Review article

Oxygen-dependent regulation of mitochondrial respiration by hypoxia-inducible factor 1

Gregg L. Semenza


The survival of metazoan organisms is dependent upon the utilization of O2 as a substrate for COX (cytochrome c oxidase), which constitutes Complex IV of the mitochondrial respiratory chain. Premature transfer of electrons, either at Complex I or at Complex III, results in the increased generation of ROS (reactive oxygen species). Recent studies have identified two critical adaptations that may function to prevent excessive ROS production in hypoxic cells. First, expression of PDK1 [PDH (pyruvate dehydrogenase) kinase 1] is induced. PDK1 phosphorylates and inactivates PDH, the mitochondrial enzyme that converts pyruvate into acetyl-CoA. In combination with the hypoxia-induced expression of LDHA (lactate dehydrogenase A), which converts pyruvate into lactate, PDK1 reduces the delivery of acetyl-CoA to the tricarboxylic acid cycle, thus reducing the levels of NADH and FADH2 delivered to the electron-transport chain. Secondly, the subunit composition of COX is altered in hypoxic cells by increased expression of the COX4-2 subunit, which optimizes COX activity under hypoxic conditions, and increased degradation of the COX4-1 subunit, which optimizes COX activity under aerobic conditions. Hypoxia-inducible factor 1 controls the metabolic adaptation of mammalian cells to hypoxia by activating transcription of the genes encoding PDK1, LDHA, COX4-2 and LON, a mitochondrial protease that is required for the degradation of COX4-1. COX subunit switching occurs in yeast, but by a completely different regulatory mechanism, suggesting that selection for O2-dependent homoeostatic regulation of mitochondrial respiration is ancient and likely to be shared by all eukaryotic organisms.

  • cytochrome c oxidase (COX)
  • hypoxia-inducible factor-1 (HIF-1)
  • mitochondrial respiratory chain
  • oxidative phosphorylation
  • reactive oxygen species (ROS)
  • tricarboxylic acid cycle


The evolution of multicellular organisms on Earth was dependent upon two critical events. The first event, ∼2.5 billion years ago, was the evolution of organisms that transduced solar energy into the chemical energy of carbon bonds [1]. During photosynthesis, CO2, water and light are utilized to generate glucose and, as a side product, O2. The resulting progressive rise in atmospheric O2 levels led ∼1.5 billion years ago to the second critical event, which was the establishment of a symbiotic relationship between single-celled organisms and internalized primitive cells that became specialized (as mitochondria) to perform a series of chemical reactions in which glucose and O2 were utilized to generate ATP with CO2 and water as side products. Compared with the simple fermentation of glucose to lactate, this process of oxidative phosphorylation greatly increased (∼18-fold) the net yield of ATP per mol of glucose consumed. The creation of the glucose–O2–ATP cycle (Scheme 1) provided biochemical conditions that were permissive for the evolution of higher organisms, as the dramatic increase in energy production resulting from oxidative phosphorylation provided the driving force required for the co-ordination of individual cells to act in concert as a multicellular organism. However, as discussed below, the utilization of O2 was not without its risks.

Scheme 1 The biochemical circle of life

Through the process of photosynthesis, plants and cyanobacteria (‘blue–green algae’) use solar energy to synthesize glucose (C6H12O6) from CO2 and water, with O2 produced as a side product. Glucose and O2 are used by most organisms to convert ADP into ATP, through the process of oxidative phosphorylation, which yields CO2 and water as side products. ‘hv’ represents solar energy [where h is Planck's constant (6.62×10−34 J·s) and v is the frequency of the photons in Hz].

The goal of oxidative phosphorylation is the transfer of electrons through a series of acceptor cytochromes in order to generate a proton gradient within the inner mitochondrial membrane (Scheme 2). The potential energy of this gradient is used to synthesize ATP. O2 is utilized as the ultimate electron acceptor, resulting in the production of water, a process that is catalysed by COX (cytochrome c oxidase; Complex IV). This process is not completely efficient, so that electron transfer to O2 may occur either at Complex I or Complex III, resulting in the generation of superoxide radicals, which are converted into H2O2 by the activity of superoxide dismutase. These ROS (reactive oxygen species) have the potential to oxidize cellular proteins, lipids and nucleic acids and, by doing so, may cause cell dysfunction or death (Scheme 3). By contrast, physiological levels of mitochondrial ROS are utilized by mammalian cells as a means of signal transduction. Most notably, increased mitochondrial ROS production under hypoxic conditions is required for activation of HIF-1 (hypoxia-inducible factor 1), as discussed below.

Scheme 2 Mitochondrial oxidative phosphorylation

The respiratory-chain complexes are as follows: I, NADH:ubiquinone oxidoreductase; II, succinate:ubiquinone oxidoreductase; III, cytochrome bc1 complex; IV, COX. IMM and OMM refer to the mitochondrial (Mito) inner and outer membranes respectively. CoQ, coenzyme Q; Cyt C, cytochrome c; O2•−, superoxide anion; TCA, tricarboxylic acid. An animated version of this Scheme can be found at

Scheme 3 Oxygen homoeostasis

The O2 partial pressure (‘tension’; PO2) in each mammalian cell is determined by the supply of O2 by diffusion from the vasculature and the local consumption of O2 by the cell and those surrounding it. In each cell, a set point is established at which consumption of O2 and production of ATP and ROS are optimized. Changes in PO2 result in decreased ATP production and/or increased ROS production, which, if uncorrected by homoeostatic mechanisms, can result in cell death.

Several recent studies that will be the main subject of the present review have revealed that mammalian cells utilize multiple homoeostatic mechanisms to modulate O2 consumption, glucose metabolism and mitochondrial respiration in response to changes in cellular O2 availability. These mechanisms serve to strike a balance between the properties of O2 as a facilitator of efficient ATP production and O2 as a generator of ROS (Scheme 3). At least one of these homoeostatic mechanisms is also utilized by yeast, suggesting that selective pressure for adaptation to changes in oxygenation has existed throughout eukaryotic evolution.


Oxygen homoeostasis represents a central organizing principle of metazoan evolution and development. At the molecular level, a critical event in metazoan evolution was the emergence of a transcription factor that functions to regulate gene expression in response to changes in O2 availability. HIF-1 consists of an oxygen-regulated HIF-1α (or HIF-2α) subunit and a constitutively expressed HIF-1β subunit. HIF-1 activates the transcription of target genes by binding to the core DNA sequence 5′-RCGTG-3′ (where R is G or A) within cis-acting hypoxia response elements and recruiting the trans-acting co-activators p300 and CBP [CREB (cAMP-response-element-binding protein)-binding protein]. Hundreds of human genes are regulated by HIF-1 [2,3] and dozens of these have been established as direct targets by identification of critical HIF-1 binding sites (Table 1).

View this table:
Table 1 Genes that are directly regulated by HIF-1

In mice and other mammals, HIF-1 is required for establishment of all three components of the circulatory system (blood, heart and vasculature) at mid-gestation, when sufficient O2 can no longer be supplied to all cells of the developing embryo by simple diffusion from the maternal circulation [46]. HIF-1 is present in the simplest metazoan organisms, such as the nematode worm Caenorhabditis elegans [7], which consists of ∼103 cells and has no specialized physiological systems for O2 uptake or delivery. As initially discovered in mammalian cells [4,5], HIF-1 induces expression of genes encoding glucose transporters and glycolytic enzymes when C. elegans is subjected to hypoxic conditions in the laboratory [8], which may mimic conditions that are encountered when worms burrow into the soil in search of food. Thus it is likely that O2-dependent regulation of glucose metabolism was a primordial function of HIF-1.


HIF-1α is continuously synthesized and degraded by the proteasome under well-oxygenated conditions, so that the cell is poised to respond instantaneously to reduced O2 availability (hypoxia) by blocking the degradation of HIF-1α, allowing its rapid accumulation [912]. The molecular mechanism by which O2 regulates the half-life of HIF-1α involves the hydroxylation of two proline residues (Pro402 and Pro564 in human HIF-1α) by dioxygenases that utilize O2 and the tricarboxylic-acid-cycle intermediate 2-oxoglutarate (α-ketoglutarate) as substrates for a reaction in which one oxygen atom is inserted into a proline residue and the other is used to convert 2-oxoglutarate into succinate [7,1316] (Scheme 4). Under hypoxic conditions, the hydroxylation reaction is inhibited, owing to substrate (O2) limitation [7] and/or oxidation of the prolyl hydroxylases, which contain Fe(II) in their active site, by ROS generated in the mitochondria by electron-transport Complex III [1719]. High NO (nitric oxide) concentrations block the induction of HIF-1α under hypoxic conditions, an effect that has been attributed to the inhibition of electron-transport Complex IV activity and O2 consumption by NO, thus relieving substrate limitation of the prolyl hydroxylases [20]. Hydroxylation of Pro402 and/or Pro564 in human HIF-1α is required for the binding of pVHL (von Hippel–Lindau protein), which is the recognition component of an E3 ubiquitin-protein ligase. Thus, under aerobic conditions, HIF-1α is rapidly hydroxylated, ubiquitinated and degraded.

Scheme 4 O2-dependent regulation of HIF-1

O2-dependent dioxygenase reactions (red box), mediated by HPH1-3 (HIF-1α prolyl hydroxylase) and the asparaginyl hydroxylase FIH-1, each result in the transfer of one oxygen atom to HIF-1α and one oxygen atom to 2-oxoglutarate (α-ketoglutarate) to form succinate (box at bottom left). Hydroxylation of proline residues Pro402 and Pro564 (P402 and P564) leads to binding of pVHL, which promotes the ubiquitination and proteasomal degradation of HIF-1α under aerobic conditions. Hydroxylation of an asparagine residue Asn803 (N803) in the C-terminal transactivation domain (TAD C) blocks the binding of the co-activators p300 and CBP. The hydroxylation reactions are inhibited under hypoxic conditions, leading to increased HIF-1α half-life and transcriptional activation. bHLH, basic helix–loop–helix motif; PAS, PER-ARNT-SIM-like sensor domain; TAD, transactivation domain.

In addition to the negative regulation of HIF-1α protein stability by prolyl hydroxylases, the transactivation function of HIF-1α is regulated by FIH-1 (factor inhibiting HIF-1) [21]. FIH-1 hydroxylates an asparagine residue (Asn803 in human HIF-1α) in another dioxygenase reaction that utilizes O2 and α-ketoglutarate [22,23]. Hydroxylation of Asn803 blocks the interaction of HIF-1α with the co-activators p300 and CBP [24]. Thus, both the half-life and specific activity of HIF-1 are directly regulated by the cellular O2 concentration through these hydroxylation reactions (Scheme 4). This pathway represents the most completely defined metazoan mechanism for oxygen sensing and signal transduction.


Since the time of Louis Pasteur (1822–1895) it has been recognized that cells preferentially metabolize glucose by respiration when O2 is available and by fermentation when O2 availability is reduced. Under hypoxic conditions, mammalian cells increase the expression of genes encoding glucose transporters as well as virtually all of the glycolytic enzymes, including hexokinase, 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase 3, aldolase, phosphoglycerate kinase, enolase and LDHA (lactate dehydrogenase A) (Table 1). The increase in glycolytic flux functions to compensate for the reduced efficiency of ATP production associated with reduced respiration. HIF-1, by co-ordinately regulating the expression of genes encoding each of the glucose transporters and glycolytic enzymes described above [4], functions as an essential mediator of the ‘Pasteur effect’ in mammalian cells [25]. Macrophages and granulocytes, which must function in the hypoxic microenvironment associated with tissue inflammation and infection, depend on glycolysis for ATP production and are rendered non-functional and energy-depleted in the absence of HIF-1 [26]. In response to stimuli other than hypoxia, several other transcription factors have been shown to regulate genes encoding glycolytic enzymes, including Myc, Sp1/Sp3 and USF2 (upstream stimulatory factor 2) [27,28]. The activity of glycolytic enzymes is also acutely regulated by NO, AMP-dependent kinase and other signal-transduction pathways [29].


Although conventional wisdom held that reduced respiration in hypoxic cells was simply a consequence of reduced substrate (O2) availability, early molecular studies had revealed that the increase in the levels of mRNAs encoding glycolytic enzymes was accompanied by a co-ordinate decrease in the levels of mRNAs encoding respiratory-chain proteins, suggesting that respiration might be actively repressed in response to hypoxia [30]. A key decision point in the metabolism of glucose occurs at pyruvate, which can be converted either into lactate by LDHA or into acetyl-CoA by the mitochondrial enzyme PDH (pyruvate dehydrogenase). The activity of the catalytic subunit of PDH is negatively regulated by phosphorylation, which is mediated by PDK (PDH kinase). Expression of the PDK1 gene is induced by hypoxia in a HIF-1-dependent manner [31,32]. Thus pyruvate metabolism is dramatically altered in hypoxic cells by the co-ordinate stimulation of LDHA and inhibition of PDH. As a result, pyruvate is shunted away from the mitochondria, reducing flux through the tricarboxylic acid cycle and thereby decreasing the delivery of reducing equivalents to the electron-transport chain.

These observations suggested that the reduction in tricarboxylic-acid-cycle activity is an adaptive response to hypoxia. When HIF-1α-deficient mouse embryo fibroblasts were maintained under hypoxic conditions for 72 h, ROS levels increased dramatically, leading to cell death [31]. However, ROS levels and apoptosis were markedly reduced in subclones transfected with an expression vector encoding PDK1. These data suggest that flux through the tricarboxlic acid cycle and, presumably, the electron-transport chain, are reduced under hypoxic conditions as an adaptive response that prevents the generation of toxic levels of ROS.


The hypothesis that alterations in electron transport chain activity might occur in mammalian cells under hypoxic conditions was supported by data from the budding yeast Saccharomyces cerevisiae demonstrating O2-dependent regulation of COX subunit composition. COX, which is located in the mitochondrial inner membrane, is a dimer in which each monomer consists of 13 subunits [33]. Subunits I, II and III (COX1–COX3 respectively), which are encoded by the mitochondrial genome and constitute the catalytic core of the enzyme, are highly conserved in all eukaryotes. The high-resolution crystal structure of bovine COX revealed that subunit IV (COX4) interacts, via its transmembrane domain, with COX1 and, via its C-terminal hydrophilic domain, with COX2 [33]. COX4 binds ATP, leading to allosteric inhibition of COX activity at high ATP/ADP ratios [34].

Molecular modelling studies indicate that the spatial relationships of the Saccharomyces homologues of COX1, COX2 and COX4 (which, in yeast, is designated COX5) are remarkably similar to the mammalian complex [35]. The yeast genome contains two genes encoding COX5 proteins, which show reciprocal patterns of gene expression: at high O2 concentrations, COX5a transcription is activated and COX5b transcription is repressed, whereas at low O2 concentrations, COX5a transcription is inactivated and COX5b transcription is de-repressed [3638]. Substitution of COX5b for COX5a increases the catalytic-centre activity (‘turnover number’; CCAmax) of COX by increasing the rate of electron transfer from haem a to the binuclear reaction centre within COX1 [39,40]. Thus yeast cells adapt to hypoxia by assembling COX complexes that have higher CCAmax values at low O2 concentrations.

Mammalian cells express a predominant COX4-1 isoform, whereas an alternative COX4-2 isoform is also expressed in certain tissues [41]. However, neither the molecular mechanisms regulating expression of the COX4I1 and COX4I2 genes that encode these proteins, nor the functional significance of alternative isoforms, was known. Analysis of cultured mouse and human cells revealed that COX4-2 mRNA and protein expression were induced by hypoxia [42]. HIF-1 heterodimers containing HIF-1β and either HIF-1α or HIF-2α bound to hypoxia response elements located in the 5′-flanking region and first intron of the COX4I2 gene within nuclear chromatin of human cells cultured under hypoxic conditions [42].

In contrast with COX4-2, COX4-1 mRNA levels did not change in response to alterations in the cellular O2 concentration [42]. However, COX4-1 protein levels were markedly reduced in response to hypoxia as a result of increased degradation of COX4-1 protein, which was HIF-1-dependent, but independent of proteasome activity. Instead, the expression of LON, a mitochondrial protease, was required for hypoxia-induced degradation of COX4-1. Remarkably, LON gene expression was induced in response to hypoxia, and multiple HIF-1 binding sites were identified in the 5′-flanking region of the human LON gene by chromatin immunoprecipitation and reporter-gene transcriptional assays [42]. Thus HIF-1 mediates co-ordinate regulation of the COX4-1 and COX4-2 subunit expression in response to changes in O2 availability.

Analysis of the effects of gain-of-function, loss-of-function and loss-of-function-with-subunit-rescue experiments for COX4-1 and COX4-2 revealed that the regulated expression of these subunits optimized the efficiency of respiration in human cells under aerobic and hypoxic conditions respectively [42]. When COX4-2 was replaced by COX4-1, there were significant decreases in O2 consumption, COX activity and ATP concentration under hypoxic conditions. When COX4-2 replaced COX4-1 under non-hypoxic conditions, O2 consumption, COX activity and ATP concentration were maintained at normal levels, but at the cost of increased ROS production and caspase activation.

Taken together, the results of these experiments indicate that the COX4 subunit switch constitutes a critical adaptive response of mammalian cells to hypoxia. In a striking example of convergent evolution, mammalian COX4 and yeast COX5 subunit expression are both regulated in an O2-dependent manner. However, the molecular mechanisms underlying this common strategy for regulating COX activity differ (Scheme 5). In Saccharomyces cerevisiae, COX5a transcription is activated by the Hap2/3/4/5 protein and COX5b transcription is repressed by the Rox1 protein under aerobic conditions, whereas, under hypoxic conditions, COX5a transcription is no longer activated and COX5b is no longer repressed [35]. The involvement of O2 in this yeast pathway is indirect: the transcription factors Hap2/3/4/5 and Hap1 [the transactivator of Rox1 (regulation by oxygen 1)] are activated by binding haem, which is only synthesized under aerobic conditions. In hypoxic mammalian cells, HIF-1, which is negatively regulated by O2-dependent hydroxylation, activates transcription of the COX4I2 and LON genes, which leads to increased COX4-2 mRNA and protein synthesis and increased COX4-1 proteolysis. COX4-2 and LON mRNA expression were induced by hypoxia in the liver and lungs of mice exposed to 10% (v/v) O2, indicating that this pathway represents a physiological response to hypoxia [42].

Scheme 5 COX subunit composition in regulated by O2 in yeast and human cells

Under aerobic conditions S. cerevisiae and Homo sapiens express COX5a and COX4-1 subunits, whereas under hypoxic conditions they express COX5b and COX4-2 respectively. The changes in subunit composition provide a mechanism by which mitochondrial respiration is optimized according to the cellular O2 concentration. Heme=Haem.

COX4 subunit switching provides a mechanism to maintain the efficiency of respiration under conditions of reduced O2 availability and may represent the initial adaptive response to hypoxia. When hypoxia is too prolonged or severe for COX subunit switching to maintain energy and redox homoeostasis, the induction of PDK1 activity may occur to shunt pyruvate away from the mitochondria and thereby reduce tricarboxylic-acid-cycle activity and flux through the respiratory chain. The extent to which PDK1 expression and/or COX4 subunit switching is utilized as a strategy for maintaining oxygen homoeostasis may be tissue-specific, as evidenced by the induction of COX4-2 and LON mRNA expression in several, but not all, tissues of chronically hypoxic mice [42]. These results are consistent with those obtained in previous studies demonstrating that the battery of HIF-1 target genes expressed in response to hypoxia differs from one cell type to another [43]. Further studies are required to investigate the adaptive significance of tissue-specific metabolic responses to hypoxia. In addition, these adaptive metabolic responses may play important roles in hibernation states in which O2 consumption is dramatically reduced.


The survival of metazoan organisms is dependent upon the utilization of O2 as a substrate for COX, which constitutes Complex IV of the mitochondrial respiratory chain. Premature transfer of electrons at Complex I or at Complex III results in the generation of ROS. The efficiency with which electrons are transferred to O2 at Complex IV (rather than at Complex I or Complex III) is dependent upon the net availability of: (i) O2, which is determined by relative rates of O2 delivery and consumption; and (ii) reducing equivalents (NADH and FADH2), which are generated through the oxidation of acetyl-CoA in the tricarboxylic acid cycle (Scheme 2). The studies reviewed here suggest that reduced O2 availability results in a reduced rate of electron transfer to O2 by COX. This mismatch between the delivery of electrons from Complex III and their transfer to O2 at Complex IV results in an increased rate of superoxide formation at Complex III, which leads to increased HIF-1 activity [1719]. HIF-1-mediated transcription results in two potential adaptive responses to hypoxia that serve to correct the putative mismatch: (i) increased PDK1 expression, leading to decreased tricarboxylic-acid-cycle activity, which reduces electron flux through Complexes I, III and IV; and/or (ii) increased COX4-2 and LON expression, leading to increased efficiency of electron transfer to O2 at Complex IV as a result of the COX4-1-to-COX4-2 subunit switch. These dual mechanisms may be important in responding to different physiological stimuli. For example, continuous hypoxia leads to increased ROS generation at Complex III [17], whereas intermittent hypoxia results leads to increased ROS generation at Complex I [44].

The studies reviewed here demonstrate that HIF-1 modulates multiple key metabolic pathways to optimize the utilization of O2 and glucose in response to changes in the availability of these substrates, in order to most efficiently generate ATP without excessive generation of ROS (Scheme 6). Although HIF-1α deficiency results in embryonic lethality, HIF-2α-deficient mice are viable on certain genetic backgrounds, but succumb to multi-organ failure with evidence of severe oxidative damage [45]. Studies reviewed above indicate that changes in oxygenation, relative to the normal physiological set point for a given cell, result in increased ROS production and provide evidence that oxygen and redox homoeostasis are inextricably linked (Scheme 3). HIF-1 plays a critical role in the maintenance of oxygen homoeostasis by mediating increased transcription of genes encoding each of the O2-regulated pathway components that have been described in the present review (Scheme 6). The remarkable similarities in the regulation of COX subunit composition in yeast and human cells indicate that the selection for O2-dependent homoeostatic regulation of mitochondrial respiration is ancient and likely to be shared by all eukaryotic organisms.

Scheme 6 Co-ordinate regulation of glucose and energy metabolism by HIF-1

In response to cellular hypoxia, increased HIF-1 activity leads to: increased glucose transport into the cell (1); increased glycolysis (2) and conversion of pyruvate into lactate (3); decreased conversion of pyruvate into acetyl-CoA (AcCoA) (4); and altered COX subunit composition that maintains efficient electron transport and minimizes superoxide (O2•−) production (5). ETC, electron-transport chain; ext, external; int, internal; TCA, tricarboxylic acid.


I am grateful to Dr Nanduri Prabhakar (Department of Medicine, Division of Biological Sciences, University of Chicago, Chicago, IL, U.S.A.) for providing helpful comments on the manuscript before its submission. Work in my laboratory was supported by funds from the Johns Hopkins Institute for Cell Engineering and by Public Health Service grants N01-HV28180, P50-CA103175 and R01-HL55338 from the National Institutes of Health.

Abbreviations: CBP, CREB (cAMP-response-element-binding protein)-binding protein; COX, cytochrome c oxidase; FIH-1, factor inhibiting HIF-1; HIF-1, hypoxia-inducible factor 1; HPH1-3, HIF-1α prolyl hydroxylase; LDHA, lactate dehydrogenase A; PDH, pyruvate dehydrogenase; PDK1, pyruvate dehydrogenase kinase 1; ROS, reactive oxygen species; VEGF, vascular endothelial growth factor


View Abstract