AP-2 (TFAP2)

The factors

Five human genes encode this family: TFAP2A, TFAP2B, TFAP2C, TFAP2D and TFAP2E, still widely called AP-2 alpha, beta, gamma, delta and epsilon. Each protein has an N-terminal proline- and glutamine-rich activation domain ^{3063603,2010091}, a basic region, and a C-terminal helix-span-helix (HSH) dimerization domain: two amphipathic helices separated by a long span. The HSH domain drives dimerization on its own and can be transplanted onto a foreign basic region, but it cannot bind DNA without an adjacent basic region ^1998122. AP-2 is a dimer in solution as well as on DNA ^2010091, and all of the paralogue pairs can heterodimerize; in neural crest development TFAP2A switches partners from TFAP2C to TFAP2B as specification proceeds, and the two heterodimers open different sets of elements ^31848212. Expression is restricted rather than housekeeping: TFAP2A and TFAP2B in neural crest, surface ectoderm, facial mesenchyme and kidney ^1989904; TFAP2C in trophoblast, germ cells and luminal breast epithelium; TFAP2D mainly in retina and midbrain; TFAP2E in olfactory epithelium. TFAP2D is the outlier, with a divergent basic region and weaker transactivation.

The motif

The functional element is a pseudo-palindrome of two triplets, GCC and GGC, with a three-base spacer: GCC(N3)GGC ^37409559. SELEX with AP-2 alpha also recovered GCC(N4)GGC and GCC(N3/4)GGG sites ^10373594, and the older consensus GCCYGCRGGC is an N4 form. The database 9-mers (GCCTGAGGC, GCCTCAGGC) are N3; the longer models are not that core plus flanks, since the extra C and G sit inside the triplets: GCCCYAGGGC is GCC(N4)GGC, and GCCCTBAGGGC reads as GCC(N4)GGG. The data consensus, GCCTCAGGGCA, is a longer form (GCC(N4)GGC, or GCC(N3)GGG), so it differs from the literature's GCC(N3)GGC in spacing, not only in flanks. The crystal structure of the TFAP2A DBD-HSH tandem domain on DNA explains the geometry: the two flat amphipathic HSH helices stack against each other through hydrophobic contacts, and a stabilized loop from each DBD inserts into one of two neighboring major grooves ^37409559. Because the two loops are a fixed distance apart, spacer length, not just spacer sequence, determines binding: a three-base gap is strongly preferred: TFAP2A bound a four-base spacer about 18-fold more weakly and two- or five-base spacers barely at all ^37409559. This makes the informative positions the GCC and GGC triplets, with the central three bases nearly free (commonly written YAG or TCA/TGA). The five paralogues are not separable from sequence. Every model in this family's data carries the same GCC and GGC/GGG triplets (7.5-14.5 bits); they differ in spacer length, but TFAP2A, TFAP2B and TFAP2C each have models of more than one spacing.

Where it acts

AP-2 sites are GC-rich, so they are enriched in GC-rich and CpG-island promoters, where they can be hard to distinguish from SP/KLF GC-boxes on loose models. The functionally important sites, however, are largely distal: TFAP2A occupies thousands of enhancers at the neural plate border and in specified neural crest, acting as a pioneer factor that remodels chromatin ahead of the lineage program ^31848212. In melanocytes TFAP2 paralogues bind in parallel with MITF at pigmentation genes ^28249010. In luminal breast cancer cells TFAP2C co-occupies enhancers with estrogen receptor and FOXA1 and regulates ESR1 itself ^24469049. Co-occurring motifs therefore depend on the tissue: SOX and PAX motifs in neural crest, MITF E-boxes in melanocytes, ERE and forkhead sites in breast. AP-2 sites are common enough in GC-rich sequence that a genome-wide scan returns very large numbers of matches; the constraint on real occupancy is TFAP2 expression, which is narrow.

Biology and disease

Tfap2a-null mice fail cranial neural tube closure and have severe craniofacial defects, matching AP-2's expression in ectoderm and migrating cranial neural crest ^{8622765,1989904}. In humans, heterozygous TFAP2A mutations, mostly missense changes in the basic region and HSH domain, cause branchio-oculo-facial syndrome: cervical and infra-auricular skin defects, ocular anomalies, cleft lip or palate, and sometimes inner ear malformation and deafness ^18423521. Mutations in TFAP2B cause Char syndrome, a dominant condition combining patent ductus arteriosus, a distinctive face and hand anomalies, with a genotype-phenotype relationship that tracks with how much DNA binding is lost ^10802654. TFAP2C is amplified and overexpressed in a subset of breast cancers, where it governs the luminal epithelial phenotype and hormone response and higher expression predicts worse outcome on anti-hormonal therapy ^24469049. TFAP2A and TFAP2C are also required in trophoblast and germ cell programs. A structural survey of disease mutations concluded that reduced or abolished DNA binding is the primary mechanism in TFAP2-associated disease ^37409559.

Reading a hit

A GCC(N3)GGC match tells you that a TFAP2 dimer could bind, and nothing about which one. The five paralogues share one matrix; assign a member from expression, and remember that the biologically active species is often a heterodimer, so the question may have no single answer. Check the spacer before believing a match: two bases is a poor site, while four binds more weakly than three in vitro but was recovered by SELEX and is common in the models here, so it is weaker evidence, not a non-site ^{37409559,10373594}. Permissive scanners report all of these spacings. Because the site is GC-rich, the main false-positive risk is confusion with other GC-binding families. A loose AP-2 model will score on SP1/KLF GC-boxes (GGGGCGGGG) and on CpG-island sequence generally; conversely, GC-box models score on AP-2 sites. Require the discontinuous GCC...GGC arrangement rather than a continuous GC run before calling a hit AP-2. The evidence base here is comparatively clean: HOCOMOCO models for AP2A and AP2C are grade A with combined ChIP-seq and in vitro selection (PSM), and the AP2B and AP2E models rest on in vitro selection alone (SM), so the family's specificity does not depend on ChIP data. The one provenance caveat is the CIS-BP TFAP2D entry, which is transferred from a mouse protein at 80% DNA-binding-domain identity ^25215497.