Huwenbo Shi
Postdoctoral Researcher, Harvard T.H. Chan School of Public Health
Enabling transcriptome-wide association studies at cell-type resolution
Transcriptome-wide association studies (TWAS) leveraging gene expression predictions from cis SNPs have identified thousands of genes associated with disease, and can be performed using summary-level GWAS data (Wainberg et al. 2019 Nat Genet). However, TWAS generally use gene expression predictions for bulk tissue (GTEx Consortium 2020 Science), and cannot pinpoint specific cell-types.
Here, we introduce a powerful approach for TWAS at cell-type resolution, leveraging the large sample size of GTEx and the cell-type resolution of mouse single-cell RNA-seq (Tabula Muris Consortium 2020 Nature). We infer cell-type proportions and cell-type specific gene expression for each Tabula Muris cell-type for each GTEx sample, enabling gene expression predictions for each cell-type.
We performed cell-type TWAS using GWAS summary statistics for 13 well-powered diseases and complex traits (average N=272,698), analyzing 343 tissue-cell-type pairs. We discovered 34,267 unique gene-trait associations at FDR<0.05, vs. 13,042 for standard TWAS across 29 GTEx tissues (+163% increase), pinpointing specific cell-types. For example, we determined that the association of SORT1 expression with LDL is significant in liver hepatocyte (P=2×10E-63), and the association of CACNA1C expression with schizophrenia is significant in brain cerebellum astrocyte and neuron (P=2×10E-8 for each), much more significant than other cell-types.
