AML cell lines (Molm13, MV4-11, and OCI-AML3) and primary AML cells (Online Supplementary Table S1) were obtained as described in the Online Supplementary Appendix. MCL-1-overexpressing (OE), MCL-1-knockdown (KD), and acquired venetoclax resistant AML cells were generated as previously described.²⁷ BAX, BAK, or BAX/BAK double KD OCI-AML3 cells using small interfering RNA (siRNA) (control or ON-TARGETplus SMARTpool siRNA for each target, Dharmacon, Chicago, IL) were generated as previously described.³² Human BM-derived MSC were isolated as previously described.³³ Cells, cultured under the conditions previously described³⁴ were treated with venetoclax, AZD5991, AZD4573, IACS-10759 (Institute for Applied Cancer Science, MD Anderson, Houston, TX), BL-8040 (BioLineRx Ltd., Israel), and various combinations without or with MSC (AML-to-MSC=4:1). For AZD4573 treatments, cells were exposed to AZD4573 for 6 hours (h), and then the drug was removed.

Viable cells and apoptosis were assessed as previously described.³⁴ For leukemia cells co-cultured with MSC, CD45⁺ cells were counted. Apoptotic cells were defined as annexin V+ and/or 7-aminoactinomycin D+ (AnnV+/7AAD+) CD45⁺ cells. For patient samples, annexin V positivity was determined in bulk (CD45⁺), CD34⁺CD38⁺/CD38^-, and CD34⁺CD38⁺/CD38^-CD123⁺ cells.

Western blot analysis was conducted as described previously.³⁴ Antibodies used are shown in the Online Supplementary Appendix.

MCL-1 and cell surface CD44 and CXCR4 were also determined by flow cytometry as shown in the Online Supplementary Appendix.

Leukemia cells’ migration and adhesion to MSC were assessed as previously described.³⁵

Mitochondrial respiration was measured using a Seahorse XF extracellular flux analyzer (Agilent Technologies, Inc., Santa Clara, CA) following the manufacturer’s instructions. The oxygen consumption rate (OCR) was expressed as pmol/min/1,000 cells or relative to a control.

Cellular and mitochondrial ROS were determined by flow cytometry after cells were stained with CellROS deep red or MitoSOX red (ThermoFisher Scientific), for 30 minutes (min) at 37°C and expressed as mean fluorescence intensity shift between stained and unstained live cells (AnnV-/DAPI-).

Total glutathione and glutathione disulfide (GSSG) were determined using a glutathione (GSH) assay kit (Cayman Chemical; Ann Arbor, MI) following the manufacturer’s instructions. Reduced GSH was calculated by subtracting the amount of GSSG from total GSH.

Tracing and subsequence metabolite analysis are detailed in the Online Supplementary Appendix.

Mouse care and experiments were performed in accordance with Institution Animal Care and Use Committee approved protocols. See the Online Supplementary Appendix for detailed experimental procedures.

Cells were stained with metal-tagged antibodies (Online Supplementary Table S2) against cell surface markers, barcoded, pooled, then stained with metal-tagged antibodies against intracellular proteins and subjected to cytometry by time of flight (CyTOF) analysis as described previously.^34,36,37 Briefly, viable single cells (cisplatin-low) were gated with FlowJo (software v10.7, FlowJo LLC) and exported as flow cytometry standard (FCS) data for subsequent analysis in Cytofkit.³⁸ Cell populations identified and embedded by RPhenoGraph in the “Cytofkit_analyzedFCS” files were gated in FlowJo to quantify marker expression. ArcSinh-transformed counts for each protein expression in desired cell populations were visualized with heat maps.

Cell line experiments were performed in triplicates. Results were expressed as the mean ± standard error of the mean (SEM). The Student t-text was used to assess differences between groups; P-values ≤0.05 were considered statistically significant. The combination index (CI),³⁹ determined using Calcusyn software, was expressed as mean CI values at effective dose (ED)50, ED75, and ED90. A CI<1 was considered synergistic; CI=1, additive; and Ci>1, antagonistic. EC50 and half maximal inhibitory concentration (IC50) values were calculated using Calcusyn. Mouse survival was estimated using the Kaplan-Meier method. Survival data were analyzed using the log-rank test.

MCL-1-KD or -OE did not alter viability of AML cell line Molm13, MV4-11, or OCI-AML3 in vitro (Online Supplementary Figure S1A). However, NSGS mice harboring MCL-1-OE Molm13 cells exhibited higher leukemia burden and succumbed sooner to AML, while mice harboring MCL-1-KD OCI-AML3 cells had decreased leukemia burden and survived longer compared with their respective controls (Online Supplementary Figure S1B), suggesting that MCL-1 levels affect leukemia expansion. In order to determine whether this effect requires the modulation of bioenergetic/ metabolic activity, we assayed mitochondrial respiration and found that MCL-1-OE increased, whereas MCL-1-KD or inhibition with AZD5991 (at doses not affecting cell viability) decreased, OCR and ATP production in OCI-AML3 cells (Figure 1A). Compared with parental MV4-11 (MV4-11P), the venetoclax-resistant MV4-11 (MV4-11R) cells that expressed increased MCL-1²⁷ had higher OCR and ATP production. In both cell types, AZD5991 inhibited OCR and ATP generation (Online Supplementary Figure S1C). In addition, compared with controls, MCL-1-OE OCI-AML3 cells exhibited higher levels of cellular and mitochondrial ROS, lower GSH levels, and lower GSH/GSSG ratios, whereas MCL-1-KD OCI-AML3 cells displayed lower ROS, higher GSH levels, and higher GSH/GSSG ratios (Figure 1B and C).

In order to further explore metabolic functions of MCL- 1, we traced ¹³C incorporation from ¹³C2-1,2-glucose and ¹³C5-glutamine into downstream metabolites⁴⁰ using an MCL-1-KD clone (shC16) with an approximately 65% MCL-1 reduction (Online Supplementary Figure S1D), selected by limiting dilution of MCL-1-KD OCI-AML3 cells and OCI-AML3 cells treated with 50 nM AZD5991 (Online Supplementary Figure S1E). Metabolomic analysis revealed markedly lower overall incorporations of ¹³C from both glucose and glutamine into key tricarboxylic acid (TCA) cycle intermediates including citrate, fumarate, and malate, in MCL-1-KD and AZD5991-treated OCIAML3 cells compared to their respective controls (Figure 1D). An examination of the patterns of fractional ¹³C enrichment from ¹³C5-glutamine into citrate, malate, and fumarate demonstrated increased M+4, but decreased M+3 fractional enrichment, in both AZD5991-treated and MCL-1-KD OCI-AML3 cells (Figure 1E), suggesting decreased glutamine entry into the TCA cycle accompanied by a smaller fraction of glutamine-derived carbon being used for anaplerotic reactions that fuel biosynthesis.

In both ¹³C2-1,2-glucose and ¹³C5-glutamine tracing experiments, the total amount of lactate secreted by AZD5991- treated or MCL-1-KD cells was statistically significantly lower than that secreted by their respective controls (Figure 1F). Notably, statistically significant ¹³C enrichment (M>0) in the secreted lactate of both the AZD5991-treated and MCL-1-KD OCI-AML3 cells, was observed from ¹³C2-1,2-glucose, but not ¹³C5-glutamine (Figure 1G). Glycolysis of ¹³C2-1,2-glucose generates lactate with two ¹³C labels (M+2 lactate). Thus, a reduction in M+2 lactate secretion suggests MCL-1 inhibition decreases glycolytic flux.

Next, we assessed whether the decrease in glycolytic flux was accompanied by a decrease in flux through the oxidative branch of the pentose phosphate pathway (oxPPP), a major pathway involved in the generation of the reducing equivalent NADPH that is used for ROS neutralization. When ¹³C2-1,2-glucose is metabolized through the oxPPP and re-enters glycolysis via the nonoxPPP, the elimination of one ¹³C carbon through decarboxylation yields lactate containing only one ¹³C. The relative oxPPP flux, which we calculated by multiplying the ratio of [M+1]/([M+1]+[M+2]) from extracellular lactate with lactate secretion rate during ¹³C2-1,2-glucose tracing, was lower in both AZD5991-treated and MCL- 1-KD OCI-AML3 cells than in their respective controls (Figure 1H).

6-Phosphogluconic acid levels were also greatly decreased under MCL-1 inhibition (Figure 1H), in accordance with oxPPP modulation.⁴¹ Consistent with the aforementioned extracellular flux assay results, AZD5991- treated, and MCL-1-KD OCI-AML3 cells, exhibited decreased ATP levels (Figure 1I). These findings indicated that MCL-1 inhibition downregulates key pathways that support cellular energetics in AML cells as summarized in Figure 1J. Furthermore, MCL-1 modulated the level of c- MYC in OCI-AML3 cells (Online Supplementary Figure S1F), which is a well-known metabolic regulator in malignant cells.

Since MSC in the BM microenvironment upregulate MCL-1, we investigated whether MCL-1 had a role in leukemia-stroma interactions. MCL-1-OE OCI-AML3 cells exhibited increased surface expression of CXCR4 and CD44, both of which participate in leukemia-MSC interactions and drug resistance, and increased OCI-AML3 migration towards, and adhesion to, MSC (Figure 2A). Conversely, MCL-1-KD OCI-AML3 exhibited decreased surface expression of CXCR4 and CD44 and decreased OCI-AML3 migration toward, and adhesion to, MSC (Figure 2B). Consistently, inhibition of MCL-1 with AZD5991 decreased OCI-AML3 surface CXCR4 and CD44 expression and interaction with MSC (Figure 2C). Like other MCL-1 inhibitors, AZD5991 increased levels of MCL-1 (Figure 2C). BCL-2 inhibition with venetoclax in OCI-AML3 cells did not appreciably decrease CD44 surface expression, AML cell migration, or adhesion to MSC. However, it decreased CXCR4 surface expression (Figure 2D). These results suggest that MCL-1 enhances surface adhesion molecule expression and promotes leukemiastroma interactions.

Molm13 and MV4-11 cells are relatively sensitive, whereas OCI-AML3 cells are resistant, to venetoclax (Online Supplementary Figure S2A). Acquired venetoclaxresistant AML cell lines have increased MCL-1, but they likely also have altered levels of many other proteins. In order to determine whether MCL-1 played a key role in venetoclax resistance and whether co-targeting BCL-2 and MCL-1 overcomes this resistance, we treated MCL- 1-KD OCI-AML3 and MCL-1-OE Molm13 and MV4-11 cells with venetoclax or AZD5991. Whereas MCL-1-KD greatly sensitized OCI-AML3 cells to venetoclax (Figure 3A), MCL-1-OE Molm13 (Figure 3B) and MV4-11 cells (Online Supplementary Figure S2B) were largely insensitive to the drug. Responses to AZD5991 were similar, though to a lesser degree (Figure 3A and B; Online Supplementary Figure S2B). We then treated OCI-AML3 and MCL-1-OE Molm13 and MV4-11 cells with venetoclax plus AZD5991. The combination synergistically induced apoptosis and decreased viable cells in the intrinsically venetoclax-resistant OCI-AML3 (Figure 3C), MCL-1-OE Molm13 (Figure 3D), and MV4-11 cells (Online Supplementary Figure S2C). This was also the case in MV4-11R cells (Figure 3E).

We then tested CDK9 inhibitor AZD4573. AZD4573 decreased MCL-1 expression and synergized with venetoclax targeting OCI-AML3 cells (Online Supplementary Figure S2D). This combination was also highly synergistic against MV4-11 and MV4-11R cells (Online Supplementary Figure S2E). In order to demonstrate that MCL-1 is indeed the critical target regulated by CDK9, we treated MCL-1- OE Molm13 and MV4-11 cells with AZD4573 and found that, similar to the MCL-1 inhibitor AZD5991, MCL-1 OE cells were more resistant than control cells to AZD4573. Furthermore, the combination of venetoclax and AZD4573 was highly synergistic in MCL-1-OE Molm13 and MV4-11 cells (Online Supplementary Figure S2F).

We next treated OCI-AML3 cells with venetoclax, AZD5991, or the combination and determined interactions between anti-apoptotic BCL-2, MCL-1, and BCL-XL with the pro-apoptotic activator BIM and effectors BAX and BAK by co-immunoprecipitation to potentially understand the mechanisms of synergy. We observed that BCL- 2 was bound to BIM and BAX (Figure 4A), MCL-1 was bound to BIM and BAK (Figure 4B), and BCL-XL was bound to all three (Figure 4C). Venetoclax treatment decreased BCL-2/BIM and BCL2/BAX interactions, but increased MCL-1/BIM, BCL-XL/BAX, and BCL-XL/BAK interactions. AZD5991 treatment decreased MCL-1/BIM interaction, but increased BCL-2/BIM and MCL-1/BAK interactions. When the two drugs were combined, BCL- 2/BIM and BCL2/BAX interactions were largely diminished, MCL-1/BIM interaction decreased, and the single agent treatment-mediated increases of BCL-XL/BAX, BCL-XL/BAK, and MCL-1/BAK interactions were abrogated (Figure 4A to C), suggesting that the combinatorial cooperative release of activator and effector BCL-2 family proteins likely contributed to the synergy in apoptosis induction (Figure 4D).

In order to determine the contribution of metabolism and leukemia-stroma interactions on the observed synergism, we treated OCI-AML3 cells with venetoclax in the presence of the OxPhos inhibitor IACS-10759⁴² or the CXCR4 inhibitor BL-8040 (4F-benzoyl-TN14003).⁴³ As expected, MSC co-culture protected AML cells from venetoclax-, AZD5991-, or IACS-10759-induced apoptosis (Figure 4E). While suppressing OCI-AML3 migration and adhesion to MSC, BL-8040 did not affect cell viability (Figure 4E and F). Under MSC co-culture, venetoclax activity was minimally enhanced by OxPhos or CXCR4 inhibition, but markedly augmented by combinatorial OxPhos and CXCR4 inhibition. Maximal apoptosis induction of OCI-AML3 was observed by combined MCL-1 and BCL- 2 inhibition (Figure 4E). These results support that cooperative release of pro- from anti-apoptotic BCL-2 family proteins and inhibition of cell metabolism and key stromal microenvironmental mechanisms all contributed to the synergism of co-targeting MCL-1 and BCL-2 in AML cells.

Primary AML cells were cultured with BM-derived MSC and treated with venetoclax, AZD5991, AZD4573, venetoclax plus AZD5991, or venetoclax plus AZD4573. Apoptosis and viable cell counts of AML blast cells and CD34⁺ stem/progenitor cells were assessed and expressed as EC50 (Figure 5A) or IC50 (Online Supplementary Figure S3) for each agent used alone or in combination, as indicated. The patient characteristics, including mutations, venetoclax treatment status, and cytogenetics/risk categories, are shown in Figure 5B and Online Supplementary Table S1. Responses to venetoclax, AZD5991, or AZD4573 varied across the patient samples. Compared with the single-agent treatments, the venetoclax plus AZD5991 and venetoclax plus AZD4573 combinations were markedly more effective in inducing apoptosis and decreasing viable cell numbers (markedly lower EC50 and IC50 values, respectively) in not only blasts, but also CD34⁺, CD34⁺CD38⁺/CD34⁺CD38^- and CD34⁺CD38⁺CD123⁺/CD34⁺CD38^-CD123⁺ stem/progenitor cells in all primary AML samples, including those from venetoclax-resistant/relapsed patients (Figure 5A and B; Online Supplementary Figure S3; Online Supplementary Table S1) regardless of mutational status and cytogenetics.

Among samples 7, 8, 11, and 12 from venetoclax or venetoclax/decitabine resistant/relapsed patients, three (7, 8, and 12) had relatively high venetoclax EC50 values. Samples 3 and 10 that exhibited relatively high venetoclax EC50 values were from resistant/relapsed patients who received venetoclax after sampling and were resistant or relapsed (Figure 5A and B; Online Supplementary Table S1), suggesting that our in vitro system mirrored clinical responses. Patient samples with high venetoclax EC50 values tended to have high AZD5991 EC50 values, suggesting that AZD5991 monotherapy would be less effective in venetoclax-resistant patients. However, more samples are needed to confirm our conclusion. We observed that samples with WT1 (3 and 12) or BCORL1 (4 and 7) mutations were the most resistant to AZD5991 and AZD4573, respectively (Figure 5A and B). Experiments with primary AML cells treated without MSC co-culture yielded similar results (Online Supplementary Figure S4), but these cells were generally more sensitive to the treatments supporting the protective role of MSC. The CI values for the combinations for each AML cell population are shown in the Online Supplementary Table S3.

In order to determine whether co-inhibition of BCL-2 and MCL-1 could overcome venetoclax resistance in vivo, we developed an AML PDX model using cells from a patient who initially responded to venetoclax/decitabine therapy, but relapsed after three cycles (Figure 6A; Online Supplementary Table S1). PDX-bearing mice were treated with venetoclax, AZD5991, AZD4573, venetoclax plus AZD5991, or venetoclax plus AZD4573 (Figure 6A). All treatments statistically significantly decreased circulating blasts compared to controls (P≤0.0001, treatment day [d] 18). The venetoclax plus AZD5991 or venetoclax plus AZD4573 group had statistically significantly fewer circulating blasts than the venetoclax (P<0.01) and AZD5991 (P<0.0001) groups or the venetoclax (P=0.0001) and AZD4573 (P<0.001) groups, respectively (Figure 6B). Analyses of BM samples (treatment d 25) yielded similar results (Figure 6C). Spleens from AZD4573- or venetoclax plus AZD4573-treated mice had a statistically significantly lower leukemia burden compared to controls, which were similar for spleens from the mice treated with venetoclax plus AZD5991 compared to the untreated or AZD5991- treated groups (treatment d 25, Figure 6D).

Although all the treatments statistically significantly extended survival in the PDX model, venetoclax (54 d, P=0.0035), as expected, and AZD5991 (53 d, P=0.0081) showed minimal effects similar to that presented for the in vitro data. AZD4573, and to greater degrees the combinations, were more effective (AZD4573: 60 d, P=0.0002; venetoclax plus AZD4573: 71 d, P=0.0005; venetoclax plus AZD5991: 76.5 d, P=0.0005 compared to controls 51 d, Figure 6E). The median survival of the venetoclax plus AZD4573 treatment group was statistically significantly longer than that of the venetoclax or AZD4573 group (P=0.0004 for both), and the venetoclax plus AZD5991 group survival was statistically significantly longer than that of venetoclax (P=0.0004) or AZD5991 group (P=0.0012) (Figure 6E). No obvious weight loss was observed during the various treatments. The mice only began to lose weight several days before they succumbed to the disease (Online Supplementary Figure S5). Of the 60 mice, one in AZD5991 group, one in venetoclax plus AZD5991 group, and two in venetoclax plus AZD4573 group died of treatment procedures and were not included in the survival analyses.

In order to elucidate the treatment effects on phenotypically- defined cell populations and on protein expression in these populations, we performed CyTOF analysis on BM samples (treatment d 25). Cells were clustered based on expressions of cell surface markers. While venetoclax had no, and AZD5991 and AZD4573 had minimal, effects in several AML cell populations, venetoclax plus AZD4573 and, to a greater degree, venetoclax plus AZD5991 had profound anti-leukemia activity in all cell populations including AML stem/progenitor cells (Figure 6F).

Compared to the vehicle treated controls, the singleagent treatments increased BCL-2, MCL-1, and c-MYC levels in most of the AML cell populations, whereas the combinations, particularly venetoclax plus AZD5991, decreased c-MYC in all populations, decreased BCL-2 in all but the CD34⁺CD38^- population, decreased MCL-1 in the CD45⁺, CD34⁺CD38⁺CD123⁺, and CD34⁺CD38^- CD123⁺ populations (Figure 7A). Interestingly, AZD5991, AZD4573, and, to a greater degree, venetoclax plus AZD5991 and venetoclax plus AZD4573 greatly decreased cell surface CXCR4 levels in all cell populations. AZD5991 and AZD4573 also decreased CD44 levels in all populations (Figure 7B). These results paralleled our in vitro findings suggesting that the inhibition of MCL-1 decreases CXCR4 and CD44 expression to suppress leukemia-stroma interactions. We also measured several signaling proteins and found that the combination treatments greatly decreased -CATENIN and p-AKT in several AML cell populations (Online Supplementary Figure S6A), particularly in CD45⁺ and CD34⁺CD38^-CD123⁺ cells (Figure 7C). Because mouse BM samples were collected 1, 3, or 2 d after venetoclax, AZD5991, or AZD4573 administration, our analyses may underestimate the treatment effects on proteins/phosphor-proteins in leukemia cells.

We also performed CyTOF analysis on BM cells collected from moribund mice. Compared with those in controls, BCL-2, MCL-1, and c-MYC levels in all leukemia cell populations, with the exception of c-MYC in CD34⁺CD38⁺ cells, were higher in venetoclax, AZD5991, and the combination treatment groups (Figure 7D). Cell signaling protein analysis (Online Supplementary Figure S6B) showed that in all but the AZD4573 treatment CXCR4 was suppressed (Figure 7E), whereas p-AKT was greatly induced especially in the combination treatment groups (Figure 7F). Although p-AKT was not induced in the AZD4573 group, the CXCR4 level recovered compared with that observed in the controls (Figure 7E and F).

In order to determine if MCL-1 regulates AML cell metabolic activity and leukemia-stromal interactions independent of its anti-apoptotic functions, we knocked down BAX, BAK, or both with siRNA in OCI-AML3 cells. Suppression of BAX and BAK levels were confirmed by western blot 48 h after siRNA transfection (Figure 8A). siRNA transfected cells were then treated with AZD5991 (50 nM), a dose that did not affect viable cell count (Online Supplementary Figure S1E). Apoptosis, mitochondrial respiration, and cell migration/adhesion to MSC were determined. As shown in Figure 8B, at the 50 nM dose of AZD5991, control siRNA-transfected cells showed no increase in apoptosis at 1 and 4 h, and only 5% more apoptotic cells over baseline at 24 h. The Bax, Bak, and Bax plus Bak siRNA-transfected cells were more resistant to AZD5991-induced apoptosis, as expected. However, inhibition of MCL-1 demonstrated the identical pattern of inhibition of mitochondrial respiration, detectable at 1 h after AZD5991 treatment, in all siRNA-transfected cells (Figure 8B). Furthermore, although AZD5991 treatment (24 h) induced low levels of apoptosis in control siRNAtransfected, but not in Bax and/or Bak siRNA-transfected OCI-AML3 cells, no apparent differences were observed in OCI-AML3 cell migration and adhesion inhibition to MSC in control, Bax, and/or Bak siRNA-transfected and AZD5991 treated cells (Figure 8C). These results suggest that MCL-1 regulates AML metabolic activity and leukemia-stromal interactions independent of its functions as an apoptosis regulator.

Finally, we treated AML patient samples (n=4) with low dose venetoclax (2.5 nM) or AZD5991 (12.5 nM) that induced low level apoptosis after 24 h (Online Supplementary Figure S7A) and found that AZD5991, but not venetoclax, effectively inhibited cellular and mitochondrial ROS production, reduced CXCR4 and CD44 levels, and diminished cell migration and adhesion to MSC (Online Supplementary Figure S7B to D), further supporting that MCL-1 regulates mitochondrial respiration and leukemia-stromal-interactions in AML.