ARTICLES

Cohesin mediates transcriptionalinsulation by CCCTC-binding factorKerstin S. Wendt1*, Keisuke Yoshida2*, Takehiko Itoh3*, Masashige Bando2, Birgit Koch1, Erika Schirghuber1,Shuichi Tsutsumi4, Genta Nagae4, Ko Ishihara6, Tsuyoshi Mishiro6, Kazuhide Yahata5, Fumio Imamoto5,Hiroyuki Aburatani4, Mitsuyoshi Nakao6, Naoko Imamoto7, Kazuhiro Maeshima7, Katsuhiko Shirahige2

& Jan-Michael Peters1

Cohesin complexes mediate sister-chromatid cohesion in dividing cells but may also contribute to gene regulation inpostmitotic cells. How cohesin regulates gene expression is not known. Here we describe cohesin-binding sites in the humangenome and show that most of these are associated with the CCCTC-binding factor (CTCF), a zinc-finger protein required fortranscriptional insulation. CTCF is dispensable for cohesin loading onto DNA, but is needed to enrich cohesin at specificbinding sites. Cohesin enables CTCF to insulate promoters from distant enhancers and controls transcription at the H19/IGF2(insulin-like growth factor 2) locus. This role of cohesin seems to be independent of its role in cohesion. We propose thatcohesin functions as a transcriptional insulator, and speculate that subtle deficiencies in this function contribute to‘cohesinopathies’ such as Cornelia de Lange syndrome.

In proliferating cells, cohesin complexes physically connect repli-cated DNA molecules (‘sister chromatids’) from S phase until thesubsequent anaphase of mitosis or meiosis. This sister-chromatidcohesion is essential for chromosome segregation and for DNAdamage repair. Cohesin is composed of four core subunits, calledSMC1, SMC3, SCC1 (also known as MDC1 and RAD21) and SCC3(also known as SA2 and STAG2)1–3. These proteins have been pro-posed to mediate cohesion by embracing sister chromatids as a ring4.

The essential role of cohesin in cohesion is well established, butevidence obtained in yeast and different animal species implies thatcohesin also contributes to gene regulation, chromatin structure anddevelopment5–13. Furthermore, certain human diseases have beenlinked to hypomorphic mutations in cohesin and in proteins thatregulate cohesin. Cornelia de Lange syndrome (CdLS) is charac-terized by growth and mental retardation, craniofacial anomaliesand microcephaly. This disease can be caused by mutations in aprotein that is required to load cohesin onto DNA, called SCC2 (alsoknown as NIPBL and delangin), or by mutations in SMC1 or SMC3(refs 14–17). Roberts/SC phocomelia syndrome (RBS/SC, OMIM26900) is a related disease that has been linked to mutations inESCO2, a protein implicated in the establishment of cohesion18,19.Surprisingly, most of these mutations do not cause obvious defects incell proliferation, implying that the resulting developmental abnor-malities are not caused by defects in cohesion but reflect a distinct, yetunknown, function of cohesion proteins.

Cohesin is expressed in differentiated postmitotic cells

In vertebrates, cohesin binds to chromatin at the end of mitosis, longbefore cohesion is established in the next cell cycle3,20,21. This suggeststhat cohesin may also have a function on unreplicated DNA, inde-pendent of its role in cohesion. To explore this possibility, we firsttested if cohesin is also expressed in postmitotic cells, which lack

cohesion. Immunoblotting experiments identified SCC1 and SMC1in numerous mouse tissues, including brain (Fig. 1a). SMC3 anti-bodies immunoprecipitated cohesin complexes from brain extracts(Fig. 1b), and by immunofluorescence microscopy (IFM), SCC1staining was observed in the nuclei of neurons (Fig. 1c and Supple-mentary Fig. 1). PDS5B, a protein that is associated with cohesin, hasrecently also been detected in mouse neurons13. Cohesin is therefore

*These authors contributed equally to this work.

1Research Institute of Molecular Pathology (I.M.P.), Dr. Bohr Gasse 7, 1030 Vienna, Austria. 2Graduate School of Bioscience and Biotechnology, Tokyo Institute of Technology, B2C4259, Nagatsuta, Midori-ku, Yokohama City, Kanagawa 226-8501, Japan. 3Research Center for Advanced Science and Technology, Mitsubishi Research Institute Inc., Chiyoda-ku,Tokyo 100-8141, Japan. 4Genome Science Division, Research Center for Advanced Science and Technology (RCAST), the University of Tokyo, Tokyo 153-8904, Japan. 5Department ofMolecular Biology, Research Institute for Microbial Diseases, Osaka University, Suita, Osaka 565-0871, Japan. 6Department of Regeneration Medicine, Institute of MolecularEmbryology and Genetics Kumamoto University, 2-2-1 Honjo, Kumamoto 860-0811, Japan. 7Cellular Dynamics Laboratory, RIKEN, Wako, Saitama 351-0198, Japan.

SMC1

CTCF

Proteasome

SCC1

SMC1

SA2

SM

C3

SM

C3

Con

trol

Con

trol

HeLacells

Mousebrain

WB

IP

SCC1WB

Tissue

10 µM

Merge SCC1 NeuN DAPI

a b

c

Bra

inH

eart

Live

rLu

ngK

idne

yM

uscl

eP

ancr

eas

Sp

leen

Sto

mac

hTe

stis

Thym

usTo

ngue

Figure 1 | Cohesin is expressed in postmitotic cells. a, Mouse tissue extractswere analysed for cohesin and CTCF expression by SDS–polyacrylamide gelelectrophoresis (SDS–PAGE) and immunoblotting with the indicatedantibodies. WB, western blot. b, SMC3 immunoprecipitates (IP) obtainedfrom mouse brain and HeLa interphase extracts were analysed bySDS–PAGE and immunoblotting with the indicated antibodies. c, Frozenthin sections from mouse brain cortex were co-stained for SCC1, theneuronal marker NeuN and DNA (4,6-diamidino-2-phenylindole, DAPI).

Vol 451 | 14 February 2008 | doi:10.1038/nature06634

796Nature Publishing Group©2008

expressed in differentiated postmitotic cells, consistent with the pos-sibility that cohesin performs functions in addition to cohesion.

Identification of cohesin-binding sites in the human genome

We next identified cohesin-binding sites in the human genome bychromatin immunoprecipitation (ChIP). We initially used Affyme-trix oligonucleotide tiling arrays designed by the ENCODE project22

(22-base pair (bp) resolution, representing 1% of the genome). InHeLa cells synchronized in the G2 phase, SCC1 antibodies identified167 sites with high statistical confidence. Most of these were alsodetected with antibodies to SMC3 and SA2, but not with unrelatedantibodies, such as anti-p53 (Fig. 2a, Supplementary Fig. 2 andSupplementary Table 1). Most if not all of these sites are cohesin-specific, because depleting SCC1 by RNA interference (RNAi) elimi-nated or reduced the signals in SMC3 ChIP-chip experiments(Fig. 2a). We confirmed 27 cohesin sites by analysing SCC1 andSMC3 ChIP samples with quantitative polymerase chain reactions(qPCRs), whereas no enrichment was found at 9 control sites (Fig. 2band Supplementary Table 2).

By ChIP-chip, we found that most SCC1 sites (82%) were identicalin the G1 phase and G2 phase, indicating that cohesin binds to thesame sites independent of cohesion. No specific sites could bedetected in mitotic cells, consistent with the finding that most cohe-sin dissociates from chromosome arms in prophase3,20,21. Most SCC1sites were also found in diploid hTERT RPE-1 and immortalizedB cells; we detected more sites in these cells than in HeLa cells(Supplementary Table 1). Different cell lines may therefore containboth common and unique cohesin sites.

To investigate the genome-wide distribution of cohesin, we ana-lysed SCC1 ChIP samples on Affymetrix arrays that represent all non-repetitive elements of the human genome with 35 bp resolution. Weidentified 8,811 sites with high confidence (P-value , 5.0 3 1028).Most of these are located in intergenic regions (49%), in introns(35%) or within 5 kilobases (kb) upstream or downstream of genes(13%; Supplementary Fig. 3). The latter sites are over-representedcompared to their frequency in the genome (Fig. 2c). The averagedistance between two sites is 340 kb and ranges from 480 bp to35.6 megabases (Mb).

For yeast, it has been proposed that cohesin can be moved alongDNA by the transcription machinery23,24; consistent with this pos-sibility, more than 80% of cohesin sites on chromosome arms areregions of convergent transcription25. In the human genome, how-ever, more than 3,000 cohesin sites are found in introns, many ofwhich are actively transcribed in HeLa cells. In humans, transcriptioncan therefore not remove cohesin permanently from genes.

Cohesin co-localizes with CTCF in the human genome

We noticed that cohesin is enriched at several sites to which thetranscriptional insulator protein CTCF26 binds, such as the H19imprinting control region (ICR) and the b-globin locus controlregion (LCR). To assess whether cohesin and CTCF co-localize alsoat other sites, we mapped CTCF sites in HeLa, hTERT RPE-1 and Bcells by ChIP-chip on ENCODE arrays (Fig. 2a and SupplementaryTable 1). Many CTCF sites were identical to SCC1 sites. For example,in HeLa cells in G2 phase, 165 out of 167 SCC1 sites are also bound byCTCF. Accordingly, the consensus sequences for SCC1 and CTCFsites are very similar (Fig. 2d). Similar results were obtained withgenome-wide arrays (Supplementary Fig. 4). We identified 13,894CTCF sites, 7,813 of which were identical to SCC1 sites, correspond-ing to 89% of all SCC1 sites. CTCF sites at which we did not detectSCC1 by ChIP-chip may nevertheless be bound by cohesin, becausewe also detected cohesin at these sites by ChIP-qPCR (Supplemen-tary Fig. 5e and Supplementary Table 2). Recently, independentstudies have identified CTCF sites by ChIP-chip, ChIP-sequencingand bioinformatics27–29. Although different cells were used, morethan 60% of our CTCF sites were also identified in these studies.

We could not co-immunoprecipitate cohesin and CTCF fromHeLa extracts (data not shown), indicating that the soluble formsof these proteins are not stably bound to each other. As predicted byChIP-chip, cohesin and CTCF antibodies yielded similar IFM pat-terns (Supplementary Fig. 6). Both showed nuclear staining in inter-phase and cytoplasmic staining from prometaphase until anaphase.Contrary to earlier findings30, we could detect little if any CTCF onmitotic chromosomes by IFM, ChIP-qPCR and immunoblotting(Supplementary Figs 6 and 7). Like cohesin, most CTCF thereforedissociates from chromosomes in mitosis and rebinds in telophase.

CTCF is required for the positioning of cohesin on DNA

Given the high degree of co-localization, we wondered whetherCTCF is required for enrichment of cohesin at specific sites. To testthis, we analysed SCC1 and SMC3 by ChIP-qPCR in CTCF-depletedHeLa cells. As a control, we analysed SCC2-depleted cells in whichcohesin loading onto DNA is reduced31. As predicted, the abundanceof cohesin sites was reduced after SCC2 depletion, but remarkably thesame was true after CTCF depletion (Supplementary Fig. 5a, b).When we analysed SCC1 ChIP samples from CTCF-depleted cellson ENCODE arrays, we could only detect 77 out of 167 cohesin sites(Fig. 2a; Supplementary Table 1). We used cells in the G2 phase forthese experiments, ruling out the possibility that cohesin binding wasreduced because cells accumulated in mitosis. CTCF is thereforerequired for enrichment of cohesin at its proper binding sites. Incontrast, in ChIP-qPCR experiments we also found a 50% reductionin CTCF binding after cohesin depletion (Supplementary Fig. 5c, d).Cohesin might therefore also contribute to CTCF positioning.

We next investigated if CTCF is required for loading of cohesinonto DNA, as is the case for SCC2. Immunoblot experiments indi-cated that this is not the case, because we observed similar SCC1signals in chromatin pellets in control and CTCF-depleted HeLaand hTERT RPE cells (Supplementary Fig. 8a and data not shown).Quantitative IFM (qIFM) confirmed that chromatin-bound cohesinlevels were not reduced after CTCF depletion, and CTCF on chro-matin was also not reduced by SCC1 or SCC2 depletion (Supplemen-tary Fig. 8b–d). Cohesin and CTCF can therefore associate with DNAindependently of each other. Accordingly, we did not observe majorcohesion defects in CTCF-depleted cells (data not shown). Thesedata suggest that cohesin is distributed more broadly on DNA inthe absence of CTCF, and may therefore not be detectable byChIP-chip.

Cohesin is required for the insulator function of CTCF

To test if cohesin is required for CTCF function, we measured tran-scriptional changes in SCC1- and CTCF-depleted HeLa cells byDNA-chip technology. In G2 cells, we identified 194 transcripts thatwere upregulated and 90 transcripts that were downregulated afterboth SCC1 and CTCF depletion (Supplementary Tables 3 and 4).Interestingly, genes within 25 kb of cohesin sites had a higher ten-dency to be upregulated than downregulated (Fig. 2e), consistentwith an insulator function for cohesin/CTCF sites32. Cohesin andCTCF may therefore regulate genes near their binding sites in similarways.

Some CTCF sites are known to function as transcriptional insula-tors. We therefore tested if cohesin depletion influences the ability ofCTCF sites to insulate a gene from a distant enhancer. We first ana-lysed the H19 ICR (Fig. 3a), which normally controls transcription atthe H19/IGF2 locus33,34. On a reporter plasmid, the H19 ICR can alsoblock a simian virus 40 enhancer from activating the H19 promoter,which drives expression of firefly luciferase35 (Fig. 3b); this effectdepends on CTCF35. We used this system to test whether transcrip-tional insulation also requires cohesin.

By ChIP-qPCR, we observed binding of CTCF and cohesin to theluciferase reporter plasmids when these were transiently transfectedinto HeLa cells (Fig. 3c). As reported35, CTCF depletion increasedluciferase activity in cell lysates, indicative of increased luciferase

NATURE | Vol 451 | 14 February 2008 ARTICLES

797Nature Publishing Group©2008

expression; remarkably, depletion of SCC1 and SMC3 had very sim-ilar effects (Fig. 3d). Luciferase activity was also decreased by deple-tion of the cohesin-loading factor SCC4, but not by depletion ofSCC2 by RNAi, perhaps owing to incomplete SCC2 depletion(Supplementary Fig. 9). Depletion of the cohesion establishment

factor sororin36 had no effect. When we used a plasmid in whichthe ICR between the luciferase gene and the simian virus 40 enhancerhad been mutated so that CTCF binding is lost35, luciferase activityincreased only insignificantly after CTCF, SCC1 or SMC3 depletion(Fig. 3b, d). The effects of cohesin and CTCF depletion therefore

Bits

a

SCC1 HeLa G1

SCC1 HeLa G2

Ensembl gene

SMC3 HeLa G2

CTCF HeLa G2

SA2 HeLa G2

SMC3 HeLa M

position

ChIP

131.50 Mb 131.60 Mb 131.70 Mb 131.80 Mb 131.90 Mb

132.00 Mb

PDLIM4SLC22A4

SLC22A5NP_001013739.1

RAD50

P4HA2 Q8N6F0 Q6ZTC2 IRF1 IL5

15

10

15

10

15

10

15

10

15

10

15

10

SMC3 ∆SCC1HeLa G2 1

510

15

10

15

10

15

10

15

10

15

10

15

10

ln (MAT score)

15

10SCC1 ∆CTCFHeLa G2

15

10CTCF ∆SCC1

HeLa G2

b

Coding exon

5′UTR

Intron

3′UTR

Upstream

Downstream

IntergenicOverlap(7,813 sites)

SCC1(8,811 sites)CTCF(13,894 sites)

Enrichment of cohesin and CTCF sites0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0

CTCF consensusCohesin consensus (SCC1)c

0

0.4

0.8

1.2

Primer number

SCC1

CTCF

Mock

81 76543 61110192 15141312

d

ChI

P e

ffic

ienc

y(p

erce

ntag

e IP

of i

nput

)

18 8212029171 2726252423 6313039222 35343332

0

e

1

2

0

1

2

Bits

5′ 1 2 3 4 5 6 7 8 9 10 11 12 13 14 153′ 5′

1 2 3 4 5 6 7 8 9 10 11 12 13 14 153′

SCC1 RNAiCTCF RNAi

0.5

1.0

1.5

2.0

2.5

3.0

Distance from cohesin- and CTCF-binding sites0 Mb0.2 0.4 0.6 0.8 1.0

SCC1 B cell

CTCF B cell

SCC1 RPE1

SMC3 RPE1

CTCF RPE1

p53 HeLa

Incr

ease

d/d

ecre

ased

tr

ansc

ripts

Figure 2 | Identification of cohesin- and CTCF-binding sites in the humangenome. a, Examples of ChIP-chip data obtained with ENCODE arrays(position 131.5–132 Mb of human chromosome 5). Vertical axes show theMAT score (log scale), which reflects the fold-enrichment of the ChIP-chipsamples. Regions in which signals were significantly enriched are coloured inorange with red flags. Cells were enriched in G1 phase or G2 phase by double-thymidine arrest/release (dTAR), and in mitosis (M) by dTAR followed by2 h nocodazole treatment and shake-off. DSCC1 and DCTCF, cells depletedby RNAi of SCC1 and CTCF, respectively. All samples were prepared frominterphase cells unless indicated otherwise. b, SCC1 and CTCF ChIP-qPCRdata obtained with primers (Supplementary Table 2) designed for 27 SCC1-positive and 9 negative sites identified by ChIP-chip. In all cases, positivesites were enriched more than 25-fold compared to negative sites (mean ofn 5 3; error bars 6 s.d.). c, Frequencies of cohesin- and CTCF-binding sitesin non-repetitive regions of the human genome (approximately 50% of theentire genome): intergenic regions, introns, exons, and 59 and 39

untranslated regions (UTRs) of Ensemble genes are shown, as well as regions5 kb upstream and downstream of 59 UTRs and 39 UTRs, respectively. Thenumber of sites in these regions was normalized against the frequencies ofthese regions in the genome and displayed as fold enrichment. d, DNAconsensus sequences for SCC1- and CTCF-binding, derived from ENCODEarray data with the MEME tool50. The height of each letter represents therelative frequency of nucleotides at different positions in the consensus.More than 70% of 362 CTCF sites and 75% of 168 SCC1 sites analysedcontain these motifs. e, Relationships between the distance of SCC1- andCTCF-binding sites from genes and transcriptional changes induced byRNAi-mediated depletion of SCC1 and CTCF. Transcriptional analyses werecarried out using MAS (microarray analysis software, provided byAffymetrix), and significantly increased and decreased genes were extractedusing the default threshold. The ratio of numbers of increased/decreasedgenes (y-axis) was plotted against the distance from each binding site(x-axis).

ARTICLES NATURE | Vol 451 | 14 February 2008

798Nature Publishing Group©2008

depend on a functional ICR. We used synchronized G2 cells in all ofthese experiments, ruling out indirect cell-cycle effects. We thereforeconclude that the insulator function of the H19 ICR requires cohesinbinding to DNA, but not the establishment of cohesion.

We also tested if cohesin is required for the function of the HS4(DNase hypersensitive site 4) insulator, which is located in theb-globin LCR37. HS4 can block the effect of enhancers on distantpromoters in a CTCF-dependent manner, and HS4 can act as abarrier to chromosomal position effects32,38. To test if the insulatorfunction of HS4 depends on cohesin, we used HeLa cell lines inwhich the expression levels of tandemly oriented marker cDNAsare increased by the presence of chicken HS4 insulators39. WhenSCC1- and CTCF-depleted cells were analysed by IFM and immuno-blotting, marker cDNA expression was reduced, whereas endogenousproteins were unaffected (Supplementary Figs 10 and 11). Cohesinmay therefore also contribute to the insulator function of HS4.

Cohesin controls transcription at the imprinted H19/IGF2 locus

The H19 ICR ensures that H19 is only transcribed from the maternalallele, whereas the neighbouring IGF2 gene is only transcribed fromthe paternal allele (Fig. 4a; reviewed in ref. 40). This imprintingdepends on CTCF, which can bind to the H19 ICR on the maternalallele where it prevents IGF2 activation from a distant enhancer,which instead activates H19. In contrast, CTCF binding to the pater-nal allele is blocked by methylation on CpG sequences. As a con-sequence, the distal enhancer activates IGF2 but cannot stimulateH19 transcription33,34.

To investigate whether cohesin is required for imprinting, we firsttested if cohesin, like CTCF, is specifically bound to the maternal H19ICR, which is located on human chromosome 11. To address this, weused mouse cells that carry either the maternal or the paternal allele ofhuman chromosome 11 (ref. 41). Chromosome-specific fluorescencein situ hybridization (FISH) confirmed the presence of a single copyof human chromosome 11 in these hybrid cells (Fig. 4b), and PCRwith reverse transcription (RT–PCR) indicated that specific tran-scripts such as KCNQ1OT1, KCNQ1 and H19 are still expressed fromhuman chromosome 11 in an imprinted fashion41,42 (Fig. 4c and datanot shown). The endogenous mouse H19 ICR could be isolated withCTCF and SMC3 antibodies from both cell lines by ChIP-qPCR, butthe same antibodies could isolate the human H19 ICR only fromcells carrying the maternal human chromosome 11 (Fig. 4d). LikeCTCF, cohesin is therefore specifically bound to the maternalallele of the H19 ICR. We confirmed this notion by re-ChIP experi-ments in which we isolated DNA fragments first with CTCF anti-bodies, eluted the ChIP samples from the antibody beads and thenre-immunoprecipitated with SCC1 antibodies, or vice versa. For alltested cohesin sites, including the human H19 ICR, we found thatcohesin and CTCF are bound to the same DNA molecules (Fig. 4e).

IP

of i

nput

(%)

A911M-2 A911P-1

a

b c

d

Mouse tubulin

Human KCNQ1OT1

Human KCNQ1

A9

A91

1M-2

A91

1P-1

0

0.1

0.2

0.3

0.01

0.03

0.05

0.07

0

H19ICR

IGF2

ICR

CTCF

e

ICR (primer 26)

LCR (primer 27)

Chr. 5 (primer 17)

Chr. 6 (primer 23)

Chr. 21 (primer 8)

SCC1

2nd ChIP:

1st ChIP:

Inpu

tSC

C1SC

C1CTC

FM

ock

CTCF

Inpu

tCTC

FSC

C1CTC

FM

ock

PCR:

Enh EnhH19 H19IGF2 IGF2

SMC3

CTCF

Mock

0.06

0.04

0.02

PCR

Cell line

Figure 4 | Cohesin co-localizes with CTCF on the maternal allele of the H19ICR. a, Schematic representation of the H9/IGF2 locus and itstranscriptional regulation. Circles represent DNA methylation. b, Detectionof human chromosome 11 in mouse–human hybrid cells by FISH. A911M-2and A911P-1 cells contain a maternal and paternal copy of humanchromosome 11, respectively41. c, Hybrid cells as in b were analysed byRT–PCR for the presence of paternal (KCNQ1OT1) or maternal-specific(KCNQ1) transcripts encoded by human chromosome 11. d, SMC3 andCTCF ChIP samples obtained from hybrid cells as in b were analysed byqPCR with primers specific for the human (primer 96, left) or the mouse(primer MmH19, right) H19 ICR (data representative of severalindependent experiments are shown). e, Re-ChIP assays to test co-localization of cohesin and CTCF on the same DNA molecules. ChIPexperiments were either performed first with SCC1 antibodies and thenanalysed by re-ChIP with CTCF antibodies, or vice versa. All samples wereanalysed by PCR with the indicated primers. Chr., chromosome.

SCC1

SMC3

CTCF 15

1015

1015

10

a

bd

CTCF

SCC1

SMC3

Soror

in

Contro

l0

Rel

ativ

e lu

cife

rase

act

ivity

pIHLIE

pIHLME

Moc

k

SM

C3

CTC

FM

ock

SM

C3

CTC

F

PC

R n

egP

CR

pos

pIHLIE w/o

Mouse ICR

Human ICR

c Plasmid transfection:

PCR

SCC1

Ensemblgene

15

10

ChIP:

Chromosome 11

pIHLIE

LuciferaseH19PICR ICR Enh

LuciferaseH19PICR ICR EnhCTCF CTCF

LuciferaseH19PICR MUT EnhCTCF

pIHLME

+CTCF

–CTCF

+CTCF

IP

2.00 Mb 2.05 Mb 2.10 MbQ8N2L8

IGF2AS

H19 IGF2 INSTH

1.980M

Figure 3 | Cohesin is required for the insulator function of the H19 ICR.a, Cohesin co-localizes with CTCF at the H19/IGF2 locus. SCC1 ChIP-chipdata are shown for the entire locus together with an enlarged view of SCC1,SMC3 and CTCF ChIP-chip data for the H19 ICR. b, Schematicrepresentation of the reporter constructs pIHLIE and pIHLME that wereused in enhancer-blocking assays. On pIHLIE, CTCF binding to the ICRrepresses luciferase expression by blocking enhancer access to the promoter.In pIHLME, H19 ICR nucleotides required for CTCF binding have beenmutated35. Enh, enhancer; H19P, mouse H19 promoter; MUT, ICR mutatedin a way that CTCF cannot bind. The black line shows the fragment amplifiedby PCR. c, SMC3 and CTCF ChIP samples obtained from HeLa cellstransfected with pIHLIE were analysed by PCR with primers specific for themouse H19 ICR (located on pIHLIE) or the endogenous human H19 ICR.neg, negative control (neither ChIP nor whole cell extract included); pos,positive control (whole extract of transfected cells included in PCR); w/o,without transfection. d, HeLa cells were depleted by RNAi of the indicatedproteins and transfected with a Renilla luciferase control plasmid togetherwith either pIHLIE or pIHLME, which contain firefly luciferase cDNAs.After synchronization in the G2 phase, the ratios of firefly versus Renillaluciferase activities were determined in the cell lysate and normalized againstcontrol RNAi (mean of n 5 3; error bars, 6 s.d.).

NATURE | Vol 451 | 14 February 2008 ARTICLES

799Nature Publishing Group©2008

To test if cohesin is required for imprinting, we analysed the levels ofH19 and IGF2 transcripts in cohesin- and CTCF-depleted HeLa cells.In control cells, both H19 and IGF2 transcripts could be detected,although IGF2 only at low levels. After CTCF depletion, H19 tran-scripts were reduced, and IGF2 transcript levels were increased,implying that IGF2 and H19 are also imprinted in HeLa cells in aCTCF-dependent manner (Fig. 5a). Remarkably, SCC1 depletionhad a very similar effect. In contrast, neither CTCF nor SCC1 depletionaltered the transcript levels of glyceraldehyde-3-phosphate dehydro-genase (GAPDH, Fig. 5a). The changes in H19 and IGF2 transcriptionwere not caused by cell-cycle effects because we used G2-synchronizedcells for these experiments (Supplementary Fig. 9a, b). These datasuggest that cohesin is required for imprinting at the H19/IGF2 locus.Importantly, we observed that SCC1 and CTCF depletion altered H19and IGF2 transcript levels in a very similar way in cells that weresynchronized in the G1 phase (Fig. 5b, c). Because sister-chromatidcohesion does not exist in the G1 phase, the role of cohesin in con-trolling H19 and IGF2 transcription seems to be independent of cohe-sin’s function in cohesion (see also Supplementary Fig. 12).

Discussion

Because hypomorphic mutations in genes required for cohesioncan cause defects in chromatin organization, transcription anddevelopment, it has long been suspected that cohesin has functionsin addition to its role in sister-chromatid cohesion5–13. However,it has been difficult to exclude the possibility that these phenotypes

are indirect consequences of subtle or rare cohesion defects, and ithas remained unknown what the molecular basis of cohesion-independent cohesin functions could be. Our results indicate thatcohesin has an important role at CTCF-binding sites, which mayfunction as transcriptional insulators or boundary elements in verte-brate genomes. Hypomorphic mutations in cohesion genes maytherefore cause defects in insulators and boundaries that could leadto transcriptional and developmental abnormalities. Recent studiesusing a mouse model of the cohesinopathy CdLS have indeedrevealed transcriptional changes that are consistent with this possibi-lity (A. Lander, personal communication).

Our observation that cohesin depletion alters H19 and IGF2 tran-scription not only in the G2 phase but also in the G1 phase, wherecohesion does not exist, implies that some of the functions of cohesinare indeed independent of its role in cohesion. This notion is sup-ported by our finding that cohesin, like CTCF, is widely expressed inmammalian tissues, most of which are predominantly composed ofpostmitotic cells. However, this does not exclude the possibility thatsister-chromatid cohesion also affects transcription, which couldexplain why mutations in the putative cohesion establishment factorESCO2 can cause RBS/SC syndromes.

The hypothesis that cohesin has important transcriptional rolesduring the G1 phase and in postmitotic cells can also provide a poten-tial explanation for why vertebrate cells remove the bulk of cohesinfrom chromosome arms in prophase and reload it onto DNA intelophase3,20,21 instead of destroying most cohesin in metaphase, asoccurs in yeast43. Because the protease separase preferentially cleaveschromosome-bound cohesin21, the prophase pathway may sparecohesin from destruction in metaphase so that cohesin can performcohesion-independent functions immediately after mitosis.

In the future it will be interesting to understand how cohesin con-tributes to transcriptional insulation at the mechanistic level. Cohesincomplexes can physically connect two distinct DNA molecules whenthey mediate cohesion between sister chromatids4. It is therefore con-ceivable that cohesin can also physically connect different sites on oneDNA molecule, thereby creating DNA loops that could control enhan-cer–promoter interactions topologically44,45. Alternatively, cohesincould physically block the spreading of transcription factors, chro-matin-remodelling enzymes and heterochromatin proteins on DNA.

Cohesin has been highly conserved in eukaryotes and related com-plexes exist in bacteria4, but CTCF has only been described in verte-brates and Drosophila46. Cohesin might therefore have adopted aninsulator function late in evolution, or cohesin might perform insu-lator functions without CTCF in lower eukaryotes. Along these lines,it is conceivable that the main function of CTCF in mammaliangenomes is to define binding sites for cohesin, and that cohesin isthe molecule that structures DNA in a way that causes insulator andboundary effects.

METHODS SUMMARY

Chromatin immunoprecipitation was performed as described47.

The immunoprecipitated DNA (ChIP) and a control for the non-enriched

DNA (whole-cell extract, WCE) were amplified by in vitro transcription, labelled

by biotin and hybridized to high-density oligonucleotide arrays (Affymetrix) asdescribed in ref. 48. After scanning and data extraction, ChIP and WCE signals

for each of the tiling arrays were normalized by the model-based analysis of

tiling-arrays algorithm (MAT)49. The MAT score was calculated and mapped

to genomic positions in the human genome assembly human genome (Hg) 18

(NCBI Build 36), and significantly enriched regions were selected using the MAT

score. False detection rates, calculated by the MAT program, were less than 2%

for all experiments (Supplementary Fig. 13).

Full Methods and any associated references are available in the online version ofthe paper at www.nature.com/nature.

Received 30 November 2007; accepted 7 January 2008.Published online 30 January 2008.

1. Michaelis, C., Ciosk, R. & Nasmyth, K. Cohesins: chromosomal proteins thatprevent premature separation of sister chromatids. Cell 91, 35–45 (1997).

b

a

0

1

2

4

3

ControlSCC1

CTCF

IGF2

Fold

cha

nge

0

0.5

1.0

1.5

2.0

Control

SCC1CTCF

H19

0

0.5

1.0

1.5

2.0

Control

SCC1CTCF

GAPDH

Fold

cha

nge

0

0.5

1.0

1.5

2.0 H19

Control

SCC1CTCF

3.0

4.0

2.0

1.0

0

Control

SCC1CTCF

Fold

cha

nge

IGF2c

G1 G1None Control SCC1None

RNAi:Syn.:

0

200

400

0 400 800

600

0

200

400

0 400 800

600

0

200

400

0 400 800FL2-A

600

Cou

nts

0

200

400

0 400 800

600

G1CTCF

Figure 5 | Cohesin controls transcription at the H19/IGF2 locus.a, RT–qPCR analysis of H19, IGF2, GAPDH and actin transcripts inSCC1- and CTCF-depleted HeLa cells synchronized in G2 phase. Thetranscript levels were normalized first to control RNAi values and secondto actin levels (mean of n 5 3 for H19 and IGF2, n 5 2 for GAPDH; errorbars, 6 s.d.). b, Fluorescence-activated cell sorting (FACS) analysis of cellsthat were enriched in G1 phase by dTAR, and removal of mitotic cells byshake-off. Syn., synchronization; FL2-A, signal intensity of the propidiumiodine stain. c, RT–qPCR analysis as in a of cells synchronized in G1phase. Transcript levels were normalized to GAPDH (mean of n 5 3; errorbars, 6 s.d.).

ARTICLES NATURE | Vol 451 | 14 February 2008

800Nature Publishing Group©2008

2. Guacci, V., Koshland, D. & Strunnikov, A. A direct link between sister chromatidcohesion and chromosome condensation revealed through the analysis of MCD1in S. cerevisiae. Cell 91, 47–57 (1997).

3. Losada, A., Hirano, M. & Hirano, T. Identification of Xenopus SMC proteincomplexes required for sister chromatid cohesion. Genes Dev. 12, 1986–1997(1998).

4. Nasmyth, K. & Haering, C. H. The structure and function of SMC and kleisincomplexes. Annu. Rev. Biochem. 74, 595–648 (2005).

5. Donze, D., Adams, C. R., Rine, J. & Kamakaka, R. T. The boundaries of the silencedHMR domain in Saccharomyces cerevisiae. Genes Dev. 13, 698–708 (1999).

6. Rollins, R. A., Morcillo, P. & Dorsett, D. Nipped-B, a Drosophila homologue ofchromosomal adherins, participates in activation by remote enhancers in the cutand Ultrabithorax genes. Genetics 152, 577–593 (1999).

7. Verni, F., Gandhi, R., Goldberg, M. L. & Gatti, M. Genetic and molecular analysis ofwings apart-like (wapl), a gene controlling heterochromatin organization inDrosophila melanogaster. Genetics 154, 1693–1710 (2000).

8. Benard, C. Y., Kebir, H., Takagi, S. & Hekimi, S. mau-2 acts cell-autonomously toguide axonal migrations in Caenorhabditis elegans. Development 131, 5947–5958(2004).

9. Rollins, R. A., Korom, M., Aulner, N., Martens, A. & Dorsett, D. Drosophila nipped-Bprotein supports sister chromatid cohesion and opposes the stromalin/Scc3cohesion factor to facilitate long-range activation of the cut gene. Mol. Cell. Biol.24, 3100–3111 (2004).

10. Dorsett, D. et al. Effects of sister chromatid cohesion proteins on cut geneexpression during wing development in Drosophila. Development 132, 4743–4753(2005).

11. Dorsett, D. Roles of the sister chromatid cohesion apparatus in gene expression,development, and human syndromes. Chromosoma 116, 1–13 (2007).

12. Horsfield, J. A. et al. Cohesin-dependent regulation of Runx genes. Development134, 2639–2649 (2007).

13. Zhang, B. et al. Mice lacking sister chromatid cohesion protein PDS5B exhibitdevelopmental abnormalities reminiscent of Cornelia de Lange syndrome.Development 134, 3191–3201 (2007).

14. Musio, A. et al. X-linked Cornelia de Lange syndrome owing to SMC1L1 mutations.Nature Genet. 38, 528–530 (2006).

15. Krantz, I. D. et al. Cornelia de Lange syndrome is caused by mutations in NIPBL, thehuman homolog of Drosophila melanogaster Nipped-B. Nature Genet. 36, 631–635(2004).

16. Tonkin, E. T., Wang, T. J., Lisgo, S., Bamshad, M. J. & Strachan, T. NIPBL, encoding ahomolog of fungal Scc2-type sister chromatid cohesion proteins and fly Nipped-B,is mutated in Cornelia de Lange syndrome. Nature Genet. 36, 636–641 (2004).

17. Deardorff, M. A. et al. Mutations in cohesin complex members SMC3 and SMC1Acause a mild variant of cornelia de Lange syndrome with predominant mentalretardation. Am. J. Hum. Genet. 80, 485–494 (2007).

18. Vega, H. et al. Roberts syndrome is caused by mutations in ESCO2, a humanhomolog of yeast ECO1 that is essential for the establishment of sister chromatidcohesion. Nature Genet. 37, 468–470 (2005).

19. Schule, B., Oviedo, A., Johnston, K., Pai, S. & Francke, U. Inactivating mutations inESCO2 cause SC phocomelia and Roberts syndrome: no phenotype-genotypecorrelation. Am. J. Hum. Genet. 77, 1117–1128 (2005).

20. Sumara, I., Vorlaufer, E., Gieffers, C., Peters, B. H. & Peters, J. M. Characterizationof vertebrate cohesin complexes and their regulation in prophase. J. Cell Biol. 151,749–762 (2000).

21. Waizenegger, I. C., Hauf, S., Meinke, A. & Peters, J. M. Two distinct pathwaysremove mammalian cohesin from chromosome arms in prophase and fromcentromeres in anaphase. Cell 103, 399–410 (2000).

22. The ENCODE Project Consortium. The ENCODE (ENCyclopedia Of DNAElements) project. Science 306, 636–640 (2004).

23. Lengronne, A. et al. Cohesin relocation from sites of chromosomal loading toplaces of convergent transcription. Nature 430, 573–578 (2004).

24. Glynn, E. F. et al. Genome-wide mapping of the cohesin complex in the yeastSaccharomyces cerevisiae. PLoS Biol. 2, E259 (2004).

25. Lindroos, H. B. et al. Chromosomal association of the Smc5/6 complex revealsthat it functions in differently regulated pathways. Mol. Cell 22, 755–767 (2006).

26. Lobanenkov, V. V. et al. A novel sequence-specific DNA binding protein whichinteracts with three regularly spaced direct repeats of the CCCTC-motif in the 59-flanking sequence of the chicken c-myc gene. Oncogene 5, 1743–1753 (1990).

27. Kim, T. H. et al. Analysis of the vertebrate insulator protein CTCF-binding sites inthe human genome. Cell 128, 1231–1245 (2007).

28. Barski, A. et al. High-resolution profiling of histone methylations in the humangenome. Cell 129, 823–837 (2007).

29. Xie, X. et al. Systematic discovery of regulatory motifs in conserved regions of thehuman genome, including thousands of CTCF insulator sites. Proc. Natl Acad. Sci.USA 104, 7145–7150 (2007).

30. Burke, L. J. et al. CTCF binding and higher order chromatin structure of the H19locus are maintained in mitotic chromatin. EMBO J. 24, 3291–3300 (2005).

31. Watrin, E. et al. Human Scc4 is required for cohesin binding to chromatin, sister-chromatid cohesion, and mitotic progression. Curr. Biol. 16, 863–874 (2006).

32. Bell, A. C., West, A. G. & Felsenfeld, G. The protein CTCF is required for theenhancer blocking activity of vertebrate insulators. Cell 98, 387–396 (1999).

33. Bell, A. C. & Felsenfeld, G. Methylation of a CTCF-dependent boundary controlsimprinted expression of the Igf2 gene. Nature 405, 482–485 (2000).

34. Hark, A. T. et al. CTCF mediates methylation-sensitive enhancer-blocking activityat the H19/Igf2 locus. Nature 405, 486–489 (2000).

35. Ishihara, K., Oshimura, M. & Nakao, M. CTCF-dependent chromatin insulator islinked to epigenetic remodeling. Mol. Cell 23, 733–742 (2006).

36. Schmitz, J., Watrin, E., Lenart, P., Mechtler, K. & Peters, J. M. Sororin is required forstable binding of cohesin to chromatin and for sister chromatid cohesion ininterphase. Curr. Biol. 17, 630–636 (2007).

37. Chung, J. H., Whiteley, M. & Felsenfeld, G. A 59 element of the chicken beta-globindomain serves as an insulator in human erythroid cells and protects againstposition effect in Drosophila. Cell 74, 505–514 (1993).

38. Recillas-Targa, F. et al. Position-effect protection and enhancer blocking by thechicken beta-globin insulator are separable activities. Proc. Natl Acad. Sci. USA 99,6883–6888 (2002).

39. Yahata, K. et al. cHS4 insulator-mediated alleviation of promoter interferenceduring cell based expression of tandemly associated transgenes. J. Mol. Biol. 374,580–590 (2007).

40. Bartolomei, M. S. & Tilghman, S. M. Genomic imprinting in mammals. Annu. Rev.Genet. 31, 493–525 (1997).

41. Kugoh, H. et al. Mouse A9 cells containing single human chromosomes foranalysis of genomic imprinting. DNA Res. 6, 165–172 (1999).

42. Lee, M. P., Hu, R. J., Johnson, L. A. & Feinberg, A. P. Human KVLQT1 gene showstissue-specific imprinting and encompasses Beckwith-Wiedemann syndromechromosomal rearrangements. Nature Genet. 15, 181–185 (1997).

43. Uhlmann, F., Wernic, D., Poupart, M. A., Koonin, E. V. & Nasmyth, K. Cleavage ofcohesin by the CD clan protease separin triggers anaphase in yeast. Cell 103,375–386 (2000).

44. Splinter, E. et al. CTCF mediates long-range chromatin looping and local histonemodification in the beta-globin locus. Genes Dev. 20, 2349–2354 (2006).

45. Kurukuti, S. et al. CTCF binding at the H19 imprinting control region mediatesmaternally inherited higher-order chromatin conformation to restrict enhanceraccess to Igf2. Proc. Natl Acad. Sci. USA 103, 10684–10689 (2006).

46. Moon, H. et al. CTCF is conserved from Drosophila to humans and confersenhancer blocking of the Fab-8 insulator. EMBO Rep. 6, 165–170 (2005).

47. Kim, T. H. et al. A high-resolution map of active promoters in the human genome.Nature 436, 876–880 (2005).

48. Liu, C. L., Schreiber, S. L. & Bernstein, B. E. Development and validation of a T7based linear amplification for genomic DNA. BMC Genomics 4, 19 (2003).

49. Johnson, W. E. et al. Model-based analysis of tiling-arrays for ChIP-chip. Proc. NatlAcad. Sci. USA 103, 12457–12462 (2006).

50. Bailey, T. L. & Elkan, C. Fitting a mixture model by expectation maximization todiscover motifs in biopolymers. Proc. Int. Conf. Intell. Syst. Mol. Biol. 2, 28–36(1994).

Supplementary Information is linked to the online version of the paper atwww.nature.com/nature.

Acknowledgements We thank K. Nakagawa, A. Watanabe and Y. Hayakawa forassistance, M. Oshimura for providing cell lines, and D. Barlow, B. Dickson,N. Galjart, A. Lander, M. Merkenschlager, C. Meyer, T. Taniguchi and members ofthe Peters and Shirahige laboratories for discussions. K.S.W. was supported by aresearch fellowship of the German Research Foundation (DFG). K.S. and H.A. weresupported by a grant of the Genome Network Project and Grant-in-Aid forScientific Research (S) from the MEXT, Japan. F.I. was supported in by InvitrogenCorporation and Grant-in-Aid for Scientific Research from the Ministry ofEconomy, Trade and Industry, Japan. K.M. and N.I. were supported by a MEXTgrant-in-aid and RIKEN institute program of Bioarchitect. Research in thelaboratory of J.-M.P. is supported by Boehringer Ingelheim, the 6th FrameworkProgram of the European Union via the Integrated Project MitoCheck, the AustrianResearch Promotion Agency, and the Austrian Science Fund via the EuroDYNAProgram of the European Science Foundation.

Author Contributions Experiments were designed and data interpreted by K.S.W.,K. Yoshida., T.I., K.S. and J.-M.P. K.S.W. performed SMC3 and SA2 ChIP-chip andChIP-qPCR, and analysed the role of cohesin at the H19 ICR. K. Yoshida performedSCC1 and CTCF ChIP-qPCR, ChIP-chip on ENCODE and whole-genome arrays, andre-ChIP. T.I. carried out bioinformatic analyses. M.B. performed RNAi, chromatinfractionation and transcriptome experiments. B.K. analysed cohesin expression inmouse tissues, carried out CTCF localization by IFM and performed cohesin/CTCFRNAi-qIFM experiments. E.S. characterized mouse–human hybrid cell lines andthe binding of CTCF to mitotic chromatin. K.M. and N.I. analysed the effect ofcohesin RNAi on chicken HS4 function using constructs provided by F.I. and K.Yahata. S.T., G.N., H.A., K.I., T.M. and M.N. prepared the initial genome-wide CTCFmap. J.-M.P., K.S.W. and K.S. wrote the manuscript.

Author Information Microarray data presented in this article have been depositedin the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo) underthe accession number GSE9613. Reprints and permissions information is availableat www.nature.com/reprints. Correspondence and requests for materials shouldbe addressed to K.S. (kshirahi@bio.titech.ac.jp) or J.-M.P.(peters@imp.univie.ac.at).

NATURE | Vol 451 | 14 February 2008 ARTICLES

801Nature Publishing Group©2008

METHODSAntibodies. Primary antibodies used were rabbit anti-CTCF (Abcam, 1:50 for

IFM), mouse anti-CTCF (BD, for IFM and immunoblotting), rabbit anti-CTCF

(Upstate, for ChIP), rabbit anti-SCC1 (Abcam, for ChIP), rabbit anti-H3S10ph

(Upstate), mouse anti-tubulin (Sigma), mouse anti-Orc2 (MBL), mouse

mAb414 for nucleoporins (Covance, 1:2,000 for immunoblotting), rabbit

anti-emerin (a gift from K. Wilson, 1:2,000 for immunoblotting) and rabbit

anti-Ki67 (a gift from M. Takagi, 1:500 for IFM). Polyclonal rabbit antibodies

against SCC1 (number 623), SMC3 (number 727) and SA2 (number 446; all for

IFM, immunoblotting and ChIP) have been described in refs 20 and 21.

Antibodies against SCC4 (number 294) have been described in ref. 31.

Cell culture. HeLa cells were cultured in DMEM (Invitrogen) supplemented

with 0.2 mM L-glutamine, 100 units ml21 penicillin, 100mg ml21 streptomycin

and 10% FCS. The hTERT RPE-1 cells were cultured in the same medium,

supplemented with non-essential amino acids. The mouse/human hybrid cells

A911P-1 and A911M-241 were cultured in the same medium as HeLa cells,

supplemented with blasticidine S (3mg ml21) to select for cells containing

human chromosome 11. HeLa CENP-A–eGFP and SMC1–eGFP cells have been

described in refs 51 and 52, respectively. Human B cells were obtained from

a healthy male donor and were immortalized by transformation with Ebstein

Barr virus.

Cell cycle synchronization. HeLa cells were synchronized by dTAR (14–16 h in

the presence of 2 mM thymidine, 8 h release, 16 h in the presence of 2 mM

thymidine) and harvested 6 h and 13 h after the second release for enrichment

in the G2 phase and in the G1 phase, respectively. For enrichment in prometa-

phase of mitosis, nocodazole (330 nM) was added 8 h after release from the

second thymidine block for 1 or 2 h, and cells were harvested by shake off.

RNA interference. The siRNA oligonucleotides were annealed according to

manufacturer’s instruction and used at a final concentration of 100–200 nM.

The siRNA transfections were performed using oligofectamine (Invitrogen).

Oligonucleotides targeting the firefly luciferase GL2 were used as controls. All

siRNA sequences are listed in Supplementary Table 5.

Subcellular fractionation. Whole-cell extracts of HeLa and hTERT RPE-1 cells

were separated into soluble supernatant and chromatin-containing pellet frac-

tions as described31.

Immunofluorescence microscopy and fluorescence in situ hybridization.HeLa cells grown on 18-mm coverslips were fixed with 4% PFA and stained

for SCC1 and CTCF. In some experiments, soluble proteins were removed before

fixation by pre-extracting cells for 3 min with 0.1% Triton X100 in PBS. Mitotic

cells that had been harvested by mitotic shake off were spun onto microscope

slides using a Cytospin centrifuge (Shandon). Images were taken on an Axioplan

2 microscope with a 633 PlanApochromat NA 1.4 objective (Zeiss) using a

CoolSnap HQ camera (Photometrics) and were processed with MetaMorph

(Universal Imaging).

For qIFM of chromatin-bound proteins, HeLa cells were transfected with

SCC1, CTCF and SCC2 siRNAs. In parallel, HeLa cells stably expressing CENP-

A–eGFP as a marker protein were transfected with control siRNA. One day after

transfection, cells were trypsinized, mixed in a 1:1 ratio, and seeded onto cover-

slips. After another 24 h, the cells were pre-extracted, fixed and stained for SCC1

and CTCF as above. Fluorescence intensities were quantified as described52.

Chromosome FISH was performed with human chromosome 11 ‘paint’

(Cambio) according to the manufacturer’s instructions.

Chromatin immunoprecipitation. Cells at 70–80% confluency were crosslinked

with 1% formaldehyde for 10 min and, after quenching with 125 mM glycine,

were prepared for ChIP as described53. ChIP was performed as described47 using

SCC1, SMC3, CTCF, SA2 and control antibodies. In brief, crosslinked cell lysates

were incubated with the antibodies for 14 h at 4 uC. After this, the lysates were

incubated for 2 h at 4 uC with pre-absorbed protein A Affiprep beads (Bio-Rad).

The beads were washed several times and eluted with elution buffer (50 mM Tris,

10 mM EDTA, 1% SDS) for 20 min at 65 uC. The eluates were treated with

proteinase K for 1 h at 37 uC and then incubated at 65 uC overnight to reverse

the crosslinks. Contaminating RNA was removed by RNase treatment. The

samples were further purified by phenol-chloroform extraction and one addi-

tional purification step using a PCR purification kit (Qiagen). The samples were

eluted from the columns with 80 ml water.

ChIP-qPCR. ChIP samples (2ml) were used for one 25 ml PCR reaction. Analyses

by qPCR were performed using a Platinum SYBR Green qPCR SuperMix UDG

(Invitrogen) on a MJ research lightcycler or on an ABI 9500 cycler. The results

were presented as fold-enrichment over control ChIP or as the percentage of

input-chromatin that was precipitated. The Re-ChIP assay was performed as

previously described54.

ChIP microarray analyses. ChIP samples were amplified by in vitro transcrip-

tion, labelled with biotin and hybridized to high-density oligonucleotide tiling

arrays (Affymetrix) as described48. A sample of DNA prepared from WCE was

prepared in the same way. ChIP and WCE samples were hybridized on arrays

according to the manufacturer’s instructions. After scanning and data extrac-

tion, enrichment values (ChIP/WCE) were calculated by using the MAT algo-

rithm49. This algorithm reduces problems that can be caused by probe-specific

biases, such as differential sequence copy numbers in the genome or variable

melting temperatures dependent on GC content. These problems can otherwise

confound high-density oligonucleotide tiling-array analyses in higher eukar-

yotes. The MAT algorithm reduces these problems by estimating probe affinity

from probe sequence and genome copy number, and can therefore eliminate

data artefacts better than the commonly used quantile normalization method.

The MAT algorithm therefore provides a powerful tool for finding enriched

regions in ChIP-chip data. The resulting MAT scores are proportional to the

logarithm of the fold-enrichment of the ChIP-chip sample49.

We mapped MAT scores to positions in human genome assembly Hg 18

(NCBI Build 36). Bandwidth, MaxGap and MinProbe parameters were set to

250, 1,000 and 12, respectively. Cutoff threshold P-values were set to 1.0 3 10210,

1.0 3 1028 and 1.0 3 1027.5 for ENCODE 1.0, ENCODE 2.0 and Human Tiling

1.0R arrays, respectively. These P-values were equivalent to MAT scores higher

than 4.85. False-discovery rates were also calculated by the MAT program. For all

experiments, the false-detection rate was less than 2% (Supplementary Fig. 13).

Transcriptome microarray analyses. For genome-wide transcription analysis,

the Affymetrix Human Genome U133 Plus 2.0 Array was used. To minimize the

possibility of artefacts by RNAi treatment, all experiments were performed with

two different siRNAs for the depletion of SCC1 and CTCF. Chip data were

analysed by MAS software provided by Affymetrix, using default settings.

Only Ensembl genes for which transcripts showed the same behaviour after

transfection with both siRNAs were considered. To analyse whether genes close

to cohesin sites have a significantly higher tendency to be upregulated or down-

regulated after SCC1 and CTCF depletion, genes in close proximity to cohesin-

binding sites were counted. The ratio of number of increased transcripts/number

of decreased transcripts in relation to the distance from the binding sites was

calculated (Fig. 2e).

Luciferase H19 ICR reporter assays. The reporter plasmids pIHLIE and

pIHLME, originally based on a pGL3 luciferase plasmid (Promega), and the

luciferase reporter assay have been described35. To test if cohesin can bind to

pIHLIE, HeLa cells were transiently transfected with this plasmid using lipofec-

tamine (Invitrogen) and harvested for ChIP 24 h later. ChIP was performed with

SMC3, CTCF and control antibodies. The ChIP samples were analysed by PCR

for the presence of the mouse H19 ICR (present on pIHLIE) and, as a positive

control, the endogenous human H19 ICR. The PCR primers are listed in

Supplementary Table 5.

To perform insulator assays, HeLa cells were transfected with cohesin, CTCF

or control siRNA oligonucleotides using lipofectamine RNAiMAX (Invitrogen)

and synchronized as indicated in Supplementary Fig. 9a. During the release

from the first thymidine arrest the firefly luciferase reporter plasmid and a

Renilla luciferase control plasmid were transfected using lipofectamine

(Invitrogen). After harvesting and lysis of the cells, the activities of both luci-

ferases were detected using a Dual Glo kit (Promega) and a Synergy 2 reader

(BioTek). To control for the transfection efficiency, the firefly luciferase activity

was normalized against the Renilla luciferase activity. All samples were per-

formed in triplicate.

Chicken HS4 insulator reporter assays. Plasmids containing two different

cDNA expression cassettes with or without cHS4 insulators were constructed

using Multi-site Gateway technology (Invitrogen). Plasmid constructions,

isolation of stable HeLa cell lines, chicken HS4(–)-1 and chicken HS4(1)-3,

siRNA transfection, IFM and chromatin insulator assays were carried out as

described39,55. ChIP of the integrated cHS4 loci was performed using three sets

of primers described in ref. 39. Cells were analysed 48 h and 60 h after siRNA

transfection by IFM and immunoblotting.

Transcript analysis by qPCR. HeLa cells were transfected with siRNA for SCC1

and CTCF for 48 h in combination with cell-cycle synchronization by dTAR

(Supplementary Fig. 9a). Cells were released from the second thymidine arrest

for 6 h and 13 h to enrich cells in G2 and G1 phase, respectively. In addition,

mitotic cells were removed by shake off before the adherent cells were harvested.

Synchronization in the G1 and G2 phases was controlled by FACS. For this

purpose, samples of cells were fixed with methanol, washed with PBS, stained

for DNA with propidium iodine, and analysed using a FACSCalibur flow cyto-

meter and Cell Quest (BD Biosciences). The histogram plots were generated

using FlowJo (Three Star). From the remaining cells, RNA was prepared using

TRIZOL (Invitrogen) according to manufacturer’s instructions. Contaminating

DNA was removed by DNase treatment followed by phenol-chloroform extrac-

tion and isopropanol precipitation. The cDNA was generated by reverse tran-

scription using an oligo-dT primer and SuperscriptII (Invitrogen) according to

doi:10.1038/nature06634

Nature Publishing Group©2008

the manufacturer’s instructions. The amount of the different transcripts wasthen compared by qPCR using the Platinum SYBR green kit (Invitrogen) on a

MJ research lightcycler and primers specific for H19, IGF2, GAPDH and actin

(Supplementary Table 5). The actin signal was then used for the normalization of

the samples, and the data are presented as fold change compared to the control

RNAi.

51. Gerlich, D., Koch, B., Dupeux, F., Peters, J. M. & Ellenberg, J. Live-cell imagingreveals a stable cohesin–chromatin interaction after but not before DNAreplication. Curr. Biol. 16, 1571–1578 (2006).

52. Kueng, S. et al. Wapl controls the dynamic association of cohesin with chromatin.Cell 127, 955–967 (2006).

53. Peters, A. H. et al. Partitioning and plasticity of repressive histone methylationstates in mammalian chromatin. Mol. Cell 12, 1577–1589 (2003).

54. Ju, B. G. et al. A topoisomerase IIb-mediated dsDNA break required for regulatedtranscription. Science 312, 1798–1802 (2006).

55. Maeshima, K. et al. Cell-cycle-dependent dynamics of nuclear pores: pore-freeislands and lamins. J. Cell Sci. 119, 4442–4451 (2006).

doi:10.1038/nature06634

Nature Publishing Group©2008