and CTCF differentially affect architecture and expression in human cells

Jessica Zuina,1, Jesse R. Dixonb,c,d,1, Michael I. J. A. van der Reijdena, Zhen Yeb,e, Petros Kolovosa, Rutger W. W. Brouwerf,g, Mariëtte P. C. van de Corputa, Harmen J. G. van de Werkena, Tobias A. Knochh,i, Wilfred F. J. van IJckenf, Frank G. Grosvelda,j, Bing Renb,e,2, and Kerstin S. Wendta,2

aDepartment of Cell Biology, iBiophysical Genomics, Department of Cell Biology, fCenter for Biomics, jCancer Genomics Center, Erasmus Medical Center, 3015 GE, Rotterdam, The Netherlands; bLaboratory of Gene Regulation, Ludwig Institute for Cancer Research, La Jolla, CA 92093; cMedical Scientist Training Program, dBiomedical Sciences Graduate Program, eDepartment of Cellular and Molecular Medicine, Institute of Genomic Medicine, Moores Cancer Center, San Diego School of Medicine, University of California, San Diego, La Jolla, CA 92093; gNetherlands Bioinformatics Centre, 6500 HB, Nijmegen, The Netherlands; and hGenome Organization and Function, Bioquant Centre/German Cancer Research Center, 69120 Heidelberg, Germany

Edited* by Richard A. Young, Massachusetts Institute of Technology, Cambridge, MA, and approved October 31, 2013 (received for review September 20, 2013) Recent studies of genome-wide chromatin interactions have Results revealed that the is partitioned into many self- Proteolytic Cleavage of RAD21 Leads to Loss of Long-Range Chromatin associating topological domains. The boundary sequences be- Interactions. To understand the contribution of cohesin to genome tween domains are enriched for binding sites of CTCC-binding organization, we generated a HEK293T cell line containing an factor (CTCF) and the cohesin complex, implicating these two fac- episome-based vector allowing doxycycline-inducible expression tors in the establishment or maintenance of topological domains. of siRNA targeting endogenous RAD21 and a RAD21-EGFP To determine the role of cohesin and CTCF in higher-order chro- variant containing a recognition site for Human rhinovirus 3C matin architecture in human cells, we depleted the cohesin com- (HRV) protease (RAD21cv) (14) (Fig. 1 A and B). Three days plex or CTCF and examined the consequences of loss of these after induction, RAD21cv replaced up to 90% the endoge- factors on higher-order chromatin organization, as well as the nous RAD21 (SI Appendix,Fig.S1A) and was incorporated in transcriptome. We observed a general loss of local chromatin SI Appendix B

the cohesin complex ( , Fig. S1 ). Subsequent trans- CELL BIOLOGY interactions upon disruption of cohesin, but the topological fection of these cells (transfection efficiency ∼80–90%) with domains remain intact. However, we found that depletion of CTCF a construct expressing HRV protease led to full cleavage of not only reduced intradomain interactions but also increased inter- RAD21cv within 24 h (Fig. 1C); TEV (tobacco etch mosaic domain interactions. Furthermore, distinct groups of be- virus) protease was used as negative control. come misregulated upon depletion of cohesin and CTCF. Taken With this system we could rapidly remove the cohesin complex together, these observations suggest that CTCF and cohesin contrib- from interphase , similar to natural RAD21 cleavage ute differentially to chromatin organization and gene regulation. by separase in mitosis (15). Similar systems have been used before to study cohesin in yeast and fly (14, 16, 17). Hi-C | transcriptional regulation | 4C | HOX cluster Fractionation of HRV- or TEV-transfected RAD21cv cells into soluble and chromatin-bound fraction showed that RAD21cv ecent studies of the topological organization of the genome had fully replaced wild-type RAD21 and bound with a comparable suggest that CTCC-binding factor (CTCF) and cohesin R fi might be involved in establishment or maintenance of topological Signi cance domains in the mammalian genome, as their binding sites are enriched at the boundaries of these domains (1). It was proposed For the 2m DNA to fit into the tiny , it is wrapped that CTCF and cohesin might work together to facilitate long- around nucleosomes and folded into loops clustering together range interactions in the genome (2). First, CTCF and the in domains. Genome function depends on this 3D-organization, cohesin complex, consisting of the core subunits SMC3, SMC1, especially on-going dynamic processes like transcription. Tech- RAD21, and STAG1/SA1 or STAG2/SA2, were found to coloc- niques studying the network of DNA contacts genome-wide alize extensively throughout mammalian genomes (3–5). Second, have recently revealed this 3D architecture, but the factors behind this are not understood. We study two both factors are involved in mediating long-range interactions that are known to help form DNA loops: cohesin and CTCC- (6–11). Finally, cohesin was shown to be important for CTCF’s binding factor (CTCF). Respective depletion and analysis of chromatin insulation function (3–5), whereas CTCF is necessary DNA contacts genome-wide show that CTCF is required to to recruit cohesin to the shared binding sites but not to chro- separate neighboring folding domains and keep cohesin in matin (3). CTCF and cohesin have also been recently correlated place, whereas cohesin is important for shaping the domains. with both interaction frequency and during Consistently, we observe different changes of gene expression. differentiation (12), indicating that they may play major roles in mediating the impacts of chromatin structure on gene regulation. Author contributions: J.Z., J.R.D., F.G.G., B.R., and K.S.W. designed research; J.Z., J.R.D., However, the exact mechanisms these factors use to contribute M.I.J.A.v.d.R., Z.Y., M.P.C.v.d.C., W.F.J.v.I., and K.S.W. performed research; J.Z., J.R.D., Z.Y., to chromatin structure and gene regulation are unclear, as de- P.K., R.W.W.B., H.J.G.v.d.W., T.A.K., and F.G.G. analyzed data; and J.Z., J.R.D., B.R., and K.S.W. wrote the paper. pletion of these factors has not yet been systematically tested on The authors declare no conflict of interest. a genome-wide basis. Whether the two factors work in concert *This Direct Submission article had a prearranged editor. or independently, through mechanisms, such as long-range en- Data deposition: The sequence reported in this paper has been deposited in the GenBank hancer looping (13) or chromatin insulation (2) to control database (accession no. GSE44267). chromatin structure and gene expression, is unknown. To de- 1 J.Z. and J.R.D. contributed equally to this work. termine the role of cohesin and CTCF in higher-order chromatin 2To whom correspondence may be addressed. E-mail: [email protected] or k.wendt@ architecture in human cells, we depleted the cohesin complex or erasmusmc.nl. CTCF and examined the consequences of loss of these factors on This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. domain structure and gene expression. 1073/pnas.1317788111/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1317788111 PNAS Early Edition | 1of6 Downloaded by guest on September 27, 2021 Scale 1 Mb Doxycycline HRV hg18 In control cells (RAD21cv/TEV) we observe, as reported be- A Cleavage site F chr11: 1,500,000 2,000,000 2,500,000 3,000,000 P Domains DB DB fore (8), that the IGF2 promoter region (Vp1) interacts strongly siRNA RAD21 RAD21 EGFP (IMR90) H19 IGF2 SMC1 SMC3 H19 IGF2 with an intergenic region between and and one B HEK293T KRTAP5 CTSD stable cell line Refseq fi Genes viewpoint there con rms this (Vp2). Further contacts of Vp1 and Expression of RAD21cv RAD21 Vp2 persist to the region upstream of H19 (Vp3) and over a 500- Depletion of wt RAD21 + dox 72 hours G ChIP/qPCR 1 23 4 5 6 7 Primer kb region until the proximal keratin associated protein gene 40_ Transfection SMC3 _ KRTAP for 24 hours TEV HRV ChIP-seq 0 cluster ( ) near the domain boundary. One viewpoint 4C with six viewpoints: 10,000 RAD21cv/TEV (Vp4) in a cohesin-depleted region shows only weak interactions. RAD21-EGFP Vp1 H RAD21cv/HRV Viewpoint A viewpoint at the upstream boundary (Vp5) shows weak

C RPM Time after HRV protease transfection: interactions with both domains and another viewpoint within the Blot: 0 4 7 9 14 20 24h EGFP RAD21-EGFP 190 neighboring domain downstream (Vp6) consistently shows inter- Cleaved product 5,000 Vp2 SMC3 actions until the domain boundary. We observed similar interaction profiles in the breast endothelial cell line 1-7HB2 (abbreviated γ-tubulin RPM HB2), indicating their conservation between cell lines (SI Appendix, Supernatant Chromatin 43 D 10,000 Transfection: Vp3 Fig. S3). TEVHRV- TEVHRV- dox induction: ++-- ++ Cleavage of RAD21 led to a global loss of interactions across Blot: RAD21-EGFP RAD21* Endogenous RAD21 RPM the entire domain at all viewpoints (Fig. 1H, red line). Correla- EGFP RAD21-EGFP fi C-terminal 296 tion analysis of the interactions identi ed by the different view- 10,000 cleavage product Vp4 SA1/2 points in the control 4C experiments revealed close correlation *This antibody does – CTCF between viewpoints Vp1 Vp5, consistent within their localiza- not recognize the RPM γ-tubulin C-terminal RAD21 cleavage product tion in the same topological domain. This correlation was re- Topo II 147 10,000 fl E Vp5 duced after RAD21 cleavage, re ecting the loss of interactions 0. 5 ChIP:Cells: SI Appendix IgG RAD21cv/TEV ( , Fig. S4). A control experiment with a cell line 0. 4

EGFP RAD21cv/TEV RPM IgG RAD21cv/HRV lacking the HRV cleavage site in RAD21-EGFP (RAD21wt) did 0. 3 EGFP RAD21cv/HRV 183 10,000 not show altered cohesin binding and long-range interactions 0. 2 Vp6 % IP of input after transfection with HRV protease (SI Appendix, Fig. S5). 0. 1 RPM These results strongly support that cohesin plays a role in higher- 0 65123478Primer Cohesin binding sites Negative site 129 order chromatin structure within this domain.

Fig. 1. Cohesin cleavage reduces long-range interactions within the H19/ RAD21 Depletion Predominantly Reduces Shorter-Range Interactions IGF2 A B domain. ( ) Scheme of the expression construct. ( ) Outline of the ex- Within Topological Domains. To investigate whether cohesin plays periment. (C) Time course showing full cleavage of RAD21cv 24 h after D − a general role in topological domain organization, we performed HRV transfection. ( ) Fractionation of uninduced cells ( dox), and trans- Hi-C experiments with control cells (RAD21cv/TEV) and after fected (TEV or HRV) RAD21cv cells into soluble and chromatin-bound RAD21-cleavage (RAD21cv/HRV). We obtained greater than fraction. Blotting for RAD21 shows full replacement of endogenous RAD21 by RAD21cv after induction (+dox). Detection of the RAD21cv C-terminal 370 million nonredundant uniquely mapping read pairs for both EGFP-tag shows full cleavage of RAD21cv after HRV transfection (+dox, control and RAD21-cleaved cells, split between two biological replicates for each condition and high correlation between the HRV) and release of RAD21cv as well as the cohesin subunits STAG1 and SI Ap- STAG2 (SA1/2) from chromatin. CTCF binding to chromatin is not affected. replicates was observed (Pearson correlation 0.90/0.90) ( (E) ChIP-qPCR with anti-EGFP targeting the RAD21cv EGFP-tag shows a re- pendix, Fig. S6). We normalized the Hi-C interaction frequencies duced ChIP signal after HRV transfection at cohesin sites. (F–H)Theeffect according to the iterative correction method (20). For each of RAD21 cleavage on long-range interactions was tested by 4C at six control and RAD21 cleavage replicate we located the topologi- different viewpoints in two topological domains of 11. (F) cal domains using a previously described algorithm (1). Of note, Domain identification in IMR90 cells (domain boundaries, blue boxes) (1). (G) the resolution of Hi-C data is dependent upon the depth of se- Cohesin sites (SMC3) determined in control cells. Primer pairs used for quencing for each sample. With our current sequencing depth, qPCR in E are indicated. (H) The 4C interaction profiles for six different we can readily identify topological domains and analyze rela- viewpoints (highlighted in green) without (RAD21cv/TEV) and with RAD21 tionships between interaction frequencies and various geno- cleavage (RAD21cv/HRV). Data are displayed as reads per million (RPM) and mic elements or DNA binding factors. However, current high- P < only interactions above a cutoff based on value 0.05 are displayed. throughput sequencing technologies are still insufficient to identify differences in interaction frequency between individual binding sites on a genomewide level using Hi-C data. Therefore, level to chromatin. Transfection of RAD21cv cells with HRV our analysis focuses on aggregate genomewide trends in in- protease led to release of RAD21cv from chromatin, but not teraction frequency and their relationship to cohesin and CTCF TEV protease. CTCF remained chromatin-bound (Fig. 1D). – binding patterns. The chromatin-bound RAD21 fraction was reduced by 70 80%, To determine genomewide cohesin binding, we performed estimated from ChIP/quantitative PCR (qPCR) experiments E ChIP-seq for the cohesin subunit SMC3 in control cells (RAD21cv/ targeting several cohesin binding sites (Fig. 1 ). This rapid de- TEV). As observed previously (1), cohesin sites were enriched at struction of cohesin allowed us to study the immediate effect of the boundaries of topological domains (SI Appendix,Fig.S7A), cohesin loss on chromatin organization and transcription, with- although this was only seen for SMC3 sites overlapping with CTCF ’ out interfering with cohesin s function in cell division, because (SI Appendix, Fig. S7 C and D). neither a shift in cell cycle distribution nor cells arrested in mi- To correlate overall Hi-C interaction frequencies with SMC3 tosis were observed (SI Appendix, Fig. S2). binding, we divided the genome into 40-kb bins and stratified To test whether the cleavage of cohesin affects long-range the interacting bin-pairs according to whether SMC3 binds to chromatin interactions, we used the 4C technique (18, 19), which both bins (“SMC3 2×”), only one bin (“SMC3 1×”), or none identifies all regions interacting with one “viewpoint.” To ex- (“none”) (Fig. 2A). We observed a higher interaction frequency amine the interior and the borders of one topological domain (1) in control Hi-C experiments between bin-pairs containing (Fig. 1F) after RAD21 cleavage, we selected six viewpoints (Vp1– SMC3 sites on both ends than when only one or no SMC3 site Vp6) overlapping cohesin/CTCF sites (Fig. 1 G and H)in was present (Fig. 2B), supporting the hypothesis that cohesin chr11p15.5 comprising the noncoding RNA H19,theinsuline- mediates long-range chromatin interactions genomewide. Upon like growth factor 2 (IGF2) gene, and other imprinted genes cleavage of RAD21, we observed an overall loss in local chromatin (here, the H19/IGF2 domain), which we previously used to es- interaction frequency, primarily occurring at distances up to 2 Mb, tablish the role of cohesin in chromatin insulation (3). with a maximum in the range between 100 and 200 kb (Fig. 2C).

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1317788111 Zuin et al. Downloaded by guest on September 27, 2021 The decrease in chromatin interaction frequencies after RAD21 Cohesin and CTCF Shape the Topological Domain Organization in cleavage was the highest when both interacting regions were Nonredundant Ways. To determine the role of CTCF in mediat- bound by SMC3 (Fig. 2C, Inset). The observed loss of interactions, ing chromatin interactions and to compare it to the effects of tested for example on bins 240-kb apart, is statistically significant RAD21 cleavage, we performed Hi-C experiments in duplicate − (P = 3 × 10 197, Wilcoxon test) (Fig. 2D). for CTCF and control siRNA knockdowns in HEK293T cells We next investigated the effects of cohesin complex de- (knockdown efficiency 80%) (SI Appendix, Fig. S9 A and B). We struction on topological domains. The positions of most topo- obtained between 95 and 288 million unique reads for each rep- logical domains did not markedly change upon cleavage of licate, with high correlation between them (Pearson correlation RAD21 and their pattern is still readily apparent in the in- 0.93/0.94) (SI Appendix,Fig.S6). Similar to SMC3, CTCF binding teraction heat maps. Although we consistently called fewer sites are enriched at the boundaries of topological domains in domains in the RAD21-depleted cells, there was a strong control cells (SI Appendix,Fig.S7B). Interestingly, CTCF sites at overlap between domain boundaries identified in the control domain boundaries are more closely aligned to the CTCF binding and RAD21-depleted cells (Fig. 2 E and F). This observation is motifs than CTCF binding sites found within topological domains not surprising because cohesin and CTCF likely are not the (SI Appendix,Fig.S7E). CTCF binding also correlates with the only factors responsible for long-range interactions (12, 13). strength of Hi-C interaction frequency, where interacting bin-pairs However, consistent with the previously described general loss bound by CTCF on each side form stronger interactions compared in interaction frequency, we also observed a clear reduction in with regions with only one or no CTCF sites (Fig. 3 A and B). interaction frequency after RAD21-depletion, both within and Upon knockdown of CTCF, we observed a loss of interactions between domains (SI Appendix, Fig. S8A). Interestingly, the de- within topological domains but, in contrast to RAD21 cleavage, gree of depletion in interaction frequency within domains was we also observed a gain of interactions (Fig. 3 C and D). Detailed most pronounced when one or both interacting bins were asso- analysis of the changes in interaction frequencies versus dis- ciated with a boundary (SI Appendix, Fig. S8B). We performed tance between bins, localization of bins in the same (intra- 3D-FISH (21) and measured distances between cosmid-based domain) or different topological domains (interdomain), and FISH probes located at boundaries of one domain comprising also whether bins have SMC3 or CTCF sites, revealed very in- the homeobox D (HOXD) genes (Fig. 2G). Consistent with teresting details (Fig. 3 E–J). RAD21 depletion appeared to most Hi-C results, we observed significantly increased distances (P = markedly affect interacting loci separated by 100–200 kb (Figs. 2C − 5,632 × 10 4, ANOVA) between the FISH probes after RAD21 and 3E), but CTCF knockdown appeared to most prominently cleavage (Fig. 2 H and I). Consequently, these results suggest affect interacting loci separated by less than 100 kb (Fig. 3H).

that the cohesin complex contributes to the self-association This finding implies that depletion of CTCF or cohesin affects CELL BIOLOGY within topological domains, in part by promoting interactions interactions within topological domains differently. However, between regions near the boundaries. However, cohesin depletion in both cases the loss of interactions was more pronounced for does not appear to contribute to the positioning and segregation intradomain interactions (Fig. 3 E and H, blue line), and in the of neighboring domains. case of RAD21 depletion, even more when interacting bins have

SMC3 2x: Control (TEV) 500 kb hg18 A B E 3.0 G SMC3 2.0 chr2: 177,000,000 12 SMC3 2x RefSeq Genes 1.8 SMC3 1x 40kb None TEV 6 10 Normalized Interaction Frequency SMC3 1x: 1.6 Normalized 0 Interaction Frequency 8 0 1.4 RAD21-depleted (HRV) 3.0 cosmids +2 or 1.2 HRV-TEV 6 ∆ Normalized Interaction Fold Change over “None” 1.0 4 Frequency -2 Interaction Frequency 0 0.5 1.0 1.5 2.0 Normalized 0 None: Distance (Mbp) Scale 5 Mb hg18 2 chr2: 155,000,000 160,000,000 Transfected RAD21cv cells:

SMC3 10 _ H 0 ChIP-Seq TEV HRV Normalized Interaction Frequency 1 _ 0 0.5 1.0 1.5 2.0 DC _ Distance (Mbp) TEV 5 -197 DI 0 - C D p = 3x10 -5 _ DC _ 0.0 7 HRV 5 0 DI 0 - 6 2μm 0.5μm -5 _ -0.1 -0.2 5 RND3 NMI FMNL2 KCNJ3 NR4A2 ERMN WDSUB1 RBMS1 TBR1 GCG RIF1 GALNT13 GPD2 ACVR1 DAPL1CD302 TANK DPP4 p=0.0005632 RBM43 PRPF40A GALNT5 TANC1 SLC4A10 1.5 4 FAP I -0.4 TNFAIP6 ARL6IP6 CYTIP PKP4 MARCH7 s (μm) -0.2 MIR4773-2 LOC100144595 ACVR1C MIR4785 IFIH1 e n=92 3 MIR4773-1 RPRM UPP2 BAZ2B PSMD14 GCA 1.0 NEB CACNB4 ITGB6 KCNH7 n=97 (HRV-TEV) -0.6 SMC3 2x CCDC148-AS1 LY75 -0.3 SMC3 1x 2 ARL5A STAM2 CCDC148 PLA2R1 None 240 kbp interactions -0.8 0.5 1 TEV1 TEV2 HRV2 TEV2 between prob

F e

0 0.5 1.0 1.5 2.0 (1,830) (1,383) ((1,830) c

-0.4 Normalized Interaction Frequency (1,747) Normalized Interaction Frequency n

∆ Distance (Mbp) 0 1,560 1,497 1,283 1,227 0

0 2 4 6810 TEV HRV TEV1 HRV1 HRV1 HRV2 Dist a TEV HRV Distance (Mbp) (1,747)) (1,674) (1,674) (1,383) Transfected RAD21cv cells

Fig. 2. Cohesin cleavage reduces interactions within topological domains. (A) Stratification of the Hi-C interaction map based on SMC3 binding. Clustering of interacting 40-kb bins for presence of SMC3 (brown peak) at both interacting bins (2×), at one bin (1×) or no SMC3 (none). (B) The normalized interaction frequency is plotted versus the distance of interacting bins for the different bin clusters (SMC3 2×,1×, and none). (Inset) In the fold-change relative to the “none” category the SMC3 2× cluster has highest interactions frequencies. (C) Cohesin destruction reduces the interactions between bins. The change of interaction frequency after RAD21 cleavage (HRV-TEV, dark-green curve) is plotted relative to the distance between interacting bins and reveals a reduction of interactions at distances up to 4 Mb. (Inset) The loss of interactions for the SMC3 2×,1×, and none categories. (D) Interaction frequencies are sig- nificantly reduced after RAD21 cleavage, shown hereattheexampleofbins240-kbapart.(E) Normalized Hi-C interaction frequencies observed in RAD21cv cells transfected with either TEV or HRV protease are shown. SMC3 ChIP-sequencing, topological domains positions (DC, domain calls) and directionality index (DI) are shown. Arrows indicate regions with significant changes. (F) Comparison of topological domain boundary calls between Hi-C replicates (TEV1/TEV2; HRV1/HRV2) and control (TEV) and RAD21-depleted cells (HRV). Variations between the respective replicates and between control and RAD21 cleavage experiments are comparable, indicating that the number of domains does not change. (G) Position of cosmid-based DNA-FISH probes at the topological domain including the HOXD . The color of the cosmid probes (red, green) corresponds to the DNA-FISH images in (H). Arrows mark the interactions visualized by DNA-FISH. (H) DNA-FISH using the cosmid probes shown in G in control cells (TEV) and after RAD21 cleavage (HRV). The marked DNA-FISH signals (white boxes) are shown enlarged at the right side of each panel. Consistent with the Hi-C experiments, we observe separation of the FISH signals after RAD21 cleavage. (I) Distances between the FISH-probes observed in the TEV and HRV experiments. The P value was calculated using an ANOVA test on the log distances.

Zuin et al. PNAS Early Edition | 3of6 Downloaded by guest on September 27, 2021 SMC3 sites (Fig. 3F). In the CTCF knockdown experiment we structure, and certain factors may influence these intradomain did not observe this (Fig. 3I). structures. Cohesin and CTCF have recently been shown to be Most remarkably, CTCF and cohesin depletion differentially enriched at the boundaries of subdomains (12). To determine altered interactions between topological domains (Figs. 3 E and how cohesin or CTCF depletion affects these substructures, H, yellow lines). RAD21 depletion again led to reduced inter- we modified our previously applied hidden Markov model (1) actions (Fig. 3E, yellow line), but CTCF depletion led to a gain to identify subdomains. We identified ∼2,600 subdomains in of interactions between neighboring domains (Fig. 3H, yellow control experiments (RAD21cv/TEV and control siRNA) line). This increase was seen over a distance scale up to 2 Mb, with a median size of 520 kb, compared with a 1,080-kb me- consistent with the range of domain sizes (SI Appendix, Fig. dian size for topological domains (SI Appendix,Fig.S11B). S11B), and was more pronounced when CTCF-binding sites were Overall, after RAD21 cleavage similar alterations in in- involved (Fig. 3 G and J). This suggests that CTCF is necessary to teraction frequencies could be observed when considering maintain topological domain boundaries. Interactions gained by both domains and subdomains for intra- and interdomain CTCF depletion could involve cohesin, which is now delocalized interactions (SI Appendix,Fig.S11C and D). However, after (3) but still chromatin-bound (SI Appendix, Fig. S9C). Consis- CTCF depletion the loss of interactions within subdomains tently, the largest gains in interdomain interaction frequency seemed to be more pronounced compared with interactions after CTCF knockdown occurred between bins containing within entire topological domains, and the gain in interactions CTCF or cohesin sites (SI Appendix,Fig.S10). Taken together, was stronger between topological domains compared with be- our observations suggest that cohesin and CTCF shape geno- tween subdomains (SI Appendix,Fig.S11C and E). Therefore, mic structure on the level of topological domains in a non- CTCF appears to function within domains less as an “insulating redundant manner. factor” and more that it appears to be contributing to non- uniform structures that exist within some topological domains. Chromatin Substructures Within Topological Domains also Depend on In contrast, cohesin appears to function, regardless of its ge- Cohesin and CTCF. Frequently, topological domains appear to har- nomic context, as a factor that contributes to the association bor substructures (SI Appendix,Fig.S11A), which have been re- between loci (SI Appendix,Fig.S11D). The observed differ- cently described as “subdomains” or “sub-TADs” (12). This ences between these two factors suggest that cohesin may suggests that some topological domains have a nonuniform interior function primarily by causing the interaction between loci

A B C D

CTCF 2x: 1.8 15 CTCF 2x CTCF CTCF 1x 1.6 None 40kb Δ (HRV-TEV) Δ (siCTCF-siControl) CTCF 1x: 10 1.4 1.2 Fold Change or 5 1.0 Normalized IF 0 0.5 1.0 1.5 2.0 None: Distance (Mbp) 0 0 0.5 1.0 1.5 2.0 +1 Distance (Mbp) +1 All All E 0.1 H 0.2 TEV)

V - -1 Interaction 0 Frequency R 0 -1 ∆ Normalized Interaction Frequency (siCTCF-siControl) (H

-0.1 ∆ Normalized -0.2 -0.2 Control (TEV) siControl -0.3 -0.4 -0.4 4 4

(HRV-TEV) -0.6 ∆ Normalized IF* ∆ Normalized IF* -0.5 Intra-Domain Intra-Domain -0.6 Inter−Domain (ciCTCF-siControl) -0.8 Inter−Domain Interaction Frequency

Normalized 0 Interaction 0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0 Frequency

Normalized 0 Distance (Mbp) Distance (Mbp) Intra-Domain Intra-Domain RAD21-cleavage (HRV) siCTCF F 0.1 I 0.2 4 0 4 0 -0.1 -0.2 -0.2 Interaction Frequency 0 Normalized Interaction -0.3 Frequency 0 -0.4 Normalized Scale hg18 2 Mb Scalehg18 2 Mb -0.4 chr6: 113,000,000 114,000,000 115,000,000 116,000,000 117,000,000 chr6: 113,000,000 114,000,000 115,000,000 116,000,000 117,000,000 (HRV-TEV) -0.6 -0.5 ∆ Normalized IF* DC DC ∆ Normalized IF* SMC3 ≥1x CTCF ≥1x 5 _ 5 _ -0.6 SMC3 None (ciCTCF-siControl) -0.8 CTCF None TEV DI 0 - siControl DI 0 - 0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0 Distance (Mbp) Distance (Mbp) -5 _ -5 _ DC DC Inter-Domain Inter-Domain 5 _ 5 _ G 0.1 J 0.2 HRV DI 0 - siCTCF DI 0 - 0 0 -5 _ -5 _ -0.1 RefSeq Genes RefSeq Genes -0.2 -0.2 -0.3 -0.4 -0.4 (HRV-TEV) -0.6 ∆ Normalized IF*

-0.5 ∆ Normalized IF* SMC3 ≥1x CTCF ≥1x -0.6 SMC3 None (ciCTCF-siControl) -0.8 CTCF None 0 0.5 1.0 1.5 2.0 0 0.5 1.0 1.5 2.0 Distance (Mbp) Distance (Mbp)

Fig. 3. CTCF depletion reduces the function of domain boundaries. (A) As in Fig.2A, interacting 40-kb bins were analyzed for the presence of CTCF at one or both interacting bins. (B) The normalized interaction frequency was plotted versus the distance between bins for each class of interactions (CTCF 2×, CTCF 1×, and none). (Inset) In the fold-change relative to the “none” category, the CTCF 2× class has a higher interaction frequency than CTCF 1× with a maximum for bins 100- to 200-kb apart. (C) The differential interaction map (HRV-TEV) displays changes in interaction frequencies after RAD21 cleavage (red, gain of interactions; blue, loss of interactions). RAD21 depletion leads predominantly to reduced interactions within domains. The domain identification (domain calls, DC; directionality index, DI) is also shown. (D) Similar to C, but showing the changes in interaction frequencies at the same region after CTCF de- pletion by siRNA. A similar pattern of reduced intradomain interaction frequency as in C is observed, visible as blue outline of the domains in the dif- ferential plot (siCTCF-siControl). CTCF depletion yields increased interdomain interactions, visible as red signals between domains. (E)Quantification of the average change of interaction frequencies after RAD21 depletion, analyzed separately for intradomain (blue) and interdomain (yellow) interactions. In both cases, RAD21 depletion leads to a reduced interaction frequency. (F and G) The frequency change of intra- (F) and interdomain (G) interactions was analyzed for the presence of SMC3 on the interacting bins. The loss of interactions is in both cases correlated with SMC3-binding (F, purple; G, orange). (H) Quantification of the average change of interaction frequencies after CTCF siRNA depletion separated for intradomain (blue) and inter- domain (yellow) interactions. CTCF depletion leads to a reduced interaction frequency within and to an increased interaction frequency between domains. (I and J) The change of interaction frequency of intra- (I) and interdomain (J) interactions was further analyzed for the presence of CTCF on the interacting bins. The gain of interactions is more pronounced for interdomain interactions (J) and is stronger when CTCF-sites are present in the inter- acting bins (J,orange).

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1317788111 Zuin et al. Downloaded by guest on September 27, 2021 RAD21cv/HRV throughout the genome, and CTCF may play a greater role in A vs RAD21cv/TEV determining the specificity of interactions and the identity of siCTCF vs siControl topological domains. Log (Treatment/Control) C HOXB-A S3

+2 -2 Myc n 1.2 o CDKN1AHoxA-AS3HoxB-AS3 i 1 Cohesin and CTCF Depletion Affect Gene Expression in Different B chr7: ss 50 kb e r 0.8 27,150,000 27,200,000

xp Ways. Loss of cohesin or CTCF seems to affect chromatin DC TEV 0.6

HRV ee _ 0.4

2 tiv structure in different ways. Given the intimate relationships be- + a 0.2 Rel TEV - 0 tween chromatin structure and gene regulation, we predict that _ TEV HRV Cont. CTCF -2 _ 2 RAD21cv siRNA loss of these two factors would also affect gene expression dif-

RNA-seq +

HRV - HOXA-AS3 ferentially. To test this, we performed RNA-sequencing (RNA- -2 _ n 1.2 o HOXA1 HOXA4 HOXA6 HOXA9 HOXA11 1 HOTAIRM1 HOXA-AS3 HOXA10 HOXA13 ssi seq) in the control (RAD21cv/TEV) and RAD21-depleted HOXA2 HOXA5 HOXA-AS4 HOTTIP e 0.8 HOXA3 HOXA-AS3 HOXA10 0.6 expr (RAD21cv/HRV) cells, as well as CTCF RNAi and mock-trea- HOXA7 HOXA11-AS

ve 0.4 i t 0.2 ted cells. In both cases we observed only modest changes in gene

D Scale 100 kb ela

R 0 SI Appendix – chr6: 131,950,000 132,000,000 132,050,000 132,100,000 TEV HRV Cont. CTCF expression ( , Tables S1 S4), consistent with earlier 50 _ RAD21cv siRNA CTCF observations (3). We observe 48 and 161 differentially expressed 0 _ 0.5 _ 1.2 H19 genes (false-discovery rate, FDR < 5%) for RAD21 and CTCF + 1 0 _ depletion, respectively, but very little overlap between these sets 0 _ 0.8 siContol 0.6 - (Fig. 4A). Among the genes with reduced expression after RAD21 0.4

-0.5 _ ive expression

0.5 _ at 0.2 HOXA11AS HOXA-AS3 RNA-seq depletion are several Hox genes ( , ,

+ Rel 0 0 _ TEV HRV Cont. CTCF HOXB-AS3, HOXB5, HOXC9) (Fig. 4B). We validated the re- 0 _

siCTCF RAD21cv siRNA - duced expression of HOXB-AS3, HOXA-AS3 and H19 by RT- -0.5 _ ARG1 ENPP3 C MED23 G p = 1.92x10e-5 PCR and qPCR (Fig. 4 ). Hox genes have been shown to be 0.4 regulated by antisense transcription as well as the topological 25 CTCF regulated 10 Up-regulated RAD21 regulated 0.3 p = 0.38 E Down-regulated F All genes organization of the locus and the contact to remote enhancers 20 All genes 8 0.2 (22, 23), but so far only for the HOXA cluster has a role 15 6 0.1 cohesin and CTCF at the barrier element been shown (24). 10 4 SMC3/Input Fraction of Interactions CTCF/Input 0 Among genes that are differentially expressed after CTCF de- 5 2 >50% >50% reduction gain pletion, we observed a clear enrichment of CTCF-binding at their 0 0 D E Cohesin-dependent genes promoters (Fig. 4 and ), with a median distance from the CELL BIOLOGY All Refseq genes -2.5kb TSS +2.5kb -2.5kb TSS +2.5kb transcription start site to the nearest CTCF binding site being only H I J 191 bp (SI Appendix,Fig.S12B). In contrast, genes that are dif- ferentially expressed after cohesin depletion are not directly bound at their promoter by SMC3 (Fig. 4F), although they are located Cohesin closer to SMC3 binding sites than would be expected at random CTCF (median distance ∼4kb)(SI Appendix,Fig.S12A). This finding Normal Cohesin cleavage CTCF RNAi indicates that altered expression of genes after RAD21 cleavage may Fig. 4. Transcriptional changes after cohesin cleavage and CTCF depletion. be a product of higher-order chromatin structural changes, for ex- (A) Changes in expression levels after RAD21 cleavage (Upper) and CTCF ample at the HOXA and HOXB cluster (SI Appendix,Fig.S13). To depletion (Lower) (FDR < 0.05) are ranked from highest to lowest. Only very validate this finding, we analyzed interactions of cohesin-regulated few genes behave similarly in both experiments. (B) Expression of HOXA geneswithDNasehypersensitivesitesasmarkersforpotentialdistal genes changes after RAD21 cleavage. Normalized RNA-seq read coverage is gene regulatory regions at a restriction fragment-level resolution. shown for RAD21cv/TEV and RAD21cv/HRV cells (+strand, purple; −strand, We observed that cohesin-regulated genes lose more interactions turquoise). HOX genes differentially expressed with FDR < 0.2 are marked in with distal DNaseI hypersensitive sites than with noncohesin- red. (C) qPCR confirmation of reduced HOXB-AS3 and HOXA-AS3 expression regulated Ref-seq genes (Fig. 4G). These results suggest that after RAD21 cleavage. CTCF depletion did not lead to a consistent reduction, cohesin may regulate gene expression by affecting the interaction as also seen in the analysis of the RNA-seq data (SI Appendix, Table S1). frequency of genes with distal regulatory elements, whereas CTCF Transcription of the H19 noncoding RNA was reduced after CTCF depletion may directly regulate genes by binding at their promoters. and also by RAD21 cleavage. (mean n = 3 ± SD). (D) Transcription of the ENPP3 gene is increased after CTCF knockdown. Normalized RNA-seq read Discussion coverage are shown for control siRNA and CTCF siRNA (+strand, purple; −strand, turquoise). CTCF binding sites are at the promoter and also intragenic. The To reveal the role of CTCF and cohesin in the 3D organization up-regulation was confirmed by RT-PCR to depend solely on CTCF knock- of the human genome, we acutely depleted either the cohesin down (SI Appendix, Fig. S9F). (E) Position of CTCF sites analyzed relative to complex or CTCF from cells and examined the changes in transcription start sites of all genes (black) and genes with altered expression chromatin organization using a combination of 4C, Hi-C, and after CTCF depletion (blue). Each line represents the average fold-enrich- 3D-FISH. We show that cohesin and CTCF contribute differ- ment of CTCF relative to input over a ±2.5-kb window surrounding the entially to the topological domain architecture. First, CTCF or promoters of genes bound by CTCF. CTCF is enriched at the transcription cohesin-bound regions are more likely to interact with each other start site of differentially expressed, in particular down-regulated genes than other genomic sites, extending recent observations at six (green), but it localizes more in the gene body at up-regulated genes. (F) genetic loci in mouse ES cells (12) and supporting a critical role Similar to E, except showing the fold-enrichment of SMC3 over input at the for these factors in the folding of the chromatin fiber. Second, promoter of genes altered after RAD21 cleavage. SMC3 does not appear to disruption of the cohesin complex by proteolytic cleavage leads be enriched at the promoter of the genes regulated by cohesin depletion. to loss of shorter-range chromatin interactions. Surprisingly, this (G) Analysis of changes in interaction frequency between restriction frag- loss of chromatin interaction is not accompanied by breaking ments containing a promoter and restriction fragments containing a distal down of topological domain organization. Third, depletion of DNaseI hypersensitive site (DHS). Shown is the fraction of genes that display CTCF also reduces the intradomain interactions but, in contrast a 50% reduction or 50% increase in interaction frequency after RAD21 to cohesin removal, it also leads to increased interactions between cleavage for either cohesin-regulated genes (orange) or all Ref-seq genes (black). Cohesin-regulated genes are enriched for a loss of interactions with restriction fragments containing distal DHS sites relative to all Ref-seq genes (Fisher’s exact test). (H–J), Models describing the different changes of chro- binding does not change and can still influence chromatin topology and mosomal interactions after cohesin cleavage (I) and CTCF depletion (J). (H) maintain domain identity. (J) CTCF depletion leads to more dynamic domains Cohesin and CTCF shape long-range interactions. (I) RAD21 cleavage and interactions across domain boundaries, normally prevented by CTCF’s destroys cohesin and leads to reduced interactions within domains. CTCF insulation function, potentially involving nonspecifically localizing cohesin.

Zuin et al. PNAS Early Edition | 5of6 Downloaded by guest on September 27, 2021 neighboring domains. In light of the interaction between cohesin observed effects might be weaker than under full depletion, which and CTCF and their colocalization, we propose that cohesin is can only be addressed in the future by more sophisticated gene- mainly involved in chromatin interactions within topological ablation methods. domains, whereas CTCF is important for their spatial segregation However, our general finding is independent from the depletion (Fig. 4 H–J). CTCF is bound stably to the chromatin (25) and efficiency and another study by Seitan et al. (27, 28), using a dif- likely maintains boundaries by determining the localization of ferent experimental system where Rad21 is depleted rather slowly cohesin. Without CTCF, cohesin would no longer localize prop- over 10 d in mouse T-cells, also found that deletion of Rad21 fails erly and, hence, form nonspecific interactions reaching beyond to interrupt topological domain formation (27). Thus, although the boundaries. This theory suggests that the removal of CTCF and cohesin would have different effects on gene expression, and in- cohesin complex is important for the chromosomal interactions deed there is very little overlap between genes affected by the within topological domains, it is not required for the formation of removal of CTCF or cohesin. Correlating these genes with these domains. Taken together, our results provide an initial model cohesin and CTCF binding sites revealed that differentially for understanding the mechanisms of higher-order chromatin or- expressed genes after CTCF depletion tended to have CTCF ganization and its relationship to gene expression. sites close to the promoter. Genes differentially expressed after cohesin cleavage do not correlate with cohesin; instead, a corre- Materials and Methods lation with neighboring DNase hypersensitive sites marking po- HEK293T stable cell lines containing episomes coding for RAD21cv or tential regulatory elements suggests that these genes have lost RAD21wt and RAD21 siRNA were grown for 3 d in the presence of doxy- contact to neighboring enhancers. Several misregulated genes cycline until RAD21 was replaced by the engineered RAD21 versions, trans- belong to the developmentally important HOX clusters, whose fected with either control protease (TEV) or cleavage protease (HRV), and properly timed and localized expression is regulated by topo- harvested after 24 h. logical domain organization and remote enhancers (22, 23). In at A detailed description of all methods can be found in the SI Appendix. least two HOX clusters (HOXA, HOXB) we observed a loss of G SI Appendix interactions after cohesin cleavage (Fig. 2 , and , ACKNOWLEDGMENTS. We thank L. Schöckel and O. Stemmann for DNA Fig. S13). This finding shows that defects in cohesin or its reg- constructs; R. Stadhouders, E. Soler, and R.-J. Palstra for advice on 4C; ulatory proteins could influence topological domain organization R. van der Linden for FACS cell cycle analysis; L. Edsall and S. Kuan for assistance of developmental genes, and hints how cohesin defects might be in next-generation DNA sequencing; A. de Klein and B. Eussen for providing linked to the phenotype observed in patients diagnosed with cosmids; and N. Galjart and J.-M. Peters for antibodies. This work was sup- “ ” ported in part by the Netherlands Organization for Scientific Research via Cornelia de Lange syndrome, a cohesinopathy that is associ- ALW2PJ/11029 (J.Z.) and the EpiGenSys project/ERASysBio+ initiative in the ated with defects in multiple organ systems (26). EU FP7 ERA-NET Plus program (F.G.G. and T.A.K.), the Erasmus Medical Cen- The finding that the topological domain structure remains ter (K.S.W.), the Dutch Royal Academy (F.G.G.), the Netherlands Organiza- largely intact after cohesin (and CTCF) depletion is surprising, tion for Health Research and Development (E-RARE network TARGET-CdLS) considering the high density of cohesin occupancy at topological (K.S.W.), the Netherlands Genomics Initiative (Zenith and Medical Epige- netics) (H.J.G.v.d.W. and F.G.G.) and the EU Systems Biology Consortium domain boundaries, and the extensive colocalization between (SyBoSS) (F.G.G.). Work in the B.R. laboratory was supported by the Ludwig CTCF and cohesin. However, protein depletion is never complete, Institute for Cancer Research, the California Institute of Regenerative Med- and the transfection efficiency is also never 100%. Therefore, the icine (RN2-00905-1), and the National Institutes of Health.

1. Dixon JR, et al. (2012) Topological domains in mammalian genomes identified by 16. Gruber S, Haering CH, Nasmyth K (2003) Chromosomal cohesin forms a ring. Cell analysis of chromatin interactions. Nature 485(7398):376–380. 112(6):765–777. 2. Phillips JE, Corces VG (2009) CTCF: Master weaver of the genome. Cell 137(7):1194– 17. Pauli A, et al. (2008) Cell-type-specific TEV protease cleavage reveals cohesin functions 1211. in Drosophila neurons. Dev Cell 14(2):239–251. 3. Wendt KS, et al. (2008) Cohesin mediates transcriptional insulation by CCCTC-binding 18. Stadhouders R, et al. (2013) Multiplexed chromosome conformation capture se- factor. Nature 451(7180):796–801. quencing for rapid genome-scale high-resolution detection of long-range chromatin 4. Parelho V, et al. (2008) functionally associate with CTCF on mammalian interactions. Nat Protoc 8(3):509–524. Cell – chromosome arms. 132(3):422 433. 19. Simonis M, Kooren J, de Laat W (2007) An evaluation of 3C-based methods to capture Proc Natl Acad Sci 5. Rubio ED, et al. (2008) CTCF physically links cohesin to chromatin. DNA interactions. Nat Methods 4(11):895–901. USA – 105(24):8309 8314. 20. Imakaev M, et al. (2012) Iterative correction of Hi-C data reveals hallmarks of chro- 6. Splinter E, et al. (2006) CTCF mediates long-range chromatin looping and local histone mosome organization. Nat Methods 9(10):999–1003. fi Genes Dev – modi cation in the beta-globin locus. 20(17):2349 2354. 21. van de Corput MP, et al. (2012) Super-resolution imaging reveals three-dimensional 7. Kurukuti S, et al. (2006) CTCF binding at the H19 imprinting control region mediates folding dynamics of the β-globin locus upon gene activation. J Cell Sci 125(Pt 19): maternally inherited higher-order chromatin conformation to restrict enhancer access 4630–4639. to Igf2. Proc Natl Acad Sci USA 103(28):10684–10689. 22. Alexander T, Nolte C, Krumlauf R (2009) Hox genes and segmentation of the hind- 8. Nativio R, et al. (2009) Cohesin is required for higher-order chromatin conformation brain and axial skeleton. Annu Rev Cell Dev Biol 25:431–456. at the imprinted IGF2-H19 locus. PLoS Genet 5(11):e1000739. 23. Noordermeer D, et al. (2011) The dynamic architecture of Hox gene clusters. Science 9. Mishiro T, et al. (2009) Architectural roles of multiple chromatin insulators at the 334(6053):222–225. human apolipoprotein gene cluster. EMBO J 28(9):1234–1245. 24. Kim YJ, Cecchini KR, Kim TH (2011) Conserved, developmentally regulated mechanism 10. Hadjur S, et al. (2009) Cohesins form chromosomal cis-interactions at the develop- couples chromosomal looping and heterochromatin barrier activity at the homeobox mentally regulated IFNG locus. Nature 460(7253):410–413. Proc Natl Acad Sci USA – 11. Hou C, Dale R, Dean A (2010) Cell type specificity of chromatin organization mediated gene A locus. 108(18):7391 7396. by CTCF and cohesin. Proc Natl Acad Sci USA 107(8):3651–3656. 25. Li Y, et al. (2013) Characterization of constitutive CTCF/cohesin loci: A possible role in BMC Genomics 12. Phillips-Cremins JE, et al. (2013) Architectural protein subclasses shape 3D organiza- establishing topological domains in mammalian genomes. 14(1):553. GeneReviews tion of genomes during lineage commitment. Cell 153(6):1281–1295. 26. Deardorff MA, Clark DM, Krantz ID (2005) Cornelia de Lange Syndrome. 13. Kagey MH, et al. (2010) Mediator and cohesin connect gene expression and chro- eds Pagon RA, et al. (Univ of Washington, Seattle). Available at http://www.ncbi.nlm. matin architecture. Nature 467(7314):430–435. nih.gov/books/NBK1104/. Accessed August 30, 2013. 14. Schöckel L, Möckel M, Mayer B, Boos D, Stemmann O (2011) Cleavage of cohesin 27. Seitan VC, et al. (2013) Cohesin-based chromatin interactions enable regulated gene rings coordinates the separation of centrioles and chromatids. Nat Cell Biol 13(8): expression within preexisting architectural compartments. Genome Res, 10.1101/ 966–972. gr.161620.113. 15. Uhlmann F, Lottspeich F, Nasmyth K (1999) Sister-chromatid separation at anaphase 28. Seitan VC, et al. (2011) A role for cohesin in T-cell-receptor rearrangement and onset is promoted by cleavage of the cohesin subunit Scc1. Nature 400(6739):37–42. thymocyte differentiation. Nature 476(7361):467–471.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1317788111 Zuin et al. Downloaded by guest on September 27, 2021