Mammalian Oxygen Sensing, Signalling and Gene Regulation

An adequate supply of oxygen is essential to all higher organisms because it serves as the terminal electron acceptor in mitochondrial oxidative phosphorylation and because several enzymatic processes require molecular oxygen as a substrate. Even slight reductions in normal oxygen concentrations (hypoxia) can cause the induction of specific genes involved in mammalian oxygen homeostasis such as erythropoietin or vascular endothelial growth factor (VEGF). Investigations of such hypoxia-inducible genes performed in many different cultured cell lines suggest that every mammalian (perhaps even every vertebrate and insect) cell possesses one or several oxygen-sensing mechanism(s), i.e. a molecular oxygen sensor. Certain cell types, which can be found in carotid bodies, airway neuroepithelial bodies and pulmonary arteries, have differentiated into specialized oxygen-sensing cells. However, the mammalian cellular oxygen sensor is not known. The beautifully elaborated principles of oxygen sensing in bacteria and yeast (for a review, see Bunn and Poyton, 1996) have still to find their counterparts in higher organisms. In the nitrogen-fixing bacterium Bradyrhizobium japonicum, for instance, the oxygen sensor is a histidine kinase termed FixL which, under anaerobic conditions, phosphorylates and thereby activates the transcription factor FixJ. Interestingly, the oxygen-sensing domain of FixL is a so-called PAS domain (see below), a versatile protein fold resembling a left-handed glove that encloses a haem cofactor (Gong et al., 1998). When oxygen is bound to ferrous (FeII) haem, this domain is in the flattened ‘off’ state, resulting in a change in the protein conformation and inactivation of the kinase domain.

So far, virtually all proteins capable of binding molecular oxygen contain iron, usually in the centre of a haem moiety. Thus, it came as no surprise when, in 1988, Bunn and co-workers suggested that the mammalian oxygen sensor could be a haem protein (Goldberg et al., 1988). Apart from hypoxia, cobalt (and to a lesser extent nickel and manganese) salts were capable of inducing erythropoietin gene expression with no additional effects under hypoxic conditions, indicating that ferrous haem iron was replaced by the non-oxygen-binding cations. According to the model, this locks the oxygen sensor in the deoxy conformation. Moreover, carbon monoxide inhibited hypoxic but not cobalt-mediated activation of erythropoietin gene expression, in line with the known properties of carbon monoxide binding to haemoglobin. Blockers of haem synthesis abolished both hypoxic and cobalt-dependent erythropoietin induction. Further evidence for the haem hypothesis was provided by the finding that iron chelators, such as desferrioxamine, are also capable of mimicking hypoxia (Ho and Bunn, 1996; Wang and Semenza, 1993b).

Although the basic findings that led to the hypothesis of a haem oxygen sensor responsible for hypoxic erythropoietin induction have been confirmed for many other hypoxia-inducible genes, several discrepancies remain. For example, differential effects of the flavoprotein inhibitor diphenylene iodonium on hypoxia-induced versus cobalt-induced gene expression have been reported. Ratcliffe and co-workers found that 50 nmol l⁻¹ and 1 μmol l⁻¹ diphenylene iodonium inhibited hypoxic but not cobalt- or iron-chelator-dependent induction of several oxygen-regulated genes (Gleadle et al., 1995). In striking contrast, Goldwasser et al. (1995) demonstrated a 50 % inhibition of cobalt chloride but not of hypoxic erythropoietin induction by 30 nmol l⁻¹ diphenylene iodonium. Schumacker and co-workers reported that diphenylene iodonium and two other mitochondrial inhibitors (rotenone and myxothiazol) blocked hypoxic but not cobalt-dependent induction of erythropoietin mRNA (Chandel et al., 1998). Ratcliffe and co-workers (Gleadle et al., 1995) and Semenza and co-workers (Agani and Semenza, 1998) also showed that the flavoprotein cytochrome P450 reductase inhibitor mersalyl inhibited hypoxia-, cobalt-chloride- and desferrioxamine-dependent induction of the erythropoietin gene. However, mersalyl induced the hypoxia-inducible factor-1α (HIF-1α, see below) and the vascular endothelial growth factor (VEGF) gene (Agani and Semenza, 1998). Mersalyl also induced a reporter gene under the control of transferrin-derived hypoxia-responsive elements (Rolfs et al., 1997) under normoxia without any further effects under hypoxia (R. Wanner and R. H. Wenger, in preparation). Finally, following careful titration of oxygen and cobalt chloride concentrations (taking into consideration that the optimal concentration of the toxic agent cobalt chloride is itself a function of the oxygen concentration), we found that there might well be additive effects on erythropoietin induction when both agents were applied simultaneously (Wenger et al., 1998b).

In summary, these discrepancies suggest either that hypoxia, cobalt and desferrioxamine interact with different steps along the oxygen signal-transduction pathway or that they each affect additional pathways that cross-talk with the oxygen pathway. These contradictory reports also reflect the difficulties of using pleiotropic agents to track down the oxygen sensor to its molecular entity. Different effects of these agents depending on concentration, target genes and cell types have to be considered. Nevertheless, the haem hypothesis is still the best model available. But what kind of haem protein is it?

Because there is no assay available to identify directly the oxygen sensor among the hundreds of known oxygen-binding proteins, many different candidate proteins have been suggested in the past to play this role as a sensor in addition to their known or proposed functions. The following section will provide some examples and discuss briefly the pros and cons for their putative role in oxygen sensing.

One of the most often cited putative haem oxygen-sensing molecules is the cytochrome b₅₅₈/NADPH oxidase complex that generates superoxide in the plasma membrane of phagocytes and B lymphocytes. The subunits of this complex have also been found in cells of non-immunological origin, including hepatoma cell lines (Görlach et al., 1993, 1994), small cell lung carcinoma (Wang et al., 1996) and pheochromocytoma PC12 cells (Hohler et al., 1999), and in oxygen-sensing cells of the carotid body (the arterial chemoreceptor; Acker et al., 1989; Cross et al., 1990; Kummer and Acker, 1995; Lahiri et al., 1999) and pulmonary neuroepithelial bodies (the airway chemoreceptors; Wang et al., 1996; Youngson et al., 1993, 1997). Therefore, Acker and co-workers put forward the hypothesis that the b₅₅₈ complex may serve as an oxygen sensor (for a review, see Acker, 1994). However, we have demonstrated normal oxygen-regulated gene expression in cell lines derived from patients suffering from chronic granulomatous disease, an inherited disease in which one of the subunits of the b₅₅₈/NADPH oxidase complex is defective (Wenger et al., 1996). Our conclusion that this complex is not likely to be the ‘universal’ oxygen sensor has recently been confirmed in knock-out mice deficient in the gp91^phox subunit of the b₅₅₈/NADPH oxidase complex (Archer et al., 1999). Despite a marked reduction in superoxide production, hypoxia caused similar levels of pulmonary vasoconstriction and inhibition of whole-cell K⁺ current in both knock-out and wild-type mice.

Another apparently promising candidate oxygen sensor is the principal oxygen-consuming organelle of the cell, the mitochondrion itself. Because the respiratory electron transport chain blocker potassium cyanide cannot induce erythropoietin gene expression, mitochondria were soon ruled out as the oxygen sensor (Goldberg et al., 1988; Tan and Ratcliffe, 1991). Nevertheless, Lahiri and co-workers proposed that mitochondrial cytochrome a₃ (Lahiri et al., 1995; Wilson et al., 1994) and Schumacker and co-workers suggested that mitochondrial cytochrome c oxidase (Chandel et al., 1997) could serve as the oxygen sensor. Interestingly, depleting cells of mitochondrial DNA (μ⁰ cells) and treatment with several mitochondrial inhibitors blocked hypoxic induction of oxygen-regulated genes (Chandel et al., 1998). However, because potassium cyanide does not interfere with hypoxic induction of HIF-1α (Jiang et al., 1996), the role of mitochondria in oxygen sensing remains unclear.

Various enzymatic processes that result in the production of different second-messenger molecules are catalyzed by haem proteins. For instance, the guanylate cyclase and cytochrome P450 of the 4A family, producing cyclic GMP (cGMP) and 20-hydroxyeicosatetraenoic acid (20-HETE), respectively, have been proposed as oxygen sensors (Harder et al., 1996; Taylor et al., 1998). Cytochrome P450 (ant)agonists have also been shown to modulate erythropoietin expression (Fandrey et al., 1990), but the involvement of cytochrome P450 could not be confirmed by spectrophotometric analysis (Ehleben et al., 1998).

As outlined above, the ‘comparative approach’ to the identification of the oxygen sensor is biased by our limited knowledge of the function and the number of candidate haem proteins. There may be many more candidates since a haem protein cannot be recognized easily from its primary amino acid sequence. What would be an unbiased ‘empirical approach’ to this problem? Oxygen binding is not a sufficient criterion for the sensor; it must connect an oxygen-dependent enzymatic activity (many of which are already known) to the regulation of hypoxia-inducible genes. Therefore, only the inactivation of the (unknown) sensor and an analysis of the consequences on gene activation can serve to identify it unambiguously. Ratcliffe and co-workers used hypoxia-response elements to direct the expression of reporter genes. Following random mutagenesis, cells were selected that had lost the ability for oxygen-regulated gene expression. This approach has been validated by the identification of an HIF-1α inactivation mutation (Wood et al., 1998). The oxygen sensor, however, remained unidentified.

Reactive oxygen species (ROS) such as superoxide and peroxide are known to serve as signal transducers in several systems (for a review, see Suzuki et al., 1997). Because oxygen is the main component of these molecules, it seems obvious that their concentration depends upon the environmental oxygen concentration. Fandrey, Jelkmann and co-workers reported a close relationship between oxygen and hydrogen peroxide concentrations and rates of erythropoietin production in hepatoma cells (Fandrey et al., 1994). Hydrogen peroxide sequestration could mimic hypoxia (Canbolat et al., 1998) and addition of hydrogen peroxide inhibited hypoxic induction of erythropoietin (Fandrey et al., 1994). This inhibition could be reversed by cobalt chloride and iron chelation (Fandrey et al., 1997). These findings have been confirmed for the oxygen-regulated genes encoding tyrosine hydroxylase (Kroll and Czyzyk-Krzeska, 1998) and several enzymes involved in glucose metabolism (Kietzmann et al., 1996, 1997). This model is also compatible with the idea that the thiol reducing agent N-(2-mercaptopropionyl)-glycine and the oxygen radical scavenger 2-acetamidoacrylic acid, but not many other antioxidants, could mimic hypoxia by inducing HIF-1α under normoxic conditions (Chilov et al., 1999; Salceda and Caro, 1997).

Enzymes generating ROS, such as the cytochrome b₅₅₈/ NADPH oxidase complex (see above), have been proposed as the source for superoxide, which is rapidly converted into peroxide by superoxide dismutase. Acker and co-workers suggested a localized Fenton reaction in perinuclear cytosolic granules containing high iron concentrations (Kietzmann et al., 1998; Porwol et al., 1998). The Fenton reaction describes the non-enzymatic reduction of hydrogen peroxide to a hydroxyl anion and the highly reactive hydroxyl radical by oxidation of the ferrous iron. This model would separate the sites of oxygen sensing and the effects of cobalt/iron chelation, and it would explain the differences observed in the function of these agents. According to this hypothesis, normoxic hydroxyl radical concentrations would inactivate HIF-1α, which is relieved from suppression following a decrease in radical concentrations under hypoxic conditions. A prototype factor for such a mechanism is the iron regulatory protein 2, which is oxidized in an oxygen-dependent manner, ubiquitinated and degraded by the proteasomes in iron-replete cells (Iwai et al., 1998).

In striking contrast to these findings, Schumacker and co-workers reported that cardiomyocytes increase their production of ROS following exposure to hypoxia (Duranteau et al., 1998; Vanden Hoek et al., 1998). Antioxidants inhibited this increase in levels of ROS and also blocked hypoxic inhibition of cardiomyocyte contraction. The mitochondrial cytochrome c oxidase was suggested as the oxygen sensor that enhanced ROS production in hypoxic cardiomyocytes (Budinger et al., 1998) and hepatocytes (Chandel et al., 1997). Support for this model was provided by using hepatoma cells depleted of functional mitochondria (μ⁰ cells), which showed no induction of ROS under hypoxia and which failed to induce oxygen-regulated gene expression (Chandel et al., 1998).

Taken together, there is no common consensus about whether ROS concentrations decrease or increase following exposure to acute hypoxia. Further experiments determining the kinetics and dependence on oxygen concentration of ROS production will be required to obtain a more definitive picture.

Apart from ROS, other metabolic processes occurring during hypoxia could potentially serve to sense and signal the cellular oxygenation state. A decrease in pH as a consequence of lactate production by anaerobic glycolysis might be responsible for hypoxic gene induction. The hypoxia-induced increase in ferritin production, for instance, might be regulated by iron release from ferritin as a result of acidification of the cytoplasm (Qi et al., 1995). Similarly, hypoxic induction of the p53 tumour suppressor protein has been reported to be dependent on acidosis (Schmaltz et al., 1998). However, a significant decrease in pH requires prolonged or severe hypoxia (anoxia) rather than the short periods of hypoxia sufficient to induce HIF-1α. Indeed, we have observed no effects of altered pH on oxygen-regulated gene expression (A. Rolfs and R. H. Wenger, unpublished observations). In general, it is difficult to understand how signal transducers such as ROS, redox state, pH, ATP, cAMP/cGMP or adenosine could trigger the ubiquitous, sensitive, specific and very rapid hypoxic induction of gene expression. A specific sensor that signals directly to HIF-1 and/or its degradation machinery seems to be more appropriate to fulfil these tasks.

Compared with the relative lack of information on the oxygen sensor and its signal-transduction pathway, much more is known about the target of these putative signalling mechanisms. In 1995, Semenza and co-workers discovered HIF-1 on the basis of its ability to bind to a hypoxia-response element (HRE) in the 3′ flanking region of the erythropoietin gene (Wang et al., 1995a). HIF-1 is a ubiquitously and constitutively expressed heterodimeric transcription factor composed of an α subunit unstable under normoxia and a common β subunit, the latter being shared by other transcription factors (for reviews, see Bunn and Poyton, 1996; Semenza, 1998; Wenger and Gassmann, 1997). HIF-1β had previously been identified as the heterodimerization partner of the dioxin receptor/aryl hydrocarbon receptor (AhR) and was hence called AhR nuclear translocator (ARNT, see Hoffman et al., 1991). HIF-1α heterodimerization with ARNT in the nucleus is required for DNA binding and transactivation (Gassmann et al., 1997; Gradin et al., 1996; Kallio et al., 1997; Salceda et al., 1996; Wood et al., 1996) but not for translocation into the nucleus (Chilov et al., 1999; Kallio et al., 1998).

The most intriguing feature of HIF-1α is a so-called oxygen-dependent degradation domain (ODD) that can confer hypoxic stabilization to HIF-1α and to various fusion partner proteins (Huang et al., 1998; Pugh et al., 1997; Srinivas et al., 1999a). Under normoxic conditions, this domain seems to receive a signal that primes the HIF-1α protein for ubiquitination and subsequent degradation in proteasomes. In fact, HIF-1α protein is usually undetectable under normoxic conditions. Proteasomal inhibitors or mutation of the ubiquitin-activating enzyme E1 stabilize the HIF-1α protein but fail to induce oxygen-regulated gene expression (Huang et al., 1998; Kallio et al., 1999; Salceda and Caro, 1997). Normoxic HIF α subunit degradation involves the von-Hippel–Lindau tumour suppressor protein (pVHL) which recognizes and links HIF α subunits, via a complex composed of several incompletely characterized proteins, to the ubiquitination machinery (Maxwell et al., 1999). Cells deficient in functional pVHL show constitutively high HIF α subunit levels and expression of many oxygen-regulated genes, leading to haemangioblastoma tumour formation.

The signal that primes the ODD domain for degradation is not known. Reduction–oxidation (redox) reactions might be involved in normoxic HIF-1α degradation, as is the case for other proteins such as iron regulatory protein 2 (Iwai et al., 1998). Certain antioxidants and suppression of the Fenton reaction by iron-chelation can induce HIF-1α protein as well as HIF-1-dependent reporter gene activation (Huang et al., 1996; Salceda and Caro, 1997). However, oxidation of HIF-1α has not been demonstrated so far. Thus, it is not known whether this is a specific mechanism linking HIF-1α to the oxygen sensor or whether the experimental data obtained are the result of nonspecific processes and/or cross-talk with other signal-transduction pathways. It is an intriguing possibility that the autonomous ODD domain of HIF-1α might sense oxygen directly. This hypothesis would predict an iron-containing oxygen-binding domain, either as a haem protein or as an iron–sulphur cluster. However, we (P. Spielmann, F. Parpan and R. H. Wenger, unpublished observations) and others (Srinivas et al., 1999b) could not detect iron in bacterially expressed fragments of HIF-1α that contain the ODD domain.

Although phosphorylation seems to play an important role in the activation of HIF-1α (Salceda et al., 1997; Wang et al., 1995b), replacement of potential phosphoacceptor sites in the critical region of the ODD domain had no effect on its activity (Pugh et al., 1997; Srinivas et al., 1999a). Two recent reports showed that the p42/p44 mitogen-activated protein kinases (MAPKs) can phosphorylate the HIF α subunits, giving it electrophoretic migration properties similar to those of the hypoxically modified protein (Conrad et al., 1999; Richard et al., 1999). However, the phosphorylated amino acid residues have not been identified. MAPK-dependent phosphorylation enhances the transcriptional activity of HIFs, but in non-excitable cells the p42/p44 MAPKs are not activated by hypoxia (Richard et al., 1999). In excitable cells, hypoxia might indirectly activate MAPKs, but HIF α protein stabilization is not related to MAPK-dependent phosphorylation (Conrad et al., 1999). Thus, p42/p44 MAPKs do not account for the most critical step in HIF α subunit activation: the stabilization of the ODD domain.

HIF-1α contains two transactivation (TA) domains, one overlapping with the ODD domain and the second at the C terminus (Huang et al., 1998; Jiang et al., 1997; Li et al., 1996; Pugh et al., 1997). Apart from conferring protein stabilization, these transactivation domains might be further activated under hypoxic conditions (Kallio et al., 1998), for example via p42/p44 MAPK-dependent phosphorylation (Conrad et al., 1999; Richard et al., 1999) or redox-dependent processes (Huang et al., 1996; Salceda and Caro, 1997). A reductive environment provided by the redox factor Ref-1 enhances the recruitment of the transcriptional coactivators SRC-1, TIF2 and CBP/p300 (Carrero et al., 2000). Fujii-Kuriyama and co-workers identified a redox-sensitive conserved cysteine residue, mutation of which inhibited the transactivation function (Ema et al., 1999). These authors proposed a model in which reduced thioredoxin translocates into the nucleus of hypoxic cells and transfers the redox signal to Ref-1 which, in turn, interacts with this cysteine. Indeed, overexpression of thioredoxin/Ref-1 has been shown previously to amplify the hypoxic signal (Huang et al., 1996). However, the functional importance of this cysteine residue could not be confirmed in other mutation experiments (O’Rourke et al., 1999), and we did not observe nuclear translocation of thioredoxin under hypoxia (U. Alt and R. H. Wenger, unpublished observations). Therefore, the specificity of thioredoxin/Ref-1 in maintaining HIF-1α in a reduced state as part of the hypoxia-signalling mechanism awaits further clarification.

Apart from the ubiquitously expressed HIF-1α, two other members of this family, HIF-2α (Ema et al., 1997; Flamme et al., 1997; Hogenesch et al., 1997; Tian et al., 1997) and HIF-3α (Gu et al., 1998), have been identified which show a more restricted expression pattern. Functional comparison between HIF-1α and HIF-2α in vitro revealed many similarities concerning genomic organization, modular protein structure, hypoxic protein stabilization, heterodimerization, DNA binding and the transactivation function of reporter genes (Ema et al., 1999; Hogenesch et al., 1998; Maxwell et al., 1999; O’Rourke et al., 1999; Wenger et al., 1997; Wiesener et al., 1998). In vivo gene targeting (knock-out) experiments in mice showed that HIF-1α is required for mesenchymal cell survival during embryonic development: knock-out animals die around midgestation, showing cardiovascular malformations and open neural tube defects (Iyer et al., 1998; Kotch et al., 1999; Ryan et al., 1998). Mice containing only one mutant HIF-1α allele develop normally but show impaired physiological responses to chronic hypoxia (Yu et al., 1999). Interestingly, in HIF-2α knock-out embryos, the organ of Zuckerkandl (the major catecholamine source during development) does not produce sufficient amounts of catecholamines, leading to embryonic lethality presumably because of bradycardia (Tian et al., 1998). The expression pattern and functional role of HIF-3α have still to be elucidated.

Another interesting feature of the HIF α subunits is the presence of PAS domains which can also be found in mammalian ARNT1, 2 and 3 and in their corresponding heterodimerization partners (HIF-1α, 2α, 3α, AhR1, 2 and CLOCK) as well as in several other transcription factors including Per1, 2, Sim1, 2 and NPAS1, 2 and 3 (for a review, see Taylor and Zhulin, 1999). PAS is an acronym standing for the first family members known: Per, AhR/ARNT and Sim. Most of these PAS proteins have been cloned from mammalian and fish species and from Drosophila, where they play important roles in various aspects of development, oxygen homeostasis and circadian rhythm.

Most of the PAS domain proteins described to date have been found in the Bacteria and Archea kingdoms, where they often serve as sensors of oxygen levels, redox state or light levels. Intriguingly, the haem-containing pocket of the FixL oxygen sensor is a PAS domain (Gong et al., 1998), and it is tempting to speculate that the HIF-1α PAS domain displays similar features. However, haem binding to the PAS domain has never been detected, and it is unlikely that HIF-1α functions by itself as an oxygen sensor. In Eukarya, the PAS domain has apparently evolved to a heterodimerization interface that is usually preceded by a basic helix–loop–helix DNA-binding and heterodimerization domain. Whether functionally altered or not, it is astonishing that this domain is conserved in proteins of higher organisms that are functionally related to their predecessors in more primitive species.

Once activated by hypoxia, HIF-1 binds to the consensus HIF-1 DNA binding site (HBS) A/(G)CGTG present in the hypoxia-response elements (HREs) of many oxygen-regulated genes (for a review, see Wenger and Gassmann, 1997). In the erythropoietin gene, this site can also be constitutively bound by the transcription factors activating transcription factor-1 (ATF-1) and cAMP-responsive element-binding-1 (CREB-1) (Kvietikova et al., 1995, 1997). Normoxic occupancy of the HBS (Hu et al., 1997) might protect the CpG dinucleotide present in the HBS from becoming methylation-inactivated (Wenger et al., 1998a). For transactivation, HIF-1 recruits via the two transactivation domains of HIF-1α as well as the C-terminal transactivation domain of ARNT transcriptional coactivators including the histone acetyltransferases CBP/p300 (Arany et al., 1996; Ebert and Bunn, 1998; Ema et al., 1999; Kallio et al., 1998) and SRC-1 and TIF2 (Carrero et al., 2000). It should be noted that an HBS is necessary, but not sufficient, for efficient gene activation. Binding of other transcription factors in the vicinity of HIF-1 or multimerization of HBSs is usually observed. In the erythropoietin HRE, for example, an unknown factor binding to a CACA box plus the hepatocyte nuclear factor-4 are required to form a functional transcriptional complex (Galson et al., 1995). Other examples include cooperation with the ATF-1/CREB-1 factor in the lactate dehydrogenase A gene (Ebert and Bunn, 1998; Firth et al., 1995) or with activator protein-1 (AP-1) binding factors in the VEGF gene (Damert et al., 1997). Alternatively, we found two adjacent HBSs in the transferrin enhancer (Rolfs et al., 1997), and two or three adjacent HBSs were also found in the genes for phosphoglycerate kinase 1, enolase 1, lactate dehydrogenase A, glucose transporter-1, haem oxygenase 1, p35srj and insulin-like growth factor binding protein-1 (IGFBP-1) (for references, see Table 1).

Several negative feedback regulatory pathways apparently exist to limit the hypoxic response. At the level of mRNA, we found a downregulation of HIF-1α mRNA concentrations by an unknown mechanism following prolonged exposure to hypoxia (Wenger et al., 1998c). At the transactivation level, HIF-1-dependent induction of p35srj has been reported, which in turn binds to the critical coactivator CBP/p300 and blocks the interaction with HIF-1α, resulting in a marked inhibition of transactivation (Bhattacharya et al., 1999).

Table 1 shows a compilation of the HIF-1 target genes identified so far. Only genes where functional HIF-1 DNA-binding or transactivation has been demonstrated are listed. These genes are involved in oxygen homeostasis at the cellular, local and systemic levels. Erythropoietin activates erythropoiesis to enhance the systemic oxygen transport capacity. Because iron is a limiting factor in haem formation, erythropoiesis is sustained by increased expression of transferrin and transferrin receptor to enhance iron supply to erythroid cells.

At the local level, HIF-1 activates vascular endothelial growth factor, as well as one of its receptors (Flt-1), which induce angiogenesis leading to an increase in the vascular density and hence a decrease in the diffusion distance for oxygen. Local blood circulation is also controlled by modulation of the vascular tone through the production of NO (nitric oxide synthase), CO (haem oxygenase 1), endothelin-1, adrenomedullin or activation of the α_1B-adrenergic receptor. It is notable that HIF-1 binding to the HREs of the endothelium-specific genes Flt-1 and endothelin-1 has not been directly demonstrated, and it seems possible that the relatively endothelium-specific HIF-2 might regulate these genes.

At the cellular level, loss of ATP production in mitochondria is compensated by anaerobic glycolysis. Therefore, glucose uptake (glucose transporters) and glycolysis (glycolytic enzymes, see Table 1) are upregulated by HIF-1. Besides hypoxia, both insulin and insulin-like growth factors (IGF)-1 and 2 induce HIF-1α expression (Agani and Semenza, 1998; Feldser et al., 1999; Zelzer et al., 1998). In addition to glucose transporters and glycolytic enzymes, HIF-1 activates IGF-2 and the IGF-binding proteins (IGFBPs) 1, 2 and 3 but not 4, 5 and 6 (Feldser et al., 1999; Tazuke et al., 1998). Other pleiotropic growth factors and cytokines capable of inducing HIF-1 include epidermal growth factor, fibroblast growth factor-2, interleukin-1β and tumour necrosis factor-α (Feldser et al., 1999; Hellwig-Burgel et al., 1999). Thus, these results indicate a complex interplay between positive and negative regulatory growth factors, although further experiments are required to provide a more detailed picture.

Fig. 1 summarizes schematically the current models and hypotheses about how a cell senses, signals and reacts to reductions in oxygen concentration. The identification of the oxygen sensor and the corresponding signal-transduction pathway that leads to the activation of HIFs will be of principal research interest for the near future. With regard to the importance of hypoxia in many major diseases (cardiovascular defects, tumour growth), the findings will be of major importance for the treatment of these diseases.

I wish to thank C. Bauer, G. Camenisch, M. Gassmann and A. Scheid for helpful discussions and critical reading of the manuscript, C. Gasser for the artwork and the Sondermassnahmen des Bundes zur Förderung des akademischen Nachwuchses and the National Science Foundation (Grant 3156743.99) for financial support.