Motif Tools
Family atlas / bHLH-PAS

HIF / ARNT

bHLH-PASRCGTG19 motifsaka HRE, Hypoxia response element, HIF binding site (HBS)
GAATCAGCTCAGGTAC
representative: AC0072:HIF1A/ARNT:bHLH · data consensus ACGTGc
members tight · 0.9stated RCGTG · data ACGTGc1 in vitro9 ChIP-only1 cross-species12 references, all PMIDs verified

The hypoxia response element is the site read by HIF, a heterodimer of an oxygen-labile alpha subunit (HIF-1alpha, HIF-2alpha/EPAS1, HIF-3alpha) and the constitutive beta subunit ARNT (HIF-1beta) or its paralogue ARNT2. Both partners are bHLH-PAS proteins. The element is an asymmetric E-box half-site pair: ARNT reads the 3' GTG and the alpha subunit reads a 5' half that is usually TAC or GAC, giving the five-base core (A/G)CGTG, most often written in context as ...TACGTGC... When oxygen is plentiful the alpha subunits are prolyl-hydroxylated and destroyed by the VHL ubiquitin ligase, so occupancy of these sites is a direct readout of oxygen availability. HIF drives erythropoietin, VEGFA and the glycolytic genes, and the pathway is constitutively active in VHL-mutant kidney cancer.

The factors

HIF is an obligate heterodimer of one oxygen-regulated alpha subunit and one constitutive beta subunit. The alpha subunits are HIF-1alpha (HIF1A), HIF-2alpha (EPAS1) and HIF-3alpha (HIF3A, which produces several splice isoforms, some of them dominant-negative); the beta subunits are ARNT (also called HIF-1beta) and the brain- and kidney-enriched paralogue ARNT2. All are bHLH-PAS proteins: a basic helix-loop-helix domain followed by PAS-A and PAS-B. Alpha-subunit prolines are hydroxylated by the PHD enzymes, creating the binding site for the VHL ubiquitin ligase; a separate asparagine hydroxylation by FIH blocks p300/CBP recruitment (Kaelin and Ratcliffe 2008) 18498744. HIF-1alpha is close to ubiquitous; HIF-2alpha is enriched in endothelium, kidney interstitium, lung and sympathoadrenal tissue. ARNT is the shared partner of the whole bHLH-PAS class: it also dimerizes with AHR, SIM1/2 and NPAS factors, and forms homodimers. Loss of Arnt in mice is embryonic lethal around E10.5 with defective angiogenesis and placental vascularization (Maltepe et al. 1997) 9121557. HIF-1 was first defined as the hypoxia-inducible activity at the erythropoietin enhancer (Semenza and Wang 1992) 1448077 and then identified as a bHLH-PAS heterodimer (Wang et al. 1995) 7539918.

The motif

The core is (A/G)CGTG, conventionally written RCGTG. Unlike the palindromic E-box CACGTG, the HIF site is an asymmetric pair of half-sites: selection experiments assign the 3' half-site GTG to ARNT, while the 5' half comes from the partner (AHR prefers T(C/T)GC and SIM prefers GT(G/A)C, so ARNT is the constant element in all of them) (Swanson et al. 1995) 7592839. For HIF the 5' half is usually TAC or GAC, so the full site reads TACGTGC or GACGTGC. One substitution, C to T at the first position, is therefore all that separates an HRE from a canonical E-box. Crystal structures of mouse HIF-2alpha-ARNT and HIF-1alpha-ARNT bound to an HRE show ARNT spiraling around the outside of the alpha subunit; the DNA-reading head rotates and extends and cooperates with a distal PAS domain, and cancer-associated alpha-subunit mutations map to residues that establish DNA binding (Wu et al. 2015) 26245371. Systematic measurements across the family find 12-15 bp footprints in which flanking-base preferences, not the core, distinguish members that share a core (Bersten et al. 2026) 41495897. The consensus computed from the 19 members here is ACGTGc, core plus a 3' C preference, matching the compiled consensus HRE (Wenger et al. 2005) 16234508. Nothing in the sequence distinguishes HIF-1alpha from HIF-2alpha (Smythies et al. 2019) 30429208. Two entries need care: JASPAR EPAS1 (MA2325.1) is written GCACGTAC, the reverse complement of GTACGTGC, so it is the same site on the other strand; JASPAR Arnt (MA0004.1) is an ARNT-homodimer palindrome, not an HRE.

Where it acts

RCGTG is expected several million times per strand in a 3-Gb genome, and only a small fraction is ever occupied. Chromatin is the dominant filter: occupancy is enriched roughly 20-fold at DNase-hypersensitive sites that were already open in normoxia, irrespective of distance from promoters (Schodel et al. 2011) 21447827. Confirmed sites number in the hundreds to low thousands per cell type and often lie far from the genes they regulate. The two isoforms partition consistently: HIF-1 is biased toward promoter-proximal sites and HIF-2 toward promoter-distant sites, each with a cell-type-specific repertoire that does not redistribute with hypoxia severity or duration (Smythies et al. 2019) 30429208. Because the core contains a CpG, HREs cluster with CpG-rich promoters and are sensitive to DNA methylation. Functional HREs generally need neighboring sequence, historically called the HIF ancillary sequence, and commonly co-occur with CREB/ATF, SP1 and AP-1 sites (Wenger et al. 2005) 16234508.

Biology and disease

HIF targets include EPO, VEGFA, the glucose transporter SLC2A1, LDHA, PDK1 and CA9, coordinating oxygen delivery with a shift to glycolysis. Biallelic VHL loss in clear-cell renal carcinoma leaves HIF-2 constitutively active, which made the PAS-B pocket seen in the HIF structures a drug target (Wu et al. 2015; Kaelin and Ratcliffe 2008) 26245371,18498744. Somatic gain-of-function EPAS1 mutations that block hydroxylation cause paraganglioma with polycythemia (Zhuang et al. 2012) 22931260, and EPAS1 carries the strongest selection signal in Tibetan high-altitude adaptation (Yi et al. 2010) 20595611. The pathway is also the target of prolyl-hydroxylase inhibitors used for renal anemia.

Reading a hit

An RCGTG match on its own licenses almost no conclusion: five informative base pairs occur millions of times, and the great majority are never bound. Treat it as a candidate only in accessible chromatin, and preferably with hypoxia-responsive expression nearby. A hit cannot tell you which alpha subunit bound, and it barely distinguishes HIF:ARNT from the other ARNT complexes: AHR:ARNT prefers GCGTG, SIM:ARNT GT(G/A)CGTG, and ARNT homodimers the palindromic CACGTG, so the discriminating information sits in one or two weakly constrained positions. Evidence is skewed: eight of the ten HOCOMOCO models here are ChIP-only (letter P) at grade B or C, supporting "this region was occupied in cells" rather than "this protein binds this sequence"; the one HT-SELEX model, ARNT2.H14CORE.2.S.C, is a palindromic CACGTGAC of the kind an ARNT2 homodimer makes. The CIS-BP ARNT entry M01719_2.00 is transferred from mouse at 100% DNA-binding-domain identity: safe, but not a human measurement. The Vierstra archetype AC0072:HIF1A/ARNT is named after the factors whose motifs clustered together and is not a claim that HIF1A and ARNT can be told apart.

How this was worked out

  1. 1992mediumElectrophoretic mobility shift, enhancer mutagenesis in reporter assays1448077

    A hypoxia-inducible nuclear activity was mapped to a short element in the 3' enhancer of the human erythropoietin gene that is required for its transcriptional induction, defining the hypoxia response element operationally before any protein was known.

    Still open then: The identity of the binding protein and the generality of the element beyond the EPO locus.

  2. 1995highProtein purification, peptide sequencing, reconstitution7539918

    HIF-1 was purified and shown to be a heterodimer of two bHLH-PAS proteins, an oxygen-regulated alpha subunit and the constitutive beta subunit now called ARNT, explaining why the element is an asymmetric pair of E-box half-sites.

  3. 1995mediumIn vitro binding-site selection (SELEX) with recombinant proteins7592839

    Binding-site selection with AHR, ARNT and SIM established the pairing rules for the class: ARNT reads the 3' GTG half-site in every complex while the partner supplies the 5' half, so the few bases 5' of CGTG carry almost all of the discriminating information.

    Still open then: Whether one or two weakly constrained positions suffice to separate HREs from AHR and SIM sites in genomic sequence.

  4. 2008highReview of biochemistry, genetics and structural work18498744

    Prolyl hydroxylation of the alpha subunits by the PHD enzymes followed by VHL-dependent degradation, with a separate asparaginyl hydroxylation blocking coactivator recruitment, was consolidated as the mechanism that makes occupancy of these sites a readout of oxygen tension rather than of sequence.

  5. 2011highChIP-seq in human cells with DNase-hypersensitivity comparison21447827

    Genome-wide ChIP-seq showed that HIF binding is concentrated at sites already DNase-hypersensitive in normoxia, roughly 20-fold enriched there and largely independent of distance from promoters, so chromatin rather than the five-base core is the main filter on which matches are used.

    Still open then: Why many accessible RCGTG matches nonetheless stay unbound.

  6. 2015highX-ray crystallography26245371

    Crystal structures of mouse HIF-2alpha-ARNT and HIF-1alpha-ARNT bound to an HRE showed ARNT spiraling around the outside of the alpha subunit, with the DNA-reading head rotated and cooperating with a distal PAS domain, and placed cancer-associated alpha-subunit mutations at residues that establish DNA binding.

  7. 2019mediumChIP-seq across cell types and hypoxic conditions30429208

    HIF-1 and HIF-2 were shown to partition consistently in chromatin, HIF-1 toward promoter-proximal and HIF-2 toward promoter-distant sites, with cell-type-specific repertoires that do not redistribute with the severity or duration of hypoxia, although neither isoform has a distinguishable consensus.

    Still open then: What establishes the isoform partition, given an identical core preference.

  8. 2026mediumHigh-throughput in vitro binding across family members41495897

    Systematic binding measurements across bHLH-PAS proteins reported footprints of 12-15 bp in which preferences for bases flanking the shared core, rather than the core itself, separate members, revising the long-standing description of these sites as 5-6 bp elements.

    Still open then: Whether the flanking preferences are strong enough to assign genomic matches to one complex.

Still unresolved

How much to trust a hit

The core is only five informative base pairs, so chance matches vastly outnumber real sites and a hit is close to uninformative without chromatin or expression evidence. HIF-1alpha and HIF-2alpha share one consensus and cannot be distinguished by sequence; neither can HIF:ARNT be cleanly separated from AHR:ARNT, SIM:ARNT or ARNT homodimers, since all read CGTG with differences at one or two flanking positions. Most HOCOMOCO models here are ChIP-only at grade B or C, so they reflect occupied regions and may include indirect binding; ChIP peaks lacking any RCGTG are well documented. The ARNT entries in JASPAR and CIS-BP include palindromic E-boxes that are homodimer, not HIF, motifs, and one JASPAR entry is written on the opposite strand.

References

19 motifs in this family — JASPAR 5 · HOCOMOCO 10 · CIS-BP 3 · Vierstra 1
motifdatabaseconsensusbits
ARNT2.H14CORE.2.S.CSCHOCOMOCOCACGTGAC14.9
M04170_2.00CIS-BPGTCACGTG12.5
ARNT2 (MA1464.2)JASPARGTCACGTG12.2
EPAS1 (MA2325.1)JASPARGCACGTAc11.6
ARNT.H14CORE.1.P.BPBHOCOMOCOtACGTGccngnnngC11.1
Arnt (MA0004.1)JASPARCACGTG11
HIF3A.H14CORE.0.P.CPCHOCOMOCOACGTGcnngnnngCacg11
HIF1A.H14CORE.1.P.BPBHOCOMOCOACGTGccnnnnnnC10.9
HIF1A.H14CORE.0.P.BPBHOCOMOCOtACGTGc10.6
M08049_2.00CIS-BPACGTGC10.3
HIF1A (MA1106.2)JASPARACGTGC10.1
ARNT.H14CORE.0.P.BPBHOCOMOCOtACGTGc10.1
M01719_2.00Mus musculus 1CIS-BPgCACGTGA9.9
ARNT2.H14CORE.1.P.BPBHOCOMOCOGbACGTGc9.8
EPAS1.H14CORE.0.P.BPBHOCOMOCOACGTGc9.7
HIF3A.H14CORE.1.P.CPCHOCOMOCOgtACGTGc9.4
ARNT::HIF1A (MA0259.2)dimerJASPARACGTG9
ARNT2.H14CORE.0.P.BPBHOCOMOCOCGTGcctdG8.7
AC0072:HIF1A/ARNT:bHLHVierstraACGTGc8.7

HOCOMOCO tags read as assay + quality grade: P ChIP-seq, S HT-SELEX, M methyl-HT-SELEX, G GHT-SELEX, I SMiLE-seq, B PBM; grade A means two or more assay types agreed. A P-only matrix is a record of where the protein was crosslinked, which need not be a sequence it binds itself.

The factors, elsewhere

how this page was made Drafted by a language model reading the primary literature, then checked: every cited PMID is resolved against PubMed and its citation text regenerated from PubMed's own record, and every claim about the motifs themselves is checked against the matrices in the four source databases. Methods and limits →

Other bHLH-PAS families