Hypoxia-Activated Prodrug TH-302 Targets Hypoxic Bone Marrow Niches in Preclinical Leukemia Models

Acute myeloid leukemia (AML) is a molecularly heterogeneous disease characterized by an expansion of immature blood cells due to impaired self-renewal, differentiation, and apoptosis. High relapse rates, thought to be due to persistence of chemoresistant leukemia cells, and refractoriness to second-line chemotherapy remain major therapeutic challenges associated with poor outcomes (1). The leukemic microenvironment has been implicated in chemoresistance and nurturing of leukemia-initiating cells through multiple mechanisms (2). Recent findings from our group and others (3–5) demonstrated that hypoxia is a prevalent feature of the leukemic, but not the normal, bone marrow microenvironment. Utilizing murine leukemia models, we showed vast expansion of hypoxia niches within the leukemic bone marrow microenvironment that paralleled leukemia progression. Similarly, hypoxia was shown to be present in multiple myeloma in vivo models and to correlate with disease progression and aggressiveness (3, 4). Pathologic hypoxia is a common feature in many solid tumors and is associated with increased tumor aggressiveness and resistance to therapy, mediated at least in part through activation of hypoxia-inducible factor 1α (HIF1α; refs. 6–8). Indeed, in acute lymphoblastic leukemia (ALL) cells, pharmacologic or genetic stabilization of HIF1α under normoxic conditions inhibited cell growth and reduced apoptosis induction by chemotherapeutic agents, whereas inhibition of HIF1α expression promoted chemosensitivity of ALL cells cultured under hypoxic conditions (9). Our previous data showed that HIF1α protein was expressed in the bone marrow biopsies of approximately 60% of Philadelphia chromosome–negative B-ALL patients but was undetectable in normal donor bone marrow, suggesting the importance of HIF1α in the leukemic bone marrow microenvironment (10). Matsunaga and colleagues reported that HIF1α promotes quiescence of leukemic stem cells residing in the endosteal niches and suggested that this could contribute to minimal residual disease in AML (11).

HIF1α induces the expression of an array of genes regulating angiogenesis, invasiveness, drug resistance, and glycolysis (12, 13). Magnetic resonance spectroscopy (MRS) is an analytic technique that allows for nonradioactive isotopes (such as ¹H, ¹³C, ³¹P) to be detected in a high-field magnet. Because nuclei in different chemical environments will have slightly different resonance frequencies, MRS can be used to determine the metabolic profile of biologic samples ex vivo or in vivo (14, 15). However, real-time metabolic imaging is not possible with conventional MRS, which requires the averaging of multiple readings to generate a well-resolved spectrum. This problem has been solved by a new method called hyperpolarized MRS (HP MRS), which achieves a >10,000-fold enhancement of signal over conventional MRS, allowing nontoxic and nonradioactive imaging of metabolic products after addition of a hyperpolarized agent to a biologic system (16–18). Hyperpolarized ¹³C-enriched pyruvate, a novel diagnostic substrate, has been utilized to interrogate glycolytic metabolism in vivo through tracing the metabolism of pyruvate to alanine, lactate, and bicarbonate (16–18). Under hypoxic conditions, cancer cells rewire their metabolism by increasing glucose uptake and utilizing anaerobic glycolysis for energy generation. This increase leads to high levels of lactate production through pyruvate being utilized by lactate dehydrogenase. Using hyperpolarized pyruvate technology, the glycolytic flux of lactate dehydrogenase can be monitored in real time. The higher the flux rates, the greater use of glycolysis in that particular tissue.

Hypoxia-activated prodrugs (HAP) are designed to deliver cytostatic or cytotoxic effects to hypoxic subregions. In most cases, the first step in their activation reaction consists of a one-electron reduction of the prodrug by generally ubiquitous one-electron reductases (19–21), resulting in a prodrug radical that undergoes redox cycling in the presence of oxygen (22). Under hypoxic conditions, the radical has a much longer lifetime and undergoes fragmentation or further reduction, leading to formation of the active drug in hypoxic cells. We have previously shown the effectiveness of HAP PR-104 in preclinical leukemia models, supporting the concept for the selective cytotoxic targeting of the hypoxic leukemic environment (23), and recently reported the results of the phase I clinical trial (24).

TH-302 is a HAP that is currently in clinical testing for a broad series of indications, including myeloma and AML/ALL. Upon reduction of a nitro group in TH-302, the prodrug releases a DNA-crosslinking mustard alkylating agent (bromo-isophosphoramide) in hypoxic regions (25, 26). In multiple myeloma models, TH-302 demonstrated profound in vitro and in vivo antitumor activity (27).

In this study, we characterized the leukemic microenvironment through hypoxia-regulated gene expression studies, dynamic MRS of HP pyruvate in vivo and interrogation of residual hypoxic leukemia cells surviving chemotherapy using the HAP TH-302. Our findings confirm the premise that hypoxia is a prominent component of leukemia and demonstrate the improved antileukemia efficacy of combining traditional therapeutics with hypoxia-targeting agents.

TH-302 was provided by Threshold Pharmaceuticals. Cytarabine, doxorubicin, 5-azacytidine, and decitabine were purchased from the pharmacy of The University of Texas MD Anderson Cancer Center (Houston, TX). Sorafenib was acquired from Selleckchem. Fluorescein isothiocyanate–conjugated mouse monoclonal anti-pimonidazole (PIMO) antibody (Hypoxyprobe, Inc.) was used for immunohistochemical analysis. Antibodies used for FACS were AnnexinV-Cy5, mouse anti-human CD45, rat anti-mouse CD45, mouse anti-human CD34, and mouse anti-human CD123 (BD Biosciences). Propidium iodide (PI) and 7-amino-actinomycin D (7-AAD) were purchased from Sigma Chemical.

KG-1, REH, and Nalm-6 cells were purchased from ATCC and MOLM-13 cells from DSMZ. KBM-5 and OCI-AML3 cells were kindly provided by Dr Miloslav Beran (Department of Leukemia, MD Anderson Cancer Center, Houston, TX) and Dr. Mark Minden (Princess Margaret Cancer Center, Toronto, Canada), respectively. All cell lines were tested and authenticated by STR DNA fingerprinting performed by the Characterized Cell Line Core facility at MD Anderson Cancer Center (Houston, TX). Additional cell line information can be found in Supplementary Table S1.

Bone marrow or peripheral blood samples for in vitro studies were obtained from patients with AML or ALL. Samples were collected during routine diagnostic procedures after written informed consent was obtained; protocols for studies in humans were approved by the Human Subjects Committee of MD Anderson Cancer Center (Houston, TX).

Mononuclear cells were separated by Ficoll-Hypaque (Sigma-Aldrich) density gradient centrifugation. Cell lines and primary samples were maintained in RPMI1640 medium containing 10% fetal calf serum (Gemini Bio-Products) and 1% penicillin-streptomycin (Life Technologies Laboratories). Patient characteristics are shown in Supplementary Table S2.

Normoxic culture conditions comprised 21% O₂ and 5% CO₂ at 37°C. For experiments involving hypoxic conditions (1% O₂ and 5% CO₂ at 37°C) cells were incubated in the hypoxic Workstation INVIVO2 400 (Ruskinn Technology).

Total RNA was amplified and hybridized to Illumina HT12 version 4 human whole-genome arrays as described previously (28). For Nanostring analysis, mononuclear cells from bone marrow and peripheral blood samples from patients with acute leukemia or healthy subjects were lysed to isolate total mRNA. For each sample, 100 ng of mRNA was used to set up Nanostring codeset 5; the genes included in the analysis are listed in Supplementary Table S3.

Imaging was performed on a 7-Tesla, 30-cm bore Biospec MR imaging system (Bruker Biospin Corp.) equipped with a Bruker gradient/shim system BGA-12 (200 mT/m maximum strength; 80 μs rise time). Prior to scanning, each mouse was anesthetized, catheterized via tail vein, and placed in the lateral decubitus position on a custom animal bed distributing 2% isoflurane in oxygen and circulating warm water. A linear, dual-tuned ¹H/¹³C radiofrequency (RF) coil with 72-mm inner diameter was used for ¹H imaging and ¹³C transmission. To enhance sensitivity and localize signal reception to the femur, a custom-built 15-mm ¹³C receive-only surface coil was placed over the femur. Vital signs were carefully monitored throughout the experiments. The imaging protocol began with T₁-weighted positioning scans, followed by rapid gradient echo multiplane scout localizers. High-resolution T₂-weighted sagittal and axial images were then acquired (echo time TE = 17 ms; repetition time TR = 2,500 ms; 156 × 156-μm in-plane resolution; 4-cm field-of-view; 10 2-mm slices; 4 averages), providing visualization of the femur and offering a means for subsequent ¹³C slice prescription.

To generate HP pyruvate, a 26-mg sample of neat [1-¹³C]pyruvic acid (Sigma-Aldrich), containing 15 mmol/L of OX063 radical (GE Healthcare) and 1.5 mmol/L Prohance (Bracco Diagnostics Inc.), was polarized in a HyperSense DNP system (Oxford Instruments) as previously described (29, 30). The frozen sample was rapidly dissolved, resulting in a final isotonic and neutral solution containing 80 mmol/L [1-¹³C]pyruvate. Please refer to Supplementary Methods for further details.

Cells were seeded at 1 × 10⁶ cells/mL, treated with TH-302 at a final concentration of 5, 7.5, 10, or 15 μmol/L, and incubated in normoxic or hypoxic (1% O₂) conditions for 6 hours. The medium was then replaced with fresh medium to yield cell suspensions at 0.25 × 10⁶ cells/mL, and cells were incubated in normoxic conditions for 48 hours until collection for FACS analysis. For combination studies, cells were seeded at 0.2 × 10⁶/mL, treated with the indicated drugs, and incubated under normoxic or hypoxic conditions for 48 or 72 hours. Combination index values were calculated by utilizing Calcusyn software (Biosoft).

Following treatment, cells were washed twice in PBS solution and resuspended in 100 μL Annexin-binding buffer containing Annexin V–Cy5 (BD Biosciences). After 15-minute incubation in the dark, cells were washed and resuspended in Annexin-binding buffer containing 7-AAD or PI. Cells were quantitated after the addition of 10,000 CountBright counting beads (Invitrogen) per sample. Annexin V–positive cells were analyzed by a FACSCalibur flow cytometer (BD Biosciences) or a FACS Array Bioanalyzer (BD Biosciences), and results are presented as percentage of Annexin V–positive cells. In the case of primary human samples, results are expressed as percentage of specific apoptosis, calculated as: (% Annexin V–positive cells in the sample–% Annexin V–positive cells in the control)/(100−% Annexin V–positive cells in the control) × 100.

All animal work was done in accordance with a protocol approved by the Institutional Animal Care and Use Committee of MD Anderson Cancer Center. Please refer to Supplementary Methods for a detailed description of each model used.

Following chemotherapy and TH-302 treatment, hypoxia probe PIMO was administered intraperitoneally at 100 mg/kg. Three hours later, mice were euthanized and their femurs and tibiae were processed for IHC.

Unless otherwise indicated, results are expressed as mean ± SEM of three independent experiments. P values were determined by one-way ANOVA followed by F statistics or paired t test (when comparing two groups). A P value less than 0.05 was considered significant. Survival analysis for in vivo studies was done using Mantel–Cox and Gehan–Breslow–Wilcoxon tests. Unless otherwise noted, all calculations were carried out by the GraphPad Prism software (GraphPad) software.

We previously used chemical probes of hypoxia to detect expansion of hypoxic niches in murine leukemia-bearing bone marrow (23), and additionally found stabilization of HIF1α in bone marrow from leukemia patients (9). To elucidate molecular changes associated with the hypoxic nature of the leukemic bone marrow microenvironment, we first established a set of hypoxia-regulated genes in leukemia cells utilizing genome-wide gene expression profiling (GEP). To this end, total RNA from REH or OCI-AML3 human leukemia cell lines, each cultured for 48 hours at 21% or 1% O₂ was used for GEP by microarrays. Using strict significance criteria (P value < 0.01, false discovery rate q statistic < 0.1) for three independent experiments, we found 21 genes consistently upregulated by hypoxia in both cell lines (Supplementary Table S3). From these genes, we selected a “hypoxia set” that included genes known to be regulated by HIF transcription factors and/or that overlap with established hypoxia signatures in solid tumors (31–33). Next, we utilized Nanostring probes to compare the hypoxia-related gene expression in samples from leukemia patients (N = 15, Supplementary Table S2) and normal healthy donors (N = 4). The expression of 9 of the 21 genes interrogated, including LDHA, EGLN1, HIF1AN, ARNT, and TGFB, was individually significantly different between leukemia patient samples and controls (Fig. 1A). When expression levels of these 21 genes were averaged to generate a “hypoxia index,” 14 leukemic samples individually exhibited significantly higher index values as compared with normal donors, and the average normalized index value for all 15 leukemic patients 443.5 ± 49.26 (mean expression ± SEM) was significantly different from that of normal donors (100 ± 17.30, P < 0.0001, Fig. 1B). The difference in hypoxic signature expression persisted when the analysis was conducted comparing peripheral blood or bone marrow from leukemic and normal donors (data not shown), likely reflecting durability of gene expression changes preserved in circulating cells upon exiting hypoxic marrow microenvironment. Our results suggest that leukemic blasts exhibit a hypoxic gene expression profile correlating with the presence of hypoxia in the setting of advanced leukemia.

As hypoxia confers metabolic shift toward glycolysis, we determined the level of glycolysis occurring in diseased (AML1/ETO–bearing mice) versus healthy mice using in vivo HP-MR. C57Bl/6 mice were transplanted with murine cells expressing the AML1/ETO fusion oncoprotein, mutant NRAS, and GFP in a p53-null background (34). These mice develop AML very rapidly and, if untreated, die of the disease with a median survival time of 27 days. A ¹³C surface coil was placed on the femur and thigh of the animal to detect resonance frequency signal from HP [1-¹³C]-pyruvate and HP [1-¹³C]-lactate (Fig. 2A). The HP lactate signal lags arrival of the concentrated bolus of HP pyruvate, and reflects direct chemical conversion of HP pyruvate into lactate through interaction with the intracellular enzyme lactate dehydrogenase. Over time, both signals decay beyond the dynamic range of the measurement due to signal losses caused by excitation, relaxation of polarization to thermal equilibrium, and washout (29, 30). The area under the HP lactate curve, normalized to the area under both lactate and pyruvate, gives insight into the relative rate of exchange of the HP ¹³C label from pyruvate to lactate. The normalized lactate value measured from thigh and femur was significantly lower in normal animals compared with those bearing AML (Fig. 2B; P < 0.01), indicating higher rates of glycolysis in the leukemic microenvironment in vivo.

To further explore the metabolic profile of pyruvic acid, in vitro AML/ETO cells were incubated in media containing 1 mmol/L of 1-¹³C pyruvic acid and compared with control (See Supplementary Methods). Aliquots were taken at time of addition of the media (t_o), 1 hour into the incubation (t₁), and 8 hours into the incubation (t₈). All media aliquot samples were analyzed using one-dimensional proton-decoupled ¹³C NMR spectroscopy. The amount of 1-¹³C pyruvate still remaining in the media after 1 and 8 hours into the incubation compared with the initial time point (t₀) was used to determine the percentage of pyruvate uptake. The values can be seen in Supplementary Table S4. On the basis of these values, approximately 11% of the pyruvate is taken up by the culture in 1 hour and increases to 89% of pyruvate in 8 hours. Spectral resonances for lactate and pyruvate were not seen in control t₈ aliquots by ¹³C NMR analysis. A resonance for lactate was observed only in t₁ and t₈ pyruvate aliquot samples. These observations confirm that AML/ETO can utilize pyruvate to generate lactic acid readily similar to what was observed in HP-MR in vivo studies.

Given our findings indicating that hypoxia is a prevalent feature of the leukemic bone marrow microenvironment, and because of the known role of hypoxia in tumor biology and chemoresistance, we studied targeting hypoxia in leukemia with the HAP TH-302.We first tested this agent's antitumor activity in vitro against several leukemic cell lines representing diverse acute leukemia subtypes, including KBM-5 (chronic myeloid leukemia), KG-1, OCI-AML3, MOLM-13 (AML), REH, and Nalm-6 (pre-B-ALL). Figure 3A shows dose-dependent cytotoxicity in the cell lines tested, summarized in Supplementary Table S5 by the median inhibitory concentrations (IC₅₀) for normoxia and hypoxia, and a hypoxia cytotoxicity ratio reflecting the hypoxia-selective cell killing. Sensitivity varied across cell lines, but was consistently much greater in hypoxia. Nalm-6 cells were the most sensitive (IC₅₀ in normoxia and hypoxia: 1.3 and 0.005 μmol/L, respectively). We extended our results to primary ALL (#1 and #2) and AML (#3) samples, which exhibited drug-dependent cytotoxicity under hypoxic but not normoxic conditions (Fig. 3B). Importantly, TH-302 was able to target the leukemic stem cell–enriched CD34+/CD123+ population in a primary AML sample, as shown in Fig. 3C. In contrast, TH-302 exhibited minimal cytotoxic activity against four normal bone marrow samples (Fig. 3D).

To better recapitulate the multidimensional bone marrow niche, we next utilized cocultures of GFP–labeled leukemic cells (OCI-AML3 or MOLM-13) with bone marrow–derived red fluorescent protein (RFP)–labeled mesenchymal stromal cells (MSC) immobilized within Matrigel to generate three-dimensional (3D) structures (spheroids). The cells within the spheroids co-proliferated over time with colonies of leukemic cells firmly attached to MSC, as shown by confocal microscopy (Supplementary Fig. S1). Extensive hypoxia was present in the MSC/AML spheroids grown at normal oxygen tension, as revealed by PIMO staining (Supplementary Fig. S2). We then determined the antileukemia activity of TH-302 in spheroid cocultures of MSCs plus MOLM-13 or OCI-AML3 cells. TH-302 induced extensive apoptosis and reduced viability of leukemia cells by >60% in 3D spheroid cocultures at 10 to 50 nmol/L (Supplementary Fig. S1A and S1B).

The antileukemia efficacy of TH-302 was next tested in combination with chemotherapy agents frequently used for treatment of AML (cytarabine, doxorubicin) or with demethylating agents 5-azacytidine or decitabine (Supplementary Fig. S3). Combinations of TH-302 with cytarabine or with decitabine were synergistic or additive in killing OCI-AML3 and MOLM-13 AML cells under hypoxic conditions (average combination index values 0.69 and 0.67 for cytarabine plus TH-302 in MOLM-13 and OCI-AML3, respectively; 0.09 and 1.1 for doxorubicin plus TH-301 in MOLM-13 and OCI-AML3, respectively; 0.3 and 0.5 for decitabine or 5-azacytidine plus TH-302 in OCI-AML3, respectively).

To validate our in vitro observations, we tested the antileukemia efficacy of TH-302 in models of AML in vivo. NOD/Scid/IL2Rγ-KO mice (NSG) transplanted with a primary AML sample were treated with vehicle or TH-302 after the presence of circulating human CD45⁺ cells (∼5%) was confirmed by FACS (day 37). TH-302 (administered by intraperitoneal injection at 50 mg/kg three times a week for 3 weeks) prolonged survival of AML-bearing mice compared with the vehicle-treated mice (Fig. 4A; median survival time: TH-302 = 75 days; Control = 56 days; P = 0.003, n = 8 mice/group) and reduced the number of circulating AML cells by week 3 (2–3 mice were characterized: controls, 17.3 and 9.2 × 10⁶ cells/mL, respectively; TH-302, 2.3, 5.5, and 0.53 × 10⁶ cells/mL, respectively; Fig. 4B). PIMO staining revealed substantial reduction of the hypoxic areas in the bone marrow of treated mice compared with control mice (Fig. 4C). To test the ability of HAPs to target leukemia-initiating cells, secondary transplant experiments were conducted in which bone marrow cells from control or TH-302–treated mice, collected after 2 weeks of therapy, were injected into secondary NSG recipient mice at 1 × 10⁴ cells/mouse (N = 5 mice). At the time of secondary transplants, both control and TH-302–treated mice had high (>70% CD45+ cells in the bone marrow) leukemia burden, despite reduction of the total white blood count. Although all transplanted mice died from leukemia, survival of animals injected with cells from TH-302–treated primary recipients was prolonged compared with that of vehicle-treated controls (median survival of mice transplanted with cells from recipients, 68 days; and of mice transplanted with cells from primary recipients treated with TH-302, 79 days; P = 0.003; Fig. 4D). This was accompanied by delayed detection of circulating huCD45+ cells in mice transplanted with cells from TH-302–treated mice, whereas huCD45+ cells were detectable in control mice by week 4 after transplantation (Supplementary Fig. S4).

We next tested the antitumor efficacy of TH-302 and the multikinase inhibitor sorafenib in the aggressive MOLM-13 FLT3/ITD in vivo model. Sorafenib and TH-302 synergistically decreased viability and induced apoptosis in MOLM-13 cells in vitro, particularly under hypoxic conditions (CI value = 0.055, Supplementary Fig. S5). Starting on day 3 after transplantation of MOLM-13-LUC cells, NSG mice were treated with TH-302 or sorafenib, alone or in combination, for 2 weeks. Disease progression was monitored by optical bioluminescence imaging (BLI). The combination of TH-302 and sorafenib had the greatest effect in prolonging mouse survival compared with vehicle treatment (P= 0.0009), but each drug alone also prolonged survival, with TH-302 proving to be more effective than sorafenib alone (mean survival time: controls, 17.8 days; TH-302, 28.3 days; sorafenib, 24.6 days; combination, 32 days; Fig. 5A). The antileukemia effects were also evidenced by reduced BLI intensity in the treated groups than in the vehicle-treated animals (Fig. 5B and C).

On the basis of recent evidence suggesting that cancer stem cells reside in hypoxic niches that protect them from chemotherapeutics and are responsible for disease recurrence after treatment (2, 35, 36), we determined whether hypoxia was still present after disease debulking using the AML1/ETO syngeneic leukemia model. On day 18 after transplantation when engraftment was confirmed by detection of circulating GFP⁺ cells by FACS, mice were randomized and treated with vehicle or chemotherapy (cytarabine plus doxorubicin, DA) for 5 days. Treatment was effective in reducing circulating GFP-positive cells (controls, 11.5 ± 4.9, N = 6; treatment, 0.52 ± 0.27, N = 8; Fig. 6A), mimicking the disease in humans. To determine the persistence of hypoxia, mice were administered the hypoxia marker PIMO 1 week after completion of DA chemotherapy and sacrificed. IHC staining of the bone marrow from these mice revealed that although hypoxia was greatly expanded in control mice with high leukemia burden, it persisted in residual cells surviving chemotherapy (Fig. 6B).

On the basis of the evidence that the leukemic niches remain hypoxic after chemotherapy, we tested whether targeting residual hypoxic leukemic cells with TH-302 had any beneficial antileukemia effect. At 1 week postchemotherapy treatment, when only minimal residual bone marrow infiltration was detectable by BLI, mice were randomized to receive vehicle or TH-302 for the next 2 weeks. While chemotherapy reduced tumor burden and extended survival, administration of TH-302 in the setting of residual disease further prolonged survival, with a few mice surviving more than 50 days (mean survival duration: vehicle, 27 days; chemotherapy plus vehicle, 36 days; chemotherapy plus TH-302, 42 days; P = 0.0002; Fig. 6C and D). To examine the effects of TH-302 on hypoxic leukemic cells in the bone marrow, we conducted an independent experiment in which AML/ETO-transplanted mice received vehicle, DA chemotherapy or TH-302 for 5 days. We next compared the fraction of PIMO-binding AML cells in control mice, mice treated with DA or TH-302, following PIMO injection and immunohistochemical staining. As reported by us previously, healthy mice did not show PIMO positivity in the bone marrow, consistent with lack of gross hypoxia (Supplementary Fig. S6A top panel). In turn, extended areas of PIMO binding were detectable in both, control and DA-treated animals, whereas selective reduction of hypoxic areas was seen in TH-302 treated mice (Supplementary Fig. S6A and S6B).

The results shown in this work highlight hypoxia's role as a major feature of the microenvironment. GEP of peripheral blood and bone marrow samples from healthy donors and patients with leukemia showed that patients with leukemia have greater “hypoxia signature” expression than controls. The fact that this difference was identifiable even in peripheral blood suggests that the expansion of the hypoxic niche that we and others have described as a common phenomenon of acute bone marrow disease translates into long-lasting gene expression changes in cells that are influenced by the microenvironment. Among the genes that were differentially expressed were the two negative regulators of HIF-1α:EGLN1 and HIFAN; and genes involved in glycolytic metabolism LDHA and PGK1. In agreement with our findings, Rapin and colleagues, recently reported common transcriptional programs in AML patients that included a striking upregulation of the hypoxic response gene signature (37). These transcriptome changes can lead to, or reflect, metabolic reprogramming of cells. A clear link between hypoxia and glycolysis has been established, whereby hypoxia-induced HIF1α promotes the expression of several enzymes involved in glycolytic pathways (12, 38, 39). In vivo metabolic imaging of the bone marrow has proven to be challenging because of technical and anatomic limitations. However, development of noninvasive HP ¹³C MR, providing 10,000-fold signal amplification over conventional MR, made possible interrogation of the in vivo glycolytic metabolic status of the bone marrow (29, 40, 41). Utilizing HP-¹³C pyruvate, we found a higher glycolytic rate in murine leukemic bone marrow compared with healthy marrow, as indicated by the greater conversion of pyruvate to lactate. We have preliminary data that the pyruvate conversion to lactate could be occurring primarily in the stromal cells compared with the AML/ETO cells (data not shown). This is consistent with recent publications indicating that increased glycolysis could be caused by stromal cells in close proximity to cancer cells and that there is metabolic crosstalk between the two cell types (42–45). This finding provides yet another piece of evidence supporting the metabolic reprogramming of bone marrow niches as a consequence of reduced oxygen availability in diseased bone marrow.

The mechanism for hypoxia development in the leukemic bone marrow remains elusive. Our working hypothesis is that leukemic blasts are intrinsically hypoxic because their high rate of proliferation depletes oxygen rapidly. Although not addressed in the present work, it has been speculated that a defective vasculature in the diseased bone marrow might also contribute to generation of a hypoxic niche. Indeed, Schaefer and colleagues reported that the microvasculature in a murine AML model was heterogeneous, resembling the defective angiogenesis observed in solid tumors and associated with disease progression (46). The role of hypoxia in promoting resistance to chemotherapy and stemness and increasing tumor invasiveness has been widely studied, and our published work together with current findings support the presence of hypoxia in the leukemic bone marrow microenvironment and provide a strong rationale for exploiting hypoxia as a therapeutic target.

Hypoxic cells can be targeted by inhibiting HIFs that are required for survival of hypoxic cells (47–49). Very recently, both HIF-1 and HIF-2 have been implicated in the maintenance of leukemia-initiating cells (50, 51), but these transcription factors remain challenging drug targets, with no selective inhibitors available in clinical practice. An alternative approach is to use HAPs that selectively target hypoxic cells. Several HAPs have advanced to clinical testing, mainly in solid tumors but more recently also in hematologic malignancies (25). The hypoxia-selective activity of TH-302 has been demonstrated in a wide range of tumor cell lines, including a multiple myeloma in vivo model (27). It is not known what genetic characteristics underlie cell line-specific differences in sensitivity or resistance to TH-302 in vitro. This same broad range of cytotoxic potency (over two logs) was observed by Meng and colleagues across 32 cell lines (26) and more recently in the Hunter and colleagues study in triple-negative breast cancer cell lines (>∼30× range of cytotoxicity) (52). Several factors have been identified that can underlie differences in sensitivity to TH-302 in vitro: (1) variations in the expression levels of the one-electron reductases necessary for the activation of the prodrug; (2) differences in the activity of the specific DNA damage response and repair pathways in response to TH-302; and (3) differences in cell fate as a consequence of the DNA damage (e.g., cell-cycle arrest, cell death). Currently in clinical development, HAP TH-302 is being tested for a broad series of indications including AML/ALL. Preliminary results of a phase I clinical trial of single-agent TH-302 in patients with advanced leukemia (N = 34) have been reported, with transient responses suggesting clinical activity (53).

Here we showed that TH-302 had impressive hypoxia-dependent activity in several leukemia cell lines and primary patient samples and in a 3D system developed to resemble the leukemic niche in vitro. Recent work by Portwood and colleagues showed that TH-302 induced double-stranded DNA breaks, cell-cycle arrest, and ultimately cell death in in vitro models of AML under hypoxic conditions (54). Our findings expand on this work by demonstrating in vivo activity of TH-302 in several other AML xenograft models. Most importantly, we found that administration of TH-302 a week after chemotherapy treatment to mice with minimal residual disease prolonged survival, suggesting that this approach may be suitable for eliminating chemotherapy-resistant leukemia-initiating cells. In addition, we determined that residual leukemic bone marrow cells in a syngeneic AML model remain hypoxic after chemotherapy; and that TH-302 selectively reduces hypoxia in AML marrows. We also demonstrated that TH-302 has synergistic and additive effects in combination with conventional chemotherapy and demethylating agents in vitro, as has been observed in preclinical solid tumor models (55). In line with the in vitro data, we also showed that TH-302 has potent activity in combination with sorafenib in the aggressive MOLM-13 FLT3/ITD in vivo model. The data included in Fig. 3D indicate only minor toxicity against normal hematopoietic cells. Yet, we believe that one can expect some degree of myelosupression as on-target side effect of HAPs, which could be in particular apparent in combinations with other cytotoxic agents. In clinical trials with TH-302 moderate myelosuppression was observed but this did not constitute dose-limiting toxicity. When TH-302 was combined with other myelosuppressive agents as part of a combination regimen in the clinical trials (e.g., TH-302 plus doxorubicin for soft-tissue sarcoma) more myelosuppression was observed than what could be expected with doxorubicin alone, but this was not dose limiting (56).

In summary, the work presented here extends our understanding of the hypoxic nature of the leukemic bone marrow microenvironment. Our in vivo and in vitro metabolic data offer hints on the metabolic rewiring occurring in the diseased marrow and provide additional “metabolic” therapeutic targets. We showed that HAPs can be utilized effectively to target leukemic cells and hypoxic leukemia-initiating cells and that combination of HAPs with chemotherapy and/or other targeted therapies may constitute a successful therapeutic regimen for patients with acute leukemia.

M. Konopleva reports receiving commercial research grants from Threshold Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Benito, N.Z. Millward, Y. Shi, P.K. Bhattacharya, M. Andreeff, M. Konopleva

Development of methodology: J. Benito, N.Z. Millward, K.G. Harutyunyan, P. Matre, H. Ma, A. Volgin, P.K. Bhattacharya, J.A. Bankson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Benito, M.S. Ramirez, N.Z. Millward, J. Velez, K.G. Harutyunyan, H. Lu, S. Konoplev, H. Mu, J. Lee, P.K. Bhattacharya, J.R. Marszalek, R.E. Davis, J. Bankson, J.E. Cortes

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Benito, M.S. Ramirez, N.Z. Millward, J. Velez, K.G. Harutyunyan, P. Matre, S. Konoplev, J. Lee, P.K. Bhattacharya, J.R. Marszalek, R.E. Davis, J.A. Bankson, J.E. Cortes, M. Konopleva

Writing, review, and/or revision of the manuscript: J. Benito, M.S. Ramirez, N.Z. Millward, J. Velez, P. Matre, R. Jacamo, S. Konoplev, J. Lee, P.K. Bhattacharya, J.R. Marszalek, J.A. Bankson, J.E. Cortes, C.P. Hart, M. Andreeff, M. Konopleva

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Benito, N.Z. Millward, P. Matre, T. McQueen, M. Protopopova, M. Andreeff

Study supervision: J. Benito

Other [realization of animal experiment. Including handling, bioluminescence to detect leukemia burden, follow up, treatment, sacrifice, extraction and processing of tissues (including flow cytometry)]: J. Velez

The authors thank Dr. Scott W. Lowe (Cold Spring Harbor) for kindly providing the AML1/ETO cells.

This research was supported in part by Threshold Pharm., Leukemia and Lymphoma Scholar in Clinical Research 2189-12 (to M. Konopleva), the NIH 5 R01 CA155056-03 (to M. Konopleva), Leukemia Spore 5 P50 CA100632-08 DRP award (to M. Konopleva), Cancer Center Support Grant P30CA016672, and the Cancer Prevention and Research Institute of Texas RP101243-P5. This research was also supported in part by Leukemia SPORE Developmental Award NCI P50 CA100632 (to P. Bhattacharya), MD Anderson Institutional Research Grant (to N.Z. Millward) and R21 CA 185536-01 (to P. Bhattacharya and N.Z. Millward), CDMRP PC110065 (to N.Z. Millward) and NCI Cancer Center Grant CA016672 for the support of the NMR Facility at MD Anderson Cancer Center.

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.