CTCF: insights into insulator function during development

As early as the 1950s, the existence of genomic insulators has been postulated based on several observations, such as the phenomenon of position effect variegation (see Glossary, Box 1), in which gene activity is dependent on its genomic location (Lewis, 1950). These locations were later identified as heterochromatic, or inactive, chromatin regions, which, upon translocation, inversion or deletion of chromosomal fragments, may repress the expression of a neighbouring gene (Baker, 1968). The interpretation of this observation was that the genomic rearrangement might have deleted insulators that normally protect genes from the repressive effect of flanking heterochromatin. To add weight to this notion, the biophysical analysis of the Drosophila genome argued for the presence of insulator sequences that separate supercoiled genomic domains (Benyajati and Worcel, 1976). More recently, functional tests with transgenes (see Glossary, Box 1) revealed the existence of three properties of insulators (see Glossary, Box 1): (1) to provide a barrier (or boundary; see Glossary, Box 1) function to prevent repressive heterochromatin from spreading into a neighbouring domain; (2) to provide an enhancer-blocking (see Glossary, Box 1) function when positioned between the enhancer and promoter (see Glossary, Box 1) (Sun and Elgin, 1999), allowing insulators to produce opposite effects either by facilitating the maintenance of a transcriptionally active state or by inhibiting the action of enhancers; (3) to allow three-dimensional looping of genomic regions, a property that is possibly inherent to insulator function, as discussed throughout this review.

The first proteins identified that bind to insulator sequences and mediate the insulating function were boundary element-associated factor of 32 kDa (BEAF32) (Zhao et al., 1995), Su(Hw) (Holdridge and Dorsett, 1991; Geyer and Corces, 1992) and zeste-white 5 (Zw5; dwg – FlyBase) (Gaszner et al., 1999) in Drosophila. In vertebrates, CCCTC-binding factor (CTCF) was first shown to mediate insulation (Bell et al., 1999) and was later demonstrated to be highly conserved and also found in Drosophila (dCTCF) (Moon et al., 2005). The DNA-binding domains of human and Drosophila CTCF are ∼60% similar. Interestingly, in Drosophila, CTCF is only one of several insulator proteins, whereas in vertebrates, only CTCF is known to mediate insulation function. DNA binding of CTCF is sequence specific and conferred by its zinc finger domain, which in vertebrates and in insects harbours 11 zinc fingers (see Glossary, Box 1) that appear to be differentially used depending on the respective binding site in the genome (Burcin et al., 1997; Boyle et al., 2010).

Here, we provide a comprehensive overview of our current knowledge of the functions of CTCF, shaped by exciting recent developments that are beginning to reveal the molecular mechanisms of CTCF with respect to the chromatin insulation activity required for proper development. Moreover, we highlight the similarities and differences between insulation in Drosophila and vertebrates using two well-analysed genomic regions: a homeotic gene locus and the imprinted Igf2/H19 locus, respectively. Finally, we discuss current information on genome-wide CTCF-binding sites and interpret these in the context of chromatin insulation.

Microscopic inspection of interphase nuclei clearly demonstrated a chromatin ‘patchwork’ in which DNA was packaged as compact heterochromatin (see Glossary, Box 1) or decondensed euchromatin (see Glossary, Box 1). Recent molecular analyses on a genome-wide scale supported the domain model of the linear genome (Fig. 1) (Filion et al., 2010; Ernst et al., 2011). These domains have to be insulated from each other in order to prevent aberrant gene expression, e.g. to prevent the inactive domain from erroneously inhibiting genes in the active domain and vice versa (Fig. 1). Within domains of potentially active genes, the activity of a given gene is controlled by enhancer sequences, which can be found either adjacent to the gene promoter or at a considerable distance either downstream or upstream of the gene. Distance at the level of the linear genome does not impose a problem for enhancer function as the intervening DNA is looped out such that promoter and enhancer will be in close contact (Fig. 1).

The question therefore arises: How does the distant enhancer locate the corresponding promoter and, in particular, how does enhancer ‘homing’ work when two genes with different expression profiles are close together and in a ‘head to head’ orientation (Fig. 1)? Enhancers or locus control regions could be envisaged to recognise and contact their corresponding promoters through a specific composition of interacting proteins bound to the enhancer and to the corresponding promoter. However, in many cases insulators are known to play a role in the three-dimensional folding of chromatin (see below), thereby allowing for or preventing functional contact between enhancer and promoter elements.

The first sequences to be functionally tested for insulator activity were chosen because of circumstantial evidence strongly suggesting an insulator function. This was clearly the case for the specialised chromatin structures scs and scs′ (Udvardy et al., 1985), which flank the Drosophila hsp70 gene and were found to be located at the junctions between decondensed and compact regions on the polytene chromosome (see Glossary, Box 1). Similarly, the facet-strawberry sequence is located at a polytene chromosome decondensed interband separating two bands of condensed chromatin (Rykowski et al., 1988). In both cases, functional tests (see Box 2) revealed their insulator activity. Moreover, genetic evidence indicated that some Drosophila retrotransposons, such as Gypsy or Idefix, mediate their biological effect by causing insulation at the site of integration, as confirmed by functional tests (Geyer and Corces, 1992; Brasset et al., 2010). In the case of the two homeotic gene clusters the Antennapedia complex (see Glossary, Box 1) and the Bithorax complex (see Glossary, Box 1), multiple regulatory regions control the segment-specific expression of homeotic genes (see below). Such a complex arrangement of segment-specific regulatory sequences argued for the presence of insulator sequences, which, again, was confirmed functionally (Hagstrom et al., 1996; Belozerov et al., 2003). Similarly, in vertebrates, the identification of complex expression patterns of adjacent genes or of neighbouring regulatory regions with different activities led to the prediction of insulators at these sites. The first one shown to mediate insulation was the chicken 5′HS4 element, which lies adjacent to the locus control region (LCR) of the β-globin gene cluster and is positioned between a compact chromatin region and the LCR-induced decondensed chromatin (Chung et al., 1993). Finally, an insulator is known to be positioned between the Igf2 and H19 genes ensuring their parent of origin-specific expression. This insulator mediates multiple functions related to reading the parental imprint and mediating the imprinted expression of these two genes. Other sequences, for which insulator function has been demonstrated, are listed in Table 1.

Given the original observation that the genome is arranged in looped genomic domains (Benyajati and Worcel, 1976) and the postulation that looped domains have to be insulated from each other, the most obvious model for insulator mechanism is that insulators themselves mediate loop formation. As depicted in Fig. 1, this insulator-mediated looping model does indeed explain how some enhancers are blocked from making contact with promoters, whereas others are ‘pulled’ into the vicinity of a promoter. An alternative model, the decoy model, postulates a competition between insulator and promoter for interaction with the enhancer. In fact, a substantial fraction of CTCF-bound insulators is found to be associated with enhancers (Handoko et al., 2011). However, these looping models cannot account for all cases of enhancer-blocking activity nor can they fully explain how insulator-mediated looped domains affect the barrier mechanisms at play at domain boundaries. Thus, additional features, such as the specific sequence, chromatin context and CTCF co-factors might play a role in insulator function and these are explained below.

The genetic information of higher organisms, such as insects and vertebrates, is stored in two copies of the genome, the parents contributing one copy each. Usually, both copies (alleles) of a single gene are expressed when the gene is activated. However, in mammalian systems, ∼100 genes exhibit a parental imprint (see Glossary, Box 1) resulting in monoallelic expression only. In other words, some genes are only expressed when inherited from the mother and others from the father. Imprinted genes are highly conserved between mouse and human and are often organised in clusters such that paternally and maternally imprinted genes are mixed (for a review, see Ferguson-Smith, 2011). One of the first identified imprinted gene loci was the insulin-like growth factor 2 (Igf2)/H19 locus (Bartolomei et al., 1991; DeChiara et al., 1991; Ferguson-Smith et al., 1991). Gene knockout studies in mice showed that the Igf2 gene enhances placental and foetal growth, whereas the H19 gene, which is expressed as a non-coding RNA, retards foetal growth. Interestingly, both genes in mouse and human are arranged such that the same set of enhancers activates the H19 gene on the maternal allele and the Igf2 gene on the paternal allele (Fig. 2A). It was thus proposed that an insulator factor is present at the imprinting control region (ICR) in order to block the enhancer function at the maternal Igf2 gene (Fig. 2A). Indeed, the insulator factor CTCF was shown by several groups to mediate enhancer blocking at the ICR (Bell and Felsenfeld, 2000; Hark et al., 2000; Kanduri et al., 2000; Szabo et al., 2000), and this site turned out to be a suitable model for unravelling several insulator features. For example, DNA methylation was found to regulate insulation; the paternal ICR is methylated, whereas the maternal ICR is not. In this case, only the unmethylated maternal ICR binds CTCF and this in turn blocks Igf2 gene activation by the enhancer. Deletion of the CTCF-binding sites and maternal transmission results in biallelic expression of Igf2, confirming the insulator function of CTCF on the wild-type maternal allele (Engel et al., 2006). Furthermore, Engel and colleagues found that the gene locus with a CTCF-binding-site deletion becomes methylated after implantation, suggesting that maintenance of the hypomethylated state requires the presence of the CTCF-binding site. CTCF-binding-site deletion not only demonstrated the enhancer-blocking function of CTCF, but also showed that the efficient initiation of the adjacent H19 gene is dependent on CTCF binding. Taken together, these observations suggest that CTCF is also involved in promoter function, a hypothesis that was substantiated by whole-genome analysis (see below). This is a simplified view emphasising insulator function. It is very likely that a concerted action of several mechanisms mediates gene activity at this locus. As an example, direct binding between the maternal ICR and the Igf2 promoter has also been shown to turn off Igf2 expression (Li et al., 2008).

Geneticists have identified imprinted genes in humans based on the parental-specific inheritance of a phenotype caused by the mutant gene, e.g. a mutant H19 gene is one of several causes of Beckwith-Wiedemann syndrome (BWS, see Glossary, Box 1) only when inherited from the mother. The syndrome is associated with foetal overgrowth and increased risk of tumour formation. However, most cases of BWS are not caused by a gene mutation, but rather by an imprinting defect. For example, one group of patients was found to gain DNA methylation at the ICR (Fig. 2B), resulting in biallelic Igf2 expression and reduced H19 activity (Azzi et al., 2009). This BWS phenotype is also observed in cases with microdeletions in the CTCF-binding site of the ICR (Sparago et al., 2004). Thus, loss of CTCF binding at the ICR by either DNA methylation or mutation causes Igf2 activation and H19 inactivation. H19 is not activated because of DNA methylation upstream of the promoter and possibly because of the lack of CTCF binding, which might be required for promoter activation. CTCF sites on both alleles flanking the Igf2/H19 locus might pull these sites together such that the enhancer can interact with the Igf2 promoter (Fig. 2B). This scenario was recently confirmed by analysing the three-dimensional arrangement of the CTCF-binding sites of BWS patients in this locus (Nativio et al., 2011). In this study, the downstream CTCF-binding site in control samples was observed to interact with both the ICR and the upstream CTCF site, whereas, in BWS cells, the interaction of this downstream CTCF-binding site with the ICR is reduced, but interaction with the upstream CTCF site is increased (Fig. 2B). Similar to the discussion above of the mouse Igf2/H19 locus, this explanation and the diagram filter out a single feature that is only part of the molecular mechanism at the human locus, which is likely to involve other interactions and mechanisms, such as DNA methylation, as well.

Further insights into CTCF function in humans came from studies of Silver-Russell syndrome (SRS, see Glossary, Box 1), which is characterised by growth retardation during foetal and postnatal development. In more than 50% of SRS patients, a loss of methylation at the paternal ICR is observed, resulting in biallelic expression of the H19 gene and reduced expression of Igf2 (i.e. the converse of the BWS pattern). Analysis of long-distance interaction in SRS cells revealed that the downstream CTCF-binding site exhibits increased interaction with the ICR and reduced contact with the upstream CTCF site (Nativio et al., 2011) (Fig. 2C).

For the mouse Igf2/H19 locus, it has been shown that three-dimensional chromatin looping is dependent on CTCF (Kurukuti et al., 2006). An interesting candidate as a mediator of a CTCF-dependent looping function is cohesin (see Glossary, Box 1). Genome-wide analyses of CTCF and cohesin binding resulted in largely overlapping patterns (Parelho et al., 2008; Wendt et al., 2008). Moreover, functional tests revealed that cohesin is bound to CTCF and is required for chromatin long-distance interaction (Wendt et al., 2008; Hadjur et al., 2009; Nativio et al., 2009). Because cohesin is also involved in sister chromatid cohesion, which is independent of CTCF, it can be envisaged that cohesin confers the linking function and CTCF is responsible for sequence-specific binding. Indeed, it has been demonstrated that the cohesin component SA2 makes direct contact with CTCF, and the ring-forming components of cohesin are recruited by SA2 (Xiao et al., 2011) (Fig. 3A). CTCF complex purification and functional tests showed that the RNA helicase p68 together with its non-coding RNA component SRA are required for insulation at the Igf2/H19 locus (Yao et al., 2010). These authors proposed that the RNA helicase stabilises cohesin-CTCF interaction and that cohesin mediates chromatin looping.

Thus, an insulation mechanism can be envisaged involving CTCF, p68, SRA-RNA, cohesin and chromatin long-distance interaction (Fig. 3A).

After identification of the specialised chromatin structures containing insulator sequences in Drosophila, namely scs and scs′ (see above), two insulator-binding proteins were identified: zeste-white 5 (Zw5) and boundary element associated factor (BEAF) (Zhao et al., 1995; Gaszner et al., 1999; Roy et al., 2006). In Drosophila, additional insulator-binding factors were identified for other insulators. For example, suppressor of hairy wing [Su(Hw)] was found to bind to the well-characterised gypsy retrotransposon (Parkhurst et al., 1988), and the GAGA-binding factor (GAF) to the eve promoter (Ohtsuki and Levine, 1998). These four proteins are all ubiquitously expressed, harbour a zinc finger domain for direct binding to DNA and are responsible for the insulator activity of specific target sites. Interestingly, the only insulator-binding factor in Drosophila with an orthologue in vertebrates is CTCF (Moon et al., 2005) with binding sites in the highly conserved homeotic gene cluster (or Hox cluster, see Glossary, Box 1) (Holohan et al., 2007). The genes within this cluster code for transcription factors that determine body patterning along the anteroposterior axis of an organism. Importantly, the linear arrangement of Hox genes on the chromosome reflects their expression along the body axis, a phenomenon called co-linearity. This arrangement of genes and of segment-specific regulatory sequences argued for the existence of insulator sequences that separate segment-specific regulatory units (Lewis, 1978; Hagstrom et al., 1996; Belozerov et al., 2003). Deletion of these insulator sequences results in a shift of segment identity, such that neighbouring segments show an identical identity (for details, see Maeda and Karch, 2006).

In Drosophila, one such Hox cluster, known as the Bithorax complex (BX-C, Fig. 4) specifies the third thoracic segment (T3) and all eight abdominal segments (A1-A8) of the fly. BX-C is ∼300 kb long and contains three homeotic genes: Ultrabithorax (Ubx), abdominal A (abd-A) and Abdominal B (Abd-B). These three genes are regulated by nine different cis-regulatory domains, each responsible for the identity of the nine segments (Fig. 4) (for a review, see Maeda and Karch, 2006). Fab-7 was the first insulator identified within BX-C and was found to separate the regulatory domains iab-6 and iab-7 from each other (Gyurkovics et al., 1990) and, subsequently, other insulators, such as Fab-8, were identified for which enhancer-blocking activity is dependent on CTCF (Moon et al., 2005). Furthermore, chromatin immunoprecipitation (ChIP, see Glossary, Box 1) of CTCF revealed binding at most of the boundaries (Fig. 4), thereby underscoring a role for CTCF in dividing the regulatory domains of the BX-C (Holohan et al., 2007; Mohan et al., 2007). Finally, the important role of CTCF for regulatory domain function in the BX-C is underscored by misexpression of Abd-B and by homeotic transformations of the abdominal segments in CTCF mutants, resulting in an additional abdominal segment 7 (Gerasimova et al., 2007; Mohan et al., 2007).

A search for co-factors involved in chromatin insulation identified the centrosomal protein 190 (CP190), which binds to Su(Hw) and is required for the enhancer-blocking activity of the gypsy retrotransposon (Pai et al., 2004). Genome-wide CP190-binding analysis clearly showed extensive colocalisation of CP190 with Su(hw), as well as with CTCF. Moreover, all of the CTCF sites in the BX-C are bound by CP190, indicating a role for CP190 in the regulation of the BX-C. In addition to the strong overlap of CP190-binding sites with CTCF and Su(Hw), the insulator-binding factors BEAF and GAF also colocalised with CP190, thus identifying CP190 as a central factor in insulator function (Bartkuhn et al., 2009; Bushey et al., 2009; Nègre et al., 2011). The RNA helicase Rm62 was also found to interact with CP190 in an RNA-dependent manner (Lei and Corces, 2006). Involvement of RNA molecules in chromatin insulation was also suggested from experiments with vertebrate CTCF, in which CTCF interaction with cohesin was found to involve RNA species as well as the RNA helicase p68 (Fig. 3A) (Yao et al., 2010). In both cases, protein-protein interaction with the Drosophila Rm62 factor or the similar human p68 factor was dependent on RNA. Whether the RNA component itself is involved in the interaction, or in changing a protein conformation required for other proteins to bind, remains to be shown.

To understand better the underlying mechanisms of insulator function, the three-dimensional structure of the locus was analysed. For this purpose, insulator elements of the BX-C harbouring CTCF-binding sites were cloned into reporter constructs and, after integration in the Drosophila genome, were found to interact physically. However, deletion of the CTCF-binding sites only partially reduced the ability of the insulator elements to interact (Kyrchanova et al., 2011). Therefore, other elements or proteins might be involved in the long-range interaction. Indeed, polycomb response elements (PREs, see Glossary, Box 1) were shown to interact in vivo in a silenced BX-C background (Lanzuolo et al., 2007), although functional and interaction assays showed that long-range contacts are independent of PREs (Li et al., 2011).

Thus, insulator function in the BX-C locus involves long-distance chromatin interactions although it is difficult to envisage how the enhancers of one domain can be blocked in a specific segment (e.g. iab-6 in abdominal segment 5), but can be active in another (e.g. iab-6 in abdominal segment 6).

Therefore, the molecular mechanism regulating the BX-C locus turns out not to be a simple insulation- or relief from insulation-based mechanism, but rather a complex picture is emerging. Complexity is in part caused by additional factors, such as polycomb group proteins, and also by the large number of genomic elements involved in the three-dimensional interaction (for a review, see Bantignies and Cavalli, 2011). Besides insulators, other sequence elements that are involved include enhancers and promoters (see below). Furthermore, promoter-targeting sequences (PTSs) have been shown to allow enhancer elements to bypass the intervening insulators (Zhou and Levine, 1999). Another element, identified in the promoter of Abd-B, is the promoter-tethering element (PTE). This element is able to bypass the insulators Fab-7 and Fab-8 and to activate Abd-B expression via the enhancer element of iab-5 (Akbari et al., 2007). Therefore, although the insulator function of CTCF and its co-factors at the BX-C locus is well studied, functional details of the underlying mechanisms enabling these factors to confer as well as to block insulation have still to be elucidated.

In addition to mediating developmental gene regulation of the Igf2/H19 and Hox gene loci, CTCF is now known to play a significant role in many other developmental processes (Table 2).

The importance of CTCF in development is demonstrated by the fact that CTCF-null mice are embryonic lethal (Splinter et al., 2006; Heath et al., 2008). Similarly, CTCF-null Drosophila strains do not develop beyond the pupal stage (Gerasimova et al., 2007; Mohan et al., 2007). Furthermore, mouse oocytes depleted of CTCF showed increased methylation of the H19 ICR, decreased developmental competence, as well as meiotic and mitotic defects in the egg and embryo, respectively. Therefore, CTCF can be considered to be a maternal-effect gene, transcription of which in the oocyte is required for normal embryonic development (Fedoriw et al., 2004; Wan et al., 2008).

Embryonic stem (ES) cells possess the potential to generate all somatic cell types. Differentiation of ES cells serves as a widely used model to analyse specific aspects of embryonic development. Recently, it has been shown that ES cells are highly enriched with the TATA-binding protein associated factor 3 (TAF3) (Liu et al., 2011) and that TAF3 is required for endoderm lineage differentiation. In addition to binding to TFIID at the core promoter (see Glossary, Box 1), TAF3 has been found at a subset of genomic sites marked by CTCF and cohesin. CTCF binds TAF3 directly and mediates DNA looping between distal, CTCF-bound sites and core promoters (Fig. 3B). In this case, CTCF confers gene activation rather than insulation; nevertheless, this example nicely illustrates the idea that CTCF-bound insulators might contact promoters as well (see below).

Embryonic development and ES cell differentiation in mammals requires inactivation of one of the two X-chromosomes in female cells. Major players in this process are two adjacent and head-on oriented genes Tsix and Xist. Xist upregulation, being essential for the inactive chromosome, is suppressed by Tsix activity on the active X-chromosome. CTCF is found to bind to a conserved sequence element positioned between the Tsix and Xist genes. Depletion of CTCF or deletion of the binding site results in aberrant X inactivation in female cells (Spencer et al., 2011). The interpretation by the authors is that the insulator function ensures the differential activity of these adjacent genes.

As discussed above, homeotic genes define segment identity during embryonic development. As observed in Drosophila, loss of CTCF leads to homeotic transformations of the abdominal segments (Mohan et al., 2007). By contrast, in mice, expression of the HoxD cluster that is responsible for limb development remains unchanged after conditional CTCF gene inactivation. However, the structure of the limbs is impaired owing to increased apoptosis (Soshnikova et al., 2010). This is surprising, as CTCF is bound at sites expected to insulate the different homeotic genes from each other. One interpretation is that establishment of the insulator function occurs early in development, before the conditional CTCF deletion was induced.

Earlier work on the transcriptional regulation of the gene coding for amyloid precursor protein (APP), which is involved in Alzheimer’s disease, in rat primary neurons suggests that CTCF is important for the upregulation of APP expression during synaptogenesis (Yang et al., 1999). The CTCF-binding site is located in the promoter, a feature that is found for a substantial number of CTCF-binding sites. Recently, the brain-specific expression of the mouse protocadherin-α (Pcdha) gene cluster has been analysed (Kehayova et al., 2011). A vast number of Pcdh isoforms contributes to the diverse repertoire of neuronal cell-surface identities. A battery of alternative promoters is controlled by a remote enhancer. Both promoters and enhancer are bound by CTCF, which is required for promoter activation (Kehayova et al., 2011). As mono-allelic and combinatorial expression of the alternative promoters is observed, a combination of CTCF-mediated promoter function and insulation could be involved in the selection of promoters in individual neurons.

CTCF is also involved in blood cell differentiation. In maturing human dendritic cells (DCs), CTCF expression is normally reduced, whereas the forced overexpression of CTCF during differentiation of murine, bone marrow-derived DCs leads to reduced proliferation and increased apoptosis (Koesters et al., 2007). During differentiation of B-cells, recombination of the Igh locus occurs, which requires locus contraction and looping. Recent studies concluded that the many CTCF-cohesin sites in the locus are responsible for locus compaction because loss of CTCF was observed to lead to a more loose structure (Degner et al., 2011; Guo et al., 2011a). In addition, immunoglobulin heavy chain (Igh) recombination is also mediated by CTCF (Guo et al., 2011b). Guo and colleagues identified a key Igh recombination regulatory region, which they termed intergenic control region 1. This region harbours two CTCF-binding sites, which, when deleted, result in impaired B-cell development. In particular, these mutations alter the usage of the variable gene segments, induce germline transcription and change the order of rearrangements. The authors propose that these CTCF sites mediate loops with downstream CTCF sites that segregate different heavy chain segments into separate regulatory domains during the heavy chain rearrangement stage of B-cell development. Simultaneously, the enhancer activation of the promoters at the variable segments before rearrangement is blocked (Guo et al., 2011b). Such a dual effect of CTCF in mediating enhancer blocking as well as immunoglobulin rearrangement has been observed for the κ light chain locus also. Here, an alternative strategy has been taken by conditionally deleting the CTCF gene in the B-cell lineage of the mouse (Ribeiro de Almeida et al., 2011). The authors observed that in the absence of CTCF, the Igκ light chain locus showed a change in the usage of Vκ gene segments as well as a change in germ line transcription. Chromosome conformation capture (3C, see Glossary, Box 1) experiments demonstrated that CTCF limits interactions of the Igκ enhancers with the proximal Vκ gene regions. In this system, no effect was seen on the heavy chain locus, which might be explained by an early epigenetic effect by CTCF, which, upon conditional deletion, does not impair the epigenetic marks set by CTCF previously.

Furthermore, T-cell differentiation and maturation also depend on functional CTCF. At the human locus coding for interferon gamma (IFNG), CTCF and cohesin sites form the basis for long-range interactions during differentiation of naive CD4 T cells into T-helper T_H1 cells (Hadjur et al., 2009). Knockout of Ctcf in mouse thymocytes leads to cell cycle arrest, thereby suggesting that CTCF acts as a regulator of cell growth and proliferation in αβ T cells (Heath et al., 2008). In addition, reducing the level of CTCF during differentiation of Th2 cells was found to impair Th2 cytokine expression and T-cell activation (Ribeiro de Almeida et al., 2009). Finally, CTCF was also found to control MHC class II gene expression and to be required for the formation of long-distance chromatin interactions at the MHC class II locus (Majumder et al., 2008).

The Wilms tumor 1 (Wt1) transcription factor regulates mesenchymal-epithelial transition during renal development. The reverse process, epithelial-mesenchymal transition, is essential for generating vascular progenitors in the epicardium. In both cases, Wt1 regulates the wingless-type MMTV integration site family, member 4 (Wnt4) locus. This locus, which is framed by CTCF-binding sites, acquires an active chromatin conformation in kidney cells upon Wt1 action, whereas in epicardial cells Wt1 switches the chromatin architecture into a repressed state (Essafi et al., 2011). The authors call this dichotomous switch a ‘flip-flop’ mechanism. Genes outside the CTCF-marked domain do not change their activities. This suggests that CTCF (and cohesin) provide a boundary function. Additional proof for this function is provided by depletion of CTCF or of cohesin, which results in spreading of the active chromatin architecture in kidney cells into the flanking region, whereas in the case of the epicardium the repressed chromatin extends into the flanking genes (Essafi et al., 2011). This not only highlights the boundary function of CTCF, but also demonstrates that both active and inactive chromatin domains are curbed by CTCF.

Common to many of these and other developmentally regulated genes is the observation that enhancer activity has to be blocked in some tissues, but conferred in others. Consequently, the enhancer blocking activity of CTCF involved in developmental gene regulation should be regulated. DNA binding of CTCF is one known level of regulation, as demonstrated by DNA methylation or transcription-induced CTCF eviction (Lefevre et al., 2008). Other mechanisms involve changes in CTCF activity without interfering with DNA binding, through synergising DNA-binding factors that have been found to modulate CTCF function (Weth et al., 2010). A recent example of insulator modulation is given by a specific CTCF-binding site in the Drosophila genome (Wood et al., 2011). The authors noticed that one of the CTCF co-factors, CP190, was only found at a particular subset of CTCF- or other insulator-binding sites upon treatment with 20-hydroxyecdysone (20-HE), a hormone that binds to the ecdysone receptor complex (ECR-C), which is a classic nuclear receptor complex. At the particular site studied, ECR-C binds upon 20-HE treatment and CP190 is recruited to a CTCF site nearby. This, in addition to a number of constitutive CTCF/CP190 regions, might shield flanking genes from 20-HE action. The authors use 3C-technology to test for long-range chromatin interactions. In fact, chromatin loop formation seems to be slightly increased upon 20-HE addition, indicating a potential increase in insulation.

Finally, post-translational modification of CTCF by phosphorylation (El-Kady and Klenova, 2005) or polyADP-ribosylation (Yu et al., 2004) regulates CTCF function, but precisely how these mechanisms work still remains to be unravelled.

The aforementioned detailed analyses of CTCF function at model gene loci led to the question of whether similar roles for CTCF can be expected at other genomic locations. Genome-wide detection of CTCF-binding sites is possible using ChIP, to detect interactions of chromatin factors or specifically modified histones with DNA in vivo. In recent years, direct sequencing of the precipitated DNA (via ChIP-Seq, see Glossary, Box 1) in order to obtain genome-wide

binding profiles for the factor of interest has become the method of choice (Johnson et al., 2007). CTCF-binding analysis revealed an almost equal distribution of CTCF sites within three classes of binding regions: intragenic, intergenic or in the vicinity of transcriptional start sites (Barski et al., 2007; Guelen et al., 2008; Cuddapah et al., 2009). Genome-wide binding studies to map histone modifications and associated factors for different cell types revealed large blocks of repressed and of active chromatin (Box 3) (Barski et al., 2007; Guelen et al., 2008). Based on their specific composition of associated factors or histone modification marks, these large blocks were further subdivided into specialised chromatin domains (Filion et al., 2010; Roy et al., 2010; Ernst et al., 2011). Combining CTCF binding and chromatin feature analysis revealed that CTCF often demarcates the boundaries of repressed chromatin associated with the nuclear lamina and/or with H3K27me3 adjacent to active regions (Fig. 5) (Barski et al., 2007; Guelen et al., 2008; Cuddapah et al., 2009). These active regions are marked by methylated H3K4 as well as acetylated H2AK5. Importantly, this demarcation of boundaries by CTCF is evolutionarily conserved across species. Despite the fact that there are some general differences in the overall distribution of CTCF profiles in human and Drosophila, similarly to the human system, Drosophila CTCF has been shown to bind to borders of H3K27me3 domains (Fig. 5). (Bartkuhn et al., 2009; Nègre et al., 2010). Additionally, an evolutionary comparison of the CTCF profiles in human, mouse and chicken revealed conserved binding of CTCF to boundaries flanking disease-associated transcription factor genes. This finding suggests a potentially important role for CTCF in diseases associated with the corresponding transcriptional programmes (Martin et al., 2011).

Finally, an additional characteristic of these boundaries is that they are often associated with promoters of actively transcribed genes. A link between promoters and boundary elements or CTCF-binding sites is potentially interesting because these two types of cis-regulatory elements appear to have common features. For example, in Drosophila, the CTCF co-factor CP190 is found at a large number of active promoters. Sites co-occupied by CTCF and CP190 are marked by a prominent loss in nucleosome occupancy (Bartkuhn et al., 2009), similar to human genomic binding sites of CTCF (Song et al., 2011). Furthermore, in human cells, CTCF-bound chromatin is associated with the H2A.Z (H2AFZ – Human Gene Nomenclature Database) and H3.3 (H3F3 – Human Gene Nomenclature Database) histone variants, which are known for their association with active promoters (Jin et al., 2009). Finally, there is evidence for CTCF-binding sites being over-represented for mono- and triple-methylated H3K4. Again, active promoters are marked by H3K4me3, whereas H3K4me1 is preferentially associated with enhancers (Fu et al., 2008). These and other findings led to the hypothesis that insulators might have evolved as specialised derivatives of promoters (Raab and Kamakaka, 2010). Indeed, two of the current models for insulator function are in line with this idea. First, promoter-like chromatin structures might function as enhancer decoys, thereby blocking illegitimate enhancer-promoter interactions. Second, promoters might serve as roadblocks to separate repressed from active chromatin. One idea is that histone depletion, as found in promoters, is incompatible with spreading of heterochromatin marks beyond a gap within a row of nucleosomes (Raab and Kamakaka, 2010).

In addition to controlling the biochemical properties of chromatin, it is evident that one of the major CTCF functions is to guide long-range chromatin interactions, which may occur either intra- or inter-chromosomally (Ling et al., 2006; Splinter et al., 2006) and might be stabilised by CTCF-CTCF self-interactions (Pant et al., 2004). Alternatively, CTCF-driven interactions might be enforced by recruitment of the cohesin complex as many studies have demonstrated strongly overlapping binding profiles between CTCF and cohesin complex members (Parelho et al., 2008; Wendt et al., 2008), and both factors are involved in establishing functional chromatin interactions (Hadjur et al., 2009; Nativio et al., 2009). These and other studies led to speculation that CTCF might be ‘the master weaver of the genome’ (Phillips and Corces, 2009) directing the framework of chromatin interactions in a given cell in order to control numerous downstream processes, such as transcription, replication and repair. This view of CTCF has been strongly supported by a recent study using a genome-wide paired-end tag (PET) approach, ChIA-PET (see Glossary, Box 1). This approach combines an initial chromatin immunoprecipitation with a subsequent 3C-based assay that allows direct sequencing of interactions associated with the immunoprecipitated factor and was recently used to map CTCF-containing chromatin interactions in mouse embryonic stem cells (Handoko et al., 2011). Although far from complete, this method allowed the simultaneous detection of ∼1800 long-range chromatin interactions involving CTCF-binding sites. Strikingly, CTCF-containing loops span and separate H3K27me3 repressive domains from H3K4 active domains, indicating that CTCF-binding sites are required for establishing or maintaining such chromatin domains. As described in a previous study (Guelen et al., 2008), many H3K27me3 domains coincide with lamin-associated domains (LADs), suggesting an active role for CTCF in the demarcation of LADs from active chromatin. Furthermore, this study revealed an interesting new aspect of CTCF function, as many of the identified loops connect a CTCF-binding site with an enhancer, where enhancers are not usually bound by CTCF directly. This finding suggests that CTCF not only blocks enhancer-promoter communications but, by contrast, might play the opposite role, too. As some of the identified interactions between enhancers and CTCF target sites involved CTCF target sites located next to promoters, it is likely that CTCF might function to tether remote enhancers to their cognate target genes.

The question then arises of how much influence CTCF has on determining the higher-order chromatin structure in the cell nucleus. To date, the answer is still not completely clear. However, although it is likely that other factors also contribute to this structural process, the finding that a large number of CTCF-associated interactions are correlated with the underlying chromatin domain landscape as well as with coordinated gene expression is a strong argument to posit CTCF as a ‘genome organiser’.

Throughout the past two decades of CTCF research, a developmental role for CTCF was postulated and was evident from early on. This role has been clearly confirmed by CTCF gene deletion in mice and Drosophila, which results in developmental arrest. Today’s genome-wide mapping of CTCF-binding sites, chromatin features and three-dimensional chromatin contacts support the view that CTCF is a ‘master’ regulator involved in many key developmental processes at both the functional and the molecular level. Currently, it remains to be shown whether the promoter-like features of CTCF-binding sites, as well as its chromatin-looping ability, are related to the barrier function of an insulator and/or to enhancer-blocking activity.

Another open question is that of the regulation of CTCF-mediated insulation. For example, as exemplified with the Bithorax locus, insulation of the Abd-B gene from segment-specific enhancer elements has to be overcome in the particular segment requiring this enhancer activity. Furthermore, there are many such situations, mostly in development, when insulator activity mediated by CTCF has to be regulated. Besides DNA binding, the ability of CTCF to interact with other factors is likely to be regulated by modification of CTCF. The mediators and mechanisms that govern these modes of regulation, however, remain to be revealed.

We thank Reinhard Dammann for critically reading the manuscript and for valuable suggestions.