Brain expression quantitative trait locus and network analyses reveal downstream effects and putative drivers for brain-related diseases.

Overview of the study.We downloaded publicly available RNA-seq and genotype data from 14 different datasets consisting of 8,613 RNA-seq measurements from seven main brain regions and 6,518 genotype samples. We created six eQTL meta-analysis datasets and performed cis-, trans- and interaction-eQTL analyses, built a brain-specific gene co-regulation network and prioritized genes using MR, co-localization and the co-regulation network. Image of sagittal cut of brain created with BioRender.com. This figure summarizes values from Supplementary Tables 1, 3, 8, 12 and 25–30.

Overview of the datasets.a, Number of samples per included cohort stratified according to the seven major brain regions. b, PCA dimensionality reduction plot of the normalized expression data after covariate correction. Each dot represents an RNA-seq sample and is colored according to the brain region. The figure shows that the samples cluster mainly on the brain region. c, Number of genotypes per cohort stratified according to ancestry. AMR, Admixed Americans; SAS, South Asian. d, Number of individuals per cohort, with each color representing an eQTL dataset. The number of individuals differ from the intersection between the number of RNA-seq samples and number of genotypes because not all samples with genotypes have RNA-seq samples and vice versa, and some individuals with genotypes have multiple RNA-seq measurements.Source data

Conditional cis-eQTLs.a, Number of conditional cis-eQTLs per eQTL dataset. b, Comparison of characteristics between primary and non-primary eQTLs. For the mean expression and pLI score values, each row compares the eQTL genes for that rank with eQTL genes from the previous rank (for example, for tertiary eQTLs, the non-significant (grey) distribution is from genes that have secondary but lack tertiary eQTLs). P values were calculated using a Wilcoxon test between significant and non-significant genes. Differences in mean gene expression levels (left), the distance between the most significant SNP–gene combination and the TSS (middle), and pLI scores (right) are shown. For primary, secondary and quaternary eQTLs, non-significant eQTLs have higher pLI scores. Vertical dotted lines indicate median of the distribution for the current rank (coloured) versus the previous rank (black). c,d, Number of overlapping eQTLs along with the Rb and standard error (s.e.) values of primary cis-eQTLs between the cortex eQTLs of different ancestries (c) and the different brain regions for the EUR datasets (d). n, sample size of the dataset. e, Correlation of effect sizes and standard error of primary cis-eQTLs of Cortex-EUR (discovery, excluding GTEx) in all of the GTEx tissues (replication). Each dot is a different GTEx tissue; the x axis indicates the number of eQTLs that are significant in both discovery and replication.Source data

Cell-type ieQTLs.a, Pearson’s correlations between the seven predicted cell-count proportions within cortex samples. b, Predicted cell-type proportions compared with cell-type proportions measured using IHC for 42 ROSMAP samples. Pearson’s correlation coefficients are provided. The cell-count predictions for most cell types closely approximate actual IHC cell counts. Shaded areas around regression lines indicate 95% confidence interval. c, Number of cell-type ieQTLs for Cortex-EUR deconvoluted cell types. The first 20 intersections with the highest overlap are shown. Oligodendrocytes have the most interactions, followed by astrocytes and other neurons. Notably, most interactions are unique for one cell type in 87.1% of the cases. d–f, Replication of cell-type ieQTLs for STMN4 (d), FAM221A (e) and CD38 (f), consisting of the scatterplot of the cell-type ieQTL in MetaBrain Cortex-EUR bulk RNA-seq (left) and a forest plot for the eQTL effect in the ROSMAP snRNA-seq data (right). Each dot in the scatterplots (left) represents a sample; colors indicate SNP genotype, with yellow being the minor allele; values under the genotypes are the Pearson’s correlation coefficients; interaction P values were determined using a one-sided F-test; eQTL P values were derived using the standard normal distribution from meta-analyzed z-scores. Forest plot (right): eQTL β values (dots) and standard error (error bars) with effect direction relative to the minor allele when replicating the eQTL effect in ROSMAP single-nucleus data (n = 38); each row denotes a cell type-specific dataset; cell types highlighted in bold reflect the equivalent to the cell type used in the ieQTL. Vertical dashed lines indicate an eQTL beta of 0. TMM, trimmed mean of M-values; AST, astrocytes; END, endothelial cells; EX, excitatory neurons; IN, inhibitory neurons; MIC, microglia; OPC, oligodendrocyte precursor cells; OLI, oligodendrocytes; NEU, other neuron; PER, pericytes.Source data

Co-localization and MR analysis of brain-related traits.a, Number of significant Wald ratio effects (blue) and those with both Wald ratio and co-localization (coloc; red) evidence for 13 brain-related traits. b, Forest plots showing the SNP and effect allele (EA), eQTL β and GWAS odds ratio for 20 MS-associated genes that are both MR and co-localization significant as well as their Wald ratio P value. The dots indicate the eQTL β or odds ratio, the error bars indicate the 95% confidence interval. WR, Wald ratio; and OR, odds ratio. Grey dotted line indicates an OR of 1. c,d, Cell-type ieQTL for CYP24A1 (c) and CLECL1 (d) showing interactions with predicted excitatory neuron and microglia proportions, respectively. Each dot represents a sample. Colors indicate the SNP genotype, with yellow being the minor allele. Values under the alleles are the Pearson’s correlation coefficients. Interaction P values were derived using a one-sided F-test; eQTL P values were derived using the standard normal distribution from meta-analyzed z-scores. TMM, trimmed mean of M-values.Source data

Trans-eQTLs in the brain.a, Location of the identified trans-eQTLs (SNP and gene positions) in the genome. The size of the dots indicates the P value of the trans-eQTL (larger is more significant). b, Two examples of convergent effects, where multiple independent SNPs affect the same genes in trans. Trans-eQTLs of rs1427407 and rs4895441 on HBG2 (top). Trans-eQTL of rs1150668 and rs106871 on ZNF31 and S100A5 (bottom). Both panels are derived from Supplementary Table 17.

Gene co-regulation.a, Genes that are co-regulated with genes that are within ALS-associated loci. Co-regulation scores between genes were calculated using the MetaBrain cerebellum and MetaBrain cortex samples as well as the combined (all) MetaBrain samples. Except for URB4, cortex and cerebellum networks find different co-regulated genes for ALS. b, Co-regulation network using all MetaBrain samples for all genes prioritized for ALS by Downstreamer. c, Top five Human Phenotype Ontology (HPO) enrichments for the Downstreamer prioritized ALS-associated genes. d, Genes that are co-regulated with genes that are within MS-associated loci. Co-regulation scores between genes were calculated using a heterogeneous multi-tissue network, MetaBrain cerebellum samples and MetaBrain cortex samples. Most genes were found using a large heterogenous co-regulation network. e, Co-regulation network of all MetaBrain samples for 33 genes prioritized by Downstreamer in cortex. f, Top five Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichments for the Downstreamer prioritized MS genes in the cortex highlight neurotrophin signaling pathway enrichment (red). c,f, Enrichment P values were calculated using a two-sided χ2 test. Panels a and d are derived from Supplementary Tables 25, 29 and 30.