A western dietary pattern during pregnancy is associated with neurodevelopmental disorders in childhood and adolescence

Abstract

Despite the high prevalence of neurodevelopmental disorders, the influence of maternal diet during pregnancy on child neurodevelopment remains understudied. Here we show that a western dietary pattern during pregnancy is associated with child neurodevelopmental disorders. We analyse self-reported maternal dietary patterns at 24 weeks of pregnancy and clinically evaluated neurodevelopmental disorders at 10 years of age in the COPSAC2010 cohort (n = 508). We find significant associations with attention-deficit hyperactivity disorder (ADHD) and autism diagnoses. We validate the ADHD findings in three large, independent mother-child cohorts (n = 59,725, n = 656 and n = 348) through self-reported dietary modelling, maternal blood metabolomics and foetal blood metabolomics. Metabolome analyses identify 15 mediating metabolites in pregnancy that improve ADHD prediction. Longitudinal blood metabolome analyses, incorporating five time points per cohort in two independent cohorts, reveal that associations between western dietary pattern metabolite scores and neurodevelopmental outcomes are consistently significant in early-mid-pregnancy. These findings highlight the potential for targeted prenatal dietary interventions to prevent neurodevelopmental disorders and emphasise the importance of early intervention.

Extended Data Fig. 1 |. Biplot of the first two principal components (PCs) from the maternal food frequency questionnaire-derived nutrient constituents.

The left panel displays box-and-whisker plots summarizing the distribution of PC2 scores, and the bottom panel shows PC1 scores, for children stratified by neurodevelopmental diagnosis status (NDD, ADHD, and autism), with boxes representing the interquartile range (IQR; 25th–75th percentile), centre lines indicating the median, and whiskers extending to 1.5 times the IQR. PC2 (Western dietary pattern) is significantly associated with any neurodevelopmental disorder (OR 1.53 (1.17–2.00), p = 0.002), ADHD (OR 1.66 (1.21–2.27), p = 0.002), and autism diagnosis (OR 2.22 (1.33–3.74), p = 0.002). PC1 is not significantly associated with any neurodevelopmental disorders (p > 0.288). Further details of the loadings for each of the 95 nutrient constituents can be viewed in supplementary table 2. Adjustments were not made for multiple comparisons.

Extended Data Fig. 2 |. Biplot of the first two principal components from the maternal FFQ-derived nutrient constituents.

Biplot of the first two principal components from the maternal FFQ-derived nutrient constituents A) Nutrient constituents are categorized into fatty acids, amino acids, sugars, minerals and vitamins. Fatty acids are a key determinant of PC2 (Western dietary pattern) B) Stratified further by fatty acid type, saturated fatty acids are most associated with PC2 (Western dietary pattern).

Extended Data Fig. 3 |. Comparison of maternal blood metabolomes at three different pregnancy time points from two mother–child cohorts.

Top panels show a Principal Component Analysis (PCA) scoreplot on all metabolites, as well as the selected metabolite scores for the COPSAC2010 vs VDAART 10–18 weeks (left) and COPSAC2010 vs VDAART 32–38 weeks (right). Bottom panels show the relative variation per metabolite computed by the ratio of sums of squares (SSQ_time/SSQ_residual) from a one-way anova model with Time/Cohort at predictor and Comparison of per metabolite standard deviation within cohort relative to 24-week gestation pregnancy time point from COPSAC2010 for the COPSAC2010 vs VDAART 10–18 weeks (left) and COPSAC2010 vs VDAART 32–38 weeks (right).

Extended Data Fig. 4 |. Linear regression associations between pregnancy dietary intake and western dietary pattern metabolite scores in the VDAART cohort.

Linear regression associations between dietary intake from food frequency questionnaires (FFQs) during pregnancy and Western Dietary Pattern Metabolite Scores (WDP-MS) at two distinct time points (10–18 weeks (n = 775) and 32–38 weeks (n = 780)) in the VDAART cohort. The depicted associations, based on the VDAART cohort, were assessed using COPSAC2010 cohort-trained models that shared overlapping metabolites. Positive and negative associations are represented by colour coding, with confidence intervals reflecting the uncertainty of the estimates. Bar plots represent model estimates with error bars indicating 95% confidence limits. Adjustments were not made for multiple comparisons.

Extended Data Fig. 5 |. Metabolites associated with a western dietary pattern and their mediation of ADHD diagnosis in the VDAART cohort.

Metabolites associated with a Western dietary pattern at VDAART gestational time points (10–18 weeks and 32–38 weeks), based on models from the COPSAC2010 cohort. The selected metabolite scores are depicted by bars, distinguished by their metabolic pathway. Striped bars indicate metabolites that mediate the association with ADHD Diagnosis in COPSAC2010, while solid bars signify non-mediating metabolites. The directionality of the bars represents the positive or negative metabolite score. Metabolite scores from VDAART are colour-coded for each time point: 10–18 weeks (green) and 32–38 weeks (purple).

Extended Data Fig. 6 |. Comparative analysis of metabolomic data across newborn dried blood spots from the COPSAC2010 and COPSAC2000 cohorts.

Comparative analysis of metabolomic data across newborn dried blood spots from the COPSAC2010 and COPSAC2000 mother–child cohorts. Principal Component Analysis (PCA) plots illustrate the distribution and separation of metabolomic data between cohorts. Panels A–C represent overlapping metabolomes used in our analysis (n = 951, n = 774, n = 626), showcasing subsets of metabolites selected based on varying degrees of correlation and distributional similarity between cohorts. A: displays the PCA for metabolites with correlations greater than 0.4 (n = 951), B: Correlations greater than 0.6 (n = 774), and C: Metabolites with a correlation greater than 0.6 passing the two-sample Kolmogorov-Smirnov test with a threshold of >=0.05 (n = 626).

Extended Data Fig. 7 |. Modulation of the western dietary pattern, stratified by child sex, on neurodevelopmental outcomes.

Odds ratio estimates of ADHD (A) and autism diagnoses (B) based on interactions of the Western dietary pattern, maternal pre-pregnancy BMI (split at median value 23.7), and child’s polygenic risk score (PRS) (median split) for ADHD and autism, stratified by child sex. Linear regression estimates for ADHD (C) and autism symptom loads (D) based on the Western dietary pattern, considering tertiles of maternal pre-pregnancy BMI ( < 22.2, 22.2–25.4, >25.4) and child’s PRS for ADHD and autism, stratified by child sex. Stars represent significance levels: * indicates p < 0.05, ** indicates p < 0.01, and *** indicates p < 0.001, with “NS” denoting non-significant results (p ≥ 0.05). Further details, including the individual associations of these modulating factors, can be found in Supplementary Table S11. Adjustments were not made for multiple comparisons.

Extended Data Fig. 8 |. Modulation of the western dietary pattern metabolite score on neurodevelopmental outcomes in COPSAC2010 cohort.

Odds ratio estimates for ADHD (A) and autism diagnoses (B) based on interactions of the Western dietary pattern metabolite score, maternal pre-pregnancy BMI (split at median value 23.7), and child’s polygenic risk score (PRS) (median split) for ADHD and autism. Linear regression estimates for ADHD (C) and autism (D) symptom loads are based on the Western dietary pattern metabolite score, considering tertiles of maternal pre-pregnancy BMI and child’s PRS for ADHD and autism. Stars represent significance levels: * indicates p < 0.05, ** indicates p < 0.01, and *** indicates p < 0.001, with “NS” denoting non-significant results (p ≥ 0.05). Further details, including the individual associations of these modulating factors, can be found in Supplementary Table S11. Adjustments were not made for multiple comparisons.

Extended Data Fig. 9 |. The western dietary patterns metabolite scores association with ADHD and autism symptom loads, considering standard clinical classifications of maternal pre-pregnancy BMI.

Panels A and D represent ADHD and autism symptom loads respectively, stratified by low/high genetic risk (median cut). Panels B and E further contextualize these associations based on maternal pre-pregnancy BMI categories. Panels C and F again contextualize these associations by stratifying by child sex (male sex shown). Panels A–C include n = 522, while Panels D–F include n = 523. Data represent model estimates with error bars indicating 95% confidence limits. Adjustments were not made for multiple comparisons.

Extended Data Fig. 10 |. Study design and validation of the western dietary pattern–neurodevelopmental association across cohorts.

Study design illustrating the discovery and validation of associations between a western dietary pattern during pregnancy and ADHD outcomes. The primary analysis was conducted in the COPSAC2010 mother-child cohort (n = 508), leveraging dietary data from pregnancy food frequency questionnaires (FFQ) and blood metabolomics profiling to derive a western dietary pattern metabolite score. Validation of findings was performed across three independent cohorts—Danish National Birth Cohort (DNBC, n = 59,625), VDAART (n = 656), and COPSAC2000 (n = 348)—utilizing distinct methodologies: FFQ-based dietary pattern analysis in DNBC, maternal metabolomic validation in VDAART, and neonatal dried blood spot (DBS) metabolomics in COPSAC2000.

Fig. 1 |. Associations between maternal dietary patterns and neurodevelopmental diagnoses in children.

Pregnancy varied and western dietary patterns, derived from nutrient constituents using PCA (PC1 and PC2), and their associations with food groups and neurodevelopmental outcomes. The top panels display scaled linear regression estimates with 95% confidence intervals for food groups positively (yellow for PC1, red for PC2) and negatively (green for PC2) associated with the dietary patterns (n = 594); adjustment was not made for multiple testing. The bottom panels present multivariable regression results showing ORs with 95% confidence intervals for the associations between dietary patterns and neurodevelopmental diagnoses (n = 508). The western dietary pattern is significantly associated with any neurodevelopmental disorder diagnosis (OR 1.53 (1.17–2.00), P = 0.002), ADHD diagnosis (OR 1.66 (1.21–2.27), P = 0.002), and autism diagnosis (OR 2.22 (1.33–3.74), P = 0.002), while the varied dietary pattern shows no significant associations (P > 0.288).

Fig. 2 |. Graphical models of western dietary pattern associations with neurodevelopmental outcomes and covariates.

a–d, Graphical models illustrate associations (partial correlations P < 0.05) between the western dietary pattern during pregnancy, neurodevelopmental outcomes ADHD diagnosis (a), ADHD symptom load (ADHD-RS) (b), autism diagnosis (c), autism symptom load (SRS-2) (d), and model covariates. Models include child ADHD PRS (a,b) and autism PRS (c,d) as additional covariates.

Fig. 3 |. Metabolites associated with the western dietary pattern and their mediation of neurodevelopmental disorders.

The 43 metabolites, selected by the sparse partial least squares model, represent those associated with the western dietary pattern during pregnancy at 24 weeks gestation in COPSAC2010. Positive metabolite scores indicate a positive association, while negative scores suggest an inverse relationship with this dietary pattern. A systematic backward elimination pinpointed 15 metabolites as mediators between the western dietary pattern and any neurodevelopmental disorders. Notably, dietary-derived compounds, like ergothioneine, suggest potential protective roles, while certain lipid-associated metabolites hint at possible detrimental impacts on neurodevelopment.

Fig. 4 |. Moderation of the western dietary pattern on neurodevelopmental outcomes in COPSAC2010 cohort.

a,b, OR estimates for ADHD (a) and autism diagnoses (b) based on interactions of the western dietary pattern, maternal pre-pregnancy BMI (split at median value 23.7) and child’s PRS (median split) for ADHD and autism. c,d, Linear regression estimates for ADHD (c) and autism (d) symptom loads based on the western dietary pattern, considering tertiles of maternal pre-pregnancy BMI and child’s PRS for ADHD and autism. ORs and estimates are in relation to a change of 1 s.d. of the western dietary pattern. Stars represent significance levels: *P < 0.05, **P < 0.01, ***P < 0.001, NS, non-significant (P ≥ 0.05). Further details, including the individual associations of these modulating factors, can be found in Supplementary Table 11. Adjustments were not made for multiple comparisons.