Harveian Oration 2020: Elucidation of molecular oxygen sensing mechanisms in human cells: implications for medicine

A widespread system of oxygen sensing in human and animal cells

We recognised that identification of this oxygen-sensitive transcriptional enhancer would allow the question of cell specificity to be addressed. Was it really true that oxygen sensing was a property that was specific to erythropoietin-producing cells? How would these sequences behave if introduced into cells that did not produce erythropoietin? Initial experiments gave an unclear result: at least the sequences did not appear to be as active as in the erythropoietin-producing hepatoma cells. This initially reinforced the idea of specific erythropoietin-producing cells and I set about what I thought was an innovative (though in retrospective very naive) experiment, attempting to ‘expression clone’ components of the ‘oxygen sensing’ pathway by transfer of genes from the ‘oxygen sensing’ hepatoma cells, to a line that would be predicted to be oxygen-insensitive (ie every cell type, according to the then-existing prejudice). To my surprise, when setting up controls for these experiments, I observed clear oxygen-regulated activity in the cells selected as recipients. Of course, this disqualified the approach, but fundamentally changed our thinking.

We found that responses to hypoxia depended on the culture conditions, but that, when rapidly dividing relatively dense cultures were used, essentially all the cultured cells we tested supported oxygen-regulated activity of the erythropoietin enhancer. Another trainee nephrologist, Patrick Maxwell, did much of this work. Almost certainly, the dependence of responses on tissue culture conditions that Patrick identified related to the dependence of the actual oxygen concentration at the cell monolayer on the oxygen consumption rate of the cells, and hence on their growth rate and density. We were excited by the implications of these findings; the same oxygen-sensing system was clearly widespread, at least in mammalian cells, and there must be other targets in non-erythropoietin-producing cells.

The findings, however, met with some scepticism and initial difficulty in publication. From my current vantage point, I have some sympathy with the editor. Part of the concern was again about incomplete knowledge. It was indeed (as was pointed out in review) only one new fact, and we hadn't then followed through by addressing the obvious next question: what were the additional oxygen-sensitive targets that were implied? I had little experience of research practice and it didn't occur to me that deriving follow-on conclusions, or a ‘more complete story’, might be important before publication. As far as I was concerned the findings were robust and the implications clear. The findings were eventually published, but not in the journal of first intent.¹³

Our findings were soon corroborated by data from Gregg Semenza, who, importantly, had identified a transcription factor in hepatoma cells that bound the erythropoietin control sequences in an oxygen-dependent manner, which he termed hypoxia-induced factor (HIF).¹⁴ In keeping with our transfection results, HIF was found to be expressed widely across different cell types.¹⁵ So what were its targets? The efforts in my laboratory initially focused on genes encoding enzymes in the glycolytic pathway, glucose transporters and angiogenic growth factors, whose regulation we considered might contribute to oxygen homeostasis. With the help of other clinical trainees, including Ben Ebert, John Firth and Jonathan Gleadle, who joined the laboratory and contributed much to the work, we were rapidly able to show that many of these genes demonstrated the characteristics that had been defined for erythropoietin (induction by both hypoxia and cobaltous ions) and possessed oxygen-regulated control sequences, with the same core DNA sequence that was present in the erythropoietin enhancer and bound to HIF.^16–18 In other work that proved important later on in the story, we went on to show that insect (Drosophila melanogaster) cell extracts also possessed a hypoxia-inducible activity that bound these sequences.¹⁹ Thus, the system was conserved in species that do not have an erythropoietin gene or even possess an oxygen-carrying blood vascular system similar to that in higher animals.

At the time we were rather surprised that so many genes, involved in different processes and manifesting very different regulatory characteristics in the intact organism, were apparently responsive to the identical transcriptional system. But as technical innovations enhanced the capacity to assay gene expression, we were truly astonished by the unexpected breadth of the HIF transcriptional cascade (Fig 2).²⁰ Many different genes with roles in processes such as pH regulation, matrix metabolism, iron transport and cell differentiation were revealed to be controlled by HIF. Studies using gene arrays and ‘next-generation sequencing’ identified hundreds or thousands of HIF transcriptional targets.^21,22 Many of these transcripts encoded transcript factors themselves; others encompassed different classes of regulatory RNA.^23,24 Together, these regulatory cascades might be predicted to impinge on practically every aspect of cell physiology, as might have been (but was not) predicted from the central role of oxygen in biological chemistry.

Other methods, particularly analyses of the phenotypes associated with genetic inactivation of components of the HIF system in the tissues and organs of recombinant mice, revealed entirely unexpected insights into the role of HIF in different aspects of integrated animal physiology. Studies by Randall Johnson and others revealed that genetic inactivation of HIF in immune and inflammatory cells is associated with phenotypes that imply important functions in these cells.^25–27 Coupled with evidence for physiological hypoxia in regions of the lymph node, bone marrow and thymus, and the pathological generation of profound hypoxia at sites of inflammation and immune activation, the work suggests a new physiology of hypoxia in immune and inflammatory control. One interesting possibility is that hypoxia (via HIF) has a direct effect on regulatory T cell function.²⁸ However, linking of the effects of ablative genetic interventions to a clearly defined regulatory role in physiology is not straightforward and much of the rapidly expanding knowledge on the HIF transcriptional cascade remains to be fully understood at the level of integrated physiology. Nevertheless, the work on erythropoietin, once considered to represent a tightly restricted response to hypoxia, clearly opened completely unexpected new fields of research. One might ask how it was that so much was overlooked? I am not sure, but our tendency to overlook things that have been overlooked by others is remarkable, something that must surely have surprised William Harvey himself.

Molecular mechanism of the oxygen sensing

I'll return to the unexpected reach of the HIF transcriptional cascade later, in the context of therapeutics. At the time our main focus was to dissect pathways that lay upstream of HIF, connecting its regulation to the availability of oxygen. Gregg Semenza had purified HIF and identified the encoding cDNAs, demonstrating that the DNA binding complex was an alpha/beta heterodimer of basic helix-loop-helix PAS (Per-AHR-ARNT-Sim) transcription factors (Fig 2).²⁹ The beta subunit (HIF-1β) had previously been identified as a dimerisation partner of the aryl hydrocarbon receptor, whereas the alpha subunit (HIF-1α) was a novel protein that is specifically involved in the response to hypoxia.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Schematic illustrating the transcriptional response transduced by HIF (hypoxia inducible factor). HIF binds as an α/β heterodimer to hypoxia responses elements that are associated with HIF transcriptional target genes. HIF target genes mediate a very broad array of cellular and systemic responses to hypoxia, which reveal that adaptation to hypoxia extends more widely than was previously foreseen.

So, what lay upstream of HIF-1α? We had no idea how far we had to travel, nor exactly what we were looking for. Based on the mimicry of the response to hypoxia by iron chelators and cobaltous ions, it was widely believed that some form of ferroprotein (with which these substances might interact) was the source of the oxygen-sensitive signal. But almost every type of ferroprotein had been considered, with the exception of those that we were later to show actually function in this way. Presuming the existence of a long and complex transduction pathway connecting this ‘sensor’ to gene transcription and not being 100% confident I possessed the necessary skills for a direct biochemical approach to its analysis, we set up almost every conceivable alternative (genetic and other) strategy to dissect the pathway in parallel. In the end it was a mixture of several of these approaches that proved successful. I'm not going to describe all the work; some was embarrassingly naïve on my part and some heroic on the part of others. Morwenna Wood and Emma Vaux, two more trainee nephrologists who joined the laboratory, approached a world record in tissue culture in their isolation of somatic mutant cells that were defective in signalling hypoxia. Both ultimately succeeded in generating mutants in the pathway,^30,31 although the means to identify the defective genes in mammalian cells were at the time insufficiently rapid for our purpose.

I'll outline some of the steps in the path we ultimately followed, which I hope illustrate my theme of how knowledge builds on knowledge in unexpected ways. Our first direct approach to the problem was the identification of sequences in HIF-1α (and later the paralogue HIF-2α) that would confer oxygen sensitivity when fused to a heterologous protein. We found three sequences that had this property; two operated by conferring oxygen-dependent protein instability on the fusion protein and one appeared to operate independently of any change in protein stability.^32,33 This ‘domain architecture’ was conserved between the paralogues HIF-1α and HIF-2α and we looked for a common consensus sequence, confidently expecting to find the postulated protein phosphorylation motif that would lead us to the upstream signalling pathway. Our efforts went as far as to mutate every phospho-acceptor in one of these domains, but the oxygen-sensitive property was unaltered.³² The work was again difficult to publish on the basis of being ‘incomplete knowledge’ and of course largely negative data. But in fact, at least for us, it was directional in excluding the ‘mainstream’ hypothesis regarding signal transduction.

The next key step was the recognition that the von Hippel-Lindau (VHL) protein, a commonly mutated tumour suppressor protein in kidney cancer, interacted with two of these domains (those that operated through protein stability). This work interfaces with contributions from my other co-laureate in Stockholm, Bill Kaelin. Bill and others had found that certain hypoxia-inducible transcripts were constitutively upregulated in VHL-defective cells, but the mechanism was unclear.^34,35 Using antibodies raised ‘in house’ to both the paralogues, HIF-1α and HIF-2α, we made the connection to HIF, revealing that VHL physically interacted with the previously defined ‘oxygen-dependent degradation’ domains in HIF-α polypeptides.^36,37 We showed that VHL was necessary for oxygen-dependent degradation of HIF-α polypeptides, acting as the recognition component of a conditional ubiquitin E3 ligase that targets HIF-α polypeptides for proteasomal destruction in the presence of oxygen.^36,37

In turn, this work framed the question as to what governed the association between HIF-α polypeptides and the VHL protein. We found that recombinant HIF-α proteins or synthetic peptides needed exposure to a cell extract in the presence of oxygen, before they could physically associate with VHL, indicating that an oxygen-dependent modification of HIF-α was essential. Our analyses of this process ultimately revealed that the key modification was the hydroxylation of specific prolyl residues in the HIF-α polypeptide. The work involved a combination of biochemical characterisation of the HIF-modify activity, genetic mutational approaches, mass spectrometry, and candidacy. Fig 3 depicts the ‘Eureka moment’ when predictions based on this work were confirmed; a synthetic peptide bearing a trans-4-hydroxyprolyl residue rather than proline at position 564 bound to VHL constitutively, without the need for any treatment with cell extract. These findings, and independent work from Bill Kaelin's laboratory, immediately suggested a mechanism of oxygen sensing.^38,39 The then known prolyl hydroxylases that catalyse procollagen prolyl hydroxylation to stabilise the collagen triple helix are 2-oxoglutarate-dependent dioxygenases. As dioxygenases (enzymes that split dioxygen and incorporate both oxygen atoms into their substrates) they have an absolute requirement for molecular oxygen. In the catalytic cycle, the oxidation of the prime substrate (in this case proposed to be the HIF peptide containing the target prolyl residue) is coupled to the oxidative decarboxylation of the Krebs cycle intermediate 2-oxoglutarate, yielding succinate and carbon dioxide.⁴⁰ In keeping with established properties of erythropoietin and HIF, these enzymes are also classically inhibited by cobaltous ions and iron chelators. The procollagen prolyl hydroxylases themselves had not seemed likely candidates, as the target prolyl peptide in HIF-α bore no resemblance to the typical PPG repeats that are targeted by the procollagen prolyl hydroxylases. In collaborative work, Johanna Myllyharju from Oulu, Finland demonstrated directly that procollagen prolyl hydroxylase had no activity on the HIF peptide. So we were clearly looking for another type of human prolyl hydroxylase, with the properties of a 2-oxoglutarate-dependent dioxygenase, which we knew from the earlier work must be conserved between human and invertebrate animals.

• Download figure
• Open in new tab
• Download powerpoint

Fig 3.

Definition of prolyl hydroxylation as the regulatory modification of HIF-a proteins governing binding to the von Hippel-Lindau protein (VHL). Binding of VHL to synthetic peptides derived from human HIF-1a sequences (556-574) was assayed by pull-down. The unmodified HIF peptide (19:WT) requires treatment with cell extract to promote binding of VHL. In contrast a HIF peptide bearing a trans-4-hydroxyprolyl substitution at residue 564 (19:Pro564Hyp) binds VHL without such treatment.

Looking back this might well have been recognised, and driven thinking, decades earlier by linking the classical (and highly distinctive) finding of induction of erythropoiesis by cobalt poisoning⁴¹ and the action of ascorbate to antagonise this effect (reported in the 1930s)⁴² to the properties of procollagen prolyl hydroxylases and related enzymes, which were identified as ascorbate-requiring enzymes that are inhibited by cobalt many years earlier.^43–45 But that is in retrospect; the work now directed the search for one or more 2-oxoglutarte dependent dioxygenases that catalyse HIF prolyl hydroxylation to generate the oxygen-dependent signal controlling VHL-dependent degradation.

To our good fortune, one of the leading authorities on this family of enzymes, Christopher Schofield, was also working in Oxford, in the Department of Chemistry. Chris's input was key in identifying the enzymes and he became a long-term collaborator in this and much subsequent work. The story of how scientific knowledge derived from apparently unconnected lines of work may come to together to answer an unanticipated question is one I'd like to emphasise. Chris had worked on 2-oxoglutarate dependent dioxygenases and related enzymes that are produced by microorganisms and involved in antibiotic synthesis.⁴⁶ But the X-ray structures he obtained on these enzymes also provided the basis for structurally informed predictions of many other putative 2-oxoglutarate dependent dioxygenases encoded by the human genome. Some of these, like the HIF system, were highly conserved across invertebrate animals. In an entirely different line of work, researchers working on the nematode worm Caenorrhabditis elegans had been painstakingly cataloguing mutants with different types of developmental abnormality. One piece of work by Bob Horvitz and colleagues identified 145 mutants with a ‘egg-laying defective’ phenotype due to neuromuscular defects that lead to an inability to extrude their eggs.⁴⁷ Following our earlier attempts to trace the phylogeny of the HIF system in Drosophila cells, a post-doc in the laboratory, John O'Rourke, identified a putative orthologue of HIF-α in C elegans. Critically, he raised an antibody to this protein to enable us to test worms bearing mutations in genes that were hypothesised to be involved in the oxygen-sensing process. The antibody worked beautifully; simply exposing worms to an atmosphere of 1% oxygen in a bell jar massively induced HIF-α, as assayed by immunoblotting. But none of the mutant worms we had tested had shown any alteration in this response.

However, very interestingly one of the putative human 2-oxoglutarate-dependent dioxygenases that Chris Schofield had indicated was a good candidate, and which was highly conserved in C elegans, corresponded to one of the mutants catalogued as ‘egg-laying defective’ by Bob Horvitz, egl-9. The mutants were duly ordered from the very well curated C elegans repository. Another of those Eureka moments followed shortly, with Andy Epstein, a trainee cardiologist and PhD student who had taken on the C elegans work, bursting through the office door with the result (Fig 4). In contrast with wild-type worms (and the multiple other mutants we had tested), all worms bearing mutant alleles in egl-9 showed constitutive upregulation of the HIF-α protein, irrespective of oxygen levels, as would be expected if their ‘oxygen sensing’ enzyme was defective.⁴⁸ We were rapidly able to confirm that egl-9 did indeed encode a HIF prolyl hydroxylase and that in humans there are three paralogues bearing the highly conserved catalytic domain, which we called PHD (prolyl hydroxylase domain) 1, 2 and 3.⁴⁹ In human HIF-1α and HIF-2α there are two sites of prolyl hydroxylation, each of which can independently promote VHL-dependent proteolysis, accounting for two of three domains in HIF-α polypeptides that had been shown to independently confer oxygen-regulated activity on heterologous proteins earlier in the work. The third oxygen-regulated domain was later shown to be subjected to hydroxylation by a second type of 2-oxoglutarate-dependent dioxygenase that catalyses the hydroxylation of specific asparaginyl residue to reduce transcriptional activity, at least in part by blocking the recruitment of transcriptional co-activators to the HIF complex.^49–51 Thus, HIF is subject to a dual mode of regulation, being both degraded and inactivated in the presence of oxygen (Fig 5).

• Download figure
• Open in new tab
• Download powerpoint

Fig 4.

Identification of the ‘oxygen sensing’ HIF prolyl hydroxylase in the nematode worm C elegans.⁴⁸ Immunoblot of C elegans HIF-α protein using an antibody produced ‘in house’ to test responses of wild-type and mutant worms to hypoxia. Whereas HIF-α is strongly induced by hypoxia in wild-type worms (wt), HIF-α protein is stabilised and upregulated irrespective of oxygen in worms bearing different mutant alleles of the gene encoding the HIF proyl hydrolylase egl-9.

• Download figure
• Open in new tab
• Download powerpoint

Fig 5.

Schematic illustrating the regulation of HIF by prolyl and asparaginyl hydroxylation. HIF prolyl hydroxylases target two sites in the oxygen-dependent degradation domain (ODDD) of HIF-α sub-units to promote association with the von Hippel-Lindau (VHL) ubiquitin E3 ligase and destruction by the ubiquitin proteasome pathway. A HIF asparaginyl hydroxylase targets the C-terminal transcriptional activation domain (tA) to prevent association with co-activators. In hypoxia, hydroxylation of HIF-α is suppressed allowing it to escape proteolytic destruction and form an active transcriptional complex.

With a few possible exceptions, all animals possess at least one representative of each of the key components of the HIF-PHD-VHL system and most also possess a HIF asparaginyl hydroxylase. Gene duplication events at the base of vertebrate evolution generated additional copies of several components.⁵² Thus, humans possess three isoforms of HIF-α, and three isoforms of the HIF prolyl hydroxylase. Given the interest in selective therapeutic manipulation of HIF pathway in human disease, an interesting question raised by these findings concerns the extent that these duplicated isoforms are specifically involved in the development or regulation of the complex oxygen delivery systems such as the heart, lungs and blood vessels that are found in higher animals and which are often the target of medical intervention. Some evidence for this is seen in the functions of the paralogues HIF-1α and HIF-2α. Two of the classical adaptations to hypoxia, increased erythropoiesis and increased ventilatory sensitivity to hypoxia (characteristics of altitude acclimatisation so beautifully defined in FitzGerald's early studies), appear to be specifically dependent on HIF-2α,^53,54 whereas a number of genes involved in the more conserved metabolic functions, such as glycolysis, appear to be specifically dependent on HIF-1α.⁵⁵ HIF-2α manifests much greater tissue specificity of expression than HIF-1α and in keeping with the above roles, HIF-2α is very highly expressed in the interstitial fibroblasts of the kidney⁵⁶ (which synthesise erythropoietin) and in the chemo-sensitive cells of the carotid body (that are important in the hypoxic control of ventilation).^57,58 Nevertheless, across the genome, in different cells, HIF-1α and HIF-2α each have a very extensive and partially overlapping set of transcriptional targets that extend well beyond these functions.⁵⁹ So even isoform-specific therapeutic manipulation of HIF would need to take account of pleiotropic biological actions.

Oxygen sensing in different kingdoms of life

As already outlined, we were initially surprised that the HIF hydroxylase system operated widely across human cells and was not confined to the regulation of erythropoietin. Given the central importance of oxygen homeostasis to most forms of life, we were next surprised that it appeared restricted to the animal kingdom. Interestingly, subsequent work has now revealed all four eukaryotic kingdoms in fact use enzymatic protein oxidation, but of different types, coupled to proteostasis to signal oxygen levels. Some of these systems appear closely related to the HIF hydroxylase pathway. Thus, the slime-mould Dictyostelium discoideum uses a similar oxygen sensitive prolyl-4-hydroxylase that targets a ubiquitin ligase directly, as opposed its substrates.⁶⁰ Fission yeast deploy a different type of prolyl hydroxylase, a prolyl-3-hydroxylase, to regulate highly oxygen-dependent biosynthetic pathways such a sterol synthesis.⁶¹ Although this system has a biochemical orthologue in human cells,⁶² we have so far been unable to define the physiological function, the oxygen-dependent regulation of cholesterol synthesis in human cells being connected to the HIF pathway.⁶³ One system which particularly interested us, as it apparently bears no molecular similarities to the HIF system, yet uses the same principle, is that which operates in plants to control the proteolytic stability of ethylene response factor (ERF)-VII transcription factors.^64,65 Here, the enzymatic oxidation is of a different type, the dioxygenation of N-terminal cysteine residues, revealed on putative substrates by methionine aminopeptidases. N-cysteine dioxygenation in plants is catalysed by a series of enzymes known as the plant cysteine oxidases;⁶⁶ following this oxidation, substrates are targeted for degradation by the N-degron pathway.⁶⁷ Thus, plants and animals use conceptually similar (yet somewhat counterintuitive) system of synthesis and oxygen-dependant degradation to signal oxygen levels. As part of examining mechanisms contributing to oxygen homeostasis in human cells more completely, we found that human cells encode an enzyme (ADO, 2-aminoethanethiol dioxygenase, otherwise known as cysteamine dioxygenase) that is similar to the plant cysteine oxidases and catalyses N-cysteine dioxygenation of a series of human proteins.⁶⁸ Among its targets, this system controls the stability of a set of G-protein regulators, RGS (regulator of G-protein signalling) 4, 5 and 16; direct proteolytic regulation of these signalling molecules potentially contributing to oxygen homeostasis on a shorter time-scale than transcriptional pathways mediated by HIF.