrofolate reductase; Ref, reference; SRS-2, Social Responsiveness Scale-2; THF, tetrahydrofolate; UMFA, unmetabolized folic acid.

^a Controlled for the following variables: child sex, child age at SRS-2 assessment, HOME score, household income, relationship status, maternal education, maternal race, maternal age, parity, smoking status, city of residence, and year of enrollment.

^b The sum of 5-formyl-THF, 5-10-methylene-THF, THF, UMFA, and 5-methyl-THF.

Using dominant and recessive models for the MTHFR 677C>T polymorphism gave consistent results (Table S2–S3). Our three-way interaction analysis found that among participants with a $\mathrm{C}|\mathrm{C}$ or $\mathrm{C}|\mathrm{T}$ genotype, first-trimester BLL-SRS-2 associations were stronger among participants with lower first-trimester plasma total folate concentrations ($<20\text{th}$ percentile: ${\mathrm{\beta }}_{adj}=2.6$; 95% CI: 0.7, 4.5; $\ge 20\text{th}$ percentile: ${\mathrm{\beta }}_{adj}=-0.1$; 95% CI: $-0.9$, 0.8; $p-\text{interaction}=0.01$) (Table S3). These differences in associations were less apparent in our primary analysis where we did not stratify by MTHFR 677C>T genotype (Table 3).

Discussion

Our study is the first, to our knowledge, to find evidence that the associations between gestational BLLs and childhood autistic-like behaviors are stronger when gestational plasma total folate concentrations or FA supplementation are lower. BLL-SRS-2 associations among participants in the middle category of these folate indicators were null (${\mathrm{\beta }}_{adj}$ ranging from $-0.3$ to 0.2). These associations were consistently stronger among participants in the lowest category of these folate indicators (${\mathrm{\beta }}_{adj}$ ranging from 0.8 to 3.3, or 1/10 to one-half of a standard deviation increase in child SRS-2 scores for each doubling of BLLs). These findings corroborate a growing body of research showing that the associations between chemicals and autism or autistic-like behaviors may be attenuated by folate.^13–17

We found no clear evidence, however, that being in the highest category of these folate indicators was associated with weaker BLL-SRS-2 associations in comparison with those in the middle category. First-trimester BLL-SRS-2 associations were slightly stronger among participants in the highest category of plasma total folate concentrations or FA supplementation in comparison with the middle category. However, results were imprecise and may be due to unmeasured confounding, and this was not observed during the third trimester of pregnancy.

Folate-mediated one-carbon metabolism provides methyl groups used in cellular methylation reactions, such as DNA methylation, and is required for de novo nucleotide synthesis and DNA repair.^20,58,61,62 In contrast, lead may impede DNA methylation^63–66 and induce DNA damage.⁶⁷ Thus, the observed modification of BLL-SRS-2 associations by folate is plausible, given their antagonistic properties.¹⁷ We investigated this hypothesis (originally proposed by Schmidt et al.¹⁷) using folate-related genotype data. The MTHFR 677C>T SNP decreases the efficiency of the MTHFR enzyme. Homozygosity for the minor allele is associated with lower folate status and reduces one-carbon metabolism, with consequences for cellular methylation potential and nucleotide synthesis.^20,21,58 Thus this SNP could modify BLL-SRS-2 associations in a similar manner as folate intakes or status.¹⁷ Third-trimester BLL-SRS-2 associations were slightly stronger among participants with two copies of the minor allele ($\mathrm{T}|\mathrm{T}$) in comparison with those with none ($\mathrm{C}|\mathrm{C}$), but the difference was not statistically significant. Interestingly, first-trimester BLL-SRS-2 associations were stronger among participants with one copy of the minor allele ($\mathrm{C}|\mathrm{T}$) in comparison with those with none ($\mathrm{C}|\mathrm{C}$). However, this trend reversed among participants with two copies of the minor allele ($\mathrm{T}|\mathrm{T}$). Again, these differences were not statistically significant. We found clearer evidence that MTHFR genotype modifies BLL-SRS-2 associations in an exploratory analysis that accounted for the fact that the stability of the MTHFR enzyme depends on folate concentrations.^56,57 Specifically, among participants with higher ($\ge 20\text{th}$ percentile) third-trimester plasma total folate concentrations, third-trimester BLL-SRS-2 associations were stronger among participants with two copies of the minor allele ($\mathrm{T}|\mathrm{T}$) in comparison with those with none ($\mathrm{C}|\mathrm{C}$). Results for participants with lower folate status were inconclusive due to small sample sizes and wide CIs. Overall, these findings suggest that impaired one-carbon metabolism may explain the modifying effect of folate on BLL-SRS-2 associations.

A key strength of our paper is our use of two indicators of folate exposure (plasma total folate concentrations and FA supplementation) to explore the modifying effect of folate on BLL-SRS-2 associations. These folate exposure ascertainment methods have unique limitations and were measured at different time points. Plasma total folate concentrations may be affected by pharmacokinetic and genetic factors.⁶⁸ In our primary analysis, we categorized plasma total folate concentrations using cutoffs based on the distribution of plasma total folate concentrations in our sample. These cutoffs may not be clinically or biologically relevant.⁶⁹ Still, an exploratory analysis where we considered the modifying effect of continuous plasma total folate concentrations yielded consistent results. In addition, plasma total folate concentrations reflect recent folate intake and are more responsive to the consumption of FA than naturally occurring dietary folate.³⁴ Most samples were nonfasted, and information on time since the last meal, snack, or FA-containing supplement was not recorded, decreasing the precision of our plasma total folate estimates.^32–34 Alternatively, we could have assessed erythrocyte folate concentrations, an indicator of longer-term folate status. Although this biomarker maybe more precise, it has its own limitations. Erythrocyte–folate concentrations are a lagging indicator^38,70 and can therefore underestimate folate status in individuals who have recently increased their folate consumption, which is frequently the case in the preconceptional and initial stages of pregnancy.

We also considered folate consumption, but this also has limitations. In our study, we did not assess dietary folate intake (e.g., from fortified grain products or foods rich in naturally occurring folate) but rather focused on FA supplementation. Still, FA supplementation is a major determinant of gestational folate status.^34,72 FA supplement use was also estimated with self-reported information, making it susceptible to bias and systematic error.⁷¹ In this respect, FA supplementation was complemented by plasma total folate concentrations being a more direct indicator of exposure. We used three different survey tools that account for slightly different time periods to estimate FA supplementation, which could have resulted in misclassification of a small proportion of participants. Participants whose FA supplementation was estimated with the baseline questionnaire (who accounted for 12% of our sample) were the most likely to have low ($<400\mathrm{\mu }\mathrm{g}/\mathrm{d}$) FA supplementation. These participants may have had lower FA supplementation because the baseline questionnaire reflects FA supplementation that occurs earlier in pregnancy, when supplementation may be lower.⁷³ These differences could also be because we used the baseline questionnaire when participants did not complete the other two FA supplementation surveys, and nonresponders may be less likely to take FA supplements due to social desirability bias.⁷¹ We also did not measure FA supplementation during the third trimester of pregnancy. In addition, we did not measure FA supplementation or folate biomarkers during the early gestational (first month of gestation) or the preconceptional period, despite previous research suggesting that folate plays an important role in autism outcomes during these times.⁷³ Finally, relatively few people were in the highest and especially the lowest categories of these two folate indicators, contributing to reduced precision when assessing interaction. Despite these limitations, the modification of BLL-SRS-2 associations by folate appears robust, given the similar pattern of results using these two different methods of folate exposure.

Our study has several other general limitations. Our study may have reduced external validity because MIREC participants were largely more educated, older, more likely to identify as White, and more affluent than the general Canadian population.²² An important aspect is that our sample was folate-replete (because FA fortification is mandatory in Canada and the vast majority of participants reported sufficient FA supplementation). Our sample also had relatively lower lead exposure than that of nonpregnant Canadian women 20–39 y of age who participated in cycle 1 (2007–2009) of the Canadian Health Measurement Survey.⁷⁴ This lower level of exposure may prevent us from observing full dose–response curves for the lead–SRS–2 relationship.⁷⁵ The SRS-2 may overestimate autistic-like behaviors in children with lower IQ scores, attention deficit hyperactivity disorder (ADHD), and/or behavioral problems.^48,76

Conclusion

In this Canadian cohort, the association between gestational lead exposure and autistic-like behaviors in preschool-age children was stronger when gestational plasma total folate concentrations or FA supplementation were lower. The benefits of prenatal FA supplementation for reducing the risk of an neural tube defect are well documented, and our results suggest that FA supplementation during pregnancy may have additional benefits in mitigating the effects of lead exposure. Our work builds on a body of research that suggests that folate may be useful for mitigating the neurotoxic effects of chemicals.^13–17,19 Further research with a wider variety of neurodevelopmental outcomes (e.g., internalizing and externalizing behaviors, IQ) and in populations with higher lead exposure and/or lower folate consumption may be necessary to explore these relationships.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.