Increased prevalence of alternative splicing in core-cohesin subunit mutated AML as compared to STAG2 mutated AML

Abstract Cohesin is a protein complex that organizes chromatin structure through DNA loop-extrusion. Subunits of the cohesin complex are recurrently mutated in acute myeloid leukemia (AML) and myelodysplastic syndrome (MDS) with core-cohesin subunit mutations (SMC3, SMC1A, RAD21) occurring more frequently in de Novo AML (10% of AML vs 2% of MDS) and STAG2 mutations showing a preference for MDS (4% of AML vs 10% of MDS). Indeed, STAG2 mutations have recently been classified as an MDS defining lesion and are more likely to co-occur with splicing factor (SF) mutations (SF3B1, SRSF2, ZRSR2). These clinical differences in cohesin subunits suggest the existence of subunit-specific molecular pathways underlying myeloid disease which could be targeted by precision medicines. Here we test this hypothesis in the setting of NPM1c+ AML and find that mutation of the core-cohesin subunits RAD21 or SMC3 lead to significantly more mRNA alternative splicing (AS) than STAG2 mutations. To test for molecular differences in core-subunit vs STAG2 mutants, Cas9 editing was performed in the human NPM1c+ OCIAML3 cell line to generate isogenic clones for RAD21+/-, SMC3+/-, and STAG2y/- (two independent clones for each genotype). Cells transfected with a non-targeting gRNA were used as controls. Western blotting for cohesin subunits demonstrated that mutations in core subunits (RAD21+/- & SMC3+/-) caused reciprocal decreases in other core-subunits (e.g. RAD21+/- clones have 50% decreases in both RAD21 and SMC1A). STAG2y/- clones had no protein level changes in core-cohesin subunits but did have increased STAG1. This indicates that RAD21 and SMC3 are required for cohesin complex stability but STAG2 loss can be compensated for by STAG1 at the protein stability level. Next, CUT&Tag was used to determine how cohesin mutations alter the genomic positioning of cohesin. RAD21+/- and SMC3+/- cells had ~16,000 shared genomic loci with significant loss of cohesin (FDR <0.05 and average Log2 Fold Change of -0.8), but STAG2y/- cells had relatively few changes in cohesin occupancy (< 300 sites with increases or decreases), consistent with compensation by STAG1. In core-mutants sites of decreased cohesin were more likely to overlap H3K27ac+ enhancers and promoters but were less likely to overlap sites of CTCF co-occupancy (pval <1e-26). Thus, active chromatin regions are more susceptible to cohesin loss in core-subunit haploinsufficiency, but CTCF can stabilize cohesin to preserve locus-specific recruitment. We performed RNA-seq to assess the impacts of cohesin mutations on transcription. Linear regression in core-mutants demonstrated a significant, albeit modest, correlation between altered cohesin binding and gene expression (R2 = 0.014 & pval = 1.6e-8). This modest correlation is likely due to the complexity of transcriptional regulation and cohesin's ability to both positively and negatively impact transcription via loop formation. Alternative splicing analysis of RNA-seq identified two times the number of significant AS events (ΔPSI > 0.1 & FDR < 0.05) in core mutants as compared to STAG2 mutants (12,374 in RAD21+/-, 12,253 in SMC3+/-, & 6,836 in STAG2y/-). This difference in splicing was linked to altered cohesin occupancy with genes losing cohesin at their promoter being more likely to undergo alternative splicing in core mutants (Odds Ratio = 1.6 & pval < 0.001). Notably, this association was true even if differentially expressed genes were omitted from the analysis. Thus, the association between altered cohesin and splicing is not simply a byproduct of quantitative changes in gene expression. This work demonstrates increased alternative splicing in SMC3 and RAD21 mutations as compared to STAG2 mutations in NPM1c+ AML, where these mutations commonly occur. Core-cohesin mutations are less likely to occur alongside SF mutations. Based on our findings, this may be due to a convergent phenotype of AS dysregulation. Indeed, it has previously been shown that multiple SF mutations are not tolerated in AML. STAG2y/- cells, by contrast, maintain cohesin protein levels and genomic positioning, via STAG1 compensation, resulting in fewer AS events. This may explain why STAG2 and SF co-mutation is enriched in AML and MDS. The AS differences described here may also have therapeutic importance for the use of PARP inhibitors and splicing inhibitors to treat AML. We are following up on the findings presented here using Cas9 editing of primary human HSPCs.