Figure 5

Clustering of architectural proteins is a conserved feature of genome organization. (a) Heatmap depicting overlap enrichment between architectural proteins mapped by ChIP-seq in mESCs. Red to blue squares represent depletion (red) or enrichment (blue), determined as the log₂ (observed/expected) frequency of overlap when compared to randomized, simulated data. (b) Example genomics viewer profile (left) of a high occupancy APBSs in mESCs, bound by CTCF, TFIIIC (-220, -110, and -90), Rad21, condensin II (CAP-D3 and CAP-H2), and PRDM5, and marked by strong DHS. Hi-C interaction matrix (right) illustrates the corresponding TAD separation observed in vivo (TAD boundary defined by black arrowhead). (c) Sites combinatorially bound by CTCF and other factors (CTCF plus three or more proteins) are significantly enriched at TAD borders in mESCs. P values (*P < 0.05, ** P < 0.01, *** P < 0.001) were calculated using permutation tests. (d) Relationship between protein occupancy, defined by the presence of CTCF, Rad21, PRDM5, TFIIIC (any or all subunits -220, -110, -90) and condensin II (CAP-H2 and/or CAP-D3), and topological domain border strength in mESCs. (e) Parallel analysis of topological domain border strength in human IMR90 fibroblasts as a function of protein occupancy at CTCF binding sites. Co-binding determined by cross-comparison of ChIP-seq datasets for transcription factors and DNA binding proteins in human K562 cells. (f) Relationship between cell-type specificity of CTCF binding sites and localization to TAD borders. CTCF ubiquity determined by cross-comparison of 62 CTCF ChIP-seq datasets across 31 human cell lines. The x-axis represents CTCF sites grouped into eight bins (approximately 15,000 sites each) of increasing ubiquity ranging from cell type-specific to constitutive. For a list of human cell lines, ubiquity scores and exact number of CTCF binding sites in each bin, see Materials and methods and Additional file 8.