The Heme-Regulated Inhibitor Pathway Modulates Susceptibility of Poor Prognosis B-Lineage Acute Leukemia to BH3-Mimetics Free

This article is featured in Highlights of This Issue, p. 541

Although cure rates for children with B-precursor acute lymphoblastic leukemia (B-ALL) now exceed 90%, responses in adults and children with poor-prognosis B-ALL (e.g., KMT2A rearrangement, BCR-ABL1 fusion, hypodiploid, etc.) are much poorer (1). Philadelphia chromosome–positive (Ph⁺) B-ALL, which results from the fusion of the BCR-ABL1 oncogene, encodes a constitutively active tyrosine kinase in B-ALL (2). Although treatment of these patients has improved with the use of tyrosine kinase inhibitors (TKI), many patients relapse and develop TKI resistance highlighting the need for improvements in treatment (3, 4). Another poor-prognosis subtype of ALL is Philadelphia chromosome–like (Ph-like) ALL; Ph-like ALL has a gene expression profile similar to Ph⁺ B-ALL but is BCR-ABL1 negative (5). Ph-like ALL is associated with high-risk clinical features and patients respond poorly to chemotherapy (5, 6). Because Ph-like ALL involves a range of mutations leading to cytokine receptor and tyrosine kinase activations, precision medicine approaches have proven to be difficult and depend on the availability of targeted therapies (5). Therefore, new approaches are warranted to improve treatment for poor-prognosis B-ALL in adults and children.

Antiapoptotic members of the BCL-2 family of proteins (e.g., BCL-2, BCL-XL, BFL-1, BCL-W, and MCL-1) are amplified in many types of cancers to prevent cell death (7, 8). MCL1 is among the most frequently amplified genes in human cancer and many cancer cells, including B-ALL cells, depend on MCL-1 expression for survival (9, 10). So-called BH3-mimetic drugs have been developed that inhibit antiapoptotic members, allowing the release of proapoptotic members to trigger apoptosis (8). Navitoclax (ABT-263), which targets BCL-2/BCL-XL/BCL-W, was the first-in-class BH3-mimetic; however, its clinical use has been hampered by dose-limiting toxicities associated with on-target apoptosis induction in mature platelets (11). Currently, the only FDA approved BH3-mimetic is venetoclax (ABT-199), which targets BCL-2 (12). Venetoclax has shown potency in a variety of malignancies, including chronic lymphocytic leukemia and acute myelogenous leukemia, especially when given in combination with standard chemotherapy (12–14). Venetoclax is well-tolerated in patients with low white blood cell counts as the most common side-effect (12). MCL-1 has been shown to be involved in resistance to ABT-199, as well as other chemotherapies, highlighting the need for MCL-1–targeting agents (15, 16). There has been notable progress in the clinical development of specific and potent MCL-1 inhibitors (e.g., S63845, AMG397, AMG176, and AZD5991); however, none are currently FDA approved, and the FDA recently placed a clinical hold on a phase I, dose escalation study of AMG397 after preliminary findings suggested some cardiac toxicity (17–19). While the extent of the cardiac toxicity is still being evaluated, genetic deletion of Mcl1 in adult mouse cardiomyocytes also triggered a rapid, fatal cardiomyopathy (20, 21).

Because direct pharmacologic inhibition of MCL-1 may not be feasible due to on-target toxicities, it is important to investigate other mechanisms of targeting MCL-1. We have previously shown that the antimalarial drug dihydroartemisinin (DHA) can repress MCL-1 translation and synergize with the BH3-mimetic ABT-263 (22). Here, we show mechanistically that DHA induces the repression of MCL-1 through activation of a cellular heme-sensing pathway that results in eIF2α phosphorylation and subsequent attenuation of protein translation. The activation of this pathway by a small molecular tool compound can also repress MCL-1 and synergize with BH3-mimetics in both Ph⁺ and Ph-like leukemia. This observation reveals for the first time that this cellular heme-sensing pathway can be targeted to repress endogenous MCL-1 expression and lead to improved treatment of poor prognosis acute lymphoblastic leukemia.

Mouse p185⁺ Arf-null B-ALL (hereafter referred to as BCR-ABL⁺ B-ALL; ref. 10) and PAX5-JAK2, RCSD1-ABL1, and RCSD1-ABL2 fusion-expressing Ph-like Ba/F3 cells obtained from the Mullighan laboratory (23) were grown in RPMI with 10% FBS, 55 μmol/L 2-mercaptoethanol, 2 mmol/L glutamine, penicillin, and streptomycin (Invitrogen). The human Ph⁺ leukemia cell lines OP-1, TOM-1, BV-173, and SUP-B15 were obtained from the Mullighan laboratory, authenticated by short tandem repeat profiling (St. Jude Hartwell Center, Memphis, TN), and cultured in RPMI with 20% FBS, 55 μmol/L 2-mercaptoethanol, 2 mmol/L glutamine, penicillin, and streptomycin. All cell lines were tested monthly for Mycoplasma contamination (MycoAlert Plus, Lonza) and were maintained in culture for no more than 1 month. Navitoclax and venetoclax were obtained from Selleckchem. DHA was obtained from AvaChem Scientific. BTdCPU was obtained from Millipore and synthesized by the Department of Chemical Biology and Therapeutics (St. Jude Children's Research Hospital, SJCRH, Memphis, TN). ISRIB (Sigma) was dissolved in DMSO for in vitro experiments.

Cas9-expressing p185⁺ B-ALL cells were transduced with the Brie CRISPR KO library (Addgene #73633) at a multiplicity of infection of 0.5 with an sgRNA coverage of 400× (24). One day after transduction, puromycin was added (1 μg/mL) to select for infected cells. Cells were then cultured in either DMSO or 10 μmol/L DHA containing media for 24 hours, followed by a 48-hour culture in drug-free media. Genomic DNA was extracted from the surviving cells (Qiagen DNeasy Blood and Tissue Kit) and amplified by PCR for Illumina sequencing of sgRNAs by Nextseq (24). MAGeCK was used to analyze the sequencing reads (25). Enrichr was used for pathway analysis of screen (26, 27). Raw sequencing data archived in NIH BioProject: Accession PRJNA681582.

A validation CRISPR KO library was generated by the Center for Advanced Genome Editing (SJCRH) containing 5 guides (different from those in the Brie library) for each of the top 35 hits in positive selection (FDR < 1%) and the top 13 hits in negative selection (FDR < 16%) from the initial screen. Cas9-expressing BCR-ABL⁺ B-ALL cells were transduced with this validation library at a multiplicity of infection of 0.5 with a sgRNA coverage of >5,000×. The drug treatment, sequencing, and analysis were carried out in the same manner as the genome-wide CRISPR screen.

BCR-ABL⁺ B-ALL knockout cell lines were generated using CRISPR-Cas9 technology. One million mouse BCR-ABL⁺ B-ALL cells were transiently cotransfected with precomplexed ribonuclear proteins (RNP) consisting of 100 pmol of chemically modified sgRNA (Synthego) and 35 pmol of Cas9 protein (SJCRH Protein Production Core). In addition, 200 ng of pMaxGFP was cotransfected via nucleofection (Lonza, 4D-Nucleofector X-unit) using solution P3 and program CM-137 according to the manufacturer's recommended protocol. Five days after nucleofection, cells were single-cell sorted by FACs to enrich for GFP⁺ (transfected) cells and verified for out-of-frame indel modifications via sequencing on a Miseq Illumina sequencer. NGS analysis was performed using CRIS.py (28). Two knockout clones for each gene were identified. BV-173 genetically modified cell pools were created by transiently cotransfecting 400,000 cells with precomplexed RNPs consisting of 100 pmol of chemically modified sgRNA (Synthego) and 35 pmol of Cas9 protein (SJCRH Protein Production Core) via nucleofection (Lonza, 4D-Nucleofector X-unit) using solution P3 and program CA-137 according to the manufacturer's recommended protocol. Five days after nucleofection, a portion of cells were harvested and sequenced via targeted NGS and analyzed using CRIS.py as described above. NGS analysis indicated 89% total indels and 80% out-of-frame indels for the BV-173 edited cell pool. Sequences of sgRNA can be found in Supplementary Table S1. MCL-1–overexpressing cells were previously generated by retroviral transduction (10, 22).

Protein expression was assessed as described previously (29). Antibodies used were: anti-MCL-1 (Rockland Immunochemical), anti-human MCL-1, anti-PERK, anti-ATF6, anti-IRE1, anti-CHOP, anti-ATF4, anti-BCL-XL, anti-phospho-eIF2α, anti-eIF2α, anti-PUMA (Cell Signaling Technology), anti-HRI (Abcam), anti-NOXA (Sigma), anti-BIM, anti-BCL-2 (BD Pharmingen), and anti-Actin (Millipore). Anti-rabbit or anti-mouse horseradish peroxidase–conjugated secondary antibodies were from Jackson Immunochemicals. Anti-hamster horseradish peroxidase-conjugated secondary antibody was from Invitrogen. Immunoblots were developed using Odyssey Fc from LI-COR Biosciences. Image Studio Lite version 5.2 software was used to select each band and quantify signal intensities by subtracting background and analyzed by GraphPad Prism. All experiments are representative of ≥3 independently performed assays.

Absorption spectra were acquired using a Nanodrop 1000 spectrophotometer (Thermo Fisher Scientific) at room temperature. All dilutions were made in enzyme dilution buffer consisting of 50 mmol/L Tris (pH 7.4), 150 mmol/L NaCl, 0.1 mmol/L EDTA, and 25% glycerol. HRI (0.05 μg/μL) was obtained from ABM (Z500115). Heme (Frontier Scientific) and DHA were dissolved in DMSO. Enzyme was brought to 5 μmol/L with heme and to the indicated concentration of DHA then incubated at room temperature for 30 minutes. All measurements were made after blanking with buffer (control) or with buffer containing the equivalent concentrations of heme and DHA as the measured sample. Absorbance readings between 220 and 750 nm were made for each sample, and Soret peak height was determined by peak absorbance (420 nm) from 3 or more technical replicates per value.

Cellular heme was measured by reverse-phase HPLC after pellets were extracted with acetone, acidified by the addition of 20% 1.6 N HCl, together with 10 pmol mesoporphyrin per sample added as an internal standard. Extracts were centrifuged at 21,100 × g for 10 minutes. after which pellets were discarded. Heme was separated from other porphyrins on a Shimadzu system (CBM-20A system controller, Shimadzu), using a mobile phase of acetonitrile in water containing 0.05% trichloroacetic acid at 1 mL/minute on a reverse-phase C18 column (Sigma), applying a 30% to 66% linear acetonitrile gradient over 5 minutes followed by a 66% to 90% linear gradient over 20 minutes. Heme was determined by measuring absorbance at 400 nm (Shimadzu, SPD-20AV). The concentration and identity of heme (t_R = 7.4 minutes) was made by comparison with hemin (Frontier Scientific, H651–9) standards extracted analogously to samples. Calculations were made by normalizing the peak area for heme with internal standard peak values for samples then calculating from a linear curve made from hemin standards similarly normalized to the internal standard.

Cells were seeded in 96-well plates and drugs (DHA, BTdCPU, ABT-199, ABT263 solubilized in DMSO or DMSO vehicle controls) were added at the indicated concentrations. Aminolevulinic acid (ALA), a precursor of heme synthesis, or succinylacetone (SA), an inhibitor of heme synthesis (Sigma), were solubilized in water and added at indicated concentrations (30). Cell viability was determined by staining with Annexin-V-APC and propidium iodide (BD Biosciences) and measured by flow cytometry as described previously (22).

Response surface modeling, implemented in Matlab version R2016a (Mathworks), was used to determine changes in the response of two drugs given in combination (31–33). A drug combination was considered either synergistic or antagonistic if the interaction term (α) describing the change in response relative to the additive model was either positive or negative, respectively. Two interaction terms (α) were considered different if their difference was statistically different from zero based on a two-tailed z-test.

Leukemia from adult patients with BCR-ABL1⁺ and EBF1-PDGFRB Ph-like ALL obtained from the Eastern Cooperative Oncology Group E2993 study (ClinicalTrials.gov identifier NCT00002514) and from the University Health Network were transplanted into un-irradiated immunodeficient NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice (The Jackson Laboratory) for 8 to 10 weeks prior to reisolation (34–36). Mice were bred and utilized in accordance with SJCRH animal care and use committee (SJCRHACUC).

BCR-ABL⁺ B-ALL were injected (2 × 10⁵) into nonconditioned, 6- to 8-week-old, female C57BL/6 recipients (The Jackson Laboratory). Five days after the transfer, recipients were treated with BTdCPU via intraperitoneal route and ABT-263 by oral gavage. Navitoclax was formulated in a mixture of 60% Phosal 50 PG, 30% PEG 400, and 10% EtOH and dosed at 100 mg/kg/day as described previously (37). BTdCPU (400 mg/kg/day) was administered in 30 μL DMSO. Treatment was given daily for 14 days (days 5–18 after leukemia injection) during and after which the mice were monitored. Mice were bred and utilized in accordance with SJCRHACUC.

All tissues were fixed in formalin, embedded in paraffin, sectioned at 4 μmol/L, mounted on slides (Superfrost Plus; Thermo Fisher Scientific), and dried at 60°C for 20 minutes before dewaxing and staining with hematoxylin and eosin (H&E) using standard methods. For IHC staining, the primary antibodies used in this study included anti-B220 (BD Biosciences) and anti-PAX5 (Abcam). Sections underwent antigen retrieval in a prediluted Cell Conditioning Solution (CC1; Ventana Medical Systems) for 32 minutes, and the OmniMap anti-rabbit HRP Kit (Ventana Medical Systems) and ChromoMap DAB (Ventana Medical Systems) were used for detection. Sections were examined by a pathologist blinded to the experimental group assignments.

Previous efforts to mechanistically address how DHA triggered the repression of MCL-1 revealed a gene expression signature consistent with the endoplasmic reticulum (ER) stress pathway (22). To further interrogate this pathway, the canonical branches of the cellular ER stress pathway were genetically ablated in mouse p185⁺ Arf-null B-ALL (hereafter referred to as BCR-ABL⁺ B-ALL) using CRISPR/Cas9 targeting (Supplementary Fig. S1A). Despite the loss of genes encoding the IRE1, ATF6, or PERK branches of the ER stress pathway, DHA treatment still triggered MCL-1 repression, indicating that none of the canonical ER stress pathway signaling arms is singularly responsible for repressing MCL-1 expression (Supplementary Fig. S1A).

To elucidate which cellular signaling pathways are activated by DHA to induce cell death in mouse BCR-ABL⁺ B-ALL cells, an unbiased genetic screen was conducted in which Cas9-expressing mouse BCR-ABL⁺ B-ALL cells were stably transduced with the Brie knockout library (78,637 sgRNAs targeting 19,674 genes; ref. 24). The cells were treated with 10 μmol/L DHA for 24 hours and cultured in drug-free media for an additional 48 hours to allow outgrowth of resistant cells (Fig. 1A). After DHA treatment, BCR-ABL⁺ B-ALL cell viability was approximately 40% in contrast to control-treated cells that maintained >95% cell viability (Fig. 1B). After treatment, genomic DNA was isolated from the viable BCR-ABL⁺ B-ALL cells, amplified and subjected to Illumina sequencing to determine sgRNAs enrichment in the surviving cells. MAGeCK identified 35 genes whose targeting sgRNAs were enriched in DHA-resistant cells at a FDR of <1% (25), indicating that loss of expression of these genes generated resistance to 10 μmol/L DHA (Fig. 1C; Table 1). Gene set enrichment analysis was performed (26, 27) using these 35 genes to identify genetic pathways implicated in DHA resistance. Multiple pathways related to heme synthesis/metabolism and apoptosis were identified (Fig. 1D; Table 1); these data implicate that either the heme synthesis or intrinsic apoptotic pathway is required for induction of apoptosis in BCR-ABL⁺ B-ALL cells in response to 10 μmol/L DHA.

A secondary screen was performed using a validation library composed of five, new sgRNA guides per gene for the top 35 hits in positive selection (FDR < 1%) and the top 13 hits in negative selection (FDR < 16%). After MAGeCK analysis, 8 of the 10 genes involved in heme synthesis and 4 of the 7 apoptosis-related genes identified in the genome-wide screen were found to be top hits in the validation screen (FDR < 12%; Fig. 1E). These data strongly suggest that DHA requires the heme synthesis pathway to kill BCR-ABL⁺ B-ALL cells.

Because sgRNAs targeting the heme synthesis pathway were associated with resistance of BCR-ABL⁺ B-ALL cells to DHA treatment, we hypothesized that repressing heme levels pharmacologically should render BCR-ABL⁺ B-ALL cells resistant to DHA treatment. To test this hypothesis, we cultured wild-type or MCL-1 overexpressing (OE) BCR-ABL⁺ B-ALL cells in SA, an inhibitor of heme synthesis (30). This rendered the wild-type cells resistant to death induced by culture with DHA (Supplementary Fig. S1B; Fig. 1F and G). Conversely, wild-type or MCL-1 OE BCR-ABL⁺ B-ALL cells were cultured in ALA, a precursor of heme synthesis (30), to increase cellular heme levels. The ALA treatment made the wild-type, but not MCL-1 OE, cells more sensitive to DHA-induced apoptosis (Fig. 1F and G). DHA, alone or in combination with ALA or SA, had little effect on MCL-1 OE cell viability, indicating that the changes in toxicity seen in wild-type cells when DHA was combined with ALA or SA are due to changes in endogenous MCL-1 expression. Because DHA acts to repress MCL-1 translation in leukemic cells (22), we tested whether pharmacologic manipulation of cellular heme levels could affect MCL-1 repression in response to DHA treatment. The combination of ALA and DHA caused MCL-1 repression like that observed with DHA alone (Fig. 1H). In contrast, the inhibition of heme synthesis by SA substantially blunted MCL-1 repression triggered by DHA (Fig. 1H). These data indicate that cellular heme is needed both for DHA to induce apoptosis and to repress MCL-1 expression.

The observation that heme is required for DHA-induced apoptosis and repression of MCL-1 prompted us to further examine how the heme pathway could trigger the translational repression of MCL-1. The phosphorylation of eIF2α inhibits cap-dependent translation by triggering the integrated stress response (38). Indeed, when BCR-ABL⁺ B-ALL cells were treated with DHA, eIF2α was phosphorylated and both ATF4 and CHOP were induced (Fig. 2A). Concomitantly, MCL-1 expression was repressed, but the expression of other antiapoptotic BCL-2 family members, such as BCL-XL or BCL-2, were unaffected (Fig. 2A). There was an increase in the proapoptotic BCL-2 family members NOXA and PUMA in response to DHA; however, BIM expression was unchanged (Fig. 2A). In addition, DHA, alone and in combination with ALA, induces the integrated stress response, as shown by the phosphorylation of eIF2α and induction of its downstream targets, ATF4 and CHOP (Fig. 1H). Decreasing cellular heme levels with SA treatment blunted the induction of the integrated stress response by DHA (Fig. 1H). There are four known eIF2α kinases, GCN2, PKR, PERK, and the heme-regulated inhibitor (HRI; ref. 39). PERK-deficient cells still repressed MCL-1 in response to DHA treatment (Supplementary Fig. S1A), thus excluding PERK and the classical ER stress pathway as being responsible for eIF2α phosphorylation triggered by DHA.

The eIF2α kinase HRI phosphorylates eIF2α in erythrocytes to prevent hemoglobin translation when cellular heme levels are limiting (40). To assess whether HRI is responsible for MCL-1 repression in BCR-ABL⁺ B-ALL cells treated with DHA, we generated Eif2ak1-deficient (Eif2ak1 encodes HRI, hereafter referred to as HRI-deficient) BCR-ABL⁺ B-ALL cells using CRISPR/Cas9. In contrast to wild-type BCR-ABL⁺ B-ALL cells, HRI-deficient BCR-ABL⁺ B-ALL cells, failed to induce either ATF4 or CHOP when treated with DHA (Fig. 2B). Notably, while DHA treatment in HRI-expressing BCR-ABL⁺ B-ALL cells triggered MCL-1 repression, the BCR-ABL⁺ B-ALL cells lacking HRI failed to repress MCL-1 expression (Fig. 2B and D). Furthermore, the addition of ISRIB, a small-molecule inhibitor of the integrated stress response (41), blocked MCL-1 repression and attenuated the induction of the integrated stress response (indicated by ATF4) in response to DHA treatment (Supplementary Fig. S2A). Although cellular heme depletion is one mechanism of integrated stress response induction, 24 hours of SA treatment alone was insufficient to activate the integrated stress response; however, culture with SA for 48 hours resulted in the HRI-dependent induction of MCL-1 repression and activation of the integrated stress response (Supplementary Fig 2B). These data indicate that HRI is required for the DHA-induced eIF2α phosphorylation and activation of the downstream integrated stress response, which represses MCL-1 expression.

In erythrocytes, HRI directly binds to heme, which represses kinase activity; however, when heme levels drop, HRI becomes active and phosphorylates eIF2α (42, 43). To address how DHA triggers HRI activation, we assessed the HRI-heme interaction using spectroscopy on purified, recombinant proteins. When HRI and heme were incubated together, a Soret peak was detected indicating binding of the two proteins. However, when DHA was incubated with heme and HRI, the Soret peak was lost indicating that DHA addition disrupts the interaction between heme and HRI (Fig. 2C). These data indicate that DHA can disrupt the HRI–heme interaction in a cell-free system.

HRI-mediated phosphorylation of eIF2α represses cap-dependent translation and blocks a wide array of protein synthesis (38). To confirm whether active HRI is upstream of the MCL-1 translational repression, wild-type or HRI-deficient BCR-ABL⁺ B-ALL cells were treated with DHA for 9 hours and then labeled with 100 μCi/mL of ³⁵S-labeled methionine and cysteine for 1 hour. Autoradiography of lysates from treated cells demonstrated that global translation was repressed in DHA-treated control cells; however, the repression of translation was substantially blunted in HRI-deficient cells in response to DHA treatment (Fig. 2D). As a positive control, culture of both cell types with cycloheximide repressed translation in an HRI-independent manner. Immunoblotting of the treated cells also revealed that MCL-1 was repressed in the control cells, but not the HRI-deficient BCR-ABL⁺ B-ALL cells (Fig. 2D). These data indicate that DHA treatment induces an HRI-dependent inhibition of translation that results in MCL-1 repression.

We previously demonstrated that DHA-induced MCL-1 repression rendered BCR-ABL⁺ B-ALL cells susceptible to synergistic cell death induced by the BH3-mimetic drug ABT-263 (22). Because HRI mediates the repression of MCL-1 in response to DHA treatment, we assessed whether HRI-deficient BCR-ABL⁺ B-ALL cells would exhibit less synergistic cell death in response to combination treatment of DHA with BH3-mimetic drugs. In control BCR-ABL⁺ B-ALL cells DHA synergized with both ABT-199 and ABT-263 as evidenced by response surface modeling (Fig. 2E; Supplementary Fig. S3A and S3B). However, the loss of HRI from BCR-ABL⁺ B-ALL cells substantially blunted the synergistic killing between DHA and both BH3-mimetics (Fig. 2E; Supplementary Fig. S3A and S3B). Although synergy with DHA and BH3-mimetics was diminished by loss of HRI, there was still sensitization to BH3-mimetics, indicating that DHA has HRI-independent effects as well. MCL-1 overexpression also attenuated the synergism between DHA and BH3-mimetics, indicating that the synergy depends on the expression levels of MCL-1 (Supplementary Fig. S3C). DHA treatment did not induce significant changes in expression of other antiapoptotic molecules, including BCL-2 or BCL-XL; however, the proapoptotic proteins NOXA and PUMA, which are known to bind to MCL-1, were induced (Fig 2A). These data indicate that the synergy between DHA and BH3-mimetics depends mainly on changes in MCL-1 expression and not on other antiapoptotic proteins.

DHA repressed MCL-1 in an HRI-dependent manner, so we asked whether a small-molecule activator of HRI, BTdCPU (44), could also induce the repression of MCL-1. Like DHA treatment, when BCR-ABL⁺ B-ALL cells were treated with BTdCPU MCL-1 expression was repressed whereas BCL-XL expression was unaffected (Fig. 3A). Consistently, much less MCL-1 repression was induced by BTdCPU treatment in the HRI-deficient leukemic cells, indicating that BTdCPU induced an HRI-dependent repression of MCL-1 (Fig. 3A). In addition, BTdCPU triggered the integrated stress response and induced the downstream targets CHOP and ATF4 in wild-type, but not in HRI-deficient BCR-ABL⁺ B-ALL cells (Fig. 3A). These data indicate that BTdCPU can activate HRI and induce MCL-1 repression.

Because BTdCPU can induce MCL-1 repression in an HRI-dependent manner, we assessed whether it can synergize with BH3-mimetics in killing BCR-ABL⁺ B-ALL cells. When BTdCPU was combined with either ABT-199 or ABT-263, it induced a synergistic cell death response in wild-type BCR-ABL⁺ B-ALL cells (Fig. 3B and C; Supplementary Fig. S4). However, in HRI-deficient BCR-ABL⁺ B-ALL cells the synergistic combination of BTdCPU with both BH3-mimetics was significantly attenuated (Fig. 3B and C; Supplementary Fig. S4). Although synergy with BTdCPU and BH3-mimetics was diminished by loss of HRI, there was still sensitization observed indicating that BTdCPU has HRI-independent effects. These data indicate that the activation of HRI by BTdCPU can induce synergistic leukemia cell killing in our murine cell model of BCR-ABL⁺ B-ALL.

To assess whether activation of HRI can synergize with BH3-mimetic agents in human Ph⁺ leukemia cells, BV-173 cells, were cultured with either DHA or BTdCPU. Like our observations in mouse BCR-ABL⁺ B-ALL cells, treatment of human leukemia cells with DHA or BTdCPU-induced eIF2α phosphorylation, expression of CHOP and ATF4, and MCL-1 repression without altering BCL-XL expression (Fig. 4A). To determine whether DHA-induced MCL-1 repression in human Ph⁺ acute leukemia cells was also HRI-dependent, EIF2AK1 (encoding HRI) was ablated in BV-173 Ph⁺ leukemia cell lines by CRISPR/Cas9 and the cells were subjected to treatment with DHA, BTdCPU, or thapsigargin (an inducer of the classical ER stress pathway; ref. 45). The loss of HRI expression prevented the repression of MCL-1 expression in cells cultured with DHA or BTdCPU and blocked the repression of ATF4 and CHOP (Fig. 4B). In contrast, thapsigargin treatment still repressed MCL-1 and induced ATF4 and CHOP in BV-173 cells lacking HRI, confirming that the ER stress pathway does not depend on HRI (Fig. 4B). These data indicate that activation of HRI-mediated eIF2α phosphorylation pathway, by either DHA or BTdCPU treatment, can cause MCL-1 repression and therefore should be able to synergize with BH3-mimetics in human Ph⁺ leukemia cells. Consistent with these findings, when BV-173 cells were cultured with either DHA or BTdCPU and combined with ABT-199 or ABT-263, the combined treatments were synergistic with both BH3-mimetics, but in BV-173 cells lacking HRI the synergism was significantly reduced (Fig. 4C; Supplementary Fig. S5A and S5B). Similar synergism between either DHA or BTdCPU and BH3-mimetics were obtained in human TOM1, OP-1, and SUP-B15 Ph⁺ leukemia cell lines treated in culture (Supplementary Fig. S5C–S5E; Supplementary Table S2).

To extend our findings to other poor prognosis subtypes of ALL, we tested DHA and BTdCPU in a mouse cell line expressing a fusion oncoprotein driver of Ph-like ALL. Mouse Ba/F3 cells stably expressing a human PAX5-JAK2 fusion were treated with DHA or BTdCPU to determine whether MCL-1 was repressed (23). Consistent with Ph⁺ leukemia cell lines, we found that MCL-1 expression was repressed in response to DHA or BTdCPU in PAX5-JAK2 Ba/F3 cells (Fig. 4D). In addition, eIF2α phosphorylation was induced, ATF4 and CHOP were expressed, and BCL-XL protein was unaffected by treatment with either DHA or BTdCPU. Because MCL-1 was repressed in the PAX5-JAK2 Ba/F3 cells, we assessed whether there was a synergistic effect when DHA or BTdCPU was combined with BH3-mimetics. As expected, synergy was observed in the PAX5-JAK2 Ba/F3 cells when DHA or BTdCPU was combined with BH3-mimetics (Fig. 4E; Supplementary Fig. S5F and S5G). In other Ba/F3 cells expressing the RCSD1-ABL1 or RCSD1-ABL2 Ph-like fusions (23), both DHA and BTdCPU also synergized with BH3-mimetics (Supplementary Fig. S5H and S5I; Supplementary Table S2).

To confirm whether DHA or BTdCPU can lead to synergistic responses when combined with ABT-199 or ABT-263 BH3-mimetics in primary patient leukemia, we took advantage of primary patient-derived xenografts (PDX) established from patients with Ph⁺ or EBF1-PDGFRB expressing Ph-like B-ALL. PDX cells isolated from recipient mice were cultured overnight with either DHA or BTdCPU and either ABT-199 or ABT-263 BH3-mimetics. Like Ph⁺ human cell lines, the combination of either BH3-mimetic agent synergized with both DHA and BTdCPU in Ph⁺ B-ALL PDX samples (Fig. 5A; Supplementary Fig. S6A). Consistently, EBF1-PDGFRB Ph-like PDX cells responded synergistically to the combination of DHA or BTdCPU with either BH3-mimetic (Fig. 5B; Supplementary Fig. S6B). In addition, in Ph⁺ and Ph-like PDX cells treated with DHA or BTdCPU MCL-1 was repressed and ATF4 was induced (Fig. 5C and D). These data indicate that primary patient leukemic cells can respond to either DHA or BTdCPU-induced synergism with BH3-mimetics agents in culture.

Because combining HRI activation by BTdCPU synergized with BH3-mimetics in cultured mouse and human leukemia, we hypothesized that combining BTdCPU with ABT-263 could prolong survival of mice bearing BCR-ABL⁺ B-ALL. To this aim, C57/BL6 mice were injected with BCR-ABL⁺ B-ALL cells followed by treatment on day 5 after transplant with either vehicle, BTdCPU (400 mg/kg), ABT-263 (100 mg/kg), or a combination of BTdCPU and ABT-263. Daily treatments continued for 14 days during which mice were monitored. Mice receiving vehicle, BTdCPU or ABT-263 alone exhibited rapidly progressing leukemia requiring euthanasia within 13 days. In contrast, recipients treated with the combination of BTdCPU and ABT-263 for 14 days exhibited significantly prolonged survival (Fig. 6A). Complete blood counts on day 9 revealed that mice receiving vehicle or BTdCPU alone had higher white blood cell counts (WBC), whereas those mice receiving the combination treatment had statistically lower WBCs (Fig. 6B). These data indicated that the combination of BTdCPU and ABT-263 reduced the leukemia burden in recipient mice.

BTdCPU triggered the repression of MCL-1 in cultured human and mouse leukemia; therefore, we investigated whether BTdCPU could induce the repression of MCL-1 expression in mice by treating leukemia bearing recipient mice daily with BTdCPU. After 8 days of treatment, the expression of MCL-1 in the bone marrow from leukemic mice treated with BTdCPU was significantly decreased when compared with that from mice treated with vehicle alone, whereas BCL-XL expression was unaffected by BTdCPU treatment (Fig. 6C). Furthermore, cells from mice treated with BTdCPU also exhibited elevated expression of ATF4 indicating the eIF2α pathway was activated (Fig. 6C).

Despite 14 days of treatment, the combination treatment group eventually succumbed to fatal leukemia as determined by IHC staining for expression of Pax5 and B220 on B-ALL cells (Fig. 6D). We therefore investigated whether the leukemia in mice treated with the combination of BTdCPU and ABT-263 had acquired treatment resistance. To complete this analysis, bone marrow from vehicle or combination-treated mice were harvested, expanded, and then re-treated in culture in a synergy assay. Leukemia cells isolated from both mice treated with vehicle and the combination of BTdCPU and ABT-263 still showed a synergistic response to the treatment when treated ex vivo, indicating that the leukemia had not acquired resistance during the in vivo treatment (Fig. 6E; Supplementary Fig. S7A). These data indicate that when combined with BH3-mimetic agents the activation of the HRI signal transduction pathway in vivo can slow leukemia progression.

The observation that MCL1 is frequently amplified in human cancers has driven efforts to develop potent and specific MCL-1 inhibitors (9). Several candidate MCL-1 inhibitors have shown on-target efficacy when tested in human hematologic malignancies in culture and xenograft models (17–19, 46). Several of these agents are in early clinical trials. Despite some promising results in patients, in the fall of 2019 the FDA placed a clinical hold on the phase I trial of one MCL-1 inhibitor (AMG397) due to evidence of cardiac toxicity. In addition, mouse genetic evidence has shown that ablation of Mcl1 in cardiomyocytes induces fatal cardiomyopathy (20, 21). These studies suggest that the direct inhibition of MCL-1 may have safety concerns in the clinic, especially when combined with standard chemotherapy that causes significant cardiac toxicity, highlighting the importance of discovering alternative mechanisms of targeting MCL-1.

Previous work from our laboratory identified that the widely used, antimalarial drug DHA triggers the repression of MCL-1 expression in mouse and human leukemic cells and synergizes with ABT-263 in culture and in vivo (22). Here, we provide mechanistic insight into how DHA induces the repression of MCL-1 translation. These efforts have revealed for the first time how a cellular heme-sensing pathway controls apoptotic sensitivity through the function of the HRI eIF2α kinase. In mouse and human leukemia, DHA treatment triggers the HRI-dependent integrated stress response which results in eIF2α phosphorylation and represses cap-dependent translation. Because MCL-1 protein is labile and undergoes rapid proteasome-dependent turnover, inhibiting new protein translation rapidly results in the loss of MCL-1 expression (47). Here, we demonstrate that HRI loss in mouse and human leukemia cells blocks the DHA-induced MCL-1 repression as well as the induction of downstream effectors ATF4 and CHOP. Notably, other prosurvival molecules like BCL-2 and BCL-XL expression were unchanged by DHA treatment. We observed the induction of the proapoptotic molecules NOXA and PUMA, which have been previously shown to be induced by the integrated stress response but did not detect changes in proapoptotic BIM expression (48). Both NOXA and PUMA can bind MCL-1 and may contribute to the cell death response. Although PKR, GCN2, PERK, and HRI can regulate eIF2α phosphorylation, our data demonstrate that loss of HRI alone prevents DHA-induced MCL-1 repression and activation of the integrated stress response. Therefore, other eIF2α kinases are incapable of replacing HRI's function in response to DHA. These data clearly implicate the HRI heme-sensing pathway as a key trigger of the integrated stress response resulting in the inhibition of MCL-1 translation.

MCL-1 is a known resistance factor for ABT-199 and ABT-263, but DHA can repress MCL-1 expression in mouse and human leukemia, thus rendering the cells more sensitive to treatment with either BH3-mimetic. Importantly, in both mouse and human leukemia the loss of HRI significantly reduced the synergistic response to these BH3-mimetics when combined with DHA demonstrating the impact of this heme-sensing pathway on apoptotic sensitivity. These data led us to wonder if other pharmacologic activators of HRI could similarly repress MCL-1 expression. To address this, we tested BTdCPU, a small-molecule activator of HRI (44). Like DHA, BTdCPU, lead to MCL-1 repression and synergized with ABT-199 and ABT-263 BH3-mimetics in mouse and human leukemia. Despite attenuating the synergism, both DHA and BTdCPU still sensitized HRI-deficient leukemic cells to BH3-mimetics, indicating that both these compounds clearly have HRI-independent effects outside inducing the integrated stress response. Consistent with in vitro data, we demonstrated that this pathway also repressed MCL-1 in vivo without any overt toxicity. This confirms the idea that activation of this heme-sensing pathway could be used to target MCL-1 and promote leukemic priming to BH3-mimetics.

The concept that perturbations of heme synthesis can affect apoptotic responses has precedent. In a metabolically focused CRISPR loss-of-function screen, acute myeloid leukemia (AML) cell lines were sensitized to ABT-199–induced killing when several heme synthesis genes were ablated (49). Wood and colleagues concluded that loss of heme synthesis exerted its apoptosis-sensitizing effects largely by disruption of electron transport chain function and loss of mitochondrial outer membrane integrity (49). However, precisely how disruptions in electron transport chain function contribute to permeabilization of the outer mitochondrial membrane was not clear from this study. Our data indicate that while simply inhibiting heme synthesis with SA for 24 hours was insufficient to activate the integrated stress response and repress MCL-1 expression, longer treatment (48 hours) with SA alone repressed MCL-1 expression in an HRI-dependent manner. This HRI-dependent loss of MCL-1 induced by repressing heme levels would promote cancer cell responses to BH3-mimetics such as ABT-199.

We reveal that the antimalarial DHA requires heme synthesis to induce the HRI-dependent integrated stress response. In contrast, Wood and colleagues showed that blocking heme synthesis sensitized AML cell lines (49). However, we demonstrate that in a cell-free system DHA can physically disrupt the interaction between heme and HRI. In killing malarial parasites, the endoperoxide bridge in DHA can react with the iron in heme to break the bond and produce reactive oxygen species (50). This reaction can lead to the formation of a covalent complex between heme and DHA (51). We hypothesize that this reaction could lead to sequestration of cellular iron and displacement of heme from HRI. Precisely how DHA disrupts the HRI and heme interaction, and why this process requires cellular heme is still unclear and will be the topic of further investigation. Our previous publication demonstrated that DHA treatment does induce ROS production, but antioxidant treatment could not prevent the death of leukemic cells or the repression of MCL-1, indicating that ROS production alone is insufficient (22). By better understanding the mechanisms of heme interactions and HRI activation one could imagine that it should be possible to develop more specific and potent drugs to target this pathway that could be used to treat a variety of types of malignancies.

Ph⁺ ALL has seen improvements in treatment; however, there is still a need for new treatment strategies as the intensive regimens have significant side-effects and are not effective in all patients (3, 4). Our work implicates an alternate therapeutic pathway that can be targeted in this subtype of leukemia. Perhaps even more importantly, we have shown that this pathway can also be targeted in Ph-like ALL, a harder to treat subtype known to have a poorer prognosis (5). We observed MCL-1 repression by both DHA and BTdCPU in mouse cell lines as well as human leukemia cells from both Ph⁺ and Ph-like ALL models. To continue to evaluate the potential of this pathway, we look forward to further validating these findings by combining DHA or BTdCPU with BH3-mimetics in NSG recipients bearing Ph⁺ or Ph-like primary patient leukemia.

We are interested in determining whether other types of cancers utilize this HRI-dependent heme-sensing pathway for regulating apoptotic sensitivity. Because cancer cells frequently have elevated levels of heme and often exhibit upregulated expression of many of the genes involved in the biosynthesis of heme (52), it is likely that our observations will be generalizable to cancer beyond B-ALL. Indeed, inhibition of heme synthesis has been reported to reduce tumor cell survival and proliferation in a variety of cancer types (30, 53, 54). The effects of inhibiting heme synthesis may be in part due to disruption of normal cellular processes including electron transport chain function, cataplerosis in the TCA cycle, p53 activity and stability, regulating trafficking of ADP and ATP, and in circadian rhythms (52). However, our novel findings have revealed mechanistically how activation of the cellular heme-sensing pathway can regulate apoptotic priming in leukemia.

It is too early to tell how much the development of potent and selective BH3-mimetic inhibitors of MCL-1 will be hampered by toxic effects induced by MCL-1 inhibition. Despite these setbacks, MCL-1 remains an important therapeutic target for cancer therapy (9). Our data indicate that simply repressing MCL-1 expression does not trigger any overt toxicities when used in mouse models (22). It is conceivable that partially repressing MCL-1 may cause a less severe insult to normal tissues than potent, pharmacologic inhibition and, thus, may avoid the associated toxicities. We speculate that activation of this HRI-dependent heme-sensing pathway could be an effective and well-tolerated alternative approach to repress MCL-1 expression and render cancer cells more susceptible to approved BH3-mimetics.

J.C. Panetta reports grants from NIH NCI P30 CA 21765 during the conduct of the study. C.G. Mullighan reports personal fees from Illumina, grants from Loxo Oncology, AbbVie, and Pfizer, and personal fees from Amgen outside the submitted work. J.T. Opferman reports grants from NIH- R01HL102175, NIH- R01CA201069, NIH- R35CA197695, NIH- P30CA021765, and grants and other from ALSAC during the conduct of the study, as well as other from AbbVie outside the submitted work. No disclosures were reported by the other authors.

K.H. Smith: Conceptualization, formal analysis, validation, investigation, writing–original draft, writing–review and editing. A. Budhraja: Conceptualization, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. J. Lynch: Investigation, methodology. K. Roberts: Resources. J.C. Panetta: Software, formal analysis, validation, visualization. J.P. Connelly: Resources, software, validation, methodology. M.E. Turnis: Resources, investigation, writing–review and editing. S.M. Pruett-Miller: Resources, software, formal analysis, supervision, validation, methodology, writing–review and editing. J.D. Schuetz: Resources, supervision, validation, writing–review and editing. C.G. Mullighan: Resources, funding acquisition, writing–review and editing. J.T. Opferman: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

We thank J.-J. Chen (MIT) for providing anti-HRI purified antibody and J. Slavish and Z. Rankovic (Department of Chemical Biology and Therapeutics, SJCRH) for synthesis of the small molecule, BTdCPU. We also thank S. Porter, S. Peters, and S. Sakurada (Center for Advanced Genome Editing, SJCRH) for generation of mutant cell lines. This research was supported by the American Lebanese Syrian Associated Charities and NIH grants to J.T. Opferman (R01HL102175 and R01CA201069); C.G. Mullighan (R35CA197695); and The Center for Advanced Genome Editing is supported by P30CA021765. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.