Fig. 5

Aggregated Hi-C data highlight a role of insulator-based long-range contacts in micro-domains. a Aggregation of Hi-C data [9, 50] highlighting genome-wide long-range interactions between all Beaf32 sites (“Type-1 insulators”) with the distinct insulators bound by GAF or dCTCF (“Type-2 insulators”) depending on their co-localization with CP190 or not (upper and lower panels, respectively). LRIs 1, 2, and 3 represent A/B compartments (LRIs-1: long-range interactions detected between two A or two B domains), TADs (LRIs-2: long-range interactions defining TAD units in the Hi-C matrices), and specific loops (LRIs-3: long-range interactions between two defines sites (e.g., Beaf32, GAF, or dCTCF)), respectively, quantified for all TADs as normalized Hi-C reads (see “Methods”). Similar results were obtained when estimating LRIs from other Hi-C data [16, 20, 63] (Additional file 4: Fig. S4). b Scheme representing an alternative mode of 3D spreading of H3K27me3 into micro-domains; specific insulator-based long-range interactions between insulators may put them in physical proximity thereby favoring spreading from a repressive H3K27me3 domain to a micro-domain (see text). c Eigen value of micro-domains compared to control regions corresponding to active A compartments (e.g., domains harboring active genes and enriched in H3K27 acecylated histone marks) or inactive/repressive B compartments (enriched in H3K27me3 repressive histone marks) chromatin domains (see “Methods”). d Gene set enrichment analysis (GSEA) testing the influence of insulator-based variations in LRIs on formation of H3K27me3 micro-domains. ΔLRIs-1/2/3 were measured as net variations between Hi-C data in Beaf32-depleted cells compared to control cells (see “Methods”) [20, 21]. The normalized differential ΔLRI scores were estimated for all loci defined by a couple of bins corresponding to one Beaf32 site interacting with any distant gene (> 5 kb) harboring a H3K27me3 micro-domain or not. Genes were classified depending on differential ΔLRI-1/2/3 levels (left, middle and right graphs, respectively) to test which variations best predict the association of genes with a micro-domain (log p values were obtained using a corrected Fisher exact test)