Computational prediction of splicing regulatory elements shared by Tetrapoda organisms

Abstract

Background: Auxiliary splicing sequences play an important role in ensuring accurate and efficient splicing by promoting or repressing recognition of authentic splice sites. These cis-acting motifs have been termed splicing enhancers and silencers and are located both in introns and exons. They co-evolved into an intricate splicing code together with additional functional constraints, such as tissue-specific and alternative splicing patterns. We used orthologous exons extracted from the University of California Santa Cruz multiple genome alignments of human and 22 Tetrapoda organisms to predict candidate enhancers and silencers that have reproducible and statistically significant bias towards annotated exonic boundaries.

Results: A total of 2,546 Tetrapoda enhancers and silencers were clustered into 15 putative core motifs based on their Markov properties. Most of these elements have been identified previously, but 118 putative silencers and 260 enhancers (~15%) were novel. Examination of previously published experimental data for the presence of predicted elements showed that their mutations in 21/23 (91.3%) cases altered the splicing pattern as expected. Predicted intronic motifs flanking 3' and 5' splice sites had higher evolutionary conservation than other sequences within intronic flanks and the intronic enhancers were markedly differed between 3' and 5' intronic flanks.

Conclusion: Difference in intronic enhancers supporting 5' and 3' splice sites suggests an independent splicing commitment for neighboring exons. Increased evolutionary conservation for ISEs/ISSs within intronic flanks and effect of modulation of predicted elements on splicing suggest functional significance of found elements in splicing regulation. Most of the elements identified were shown to have direct implications in human splicing and therefore could be useful for building computational splicing models in biomedical research.

Figure 1

Four segments for testing PU values for the predicted elements. Next to the 5'SS and 3'SS segments were chosen to extend 50 nt context, not including ± 30 nt, inside intron from the corresponding exonic boundaries.

Figure 2

Location of genomic regions used for comparative analysis. (A) Statistical significance tests for intronic enhancing/silencing elements surrounding exon. Blue is the null-hypothesis region and red is the region of statistical significance associated with the exon proximity. The red region is specifically located outside the area associated with donor or acceptor signal consensuses [36]. (B) Statistical significance test for the ESEs/ESSs elements supporting the exonic definition. This strategy allows canceling the statistical biases associated with the protein coding potential best characterized by the hexamer statistics [41] and focusing at the essential difference between the exonic flanks, normally enriched with ESEs [42], and the middle section supposedly depleted of such elements. (C) The differential strategy allows detecting enhancing and silencing elements that have substantially different concentration in vicinity of a strong vs. weak SS as defined by the Bayesian SS sensor [36]. The score from the sensor is measured on a discrete scale from 1 to 5, where 1 stands for a weak signal and 5 stands for strong.

Figure 3

Location of the counting regions used for oligonucleotide scoring relative to exonic flanks. All short exons that were not able to accommodate the regions are disregarded. (A) The region arrangement for the counting strategies shown in Figures 2 (A) and (B), where the Skip value is set to 0 nt for the first comparative measurement and 29 nt for the second. The second comparative measurement is necessary to predict active intronic elements that have maximum enhancing/silencing potential at certain optimal distance from the exonic boundary, such as polyG signals [26]. The second measurement also trades the smaller number of longer exons considered for the greater chance of detecting element density discrepancy between the middle of the exons and the flanks. (B) The region arrangement corresponding to differential test strategy shown in Figure 2 (C). (C) The tiling strategy within a region increases the variety of elements sampled in a counting round. Tree different colors used to show which oligo within a region gets sampled in a three consecutive statistical tests (red in the first test, green in the second test, blue in the third test). This strategy reduces chances for multiple sampling of the same oligo conserved at a certain position in closely related organisms.