SF3B1 mutations in different cancer types cause recognition of sterically hindered cryptic splice sites downstream of the branch point

One of the biggest surprises to emerge from the growing catalog of somatic mutations in various cancer types is the recurrent mutation of genes encoding the RNA spliceosome. Recurrent mutations in the highly conserved HEAT 5-9 repeats of splicing factor 3B subunit 1 (SF3B1) have been reported in myelodysplastic syndrome (MDS), chronic lymphocytic leukemia (CLL), breast cancer, uveal melanoma (UM), and pancreatic cancer. Interestingly, SF3B1 mutation is associated with poor prognosis in CLL but improved prognosis in myelodysplasia and UM. Prior studies have shown that mutated SF3B1 CLL samples use canonical 5’ splice sites but cryptic 3’ splice sites. However it is unknown whether SF3B1 mutation causes the same 3’ splicing defects in different cancers. The mechanism by which SF3B1 mutations cause cryptic 3’ splicing and the functional consequences thereof remain unresolved as well.

Here we define the specific sequence requirements needed for cryptic 3’ splicing in tumors with mutated SF3B1. We examined splice junction usage in transcriptome data from SF3B1 mutant and unmutated CLL, UM and BRCA cases and found that SF3B1 mutants use as cryptic acceptors AG dinucleotides ~13-17 bp downstream of the branch point that are likely sterically hindered when SF3B1 is unmutated. The cryptic acceptors are also located >10 bp upstream of nearby canonical acceptors and thus avoid competing with them for splicing. In our genome-wide analysis only 617 AG dinucleotides met these specific sequence requirements and were used as cryptic acceptors. The same cryptic 3’ splicing signature was observed in different cancers but only in samples with mutations in ~10 amino acid hotspots in the SF3B1 HEAT 5-9 repeats.

We assessed the functional impact of SF3B1 mutation and found that the cryptic acceptors are typically used at low frequency in the SF3B1 mutants (<10% relative to the canonical splice site) and are sometimes present in the SF3B1 unmutated tumors but at an even lower frequency (<0.5% relative to the canonical splice site). Nonetheless, we identified three genes previously implicated in cancer TTI1, MAP3K7 and FXYD5 and four others (YIF1A, ORAI2, ZNF91, RP11-1280I22.1) with cryptic acceptors that were consistently preferred to the associated canonical acceptor in the CLL SF3B1 mutant samples.

Our study suggests that cryptic 3’ splicing in SF3B1 mutants results from altered sterics of SF3B1 and other proteins bound at the branch point allowing for the usage of acceptors that are normally hindered and provides a framework for understanding the effects of SF3B1 mutations on the pathophysiology of various cancers.