Cohesin-based chromatin interactions enable regulated gene expression within preexisting architectural compartments

Abstract

Chromosome conformation capture approaches have shown that interphase chromatin is partitioned into spatially segregated Mb-sized compartments and sub-Mb-sized topological domains. This compartmentalization is thought to facilitate the matching of genes and regulatory elements, but its precise function and mechanistic basis remain unknown. Cohesin controls chromosome topology to enable DNA repair and chromosome segregation in cycling cells. In addition, cohesin associates with active enhancers and promoters and with CTCF to form long-range interactions important for gene regulation. Although these findings suggest an important role for cohesin in genome organization, this role has not been assessed on a global scale. Unexpectedly, we find that architectural compartments are maintained in noncycling mouse thymocytes after genetic depletion of cohesin in vivo. Cohesin was, however, required for specific long-range interactions within compartments where cohesin-regulated genes reside. Cohesin depletion diminished interactions between cohesin-bound sites, whereas alternative interactions between chromatin features associated with transcriptional activation and repression became more prominent, with corresponding changes in gene expression. Our findings indicate that cohesin-mediated long-range interactions facilitate discrete gene expression states within preexisting chromosomal compartments.

Figure 1.

Chromosomal compartments are resilient to the depletion of the cohesin subunit RAD21 from noncycling thymocytes in vivo. (A) Eigenvector analysis of chromosomal organization. (B) Compartment tracks (top) and interaction matrices (bottom) for chr13 at 140-kb resolution in control and cohesin-deficient thymocytes. Of 17,548 regions evaluated, only two showed a consistent compartment change from A to B and two from B to A (change in eigenvector value >0.03): chr18: 3640001 and chr18: 33600001. (C) Hi-C compartment data are highly correlated for control and cohesin-deficient thymocytes.

Figure 2.

Cohesin binding predicts perturbed long-range interactions in cohesin-deficient thymocytes. (A) Schematic of approach for the mapping of differentially interacting regions. (B) Circos plot (Krzywinski et al. 2009) illustrating the chromosomal position of differential interactions in the context of chromosomal compartments. The HOMER software suite was used to determine significant interactions between 100-kb genomic regions in either control or cohesin-deficient thymocytes (FDR = 0.1; replicates pooled). Of 10,917 interactions that were significantly altered in the pooled samples, 1476 interactions changed in replicate 1 and 5004 in replicate 2 (P < 0.05; the remaining interactions were only seen in the pooled data). Of 502 differential interactions that were shared between replicates, 278 were decreased (blue) and 224 were increased (red). All differential interactions were intra-chromosomal. Compartment assignment is indicated at the base of the interactions: black for open compartments and gray for closed compartments. (C) Differential interactions in cohesin-deficient cells are largely contained within preexisting chromosomal compartments. (A) Differential interactions entirely contained within the same A compartment. (B) Differential interactions entirely contained within the same B compartment (the frequency of intra-B interactions was <2% among down-regulated interactions). (A-B) Differential interactions bridging A and B compartments. (Other) interacting regions are either unassigned (i.e., inconsistent between replicates) or bridge two distinct A or B compartments. Down-regulated interactions (top) and up-regulated interactions (bottom) are shown separately. (D) Features enriched in 100-kb regions that show differential interactions in cohesin-deficient thymocytes. Differential interactions between control and cohesin-deficient samples involved 946 unique 100-kb regions (510 involved in decreased interactions, 427 in increased interactions, and 9 involved in both) that participated in 502 unique differential interactions (278 that were decreased and 224 that were increased) and that were shared between replicates (P < 0.05). We tested whether differentially interacting 100-kb regions were enriched for the presence of Rad21, CTCF, NIPBL, MED1, H3K4me3, and RNAP2 binding events. See Supplemental Figure S1C,D for validation of the MED1 and NIPBL antibodies and ChIP, and Supplemental Figure S1E for an analysis of the relationship between RAD21, CTCF, and NIPBL ChIP-seq peaks. Differentially interacting regions were significantly enriched for the binding of the cohesin subunit RAD21, both with and without CTCF (cohesin-non-CTCF; CNC), and of the cohesin-associated factors CTCF and NIPBL, and features of transcriptional activity including MED1, H3K4me3, and RNA Pol II. Differential interactions were further classified into decreased and increased interactions. (E) Strength distribution of cohesin-dependent interactions. Using the number of Hi-C reads as an indicator of the strength of interactions, differential interactions that are decreased in cohesin-depleted thymocytes are similar in strength to unchanged interactions in control cells, whereas increased interactions tend to be weak before cohesin depletion (Mann-Whitney U-test P < 10⁻¹⁵). Outliers are not depicted.

Figure 3.

Gene expression in cohesin-deficient thymocytes. (A) Volcano plot of RNA-seq data. Of the 17,849 genes assayed, 1153 are significantly differentially expressed in cohesin-deficient thymocytes (FDR = 0.05), 450 of which are up-regulated (orange) and 703 down-regulated (green). The red dot represents Rad21. Eight genes above the −log10(P-value) = 50 threshold (all up-regulated) were omitted from the plot. (B) Validation of RNA-seq data for 15 transcripts over a wide range of expression levels. Shown is fold-change measured by qRT-PCR (x-axis, two independent biological replicates) against the fold-change measured by RNA-seq (y-axis) and the Pearson's correlation coefficient (r) for two independent RNA-seq experiments. (C) Genes affected by cohesin depletion are associated with open compartments. Approximately 98% of differentially expressed genes that could be assigned to compartments (1121 of 1142 autosomal and X-linked genes) reside in open (A) compartments, compared with 91.2% (16,255 of 17,819) of genes included in our analysis. Genes spanning more than one compartment were assigned to A when they overlapped at least partially with open compartments; genes overlapping compartments that could not be clearly defined as A or B because they were inconsistent between replicates were called unassigned. Only 21 deregulated genes were found outside open compartments. Of these, 11 were located in unassigned compartments and just 10 of 1142 deregulated genes were located in closed (B) compartments. This corresponds to a highly significant depletion of deregulated genes in B compartments (P < 10⁻¹⁵, odds ratio = 0.12). (D) Cohesin-regulated genes are bound by cohesin and associated factors. Associations are shown for ChIP-seq peaks for RAD21, NIPBL, and CTCF (Shih et al. 2012) within 2.5 kb of transcription start sites (TSS) and within 10 kb of canonical gene bodies. (E) Gene Ontology (GO) analysis of genes that are differentially expressed in cohesin-deficient thymocytes. Representative GO biological process terms with adjusted P < 10⁻⁸ are shown (the complete list can be found in Supplemental Table 2).

Figure 4.

Features and predictors of cohesin-dependent gene expression. (A) The T cell receptor locus Tcra and 3′ flanking region of chr14 are shown alongside Hi-C heat maps at 140-kb resolution for control (top) and cohesin-deficient thymocytes (bottom), epigenomic features, and genes that are up- (red) or down-regulated (blue, no examples in this region). Differentially interacting (DI) regions as determined by HOMER analysis of Hi-C data are shown in red (up-regulated) or blue (down-regulated). Dashed gray lines demarcate increased interactions with the region 3′ of Tcra. (B) HOMER-identified increased and decreased interactions around the Cd3 gene cluster on chr9. (Top) increased interactions marked as (i) chr9:44300000–44400000 and (ii) chr9:44800000–44900000 contain the Bcl9 and Mpzl2 genes that are up-regulated in cohesin-deficient thymocytes. (Bottom) Gene expression and 3C analysis of region (ii). Hind III fragments containing CTCF and RAD21 binding sites are shaded gray. (Left inset) Primary 3C data (mean ± SD, n = 3). (Right inset) RT-PCR validation of Mpzl2 expression (mean ± SD, n = 3). (C) Multinomial logistic regression model integrating gene expression, Hi-C, and ChIP-seq data to predict up-regulated, down-regulated, and unchanged genes. We tested each gene for the following features: (1) the presence or absence of ChIP-seq peaks near the promoter (TSS ± 2.5 kb); (2) location within 100-kb differentially interacting regions (divided into interactions that were stronger, DI region [Up], or weaker, DI region [Down], in cohesin-deficient thymocytes); (3) the presence of the H3K4me3 histone modification; (4) the binding of RAD21, CTCF, NIPBL and Mediator (MED1), RNA Pol II (RNAP2), paused RNA Pol II at the promoter (Hendrix et al. 2008); (5) the presence of a promoter CpG island (CGI); and (6) gene length. Error bars represent 95% confidence intervals. Variables are ranked by coefficient significance from left to right. Of 17,849 genes assayed by RNA-seq, 1461 resided in DI regions. Of 450 down-regulated genes, 83 resided in DI regions, which represents a strong enrichment (P-value < 10⁻¹¹, odds ratio 2.63). Up-regulated genes were only slightly enriched in DI regions (P < 0.05, odds ratio 1.27).

Figure 5.

The chromatin landscape of cohesin-deficient thymocytes is characterized by the loss of cohesin-based interactions and the detection of alternative interactions. (A) Outline of approach used to score interactions between specific chromatin features located within architectural compartments as assigned by eigenvector analysis of chromosomal organization (SIMA) (Lin et al. 2012). Open compartments (1, 2, …, n) are indicated by gray boxes. Orange, green, and black symbols (features 1, 2, etc.) represent chromatin features, and purple lines indicate “featureless” control sites. Hi-C interactions between features are indicated by dashed lines. For each compartment, we counted Hi-C reads connecting features of the same type (homotypic interactions, illustrated for feature 1), and Hi-C reads connecting different features (heterotypic interactions, illustrated for features 2 and 3), assigning Hi-C reads that mapped within 10 kb of each feature (Lin et al. 2012). We compared interactions in control and cohesin-deficient cells for each feature and pair of features within each open compartment and combined these scores as a measure for the cohesin dependence of long-range interactions between specific features. (B) Impact of cohesin deficiency on homotypic cohesin-based and alternative interactions. SIMA was used to determine the enrichment of Hi-C reads in interactions connecting “like” features in control and cohesin-deficient thymocytes. The difference between normalized enrichment ratios in each condition was assessed by the Wilcoxon signed-rank test (see Methods). (C) Impact of cohesin deficiency on all pairwise feature-based interactions. SIMA results for homotypic interactions are shown together with those associated with interactions connecting different features (heterotypic interactions). Refer to B for details. Homotypic RAD21-RAD21 interactions are the most decreased, followed by CTCF-RAD21 interactions. Interactions between features associated with active transcription are strongly increased in cohesin-deficient thymocytes, as are interactions involving the repressive histone modification H3K27me3 (Zhang et al. 2012) with marks of active transcription. (D) Cytoscape representation of SIMA results in C. Edge color and width correspond to the Wilcoxon signed-rank test effect size and significance, respectively. (E) Length scale of lost and gained cohesin-dependent feature-based interactions. Boxplots are based on SIMA comparisons stratified into three classes according to compartment size (≤1 Mb, 1–3 Mb, >3 Mb). Effect sizes for decreased interactions (RAD21-RAD21, CTCF-RAD21, CTCF-CTCF) and increased interactions (remainder) were grouped and are indicated separately. The effect of reduced cohesin-based interactions is most pronounced within smaller compartments and decreases when larger compartments are considered (Pearson's correlation coefficient r = −0.73, P < 0.05), whereas alternative interactions increased with compartment size (r = 0.7, P < 0.001). Outliers are not depicted.

Figure 6.

Cohesin depletion compresses the dynamic range of gene expression. (A) Genes were stratified into 10 equally sized log intervals from low (0–1) to high (>9) based on the average gene expression of control and cohesin-deficient thymocytes. Boxplots indicate the distribution of gene expression fold changes in cohesin-deficient thymocytes. The number of genes in each bin is indicated (bins 1 and 2 are empty). Note that lowly expressed genes are frequently up-regulated, whereas highly transcribed genes tend to be down-regulated. Note that this pattern does not result from ascertainment bias in which lowly expressed genes can only be up-regulated and vice versa, because we stratified genes according to the mean of their expression in control and cohesin-depleted cells. Therefore, the direction of regulation is the inverse when control cells are compared to cohesin-deficient cells (not shown). P-values are based on one-sample Wilcoxon signed-rank tests and indicate significant difference from zero (no change). Genes with zero mean expression in both cohesin-deficient and control thymocytes are excluded and outliers are not depicted. (B) Bar plot indicating the proportion of up- and down-regulated genes in each gene expression interval (see A). The proportion of up-regulated genes is anti-correlated with interval rank (Pearson's correlation coefficient r = −0.97, P < 10⁻⁴).