Maternal Vitamin D Levels During Pregnancy in Association With Autism Spectrum Disorders (ASD) or Intellectual Disability (ID) in Offspring; Exploring Non-linear Patterns and Demographic Sub-groups

Abstract

Increasing vitamin D deficiency and evidence for vitamin D’s role in brain and immune function have recently led to studies of neurodevelopment; however, few are specific to autism spectrum disorder (ASD) and vitamin D in pregnancy, a likely susceptibility period. We examined this in a case–control study of 2000–2003 Southern Californian births; ASD and intellectual disability (ID) were identified through the Department of Developmental Services and controls from birth certificates (N = 534, 181, and 421, respectively, in this analysis). Total 25-Hydroxyvitamin D (25(OH)D) was measured in mid-pregnancy serum, categorized as deficient (<50 nmol/L), insufficient (50–74 nmol/L), or sufficient (≥75 nmol/L, referent category), and examined continuously (per 25 nmol/L). Crude and adjusted odds ratios (AORs) and 95% confidence intervals (95% CI) were calculated. Non-linearity was examined with cubic splines. AORs (95% CI) for ASD were 0.79 (0.49–1.3) for maternal deficiency (9.5%), 0.93 (0.68–1.3) for insufficiency (25.6%), and 0.95 (0.86, 1.05) for linear continuous 25(OH)D. Results were similarly null for ASD with or without ID, and ID only. Interactions were observed; non-Hispanic whites (NHW) (AOR = 0.82, 95% CI = 0.69–0.98) and males (AOR = 0.89, 95% CI = 0.80–0.99) had protective associations for ASD with continuous 25(OH)D. A positive association with ASD was observed in females (AOR = 1.40, 95% CI = 1.06–1.85). With splines, a non-linear inverted j-shaped pattern was seen overall (P = 0.009 for non-linearity), with the peak around 100 nmol/L; a non-linear pattern was not observed among NHW, females, nor for ID. Our findings from a large study of ASD and prenatal vitamin D levels indicate that further research is needed to investigate non-linear patterns and potentially vulnerable sub-groups.

Lay Summary:

We studied whether mothers’ vitamin D levels during pregnancy were related to their children having autism (or low IQ) later. Low vitamin D levels were not related to greater risk of autism or low IQ in children overall. With higher levels of mothers’ vitamin D, risk of autism went down in boys, but went up in girls. Risk of autism also went down in children of non-Hispanic white mothers with higher vitamin D levels, but we did not find a relation in other race/ethnic groups.

Introduction

Autism spectrum disorders (ASDs) include a range of serious developmental disabilities with lifelong consequences, the etiology of which is complex and not well-elucidated [Lyall et al., 2017a]. Recent research indicates environmental factors play a role in ASD, potentially acting in combination with genetic susceptibilities [Grandjean & Landrigan, 2014; Lyall et al., 2017b].

Furthermore, maternal immune system aberrations have been associated with ASD, including chronic inflammation, elevated pro-inflammatory cytokine profiles during mid-gestation, the presence of fetal brain-reactive maternal autoantibodies, and changes in immune cell function in affected children and mothers [Gesundheit et al., 2013; Jones et al., 2017; Lyall, Ashwood, Van de Water, & Hertz-Picciotto, 2014; Meltzer & Van de Water, 2017].

Vitamin D deficiency or insufficiency has been increasing in the last 20 years [Centers for Disease Control and Prevention, 2012], with rates reported between 25 and 80%, including among pregnant women [Hamilton et al., 2010; Kiely, Hemmingway, & O’Callaghan, 2017; Marshall, Mehta, Ayers, Dhumal, & Petrova, 2016]. There is evidence of associations of vitamin D deficiency with a wide range of maternal and child health conditions in addition to its known role in calcium metabolism [Basit, 2013; Dror, King, Durand, & Allen, 2011; Holick, 2007; Marshall et al., 2016]. Vitamin D plays a well-documented role in mediating immune function [Cantorna, Zhu, Froicu, & Wittke, 2004; Hayes, Nashold, Spach, & Pedersen, 2003; Hewison, 2012] and in brain development [Eyles, Burne, & McGrath, 2013; Eyles, Smith, Kinobe, Hewison, & McGrath, 2005; Macova, Bicikova, Ostatnikova, Hill, & Starka, 2017]. This, along with parallel increases in prevalence of ASD [Christensen et al., 2016; Lyall, Croen, Daniels, et al., 2017a], led to the suggestion of a possible association of vitamin D deficiency with ASD a decade ago [Cannell, 2008].

“Vitamin D” refers to a group of steroid molecules, with formation of vitamin D₃, the major circulating form in the body, dependent on sunlight (UVB) exposure. Lower vitamin D concentrations are typically found in individuals that have darker-pigmented skin, who live at higher latitudes, avoid sun exposure (via use of sunscreen, clothing, or indoor activities), or when measured in the winter [Holick, 2007; Humble, Gustafsson, & Bejerot, 2010]. Other sources of vitamin D include supplements and a few foods, such as fatty fish, mushrooms, and eggs. Levels may also vary by age, parity and obesity [Bodnar et al., 2007; Dror et al., 2011]. Maternal vitamin D levels during pregnancy are of particular concern because they are the sole source of vitamin D for the fetus or neonate [Hollis, Johnson, Hulsey, Ebeling, & Wagner, 2011; Kiely et al., 2017; Marshall et al., 2016], and have been proposed to be involved in maternal immuno-tolerance to the foreign fetus [Wagner & Hollis, 2018]. Yet the optimal level of vitamin D to prevent adverse pregnancy outcomes is not clear and separate standards of intake for pregnant women are not well established, although there is movement toward supplementation of pregnant women [ACOG Committee on Obstetric Practice, 2011; Hollis et al., 2011; Kiely et al., 2017; Urrutia & Thorp, 2012].

Initial studies of potential associations between vitamin D and ASD often examined surrogates of vitamin D status, such as latitude or season, or measured vitamin D levels after ASD diagnosis, so likely reflected current lifestyle [Du et al., 2015; Fernell et al., 2010; Humble et al., 2010; Meguid, Hashish, Anwar, & Sidhom, 2010; Molloy, Kalkwarf, Manning-Courtney, Mills, & Hediger, 2010; Mostafa & Al-Ayadhi, 2012]. Only recently have vitamin D levels (typically as total 25-hydroxyvitamin D, e.g. 25(OH)D) been measured pre- or peri-natally, a critical period for brain development. A recent meta-analysis that included five studies reported a pooled odds ratio of 0.42 (95% CI 0.25, 0.71) for highest vs. lowest prenatal vitamin D and “autismrelated traits” [Garcia-Serna & Morales, 2019]. The authors also calculated summary protective associations for attention deficit and hyperactivity disorder (ADHD), but not for global IQ/cognitive development or psychomotor development, with studies on language development and behavioral difficulties even more inconsistent. Additional studies of 25(OH)D measured in newborn blood spots have been published since, including our study conducted in the same sample as the current analyses and another California study, neither of which found associations with ASD and ID or broader developmental delay [Schmidt, Niu, Eyles, Hansen, & Iosif, 2019; Windham et al., 2019]. Both studies, conducted independently on different study samples in California, showed some interactions with race/ethnicity and sex, but in different directions. Even more recently, a study from Sweden reported higher risk of ASD among children who had deficient vitamin D levels as neonates, as well as a protective association as maternal levels increased, only among the Nordic (i.e. “white”) sub-set [Lee et al., 2019].

Studies conducted thus far suggest that 25(OH)D in early to mid-pregnancy may be the most important period for the effects of vitamin D on brain developmental processes of most relevance to ASD [Garcia-Serna & Morales, 2019]. Therefore, our objective for this analysis was to determine the association of ASD with measured vitamin D levels during an earlier window of development than our prior newborn analysis, i.e. mid-gestation, and to examine interactions by sex and race/ethnicity, as well as potential non-linear effects. To evaluate the specificity of any such associations, we also examined the association with intellectual disability (ID), both among children with ASD and without.

Methods

Study Population

As in our prior study of vitamin D examining neonatal levels [Windham et al., 2019], we used data from the Early Markers for Autism (EMA) study, a population-based case control study designed to identify biologic markers of autism using archived prenatal and newborn blood samples from California state-wide genetic screening programs [Croen et al., 2008]. The EMA sample was drawn from the population of children born in three counties in Southern California from January 2000 through June 2003, who had live birth records linked to banked prenatal and newborn specimens. For this study, we used samples from phase 3 of EMA, which had excluded observations included in earlier phases (conducted in only one of the three counties, births July 2000 to September 2001). Study activities were approved by the Institutional Review Board at Kaiser Permanente and the State of California Health and Human Services Agency Committee for the Protection of Human Subjects.

Case Ascertainment

Diagnostic information was obtained from the California Department of Developmental Services (DDS), which provides services to individuals with substantial disabilities—including ASD and other developmental disabilities (http://www.dds.ca.gov). DDS data were linked to eligible live birth records for the study region and birth years to identify, based on DDS service codes: children with ASD or children with ID without autism (originally: “mental retardation of unknown etiology,” thus excluding Down Syndrome for example). General population (GP) co ntrols were identified by random sampling from eligible birth certificates, frequency matched to original ASD cases by sex, birth month and birth year. Children who died in the first year of life or became clients of DDS were excluded from the eligible control population.

Final case status was determined by expert clinician review of diagnostic and clinical data abstracted from DDS records, using an approach based on national surveillance methodology [Christensen et al., 2016; YearginAllsopp et al., 2003] and previously described [Windham et al., 2019]. A final diagnostic classification of ASD was made if DSM-IV-TR criteria were considered met. Among those with a prenatal specimen for vitamin D measure and a final case status, 431 children originally identified as ASD in DDS met these diagnostic criteria. Additionally, a total of 139 individuals originally identified as mental retardation were reclassified as ASD following expert review. Final classification of ID (ID without ASD) was based on standardized test score results found in records (composite developmental/cognitive score <70) in the absence of meeting ASD criteria, yielding 193 children in this sample. ASD cases were further qualified by presence/absence of co-occurring ID. Case groups were compared to 441 controls.

Prenatal Samples and Vitamin D Measurements

Maternal mid-pregnancy specimens were retrieved from the California Department of Public Health’s prenatal screening specimen archive, which consists of maternal serum and blood cell pellet specimens. These specimens were obtained as part of routine prenatal expanded alpha-fetoprotein (XAFP) screening during the second trimester. Maternal specimens were collected in serum separator tubes by obstetrical care service providers and underwent XAFP testing within 7 days of collection at a central laboratory (median time = 3 days). After 1–2 days of refrigeration, leftover specimens were stored at −20°C at the Genetic Disease Screening Program. Banking began in November 1999, thereby affecting availability for early 2000 births. Consent forms for the XAFP Screening program stipulated that specimens and results from prenatal testing could be used for legitimate research purposes given appropriate IRB approval.

Maternal serum was assayed for a number of markers in EMA; specimens previously used to measure immunoglobulins/cytokines [Jones et al., 2017], with the remainder stored frozen at the University of California, Davis laboratory, were aliquoted (11 uL/sample) in March 2016 and shipped frozen to the laboratory at the University of Queensland, Brisbane, Australia.

“Vitamin D” is a general term that actually describes a family of steroid compounds that are related structurally to cholesterol. The major circulating form, 25-hydroxyvitamin D (25(OH)D), also known as 25-hydroxycholecalciferol, is a reliable indicator of cumulative production and is therefore used clinically to measure vitamin D levels. Total 25-Hydroxyvitamin D (sum of 25(OH)D₂ and 25(OH)D₃, hereinafter referred to as 25(OH)D, was measured by a sensitive isotope dilution liquid chromatography–tandem mass spectrometry method (LC/MS/MS) in 3 uL aliquots [Eyles et al., 2009; Kvaskoff et al., 2016]. 25(OH) D₃ was measurable in all specimens and 18% had detectable 25OH D₂. Method limit of reporting was 1 and 10 for 25(OH)D₃ and 25(OH)D₂, respectively. Coefficients of variation (CV’s) for international 25(OH)D reference materials [National Institute for Standards and Technology (NIST SRM972a)] centered around 10% (8–12% depending on level and whether 25(OH)D₃ vs. 25(OH)D₂).

Covariate Data

Potential covariates, including demographics, and pregnancy-related characteristics were obtained from birth certificates, the CA Genetic Disease Screening Program, and DDS. Potential covariates were selected on the basis of a priori associations with ASD and/or vitamin D. Birth characteristics included the original matching variables: birth year, birth month categorized into seasons of Winter (Dec.-Feb.), Spring (March–May), Summer (June–August), and Fall (Sept.-Nov), and child sex. Maternal characteristics, all categorized, included race/ethnicity (non-Hispanic white [NHW], Hispanic white, and Black, Asian or other/unknown), age, education, parity, insurance status at delivery, weight at time of blood draw (quartiles) and season at blood draw (associated with birth month, but roughly opposite seasons).

Statistical Analysis

Additional selection criteria included restricting to singletons, one child per family (with ASD index child given priority, and ID next), and one prenatal screening sample per child collected at ≤20 gestational weeks (preferentially the earliest specimen within 15–20 weeks), leaving 534 classified as ASD, 181 as ID, and 421 controls. We categorized vitamin D levels as deficient (<50 nmol/L), insufficient (50–74 nmol/L) and sufficient (≥75 nmol/L), based on IOM recommendations [Institute of Medicine, 2011], and also examined continuous levels (reported per 25 nmol/L 25(OH)D). Covariates were compared between the ASD, ID, and control groups by chi-square tests. Geometric mean concentrations of total 25(OH)D and their 95% confidence intervals (CIs) were calculated and compared by case group and the potential confounders. The categorical distribution of 25(OH)D levels was also examined by covariates among controls, and assessed by chi-square test.

Logistic regression was used to calculate crude and multivariable adjusted odds ratios (AOR) and 95% CIs for the association between 25(OH)D and ASD in comparison to controls. The original frequency matching variables and those associated with ASD and/or 25(OH)D were included in models singly and together to determine the change in the effect estimates for final covariate selection. None altered associations greatly but several were retained in fully adjusted models due to strong associations with ASD or inclusion in prior papers: birth year, blood draw season, and child sex (as matching variables), as well as maternal age (five categories), education (four categories), race/ethnicity (three categories, as above) and parity (0, 1, ≥2).

We examined whether odds of ASD differed according to presence of co-occurring ID by modeling ASD with and without ID separately. In parallel analyses, we examined odds of ID (without ASD) relative to GP controls. We explored interactions of 25(OH)D with child sex, with maternal race/ethnicity, and parity. To maintain adequate sample size, results for interaction analyses are presented for continuous 25(OH)D, with dichotomous terms (e.g. male vs. female, NHW vs. all other groups, and nulliparous vs. multiparous), and P-values <0.10 considered supportive of interactions.

The EMA3 study sample excluded subjects who had been included in prior data collection phases (EMA 1–2). To account for the disproportionate impact of that exclusion on the geographic and temporal distribution of cases (vs. controls) in the EMA3 sampling frame, we further excluded any births later captured in EMA3 from the EMA 1–2 single county and time period (July 1, 2000 and September 30, 2001), in a sensitivity analysis (n = 44 ASD, 12 ID, and 63 controls). Because adjustment for birth month and year may over-control for temporal trends related to vitamin D, we re-ran our models with analytic weights accounting for frequency matching on birth month, birth year, and child sex, as well as with the sampling exclusion of EMA phase 1–2 subjects above.

Other sensitivity analyses included examining 25(OH) D₃ only, as 25(OH)D₂ is usually only obtained from supplements. Preterm delivery may be on the pathway to ASD, although not associated with 25(OH)D in our data, so we did not adjust for it in our models. However, to assess potential modification, we re-ran key analyses among term births (≥37 weeks) only. We also examined potential non-linearity in the association between 25(OH)D and neurodevelopmental outcomes using restricted cubic splines (RCS) [Durrleman & Simon, 1989; Govindarajulu, Spiegelman, Thurston, Ganguli, & Eisen, 2007]. Models included three knots with specification at the 10th, 50th and 90th percentiles of the data in the specific analysis (number of knots chosen based on lowest AIC), adjusted as in primary models. Nonlinear associations were examined in strata defined by sex, race-ethnicity, and parity groups, and by ASD subtype according to co-occurring ID. AORs and CIs were plotted using a reference value at the median of the overall RCS model, and spline interactions were explored by sex and race-ethnicity (as none of the parity groups showed non-linear associations). To determine if outliers were affecting non-linear associations, we re-ran these models in a sample excluding the lower and upper 1% distribution of case- and control-specific 25(OH)D levels.

Results

Compared to mothers of control children, mothers of children with ASD were more likely to be highly educated, of lower parity, and somewhat more likely to be Asian and older (Table 1). Mothers of children with ID were more likely to be Hispanic, younger, less educated, insured by government programs, heavier at screening, deliver preterm, and to have blood drawn in the summer, while the children were more likely to be female (due to matching protocol).

Table 1.

Demographic Characteristics by Case Status in the EMA Study, Maternal Vitamin D Analysis

Characteristic	ASDa (N = 534)		IDa (N = 181)		GPa controls (N = 421)		ASD vs. GP	ID vs. GP
	N	Percent (%)	N	Percent (%)	N	Percent (%)	P-valueb	P-valuec
Birth year
2000	96	18.0	41	22.7	76	18.1	0.99	0.33
2001	141	26.4	51	28.2	110	26.1
2002	212	39.7	69	38.1	170	40.4
2003	85	15.9	20	11.0	65	15.4
Blood draw season
Winter (Dec–Feb)	155	29.0	45	24.9	117	27.8	0.88	0.03
Spring (Mar–May)	131	24.5	35	19.3	106	25.2
Summer (Jun–Aug)	112	21.0	55	30.4	83	19.7
Fall (Sep–Nov)	136	25.5	46	25.4	115	27.3
Child gender
Male	436	81.6	100	55.2	349	82.9	0.62	<0.01
Female	98	18.4	81	44.8	72	17.1
Maternal race/ethnicity
non-Hispanic white	180	33.7	34	18.8	137	32.5	0.14	<0.01
White Hispanic	218	40.8	124	68.5	196	46.6
Black, Asian, Other, unknown	136	25.5	23	21.7	88	20.9
Maternal age (years)
< 20	17	3.2	25	13.8	23	5.5	0.13	<0.01
20–24	80	15.0	42	23.2	71	16.9
25–29	150	28.1	49	27.1	128	30.4
30–34	194	36.3	43	23.8	144	34.2
≥ 35	93	17.4	22	12.2	55	13.1
Maternal education
<High school	96	18.3	74	41.3	102	24.5	<0.01	<0.01
High school grad	117	22.2	48	26.8	115	27.6
College	214	40.7	46	25.7	143	34.3
Postgraduate	99	18.8	11	6.1	57	13.7
Parity
0	245	45.9	62	34.3	162	38.5	<0.01	0.61
1	203	38.0	58	32.0	128	30.4
≥ 2	86	16.1	61	33.7	131	31.1
Preterm Delivery
Yes (<37 weeks)	76	14.2	43	23.8	54	12.8	0.53	<0.01
No (Term)	458	85.8	138	76.2	367	87.2
Maternal weight quartiles
Less than 127 lbs	137	25.7	31	17.1	106	25.2	0.97	0.02
127–144 lbs	135	25.3	37	20.4	112	26.6
145–167 lbs	129	24.2	57	31.5	102	24.2
167 lbs+	133	24.9	56	30.9	101	24.0
Delivery payment source
Insurances	279	52.2	49	27.1	207	49.2	0.59	<0.01
Government programs	236	44.2	125	69.1	196	46.6
Self and others	19	3.6	7	3.9	18	4.3

^aASD = autism spectrum disorders, ID = intellectual disability, GP = general population, EMA = Early Markers for Autism.

^bP-value for the Chi-square test of Association for ASD vs. GP controls; first 3 were matching variables.

^cP-value for the Chi-square test of Association for ID vs. GP controls.

Total 25(OH)D was considered deficient in 9.6% and insufficient in 25.6% of mothers. Levels were similar between mothers of ASD cases and population controls, examined as the median (86.7 vs. 85.6 nmol/L, respectively) or categorized (Table 2). Among the controls (Table S1), total 25(OH)D showed the most variation by maternal race/ethnicity, with NHW mothers least likely to be deficient (3.7%), as expected. Vitamin D deficiency was present much more often in serum collected in winter months (18.8%). There was slight variation by birth year, but not in a monotonic pattern, and little variation by other characteristics, including maternal weight or preterm delivery. Geometric mean levels generally followed similar patterns with NHW mothers having a much higher geometric mean level (103 nmol/l) than mothers who were Hispanic (78 nmol/l) or Black, Asian, and other races (72 nmol/l); in addition, levels were somewhat higher in mothers carrying male vs. female infants and by later birth year; the later differences likely reflect truncation of months included for the first and last years of the study.

Table 2.

Maternal Vitamin D Levels by Case Status, EMA^a Study

Total 25(OH)Da variable (nmol/L)	ASD casesa (n = 534)		IDa (n = 181)		Controls (n = 421)
Continuous:
GMa (95% CIa)	84.3	81.7–87.0	82.5	77.5–87.8	83.8	80.4–87.3
Median (IQRa)	86.7	39.9	84.4	46.2	85.6	46.7
Categorizedb:	N	Percent (%)	N	Percent (%)	N	Percent (%)
Deficient (<50)	44	8.2	23	12.7	42	10.0
Insufficient (50 to <75)	136	25.5	41	22.7	114	27.1
Sufficient (≥75)	354	66.3	117	64.6	265	62.9

^aEMA = Early Markers for Autism, ASD = autism spectrum disorders, ID = Intellectual disability, 25(OH)D = 25hydroxyvitamin D, GM = geometric mean, CI = confidence interval, IQR = interquartile range.

^bChi-square P-values = 0.67 for ASD and 0.35 for ID.

Little association was seen between ASD and maternal vitamin D level; either categorically, with AORs (95% CI) of 0.79 (0.49–1.3) for deficiency and 0.93 (0.68–1.3) for insufficiency, or continuously (0.95; 0.86–1.05, per 25 nmol/L) (Table 3). Results did not vary greatly for ASD with or without ID. Nor was there much association for ID only, with AORs (95% CI) of 1.15 (0.60–2.19) for deficiency, 0.75 (0.47–1.12) for insufficiency, and 1.01 (0.88–1.16) continuously.

Table 3.

Risk of Autism Spectrum Disorder (ASD) or Intellectual Disability (ID) Compared to Controls, by Maternal Vitamin D

	Deficient 25OHD (<50 nmol/L)a		Insufficient 25OHD (50- < 75 nmol/L)a		Total 25OHD (per 25 nmol/L)
Model	ORa	95% CI	ORa	95% CI	OR	95% CI
ASD (N = 534)
Unadjusted	0.78	(0.50, 1.23)	0.89	(0.67, 1.20)	0.97	(0.88, 1.06)
Adjustedb	0.79	(0.49, 1.30)	0.93	(0.68, 1.28)	0.95	(0.86, 1.05)
ASD w/ ID (N = 283)
Model
Unadjusted	0.93	(0.55, 1.57)	1.05	(0.75, 1.48)	0.92	(0.83, 1.03)
Adjustedb	0.95	(0.54, 1.67)	1.10	(0.76, 1.58)	0.92	(0.82, 1.04)
ASD w/o ID (N = 216)
Unadjusted	0.63	(0.34, 1.18)	0.79	(0.54, 1.16)	1.01	(0.90, 1.13)
Adjustedb	0.67	(0.34, 1.33)	0.82	(0.54, 1.25)	0.98	(0.86, 1.11)
ID only (N = 181)
Unadjusted	1.24	(0.71, 2.16)	0.82	(0.54, 1.24)	0.97	(0.86, 1.09)
Adjustedb	1.15	(0.60, 2.19)	0.75	(0.47, 1.12)	1.01	(0.88, 1.16)

^aCompared to sufficient (25(OH)D ≥ 75 nmol/L).

^bAdjusted for birth year (2000 = ref, 2001, 2002, 2003), blood draw season (Winter (Dec–Feb) = ref, Spring (Mar–May), Summer (Jun–Aug), Fall (Sep–Nov)), child sex (male = ref, female), maternal age (<20, 20–24, 25–29 = ref, 30–34, ≥35), maternal education (HS or less, HS = ref, some college, post-graduate), maternal race/ethnicity (Non-Hispanic White = ref; White Hispanic; Black, Asian, Other, and Unknown), and parity (0 = ref,1, ≥2).

There was an interaction of the association of maternal vitamin D and ASD by race/ethnicity (p_interaction = 0.02) (Table 4a); children of NHW mothers showed the hypothesized protective effect for ASD as 25(OH)D level increased in the continuous model (AOR 0.82, 95% CI 0.70–0.97, per 25 nmol/L), while the “other” race group showed little association (AOR 1.05, 95% CI 0.92–1.2). The interaction was present in the ASD without ID group (P-interaction <0.01) with AORs (95% CI) of 0.81 (0.66–0.98) among NHWs and 1.2 (0.97–1.37) among all others. While similar patterns were seen in the ASD with ID group the race/ethnicity difference was less (p_interaction = 0.22), and was not observed for ID only.

Table 4a.

Interactions for Continuous Vitamin D (per 25 nmol/L) and Autism Spectrum Disorder (ASD) or Intellectual Disability (ID); with Maternal Race/ethnicity or Child Sex

By maternal race/ethnicitya
	non-Hispanic White		Other race/ethnicityb
Model	OR	95% CI	OR	95% CI	Interaction P-Value
ASD
Unadjusted	0.82	(0.70, 0.95)	1.06	(0.94, 1.20)	0.009
Adjustedc	0.82	(0.70, 0.97)	1.05	(0.92, 1.20)	0.019
ASD w/ ID
Unadjusted	0.83	(0.69, 1.02)	1.01	(0.88, 1.15)	0.128
Adjustedc	0.83	(0.67, 1.03)	0.98	(0.84, 1.13)	0.215
ASD w/o ID
Unadjusted	0.80	(0.66, 0.96)	1.13	(0.96, 1.32)	0.006
Adjustedc	0.81	(0.66, 0.98)	1.16	(0.97, 1.37)	0.006
ID Only
Unadjusted	0.94	(0.75, 1.19)	1.08	(0.93, 1.24)	0.339
Adjustedc	0.96	(0.75, 1.23)	1.05	(0.89, 1.23)	0.575

^aN for adjusted models: non-Hispanic white is 314 in ASD, 210 in ASD w/ ID, 230 in ASD w/o ID, 170 in ID only; for others is 629 in ASD, 484 in ASD w/ID, 402 in ASD w/o ID, 426 in ID only.

^bOther includes Hispanic whites, Blacks, Asians, others, or unknown.

^cAdjusted for birth year (2000 = ref, 2001, 2002, 2003), blood draw season (Winter (Dec–Feb) = ref, Spring (Mar–May), Summer (Jun–Aug), Fall (Sep–Nov)), maternal age (<20, 20–24, 25–29 = ref, 30–34, ≥35), maternal education (HS or less, HS = ref, some college, postgraduate), and parity (0 = ref,1, ≥2); as well as maternal race/ethnicity (Non-Hispanic White = ref; White Hispanic; Black, Asian, Other, Unknown) or child sex (male = ref, female) in 4a and 4b, respectively.

There was also an interaction by child sex (p_interaction < 0.01) (Table 4b), with a protective association of vitamin D and ASD among males (AOR 0.89, 95% CI 0.80–0.99 per 25 nmol/L 25(OH)D), but higher odds of ASD among females (AOR 1.40; 95% CI 1.06–1.85), though this latter estimate was based on smaller numbers (n = 166 females vs. 777 males). The sex interaction was observed for ASD by ID, although not for ID only. No significant interactions were observed by parity (data not shown). In general, odds ratios and interactions changed little in sensitivity analyses; i.e. when examined in term births only, after applying analytic weights or excluding the EMA Phase 1–2 sampling frame (data not shown).

Table 4b.

Interactions for Continuous Vitamin D (per 25 nmol/L) and Autism Spectrum Disorder (ASD) or Intellectual Disability (ID); with Maternal Race/ethnicity or Child Sex

By child sexa
	Males		Females
Model	OR	95% CI	OR	95% CI	Interaction P-Value
ASD
Unadjusted	0.92	(0.83, 1.01)	1.36	(1.04, 1.77)	0.006
Adjustedb	0.89	(0.80, 0.99)	1.40	(1.06, 1.85)	0.003
ASD w/ID
Unadjusted	0.86	(0.77, 0.97)	1.47	(1.06, 2.03)	0.003
Adjustedb	0.86	(0.75, 0.98)	1.51	(1.08, 2.10)	0.002
ASD w/o ID
Unadjusted	0.97	(0.86, 1.10)	1.26	(0.95, 1.67)	0.099
Adjustedb	0.93	(0.80, 1.07)	1.24	(0.91, 1.70)	0.089
ID Only
Unadjusted	0.92	(0.79, 1.07)	1.18	(0.94, 1.46)	0.067
Adjustedb	0.94	(0.79, 1.11)	1.17	(0.92, 1.49)	0.128

^aN for adjusted models: males is 777 in ASD, 571 in ASD w/ ID, 522 in ASD w/o ID, 444 in ID only; for females is 166 in ASD, 123 in ASD w/ ID, 110 in ASD w/o ID, 152 in ID only.

^bAdjusted for birth year (2000 = ref, 2001, 2002, 2003), blood draw season (Winter (Dec–Feb) = ref, Spring (Mar–May), Summer (Jun-Aug), Fall (Sep–Nov)), maternal age (<20, 20–24, 25–29 = ref, 30–34, ≥35), maternal education (HS or less, HS = ref, some college, postgraduate), and parity (0 = ref,1, ≥2); as well as maternal race/ethnicity (Non-Hispanic White = ref; White Hispanic; Black, Asian, Other, Unknown) or child sex (male = ref, female) in 4a and 4b, respectively.

When exploring potential non-linear relationships using cubic splines, there was a significant, non-linear association of ASD and 25(OH)D (P-value = 0.02 and for test of non-linearity P = 0.009, Table S2), appearing as an inverted j-shape with the peak around 90–100 nmol/L 25(OH)D (Fig. 1). Similar patterns were observed for ASD with or without ID, but only the latter was significantly non-linear (p_nonlinear = 0.02). Stratified by race/ethnicity (Fig. 2A), the “Other” group showed a suggestive non-linear pattern (p_nonlinear = 0.08) although flatter than overall (P = 0.15), while the NHWs still showed a “protective” association, although somewhat attenuated (P = 0.07), consistent with linearity (p_nonlinear = 0.49). A test for interaction was not statistically significant (p_interaction = 0.30). Stratified by child sex (Fig. 2B), there were associations of ASD and 25(OH)D among both groups, but in opposite directions. The non-linear, inverted j-shape was apparent for males (p_nonlinear = 0.013), but not for females (p_nonlinear = 0.25), who still showed increasing odds of ASD with increasing 25(OH)D (P = 0.025). The interaction with sex was significant (p_interaction = 0.02). Non-linear patterns were not seen for ID only. The bulk of the 25(OH)D values lie between about 50 and 150 nmol/L, so some caution should be applied in interpretation of curves beyond these values. Excluding outliers did not alter patterns of association of ASD and vitamin D for any sub-groups or nonlinear findings (data not shown).

Figure 1.

Association of autism spectrum disorder with maternal mid-pregnancy Vitamin D Levels (continuous), using restricted cubic splines with three knots. Curved line represents adjusted odds ratios and band shows 95% Confidence Intervals. Adjusted as in primary models for birth year, blood draw season, maternal age, maternal education, parity, maternal race/ethnicity, and child sex.

Figure 2.

Association of autism spectrum disorder with maternal mid-pregnancy Vitamin D levels (continuous), among sub-groups using restricted cubic splines with three knots. Curved line represents adjusted odds ratios and band shows 95% confidence intervals. Adjusted as in primary models for birth year, blood draw season, maternal age, maternal education, and parity, as well as maternal race/ethnicity or child sex, as appropriate. (A) Stratified by maternal race/ethnicity; purple represents Hispanic, Black, Asian, and other/unknown, and blue line represents non-Hispanic Whites. (B) Stratified by Child Sex; green represents females, and beige/brown represents males.

Discussion

In one of the few studies to have both ASD diagnoses and vitamin D measured at the appropriate susceptibility window reflecting in utero development, our results did not indicate the hypothesized increased odds of ASD, or ID, with maternal vitamin D deficiency, nor a protective effect of increasing vitamin D levels, overall. However, a protective association for ASD, consistent with a linear effect, was observed with increasing maternal vitamin D among children of NHW mothers. Furthermore, some non-linear patterns were observed, indicating inverse (“protective”) associations with ASD at higher levels of vitamin D (>100 nmol/L) overall, as well as among males. The corresponding “protective” patterns at lower levels of vitamin D, while lower in magnitude, are difficult to explain. In contrast, there was a suggestion of increased risk of ASD with increasing maternal vitamin D levels in females. Our results appeared specific to ASD, as little association was found with ID alone, although based on much smaller numbers.

A possible association of vitamin D deficiency with ASD was proposed a decade ago [Cannell, 2008], with early studies based primarily on examining surrogates of vitamin D status or measured levels after ASD diagnosis. Nevertheless such an association is supported by evidence of vitamin D’s role in mediating immune function [Cantorna et al., 2004; Hayes et al., 2003; Hewison, 2012] and in brain development [Eyles et al., 2005; Eyles et al., 2013; Macova et al., 2017]. A recent animal study tested rat pups of different postnatal ages using a battery related to the ASD behavioral phenotype and found several differences in offspring of vitamin D deficient dams (Ali et al., 2019). Genetic polymorphisms may also play a role; a recent meta-analysis [Yang & Wu, 2020] reported that allelic differences in two vitamin D receptor genes involved in binding or transport of 25(OH)D were associated with ASD and another recent study [Yu, Zhang, Liu, Hu, & Liu, 2020] found polymorphisms in vitamin D metabolism-related enzymes associated with childhood ASD and its severity.

Additional epidemiologic studies of 25(OH)D measured during the relevant peri-natal period have also found associations, but not consistently. The recent meta-analysis on neurodevelopmental effects of prenatal vitamin D [Garcia-Serna & Morales, 2019] included five reports on “autism-related traits” (including autism) and calculated a pooled odds ratio of 0.42 (95% CI 0.25, 0.71) for highest vs. lowest prenatal vitamin D as defined in the respective studies. Not all of the studies had clinical ASD as an outcome, and data from two analyses of the same study population were both included [Vinkhuyzen et al., 2017; Vinkhuyzen et al., 2018], whereas at least one other published study was not [Wu et al., 2018]. Only two of the prior studies included measured maternal vitamin D and cases with ASD (although with <75 cases in each) and both showed increased risk of ASD with mid-pregnancy vitamin D levels either below the median [Chen, Xin, Wei, Zhang, & Xiao, 2016] or considered “deficient” (defined as <25 nmol/L) [Vinkhuyzen et al., 2018]. Interestingly, there was no association with 25(OH)D examined continuously in the latter study. This Generation R study from the Netherlands also examined associations with a subset of items from the Social Responsiveness Scale (SRS) scores in their large cohort and reported higher scores (indicative of more autistic traits/greater social deficits) in children whose mothers had deficient vitamin D mid-pregnancy [Vinkhuyzen et al., 2017]. New studies not included in the meta-analysis included one with first trimester vitamin D levels measured in mothers who delivered in Sweden during 1996–2000 and found no association with categorical or continuous levels and ASD overall [Lee et al., 2019]. However, among the sample with Nordicborn mothers, a protective association was seen, consistent with a linear pattern as vitamin D level increased, similar to our finding for white mothers. A study from Spain measured plasma D₃ (vs. total vitamin D) in pregnancy and found no association with ASD symptoms (measured by the CAST, or Childhood Autism Spectrum Test) at age 5, after adjustment [López-Vicente et al., 2019].

We previously examined newborn vitamin D levels (from dried blood spots) in the same mother-infant pairs as this study [Windham et al., 2019], representing a slightly later point in the developmental period. Similar to current results, we saw little association between vitamin D and ASD overall, but also some interactions with sex and race/ethnicity, although less strongly than for maternal vitamin D. We did not observe non-linear patterns with newborn vitamin D. A few other studies have examined ASD in relation to newborn vitamin D and further to allow for across-study comparison several investigators made an attempt to align some statistical analyses (categorical as well as continuous, at the same increment, vitamin D levels; ethnic sub-grouping; non-linear models) [Lee et al., 2019; Schmidt et al., 2019; Windham et al., 2019]. The other recent study in California with neonatal blood spots, but children examined in-person by study staff, also did not report an association with ASD [Schmidt et al., 2019]. The Swedish study reported higher risk of ASD in their larger sample with neonatal measurements considered deficient (<25 nmol/L) [Lee et al., 2019]. The Netherlands cohort did not report an association of ASD with deficient vitamin D in cord blood, but did find higher scores on the SRS subset examined [Vinkhuyzen et al., 2017; Vinkhuyzen et al., 2018]. Two other studies reported associations with lower vitamin D levels measured in newborn blood spots and increased risk of ASD [Fernell et al., 2015; Wu et al., 2018], although the former was rather small and did not adjust for many factors.

Prior studies that reported lower vitamin D levels in children already diagnosed with ASD than in comparison children tended to be limited in sample size and may be reflecting lifestyle differences, or reverse causation [Fernell et al., 2010; Kocovska et al., 2014; Meguid et al., 2010; Mostafa & Al-Ayadhi, 2012].

The authors of the recent meta-analysis [Garcia-Serna & Morales, 2019] also calculated summary protective associations (pooled OR = 0.72, 95% CI 0.59, 0.89) for ADHD. The scores for global IQ/cognitive development or psychomotor development showed weakly positive, or no association, respectively, while studies on language development and behavioral difficulties were even more inconsistent or null, so they were not pooled. Several studies with various assessments for developmental delays conducted at ages 8 or older did not find lasting effects of prenatal vitamin D level [Darling et al., 2017; Gale et al., 2008; López-Vicente et al., 2019; Strom et al., 2014]. In our prior study, we reported some increased risk for ID with increasing newborn vitamin D (opposite the hypothesized direction), which was not seen here with maternal vitamin D. The other recent California study did not see an association of newborn vitamin D and a broader range of developmental delays, nor with the Mullen Scales of Early Learning composite or Vineland Adaptive Behavior Scales scores, in adjusted models [Schmidt et al., 2019]. Results may vary across studies for a number of reasons, including the outcome examined, the range of vitamin D levels, gestational timing, and adequate control for possible confounders. The authors of the meta-analysis suggested that studies with vitamin D levels measured in early to mid-pregnancy compared to later were more likely to show associations with ASD; this was somewhat born out in the subsequently published Swedish study, but not entirely.

Our results indicated race/ethnic differences, with the hypothesized protective associations of higher 25(OH)D levels on ASD observed only among NHWs, consistent with a linear effect and with our prior study of newborn 25(OH)D levels. Many of the studies of perinatal 25(OH) D were conducted in less racially diverse populations than California, primarily Caucasian [Strom et al., 2014; Vinkhuyzen et al., 2017; Vinkhuyzen et al., 2018; Whitehouse et al., 2013], although some were conducted in Asian countries or included darker-skinned immigrants [Chen et al., 2016; Wu et al., 2018]. Notably, the recent Swedish study, with lower mean vitamin D levels, also found stronger (or only) associations in the Nordic-born mother subset, for both maternal and neonatal vitamin D levels [Lee et al., 2019]. The other California study also found a protective effect only among NHW children, but for developmental delay, not ASD [Schmidt et al., 2019]. Typically, non-white race/ethnic groups have lower vitamin D levels, in part due to more melanin in the skin reducing sunlight-induced production of the pre-vitamin D precursor. Therefore, data may be sparser at higher levels for detecting a significant association. We did not have sufficient numbers of Black mothers (N = 37) to examine separately, but the white, Hispanic mothers had vitamin D levels much closer to Blacks and Asians than to NHWs, so this grouping seemed most appropriate. This “other” race group includes mothers with the lowest levels of the vitamin D distribution in our study, which appears to be the source of the inverted, J-shaped, pattern of association. Perhaps this is explained by genotypic differences in vitamin D regulation during pregnancy, or by some other health-protective behavior or effect, but it is difficult to hypothesize what. Additional studies are warranted to pursue these differences in finer detail and potentially target interventions.

Our results also indicated an interaction of 25(OH)D with child sex, with lower risks of ASD at higher 25(OH) D levels seen only in males, and associations in the opposite direction in females. These findings are again consistent with our newborn 25(OH)D results. In contrast, the other California study of newborn vitamin D reported inverse associations among females and no association in males [Schmidt et al., 2019], and this was not examined in other studies to our knowledge. NHANES data from the U.S. [Centers for Disease Control and Prevention, 2012] show slightly lower vitamin D levels in females than males, as did our data from neonates. The “active” form of vitamin D (1,25 di-hydroxy vitamin D), is a neurosteroid with transcriptional control over numerous genes, similar to other neurosteroids such as estrogen and testosterone. It has been reported that estrogen may protect against vitamin D deficiency and that immunemodulatory effects of vitamin D vary by sex, leading to the suggestion that males may be more susceptible to effects of vitamin D deficiency (A. Ali, Cui, Alexander, & Eyles, 2018; Cannell, 2017; Correale, Ysrraelit, & Gaitan, 2010). An Australian study [Wilson et al., 2018] reported an interaction of fetal sex and gestational 25(OH)D on gestational diabetes mellitus, with female “pregnancies” showing an association in the unexpected direction. This is similar to our findings by sex and the protective association of higher vitamin D in males only may suggest directions for future mechanistic research. Studies of other environmental risk factors and ASD also suggest sex-specific effects [Kern et al., 2017; Lyall, Croen, Daniels, et al., 2017a; Roberts et al., 2013; Schaafsma & Pfaff, 2014]. Caution in interpretation of differences is warranted as small numbers of affected females often limit the ability to study sex differences in ASD etiology and potential diagnostic bias by sex has been observed [Giarelli et al., 2010], so further research in larger samples is needed.

Our study sample has one of the largest number of documented ASD cases and measured maternal levels in any investigation of vitamin D. Additional strengths of our study include consideration of another neurodevelopmental endpoint (ID), and vitamin D measured in an appropriate susceptibility window during in utero development, and use of an ethnically diverse population allowing for examination of racial/ethnic differences. We also had the ability to adjust for a number of possible confounders and explore sex interactions, as well as non-linear effects. Potential limitations include having a smaller proportion of the sample considered vitamin D deficient compared to some other studies, perhaps limiting our ability to detect an effect, especially if driven by severe deficiency. However, studies such as ours with similarly high 25(OH)D levels, may help discern the bounds of possible beneficial effects. There may also be residual confounding from factors that we did not have the ability to adjust for, such as, pre-pregnancy BMI, which was included in some recent studies, or other dietary factors and medications, which are not generally accounted for in other studies either. Furthermore, results are based on one measure of vitamin D, which may change during pregnancy, especially if the gravida begins supplementing or due to other lifestyle changes. We did have newborn levels as well, reflecting late pregnancy; correlations with mid-gestation were moderate (0.37, P < 0.001), but samples were collected 5–6 months apart, reflecting different seasons. Results were generally consistent for newborn as for maternal vitamin D overall, although non-linear effects were not seen and the NHW association was not as strong [Windham et al., 2019].

In conclusion, given continued reports of widespread vitamin D deficiency and insufficiency and potential benefits of vitamin D supplementation, it is important to determine corresponding health effects, especially for pregnant women. Our findings on vitamin D measured in mid-gestation did not support a general increased risk of ASD at deficient levels, which might have contributed to an explanation for the continued rise in ASD prevalence. However, a protective effect on ASD with increasing vitamin D was seen among NHWs, as well as at higher vitamin D levels (> ~ 100 nmol/L) in males, who typically comprise 75–80% of ASD cases in a population. More studies on ASD are needed that measure vitamin D at multiple time points to identify specific susceptibility windows, and that are large enough to allow examination of sub-groups of ASD, including by sex, race/ethnicity and phenotype, as well as pooling to examine genetic polymorphisms.

Supplementary Material

supinfo

Table S1. Vitamin D Distribution by Maternal Characteristics among General Population Controls in the EMA Study, Geometric means (GM) and categorized

Table S2. Summary of analyses of non-linear effects using Restricted Cubic Splines*; p-values for Maternal Vitamin D and Autism