Nucleosome remodelling is an essential principle to assure that the packaging of eukaryotic genomes in chromatin remains flexible and adaptable to regulatory needs. Nucleosome remodelling enzymes spend the energy of ATP to alter histone–DNA interactions, to catalyse nucleosome displacement and reassembly, on histone exchange and on the relocation of histone octamers on DNA. Despite these dynamics, chromatin structures encode ‘epigenetic’ information that governs the expression of the underlying genes. These information-bearing structures must be maintained over extended periods of time in resting cells and may be sufficiently stable to resist the turmoil of the cell cycle to be passed on to the next cell generation. Intuitively, nucleosome remodelling should antagonize the maintenance of stable structures. However, upon closer inspection it becomes evident that nucleosome remodelling is intimately involved in the assembly of stable chromatin structures that correspond to functional states. Remodellers may even contribute structural information themselves. Their involvement can be seen at several structural levels: at the levels of positioning individual nucleosomes, homoeostasis of linker histones, histone variants and non-histone proteins, as well as the differential folding of the nucleosome fibre. All of them may contribute to the assembly of heritable epigenetic structures.
Introduction: building structures on shaky grounds
It is probably one of the most central themes in molecular biology to understand how information is encoded in molecular structures and how this information is read and translated into cellular functions. In the context of epigenetics, chromatin turned out to provide a multi-facetted information system ‘on top of’ DNA-encoded information that may become very stable and may even be passed on to the next cell generation. How stable structures – and the information they encode – are maintained in a nuclear milieu whose main feature is to be dynamic and adaptable is one of the interesting questions of current research. The dynamic aspects of chromatin organization are mediated to a large extent by nucleosome remodellers. Even in the most simple eukaryotic cells there is a great variety of these molecular machines that couple ATP hydrolysis to energy-consuming alterations of histone–DNA interactions, which in turn may lead to the assembly, relocation and disassembly of nucleosomes, or to the exchange of histones . Intuitively, remodelling factors should oppose the assembly of stable information-bearing structures. At closer range, however, one realizes that nucleosome remodelling is fundamentally involved in the assembly of all chromatin structures in vivo. In this chapter we highlight the roles of remodellers in generating epigenetically stable chromatin states and consider whether they themselves contribute structural information.
The interpretation of genetic information by regulators and by the transcription machinery is a matter of direct DNA binding. Therefore the molecular correlate to chromatin-encoded information boils down to the degree of accessibility of a particular DNA sequence. This accessibility is influenced at several levels of organization. The packaging of DNA into nucleosomes usually renders DNA much less accessible, although some ‘pioneer’ DNA-binding factors may prefer or even require the bent and rotationally phased arrangement of nucleosomal DNA for binding. It is decisive for the access to genetic information where nucleosomes are positioned and where DNA regions are left nucleosome-free. Access is also modulated by the folding of nucleosomal arrays into a series of fibres approximately 30 nm wide. The compaction at this level can be tuned by the spacing of nucleosomes, by post-translational histone modifications (see Chapter 3) and by the replacement of canonical histones with histone variants that affect the bridging interactions between close-by nucleosomes (see Chapter 5). Interactions of dedicated non-histone proteins with (modified) nucleosomes, such as the heterochromatin protein HP1 or polycomb proteins, or indeed nucleosome remodelling enzymes, may be involved in setting up specific chromatin states (see Chapter 3). These may be locally confined and effective, for example at a promoter region, but may also extend over large chromosomal domains. The heterochromatic domains of the inactivated X-chromosomes in female mammalian cells may serve as an example of the latter case.
If one considers the transition from one chromatin state to another in terms of a simple reaction diagram (Figure 1), then the direction of the transition should be determined by the relative thermodynamic stability of the chromatin states (transition from state A to B). It seems clear that structural alterations, like histone modifications, histone variants or associated protein or RNA factors have the potential to significantly alter the thermodynamic stability of chromatin states (state A compared with A′ in Figure 1). Such alterations, therefore, may carry epigenetic information in the sense that they stabilize a certain chromatin state, which in turn correlates to a certain degree of DNA accessibility. What is the role of nucleosome remodellers in this scheme? Do they just catalyse the transition between structural states with no ‘sense of direction’ or do they also contribute a layer of epigenetic information?
Nucleosome remodelling may both mediate the transition between chromatin states as well as influence the states themselves
Groundwork: remodellers and nucleosome positioning
One primary mechanism by which remodelling factors keep chromatin flexible is to move histone octamers on DNA . To what extent do remodelling factors determine nucleosome positions? Even though all DNA sequences may be assembled into nucleosomes, some DNA sequences form more stable nucleosomes than others, for example, because they are more bendable. Conceptually, scanning a DNA sequence reveals an energy landscape with hills and valleys, i.e. with regions where the energetic cost to adopt the strongly bent conformation of nucleosomal DNA is intrinsically higher or lower. This energy landscape may be probed by biophysical nucleosome reconstitution methods, like salt gradient dialysis where nucleosomes form slowly in a solution of DNA and histones while an initially high amount of salt is gradually dialysed away. A comparison of genome-wide nucleosome positioning of such in vitro reconstituted chromatin with mapped in vivo positions has led to conflicting conclusions, either claiming that nucleosome occupancy may be largely encoded in the genome  or, conversely, that DNA sequence is not sufficient to determine nucleosome positions in vivo . This discrepancy mainly originates from the different measures ‘nucleosome occupancy’ (the probability of a given base pair to be anywhere in a nucleosome) and ‘nucleosome positioning’ (the probability of a base pair to be in the nucleosome centre). Furthermore, the identification of nucleosome positions in vivo and the outcome of in vitro reconstitutions considerably depend on the details of the respective methodologies applied, e.g. regarding the salt concentrations or temperature, and it is questionable which conditions faithfully recapitulate the in vivo situation . Integral to this ongoing debate over the degree of DNA-intrinsic nucleosome positioning is the question of how much nuclear factors such as DNA-binding proteins, and especially nucleosome remodellers, modulate nucleosome positioning in vivo. All remodellers can relocate nucleosomes and due to the energy input they may move nucleosomes to positions with intrinsically unfavourable DNA sequence (transition from state B to A’ in Figure 1). This has been beautifully illustrated for the yeast Isw2 complex by Whitehouse et al. , who showed that this remodeller was involved in moving nucleosomes all over the yeast genome on to repellent A/T-rich sequences at the 5′ and 3′ ends of genes. This active positioning suppresses cryptic initiation of transcription from otherwise larger NFRs (nucleosome-free regions). The activity of Isw2 can be harnessed for specific repression of the POT1 gene. The DNA-encoded nucleosome position leaves the POT1 promoter free and constitutively active, unless Isw2 shifts the nucleosome on to the unfavoured position, effectively occluding the access of the transcription machinery to the transcription start site . The maintenance of stable repression of the POT1 gene in rich medium may require the continuous presence and action of the remodeller, which in this scenario becomes part of a specific chromatin structure and alters the thermodynamics (state A’ instead of state A in Figure 1). Alternatively, the nucleosome could be kinetically trapped after dissociation of the remodeller in the repellent position, as relocation along DNA is slow in the absence of destabilizing factors such as salt, temperature or remodellers (state A in Figure 1 would result from state B through ATP-dependent remodeller action, and remain kinetically stable even in the absence of the remodeller). In either case, the remodeller itself would add information where a nucleosome shall be positioned (Figure 2A), rather than simply equilibrating positions dictated by the DNA sequence only.
Nucleosome remodelling catalyses transitions between diverse chromatin structures
Such information may explain the distinct roles that different remodellers play in the different regulatory logics of promoters [7,8]. In contrast with Isw2, for example, the yeast RSC remodeller does not repress POT1 transcription, but appears to mobilize nucleosomes at many promoters to ensure that NFRs remain open [9,10]. These diverse outcomes of remodelling reactions may be attributed to some extent to differences in their substrate interactions. Since remodelling requires the enzyme to contact DNA at the entry site into the nucleosome, remodelling enzymes may be differentially sensitive to sequence. For example, seven different remodellers generated widely different positions starting from the same nucleosomes in vitro . According to Längst and colleagues , the diversity in substrate interaction of the remodelling enzymes affects the resultant nucleosome positions. In vitro experiments usually use mononucleosomes or short arrays as substrate and may be confounded by effects of DNA ends. But even studies with minicircles point to remodeller-specific nucleosome positioning (for a further review see  and references therein). Nonetheless, it is unclear if this is based on different substrate interactions. Partensky and Narlikar  showed biochemically that the remodellers RSC and ACF remodel nucleosomes of very different stabilities with similar efficiency. They suggest that remodellers provide sufficient energy to convert the canonical histone–DNA interactions into a high-energy intermediary state that is largely indifferent to the effect of DNA sequence and allows moving around nucleosomes globally and indiscriminately. In this view, remodellers would not have much specific read on the DNA sequence, but DNA-intrinsic sequence preferences would determine nucleosome positioning very locally upon collapse of the high-energy intermediate. Dimitrov and colleagues  recently visualized a stable, but probably high-energy, remodelling intermediate for the RSC complex, where apparently 180 bp instead of 147 bp of DNA are co-ordinated by the histone octamer . Still, for this model it remains to be understood what triggers the collapse of a high-energy intermediate and if there are differences between remodellers, which in turn would lead to nucleosome deposition in different locations.
Staking claims: targeting of nucleosome remodellers
Even if a remodeller carried no information about the direction of the process it catalyses, it may still contain information about its targeting to genomic loci. Targeting by genetic principles is mediated by DNA-bound transcription factors that may recruit remodellers selectively to target areas . Intriguingly, there is now a first example of a remodeller with ‘built-in’ targeting to genetic (and not epigenetic) sites. The Rsc3 and Rsc30 subunits of the well-known RSC complex are able to recognize a specific DNA sequence . As mentioned above, the presence of such Rsc3/30-binding sites may be the basis for the generation of many NFRs in the yeast genome.
Still, the better known ‘built-in’ targeting principle functions through ‘epigenetic’ marks, namely certain histone modifications (acetylation, methylation) that are bound by dedicated domains [bromodomains, chromodomains and PHD (plant homeodomain) fingers respectively] . Recently, a new epigenetic target has been discovered. Poly(ADP-ribose) is recognized by the so called ‘macrodomain’, named for its homology with a domain in the histone variant macro-H2A, and was found in the remodeller ALC1 [16,17]. Poly-ADP ribosylation is an abundant histone modification and may target ALC1 especially to sites of DNA repair.
A particularly instructive example for recruitment of a remodelling activity to a single target nucleosome is the case of the NoRC (nucleolar remodelling complex) that is recruited both by interaction with the sequence-specific DNA-binding factor, TTF1, and by non-coding pRNA (promoter RNA) to rDNA (ribosomal DNA) loci in the mouse . Movement of the promoter nucleosome by just 25 bp prevents pre-initiation complex formation and leads to DNA methylation and epigenetic silencing .
Bringing up the walls: nucleosome mobility lays the foundation for higher-order structures
The local action of nucleosome remodellers is determined by their targeting, but there is increasing awareness of more fundamental (and perhaps global) roles in the formation of stable epigenetic structures for a subset of factors whose activities mediate different aspects of chromatin assembly. Several of the remodelling ATPase ISWI-containing complexes, such as CHRAC, ACF and RSF, appear to be particularly involved in the assembly of nucleosomes and in shuffling histone octamers on DNA until regular arrays without gaps are created . The integrity of the nucleosomal fibre is of vital importance to a cell, not only because a tight chromatin organization protects the precious DNA from damage, but also because it is fundamental for the overall repressive environment that characterizes higher eukaryotic genomes. Regular uninterrupted nucleosomal arrays fold readily into various types of chromatin fibres of approximately 30 nm width that are a prerequisite for any further chromatin compaction (Figure 2B). Depending on the precise distances between nucleosomes (their spacing), the resulting fibres are predicted to be more or less compact  (for a more recent discussion, see  and references therein). The discussion about the predominant physiological state of chromatin is still ongoing. Travers and colleagues  suggested, on the basis of data from the Rhodes and Richmond laboratories, that active euchromatin and more compact repressed heterochromatin may already be determined at the level of the nucleosomal repeat length, which will define the local geometry of the nucleosomal fibre (Figure 2D).
ACF, a simple remodeller consisting of ISWI and ACF1, is a prototype of a nucleosome ‘spacing factor’ that is involved in optimizing chromatin packaging. Recently, Narlikar and colleagues [23,24] have made significant progress towards deciphering the mechanism of spacing. They also showed that ACF preferentially generates the 50–60 bp linker length characteristic of silent chromatin and kinetically discriminates against shorter linker lengths that dictate a more open fibre in vitro . Conceivably, ACF defines the nucleosome repeat length through its interactions with the nucleosomal fibre – it may use a ‘built-in’ molecular ruler. The molecular geometries of different nucleosome remodelling factors may thus further explain the differential mobilization of nucleosomes by different remodelling factors (see above) .
Mutation of ACF1 in Drosophila leads to irregular, ‘sloppy’ chromatin in embryos and as a consequence the stability and potential heritability of ‘higher-order’ chromatin structure, such as heterochromatin, is compromised , (M. Chioda and P. Becker, unpublished work). The dependence of heterochromatin silencing on the regularity of the underlying nucleosomal array may be a widely conserved feature. The genome of the fission yeast Schizosaccharomyces pombe curiously does not contain a remodelling ATPase of the ISWI type, but there the Mi-2-related ATPase Mit1 could fulfil similar functions. Deletion of the mit1 gene changes the nucleosome positioning at the silent mating-type locus and alleviates heterochromatin silencing . Furthermore, Mit1 is actively engaged in setting the regularity of nucleosome arrays that are found phased with respect to active promoters . It remains to be seen if the role of Mit1 in shaping repressive heterochromatin is also linked to setting up regularly spaced arrays.
The folding of nucleosomal arrays into fibres is determined by nucleosome spacing but also promoted by the association of linker histones . The presence of linker histones will affect the activity of nucleosome remodellers differentially: some enzymes are unable to work on nucleosomes if they are ‘locked in’ by H1 association, but others, notably the ISWI-containing ACF, are able to move nucleosomes even with H1 bound. ACF can even facilitate the incorporation of H1 into chromatin in vitro . These biochemical observations may have physiological relevance as remodellers of the ISWI-type have been suggested to affect the steady-state association of H1 with chromosomes (Figure 2C) [30,31]. Conceivably, the DNA translocase activity nucleosome remodelling factors share with helicases  could be used not just to disrupt canonical nucleosomes, but to weaken other protein–DNA interactions as well. Verrijzer and coworkers  found the nucleosome remodelling complex SWI/SNF (switch/sucrose non-fermentable) to be involved in the eviction of the epigenetic repressor polycomb, effectively reprogramming the target locus . It remains to be seen whether histone H1 and the polycomb complex are direct substrates for remodelling factors. Whether or not this is the case, the effect of these ATPases on chromatin dynamics clearly reach far beyond moving single nucleosomes.
A third mechanism through which nucleosome remodellers may affect higher-order chromatin structure is the incorporation of histone variants with particular features. For example, the nucleosome remodelling ATPase dCHD1 is uniquely required for the replication-independent incorporation of the H3.3 variant during chromatin assembly on the male pronucleus in Drosophila . Variants of histone H2A differ in their interactions with the H4 N-termini of close-by nucleosomes and their presence in a nucleosomal fibre modulates the tightness of compaction  (Figure 2E). The ISWI-containing remodeller RSF is involved in incorporating the variant H2Av in Drosophila . RSF combines histone chaperone and nucleosome spacing activity in vitro and is therefore ideally suited to promote the assembly of regular nucleosomal arrays. In Drosophila, H2Av incorporation is an early requirement for the constitution of heterochromatin, and inactivation of the RSF1 subunit of RSF impairs heterochromatin formation . Evidently, the assembly of the tightest, heritable chromatin structures requires the activities of nucleosome remodelling factors for variant histone incorporation into regular nucleosomal arrays.
It is likely that the activity of remodelling factors is influenced by the accessibility of the nucleosome substrate, but so far the evidence for this notion is limited. Nightingale et al.  observed that the unfolding of chromatin associated with strong histone acetylation increased the access of the remodeller ACF to its nucleosome substrate, which in the context of the assay led to enhanced site-specific recombination of chromatin segments. By contrast, site-specific acetylation of H4K16 reduces the ATPase activity of ISWI in vitro . Deletion of ISWI leads to a collapse of the polytene organization of the H4K16-acetylated male X chromosome in Drosophila larvae in a process that correlates with H1 depletion. The mechanistic aspects of the structural and functional fine-tuning involving H4K16 acetylation, H1 levels and ISWI activity is currently a mystery .
Brick layer or building block: remodellers as structural entities
Collectively, or in some cases even individually, nucleosome remodelling enzymes are relatively abundant factors . According to the persuasive ‘three-dimensional sampling’ model of Hager, McNally and Misteli , remodelling complexes find their target sites by diffusion through the nuclear interior until interaction with a chromatin-bound recruitment domain leads to local concentration. The search for the specific target typically involves frequent non-specific chromatin encounters, which in the context of a transcription factor may simply be transient and non-productive. In the case of nucleosome remodelling factors these random encounters with the chromatin substrate may trigger a ‘basal’ remodelling activity and significantly contribute to nucleosome dynamics. For those remodellers, such as ACF and RSF, for which no targeting principles are known to date, ‘random sampling’ may be the mechanism to survey the genome for perturbations of the nucleosomal array. Considering the effective concentrations of nucleosome substrates in the nucleus, we envisage that remodellers are chromatin-associated for most of the time. ‘Spacing factors’ such as ACF, which sample the linker length on either side of a core nucleosome with remarkable processivity , may be intimately engaged with the fibre to an extent at which their mere presence affects the structure of the chromatin fibre . The non-catalytic structural contributions of abundant chromatin-bound remodellers remain to be explored.
Remodellers are chiefly known for their dynamic input in mediating transitions between different chromatin structures. Nonetheless, there is growing evidence that remodeller complexes also harbour ‘built-in’ structural information. They may read specifically DNA sequence preferences for nucleosome positioning, their remodelling mechanism in combination with remodeller geometry may encode specific nucleosome spacing, or their subunits may determine the targeting of their action to specific genomic loci. The great variety and abundance of remodelling complexes contributes also structurally to the establishment of chromatin states. As stable chromatin structures are among the prime vehicles for epigenetic information, the local and global actions of nucleosome remodellers have far-reaching implications for the epigenetic processes during cell growth and differentiation .
Nucleosome remodellers keep chromatin dynamic.
Remodellers may add structural information and confer epigenetic stability to chromatin on several levels.
Remodellers carry their own targeting modules.
Remodellers co-determine nucleosome positioning.
Remodellers control nucleosome spacing and thus chromatin fibre folding.
Remodellers influence the association of H1 or non-histone proteins with chromatin.
The high abundance and ubiquitous chromatin interactions of remodellers may provide the groundwork for higher-order structures.
Research in the groups of the authors is supported by grants from the Deutsche Forschungsgemeinschaft and by the European Union via the Network of Excellence ‘The Epigenome’ (FP6-503433).