Computational identification of tissue-specific alternative splicing elements in mouse genes from RNA-Seq

Abstract

Tissue-specific alternative splicing is a key mechanism for generating tissue-specific proteomic diversity in eukaryotes. Splicing regulatory elements (SREs) in pre-mature messenger RNA play a very important role in regulating alternative splicing. In this article, we use mouse RNA-Seq data to determine a positive data set where SREs are over-represented and a reliable negative data set where the same SREs are most likely under-represented for a specific tissue and then employ a powerful discriminative approach to identify SREs. We identified 456 putative splicing enhancers or silencers, of which 221 were predicted to be tissue-specific. Most of our tissue-specific SREs are likely different from constitutive SREs, since only 18% of our exonic splicing enhancers (ESEs) are contained in constitutive RESCUE-ESEs. A relatively small portion (20%) of our SREs is included in tissue-specific SREs in human identified in two recent studies. In the analysis of position distribution of SREs, we found that a dozen of SREs were biased to a specific region. We also identified two very interesting SREs that can function as an enhancer in one tissue but a silencer in another tissue from the same intronic region. These findings provide insight into the mechanism of tissue-specific alternative splicing and give a set of valuable putative SREs for further experimental investigations.

Figure 1.

(a) Schematic flow chart for the identification of tissue-specific SREs. (b) Example of genes with more than two isoforms that were selected or excluded in our analysis. The left one was selected for further analysis since the ASE was either included or skipped in each isoform. The right one was not selected in our analysis because isoform 3 does not strictly skip the ASE.

Figure 2.

z-scores for all hexamers in liver and muscle. (a) z-scores in exons. ESE_C, ESE_L and ESE_M stand for common ESE, liver-specific ESE and muscle-specific ESE, respectively. (b) z-scores in 400 nt intronic sequences upstream of the exons. (c) z-scores in 400 nt intronic sequences downstream of the exons.

Figure 3.

Venn diagram for the number of SREs identified in three studies.

Figure 4.

Position distribution of top eight SREs with smallest P-values in the position bias test. Each bar represents the average number of SREs falling into a region of 10 nt divided by the number of intron or exon sequences used in analysis.

Figure 5.

position distribution comparison for the us' ISS_−LM CUCUCU in upstream introns of the exclusion set and the inclusion set of brain, liver and muscle. Each bar represents the average number of SREs falling into a region of 10 nt normalized by the number of intronic sequences used in analysis.

Figure 6.

Comparison of frequencies of different SREs in different data sets. The first bar in each group stands for the ratio of the frequency in constitutive data to the frequency in the positive data. The second bar stands for the ratio of frequency in the negative data to the frequency in the positive data. Number of SREs used in each comparison is shown in parenthesis.

Figure 7.

Probability density of P-values of the SREs with or without experimental validation. SREs are computationally identified at a significance level of 0.05.