The bacterial chromosomal DNA is folded into a compact structure called nucleoid. The shape and size of this ‘body’ is determined by a number of factors. Major players are DNA supercoiling, macromolecular crowding and architectural proteins, associated with the nucleoid, which are the topic of this MicroReview. Although many of these proteins were identified more than 25 years ago, the molecular mechanisms involved in the organization and compaction of DNA have only started to become clear in recent years. Many of these new insights can be attributed to the use of recently developed biophysical techniques.
Bacteria, like other organisms, are faced with the task of compacting their chromosomal DNA. In the case of Escherichia coli, 4.6 Mb of DNA with a contour length of approximately 1.6 mm is fitted into a cell that measures about 2 µm along its major axis and 1 µm along the other axes. The combined action of several factors (DNA supercoiling, macromolecular crowding and nucleoid-associated proteins) results in a nucleoid that occupies only a small part (about one-fourth) of the intracellular volume. The relative contributions of each of these factors to DNA compaction are as yet unclear.
Rapidly growing cells have a very active DNA metabolism. Replication, chromosome segregation and cell division occur within ∼20 min during fast growth. Unlike in eukaryotic organisms, almost the entire chromosome is competent for transcription in bacteria during the entire cell cycle. Transcriptional activity in some regions, e.g. in stable RNA operons, is very high, whereas transcription in other regions is limited. Nevertheless, even genes that are quiescent can be rapidly activated in response to sudden environmental changes. This requires an organization of the chromosomal DNA that is very dynamic in nature, and which guarantees access for proteins involved in DNA transactions. It also suggests that the nucleoid is not homogeneous in structure and that there is a correlation between transcriptional activity and local nucleoid-organization. These properties must be kept in mind when trying to understand the organization of the bacterial nucleoid.
Structure of the bacterial nucleoid
Our view of nucleoid structure has been strongly shaped by electron micrographs recorded in the early seventies of the last century. These micrographs showed E. coli nucleoids that were obtained upon lysis of the cell, in which large DNA loops emanate from a central core. These loops were plectonemically interwound – indicating the presence of supercoiling – but did not seem to have any proteins bound that could be involved in a higher order type of organization of the bacterial chromatin (Delius and Worcel, 1974; Kavenoff and Bowen, 1976) similar to that conferred by eukaryotic chromatin (Olins and Olins, 1974; Lohr and Van Holde, 1975). The proposal that the bacterial chromosome is organized in supercoiled loops is based primarily on these images. In fact, supercoiling and branch formation of DNA by itself is an effective way to reduce its radius of gyration. These visual data found firm support in biochemical evidence for the existence of loops in isolated nucleoids (Worcel and Burgi, 1972) and in vivo (Sinden and Pettijohn, 1981; Pettijohn, 1982). The existence of independent (superhelical) domains that are dynamic with regard to their exact size and the positioning on the chromosome of their boundaries was later demonstrated using genetic methods (Higgins et al., 1996). Initially, it was believed that these domains were up to several hundreds kb in size (Kavenoff and Bowen, 1976; Sinden and Pettijohn, 1981). The most recent estimates indicate that these domains are probably closer to ∼10 kb in size (Cunha et al., 2001; Postow et al., 2004).
In the period when Kavenoff and Bowen made their images of the nucleoid, another study employing electon microscopy (EM) provided evidence of a eukaryotic-type compaction into nucleosome-like structures or ‘compactosomes’ (Griffith, 1976). It was proposed that these structures resulted from the binding of members of a group of small, mostly basic, DNA binding proteins that were soon to be identified in E. coli (Varshavsky et al., 1977). However, the existence of such structures (or any other type of regular organization) has long been the subject of controversy (Kellenberger, 1988) and, up to a few years ago, relatively little was known about the structural role of any of these proteins. This is attributed to the fact that biochemical techniques do not unambiguously resolve any regular structure in complexes between these proteins and DNA (Broyles and Pettijohn, 1986). An additional reason for scepticism was that the early EM methods employed were prone to artefacts (Eickbush and Moudrianakis, 1978; Kellenberger, 1988).
The introduction of novel physical techniques into biology has opened up new horizons in general and has instigated renewed interest in aspects of nucleoid organization, in particular. The use of improved nuclear magnetic resonance (NMR) and X-ray crystallographic analysis, in combination with scanning force microscopy (Bustamante and Rivetti, 1996), optical or magnetic tweezers (Allemand et al., 2003; Amit et al., 2004) and (single-molecule) fluorescence methods (Ha, 2001; Hillisch et al., 2001) has led to new molecular insights into the role of nucleoid-associated proteins in E. coli.
The role of nucleoid-associated proteins
Nucleoids can be isolated after cell lysis under stabilizing conditions, for instance in a buffer that mimics the macromolecular crowding conditions found in the cytoplasm. In order to obtain an understanding of the role of architectural proteins, the protein content of such isolated nucleoids can be determined. Depending on the particular conditions employed, a number of proteins are found associated with the nucleoid. The early preparations of Varshavsky et al. revealed the presence of two abundant proteins, H-NS (histone-like nucleoid structuring protein) and HU (heat unstable protein) (Varshavsky et al., 1977). More recently, nucleoids isolated in a similar fashion were shown to contain H-NS, HU, Fis (factor for inversion stimulation) and IHF (integration host factor) (Murphy and Zimmerman, 1997). Attributed to their high intracellular abundance and DNA binding properties, these four proteins are generally believed to be the most important players in the organization and compaction of bacterial chromatin. Two other proteins that might play such a role and recently attracted biophysical interest, Lrp (leucine responsive protein) and MukBEF, will also be discussed in this review.
The expression level of nucleoid-associated proteins is dependent on growth phase. Analysis of a group of 12 proteins associated with the nucleoid as a function of growth phase reveals that certain proteins are absent during exponential growth and abundant during stationary growth or vice versa (Talukder et al., 1999). This allows the cell to cope, for instance, with the requirement for high levels of transcription and translation during exponential growth by stimulation of transcription from stable RNA operons with Fis (Schneider et al., 2003). Similarly, during the stationary phase the cell can protect its genome by the extensive binding of Dps (DNA binding protein from starved cells) and, thus, shut off transcription (Almiron et al., 1992; Frenkiel-Krispin et al., 2001). The variation in expression levels of the nucleoid-associated proteins can thus provide a means to modulate (locally) the structure of the nucleoid, depending on growth conditions. Interestingly, in order to switch effectively between compact and less compact states, some of the nucleoid-associated proteins seem to have a clear-cut role in compaction, whereas others seem to act both as compacting agents and as antagonists of compaction.
H-NS consists of a dimerization (N-terminal) and a DNA binding (C-terminal) domain connected by a flexible linker (Dorman et al., 1999; Rimsky, 2004). The three-dimensional structures of the N- and C-terminal domains have been solved separately; resolution of the full-length structure is presumably hampered by the flexibility of the linker domain. At present, conflicting results exist about the orientational alignment of the N-terminal dimerization domains. NMR analysis of the N-terminal domain of Salmonella typhimurium H-NS suggests parallel dimerization (Esposito et al., 2002), whereas NMR analysis of the N-terminal domain of E. coli H-NS and X-ray analysis of the same domain in Vibrio cholerae H-NS indicate antiparallel dimerization (Bloch et al., 2003; Cerdan et al., 2003). Independent of the orientation of these domains, the fact that oligomerization leads to the simultaneous availability of two or more DNA binding domains suggests that interactions might involve multiple DNA molecules. This idea is in good agreement with scanning force microscopic observations of H-NS–DNA complexes, which already indicated that H-NS can form bridges between adjacent tracts of double-stranded DNA (Dame et al., 2000) (see Fig. 1A). This property appears to be conserved among H-NS like proteins with varying degrees of sequence homology, but with a similar two-domain structure (Dame et al., 2005).
What is the molecular basis underlying DNA bridging? Different models for the interaction with DNA should be considered to account for the differences in the three-dimensional structure of H-NS. Self association of H-NS in a parallel fashion has been proposed to result in the formation of dimers or large oligomers in which multiples of two DNA binding domains are available to interact with DNA on opposite sides of the multimer (Esposito et al., 2002) (see Fig. 1B). In this model, dimers are stacked on top of each other through interactions between the N-terminus and part of the flexible linker region. In principle, a model with stacked dimers could also account for bridging by antiparallel associated H-NS. However, Bloch et al. (Bloch et al., 2003) found that two arginine residues at the N-terminus somehow affect the binding of H-NS to DNA. These residues might provide direct interaction with DNA supplementary to the binding of the C-terminus (Bloch et al., 2003). The central part of the dimer (encompassing the arginine residues) and the C-terminal domains could bind side-by-side to DNA. In that scenario, bridge formation between adjacent DNA tracts might result from interaction between the flexible linker regions (see Fig. 1C) (Bloch et al., 2003; Dorman, 2004). Both models can account for the observed co-operativity of H-NS binding.
H-NS–DNA complexes have been analysed recently using magnetic tweezers (Amit et al., 2003). In magnetic (or optical) tweezers, a single DNA molecule is suspended between a surface and a bead (or between two beads). The mechanical response of the DNA tether is then probed by varying the end-to-end distance of the DNA. Comparison of force-extension curves of protein–DNA complexes with those of bare DNA molecules gives an indication of the architectural effects of such proteins (Finzi and Dunlap, 2003). In the case of H-NS, this type of study does not give any evidence of the DNA loop formation involving bridging suggested by earlier scanning force microscopy (Dame et al., 2000) and electron microscopy studies (Schneider et al., 2001). In fact, the binding of H-NS results in stiffening of the DNA. The question is how these two different observations can be reconciled. It has been suggested that these results can be explained by the typical circumstances found in single molecule measurements, i.e. the enormous excess of protein over DNA (Dame and Wuite, 2003). A second possibility is that the extension of DNA that results from the liquid drag when H-NS is flushed into the sample chamber biases the experiment towards side-by-side binding of the protein and disfavours DNA bridging. Whether either of these two explanations is actually correct would require experimental verification.
Independent of how these differences can be explained, we believe that the data can be reconciled in following manner. The intrinsic binding affinity of H-NS is assumed to be imposed by structural features of the DNA such as flexibility or curvature, and by sequence composition. For this reason (and as there is insufficient H-NS, about 20 000 molecules per cell, to cover the full length of the DNA) it seems likely that the protein is not uniformly bound, but rather binds in patches along the DNA. Lateral (co-operative) H-NS binding occurring preferentially along A/T-rich tracts (but extending into flanking DNA sequences) (Rimsky et al., 2001) might account for local rigidification (the formation of an H-NS patch), similar to that observed in the magnetic tweezers experiments (Amit et al., 2003). Such patches of H-NS on the DNA might then interact with other patches or with naked DNA. Further extension of the DNA bridged areas is then believed to be co-operative, attributed to the close proximity of the DNA directly aside of the region bridged by H-NS (Dame et al., 2000).
Binding of H-NS has also been implicated in changes in DNA supercoiling in vivo[see for instance the study by Higgins et al. (1988)]. However, only in vitro experiments give a clear picture of the effect of H-NS: it constrains negative supercoils (Spassky et al., 1984; Tupper et al., 1994). Assuming that DNA is bridged by H-NS, supercoiling might conversely also promote the formation of bridges within a plectonemic superhelix attributed to the severe spatial constraints imposed by the supercoiled conformation.
The formation of bridges is relevant not only to the compaction of DNA, but also to the regulation of transcription. For instance, it has been suggested that repression of transcription from virF is determined by the bridging of two H-NS binding sites, which occludes RNA polymerase from the promoter (Prosseda et al., 2004). More complex mechanisms have been proposed for repression at the rrnB P1 promoter and of the bgl operon. At the rrnB P1, RNA polymerase is physically trapped in a transcription competent open complex by the bridging of up- and downstream regions (Schroder and Wagner, 2000; Dame et al., 2002). A similar mechanism has been proposed for the hdeAB promoter, at which the trapping (and thus the repression by H-NS) seems to be dependent on whether the RNAP holoenzyme contains σ38 or σ70 (H.E. Choy, pers. comm.). Repression at the bgl operon occurs during elongation and results from downstream bound H-NS molecules that act as a roadblock. As binding of H-NS to areas upstream and downstream of the promoter is needed for repression, these regions were proposed to be bridged to form a loop (Dole et al., 2004). As with regard to the role of H-NS as a repressor other mechanisms that do not necessarily involve bridging have also been proposed. At some promoters lateral binding of H-NS initiating at a preferential binding site (consisting of curved or flexible DNA) that extends over (part of) the promoter region might be sufficient for repression (Rimsky et al., 2001). Also, the proposed effect of H-NS on DNA supercoiling might be an indirect means to regulate transcription from a subset of promoters sensitive to changes in supercoiling (Higgins et al., 1988).
HU and its sequence specific homologue IHF consist of two (identical or homologous) subunits (Swinger and Rice, 2004). The structure of Bacillus stearothermophilus HU was solved using X-ray crystallographic analysis about 20 years ago, while that of IHF (in a cocrystal with DNA) was solved only recently. HU and IHF both consist of a compact body of several intertwined α-helices from which two protruding β-ribbon arms wrap around the minor groove of DNA upon binding. The conserved proline residues at the tip of each of the arms intercalate between base pairs. A positively charged surface extends down the sides of both HU and IHF, which has been implicated in DNA binding, and as being important for DNA binding over a variable distance (Tanaka et al., 1984; Rice et al., 1996; Swinger et al., 2003). As a consequence, two kinks are created or stabilized in the DNA where the arms are inserted into the minor groove. The estimates for the bending angle induced by IHF differ from >140° (Thompson and Landy, 1988) and close to 180° using biochemical bulk techniques to 160–180° in the DNA cocrystal (Rice et al., 1996). In fact, footprinting experiments also suggest that the bending is variable and depends on the particular site to which IHF is bound (Yang and Nash, 1989). In the case of IHF, the bend is stabilized by interactions between the DNA and the body of the protein (Rice et al., 1996).
Although HU shows little sequence specificity, it binds preferentially to supercoiled DNA (Shindo et al., 1992) and constrains negative supercoiling in vitro (Rouviere-Yaniv et al., 1979; Broyles and Pettijohn, 1986). This observation can be rationalized by looking into the HU–DNA cocrystal structure, in which, besides bending, underwinding and negative writhe are induced (Swinger et al., 2003). Furthermore, HU binds preferentially to structural distortions in DNA, such as nicks, gaps and three- or four-way junctions, recognizing and stabilizing a pre-existing bend, rather than actively inducing the bend (Pontiggia et al., 1993; Castaing et al., 1995; Kamashev et al., 1999; Kamashev and Rouviere-Yaniv, 2000; Balandina et al., 2002). Exploiting the ability of HU (converted into a chemical nuclease) to bind preferentially to nicked DNA and using circular permutation analysis a bending angle of 65° could be determined for the nicked DNA–HU complex (Kamashev et al., 1999). Ensemble-Förster Resonance Energy Transfer (FRET) and single-molecule FRET studies indicate bending on DNA substrates with two TT:TT mismatches 9 bp apart (inducing severe kinks) that are quantitatively similar to that induced by IHF (Sagi et al., 2004). X-ray studies of Anabaena HU–DNA cocrystals reveal bending angles between 105° and 140° on that type of substrates (Swinger et al., 2003). The bending angle induced by HU into undistorted DNA cannot be easily estimated because of the lack of sequence specificity. In biochemical assays, the protein is therefore not necessarily continuously associated (to the same site) during the course of the assay. However, considerable progress has been made recently using short DNA substrates in FRET studies. In this type of study, DNA is labelled on each of its extremities with a fluorescent tag. These labels are selected on the basis of their ability to exhibit FRET if they are in close proximity. In practice, this means that the fluorescence intensity of a donor fluorophore is reduced and that of the acceptor fluorophore is increased when they are brought together. The efficiency with which this occurs is an indication of the distance between the two fluorophores and allows bending angles to be calculated. A bend angle of approximately 60° was observed in ensemble FRET studies of naturally mildly curved DNA (Wojtuszewski and Mukerji, 2003). Although longer substrates (accommodating more HU molecules simultaneously) were used in single-molecule FRET studies, a comparable bend angle (∼53°) was extracted from those data, assuming that the bending by each HU dimer occurs in the same plane (Sagi et al., 2004). A more detailed picture was obtained in scanning force microscopy (SFM) studies. The imaging of individual HU–DNA complexes showed a broad distribution of angles (from 0° to ∼180°, with an average of ∼100°) (see Fig. 2A and B). The large width of the distribution provided further evidence that HU acts as a flexible hinge that can accommodate a range of different bending angles (van Noort et al., 2004), a term coined earlier by Rice and coworkers (Swinger et al., 2003). The average of these different bending angles is measured in bulk studies. The functional significance of strong DNA bending was recently exemplified in a magnetic tweezers experiment in which the interaction between two GalR molecules separated by 115 bp distance was shown to be directly stimulated by the presence of HU (Lia et al., 2003).
Soon after the observation of ‘compactosomes’ (Griffith, 1976) on the DNA from lysed cells imaged by electron microscopy, similar types of structures (described as being ‘nucleosome-like’) formed by HU were also observed (Rouviere-Yaniv et al., 1979). Therefore, HU was long considered as the primary protein factor involved in the organization of bacterial chromatin. It was proposed that HU dimers bind side-by-side on the DNA and that the DNA would attain a superhelical structure (that would easily explain the observed effect of HU on DNA topology) around a HU protein core (Tanaka et al., 1984). As it turned out to be very difficult to find direct evidence for the formation of this type of structures using biochemical methods, these findings were long considered controversial (Dame and Goosen, 2002).
In an attempt to settle this controversy, studies employing SFM (which, unlike EM, does not require fixation methods that are a potential source of artefacts) have been undertaken. No indication could be found for the existence of nucleosome-like structures induced by the binding of HU (Dame and Goosen, 2002). Several recent studies provide a new perspective of the modulation of DNA architecture by HU. By using magnetic tweezers, it was demonstrated independently by two groups that there are two concentration regimes in which HU exhibits different behaviour. DNA is compacted at low HU concentrations, but becomes extended at high HU concentrations (van Noort et al., 2004; Skoko et al., 2004). SFM imaging revealed that the observed compaction can be attributed to bending by HU (resulting in an increased effective flexibility), rather than to DNA wrapping (see Fig. 2A). Extension of the DNA was proposed to result from side-by-side binding of HU and the formation of a regular helical HU–DNA filament (van Noort et al., 2004) (see Fig. 2C–E). The structural model of the HU–DNA filament as originally proposed (see Fig. 2D), shows HU dimers that are bound on the outside of the DNA superhelix and limited DNA bending. In an alternative structural model (in which HU is allowed to bend the DNA), the HU dimers are bound on the inside of the DNA superhelix (see Fig. 2E). In both models the ability of HU to induce or constrain supercoiling is explicitly taken into account (Rouviere-Yaniv et al., 1979; Broyles and Pettijohn, 1986; Shindo et al., 1992). It should be noted that the force-extension measurements on HU–DNA complexes were in both cases performed with topologically unconstrained DNA. Therefore, with the present knowledge one cannot yet exclude that supercoiling of the DNA would induce wrapping around an HU protein core, as originally proposed (Rouviere-Yaniv et al., 1979; Tanaka et al., 1984). Importantly, FRET studies (Sagi et al., 2004) and UV resonance Raman spectroscopy (Wojtuszewski and Mukerji, 2004) lend further support for the unexpected observation that HU has two binding ‘modes’. To date, crystallographic evidence that could provide detailed insight into the structure of the proposed HU–DNA filaments is not available.
So what does DNA compacted by HU look like? Most of the present data suggest that HU is bound at different positions along the DNA and induces bends. The extent to which DNA is compacted by HU depends on the angle induced by the protein and the number of HU molecules bound (van Noort et al., 2004; Skoko et al., 2004). It remains to be solved whether or not the DNA stiffening mode is of physiological significance, as the ratio of HU to DNA in vivo (1 HU dimer per 100 bp) (Azam et al., 1999) is much lower than that at which stiffening is observed (van Noort et al., 2004; Skoko et al., 2004). However, the co-operativity of filament formation makes it likely that in vivo short regions of reduced flexibility can be formed within the bacterial nucleoid, which do not lead to compaction (van Noort et al., 2004).
Whereas HU only exhibits sequence unspecific binding, IHF-binding can also occur with sequence specificity. About 1000 specific IHF binding sites have been identified in the E. coli chromosome. Most of these sites are located in close vicinity of promoters and are probably involved in transcription activation (Goosen and van de Putte, 1995; Ussery et al., 2001). SFM studies have confirmed that the DNA becomes strongly bent (∼120°) at specific sites (Seong et al., 2002; R. T. Dame et al., submitted). Ensemble and single-pair FRET measurements also indicate that IHF induces a DNA bend of on average ∼160° (Lorenz et al., 1999; Sagi et al., 2004). IHF, in fact, was the first nucleoid-associated protein to be analysed in single-molecule elasticity measurements (Ali et al., 2001). These experiments showed that IHF induces a compaction of the DNA of maximally 30%. This compaction can be attributed to non-specific binding of the protein at many sites simultaneously, rather then to specific binding and bending. The majority of the 1000 specific IHF sites on the E. coli chromosome (Ussery et al., 2001) is believed to be continuously occupied (Yang and Nash, 1995). This means that the remaining IHF molecules [30 000–60 000 per cell (Talukder et al., 1999)] are available for non-specific binding to DNA, which might be significant for (HU-like) compaction of the bacterial chromosome.
Evidence has been put forward that IHF can also bind non-specifically side-by-side (Holbrook et al., 2001) and might form filaments as described above for HU. Although not mentioned as such in the report, in fact, this type of binding might be reflected in the single-molecule elasticity measurements of Ali et al. (2001). The extension of the DNA tether in their experiments is reduced maximally about 30% at a certain concentration and then the effect levels off. This concentration range could well correspond to the broad extension minimum observed in the case of HU (van Noort et al., 2004), in which bending and local rigidification presumably co-exist. Clarification of this issue requires similar types of experiments in which this possibility is explicitly taken into consideration.
Like HU and IHF, Fis can bend DNA. The protein recognizes a poorly conserved 15 bp ‘core’ binding site and has considerable affinity for non-specific DNA. Fis consists of two identical subunits, each of which contains a prototypical helix–turn–helix (HTH) element that is involved in DNA binding. These HTH-elements are presumably inserted into adjacent major grooves of the DNA helix. DNA bending is believed to be the result of that the spacing between these elements is too short to be fit to a straight B-DNA helix (Kostrewa et al., 1991; 1992; Yuan et al., 1991). Estimates of the DNA bend induced range from 50° to 90° depending on the exact sequence bound and its context (Pan et al., 1996). SFM studies confirm that DNA bends by ∼90° at specific sites upon binding of Fis. (Zhang et al., 2004). The non-specific binding of low amounts of Fis leads to the formation of branches on supercoiled DNA molecules. In these complexes, Fis is bound in clusters at crossovers and the apical loops. Increasing the amount of Fis induces strong compaction of the DNA (Schneider et al., 2001). Bioinformatic analysis has revealed that the E. coli chromosome contains up to ∼68 000 sites, or one site per 230 bases (Hengen et al., 1997). A large fraction of these sites is probably occupied during exponential growth, when Fis expression levels reach a maximum of ∼60 000 molecules (Talukder et al., 1999).
Lrp type proteins seem to have two modes of DNA binding. The LrpC protein from Bacillus subtilis can bridge DNA (Tapias et al., 2000), in a fashion similar to H-NS, as revealed by electron microscopy (Beloin et al., 2003). In this scenario Lrp-dimers on one tract of DNA are believed to interact with Lrp-dimers on the other tract (Beloin et al., 2003). This could be a means to stabilize DNA loops. In addition, the protein can assemble into an octameric, globular, ‘nucleosome-like’ structure around which DNA can be wrapped. This occurs specifically at the promoter of its structural gene to attain autorepression (Jafri et al., 1999; Beloin et al., 2003), but also non-specifically on supercoiled DNA substrates (Beloin et al., 2003). Both mechanisms probably act in parallel in vivo and are dependent on local sequence context.
Although MukBEF (the E. coli SMC homologue) is usually not considered a nucleoid-associated protein sensu strictu, it should be considered as such. This protein, which is much larger (about 600 kDa) than any of those discussed so far, consists of two DNA binding domains connected by a long coiled-coil, with a flexible hinge at its centre (Graumann, 2001). SMC-like proteins are therefore capable of binding DNA simultaneously at more than one site. Force-extension measurements on MukBEF–DNA complexes reveal strong DNA compaction, which can be released in a stepwise fashion (55 nm steps) by stretching the filament (Case et al., 2004). These steps are believed to reflect the extension of the hinge, while the DNA binding domains remain bound. In this configuration, the large size of the protein allows a distance of ∼170 base pairs to be spanned. MukBEF is proposed to be involved in the creation of large topological loops by bridging two DNAs at the base of the stem of such loops (Case et al., 2004).
A balance of forces
The primary focus of this review is the role played by architectural proteins in shaping the bacterial nucleoid. However, several other factors are believed to be involved. Supercoiling of the DNA as mediated by the action of topoisomerases provides a means of compaction (Vologodskii and Cozzarelli, 1994). About half the amount of supercoiling is not constrained by proteins and is thought to be present in the form of (branched) plectonemes. An important role is also attributed to macromolecular crowding forces exerted by the cytoplasm (Zimmerman and Murphy, 1996; Odijk, 1998). Finally, coupled transcription and translation of transmembrane proteins (also called transertion) (Norris, 1995; Woldringh et al., 1995) counteracts compaction by pulling DNA loops out from the nucleoid body towards the cell membrane. The resultant shape of the nucleoid stems from a delicate balance of these forces (Woldringh et al., 1995). The role of nucleoid-associated proteins in the Woldringh model is restricted to providing a means of compaction. If the effects of all nucleoid-associated proteins on the nucleoid as a whole are taken together, this is probably justified, but locally these proteins might exert opposite effects.
The abundance of nucleoid-associated proteins varies dramatically depending on the growth phase (Talukder et al., 1999). As most of the nucleoid-associated proteins function as transcription factors (either directly or indirectly as activators or repressors), the expression of many genes is regulated in a growth phase-dependent manner (Ishihama, 1999). The picture that is now emerging is that the regulation of a subset of genes depends on the antagonistic action of combinations of nucleoid-associated proteins. For instance, the sequence specific DNA bending proteins IHF and Fis can relieve repression by H-NS at specific promoters (Tippner et al., 1994; van Ulsen et al., 1996; Afflerbach et al., 1998; Falconi et al., 2001). Similarly, HU is proposed to counteract the effects of H-NS, possibly by competing for preferential binding sites (Dame and Goosen, 2002; van Noort et al., 2004 and references therein). It should be noted, that more complex arrangements, in which the proteins involved are not limited to acting in an antagonistic manner, but also engage in co-operative activities that might lead to repression of transcription, can be envisaged. The few known examples include the co-operative action of Fis, IHF and H-NS that leads to repression at the E. coli nir promoter, and the action of IHF, H-NS and a third regulatory protein OmpR at the S. typhimurium csgD promoter (Browning et al., 2000; Gerstel et al., 2003). We suggest that the antagonistic effects described above also play a role at the level of global nucleoid organization, and that this is responsible for a considerable degree of heterogeneity within the nucleoid. The action of HU, IHF or FIS could result in regions that are less compact than others. Local reorganization of the nucleoid could result in an effect on the expression of genes that are hundreds of bases away, and would in part explain the very pleiotropic effects of these proteins (Arfin et al., 2000; Hommais et al., 2001; Kelly et al., 2004). Besides being favoured in the context of a locally more ‘open’ nucleoid structure, transcription may itself be responsible (at least transiently) for locally disturbing nucleoid compactness. It is not hard to imagine that in order for transcription to proceed, it is essential that DNA bound proteins (bridging or bending it) ahead of the progressing RNA polymerase are driven off the DNA. In particular, ‘open’ nucleoid structure is thus expected at sites where heavy transcription takes place, such as within ribosomal RNA operons during exponential growth. Finally, the effects could be indirect, for instance when overexpression of H-NS leads to the ‘complete’ shut down of transcription, resulting in strongly reduced transertion mediated nucleoid expansion and, thus, compact nucleoids (Spurio et al., 1992).
In the light of all these considerations, we propose the following model for the bacterial nucleoid. The nucleoid is organized in loops of ∼10 kb (Deng et al., 2004; Postow et al., 2004), which are connected and topologically closed because of the binding of proteins that create cross-links between DNA tracts, such as MukBEF (Case et al., 2004). In vivo, E. coli has sufficient MukBEF molecules to bind once per ∼30 kb, which is of the same order as the number of loops that has been proposed to exist. In addition, a number of other factors are probably also involved. H-NS-like proteins (Dame et al., 2000; 2005) and Lrp (Tapias et al., 2000; Beloin et al., 2003) could be involved at this level but, although H-NS is very abundant, the Lrp levels (about 1000 dimers per cell) are relatively low (Talukder et al., 1999). The presence of a small number of boundaries (about 20) is attributed to transertion sites within the membrane (Woldringh, 2002). Furthermore, transcription from highly active promoters (independent of membrane translocation) can also introduce new domain boundaries (Deng et al., 2004). The fact that transcription of rRNA operons takes place in a few clusters (‘transcription factories’) in the cell, which has been linked to a role in DNA compaction, might be related to this observation (Cabrera and Jin, 2003). Rather than being important for the stabilization or formation of large DNA loops, nucleoid-associated proteins could be essential for the local structure of smaller loops within these larger ones. The structure and number of these smaller loops thereby determines, in part, the compact state of the nucleoid.
Protein-induced or intrinsic DNA bends tend to localize at the apex of supercoiled DNA loops (Laundon and Griffith, 1988; ten Heggeler-Bordier et al., 1992). In the absence of protein, the apex of a supercoiled loop will not be very specifically localized, but will correspond mostly to a region of higher DNA flexibility. The high-affinity binding of a protein such as IHF or Fis to its specific site will thus impose a defined configuration upon a supercoiled loop. The non-specific binding of these proteins and HU is expected to be of a more dynamic, less spatially defined, nature. The higher the number of DNA bending proteins bound within a supercoiled loop, the more branched (Tan et al., 1996; Wedemann et al., 1998) and, thus, the more compact it will become. Assuming that a large fraction of the DNA bending proteins are more or less continuously bound, this suggests that there are on the order of >10 000 branches within the supercoiled loop organization of the bacterial nucleoid. In the case of HU, a second scenario can be envisaged: HU might recognize the often A/T-rich apex of a loop as a region of preferential binding (Tanaka et al., 1993; Shimizu et al., 1995) and bind there co-operatively (Pinson et al., 1999; van Noort et al., 2004). As this will result in rigidification of this tract, an apical localization is no longer favourable, and the tract will relocate to the stem of the loop. As discussed above, it seems likely that H-NS is not uniformly bound, but rather binds in patches along the DNA that can interact with other patches or with naked DNA. If this occurs within a supercoiled loop, that particular configuration of the loop will be stabilized, at least temporarily.
An isolated view on the action of the different nucleoid-associated proteins does not do justice to the complex interplay between them. For the sake of simplicity we now consider the nucleoid as being in a certain state under a given growth condition. If these growth conditions change, the ratio between proteins such as H-NS on one side, and IHF, Fis and HU on the other side will become different. These proteins can act specifically as antagonists during the regulation of transcription by competition for approximately the same binding sites, as described above. Local disruption of an H-NS–DNA complex (de-compaction) results in transcription activation. We believe that this mechanism plays only a minor role at the global organizational level. We envisage a role for DNA bending by nucleoid-associated proteins that is functionally analogous to chromatin remodelling in eukaryotic organisms. The induction of a bend dictates the location of the apex of a supercoiled loop and, thus, defines the spatial vicinity of sites within the loop. A bend in the DNA might thereby facilitate or disfavour DNA bridging (Fig. 3A). This will change the relative location of areas previously bridged by H-NS. A patch with H-NS bound will relocate and might form new bridges at a different position where it encounters a second H-NS patch or bare DNA (depending on which mechanism of bridging is employed – see Fig. 1B and C). By introducing an additional bend, a loop will be reorganized and attain a branched configuration, with concomitant further changes in spatial vicinity primarily of sites that are within the loop attributed to slithering (Wedemann et al., 1998). An example of a consequence of such dramatic reconfigurations is that even promoter regions distant from the DNA bending protein can become subject to differential accessibility to RNA polymerase.
In the case of HU, a second scenario that involves direct competition might be envisaged. Preferential and possibly co-operative (see above) binding of HU to an A/T-rich tract (as is also preferred by H-NS) dictates location within the stem of the loop (see Fig. 3C). If HU covers an extended area, this whole tract cannot be bridged (and thus compacted) by binding of H-NS to a second DNA tract, and thus part of the loop will be forced into a more ‘open’ configuration.
In reality, the loops are also disrupted (and then reformed) because of ongoing transcription and replication. If there are no specific determinants in a certain region, such as specific sites for DNA bending proteins, it is likely that the loop will subsequently reform in a different manner, reflecting the dynamics of bacterial chromosome organization.
The introduction of single-molecule techniques into the field of bacterial nucleoid studies has been a boost towards a better understanding at least of the individual components of the nucleoid. The single-molecule techniques described in this MicroReview have matured a lot over the last years, and the way is now open to more complex experiments. This means that one could start to investigate the combinations of nucleoid-associated proteins that have been shown to act antagonistically or in concert, and in particular to investigate dynamic aspects of these systems, in order to determine their respective roles. Furthermore, it will be of interest to analyse transcription at the single-molecule level within a context of bound nucleoid-associated proteins, and to investigate to what extent these proteins act as a barrier to a progressing RNA polymerase. It might be possible to analyse a sequence of events, e.g. transcriptional repression, relief of repression (activation), followed by active transcription, in a quasi-physiological context at the single-molecule level by combining these ideas in a single experiment. When single-molecule experiments can deal with that level of complexity, molecular biology and biophysics will be fully complementary and a mutual active exchange of insights may become reality.
We wish to acknowledge Maarten Noom for discussion and preparation of the figures. Furthermore we wish to thank Conrad Woldringh, Rolf Wagner and Gijs Wuite for discussion and/or critical reading of the manuscript and anonymous referees for their constructive comments. We are grateful to Hyon Choy for communicating results prior to publication. This work was supported by a Nederlandse Organisatie voor Wetenschappelijk Onderzoek ALW open competition grant.