Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 171 (2), 305-320.e24

Cohesin Loss Eliminates All Loop Domains


Cohesin Loss Eliminates All Loop Domains

Suhas S P Rao et al. Cell.


The human genome folds to create thousands of intervals, called "contact domains," that exhibit enhanced contact frequency within themselves. "Loop domains" form because of tethering between two loci-almost always bound by CTCF and cohesin-lying on the same chromosome. "Compartment domains" form when genomic intervals with similar histone marks co-segregate. Here, we explore the effects of degrading cohesin. All loop domains are eliminated, but neither compartment domains nor histone marks are affected. Loss of loop domains does not lead to widespread ectopic gene activation but does affect a significant minority of active genes. In particular, cohesin loss causes superenhancers to co-localize, forming hundreds of links within and across chromosomes and affecting the regulation of nearby genes. We then restore cohesin and monitor the re-formation of each loop. Although re-formation rates vary greatly, many megabase-sized loops recovered in under an hour, consistent with a model where loop extrusion is rapid.

Keywords: 4D Nucleome; CTCF; Hi-C; chromatin loops; cohesion; gene regulation; genome architecture; loop extrusion; nuclear compartments; superenhancers.


Figure 1
Figure 1. Tagging of endogenous RAD21 with an auxin-inducible degron allows for rapid, near complete cohesin loss
(A) In HCT-116-RAD21-mAC cells, both RAD21 alleles are tagged with auxin-inducible degrons and an mClover reporter, and the OsTIR1 gene is integrated at the AAVS1 locus. Auxin treatment leads to proteasomal degradation of RAD21. (B) Live cell imaging after Hoechst 33342 staining to label nuclei. Nuclear mClover fluorescence corresponding to tagged RAD21 was lost after 1 hour of auxin treatment. (See Fig. S1.) (C) SMC1 and CTCF ChIP-Seq signal with and without auxin treatment. (D) RAD21, SMC1 and CTCF ChIP-Seq signal (left, middle, right) across all peaks called for each of the proteins in untreated RAD21-mAC cells. (Top) Average enrichments for each protein. After RAD21 degradation, the cohesin complex no longer binds to chromatin. CTCF binding is unaffected.
Figure 2
Figure 2. Cohesin degradation eliminates loop domains
(A) Contact matrices show that loop domains in untreated RAD21-mAC cells (top) disappear after auxin treatment (bottom). Three representative loci are shown (at 10kb resolution): chr8:133.8–134.6Mb (left), chr4:40.8–42.1Mb (middle) and chr1:91.9–95.8Mb (right). (B) Aggregate peak analysis (APA) was used to measure the aggregate strength of the links associated with all loop domains in low-resolution Hi-C maps generated across a time course of auxin treatment and withdrawal. (Top) APA scores; values greater than 1 indicate the presence of loops. (Bottom) APA plots; loop strength is indicated by the extent of focal enrichment at the center of the plot (See Fig S2B). (C) Individual loop reformation curves for each of 1,988 loop domains (blue lines); the number of contacts in the untreated map corresponds to a value of 1, and the number of contacts in the auxin-treated map corresponds to 0. We highlight the media (black), the 5th percentile (red) and the 95th percentile (green) in terms of speed of recovery, see Methods. Error bars indicate 25th and 75th percentile within each subset. (D) Enrichment of epigenetic features within a loop domain vs. speed of recovery. Enrichment is with respect to all intervals spanned by loop domains. (E) Regions containing fast loop domains (1st row: chr18:67.6–68.4Mb; 2nd row: chr14:68.2–69.5Mb) and slow loop domains (3rd row: chr5:95.5–96.15Mb; 4th row: chr12:91.15–91.95Mb) are shown, along with ChIP-Seq tracks (from auxin-treated cells) for NIPBL, H3K4me1, H3K4me3, and H3K27Ac. For fast loop domains, reformation is apparent by 20–40 minutes after auxin withdrawal, whereas for slow loop domains, reformation is not seen until 3 hours after auxin withdrawal. An interactive version of this figure is available at:
Figure 3
Figure 3. Genome compartmentalization is strengthened after cohesin degradation
(A) Contact matrices of chromosome 8 at 500kb resolution. The plaid pattern in the Hi-C map, indicating compartmentalization, is preserved after auxin treatment. (B) Strength of contact domains called in untreated cells versus random intervals measured using the corner score (see Methods) in untreated (top) and treated cells (middle). Contact domain strength is reduced, but does not disappear. The remaining signal comes from compartment domains (bottom). The signal in treated maps from contact domains where both boundaries are contained completely inside a compartment interval (‘other domains’) is not enriched vs. random pixels. (C,D) Examples (C: chr21:32.4–39Mb and D: chr1:167–177Mb) showing that the loss of cohesin-associated loops after auxin treatment results in increased fine-scale compartmentalization. Top: Sliding correlation scores; valleys imply strong differences in long-range contact pattern observed at a locus as compared to neighboring loci, indicating a change in compartment (see Methods). Middle: Observed contact matrices. Bottom: Pearson’s correlation maps for the local region shown (see Methods). Deeper valleys in the sliding correlation score and increased plaid patterning in the observed and Pearson’s correlation maps indicate stronger fine-scale compartment interactions after auxin treatment. Blowouts: loss of a loop domain results in strengthening of a compartment boundary spanned by the loop. Blown-out regions are indicated on zoomed out maps for both the observed (black upper triangle) and Pearson’s correlation maps (white rectangle). Observed and Pearson’s correlation maps are both shown at 25kb resolution for the zoomed out matrices and 10kb and 25kb resolution respectively for the blown-out matrices. (E) Sliding correlation scores before and after auxin treatment for compartment boundaries which either coincide with loop domain anchors (left) or are located in the interior of a loop domain (right). (F) Sliding correlation scores before and after auxin treatment for H3K27ac boundaries in untreated cells which either coincide with loop domain anchors (left) or are located in the interior of a loop domain (right). H3K27Ac modification patterns are unchanged after auxin treatment (top and middle). Interactive figure:
Figure 4
Figure 4. Cohesin loss causes superenhancers to co-localize, forming hundreds of links within and across chromosomes
(A) A network of intra- and interchromosomal cohesin-independent links between superenhancers on chr6, chr4, and chr2. H3K27 acetylation does not change with auxin treatment, but cohesin-independent links are significantly strengthened upon treatment. Intrachromosomal matrices are shown at 25kb (on-diagonal) and 50kb (off-diagonal) resolutions; interchromosomal matrices are shown at 100kb resolution. Maximum color intensities are 28 reads for the offdiagonal intrachromosomal matrices and 20 reads for the interchromosomal matrices. (B) Length distribution of cohesin-associated loops (green) versus cohesin-independent loops (blue). (C) CTCF binding patterns at cohesin-associated (top) versus cohesin-independent loop anchors (bottom). (D) Percent of cohesin-independent loop anchors bound versus fold enrichment for 36 DNA-binding proteins and histone modifications. (E) APA for intrachromosomal (blue) and interchromosomal (red) cohesin-independent links across a time course of auxin treatment and withdrawal (top: APA scores; bottom: APA plots). Interactive figure:
Figure 5
Figure 5. In the absence of cohesin, a clique spanning more than 20 superenhancers forms pairwise links and higher-order hubs
(A) The interactions between 20 cohesin-independent loop anchors spread across 9 chromosomes are shown before (lower triangle) and after (upper triangle) auxin treatment. Each matrix shows a 2 Mb by 2 Mb matrix centered on the respective anchors. Intrachromosomal interactions are shown at 25kb resolution; interchromosomal interactions are shown at 100kb resolution. The anchors are strongly enriched for H3K27 acetylation both before and after auxin treatment. (ChIP-Seq data is shown at 25kb resolution.) Cohesin loss causes the anchors to form a clique, with focal interactions seen between nearly all pairs of loop anchors, regardless of whether they lie on the same chromosome. (B) In addition to pairwise contacts, in situ Hi-C generates concatemers spanning three or more fragments. There are millions of triples (chimeric reads which align to three loci) and quadruples (chimeric reads which align to four loci) in both our untreated and auxin-treated in situ Hi-C data sets for RAD21-mAC cells. The numbers in parentheses indicate the number of n-mer contacts observed in the untreated (left) and auxin-treated (right) data. (C) 3D tensor showing collisions between three loci on chromosome 6 at 1Mb resolution (see Methods). (D) (Left) 3D aggregate peak analysis (APA) using the untreated in situ Hi-C data for all 131 intrachromosomal trios of cohesin-independent loop anchors, chosen so that each anchor in a trio lies on the same chromosome as the other two anchors, but no two anchors in a trio lie within 10Mb of one another. To create a 3D APA cube, we excise a 3.9 × 3.9 × 3.9Mb subtensor centered on each trio, and superimpose the results. The cube is shown at 300kb resolution (i.e., each voxel corresponds to all collisions between three loci, each 300kb in length). The subtensors are oriented such that the locus closest to the p-terminus of a chromosome is always located on the z-axis, the one closest to the q-terminus is located on the y-axis, and the locus in between is located on the x-axis. The number of collisions in a voxel is indicated by its color; the histogram above the color scale shows the number of voxels of each color. No voxel contains more than 5 collisions, and the center voxel – reflecting all collisions between three cohesin-independent loop anchors – contains no collisions at all. (Right) Top Row: The central cross-section in z is shown, flanked by the two adjacent cross-sections. Middle Row: The central cross-section in y, flanked by the adjacent cross sections. Bottom Row: The central cross section in x, flanked by the adjacent cross sections. There is no enrichment at the center of the 3D APA cube. (E) The preceding analysis is repeated using the auxin-treated data. Now, the center voxel contains 11 collisions, whereas no other voxel contains more than 5 collisions. These findings indicate that, in the absence of cohesin, cohesin-independent loop anchors tend to co-localize to form hubs containing three or more anchors. (F) Histogram of number of voxels vs. number of collisions for the two 3D-APA cubes shown in 5D and 5E, as well as for 52 control 3D-APA cubes obtained by shifting one or more of the loci in each of the above trios by 3.9Mb. With the exception of the central voxel in the auxin-treated 3D APA cube, which contains 11 collisions, no voxel contains more than 8 collisions. This indicates that the observation of 11 collisions purely by chance is exceedingly unlikely. (G) Under normal circumstances, loop extrusion facilitates short-range contacts between superenhancers and neighboring loci. Upon cohesin loss, superenhancers begin to co-localize, even when located on different chromosomes, and thereby form a subcompartment. Interactive figure:
Figure 6
Figure 6. Molecular dynamics simulations combining extrusion and compartmentalization can recapitulate Hi-C experimental results
(A) We use loop extrusion and compartmentalization to simulate a 2.1 Mb region on chromosome 3 in RAD21-mAC cells before (left) and after (right) auxin treatment. CTCF and SMC1 ChIP-Seq signals are normalized and converted into binding probabilities for the simulated extrusion complex (first and second rows). Each peak is assigned a forward (green) or reverse (red) orientation based on the corresponding CTCF motif. ChIP-Seq data for 9 histone modifications were used to classify loci into two compartments (red and blue, fifth row). Histone modification data for H3K36me3 and H3K4me1 is shown, illustrating the correspondence between the classification tracks and the underlying ChIP-Seq signals (third and fourth rows). The simulations yield an ensemble of polymer configurations. We show contact maps from the simulated ensemble (top) and from the corresponding Hi-C experiments (bottom). (B) Examples of globules from simulations of compartmentalization with extrusion (left) and without (right). The globule without extrusion shows stronger segregation of compartment types. Interactive figure:
Figure 7
Figure 7. Cohesin degradation results in strong down-regulation of genes near superenhancers but does not result in widespread ectopic gene activation
(A) Scatter plot of gene-wide PRO-Seq counts in RAD21-mAC cells before (x-axis) and after (y-axis) treatment. (B) Genes that are expressed in untreated cells rarely undergo substantial changes in expression level after cohesin loss. (C) An example of a strongly down-regulated gene near a superenhancer. In untreated cells, a series of cohesin-associated loops form between the IER5L promoter and nearby superenhancers. Upon auxin treatment, these loops are lost and IER5L expression is 2.6-fold down-regulated. (D) Cumulative probability distributions of distances to the nearest superenhancer for 1.75-fold down-regulated genes after auxin treatment (red) versus random genes (black). (E) A model of how extrusion and compartmentalization combine to shape the spatial organization of the genome inside the nucleus. Intervals of chromatin with similar patterns of histone modification co-localize in nuclear subcompartments. Loop extrusion facilitates short-range contacts between nearby loci as the two subunits of the cohesin-based extrusion complex translocate in opposite directions on chromatin. The extrusion subunits halt at CTCF motifs facing inward, thus forming a loop domain between a pair of motifs in the convergent orientation. Loop domains represent dynamic structures that are maintained by cohesin; only a subset of them may be present at any given time. When the loop anchor motifs span multiple compartment intervals, the dynamics of loop extrusion interfere with compartmentalization by facilitating contacts between loci in different compartments. Loss of cohesin leads to the disappearance of loop domains and to a closer correspondence between genome compartmentalization patterns and histone modification patterns. Interactive figure:

Similar articles

See all similar articles

Cited by 198 PubMed Central articles

See all "Cited by" articles

LinkOut - more resources