Repressed Chromatin Drives Leukaemogenesis in Mutant IDH2 Acute Myeloid Leukaemia Via Inhibition of Granulocyte Differentiation and Cell Cycle Progression

Differentiation arrest in acute myeloid leukaemia (AML) results in accumulation of leukaemic progenitors (L-Prog) and bone marrow failure. Mutant isocitrate dehydrogenase enzyme produces d-2-hydroxyglutarate (2HG), which inhibits α-ketoglutarate-dependent dioxygenases, including Jumonji histone demethylases (JKDM) and TET2, but how this causes AML is unclear. Inhibitors of mutant IDH enzyme (mIDHi) restore differentiation in IDH-mutant (mIDH) AML (Amatangelo et al., 2018). Here, we studied transcriptional networks involved using single-cell (SC) gene expression (GEX) and transcription factor (TF) motif accessibility in primary AML treated with the mIDH2 inhibitor enasidenib (ENA) and found that ENA activates cell cycle (CC) and pro-differentiation programmes through increased promoter accessibility of granulocyte-monocyte (GM)-TF targets.

We treated patient L-Prog in vitro with ENA or vehicle, and performed SC RNA-seq (Chromium 10x) in 4 responsive (R), and one non-responsive (NR) patient samples in early, mid and late timepoints. GEX signatures were used to annotate cells according to function (undifferentiated [U], early and late GM [EGM and LGM]) and CC states. In R samples, ENA yielded more dividing late-GM at mid-late timepoints than DMSO (18% vs 6.5%), and more terminally differentiated neutrophils at late timepoints (46% vs 16%). Using SCENIC (Aibar et al., 2017) to assign highly differentially-expressed genes to TF motifs, we computed regulatory networks (regulons, '^R'). Expression of the SP1 ^R was strongly correlated with active proliferation and ENA conditions led to generation of more cells that co-expressed CEBPA ^R or CEBPE ^R with SP1 ^R, emphasising simultaneous engagement of CC and GM programmes. SP1 function is associated with CC and GM differentiation, and silencing of its binding to its targets contributes to AML pathogenesis (Maiques-Diaz et al., 2012). Control and NR samples failed to produce neutrophils, had reduced co-expression of CEBPE/SP1 ^R and yielded more poorly differentiated cells expressing GATA2 ^R.

At the individual gene level, ENA stimulated downregulation of GATA2, GFI1B, IKZF1/2, and RUNX3 together with upregulation of immediate early genes which respond to cytokine and mitogenic stimuli (EGR1, IER2, AP-1) in early-mid phase. Later there is upregulation of CEBP TFs and effector genes FUT4, ELANE, AZU1 and PRTN3. Interestingly, expression of some GM-TFs (RUNX1, SPI1/PU.1, GFI1) was similar between ENA and DMSO, indicating that gene expression alone was insufficient for GM differentiation.

Given the effects of 2-HG on JKDM, we assessed chromatin accessibility and TF binding using SC ATAC-seq. Overall, we had 25% of differentially accessible (DA) peaks, from which 75% were more accessible in ENA than in DMSO. ENA DA peaks were highly enriched in promoters. Using ArchR (Granja et al., 2021), we clustered cells and used ELANE expression levels to compute trajectories in parallel with SC RNA-seq data. ENA peaks were sequentially enriched for CBF/RUNX and GATA families, followed by AP-1 (JUN/FOS) and EGR/CEBP/KLF motifs. Footprinting analysis showed sequential decrease and increase of TF binding for GATA2 and CEBPA/E respectively during ENA-induced differentiation. Although it did not cause higher expression of SPI1/PU.1, ENA induced increased accessibility of its target binding sites at promoters, which included CEBPA/E and GM effectors (MPO, FUT4, PRTN3). This provides a novel mechanism by which ENA induces differentiation of L-prog.

Regulatory network analysis around active, differentially expressed TFs at different phases of ENA-induced differentiation showed a switch from a repressive transcriptional landscape driven by stem-progenitor TFs, to one where AP-1 and GM-TFs activate expression of GM-effector genes. We postulate a model where MYC, E2F8 and EGR1 upregulate the CEBP family in early-mid differentiation. In addition to stimulation of promoter accessibility of TFBS, we find that ENA increases accessibility of cis-regulatory elements of CEBP TFs, adding another mechanism by which differentiation of L-Prog occurs.

Our data on the mechanism of action of ENA suggest that differentiation arrest in IDHm AML involves suppression of CC and GM differentiation programs in a repressive chromatin landscape, likely via inhibition of KDM6A and demethylation of repressive H3K27me3 marks.