Cell-Type Interaction QTLs
Summary¶
Kim-Hellmuth, Aguet, Oliva, et al. (2020) computationally recover cell-type-specific genetic regulation from bulk GTEx tissue RNA-seq, without single-cell sequencing, by testing for a statistical interaction between genotype and in silico-estimated cell-type abundance (via xCell) for 43 cell-type/tissue combinations spanning 7 cell types and 35 tissues. These "cell-type interaction QTLs" (ieQTLs for expression, isQTLs for splicing; iQTLs jointly) recover regulatory effects invisible to standard bulk cis-eQTL mapping, explain a meaningful share of eQTL tissue-specificity, and improve GWAS colocalization discovery by hundreds of loci relative to bulk tissue alone.
Method¶
- Cell-type deconvolution: xCell (benchmarked against several alternatives on correlation with blood cell counts, marker-gene expression, and histology/pathology concordance) estimated the enrichment of each cell type per bulk RNA-seq sample; xCell scores correlated strongly with the PEER factors already used to correct GTEx expression data for unobserved confounders, indicating cell-type composition explains much of that "technical" variance.
- Interaction testing: for each tissue where a given cell type was well-enriched, a linear regression tested genotype × cell-type-abundance interaction effects on gene expression (ieQTL) or splicing (isQTL).
Findings¶
- Coverage: 3,347 protein-coding/lincRNA genes with an ieQTL and 987 genes with an isQTL (5% FDR) across the 43 cell-type/tissue combinations. 21% of ieQTLs were not in LD (R²<0.2) with any conditionally independent standard bulk cis-eQTL for the same gene — genuinely new signal invisible to standard eQTL mapping, not just a re-discovery.
- Validation: allele-specific expression data corroborated a median 62% of ieQTLs (mean π1 = 0.75 for well-powered cell-type/tissue pairs); external cohort replication (GAIT2 whole blood, purified neutrophils, TwinsUK adipose/skin, Mayo temporal cortex) ranged π1 = 0.32–0.67.
- Tissue specificity: cell-type ieQTL status was a strong negative predictor of cross-tissue eQTL sharing (comparable in magnitude to enhancer annotation), and tissue-specific eGenes had significantly higher rates of cell-type ieQTLs than tissue-shared eGenes (P=1.9×10⁻⁵) — cell-type-specific regulation is a substantial driver of the tissue specificity long observed in bulk eQTL catalogs.
- GWAS relevance: across 87 complex-trait GWAS, both ieQTLs and isQTLs were enriched for GWAS signal (median 1.3–1.4-fold), comparable to standard QTLs, and colocalization analysis using iQTLs recovered hundreds of GWAS loci that were not colocalized using bulk-tissue QTLs alone — cellular heterogeneity within a sampled tissue can mask real disease-relevant regulatory signal from standard bulk analysis.
Availability¶
Data and visualizations: GTEx Portal (open-access) and dbGaP (phs000424.v8, protected). QTL mapping pipeline: github.com/broadinstitute/gtex-pipeline; TensorQTL: github.com/broadinstitute/tensorqtl.
See Also¶
- Statistical Colocalization — the analysis these iQTLs feed into to recover additional GWAS loci.
- MetaXcan — the broader GTEx-based TWAS ecosystem this cell-type-resolution refinement complements.
- lncRNA Genetic Regulation and Disease Mapping — a companion GTEx v8 analysis applying the same colocalization approach to long non-coding RNAs.