Physiologic hypoxia promotes maintenance of CML stem cells despite effective BCR-ABL1 inhibition

Chronic myeloid leukemia (CML) is a disease of hematologic stem cells caused by the reciprocal translocation t(9:22).^1,2  This produces the chimeric breakpoint cluster region (BCR)–c-abl oncogene 1, nonreceptor tyrosine kinase (ABL1) oncoprotein that drives the pathogenesis of CML, and as a consequence, BCR-ABL1 represents the primary target of therapy.¹  The development of the tyrosine kinase inhibitor (TKI), imatinib mesylate (imatinib, Gleevec), has drastically improved the treatment of CML with the majority of chronic phase (CP) patients achieving complete cytogenetic responses.³  However, although imatinib is effective in controlling CML and preventing progression to blast crisis (BC), it is not curative. Most patients who achieve complete cytogenetic responses continue to have detectable BCR-ABL1 transcripts,⁴  and in almost all cases, CML stem cells can still be detected in patient bone marrow (BM).⁵  Even for patients with complete molecular responses and no detectable BCR-ABL1 transcripts, cessation of TKI therapy frequently results in disease relapse.^6,7  This suggests that despite TKI treatment, leukemic stem cells (LSCs) are able to survive and persist in patients, necessitating long-term TKI therapy. Identifying the mechanisms that support the persistence of LSCs is therefore of utmost importance for their eradication.

The mechanism of LSC persistence is not well-characterized. Although BCR-ABL1-dependent factors, including kinase domain mutations⁸  and BCR-ABL1 overexpression,⁹  have been proposed to play a part, this is not conclusively supported by both in vitro work and clinical observations. For example, several studies have shown that primitive quiescent CML cells are intrinsically resistant to TKIs.^10,11  Further, in patients with stable complete cytogenetic responses, the presence of BCR-ABL1-kinase domain mutations is infrequent and does not predict relapse.¹²  Together, these studies^10-12  indicate that LSCs may not require BCR-ABL1 for persistence and may thus resemble normal hematopoietic stem cells (HSCs) in their requirements for survival.¹³  Of note, studies in normal hematopoiesis have indicated that the BM microenvironment, which includes the stem cell niche, is crucial for normal HSC function.¹⁴  Accordingly, and because LSCs likely arise from HSCs or primitive progenitors, it is plausible that the BM microenvironment also contributes to LSC persistence. Indeed, recent studies indicate that physiologic growth factor signaling, by stimulating prosurvival pathways in CML progenitors, may render them independent of BCR-ABL1-mediated survival signals.^15-17 

Physiological hypoxia is said to occur when the oxygen (O₂) tension falls below 2%.¹⁸  In humans, including patients with leukemia, the average BM O₂ has been measured at 6% to 7%,^19,20  whereas in mouse, the oxygen tension is estimated to be below 1% in the stem cell niche.^21,22  Of note, hypoxia has long been regarded as a pathological phenomenon in ischemic solid tumors that is associated with radio- and chemoresistance.^23,24  Further, recent studies in embryonic¹⁸  and hematopoietic²⁵  systems have revealed that physiologic hypoxia also plays important roles in regulating stem cell functions.²⁶  Although hypoxia is emerging as a key factor in the biology of leukemia,²⁷  its role in CML LSCs and therapy resistance has not been thoroughly investigated. Although earlier studies have demonstrated that primary CML cells and CML cell lines can survive hypoxic conditions in vitro,^28-30  and that hypoxia confers both stem-like characteristics and imatinib resistance in cell lines,^29,30  it is unclear whether such features extend to primary CML LSCs.

To address these issues, we characterized the interplay between physiologic hypoxia and imatinib in primary human CML stem and progenitor populations, including their ability to function as LSCs. We found that although imatinib inhibits BCR-ABL1 effectively, hypoxia promotes CML cell maintenance, primarily through the upregulation of hypoxia-inducible factor 1 α (HIF1-α). We also uncover a set of prosurvival genes that are preferentially upregulated by hypoxia in CML versus normal cord blood CD34⁺ cells. Our study indicates that cosuppression of BCR-ABL1 and HIF1-α or HIF1-α-regulated targets represents a rational strategy for elimination of LSCs.

Primary CP CML samples were obtained from patients at the Singapore General Hospital. Written informed consent was obtained from all patients under protocols approved by the institutional review board. Cord blood samples were obtained from the Singapore Cord Blood Bank. This study was conducted in accordance with the Declaration of Helsinki.

Mononuclear cells were purified by Ficoll-Paque (Sigma-Aldrich) density gradient centrifugation. CD34⁺ cells were purified using either a CD34 MicroBead kit or CD34 multiSort/CD38 MicroBead kits (Miltenyi Biotec) and subsequently grown in serum-free media supplemented with cytokines.³¹  Hypoxic treatment (0.5% oxygen, 5% carbon dioxide) was performed in a humidified incubator (Innova CO-170; New Brunswick Scientific). Cell numbers were determined by trypan blue exclusion and flow cytometry.

Cells were plated in drug-free methylcellulose (H4434; StemCell Technologies). The number of colonies (colony-forming unit-granulocyte macrophages and colony-forming unit-erythroid) was scored 14 days later.

These were performed as previously described.³² 

Staining was performed according to protocols from Cell Signaling Technology. Cells were stained with Viability Dye eFluor 780 (eBioscience), followed by CD34-fluorescein isothiocyanate and CD38-allophycocyanin (APC) (BD Biosciences). Isotype and phosphorylated–v-crk avian sarcoma virus CT10 oncogene homolog-like (pCrkL) signals were established using normal rabbit immunoglobulin G (IgG; sc-3888; Santa Cruz) and anti-pCrkL (1:50 dilution; Tyr207; 3181; Cell Signaling Technology), respectively. Cells were then stained with a secondary antibody (Alexa Fluor 594 goat anti-rabbit IgG heavy + light chain ([H+L]); Invitrogen) and analyzed by flow cytometry.

Limiting dilution was performed using reagents and protocols from Stemcell Technologies. Cells were plated on irradiated M2-10B4 cells in 96-well plates at 5 different densities with 8 replicates each. At the end of 7 weeks of incubation, wells were scored as positive or negative for the presence of colonies, and the frequency of long-term culture-initiating cells (LTC-ICs) was calculated using L-Calc. Colonies from each condition were pooled and analyzed for the presence of BCR-ABL by fluorescence in situ hybridization (FISH).

All animal experiments were approved and performed in accordance with the guidelines required by the SingHealth Institutional Animal Care and Use Committee. CD34⁺ CML cells (2 × 10⁶ cells/mouse) were treated in vitro before they were injected via tail vein into irradiated (240 cGy) immunodeficient nonobese diabetic/severe combined immunodeficiency interleukin 2 receptor γ chain-deficient mice [NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ or NSG mice] The Jackson Laboratory). Mice were euthanized after 6 weeks, and the contents of BM (both femurs and tibias), peripheral blood, and spleen were checked for human engraftment by staining with anti-human CD45-APC and anti-mouse CD45-fluorescein isothiocyanate (both from BD Biosciences). Engrafted cells from each mouse were subjected to FISH analysis for the presence of BCR-ABL1.

Immunohistochemistry was performed using HIF1-α (ADI-OSA-602-E; Enzo) and HIF2-α (Ab8365; Abcam) antibodies. Immunofluorescence was performed on fixed cells on micro cover glasses (12 mm; VWR). Isotype and HIF1-α signals were established using normal rabbit IgG (sc-3888; Santa Cruz Biotechnology) and anti-HIF1-α (Ab51608; Abcam), respectively. Cells were than stained with a secondary antibody (Alexa Fluor 594 goat anti-rabbit IgG [H+L]), mounted (P36931 ProLong Gold antifade reagent; Invitrogen), and visualized by confocal microscopy (LSM710; Carl Zeiss).

Using the Illumina TotalPrep Amplification kit (Ambion), 120 ng total RNA was subjected to cDNA synthesis. The cDNA was purified using the MinElute Reaction Cleanup kit (Qiagen). In vitro transcription was performed on the cDNA to generate biotin-labeled cRNA, according to manufacturer recommendations, followed by purification using the RNeasy Mini kit (Qiagen). The integrity and size distribution of the cRNA libraries were verified by RNA Nano 6000 chip profiling on the 2100 Bioanalyzer (Aligent). High-quality cRNA libraries were then hybridized on Human HT-12 beadchips (Illumina) and scanned on the BeadArray Reader (Illumina) at scan factor 2. The raw microarray data were normalized using the cross-correlation method.³³  Differential gene expression was identified on the basis of a fold change cutoff of more than 1.5 compared with the dimethylsulfoxide (DMSO)/normoxia control for each patient and each time. The microarray data were deposited in the Gene Expression Omnibus database (accession number GSE48294).

Data were reported as mean ± standard error of the mean. Statistical significance was determined by Student t test.

Additional methods can be found in the supplemental information, available on the Blood Web site.

To investigate the effect of hypoxia on primary patient-derived CML progenitors, we first performed colony-forming assays to enumerate the number of committed CML progenitors under hypoxic conditions. CML cells were first treated under normoxic or hypoxic conditions, after which they were plated in drug-free methylcellulose, and colony-forming cells (CFCs) were counted at 14 days (Figure 1A-B). This 2-step procedure allowed us to assess the specific effect of the various conditions on CFCs exclusively during the initial culture period.³¹  We found that hypoxia by itself resulted in a 2-fold increase in the number of CFCs (Figure 1C; supplemental Table 1). In addition, hypoxia also enhanced the ability of CML progenitors to persist in the presence of imatinib by 2- to fivefold (Figure 1C). Importantly, we observed the same trend in CP samples obtained from peripheral blood (n = 6) or BM (n = 5), indicating that the effects of hypoxia are intrinsic to CML progenitors (Figure 1C). Furthermore, the increase in the number of CML progenitors was apparent even when cells were not subjected to the initial 96-hour in vitro culture period (supplemental Figure 1A). Interestingly, we also observed that the effect on CFCs was dependent on the degree of hypoxia, as lower oxygen concentrations (0.5% vs 5%) resulted in greater increases in CFC numbers (supplemental Figure 1B). In contrast, hypoxia and imatinib had no effect on cord blood CFCs (Figure 1C). Our findings demonstrate that hypoxia specifically enhances the in vitro maintenance of CML CFC function by twofold and that this is associated with a 2- to fivefold relative resistance to imatinib compared with normoxia.

To understand how hypoxia might be enhancing CFC maintenance and imatinib resistance, we next examined the growth of the CML CD34⁺CD38⁺ (enriched for committed progenitors) and CD34⁺CD38⁻ (enriched for stem cell progenitors) subpopulations under normoxic (21% O₂) or hypoxic (0.5% O₂) conditions. Here, we determined the change in cell numbers by trypan blue exclusion and flow cytometry after 4 days of culture. We found that both CD34⁺CD38⁺ and CD34⁺CD38⁻ CML progenitors exhibited 30% to 50% slower growth under hypoxia (Figure 2A). We also observed that imatinib greatly reduced the growth of CML cells, regardless of the oxygen concentration, although there was a small but statistically significant increase in viable CD34⁺CD38⁺ and CD34⁺CD38⁻ cells after imatinib treatment in hypoxic vs normoxic conditions (Figure 2A). Because changes in cell number may be caused by changes in either proliferation and/or cell death, we next monitored the degree of apoptosis by Annexin V and PI staining. We found that hypoxia did not significantly alter the proportion of apoptotic cells in the absence of drug but that it consistently suppressed imatinib-induced apoptosis in both CD34⁺CD38⁻ and CD34⁺CD38⁺ populations at clinically achievable imatinib concentrations (Figure 2B).³⁴  To exclude the possibility that hypoxia was impairing the ability of imatinib to inhibit BCR-ABL1 kinase, we also examined the level of pCrkL, a surrogate marker for BCR-ABL1 kinase activity,¹¹  in CML progenitors under hypoxia. In both CD34⁺CD38⁻ and CD34⁺CD38⁺ subpopulations, imatinib reduced the level of pCrkL to a similar extent, regardless of oxygen concentration (Figure 2C-D). Next, we used carboxyfluorescein diacetate succinimidyl ester tracing to measure proliferation.³¹  Consistent with the decreased growth observed under hypoxia, we found that both CD34⁺CD38⁻ and CD34⁺CD38⁺ CML progenitors proliferated less at 0.5% vs 21% O₂ and that the addition of imatinib further inhibited cell division (supplemental Figure 2). Together, our findings indicate that hypoxia inhibits the proliferation of committed and primitive CML progenitors while also impairing the ability of imatinib to induce apoptosis in these cells, despite effective BCR-ABL1 inhibition.

Colony-forming assays reflect primarily the clonogenic capacity of more mature committed progenitors, but not that of the more primitive progenitors, which are enriched for LSCs. To test whether hypoxia promotes the maintenance of primitive progenitors, we performed LTC-IC-limiting dilution assays.³¹  Here, we treated CML cells and incubated them on stromal layers for an additional 5 weeks before plating for colony formation. In line with the colony-forming assay results, hypoxia increased the progenitor frequency, both in the presence or absence of imatinib (Figure 3A, n = 2; supplemental Table 1). We confirmed that the colonies recovered from the LTC-IC were of CML origin, as determined by BCR-ABL1 FISH (Figure 3A, 87%-100%; average, 96%). Interestingly, hypoxia alone also increased the progenitor frequency of cord blood cells (Figure 3B, n = 2), suggesting that hypoxia promotes the maintenance of both normal cells and LSCs.

To further confirm the in vivo relevance of hypoxia on LSCs, we determined the effect of physiological hypoxia on the ability of CML LSCs to engraft NSG mice.^35,36  CD34⁺ CML cells from 5 different patients with CP CML (2 × 10⁶ cells/mouse) were treated in vitro for 96 hours before being injected into sublethally irradiated NSG mice. Six weeks after injection, the mice were euthanized and engraftment of human cells was assessed in BM, peripheral blood, and spleen. Among the 5 samples used, we were only able to successfully engraft CP CML cells from 2 samples (CML#3 and CML#4). For CML#3, we found that hypoxia consistently enhanced CML cell engraftment in all 3 hematopoietic compartments and that CML cells engrafted most efficiently in BM (Figure 3C; supplemental Figure 3). Specifically, we found that samples maintained in vitro at 21% oxygen exhibited impaired engraftment compared with samples that were injected immediately before any in vitro culture, and that hypoxia treatment maintained the ability of CML progenitors to engraft mice (Figure 3C). Interestingly, the engraftment ability of the imatinib-treated groups was consistently higher than that of the corresponding DMSO control groups. The decreased engraftment of the DMSO 21% O₂ control group suggested gradual loss of engraftment potential during the in vitro treatment, whereas engraftment potential could be maintained by either imatinib and/or hypoxia, at least for the CML cells of this particular patient. Again, the engrafted cells in each mouse were found to be of CML origin, as confirmed by BCR-ABL1 FISH (Figure 3D, 90%-100%; average, 97%). For CML#4, we again observed a trend that hypoxic incubation promotes BM engraftment in the imatinib-treated group (0.36% in imatinib-treated cells grown in 21% O₂ vs 1.25% in imatinib-treated cells grown in 0.5% O₂; P = .16) (Figure 3C). Taken together, our in vitro and in vivo studies of CML stem cell function, despite patient-to-patient variability with respect to engraftment potential, are consistent with a role for physiologic hypoxia in CML stem cell maintenance. Next, we set out to understand the molecular underpinnings of this process.

The cellular hypoxic response is primarily driven by HIFs, and both HIF1-α and HIF-2α play key roles in cancer stem cells.^37,38  We therefore sought to analyze HIF expression in CML BM from patients in CP and BC, using immunohistochemistry. In CP samples, we observed scattered HIF1-α expression in cells corresponding to immature myeloid precursors, whereas in BC samples, HIF1-α expression was more widespread and consistent with the increased blast population in BC (Figure 4A; supplemental Figure 4A; antibody controls in supplemental Figure 4B). Interestingly, we observed ubiquitous HIF2-α expression in both CP and BC BM (Figure 4A). We conclude that both HIF1-α and HIF2-α are expressed in CML progenitors in vivo.

Because HIF1-α has been suggested to play important roles in CML,^39,40  we next analyzed HIF1-α expression in CML progenitors in vitro. Using immunofluorescence and western blot, we found that activated nuclear HIF1-α could only be observed under hypoxia (Figure 4B-C; control in supplemental Figure 4C) and disappeared on reexposure to normoxia (supplemental Figure 4D), and further, that hypoxia-induced activation of HIF1-α was partially attenuated by imatinib (Figure 4B-C). We also found that HIF1-α downregulation by imatinib was largely posttranscriptional, as the level of HIF1-α transcripts remained relatively constant across treatments at different times (Figure 4D, n = 3). Several HIF1-α regulated genes, including BNIP3, VEGFA, ALDOC, and GLUT1,⁴¹  were consistently upregulated by short- and long-term hypoxia, an effect that was variably attenuated by imatinib (Figure 4D). Thus, at 96 hours, imatinib was mostly able to prevent the hypoxia-mediated increase in BNIP3, ALDOC, and GLUT1, but not that of VEGF (Figure 4D). Importantly, and consistent with the ability of imatinib to inhibit BCR-ABL1 in hypoxia (Figure 2C), imatinib consistently downregulated phosphorylated signal transducer and activator of transcription 5A (pSTAT5) and pCrkL (Figure 4C; supplemental Figure 4E) and WT1 expression, a gene known to be upregulated in CML cells in a BCR-ABL1-dependent manner (Figure 4D; supplemental Figure 4F).⁴²  Together, our data demonstrate that BCR-ABL1 augments hypoxia-induced HIF1-α activation in CML progenitors and that imatinib only partially attenuates HIF1-α activation and HIF target gene transcription.

To better understand the relationship between hypoxia and imatinib exposure at the gene regulation level, we performed microarray analysis on CP CD34⁺ CML progenitors. Cells were treated for 24 or 96 hours, and global transcriptional changes were identified after comparison with control treatments. Both imatinib and hypoxia alone induced disparate transcriptional changes, with imatinib regulating a much larger subset of targets than hypoxia (Figure 5A-B). During combined imatinib and hypoxia treatment, there was transactivation of both hypoxia and imatinib target genes, with the overall transcriptional profile being predominantly similar to that of imatinib alone (Figure 5A-B). There were few synergistic or antagonistic effects between imatinib and hypoxia, as evidenced by the small number of coregulated targets (4.38% and 6.01% of the total differentially expressed genes at 24 and 96 hours, respectively; Figure 5B). Importantly, many of the hypoxia-induced target genes were attenuated by imatinib, which concurs with our observation that imatinib inhibits BCR-ABL1 kinase activity effectively (Figure 2C; supplementary Figure 5) but also inhibits HIF1-α activity partially (Figure 4B-C).

Because hypoxia impairs imatinib-induced apoptosis and enhances clonogenicity, we determined the hypoxia-induced targets that might be responsible. Detailed examination of persistently upregulated targets by hypoxia at both 24 and 96 hours with and without imatinib highlighted the enhancement of 14 antiapoptotic/survival genes (Figure 5B, overlap between 0.5% O₂ and imatinib/0.5% O₂; supplemental Table 2). Next, we conducted gene ontology analysis on targets (Figure 5A, yellow boxes; Figure 5C; supplemental Table 3) that were predominantly dysregulated in hypoxia treatments and partially reduced by imatinib but could still be independently responsible for sustaining CML progenitors. Carbohydrate metabolism was the most significantly upregulated, a typical feature of glycolytic adaption under hypoxia in the acute 24-hour phase. Cell cycle and cellular growth and proliferation signatures were generally downregulated, which is consistent with the decreased cell numbers and proliferative index measurements we observed (Figure 2A; supplemental Figure 2). Interestingly, the biological functions of cell-to-cell signaling and interaction, as well as hematological system and development, were upregulated with an overrepresentation of genes involved in increased adhesion, migration, and infiltration of hematopoietic lineages in long-term cultures. Overall, this analysis indicates a possible role for hypoxia in leukemogenesis through the concerted transactivation of multiple survival and adhesion factors that may contribute to decreased apoptosis and improved colony-forming potential and engraftment.

Imatinib partially attenuates HIF1-α signaling, but residual HIF1-α transcriptional activity may still be sufficient to support the survival and maintenance of CML cells under hypoxia. To determine whether HIF1-α plays a role in supporting CML progenitors under imatinib and hypoxia, we knocked down HIF1-α expression under these conditions. Here, CML CD34⁺38⁻ cells were transduced with lentivirus-expressing control or HIF1-α-specific short hairpin RNAs (shRNAs; 2 validated clones, sh-HIF1-α 3811 and sh-HIF1-α 10819), followed by puromycin selection (Figure 6A). We were able to recover enough cells from those transduced with control and sh-HIF1-α 3811 (supplemental Table 4) and confirmed the specific knockdown of HIF1-α in these cells by quantitative real-time polymerase chain reaction (Figure 6B; supplemental Figure 6). As expected, hypoxia increased the number of CFCs in the control cells, but this was abolished in the HIF1-α knockdown cells, both in the presence and absence of imatinib (Figure 6C). These results suggest that HIF1-α is primarily responsible for the maintenance of CML progenitors under hypoxic conditions, both in the presence and absence of imatinib.

Because hypoxia enhances colony formation of CML progenitors but not normal progenitors (Figure 1C), as well as improves CML progenitor survival in the presence of imatinib (Figure 2B), our data suggested hypoxia may differentially upregulate prosurvival genes in CML vs cord blood progenitors. Specifically, we compared the expression of 27 hypoxia-induced genes (identified in Figure 5 and supplemental Table 3) between cord blood and CP CML samples. As depicted in Figure 7A, we found that 14 genes were equivalently upregulated in cord blood and CML cells, whereas 12 genes were preferentially upregulated in CML vs cord blood cells. Notably, among these 12 genes, 5 (ARG2, CHI3L1, FAM134B, KRT17, and CD6) have prosurvival functions and/or are associated with transformation and poorer clinical outcome in human cancers.^43-49  Next, we confirmed that a subset of these genes is dependent on HIF1-α for hypoxia-induced upregulation. As shown in Figure 7B, knockdown of HIF1-α resulted in abrogation of the hypoxia-induced upregulation of ARG2, BNIP3, SULF2, and NEDD9.

CML LSCs are thought to serve as a reservoir for resistance and relapse, and the inability of BCR-ABL1 TKIs to eliminate LSCs may explain why most patients require long-term therapy.⁵⁰  These notions suggest the existence of BCR-ABL1-independent factors mediating LSC maintenance and survival.^10-13  Indeed, recent work has identified extracellular signals from stromal cells and growth factors as important contributors to LSC survival that confer BCR-ABL1-independence.^11,13  Recent studies also underscore the importance of physiologic hypoxia in maintaining HSC function.²⁶  However, the relationship between hypoxia and CML LSCs remains poorly understood. In the current work, we explored the effect of physiologic hypoxia on CML progenitor and LSC function, as well as the relationship between hypoxia, LSC function, and imatinib exposure. By doing so, we identify physiologic hypoxia as an important contributor to CML LSC persistence, despite BCR-ABL1 inhibition, and suggest therapeutic approaches to extinguish LSC function.

As a first step in establishing therapeutic strategy for LSCs, we evaluated the effect of hypoxia on committed and primitive progenitor function, as well as LSCs in the presence and absence of imatinib. We found that hypoxia enhanced the ability of primary CP CML cells to function in all 3 assays and conclude that physiologic hypoxia contributes to CML progenitor maintenance and LSC function. By performing the experiments in the presence of clinically relevant imatinib concentrations, we find that hypoxia protects against imatinib-induced apoptosis in both CD34⁺CD38⁺ and CD34⁺CD38⁻ CML progenitors, despite BCR-ABL1 kinase inhibition, and that hypoxia maintains both LTC-IC formation and engraftment capacity. By assessing cell division of CD34⁺CD38⁺ and CD34⁺CD38⁻ CML progenitors, we find that hypoxia slows cell division. These observations are consistent with previous reports describing an association between increased proliferation or loss of quiescence and so-called stem cell exhaustion.^51-53  The decreased proliferation is also congruent with the idea that imatinib-resistant, quiescent CML progenitors are the populations that contribute to disease persistence.⁵⁴  Consistent with the effects of hypoxia on cell division, we found that hypoxia downregulated genes involved in cell cycle and cellular growth and proliferation.

Decreased intracellular levels of imatinib, possibly through the actions of drug transporter pumps, have been suggested as a mechanism for imatinib resistance.⁹  However, the above might not be relevant here, as BCR-ABL1 kinase activity remains suppressed by imatinib under hypoxia. To measure the effect of hypoxia on BCR-ABL1 kinase activity, we monitored the level of pCrkL in the CML stem and committed populations. We found that even under prolonged treatments (96 hours), hypoxia alone does not reduce the levels of pCrkL in both stem and committed progenitors (Figure 2). This is in contrast to previous reports that in CML cell lines, hypoxia reduces the level of BCR-ABL1 phosphorylation²⁹  or BCR-ABL1 protein.³⁰  The reason behind this discrepancy is unclear, but it may be a result of differences in model systems between CML cell lines and primary human CML cells. In support of this possibility, we found that CML cell lines were less able to form colonies under hypoxia (unpublished data), further emphasizing the difference in hypoxic responses between primary and CML cell lines.

Given that cells within the stem cell niche communicate through cytokines, and cytokines can induce BCR-ABL1-independent imatinib resistance,^16,17,55  it would be of interest to determine the combined effect of hypoxia and cytokines on CML progenitors. Our unpublished data also indicate that hypoxia and stromal cell coculture enhance colony formation, with or without imatinib. Thus, it is likely that several components of the BM microenvironment, including hypoxia, cytokines, and cell–cell interactions,⁵⁶  play a part in LSC maintenance.

We also explored the molecular mechanisms underlying the effects of hypoxia on CML progenitors. Because in most cell types the canonical hypoxic response is elicited by HIFs, we specifically explored the role of HIFs.⁴¹  First, we show in primary CML LSCs that HIF1-α is stabilized and localized to the nucleus only under hypoxia, in contrast to a previous report using mouse cell lines.⁴⁰  We further demonstrate that imatinib can partially attenuate the expression of HIF1-α, and by silencing HIF1-α, we show that the maintenance of CML progenitors under hypoxia is dependent on HIF1-α expression.

Interestingly, despite the transient nature of HIF1-α expression produced by our protocol (supplemental Figure 4D), the functional effects were long-lasting. Several mechanisms may account for the enduring effects of transient HIF1-α expression and include induction of epigenetic changes⁵⁷  and, possibly, HIF1-mediated expression of long-lived proteins. The situation in CML cells is also reminiscent of normal HSCs, as a previous report using human BM cells also showed that severe combined immunodeficiency-repopulating cells are preferentially preserved under hypoxic incubation.⁵⁸  Our findings are also in line with recent studies in mouse CML models demonstrating that HIF1-α is required for the long-term maintenance of LSCs.³⁹  Interestingly, these investigators concluded that LSCs seem to be more dependent on the HIF1-α pathway for survival than HSCs.³⁹  Although our work did not specifically address the role of HIF in BC CML, we observed massive expression of HIF1-α in BC CML BM specimens. Notably, others have reported that in BC cell lines selected for imatinib resistance, BCR-ABL1 upregulation is associated with enhanced HIF1-α signaling.⁵⁹  Interestingly, these biological effects of HIF1-α were evident under normoxic conditions, and therefore support our observation that HIF1-α silencing can impair CML colony formation under normoxia as well as hypoxia (Figure 6C). It also has been reported that engraftment of acute myeloid leukemia cells results in the appearance of widespread areas of hypoxia in the BM, whereas engraftment of normal HSCs only produces discrete areas of hypoxia.⁶⁰  As BCR-ABL1 kinase activity augments HIF1-α signaling under hypoxia, we predicted, and indeed observed, a much higher level of HIF1-α activity in CML BC cells compared with CP, as BC cells overexpress BCR-ABL1.⁶¹ 

Further mechanistic insights were revealed through our transcriptome analysis. Decreased cell death during hypoxia treatment of CML cells may be a result of the upregulation of antiapoptotic factors that counteract imatinib eradication. Indeed, antiapoptotic function is a prominent feature of hypoxia that mediates survival during ischemic stress.⁶²  Separately, the hypoxic regulation of diverse cell adhesion factors is known to enhance binding of leukocytes to endothelial cells,⁶³  which is important for extravasation and may explain the enhancement of clonogenicity and engraftment of hypoxia-treated CML progenitors. These findings connect multiple mechanisms involving survival and adhesion factors driven by hypoxia with the establishment of long-term CML resistance against imatinib.

In summary, our study suggests that under physiologic hypoxia, HIF-dependent signaling is incompletely abolished by imatinib and is able to support LSC maintenance and potency (Figure 7C). Thus, a combination of anti-BCR-ABL1 and anti-HIF1-α therapy might represent a promising strategy to eliminate LSCs and cure CML. Some of the hypoxia-regulated genes in CML cells (Figures 5B and 7A; supplemental Tables 2 and 3), particularly those extensively involved in survival and cell adhesion, could also represent therapeutic targets in CML progenitors, especially if direct targeting of HIF1-α should prove too toxic for normal hematopoiesis.⁶⁴ 

We are grateful for insightful discussions with Dr Chai Chou. We thank Zheng Lin and Norazean Zaiden for technical support and Dr Sharon Lim for advice on immunofluorescence. This study was supported by grants from the Singapore Ministry of Health’s National Medical Research Council, Biomedical Research Council of the Agency for Science, Technology and Research, Singapore National Research Foundation, Singapore Ministry of Education under the Research Center of Excellence Programme, Swedish Research Council, the Swedish Cancer Society, and the European Union.

Contribution: K.P.N. and K.L.L. designed and performed research, interpreted data, and wrote the paper; W.H. and A.M. performed experiments; S.Y.T. interpreted data; C.T.H.C. contributed vital material and clinical information, and participated in discussions; L.P. designed research and analyzed data; and S.T.O. designed research, interpreted data, and wrote the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: S. Tiong Ong, Duke-National University of Singapore Graduate Medical School, 8 College Rd, Singapore, 169857; e-mail: sintiong.ong@duke-nus.edu.sg.