Mutations in Cohesin Subunits STAG2 and SMC3 Are Non-Synonymous in NPM1-Mutated Acute Myeloid Leukemia

Mutations affecting genes encoding cohesin complex subunits (STAG1, STAG2, SMC3, SMC1A, RAD21) occur in ~15% of acute myeloid leukemia (AML) and ~20% of myelodysplastic syndrome (MDS), but the molecular impacts of these mutations and their contributions to leukemia are poorly understood. Clinical analyses have shown that the STAG2 subunit of cohesin is disproportionately mutated in MDS leading to the recent classification of STAG2 mutations as MDS defining lesions. This is intriguing since STAG2 is frequently mutated in non-MDS subtypes of AML like NPM1-mutated. Given its differing prevalence in myeloid leukemias, we hypothesized that STAG2 mutations affect different molecular pathways than mutations in core cohesin subunits (SMC3, SMC1A, and RAD21). Here we test this hypothesis in the setting of NPM1-mutated AML and find that STAG2 and SMC3 mutations have disparate impacts on cohesin’s genomic positioning, its role in transcriptional regulation, and its ability to prevent DNA damage.

Cas9 editing of the human NPM1-mutated OCIAML3 cell line was used to generate SMC3^+/- and STAG2^y/- cells. Immunoblotting for cohesin subunits showed that SMC3^+/- cells had a ~45% decrease in core-subunit cohesin proteins including SMC3. STAG2^y/- cells, by contrast, did not have altered levels of these core subunits but did show increased protein levels of STAG1. This suggested that SMC3 is required for cohesin complex formation but STAG2 is not. Although STAG2 knockout did not affect core cohesin protein levels, it was possible that cohesin’s DNA-binding was altered. To address this, we performed Cleavage Under Target and Tagmentation (CUT&Tag) for the cohesin subunit RAD21 followed by differential binding analysis. Compared to non-targeting gRNA control cells, SMC3^+/- cells had 5,903 loci with significant loss of RAD21 (FDR <0.05), however STAG2^y/- cells had relatively few changes in RAD21 occupancy (343 sites with gain or loss). SMC3^+/- differentially bound regions (DBRs) were unenriched for the CTCF motif and CUT&Tag for CTCF showed that RAD21 DBRs were less likely to have CTCF co-occupancy (p-val = 2.2e^-16).

SMC3^+/- and STAG2^y/- DBRs were more likely to overlap active transcription start sites or enhancers (H3K4me3+ and / or H3K27ac+) leading us to perform peak-to-gene linkage followed by hallmark gene set enrichment analysis (GSEA). When comparing the DBR-linked genes for SMC3^+/- versus SA2^y/- GSEA demonstrated significant enrichments in G2M Checkpoint, MYC Targets, E2F Targets, p53 Pathway, and Apoptosis (p-adj <0.05). To follow this up RNA-sequencing was used to identify 1,498 and 1,906 genes with altered expression in SMC3^+/- and STAG2^y/- cells, respectively. GSEA for RNA-seq identified increased expression of G2M checkpoint, E2F Targets, and MYC Targets in SMC3^+/- cells, but these enrichments were not seen in STAG2^y/-. Instead, there was a positive enrichment for the p53 pathway, DNA-damage Response and Apoptosis. Linear regression demonstrated a statistically significant, albeit modest, correlation between gene expression and RAD21 binding (R² = 0.024 & p-val = 2.6e^-5). This modest correlation likely reflects cohesin’s ability to both positively and negatively regulate the expression of distant genes via loop formation. Proliferation and cell cycle analyses revealed that more STAG2^y/- cells are in G2, but this does not translate to increased proliferation. This data suggests that DNA damage and p53 could be arresting STAG2^y/- cells at the G2M checkpoint. Changes in cell cycle were not observed for SMC3^+/-.

This work demonstrates significant differences between SMC3 and STAG2 mutations in NPM1-mutated AML, where both mutations commonly occur. STAG1 compensation allows STAG2^y/- cells to maintain cohesin protein levels and genomic positioning, however, cohesin’s role in DNA damage repair and/or replication fork stabilization is compromised leading to increased p53 expression. Mutation of SMC3 caused decreases in cohesin at many genomic positions but did not increase transcription of p53 or DNA damage pathways. These differences may have therapeutic importance, especially for the use of PARP-inhibitors to treat cohesin-mutant AML and MDS. We are continuing to investigate these findings using single-cell epigenomic profiling (scCUT&Tag) to determine genome-wide cohesin dynamics and by analyzing Cas9 edited primary human samples as verification.