Combinatorial BCL2 Family Expression in Acute Myeloid Leukemia Stem Cells Predicts Clinical Response to Azacitidine/Venetoclax

Abstract

The BCL2 inhibitor venetoclax (VEN) in combination with azacitidine (5-AZA) is currently transforming acute myeloid leukemia (AML) therapy. However, there is a lack of clinically relevant biomarkers that predict response to 5-AZA/VEN. Here, we integrated transcriptomic, proteomic, functional, and clinical data to identify predictors of 5-AZA/VEN response. Although cultured monocytic AML cells displayed upfront resistance, monocytic differentiation was not clinically predictive in our patient cohort. We identified leukemic stem cells (LSC) as primary targets of 5-AZA/VEN whose elimination determined the therapy outcome. LSCs of 5-AZA/VEN-refractory patients displayed perturbed apoptotic dependencies. We developed and validated a flow cytometry-based "Mediators of apoptosis combinatorial score" (MAC-Score) linking the ratio of protein expression of BCL2, BCL-xL, and MCL1 in LSCs. MAC scoring predicts initial response with a positive predictive value of more than 97% associated with increased event-free survival. In summary, combinatorial levels of BCL2 family members in AML-LSCs are a key denominator of response, and MAC scoring reliably predicts patient response to 5-AZA/VEN.

Significance: Venetoclax/azacitidine treatment has become an alternative to standard chemotherapy for patients with AML. However, prediction of response to treatment is hampered by the lack of clinically useful biomarkers. Here, we present easy-to-implement MAC scoring in LSCs as a novel strategy to predict treatment response and facilitate clinical decision-making. This article is highlighted in the In This Issue feature, p. 1275.

Trial registration: ClinicalTrials.gov NCT04801797 NCT05177731.

Figure 1.

Monocytic characteristics indicate poor response to 5-AZA/VEN in cultured AML cells, but fail to predict clinical response in patients. A, Nineteen AML cell lines classified as primitive (Prim-AML, n = 11) or monocytic (Mono-AML, n = 8) based on CD64 surface expression (Mono-AML: MFI > 3,500, Prim-AML: MFI < 1,000) were treated ex vivo with 1.5 µmol/L of 5-AZA and increasing concentrations of VEN for 72 hours. Representative data of two independent replicates. Mean ± SEM of technical replicates. B, Mononuclear cells of patients with AML (N = 12) were treated ex vivo for 72 hours on a drug matrix with increasing 5-AZA and VEN concentrations. Unsupervised clustering was performed based on viability. Each quadrant represents one well with a specific 5-AZA/VEN combination on the drug matrix. C–F, Fifty-four untreated naive patients with AML treated with 5-AZA/VEN as first-line therapy were retrospectively assessed for risk factors of response/refractoriness to therapy. C, Univariate logistic regression was performed for every parameter. Multivariate logistic regression was performed on parameters with P < 0.1 in the univariate analysis. D, Percentages of CD64⁺ cells in pretreatment AML samples. E, Number of responding or refractory patients associated with differentiation state (based on the percentage of CD64⁺) or ELN classification groups. F, Abundance of different mutations within responding and refractory patients. ORR, objective response rate. Parts of the figure were created with BioRender.com.

Figure 2.

LSC-like cells as defined by functional and transcriptomic parameters are predominantly located in the immature GPR56⁺ but not in the CD64⁺CD11b⁺ mature subpopulation. A, FACS gating strategy for mature, non-LSC, and LSC-like subpopulations. Displayed are AML bulk cells from primitive CD34⁺ (NPM1-wild-type), CD34⁻ (NPM1-mutated), and monocytic (NPM1-mutated) samples. B, Percentages of mature, non-LSC-like, and LSC-like subpopulations among bulk AML cells in 72 diagnostic AML samples sorted by the frequency of the mature population. C, Schematic overview of the experimental setup for xenotransplantation experiments and RNA sequencing (RNA-seq) of FACS-sorted subpopulations. D, Percentage of human leukemic engraftment obtained from mature, non-LSC, and LSC-like subpopulations of 14 AML samples at endpoints in the bone marrow of NSG mice. Each dot represents an individual mouse with the line marking mean engraftment levels. E, Mean percentage of human engraftment per mouse obtained from mature, non-LSC, and LSC-like subpopulations of 14 AML samples at endpoints in the bone marrow of NSG mice. Each dot represents an individual patient with AML with the line marking mean engraftment levels. Friedmann test was used to compare LSC-like with non–LSC-like and mature subpopulations. F, PCA plot of bulk RNA-seq data from LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 14) and Mono-AML (n = 9) annotated based on subpopulation and AML subclass. Each dot represents a population from one AML sample. G, LSC17 score in LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 14) or Mono-AML (n = 9) patient samples. LSC17 score was calculated for each AML sample as the mean expression of the 17 LSC signature genes. H–J, Normalized counts of BCL2 (H), MCL1 (I), and BCL2L1 (J) expression in LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 14) or Mono-AML (n = 9) patient samples. K, Schematic representation of the experimental setup used in L–O to measure intracellular BCL2, MCL1, and BCL-xL protein expression by flow cytometry. L and M, Mean fluorescence intensity (MFI) of BCL2 in (L) AML bulk and (M) LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 11) or Mono-AML (n = 7) patient samples. N and O, MFI of MCL1 in (N) AML bulk and (O) LSC-like, non-LSC, and mature subpopulations from Prim-AML (N = 11) or Mono-AML (N = 7) patient samples. P and Q, Representative tSNE plots of (P) AML26 (Prim-AML) and (Q) AML50 (Mono-AML) showing expression of CD64, CD34, GPR56, MCL1, BCL2, and BCL-xL. AML bulk is defined as mononuclear cells from patients with AML after the exclusion of dead cells, doublets, lymphocytes, and nucleated erythrocytes. Two-way ANOVA with Tukey correction for multiple comparisons test was used to compare groups of four, and Mann–Whitney test was used to compare groups of two unless specified otherwise. Each dot represents an AML patient sample with the line marking the mean unless specified otherwise. Fix, fixation; IC-staining, intracellular staining; Mut, mutated; NL, non-LSC; NSG, NOD.Prkdc^scid.Il2rg^null; Perm, permeabilization; WT, wild-type. Parts of the figure were created with BioRender.com.

Figure 3.

LSC-like and mature subpopulations show distinct dependencies on antiapoptotic proteins and response to 5-AZA/VEN therapy independent of AML subclass. A, Workflow for C–F. Mononuclear cells of diagnostic AML patient samples were stained with surface antibodies, followed by BH3 profiling and quantification of AUC to assess apoptotic susceptibility in bulk and pregated subpopulations. B, Overview of assessed BH3 mimetics and their target proteins. C and D, AUC of VEN mediated cytochrome C release in (C) AML bulk and (D) LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 11) or Mono-AML (n = 7) patient samples. E and F, AUC of MS1 mediated cytochrome C release in (E) AML bulk and (F) LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 11) or Mono-AML (n = 7) patient samples. Each dot represents an individual AML patient sample with the line marking the mean. Mann–Whitney test was used to compare groups of two. Two-way ANOVA with Tukey correction for multiple comparisons test was used to compare groups of four. G, Schematic representation of ex vivo treatment strategy for H–J. Mononuclear cells of diagnostic AML patient samples (N = 18) were treated ex vivo for 24 hours at 1.5 µmol/L 5-AZA and 100 nmol/L VEN. H, Relative viability of LSC-like, non-LSC, and mature subpopulations from Prim-AML (n = 11) or Mono-AML (n = 7) patient samples was compared using two-way ANOVA with Tukey correction for multiple comparisons test. I and J, Representative tSNE plots of (I) AML26 (Prim-AML) and (J) AML50 (Mono-AML) highlighting expression of CD64 and GPR56 in 5-AZA/VEN-treated and untreated controls. K, Schematic representation of PBMC collection strategy of patients with AML undergoing 5-AZA/VEN therapy. L, Quantification of mature, non-LSC, and LSC-like cell counts from PBMCs relative to pretherapy in the first week of 5-AZA/VEN treatment in 3 patients undergoing therapy initiation. Each line represents an individual patient with each dot on the line representing an individual timepoint of the patient. M, Representative gating strategy highlighting population dynamics of LSC-like and mature AML cell frequencies during the first week of 5-AZA/VEN treatment. All percentages represent the fraction of total live-singlet AML cells. N, Schematic representation of ex vivo treatment strategy for O. Mononuclear cells of 5-AZA/VEN first-line–treated patients with AML (N = 24) were treated ex vivo for 24 hours at 1.5 µmol/L 5-AZA and 100 nmol/L VEN. O, Relative viability of LSC-like, non-LSC, and mature subpopulations was compared using the Mann–Whitney test. Each dot represents an AML patient sample with the line marking the mean unless specified otherwise. Ab, antibody. Parts of the figure were created with BioRender.com.

Figure 4.

Response to 5-AZA/VEN therapy in patients with AML can be predicted by MAC scoring in LSC-like cells. A, Schematic representation of the experimental design for B–G. Mononuclear cells of AML patient samples treated first-line with 5-AZA/VEN from three independently processed cohorts (cohort 1: n = 17, cohort 2: n = 18, and vohort 3: n = 24) were stained with surface antibodies, followed by intracellular staining of three BCL2 family proteins. MAC-Score was calculated based on normalized BCL2 family protein expression levels in LSC-like, non-LSC, mature, and total blast cells. B, Expression of BCL2, MCL1, and BCL-xL in LSC-like cells of patients with AML from cohorts 1 and 2 combined and associated 5-AZA/VEN therapy outcome. Protein expression is shown as MFI z-scores. C, MAC-Score in LSC-like cells of patients with AML from cohorts 1 and 2 combined and association to 5-AZA/VEN therapy outcome. D, Comparison of MAC-Score in LSC-like, non-LSC, mature, and total blast cells of patients with AML from cohorts 1 and 2 and association to 5-AZA/VEN therapy outcome. E, EFS of first-line 5-AZA/VEN AML patients from cohorts 1 and 2 combined with above and below median MAC-Score, BCL2 expression, MCL1 expression, or BCL-xL expression in LSC-like cells. F, MAC-Score in LSC-like cells of patients with AML from cohort 3 and associated 5-AZA/VEN therapy outcome. G, EFS of first-line 5-AZA/VEN AML patients from cohort 3 with above (>0.4) and below (<0.4) median MAC-Score in LSC-like cells. H, Schematic representation of the experimental design for I–J. Mononuclear cells of patients with relapsed/refractory AML who received 5-AZA/VEN as a salvage therapy (cohort 4: n = 23) were stained with surface antibodies, followed by intracellular staining of BCL2 family proteins. I, MAC-Score in LSC-like cells of patients with AML from cohort 4 and associated 5-AZA/VEN therapy outcome. J, EFS of salvage-treated 5-AZA/VEN AML patients from cohort 4 with above (>0.4) and below (<0.4) median MAC-Score determined in LSC-like cells. Each dot represents an AML patient sample with the line marking the mean unless specified otherwise. Mann–Whitney test was used to compare groups and log-rank test to compare therapy durations of AML patients. R/R, relapsed/refractory to standard induction. Parts of the figure were created with BioRender.com.

Figure 5.

MAC-Score in LSC-like cells predicts response to 5-AZA/VEN with high accuracy. A, MAC-Score in LSC-like cells of patients with AML from all first-line 5-AZA/VEN AML patients combined (cohorts 1–3) and association to 5-AZA/VEN therapy outcome. B, Receiver operating characteristic (ROC) curve of MAC-Score and therapy outcomes of all first-line 5-AZA/VEN AML patients combined (cohorts 1–3). C, EFS of all first-line 5-AZA/VEN AML patients combined (cohorts 1–3) with above and below median MAC-Score. D, EFS of all first-line 5-AZA/VEN AML patients who achieved complete remission from combined cohorts (cohorts 1–3) with above (>0.4) and below (<0.4) median MAC-Score. E, Patient characteristics of first-line 5-AZA/VEN cohorts with retrospectively assessed risk factors of refractoriness to therapy. Univariate logistic regression was performed for every parameter. Multivariate logistic regression was performed on parameters with P < 0.15 in the univariate analysis. F, EFS from combined cohorts (cohorts 1–3) with above (>0.4) and below (<0.4) median MAC-Score based in patients with complex karyotype, RUNX1, or NPM1 mutation. G, MAC-Score in LSC-like cells from diagnostic AML patients receiving first-line standard induction chemotherapy and association to therapy outcome. H, Schematic representation of the experimental design for I–J. MAC-Score was calculated based on normalized BCL2 family protein expression levels in LSC-like cells from diagnostic AML patients independent of received therapy (n = 95). I, MAC-Score of patients with AML differentiated by the number of structural variants. J, MAC-Score of patients with AML with different recurrent AML mutations. K, Schematic model outlining the MAC-Score concept for predicting clinical response to 5-AZA/VEN. Each dot represents an AML patient sample with the line marking the mean unless specified otherwise. Mann–Whitney test was used to compare groups and the log-rank test to compare therapy durations of patients with AML. Resp, responder. Parts of the figure were created with BioRender.com.