Influence of AML Differentiation State in Risk Stratification of Frontline Therapy with Hypomethylating Agents + Venetoclax in AML

Background:

Acute myeloid leukemia (AML) with monocytic differentiation can be associated with resistance to hypomethylating agents plus venetoclax (HMA+VEN). Using a multi-modal analysis in newly diagnosed patients with AML treated with HMA+VEN, we assessed the influence of AML differentiation on clinical outcomes to determine modifying factors of this key variable.

Methods:

This retrospective included two cohorts. The Beat AML cohort (N=228) was used to comprehensively establish the influence of genetic mutations and AML differentiation state (using transcriptional phenotype scores based on RNA sequencing) with ex vivo VEN sensitivity (measured using the area under the curve [AUC] where increasing AUC values correspond with decreased drug response).

An independent clinical cohort (N=144) was next utilized to clinically confirm associations identified within the Beat AML cohort and assess clinical endpoints, including response and overall survival (OS). The differentiation state for this cohort was assessed using unbiased clustering diagnostic flow cytometry data.

Comparisons of categorical and continuous variables utilized the Chi-square or Fisher exact test and Wilcoxon rank-sum or Kruskal-Wallis test as appropriate. Time-to-event endpoints were evaluated utilizing the Kaplan-Meier method with Cox proportional hazards modeling for multivariable analysis (MVA). Allogeneic hematopoietic cell transplant (HCT) was included as a time-dependent covariate.

Results

In the Beat AML cohort (N=228), variability in VEN AUC was observed based on genetics and differentiation. IDH1 (median AUC: 111) and IDH2-mutated AML samples (median AUC: 120) had the lowest VEN AUC and were significantly lower compared with K/NRAS (median AUC: 194) or PTPN11-mutated samples (median AUC: 231). Increasing monocyte-like gene expression positively correlated with VEN AUC (R=0.72, p < 0.001), while progenitor-like gene expression negatively correlated with VEN AUC (R=-0.58, p < 0.001).

In MVA inclusive of clinical factors, genetics, and myeloid cell differentiation state scores, IDH1/2 mutations correlated with increased VEN sensitivity (p: 0.019), while increased monocyte-like gene expression correlated with decreased response to VEN (p < 0.001) ex vivo. Incorporating AML differentiation (specifically, monocyte-like gene expression) in the MVA model of VEN AUC increased the adjusted R² 4-fold (0.54 vs. 0.13) compared to the selected model that considered only clinical and genetic variables.

In the clinical cohort treated with HMA+VEN (N=144), patients (pts) with diagnostic expression profiles compatible with monocytic differentiation (expressing CD33, CD64, and CD11b, and predominantly negative for CD34 and CD117) (N=23) had a numerically higher rate of refractory disease (26% vs. 12%, p: 0.11) and experienced increased 30-day mortality (22% vs. 5%, p: 0.016). After a median follow-up of 38 months (mos; range 0-65 months), pts with monocytic vs. non-monocytic AML experienced modest differences in OS (median 7.8 vs. 10.5 mos, p: 0.31). However, OS was influenced by underlying genetic mutations across differentiation states. Pts with monocytic AML and wild type vs. mutated-NPM1 experienced 12-month OS of 20% vs. 75%, respectively (p: 0.017). Pts with monocytic AML and wild-type NPM1 also had inferior OS compared to pts with non-monocytic AML and wild-type NPM1 (p: 0.026).

To account for both AML differentiation state and underlying genetics, MVA was performed adjusting for HCT, myelomonocytic/monocytic immunophenotype, and mutations associated with VEN sensitivity (IDH1, IDH2, NPM1) or resistance (FLT3-ITD, N/KRAS, PTPN11, TP53). Myelomonocytic (HR: 3.82, 95% CI: 1.91-7.63, p <0.001) and monocytic (HR: 1.95, 95% CI: 1.01-3.79, p: 0.047) immunophenotypes remained independent predictors of OS in addition to genetics. Compared to MVA inclusive of the mPRS groups alone, inclusion of AML differentiation state and additional mutations (IDH1, IDH2, NPM1, PTPN11) improved model performance (c-index 0.76 vs. c-index 0.69).

Conclusions

Response to HMA+VEN is independently modified by AML differentiation state and associated genetics. Prognostic models accounting for AML differentiation state, in addition to key genetic changes, may enable more accurate risk stratification of HMA+VEN outcomes and identify patients most likely to benefit from alternative treatment approaches.