Elucidating Transcriptional Heterogeneity in Venetoclax Resistant AMLs

Background: Inhibition of the anti-apoptotic protein BCL2 with venetoclax (VEN) has revolutionized clinical care of acute myeloid leukemia (AML). Existing and acquired resistance to VEN and its combination therapies is common and limits the clinical efficacy of VEN-based therapies. While numerous VEN resistance mechanisms have been identified, their inter and intra tumor heterogeneity is poorly understood. Here we use integrative analysis of drug-response and transcriptomic data to elucidate the transcriptional heterogeneity underpinning VEN resistance in AML.

Methods: A gene-expression signature associated with VEN resistance was derived from AML cell-lines in the cancer cell line encyclopedia (CCLE). Non-negative matrix factorization was used to decompose this signature in VEN resistant AML patients in BeatAML (n = 186) to identify 4 transcriptionally distinct clusters/states of VEN resistant patients (VR_C1-4). Drug response patterns were also validated in vitro in AML cell-lines and VEN-exposed relapse/refractory primary AML patient samples.

Results: VR_C1, are characterized by DNMT3A and RAS (NRAS and KRAS) mutations and enrichment for monocytic blasts. Transcriptionally, VR_C1 is characterized by high oxidative phosphorylation (OXPHOS), fatty acid metabolism and PI3K-mTOR signaling indicating improved metabolic fitness. These tumors also show higher sensitivity to CDK inhibitor SNS-032 (SNS), in part through suppression of OXPHOS and MTORC1 signaling. VR_C3 is characterized by TP53 and SRSF2 mutations, enrichment for erythroid blasts. They also have high cytotoxic T lymphocyte infiltration and expression of inflammatory pathways and immune checkpoint and effector genes, indicating an inflamed tumor. VR_C3 tumors also have active JAK-STAT and MAPK signaling, which are upstream of pro-survival pathways and proliferation, consequently these tumors are sensitive to inhibition of JAK signaling. VR_C3-like cell-lines are sensitive to the JAK inhibitor ruxolitinib (RUXO). In a subset of cases RUXO also synergizes with VEN. VR_C2 samples are characterized by NRAS and TP53 mutations, they also show a distinctive suppression of HOX gene expression and transcription factor activity which has been linked to VEN resistance. Finally, VR_C4 samples show an enrichment for DNMT3A and IDH1 mutations. Transcriptionally, they share similarity with VEN sensitive samples, other than high interferon signaling. We validated these trends by projecting VR_C definitions on samples from TCGA (n =173) and BeatAML samples not used in discovery (n = 182). Cell-lines were also used to confirm drug response characteristics in vitro.

To evaluate whether SNS and RUXO can be used to over-come VEN resistance in a clinical setting of therapeutic relapse, we assessed drug responses ex vivo in primary samples from 16 AML patients. 15 of these samples were from relapsed/refractory patients, 14 of whom had a prior history of VEN-therapy. ASXL1, NRAS and TP53 were the most frequently mutated genes in these samples.12/16 were resistant to VEN ex vivo, and SNS or RUX monotherapies showed modest/heterogenous improvements in efficacy. In contrast, while VEN+RUX was effective in n=6/11 (54.5%) patients, VEN+SNS demonstrated broad statistically significant efficacy in n=13/16 (81%) of relapse/refractory patients with p=0.0026 (VEN vs. VEN+SNS), and p=0.0049 (SNS vs. VEN+SNS) respectively.

Finally, we use single cell RNAseq to project VR_C definitions on blast cells of a patient who relapsed on VEN-therapy. Pre-treatment blasts cells existed in both VEN sensitive (41%) and resistant (59%) transcriptional states. Post-treatment VEN-sensitive cells were eliminated (2%), and patient tumor was dominated by VEN-resistant cells (VR_C1: 5% and VR_C2: 93%). Transcriptional analysis indicated that VR_C1-like cells had higher expression of metabolic pathways. VR_C2-like cells in contrast were more proliferative and came to dominate the tumor post therapeutic relapse.

Conclusion: We identified and characterize transcriptionally distinct subset/states of VEN resistant AMLs that can facilitate interrogation of inter and intra tumor heterogeneity associated with VEN resistance. The granularity captured by these states can guide development of new therapeutic approaches and their effective application in the clinical setting.