MEF2

The factors

MEF2A, MEF2B, MEF2C and MEF2D each carry a MADS box - a 56-58 residue domain shared with SRF and with the plant and yeast MADS proteins ^7744019 - immediately followed by a MEF2-specific extension of about 30 residues. The two regions together form the DNA-binding and dimerisation surface; MEF2 proteins bind only as homodimers or as heterodimers among themselves ^1748287. The crystal structure of the MEF2A core on DNA at 1.5 A resolution and an NMR structure of the same complex showed that MEF2 lacks the N-terminal extension SRF uses, and that the MEF domain C-terminal to the MADS box adopts a conformation substantially different from the equivalent region in SRF and yeast MCM1 - the structural basis for a different sequence preference and a different degree of DNA bending within one fold ^{10715212,10835359}. All four genes are broadly expressed with different emphases: MEF2A and MEF2D are near-ubiquitous, MEF2C is prominent in brain, cardiac and skeletal muscle and in B lymphocytes, MEF2B in germinal-centre B cells ^23974956. Activity is regulated post-translationally rather than by abundance: class IIa histone deacetylases (HDAC4, HDAC5, HDAC7, HDAC9) bind the MADS/MEF2 domain and repress it until calcium-CaMK signalling drives their nuclear export, and p38 MAPK, ERK5, sumoylation and the corepressor CABIN1 tune the same surface ^{17959722,23974956}.

The motif

The consensus came with the cloning of the three human SRF-related proteins, later named MEF2, which recognize CTA(A/T)4TAG, distinct from the SRF site CC(A/T)6GG ^1748287. Site selection with anti-MEF2A antibodies and skeletal muscle, heart and brain extracts later returned the same core, 5'-CT(A/T)(A/T)AAATAG-3' ^7559475; it is now written CTA(A/T)4TAG or YTA(A/T)4TAR, a 10-bp near-palindrome. Each MADS monomer reads one half, contacting the bases through an antiparallel coiled coil of two alpha-helices laid in the widened major groove, with basic residues gripping the flanking phosphates ^{10715212,10835359}. Because the centre is a run of A and T, essentially all the sequence information lives in the outer CTA and TAG - which is why an A/T-rich stretch with the wrong flanks scores poorly, and why MEF2 sites are easy to over-call in AT-rich genomes. That same selection experiment found that MEF2 complexes from brain extracts imposed additional constraints outside the decamer, a tissue effect that has never been resolved into a distinct PWM ^7559475. The four paralogues are not separable: in these databases the JASPAR and CIS-BP models for MEF2A, MEF2B, MEF2C and MEF2D are nearly identical (family cohesion 0.86), and no in vitro study has reported a usable difference ^23332764. One real complication is visible in the data. Alongside the canonical CTATTTTTAG models sits a second flavour - CCAAATATGG, CCNAATTTGGAA and similar - which comes almost entirely from HT-SELEX and methyl-HT-SELEX experiments and which is, read literally, a CArG box: CC plus an A/T core plus GG. These entries have lower label agreement (0.6-0.7) and their nearest non-MEF2 neighbours are MADS-box motifs from SRF. Whether this is a genuine secondary MEF2 specificity or an artefact of selection is unsettled; either way, scanning with those models partly scans for SRF.

Where it acts

MEF2 motifs are enhancer elements, not promoter elements, and are depleted from CpG islands - unsurprising for an A/T-rich site. In differentiating muscle they mark the same enhancers as the myogenic bHLH factors MYOD1 and MYF5, and MEF2-E-box pairs are the canonical muscle enhancer architecture ^17959722; MEF2 does not have a named composite element of its own but the MEF2-plus-E-box arrangement is close to one. In neurons MEF2 sites mark activity-regulated enhancers; in cardiomyocytes they co-occur with GATA4, NKX2-5, TBX20 and SRF sites. Paralogue occupancy is largely redundant where two MEF2 proteins are co-expressed, and which paralogue compensates at a given site depends on the local chromatin environment ^31722213.

Biology and disease

MEF2 factors are required for differentiation rather than for lineage specification. Mef2c-null mouse embryos fail cardiac morphogenesis, with a single ventricular chamber and absent right ventricle, and die around E9.5 ^9162005. In neurons, activity-dependent MEF2 activation restricts excitatory synapse number, so MEF2 acts in synapse elimination rather than formation ^16484497. In humans, MEF2C haploinsufficiency - by 5q14.3 microdeletion or point mutation - causes severe intellectual disability with stereotypic movements, epilepsy and absent speech ^19592390. MEF2B is somatically mutated in roughly 11% of diffuse large B-cell lymphomas and 12% of follicular lymphomas; the mutations increase MEF2B activity, either by disrupting its interaction with the corepressor CABIN1 or by escaping inhibitory phosphorylation and sumoylation, and deregulate the proto-oncogene BCL6 ^23974956. MEF2D fusions define a subtype of B-cell acute lymphoblastic leukemia ^27824051, and a 21-bp MEF2A deletion was reported to cause an autosomal dominant form of coronary artery disease ^14645853, a link not supported by later resequencing: in about 300 patients with premature disease no causative mutations were found, the same deletion appeared in unaffected controls, and it did not cosegregate with disease ^15841183.

Reading a hit

The main risk here is the base composition of the genome, not the paralogue problem. An A/T-rich decamer is common in intergenic and intronic DNA, so scan with a model that actually constrains the CTA and TAG flanks and check the flanks in the match itself; a hit on a poly-A or poly-T stretch with soft flanks is almost certainly noise. A hit identifies MEF2, not MEF2A versus MEF2B, MEF2C or MEF2D - there is no reported sequence-level discrimination, and expression plus chromatin decide the occupant. Be explicit about which model you used, because the collections contain two incompatible flavours: canonical CTA(A/T)4TAG models, and CArG-like CCAAATATGG models from in vitro selection whose nearest neighbours are SRF motifs. If a putative MEF2 hit came from the second kind, consider SRF as an equally good explanation. The ChIP-derived MEF2A and MEF2C models here are grade B and reflect occupancy in specific cell types; because MEF2 sits beside E-boxes at muscle enhancers, E-box information can leak into ChIP-derived MEF2 models and vice versa.