Alternative splicing induces sample-level variation in gene–gene correlations

Abstract

Abstract Background The vast majority of genes in the genome are multi-exonic, and are alternatively spliced during transcription, resulting in multiple isoforms for each gene. For some genes, different mRNA isoforms may have differential expression levels or be involved in different pathways. Bulk tissue RNA-seq, as a widely used technology for transcriptome quantification, measures the total expression (TE) levels of each gene across multiple isoforms in multiple cell types for each tissue sample. With recent developments in precise quantification of alternative splicing events for each gene, we propose to study the effects of alternative splicing variation on gene–gene correlation effects. We adopted a variance-component model for testing the TE–TE correlations of one gene with a co-expressed gene, accounting for the effects of splicing variation and splicing-by-TE interaction of one gene on the other. Results We analyzed data from the Genotype-Tissue Expression (GTEx) project (V8). At the 5% FDR level, 38,146 pairs of genes out of ∼10 M examined pairs from GTEx lung tissue showed significant TE-splicing interaction effects, implying isoform-specific and/or sample-specific TE–TE correlations. Additional analysis across 13 GTEx brain tissues revealed strong tissue-specificity of TE-splicing interaction effects. Moreover, we showed that accounting for splicing variation across samples could improve the reproducibility of results and could reduce potential confounding effects in studying co-expressed gene pairs with bulk tissue data. Many of those gene pairs had correlation effects specific to only certain isoforms and would otherwise be undetected. By analyzing gene–gene co-expression variation within functional pathways accounting for splicing, we characterized the patterns of the “hub” genes with isoform-specific regulatory effects on multiple other genes. Conclusions We showed that splicing variation of a gene may interact with TE of the gene and affect the TE of co-expressed genes, resulting in substantial tissue-specific inter-sample variability in gene–gene correlation effects. Accounting for TE-splicing interaction effects could reduce potential confounding effects and improve the robustness of estimation when estimating gene–gene correlations from bulk tissue expression data.

Keywords

• Alternative splicing
• Total expression
• Isoform
• Gene–gene correlation
• Sample-level variation