Tissue-specific regulation of gene expression via unproductive splicing

Abstract

Eukaryotic gene expression is regulated post-transcriptionally by a mechanism called unproductive splicing, in which mRNA is triggered to degrade by the nonsense-mediated decay (NMD) pathway as a result of regulated alternative splicing (AS). Only a few dozen unproductive splicing events (USEs) are currently documented, and many more remain to be identified. Here, we analyzed RNA-seq experiments from the Genotype-Tissue Expression (GTEx) Consortium to identify USEs, in which an increase in the NMD isoform splicing rate is accompanied by tissue-specific down-regulation of the host gene. To characterize RNA-binding proteins (RBPs) that regulate USEs, we superimposed these results with RBP footprinting data and experiments on the response of the transcriptome to the perturbation of expression of a large panel of RBPs. Concordant tissue-specific changes between the expression of RBP and USE splicing rate revealed a high-confidence regulatory network including 27 tissue-specific USEs with strong evidence of RBP binding. Among them, we found previously unknown PTBP1-controlled events in the DCLK2 and IQGAP1 genes, for which we confirmed the regulatory effect using small interfering RNA (siRNA) knockdown experiments in the A549 cell line. In sum, we present a transcriptomic pipeline that allows the identification of tissue-specific USEs, potentially many more than were reported here using stringent filters.

Figure 1.

The workflow. The pipeline is designed to identify USEs and categorize them based on their splicing and expression patterns and evidence of regulation by RBPs. (1) An initial catalog of USEs is extracted from the transcriptome annotation. (2) Significant USEs are selected based on the association between ψ and e_g. (3) Opposite-sign deviations of ψ and e_g are expected in tissue-specific USEs. (4) Opposite-sign deviations of ψ and e_g are expected in RBP perturbation experiments. The mode of regulation (NMD-inhibiting or NMD-promoting) is predicted for each RBP–USE pair. (5) Concordant changes of ψ and e_r are expected in tissue-specifically regulated USEs. (6) Selection of USEs with CLIP support in the gene and local CLIP support.

Figure 2.

Significant validated USEs. (A) Significant USEs are characterized by the z < −5. ψ_H − ψ_L denotes the difference in the medians in the upper and the lower quartiles of the ψ distribution. (B) The distribution of ψ values for selected USEs from (A). The top 25% and the bottom 25% of ψ values are colored in orange and blue, respectively. Box plots represent the distribution of e_g in the two groups. Shown in green are Δe_g values. Asterisks (***) denote statistically discernible differences at the 0.1% significance level.

Figure 3.

Tissue-specifically regulated USEs. (A) A clustering diagram of 27 tissue-specifically regulated USEs with local CLIP support. The clusters are characterized by decreased ψ in the brain (cluster 1), increased ψ in the brain (cluster 2), decreased ψ in blood (cluster 3) and decreased ψ in skeletal muscle and heart (cluster 4). (B) The predicted network of tissue-specifically regulated USEs with local CLIP support. The nodes represent USEs listed in (A). The edges represent NMD-promoting and NMD-inhibiting regulatory connections.

Figure 4.

Examples of novel tissue-specific USEs. (A and B) The results for DCLK2 and IQGAP1 genes, respectively. In each case, boxplots show (left to right) the distribution of ψ, e_g, the expression levels of the regulator (PTBP1) and the protein expression level. Statistically significant deviations from the pooled median are marked with asterisks (***FDR <0.001; **FDR <0.01; *FDR <0.05; NS not significant). The color of an asterisk shows whether the deviation is in the expected (green) direction. Tissues are colored by GTEx color codes (Supplementary Table S3) and sorted in ascending ψ order. The ideograms in each panel show the USE and CLIP peaks of PTBP1. The subpanels on the right show the results of RT-PCR (top) and RT–qPCR (bottom) experiments. ‘PE+’ and ‘PE–’ denote AS isoforms with and without the poison exon, respectively. The lanes are (left to right) non-treated control (con), treatment with siRNA targeting the firefly luciferase gene (luc), treatment with siRNA targeting PTBP1, CHX treatment, CHX and luc treatment, and treatment with CHX and siRNA targeting PTBP1. The code for asterisks is as in the left subpanels.