Transcriptomic Signature of Resistance to Decitabine/Venetoclax in AML Patients

There is an urgent need to elucidate the molecular mechanisms and markers of resistance to combination therapy with the BCL-2 inhibitor venetoclax (VEN) and hypomethylating agents (such as Decitabine, DEC) for acute myeloid leukemia (AML). The aim of this study was to systematically identify molecular determinants of therapeutic resistance utilizing transcriptomic analysis by single cell (sc)RNA-seq and bulk RNA-seq analysis of samples from patients enrolled in a phase II clinical trial of VEN-DEC (NCT03404193). For scRNA-seq, we utilized bone marrow (BM) and peripheral blood (PB) samples from 15 newly diagnosed (ND) or relapsed/refractory (RR) AML patients. Patients were classified as responders, including complete remission (CR) and subsequent relapse; and non-responders. Bulk RNA-seq data were obtained from pretreatment samples of separately selected 18 ND patients (16 responders and 2 non-responders) and 16 RR patients (10 responders and 6 non-responders). Data from the ND and RR groups were combined and analyzed for responders and non-responders. We first investigated differentially expressed genes in responders vs non-responders using scRNA-seq analysis of pre-treatment samples of patients on DEC-VEN therapy. Gene expression in three AML clusters at different maturation stages observed in BM (LSC-like cluster, primitive cell-like cluster, and progenitor cell-like cluster) and one AML cluster observed in PB were analyzed. The results showed that in all four AML clusters, nine genes—NRIP1, ERG, MYB, CD34, KMT2A, PROM1, NOTCH1, BCL11A, and BCL2L1—were consistently highly expressed in non-responders compared to responders. Random forest machine learning and Shapley additive explanation (SHAP) value analysis yielded a ranking of the contributions of these candidate genes to treatment response. The SHAP method displays the importance ranking and provides a way to visualize the marginal contribution of each candidate gene to treatment response. The importance ranking obtained from the SHAP model was consistent with the output of the random forest importance function. Logistic regression modeling yielded a prediction equation for treatment resistance based on the expression levels of the top three genes: BCL2L1 encoding the anti-apoptotic BCL2 family protein BCL-xL, NOTCH1 involved in the regulation of LSC quiescence, and the retinoic acid (RA) receptor co-factor NRIP1 functioning to limit RA signaling and tumor cell differentiation (P = 1 / (1 + e^(-6. 00 + 0.46 * NOTCH1 + 0.40 * BCL2L1 + 1.01 * NRIP1); AUC 0.96).

We further performed knockdown (KD) of BCL2L1 and NOTCH1 using lentiviral vectors in MV4-11 and HL60 AML cell lines. Apoptosis induction by VEN and/or DEC was measured by Annexin V flow cytometry. Silencing of BCL2L1 was associated with higher apoptosis induction by VEN and/or DEC treatment compared to control vector-transfected cells in both MV4;11 and HL60, with more pronounced effects in MV4-11 (% increase in specific apoptosis compared to control; MV4;11: VEN 25.5±0.6%, DEC 11.5±0.2%, VEN+DEC 40.5±0.4%, HL60: VEN 6.1±1.6%, DEC 9.1±1.1%, VEN-DEC 13.7±0.8%). In contrast, NOTCH1 KD did not impact apoptosis induction after exposure to DEC and/or VEN compared to control cells in both cell lines. Notably, however, in both MV4;11 and HL 60 cells, VEN-induced apoptosis was enhanced by BCL2L1/NOTCH1 double KD compared to BCL2L1 only KD (MV4;11: 31.4±1.7%, HL60: 21.5±3.0%).

In summary, we propose here a predictive model for VEN-DEC treatment resistance that encompasses BCL2L1, a pro-survival and VEN resistance factor; and NOTCH1 and NRIP1, which are involved in control of the quiescent state in LSCs. Further studies are needed to determine the contribution of NOTCH1 and NRIP1 to proliferation and differentiation of AML LSCs. While these putative resistance markers need further validation in larger studies, we hypothesize that combined inhibition of BCL-xL and of Notch1 signaling may induce apoptosis and eradiate VEN-resistant quiescent LSCs.