Cell type specificity of chromatin organization
mediated by CTCF and cohesin
Chunhui Hou, Ryan Dale, and Ann Dean1
Laboratory of Cellular and Developmental Biology, National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda,
Edited* by Gary Felsenfeld, National Institutes of Health, Bethesda, MD, and approved January 7, 2010 (received for review October 20, 2009)
CTCF sites are abundant in the genomes of diverse species but their
function is enigmatic. We used chromosome conformation capture
to determine long-range interactions among CTCF/cohesin sites
over 2 Mb on human chromosome 11 encompassing the β-globin
locus and flanking olfactory receptor genes. Although CTCF occu-
the long-range interaction frequencies among the sites are highly
the absence of globin gene activity. Both CTCF and cohesins are
requiredforthe cell-type-specific chromatinconformation. Further-
more, loss of the organizational loops in K562 cells through reduc-
tion of CTCF with shRNA results in acquisition of repressive histone
marks in the globin locus and reduces globin gene expression
whereas silent flanking olfactory receptor genes are unaffected.
These results support a genome-wide role for CTCF/cohesin sites
through loop formation that both influences transcription and con-
tributes to cell-type-specific chromatin organization and function.
insulators are proposed to play a role in the establishment of such
domains within which proper enhancer–gene interactions occur
and improper ones are excluded (2). Insulators can function as
boundary elements between active and silent chromatin domains
and can also interfere with enhancer–gene interaction when
placed between them. Evidence also suggests that insulators are
involved in higher-order chromatin organization by clustering
together with other insulators. Such “insulator bodies” can be
visualized at the nuclear periphery of certain cells in Drosophila
and are thought to coalesce through the interaction of proteins
such as Su(Hw) and Drosophila CTCF (3, 4).
In mammalian cells, insulators are bound by CTCF, a protein
with 11 zinc fingers through which it can bind to a range of DNA
sequences, to itself, and to nuclear structural proteins such as
nucleophosmin and matrix attachment regions (MARs) (2).
Although insulators would be expected to reside primarily in
intergenic regions where they could provide boundary or
enhancer blocking activity, genome scans of CTCF enrichment
show that CTCF sites are overrepresented in genes and promoter
regions, with only 41–56% at intergenic locations (5). In Dro-
in intergenic regions (4). Thus, although CTCF sites are proposed
to be insulators, their in vivo functions may not be limited to
insulation. Recently, cohesin has been reported to colocalize at
most sites of CTCF enrichment and to play a role in transcrip-
tional insulation, but the significance of this joint occupancy to
higher-order chromatin structures is unknown (6).
In the Igf2/H19 locus, CTCF binding to the imprinting control
region on the maternal allele is required to loop out the Igf2 gene
and prevent interaction between Igf2 and enhancers (7, 8). Sim-
ilarly, CTCF binding and chromatin organization is necessary for
activation of IFNG during T-cell differentiation (9). However, in
ukaryotic chromatin is dynamically organized to form distinct
transcriptionally active or silent domains (1). Chromatin
be required for normal globin gene regulation in erythroid pro-
genitor cells (10). These functions of CTCF all involve CTCF-
mediated long-range chromatin organization, which can be man-
ifest intra- or interchromosomally (11, 12).
In a β-globin transgenic mouse model, we showed that CTCF-
When an extra copy of HS5 was placed between the locus control
region (LCR) and downstream genes, it formed a new insulator
loop with endogenous HS5 that topologically isolated the LCR
and nullified its activity. Here we show that CTCF/cohesin sites
over a 2-Mb region of human chromosome 11, primarily among
silent odorant receptor genes up- and downstream of the β-globin
enrichment is maintained in K562 cells and nonerythroid 293T
cells; however, the interactions between and among the sites are
highly cell type specific and require both CTCF and cohesin.
Furthermore, general loss of CTCF site interactions in a human
erythroid background results in diminished globin gene tran-
scriptioncoincident withanincrease in H3lysine 9di-methylation
interactions function generally in chromosome organization,
occurring not only at sites of insulator function.
Results and Discussion
Numerous Sites of CTCF Enrichment Reside Among Silent Olfactory
Receptor Genes Surrounding the Human β-Globin Locus. Although
several reports have documented sites of CTCF enrichment
genome-wide, whether these sites participate generally in chro-
matin organization through mutual interactions is unknown. To
explore this issue, we selected a 2-Mb test region on chromosome
11 surrounding the β-globin locus that is engaged in active tran-
both cell types, the clusters of olfactory receptor genes up- and
downstream of the β-globin locus are silent (14).
To accurately predict sites of CTCF interaction, we carried out
CTCF chromatin immunoprecipitation (ChIP)-chip analysis for
K562 cells and 293T cells using a NimbleGen tiling array of
nonrepetitive sequences on human chromosomes 10 and 11. We
then analyzed the data using the ACME algorithm (15) with
parameters chosen to maximize peak designation to a “training
set” of CTCF sites confirmed by ChIP assay (Fig. 1A and Fig. S1).
Importantly, theoptimized ACME parameters identified peaks at
the locus flanking β-globin HS5 and 3′HS1 sites where numerous
reports have documented CTCF binding in vivo(10, 13,16); these
sites had not been consistently identified as peaks in earlier
genome scans (17, 18). In other respects, such as the degree of
overlap in CTCF sites between cell types and the CTCF motif
Author contributions: A.D. and C.H. designed research; C.H. performed research; R.D.
contributed new reagents/analytic tools; C.H., R.D., and A.D. analyzed data; and C.H., R.
D., and A.D. wrote the paper.
The authors declare no conflict of interest.
*This Direct Submission article had a prearranged editor.
1To whom correspondence should be addressed. E-mail: email@example.com.
This article contains supporting information online at www.pnas.org/cgi/content/full/
| February 23, 2010
| vol. 107
| no. 8
shared by the sites, there was a high level of agreement between
this analysis and the earlier studies (Fig. S2).
We found that the cohesin complex members Rad21, SMC1,
consistent with reports that cohesin colocalizes with CTCF at
more than 80% of sites genome-wide (19–21) (Fig. S3). These
results validate numerous CTCF/cohesin sites at intervals
between 30 and 400 kb flanking the β-globin locus and within the
silent olfactory receptor gene clusters up- and downstream on
human chromosome 11.
Cell Type–Specific Interactions Among CTCF Sites Are Characteristic
Across an Extensive Chromosomal Region. A crucial step in
understanding the chromatin organizational potential of CTCF
binding sites is to determine interaction among them. Thus, we
carried out chromosome conformation capture (3C) analysis.
Primers for each EcoRI fragment with a CTCF enrichment site
were designed, and each was used as an anchor primer in tests of
proximity in chromatin with every other quantitative PCR
(qPCR)-validated CTCF site.
Of all possible (78) combinations, 10 interactions between
CTCF sites were confirmed in K562 cells, including the known
interaction between β-globin HS5 and 3′HS1 (Fig. 1B, Upper). Six
of the 10 interactions occurred between adjacent sites, with 3
clusters of 4 sites. The average distance between interacting sites
was about 70 kb with the largest interval being 185 kb. No inter-
actions were detected for the sites most distant from the globin
locus; however, these sites might interact with sites more 5′ or 3′
that were not analyzed. To confirm the specificity of the contacts,
a primer for each interacting site was paired with one from the
non-CTCF fragment adjacent to each of its partner CTCF sites
(Fig. S4). In each case, reduced interaction frequencies (18% or
less compared to partner CTCF frequencies) between CTCF sites
and non-CTCF sites were observed.
ChIP experiments confirmed that the CTCF sites occupied in
K562 cells were also occupied in 293T cells (Fig. S3). To test
whether the interactions between CTCF sites are also conserved,
3C was carried out on 293T cells. Fig. 1B, Lower, shows that 10
interactions were confirmed of which 4 were conserved with K562
cells, including the interaction between β-globin HS5 and 3′HS1,
and 6 were specific for 293T cells. The average interval between
cells derived from ChIP-chip experiments. CTCF peaks were predicted using the ACME algorithm optimized to the training set (C1–C31) of sites validated by
ChIP. RefSeq genes are depicted at the bottom in red (globin genes), blue (odorant receptor genes), or gray (other genes). (B) Long-range interactions among
CTCF/cohesin sites vary between cell types. (Upper) 3C was carried out with K562 cell chromatin. Specific interactions between and among CTCF sites (black
circles; gray circles represent CTCF sites not analyzed by 3C) are depicted by blue lines with the height of the curve corresponding to the cross-linking fre-
quency. (Lower) Interactions among CTCF sites in 293T cells are depicted by red curves.Cross-linking is plotted relative to the signal for two fragments in the
α-tubulin gene. Note the log scale of the y axis. Error bars represent SD.
CTCF occupancy and chromatin loop formation over 2 Mb of chromosome 11. (A) CTCF localization on chr11:4,000,000–6,000,000 in K562 and 293T
| www.pnas.org/cgi/doi/10.1073/pnas.0912087107 Hou et al.
these sites was 190 kb, which is considerably greater than in K562
cells. Five interactions were over even greater distances, up to
nearly 1 Mb. There were four clusters where multiple CTCF sites
came together, including three with four contacts and one with six
contacts, whereas only four interactions were with nearest neigh-
bors. Thus, clustered and longer-range interactions of CTCF/
cohesinsitespredominated in 293Tcells,wheretheβ-globinlocus
This clustering is reminiscent of the insulator bodies described in
Drosophila cells and might facilitate recruitment of the locus to a
nuclear territory inhospitable to transcription (22).
These data show that CTCF sites have the capacity to interact
with each other over a wide genomic area in both paired and
to be functional insulators as they reside among silent odorant
receptor genes where they do not mark boundaries between
regions of differing gene activity (see below). It is important to
note that the interactions detected by 3C are a composite of those
occurring in a large population of cells; in individual cells, a
subpopulation of the interactions may occur, as well as others that
a common phenomenon in the genome that exists in both active
and repressed gene neighborhoods.
Of the interactions that we observed, 60% were erythroid cell
type specific. A chromosome conformation capture-on-chip (4C)
study revealed that the β-globin locus interacts over very long
distances with a different set of loci on mouse chromosome 7 in
an erythroid versus a nonerythroid background (24); this study
was unable to resolve looping interactions within 10 Mb of the
globin locus itself, and the factors that participate in the inter-
actions are unknown. The association of CTCF with boundaries
of active and repressive genomic regions was also found to be
cell type specific (25). How the specificity of CTCF-dependent
chromatin loops is established is not known but might involve
cell-type-specific proteins that support more frequent or more
strongly stabilized higher-order chromatin interactions or might
relate to lineage-specific recruitment of cohesin (26).
CTCF Is Required for Long-Range Interaction Among Its Binding Sites
and for Enrichment of Cohesin at These Sites. To ascertain the
dependence of the long-range interactions on CTCF protein,
lentivirus-mediated shRNA knock-down of CTCF was conducted
in K562 cells. Western blots showed that CTCF protein was very
significantly reduced compared to nontransduced cells or cells
transduced with a control shRNA whereas the levels of Rad 21
and SMC3 were unaffected (Fig. 2A). ChIP indicated that
enrichment of CTCF was reduced fivefold on average compared
to K562 cells infected with control virus (Fig. 2B). Long-range
interaction frequencies observed in 3C experiments in CTCF
knock-down cells (Fig. 2C, gray curves) were significantly reduced
(5–10-fold, although at 2 sites the reduction was only 2-fold)
compared to those in control cells (blue curves), confirming the
importance of CTCF protein to the interactions.
ChIP was next carried out on CTCF knock-down K562 cells
with antibodies against cohesin. Although the expression levels
of Rad21 and SMC3 were unaffected by CTCF reduction (Fig.
2A), ChIP showed that enrichment of these proteins at CTCF
sites decreased on average fivefold, similar to the CTCF reduc-
tion, compared to control K562 cells (Fig. S5). Thus, the pres-
ence of CTCF is required for cohesin binding to the sites that we
studied, consistent with earlier work indicating that CTCF is
required to recruit cohesin complex members to shared sites
(19–21). The direct interaction of CTCF with cohesin subunit
SCC3 may underlie such recruitment (27).
Cohesin Can Influence Stable CTCF Binding, and Its Reduction Results
in Loss of Long-Range Interactions Among CTCF Sites. Some studies
reported that knock-down of Rad21 had no effect on CTCF
enrichment, whereas others reported that Rad21 knock-down
reduced CTCF enrichment about 50% at some sites (19, 20). To
shed light on this issue, Rad21 was knocked down with lentivirus
shRNA in K562 cells. Western blotting showed that the expression
of Rad21 was nearly abolished whereas the protein level of CTCF
was not affected (Fig. 3A). ChIP experiments showed that the
enrichment of Rad21 at CTCF sites was significantly reduced
(average of fivefold) compared to that in control K562 cells (Fig.
3B). Significantly, enrichment of CTCF was also reduced (average
of threefold) in K562 cells subjected to Rad21 knock-down com-
to stable CTCF binding at sites surrounding the β-globin locus.
Because reduction of Rad21 leads to lower CTCF enrichment
at the shared sites, the interaction frequencies between CTCF
sites would be expected to be reduced because of the requirement
for CTCF. To examine this prediction, 3C was carried out on
Rad21 knock-down cells. As shown in Fig. 3D, the interaction
frequencies of CTCF sites in K562 cells with reduced Rad21 (gray
curves) were significantly lower (average reduction of threefold)
than those in control cells (blue curves): CTCF site interactions
were also reduced in SMC3 knock-down cells (Fig. S6). No
changes in cell cycle distribution, growth rate, or apoptosis were
and among its binding sites. (A) Western blot indicating reduction of CTCF
protein by shRNA without alteration of Rad21 or SMC3 levels. (B) K562 cells
were transduced with control or CTCF shRNA. ChIP was performed with an
antibody to CTCF. Error bars represent SEM. (C) 3C was carried out using
chromatin from K562 cells transduced with control or CTCF shRNA. Inter-
actions between and among CTCF sites are indicated by blue curves, and
reduction of these interactions after knock-down of CTCF is indicated by
gray curves. Error bars represent SD. **P < 0.01, *P < 0.05.
Reduction of CTCF results in loss of long-range interactions between
Hou et al.PNAS
| February 23, 2010
| vol. 107
| no. 8
observed for cells treated with cohesin (or CTCF) shRNAs (Fig.
S7). Thus, our data suggest the importance of both cohesin and
CTCF tolong-range interactionsbetween andamongCTCF sites.
In contrast, two CTCF/cohesin site looping interactions at the
IFNG cytokine locus were reduced after Rad21 protein reduction
even though CTCF remained bound (9).
CTCF Reduction Alters the Epigenetic Environment Within the
β-Globin Locus and Negatively Affects γ-Globin Expression. Recent
studies in which one or a few specific CTCF sites were lost due to
deletion or mutation have revealed both positive and negative
effects on transcription of local genes (8, 21, 28, 29). However,
deletion of the mouse β-globin HS5 and 3′HS1 CTCF sites indi-
viduallyorin combination didnot affectβ-globingene expression,
suggesting that, although these sites interact in vivo, they are not
required to function as insulators at their endogenous location in
the mouse genome (30–32). In other studies, partial knock-down
of CTCF protein or mutation of the 3′HS1 CTCF site were found
to have no effect on β-globin transcription in mouse erythroid
cells before the onset of active globin gene transcription (10).
CTCF knock-downin K562cells reduced γ-globintranscription
by 50% (Fig. 4A). Transcription of α-globin was also reduced,
suggesting that CTCF function is required in this unlinked locus.
RNA pol II enrichment was reduced at γ-globin (Fig. 4B) and at
HS3 and HS4 of the LCR, sites known to recruit pol II (33). To
ask if reduction of CTCF looping interactions changed the epi-
genetic landscape within the β-globin locus, we determined
occupancy of histone H3K4 di-methylation (H3K4me2), a mark
of active transcription and of histone H3 K9 di-methylation
(H3K9me2), a repressive mark. There is an inverse pattern of
these marks over silent odorant receptor genes flanking the locus
(lines, Fig. 4C), which is punctuated by sites of CTCF and Rad21
binding (bars, Fig. 4C). Between the HS5 and 3′HS1 CTCF sites
flanking the globin locus, the two modifications vary as a function
of gene activity: H3K4me2 marks the LCR and γ-globin gene
whereas H3K9me2 is low at these positions but high at the silent
δ- and β-globin genes. Upon reduction of CTCF, H3K4me2 was
reduced about twofold at γ-globin, and other changes throughout
the locus were not observed (Fig. 4D). In contrast, there was an
increase in H3K9me2 (twofold to over fivefold at HS1-4) over the
LCR, and this mark was detected at γ-globin (Fig. 4E), in parallel
with reduced transcription and reduced pol II enrichment.
Thus, global loss of CTCF site interactions negatively affects
active globin gene transcription coincident with an increase in
H3K9me2 in the locus. We consider it likely that this effect is
directly due to the loss of CTCF occupancy. This conclusion is
supported by Western blot analysis, indicating that CTCF
reduction does not affect the protein levels of the important
globin gene activators GATA-1 and NF-E2 (Fig. S8). Inactive
genes up- and downstream of the globin locus were unaffected by
CTCF reduction (Fig. S8) in agreement with earlier work (10).
The differing effect of CTCF reduction on human globin gene
transcription compared to loss of either or both of the mouse HS5
in our study, CTCF site interactions are lost globally, perhaps
allowing more generalized spread of negative histone mod-
ifications into the globin locus. Second, it is conceivable that fetal
β-globin transcription. Finally, solitary ERV9-LTR insertions
flank human HS5 and 3′HS1, and adjacent, unique ChIP ampli-
cons (ERVdn and OR51V1 in Fig. 4E) are highly enriched for
genes, similar to the role proposed for the HS4 insulator (HS5
ortholog) lying between the chicken globin locus and a 16-kb
heterochromatic region (34). The mouse locus lacks these retro-
viral insertions and may not require the insulator activity (35, 36).
Because CTCF sites have an intrinsic loop-forming capability
even at ectopic locations (13), we favor the idea that loops are
central to CTCF site function even though in these experiments
we cannot separate CTCF occupancy, per se, from CTCF site
loop formation. Collectively, the results suggest that loss of
chromosomal CTCF organizational structure through disruption
of loops exerts a negative effect on active globin gene tran-
scription at an endogenous locus. Interestingly, this effect seems
actions among CTCF sites. (A) Western blots showing reduction of Rad21 by
shRNA and lack of effect of this reduction on CTCF protein. (B) K562 cells
were transduced with control or Rad21 shRNA. ChIP was performed with an
antibody to Rad21. Error bars represent SEM. (C) ChIP was carried out as
described in B with antibodies to CTCF. (D) 3C was carried out using chro-
matin from K562 cells transduced with control or Rad21 shRNA. Interactions
between and among CTCF sites are indicated by blue curves, and reduction
of interactions after knock-down of Rad21 is indicated by gray curves. Error
bars represent SEM. **P < 0.01, *P < 0.05.
Reduction of cohesin components diminishes long-range inter-
| www.pnas.org/cgi/doi/10.1073/pnas.0912087107Hou et al.
to be dominant over LCR activity because the silent genes sur-
rounding the globin locus remain silent; this outcome could also
relate to the absence of odorant receptor gene-activating factors
in erythroid cells. Considering the broad and diverse distribution
of CTCF sites revealed in genome scans, we suggest that the
insulation function of CTCF/cohesin sites may be secondary to
their broader organizational role in the genome. The intrinsic
insulation/looping capacity of the sites might then be manifest at
particular positions where the chromatin surrounding an active
locus is not conducive to gene transcription.
Materials and Methods
Cell Culture. K562 cells and 293T cells were cultured in RPMI1640 and DMEM,
respectively, supplemented with 10% FBS. Cells were maintained at 37 °C in a
ChIP-chip and Data Analysis. ChIP DNA was amplified as recommended
(Nimblegen), and DNA samples were hybridized to a human genome tiling
array consisting of 50 mers positioned every 100 bp along nonrepeating
Onthebasisoftheinspection ofindividualdatatracksatafalsediscovery rate
(FDR) of <0.2, 31 primer combinations were chosen to represent CTCF peaks
and negative positions. CTCF ChIP and real-time qPCR (Fig. S1) results vali-
type, biweight-scaled log2 ratios from three biological replicates were aver-
aged to reduce biological variation. The averaged log2 ratios were used as
input to the ACME tiling array algorithm. To call significant peaks, ACME uses
a χ2test on a 2 × 2 contingency table by using the number of probes above
and below a threshold both inside and outside a moving window. The final
ACME parameters (threshold of 0.984, p value 7.58e-14 for K562; threshold of
0.9890, p value 1.67e-9 for 293T; window size 1000 bp) were chosen by an
iterative process to obtain the best fit to sites of positive CTCF binding as
determined by the qPCR data.
Results of three RNA preparations are shown ±SEM. (B) ChIP performed on chromatin from K562 cells after treatment with control or CTCF shRNA using an
antibody against RNA pol II. Error bars represent SEM. Amplicons include ERVin and ERVdn, within and downstream (unique amplicon) of ERV9-LTR; HS5 and
3′HS1, globin locus-flanking CTCF sites; HS1–HS4, locus control region DNase I hypersensitive sites; globin gene promoters and exons as indicated; and
OR51V1, odorant receptor gene in the globin locus 3′ flank. (C) ChIP was performed using K562 cell chromatin as described in B with antibodies to H3K4me2,
H3K9me2, CTCF, and SMC1. H3 methylation is plotted with the highest value for each modification set to 1 in the line graphs. CTCF or cohesin ChIP
enrichment is represented by superimposed bars. Diagram at the top: orange lines, globin genes; gray box, LCR; blue lines, odorant receptor genes; gray lines,
other silent genes; green lines, ERV9-LTR. Note the depiction of chromosome 11 in the 5′ to 3′ direction, which is reversed compared to the Genome browser
view in Figs. 1–3. Panel C is drawn to scale but the numbers below the graph are arbitrary. (D) ChIP was performed as described in B with antibodies to
H3K4me2. (E) ChIP was performed as described in B using an antibody to H3K9me2.
Effect of CTCF shRNA on K562 cells. (A) RT–PCR analysis of globin gene transcription in K562 cells after transduction with control or CTCF shRNA.
Hou et al. PNAS
| February 23, 2010
| vol. 107
| no. 8