Introduction
Neurodevelopmental conditions include attention-deficit hyperactivity disorder (ADHD) and autism. They are characterized by early-onset delays and differences in developmental domains such as language and motor skills, social communication, flexibility, attention, and activity regulation (Thapar, Cooper, & Rutter, Reference Thapar, Cooper and Rutter2017). Neurodevelopmental traits are normally distributed throughout the population, and are etiologically complex, involving a large number of genetic and environmental predisposing factors (Thapar et al., Reference Thapar, Cooper and Rutter2017). One proposed modifiable environmental predisposing factor is maternal nutritional deficiencies during pregnancy (Georgieff, Ramel, & Cusick, Reference Georgieff, Ramel and Cusick2018). Many prenatal micronutrient deficiencies have been associated, including vitamin-D levels and omega-3 fatty acids (specifically Docosahexaenoic acid, DHA) (García-Serna & Morales, Reference García-Serna and Morales2019; Nevins et al., Reference Nevins, Donovan, Snetselaar, Dewey, Novotny, Stang and Raghavan2021). Nutrient availability during the prenatal period plays an important role in brain development. For example, vitamin-D receptors are present across several regions of the neonatal brain (Eyles, Smith, Kinobe, Hewison, & McGrath, Reference Eyles, Smith, Kinobe, Hewison and McGrath2005) and play a role in crucial processes of brain development including neuronal differentiation and axonal connectivity (Eyles, Burne, & McGrath, Reference Eyles, Burne and McGrath2013). DHA (along with other long-chained polyunsaturated fatty acids) is necessary for processes of brain development including neurogenesis, synaptogenesis, and neuronal migration (Innis, Reference Innis2008). Therefore, maternal deficiencies for both vitamin-D and DHA have been proposed to increase likelihood of neurodevelopmental conditions through changes to the structure or functions of the developing fetal brain.
However, results from observational studies are inconsistent and can be biased by residual environmental or genetic confounding and reverse causation (García-Serna & Morales, Reference García-Serna and Morales2019; Nevins et al., Reference Nevins, Donovan, Snetselaar, Dewey, Novotny, Stang and Raghavan2021). Thus, the causal role of these exposures on neurodevelopmental traits is unclear. Furthermore, it is difficult to use RCTs to investigate the effects of intrauterine exposures as it is often unethical or impractical to modify levels experimentally, and long-term follow-up is expensive and time-consuming. A systematic-review of RCTs for DHA supplementation during pregnancy showed very little evidence for an effect on offspring neurodevelopment (Nevins et al., Reference Nevins, Donovan, Snetselaar, Dewey, Novotny, Stang and Raghavan2021). For vitamin-D, only one such RCT exists, which reported that third-trimester supplementation did not associate with neurodevelopmental outcomes up to 6 years (Sass et al., Reference Sass, Vinding, Stokholm, Bjarnadóttir, Noergaard, Thorsen and Bisgaard2020). Overall, the current RCT evidence is low quality (consisting of under representative and small samples) (Nevins et al., Reference Nevins, Donovan, Snetselaar, Dewey, Novotny, Stang and Raghavan2021).
Mendelian randomization (MR) is a complementary approach to assess causality in observational studies (Carnegie et al., Reference Carnegie, Zheng, Sallis, Jones, Wade, Evans and Martin2020). MR examines whether exposure–outcome relationships are consistent with causality using genetic instruments as proxies for modifiable exposures (Sanderson et al., Reference Sanderson, Glymour, Holmes, Kang, Morrison, Munafò and Davey Smith2022). Instead of statistically randomizing individuals to receive the exposure (e.g. supplementation) or control, MR makes use of naturally occurring random variation in genetic variants, which influence levels of the exposure (e.g. higher vitamin-D or DHA levels) (Carnegie et al., Reference Carnegie, Zheng, Sallis, Jones, Wade, Evans and Martin2020). MR allows for unbiased estimation of causal effects provided that the following assumptions are satisfied: (1) the genetic variants are robustly associated with the exposure; (2) there is no confounding of the genetic instrument and the outcome; and (3) the genetic variants only influence the outcome through the exposure (Sanderson et al., Reference Sanderson, Glymour, Holmes, Kang, Morrison, Munafò and Davey Smith2022). Assumptions can be violated by horizontal pleiotropy, where genetic instruments for the exposure are associated with other traits that are not on the causal pathway from exposure to outcome. Primers on the MR method in general and in application to nutritional psychiatry are available elsewhere (Carnegie et al., Reference Carnegie, Zheng, Sallis, Jones, Wade, Evans and Martin2020; Davies, Holmes, & Smith, Reference Davies, Holmes and Smith2018; Sanderson et al., Reference Sanderson, Glymour, Holmes, Kang, Morrison, Munafò and Davey Smith2022). Two previous MR studies have explored the impact of vitamin-D levels on neurodevelopment, finding no strong evidence for a causal effect on autism (Revez et al., Reference Revez, Lin, Qiao, Xue, Holtz, Zhu and McGrath2020) nor ADHD (Libuda et al., Reference Libuda, Naaresh, Ludwig, Laabs, Antel, Föcker and Peters2021).
In the current study, we aim to test the possible causal effect from mothers on their offspring. Such a causal question requires a further extension to the MR design to account for shared genetic predisposition. Without such adjustment, MR estimates could reflect a direct intrauterine effect, but the association could alternatively be due to genetic variants directly transmitted from the mother to offspring that influence the child's own nutrient levels (Lawlor et al., Reference Lawlor, Richmond, Warrington, McMahon, Davey Smith, Bowden and Evans2017). Here, we extend the standard MR approach using mother–father–child family data in a trio-MR design. We control for mediation via offspring genotype to isolate the effect through the maternal genotype (as applied elsewhere: [Moen et al., Reference Moen, Brumpton, Willer, Åsvold, Birkeland, Wang and Evans2020; Taylor et al., Reference Taylor, Wootton, Yang, Oddie, Wright, Yang and Lawlor2023]). This approach is not common, given the need for large samples of at least mother-offspring pairs (Evans, Moen, Hwang, Lawlor, & Warrington, Reference Evans, Moen, Hwang, Lawlor and Warrington2019). Here it was possible thanks to the recent release of large-scale trio data from the Norwegian Mother, Father and Child Cohort Study (MoBa). Having access to father's genotype data provides two additional benefits: (1) adjusting for father's genotype prevents collider bias (Lawlor et al., Reference Lawlor, Richmond, Warrington, McMahon, Davey Smith, Bowden and Evans2017) and (2) fathers provide a negative control to explore whether the maternal effects are likely pregnancy-related or explained by early life environments provided by the parents. One MR study to date has explored maternal effects of vitamin-D on offspring autism accounting for child genetic liability which did not support the presence of a causal effect (Madley-Dowd et al., Reference Madley-Dowd, Dardani, Wootton, Dack, Palmer, Thurston and Rai2022). To the best of our knowledge, there are no other MR studies of vitamin-D nor DHA on neurodevelopmental outcomes using designs able to account for shared genetic liability.
If associations between maternal nutrient levels and offspring neurodevelopmental traits are not due to causal effects, an alternative explanation could be reverse causation. Evidence suggests that individuals with a genetic predisposition to neurodevelopmental conditions eat more restricted diets (Smith, Rogers, Blissett, & Ludlow, Reference Smith, Rogers, Blissett and Ludlow2020), which could in turn lead to lower nutrient levels. Mothers with a genetic predisposition to neurodevelopmental conditions might therefore have lower nutrient levels, and (independently of nutrient levels) pass on a genetic predisposition to neurodevelopmental conditions to their offspring. We estimate possible reverse causation using an MR design, this time using genetic predisposition to ADHD and autism diagnoses as the exposure, and vitamin-D and DHA levels as the outcome.
In this study we used maternal genetically predicted vitamin-D and DHA levels to estimate the causal effects of maternal vitamin-D and DHA on offspring neurodevelopment, while accounting for both offspring and paternal genetics. Second, we used paternal genetically predicted vitamin-D and DHA levels as a negative control to explore whether maternal effects are likely pregnancy-related. Third, we used MR to test for possible reverse causation, estimating causal effects of genetic predisposition to ADHD and autism on vitamin-D and DHA levels.
Methods
Sample
MoBa is a population-based pregnancy cohort study conducted by the Norwegian Institute of Public Health (Magnus et al., Reference Magnus, Birke, Vejrup, Haugan, Alsaker, Daltveit and Stoltenberg2016). Participants were recruited from across Norway from 1999 to 2008. The women consented to participate in 41% of pregnancies. The cohort includes 114 500 children, 95 200 mothers, and 75 200 fathers. The current study is based on version 12 of the quality-assured data released for research in January 2019. Biological material, including DNA samples were collected from both parents during pregnancy and from children (umbilical cord) at birth (Paltiel et al., Reference Paltiel, Anita, Skjerden, Harbak, Bækken, Kristin and Magnus2014).
Details of the genotyping and QC procedures in MoBa are available elsewhere (Corfield et al., Reference Corfield, Frei, Shadrin, Rahman, Lin, Athanasiu and Havdahl2022). After restricting to unrelated individuals between trios (proportion of the genome shared identity-by-descent < 0.15) of European ancestry with genotype data passing QC and phenotype data available, we had 23 713 mothers, 17 990 fathers, 26 646 children, and 16 298 trios (N genotyped individuals/trios for each outcome in online Supplementary Table S1).
Exposures
Vitamin-D
A GWAS of 401 460 individuals (Manousaki et al., Reference Manousaki, Mitchell, Dudding, Haworth, Harroud, Forgetta and Richards2020) identified 38 genome-wide significant single nucleotide polymorphisms (SNPs) that explained 3.1% variance in an independent sample. These variants have been validated for use in pregnancy, in the UK Avon Longitudinal Study of Parents and Children (ALSPAC) where they explained 1.05% of the variance (Madley-Dowd et al., Reference Madley-Dowd, Dardani, Wootton, Dack, Palmer, Thurston and Rai2022). Another GWAS of Vitamin-D (Revez et al., Reference Revez, Lin, Qiao, Xue, Holtz, Zhu and McGrath2020) published at the same time was used as a sensitivity analysis (online Supplementary Note S1). Given possible pleiotropy (Fang, Zhao, Yang, Zhang, & Giovannucci, Reference Fang, Zhao, Yang, Zhang and Giovannucci2024) we also conducted a sensitivity analysis using SNPs with known functional effects on vitamin-D levels (online Supplementary Note S4).
DHA
We used the largest available GWAS of DHA which identified 61 genome-wide significant SNPs in a sample of 114 999 individuals (Borges et al., Reference Borges, Haycock, Zheng, Hemani, Holmes, Davey Smith and Lawlor2022). We estimated that these SNPs explain 6.6% of the variance in the discovery sample. To validate these SNPs for use during pregnancy, we tested their prediction in the ALSPAC cohort (online Supplementary Note S2), where they explained 0.85% of variance in DHA levels during pregnancy. We also conducted a sensitivity analysis using SNPs with known functional effects on DHA levels (online Supplementary Note S4).
Outcomes in MoBa
We focused on clinically relevant neurodevelopmental traits, reported by the mothers at child ages 3, 5, and 8 years (further details in online Supplementary Note S3):
ADHD-related traits of inattention and hyperactivity/impulsiveness were assessed using two scales: a reduced Child-Behavior Checklist at 3 and 5 years (Achenbach, Reference Achenbach1992), and the Disruptive Behaviour Disorder Rating Scale (RS-DBD) (Silva et al., Reference Silva, Alpert, Pouget, Silva, Trosper, Reyes and Dummit2005) at 8 years.
Autism-related traits of social communication differences and restricted and repetitive behaviors and interests (RRB) were assessed at 3 and 8 years using the Social Communication Questionnaire (Rutter, Bailey, & Lord, Reference Rutter, Bailey and Lord2003).
Language delay was measured using the Ages and Stages Questionnaire (ASQ) (Squires, Potter, & Bricker, Reference Squires, Potter and Bricker1999) language items at ages 3 (6-items) and 5 years (7-items) years.
Motor delay was measured by the ASQ at age 3 (4-items) and by the Child Development Inventory at age 5 years (12-items).
Statistical analysis
We conducted two different methods to strengthen causal inference: two-sample MR and a trio-MR analysis both using genome-wide significant SNPs (p < 5 × 10−8). Online Supplementary Fig. S11 provides an overview of the analyses conducted. We performed a Bonferroni adjustment for the number of tests conducted in the primary analyses (p = 0.05/(19 × 2) = 0.001).
Analysis 1 – Two-sample MR estimates of nutrient levels on offspring neurodevelopment
In MR analysis, we use genetic variants as instruments to estimate a causal effect of an exposure on an outcome. In two-sample MR, estimates of the SNP-exposure and SNP-outcome effects come from summary statistics of two independent GWAS (Hartwig, Davey Smith, & Bowden, Reference Hartwig, Davey Smith and Bowden2017). As outlined in the introduction, three core assumptions must hold for valid causal inference (Davies et al., Reference Davies, Holmes and Smith2018). To satisfy assumption 1, we selected only independent genome-wide significant SNPs, where predictive validity during pregnancy had been established (online Supplementary Note S2). SNPs passing QC were clumped for independence (r 2 < 0.001, 10 000 kilobases). Measures of nutrient levels were not available for pregnant mothers in the MoBa cohort, so genetic variants were validated as a robust predictor of nutrient levels during pregnancy in the ALSPAC cohort (see online Supplementary Note S2 and [Madley-Dowd et al., Reference Madley-Dowd, Dardani, Wootton, Dack, Palmer, Thurston and Rai2022]). Furthermore, we conducted a sensitivity analysis using only SNPs for vitamin-D and DHA where biological function is understood (online Supplementary Note S4).
SNP-outcome effects using maternal genotype data were estimated in MoBa with linear regression (adjusted for 10 principal components (PCs) to account for population structure and genotype batch). Our primary method for estimating causal effects was the inverse–variance weighted (IVW) estimator. To explore the validity of assumptions 2 and 3, we conducted three additional sensitivity methods with different assumptions about the nature of possible pleiotropy: weighted median (Bowden, Davey Smith, Haycock, & Burgess, Reference Bowden, Davey Smith, Haycock and Burgess2016a), MR Egger (Bowden, Davey Smith, & Burgess, Reference Bowden, Davey Smith and Burgess2015), and weighted mode (Hartwig et al., Reference Hartwig, Davey Smith and Bowden2017) (see online Supplementary Note S5 for further details). The MR Egger intercept provides an estimate of bias from directional horizontal pleiotropy (Bowden et al., Reference Bowden, Davey Smith and Burgess2015). Heterogeneity of genetic instruments was assessed using Cochrans Q, and instrument strength was checked using the F statistic (F > 10 indicates low risk of weak instrument bias). Simulation-extrapolation MR was performed if violations of the no measurement error assumption were detected from I 2GX (Bowden et al., Reference Bowden, Del Greco, Minelli, Davey Smith, Sheehan and Thompson2016b). Where there was evidence for a causal effect, we conducted leave-one-out-analysis to identify potential outliers and Steiger filtering to test for possible reverse causation (Hemani, Tilling, & Davey Smith, Reference Hemani, Tilling and Davey Smith2017) (online Supplementary Fig. S11, Analysis 1.1).
We conducted a replication of our MoBa MR analysis using GWAS summary statistics from the Psychiatric Genomics Consortium (PGC) for autism (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) and ADHD (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019) diagnoses. The GWAS of autism comprised 18 381 cases and 27 969 controls (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019). The GWAS of ADHD comprised 20 183 cases and 35 191 controls (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019) (online Supplementary Fig. S11, Analysis 1.2).
We conducted a sensitivity analysis using child genotype for SNP-outcome effects to test for effects of child's own nutrient levels on their neurodevelopmental traits (online Supplementary Fig. S11, Analysis 1.3). Our power calculation demonstrated that the genetic instruments provided sufficient power for MR analyses (online Supplementary Note S6).
Analysis 2 – Trio-MR analysis in MoBa
We used independent (r 2 = 0.25, 500 kb) genome-wide significant SNPs from the respective GWAS summary statistics to construct PGS. PGS were computed as the weighted sum of effect alleles for the respective exposure for MoBa mothers, fathers and children using PRSice (Euesden, Lewis, & O'Reilly, Reference Euesden, Lewis and O'Reilly2015).
First, we looked at the effect of maternal PGS on neurodevelopmental outcomes, with adjustment for offspring and paternal genotype to determine if the observed association is due to maternal causal effects, genetic confounding or bias. These analyses were adjusted for child sex, population structure (using the first 10 PCs), and genotype batch. Adjusting for the child PGS alone risks opening up a back-door path through paternal genotype, creating an additional source of bias. Therefore, we also adjusted for paternal genotype (online Supplementary Fig. S11, Analysis 2.1). Due to different rates of diagnoses in males and females, we also conducted analyses stratified by child sex.
Additionally, we explored possible effects of child vitamin-D and DHA levels on child neurodevelopmental traits. Here, our model first looked at the association of child PGS for vitamin-D and DHA only, then additionally adjusted for maternal and paternal PGS (online Supplementary Fig. S11, Analysis 2.2).
Finally, we conducted a paternal negative control analysis. If we hypothesize that causal effects are maternal specific (e.g. through the intrauterine environment), rather than through other environmental effects (such as nutrition provided to the child in early life), then we would expect to see an effect of maternal PGS, but not paternal PGS. We therefore repeated analyses using the paternal PGS only and subsequently controlled for the maternal and child PGS (online Supplementary Fig. S11, Analysis 2.3). We performed a sensitivity analysis to check for assortative mating on vitamin-D and DHA levels in the MoBa cohort by looking at the correlation between maternal and paternal PGS.
Analysis 3 – Two-sample MR testing for possible reverse causation of neurodevelopment on nutrient levels
SNP-exposure estimates were obtained from the PGC GWAS of autism (Grove et al., Reference Grove, Ripke, Als, Mattheisen, Walters, Won and Børglum2019) and ADHD diagnoses (Demontis et al., Reference Demontis, Walters, Martin, Mattheisen, Als, Agerbo and Neale2019). We used the same GWAS of vitamin-D and DHA, this time as the outcome (see online Supplementary Fig. S11, 3.1). Given the small number of genetic instruments available, we used a relaxed p value threshold of p < 5 × 10−6, resulting in 34 SNPs for autism and 63 SNPs for ADHD. In this direction, we report effect sizes as the average change in the outcome (nutrient levels) per doubling (2-fold increase) in the liability to the exposure (ADHD or autism) (Burgess & Labrecque, Reference Burgess and Labrecque2018; Howe, Tudball, Davey Smith, & Davies, Reference Howe, Tudball, Davey Smith and Davies2022).
Results
Testing for causal effects of DHA and vitamin-D on neurodevelopmental traits in MoBa
Two-sample MR analysis in the MoBa cohort provided little evidence for a causal effect of vitamin-D or DHA on any neurodevelopmental traits after adjustment for multiple testing (Fig. 1, online Supplementary Table S2). There was weak evidence for an effect of higher maternal vitamin-D levels on decreased ADHD hyperactivity traits at age 5 years (−0.015, 95% CI −0.024 to −0.006) and total ADHD-related traits at age 5 years (−0.018, 95% CI −0.030 to −0.006). These effects were consistent in direction, with similar effect sizes across MR sensitivity methods but did not survive correction for multiple testing (online Supplementary Table S2). SNPs had sufficient instrument strength (online Supplementary Table S3) and there was minimal evidence of regression dilution bias, suggesting suitability of the MR Egger method (online Supplementary Table S3). There was little evidence of heterogeneity (online Supplementary Table S4), and the MR Egger intercept did not suggest bias from horizontal directional pleiotropy (online Supplementary Table S5). There was similarly weak evidence for causal effects using canonical SNPs for vitamin-D and DHA (see online Supplementary Note S4 for results and discussion).
Figure 1. Two-sample MR results for vitamin-D and DHA on neurodevelopmental outcomes in the MoBa sample. All estimates are from the inverse-variance weighted method. Units can be interpreted as per s.d. increase in exposure on the scale of the outcome traits.
Two-sample Mendelian randomization analyses using child genotypes
There was very limited evidence for causal effects of genetic predisposition to DHA or vitamin-D on any of the neurodevelopmental traits (online Supplementary Table S6). There was some weak evidence for an effect of child vitamin-D levels on decreased ADHD hyperactivity traits at age 5 years (−0.014, 95% CI −0.022 to −0.005), total ADHD traits at age 5 years (−0.017, 95% CI −0.029 to −0.005), and restrictive-repetitive behaviors at age 3 years (−0.017, 95% CI −0.028 to −0.006). There was weak evidence for an effect of DHA levels on decreased ADHD hyperactivity traits at age 8 years (−0.310, 95% CI −0.553 to −0.066) and decreased restrictive and repetitive behaviors at age 8 years (−0.078, 95% CI −0.153 to −0.003). These effects had consistent direction and relatively similar effect sizes across MR sensitivity methods (online Supplementary Table S2) but did not survive correction for multiple testing.
Testing for causal effects of DHA and vitamin-D on autism and ADHD diagnoses
We observed little evidence for causal effects of either vitamin-D or DHA on autism or ADHD diagnoses (Fig. 2, online Supplementary Table S7). F statistics were greater than 10 suggesting no weak instrument bias (online Supplementary Table S8). There was evidence of heterogeneity (online Supplementary Table S9), and the MR Egger intercepts suggested possible bias from horizontal pleiotropy for DHA on ADHD and vitamin-D on autism (online Supplementary Table S10).
Figure 2. Two-sample MR results for vitamin-D and DHA on diagnoses of ADHD and autism. All units can be interpreted as per s.d. increase in exposure on the odds of outcome diagnosis.
Testing maternal effects using trio-MR
We did not find evidence to support maternal effects of vitamin-D or DHA on any offspring neurodevelopmental traits after adjusting for child and paternal PGS (Figs 3 and 4, online Supplementary Table S11). Prior to adjustment, there was evidence that maternal vitamin-D PGS was associated with offspring ADHD hyperactivity traits at age 5 years (−0.045, 95% CI −0.070 to −0.021, Fig. 3, online Supplementary Table S13) but this association was attenuated after controlling for child and father PGS (−0.016, 95% CI −0.051 to 0.019) (online Supplementary Table S11). Correlations between maternal and paternal PGS were weak (DHA: r = 0.006, 95% CI −0.005 to 0.016, p = 0.29; vitamin-D: r = −0.004, 95% CI −0.015 to 0.006, p = 0.42) suggesting results are unlikely biased by assortative mating. Results stratified by child sex did not show evidence for association (online Supplementary Figs S3 and S4, Tables S12 and S13).
Figure 3. Trio-Mendelian randomization analysis of maternal polygenic score for vitamin-D levels on neurodevelopmental outcomes in the MoBa sample, with and without adjustment for child and paternal polygenic scores. All units can be interpreted as per s.d. increase in polygenic score on the scale of the outcome traits.
Figure 4. Trio-Mendelian randomization analysis of maternal polygenic score for DHA levels on neurodevelopmental outcomes in the MoBa sample, with and without adjustment for child and paternal polygenic scores. All units can be interpreted as per s.d. increase in polygenic score on the scale of the outcome traits.
Testing effects of child's nutrient levels using trio-MR
There was little evidence for associations between child DHA PGS and any neurodevelopmental traits (online Supplementary Fig. S5, Table S14). Prior to adjustment, there was some evidence for an association between child PGS for higher vitamin-D levels and lower hyperactivity scores at age 5 years (−0.046, 95% CI −0.068 to −0.023) (online Supplementary Fig. S6, Table S14) but this was somewhat attenuated after adjustment for the parental polygenic scores (−0.035, 95% CI −0.075 to 0.005). There was strong evidence for an association between child PGS for higher vitamin-D levels and lower total ADHD traits at both age 3 (−0.049, 95% CI −0.093 to −0.005) and age 5 years (−0.054, 95% CI −0.085 to −0.023), the latter of which was only partially attenuated after adjustment for parental PGS (3 years: −0.068, 95% CI −0.148 to 0.011; 5 years: −0.058, 95% CI −0.112 to −0.003). Finally, there was strong evidence for an association between child PGS for higher vitamin-D levels and lower restrictive and repetitive behaviors at age 3 (−0.043, 95% CI −0.073 to −0.013) which again was attenuated by adjustment (−0.027, 95% CI −0.081 to 0.027).
Paternal negative control analysis
We did not find evidence for associations between paternal PGS and child neurodevelopmental outcomes before or after adjustment for maternal and child PGS (online Supplementary Figs S7 and S8, Table S15).
Testing for possible reverse causation using PGC diagnostic GWAS
There was consistent evidence for a causal effect of higher genetic predisposition to autism on lower vitamin-D levels across sensitivity methods (IVW: −0.011 unit increase in vitamin-D per doubling of autism odds, 95% CI −0.020 to −0.003; Fig. 5, online Supplementary Table S7). There was some evidence for an effect of higher genetic predisposition to ADHD on lower DHA levels (IVW: −0.028), (95% CI −0.043 to −0.013), which was consistent for weighted median, but not weighted mode (Fig. 5, online Supplementary Table S7). We did not conduct MR Egger due to substantial regression dilution bias with the exception of ADHD on vitamin-D (online Supplementary Table S8). F statistics suggested that instruments were not weak (online Supplementary Table S8). For autism on vitamin-D, all 34 SNPs explained more variance in the exposure than the outcome, suggesting reverse causation was unlikely. For ADHD on DHA, all but one of the 64 SNPs explained more variance in the exposure than the outcome, again suggesting reverse causation was unlikely. Iteratively removing genetic variants provided virtually the same results (online Supplementary Figs S1 and S2).
Figure 5. Two-sample MR results for genetic predisposition to ADHD and autism on vitamin-D levels and DHA. Units can be interpreted as the average change in standardized units of the outcome per doubling (2-fold increase) in the prevalence of the exposure.
Discussion
Across two different study designs (two-sample MR and trio-MR), the evidence did not support causal effects of maternal vitamin-D or DHA levels on offspring neurodevelopmental outcomes (both traits and diagnoses). Trio-MR analysis suggested that two-sample MR results were biased by shared genetic liability, as associations between maternal vitamin-D and offspring ADHD-related traits at age 5 were attenuated once child and father genetic predisposition had been accounted for. This suggests that associations were largely explained by genetic confounding. Similarly, triangulating across our different study designs and samples, it seems likely that observed associations do not reflect maternal causal effects, highlighting the benefits of using a genetically informed family-based cohort.
The current findings add to a growing body of evidence that suggests that if maternal effects of vitamin-D or DHA deficiencies exist for neurodevelopmental outcomes, that they are likely to be small in magnitude. Randomized control trials, MR, trio-MR, paternal negative control studies, and family studies all have different sources of bias, hence we can strengthen our causal inferences by comparing evidence across them (Munafò, Higgins, & Davey Smith, Reference Munafò, Higgins and Davey Smith2021). Randomized control trials have found limited evidence for intrauterine effects (Nevins et al., Reference Nevins, Donovan, Snetselaar, Dewey, Novotny, Stang and Raghavan2021; Sass et al., Reference Sass, Vinding, Stokholm, Bjarnadóttir, Noergaard, Thorsen and Bisgaard2020) and previous MR studies of vitamin-D did not find evidence for an effect (Libuda et al., Reference Libuda, Naaresh, Ludwig, Laabs, Antel, Föcker and Peters2021; Madley-Dowd et al., Reference Madley-Dowd, Dardani, Wootton, Dack, Palmer, Thurston and Rai2022; Revez et al., Reference Revez, Lin, Qiao, Xue, Holtz, Zhu and McGrath2020). This is the first study using a trio-MR analysis to explore possible maternal effects of vitamin-D and DHA on neurodevelopment, while accounting for shared genetics among family members. Consistent with previous findings, our results did not support a causal maternal effect after controlling for shared genetic predisposition. Triangulating evidence across these diverse study designs, with different sources of bias, we conclude that causal maternal effects of vitamin-D and DHA on offspring neurodevelopmental outcomes are unlikely to play a significant role. Consequently, supplementation during pregnancy is unlikely to influence predisposition to neurodevelopmental difficulties in the offspring.
Observational associations between maternal nutrient levels and offspring neurodevelopment could instead be due to the influence of child's own nutrient levels. Maternal and child nutrient levels are likely to be correlated through shared diet and shared genetic predisposition. Nutrient levels during early childhood could be more important for neurodevelopmental outcomes than levels during pregnancy. Previous studies have found associations between child nutrient levels and child neurodevelopmental outcomes (Gan, Galer, Ma, Chen, & Xiong, Reference Gan, Galer, Ma, Chen and Xiong2019; Li et al., Reference Li, Xu, Zhang, Zhang, Wu, Wang and Zhu2020; Ryan et al., Reference Ryan, Astwood, Gautier, Kuratko, Nelson and Salem2010). For example, a meta-analysis of RCT studies of child vitamin-D supplementation showed very small improvements in ADHD scores (Gan et al., Reference Gan, Galer, Ma, Chen and Xiong2019) and a review of vitamin-D supplementation studies in children with autism showed a small improvement in hyperactivity scores (Li et al., Reference Li, Xu, Zhang, Zhang, Wu, Wang and Zhu2020). A review of DHA supplementation studies in children reported mixed evidence, with some improvement in cognitive outcomes, and a need for higher quality study designs (Ryan et al., Reference Ryan, Astwood, Gautier, Kuratko, Nelson and Salem2010). In the current study, we found some evidence that the association of child's vitamin-D levels with ADHD and autism-related traits could be due to shared genetic predisposition in trio-MR analyses.
Previous observational associations between maternal nutrient levels and offspring neurodevelopment could instead be due to reverse causation. Our bi-directional MR found evidence for genetic predisposition to autism predicting lower vitamin-D levels and genetic predisposition to ADHD predicting lower DHA levels. Individuals with autism often exhibit picky eating and resistance to new foods leading to a restricted diet (Guo et al., Reference Guo, Zhu, Yang, Lai, Lei, Chen and Li2019), which might lead to nutritional deficiencies during pregnancy. Genetic predisposition to neurodevelopmental conditions in the mother will also increase odds of neurodevelopmental conditions in the child through both direct and indirect genetic transmission. This transmission might appear as an association between nutrient levels and neurodevelopmental outcomes when shared genetic predisposition and reverse causation are not sufficiently accounted for in study design. Emerging evidence supports the hypothesis that neurodevelopmental genetic predisposition precedes difference in nutrient levels rather than the other way around. A recent analysis of the microbiome in individuals with autism found evidence to suggest that differences in composition were due to differences in diet, rather than causal effects of microbiome on autism predisposition (Yap et al., Reference Yap, Henders, Alvares, Wood, Krause, Tyson and Gratten2021). Such reverse effects have also been observed in studies of genetic confounding in MoBa, where mothers with higher genetic predisposition for ADHD were less likely to take vitamin supplements during pregnancy (Havdahl et al., Reference Havdahl, Wootton, Leppert, Riglin, Ask, Tesli and Stergiakouli2022). As a result, supplementation interventions that ensure pregnant women with neurodevelopmental conditions are not experiencing nutritional deficiencies could still be beneficial for other maternal and offspring outcomes where evidence for a causal effect is more robust.
Strengths and limitations
There were several strengths to our study design. First, we had a large sample of mother–father–child trios, with genotype data and detailed measures of child neurodevelopmental traits. To the best of our knowledge, this is the first implementation of a trio-MR design to test maternal effects of vitamin-D or DHA on offspring neurodevelopment. Previous studies have tried to control for shared genetic predisposition using only mother–child pairs however, adjusting for child PGS alone can lead to collider bias via father's genotype. We have also triangulated the findings with an MR study where we conducted several sensitivity analyses to assess possible bias from horizontal pleiotropy, and were able to bi-directionally assess reverse causation.
There were also several limitations to the current study, including the sample size of full trios. Power analyses suggested that we had 80% power to detect ORs < 1.105, so we were underpowered to detect smaller effects. However, such small maternal effects are unlikely to have a meaningful impact on neurodevelopmental outcomes. Second, we were not able to validate the genetic instruments for vitamin-D and DHA during pregnancy in the MoBa cohort. These genetic variants were identified in non-pregnant adults, and changes to metabolism during pregnancy might affect their suitability. However, we did validate the genetic instruments in a separate pregnancy cohort (ALSPAC), conducted a replication analysis using independent GWAS of diagnoses, and reran MR analyses using only canonical SNPs. These analyses also suggested that causal maternal effects were unlikely, increasing our confidence that the results are not limited to MoBa. Third, because SNP effects could not be regressed onto observed nutrient levels during pregnancy, maternal effects on offspring outcomes are not specific to pregnancy, but could also occur in the child's early life (e.g. through the food provided by the mother for the child). If effects of maternal genetic liability to nutrient levels act in the opposite direction during pregnancy than after birth, it could be the case that these effects cancel out, resulting in the lack of evidence observed here. However, such opposite effects seem unlikely, and we employed a paternal negative control design, in which we did not see evidence for effects of paternal nutrient levels on offspring outcomes via pathways occurring in early life. Fourth, the genetic variants for maternal circulating nutrients only proxy small variations and our analyses assume linear effects, meaning we might be underpowered to detect effects of extreme deficiencies. Finally, the MoBa sample is relatively homogeneous, and our analyses are restricted to individuals of European ancestry, hence, results might not be generalizable to other populations.
Conclusion
Triangulating across complementary study designs, we did not find evidence to support causal maternal effects of vitamin-D and DHA on offspring neurodevelopmental outcomes, after accounting for shared genetic predisposition. There was evidence suggesting possible reverse causation, with genetic predisposition to autism associated with lower vitamin-D levels and genetic predisposition to ADHD associated with lower DHA levels. Previous observational associations might have been (at least in part) due to shared genetic predisposition to neurodevelopmental conditions, which also influences nutrient levels. Future studies exploring the role of intrauterine effects on offspring neurodevelopment should control for shared genetic predisposition and use designs robust to reverse causation.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291724001466
Acknowledgements
We are grateful to Pål Surén for his input on the analysis plans. This work was performed on the TSD (Tjeneste for Sensitive Data) facilities, owned by the University of Oslo, operated and developed by the TSD service group at the University of Oslo, IT-Department (USIT). This project was conducted using Sigma2 (NS9791S).
The Norwegian Mother, Father and Child Cohort Study is supported by the Norwegian Ministry of Health and Care Services and the Ministry of Education and Research. We are grateful to all the participating families in Norway who take part in this on-going cohort study. We thank the Norwegian Institute of Public Health (NIPH) for generating high-quality genomic data. This research is part of the HARVEST collaboration, supported by the Research Council of Norway (NRC) (#229624). We also thank the NORMENT Centre for providing genotype data, funded by NRC (#223273), South East Norway Health Authority and KG Jebsen Stiftelsen. Further, we thank the Center for Diabetes Research, the University of Bergen for providing genotype data and performing quality control and imputation of the data funded by the ERC AdG project SELECTionPREDISPOSED, Stiftelsen Kristian Gerhard Jebsen, Trond Mohn Foundation, NRC, the Novo Nordisk Foundation, the University of Bergen, and the Western Norway Health Authorities (Helse Vest).
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists, and nurses.
Funding statement
Havdahl, Wootton, and Corfield are supported by South-Eastern Norway Regional Health Authority grants 2020024, 2018059, 2021045, and 2020022. Wootton and Riglin were supported by grant 204895/Z/16/Z from the Wellcome Trust, awarded to Thapar, Davey Smith, and Stergiakouli. For the purpose of Open Access, the authors have applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. Wootton, Dack, Borges, Jones, Madley-Dowd, Pagoni, Davey Smith and Stergiakouli work in a unit that receives funding from the University of Bristol and the UK Medical Research Council (MRC Integrative Epidemiology Unit, University of Bristol) (MC_UU_00032/01, MC_UU_00032/02, MC_UU_00032/05, MC_UU_00032/06) (https://mrc.ukri.org/). Havdahl and Corfield are supported by Research Council of Norway (RCN) grants 274611 and 336085. Havdahl's contribution was supported by the European Union's Horizon Europe Research and Innovation Programme (FAMILY, grant agreement No 101057529). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union, or the other funding agents. Neither the European Union nor the granting authorities can be held responsible for them. Borges has received support from the MRC Skills Development Fellowship (MR/P014054/1) and the University of Bristol Vice-Chancellor's Fellowship. Ask was supported by RCN grant 248983. This work was supported by the Wolfson Centre for Young People's Mental Health, established with support from the Wolfson Foundation. The Norwegian Mother, Father and Child Cohort Study is supported by the Norwegian Ministry of Health and Care Services, and the Ministry of Education and Research. The genotype data was provided by the HARVEST collaboration (supported by the Research Council of Norway (RCN) (#229624)), the NORMENT Centre (RCN #223273, South-Eastern Norway Regional Health Authority (SENRHA) and Stiftelsen Kristian Gerhard Jebsen) in collaboration with deCODE Genetics, and the Center for Diabetes Research at the University of Bergen (funded by the ERC AdG project SELECTionPREDISPOSED, Stiftelsen Kristian Gerhard Jebsen, Trond Mohn Foundation, the RCN, the Novo Nordisk Foundation, the University of Bergen, and the Western Norway Regional Health Authority). The UK Medical Research Council and Wellcome (Grant ref: 217065/Z/19/Z) and the University of Bristol provide core support for ALSPAC. Genotyping of ALSPAC mothers was funded by the Wellcome Trust (WT088806). This publication is the work of the authors – Robyn Wootton and Hannah Jones will serve as guarantors for the ALSPAC analyses in this paper. This study was supported by the National Institute for Health and Care Research Bristol Biomedical Research Centre. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.
Competing interests
The authors declare no conflicts of interest.
Ethical standards
The establishment and data collection in MoBa was previously based on a license from the Norwegian Data protection agency and approval from The Regional Committee for Medical Research Ethics. It is now based on regulations related to the Norwegian Health Registry Act. The current study was approved by The Regional Committees for Medical and Health Research Ethics (2016/1702). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.
Open access
