Motif Tools
Family atlas / Forkhead

Forkhead (FOX)

ForkheadRYAAAYA106 motifsaka forkhead, FKH site, winged-helix site
CAGGCATCACAGATCCA
representative: M03022_2.00 · data consensus GTAAACA
members tight · 0.78stated RYAAAYA · data GTAAACA24 in vitro6 ChIP-only4 cross-species14 references, all PMIDs verified

Forkhead box (FOX) proteins are a family of about 50 human transcription factors that bind DNA as monomers through a ~100-residue winged-helix domain, named for the two loops, or wings, that flank its three-helix core. Nearly all of them read the same seven-base element, RYAAAYA, usually written GTAAACA or, on the opposite strand, TGTTTAC. The family includes the FOXA pioneer factors that engage nucleosomal DNA and open liver, prostate and breast enhancers; the insulin- and stress-controlled FOXO proteins; FOXC, FOXE, FOXG and FOXL factors needed for eye, thyroid, forebrain and ovary development; and the proliferation factor FOXM1. Because the DNA-contacting residues are almost invariant across the family, a forkhead match tells you that a FOX protein could bind there, but rarely which one.

The factors

The human genome carries about 50 forkhead box genes in 19 subfamilies, FOXA through FOXS 20650821. Each encodes one copy of the forkhead (winged-helix) domain: roughly 100 residues in which three alpha-helices pack against a short three-stranded beta-sheet. Helix H3 lies in the major groove and supplies the base-specific contacts, while two extended loops - wing 1, between strands beta2 and beta3, and wing 2 at the C-terminus - fold over the phosphate backbone and reach into the flanking minor groove. The first co-crystal structure, of HNF-3gamma (FOXA3) on DNA, showed that this fold is essentially that of linker histone H5 and that it bends DNA by about 13 degrees 8332212. Binding is monomeric in the usual case, but FOXO1 (and, in the same assays, FOXO3, FOXM1 and FOXI1) also homodimerizes on a palindromic element using wing 1 33577686. Expression ranges from near-ubiquitous (FOXK1/2, FOXO1/3, FOXJ3) to sharply restricted: FOXA1-3 in endoderm- derived epithelia, FOXE1 in thyroid, FOXG1 in telencephalon, FOXL2 in granulosa cells and eyelid, FOXI1 in inner ear and renal intercalated cells, FOXM1 in cycling cells.

The motif

The literature consensus is RYAAAYA (R = A/G, Y = C/T), seven base pairs, most often realized as GTAAACA 7957066,23836653; the medoid model of the 106 collected here is exactly GTAAACA, matching the databases. Specificity comes almost entirely from H3 contacts to the central AAA run and the following pyrimidine, several of them water-mediated, which is why that core tolerates so little variation and the flanks so much 8332212,7957066. The FOXO field calls the same element the DAF-16 binding element, core TTGTTTAC, and distinguishes it from the weaker insulin-response sequence in the IGFBP1 promoter 10880363. Longer models in the data are one of three things: an extended monomeric site with AT-rich flanks (HOCOMOCO FOXA1, TTGTTTACTTAA); a perfect palindrome, GTAAATATTTAC, which is two RYAAAYA half-sites head-to-head with no spacer and corresponds to the wing-1 homodimer site (HOCOMOCO FOXD3, FOXD2, FOXB1, FOXC1; CIS-BP FOXC2); or a composite with a partner, as in the JASPAR models FOXO1::ELK1, FOXO1::ELK3, FOXO1::ELF1 and FOXO1::FLI1, where most of the information content is the ETS GGAA core, not the forkhead site. Two members are real exceptions. FOXM1 models carry almost no information (2.1 bits in CIS-BP) because FOXM1 reaches chromatin mainly through cofactors and non-canonical sequences 26100407. FOXH1 (FAST-1/FAST-2) models read AATCCACA, equivalently TGTGGATT - the activin/Smad response element - and not RYAAAYA at all 8878477. Otherwise FOXA, FOXB, FOXC, FOXD, FOXI, FOXK, FOXL, FOXO, FOXQ, FOXR and FOXS are not separable by sequence 23836653.

Where it acts

Forkhead sites are AT-rich and enhancer-biased: depleted from CpG-island promoters, enriched in distal tissue-specific elements, and frequently on nucleosomal rather than naked DNA. FOXA1 and FOXA2 bind a nucleosomal site that overlaps the linker-histone position 9427758, displace H1, and hold liver enhancers in an accessible but nucleosome-occupied state 27058788. ChIP-seq typically recovers tens of thousands of FOXA1 sites per cell type, most of them distal, and which ones are used depends on the partner present: androgen receptor in prostate, estrogen receptor alpha and GATA3 in luminal breast, HNF4A and CEBP in liver. FOXA1 is required for almost all ER-chromatin contacts and for tamoxifen-dependent ER activity 21151129. FOXO sites sit at insulin- and oxidative-stress- responsive promoters and enhancers (IGFBP1, G6PC, SOD2) and are switched off by AKT-driven nuclear export, not by loss of the site. RYAAAYA is short - with two degenerate positions it is expected on the order of three million times in a 3.1 Gb double-stranded genome - so occupancy is set by chromatin and partners, not by the element.

Biology and disease

Forkhead loss-of-function is usually severe. FOXC1 mutations cause Axenfeld-Rieger malformations and early-onset glaucoma 9620769; FOXL2 mutations cause blepharophimosis-ptosis-epicanthus inversus syndrome, and the recurrent FOXL2 C134W substitution is present in most adult granulosa cell tumors 19516027; FOXE1 variants affect thyroid development; FOXG1 mutations produce a Rett-like encephalopathy. FOXA1 is the clearest cancer story. It is altered in roughly a third of prostate cancers, and the most common class of activating mutation clusters inside the wing-2 loop of the forkhead domain, increasing chromatin mobility and driving a luminal androgen-receptor program; other classes truncate the C-terminus or rearrange the locus 31243372. FOXA1 is also amplified or mutated in breast, bladder and salivary tumors and sets the response to endocrine therapy. FOXM1 is overexpressed in most carcinomas and drives cell-cycle gene expression 26100407.

Reading a hit

A GTAAACA/TGTTTAC match identifies a forkhead site, not a forkhead protein. All of FOXA, FOXO, FOXK, FOXC and FOXD read this element with the same residues, the models in all four databases collapse onto one cluster (cohesion 0.78, no member with out-of-family nearest neighbors), and the archetype names make the point explicitly: AC0042 is "FOXA/FOXB", AC0039 is "FOXD/FOXK", AC0093 is "FOXC/HOXA". Deciding which FOX protein is involved requires expression data, ChIP, or a partner site. The element is common enough to occur millions of times by chance, so a single match in a region of unknown function is weak evidence. Prefer the in vitro models when you need "this protein binds this sequence": 24 of the 30 HOCOMOCO entries here carry HT-SELEX or methyl- HT-SELEX evidence (S or M in the fourth field), and only six are ChIP-only. Treat the FOXO1::ETS and FOXJ2::ELF1 composite models as ETS motifs with a forkhead flank, and treat FOXM1 and FOXH1 models as not representing RYAAAYA.

How this was worked out

  1. 1993highX-ray crystallography8332212

    The co-crystal structure of the HNF-3gamma (FOXA3) forkhead domain on DNA shows a monomeric winged-helix fold resembling linker histone H5, with helix H3 in the major groove making direct and water-mediated base contacts and a 13-degree DNA bend.

    Still open then: Explained the core contacts but not how paralogues with near-identical recognition helices come to prefer different flanks.

  2. 1994mediumSelection from random oligonucleotides, domain swaps, circular permutation7957066

    Binding-site selection with four human forkhead proteins gives a shared core, RTAAAYA, with the paralogues differing in the flanking positions; domain swaps map those flank differences to two subregions of the forkhead domain.

    Still open then: The 80-90 degree bend it measured by circular permutation did not match the 13-degree bend seen in the crystal structure.

  3. 1998mediumNucleosome reconstitution, footprinting in vitro and in vivo9427758

    HNF3 (FOXA) binds its site at the side of a nucleosome core, where linker histone sits, and occupies such a site at the albumin enhancer in active chromatin: the starting point of the pioneer-factor model.

  4. 2000mediumBinding-site selection, EMSA10880363

    DAF-16 and its three mouse homologues (FOXO1, FOXO3, FOXO4) share the core TTGTTTAC, the DAF-16 binding element, and bind it more strongly than the IGFBP1 insulin-response sequence that had been used as their site.

  5. 2013mediumProtein-binding microarrays, phylogenetic inference23836653

    Protein-binding-microarray profiling across the forkhead phylogeny finds the canonical site shared by most subfamilies, while alternative specificities, including the FHL site, arose repeatedly and mostly without changes at the known DNA-contacting residues.

  6. 2015mediumChIP-seq of wild-type and DNA-binding-deficient FOXM1, fluorescence polarization, interactome analysis26100407

    FOXM1 ChIP-seq peaks lack the forkhead motif, yet a DNA-binding-deficient FOXM1 is not recruited, supporting direct contact with non-canonical sequence in addition to cofactor interactions.

    Still open then: The non-canonical sequences themselves were not defined.

  7. 2016mediumMNase titration, core histone mapping, ChIP-seq27058788

    In mouse liver, FoxA displaces linker histone H1 and keeps enhancer nucleosomes accessible rather than evicting them, revising the picture of active enhancers as nucleosome-free.

  8. 2021highX-ray crystallography, EMSA, ITC, mutagenesis33577686

    The FOXO1 DNA-binding domain crystallizes as a wing-1-mediated homodimer on the head-to-head palindrome GTAAACATGTTTAC (DIV2), which it binds more tightly than a single site; FOXO3, FOXM1 and FOXI1 also bind it as dimers, FOXC2 only as a monomer.

    Still open then: Did not test the overlapping palindrome GTAAATATTTAC that appears in many HT-SELEX models of FOXA, FOXB, FOXC and FOXD proteins.

Still unresolved

How much to trust a hit

High confidence that a match is a forkhead site; very low confidence about which of the ~50 FOX proteins binds it. Sequence cannot separate FOXA from FOXO, FOXC, FOXD, FOXI, FOXK, FOXL or FOXQ, and the Vierstra archetype names (FOXA/FOXB, FOXD/FOXK, FOXC/FOXK) reflect that rather than resolving it. The element is short and AT-rich, so it occurs millions of times by chance; treat matches as candidates that need chromatin or expression evidence. Two labels in this family do not represent RYAAAYA: FOXM1 models are near-uninformative, and FOXH1 models are the activin/Smad response element. Four CIS-BP entries are mouse-derived (FOXJ1, FOXM1, and the inferred FOXS1 and FOXB2 models at 79% and 93% DBD identity).

References

106 motifs in this family — JASPAR 29 · HOCOMOCO 30 · CIS-BP 29 · Vierstra 18
motifdatabaseconsensusbits
AC0036:FOXO:ForkheadVierstraGTAAACATGTTTAC22.6
M04831_2.00CIS-BPccTTAAGTAAACAAA22.2
M03049_2.00CIS-BPGTAAACATGTTTAC21.2
M04837_2.00CIS-BPtatGtAAATATTgACaTa20.5
FOXD3.H14CORE.1.S.BSBHOCOMOCOAhgtaAATATTTACtTAa19.3
M04857_2.00CIS-BPAtATAaACATAAATa18.4
FOXD2.H14CORE.0.S.BSBHOCOMOCOGTAAATATTTActTa18.3
AC0043:FOXR:ForkheadVierstrataTAaACATAAATa18.3
FOXO1::FLI1 (MA1956.2)dimerJASPARGtAAACAGGAAGT18.2
FOXL1.H14CORE.0.S.BSBHOCOMOCOTGTTTgTTTAttTt17.8
AC0362:FOXG/FOXK:ForkheadVierstraCGGACACAAT17.6
FOXA1.H14CORE.1.S.CSCHOCOMOCOTTGTTTACtTAa17.4
FOXI1.H14CORE.0.SM.BSMBHOCOMOCOTTGTTTACatTa17.1
M04845_2.00CIS-BPctTAAAtAAACAa17.1
AC0037:FOXJ:ForkheadVierstraTgTTTACAACA17.1
FOXB1.H14CORE.0.SM.BSMBHOCOMOCOtaAATATTTACatA16.8
FOXC1.H14CORE.0.S.BSBHOCOMOCOTvAATATTTACaTA16.4
FOXA2.H14CORE.1.SM.BSMBHOCOMOCOtcAATATTTAC15.9
FOXO1::ELK3 (MA1955.2)dimerJASPARgtaAACAGGAAGT15.3
FOXO1::ELK1 (MA1954.2)dimerJASPARatcAACAGGAAG15
FOXC2.H14CORE.0.SM.BSMBHOCOMOCOTTGTTTActTa15
FOXE1 (MA1487.3)JASPARcTAAAtAAACAA14.8
M02680_2.00CIS-BPAacGTAAACAa14.8
AC0042:FOXA/FOXB:ForkheadVierstraGTaAATATTtaC14.6
FOXE1.H14CORE.0.SM.BSMBHOCOMOCOTTGTTttTtTAggg14.5
FOXR2.H14CORE.0.M.CMCHOCOMOCOATTTATgTaTATaT14.4
FOXD2 (MA0847.4)JASPARtTAAAcAAACA14.1
AC0332:FOXJ:ForkheadVierstrataATAATATTT14.1
Foxq1 (MA0040.2)JASPARATTGTTTATt14
FOXF2 (MA0030.2)JASPARcGTAAACAa13.8
M08119_2.00CIS-BPAATCCACA13.8
FOXA3.H14CORE.1.M.CMCHOCOMOCOgcaAATATTTACttA13.7
AC0331:FOXD:ForkheadVierstraAACAATAACAt13.6
FOXH1 (MA0479.2)JASPARAATCCACA13.4
FOXO1::ELF1 (MA1953.2)dimerJASPARgtaAACAGGAAGt13.4
FOXG1 (MA0613.1)JASPARGTAAACAa13.3
M01978_2.00CIS-BPGTAAACAa13.3
M04813_2.00CIS-BPtCCCCACAC13.2
M07960_2.00CIS-BPctaaGTAAACA13.1
FOXA2.H14CORE.0.PSM.APSMAHOCOMOCOTGTTTACatA13
AC0610:FOXO:ForkheadVierstraTTCCCCACAC12.9
FOXH1.H14CORE.0.P.BPBHOCOMOCOTGTGGATT12.5
FOXF1.H14CORE.0.P.CPCHOCOMOCOTGTTTAt12.4
FOXA3.H14CORE.0.PS.APSAHOCOMOCOTGTTTACtta12.3
FOXB1 (MA0845.1)JASPARTATGTAAATAt12.2
AC0639:FOXO1::ELK/ERF::FOXO:['Tryptophan_cluster_factors',_'Fork_head/winged_helix_factors']VierstraaAACAGGAAG12.2
M03036_2.00CIS-BPGTcAATATTtAC12.1
FOXO6 (MA0849.1)JASPARGTAAACA12
M07959_2.00CIS-BPaaaGTAAACA12
FOXC1 (MA0032.2)JASPARTAtGTAAAtAt11.9
FOXL2.H14CORE.0.PSM.APSMAHOCOMOCOTGTTTAcatT11.9
FOXO6.H14CORE.0.S.BSBHOCOMOCOTGTTTAC11.9
M02681_2.00CIS-BPGTAAACA11.9
M03042_2.00CIS-BPgTaAATATTtwt11.9
FOXD3 (MA0041.3)JASPARaangtAAATAAACa11.8
FOXD1 (MA0031.2)JASPARGTAAACA11.7
M00161_2.00Mus musculus 1CIS-BPtAAACAAA11.7
FOXI1 (MA0042.2)JASPARGTAAACA11.6
FOXE1.H14CORE.1.S.CSCHOCOMOCOTTGTTT11.6
M03026_2.00CIS-BPGTAAACA11.6
Foxl2 (MA1607.2)JASPARATGTAAACA11.5
FOXC1.H14CORE.1.S.CSCHOCOMOCOtTgTTTACtt11.5
M08120_2.00CIS-BPGTAAACA11.5
FOXC2 (MA0846.2)JASPARtAaGTAAACAa11.4
FOXO3.H14CORE.0.PS.APSAHOCOMOCOTGTTTAC11.2
M04823_2.00CIS-BPtTAtaTAAACa11.2
FOXD2.H14CORE.1.SM.BSMBHOCOMOCOTGTTATTgTaa11.1
AC0345:FOXH:ForkheadVierstraAATCCACA11.1
FOXA1 (MA0148.5)JASPARGTAAACAT11
FOXK1.H14CORE.0.PS.APSAHOCOMOCOTGTTTAC11
FOXS1 (MA2118.1)JASPARTGTAAACA10.8
FOXO4 (MA0848.1)JASPARGTAAACA10.7
FOXK1 (MA0852.3)JASPARGTAAACA10.7
FOXK2.H14CORE.0.P.BPBHOCOMOCOTGTTTAC10.7
FOXL1 (MA0033.2)JASPARgTAAACA10.6
M03022_2.00CIS-BPGTAAACA10.6
M03034_2.00CIS-BPgTAAACA10.6
FOXA2 (MA0047.4)JASPARGTAAACA10.5
M08118_2.00CIS-BPGTAAACA10.5
AC0038:FOXO/FOXP:ForkheadVierstraGTAAACA10.4
Foxf1 (MA1606.2)JASPARgTAAACA10.2
FOXM1.H14CORE.0.P.BPBHOCOMOCOTGTTTaCttt10.2
FOXA1.H14CORE.0.P.BPBHOCOMOCOTGTTTACtta10.1
FOXG1.H14CORE.0.PSM.APSMAHOCOMOCOTGTTTdc10
Foxo3 (MA0157.4)JASPARGTAAACA9.9
FOXK2 (MA1103.3)JASPARGTAAACA9.8
FOXA3 (MA1683.2)JASPARGTAAACA9.8
FOXO4.H14CORE.0.SM.BSMBHOCOMOCOTGTTTAc9.8
AC0040:FOXA/FOXC:ForkheadVierstraGTAAACA9.7
FOXO1.H14CORE.0.PS.APSAHOCOMOCOTGTTTaC9.6
Foxo1 (MA0480.3)JASPARGTAAACA9.3
M00253_2.00CIS-BPATAAayA8.8
AC0093:FOXC/HOXA:ForkheadVierstraATAAA8.2
M04847_2.00CIS-BPGtAAACA8.1
M04855_2.00CIS-BPaygTAAAC8
M04849_2.00CIS-BPcGTaAACA7.9
FOXD3.H14CORE.0.P.CPCHOCOMOCOTGTTTnCtta7.7
AC0039:FOXD/FOXK:ForkheadVierstraTGTTTAC7.5
M00164_2.00Mus musculus 0.79CIS-BPTAAACA7
M01983_2.00Mus musculus 0.93CIS-BPTgTTTACata6.6
AC0107:FOXC:['Fork_head/winged_helix_factors']VierstraGTA6.3
AC0662:FOXL:['Fork_head/winged_helix_factors']VierstraacATA6.1
M01008_2.00CIS-BPCAAAC3.9
AC0152:FOXC/FOXK:ForkheadVierstraaAAcA3.8
AC0014:FOXR/LCORL:ForkheadVierstraAAa3.4
M01009_2.00Mus musculus 1CIS-BPcaT2.1

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 Forkhead families