Oxygen Sensing, Hypoxia-Inducible Factor-1 and the Regulation of Mammalian Gene Expression

Reduction of oxygen provides one of the most important sources of metabolic energy for living organisms, and the full (four electron) reduction of dioxygen (O₂) to water has unusual properties which are well suited to this role. First, the energy available from reduction by biological substrates is large and, second, energy barriers to the activation of dioxygen mean that the reduction can be enzymatically controlled and that relatively high concentrations of oxygen can be tolerated. Following a major increase in oxygen production by photosynthesis 1–2 billion years ago, atmospheric oxygen concentrations have been very much greater than those required for maximal rates of mitochondrial electron transfer, and this has enabled multicellular organisms to utilize large oxygen gradients for transfer purposes. Even bigger organisms have evolved special transport systems (e.g. lungs, heart and blood) to augment the potential of these gradients to support oxygen transfer over large distances. The intricacy of the anatomical organization underlying oxygen transfer by these organs is striking, as is the microscopic organization of the oxygen-consuming tissues. This is seen, for instance, in analyses of blood capillary density, in the zonation of metabolic enzyme activities and in the intracellular disposition of organelles such as mitochondria (Weibel et al. 1992; Jungermann, 1995). In fact, the whole of the macro- and micro-anatomy of a large animal have to be constrained by the need for adequate provision of oxygen.

Such processes must be dependent on mechanisms for the sensing of oxygen and the regulation of gene expression. One potential mechanism of sensing is through the adequacy of high-energy phosphate supply from mitochrondrial oxidative phosphorylation. Many processes are probably responsive to such ‘metabolic’ oxygen sensing (Adair et al. 1990), but it has long been clear that there are difficulties in proposing the operation of such a system in isolation. First, it would require a level of continuing metabolic compromise to generate an ‘error’ signal. Second, the multiple processes involved in oxygen delivery appear to be tightly coordinated so that capacity at different points is well matched, and it is difficult to understand how sensing at a single distal point could provide the necessary control. Third, analyses of two systems long known to be involved in the regulation of mammalian oxygen supply (the control of ventilation by the carotid body and the regulation of red cell production by erythropoietin) have strongly suggested the operation of other processes. In the case of erythropoietin production, the response to hypoxia cannot be mimicked by inhibitors of oxidative phosphorylation (Necas and Neuwirt, 1972). For the carotid body, it is clear that, although some response to cyanide exposure is observed, the operating is very much higher than that limiting mitochondrial respiration (Gonzalez et al. 1994).

The rapid dynamic response and extreme sensitivity of these systems has attracted great interest in the underlying mechanism(s) of oxygen sensing. Though each had traditionally been regarded as involving the operation of a highly specialized and tissue-restricted ‘sensor’, recent evidence indicates that at least some of the underlying processes are much more widely distributed. Oxygen-sensitive K⁺ channels, first recognized in the carotid body, have now been described in a number of different vascular cells, and oxygen sensitivity has been recognized for a variety of different ion channels (Weir and Archer, 1995). Studies of erythropoietin gene regulation have provided even more powerful evidence of a general oxygen-sensing system with a widely operative role in the regulation of gene expression (Maxwell et al. 1993; Bunn and Poyton, 1996).

Erythropoietin is the major regulator of erythropoiesis. Production by the kidneys and liver is increased in response to a reduction in haematocrit, arterial hypoxaemia or an increased haemoglobin oxygen-affinity. In this way, a feedback loop is completed in which a reduction in blood oxygen availability is sensed and corrected by stimulating red cell production in the bone marrow. Increased erythropoietin production parallels increases in erythropoietin mRNA production, and both are induced more than 100-fold by severe stimulation (Ratcliffe, 1993). In addition to this dramatic response to severe perturbations of oxygen delivery, small increases in erythropoietin production are induced by much more minor stimuli, such as the donation of a unit of blood (Lorentz et al. 1991). Furthermore, the basal plasma level is suppressed in cases of primary polycythaemia, demonstrating that regulation is operative in the normal physiological range of oxygen delivery (Cotes et al. 1986).

In enabling a molecular approach to the underlying mechanism, an important step was made by Goldberg et al. (1987), who demonstrated oxygen-regulated gene expression and erythropoietin production by certain human hepatoma cells. As had been observed in vivo, production was also induced by particular transition metal ions (cobalt II, nickel II and manganese II). When cells were exposed to carbon monoxide, stimulation of erythropoietin production by hypoxia, but not by cobaltous ions, was much reduced. This led the authors to propose the existence a haemoprotein-sensing molecule which (like haemoglobin) would respond to the liganding of oxygen (or carbon monoxide) with a conformational change. Arguing from knowledge that cobaltous ions are a substrate for ferrochelatase and that cobalt protoporphyrin does not bind oxygen, it was proposed that cobalt could substitute for iron in such a sensor, leading to a constitutive deoxy form which would stimulate gene expression even in normoxic cells (Goldberg et al. 1988).

Other data have suggested the operation of a redox process in the oxygen-sensing mechanism. For instance, induction of erythropoietin gene expression is very sensitive to hydrogen peroxide. Hydrogen peroxide production is itself oxygen-dependent, being reduced during hypoxia. Fandrey et al. (1994) have therefore proposed that this molecule is a signal pathway intermediate and have provided supporting evidence from the action of compounds with known effects on hydrogen peroxide production. However, it is not yet known whether these compounds act by perturbing hydrogen peroxide level within the concentration range observed in the normal normoxic to hypoxic transition, so that the evidence for hydrogen peroxide level as a sensor is incomplete. Another interesting characteristic is sensitivity to iodonium compounds (Gleadle et al. 1995b). These compounds are believed to operate as redox-activated flavoprotein inhibitors (O’Donnell et al. 1993). Since flavoproteins are often linked to haem proteins in electron transport systems, it might be that the interaction of oxygen with such a system forms the basis of sensing. Though such a model might accommodate the haem protein hypothesis, a different mode of operation is required. If it is proposed that a reduced electron flow to oxygen signals hypoxia and somehow activates the system, why do iodonium compounds and carbon monoxide (which should also act to block electron flow) not activate the system? A possible explanation is that the interaction of oxygen with a haem group regulates or even diverts electron flow from a signalling redox couple by providing an alternative electron acceptor.

Yet another characteristic requiring explanation is the stimulation of erythropoietin gene expression by iron chelators (Wang and Semenza, 1993a; Gleadle et al. 1995a). Here again, two quite different explanations have been suggested. Haem iron is not chelatable, but iron chelation might interfere with the synthesis of a rapidly turning over haem protein or with the function of a ferroprotein with a chelatable iron moiety. Alternatively, iron chelation might interfere with the generation of signalling active oxygen species through the Haber–Weiss reaction.

To date, the complexity of redox systems and our incomplete understanding of their many potential interactions with oxygen in mammalian cells has made it difficult to prove or disprove these interesting hypotheses. Together with the definition of the hepatoma-based tissue culture system, molecular cloning of the erythropoietin gene provided another new approach to the problem – from the gene outwards. Isolation of the DNA control sequences which mediate oxygen-regulated expression could be followed proximally from the DNA-binding transcription factors to the signal transduction molecules, and so on, to the regulator. In pursuit of this, sequences from the Epo (erythropoietin) locus were assayed for oxygen-regulated activity after transfection of the Epo-producing cell lines Hep3B or HepG2. These studies identified a powerful regulatory element lying 3´ to both the human and murine genes (the Epo 3´ enhancer) (Beck et al. 1991; Pugh et al. 1991; Semenza et al. 1991). Detailed studies of this enhancer defined several binding sites, one of which was critical for hypoxia-inducible function and bound a complex termed hypoxia-inducible factor-1 (HIF-1) (Semenza and Wang, 1992). In keeping with an important regulatory role, activation of HIF-1 was found to share several key physiological characteristics with erythropoietin regulation: progressive activation by hypoxia of graded severity, activation by cobaltous ions and sensitivity to the protein synthesis inhibitor cycloheximide. Subsequently, affinity-purification of the complex enabled Semenza and colleagues to identify encoding cDNAs and to demonstrate that the DNA-binding complex was a heterodimer of two basic-helix–loop–helix proteins of the PAS family, HIF-1α and HIF-1β (Wang et al. 1995). HIF-1α was a newly defined member of this family. However, HIF-1β had previously been identified as a dimerization partner for another basic-helix–loop–helix PAS protein, the aryl hydrocarbon receptor (AHR), where it is essential for the xenobiotic response and is termed the aryl hydrocarbon nuclear receptor translocator (ARNT) (Reyes et al. 1992). The term PAS is an acronym derived from the names of the first members of this gene family Per, AHR, ARNT and Sim; Per and Sim are Drosophila genes involved in the control of periodic and midline gene expression, respectively. The defining characteristic is the PAS domain, an imperfectly duplicated sequence of approximately 50 amino acids containing the characteristic motif His-X-X-Asp (Huang et al. 1993). Recent data base searches and cloning experiments have greatly expanded the family (Hogenesch et al. 1997; Tian et al. 1997; Zhou et al. 1997). At least one other member, endothelial PAS protein-1 (EPAS-1), appears to be directly involved in hypoxic gene regulation, whilst several others appear to be able to interact with the HIF-1 dimerization partners.

What was not expected when the analysis of erythropoietin gene regulation began was that this would uncover a general mechanism of cellular oxygen sensing and transcriptional control. The first clear evidence that this system was operative outside the context of erythropoietin regulation was provided by transfection studies of the erythropoietin enhancer (Maxwell et al. 1993). In contrast with the tight tissue restriction of erythropoietin gene expression, transiently transfected reporter genes linked to the erythropoietin enhancer showed similar hypoxia-inducible activity in both the hepatoma cells and a wide variety of non-erythropoietin-producing cells. In keeping with this, HIF-1 was found to be widely expressed (Wang and Semenza, 1993b), and functionally critical HIF-1 sites were defined in other genes, the first to be recognized encoding the glycolytic enzymes phosphoglycerate kinase-1 and lactate dehydrogenase A (Firth et al. 1994). The definition of a common regulatory mechanism for genes with functions quite different from erythropoietin suggested that many other HIF-1-responsive genes might exist, and a large number have now been identified through functional similarities with the induction of erythropoietin gene expression, reduced expression in HIF-1-deficient cells or the functional definition of HIF-1-binding sites in cis-acting sequences. These include genes involved in various aspects of energy metabolism (glucose transporters, glycolytic and gluconeogenic enzymes), iron metabolism (transferrin), catecholamine metabolism (tyrosine hydroxylase), vasomotor control (nitric oxide synthases, endothelin-1) and angiogenesis (vascular endothelial growth factor, platelet-derived growth factor) (see Table 1 for references). Though these studies strongly suggested a major role for HIF-1 in many physiological and pathophysiological processes, this cannot be immediately deduced from studies of gene expression in tissue culture.

In the whole organism, hypoxia most commonly occurs in the more complex setting of ischaemia and, in assessing the importance of HIF-1, it is important to understand the circumstances under which activation occurs in vivo, the effects on gene expression in this setting and the pathophysiological consequences. Such studies will ultimately require the development of techniques for the assay of HIF-1 activation that can be applied to heterogeneous tissues in whole animals. In the meantime, we have addressed this issue (at least in the tumour setting) by analysis of tumour xenografts from tissue culture cells in nude mice (Maxwell et al. 1997). We examined a series of hepatoma (mouse hepa-1) cell lines which were originally selected for a defective xenobiotic response through resistance to benzo[a]pyrene (Hankinson, 1979). Cells from one complementation group of resistant cells are deficient in the common heterodimerization partner HIF-1β/ARNT (Hoffman et al. 1991) and, in addition to manifesting a deficient xenobiotic response, they are unable to form the HIF-1 complex and show defective oxygen-regulated gene expression in tissue culture (Wood et al. 1996). We therefore compared gene expression, vascularization and growth in tumours from wild-type cells, a HIF-1β/ARNT-deficient line (c4), a revertant selected from c4 which has regained wild-type levels of HIF-1β/ARNT (Rc4) and a line (c31) that has a different defect in the xenobiotic response and is functional for HIF-1.

Fig. 1A shows an in situ hybridization study for the glucose transporter Glut-3 and vascular endothelial growth factor (VEGF) mRNAs in tumours from wild-type and HIF-1β/ARNT-defective (c4) cells. In wild-type tumours, areas of intense focal expression of Glut-3 and VEGF mRNA were often observed, and these were often close to necrotic regions. This intense focal gene expression was not seen in tumours derived from the mutant (c4) cells, but was observed in the revertant line (Rc4) and in the c31 cells, both of which form HIF-1 normally. This pattern of VEGF expression around (presumably hypoxic) necrotic zones has been noted previously and has led to the suggestion that hypoxia is a major regulator of tumour angiogenesis (Plate et al. 1992; Shweiki et al. 1992). Our experiments show that, at least in the context of these hepatoma cells, activation of HIF-1 is largely responsible for this pattern of gene expression within the tumour. When vascularity and growth were examined, substantial differences were observed which correlated with the presence or absence of a functional HIF-1β/ARNT gene product. In particular, tumours derived from the HIF-1β/ARNT-deficient (c4) cells showed a striking reduction in vascularity (Fig. 1B).

These experiments strongly suggest that HIF-1 is activated within solid tumours and provide evidence for an important pathophysiological role in this setting. Although the evidence linking hypoxia and erythropoietin regulation, through HIF-1, to hypoxia and tumour angiogenesis is rather strong, it is necessary to recognize the importance of other factors in shaping the pattern of gene expression. The genes listed in Table 1 are apparently linked to a common mechanism of regulation by oxygen, yet the pattern of inducible expression is in each case quite different. For instance, the VEGF gene is also expressed in normal tissues – yet in many organs, including kidney, it is induced modestly, if at all, by anaemic stimulation sufficient to generate a massive increase in erythropoietin production (Sandner et al. 1996). These differences cannot easily be explained by cell-specific gene expression and heterogeneous organ oxygenation. Rather, they suggest the operation of cooperative effects at the cellular level, in which transcriptional activation by HIF-1 requires ancillary factors that may be cell-type-specific or responsive to some other aspect of cellular physiology. For example, VEGF expression is also increased by a number of activated oncogenes, and it may be that an interaction with such molecules constrains the circumstances under which the gene is inducible by HIF-1 (Mazure et al. 1996). Other insights into the mechanisms underlying the selective responses to HIF-1 have come from studies of sequences close to HIF-1 binding sites. For instance, the function of the Epo enhancer is also dependent on at least two other sites adjacent to the HIF-1-binding site (Blanchard et al. 1992; Semenza and Wang, 1992; Pugh et al. 1994). One of these has been shown to bind the tissue-specific transcription factor hepatic nuclear factor (HNF-4) (Galson et al. 1995). In the lactate dehydrogenase A gene promoter, a functional interaction between an HIF-1 site and a cyclic AMP response element has been defined (Firth et al. 1995). Thus, the assembly of a functional transcriptional complex requires factors other than HIF-1, and it appears likely that this requirement for cooperative factors is used to define the individual features of an inducible response.

In pursuing the mechanism of oxygen sensing, a great deal of attention has been turned towards analysis of the mechanism of activation of HIF-1. In keeping with the demonstration of HIF-1 DNA-binding activity in a wide variety of cell types, mRNAs for HIF-1α and HIF-1β have been found in all cells and tissues examined (Wenger et al. 1996; Wiener et al. 1996). In the majority of studies, the mRNAs themselves were not found to be inducible by hypoxic stimulation, indicating that the activation of HIF-1 involves post-translational and/or translational mechanisms (Huang et al. 1996; Wenger et al. 1996; Wood et al. 1996). We were unable to observe translational regulation in fusions containing the HIF-1 promoter and 5´ untranslated region fused to a luciferase reporter and focused our attention on analyses likely to define post-translational mechanisms of regulation. As with other transcription factors, studies of regulatory mechanisms are potentially confounded by dependence on a series of interrelated events (e.g. dimerization, DNA binding, interaction with transcriptional activators) for an assayable response. For this reason, we have sought to define regions of HIF-1 genes that can independently confer oxygen-regulated activity on heterologous transcription factors. In this work, we have used two types of chimeric gene: those in which the heterologous transcription factor encoded a DNA-binding activity but lacked trans-activation; and others in which an activation domain was either included with the heterologous DNA binding domain or added from a second heterologous gene. This allowed for the analysis of regulatory domains from HIF-1 genes which did not necessarily contain intrinsic activation potential. In these experiments, we found that sequences from HIF-1α but not from HIF-1β could convey the hypoxia-inducible property, indicating a regulatory role for the α subunit (Pugh et al. 1997).

An example of such an experiment is shown in Fig. 2. The fusion of the powerful activation domain (amino acids 410–490) from the herpes simplex virus protein VP16 to a DNA-binding domain derived from amino acids 1–147 of the yeast transcription factor Gal4 creates a powerful transcription factor which activates expression from promoters containing a Gal4 binding site. When co-transfected into Hep3B cells with a gene containing such a promoter linked to a luciferase reporter, this Gal4/VP16 fusion is highly active. When particular sequences derived from the HIF-1α gene are encoded between the Gal4 and the VP16 sequences, two striking effects are observed. First, in normoxic cells, the overall level of activity is much reduced. Second, when cells are exposed to hypoxia, cobaltous ions or desferrioxamine, the apparent repression is greatly relieved, producing a regulated activity which reflects that of HIF-1. The experiment illustrated in Fig. 2B demonstrates a regulatory domain of HIF-1α between amino acids 530 and 652. In other work, we have defined sequences lying further towards the amino terminus of HIF-1α that also have this property (J. F. O’Rourke et al., unpublished work).

As a first analysis of the mechanisms by which such regulatory domains operate, we assayed levels of protein product in cells transfected with similar fusion genes. Studies so far have demonstrated that, at least with the regulatory sequences between amino acids 530 and 652, a major effect is observed on the level of expressed fusion protein. In normoxic cells, the fusion protein levels are much reduced by the inclusion of this sequence, with this reduction being relieved not only by exposure to hypoxia, cobaltous ions and desferrioxamine, but also by a variety of inhibitors of proteosomally mediated proteolysis. This suggests that one mechanism of regulation might be mediated by proteolysis of HIF-1α, with activation involving an inducible blockade of a constitutively high rate of degradation targetted through these sequences.

However, it is unlikely that this is the only mechanism of regulation. In our chimeric transcription factor assays, regulated activity was also observed which appeared to be independent of the level of expressed protein. This was particularly the case with the most C-terminal HIF-1α sequences. When fused to the Gal4 DNA-binding domain, C-terminal sequences from HIF-1α demonstrated oxygen-regulated trans-activation. Detailed analysis of these sequences showed that, while HIF-1α amino acids 786–826 showed constitutive trans-activation, amino acids lying immediately N-terminal to position 786 suppressed total activity and conferred inducibility (J. F. O’Rourke et al., unpublished work). Thus, a Gal4 fusion bearing amino acids 775–826 showed reduced activity in normoxic cells which was relieved by hypoxia. In this case, the fusion protein appeared to be expressed at a similar level in normoxic and stimulated cells.

Overall, these findings suggest a model involving at least two mechanisms of HIF-1 activation, with induction or derepression of activation domains being amplified by regulation of transcription factor abundance, occurring through changes in protein stability (Pugh et al. 1997). Such a model would be consistent with assays of endogenous HIF-1 immunoactivity. These demonstrate rapid nuclear accumulation of HIF-1α during hypoxia from very low levels in unstimulated cells (Wang et al. 1995; Huang et al. 1996). Upon reoxygenation, levels decline within minutes, indicating a very short half-life in normoxic cells. Similar studies of HIF-1β/ARNT in whole-cell extracts have shown modest or even absent induction by hypoxia, with considerable levels being present in unstimulated cells (Huang et al. 1996). Consistent with a constitutive excess of the β subunit, forced overexpression of the α subunit in normoxic cells is itself sufficient to drive HIF-1-dependent reporter gene expression. However, in keeping with a dual mechanism of activation, hypoxic stimulation of these cells further enhances activity (Jiang et al. 1996).

The definition of regulatory and activation domains for HIF-1α now points the way to defining the next steps in regulation. Quite possibly, this will involve regulated cofactors. For instance, the proteosome is constitutively active, and any regulated activity on HIF-1α could be modulated by a cofactor or by a modification of the target. Similarly, the known activation domains possess no clear homology with classical activator sequences, and it may be that they operate to recruit a cofactor. Interestingly, one known transcriptional coactivator, p300, is known to affect the activity of the system (Arany et al. 1996).

One important question regarding a system which appears to play such a central role in mammalian gene regulation by oxygen is what its role might be in non-mammalian cells – in particular, in simpler organisms which might facilitate genetic analysis of the sensing mechanism. To address this, we have studied the binding of nuclear extracts from non-mammalian cells to an oligonucleotide from the mouse erythropoietin enhancer (Nagao et al. 1996). So far, our clearest results have been from Drosophila melanogaster SL2 cells and are shown in Fig. 3. These cells contain a sequence-specific hypoxia-inducible DNA-binding activity which has a similar electrophoretic mobility to HIF-1, making it highly likely that a homologous system is operative in insects. In ongoing experiments, we are exploring the nature of this activity, but it is clearly of interest that the first basic-helix–loop–helix PAS proteins were defined in Drosophila and that more recent studies have isolated new genes, sima (Nambu et al. 1996) and trachealess (Wilk et al. 1996), which show even better sequence similarity and (at least in the case of trachealess) have a clear role in the organization of oxygen-delivery systems in the fruit fly. Thus, it seems likely that the uncovering of this widespread system of oxygen-regulated gene expression in mammalian cells will soon also lead to the recognition of a connection with oxygen-regulated gene expression in non-mammalian systems.

The work conducted in the authors’ laboratory was supported by the Wellcome Trust and the Medical Research Council, UK.