Researchers have leveraged a large single‑nucleus RNA sequencing (snRNA‑seq) dataset from the dorsolateral prefrontal cortex (DLPFC) to create cell‑type‑specific expression models and apply them to genome‑wide association studies (GWAS) of neuropsychiatric and neurodegenerative disorders. The effort, part of the PsychAD Consortium, integrates genetic and transcriptomic data from 1,494 donors, encompassing European (EUR), African (AFR) and admixed American (AMR) ancestries, and over six million nuclei.
Building ancestry‑aware, cell‑type‑specific models
Using quality‑controlled genotype and snRNA‑seq data, the team trained 94 single‑nucleus transcriptomic imputation models (snTIMs) across 32 cellular populations in the DLPFC. Robust cross‑validated performance was required for downstream analysis. For protein‑coding genes, reliable models were obtained for 12,289 genes (74.1% of 16,594 autosomal genes) in the EUR cohort (N=920), 10,375 genes (62.6%) in AFR (N=321), and 5,543 genes (33.4%) in AMR (N=118). By contrast, a bulk‑tissue model derived from pooled nuclei (snBulk) and a conventional homogenate‑based bulk model each imputed roughly 9,000 genes, with a 66.7–68.1% overlap. Increasing resolution from snBulk to broader cellular classes and then to subclasses added 3,330 uniquely imputable genes, yielding totals of 11,021 and 11,052 genes at class and subclass levels respectively. Higher‑resolution models more frequently achieved the top cross‑validated R², indicating improved predictive performance.
Cross‑ancestry validation and discovery of gene‑trait associations
To assess out‑of‑sample accuracy, the models were tested against two independent datasets: an EUR snRNA‑seq cohort from the Religious Orders Study/Memory and Aging Project (ROSMAP) and bulk RNA‑seq from fluorescence‑activated cell sorting‑isolated microglia (FACS‑MG). Both showed strong agreement between predicted and observed expression, supporting reproducibility. Correlations of model performance across ancestries were moderate but significant (EUR‑AMR ρ=0.58, EUR‑AFR ρ=0.54, AFR‑AMR ρ=0.49), suggesting conserved genetic regulation of brain gene expression despite differing SNP predictors.
Applying the EUR bulk and snTIMs to summary‑level GWAS for 12 neuropsychiatric and neurodegenerative traits, researchers identified 1,494 significant gene‑trait associations (GTAs) with the bulk model and 1,254 with snBulk. Of the genes imputable by both, 882 were significant in snBulk and 963 in bulk, with 529 overlapping (Jaccard = 0.40). Finer cellular resolution uncovered many additional links: subclass models revealed 3,003 unique GTAs across all traits, class models 2,470, and snBulk 1,254. Gene‑set enrichment analyses demonstrated that bulk and snBulk results were significantly enriched for known central‑nervous‑system disease genes in five and four of the twelve disorders respectively, underscoring the clinical relevance of the expanded cell‑type‑specific findings.
Implications for brain disorder genetics and therapeutics
The study provides an ancestry‑aware, cell‑type‑resolved atlas of genetically regulated expression in the human prefrontal cortex. By expanding the set of imputable genes and improving prediction accuracy through single‑nucleus resolution, the approach uncovers gene‑trait connections that bulk tissue analyses miss. The cross‑ancestry consistency of regulatory patterns supports the generalizability of findings to diverse populations. Ultimately, these results illustrate how single‑nucleus transcriptomics can sharpen gene discovery, clarify causal pathways, and aid therapeutic target prioritization for complex brain disorders.
Mitchell Landsberg is a Senior Technology Correspondent at News Raise. He covers consumer electronics, artificial intelligence, software developments, and digital privacy trends.
