Multiancestry brain pQTL fine-mapping and integration with genome-wide association studies of 21 neurologic and psychiatric conditions

Abstract

To understand shared and ancestry-specific genetic control of brain protein expression and its ramifications for disease, we mapped protein quantitative trait loci (pQTLs) in 1,362 brain proteomes from African American, Hispanic/Latin American and non-Hispanic white donors. Among the pQTLs that multiancestry fine-mapping MESuSiE confidently assigned as putative causal pQTLs in a specific population, most were shared across the three studied populations and are referred to as multiancestry causal pQTLs. These multiancestry causal pQTLs were enriched for exonic and promoter regions. To investigate their effects on disease, we modeled the 858 multiancestry causal pQTLs as instrumental variables using Mendelian randomization and genome-wide association study results for neurologic and psychiatric conditions (21 traits in participants with European ancestry, 10 in those with African ancestry and 4 in Hispanic participants). We identified 119 multiancestry pQTL-protein pairs consistent with a causal role in these conditions. Remarkably, 29% of the multiancestry pQTLs in these pairs were coding variants. These results lay an important foundation for the creation of new molecular models of neurologic and psychiatric conditions that are likely to be relevant to individuals across different genetic ancestries.

Fig. 1. Study design.

We performed population-stratified pQTL mapping, multiancestry pQTL fine-mapping and causal inference.

Fig. 2. Causal pQTLs specific to or shared among populations.

a,b, LocusZoom plots of two loci identified as causal pQTLs specific to a population (a) and shared among three populations (b) based on MESuSiE pQTL fine-mapping results: the DDX58 locus on chromosome 9 for each population (a) and the PPIL3 locus on chromosome 2 for each population (b). Tested SNPs within each locus are plotted by position on the x axis and −log P value on the y axis for the association with the abundance of the respective protein. The putative causal site is labeled and denoted by a purple diamond, and each tested SNP is colored by its $r^2$ with the causal site. The statistical test used was the restricted likelihood ratio test with multiple testing correction, followed by the Bayesian framework. Full lists of these causal pQTLs are provided in Supplementary Tables 3 and 4.

Fig. 3. Candidate causal proteins in neurologic and psychiatric conditions derived from multiancestry causal pQTLs using Mendelian randomization.

Using the 858 multiancestry causal pQTLs as the instrument variable, we performed Mendelian randomization using population-matched GWASs. Proteins with SMR FDR < 0.05 and HEIDI P > 0.05 were considered to be candidate causal proteins. In total, 119 multiancestry causal pQTL–protein–trait triads were found in the NHW population. None was found in the AA or Hispanic population, likely owing to smaller GWASs with limited power. Fifteen of the 21 tested brain conditions had multiancestry causal pQTL–protein–trait triads. a, Bar plot listing the causal proteins for each trait on the horizontal axis. Colors denote whether the protein was identified as causal for two or more conditions or targeted by an existing drug compound. b, Upset plot illustrating traits for the causal proteins identified as causal for ≥2 traits. For example, POR is a causal protein in alcohol use disorder targeted by existing drug compounds; it is also causal in major depression, and neuroticism. Detailed results are provided in Supplementary Table 17.

Fig. 4. Causal proteins and their physical PPIs and gene set enrichment results for Alzheimer’s disease and anxiety symptoms.

Candidate causal proteins were identified through population-matched integration of brain proteomes and GWASs using PWASs and SMR/HEIDI. Each gene is plotted as a circle, with the color denoting an enriched process (gray indicates genes not in an enriched gene set). Solid lines connecting the genes denote evidence of PPI. a, For Alzheimer’s disease, 64 causal proteins were identified. Of these, 29 had evidence of PPIs among themselves, and there was gene set enrichment for endosomal proteins (FDR < 0.05). b, For anxiety symptoms, ten causal proteins were identified, and seven had PPIs with each other. Some of these anxiety causal proteins were enriched in cellular components of the cytoplasmic region, dendritic tree, somatodendritic neuron projection, synapse and glutamatergic synapse (FDR < 0.05). Full results are provided in Supplementary Tables 18 and 19.

Extended Data Fig. 1. Cross-ancestry prediction of genetically regulated protein expression for the pGenes corresponding to the 858 multi-ancestry causal pQTLs.

The plot depicts accuracy of cross-ancestry prediction of genetically regulated protein abundance for the pGenes corresponding to the 858 multi-ancestry causal pQTLs. The x-axis shows the ratio of the variance explained (R²) of cross-ancestry to same-ancestry prediction. Higher ratios indicate more accuracy. The y-axis reflects the number of pGenes for a particular ratio bin. The data underlying this plot are provided as Source Data. Source data