Using the hypoxia marker pimonidazole, we have shown in several in vivo
models of human and murine leukemia that the BM becomes highly hypoxic in advanced stages of the leukemia process. Furthermore, BM from ALL patients stained for HIF-1α showed very strong positivity at diagnosis that was impressively reduced or eliminated when the patients achieved CR. This is the first report indicating that the expansion of hypoxic areas is one of the essential characteristics of the leukemic microenvironment. The mechanism responsible for this expansion remains elusive at this point, but one possible explanation is that accumulation of leukemic blasts in the BM leads to increased oxygen consumption, thus lowering steady-state oxygen concentrations. Time-course studies using a blast crisis CML model support this hypothesis, clearly demonstrating that evolving hypoxic areas are associated with leukemia progression. Although not thoroughly investigated in our studies, it is also likely that leukemia propagation is associated with derangements and non-functionality of vascular architecture which might contribute to progressive hypoxia despite stimulation of angiogenesis through HIF-1α. This notion is supported by findings by Schaefer et al. 
demonstrating time-dependent alterations in the microvasculature in the in vivo leukemia models, with the early angiogenic wave followed by decreased vessel density and reduced tissue perfusion at the late stages of leukemia. We have observed that not only leukemic cells but also the surrounding stroma expresses HIF-1α (Fig. S3B
), implying that hypoxia is an intrinsic property of the leukemia microenvironment. While HIF-1α expression is frequently promoted in normoxic tumor cells, for example through oncogenic stimuli, generating molecular signatures resembling hypoxic response in the absence of hypoxia (coined as “pseudohypoxia” 
), hypoxic marker Pimonidazole can only be metabolized and form intracellular adducts at low oxygen concentrations, thus reflecting true hypoxia as opposed to pseudohypoxia. In a recent report, Hu et al. investigated the BM microenvironment in multiple myeloma models and observed vast hypoxic expansion (revealed by PIM staining), while naïve mice did not exhibit PIM positivity 
. In agreement with this, Colla et al showed that BM hypoxia and high HIF-1α expression is a characteristic of multiple myeloma patients 
. Therefore, a hypoxic microenvironment may be a common phenomenon in bone marrows with expanding tumor cell populations. We also noted only infrequent HIF-1α positive cells in normal BMs supporting the notion that under physiological conditions the BM harbors only discrete areas of hypoxia.
Our exploration of the role of hypoxia in leukemic cells showed that, in vitro
, 1% O2
levels conferred resistance against selective chemotherapeutic agents in the ALL cell lines tested. The mechanism responsible for this effect was not addressed in the present work. However, a potential candidate is HIF-1α given its role as master regulator of the hypoxic response. Many known HIF-1α targets could mediate the protection conferred by hypoxia, and a number of those have been validated as chemoresistance factors in leukemias (i.e., MDR-1, Nur-77, CXCR4 and others) 
. In this regard, we have previously reported that hypoxia increases CXCR4 expression leading to increased migration and survival of leukemic cells 
. HIF-2α is another HIF family member induced by chronic hypoxia and expressed at high levels in primary tumors and their metastases 
. HIF-2α targets overlap with but seem to be different from those regulated by HIF-1α 
. Which HIF-1α targets mediate hypoxia-induced chemoprotection in leukemia cells and whether there is any contribution from HIF-2α requires further investigation. In preliminary studies, HIF-2α was not found to be expressed in the ALL cell lines used for this study. Furthermore, HIF1α-independent mechanisms have been described as mediating hypoxia responses in various systems. Among them, NF-κB is of particular interest given that it can be induced by hypoxia and can regulate important pathways including cell proliferation, angiogenesis, metastasis and survival 
. Its importance in hematological malignancies is becoming clearer 
and in fact, two groups 
recently demonstrated that modulation of the NF-κB pathway sensitized leukemic cells to chemotherapy and inhibited leukemia cells growth, respectively. While hypoxia is a known factor mediating chemoresistance in solid tumors, our data presented here for the first time indicate the relevance of these processes in leukemia pathophysiology.
Various approaches have been explored to target the hypoxic microenvironment and thus render tumor cells susceptible to chemotherapy. Some approaches seek to directly inhibit HIF-1α activity, while others take advantage of the hypoxic microenvironment via hypoxia-activated prodrugs. In this study we used PR-104, a nitrogen mustard that under very low oxygen concentrations is reduced to its amine and hydroxyl amine metabolites which function as alkylating agents, leading to cell death. In vitro, the alcohol form of PR-104 had good cytotoxic activity against B-lineage ALL cell lines REH and Nalm6 and primary ALL cells. As expected, the cell killing effect was more pronounced in hypoxic compared to normoxic conditions. In cord blood cells used here as normal controls, the drug caused little cell death. The oxygen concentration required for 50% inhibition of PR-104A cytotoxicity in SiHa cells is only 0.1 µM, corresponding to 0.01% oxygen in the gas phase 
. The true oxygen concentration in cells with 1% oxygen in the gas phase is unknown, but is probably not low enough to fully activate PR-104A. However, given that PIM and PR-104A are both nitro compounds, and probably have broadly similar oxygen dependencies, we tested PIM activation in Nalm6 and REH under decreasing oxygen concentrations and observed hypoxia dependent binding at 1%O2
and lower (Figure S1
). These findings suggest that PR-104 could similarly, although not fully, be activated at 1% oxygen and that therefore the in vitro
activity probably underestimates activity in severely hypoxic tissues.
Our results in vitro prompted us to test PR-104 in several in vivo leukemia models, where it showed remarkable antitumor activity as a single agent. Murine xenograft models harboring Nalm6-Luciferase ALL or primary ALL treated with PR-104 showed responses with significant decreases in the percentage of circulating CD45+ cells and prolongation of survival. Furthermore, the agent is likely effective against a broad range of acute leukemias as PR-104 was effective in reducing tumor burden when primary AML cells were used in the murine xenograft model. There were however differences in the overall responses to the treatment since mice injected with the aggressive MLL-ALL primary cells exhibited a gradual increase in circulating leukemia cells once the treatment was stopped, an event that was not observed in mice injected with primary AML.
Dose response curves were also evaluated in two xenograft models of primary ALL. PR-104 significantly delayed the progression of ALL-8 in a dose-dependent manner at all doses tested. Treatment at the three highest doses of this drug resulted in an objective response measure (ORM) of maintained complete response (MCR). Two of these three doses correspond to the plasma levels of the drug that can be achieved in humans; recent comparison of plasma pharmacokinetics has shown that the mouse dose equivalent to the MTD in humans with solid tumors (1100 mg/m2
) is ~125 mg/kg 
. PR-104 also significantly delayed the progression of ALL-19 in a dose-dependent manner at the three highest doses tested. However, only the highest dose, which leads to plasma drug concentrations well above those achievable in the clinic, resulted in an ORM of MCR suggesting that combination therapies are needed. Since both ALL-8 and ALL-19 are derived from patients who experienced early relapse (within 2 years of diagnosis), possible differences in leukemia subtype specificity of PR-104 between T-lineage ALL (ALL-8) and B-cell-precursor ALL (ALL-19) should be explored in future studies. Interestingly, given the remarkable antitumor activity of PR-104, even at concentrations 50% or lower than the MTD, there seems to be higher activity of the agent in the leukemia models than that observed in the pediatric 
or adult 
solid tumours. The reasons for the observed differences are not clear at this point and would not appear to simply relate to AKR1C3 expression levels, since the levels in the 6 solid tumors varied from as high as the T-ALL xenograft which showed the highest sensitivity to PR-104, to very low levels of expression. Therefore, one could speculate that it might relate to: (1) the known sensitivity of ALL to alkylating agents e.g. cyclophosphamide; (2) the homologous recombination repair status of the cells; (3) the difference in models tested (systemic model of leukemia versus subcutaneous solid tumors) and differences in PR-104 biodistribution that might impinge on this; (4) other unknown factors intrinsic to the different tumor types. Notably, another hypoxia-activated alkylating agent TH-302 exhibited remarkable anti-tumor effects on multiple disease parameters in a multiple myeloma model 
In all three models tested here, restoration of normal hematopoiesis resulted in drastic decline in the areas of PIM positivity possibly indicating normalization of BM vasculature and restoration of higher physiologic oxygen tension. It is conceivable to hypothesize that application of hypoxia-activated prodrugs would selectively eliminate leukemic cells in most hypoxic areas. We also propose that leukemic stem cells may have a propensity similar to HSC to reside in these areas resulting in the emergence of resistant clones. While these hypotheses require further experimental proof, they may develop into rational combinations of PR-104 with chemotherapeutic or other targeted agents. Another attractive approach would be the utilization of this class of drugs in the setting of minimal residual disease. However, given the recent finding that normal long-term repopulating cells may likewise reside in these areas, their elimination may result in prolonged myelosuppression. Indeed, myelosupression has been reported for PR-104 in a Phase I clinical trial in solid tumors 
. It is arguable however whether this side effect is attributable to the ability of PR104 to target human progenitor cells residing in hypoxic areas, or to the fact that CD34 positive progenitors express AKR1C3, the enzyme responsible for aerobic activation of PR-104 
. In this respect, PR104-A indeed caused significant inhibition of the clonogenic survival of human progenitor cells in vitro, which was in turn ameliorated by AKR1C3 inhibitor naproxen (personal communications, J. Down and K. Parmar). Further studies with more selective hypoxia-activated prodrugs or HIF inhibitors will likely yield the answer to this clinically relevant question.
In summary, we have shown for the first time that the leukemic BM microenvironment is hypoxic and provided a rationale for targeting it. Our results with the bioreductive drug PR-104 suggest that targeting hypoxia is feasible and could have significant impact in the treatment of leukemias. Several in vivo murine leukemia models validated hypoxia as a potential target; as a consequence, PR-104 is undergoing a Phase I clinical trial in patients with relapsed and refractory AML and ALL. In the future, the ultimate goal would be to use bioreductive drugs in combination with conventional chemotherapies or novel targeted agents to eradicate resistant leukemic blasts that persist in the hypoxic BM microenvironment.