A CEBPB/IL-1β/TNF-α feedback loop drives drug resistance to venetoclax and MDM2 inhibitors in monocytic leukemia Free

The p53-MDM2 interaction is an important target in antitumor drug development. TP53 is a master tumor suppressor and plays a critical role in hematopoiesis and leukemogenesis.^1-3 Wild-type (WT) p53 promotes cell cycle arrest, senescence, and apoptosis, thereby preventing cells with mutated or damaged DNA from dividing and oncogenic transformation from occurring. MDM2 is a major negative regulator of TP53. It is an E3 ubiquitin ligase that forms a complex with TP53, inhibits its transactivation, ubiquitinates it for degradation, and facilitates its translocation from the nucleus to the cytoplasm, all of which prevent TP53 from exerting its tumor suppressive function.^4,5,TP53 mutations or deletions are infrequent in de novo adult acute myeloid leukemia (AML) and account for 8% to 13% of adult AML cases.⁶ Inactivation of p53 through abnormalities in p53-regulatory proteins, such as overexpression of MDM2/MDMX, is common in AML. Reactivation of p53 activity has therefore been raised as an attractive therapeutic strategy for treating tumors with WT TP53. MDM2 expression is upregulated in ∼50% of AML cases, supporting the rationale of targeting the p53-MDM2 interaction as a therapeutic strategy.^7-9 There are a growing number of small molecule inhibitors or peptides designed to disrupt the MDM2 and TP53 interaction. Among others, idasanutlin is a potent and selective small-molecule MDM2 antagonist.^10,11 Idasanutlin was shown to induce cell death in leukemia cell lines without TP53 mutations.^12,13 In clinical trials, idasanutlin treatment led to varied responses in AML, polycythemia vera, and solid tumors.^14-16 Besides TP53 mutations, the underlying drug resistance mechanisms are unclear.

Emerging studies have shown that the differentiation status of leukemia cells affects their sensitivity to targeted therapies. Based on morphology and cytochemical staining, AML cells are classified into 8 French-American-British (FAB)^17,18 or 8 World Health Organization subtypes that are indicative of lineage branching and maturation.^19,20 Previous studies have shown that the differentiation status of leukemia cells affects their sensitivity to BCL2 inhibitors.^21-28 In this study, we found that, like their resistance to venetoclax-based therapy, monocytic leukemia cells also exhibit resistance to MDM2 inhibitors.

To unravel the underlying mechanisms, we explored both intrinsic and extrinsic factors that contribute to monocytic leukemia’s resistance to venetoclax and MDM2 inhibitors. We are particularly interested in investigating whether myeloid transcription factors (TFs) and the myriad of factors emitted by the surrounding environment and immune cells play a role in driving drug resistance. These elements orchestrate the differentiation and maturation of hematopoietic stem and progenitor cells (HSPCs), thereby ensuring the proper function of the HSPCs and immune system. In the context of AML, TFs are often dysregulated and contribute to the onset of leukemia and the development of drug resistance. The AML microenvironment,^29-31 especially in cases of monocytic leukemia, is enriched in inflammatory cytokines and growth factors from the surrounding tissue and is further influenced by intracellular pathways that are triggered by cell interactions.³² We aimed to identify specific alterations in TFs and environmental cues in monocytic leukemia that drive the drug resistance to BCL2 and MDM2 inhibitors while also examining the potential interplay between these resistance mechanisms.

We conducted differential expression (DE) gene/protein analysis of the TFs in M4/M5 samples in comparison with M0/M1 samples from the Beat AML and The Cancer Genome Atlas (TCGA) AML cohorts. We overexpressed M4/M5 upregulated myeloid TFs and performed competitive drug assays. We treated M4/M5 and M0/M1 samples with venetoclax and idasanutlin in the presence of a panel of cytokines and measured cell viabilities using a transwell coculture system. To identify dysregulated cytokines, a Luminex assay was performed on supernatants from CD14⁺ cells isolated from the M4/M5 and M0/M1 AML samples. Furthermore, RNA sequencing (RNA-seq) and DE gene and immunoblot analyses were performed on CEBPB-overexpressing cell lines, interleukin (IL)-1–treated AML samples, and idasanutlin- and venetoclax-treated cell lines. An in vivo leukemia xenograft NOD/SCID/IL2rγ^null-SGM3 (NSGS) mouse model was established to evaluate the drug combination effect. Detailed methods are provided in the supplemental Methods (available on the Blood website).

Because of the small clinical trial size and a lack of the matched clinical and genomic data from the currently available idasanutlin clinical trials, we analyzed ex vivo functional screening data from the Beat AML cohort^33-37 (Figure 1A; supplemental Table 1) to explore potential biomarkers for idasanutlin sensitivity. We observed that the area under the curve (AUC) for idasanutlin was positively correlated with the AUC of venetoclax (Figure 1B). Like with venetoclax, primary samples with high expression levels of BCL2A1 and KRAS/NRAS mutations are resistant to idasanutlin²⁶ (Figure 1C-D). We, therefore, overexpressed RAS mutations and BCL2A1 and performed a drug assay. Surprisingly, no changes in idasanutlin sensitivity in the RAS mutants or BCL2A1 overexpressing cells were observed, although KRAS mutations and BCL2A1 overexpression conferred venetoclax resistance (supplemental Figure 1A). As expected, patient samples with TP53 mutations demonstrated resistance to idasanutlin (Figure 1D), which we confirmed through cell line validation experiments (supplemental Figure 1A). We further explored the relationship between gene/protein expression and sensitivity to idasanutlin/venetoclax. A significant correlation between resistance to idasanutlin/venetoclax and the expression of genes/proteins involved in immune functions, notably interleukins, cytokines (especially interleukin-1 [IL-1]), and the NF-κB pathway (Figure 1E; supplemental Table 2), was observed.

We further analyzed the association between the idasanutlin AUC and clinical parameters. We found that idasanutlin resistance was significantly associated with low blast counts, high monocyte, neutrophil, and T cell counts, and the FAB M4/M5 subsets (Figure 1F-G; supplemental Figure 1B).

To confirm the resistance of monocyte leukemia cells to MDM2 inhibitors, we compared the sensitivity of these drugs across different cell types isolated from the same leukemia sample, including leukemia blasts, monocytes, granulocytes, and T cells. We employed either fluorescence-activated cell sorting (FACS) phenotyping to track the percentage changes of each cell population upon drug treatment in unsorted cell populations or performed drug assays on presorted cell types. We observed that, indeed, leukemia blast cells were more sensitive to MDM2 inhibitors than monocytes, neutrophils, and T cells isolated from the same leukemia sample in both drug assays (Figure 1H-I; supplemental Figure 1C). Furthermore, idasanutlin induced significantly less apoptosis in M4/M5 than in M0/M1 leukemia cells (Figure 1J). Conversely, CD34⁺ blast cells were more sensitive to MDM2 inhibitors (supplemental Figure 1D).

To investigate whether M4/M5 leukemia samples exhibit primary drug resistance, specifically reduced expression of direct or major downstream targets of MDM2 inhibitors, we used immunoblotting to measure the protein levels of p53, MDM2, and key p53 downstream targets. M4/M5 leukemia samples showed comparable or higher cytoplasmic or nuclear MDM2, p53, p21, and/or PUMA levels when compared with M0/M1 samples (Figure 1K; supplemental Figure 1E). Upon idasanutlin treatment, both non-M4/M5 and M4/M5 leukemia cells exhibited an increase in p53, MDM2, and p53 downstream targets, indicating the activation of p53 pathways (Figure 1I). These findings suggest that the resistance of monocytic leukemia to MDM2 inhibitors is not driven by reduced expression or inactivation of the p53 pathway.

Given the drug resistance observed in M4/M5 AML when compared with M0/M1 AML for idasanutlin and venetoclax^26,27,38,39 (Figure 1F-J; supplemental Figure 2A), we conducted a DE gene analysis of M4/M5 samples in comparison with M0/M1 samples. We were particularly interested in the TFs that are differentially expressed between M4/M5 and M0/M1 cells. Among the 1281 TFs, 97 were significantly upregulated at a log2 fold change of ≥0.5 and a false discovery rate of <0.05 in M4/M5 (Figure 2A-B; supplemental Table 3). In addition, 131 TFs showed a significant positive correlation with the idasanutlin and venetoclax AUCs (Figure 2C; supplemental Table 4). Furthermore, 53 TFs were significantly upregulated in idasanutlin- and venetoclax-resistant samples when compared with those in sensitive samples (Figure 2D; supplemental Table 5). In total, 32 TFs were shared among the 3 analyses, including 5 myeloid maturation-associated TFs (CEBPB, CEBPD, MAFB, IRF8, and KLF4; supplemental Figure 2B). Four of these 5 TFs were also significantly upregulated in M4/M5 when compared with M0/M1 at the protein level⁴⁰ (supplemental Table 6; Figure 2E).

We overexpressed each TF in 3 TP53 WT cell lines (MOLM14, OCIAML2, and OCIAML3; supplemental Figure 2C-D). Overexpression of CEBPB induced resistance to both venetoclax and idasanutlin in all cell lines as determined by competitive FACS and methyl tetrazolium salt (MTS)-based drug assays (Figure 2F-G). CRISPR-mediated knockout of CEBPB sensitized cells to these drugs (Figure 2H). Furthermore, CEBPB-overexpressing cells conferred broader resistance to other BH3 mimetics, MDM2 inhibitors (Figure 2I), and most combinations evaluated (Figure 2J). Consistent with these findings, CEBPB expression was positively correlated with the AUC for venetoclax combinations (Figure 2K) in the Beat AML cohort, indicating that samples with high CEBPB expression were less sensitive to these drugs than samples with low CEBPB expression.

CEBPB protein expression was examined using immunoblotting. Despite the intersample variability, M4/M5 AML cells displayed significantly higher nuclear levels of CEBPB (Figure 2L; supplemental Figure 2E). These data suggest that, in M4/M5 leukemia samples, there is an inherent upregulation of CEBPB in the nucleus, which may contribute to drug resistance against both BCL2 and MDM2 inhibitors.

To investigate how CEBPB drives resistance to BCL2 and MDM2 inhibitors, we first examined CEBPB RNA expression correlations using patient data from the TCGA and Beat AML cohorts. In line with its role in driving myeloid differentiation and maturation, CEBPB expression was positively correlated with myeloid maturation markers, such as ITGAM (also named CD11b) and CD14 (Figure 3A; supplemental Figure 3A; supplemental Table 7). This was confirmed by our experiments that showed an upregulation of CD14 and CD11b levels when CEBPB was overexpressed in AML cell lines (Figure 3B; supplemental Figure 3B).

We next analyzed the correlation between CEBPB and the p53 pathway and apoptosis related genes. CEBPB was positively correlated with the antiapoptotic genes MCL1 and BCL2A1 and with extrinsic apoptosis pathway genes, such as FAS, TNF, and their receptors, while being negatively correlated with TP53, BCL2, PMAIP1, CASP3, and CASP6 (Figure 3A; supplemental Figure 3A). Intriguingly, the DE gene network analysis showed that M4/M5 AML samples expressed significantly higher levels of BCL2A1 and tumor necrosis factor (TNF) family members and lower levels of TP53, BCL2, CASP3, and CASP6 than the M0/M1 AML samples at the messenger RNA or protein levels (Figure 3C; supplemental Figure 3C). Mining the gene expression of normal hematopoiesis,⁴¹ we observed that multiple extrinsic apoptosis–related genes, including TNF, FAS, TNFSF10, TNFSF12, TNFSF13, TNFRSF1B, DISC1, and TNFAIP2, were upregulated, whereas several caspases, notably CASP6, was downregulated during monocytic differentiation (Figure 3D; supplemental Figure 3D). These findings indicate that the upregulation of the extrinsic apoptosis pathway and the downregulation of specific caspases in monocytic leukemia are consistent with the patterns observed during normal hematopoiesis.

Aligned with the correlation analysis, the RNA-seq DE gene analysis (CEBPB vs empty vector overexpression) revealed alterations in the apoptosis-related genes, enhanced cytokine/growth factor signaling, and markers of myeloid maturation. Specifically, CEBPB overexpression increased BCL2A1, MCL1, IL-1/TNF/NF-κB pathway genes, CD14, and CD11b, whereas BCL2 and downstream targets of the p53, including TP53, PMAIP1, BBC3, and CDNK1A, were decreased (Figure 3E; supplemental Figure 3E; supplemental Table 8). Furthermore, CEBPB overexpression downregulated CASP3 and upregulated IL-1β, its receptor CD121a, CD120b, the granulocyte-macrophage colony-stimulating factor receptor CD116, and MCL1 at the translational level (Figure 3F; supplemental Figure 3F-H). Immunoblot analysis revealed increased phosphorylated NF-κB and p38 in M4/M5 when compared with M0/M1 AML patient samples (supplemental Figure 3I). Notably, CASP3/6, BCL2 negatively and MCL1/BCL2A1 and multiple TNF family genes/proteins were positively correlated with the idasanutlin and/or venetoclax AUC (Figure 3G; supplemental Figure 3J). CASP3 and CASP6 were also significantly downregulated in the idasanutlin- and venetoclax-resistant patient samples (with top 20% drug AUC) when compared with the sensitive samples (with bottom 20% drug AUC) in the Beat AML cohort (supplemental Figure 3K). To validate this, we performed overexpression and knockout experiments. Overexpression of CASP3 and CASP6 increased drug sensitivity (Figure 3H), whereas CRISPR-mediated knockout of CASP3 and CASP6 conferred resistance (Figure 3I). Conversely, overexpression of MCL1 induced resistance to both drugs (Figure 3I).

These data suggest that monocytic leukemia intrinsically expresses high levels of CEBPB, which leads to elevated BCL2A1, MCL1, and NF-κB pathway members and reduced BCL2 and caspases, thereby contributing to the BCL2 and/or MDM2 inhibitor resistance.

Immature monocytes accumulate in M4/M5 AML because of differentiation skewing and blockage.³² We subsequently investigated whether these cells protected leukemia cells from MDM2/BCL2 inhibition by producing cytokines.

Leukemia blast cells were treated with MDM2 inhibitors or venetoclax in the presence or absence of leukemic granulocytes, monocytes, and T cells isolated from different leukemia samples in a transwell coculture plate (Figure 4A). Remarkably, the viability of leukemia blast cells was significantly enhanced when cocultured with monocytes isolated from monocytic leukemia but not when cocultured with granulocytes or T cells from the same leukemia patient or with monocytes from healthy donors (Figure 4B-E). This suggests that abnormal monocytes in monocytic leukemia produces a specific prosurvival effect.

Because there was no cell-to-cell contact in this coculture apparatus, we investigated the extrinsic factors that contributed to apoptosis protection by the leukemia monocytes. We measured cytokines and growth factors in the cell culture medium that were secreted by CD14⁺ monocytes or their precursors isolated from different leukemia samples or healthy monocytes. We observed that CD14⁺ cells from M4/M5 samples secreted significantly higher levels of IL-1α, IL-1β, and TNF-α than CD14⁺ cells from non-M4/M5 and healthy monocytes (Figure 4G).

To investigate whether IL-1 and TNF-α affected the drug response, we treated primary leukemia cells with venetoclax or MDM2 inhibitors in the presence of these factors. We also assessed 8 other leukemia-enriched factors. We observed that the response of leukemia cells to these factors exhibited considerable variability among the FAB subtypes. IL-3 and granulocyte-macrophage colony-stimulating factor (GM-CSF) decreased drug-mediated apoptosis in all leukemia and cord blood (CB) HSPCs (Figure 4H; supplemental Figure 4A). IL-1α, IL-1β, and TNF-α uniquely protected M4/M5 cells, whereas stem cell factor (SCF) significantly increased the viability of CB HSPCs and M0/M1 cells in the presence of idasanutlin and venetoclax (Figure 4H; supplemental Figure 4A). Interestingly, TNF-α augmented idasanutlin- and venetoclax-mediated cell death in most M0/M1 cells (Figure 4H; supplemental Figure 4A). Consistently, the plasma IL-1β, IL1RA, and/or TNF-α levels were positively correlated with the idasanutlin and venetoclax AUC in the Beat AML samples⁴² (Figure 4I). Gene expression analysis showed that the M4/M5 AML samples demonstrated significantly higher levels of TNF and cytokine receptors, including CSF2RA, TNFRSF1B, and IL1R2, than the non-M4/M5 samples (Figure 3C). In addition, the expression of these cytokine receptors showed a positive correlation with the AUCs of these drugs (supplemental Figure 4B) and were upregulated in drug-resistant samples when compared with the drug-sensitive samples (supplemental Figure 4C). As depicted in Figure 3E, the overexpression of CEBPB concurrently enhanced the expression of IL-1/TNF and their receptors. These data suggest that CD14⁺ monocytes in M4/M5 leukemia secrete elevated levels of IL-1 and TNF-α that confer extrinsic drug resistance to M4/M5 leukemia blast cells that express high levels of IL-1/TNF-α receptors that are driven, at least, partially by the elevated CEBPB expression in these cells.

In agreement, the single-cell RNA-seq data⁴³ showed that M4/M5 promonocytic and various differentiated cells expressed higher levels of CEBPB and TNF than those from M0/M1 samples (Figure 4J-K; supplemental Table 9). Furthermore, these cells, along with the AML progenitor cells in M4/M5 cells, expressed higher levels of cytokine receptors and lower levels of CASP3 and CASP6 than their M0/M1 counterparts (Figure 4J-K; supplemental Table 9). These data highlight the coexistence and potential interplay between CEBPB-mediated intrinsic and IL-1/TNF-α–mediated extrinsic drug resistance mechanisms in monocytic leukemia.

To elucidate the mechanisms of IL-1– and TNF-α–mediated MDM2 inhibitor resistance, we performed immunoblotting on idasanutlin-treated AML samples in the presence or absence of these factors. When TNF-α or IL-1α/β was added, p53, MDM2, and key p53 downstream effectors remained upregulated; indeed, their levels were higher in the presence of idasanutlin when combined with IL-1β and TNF-α than when exposed to idasanutlin alone (Figure 5A-B; supplemental Figure 5A). These results suggest that the IL-1/TNF-α–driven resistance operates via a p53 pathway-independent mechanism.

We then performed RNA-seq on samples from patients with AML that were treated with idasanutlin, with or without IL-1β. Consistent with the immunoblot results, most p53 pathway targets were upregulated after idasanutlin treatment regardless of the presence of IL-1β (Figure 5C; supplemental Figure 5B; supplemental Table 10). We subsequently performed DE analysis on IL-1β treated AML cells (IL-1β vs untreated; IL-1β + idasanutlin vs untreated; and IL-1β + idasanutlin treatment vs idasanutlin treatment alone; Figure 5D). IL-1β treatment upregulated 3295, 1353, and 680 genes (≥1 log2 fold change), and most of these genes (82%, 62.4%, and 47.6%) remained upregulated in the presence of idasanutlin in 3 patients, respectively. Pathway analysis showed that the upregulated genes were associated with NF-κB, the IL-1 family, immune function, inflammation, and myeloid maturation (supplemental Figure 5C). Interestingly, IL-1β induced its own expression and other prosurvival cytokines, including TNF, CSF2 (GMCSF), IL-6, and CSF3 (GCSF), and the respective receptors and/or coactivators (supplemental Table 10). We verified that IL-1β and TNF-α induced the upregulation of their membrane-bound forms, their receptors, CSF2RB, and the monocytic marker CD14 as determined by FACS immunophenotyping (Figure 5E; supplemental Figure 5D). Furthermore, IL-1β upregulated many proinflammatory chemokines, including CXCL13 and CXCL2, which have been shown to activate monocytes and enhance their release of inflammatory cytokines. This autocrine loop of activation of multiple prosurvival signaling pathways may, at least, partially contribute to the IL-1–mediated drug resistance.

Interestingly, IL-1 and TNF-α induced a marked increase in CEBPB only in M4/M5 but not in M0/M1 samples (Figure 5F-H). The upregulation of CEBPB was accompanied by variable upregulation of MDM2 and MCL1 and downregulation of CASP6 and/or BCL2 (Figure 5G-H). These data highlight the pleiotropic nature of IL-1/TNF and the heterogeneous responses of AML samples to these factors. Moreover, these findings suggest a synergistic interaction between CEBPB-mediated intrinsic drug resistance and IL-1/TNF-α–mediated extrinsic drug resistance mechanisms, which likely contribute to the observed drug resistance to MDM2 and BCL2 inhibitors in monocytic AML.

MDM2 inhibitors upregulate both the intrinsic and extrinsic apoptotic pathways.^44-47 We performed RNA-seq on idasanutlin-treated cells. Consistent with previous studies, we observed an upregulation of not only key p53 downstream targets related to the intrinsic apoptosis pathway but also of extrinsic apoptosis pathway genes, including the TNF family members (supplemental Table 11; Figure 6A). Interestingly, idasanutlin treatment also induced IL1B, CEBPB, and myeloid differentiation markers, such as CD14 and CD11b (Figure 6A), suggesting a feedback protection mechanism.

To pinpoint the downstream targets of venetoclax and explore the potential emergence of similar protective mechanisms during its treatment, we conducted a DE gene analysis (venetoclax vs dimethyl sulfoxide [DMSO]) by comparing venetoclax-treated cells with DMSO-treated vehicle controls. Venetoclax treatment led to the upregulation of p53 downstream targets, particularly genes associated with cell cycle regulation and DNA damage response (supplemental Figure 6A; supplemental Tables 12 and 13), indicating the activation of p53-mediated antiproliferation and apoptosis pathways. Interestingly, BCL2A1 expression markedly increased upon venetoclax treatment (Figure 6B). Concurrently, several NF-κB pathway components, the IL-1 family, and a suite of immune-related genes were upregulated, indicating that venetoclax has a broad impact on cellular stress responses and immune activation (supplemental Figure 6A). Upon analyzing the protein network associated with the significantly upregulated genes (log fold change ≥0.5), we found that 62 of the 92 genes were interconnected and central within the TNFA/IL1B network (supplemental Figure 6B), indicating a pivotal role for TNF-α and IL-1β in regulating these genes.

To confirm our observations, we examined the protein levels of CEBPB, TNF-α/IL-1β, and their corresponding receptors in primary patient samples following drug exposure using immunoblotting and FACS immunophenotyping (Figure 6B-F). Notably, venetoclax and idasanutlin treatment induced a dose-dependent upregulation of CEBPB protein (Figure 6C) in M0/M1, M2, M4/M5 samples, and CB HSPCs (Figure 6D-E). The drug treatment also induced myeloid differentiation (increased CD14 expression) and upregulated TNF-α/IL-1 and/or their receptors in AML cell lines and primary AML samples, albeit with variability (Figure 6F-G; supplemental Figure 6C). Consistent with CEBPB’s role in regulating caspases, MDM2, and MCL1, idasanutlin and venetoclax treatment variably increased MDM2 or MCL1 and decreased the CASP3 and CASP6 levels in patient samples (Figure 6H; supplemental Figure 6D). These data suggest that idasanutlin and venetoclax treatment can further enhance CEBPB-mediated intrinsic and IL-1/TNF-α–mediated extrinsic drug resistance.

Because the IL-1/TNF-α/CEBPB feedback loop drives idasanutlin/venetoclax resistance, we investigated whether blocking the IL-1/TNF-α receptor or inhibiting IL-1/TNF-α–mediated signaling could restore sensitivity to these therapies.

We performed an in vitro drug assay and observed that combining inhibitors or antagonists of MAPK14 (p38; doramapimod), the IRAK4 inhibitor (CA-4948), IL-1 (anakinra or IL1RA), or TNF-α (IW927) modestly decreased the viability of M4/M5 cells when compared with cells treated with idasanutlin/venetoclax alone as determined by MTS assays (Figure 7A). Notably, the synergistic effect (reduced AUCs when the combination was compared with venetoclax) between doramapimod and venetoclax was only observed in the M4/M5 AML samples (22 of 23 samples; P < .0001) and not the M0/M1/M2 samples (14 out of 23; P = .35) from the Beat AML cohort²¹ (Figure 7B). Furthermore, these combinations reduced monocyte protection in coculture experiments as indicated by the reduced viabilities (Figure 7C). A trend toward increased survival was observed in mice that received CA-4948, anakinra, and venetoclax combination (P = .08) when compared with those treated with venetoclax alone, although this did not reach statistical significance (Figure 7D). Future studies will investigate the potential benefits of extending the treatment duration or combining these agents with other therapies, such as doramapimod or a TNF-α antagonist, to overcome drug resistance.

Recent research suggests a link between a leukemia cell’s maturation status and its response to targeted treatments. We showed that M4/M5 leukemia samples are resistant to MDM2 inhibitors in this study in agreement with the resistance to venetoclax observed in previous studies.^23,26,38 We demonstrated that the upregulation of CEBPB and the IL-1/TNF family members in M4/M5 samples induced intrinsic and extrinsic drug resistance to these inhibitors. We also uncovered a drug-induced compensatory protection mechanism that involved these same factors.

C/EBPs are a family of TFs that participate in a variety of physiological activities, including energy metabolism, fat storage, tissue differentiation, hematopoiesis, and immune responses.^48-55 CEBPB is known to promote myeloid differentiation during basic and acute-phase emergency hematopoiesis. Therefore, it is not surprising that CEBPB increases the expression of monocyte markers, such as CD11b and CD14. CEBPB is distinct from other CEBPs because of its ability to respond to a variety of extracellular and intracellular signals, thereby triggering numerous defensive and feedback responses.^51,56 Studies have shown that CEBPB regulates the expression of numerous genes/proteins associated with the infection-related acute-phase and cancer cachexia-inducing secretome, such as TNF, IL-8, IL-6, CSF3, and adiponectin, which contribute to poor outcomes and drug resistance in cancer.^57-64 In this study, we observed that CEBPB is upregulated and orchestrates the expression of IL-1/TNF and other inflammatory factors in monocytic leukemia (Figure 3C,E; supplemental Figure 3D). In addition, CEBPB also modulates the expression of apoptosis pathway members (Figure 3F; supplemental Figure 3F).

The expression of intrinsic apoptosis genes is dynamic during hematopoiesis and heterogeneous in leukemia samples. We and others have shown that BCL2 have high expression levels in healthy HSPCs and primitive M0/M1 leukemia samples.^26,65 In contrast, BCL2A1 levels increase in more differentiated cells and in M4/M5 AML.^26,38 In this study, we showed that extrinsic apoptosis-related genes and caspases also vary dynamically during hematopoiesis, aligning with the AML maturation status (Figure 3C-D; supplemental Figure 3D-E). An array of TNF and other death receptor members are upregulated, whereas CASP6 and CASP3 are downregulated in M4/M5 AMLs (Figure 3C; supplemental Figure 3D). Our data further suggest that CEBPB participates in modulating this switch. CEBPB enhances MCL1, BCL2A1, and IL-1/TNF members while suppressing BCL2 and CASP3, thereby contributing to drug resistance against venetoclax and MDM2 inhibitors by decreasing the intrinsic apoptosis susceptibility and increasing TNF extrinsic apoptosis.

Beyond CEBPB-driven intrinsic drug resistance, we discovered that monocytic leukemia cells produce higher levels of IL-1 and TNF-α, which leads to extrinsic drug resistance. Leukemia cells exhibit diverse responses to cytokines and growth factors. Our data underscored the pivotal role of leukemia differentiation status in shaping the responses to these factors. TNF-α acts synergistically with BCL2 and MDM2 inhibition to enhance apoptosis in most M0/M1 AML cells (Figure 4H). In contrast, IL-1 and TNF-α selectively protect M4/M5 cells against apoptosis induced by BCL2 and MDM2 inhibition (Figure 4H; supplemental Figure 4A). Furthermore, IL-1β and TNF-α further enhanced CEBPB expression in M4/M5 but not in M0/M1cells (Figure 5F). Therefore, IL-1β, TNF-α, and CEBPB form an autocrine forward-feedback loop in M4/M5, which reinforces each other and may at least partially explain the drug resistance in M4/M5.

Alongside the CEBPB-mediated intrinsic and IL-1/TNF-α–mediated extrinsic drug resistance, we have uncovered a drug-induced compensatory protective mechanism. Previous studies have shown that when BCL2 is inhibited by venetoclax, MCL1 is upregulated.⁶⁶ Similarly, MDM2 expression increased following its own inhibition caused by the p53-MDM2 negative feedback loop.⁶⁷ Tseng et al showed that Nutlin3 treatment delayed MCL1 degradation, thereby contributing to the apoptosis protection and docetaxel resistance in melanoma cells.⁴⁵ Consistent with these findings, we observed elevated MDM2 and MCL1 levels following idasanutlin and venetoclax treatment (Figure 6G). Intriguingly, we also observed a dose-dependent increase in CEBPB messenger RNA and protein levels following drug exposure (Figure 6A,D). This increase coincided with elevated levels of the IL-1/TNF-α signaling pathway, monocyte differentiation, and other extrinsic apoptosis pathway members. As discussed earlier, CEBPB and IL-1/TNF-α confer drug resistance to M4/M5 leukemia, whereas TNF-α promotes apoptosis in M0/M1 leukemia. These findings suggest that venetoclax- and idasanutlin-induced extrinsic apoptosis pathways synergize with the mitochondrial intrinsic apoptosis pathway to induce cell death in M0/M1 cells, whereas the induction of extrinsic apoptosis in the M4/M5 cells by these inhibitors seems to confer resistance and mitigate the impact of these inhibitors. Our data also provide a potential mechanism for acquired resistance in patients who are receiving venetoclax-based therapies.

The upregulation of CEBPB, inflammatory cytokines, and receptors, the switch in intrinsic apoptosis gene expression, the prosurvival role of the extrinsic apoptosis pathway, and a drug-induced compensatory protective mechanism collectively contribute to the drug resistance observed in M4/M5 leukemia cells. Considering the presence of multiple drug resistance mechanisms in monocytic leukemia, a single-agent combination treatment is unlikely to achieve a robust and sustained response. We observed only modest synergy between venetoclax/idasanutlin and inhibitors or antagonists that targeted individual pathway. These findings underscore the urgent need for further research to investigate the safety and efficacy of employing multiple combination therapies and to explore the potential of directly targeting CEBPB through innovative approaches, such as molecular glue degraders or proteolysis targeting chimeras (PROTACs).

Our findings highlight the multifaceted roles of CEBPB, IL-1/TNF-α, and monocyte differentiation in driving AML drug resistance. This suggests that targeting both intrinsic and extrinsic resistance mechanisms may improve the efficacy of AML combination therapies.

The authors thank their colleagues in the Druker, Tyner, and Agarwal laboratories for sharing reagents and insightful discussions, and the Carolinas Cord Blood Bank for providing the cord blood. The authors thank their patients for their generous donations of tissue samples. The authors express their appreciation to the dedicated patient sample processors in the Oregon Health & Science University (OHSU) hematologic malignancy neighborhood for their diligent work in processing patient samples. The authors acknowledge the OHSU Massively Parallel Sequencing Shared Resource and the Flow Cytometry Shared Resource for their technical assistance, exceptional support and expertise, and valuable guidance throughout the study.

H.Z. was supported by a National Cancer Institute (NCI)/National Institutes of Health (NIH) R37CA284012 and R00 (5K99CA237630/5R00CA237630-05) grant, the Oregon Medical Research Foundation New Investigator Award, the Translational Oncology Program Pilot Award (TOP-2023-002), and a Cancer Early Detection and Research (CEDAR) pilot award. J.W.T. was supported by the V Foundation for Cancer Research, Gabrielle’s Angel Foundation for Cancer Research, the Mark Foundation For Cancer Research, the Silver Family Foundation, and the NCI/NIH (5R00CA151457-04 and 1R01CA183947-01). R.M. was supported by NIH/NCI Grant 1R01CA251331 and the Stanford Ludwig Center for Cancer Stem Cell Research and Medicine.

Contribution: B.A., A.W., R.C., L.S., T.L., K.J., and H.Z. performed the in vitro experiments, analyzed and interpreted the data, and wrote the manuscript; H.-Y.L., I.K., A.S., and B.H.C. performed the in vivo experiments; T.K. created the figures and revised the manuscript; N.L. summarized the patient sample information; D.B., P.R., and A.K. analyzed the RNA sequencing (RNA-seq) data and participated in data interpretation and discussion; A.C. and R.S. performed the library preparation and RNA-seq; P.D.P. and S.J.C.G. participated in the proteomics data analysis and discussion; C.T., S.K.M., J.W.T., B.J.D., R.M., M.B., B.H.C., A.A., C.A.E., S.E.K., E.M., and H.Z. participated in the study design, data analysis, discussion, and oversight; and all the authors had access to the data, reviewed the manuscript, and approved the final version of the manuscript for submission.

Conflict-of-interest disclosure: J.W.T. reports receiving research support from Aptose, Array, AstraZeneca, Constellation, Genentech, Gilead, Incyte, Janssen, Seattle Genetics, Syros, and Takeda. R.M. reports serving on the advisory boards of Kodikaz Therapeutic Solutions, Orbital Therapeutics, Pheast Therapeutics, and 858 Therapeutics; and reports being the cofounder of and having equity in Pheast Therapeutics, MyeloGene, and Orbital Therapeutics. C.T. reports receiving support from Notable Labs. B.J.D. reports serving on the senior advisory boards (SABs) of Adela Bio, Aileron Therapeutics (inactive), Therapy Architects/ALLCRON (inactive), Cepheid, Labcorp, Nemucore Medical Innovations, Novartis, and RUNX1 Research Program; serving on the SABs of and owning stock in Aptose Biosciences, Blueprint Medicines, Enliven Therapeutics, Iterion Therapeutics, GRAIL, and Recludix Pharma; serving on the board of directors of and owning stock in Amgen and Vincerx Pharma; serving on the board of directors of Burroughs Wellcome Fund and CureOne; serving on the joint steering committee of Beat AML LLS; serving on the advisory committee of Multicancer Early Detection Consortium; being the founder of VB Therapeutics; having a sponsored research agreement with AstraZeneca, DELiver Therapeutics, Immunoforge, Enliven Therapeutics (inactive), Recludix Pharma (inactive); receiving clinical trial funding from Novartis and AstraZeneca; receiving royalties from patent 6958335 (Novartis exclusive license), OHSU and the Dana-Farber Cancer Institute (Merck exclusive license, CytoImage, Inc exclusive license, DELiver Therapeutics nonexclusive license, Sun Pharma Advanced Research Company nonexclusive license), and the US patents 4326534, 6958335, 7416873, 7592142, 10473667, 10664967, and 11049247. The remaining authors declare no competing financial interests.

Correspondence: Haijiao Zhang, Division of Oncological Sciences, Knight Cancer Institute, Oregon Health & Science University, 3181 Sam Jackson Park Rd, Portland, OR 97239; email: zhahai@ohsu.edu.