Saturation mutagenesis reveals manifold determinants of exon definition

To illuminate the extent and roles of exonic sequences in the splicing of human RNA transcripts we conducted saturation mutagenesis of a 51 nt internal exon in a 3-exon minigene. All possible single and tandem dinucleotide substitutions were surveyed. Using high throughput genetics, 5560 minigene molecules were assayed for splicing in HEK293 cells. Over 70% of mutations produced substantial (>2X) phenotypes of either increased or decreased splicing. Of all predicted secondary structural elements only a single 15 nt stem-loop, showed a strong correlation with splicing, acting negatively. The in vitro formation of exon-protein complexes between the mutant molecules and proteins associated with spliceosome formation (U2AF35, U2AF65, U1Aa, and U1-70K) correlated with splicing efficiencies, suggesting exon definition as the step affected by most mutations. The measured relative binding affinities of dozens of human RNA binding protein domains as reported in the CISBP-RNA database were found to correlate either positively or negatively with splicing efficiency, more than could fit on the 51 nt test exon simultaneously. Surprisingly, such correlations extended to weak relative protein-sequence affinities. These myriad protein binding correlations point to a dynamic and heterogeneous population of pre-mRNA molecules, each responding to a particular collection of binding proteins. Overall design: In the work described here we have used deep sequencing to examine the splicing phenotypes produced by saturating a model exon with single and double base substitutions at every position We reasoned that variations on the wild type (WT) theme would provide insights into the prevalence and roles of the sequences that reside in a natural exon. We found that over 70% of these mutations produced substantial (>2-fold) effects on splicing, supporting the idea that most of the computationally predicted ESRs do in fact play a role in splicing. The stability of specific secondary structures was also found to correlate with splicing. We found many strong correlations between splicing and the binding of RNA binding proteins (RBPs) to mutant exons in nuclear extracts, implying that most exonic mutations affected exon definition. Finally, we found significant correlations between splicing and reported binding affinity for a large number of human RBPs. The large scale RBP correlations raise the unanticipated possibility that sequences exhibiting relatively weak affinities can nonetheless influence splicing efficiency. We constructed a 3-exon minigene (Ke et al. 2011) comprised of a 51 nt central target exon WT1-5 (exon 5 of the human Wilms' tumor gene) surrounded by terminal exons and intronic sequences derived from the Chinese hamster dhfr gene. Thousands of DNA exons were synthesized to specification by primer-extension of a custom DNA microarray. Minigene libraries that incorporated these oligomers into a central exon in a 3-exon minigene were then prepared (Fig. 1A). Key features of the minigene framework were the provision of strong promoter (CMV) and polyadenylation (SV40) sites and the removal of all start codons from the first exon (Arias et al. 2015) to minimize the chance of nonsense mediated decay (NMD), already unlikely due to the modest size of the exon (Maquat 2004). At each exonic position from 2 to 47 each dinucleotide in the WT1-5 exon was changed to every other possible dinucleotide (Fig. 1B). Positions 1 and 49 to 51 were left unmodified so as not to consider mutations affecting the splice site sequences themselves. The resulting mutant library comprised 555 mutations: 414 double base substitutions (DBSs) and 141 (SBSs) at each position from 2 to 47 and 3 SBSs mutations at position 48. Including the wild type (WT), the total number of distinct molecules numbered 556 (Fig. 1B). Splicing of the WT exon takes place with an efficiency (percent spliced in, psi; presented here as a proportion) of 0.065 (Ke et al. 2011); this low level is the result of depriving WT1-5 of its natural flanking intronic sequences (Zhang et al. 2005c). The modest splicing efficiency of the WT minigene was purposely engineered to allow detection of mutations that increase (up to 16X) as well as decrease (to 0) splicing efficiency.