Abundant pleiotropy across neuroimaging modalities identified through a multivariate genome-wide association study

Abstract

Genetic pleiotropy is abundant across spatially distributed brain characteristics derived from one neuroimaging modality (e.g. structural, functional or diffusion magnetic resonance imaging [MRI]). A better understanding of pleiotropy across modalities could inform us on the integration of brain function, micro- and macrostructure. Here we show extensive genetic overlap across neuroimaging modalities at a locus and gene level in the UK Biobank (N = 34,029) and ABCD Study (N = 8607). When jointly analysing phenotypes derived from structural, functional and diffusion MRI in a genome-wide association study (GWAS) with the Multivariate Omnibus Statistical Test (MOSTest), we boost the discovery of loci and genes beyond previously identified effects for each modality individually. Cross-modality genes are involved in fundamental biological processes and predominantly expressed during prenatal brain development. We additionally boost prediction of psychiatric disorders by conditioning independent GWAS on our multimodal multivariate GWAS. These findings shed light on the shared genetic mechanisms underlying variation in brain morphology, functional connectivity, and tissue composition.

Fig. 1. Overview of the study.

a This study investigates statistical pleiotropy by overlapping loci and genes across unimodal multivariate GWAS, and additionally performing a multimodal multivariate GWAS. b Based on the overlapping patterns we classify single-modality and cross-modality loci and genes, for which we investigated potential differences in biological convergence. c Lastly, we use the multimodal multivariate GWAS statistics to adapt standard polygenic scores for five psychiatric conditions. This Figure includes images obtained from Servier Medical Art (https://smart.servier.com/), licensed under a Creative Commons Attribution 4.0 Unported License (https://creativecommons.org/licenses/by/4.0/).

Fig. 2. Abundant pleiotropy of loci and genes across neuroimaging modalities.

Overlap of genome-wide significant a loci and b genes observed across neuroimaging modalities in single-modality and joint multimodal analyses with MOSTest (Bonferroni corrected p < 5 × 10⁻⁸/3 and 8.83 × 10⁻⁷ respectively). When sMRI, fMRI and dMRI-derived phenotypes are jointly analyzed in MOSTest (multimodal analysis), a boost in discovery of pleiotropic loci and genes is observed (yellow). This pattern partially replicates in the ABCD cohort (Supplementary Fig. 5). For 10 example lead SNPs from different parts of a) the univariate GWAS z-scores of all phenotypes used in this study (sMRI-derived phenotypes in green, dMRI-derived phenotypes in red, fMRI-derived phenotypes in blue) are plotted in c as the distance from the center of each circular plot. The dashed black line corresponds to p = 5 × 10⁻⁸ (z = 5.45) and z-scores that pass this threshold are depicted larger. Positive effect direction is shown as a filled circle and negative effect direction as a white circle.

Fig. 3. Comparing properties of single-modality and cross-modality lead SNPs and genes.

a Gene-ontology biological processes, molecular functions and cellular components that were (Bonferroni-corrected) significantly enriched for cross-modality, sMRI-modality and/or dMRI-modality genes (none of the GO terms tested showed enrichment for the 6 fMRI-modality genes). Node size reflects gene-set size, edges reflect pathway similarity scores (Methods). b Functional consequences of single-modality and cross-modality lead SNPs as annotated with ANNOVAR. When the null hypothesis (OR = 1) could be rejected after Bonferroni correction (p < 1.14 × 10⁻³), the solid line indicates significant enrichment of the annotation. Annotations of 43,492 (unique) lead SNPs derived from 558 traits across 24 trait domains from Watanabe et al. were used as reference for Fisher Exact Test (Supplementary Data 9). ncRNA non-coding RNA, UTR untranslated region. c Mean-normalized expression (y-axis) of cross-modality and single-modality genes over developmental timepoints (x-axis; log10 scale). Gray shading indicates 95% confidence intervals. The mean-normalized expression of fMRI-modality genes is displayed in Supplementary Fig. 8, since the number of genes (n = 5) was low and therefore created an unreliable pattern. d Cell-type enrichment analysis with Fisher Exact test for fetal brain tissue from Bhaduri et al.. Bonferroni corrected significant results (p < (0.05/30 = ) 1.67 × 10⁻³) are indicated by an asterisk (*). OR odds ratio, ipc intermediate progenitor cells, oligo oligodendrocytes.

Fig. 4. Phenotypic variance explained for five psychiatric disorders explained by polygenic scores based on original disorder GWAS summary statistics (original PGS) and conditional disorder-multimodal GWAS summary statistics (pleioPGS).

Note that PGS using top 10 and top 100 SNPs had too low R² values to be visible and are therefore not plotted, but can be found in Supplementary Data 17. Independent target samples were used for prediction (see Methods). A likelihood ratio test was applied to compare the model which includes only the best PGS (highest R²) with the model including both PGS for each disorder, and is indicated with an asterisk (*) if significant (p < 0.05). MDD major depressive disorder, BD bipolar disorder, SCZ schizophrenia, ASD autism spectrum disorder, ADHD attention deficit hyperactivity disorder.