Venetoclax plus gilteritinib is effective in preclinical models of FLT3-mutant BCL11B-a lineage ambiguous leukemia

Abstract

Aberrant activation of BCL11B (“BCL11B-a”) defines a subtype of lineage ambiguous leukemias with T-lymphoid and myeloid features, co-occurring activating FLT3 mutations, and a stem/progenitor immunophenotype and gene expression profile. As with other lineage ambiguous leukemias, optimal treatment is unclear and there are limited targeted therapeutic options. Here, we investigated the efficacy of BCL-2 and FLT3 inhibition with venetoclax and gilteritinib, respectively, in preclinical models of BCL11B-a leukemia. Despite variation in response to single agent therapies, the combination of venetoclax plus gilteritinib (VenGilt) was highly effective in all models evaluated. BH3 profiling suggested that resistance to venetoclax monotherapy was due to the tumor-intrinsic dependence on additional BCL-2 family proteins prior to drug treatment. Longitudinal single cell RNA-seq analysis identified mitochondrial pathways and a pro-lymphoid gene expression signature as potential drivers of rare cell survival on VenGilt therapy. These data support clinical evaluation of venetoclax in combination with gilteritinib in BCL11B-a lineage ambiguous leukemias.

Introduction

The management of lineage ambiguous leukemias with myeloid and T-lymphoid features is challenging due to lack of targeted therapies. The BCL11B-activated (“BCL11B-a”) subtype comprises up to 40% of early T cell precursor acute lymphoblastic leukemia (ETP-ALL) and T/myeloid mixed phenotype acute leukemia (T/M MPAL)^1,2 and is recognized as a distinct entity³. BCL11B-a leukemia has a gene expression signature consistent with a hematopoietic progenitor cell of origin and a high frequency (80–100%) of co-occurring FLT3 mutations^1,2. Maturation stage is an important determinant of dependence on different BCL-2 family anti-apoptotic proteins as exemplified by the BCL-2-specific BH3 mimetic drug, venetoclax, which shows preferential activity in leukemic cells with an immature phenotype, including ETP-ALL^4,5. The combined use of venetoclax with the FLT3 inhibitor gilteritinib showed preclinical efficacy in acute myeloid leukemia (AML) preclinical models, including FLT3 wild-type tumors^6–10, and a recent phase 1b clinical trial demonstrated strong molecular response rates in FLT3-mutant relapsed/refractory AML¹¹. Given the high rate of activating FLT3 mutations and the immature phenotype of BCL11B-a leukemias, we investigated the efficacy of venetoclax and gilteritinib in preclinical patient-derived xenograft (PDX) models of this subtype.

Study design

For full details of methods, please refer to supplemental Table 1 and Supplemental Information, available online.

Results & Discussion

All BCL11B-a leukemias exhibit significantly higher FLT3 expression compared to non-BCL11B-a MPAL, ETP-ALL, AML or T-ALL, including FLT3 wild-type cases (supplemental Figure 1A). Moreover, among the anti-apoptotic BCL-2 family genes, BCL2 was most significantly elevated in the BCL11B-a group (supplemental Figure 1B), suggesting susceptibility to FLT3 and BCL-2 inhibition. No BCL11B-a leukemia cell lines exist; therefore, we generated or used existing¹² PDX models of BCL11B-a T/M MPAL, AML and ETP-ALL (Figure 1A,B and supplementary Table 2). Single cell (sc)RNA-seq analysis of each model confirmed the predominant progenitor gene expression signature and identified heterogeneity among the models, most notably multiple B cell-like subpopulations within the MPAL PDX (supplemental Figure 1C).

Figure 1. BH3 profiling demonstrates baseline dependence on BCL-2 in BCL11B-a PDX models.

(A) Summary created using Biorender.com of the three BCL11B-a PDX models studied. The BCL11B structural variant and enhancer hijacking partner are indicated, along with co-occurring mutations identified in each diagnostic sample (BETA; BCL11B enhancer tandem amplification). Luc⁺ indicates the PDX cells express luciferase to enable in vivo bioluminescent imaging. (B) Representative flow plots of each PDX’s immunophenotype are shown. cyCD3, cytoplasmic CD3; cyMPO, cytoplasmic myeloperoxidase. See also supplemental Table 2. (C) Schematic of BH3 profiling. Shaded boxes indicate the affinity of each BH3 peptide or mimetic (rows) to bind and inhibit each BCL-2 family protein (columns). The histograms below show representative results of Cytochrome c release following treatment with BAD peptides. As BAD peptide inhibits both BCL-2 and BCL-xL, sensitivity to BCL-2 inhibition is indicated by increased depolarization following treatment with BAD peptide compared to HRK peptide (D,E) Baseline mitochondrial depolarization induced by each treatment in each PDX model. HEL cells are an erythroleukemia cell line with known sensitivity to BCL-xL inhibition¹⁴ and were included as a positive control for HRK peptide activity. The bar plot (D) shows the percentage of PDX blasts in the Cytochrome c-negative gate following peptide exposure while the heatmap in (E) shows delta priming values of the same experiments (delta priming = % Cytochrome c release [treatment] - % Cytochrome c release [DMSO]). Ala, alamethicin (positive control). Data shown for two independent experiments on PDX cells from two different mice. (F) Heatmap of delta priming values for PDX samples treated with BH3 mimetic drugs. Data are representative of 3 replicates. (G) Ex vivo cell viability of each PDX following 48 hours of exposure to the BH3 mimetic drugs ABT-199 (venetoclax), ABT-263 (navitoclax), S63845 (MCL1 inhibitor), and A-1155463 (BCL-xL inhibitor). Data are representative of two independent experiments.

To determine which anti-apoptotic proteins these models depend on for survival, we treated PDX cells with BH3-only peptides and mimetics specific for BCL-2, BCL-xL or MCL-1, followed by assessment of mitochondrial depolarization (BH3 profiling)¹³ (Figure 1C). The erythroleukemia cell line HEL was included as a positive control for BCL-xL sensitivity.¹⁴ All three models were sensitive to BCL-2 inhibition (Figure 1D,E), and the MPAL PDX was sensitive to BCL-xL and MCL-1 inhibition (Figure 1D,E and supplemental Figure 1D). BH3 profiling with BH3 mimetics, including the BCL-2-specific inhibitor ABT-199 (venetoclax), the dual BCL-2/BCL-xL inhibitor ABT-263 (navitoclax), the BCL-xL-specific inhibitor DT2216, and the MCL-1 inhibitor S63845 showed that in contrast to BH3 peptides, all BH3 mimetics induced mitochondrial depolarization in each model, particularly the MCL-1 inhibitor S63845 (Figure 1F). This was supported by ex vivo cell viability screening (Figure 1G). Discrepancy in apoptotic responses between the MS-2 peptide and S63845 may be due to the greater stability and potency of S63845 compared to the MS-2 peptide. These data indicate that BCL11B-a PDX models depend on BCL-2 for survival, with varying dependence on additional anti-apoptotic proteins. We therefore assessed in vivo efficacy of the BCL-2-specific inhibitor venetoclax in combination with the FLT3 inhibitor gilteritinib, hereafter termed VenGilt therapy.

We treated leukemia-bearing PDX mice with each drug alone or in combination for 8 weeks (Figure 2A and supplemental Table 3). The AML and ETP-ALL models were sensitive to venetoclax and gilteritinib monotherapy (Figure 2B–D and supplemental Figure 2A), while the MPAL PDX was resistant to venetoclax and showed a blunted response to gilteritinib. Notably, within 5 weeks of concluding gilteritinib monotherapy in the MPAL PDX, the disease rapidly progressed (Figure 2E,F and supplemental Figure 2B,C).

Figure 2. VenGilt therapy is effective in vivo in BCL11B-a PDX models.

(A) Schematic created using Biorender.com of in vivo drug treatment of BCL11B-a PDX models. See also supplemental Table 3. (B,C) In vivo bioluminescent imaging (B) and quantitation (C) from SJAUL068292 PDX mice. Due to unexpected loss of animals from the venetoclax-only arm, 2 animals from each treatment group were sacrificed at 4 weeks to have matched engraftment and immunophenotype data (see Supplemental Table 3). (D) Leukemic burden in the BM and SPL at the indicated time points of SJTALL005187 PDX mice. (E,F) In vivo bioluminescent imaging (E) and quantitation (F) from SJMPAL011911 PDX mice Circled data points in panel F indicate re-treatment. (G) Immunohistochemistry for human (h)CD7 and hCD33 on BM sections from a representative SJMPAL011911 PDX mouse of each treatment group following 4 weeks of treatment. (H) Leukemic burden in the BM of SJMPAL011911 PDX mice following 4 weeks of treatment. This cohort represents an independent study from that shown in panels E,F. Two representative samples from this study were used in BH3 profiling experiments (panels I-K). (I-K) Results from BH3 profiling of residual leukemic cells from the BM of SJMPAL011911 PDX mice following 4 weeks of treatment. Data shown are for two biological replicates. (I) Percentage of cells showing evidence of mitochondrial depolarization (% Cytochrome c release) following treatment with each indicated peptide; (J,K) Log₂ fold change in delta priming values (% Cytochrome c release [treatment] - % Cytochrome c release [DMSO]) of each treatment group compared to vehicle. Panel K shows these data for the CD33+ and CD33- population separately. o.d., omne die (once daily); IP, immunophenotype; BM, bone marrow; SPL, spleen; H&E, Hematoxylin and eosin.

Despite variation in response to monotherapies, all three models showed excellent responses to VenGilt, with less than 2% residual bone marrow blasts (Figure 2D and supplemental Figure 2A,C), including in animals treated with a high leukemic burden (supplemental Figure 2D–F). Despite the extremely low tumor burden following treatment, leukemic growth gradually increased post-treatment in the MPAL PDX, although re-treatment of one animal showed the disease remained sensitive to VenGilt (Figure 2F). These data demonstrate the high efficacy of VenGilt in multiple BCL11B-a models despite different responses to monotherapy.

To examine potential synergy with conventional chemotherapeutic agents, we treated PDX cells ex vivo with VenGilt in combination with azacitidine, methotrexate, daunorubicin, cytarabine or vincristine (supplemental Figure 3A–F). No combinations exhibited synergy, whereas VenGilt alone achieved the lowest IC₅₀ of all agents tested, aside from daunorubicin in SJMPAL011911 and SJTALL005187. However, these PDX show limited proliferation ex vivo, which may underestimate the activity of cytotoxic agents.

Due to the varied responses of the MPAL PDX to monotherapy, we investigated this model further. We did not observe acquired mutations in genes known to confer resistance to venetoclax or gilteritinib (supplemental Tables 4,5) and the FLT3-ITD mutation remained post-treatment (supplemental Figure 4A). The only significant immunophenotypic shift observed was expansion of the CD33+ population in gilteritinib-treated mice (supplemental Figure 4B,C), although both CD33+ and CD33- populations recapitulated the immunophenotype of the original tumor in subsequent transplants, underscoring this tumor’s phenotypic plasticity (supplemental Figure 4D–F).¹²

We hypothesized that refractoriness to venetoclax in the MPAL model was due to intrinsic dependence of this tumor on multiple BCL-2 family proteins (Figure 1D,E). To test this, we performed BH3 profiling on residual bone marrow-derived leukemic cells after 4 weeks of treatment, a time point at which VenGilt resulted in a bone marrow burden of less than 2% residual leukemic cells (Figure 2G,H). As expected, leukemic cells from venetoclax-treated mice showed increased dependence on BCL-xL and MCL-1 and were in a higher state of apoptotic priming compared to vehicle as assessed by sensitivity to BIM (Figure 2I,J). This higher apoptotic priming was more pronounced in residual cells of VenGilt-treated mice, but not in mice receiving gilteritinib only. Gilteritinib-resistant leukemic cells showed decreased dependence on both BCL-2 and BCL-xL and increased dependence on MCL-1, particularly within the expanded CD33+ population which may be less dependent on BCL-2/BCL-xL pro-survival pathways¹⁵. Although VenGilt therapy eradicated most of the tumor, rare cells surviving therapy showed increased dependence on MCL-1 (Figure 2J,K), suggesting that gilteritinib was not able to overcome these known mechanisms of acquired venetoclax resistance.

To better interrogate molecular responses to each treatment and subsequent survival mechanisms, we performed scRNA-seq on drug-treated MPAL PDX mice at 24 hours, 4 weeks, and 3–5 weeks after cessation of treatment. Clustering identified 12 cell groups (Figure 3A), including those characterized by cell cycle and metabolic pathways (Figure 3B and supplemental Figures 5A–D). VenGilt caused increased cell cycle arrest after 24 hours compared to monotherapy, consistent with prior reports in AML cell lines⁹ (supplemental Figure 5E). One cluster (Cluster 8) harbored a dominant B-lymphocyte progenitor signature (Figure 3B and supplemental Figure 5F) which expanded post-VenGilt treatment (Figure 3C–E and supplemental Figure 5G). Expansion of cells expressing Pro-B gene programs was corroborated by projection onto a normal bone marrow reference map¹⁶ (supplemental Figure 5H). In general, however, we only observed minor changes in cluster composition across the treatment groups and time points.

Figure 3. Single-cell profiling of BCL11B-a acute leukemia PDX reveals transcriptional pathways associated with drug response.

(A) Uniform Manifold Approximation and Projection (UMAP) visualization of all scRNA-seq samples and time points from the SJMPAL011911 PDX in vivo treatment study. Cells are colored according to their Seurat-defined gene expression cluster. (B) Gene set enrichment analysis (GSEA) of differential expression between each cluster from Seurat analysis and all other clusters. [R] indicates Reactome Gene Set; [H] indicates Hallmark Gene Set (C) Stacked bar plot of cluster composition within each sample. (D) GSEA result of a B-lymphocyte gene program¹⁷ that is enriched in Cluster 8 compared to all other clusters. (E) The percentage of cells from each sample mapped to Cluster 8. (F) Selected GSEA results of each treatment compared to the time-point-matched vehicle sample. Gene sets within each category are ranked by the significance of the VenGilt vs. vehicle 24-hour timepoint. See also supplemental Table 6. (G) Schematic showing that residual SJMPAL011911 leukemic cells from one mouse of each treatment group were harvested and transplanted into subsequent recipients (N=3 recipient mice per drug-treated donor sample; see Methods) to measure engraftment kinetics, resulting tumor immunophenotype, and subsequent ex vivo drug sensitivity. (H) Engraftment as measured by total bioluminescent signal at 6- and 10- weeks post-transplant. (I) Bar plots quantifying the percentage of live, single hCD45+ cells positive for each marker shown across all recipient mice. Significance assessed with a two-way ANOVA with multiple testing correction using the Tukey method. *p<0.05 **p<0.01 ***p<0.001. (J) Representative flow cytometry plots of one mouse per recipient group. Data shown are gated on live, single, hCD45+ cells. (K) Ex vivo drug treatment of engrafted tumors from the bone marrow of recipient mice. Data shown represent the percentage of live (Annexin-V and propidium iodine double-negative) hCD45+ cells relative to DMSO treatment alone. Bars are colored according to the in vivo drug treatment of the original tumor. X-axis labels indicate ex vivo drug treatment. Concentrations used for all treatments were: venetoclax (ven), 50 nM; gilteritinib (gilt), 1 μM; S63845 (S6), 50 nM. Cells were treated for 48 hours. Mito., mitochondria; DNA dam., DNA damage; Hypx, hypoxia; Onc; Oncogene

Because gene programs can be shared across clusters, we also performed differential expression and gene set enrichment analysis (GSEA) at the sample level. Consistent with in vivo refractoriness, venetoclax treatment had minimal effects on gene expression (Figure 3F and supplemental Figure 6A). Most expression changes occurred at 24 hours with concordant changes in cell cycle arrest and metabolism induced by gilteritinib and VenGilt. However, DNA damage and MYC-regulated pathways were significantly more repressed following acute VenGilt treatment, indicating that venetoclax can enhance the effect of blocked FLT3 signaling on these FLT3-regulated pathways (Figure 3F and supplemental Table 6). The gene expression changes induced by VenGilt and gilteritinib were no longer concordant at 4 weeks; most notably, residual cells surviving prolonged VenGilt therapy showed significantly stronger downregulation of mitochondrial pathways compared to residual gilteritinib-treated cells (Figure 3F and supplemental Figure 6B). Interestingly, only gilteritinib monotherapy caused significant gene expression changes post-treatment. Thus, despite the profound anti-leukemic effect of VenGilt, rare cells that survive therapy appear to do so without any major shifts in tumor cell identity or biology. Indeed, transplantation of residual leukemic cells from drug-treated mice resulted in engraftment of a tumor of the same immunophenotype (Figure 3G–J) and with similar ex vivo sensitivity to venetoclax, gilteritinib, and the MCL-1 inhibitor S63845 alone or in combination (Figure 3K). We speculate that the plasticity of this tumor is intrinsic to all populations which, encouragingly, appear equally susceptible to VenGilt. The clinical relevance of residual blasts post-treatment requires future investigation.

In summary, BCL11B-a PDX models were highly sensitive to VenGilt treatment, in line with the high rate of FLT3 mutations and high BCL2 expression observed in this subtype. We tested VenGilt in all currently available PDX models of BCL11B-a leukemia and provide experimental evidence of variable responses to single agent treatment which is likely due to underlying differences in dependence on BCL-2 pro-survival pathways. Regardless of this variation, all three models responded exceptionally well to VenGilt, supporting its clinical consideration in patients with the BCL11B-a subtype of lineage ambiguous leukemia.

Key points.

1. BCL11B-a lineage ambiguous leukemia is highly sensitive to the combination of venetoclax plus gilteritinib in preclinical PDX models.

2. Differences in BCL-2 family dependence do not affect response to venetoclax plus gilteritinib, despite variable responses to single agents.

Data accessibility

WES and scRNA-seq data were deposited to the European Genome-Phenome Archive (EGA) under accession number EGAS50000000978. WES data from SJMPAL011911 at diagnosis was previously published.¹²

Data Sharing Statement:

All sequencing data have been deposited to the European Phenome-Genome Archive (EGA) under accession number EGAS50000000978.