Fig. 3: Cobinding predictions in memory of colitis identifies cooperativity between AP-1 and FOX TFs.

a, Schematic for derivation of de novo motifs from footprinting data. b, Left, mean accessibility change over control for all de novo and known motifs. Right, sequence information content of de novo derived AP-1 composite motifs. c, Top, FOX/AP-1 composite motif accessibility change over control through colitis progression. Mean across n = 9 control, 4 acute, 5 chronic and 5 recovered mice. Bottom, predicted effect of motif presence on footprint scores. The x axis represents distance from motif, the y axis represents size of footprint and colour indicates predicted change of footprint score when motif is added. d, Disruption of cobinding with AP-1 for select TF families following 24 h of AP-1 inhibition. e, Cobinding of select TF families at IBD-specific AP-1 footprints in human organoids. f, Comparison of cobinding scores between mouse and human organoids. g, Schematic for quantification of in vitro binding capability. h, Example locus assayed by the in vitro binding assay. Top, seq2PRINT footprint scores from control and colitis-recovered mice. Bottom, in vitro TF binding score at the same sequence. i, In vitro binding scores for AP-1 and FOX TFs alone and in combination. The y axis represents the binding score at the AP-1 motif normalized to the FOS–JUN heterodimer alone. Error bars represent s.e.m. across all tested loci (n = 29 sequences for AP-1 alone, n = 34 sequences for FOX/AP-1). j, AlphaFold3 predicted structure for FOXP1, composite motif DNA and FOS–JUN dimer (left) or JUN alone (right). All error bars represent s.e.m. Significance values are from two-sided t-tests unless otherwise indicated. For box and whisker plots, the centre line represents median, the box upper and lower quartiles and the whiskers 1.5× IQR. Panels d and e created in BioRender; Nagaraja, S. https://biorender.com/or0ceke (2026).