CDK4/6 inhibition overcomes venetoclax resistance mechanisms with enhanced combination activity in acute myeloid leukemia

Summary

Venetoclax (ven) combined with azacytadine is a widely used therapy for acute myeloid leukemia (AML). However, most patients develop resistance. To identify more effective combinations, we analyze 302 AML patient samples and find ven plus palbociclib (ven+palbo), a cyclin dependent kinase (CDK)4/6 inhibitor, to be highly effective. Ven+palbo shows synergistic activity in AML cell lines and patient-derived xenograft mouse models. Patient samples exhibiting a synergistic response to ven+palbo show downregulation of genes involved in protein synthesis. Genome-wide (CRISPR) screening shows that loss of translational genes uniquely confers sensitivity to ven but not to ven+palbo. AML cells exposed to ven exhibit an adaptive increase of protein synthesis that is overcome by ven+palbo through regulation of translational machinery. Additionally, ven+palbo mitigates resistance mechanisms observed with single-agent ven (BAX loss) and palbo (RB1 loss). Finally, we identify the loss of IKZF1 as a mechanism of resistance to ven+palbo and show that targeting AXL is effective in IKZF1-mutated AML.

Graphical abstract

Highlights

• •Venetoclax plus CDK4/6 inhibitor palbociclib shows enhanced efficacy in AML
• •Targeting cell cycle and protein synthesis pathways underpin ven+CDK4/6i efficacy
• •Protein synthesis increases after ven challenge and is mitigated by combination
• •Resistance to ven (BAX loss) and palbo (RB1 loss) is mitigated with combination

Stewart et al. demonstrate that combining venetoclax with the CDK4/6 inhibitor palbociclib enhances efficacy in acute myeloid leukemia by co-targeting cell cycle and protein synthesis pathways. The combination mitigates venetoclax-induced increases in protein synthesis and overcomes resistance driven by the loss of BAX or RB1.

Introduction

In recent years, initial remission rates for elderly patients with acute myeloid leukemia (AML) have seen improvement due to the combination therapy of the BCL2 inhibitor venetoclax (ven) with hypomethylating agents azacitidine (aza) and decitabine.^1,2 However, drug resistance and disease relapse continue to be major hurdles in the long-term management and overall survival of patients with AML due to the heterogeneity of genetic lesions and tumor cell differentiation states that affect how tumor cells respond to a given drug regimen.^{3,4,5,6,7,8,9,10,11,12} Therefore, understanding the complex genetic and biological factors driving the drug response, both before and after drug treatment, is required to advance therapeutic options for this highly heterogeneous disease.

Previous work from our lab demonstrated that ex vivo drug sensitivity assays using primary AML patient samples recapitulate clinical experience with the frontline therapy ven+aza.³ In addition, a recent clinical trial (VenEx, ClinicalTrials.gov: NCT04267081) reported that ex vivo ven sensitivity was the strongest predictor of treatment response and patient survival.¹³ Using ex vivo drug sensitivity assays of 25 drugs in combination with ven, we identified ven plus the cyclin dependent kinase (CDK)4/6 inhibitor palbociclib (palbo) to be one of the most potent and efficacious combinations.³

Palbo is a US Food and Drug Administration (FDA)-approved drug currently used in the treatment of hormone receptor (HR)-positive, human epidermal growth factor receptor 2 (HER2)-negative advanced or metastatic breast cancer.¹⁴ During cell cycle progression, CDK4/6 binds to cyclin D1, which results in the hyperphosphorylation of RB1. Palbo prevents CDK4/6-cyclin D1 interactions, which block the phosphorylation of RB1, preventing E2F1 release from RB1, leading to cell-cycle arrest at the G1 phase.^15,16 The use of CDK4/6 inhibitors in hematological malignancies has not been studied in detail.

Several mechanisms have been shown to underlie resistance to ven. The most studied mechanisms include mutations in FLT3-ITD, RAS pathway genes, and TP53¹⁷; upregulation of alternative pro-survival BCL2 family members, including MCL1, BCL2L1 (BCLXL), or BCL2A1 (BFL1)^12,18,19; and differentiation of AML cells toward a monocytic or megakaryocytic/erythroid cell state.^{3,4,5,6,7,8,9,10,11,12} Interestingly, our ex vivo screening data showed both palbo and ven single agents to be more effective against more primitive, progenitor-like AML cells. However, when used in combination, ven plus palbo (ven+palbo) had greater efficacy on more differentiated, monocytic AML cells.³ The mechanism by which this shift in cell state efficacy of the combination relative to both single agents is unknown.

Results

Ven, in combination with the CDK4/6 inhibitor palbo, inhibits growth, stalls cell cycle progression, and reduces tumor burden in AML models

Our ex vivo drug sensitivity data from AML patient samples suggested that ven+palbo may be more potent and effective than the current frontline therapy of ven+aza.³ To further examine this combination, OCI-AML2 cells were treated with palbo, ven, or the combination for 5 days, and apoptosis was measured by Annexin V staining. Combination-treated cells showed a significant increase in the number of cells in late apoptosis compared to ven alone (Figure 1A). Analysis of cell cycle progression 5 days post-treatment showed that palbo was effective at stalling AML cells in G0/G1, and this effect was maintained in combination with ven (Figure 1B). Ven+palbo treatment resulted in a significant decrease in the number of cells in the S and G2/M phases of the cell cycle compared to the ven single agent (Figure 1B), suggesting a decrease in cell growth with the combination. To assess cell growth, equal numbers of OCI-AML2 cells were treated with DMSO, single agents, or the combination. DMSO- and palbo-treated cells showed comparable viability throughout treatment (Figure S1A). However, total viable cell counts were greatly diminished with palbo treatment (Figures 1C and S1B). Ven treatment reduced cell viability, with a further decrease when in combination with palbo (Figures 1C and S1B). Collectively, these data show that ven+palbo can stall cell cycle progression, increase apoptosis, and inhibit cell growth more effectively than either single agent alone in AML cells.

Figure 1.

Venetoclax, in combination with the CDK4/6 inhibitor palbociclib, inhibits growth, stalls cell cycle progression, and reduces tumor burden in AML models

(A–C) OCI-AML2 cells assessed after a 5-day treatment with palbociclib (1 μM), venetoclax (200 nM), or the combination (equimolar to single agents) to determine levels of apoptosis (A), cell cycle progression (B), and percentage of viable OCI-AML2 cells/mL of media (C) Data represent the mean ± SD for 3 replicates (∗∗p ≤ 0.01).

(D) Schematic depicting two independent PDX experiments. PDX model 1: evaluation of disease burden after injection of AML patient tumor cells. PDX model 2: survival studies after injection of AML patient tumor cells. The schematic was generated using BioRender.

(E–G) Flow cytometry analysis at time of euthanization for PDX model 1 showing human (h)CD45 blasts in peripheral blood (E), spleen weight of animals (F), and percentage of hCD45 chimerism in spleen tissue (G). Mean values ± SEM are shown unless otherwise stated. Two-tailed Student’s t tests were used for comparisons (∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001).

(H) Survival Kaplan-Meier curves for PDX model 2 (log rank [Mantel-Cox] test, ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001).

We next assessed the efficacy of ven+palbo in vivo using two different patient-derived xenograft (PDX) mouse models (Figure 1D). Clinical and genetic features of patient samples used in these studies are listed in Table S1. For PDX model 1, 4 weeks after tumor cell injection, animals were randomly assigned to control, palbo, ven, or ven+palbo treatment and showed similar percentages of hCD45 chimerism (Figures S1F and S1G). Total circulating human cells assessed at 2 and 4 weeks showed fewer circulating human cells in the ven+palbo arm compared to the control arm (Figures S1C, S1D, S1F, and S1G). After the 4-week treatment, control-treated mice showed a higher average disease burden measured by circulating hCD45+ tumor cells (control = 5.5 × 10³/mm³, palbo = 1.4 × 10³/mm³, ven = 2 × 10³/mm³, and ven+palbo = 0.5 × 10³/mm³; Figure 1E), spleen weight (control = 250 mg, palbo = 145 mg, ven = 154 mg, and ven+palbo = 93 mg; Figure 1F), and percentage of chimerism of the spleen (control = 95%, palbo = 92%, ven = 95%, and ven+palbo = 84%; Figure 1G). Animals treated with single agents showed a significant reduction in disease, while the combination treatment showed an even greater reduction (Figures 1E–1G, S1C, S1D, S1F, and S1G). Total body weight was not affected by any treatment regimen (Figure S1E), suggesting minimal toxicity. Additionally, we saw minimal effects in various blood cell compartments when comparing normal NSGS mice to treated mice, suggesting minimal effects on normal murine hematopoiesis (Figure 1, Figure 2, Figure 3, Figure 4, Figure 5, Figure 6, Figure 7–SM).

Figure 2.

Protein synthesis pathways are downregulated in AML patient samples that show a synergistic response to ven+palbo ex vivo

(A) Ex vivo drug sensitivity results from a cohort of 302 patient samples treated with ven, palbo, and ven+palbo for 3 days using a 7-point dose range curve from 10 nM to 10 μM. The results are shown as the area under the dose response curve (AUC).

(B) Chart depicting grouping of synergy (n = 71, blue) and non-synergy (n = 30, orange) primary patient samples.

(C) Slope chart plotting AUC% values from patient samples within the synergy and non-synergy groups, demonstrating response to the ven single agent vs. ven+palbo.

(D) Odds ratios from correlating clinical and genetic features of patients with the synergy group. Clinical or genetic features that reached significance are shown with a red dot.

(E and F) Pathway enrichment plot for upregulated genes (E) and downregulated genes (F) in the ven+palbo synergy compared to the non-synergy Beat AML patient dataset. Dot size indicates the number of evaluated genes in the Reactome pathway. The significance of pathway enrichment within the differentially expressed genes is shown as a log-transformed, FDR-corrected p value (x axis).

Figure 3.

Genetic targeting of protein synthesis pathways sensitizes AML cells to venetoclax

(A) Volcano plots of genome-wide CRISPR screen. All 480 genes uniquely sensitizing to ven with no evidence of an effect by combination treatment (non-significant) are shown as black dots.

(B–D) Reactome pathways from differential expression analysis (DEA) of the 480 gene set show enrichment in pathways related to cell cycle (B), metabolism of RNA (C), and metabolism of proteins (D). The dot size indicates the number of evaluated genes in the Reactome pathway.

Figure 4.

A combination of ven+palbo leads to changes in protein synthesis rate and translational machinery

(A) Bulk RNA-seq for 560 primary AML samples, showing Pearson correlations between AML cellular state eigengenes (columns) and Reactome Translation pathway eigengenes (rows). Protein synthesis pathways positively correlate (red) with a progenitor-like cell-state signature and negatively correlate (blue) with a monocyte-like cell-state signature.

(B) MTS assays confirming drug responsiveness in parental/venetoclax-sensitive cells (VenS/Par) and venetoclax-resistant cells (VenR) for OCI-AML2 cell lines. Data points denote the mean normalized cell viability ± SD for 3 replicates.

(C) SUnSET assay to assess protein synthesis in OCI-AML2 cells treated with drug for 24 h. Puromycin incorporation into newly synthesized proteins was measured in ven-sensitive (VenS/Par [parental]) and -resistant (VenR) cells. A representative image of n = 4 immunoblots is shown.

(D) Estimation plots of intensity measurements of anti-puromycin signal from 4 separate immunoblot experiments.

(E) Immunoblots of OCI-AML2 cells following 24-h or 5-day drug treatments. Vinculin is used as a loading control.

(F) SUnSET assay immunoblots showing puromycin incorporation in progenitor-like or monocyte-like patient samples after drug treatment.

(G) Flow-cytometry-based SUnSET assay detecting CD64⁺, CD11b⁺, CD33⁺ monocytes. The percentage of puromycin-positive cells reflects active protein synthesis within this monocyte population.

Figure 5.

Combining ven+palbo mitigates single-agent resistance due to clinically observed mutations

(A) Enrichment of individual sgRNAs for RB1, BAX, and IKZF1 shown as fold change over DMSO control following a 21-day exposure to palbo, ven, or ven+palbo in OCI-AML2 Cas9 C6 cells.

(B) Immunoblot showing efficiency of knockdown of RB1, BAX, and IKZF1 proteins in OCI-AML2 cell lines. A cell line expressing an NT sgRNA was used to generate a control cell line. Vinculin was used as a protein loading control. Par, parental; NT, non-targeting.

(C–F) Dose-response curves for OCI-AML2 NT and KO cell lines evaluated for drug sensitivity to palbo, ven, or the combination. Data points denote the mean normalized cell viability ± SD for 3 replicates.

(G) IC₅₀ values derived from dose-response curves of OCI-AML2 cell line drug sensitivity assays shown in (C)–(F). Data represent the mean IC₅₀ ± SD for 3 replicates (∗p ≤ 0.05 and ∗∗p ≤ 0.01 by Student’s t test).

(H–J) Outgrowth of OCI-AML2 Non-targeting (H), OCI-AML2 Bax KO (I), and OCI-AML3 cell lines (J) treated with palbo, aza, and ven single agents, duplicate combinations and the triplet. Total viable cells over a 14-day drug treatment are shown. Data points denote the mean total number of viable cells ± SD for 3 replicates. One-way ANOVA with Tukey’s post-test for multiple comparisons was used for day 7 and day 14 time points as indicated. (∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001)

(K) Immunoblot of apoptotic proteins in OCI-AML2 cells, drug treated for 14 days.

Figure 6.

Loss of IKZF1 leads to increased expression of AXL and is increased in IKZF1-mutated AML patient samples

(A) Volcano plot highlighting genes of interest from RNA-seq in IKZF1-KO OCI-AML2 cells, color coded based on primary known function.

(B) qPCR validation of RNA-seq results showing mean fold change ± SD for 3 replicates.

(C) Immunoblot shows upregulation of AXL protein with loss of IKZF1 in OCI-AML2 cells.

(D) AXL mRNA is overexpressed in AML primary patient samples harboring IKZF1 mutations (n = 9) compared to WT samples (n = 662).

(E) OCI-AML2 IKZF1-KO cells show resistance (red dots) to palbo, ven, and ven+palbo and retain drug sensitivity to several AXL inhibitors (blue dots). Sensitivity is shown as a percentage of the maximum area under the dose response curve (AUC) derived for a 7-point concentration series ranging from 10 μM to 10 nM.

(F) AUC values from ex vivo drug sensitivity assays for 4 primary AML samples, each harboring the IKZF1 hotspot mutation N159S (∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001 by Student’s t test).

Figure 7.

Model schematic

Primitive AML cells have a high basal protein synthesis rate, which increases after exposure of cells to ven; however, since primitive cells already undergo high levels of protein synthesis at baseline, this is not an adaptation available for these cells, rendering them sensitive to the ven single agent. Therefore, the addition of palbo does not provide any added benefit, resulting in non-synergy. In contrast, monocytic AML cells have a lower basal protein synthesis rate. Protein synthesis rates increase after exposure of cells to ven, an adaptation that monocytic cells employ, rendering them resistant to the ven single agent. The addition of palbo therefore provides an added benefit in blocking the ven-induced increase in protein synthesis, leading to synergy with ven+palbo. The model schematic was made using BioRender.

For the second PDX model, we chose a patient sample harboring FLT3-ITD and RAS mutations, two genetic lesions that confer resistance to ven.^20,21,22 In an alternate approach to PDX model 1, treatment began 4 weeks after tumor injection and continued for 4 weeks, followed by monthly monitoring for 12 months to assess survival. The percentages of hCD45 chimerism for individual PDX2 mice in each treatment group were similarly low at the start of treatment, less than 2% of the circulating cells (Figure S1H). Overall survival curves that followed the 1-month treatment period showed that ven-treated mice had survival outcomes indistinguishable from control-treated mice, with a median of 3 months (Figure 1H). Palbo-treated mice had improved survival (p = 0.008 palbo vs. control/ven), with a median of 7 months. The combination treatment group had even longer survival (p = 0.008 combination vs. ven; p = 0.030 combination vs. palbo), with a median of 11 months and one mouse showing no evidence of disease when the study was concluded at 12 months. We observed minimal effects in healthy blood cell compartments in treated mice compared to controls (Figures S1N–S1R). Taken together, these data show that the effectiveness of ven+palbo translates from our ex vivo drug assays into an in vivo disease model.

Protein synthesis pathways are downregulated in AML patient samples that show a synergistic response to ven+palbo

Given the modest effects of the combination on the known functions of each single agent yet the striking survival outcomes observed in our in vivo studies, we hypothesized that additional mechanisms may contribute to ven+palbo efficacy. To investigate this, we analyzed our Beat AML dataset of patient samples treated ex vivo with ven+palbo and their constituent single agents.³ This dataset was expanded to include samples newly collected since our previous study,⁴ resulting in a total cohort of 302 patient samples. Area under the dose response curve (AUC) values were calculated after a 3-day ex vivo drug exposure ranging from 10 μM to 10 nM. We found that ven+palbo produced markedly lower AUC values than either single agent alone (Figure 2A), evidence of highest single agent (HSA) synergy and in agreement with our previous findings.³

We next divided the patient cohort into two groups based on the degree of combination synergy observed in our ex vivo drug testing. Of the 101 patient samples with corresponding RNA sequencing (RNA-seq) data, 70 showed synergistic sensitivity to ven+palbo compared to single agents (“synergy” group), while 31 did not (“non-synergy” group) (Figure 2B) (see STAR Methods for group definitions). Notably, the non-synergy samples tended to be quite sensitive to the ven single agent (Figure 2C), which was one key reason these samples did not further respond when adding palbo for the combination treatment.

To identify genetic or clinical features distinguishing synergy from non-synergy groups, we used logistic regression and found that patient samples with N/KRAS mutations were significantly more likely to exhibit ven+palbo ex vivo synergy (odds ratio [OR] = 9.6, p = 0.035), while FLT3-ITD mutations predicted reduced synergy (OR = 0.2, p = 0.001) (Figure 2D; Table S2). Samples with blasts deemed positive for the monocytic marker HLA-DR were significantly more likely to exhibit synergy (OR = 4.7, p = 0.005). Blast expression of CD14, another common monocytic marker, was more frequent in the synergy group (25% vs. 5% in the non-synergy groups), with a trend toward enrichment (OR = 6.8, p = 0.072). These results support prior ex vivo findings that ven+palbo is more effective in monocytic AML.³

To elucidate the potential pathways underlying ven+palbo-enhanced efficacy, we used RNA-seq results from our synergy and non-synergy groups to perform differential expression analysis (DEA). We then queried the full list of differentially expressed genes for pathway enrichments using the Reactome Pathway Database.²³ Pathways enriched for upregulated genes in the synergy group included interleukin (IL)-10 signaling, neutrophil degranulation, IL-4 and IL-13 signaling, and interferon gamma signaling, among others (Figure 2E). Alternatively, genes downregulated in the synergy group compared to the non-synergy group showed enrichment of several pathways related to various aspects of translation and RNA processing (Figure 2F). Genes involved in protein synthesis, including several elongation initiation factors (EIFs), are largely downregulated in the synergy group (Figure 2G), suggesting a role for protein synthesis in ven+palbo synergy.

Genetic targeting of protein synthesis pathways sensitizes AML cells to ven

To further understand the mechanisms underlying the response to ven that might be contributing to the enhanced efficacy of the combination treatment, we performed a genome-wide CRISPR screen to identify genes and pathways that, upon genetic deletion, uniquely confer sensitivity to ven but not ven+palbo (Figure S2A). An OCI-AML2 Cas9-expressing cell line was generated, and Cas9 functionality was validated by flow cytometry.^24,25 A single clonal line with high Cas9 functionality was chosen for use in the screen (Figure S2B). Drug sensitivity profiling across 193 compounds showed no significant differences compared to bulk Cas9 cells (Figure S2C).

OCI-AML2 Cas9 cells were transduced with a genome-wide CRISPR library targeting 18,010 genes.^24,25 After 7 days of puromycin selection, a day 0 baseline, pre-treatment pellet was collected. Cells were then treated with DMSO, single agents, or ven+palbo for 21 days. Genomic DNA from all samples was processed for barcoded Illumina sequencing. Normalized counts of single guide RNAs (sgRNAs) and read count distributions were consistent among replicates of each treatment (Figure S2D).

We identified 480 genes that uniquely conferred sensitivity to ven with no evidence of an effect from combination treatment, suggesting that loss of these genes leads to sensitization to ven (Figures 3A, S2E, and S2F; Table S4). The absence of sensitization when targeting these genes in the context of the combination suggests that these gene targets could underpin the improved activity of ven+palbo compared to ven alone. Accordingly, this gene set was used for pathway enrichment analysis in Reactome, identifying several enriched pathways (Figure S2G). Many of the most highly enriched pathways were related to the cell cycle (Figure 3B), showing proof of principle of the analytical approach, since blockade of the cell cycle is a primary feature of CDK4/6 inhibition. In addition, we found several enriched pathways related to the metabolism of RNA (Figure 3C) and the metabolism of proteins (Figure 3D), the most significant of which was translation.

Combination of ven and palbo leads to changes in protein synthesis rate and translational machinery

Given the differential enrichment of protein synthesis pathways seen in our ex vivo patient sample synergy vs. non-synergy groups and the CRISPR screening data, we investigated if protein synthesis is affected by ven, palbo, and ven+palbo treatments. As single-agent and combination drug sensitivities can be dependent on AML cell differentiation state, we examined whether expression of translational machinery was correlated with differentiation state in AML cells. Using the Beat AML dataset, we correlated eigengenes for both AML cell-state-specific gene signatures^4,26 and translation pathways²³ (Reactome). The scores of six AML cell states were calculated from bulk RNA-seq from AML patient samples by the eigengene of 30 genes designed to enrich each state as performed previously.^4,26 Similarly, eigengenes were calculated for the Translation Reactome pathway (R-HAS-72766) and all 21 daughter pathways, ensuring that each principal component (PC) had a positive mean rotation to aid interpretation. Correlating these two metrics across 560 AML samples showed a positive correlation between the progenitor-like cell state and all translation pathway eigengenes (Pearson’s correlation minimum = +0.10, median = +0.78; false discovery rate [FDR] maximum = 0.017, median <1 × 10⁻¹¹⁷), and a negative correlation with the monocyte-like state (Pearson’s correlation maximum = −0.26, median = −0.51; FDR maximum <1 × 10⁻⁹; Figures 4A and S3A). This suggests a dramatic difference, with high steady-state expression of translation pathways in more primitive AML cells and universally lower expression of the same pathways in more monocytic AML cells.

Monocytic AML samples with lower protein synthesis also appear most responsive to ven+palbo. Paradoxically, CRISPR data showed that genetic knockout (KO) of protein synthesis genes sensitizes AML cells to ven but not to ven+palbo, suggesting that targeting protein synthesis may drive the improved activity of the combination. To reconcile these findings that AML cells with lower baseline protein synthesis rates might benefit from targeting of protein synthesis, we hypothesized that ven-resistant, monocyte-like AML cells with a low basal translation rate may upregulate protein synthesis rates as an adaptive response to ven and that CDK4/6 inhibition may mitigate this adaptation.

To assess the direct effects of ven, palbo, and ven+palbo on protein synthesis, we performed a surface sensing of translation (SUnSET) assay, a pulse-chase technique that uses puromycin incorporation into elongating peptide chains in live cells to measure relative rates of protein synthesis.^27,28 OCI-AML2 parental cells that are sensitive to ven (VenS/Par) or that were generated to become resistant to ven through long-term culture (VenR) (Figure 4B) were treated with single agents or the combination for 24 h, followed by a 15 min puromycin pulse. DMSO-treated controls showed robust labeling of proteins, with VenR cells displaying lower basal translation than VenS/Par cells (Figures 4C and 4D). Ven increased protein synthesis in VenR cells but not in VenS cells, an effect blocked by ven+palbo. Palbo alone reduced translation in both cell types (Figures 4C and 4D). These findings support our hypothesis that ven-resistant AML cells upregulate translation in response to ven, which is mitigated by CDK4/6 inhibition.

Given that palbo appeared to reduce levels of protein synthesis, we investigated how CDK4/6 inhibition might regulate translational machinery. CDK4/6 inhibition inactivates mTORC1, which promotes translation via the phosphorylation of 4EPB1. This phosphorylation causes 4EBP1 to release eIF4E, enabling its assembly with other initiation factors at the 5′ cap of mRNA to initiate translation.^29,30 We analyzed total and phospho-4EBP1 in OCI-AML2 cells by immunoblot following 24-h and 5-day drug treatment. Ven increased both forms of 4EBP1, while ven+palbo led to a greater reduction than palbo alone (Figures 4E and S3D). These results suggest that ven+palbo reduces mTORC1 activity, consistent with the observed suppression of protein synthesis following ven+palbo treatment. Collectively, these data suggest that ven induces adaptive translation, which is suppressed when combined with palbo, likely through reduced mTORC1/4EBP1 inhibition.

We next investigated if these observations extended to AML patient samples. Using immunophenotyping of samples at the time of collection (Table S3), we evaluated the samples determined to be more primitive or monocytic exhibiting non-synergistic and synergistic ex vivo drug sensitivities to ven+palbo, respectively (Figure S3E). SUnSET assays followed by immunoblot showed similar results observed in cell lines, with the progenitor-like sample exhibiting higher puromycin incorporation than the monocyte-like sample with DMSO treatment (Figure 4F). In the monocytic sample, ven treatment resulted in sustained levels of puro incorporation, which was reduced with palbo and the combination (Figures 4F and S3F). We used flow cytometry to identify primary AML CD64⁺, CD11b⁺ monocytes following treatment with ven+palbo and saw a marked reduction in the population of monocytes that were puromycin positive compared to DMSO-treated monocytes (Figures 4G and S3G; 97.4% DMSO vs. 54.9% ven+palbo), suggesting that combination treatment reduces translation in this cell population.

Combining ven+palbo mitigates ven and palbo single-agent resistance due to clinically observed mutations

Clinical resistance to both ven and palbo has been linked to mutations in genes that directly interact with their respective drug targets. Loss-of-function acquired mutations in BAX, seen in ∼17% of patients following ven-based treatment, led to ven resistance.³¹ These mutations were originally predicted from genome-wide CRISPR screens in ven-treated AML cell lines.^3,25 In our current screen, we identified similar ven-resistance genes, including BAX, BCL2, and PMAIP (NOXA), among others (Figure S4B), giving us confidence that our screen performed as expected. Similarly, clinical resistance to palbo has been associated with the loss of cell-cycle-related genes such as RB1.³² Our CRISPR screen showed that RB1 and other cell cycle genes conferred resistance to palbo (Figures 5A and S4A). Interestingly, while loss of BAX and RB1 conferred resistance to the single agent ven or palbo, respectively, the ven+palbo combination was unaffected by the loss of either gene, suggesting that the presence of both drugs may mitigate these clinically relevant resistance mechanisms that occur in the context of single-agent exposures (Figures 5A and S4C).

To analyze these findings further, we generated KO lines using a single vector containing sgRNA/Cas9 targeting either RB1 or BAX in OCI-AML2 and MOLM13 cells. A non-targeting (NT) sgRNA/Cas9 construct was used as a control. Efficient KO of the targeted protein was confirmed by immunoblot (Figures 5B and S4D). To validate the drug sensitivity of selected CRISPR hits, KO cell lines were treated with single agents or ven+palbo for 3 days using a 7-point dose curve (10 nM–10 μM), and cell viability was assessed. In line with the CRISPR screen results, KO of RB1 led to resistance to palbo in both cell lines (Figures 5C–5G and S4E–S4G) while retaining sensitivity to BCL2 inhibition by ven. Consistent with the CRISPR screen, loss of BAX conferred resistance to ven, while the BAX-deficient cells were sensitive to palbo and displayed further sensitivity to the ven+palbo combination (Figures 5F, 5G, and S4E–S4G).

Given the sensitivity of BAX-KO cells to ven+palbo, we tested whether this effect was unique to palbo or shared among the class of CDK4/6 inhibitors. Three CDK4/6 inhibitors—palbo, ribociclib, and abemaciclib—are FDA approved and currently used in clinical treatment of HR+, HER2− breast cancer.^33,34 Lerociclib, another CDK4/6 inhibitor, was recently tested in a phase 3 clinical trial for breast cancer (LEONARDA-1, ClinicalTrials.gov: NCT05054751).³⁵ Using OCI-AML2 NT and BAX-KO cells, we tested drug sensitivity to CDK4/6 inhibitors alone or combined with ven. Single-agent CDK4/6 inhibitors had minimal effects on cell viability in NT and ven-resistant BAX-KO cells. However, the addition of CDK4/6 inhibitors with ven increased sensitivity in BAX-deficient cells, enhancing cell death across all inhibitors tested (Figures S4H and S4J).

OCI-AML3 cells are highly resistant to ven and harbor a loss-of-function E41Gfs∗33 mutation in exon 3 of BAX that has been identified in patients after ven treatment.³¹ All CDK4/6 inhibitors tested in OCI-AML3 cells, alone or in combination, suppressed cell proliferation, with lower IC₅₀ values compared to ven (Figures 5H, S4I, and S4J). These data suggest that CDK4/6 inhibitors could be a beneficial addition to treatment for patients who exhibit resistance to ven, particularly those with a BAX mutation. To further support this, long-term outgrowth experiments showed that palbo and ven+palbo could prevent the growth of both BAX-KO cells and ven-resistant OCI-AML3 cells over a 2-week treatment time course (Figures 5H–5J). Since BAX mutations have never been observed in patients with newly diagnosed AML and have only been seen to arise in the context of clinical ven-based therapy,³¹ these data suggest that the addition of a CDK4/6 inhibitor to a ven-based regimen could help prevent the outgrowth of BAX-mutant, ven-resistant clones.

Resistance to ven can involve increased tumor dependency on MCL1, which blocks BIM and BAX mitochondrial membrane localization, or altered BCL2 and BAX expression, promoting cell survival.³⁶ To test if these were possible mechanisms by which the addition of palbo to ven may overcome resistance to the ven single agent mediated by loss of BAX, we evaluated the levels of several apoptotic proteins after a 2-week exposure to the drug, reasoning that this duration is more comparable to that of our CRISPR screens and a patient treatment time course. We found that MCL1 protein levels greatly decreased with ven+palbo, while single agents showed slight reductions compared to DMSO controls (Figure 5K). Pro-apoptotic PMAIP expression increased in ven+palbo-treated samples compared to ven alone, which nearly abolished PMAIP expression. BCL2L1 (BCLXL) and BAX levels were unchanged in all conditions. These data suggest that palbo may sensitize cells to ven-mediated apoptosis by upregulating PMAIP1, which binds and degrades MCL1, enabling ven+palbo to bypass BAX-mutation-driven resistance.^37,38

Loss of IKZF1 leads to increased expression of AXL and is increased in IKZF1-mutated AML patient samples

While the ven+palbo combination appeared to mitigate common resistance mechanisms of BAX or RB1 loss, the CRISPR screen showed that loss of IZKF1, a transcription factor involved in myeloid and lymphoid differentiation, conferred resistance to both palbo and the combination (Figures 5A and S4A–S4C). IKZF1 is well established in the pathogenesis of acute lymphoblastic leukemia (ALL)³⁹ and has been identified as having an emerging role in AML.^40,41 In a study of outcomes of relapsed or refractory (R/R) AML after treatment with ven+aza, 30% of mutations gained at the time of R/R disease were in transcription factors, including IKZF1.⁴² To further study this potential liability for ven+palbo, we generated IKZF1-KO lines in the same fashion as for BAX and RB1 previously. Loss of IKZF1 led to resistance to single agents and ven+palbo (Figures 5E, 5G, S4E, and S4F).

IKZF1 mutations occur in approximately 3%–5% of AML cases and are associated with poor treatment responses and reduced overall survival.⁴³ To identify the targetable pathways for this subset of patients, we performed RNA-seq on OCI-AML2 IKZF1-KO cells. DEA showed upregulation of several AML-relevant receptors in IKZF1-KO cells compared to NT control cells (Figure 6A). qPCR validated increased expression of CSF1R, CLEC10A, AXL, and CX3CR (Figure 6B). AXL was of particular interest due to its role in AML and solid tumors, including its effects on proliferation and stem cell maintenance.⁴⁴ AXL overexpression also mediates resistance to targeted therapies, including FLT3, PI3K, and EGFR inhibitors.⁴⁵ Immunoblot confirmed increased AXL protein in IKZF1-KO cells (Figure 6C). Next, we wanted to see if patients who harbored an IKZF1 mutation also showed an increase in AXL expression. Using RNA-seq data from our Beat AML cohort, we found that wild-type (WT) samples (n = 662) had a lower expression level of AXL mRNA (log2-transformed reads per kilobase per million mapped reads [RPKM] = −0.7) compared to mutant samples (n = 9) harboring an IKZF1 mutation (normalized RPKM = 1.6) (p = 0.006) (Figure 6D).

To test the effects of AXL inhibitors ex vivo, we treated NT and IKZF1-KO OCI-AML2 cells for 3 days with a panel of AXL inhibitors alone or in combination with ven and measured cell viability by MTS assay. IKZF1-KO cells showed resistance to palbo, ven, and ven+palbo, but AXL inhibitors maintained their effectiveness (Figure 6E). Lastly, we cultured primary patient samples harboring a hotspot N159S IKZF1 mutation and evaluated ex vivo drug sensitivity. Patient samples exhibited strong resistance to ven yet demonstrated substantial tumor cell killing in response to all AXL inhibitors tested (Figure 6F). These findings suggest that AXL inhibitors, as single agents, may represent an effective strategy to mitigate IKZF1-mediated resistance to ven and ven+palbo.

Discussion

In recent years, ven in combination with aza has become the standard of care for elderly patients with AML unfit for intensive chemotherapy. We identified the combination of the FDA-approved CDK4/6 inhibitor palbo with ven as a more potent and efficacious combination than ven+aza in ex vivo studies.³ The triple drug treatment of ven+aza+palbo has been examined in the AML cell lines THP-1 and KG-1 and resulted in increased apoptosis in vitro and a reduced tumor load in a KG-1 cell-line derived mouse model.⁴⁶ This current work further supports the potential therapeutic benefit of ven+palbo as well as uncovers a previously unknown mechanism of resistance to ven involving the regulation of translation. We propose a model (Figure 7) in which AML cells undergo an adaptive response to increase the protein synthesis rate after exposure to ven. In the context of primitive AML cells with a progenitor-like differentiation state, which already have high basal levels of protein synthesis, there is no capacity for further increasing protein synthesis following ven challenge. Consequently, this adaptation is not available to primitive cells, rendering them sensitive to single-agent ven. In this case, the addition of palbo does not provide added benefit, resulting in non-synergy of ven+palbo. In contrast, monocytic AML cells have a lower basal protein synthesis rate. In these more differentiated AML cells, protein synthesis rates can increase after the exposure of cells to ven, rendering them resistant to ven. The addition of palbo prevents the ven-induced increase in protein synthesis, blocking this adaptive drug resistance mechanism and leading to synergy of the ven+palbo combination. While the precise mechanism underlying the adaptive increase in protein synthesis following ven exposure remains to be fully elucidated, our findings highlight an intriguing aspect of ven-resistance biology that warrants further investigation.

Pharmacological targeting of dysregulated translational activity in cancer has been highlighted in several recent studies. One study found that enzalutamide-resistant PKCλ/ι-deficient prostate cancer cells have increased translation and that exposure to an inhibitor of translation restored enzalutamide sensitivity.⁴⁷ Modulation of protein synthesis pathways and machinery appears to be an emerging mechanism for overcoming resistance to ven as well. Changes in cap-dependent protein synthesis may potentially play a role in drug resistance by enabling cells to produce proteins that promote cell survival.⁴⁸ Targeting the translational machinery itself or its regulators could be an effective approach for overcoming resistance to cancer therapies, including ven, as shown with the eIF4A inhibitor zotatafin.⁴⁹ A recent phase 1 study showed that combining ven with pegcrisantaspase (PegC), an enzyme that depletes asparagine and glutamine, resulted in decreased cap-dependent protein synthesis.⁵⁰ This combination led to reduced levels of MCL1 linked to decreased phosphorylation of translation initiation factors, indicating inhibited cap-dependent translation. In addition, treatment of HEK293T cells and MCF7 breast cancer cells with the CDK4/6 inhibitor abemaciclib inactivates mTORC1,³⁰ a master regulator of protein synthesis, suggesting a potential link between CDK4/6 inhibition and translation, in addition to inhibition of RB1 and the cell cycle.

Interestingly, ven+palbo effectively downregulated MCL1 over a sustained treatment period in our study. Efforts to directly target MCL1 in cancer have been of extreme interest in recent years, but the use of MCL1 inhibitors in clinical settings has been hindered by cardiotoxic effects.^51,52,53 Thus, targeting regulators of MCL1 expression could provide potential therapeutic targets for cancer treatment. Interestingly, MCL1 expression is tightly controlled at the transcriptional level, with key transcription factors regulating its expression during progression through the cell cycle, with the lowest levels observed during the G1 phase, the phase at which palbo-treated cells are stalled (Figure 1B). E2F transcription factors, which control cell cycle progression, have been shown to promote MCL1 expression in the G1/S transition.⁵⁴ CDK4/6 inhibition blocks phosphorylation of RB1, which results in the sequestering of E2F family members by hypophosphorylated RB1, ultimately reducing E2F-mediated transcriptional activity.⁵⁴ Data from Beat AML show that E2F2 overexpression positively correlates with ven resistance (r = 0.41, p = 1.6 × 10⁻¹⁶; Figure S5), suggesting that inhibiting E2F family members may contribute to CDK4/6 inhibitor efficacy in AML, ultimately through the downregulation of MCL1 expression. MCL1 protein expression is also regulated via mTORC1 cap-dependent translation.^29,55 Accordingly, CDK4/6 inhibitors may confer an additional therapeutic advantage by indirectly targeting MCL1 levels through induction of G1-phase arrest, as opposed to approaches that reduce MCL1 solely by inhibiting its translation.

IKZF1 mutations are somewhat rare in AML, occurring in 3%–5% of patients, specifically in patients with therapy-related AML, and are associated with poor treatment response and lower overall survival.⁴³ Our data suggest that this subpopulation would likely not benefit from the ven+palbo combination. We found that loss of IKZF1 leads to upregulation of the AXL receptor, the overexpression of which leads to resistance to various targeted therapies.⁴⁵ We demonstrated that AXL inhibitors retained their effectiveness with loss of IKZF1, suggesting a potential treatment strategy warranting further investigation for this hard-to-treat subpopulation. Future studies will be needed to determine the mechanism that links IKZF1 to AXL expression in AML and whether this translates to ALL, where IKZF1 mutations are much more frequent, ranging from 30% to 80% depending on the subtype of ALL.^40,56

The combination of ven+palbo has not been evaluated in clinical trials for AML, and only one phase 1 trial has tested ven+palbo with letrozole in ER+, BCL2+ breast cancer.⁵⁷ Notably, a recent case study reported on a patient with R/R, therapy-related AML, who relapsed during ven+aza therapy with HLA-DR+ blasts, a feature we identified as predictive of ven+palbo synergy ex vivo. The patient received palbo+ven+aza and achieved morphological, immunophenotypic, and molecular complete remission after one cycle, with no serious complications.⁵⁸ This study highlights that palbo+ven may work synergistically in human patients to improve tumor cell killing and warrants further clinical evaluation, likely as a triplet regimen. In addition, CDK4/6 inhibitors may offer a more favorable toxicity profile compared to other CDK-targeting agents. For example, CDK9 inhibitors (e.g., alvocidib and voruciclib) are being studied in combination with ven, based on the rationale that CDK9 inhibition suppresses MCL1.⁵⁹ However, CDK9 inhibitors have shown off-target effects and toxicity with significant myelosuppression.⁶⁰ Given differences in toxicity and the mechanism of action, CDK4/6 inhibitors may show improved results in clinical trials compared to CDK9 inhibitors. In sum, our findings establish ven+palbo as a promising drug regimen and provide further insight into which subpopulations of patients with AML are most likely to benefit from the combination, specifically patients with monocytic disease, as well as patients lacking IKZF1 mutations or harboring acquired RB1 or BAX mutations.

Limitations of the study

Drug-treated PDX mouse model cohorts consisted of 4–5 mice each. We acknowledge that using larger cohorts with a higher peripheral blood leukemia burden at treatment initiation could yield more robust results. Our PDX mouse models demonstrated a minimal apparent impact of ven+palbo on healthy hematopoiesis. However, additional studies using syngeneic models and/or in phase 1 clinical trials will be needed to fully characterize the toxicity profile of this drug combination, including its potential effects on normal hematopoiesis. Also, further studies are needed to determine how IKZF1 loss contributes to resistance to ven and ven+palbo and to uncover the link between IKZF1 and AXL expression in AML. Lastly, the exact mechanism driving the adaptive increase in protein synthesis after ven exposure is not yet fully understood; however, our findings reveal an aspect of ven-resistance biology that merits further studies.

Resource availability

Lead contact

Requests for further information, resources, and reagents may be directed to and will be fulfilled by the lead contact, Dr. Jeffrey W. Tyner (tynerj@ohsu.edu).

Materials availability

The study did not generate unique reagents.

Data and code availability

• •Raw and processed count data for the cell line CRISPR screens and RNA-seq experiments are available in GEO: GSE291338 and GSE291339, respectively.
• •This study does not report custom computer code.
• •Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request.

Acknowledgments

The authors thank all the patients for the generous use of their samples. We thank the Massively Parallel Sequencing Shared Resource for Illumina sequencing support and the flow cytometry core at the Oregon Health and Science University. This work was supported by the Acquired Resistance to Therapy Network (ARTNet) of the National Institutes of Health (NIH), National Cancer Institute (NCI) grant U54CA224019, NCI award R01CA262758 (J.W.T. and S.E.K.), the Waldman Family Fund for AML Research (J.W.T.), the George Ettelson Endowed Professorship in Acute Myeloid Leukemia Research (J.W.T.), the Mark Foundation for Cancer Research (J.W.T.), the Silver Family Foundation (J.W.T.), and the Marsha and Richard Wright Sr. Family Endowed Professorship of Pediatric Oncology (B.H.C.).

Author contributions

Conceptualization, M.L.S. and J.W.T.; validation, M.L.S., J.G., and A.H.; data analysis, M.L.S., J.G., A.N., A.H., L.S., D.B., B.H.C., and J.W.T.; writing – original draft, M.L.S.; writing – review & editing, M.L.S., J.G., K.W.-S., A.N., A.K., C.A.E., N.L., J.N.S., L.S., D.B., S.K.M., B.H.C., and J.W.T.; visualization, M.L.S.; formal analysis, K.W.-S. and A.K.; methodology, K.W.-S., A.K., and D.B.; mouse studies, I.K., K.T.-K., and B.H.C.; data curation, C.A.E., S.E.K., and N.L.; VenR cell line production, J.N.S.; CRISPR formal analysis and methodology, S.K.M.; supervision, S.K.M. and J.W.T.; resources, J.W.T.; funding acquisition, J.W.T.

Declaration of interests

J.W.T. received research support from Acerta, Agios, Aptose, Array, AstraZeneca, Constellation, Genentech, Gilead, Incyte, Janssen, Kronos, Meryx, Petra, Schrodinger, Seattle Genetics, Syros, Takeda, and Tolero and serves on the advisory boards for Recludix Pharm, AmMax Bio, and Ellipses Pharma.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti-tubulin	Cell Signaling Technology	Cat#3873; RRID:AB_1904178
Rabbit monoclonal anti-Axl	Cell Signaling Technology	Cat#8661; RRID:AB_11217435
Rabbit polyclonal anti-Bax	Cell Signaling Technology	Cat#2772; RRID:AB_10695870
Rabbit monoclonal anti-Bcl-XL	Cell Signaling Technology	Cat#2764; RRID:AB_2228008
Rabbit monoclonal anti-4EBP1	Cell Signaling Technology	Cat#9644; RRID:AB_2097841
Rabbit polyclonal anti-Ikaros	Cell Signaling Technology	Cat#5443; RRID:AB_10691693
Rabbit polyclonal anti-Mcl-1	Cell Signaling Technology	Cat#4572; RRID:AB_2281980
Mouse monoclonal anti-Noxa	Santa Cruz Biotechnology	Cat#sc-515840
Rabbit monoclonal anti-4E-BP1, phospho (Thr37/Thr46)	Cell Signaling Technology	Cat#2855; RRID:AB_560835
Mouse monoclonal anti-puromycin, clone 12D10	Sigma-Aldrich	Cat#MABE343; RRID:AB_2566826
Mouse monoclonal anti-Rb1	Cell Signaling Technology	Cat#9309; RRID:AB_823629
Rabbit monoclonal anti-vinculin	Cell Signaling Technology	Cat#13901; RRID:AB_2728768
anti-human CD33-RY610	BD Biosciences	758223; RRID:AB_3690372
anti-human CD45-Pacific Orange	Exbio	PO-684-T025; RRID:AB_10954107
anti-human CD117-Pacific Blue	Biolegend	375209; RRID:AB_2890811
anti-human CD64-PE	Biolegend	305007; RRID:AB_314491
anti-human CD11b-BUV737	BD Biosciences	568332; RRID:AB_3684188
anti-puromycin Alexa 647	Millipore Sigma	MABE343-AF647
Bacterial and virus strains
One Shot Stbl3 Chemically Competent E.coli	ThermoFisher	C737303
Biological samples
Patient-derived xenografts (PDX)	Oregon Health & Science University (OHSU)	OHSU; IRB #4422
AML Patient Samples	Oregon Health & Science University (OHSU)	OHSU; IRB #4422
Chemicals, peptides, and recombinant proteins
Venetoclax	Selleckchem	S8048
Palbociclib	Med Chem Express	HY-50767
Lerociclib	Med Chem Express	HY-112272A
Abemaciclib	Med Chem Express	Cat#S5716
Ribociclib	Med Chem Express	HY-15777
Azacytidine	Med Chem Express	HY-10586
Bemacentinib	Med Chem Express	HY-15150
Dubermatinib	Med Chem Express	HY-12963
IL-3	Peprotech	200–03
IL-6	Peprotech	200–06
SCF	Peprotech	300–07
FLT3-L	Peprotech	300–19
TPO	Peprotech	300–18
SR1	Stem Cell Technologies	72342
UM729	Stem Cell Technologies	72332
Fixable Viability Stain FVS780	BD Horizon	565388
Critical commercial assays
RNeasy Mini Kit	Qiagen	Cat#74104
High-Capacity cDNA Reverse Transcription Kit	Applied Biosystems	Cat#4368814
TruSeq Stranded mRNA kit	Illumina Inc.	Cat#20020594
BD Cytofix/Cytoperm Fixation/Permeabilzation kit	BD Biosciences	554714
Applied Biosystems Power SYBR Green PCR Master Mix	Thermo Fisher	43-676-59
Mycoplasma PCR Detection Kit	Applied Biological Materials Inc.	Cat#G238
Deposited data
RNA sequencing data- BeatAML	Bottomly et al.4	vizome.org/aml2, dbGaP study ID is 30641 and accession ID is phs001657.v2.p1
Raw and processed data- CRISPR screens	This paper	GEO: GSE291338
Raw and processed data- RNA-seq	This paper	GEO: GSE291339
Experimental models: Cell lines
Human: OCI-AML2	DSMZ	DSMZ no:ACC 99
Human: MOLM-13	DSMZ	DSMZ no:ACC 554
Human: OCI-AML3	DSMZ	DSMZ no:ACC 582
HEK 293T/17	Laboratory of Brian Druker	N/A
Experimental models: Organisms/strains
Mouse: NOD.Cg-PrkdcscidIl2rgtm1WjlTg(CMVIL3,CSF2,KITLG) 1Eav/MloySzJ (NSGS)	Jackson Laboratories	N/A
Oligonucleotides
qPCR oligonucleotides	See Table S3	N/A
Single CRISPR guides	See Table S3	N/A
Recombinant DNA
psPAX2	Addgene	Cat#12260
VSVG	Life Technologies	N/A
pKLV2-U6gRNA5(Empty)-PGKmCherry2AGFP-W	Addgene	Cat#67981
pKLV2-U6gRNA5(gGFP)-PGKmCherry2AGFP-W	Addgene	Cat#67982
lentiCRISPRv2	Addgene	Cat#52961
lentiCas9-Blast	Addgene	Cat#52962
human CRISPR library plasmid DNA	Addgene	Cat#67989
Software and algorithms
Reactome	https://reactome.org	N/A
Fiji	https://imagej.net/software/fiji/	N/A
Biorender	https://www.biorender.com/	N/A
FlowJo	https://www.flowjo.com/	N/A
SpectroFlow	https://cytekbio.com/pages/spectro-flo	N/A

Experimental model and study participant details

Cell lines

Human AML cell lines (OCI-AML2, OCI-AML3, MOLM13) were purchased from DSMZ. AML cell lines were grown in RPMI 1640 (Life Technologies Inc.) supplemented with 20% FBS (Cytiva), 2% L-glutamine (Life Technologies Inc.) and 1% penicillin/streptomycin (Life Technologies Inc.). HEK 293T/17 cells were cultured in DMEM (Life Technologies Inc.) supplemented with 10% FBS (Cytiva), 2% L-glutamine (Life Technologies Inc.) and 1% penicillin/streptomycin (Life Technologies Inc.). VEN-resistant (VR) OCI-AML2 cells were generated by continuous incubation with escalating doses of ven. Viability was monitored by Guava EasyCyte three times per week. Once viability returned to ∼80%, cell sensitivity was remeasured to confirm ven resistance. Cells were maintained by co-culture with 1μM of ven added twice weekly. All cell lines were tested biweekly for mycoplasma using the Mycoplasma PCR Detection Kit (Applied Biological Materials Inc, Cat.#G238) and authenticated by the DNA sequencing core at OHSU using Short Tandem Repeat (STR) profiling.

Patient derived xenograft mouse models

NOD.Cg-Prkdc^scid Il2rg^tm1WjlTg(CMVIL3,CSF2,KITLG) 1Eav/MloySzJ (NSGS) (Jackson Laboratories) female mice were used for primary AML xenograft experiments. All murine studies were performed in accordance with OHSU IACUC (IP00000179), maintained in barrier conditions and monitored through the Department of Comparative Medicine at OHSU. Two independent patient-derived xenograft (PDX) experiments were conducted in this study. In the first experiment (PDX1), primary mononuclear cells (MNCs) from a newly diagnosed AML patient sample (#07–00335, AML M2 with NPM1 mutation and FLT3-ITD) were injected into six-week-old female NSGS mice with 2 × 10ˆ6 cells per animal 24h after receiving 150cGy X-ray irradiation. Engraftment was monitored in peripheral blood using flow cytometry against mouse CD45-erCP-Cy5.5 to human CD45-FITC, human CD33-APC, and human CD3-PE. No human CD3 was detected. One month after tumor cell injection, peripheral blood was assessed for percen human CD45/CD33 expression. At that time mice were divided into 4 treatment groups (n = 4–5 mice per group). Cohorts began treatment as such: Control = vehicles; palbociclib = 25mg/kg/dose daily oral; ven = 25mg/kg/dose daily oral; palbo/ven = same dose. Animals were treated daily for 4 weeks, then euthanized. Drug stocks were prepared as follows: palbo (5mg/mL in 0.5% methylcellulose and PBS),⁶¹ ven (5mg/mL in 60% Phosal 50 propylene glycol, 30% polyethylene glycol 400, 10% ethanol).⁶² Peripheral blood samples were obtained at 2 weeks and at terminal blood draw at 4 weeks. Dosage of 25mg/kg venetoclax was selected based on prior studies demonstrating therapeutic potential, particularly in combination settings^63,64 and were chosen to balance efficacy with minimized toxicity. A similar rationale supports the use of palbociclib at 25mg/kg, as shown by Vijayaraghavan et al.⁶¹ For the second mouse model (PDX2) we assessed survival in a different patient sample (#07–00292, AML M2 with FLT3-ITD and NRAS mutation). Six-week-old female mice were injected with 1 × 10ˆ6 cells per animal 24h after receiving 15cGy X-ray irradiation. One month after injection, peripheral blood chimerism was assessed and cohorts of animals were treated for 4 weeks as described above. After treatment, animals were monitored monthly in peripheral blood using flow cytometry against mouse CD45 to human CD45, human CD33, and human CD3 (same antibodies described above). A survival event was defined as either having ≥20% human CD45 positive cells in the blood, or moribund. Statistical analysis was conducted using GraphPad Prism Version 6.0.

Clinical samples

All clinical specimens used in this study were collected to the OHSU Biorepository with written informed consent from patients according to the protocol approved by the Oregon Health & Science University Institutional Review Board (IRB #00004422). Mononuclear cells were isolated by Ficoll gradient from peripheral blood or bone marrow drawn from patients with acute myeloid leukemia (AML). Mononuclear cells were used immediately in drug sensitivity assays or for RNA isolation. If frozen, clinical specimens were thawed and maintained in Iscove Modified Dulbecco Medium (Life Technologies Inc.) supplemented with 20% FBS (Cytiva), IL-3 (Peprotech #200-03), IL-6 (Peprotech #200-06), SCF (Peprotech #300-07), FLT3-LG (Peprotech #300-19), TPO (Peprotech #300-18), SR1 (Stem Cell Technologies #72342), and UM729 (Stem Cell Technologies #72332).⁶⁵ For ex vivo drug studies, 302 samples were analyzed. Of those, 101 samples were used for RNA-seq analysis from data generated in prior studies.⁴ Patient samples with the diagnosis of Acute Myeloid Leukemia (AML) or Acute Leukemia of Mixed or Ambiguous Lineage (ALMAL) were included without age restrictions or disease stage restrictions (both newly diagnosed and relapsed/refractory patients were included). We computed an AUC “combination ratio” (CR) for each sample, by dividing the AUC of the drug combination by the AUC of the more effective (i.e., lower AUC) single agent.⁶⁶ Hence, a CR < 1 indicates the combination outperforms both single agents, meeting the definition of Highest Single Agent synergy. For our “synergy” group, we set a higher bar and required samples to have an AUC CR < 0.7 and a ven+palbo AUC% that was >10 percentage points lower than both the ven and palbo single agent AUC%. When estimating the effect of individual clinical or genetic features on ven+palbo ex vivo synergy (as defined above) within a univariable logistic regression model, odds ratios refer to the relative odds (e.g., compared to a reference group) of a patient sample exhibiting synergy. Multiplicity adjustment of p-values across univariable logistic regression models was not performed. No sex-based differences were observed between the synergy and non-synergy groups.

Method details

Apoptosis, cell cycle and viability assays

One million cells/mL of OCI-AML2 cells were plated in a 12-well plate and incubated with DMSO, 1μM palbociclib, 200nM venetoclax, or a combination of 1μM palbociclib and 200nM venetoclax for 72 h. For apoptosis assays, cells were mixed with Guava Nexin Reagent (Luminex Corp.) and Annexin V positivity was assessed on the Guava Muse Cell Analyzer (Luminex Corp.). For cell cycle assays, 100μL of cells for each treatment were transferred in triplicate to 96-well plate and fixed with 70% ethanol before incubation at 4°C for 3-12h. Fixed cells were stained with Guava Cell Cycle Reagent (Luminex Corp.) and analyzed on the Guava Muse Cell Analyzer (Luminex Corp.). Cell viability was assessed with Guava via count Reagent (SKU 4000-0041) which distinguishes viable and non-viable cells based on differential permeabilities of two DNA-binding dyes in the Guava ViaCount reagent. The nuclear dye stains only nucleated cells, while the viability dye brightly stains dying cells. Reagent used at a 1:10 dilution and cells were analyzed on a Guava Muse Cell Analyzer.

RNA sequencing

Synergy vs. non-synergy pathway analysis was performed using gene expression data from Beat AML⁴ to identify differentially expressed genes between synergy and non-synergy patient samples. Welch’s t test was used to compare normalized expression levels, with false discovery rate (BH FDR) correction applied. Genes meeting the significance threshold (FDR <0.05) were separately submitted as up or downregulated lists to Reactome for pathway analysis to identify enriched biological pathways. All analyses were conducted in R version 4.4.1 using broom, dplyr, ggrepel, purrr, readr, readxl, and stringr.

For cell line work, RNA was isolated from OCI-AML2 parental and CRISPR-derived knockout cells in triplicate using the RNeasy Mini Kit (QIAGEN Inc.). The following RNA-seq tasks were performed by the OHSU Massively Parallel Sequencing Shared Resource. RNA quality was assessed using the Agilent 2100 Bioanalyzer. Libraries were prepared using the TruSeq Stranded mRNA kit (Illumina Inc.) and sequenced on a NovaSeq 6000 (Illumina Inc.). Base calling was done by RTA v3.4.4. Fastq files were assembled using bcl2fastq v2.20.0.422 (Illumina Inc.). Paired reads were trimmed using Trimmomatic 0.36 and the “TruSeq3.fa” adapter file. Trimmed reads were aligned to GRCh38.89 and Ensembl ID-annotated gene counts (i.e., reads per gene) were calculated using STAR 2.5.3a.

For differential expression analysis (DEA) of IKZF1-knockout vs. NT control OCI-AML2 cells, gene-level reads were low count-filtered and normalized by the Trimmed Mean of M-values method (TMM; ‘edgeR’ R/Bioconductor package) before application of surrogate variable analysis (‘sva’ R/Bioconductor package) to identify latent factors – unknown sources of variation – that were controlled for when estimating gene expression differences between IKZF1-KO and NT cells.^67,68 Applying the limma-voom method (‘limma’ R/Bioconductor package), observation-level weights and empirical Bayesian moderation of residual variances were utilized for gene-wise linear modeling, which resulted in log₂ fold changes and associated Benjamini-Hochberg adjusted p-values (controlling the expected False Discovery Rate).⁶⁹

Lentiviral production and transduction

HEK 293T/17 cells were transfected using Lipofectamine-2000 (Life Technologies Inc.) with single transfer vectors in combination with packaging plasmids psPAX2 (Addgene #12260) and VSVG (Life Technologies Inc.). The following lentiviral transfer vectors were purchased from Addgene: pKLV2-U6gRNA5(Empty)-PGKmCherry2AGFP-W (#67981), pKLV2-U6gRNA5(gGFP)-PGKmCherry2AGFP-W (#67982), lentiCRISPRv2 (#52961), and lentiCas9-Blast (#52962). Individual guide sequences purchased from Integrated DNA Technologies provided Table S4. Viral supernatants were collected and 0.5-1mL was used for transduction with 3e6 cells per condition in the presence of 14mM HEPES and 14 μg/mL Polybrene (Santa Cruz). Cells were centrifuged at 2400 RPM for 1.5h at 27°C and centrifugation was repeated 24h later followed by selection with puromycin (2 μg/mL) or blasticidin (10 μg/mL) for 5–7 days. Virus for sgRNA YUSA library was produced by transfecting HEK 293T/17 cells with Calcium chloride, HEPES buffer saline, VSVG, psPAX2 and human CRISPR library plasmid DNA (Addgene #67989). Viral supernatant was collected and concentrated via ultra centrifugation. Titered virus was used to transduce 180 × 10⁶ cells at a multiplicity of infection (MOI) of 0.3. Cells were centrifuged at 2400 RPM for 1.5h at 27°C and centrifugation was repeated at 24h before selection with puromycin (2μg/mL) for 5–7 days.

Genome-wide CRISPR screen

Cas9-expressing OCI-AML2 cells were generated using lentiCas9-Blast (Addgene #52962) as described above and single cell sorted to create a clonal Cas9 line with high functionality. Cas9 functionality was tested by transducing Cas9 containing cells with pKLV2-U6gRNA5(Empty)-PGKmCherry2AGFP-W (Addgene #362132) or pKLV2-U6gRNA5(gGFP)-PGKmCherry2AGFP-W (Addgene #362132) vectors. After 5 days, cells were analyzed for GFP and mCherry positivity by FACS. Clonal OCI-AML2 Cas9 C6 cells were used for genome wide knockout with sgRNA library (Addgene #67989) as described above. After 5–7 days of puromycin selection, transduced cells were treated with DMSO, 1μM Palbociclib, 200nM Venetoclax, or a combination of 1μM Palbociclib and 200nM Venetoclax. Cells were collected at days 0 and 21 for library preparation. Briefly, genomic DNA was extracted, and libraries amplified via PCR. Short read Illumina sequencing assays were performed by the OHSU Massively Parallel Sequencing Shared Resource. The sgRNAs were filtered for low representation (those not seen in the Day0 or those with ≤100 counts per million in more than half the samples) and then were normalized using TMM⁶⁷ for the palbociclib vs. DMSO contrast. However, since the combination and venetoclax samples have very different read count distributions as compared to DMSO samples (due to large effect sizes from treatment) no normalization could be performed. The edgeR package⁷⁰ was used for each linear model, generating log₂ fold changes (LFC) and 2-sided p-values per sgRNA. Gene-level tiering was performed as described previously.²⁵

Ex vivo drug sensitivity assays

Small molecule inhibitors purchased from Selleck Chemicals (Houston, TX) or MedChemExpress were reconstituted in DMSO and stored at −80°C. Cells were seeded in 384-well plates using a Multidrop Combi Reagent Dispenser (ThermoFisher Scientific Inc.) at a density of 1,250 cells/well in 50μL RPMI supplemented with 20% FBS (Cytiva), 2% L-glutamine (Life Technologies Inc.) and 1% penicillin/streptomycin (Life Technologies Inc.). Inhibitors were added using an HP D300e Digital Dispenser (Tecan, Mannedorf, Switzerland). After 72 h, cell viability was assessed using tetrazolium (MTS)-based colorimetric assay (CellTiter96 Aqueous One Solution; Promega). Absorbance (490 nm) was read at 2–24 h using a BioTek Synergy 2 plate reader (BioTek, Winooski, VT). Absorbances of inhibitor treated cells were normalized to the absorbances of untreated cells and wells containing only media. Cell viabilities were averaged across replicates at the inhibitor-specific dose level.

Immunoblotting

OCI-AML2 or MOLM13 parental and CRISPR-derived knockout cell lines were lysed with Cell Lysis Buffer (Cell Signaling Technologies Inc. #9803) containing a Complete Mini Protease Inhibitor Cocktail Tablet, Phosphatase Inhibitor Cocktail 2, and phenylmethylsulfonyl fluoride (PMSF) solution (Sigma-Aldrich Inc.) and clarified by centrifugation. Protein was quantified by Bicinchoninic acid (BCA) assay (ThermoFisher Scientific Inc.). An equal amount of each protein lysate per lane was run on precast 4–15% Tris-Glycine gradient gels (Bio-Rad) and transferred onto Immobilon-P PVDF membranes (Millipore Inc.). Membranes were blocked in a 5% BSA TBS-T solution before O/N incubation at 4°C with primary antibodies. Membranes were washed and probed with species-specific HRP-conjugated IgG antibodies (Promega), then exposed to Clarity Western ECL Substrate (Bio-Rad) and imaged with the Bio-Rad Gel Doc Imaging System.

Sunset assay

OCI-AML2 parental (ven-sensitive) and ven-resistant cell lines were treated with DMSO, 1μM Palbociclib, 200nM Ven, or 1μM Palbociclib +200nM Ven for 24 h. Cells were treated with 5μg/mL puromycin (ThermoFisher Scientific Inc. #A11138-03) for 15 min at 37°C and 5% CO₂ to allow for incorporation into nascent peptide chains, followed by 2X washes with PBS to remove unincorporated puromycin. Cells were harvested by centrifugation at 1200rpm for 5 min and washed in ice-cold PBS before lysing for immunoblotting as described above. Densitometry of puromycin signal was performed using FIJI software. For Sunset assays using primary patient samples, frozen viables were thawed in cytokine rich media as described above. Cells were grown for a period of 3–7 days to ensure viability and adequate cell numbers for experimentation. For primary cell SUnSET assays followed by immunoblot, cells were treated with 10 μg/mL of puro for 30 min (pulse) followed by 2 washes with IMDM+20% FBS. Cells were resuspended in fresh IMDM+20% FBS and incubated at 37°C, 5% CO₂ for 30 min (chase). Cells were washed in PBS, pelleted and prepared for immunoblot as described above. Densitometry of puromycin signal was performed using FIJI software. For primary cell SUnSET assays followed by flow experiments, cells were treated with 10 μg/mL of puro for 20 min (pulse) followed by 3 washes in media and incubated for 40 min at 37°C and 5% CO₂. Cells were stained with Fixable Viability stain (FVS780, BD Horizon Cat.#565388) to determine live/dead cell populations. Cells were blocked with human Fc block (BD Biosciences, Cat# 564219) for 15 min at RT. Cells were incubated for 30 min with an antibody cocktail against surface antigens including anti-human: CD33-RY610 (BD Biosciences, Cat#758223), CD117-Pacific Blue (Biolegend, Cat# 375209), CD45-Pacific Orange (exbio, Cat#PO-684-T025), CD64-PE (Biolegend, Cat#305007), and CD11b-BUV737 (BD Biosciences, Cat#568332). Cells were then fixed/permeabilized (BD Cytofix/Cytoperm Fixation/Permeabilzation kit, Cat#554714) for 20 min, washed and blocked again for 20 min prior to intracellular staining with Puromycin-Alexa647 antibody (Millipore Sigma, Cat#MABE343-AF647). Cells were incubated for 30 min, washed extensively to remove unbound puro antibody and resuspended in 0.5% BSA in PBS until analysis. Cells were analyzed using a Cytek Aurora spectral cytometer and SpectroFlo software. Final data analysis was done using FlowJo software.

Quantitative PCR

RNA was isolated from OCI-AML2 parental and CRISPR-derived knockout cells in triplicate using the RNeasy Mini Kit (QIAGEN Inc.). cDNA was synthesized using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Cat# 4368814). QPCR was run on the QuantStudio7 (ThermoFisher Scientific Inc.) using Applied Biosystems Power SYBR Green PCR Master Mix (Thermo Fisher, Cat#43-676-59) and primers purchased from Integrated DNA Technologies provided in a Table S4.

Quantification and statistical analysis

Data are presented as mean ± SD or SEM as indicated in the figure legends for n = 3 experiments. Two-tailed Student’s t tests were used to calculate p values and are reported in figure legends, within figures and in the Results section. One-way ANOVA with Tukey’s post-test for multiple comparisons was used for cellular outgrowth experiments and are reported in the corresponding figure legend. For in vivo xenograft studies, tumor burden was compared between treated arms (n = 4–6 mice per treatment group) by one-way ANOVA with Tukey’s post-test for multiple comparisons and reported in the figure, figure legends and results section. For in vitro drug profiling of select CRISPR gene knockout cell lines, all treatment conditions were tested in triplicate and fit by four-parameter non-linear regression curves. For differential expression analysis gene-level reads were low count-filtered and normalized by the Trimmed Mean of M-values method (TMM; ‘edgeR’ R/Bioconductor package) before application of surrogate variable analysis (‘sva’ R/Bioconductor package) to identify latent factors (unknown sources of variation).^67,68 Applying the limma-voom method (‘limma’ R/Bioconductor package), observation-level weights and empirical Bayesian moderation of residual variances were utilized for gene-wise linear modeling, which resulted in log₂ fold changes and associated Benjamini-Hochberg adjusted p-values (controlling the expected False Discovery Rate).^69,71

Published: December 29, 2025