Insulators: exploiting transcriptional and epigenetic mechanisms

Insulators are DNA sequence elements that prevent inappropriate interactions between adjacent chromatin domains. One type of insulator establishes domains that separate enhancers and promoters to block their interaction, whereas a second type creates a barrier against the spread of heterochromatin. Recent studies have provided important advances in our understanding of the modes of action of both types of insulator. These new insights also suggest that the mechanisms of action of both enhancer blockers and barriers might not be unique to these types of element, but instead are adaptations of other gene-regulatory mechanisms.

The organization of eukaryotic genomes necessarily results in the proximity of domains with distinct functions. A domain containing genes that are transcriptionally active in a particular cell type might lie close to another domain containing genes that are not active. Elsewhere, an active gene might be surrounded by constitutively silenced chromatin structures. To an extent, the identity of these domains is maintained by classical transcriptional regulatory elements, such as enhancers, silencers and upstream activating sequences (UASs). In other cases, however, specific DNA sequences and their associated binding proteins have a role in establishing or maintaining discrete inter-domain boundaries. Such sequence elements have been given the name insulators.
In several cases endogenous insulator sites have been shown to have important functions in the regulation of gene expression. Much of our knowledge about insulator function, however, comes from experiments with transgene constructs. Most transgenes in Drosophila are affected by endogenous enhancers and silencers, a result of the combined effects of a largely euchromatic genome and the method of transgene delivery 1 . By contrast, most transgenes in vertebrates are subject to chromatin-mediated silencing, a reflection of the high heterochromatin content of the genome. The biochemical activities that underlie the protection from these two types of regulatory interference are different, and the elements that are involved have distinct names: those that protect from activation by enhancers are known as enhancer-blocking insulators, whereas those that protect against heterochromatin-mediated silencing are known as barrier insulators 2 . Some compound insulators possess both enhancer-blocking and barrier activities, and enhancer-blocking insulators also protect against certain types of transcriptional repressor.
Here, we discuss both enhancer-blocking and barrier insulators. Many recent studies have been devoted to the identification and characterization of insulators. Among the reasons for this intense activity is the recent shift in emphasis from transcriptional control of gene expression by nearby regulatory elements to a recognition of the role of long-range interactions and three-dimensional organization of chromatin within the nucleus. Insulation has also emerged as a major mechanism for epigenetic control of gene expression, especially at imprinted loci. Studies of insulators have led to new models of their modes of action. These studies also suggest that both kinds of insulator element exploit the function of other regulatory elements within the nucleus. This leads us to suggest that the established distinctions between various regulatory elements in eukaryotes might be too constraining because all such elements -enhancers, silencers, promoters and insulators -make use of a shared set of strategies to regulate transcription, with a considerable overlap of function.
Enhancer-blocking elements Enhancers are typically located some distance upstream or downstream of the genes that they regulate, and serve to raise the level of basal transcription from that gene. Enhancer-blocking insulators are defined by their ability to interfere with enhancer-promoter interactions when placed between the two, but not from a flanking position (FIG. 1). Enhancer-blocking insulators were initially thought to be rare, but recent results, especially in vertebrates, show that enhancer blocking has a crucial role in a wide range of regulatory systems.

Enhancer
A cis-acting regulatory sequence that markedly increases expression of a neighbouring gene. Enhancers are typically capable of operating over considerable distances (sometimes ~50 kb) upstream or downstream of the gene, and in either orientation.

Silencer
A cis-acting regulatory sequence that decreases expression of a neighbouring gene.

Upstream activating sequence
The scs/scs′ paired elements, which flank the endogenous Hsp70 (heat-shock protein 70) locus at cytological position 87A7, are another well-studied insulator system from Drosophila that provides support for the importance of loop domains 10,11 . The proteins Zw5 (Zeste-white 5, also known as Deformed wings) 12 and the A and B isoforms of BEAF32 (boundary-element-associated factor of 32 kD) 13 bind to scs and scs′ respectively. The two BEAF32 isoforms form heterocomplexes at scs′ in vitro 14 , and an interaction between BEAF32 and Zw5 has been shown to stabilize loop-domain formation at opposite ends of the 87A7 hsp70 locus in vivo 15 . Chromatin loop formation that is mediated by scs/scs′, similar to that observed with Su(Hw), is presumably responsible for the enhancer-blocking activity of these elements.
Are similar structures generated by enhancerblocking insulators in vertebrates? The first vertebrate enhancer-blocking insulator to be identified was cHS4, a complex element that combines enhancer-blocking and barrier activity and lies at the 5′ end of the chicken β-globin locus 16,17 (FIG. 3). The powerful enhancer-blocking activity of this element is associated with a strong binding site for CTCF, a ubiquitously expressed vertebrate zincfinger protein 18,19 . Since this initial discovery, there has been great interest in identifying other binding sites for CTCF and understanding the basis of its function.
Fortunately, one of the first searches for other CTCF binding sites indicated a role for CTCF enhancerblocking activity in a known regulatory system, the imprinted Ig f2 (insulin-like growth factor 2)-H19 locus. In embryonic tissues Igf2 is expressed only from the paternal allele, under the control of downstream endodermal enhancers, and expression is correlated with paternal-allele-specific CpG methylation over the imprinted control region (ICR) that lies between the enhancers and the Igf2 promoter. In both mouse and human genomes the ICR contains multiple CTCF-binding sites that function as an insulator, and paternal-specific methylation of these sites abolishes both CTCF binding and insulator activity, which allows imprinted Igf2 expression . These results established a role for enhancer-blocking insulators in the regulation of endogenous loci, which has now been demonstrated at many other sites [23][24][25][26] .
What is the molecular basis of CTCF insulator activity? The same model that is proposed for Su(Hw) in Drosophila seems likely to apply: CTCF molecules can interact with each other to form clusters and therefore generate closed loop domains 27 . It has been proposed that CTCF can also tether the chromatin fibre to the nucleolar surface through interactions with the nucleolar protein nucleophosmin (also known as B23). This would create 'open' loop domains in which CTCF sites contact the nucleolar surface but not each other. Such a model is supported by the finding that multicopy, stably integrated cHS4 transgenes are localized to the nucleolar surface in a CTCF-dependent manner. It should be emphasized that the essential property of this model of enhancer-blocking insulators is the ability to position the enhancer and promoter in separate domains. An enhancer-blocking insulator is expected to interfere with enhancer-promoter communication in a position-dependent manner. An enhancer-blocking insulator will block transcriptional activation only when it lies between a promoter and an enhancer (as in the case of promoter 1); in other situations (such as for promoter 2), activation is not blocked. A transcriptional repressor, by contrast, would reduce the level of transcription from both promoters when placed in the same position. In the example shown, the test construct contains both the experimental enhancer-promoter-1 and control enhancer-promoter-2 pairs; in other cases they can reside in separate transgenes.

How do insulator-induced domains prevent enhancer action?
Understanding how enhancer-blocking insulators work requires knowledge of how enhancers work, and vice versa. In the past it has been difficult to explain the position dependence of enhancer-blocking insulators: unlike silencers, they exert their effect only when they lie between the promoter and enhancer. The insulator-mediated loop-domain model has led to two possible explanations of the source of this position dependence, each based on a specific model of enhancer action (FIG. 4). First, enhancers might function by directly interacting with their designated promoters (the direct-contact model). In this case, enhancer-blocking insulators might have a steric effect that prevents enhancers from contacting other promoters, either by favouring intra-loop enhancer-promoter interactions or preventing inter-loop contacts. Alternatively, there could be an activating signal that travels processively from enhancer to promoter (the tracking model of enhancer action). This signal could be, for example, a helicase complex that modifies histones or alters nucleosome structure, or it could be RNA polymerase itself, launched from the enhancer. This signal could be blocked by an enhancerblocking insulator as it tries to traverse the nucleoprotein structure at the base of the loop that the insulator generates. An extensive body of literature is concerned with the mechanism of action of UASs and enhancers 28,29 .
Evidence exists for both direct-contact and tracking models of enhancer action, which need not be mutually exclusive.
Although it is fairly simple to imagine how a tracking signal might be stopped by an insulator complex, it is more difficult to visualize how chromatin loop formation can influence direct contact between enhancer and promoter. A potential mechanism has been described in studies of the direct contact between the Escherichia coli glnAp2 promoter (a downstream promoter in the gene that encodes glutamine synthetase, glnA) and an enhancer that is dependent on the nitrogenregulation protein NtrC (also known as GlnG), which were separated by 2.5 kb on a closed circular supercoiled template 30,31 . Expression was inhibited when the template was relaxed, or when the lac repressor, LacI, bound at sites that are present on either side of the elements, dimerized to place the enhancer and promoter in separate loops. The data support a model in which enhancer action depends on direct contact between the enhancer and promoter that is achieved by a slithering mechanism in which the interwound supercoil explores various conformations until contact is made. Relaxation or fixation of the DNA structure by the LacI interactions is postulated to prevent this contact. A related mechanism has been suggested in eukaryotes from a study that used an SV40 minichromosome carrying an enhancer and enhancer-blocking elements 32 .
Enhancer blocking and chromatin hubs. These traditional views of enhancer-promoter communication by direct contact, which are focused on the regulation of single genes, might need to be modified in light of recently proposed models for the regulation of eukaryotic gene expression 33,34 . Studies that were initially carried out at human and mouse β-globin loci showed that promoters, gene-proximal enhancers and farupstream activators -which can be separated by many kilobases -tend to be co-localized within the nucleus in so-called chromatin hubs. The genes that are controlled by these elements are transcribed when the hubs make contact with RNA polymerase II molecules, which are distributed as multimolecular aggregates 34,35 within the nucleus that form 'factories' for transcription. In recent studies, interactions that are even more distant have been detected at these factories, both within and between chromosomes 34,36,37 .
How do active promoters find their way to these hubs? Although they might do this simply by a process of random encounter over large distances (direct contact), West and Fraser 38 have suggested that distant )-containing nucleoprotein complexes that are located at distant chromosomal binding sites. As a consequence of the formation of these bodies, the chromatin fibre that separates individual complexes is sequestered into a loop. Insulator bodies are preferentially located at the nuclear periphery because of their binding of nuclear lamin. When considering a mechanistic role for insulator bodies, the following aspects need to be considered: the topological consequences of loop formation, which might limit contact between an enhancer and promoter; preferential localization at the nuclear periphery, which could in some cases help to stabilize loop formation; and the nucleoprotein complex at the base of the loop, which is not restricted to Su(Hw) and associated factors, but also includes activities that are recruited by sequence elements near the chromosomal Su(Hw)-binding sites. Not all insulator bodies are the same; some protein components that are depicted in the figure are associated only with a fraction of them. It is not known whether this heterogeneity has any functional consequences. CP190, centrosomal protein 190 kD; Mod(mdg4), modifier of mdg4; Topors, topoisomerase-I-interacting protein. enhancers and promoters might be brought together by action of the RNA polymerase in the course of transcribing the intergenic regions that lie between the enhancer and the promoter. Intergenic transcripts, which were first described some years ago 39,40 , are well documented; in this new view they are by-products of a mechanism that enables contact between promoters and even very distant enhancers.
As the polymerase is fixed in the factory it would pull the promoter towards it as it transcribes, in contrast to simpler tracking models in which the polymerase moves freely along the DNA from enhancer to promoter. In either case the presence of an enhancerblocking insulator could interfere with tracking or with the architecture of the chromatin hub in a variant of the direct-contact model. The model that is described above is only one example of the more general class of tracking mechanism that can be envisioned. A variant that has been discussed at length elsewhere 41 proposes a role for the Drosophila protein Chip in stabilizing the binding of certain homeodomain proteins to distributed sites in the region between promoter and enhancer, creating clusters that bring enhancers and promoters closer together 42,43 . It was suggested that Su(Hw) sites would interfere with this process. As has been pointed out 41 , the clustering process must be processive to account for the enhancer-blocking insulator properties of Su(Hw). The best direct evidence for tracking comes from a study of CTCF. This insulator protein can block the advance of RNA polymerase II, as has been shown using stably replicated minichromosomes that carry the cHS4 insulator: the polymerase accumulates upstream of and within the insulator 44 .

Challenges to simple models of enhancer-blocking action.
It is not known whether hubs exist in Drosophila, but the more general tracking and direct-contact models that are discussed above still apply. Both of these possibilities, however, seem too simplistic because neither one can explain all of the properties of the gypsy element. For example, the tracking model must confront the fact that gypsy elements that are inserted into an intron of yellow effectively block activation by an enhancer that is located further downstream, but do not interfere with the action of enhancers that are upstream of the promoter (FIG. 5a). If the anchoring complex always blocked a tracking polymerase, it is hard to see why it would not block formation of the yellow transcript at those developmental stages in which the upstream enhancers were active. One possibility is that polymerases that originate from enhancers are more easily blocked than ones that are initiated at promoters; in support of this, it has been shown that the insulator strength of a gypsy element is inversely proportional to the strength of an upstream enhancer 45 .
A second difficulty for both tracking and directcontact models is raised by the striking observation that when two gypsy elements are introduced tandemly, the enhancer-blocking activity is nullified 46,47 (FIG. 5b). It has been suggested that the adjacent clusters of Su(Hw)-binding sites interact with each other to form a micro-loop, therefore preventing them from interacting with more distant sites to form a large loop domain. It is not clear, however, why such a micro-loop would not form an effective block against a tracking polymerase. Nor is it clear, in terms of the direct-contact model, why such a micro-loop would not join a cluster of other gypsy sites to maintain the domain organization. The situation is further complicated by the findings that the insertion of a third gypsy element between the enhancer and promoter restores enhancer-blocking activity in some cases but not others 48,49 At least in the case of gypsy, other structures or components must be involved in establishing an active insulator site.
Could the putative mechanisms of enhancer-blocking insulator action that are described above also result in interference with certain types of repressor activity, as well as with activities that stimulate gene expression? The models that have been suggested could account for the functions of those repressors in which a distal silencer has to make physical contact with the promoter in order (for example) to deliver a histone deacetylase. In vertebrates, however, the preponderance of silencing activities arise differently -from the expansion of heterochromatin into surrounding areas. This type of silencing is not affected by enhancer-blocking insulators, but is prevented by barrier insulators.

CTCF -a context-dependent enhancer-blocking insulator?
Given that CTCF often binds to regions of the genome that are adjacent to binding sites for other regulatory factors, it would not be surprising if its function were context dependent. Like many regulatory proteins, CTCF also is the target of modifications that can affect its properties. For example, it can be poly(ADP-ribosyl)ated, Figure 3 | The chicken β-globin locus. In chicken (Gallus gallus), the 5′ HS4 and 3′ HS insulator elements define the limits of a chromatin domain that encompasses the developmentally regulated β-globin gene cluster and its locus-control region (LCR), which is comprised of the HS1-3 and β A/ε enhancers. This domain is flanked by a region of condensed chromatin and a cluster of chicken olfactory receptor genes (CORs) at its 5′ and 3′ ends, respectively. The HS4 element possesses both enhancer-blocking and barrier activity, presumably to prevent the LCR from inappropriately activating genes outside the domain and at the same time protecting the globin cluster against silencing that emanates from the flanking condensed-chromatin region. Enhancer blocking is mediated by CTCF, whereas barrier activity results from the combined effect of USF1 and USF2 and the as yet uncharacterized FI-, FIII-and FV-binding proteins. 3′ HS binds CTCF and functions only as an enhancer-blocking insulator. and inhibition of this modification impairs its ability to function as an insulator 50 . Importantly, CTCF can also function as a classical transcription factor, although this behaviour appears to be restricted to certain sites (see, for example, REF. 51). Because CTCF is an 11-zincfinger protein for which different binding sites in DNA engage different subsets of fingers 52 , with the potential of exposing different regulatory surfaces on CTCF, its stimulatory effect on transcription could be unrelated to its enhancer-blocking function. On the other hand it is quite possible that under some circumstances the ability of CTCF to form or enter chromosomal clusters could allow it to bring promoters close to active transcription hubs. This raises the possibility that enhancer-blocking activity is only one manifestation of more general mechanisms that bring genes and regulatory elements together at multi-component sites within the nucleus.

Barrier elements
Barrier insulators protect against position-effect variegation (PEV), which is the stochastic, meta-stable and heritable silencing of a euchromatic gene through the spread of heterochromatin formation (FIG. 6). Barrier elements have been isolated from several organisms (for a partial list, see REF. 41), and recent studies have led to a much more detailed understanding of their molecular activities that sheds light on the cellular mechanisms that are used to maintain the epigenetic characteristics of chromatin domains.

Heterochromatin and euchromatin.
Barrier activity can only be discussed in the context of heterochromatin and euchromatin. Biochemically, heterochromatin is the more condensed form of chromatin, as demonstrated by its reduced sensitivity to nuclease digestion, which is a reflection of the positioning of nucleosomes at regular, short intervals. Characteristic histone modifications that are seen in heterochromatic regions are high levels of methylation at the histone H3 Lys9 (H3K9) and Lys27 (H3K27) residues, combined with a lack of acetylation marks; heterochromatin is also marked by the presence of heterochromatin protein 1 (HP1). Heterochromatic DNA in vertebrates, as well as in plants, also shows extensive CpG methylation. Recent advances have provided insight into the biochemistry of initiation and maintenance of heterochromatic structures (REF. 53). At the centre of both processes is a self-perpetuating cycle of reactions: methylation of H3K9 leads to the HP1-mediated recruitment of additional histone methyltransferase (HMT) activity. Two pathways for targeting heterochromatin formation have been described. The first targets H3K9 methylation Figure 4 | Models for enhancer-blocking activity. a | One set of models focuses on the formation of topologically closed looped chromatin domains. They posit that enhancer-blocking elements and the enhancer-blocking (EB) proteins that bind them prevent enhancer-promoter communication by partitioning the promoter and the enhancer that it is to be shielded from into separate looped domains (here, promoter 2 is shielded from the enhancer (E), whereas promoter 1 is not). An assumption of these models is that the frequency of intra-loop enhancer-promoter interactions is higher than that of inter-loop interactions; this can be achieved by the existence of a mechanism that either facilitates intra-loop interactions or inhibits inter-loop interactions. The dashed double arrows mark enhancer-promoter interactions. Arrow thickness denotes the probability of the interaction b | A second group of models postulates that enhancer blocking stems from the targeting of a specific nucleoprotein complex (for example, an insulator body) to the region of the chromatin fibre that separates the enhancer and promoter (again, promoter 2 is shielded from the enhancer, whereas promoter 1 is not). Inherent to tracking models is the idea that transcriptional activation involves the processive transfer of a signal from the enhancer to the promoter. According to these models, specific interactions between the activation signal and the enhancer-blocking complex disrupt the transfer. The models can accommodate both a fixed chromatin template combined with a mobile signal (for example, RNA polymerase tracking along chromatin) and a mobile template with a fixed signal (for example, specific regions of chromatin moving into RNA polymerase II transcription factories). to repetitive sequence elements using small RNA molecules that are complementary to the target 54 . The second relies on sequence-specific DNA-binding proteins to deliver HMT activity to specific genomic locations 55 . The same biochemical cycle has been invoked to explain the spreading of heterochromatin: H3K9 methylation of nucleosomes near the initiation site leads to an extension of the HP1-bound domain (FIG. 7a). This pathway of heterochromatin formation is conserved in most organisms. A notable exception is S. cerevisiae, which uses a biochemically different but conceptually similar pathway for chromatin-mediated silencing 56 . Yeast heterochromatin consists of deacetylated nucleosomes that are spaced at short, regular intervals. It is restricted to the HML and HMR silent mating loci, telomeres and rDNA repeats, and PEV is experienced by transgenes that are inserted within or near silenced chromatin regions. Silenced chromatin formation at HML and HMR is initiated by the E and I silencers, at which the binding of sequencespecific factors recruits the Sir2 histone deacetylase. Deacetylation of histone tails by Sir2 promotes the binding of the Sir3-Sir4 complex to the nucleosome, which subsequently recruits additional Sir2 molecules, stabilizing yeast heterochromatin and allowing it to spread.
Euchromatin comprises the transcriptionally active portions of the genome in which DNA is more accessible to nucleases and nucleosomes are irregularly spaced. A characteristic feature of euchromatin is the presence of nuclease-hypersensitive sites that mark the presence of sequence-specific DNA-binding proteins. Nucleosomes within euchromatin carry a combinatorial pattern of many post-translational modifications, which include high levels of acetylation and methylation of H3K4 and H3K79. Euchromatin formation is aided by processes that are associated with the activation of transcription. These include various histone modifications and nucleosome remodelling, as well as deposition of histone variants. Unlike heterochromatin, euchromatin probably does not spread through a linear polymerization-like process but instead occurs through destabilization of heterochromatic structures.

Barrier activity: breaking the nucleosome chain.
Given the array of genetic tools available, it is not surprising that the most detailed characterization of barrier activity has been carried out in yeast cells. Most of these experiments have used constructs that are based on the simple architecture of a silencer (an E or I element) that is separated from the reporter transgene by a putative barrier, and screens have been carried out to identify sequence elements 57 and enzymes that are associated with barrier activity 58 . Screens for enzymes have made use of a GAL4 DNA-binding-domain fusion library and four GAL4-binding sites in the barrier position. These experiments linked barrier activity to the localized disruption of the polymerization-like reaction cycle at the heart of heterochromatin spreading. Barriers function as chain terminators by modifying the nucleosomal substrate of this processive reaction (FIG. 7b). The most extreme modification of the template is nucleosome removal; various nucleosome-excluding sequence elements were shown to disrupt the spread of chromatin-mediated silencing 59 . Other forms of modification are achieved through the targeted recruitment of histone acetyltransferase (HAT) and ATP-dependent nucleosome-remodelling complexes 58 . Whether barrier activity is only associated with a specific subset of these enzymes remains to be seen. As discussed below, both nucleosome exclusion and the recruitment of histone-or nucleosome-modifying complexes have important roles at endogenous yeast barrier elements.  Hw)) and gypsy activity are not explained by current enhancer-blocking models. a | An intronic gypsy enhancer-blocking insulator blocks enhancers that are located in the 3′ direction from communicating with the promoter. It does not, however, prevent the passage of the transcribing RNA polymerase II (Pol II) complex in the forward direction. If the enhancer-blocking insulator is blocking a processive signal that originates at the downstream enhancer, why does it not also block the polymerase? b | Although a single copy of a Su(Hw)-gypsy insulator blocks enhancerpromoter communication (top), two copies inserted between the enhancer and the promoter cancel enhancer blocking (middle) 46,47 . The proposed explanation is that the elements participate in a non-productive, local micro-loop formation instead of targeting the region to an insulator body. The validity of this explanation was challenged by the finding that insertion of a third copy of a Su(Hw)-gypsy element restores enhancer blocking (bottom) 48 . A more recent report suggests that the non-productive selfinteraction of three Su(Hw)-gypsy elements is influenced by their relative spacing, as well as the presence of transcriptional control elements in the space that separates them 49 . It seems that the models for enhancer blocking will need further refinement to accommodate all the new data. The behaviour described above might be specific to the Su(Hw) elements: by contrast, two copies of the scs element inserted between an enhancer and a promoter do not abolish its enhancer-blocking activity 48 .

Nuclear pore complex
A large multiprotein complex that forms a channel in the nuclear envelope of eukaryotic cells. It joins the inner and outer nuclear membranes and allows transport of proteins and nucleic acids to and from the nucleus.
Studies of cHS4, the complex vertebrate insulator that we discussed above, support the idea that a similar molecular mechanism has an important role in barrier activity in higher eukaryotes, although different enzymes are involved. cHS4 lies at the extreme 5′ end of the chicken β-globin locus 16,60 , and has both enhancerblocking 18 and barrier activities 17 . Whereas enhancer blocking by cHS4 is mediated by the CTCF protein, its barrier activity is independent of this protein 61 . The chromosomal position of cHS4 is marked in all cell types by a nuclease-hypersensitive site and a peak of euchromatin-specific histone modifications (histone acetylation and H3K4 methylation) 62,63 . The peak of histone modifications at cHS4 is due to recruitment of HATs and HMTs by the upstream transcription factor 1 (USF1) and USF2 sequence-specific DNA-binding proteins 64 . cHS4 mutations that disrupt the binding of USF1 and USF2 not only eliminate the recruitment of HATs and HMTs, but also abolish barrier activity. These findings led to the proposal that cHS4-mediated acetylation and H3K4 methylation of nucleosomes renders them resistant to H3K9 methylation and HP1 binding and therefore stops the spread of heterochromatin formation. Similar mechanisms could be based on propagation of other heterochromatin-associated histone modifications and their co-factors. It is not yet known whether all the histone modifications that are seen at cHS4 contribute equally to the barrier activity.

Can barriers function through subnuclear targeting?
The idea that higher-order chromatin structures, especially chromatin loops, can limit the spread of heterochromatin has been discussed for some time. Interest in this topic was reinvigorated by two recent reports that link barrier activity and the ability to anchor the chromatin fibre, either through binding to the nuclear pore 65 or homotypic protein-protein interactions 66 . These reports indicate that anchoring limits heterochromatin spreading through a molecular mechanism that is different from the one described in the previous section, although no specific mechanism has been proposed. The data, however, are consistent with the possibility that the weak barrier activities of the elements that were investigated in these studies result from a combination of nucleosome exclusion and targeting of the reporter to a subnuclear compartment 67 . For example, targeting to the nuclear pore complex would provide an environment that is unfavourable to chromatin-mediated silencing because of the known high concentration of transcriptional activators that favour euchromatin formation.
Can enhancer-blocking proteins also function as barriers by tethering a locus to a subnuclear compartment that is unfavourable to heterochromatin formation or spreading? Su(Hw) has been reported to partially protect transgenes from heterochromatin-mediated silencing in D. melanogaster 68 . The molecular mechanism of this activity is unknown; however, it is tempting to speculate that the demonstrated ability of Su(Hw) to target the chromatin fibre to insulator bodies is involved. In vertebrates, there are no reports so far of CTCF directly protecting a locus against heterochromatin-mediated silencing. In fact, CTCF fails to function as a barrier in a chicken cell-based transgene assay 61 . cHS4, and therefore CTCF, also fails to protect a transgene that is located on the inactive mouse X chromosome from being silenced 69 . There is much continued interest in this topic, however, given the recent identification of novel CTCF-binding sites at or close to transition points between silenced and active chromatin structures. One study describes such sites at the 5′ end of genes that escape silencing on the inactive X chromosome 70 . Another, discussed above, shows that CTCF can selectively block elongation of a transcript that is initiated at an enhancer element 71 . Neither report, however, provides evidence that CTCF directly limits the spread of heterochromatin, although this might ultimately prove to be true. tRNA genes can function as barriers. Studies that show that tRNA genes can function as a barrier further challenge the concept that barrier insulation is a property of a distinct class of unique regulatory element. Sir3, a marker for silenced chromatin, is localized to a ~4 kb region that encompasses the silent α genes and the flanking E and I silencers at the S. cerevisiae HMR locus 72 . Transition between silenced and active chromatin takes place over a ~1 kb region at the centromere-distal end of HMR 73 . Using a transgene assay, barrier activity within this fragment was mapped to a unique tRNA Thr gene, with a minor contribution from flanking sequences 74 . Deletion of conserved promoter elements abolished barrier activity, and mutations in the trans-acting factors TFIIIB and TFIIIC (which are basal transcription factors for RNA polymerase III) had the same effect. The tRNA Thr barrier was also sensitive to mutations in a subset of genes that encode HATs. The available data support the hypothesis that high-level transcription of the tRNA promoter is necessary for its boundary activity. A possible explanation for this requirement comes from the observation that high transcription levels lead to the formation of a nucleosome-free gap at the promoter. A barrier insulator protects a transgene against heterochromatin-mediated silencing in long-term tissue culture assays. The graphs show transgene activity over time in assays of this type. The barrier must flank both 5′ and 3′ ends of the construct as a randomly integrated transgene might experience heterochromatic encroachment from either direction. Yeast experimental constructs are an exception to this rule. They contain a single barrier that is positioned between the reporter gene and the known source of chromatin-mediated silencing activity. A recent study analysed the effects of specific mutations on the in situ activity of the tRNA Thr barrier at HMR 75 . Single mutations that affected a HAT or the tRNA Thr promoter alone did not result in increased spreading of Sir3 protein beyond the wild-type boundary. Disruption of the barrier required a combination of two mutations: one in the tRNA Thr promoter to eliminate the nucleosome-free gap and a second one in either the ada2, eaf3 or sas2 genes that encode histone acetyltransferases. There is no report of these HATs being specifically recruited to tRNA promoters, so it is more likely that they contribute to barrier activity through increasing global acetylation levels. Reduced global acetylation levels in mutant strains increase the stability of silenced chromatin structures by lowering the Sir2 deacetylase activity that is required to maintain the acetylation-free status of a nucleosome.
tRNA genes also serve as barriers in other organisms. High-resolution analysis of chromatin architecture in Schizosaccharomyces pombe showed that the transition points between active and silent chromatin regions at the centromeres co-localize with a cluster of tRNA genes 76,77 . A recent study demonstrates that an active tRNA gene can function as a barrier to heterochromatin spreading in S. pombe, although the mechanism of action is not yet known 78 . Interestingly, deletion of the tRNA barrier not only disrupts wild-type chromatin organization at the centromere but -presumably as a consequence of this effect on chromatin -also leads to abnormal centromere function. The ability of barrier insulators to determine heterochromatic boundaries can therefore affect large-scale chromatin organization as well as local transcriptional activity. RNA polymerase III promoters (which transcribe tRNA) can also function as barriers in vertebrates, in which they have been implicated in the barrier activity of Alu elements that flank the human gene that encodes keratin 18 79 .
A recent study has shown that clusters of Box B binding sites for TFIIIC that flank the silent mating locus of S. pombe can function as barriers in the absence of additional Pol-III-promoter-associated factors. This report provides further insight into the potential molecular mechanisms of barriers 80 . Genome-wide chromatin immunoprecipitation analysis revealed the existence of a number of TFIIIC-bound Box B clusters (chromosomeorganizing clamp or COC sites) that are not associated with RNA polymerase III. Microscopic examination of S. pombe nuclei showed that distant COC sites come together to form a limited number of clusters at the nuclear periphery. How does the formation of this structure, which has a striking similarity to Su(Hw)-mediated insulator bodies, facilitate barrier activity? The answer might come from the observation that a large fraction of COC sites are located at the 5′ ends of highly active, divergently transcribed gene pairs. Perhaps the targeting of COC sites that flank the silenced MAT locus (which controls mating type) to a subnuclear region with high transcriptional activity prevents further spreading of heterochromatin-associated structures. Further experiments will be required to determine whether TFIIIC, when it functions alone, acts through the same mechanisms as intact tRNA gene barriers.

Transition without a fixed barrier.
Studies of the fourth chromosome in Drosophila revealed an organization that does not seem to depend on fixed barrier elements to separate euchromatic and heterochromatic regions. Many fourth-chromosome-linked lines were generated using a transgene construct that was specially designed to report on the chromatin status at its site of insertion 81 . Analysis of the lines showed that the gene-rich fraction of the fourth chromosome has interspersed Figure 7 | A local balance of activities determines the extent of heterochromatin propagation. a | A predominance of heterochromatin-promoting enzymatic activities (for example, histone deacetylases (HDACs) and histone methyltransferases (HMTs) that promote methylation at lysine 9 of histone 3 (H3K9) and H3K27) combined with high levels of heterochromatin-specific structural proteins (for example, heterochromatin protein 1 (HP1)) leads to the propagation of heterochromatin. High local levels of enzymatic activities that are linked to euchromatin or a depletion of heterochromatin components, on the other hand, prevent further spreading into euchromatic regions. b | Barrier insulators change the local balance by recruiting (directly or indirectly) euchromatin-promoting enzymatic activities (histone acetyltransferases (HATs), H3K4 and H4R3 HMTs and ATP-dependent chromatin-remodelling enzymes). They might also work through tethering the chromatin fibre to a subnuclear compartment, the protein composition of which is unfavourable to heterochromatin formation, or by disrupting heterochromatin propagation by displacing nucleosomes (not shown). Ac, acetyl group; Me, methyl group.

Silenced promoter
Active promoter heterochromatic and euchromatic regions, with some of the genes located in the heterochromatic parts 82 . A similar arrangement is likely to be present in the pericentric region of other chromosomes. When considering the implications of this result one needs to keep in mind that in the specific tissues in which the endogenous genes are active the chromatin organization might be different from that of the eye, which is the target tissue of the reporter transgene.
Results from genetic manipulation of individual transgenes lead to the conclusion that, on the fourth chromosome, the boundary between heterochromatin and euchromatin is not regulated by fixed barriers 83 . Instead, it is determined by the local balance between the strength of activities that promote either heterochromatin or euchromatin formation (FIG. 8). As these activities show slight cell-to-cell variation, so does the position of the boundary, which leads to the variegated expression of a transgene that has been inserted into the region of transition. The proposed initiation site of heterochromatin formation in this case is the 1630 subclass of repetitive elements; transgenes that are inserted within ~10 kb of a 1630 element are subject to PEV. Spreading of heterochromatin beyond the ~10 kb limit must be curtailed by activities that are associated with euchromatin formation. These heterochromatin-limiting 'positive' signals are likely to be a combination of the global activity and targeted recruitment of histone-and/or nucleosome-modifying enzymes. Presumably none of the individual local-recruitment elements is strong enough to be dominant over heterochromatin spreading in a way that would allow it to behave as a fixed barrier.
The picture that has emerged from various closely related fields of chromatin research shows that the epigenetic state of a given genomic region is determined by the local balance between opposing activities that promote heterochromatin and euchromatin formation 84 . These activities include both those that affect global components (such as HP1 protein levels) and targeted modifications (for example, HATs or HMTs that are recruited by USF1 or USF2 to cHS4), and can function either to stabilize or destabilize euchromatin or heterochromatin. Tight regulation of the chromatin state seems to be limited to genomic regions in which the presence of one form is clearly desirable. Examples include not only the formation of euchromatin at active genes and their regulatory regions but also the maintenance of heterochromatin at pericentromeric regions and repressed genes (for example, silenced copies of mating-type genes at the HML and HMR loci). A sharp transition between heterochromatin and euchromatin is observed only in those cases in which two tightly regulated regions are juxtaposed (for example, the chicken β-globin locus). In these situations the transition point is determined by the position of a barrier that functions as a dominant recruitment site for euchromatin-promoting activities. In other cases the transition seems gradual when examined with experimental methods, such as chromatin immunoprecipitation, that reflect the average properties of a cell population (for example, the S. cerevisiae HML locus) 72 . The apparent gradual transition is probably a consequence of cell-to-cell variation in the extent of heterochromatin spreading. Even in these cases weak euchromatin-recruitment sites have a role; they are not strong enough, however, to serve as a defined barrier in all cells.
Rethinking the definition of insulators As we have described, recent experiments strongly support the existence of a connection between enhancer-blocking insulation and the organization of chromatin structure within the nucleus. Despite this, none of the plausible models that arise from these observations is entirely satisfactory in explaining how enhancer-blocking insulation works, perhaps because more than one mechanism might be involved, both among and within organisms. By contrast, the mechanisms for barrier-insulator function seem to share the more obvious common shown schematically for the same chromosomal region in six genetically identical diploid cells from the same tissue. In the absence of a dominant barrier, the exact transition point between heterochromatin (orange) and euchromatin (green) is determined by a chromosomespecific balance of activities. Just as this balance shows chromosome-to-chromosome variation, so does the point of transition. When analysed with chromatin immunoprecipitation, the region of transition shows a gradual change from heterochromatin to euchromatin. The bimodal nature of the transition can only be revealed by methods that report on the status of individual cells; a gene that is located in the region of transition will have a variegated phenotype.
Locus-control region (LCR). Originally defined as a cis-acting sequence element that confers tissue-specific, copy-number-dependent expression on a transgene. Molecular dissection of some LCRs showed them to be composite structures that are comprised of transcriptional activators and insulator elements.
theme of maintaining histone modifications at the boundary that are associated with 'active' chromatin. For both types of insulator, these mechanisms seem to reflect not a unique, highly specialized apparatus that is devoted to insulation, but rather a modification or extension of existing regulatory elements.
This blurring of the divisions between the mechanisms of insulators and other regulatory elements is quite clear in the case of barrier insulators, which recruit a wide variety of histone-modifying factors (HATs and HMTs) of a kind that are also found at enhancers. Strong enhancers might to a greater or lesser extent also confer protection against position effects, presumably through maintenance of positive histone modifications. Barrier insulators differ from enhancers in that barrier elements lack the ability to work as activators in transient transfection experiments. We note, however, that locus-control regions (LCRs), which can confer strong position-independent expression on transgenes, consist of multiple regulatory regions, some of which might function in transient assays, and others only when stably integrated into the genome. We suggest that the latter elements might have barrier-insulator properties. Whether or not this is the case, it is clear that barrier insulators and enhancers share many of the same factors and modes of action.
A similar argument can be made about the relationship of enhancer-blocking insulators to other types of transcriptional regulator and chromatin architectural element: as we noted above, the loop architecture that is connected with this kind of insulator action might be one specialized application of a more general set of regulatory mechanisms that assist in bringing together distant regulatory elements and genes (for example, in chromatin hubs), and stabilizing interchromosomal interactions. Proteins such as CTCF would be well suited to a role in these mechanisms, which would be quite different from their modes of action at enhancer-blocking insulators.
It is clear that insulators of both kinds share the same bag of tricks with other regulatory elements -enhancers, silencers and LCRs -which they combine in various ingenious ways to acquire specific properties. Although it is still useful to maintain the different categories of regulatory element, it should be kept in mind that there might be considerable overlap in their functions in vivo. As with the other elements, we should not let our classification schemes for insulators obscure our understanding of their potential versatility in controlling chromatin organization within the nucleus.