Conserved, developmentally regulated mechanismcouples chromosomal looping and heterochromatinbarrier activity at the homeobox gene A locusYoon Jung Kim, Katharine R. Cecchini, and Tae Hoon Kim1

Department of Genetics, Yale University School of Medicine, 333 Cedar Street, New Haven, CT 06520-8005

Edited by Mark T. Groudine, Fred Hutchinson Cancer Research Center, Seattle, WA, and approved March 22, 2011 (received for review December 6, 2010)

Establishment and segregation of distinct chromatin domains areessential for proper genome function. The insulator protein CCCTC-binding factor (CTCF) is involved in creating boundaries that seg-regate chromatin and functional domains and in organizing high-er-order chromatin structures by promoting chromosomal loopsacross the vertebrate genome. Here, we investigate the insulationproperties of CTCF at the human and mouse homeobox gene A(HOXA) loci. Although cohesin loading at the CTCF binding site isrequired for looping, we found that cohesin is dispensable for chro-matin barrier activity at that site. Using mouse embryonic stemcells in both a pluripotent and differentiated neuronal progenitorstate, we determined that embryonic stem cell pluripotency factorOCT4 antagonizes cohesin loading at the CTCF binding site. Loss ofOCT4 in the committed and differentiated neuronal progenitorcells results in loading of cohesin and chromosome looping, whichcontributes to heterochromatin partitioning and selective gene ac-tivation across theHOXA locus. Our analysis reveals that chromatinbarrier activity of CTCF is evolutionarily conserved and is respon-sible for the coordinated establishment of chromatin structure,higher-order architecture, and developmental expression of theHOXA locus.

The establishment and segregation of different chromatin do-mains (heterochromatin and euchromatin) and maintenance

of these structures are essential for proper genome function.Segregation of euchromatin and heterochromatin structures inthe genome is mediated by noncoding DNA elements knownas insulators (1), which are bound by CCCTC-binding factor(CTCF). CTCF binding sites (CBSs) across the genome definethe boundaries of silent heterochromatin domains exhibiting tri-methylated lysine residue 27 on histone H3 (H3K27me3) (2, 3).Potential CTCF cofactors have been identified (4–8); amongthese, suppressor of zeste 12 (SUZ12), a component of the poly-comb repressive complex (PRC), seems highly relevant to theestablishment of heterochromatin barrier activity (9).

CTCF has also been implicated in mediating long-range inter-and intrachromosomal interactions (10–12). DNA looping hasbeen identified as a potential mechanism of transcriptional insu-lation at the apolipoprotein cluster and interferon gamma locus(13, 14). At these loci and numerous other sites in the genome(15), CTCF partners with cohesin to mediate looping, enhancer-blocking, and establishment of barrier sites; although cohesin hasbeen implicated as a regulator in these studies, both CTCF andcohesin are present at these sites. In fact, cohesin is present at asignificant majority of CTCF sites, and vice versa, making the ex-act role of each protein in barrier activity or the formation ofchromatin loops unclear. In Drosophila, cohesin was recentlyidentified as a member of the trithorax group (TRX), a familyof transcriptional regulators that generally promote gene expres-sion through chromatin remodeling (16). TRX proteins were dis-covered because of their ability to regulate the homeobox (HOX)genes, where they were usually associated with establishmentof domains characterized by trimethylated lysine 4 on histoneH3 (H3K4me3) (17). Coordination of CTCF and cohesin in

establishing expression patterns and chromatin domains at HOXgenes is obviously significant to development.

The HOX genes are arranged into four clusters (HOXA, B, C,andD) in vertebrates. Each locus spans several hundred thousandbase pairs of exceptionally high evolutionary conservation in bothcoding and noncoding intergenic regions. TheHOX genes encodemaster transcription factors responsible for patterning the ante-rior-posterior body axis during development. Colinearity betweenthe order of genes in the cluster and their expression patternalong the anterior–posterior body axis has been observed acrossthe animal phyla (18). In human lung fibroblasts, the upstreamhalf of the HOXA locus lacks H3K27 trimethylation as well asassociation with repressive PRC component SUZ12, whereasthe downstream half of the locus exhibits extensive H3K27 tri-methylation and SUZ12 association (9). In foot fibroblasts, thispattern is completely reversed (19). In either tissue location, theabsence of heterochromatin is marked by extended deposition ofdimethylation (H3K4me2) and trimethylation (H3K4me3) of thelysine residue 4 of histone H3 at the promoters of the HOXAgenes. The developmentally crucial HOXA locus demonstrateswell-delineated chromatin domains, and we have utilized thisproperty of the locus to elucidate the specific function of CTCFand cohesin in the establishment of chromatin domains, chromo-somal looping, and gene expression across a large genomic region.

Unlike normal fibroblast cells, embryonic stem cells have abivalent chromatin structure composed of both euchromatin andheterochromatin, exhibiting no segregation in the HOXA locus(20, 21). The mechanistic basis of this cell-type specificity in chro-matin structure is unclear. According to a recent study, inducedpluripotent stem cells can be generated by expressing four repro-gramming factors: octamer binding factor 4 (OCT3/4), sex deter-mining region Y-box 2 (SOX2), avian myelocytomatosis viraloncogene homolog (MYC), and Krueppel-like factor 4 (KLF4)(22). Among these factors, only OCT4 has been shown to interactwith cohesin at enhancer and promoter sites to activate gene ex-pression (23) and also interact with CTCF in X chromosome in-activation (24). Therefore, we hypothesized that OCT4 might beresponsible for the bivalent chromatin structure observed in em-bryonic stem cells (ES cells). In this study, we use normal humanlung fibroblasts (IMR90 cells) as well as mouse ES cells, whichcan further be differentiated into neural progenitor cells (NPC).Using this system, we are able to discriminate the specific func-tions of CTCFand cohesin in maintenance of chromatin domainsand secondary higher-order structure. Through the use of ES

Author contributions: Y.J.K. and T.H.K. designed research; Y.J.K. and K.R.C. performedresearch; Y.J.K. and K.R.C. contributed new reagents/analytic tools; Y.J.K., K.R.C., andT.H.K. analyzed data; and Y.J.K., K.R.C., and T.H.K. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. E-mail: taehoon.kim@yale.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1018279108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1018279108 PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7391–7396

BIOCH

EMISTR

Y

cells, we are able to further connect the activities of both CTCFand cohesin to the role of pluripotency factor OCT4, which wedemonstrate can disrupt chromatin looping, an activity which hasnot been previously described for this factor. This model systemallows us to offer evidence that this integrated mechanism ofchromatin domain formation and characterization is both con-served and developmentally regulated.

ResultsCTCF Binding Site 5 (CBS5) Is a Conserved Mammalian ChromatinBarrier. We used available chromatin immunoprecipitationcoupled to deep sequencing (ChIP-Seq) data (21, 25, 26) to in-vestigate the chromatin structure and organization of the HOXAlocus and to gain insight into the role of CTCF in establishinglarge-scale chromatin and expression patterns. We had previouslyidentified seven CTCF binding sites at the HOXA locus in theIMR90 cell line (27). One of the seven CTCF binding sites, CBS5,demarcates an extended H3K27me3 domain (Fig. 1A andFig. S1) that encompasses all downstream HOX genes (HOXA9-13). In the absence of heterochromatin, HOXA genes upstreamof CBS5 (HOXA1-7) are marked by dimethylation and trimethy-lation of histone H3K4 (H3K4me2, H3K4me3) at the promotersand trimethylation of histone H3K36 (H3K36me3) at the genebodies, indicating expression of these genes in IMR90 cells(Fig. S2). Additionally, examination of genomic alignments (28)across vertebrates reveals that conservation of the sequence ofCBS5 is limited largely to mammals, from human and mouseto platypus (Fig. 1 B and C). Therefore, at this locus, heterochro-matin and euchromatin domains are arranged around a con-served CBS, suggesting potential developmentally regulatedsegregation of chromatin domains.

To confirm the chromatin barrier function of the CBS5 DNAelement, we used a previously developed reporter system contain-ing 256 Lac operator (LacO) sites (29). This LacO/LacI reportersystem was recently utilized to demonstrate that tethering ofthe reporter to the nuclear periphery results in silencing of thereporter gene by de novo heterochromatin deposition and DNAmethylation (30). We cloned CBS5 between the LacO sites and theEGFP reporter gene. We generated single copy stably integratedclones of HEK293 cells that contain either the control or theCBS5-containing reporter. Using these stably integrated clones, wetransfected LacI-Emerin (EMD) fusion protein, previously shownto repress the reporter by localizing it to the nuclear periphery(30). As expected, the control reporter was efficiently repressedby expression of LacI-EMD, whereas CBS5 preserved EGFP re-porter expression (Fig. 1D). This result suggests that CBS5 canfunction as a barrier element in a synthetic reporter system, likelyby inhibiting the intrusion of heterochromatin.

Barrier Site CBS5 Is Developmentally Regulated. We further exam-ined the available ChIP-Seq data for evidence that establishmentof this barrier at theHOXA locus is appropriately developmentallyregulated. Both mouse and human embryonic stem (mES andhES, respectively) cells exhibit bivalent chromatin, composed ofboth euchromatin (H3K4me3) and heterochromatin (H3K27me3)marks across the entire HOXA locus (Fig. 1A and Fig. S1), as pre-viously demonstrated (20, 21). Upon differentiation of mouse EScells to NPC, a complete reorganization of the HOXA chromatinstructure is observed, with CBS5 now segregating the large ex-tended H3K27me3 heterochromatin structure comparable to thatseen in human lung fibroblasts, providing direct evidence thatthese chromatin domains are developmentally regulated.

CTCF Is Required for Barrier Activity at CBS5. Recent studies havefound that cohesin colocalizes at a majority of CTCF binding sitesin the human and mouse genomes (15, 31). We first investigatedthe individual role of CTCF in chromatin barrier activity at CBS5by reducing the levels of CTCF using lentiviral vectors expressing

specific shRNA. We verified CTCF binding at CBS5 using ChIPanalyzed by quantitative real-time PCR (ChIP-qPCR, Table S1),and found that both CTCFand radiation repair protein 21 homog(RAD21, cohesin subunit) bound this site in IMR90 humanfibroblasts (Fig. 2A, Left). Upon CTCF knockdown using the len-tivirus-delivered shRNA in IMR90 human fibroblasts, CTCFbinding at CBS5 was reduced to background levels and RAD21binding at CBS5 was also reduced (Fig. 2A, Right). Our ChIP-qPCR experiments confirm the heterochromatin (H3K27me3)domain was bounded by CBS5 in IMR90 cells as observed inthe ChIP-Seq data. Upon CTCF knockdown, H3K27me3 levelsbetween CBS4 and CBS5 significantly increased (Fig. 2B), indi-cating an upstream spread of heterochromatin. We also analyzedexpression of HOXA genes under these conditions by reversetranscription quantitative real-time PCR (RT-qPCR, Table S2).Normally, genes downstream of CBS5 are repressed; after knock-down of CTCF, these genes (HOXA9–10) significantly increased,while upstream genes (HOXA6 and 7) decreased in expression(Fig. 2C). These results suggest a loss of chromatin integrity atCBS5 as a result of loss of CTCF and, indirectly, cohesin.

A

B

C

E

D

Fig. 1. CBS5 functions as a chromatin barrier in the HOXA locus. A repre-sents the HOXA gene expression status and CTCF binding sites in ES anddifferentiated cells. B is the result of CBS5 sequence alignment; the redbox is the core 20 bp of CBS5 and the blue box is a candidate site forOCT4 binding in mouse HOXA locus. C is a phylogenetic tree indicating con-servation of the CBS5 sequence primarily in mammals. D is the percentageof EGFP expression levels normalized to the gamma actin (ACTG1) geneexpression, determined by RT-qPCR, upon expression of LacI or LacI-EMD pro-teins, for cells containing either a control reporter or a reporter containingthe CBS5 element between LacO and the EGFP gene. Schematics of the CBS5and control reporters are shown in E.

7392 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1018279108 Kim et al.

Cohesin Is Required for Appropriate Gene Expression, but Not BarrierActivity. Because loss of CTCF binding resulted in concomitantloss of RAD21 binding at CBS5, we knocked down RAD21 inIMR90 lung fibroblasts using the same lentiviral shRNA system.Accordingly, RAD21 was no longer bound at CBS5, but CTCFbinding at this site was not perturbed, confirming previous obser-vations that CTCF binding is independent of the presence ofcohesin (15, 31) (Fig. 3A). ChIP-qPCR analysis of H3K27me3indicated no significant change in the heterochromatin at andaround CBS5 (Fig. 3B), indicating that barrier activity remainedintact. Finally, mRNA levels of the euchromatin HOXA genesadjacent to CBS5 (HOXA6 and 7) were uniformly decreasedafter cohesin knockdown, whereas HOXA9 and HOXA10 genesresiding within heterochromatin were not significantly affected(Fig. 3C). Our results indicate that cohesin is not required forthe maintenance of heterochromatin barrier sites, but is neces-sary for appropriate gene expression at this locus.

CBS5 Is Directly Involved in Forming Heterochromatin Loops. Previousstudies have implicated that CTCF can mediate formation of longextended loops that can segregate distinct chromatin environ-ments (11). We investigated whether detectable looping orga-nized at the CBS5 barrier site is present across the HOXA locususing chromosome conformation capture (3C) assays (32–34).We designed a series of primers positioned at all unique EcoRIrestriction digestion sites across the entire locus (Fig. S3 and

Table S3). Using an anchor primer located at CBS5, we per-formed qPCR assays to determine the proximity of different re-gions of the HOXA locus to CBS5 in IMR90 cells. We identifiedtwo sites along the HOXA locus that interact with CBS5. The ma-jor interaction site is located directly over the heterochromatinboundary downstream of the EVX1 gene (p40) (Fig. 4A). Addi-tionally, we detected a weak interaction upstream near the endof the euchromatin boundary of theHOXA locus. We further con-firmed these results by performing 3C using a different restrictionenzyme (MseI) (Fig. S4). The higher-order architecture of thislocus in IMR90 cells appears to be one upstream looped regioncomposed of euchromatin and one downstream looped region ofheterochromatin, organized around CBS5.

Both CTCF and Cohesin Are Required for Heterochromatin LoopFormation. To determine the specific roles of CTCF and cohesinin formation of the loops at the HOXA locus, we again utilizedCTCF and RAD21 shRNA lentiviruses in IMR90 cells. We per-formed 3C assays to detect the interaction between CBS5 andthe downstream heterochromatin barrier (p40) after separateCTCF and RAD21 knockdown. We found that independentknockdown of CTCF and RAD21 both resulted in loss of loopingbetween CBS5 and the downstream heterochromatin barriersite p40 (Fig. 4B). Our results are consistent with the hypothesisthat CTCF is required to load cohesin, which would be the majormediator of looping at this site.

Heterochromatin Looping Mediated by CBS5 Is Conserved and Devel-opmentally Regulated in Mouse Cells. We investigated whether theheterochromatin architecture of the HOXA locus is conserved

-3

-2

-1

0

1

2

3

4

CTCF RAD21 CTCF RAD21En

rich

men

t -

log

2 (C

hIP

/inp

ut)

CTCF and RAD21 ChIP

CBS5 Control

IMR90 CTCF KD

A

0

20000

40000

60000

80000

100000

120000

HOXA6 HOXA7 HOXA9 HOXA10

Rel

ativ

e co

py

nu

mb

er

CTCF KD - mRNA expression CTCF KD pGIPzC

0

0.5

1

1.5

2

2.5

3

3.5

CBS4 CBS4.5 CBS5 CBS5.5 CBS6 CBS7En

rich

men

t -

log

2 (C

hIP

/inu

t)

CTCF KD - H3K27me3 ChIPCTCF KD pGIPzB

α-actin

α-CTCF

controlCTCFshRNA

Fig. 2. CTCF function in chromatin barrier activity and looping. CTCF knock-down was performed in IMR90 cells to investigate the function of CTCF inchromatin barrier activity and looping. A shows qRT-PCR results of CTCFand RAD21 (cohesin subunit) ChIP analysis of chromatin isolated fromIMR90 cells infected with either a control virus, pGIPz (IMR90), or a CTCFknockdown virus (CTCF KD). Dark-gray bars represent enrichment level of tar-get protein at CBS5, whereas light-gray bars represent the enrichment at arandom control site. Inset is Western blot analysis showing the efficacy ofCTCF KD in IMR90 cells. B is the ChIP-qPCR results using H3K27me3 antibody,and shows the H3K27me3 levels at CTCF binding sites (CBS4 to CBS7) includ-ing CBS5 in the HOXA locus. C shows the relative expression level of HOXAgenes near CBS5 (two upstream genes, HOXA6 andHOXA7, and downstreamgenes, HOXA9 and HOXA10), in IMR90 cells infected with CTCF KD andcontrol (pGIPz) virus, respectively. The y axes represent the cDNA copy num-ber of HOXA genes normalized by the cDNA copy number of the controlACTG1 gene (HOXA copy no.∕ACTG1 copy no. × 108).

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

CTCF RAD21En

rich

men

t -

log

2 (C

hIP

/inp

ut)

RAD21 KD - CTCF and RAD21 ChIPCBS5 Control

A

α-actin

α-RAD21

controlRAD21shRNA

0

20000

40000

60000

80000

100000

120000

HOXA6 HOXA7 HOXA9 HOXA10

Rel

ativ

e co

py

nu

mb

er

RAD21 KD - mRNA expression RAD21 KD pGIPz

C

0

0.5

1

1.5

2

2.5

CBS4 CBS4.5 CBS5 CBS5.5 CBS6 CBS7En

rich

men

t -

log

2 (C

hIP

/inp

ut)

RAD21 KD - H3K27me3 ChIPRAD21 KD pGIPz

B

Fig. 3. Cohesin function in chromatin barrier activity and looping. RAD21knockdown was performed in IMR90 to determine the role of cohesin inchromatin barrier activity and looping independent of CTCF. A and B areChIP-qPCR results using CTCF, RAD21, and H3K27me3 antibody in IMR90 cellsinfected with RAD21 knockdown virus (RAD21 KD). Inset is Western blot ana-lysis showing the efficacy of RAD21 KD in IMR90 cells. In B, the H3K27me3levels at CTCF binding sites (CBS4 to CBS7) in the HOXA locus were deter-mined by ChIP-qPCR. C shows the mRNA expression level (relative copy num-ber) of HOXA genes (HOXA6–HOXA13) in control (pGIPz) and RAD21 KDcells. The y axes represent the cDNA copy number of HOXA genes normalizedby the cDNA copy number of the control gamma actin gene, ACTG1(HOXA copy no.∕ACTG1 copy no. × 108).

Kim et al. PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7393

BIOCH

EMISTR

Y

between human and mouse. We performed 3C analysis of CBS5interactions in mES cells and mNPC (Fig. S3B and Table S4).Interestingly, except for a strong proximal ligation signal most

likely due to self-ligation, which did not change in its intensityupon differentiation of mES cells to mNPC, we observed thatmES cells display very weak long-range interactions across theHOXA locus at CBS5 (Fig. 4C). However, upon differentiationinto mNPC, we detected strong induced interactions, indicatingtwo large loops corresponding to euchromatin and heterochro-matin domain boundaries and centered around CBS5 (Fig. 4D).Except for the strong proximal interaction, looping of the hetero-chromatin domain in theHOXA locus in mNPC resembles closelythe architecture observed in IMR90 cells.

OCT4 and Cohesin Bind at CBS5. Our results demonstrate that per-turbation of insulator function causes the architecture of theHOXA locus in IMR90 cells to closely resemble that of ES cells.We hypothesized that mES cell-specific factors might regulateloading of cohesin at CBS5. We found an experimentally deter-mined binding site for OCT4 within 20 bp of CBS5 in the mouseHOXA locus (35) (Fig. 1B). We first used ChIP-qPCR to verify andfurther analyze CTCF, RAD21, and OCT4 binding at CBS5 inmES and mNPC cells. We found that, although CTCF was boundat CBS5 in bothmES cells and mNPC, RAD21 was bound at CBS5only in NPC (Fig. 4 E and F). By ChIP-qPCR, we determined thatOCT4was bound at CBS5 only inmES but not inmNPC (Fig. 4G).To verify this interaction, we performed sequential ChIP assays(ReChIP) with CTCF and RAD21 antibodies after the primaryOCT4 ChIP in ES cells (Fig. 4H). Again, OCT4 interacted withCTCF but not RAD21. Therefore, it appears that binding ofcohesin and OCT4 at CBS5 is mutually exclusive at CBS5.

Exogenous OCT4 Antagonizes Cohesin Binding and Loop Formation atCBS5.To analyze the role of OCT4 in higher-order chromatin struc-ture, we overexpressed OCT4 in human IMR90 fibroblasts bylentiviral infection and performed 3C, ChIP, and gene expressionanalyses in parallel. Ectopic OCT4 bound at CBS5 and resulted insignificant loss of RAD21, but not CTCF binding (Fig. 5 A–C).ChIP-qPCR using H3K27me3 antibody revealed that the hetero-chromatin domain also remained intact after OCT4 overexpres-sion (Fig. 5E). However, 3C analysis of OCT4-overexpressingcells showed loss of looping across the major heterochromatin do-main (Fig. 5D). RT-qPCR analysis of HOXA transcripts showedexpression of adjacent genes HOXA6 and HOXA7 is reduced(Fig. 5F). Exogenous expression of OCT4 seems to replace cohesinat CBS5 and result in chromatin looping and gene expressionchanges similar to those observed as a result of RAD21 knockdown.Combined with our 3C and ChIP data from mES and mNPC, ourresults suggest that OCT4 prevents chromatin looping at theHOXAlocus by antagonizing cohesin loading at CBS5.

Looping Reinforces the Heterochromatin Stability.We expanded ouranalyses of HOXA mRNA expression to other genes within theheterochromatin domain (HOXA9–13) to determine whetherchanges in looping across the heterochromatin domain havelong-range effects on expression of genes within the heterochro-matin domain. CTCF knockdown in IMR90 resulted in domain-wide activation of HOXA9, HOXA10, HOXA11, and HOXA13(Fig. S5A). In contrast, RAD21 knockdown resulted in HOXA13up-regulation only (Fig. S5B), although other silent genes werenot affected. Similarly, in the IMR90 cells overexpressing OCT4,only HOXA13 was derepressed (Fig. S5C). Because OCT4 over-expression and RAD21 knockdown affects only the looping andnot the chromatin barrier activity, activation of HOXA13 is mostlikely due to loss of looping. These results suggest that loopingacross the heterochromatin domain provides an additional me-chanism for stable repression of HOXA genes.

DiscussionWe have elucidated the mechanism of a functionally conservedmammalian chromatin barrier at CBS5 in the HOXA locus of

Fig. 4. Chromosome architecture of the human and mouse HOXA locus. In-teraction frequencies between the anchor primer (noted by a yellow circle)and other primers in the locus were determined by real-time PCR and normal-ized to a random ligation library generated from EcoRI restriction digestionfragments from the bacterial artificial clones (BAC) covering the HOXA locus(CTD-3054H22 and CTD-2536K9) of this locus.A shows the interaction frequen-cies observed with the anchor primer located at CBS5 (blue vertical bar) withthe all the other primers in the locus in the top panel. The middle panel showsthe interaction frequencies observed with a control anchor primer located12-Kb downstream from CBS5 with all the other primers in the locus. The bot-tom panel shows the corresponding H3K27me3 ChIP-seq density across theHOXA locus in IMR90 cells. B is the result of the 3C assay in IMR90 cells infectedwith either CTCF or RAD21 knockdown (KD) viruses, or control pGIPz virus,tested to determine the interaction between CBS5 and the heterochromatinboundary (p40), which was detected as the strongest specific long-rangeinteraction site inA. C andD show the interaction frequency of CBS5 or controlprimer (noted by yellow circles) with all the other 3C primers in mES cells andmNPC across the mouse HOXA locus. The interaction frequencies were calcu-lated as in A. The control random ligation library for the mouse HOXA locuswas generated using the BAC, RP23-33N14. The H3K27me3 ChIP-Seq densitiesacross the HOXA locus from mES and mNPC are shown in the bottom panel.A strong interaction frequency observed between the close proximal site(within 10-Kb downstream to CBS5) and the anchor site at CBS5 likely resultsfrom self-ligation and thus the strong signal may not indicate long-rangeinteraction between the two sites. E–G are the results of ChIP-qPCR at CBS5performed using CTCF, RAD21, and OCT4 antibodies, respectively, and usingmES andmNPC chromatin.H shows the results of ReChIP performedwith CTCFand RAD21 antibodies following a primary OCT4 ChIP in ES cells.

7394 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1018279108 Kim et al.

both human and mouse. Our combined study of chromatin struc-ture, architecture, and gene expression has defined distinct rolesfor CTCF and cohesin at barrier sites and implicates chromoso-

mal looping in the stable maintenance of heterochromatin. CTCFis necessary for restricting the heterochromatin domain of theHOXA locus, and reduction of CTCF levels results in loss ofchromatin barrier activity (Fig. S6). In addition, CTCF providesa docking site for the cohesin complex, which results in the for-mation of higher-order loops. Alternately, cohesin is responsiblefor formation of chromosomal loops and proper regulation ofexpression at the HOXA locus, but displays a limited role inchromatin barrier activity. In addition, chromosomal loopingmediated by cohesin may contribute to a more stable heterochro-matin domain, as indicated by modest derepression of one of theHOXA genes in the silent heterochromatin domain as a result ofthe reduction in cohesin levels. In contrast to its role in loopingof the heterochromatin domain, cohesin seems to have a directeffect on transcriptional activation, independently of CTCF, ofgenes that are present within the euchromatin domain.

Previously, the mechanisms governed by CTCFand cohesin thatcontribute to chromatin state have been investigated indepen-dently (10–14, 36). Our study, however, investigates how thesefunctions (barrier activity, transcriptional regulation, and looping)are accomplished in parallel at a conserved locus through the dis-section of each responsible components (CTCF and cohesin).Thus, we provide a conserved regulatory mechanism that linksgene expression patterns with higher-order chromatin state.

Our work with ES cells demonstrates that this model is alsodevelopmentally regulated and provides a context for the estab-lishment of these expression domains. We identified OCT4, plur-ipotency factor and the principal regulator of ES cell state (22),as a negative regulator of loop formation mediated by cohesin ata CTCF binding site in mouse ES cells. OCT4 has been shownrecently to interact separately with cohesin at enhancer andpromoter sites where cohesin and mediator, a transcriptionalcoactivator, interact to promote gene activation by looping theintervening DNA sequences between these elements (23). OCT4has also been shown to interact with CTCF when regulatingX chromosome inactivation (24), We observe that OCT4 coloca-lizes with CTCF at CBS5, and reduction of OCT4 and loss of itsbinding during mNPC differentiation allows cohesin to bind toCBS5, which facilitates looping and heterochromatin segrega-tion. Additionally, our experiments show that ectopic OCT4 ex-pression in differentiated cells causes the deregulation of HOXAand other developmental genes by antagonizing cohesin andlooping. Detailed mechanisms underlying mutually exclusivebinding of OCT4 and cohesin at CBS5 remain to be elucidated.A possible mechanism may involve the epigenetic and sequencefeatures of the CBS5 chromatin barrier that is distinct fromOCT4 binding sites that function as enhancer elements. Alterna-tively, the previously demonstrated direct interaction betweenCTCF and OCT4 (24) may mask the domains on both proteinsthat promote cohesin loading at CBS5. This function of OCT4 islikely representative of the mechanism of maintaining cells in apluripotent state by regulation of developmentally regulated loci.

Based on our observations and previous studies, we propose athree-dimensional model for the heterochromatin folding andgene expression in HOXA locus (Fig. 5G). We have investigatedthe functions of two important epigenetic regulators, CTCF andcohesin, comprehensively and in parallel, to form a definitivemodel for establishment and maintenance of lineage-associatedchromatin domains, which is in turn important for proper geneexpression. Combined with previous related studies, our resultssuggest deregulated expression of the HOXA genes (37, 38) andmutations in the sequence of CTCF or CBSs could result in can-cer (39–41) via the mechanisms outlined in this paper.

MethodsCell Lines and Antibodies. IMR90 cells and mouse ES cells (AB2.2) were ob-tained from American Type Culture Collection and cultivated according tothe supplier’s instructions. Neural progenitor cells were obtained by in vitrodifferentiation of the mouse ES cells using the previously published protocol

-1

-0.5

0

0.5

1

1.5

2

OCT4 OE LacZ

En

rich

men

t -

log

2 (C

hIP

/inp

ut)

RAD21 ChIP

CBS5 control

DC

0

0.5

1

1.5

2

2.5

OCT4 OE LacZ

Fre

qu

ency

Major looping (CBS5-p40)

-1

-0.5

0

0.5

1

1.5

2

OCT4 OE LacZ

En

rich

men

t -

log

2 (C

hIP

/inp

ut)

OCT4 ChIP

CBS5 control

0

1

2

3

4

OCT4 OE LacZ

En

rich

men

t -

log

2 (C

hIP

/inp

ut)

CTCF ChIP

CBS5 Control

A B

-OCT4

Control OCT4 OE

G

heterochromatin

CBS5 p40

differentiation

reprogramming

ES cells

Differentiated cell

E

-1-0.5

00.5

11.5

22.5

33.5

CBS4 CBS5 CBS6 CBS7

En

rich

men

t -

log

2 (C

hP

/inp

ut)

OCT4 OE - H3K27me3 ChIP

OCT4 OE LacZ

0

20000

40000

60000

80000

100000

120000

HOXA6 HOXA7 HOXA9 HOXA10

Rel

ativ

e co

py

nu

mb

er

OCT4 OE - mRNA expression

OCT4 OE LacZ

F

Fig. 5. Effect of OCT4 overexpression on chromatin structure and architec-ture. A–C are the results of ChIP-qPCR analysis of CTCF, OCT4, and RAD21,respectively, at the CBS5 site using chromatin isolated from IMR90 cells over-expressing OCT4 (OCT4 OE) or a control protein (LacZ). Western blot analysisshowing the efficacy of OCT4 OE in IMR90 cells is displayed in the inset of B. Dshows the interaction frequency for major looping between CBS5 and theheterochromatin boundary site (p40) in IMR90 cells overexpressing OCT4and LacZ. In E, H3K27me3 ChIP-qPCR analysis of chromatin isolated fromIMR90 cells overexpressing OCT4 (OCT4 OE) was carried out at four CTCFbinding sites, CBS4, 5, 6, and 7 in the HOXA locus. F is the mRNA expressionanalysis on the HOXA genes (HOXA6, 7, 9, and 10) after OCT4 overexpressionin IMR90 cells. Values of the y axis represent the relative copy number ofHOXA cDNAs normalized to ACTG1 cDNA. G is the proposed model of regu-lated chromatin looping at HOXA locus.

Kim et al. PNAS ∣ May 3, 2011 ∣ vol. 108 ∣ no. 18 ∣ 7395

BIOCH

EMISTR

Y

(42). Antibodies of H3K27me3, H3K4me2, H3K4me3, CTCF, and RAD21 wereobtained from Millipore. Oct3/4 and pan-actin antibodies were obtainedfrom Santa Cruz Biotechnology.

LacO Reporter Assay. In order to analyze the chromatin barrier functionsat CBS5, we constructed EGFP reporter plasmids containing the 50 bpCBS5 element (including the core 20 bp CBS) flanking the EGFP expressioncassette and the LacO256 repeat elements located upstream (Fig. 1E). StableHEK293 clones harboring a single copy reporter were identified and used inthe LacI transfection experiments. To accelerate the EGFP transgene silen-cing, we transfected a LacI-EMD fusion protein expression vector (30) andcompared the resulting EGFP expression levels against the control LacI ex-pression vector without EMD. EGFP expression was determined by RT-qPCRusing the following primer pair: forward 5′-ACGACGGCAACTACAAGACC-3′and reverse 5′-ACCTTGATGCCGTTCTTCTG-3′.

Chromatin Immunoprecipitation. ChIP was performed as described previously(27). The DNA concentration of the recovered ChIP samples were determinedusing Quant-iT PicoGreen reagent (Invitrogen) using a set of DNA concentra-tion standards generated from the total chromatin sample. ReChIP assayswere performed as previously described (20).

RNA Preparation and Reverse Transcription. Total RNA (extracted using Trizolreagent, Invitrogen) was treated with DNase I (Roche) for 30 min at 37 °C andfurther extracted with acidic phenol:chloroform and precipitated with ammo-nium acetate and ethanol. The RNA was reverse transcribed using SuperScriptIII (Invitrogen). The resulting cDNA was incubated with 10 μg of RNaseA for30 min at 37 °C and purified using the QIAquick PCR purification kit (Qiagen).

Chromosome Conformation Capture. The 3C assays were performed asdescribed previously (33, 34, 43). One million formaldehyde cross-linked nu-clei were digested with either EcoRI or MseI (New England Biolabs) overnightat 37 °C. The subsequent steps were carried out as previously described.

Quantitative real-time PCR was performed to confirm the specific ligationbetween two DNA fragments in the test and control 3C libraries. Interactionfrequencies were calculated by dividing the amount of PCR product obtainedwith the test 3C library constructed from nuclei by the amount of PCRproduct obtained with the control library DNA generated from ligating EcoRIor MseI fragments from the corresponding bacterial artificial clones. All 3Canalyses were performed, at a minimum, in triplicate. The position of primersdesigned for 3C assay is represented in Fig. S3 and their sequence informationis in Tables S3 and S4.

Lentiviral shRNA and cDNA Expression. Plasmids containing shRNAs targetingeither RAD21 or CTCF were obtained from Open Biosystems. Each shRNAplasmid was cotransfected into 293T cells with Trans-Lentiviral Packagingmix (Open Biosystems) according to the manufacturer’s protocol. The resul-tant viruses were titered and used to infect IMR90 cells at a multiplicity ofinfection MOI ¼ 5 (for CTCF knockdown) or MOI ¼ 1 (for RAD21 knockdown)for 24 h in the presence of 10 μg∕mL polybrene (Specialty Media Products),and infected cells were harvested for RNA extraction, chromatin extraction,or 3C analysis at 48 h after infection. For OCT4 expression, we clonedfull-length human OCT4 (POU5F1) into the pLenti6.3/V5-TOPO lentiviral ex-pression vector (Invitrogen) and transfected into 293T cells according to theprotocol of the ViraPower HiPerform Lentiviral Expression System (Invitro-gen). A control virus carrying LacZ was also generated according to thisprotocol. IMR90 cells were infected with either OCT4 or LacZ lentivirus atan MOI ¼ 2 for 24 h and harvested for RNA extraction, chromatin extraction,or 3C analysis at 72 h after infection.

ACKNOWLEDGMENTS. We thank Andrew Belmont (University of Illinois atUrbana-Champaign, Urbana, IL) for providing the LacO construct. We alsothank the members of the laboratory for their comments on the manuscriptand Amanda Silverio for her help with initial characterization of lentiviruses.This work was supported by The Rita Allen Foundation, Sidney Kimmel Foun-dation for Cancer Research, and National Cancer Institute (R01CA140485).

1. Phillips JE, Corces VG (2009) CTCF: Master weaver of the genome. Cell 137:1194–1211.2. Cuddapah S, et al. (2009) Global analysis of the insulator binding protein CTCF in chro-

matin barrier regions reveals demarcation of active and repressive domains. GenomeRes 19:24–32.

3. Simon JA, Kingston RE (2009) Mechanisms of polycomb gene silencing: Knowns andunknowns. Nat Rev Mol Cell Biol 10:697–708.

4. Defossez PA, et al. (2005) The human enhancer blocker CTC-binding factor interactswith the transcription factor Kaiso. J Biol Chem 280:43017–43023.

5. Docquier F, et al. (2009) Decreased poly(ADP-ribosyl)ation of CTCF, a transcriptionfactor, is associated with breast cancer phenotype and cell proliferation. Clin CancerRes 15:5762–5771.

6. DonohoeME, Zhang LF, Xu N, Shi Y, Lee JT (2007) Identification of a Ctcf cofactor, Yy1,for the X chromosome binary switch. Mol Cell 25:43–56.

7. Yu W, et al. (2004) Poly(ADP-ribosyl)ation regulates CTCF-dependent chromatininsulation. Nat Genet 36:1105–1110.

8. Ishihara K, OshimuraM, NakaoM (2006) CTCF-dependent chromatin insulator is linkedto epigenetic remodeling. Mol Cell 23:733–742.

9. Squazzo SL, et al. (2006) Suz12 binds to silenced regions of the genome in a cell-type-specific manner. Genome Res 16:890–900.

10. Hou C, Zhao H, Tanimoto K, Dean A (2008) CTCF-dependent enhancer-blocking byalternative chromatin loop formation. Proc Natl Acad Sci USA 105:20398–20403.

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

12. Ling JQ, et al. (2006) CTCF mediates interchromosomal colocalization between Igf2/H19 and Wsb1/Nf1. Science 312:269–272.

13. Hadjur S, et al. (2009) Cohesins form chromosomal cis-interactions at the developmen-tally regulated IFNG locus. Nature 460:410–413.

14. Mishiro T, et al. (2009) Architectural roles of multiple chromatin insulators at thehuman apolipoprotein gene cluster. EMBO J 28:1234–1245.

15. Wendt KS, et al. (2008) Cohesin mediates transcriptional insulation by CCCTC-bindingfactor. Nature 451:796–801.

16. Hallson G, et al. (2008) The Drosophila cohesin subunit Rad21 is a trithorax group(trxG) protein. Proc Natl Acad Sci USA 105:12405–12410.

17. Cao R, et al. (2002) Role of histone H3 lysine 27 methylation in Polycomb-groupsilencing. Science 298:1039–1043.

18. Alexander T, Nolte C, Krumlauf R (2009) Hox genes and segmentation of the hindbrainand axial skeleton. Annu Rev Cell Dev Biol 25:431–456.

19. Rinn JL, et al. (2007) Functional demarcation of active and silent chromatin domains inhuman HOX loci by noncoding RNAs. Cell 129:1311–1323.

20. Bernstein BE, et al. (2006) A bivalent chromatin structure marks key developmentalgenes in embryonic stem cells. Cell 125:315–326.

21. Mikkelsen TS, et al. (2007) Genome-wide maps of chromatin state in pluripotent andlineage-committed cells. Nature 448:553–560.

22. Takahashi K, Yamanaka S (2006) Induction of pluripotent stem cells from mouseembryonic and adult fibroblast cultures by defined factors. Cell 126:663–676.

23. Kagey MH, et al. (2010) Mediator and cohesin connect gene expression and chromatinarchitecture. Nature 467:430–435.

24. Donohoe ME, Silva SS, Pinter SF, Xu N, Lee JT (2009) The pluripotency factor Oct4interacts with Ctcf and also controls X-chromosome pairing and counting. Nature460:128–132.

25. Lister R, et al. (2009) Human DNA methylomes at base resolution show widespreadepigenomic differences. Nature 462:315–322.

26. Hawkins RD, et al. (2010) Distinct epigenomic landscapes of pluripotent and lineage-committed human cells. Cell Stem Cell 6:479–491.

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

28. Blanchette M, et al. (2004) Aligning multiple genomic sequences with the threadedblockset aligner. Genome Res 14:708–715.

29. Belmont AS, Li G, Sudlow G, Robinett C (1999) Visualization of large-scale chromatinstructure and dynamics using the lac operator/lac repressor reporter system. MethodsCell Biol 58:203–222.

30. Reddy KL, Zullo JM, Bertolino E, Singh H (2008) Transcriptional repression mediatedby repositioning of genes to the nuclear lamina. Nature 452:243–247.

31. Parelho V, et al. (2008) Cohesins functionally associate with CTCF on mammalianchromosome arms. Cell 132:422–433.

32. Dostie J, et al. (2006) Chromosome conformation capture carbon copy (5C): A mas-sively parallel solution for mapping interactions between genomic elements. GenomeRes 16:1299–1309.

33. Dekker J (2006) The three ‘C’s of chromosome conformation capture: Controls,controls, controls. Nat Methods 3:17–21.

34. Dekker J, Rippe K, DekkerM, Kleckner N (2002) Capturing chromosome conformation.Science 295:1306–1311.

35. Chen X, et al. (2008) Integration of external signaling pathways with the core tran-scriptional network in embryonic stem cells. Cell 133:1106–1117.

36. Rubio ED, et al. (2008) CTCF physically links cohesin to chromatin. Proc Natl Acad SciUSA 105:8309–8314.

37. Andreeff M, et al. (2008) HOX expression patterns identify a common signature forfavorable AML. Leukemia 22:2041–2047.

38. Gupta RA, et al. (2010) Long non-coding RNA HOTAIR reprograms chromatin state topromote cancer metastasis. Nature 464:1071–1076.

39. Filippova GN, et al. (2002) Tumor-associated zinc finger mutations in the CTCF tran-scription factor selectively alter tts DNA-binding specificity. Cancer Res 62:48–52.

40. Sparago A, et al. (2004) Microdeletions in the human H19 DMR result in loss of IGF2imprinting and Beckwith-Wiedemann syndrome. Nat Genet 36:958–960.

41. Cui H, et al. (2003) Loss of IGF2 imprinting: A potential marker of colorectal cancer risk.Science 299:1753–1755.

42. Bain G, Kitchens D, YaoM, Huettner JE, Gottlieb DI (1995) Embryonic stem cells expressneuronal properties in vitro. Dev Biol 168:342–357.

43. Dostie J, Dekker J (2007) Mapping networks of physical interactions between genomicelements using 5C technology. Nat Protoc 2:988–1002.

7396 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1018279108 Kim et al.