RUNX

The factors

RUNX1 (AML1, CBFA2), RUNX2 (CBFA1) and RUNX3 (CBFA3) each carry one Runt domain near their N terminus. The domain is about 128 residues and adopts an s-type immunoglobulin fold, placing RUNX in the same structural superfamily as NF-kB, NFAT, STAT and p53 rather than with any classical helix-turn-helix family ^10404214. It binds DNA as a monomer, but not effectively on its own: the shared subunit CBFB (CBFbeta, PEBP2beta) makes no DNA contact yet is required in practice. Two crystal structures of the Runt domain-CBFB-DNA ternary complex showed how - CBFB stabilises the DNA-bound conformation of the domain's C-terminal "tail" and "wing", which together clamp the phosphate backbone across both grooves, and the extended tail contributes most of the NF-kB-like base-specific major-groove contacts ^{11257229,11276260}. Expression is largely complementary: RUNX1 in hematopoietic stem and progenitor cells and lymphocytes, RUNX2 in osteoblasts, chondrocytes and some epithelia, RUNX3 in gut epithelium, sensory neurons and T and NK cells. All three interact with CBFB and with the same corepressor and coactivator machinery, so which one is present is usually a matter of cell type rather than of DNA sequence.

The motif

The core is six bases. Binding-site selection with core-binding factor gave the consensus PyGPyGGTPy ^8445737, most compactly written TGYGGT, and its reverse complement ACCRCA is equally common in the literature. The Y at position three is the one genuinely ambiguous base, and it explains a discrepancy visible in these databases: ChIP-derived models for all three paralogues read TGTGGTT (ACCACA), whereas HT-SELEX and methyl-SELEX models for RUNX2 and RUNX3, the JASPAR RUNX2 model, and the Vierstra archetypes read TGCGGTT (ACCGCA) ^23332764. Both are correct readings of TGYGGT; in vitro selection favours the C variant, genomic occupancy the T variant. Structurally, the guanines of the GG step are read by arginine and lysine side chains from the Runt domain tail in the major groove, while the flanking T and the distorted, bent B-form geometry of the site account for the rest of the specificity ^{11257229,11276260}. Six informative positions is little, and the databases compensate in two ways. Some models extend one or two positions into the flank; others are tandem repeats of the core (AACCGCAAAAACCGCA, GTGGTTTTTGTGGTTT). RUNX enhancers do tend to carry paired sites, but the tandem models are not read off enhancers: the HOCOMOCO ones come from HT-SELEX (RUNX3.H14CORE.1.S.B) and methyl-HT-SELEX (RUNX2.H14CORE.2.M.C) on random ligands, which recovered two cores at a fixed spacing ^23332764. Every structure of the complex shows a monomeric Runt domain ^{11257229,11276260}, so whether a pair of cores is bound cooperatively is unresolved rather than excluded. No model distinguishes RUNX1, RUNX2 and RUNX3 - family cohesion is 0.86 and the three ChIP-derived HOCOMOCO consensuses are character-for-character identical. Note also that one CIS-BP RUNX1 entry was measured on the mouse protein, whose Runt domain is identical to the human one at 100% identity.

Where it acts

RUNX motifs are enhancer features more than promoter features, and are not enriched in CpG islands. In hematopoietic cells RUNX1 occupies tens of thousands of sites, and its motif is one of the commonest in the accessible chromatin of blood lineages; RUNX2 shows the same pattern in osteoblasts. The characteristic arrangement is a composite with ETS. Cooperative binding of ETS1 and core-binding factor to adjacent sites was described at the T-cell receptor enhancers ^8264651, and genome-wide work later showed that the ETS-RUNX composite is what distinguishes ETS1's T-cell-specific enhancer occupancy from its redundant, consensus-driven occupancy at housekeeping promoters ^20019798. RUNX sites also co-occur with GATA (the GATA-RUNX pairing at hematopoietic enhancers), with SPI1/PU.1 in myeloid cells, with TCF/LEF and with AP-1.

Biology and disease

Runx1-null mice fail definitive hematopoiesis and die in mid-gestation with central nervous system hemorrhage ^8565077; the Cbfb knockout reproduces this, which was the genetic demonstration that the partner subunit is required in vivo ^8929538. Runx2-null mice have no bone formation at all because osteoblasts arrest before maturation ^9182763, and heterozygous loss-of-function mutations in human RUNX2 cause cleidocranial dysplasia ^9182765. RUNX3 silencing by promoter hypermethylation and loss of expression was linked causally to gastric carcinoma ^11955451. RUNX1 is among the most frequently rearranged genes in human leukemia: t(8;21) creates RUNX1-RUNX1T1 in acute myeloid leukemia, t(12;21) creates ETV6-RUNX1 in the commonest subtype of childhood B-cell acute lymphoblastic leukemia, and inv(16) fuses CBFB to MYH11; germline RUNX1 mutations cause familial platelet disorder with predisposition to myeloid malignancy ^28179279. Most of these lesions keep an intact Runt domain and convert RUNX1 into a repressor at its normal sites ^28179279.

Reading a hit

TGYGGT is short. Expect roughly one match every few hundred base pairs per strand, so an isolated hit is close to meaningless; weight matches by clustering, by conservation, and above all by the presence of a partner site. An adjacent ETS core (GGAA/GGAT) within a few base pairs, or a nearby GATA site, turns a weak RUNX match into a plausible composite element. The hit cannot name the paralogue, and it cannot distinguish RUNX from CBFB-containing complexes generally, since CBFB has no motif of its own - a "CBFB site" in any database is a RUNX site. Decide which variant to scan with deliberately: an in vitro TGCGGT model will score differently from a ChIP-derived TGTGGT model on the same sequence, and neither is wrong. The HOCOMOCO RUNX1, RUNX2 and RUNX3 primary models are ChIP-only (grade B); because RUNX travels in composites with ETS and GATA, ChIP-derived RUNX models can absorb flanking partner information, and conversely ETS models derived from T-cell ChIP can absorb the RUNX core.