HIF oxygen sensing — prolyl hydroxylation of HIF-alpha by PHD1-3, CRL2(VHL) ubiquitination and degradation, FIH asparaginyl hydroxylation, and HIF-alpha/ARNT dimerisation with p300/CBP; EGLN1/EGLN2/EGLN3/OS9/VHL/ELOB/ELOC/CUL2/RBX1/SAT2/HIF1AN/HIF1A/EPAS1/HIF3A/ARNT/ARNT2/EP300/CREBBP/RACK1/SAT1

Hypoxia-inducible factor (HIF) is the master transcriptional regulator of the metazoan response to low oxygen. It is a heterodimer of an oxygen-labile alpha subunit (HIF-1alpha, HIF-2alpha/EPAS1, or HIF-3alpha) and a constitutive bHLH-PAS beta subunit (ARNT, also called HIF-1beta, or its paralog ARNT2). The oxygen signal is read by two families of Fe(II)- and 2-oxoglutarate-dependent dioxygenases that use molecular O2 as a cosubstrate, so their activity falls as oxygen falls. The prolyl hydroxylases PHD1, PHD2 and PHD3 (EGLN2, EGLN1 and EGLN3) hydroxylate two prolines in the oxygen-dependent degradation domain of the alpha subunit. trans-4-Hydroxyproline creates the recognition mark for the von Hippel-Lindau protein (VHL), the substrate receptor of a cullin-2 RING ubiquitin ligase (CRL2) in which the Elongin B/C module (ELOB, ELOC) bridges VHL's VHL-box to the CUL2 scaffold and the RING subunit RBX1 transfers ubiquitin. Polyubiquitinated HIF-alpha is destroyed by the proteasome, giving the protein a half-life of minutes in oxygenated cells. In parallel, the asparaginyl hydroxylase FIH (HIF1AN) hydroxylates a conserved asparagine in the C-terminal transactivation domain, which blocks binding of the coactivators p300 (EP300) and CBP (CREBBP) and so silences the fraction of alpha subunit that escapes degradation. When oxygen is scarce both hydroxylations fail: the alpha subunit accumulates, enters the nucleus, dimerises with ARNT through its bHLH and PAS domains, binds hypoxia response elements and recruits p300/CBP to activate transcription of genes for angiogenesis, erythropoiesis, glycolysis and iron handling. Several proteins tune the switch without sensing oxygen themselves: OS-9 scaffolds HIF-1alpha with the prolyl hydroxylases, SSAT2 (SAT2) stabilises the VHL-Elongin C interaction, and RACK1 with SSAT1 (SAT1) recruits the Elongin B/C ligase to HIF-1alpha independently of hydroxylation and of VHL. Loss of VHL abolishes degradation and gives constitutive HIF activity, the basis of von Hippel-Lindau disease and of clear-cell renal carcinoma; EPAS1 and EGLN1 mutations cause erythrocytosis and paraganglioma.

Derived from the RGD Pathway Ontology diagram PW:0000360 through the RGD_PATHWAY_ONTOLOGY project's diagram->module route. RGD curates 20 human genes to PW:0000360 (ARNT ARNT2 CREBBP CUL2 EGLN1 EGLN2 EGLN3 ELOB ELOC EP300 EPAS1 HIF1A HIF1AN HIF3A OS9 RACK1 RBX1 SAT1 SAT2 VHL). All 20 are modelled here. The mapped GO process GO:0097411 is the module's core concept and the broader GO:0071456 is kept as prose; both mappings come from the project's curated SSSOM file. The fly counterpart is modules/dmel_hif_complex.yaml, which models only the Sima-Tgo heterodimer (the FlyBase HIF gene group) and explicitly leaves the oxygen sensor Fatiga (Hph) and the VHL ligase out as regulators; this mammalian module takes the opposite boundary, because the RGD pathway is the sensing-and-degradation switch and the heterodimer is one step inside it. Nothing is duplicated between the two: the fly module is referenced here only in prose, since cross-species inclusion by module_ref would assert an identity between the fly complex and the human one that this module does not need. Molecular functions were taken from each protein's experimental GO annotations in QuickGO (checked 2026-10-28) and, where a gene review exists, cross-checked against its core functions. EGLN1, EGLN2 and EGLN3 all carry GO:0160082 hypoxia-inducible factor-proline dioxygenase activity (EGLN1 EXP/IDA and the core function of the completed EGLN1 review; EGLN2 and EGLN3 EXP), whose definition and Rhea xref RHEA:48400 are exactly the reaction this module's first step performs; the parent GO:0031545 peptidyl-proline 4-dioxygenase activity was not used for the HIF reaction, although it is the term the Reactome HIF reactions map to and the term the EGLN3 GO-CAM uses for its GPX4 substrate. HIF1AN carries GO:0036140 [protein]-asparagine 3-dioxygenase activity (IDA/IMP; RHEA:54260). VHL carries GO:1990756 ubiquitin-like ligase-substrate adaptor activity (IDA), ELOC GO:0030674 protein-macromolecule adaptor activity (IBA; core in the ELOC review), ELOB GO:0031625 ubiquitin protein ligase binding (IDA; core in the ELOB review), CUL2 GO:0160072 ubiquitin ligase complex scaffold activity (IDA) and RBX1 GO:0061630 ubiquitin protein ligase activity (IDA; core in the RBX1 review). The alpha subunits carry GO:0000981 DNA-binding transcription factor activity, RNA polymerase II-specific (HIF1A IDA, HIF3A IDA, EPAS1 IBA/ISA), the term both human HIF1A GO-CAMs use; the more specific GO:0001228 DNA-binding transcription activator activity, RNA polymerase II-specific is available for HIF1A (IDA/IMP) but not experimentally for EPAS1 or HIF3A, and HIF3A's IPAS variant is a repressor, so the shared parent is used across the variant set. ARNT and ARNT2 carry GO:0046982 protein heterodimerization activity (both IDA; core in the ARNT review) — the beta subunit's role here is to provide the dimerisation partner and the composite DNA-binding surface, and its own GO:0000981 annotation is contributes_to (IDA) rather than enables. EP300 and CREBBP carry GO:0003713 transcription coactivator activity (IDA; core in both reviews). Boundary decisions. (1) Target genes are out of scope. VEGFA and EPO are the outputs the pathway acts on, not participants: they bind no HIF, hydroxylate nothing and carry no signalling activity, exactly the substrate-versus-participant distinction CLAUDE.md requires. RGD draws them behind a single "Hif target genes" sub-page link rather than as genes, and neither is in the PW:0000360 gene list. (2) OS9, SAT2, RACK1 and SAT1 are modelled as regulatory parts rather than relegated to context, because each has a primary paper with gain- and loss-of-function evidence placing it in a physical complex with HIF-1alpha and with a named component of the degradation machinery (OS-9 with the prolyl hydroxylases, PMID:15721254; SSAT2 with VHL and Elongin C, PMID:17558023; RACK1 with Elongin C, PMID:17244529; SSAT1 with RACK1, PMID:17875644). None of the four has a GO molecular function that describes this role: OS9's only experimental MF is GO:0140032 glycosylation-dependent protein binding (IDA), from its ERAD lectin function; SAT1 and SAT2 carry only their polyamine acetyltransferase activity GO:0004145 (and SAT2 also a NOT|enables GO:0004145 IDA row, reflecting the dispute over whether it acetylates spermidine at all), which is not the activity at work here; RACK1's adaptor role is captured only by the broad GO:0060090 molecular adaptor activity (TAS; core in the RACK1 review, which already records HIF1A as a client). Their annotons therefore carry a preferred_term and evidence but no GO identifier, except RACK1 where the broad adaptor term is used, rather than asserting a plausible-looking term the ontology does not support. This is a candidate for proposed new terms once the scaffolding roles are better characterised. (3) The oxygen-independent arm (RACK1 + SAT1) is marked optional: it operates on the same substrate but is not required for the oxygen switch, and it was defined largely through HSP90 inhibition. (4) HIF3A is modelled as a third alpha variant rather than as a separate inhibitory module. Its full-length product is a hydroxylation-regulated, ARNT-dimerising transcription factor (GO:0000981 IDA, GO:0000978 IDA) that lacks a C-terminal transactivation domain, and its short IPAS splice variant is a dominant-negative that sequesters HIF-1alpha (PMID:11734856). Both readings are recorded on the annoton; splitting the locus into two modules would overstate how well the isoform repertoire is resolved. (5) HSP90, which competes with RACK1 for the PAS-A domain, is left out: it is a general chaperone whose involvement is shown mainly through inhibitor treatment, and including it would pull the HSP90 cycle into the module. The same applies to the HIF target-gene feedback on EGLN1 (PHD2 is itself HIF-inducible) and to the 2-oxoglutarate and ascorbate supply that all four dioxygenases need. (6) The Elongin B/C module is modelled as two explicit annotons, not folded into a complex descriptor, because both ELOB and ELOC have completed gene reviews whose core functions are exactly these adaptor activities, and because the same module serves CRL5 ligases elsewhere (see modules/reelin_signaling.yaml, whose CRL5(SOCS7) part omits Elongin B/C following its GO-CAM). The generic cullin-RING machinery is not re-modelled here: neddylation, the CAND1/CSN cycle and the proteasome itself are outside the boundary, and the 26S proteasome step is recorded as a process term on the ligase annotons rather than as a part. GO-CAM and GOA observations for upstream review. No production GO-CAM covers the oxygen sensor: gocams/index.tsv (October 2026) has no model containing VHL, HIF1AN, EGLN1, EGLN2, ELOB, ELOC, EPAS1, HIF3A, OS9, SAT1 or SAT2, and the two HIF1A models (gomodel:689e7a5d00004061, gomodel:689e7a5d00004412) use HIF1A only as the transcription factor downstream of COXFA4L2. This is the largest curation gap the module reveals. In GOA, GO:0097411 hypoxia-inducible factor-1alpha signaling pathway is carried in human, mouse and rat by CYBB, PDK1, PDK3, mouse Hif1a and miR-210, but by none of EGLN1-3, VHL, HIF1AN, ARNT, EP300 or CREBBP; EGLN1 and mouse Vhl instead carry the regulatory term GO:1902072 negative regulation of hypoxia-inducible factor-1alpha signaling pathway, which is the convention this module follows (the hydroxylase and the ligase regulate the pathway rather than being part of it), so no NEW pathway annotations are proposed for them here. SAT2's human GOA carries both enables and NOT|enables GO:0004145 by IDA from different references, which is worth an upstream look. Gene reviews exist for ARNT, EGLN1, ELOB, ELOC, RBX1, EP300, CREBBP, RACK1 and SAT1; HIF1A, EPAS1, HIF3A, ARNT2, VHL, EGLN2, EGLN3, HIF1AN, CUL2, OS9 and SAT2 have none, and PANTHER family ids and PTNs are deliberately not asserted.