Architecture of the Escherichia coli nucleoid

How genomes are organized within cells and how the 3D architecture of a genome influences cellular functions are significant questions in biology. A bacterial genomic DNA resides inside cells in a highly condensed and functionally organized form called nucleoid (nucleus-like structure without a nuclear membrane). The Escherichia coli chromosome or nucleoid is composed of the genomic DNA, RNA, and protein. The nucleoid forms by condensation and functional arrangement of a single chromosomal DNA with the help of chromosomal architectural proteins and RNA molecules as well as DNA supercoiling. Although a high-resolution structure of a bacterial nucleoid is yet to come, five decades of research has established the following salient features of the E. coli nucleoid elaborated below: 1) The chromosomal DNA is on the average a negatively supercoiled molecule that is folded as plectonemic loops, which are confined into many independent topological domains due to supercoiling diffusion barriers; 2) The loops spatially organize into megabase size regions called macrodomains, which are defined by more frequent physical interactions among DNA sites within the same macrodomain than between different macrodomains; 3) The condensed and spatially organized DNA takes the form of a helical ellipsoid radially confined in the cell; and 4) The DNA in the chromosome appears to have a condition-dependent 3-D structure that is linked to gene expression so that the nucleoid architecture and gene transcription are tightly interdependent, influencing each other reciprocally. Current advents of high-resolution microscopy, single-molecule analysis and molecular structure determination of the components are expected to reveal the total structure and function of the bacterial nucleoid.


Introduction
In many bacteria, the chromosome is a single covalently closed (circular) double-stranded DNA molecule that encodes the genetic information in a haploid form. The size of the DNA varies from 500,000 to several million base-pairs (bp) encoding from 500 to several thousand genes depending on the organism. The chromosomal DNA is present in cells in a highly condensed, organized form called nucleoid (nucleus-like), which is not encased by a nuclear membrane as in eukaryotic cells. The isolated nucleoid contains 80% DNA, 10% protein, and 10% RNA by weight [1,2]. In this exposition, we review our current knowledge about (i) how chromosomal DNA becomes the nucleoid, (ii) the factors involved therein, (iii) what is known about its structure, and (iv) how some of the DNA structural aspects influence gene expression, using the gram-negative bacterium Escherichia coli as a model system. We also highlight some related issues that need to be resolved. This exposition is an extension of past reviews on the subject [3,4].
There are two essential aspects of nucleoid formation; condensation of a large DNA into a small cellular space and functional organization of DNA in a three-dimensional form [5,6]. The haploid circular chromosome in E. coli consists of~4.6 x 10 6 bp. If DNA is relaxed in the B form, it would have a circumference of~1.5 millimeters (0.332 nm x 4.6 x 10 6 ) ( Fig 1A). However, a large DNA molecule such as the E. coli chromosomal DNA does not remain a straight rigid molecule in a suspension. Brownian motion will generate curvature and bends in DNA. The maximum length up to which a double-helical DNA remains straight by resisting the bending enforced by Brownian motion is~50 nm or 150 bp, which is called the persistence length. Thus, pure DNA becomes substantially condensed without any additional factors; at thermal equilibrium, it assumes a random coil form. The random coil of E. coli chromosomal DNA (Fig 1B) would occupy a volume (4/3 π r 3 ) of~523 μm 3 , calculated from the radius of gyration (Rg = ( p N a)/ p 6) where a is the Kuhn length (2 x persistence length), and N is the number of Kuhn length segments in the DNA (total length of the DNA divided by a). Although DNA is already condensed in the random coil form, it still cannot assume the volume of the nucleoid which is less than a micron ( Fig 1C). Thus, the inherent property of DNA is not sufficient: additional factors must help condense DNA further on the order of~10 3 (volume of the random coil divided by the nucleoid volume). The second essential aspect of nucleoid formation is the functional arrangement of DNA. Chromosomal DNA is not only condensed but also functionally organized in a way that is compatible with DNA transaction processes such as replication, recombination, segregation, and transcription ( Fig 1C). Almost five decades of research beginning in 1971 [1], has shown that the final form of the nucleoid Arrows represent bi-directional DNA replication. The genetic position of the origin of bi-directional DNA replication (oriC) and the site of chromosome decatenation (dif) in the replication termination region (ter) are marked. Colors represent specific segments of DNA as discussed in C. B. An illustration of a random coil form adopted by the pure circular DNA of E. coli at thermal equilibrium without supercoils and additional stabilizing factors [5,6]. C. A cartoon of the chromosome of a newly born E. coli cell. The genomic DNA is not only condensed by 1000-fold compared to its pure random coil form but is also spatially organized. oriC and dif are localized in the mid-cell, and specific regions of the DNA indicated by colors in A organize into spatially distinct domains. Six spatial domains have been identified in E. coli. Four domains (Ori, Ter, Left, and Right) are structured and two (NS-right and NS-left) are non-structured (See section 4 of the main text for details). The condensed and organized form of the DNA together with its associated proteins and RNAs is called nucleoid. Drawings are not in scale with each other.
https://doi.org/10.1371/journal.pgen.1008456.g001 arises from a hierarchical organization of DNA. At the smallest scale (1 -kb or less), nucleoidassociated DNA architectural proteins condense and organize DNA by bending, looping, bridging or wrapping DNA. At a larger scale (10 -kb or larger), DNA forms plectonemic loops, a braided form of DNA induced by supercoiling. At the megabase scale, the plectonemic loops coalesce into six spatially organized domains (macrodomains), which are defined by more frequent physical interactions among DNA sites within the same macrodomain than between different macrodomains [7]. Long-and short-range DNA-DNA connections formed within and between the macrodomains contribute to condensation and functional organization. Finally, the nucleoid is a helical ellipsoid with regions of highly condensed DNA at the longitudinal axis [8][9][10]. We discuss these organizational features of the nucleoid and their molecular basis below.

DNA condensation and organization by nucleoid-associated proteins (NAPs)
In eukaryotes, genomic DNA is condensed in the form of a repeating array of DNA-protein particles called nucleosomes [11][12][13].
A nucleosome consists of~146 bp of DNA wrapped around an octameric complex of the histone proteins. Although bacteria do not have histones, they possess a group of DNA binding proteins referred to as nucleoid-associated proteins (NAPs) that are functionally analogous to histones in a broad sense. NAPs are highly abundant and constitute a significant proportion of the protein component of the nucleoid [14].
A distinctive characteristic of NAPs is their ability to bind DNA in both a specific (either sequence-or structure-specific) and non-sequence specific manner. As a result, NAPs are dual function proteins. The specific binding of NAPs is mostly involved in gene-specific transcription, DNA replication, recombination, and repair. At the peak of their abundance, the number of molecules of many NAPs is several orders of magnitude higher than the number of specific binding sites in the genome. Therefore, it is reasoned that NAPs bind to the chromosomal DNA mostly in the non-sequence specific mode and it is this mode that is crucial for chromosome compaction. It is noteworthy that the so-called non-sequence specific binding of a NAP may not be completely random. There could be low-sequence specificity and or structural specificity due to sequence-dependent DNA conformation or DNA conformation created by other NAPs.
Although molecular mechanisms of how NAPs condense DNA in vivo are not well understood, based on the extensive in vitro studies it appears that NAPs participate in chromosome compaction via the following mechanisms: NAPs induce and stabilize bends in DNA, thus aid in DNA condensation by reducing the persistence length (Fig 2A). NAPs condense DNA by bridging, wrapping, and bunching that could occur between nearby DNA segments or distant DNA segments of the chromosome (Fig 2C, 2D and 2E). Another mechanism by which NAPs participate in chromosome compaction is by constraining negative supercoils in DNA thus contributing to the topological organization of the chromosome (see section 3).
There are at least 12 NAPs identified in E. coli [15]. Here, we focus on the most extensively studied NAPs, HU, IHF, H-NS, and Fis. Their abundance and DNA binding properties are summarized in Tables 1 and 2. Current models of how each NAP condenses and organizes DNA are discussed in detail below.
HU. Histone-like protein from E. coli strain U93 (HU) is an evolutionarily conserved protein in bacteria [28,29]. HU exists in E. coli as homo-and heterodimers of two subunits HUα and HUβ sharing 69% amino acid identity [30]. Although it is referred to as a histone-like protein, close functional relatives of HU in eukaryotes are high-mobility group (HMG) proteins, and not histones [31,32]. HU is a non-sequence specific DNA binding protein. It binds with low-affinity to any linear DNA. However, it preferentially binds with high-affinity to structurally distorted DNA ( In the structurally specific DNA binding mode, HU recognizes a common structural motif defined by bends or kinks created by distortion [17,18,39], whereas it binds to a linear DNA by locking the phosphate backbone [40]. While the high-affinity structurally-specific binding is required for specialized functions of HU such as site-specific recombination, DNA repair, DNA replication initiation, and gene regulation [41][42][43], it appears that the low-affinity general binding is involved in DNA condensation [40]. In chromatin-immunoprecipitation coupled with DNA sequencing ( ChIP-Seq), HU does not reveal any specific binding events. Instead, it displays a uniform binding across the genome presumably reflecting its mostly weak, non-sequence specific binding, thus masking the high-affinity binding in vivo (Fig 3).
In strains lacking HU, the nucleoid is "decondensed", consistent with a role of HU in DNA compaction [44]. The following in vitro studies suggest possible mechanisms of how HU might condense and organize DNA in vivo. Not only HU binds stably to distorted DNA with bends, but it also induces flexible bends even in a linear DNA at less than 100 nM concentration (Fig 2A) [45]. In contrast, HU shows the opposite architectural effect on DNA at higher physiologically-relevant concentrations [40,45]. It forms rigid nucleoprotein filaments causing the straitening of DNA and not the bending (Fig 2B). The filaments can further form a DNA network (DNA bunching) expandable both laterally and medially because of the HU-HU multimerization triggered by the non-sequence-specific DNA binding ( Fig 2D) [40]. How are these behaviors of HU relevant inside the cell? The formation of filaments requires high-density binding of HU on DNA, one HU dimer per 9-20 bp DNA [40,45]. But there is only one HU dimer every~150 bp of the chromosomal DNA based on the estimated abundance of 30,000 HU dimers per cell (4600000 bp /30,000) [16]. This indicates that flexible bends are more likely to occur in vivo. The flexible bending would cause condensation due to a reduction in the persistence length of DNA as shown by magnetic tweezers experiments [45], which allow studying condensation of a single DNA molecule by a DNA binding protein [48]. However, because of the cooperativity, the rigid filaments and networks could form in some regions in the chromosome. The filament formation alone does not induce condensation [45], but DNA networking or bunching can substantially contribute to condensation by bringing distant or nearby chromosome segments together [40].
IHF. Integration host factor (IHF) is structurally almost identical to HU [49] but behaves differently from HU in many aspects. Unlike HU, which preferentially binds to a structural motif regardless of the sequence, IHF preferentially binds to a specific DNA sequence even though the specificity arises through the sequence-dependent DNA structure and deformability. The specific binding of IHF at cognate sites bends DNA sharply by >160˚bend angle [49]. An occurrence of the cognate sequence motif is about 3000 in the E. coli genome [47]. The estimated abundance of IHF in the growth phase is about 6000 dimers per cell (Table 1). Assuming that one IHF dimer binds to a single motif and nucleoid contains more than one genome equivalent during the exponential growth phase, most of the IHF molecules would occupy specific sites in the genome and likely only condense DNA by inducing sharp bending (Fig 2A).
Besides preferential binding to a specific DNA sequence, IHF also binds to DNA in a nonsequence specific manner with the affinities similar to HU ( Table 2). The role of the non-specific binding of IHF in DNA condensation appears to be critical in the stationary phase  Fig  1A) additionally depicting the genome occupancy of indicated NAPs in the growth phase. B. The genome occupancy of indicated NAPs in the stationary phase. The genome layout is the same as in A. The genome occupancy of each NAP, determined by ChIP-Seq, is plotted as a histogram (bin size 300 bp) in which the bar height is indicative of relative binding enrichment. The figures were prepared in Circos/0.69-6 using the data from [46,47].
https://doi.org/10.1371/journal.pgen.1008456.g003 because the IHF abundance increases by five-fold in the stationary phase (Table 1) and the additional IHF dimers would likely bind the chromosomal DNA non-specifically [16,50,51]. Unlike HU, IHF does not form thick rigid filaments at higher concentrations. Instead, its nonspecific binding also induces DNA bending albeit the degree of bending is much smaller than that at specific sites and is similar to the flexible bending induced by HU in a linear DNA at low concentrations [52]. In vitro, the bending induced by non-specific binding of IHF can cause DNA condensation and promotes the formation of higher-order nucleoprotein complexes depending on the concentrations of potassium chloride and magnesium chloride [52]. Whether the higher-order DNA organization by IHF occurs in vivo needs further investigation.
H-NS. A distinguishable feature of histone-like or heat-stable nucleoid structuring protein Furthermore, H-NS is best known as a global gene silencer that preferentially inhibits transcription of horizontally transferred genes and it is the rigid filament that leads to gene silencing [68,69]. Taken together, it appears that the formation of rigid filaments is the most likely outcome of H-NS-DNA interactions in vivo that leads to gene silencing but does not induce DNA condensation. Consistently, the absence of H-NS does not change the nucleoid volume [70]. However, E. coli may experience high-magnesium concentration under some environmental conditions. In such conditions, H-NS can switch from its filament inducing form to the bridge inducing form that contributes to DNA condensation and organization.
Fis. Factor for Inversion Stimulation (Fis) is a sequence-specific DNA binding protein that binds to specific DNA sequences containing a 15-bp symmetric motif (Table 2) [24, 25, 71]. Like IHF, Fis induces DNA bending at cognate sites. The ability to bend DNA is apparent in the structure of Fis homodimer. A Fis homodimer possesses two helix-turn-helix (HTH) motifs, one from each monomer. An HTH motif typically recognizes the DNA major groove. However, the distance between the DNA recognition helices of the two HTH motifs in the Fis homodimer is 25 A˚, which is~8 A˚shorter than the pitch of a canonical B-DNA, indicating that the protein must bend or twist DNA to bind stably [72,73]. Consistently, the crystal structure of Fis-DNA complexes shows that the distance between the recognition helices remains unchanged whereas DNA curves in the range of 60-75˚bend angles [25]. There are 1464 Fis binding regions distributed across the E. coli genome and a binding motif, identified computationally, matches with the known 15-bp motif [46,74]. Specific binding of Fis at such sites would induce bends in DNA, thus contribute to DNA condensation by reducing persistence length of DNA. Furthermore, many Fis binding sites occur in tandem such as those in the stable RNA promoters, e.g., P1 promoter of rRNA operon rrnB. The coherent bending by Fis at the tandem sites is likely to create a DNA micro-loop ( Fig 2E) that can further contribute to DNA condensation [75].
Besides high-affinity specific binding to cognate sites, Fis can bind to a random DNA sequence ( Table 2). The non-specific DNA binding is significant because Fis is as abundant as HU in the growth phase (Table 1). Therefore, most of Fis molecules are expected to bind DNA in a non-sequence specific manner. Magnetic tweezers experiments show that this non-specific binding of Fis can contribute to DNA condensation and organization [76,77]. Fis causes mild condensation of a single DNA molecule at <1 mM but induces substantial folding through the formation of DNA loops of an average size of~800-bp at >1 mM. The loops in magnetic tweezers experiments are distinct from the micro-loops created by coherent DNA bending at cognate sites, as they require the formation of high-density DNA-protein complexes achieved by sequence-independent binding. Although occurrence of such loops in vivo remains to be demonstrated, high-density binding of Fis may occur in vivo through the concerted action of both specific and non-specific binding. The in-tandem occurrence of specific sites might initiate a nucleation reaction similar to that of H-NS, and then non-specific binding would lead to the formation of localized high-density Fis arrays. The bridging between these localized regions ( Fig 2C) can create large DNA loops [77]. Fis is exclusively present in the growth phase and not in the stationary phase [78,79]. Thus, any role in chromosomal condensation by Fis must be specific to growing cells.

DNA condensation and organization by nucleoid-associated RNAs (naRNAs)
Early studies examining the effect of RNase A treatment on isolated nucleoids indicated that RNA participated in the stabilization of the nucleoid in the condensed state [80]. Moreover, treatment with RNase A disrupted the DNA fibers into thinner fibers, as observed by atomic force microscopy of the nucleoid using the "on-substrate lysis procedure" [81]. These findings demonstrated the participation of RNA in the nucleoid structure, but the identity of the RNA molecule(s) remained unknown until recently [44]. Most of the studies on HU focused on its DNA binding. However, HU also binds to dsRNA and RNA-DNA hybrids with a lower affinity similar to that with a linear dsDNA [82]. Moreover, HU preferentially binds to RNA containing secondary structures and an RNA-DNA hybrid in which the RNA contains a nick or overhang [82,83]. The binding affinities of HU with these RNA substrates are similar to those with which it binds to distorted DNA. An immunoprecipitation of HU-bound RNA coupled to reverse transcription and microarray (RIP-Chip) study as well as an analysis of RNA from purified intact nucleoids identified nucleoid-associated RNA molecules that interact with HU [44]. Several of them are non-coding RNAs, and one such RNA named naRNA4 (nucleoidassociated RNA 4), is encoded in a repetitive extragenic palindrome (REP325). In a strain lacking REP325, the nucleoid is decondensed as it is in a strain lacking HU [44]. naRNA4 most likely participate in DNA condensation by connecting DNA segments in the presence of HU [84]. Recent studies provide insights into the molecular mechanism of how naRNA4 establishes DNA-DNA connections. The RNA targets regions of DNA containing cruciform structures and forms an RNA-DNA complex that is critical for establishing DNA-DNA connections [85]. Surprisingly, although HU helps in the formation of the complex, it is not present in the final complex, indicating its potential role as a catalyst (chaperone). The nature of the RNA-DNA complex remains puzzling because the formation of the complex does not involve extensive Watson/Crick base pairing but is sensitive to RNase H, which cleaves RNA in an RNA-DNA hybrid. Moreover, the complex binds to an antibody specific to RNA-DNA hybrids.

DNA condensation and organization by supercoiling
Supercoiling. Because of its helical structure, a double-stranded DNA molecule becomes topologically constrained in the covalently closed circular form which eliminates the rotation of the free ends [86]. The number of times the two strands cross each other in a topologically constrained DNA is called the linking number (Lk), which is equivalent to the number of helical turns or twists in a circular molecule (Fig 4). The Lk of a topological DNA remains invariant, no matter how the DNA molecule is deformed, as long as neither strand is broken.
The Lk of DNA in the relaxed form is defined as Lk 0 . For any DNA, Lk 0 can be calculated by dividing the length (in bp) of the DNA by the number of bp per helical turn. This is equal to 10.4 bp for the relaxed B-form DNA. Any deviation from Lk 0 causes supercoiling in DNA. A decrease in the linking number (Lk<Lk 0 ) creates negative supercoiling (Fig 4) whereas an increase in the linking number (Lk>Lk 0 ) creates positive supercoiling (see [87,88] for more detail of supercoiling).
The supercoiled state (when Lk is not equal to Lk 0 ) results in a transition in DNA structure that can manifest as a change in the number of twists (negative <10.4 bp/turn, positive >10.4 bp per turn) and/or in the formation of writhes, called supercoils (Fig 4). Thus, Lk is mathematically defined as a sign dependent sum of the two geometric parameters, twist and writhe. A quantitative measure of supercoiling that is independent of the size of DNA molecules is the supercoiling density (σ) where σ = ΔLk/Lk 0 .
Writhes can adopt two structures; plectoneme and solenoid or toroid (Fig 4). A plectonemic structure arises from the interwinding of the helical axis (Fig 4). Toroidal supercoils originate when DNA forms several spirals, around an axis and not intersecting with each other, like those in a telephone cord. The writhes in the plectonemes form are right-and left-handed in negatively or positively supercoiled DNA, respectively. The handedness of the toroidal supercoils is opposite to those of plectonemes. Both plectonemes and toroidal supercoils can be either in a free form or restrained in a bound form with proteins. The best example of the bound toroidal supercoiling in biology is the eukaryotic nucleosome in which DNA wraps around histones (Fig 5) [12].
The E. coli genome is organized as plectonemic supercoils. In most bacteria, DNA is present in a supercoiled form. The circular nature of the E. coli chromosome makes it a topologically constrained molecule that is mostly negatively supercoiled with an estimated average supercoiling density (σ) of -0.05 [89]. In the eukaryotic chromatin, DNA is found mainly in the toroidal form that is restrained and defined by histones through the formation of nucleosomes. In contrast, in the E. coli nucleoid, about half of the chromosomal DNA is organized in the form of free, plectonemic supercoils ( Fig 5) [90][91][92]. The remaining DNA is restrained in either the plectonemic form or alternative forms (see section 5.3.3), including but not limited to the toroidal form, by interaction with proteins such as NAPs. Thus, plectonemic supercoils represent effective supercoiling of the E. coli genome that is responsible for its condensation and organization. Both plectonemic and toroidal supercoiling aid in DNA condensation. It is noteworthy that because of the branching of plectonemic structures, it provides less DNA condensation than does the toroidal structure. The same size DNA molecule with equal supercoiling densities is more compact in a toroidal form than in a plectonemic form. In addition to condensing DNA, supercoiling aids in DNA organization. It promotes DNA disentanglement by reducing the probability of catenation [93]. Supercoiling also helps bring two distant sites of DNA in proximity thereby promoting a potential functional interaction between different segments of DNA. Rules of DNA topology are explained using such a molecule (ccc-DNA) in which a numerical parameter called the linking number (Lk) defines the topology. Lk is a mathematical sum of two geometric parameters, twist (Tw) and writhe (Wr). A twist is the crossing of two strands, and writhe is coiling of the DNA double helix on its axis that requires bending. Lk is always an integer and remains invariant no matter how much the two strands are deformed. It can only be changed by introducing a break in one or both DNA strands by DNA metabolic enzymes called topoisomerases. B. A torsional strain created by a change in Lk of a relaxed, topologically constrained DNA manifests in the form of DNA supercoiling. A decrease in Lk (Lk<Lk 0 ) induces negative supercoiling whereas an increase in Lk (Lk>Lk 0 ) induces positive supercoiling. Only negative supercoiling is depicted here. For example, if a cut is introduced into a ccc-DNA and four turns are removed before rejoining the two strands, the DNA becomes negatively supercoiled with a decrease in the number of twists or writhe or both. Writhe can adopt two types of geometric structures called plectoneme and toroid. Plectonemes are characterized by the interwinding of the DNA double helix and an apical loop, whereas spiraling of DNA double helix around an axis forms toroids. Topoisomerases. Topoisomerases are a particular category of DNA metabolic enzymes that create or remove supercoiling by breaking and then re-ligating DNA strands [94]. E. coli possesses four topoisomerases (Table 3). DNA gyrase introduces negative supercoiling in the presence of ATP and removes positive supercoiling in the absence of ATP [95]. Across all forms of life, DNA gyrase is the only topoisomerase that can create negative supercoiling and it is because of this unique ability that bacterial genomes possess free negative supercoils; DNA gyrase is found in all bacteria but absent from higher eukaryotes. In contrast, Topo I opposes DNA gyrase by relaxing the negatively supercoiled DNA [96,97]. There is genetic evidence to suggest that a balance between the opposing activities of DNA gyrase and Topo I are responsible for maintaining a steady-state level of average negative superhelicity in E. coli [98]. Both enzymes are essential for E. coli survival. A null strain of topA, the gene encoding Topo I, survives only because of the presence of suppressor mutations in the genes encoding DNA gyrase. These mutations result in reduced gyrase activity, suggesting that excess negative supercoiling due to the absence of Topo I is compensated by reduced negative supercoiling activity of DNA gyrase. Topo III is dispensable in E. coli and is not known to have any role in supercoiling in E. coli [99]. The primary function of Topo IV is to resolve sister chromosomes. However, it has   [102]. According to their model, transcribing RNA polymerase (RNAP) sliding along DNA forces the DNA to rotate on its helical axis. A hindrance in the free rotation of DNA might arise due to a topological constraint, causing the DNA in front of RNAP to become over-twisted (positively supercoiled) and the DNA behind RNAP would become under-twisted (negatively supercoiled) (Fig 6). It has been found that a topological constraint is not needed because RNAP generates sufficient torque that causes supercoiling even in a linear DNA template [103]. If DNA is already negatively supercoiled, this action relaxes existing negative supercoils before causing a buildup of positive supercoils ahead of RNAP and introduces more negative supercoils behind RNAP. In principle, DNA gyrase and Topo I should remove excess positive and negative supercoils respectively but if the RNAP elongation rate exceeds the turnover of the two enzymes, transcription contributes to the steady-state level of supercoiling.
Control of supercoiling by NAPs. In the eukaryotic chromatin, DNA is rarely present in the free supercoiled form because nucleosomes restrain almost all negative supercoiling through the tight binding of DNA to histones. Similarly, in E. coli, nucleoprotein complexes formed by NAPs restrain half of the supercoiling density of the nucleoid [89,92]. In other words, if a NAP dissociates from a nucleoprotein complex, the DNA would adopt the free, plectonemic form. DNA binding of HU, Fis, and H-NS has been experimentally shown to restrain negative supercoiling in a relaxed but topologically constrained DNA [104][105][106][107][108]. They can do so either by changing the helical pitch of DNA or generating toroidal writhes by DNA bending and wrapping (Fig 2). Alternatively, NAPs can preferentially bind to and stabilize other forms of the underwound DNA such as cruciform structures and branched plectonemes. Fis has been reported to organize branched plectonemes through its binding to cross-over regions ( Fig 5) and HU preferentially binds to cruciform structures [108].
NAPs also regulate DNA supercoiling indirectly. Fis can modulate supercoiling by repressing the transcription of the genes encoding DNA gyrase [109]. There is genetic evidence to suggest that HU controls supercoiling levels by stimulating DNA gyrase and reducing the activity of Topo I [110,111]. In support of the genetic studies, HU was shown to stimulate DNA gyrase-catalyzed decatenation of DNA in vitro [112]. It is unclear mechanistically how HU modulates the activities of the gyrase and Topo I. HU might physically interact with DNA gyrase and Topo I or DNA organization activities of HU such as DNA bending may facilitate or inhibit the action of DNA gyrase and Topo I respectively.
Plectonemic supercoils organize into multiple topological domains. One of the striking features of the nucleoid is that plectonemic supercoils are organized into multiple topological domains (Fig 7) [113]. In other words, a single cut in one domain will only relax that domain and not the others. A topological domain forms because of a supercoiling-diffusion barrier. Independent studies employing different methods have reported that the topological domains are variable in size ranging from 10-400 kb [91,113,114]. Random placement of barriers commonly observed in these studies seems to explain the wide variability in the size of domains.
Although identities of domain barriers remain to be established, possible mechanisms responsible for the formation of the barriers include: (i) A domain barrier could form when a protein with an ability to restrain supercoils simultaneously binds to two distinct sites on the chromosome forming a topologically isolated DNA loop or domain. It has been experimentally demonstrated that protein-mediated looping in supercoiled DNA can create a topological domain [115,116]. NAPs such as H-NS and Fis are potential candidates, based on their DNA looping abilities and the distribution of their binding sites. (ii) Bacterial interspersed mosaic elements (BIMEs) also appear as potential candidates for domain barriers. BIMEs are palindromic repeats sequences that are usually found between genes. A BIME has been shown to impede supercoiling diffusion in a synthetically designed topological cassette inserted in the E. coli chromosome [117]. There are~600 BIMEs distributed across the genome, possibly dividing the chromosome into 600 topological domains [118]. (iii) Barriers could also result from the attachment of DNA to the cell membrane through a protein that binds to both DNA and membrane or through nascent transcription and the translation of membrane-anchored proteins (see section 5.2). (iv) Transcription activity can generate supercoiling-diffusion barriers. An actively transcribing RNAP has been shown to block the dissipation of plectonemic supercoils, thereby forming a supercoiling-diffusion barrier [119][120][121].

Spatial organization of the nucleoid
Chromosomal interaction domains. In recent years, the advent of a molecular method called chromosome conformation capture (3C) has allowed studying a high-resolution spatial organization of chromosomes in both bacteria and eukaryotes [122]. 3C and its version that is coupled with deep sequencing (Hi-C) [123] determine physical proximity, if any, between any two genomic loci in 3D space (Fig 8A and 8B). A high-resolution contact map of bacterial chromosomes including the E. coli chromosome has revealed that a bacterial chromosome is segmented into many highly self-interacting regions called chromosomal interaction domains (CIDs) (Fig 8B) [124][125][126]. CIDs are equivalent to topologically associating domains (TADs) observed in many eukaryotic chromosomes [127], suggesting that the formation of CIDs is a general phenomenon of genome organization. Two characteristics define CIDs or TADs. First, genomic regions of a CID physically interact with each other more frequently than with the genomic regions outside that CID or with those of a neighboring CID. Second, the presence of a boundary between CIDs that prevents physical interactions between genomic regions of two neighboring CIDs.
The E. coli chromosome was found to consist of 31 CIDs in the growth phase [124]. The size of the CIDs ranged from 40 to~300 kb. It appears that a supercoiling-diffusion barrier responsible for segregating plectonemic DNA loops into topological domains functions as a CID boundary in E. coli and many other bacteria. In other words, the presence of a supercoiling-diffusion barrier defines the formation of CIDs. Findings from the Hi-C probing of chromosomes in E. coli [124], Caulobacter crescentus [125], and Bacillus subtilis [126] converge on a model that CIDs form because plectonemic looping together with DNA organization activities of NAPs promotes physical interactions among genomic loci, and a CID boundary consists of a plectoneme-free region (PFR) that prevents these interactions (Fig 8C). A PFR is created due to high transcription activity because the helical unwinding of DNA by actively transcribing RNAP restrains plectonemic supercoils. As a result, supercoil dissipation is also blocked, creating a supercoiling-diffusion barrier. Indirect evidence for this model comes from an observation that CIDs of bacterial chromosomes including the E. coli chromosome display highly transcribed genes at their boundaries, indicating a role of transcription in the formation of a CID boundary [124,125]. More direct evidence came from a finding that the placement of a highly transcribed gene at a position where no boundary was present created a new CID boundary in the C. crescentus chromosome [125]. However, not all CID boundaries correlated with highly transcribed genes in the E. coli chromosome suggesting that other unknown factors are also responsible for the formation of CID boundaries and supercoiling diffusion barriers.
Macrodomains. Plectonemic DNA loops organized as topological domains or CIDs appear to coalesce further to form large spatially distinct domains called macrodomains [128,129]. In E. coli, macrodomains were initially identified as large segments of the genome whose DNA markers localized together (co-localized) in fluorescence in situ hybridization (FISH) studies. A large genomic region (~1-Mb) covering oriC (origin of chromosome replication) locus co-localized and was called Ori macrodomain. Likewise, a large genomic region (~1-Mb) covering the replication terminus region (ter) co-localized and was called Ter macrodomain. Macrodomains were later identified based on how frequently pairs of lambda att sites Chromosome conformation capture (3C) methods probe 3D genome organization by quantifying physical interactions between genomic loci that are nearby in 3D-space but may be far away in the linear genome. A genome is cross-linked with formaldehyde to preserve physical contacts between genomic loci. Subsequently, the genome is digested with a restriction enzyme. In the next step, a DNA ligation is carried out under diluted DNA concentrations to favor intra-molecular ligation (between cross-linked fragments that are brought into physical proximity by 3D genome organization). A frequency of ligation events between distant DNA sites reflects a physical interaction. In the 3C method, ligation junctions are detected by the semiquantitative PCR amplification in which amplification efficiency is a rough estimate of pairwise physical contact between genomic regions of interests and its frequency. The 3C method probes a physical interaction between two specific regions identified a priori, whereas its Hi-C version detects physical interactions between all possible pairs of genomic regions simultaneously. In the Hi-C method, digested ends are filled in with a biotinylated adaptor before ligation. Ligated fragments are sheared and then enriched by a biotin-pull down. Ligation junctions are then detected and quantified by the paired-end next-generation sequencing methods. B. Hi-C data are typically represented in the form of a two-dimensional matrix in which the x-axis and y-axis represent the genomic coordinates. The genome is usually divided into bins of a fixed size, e.g., 5-kb. The size of bins essentially defines the contact resolution. Each entry in the matrix, m ij , represents the number of chimeric sequencing reads mapped to genomic loci in bins i and j. A quantification of the reads (represented as a heatmap) denotes the relative frequency of contacts between genomic loci of bins i and j. A prominent feature of the heatmap is a diagonal line that appears due to more frequent physical interaction between loci that are very close to each other in the linear genome. The intensity as we move away from this diagonal line represents the relative frequency of physical interaction between loci that are far away from each other in the linear genome. Triangles of high-intensity along the diagonal line represent highly self-interacting chromosomal interaction domains (CIDs) that are separated by a boundary region that consists of a smaller number of interactions. C. In many bacterial species including E. coli, it appears that supercoiled topological domains organize as CIDs. Plectonemic supercoiling promotes a high level of interaction among genomic loci within a CID, and a plectoneme-free region (PFR), created due to high transcription activity, acts as a CID boundary. Nucleoid-associated proteins, depicted as closed circles, stabilize the supercoiling-mediated interactions. The actively transcribing RNA polymerase (depicted as a green sphere) in the PFR blocks dissipation of supercoiling between the two domains thus acts as a supercoiling diffusion barrier. The size of the CIDs ranges between 30-400 kb. Several triangles (CIDs) merge to form a bigger triangle that represents a macrodomain. In other words, CIDs of a macrodomain physically interact with each other more frequently than with CIDs of a neighboring macrodomain or with genomic loci outside of that macrodomain. A macrodomain may comprise of several CIDs. For simplicity, a macrodomain comprising of only two CIDs is shown.
https://doi.org/10.1371/journal.pgen.1008456.g008 that were inserted at various distant locations in the chromosome recombined with each other. In this recombination-based method, a macrodomain was defined as a large genomic region whose DNA sites can primarily recombine with each other, but not with outside of that macrodomain. The recombination-based method confirmed the Ori and Ter macrodomains that were identified in FISH studies and identified two additional macrodomains [7,130].
The two additional macrodomains were formed by the additional~1-Mb regions flanking the Ter and were referred to as Left and Right. These four macrodomains (Ori, Ter, Left, and Right) comprised most of the genome, except for two genomic regions flanking the Ori (Fig  1). These two regions (NS-L and NS-R) were more flexible and non-structured compared to a macrodomain as DNA sites in them recombined with DNA sites located in macrodomains on both sides. The genetic position of oriC appears to dictate the formation of macrodomains, because repositioning of oriC by genetic manipulation results in the reorganization of macrodomains. For example, genomic regions closest to the oriC always behave as an NS regardless of DNA sequence and regions further away behave as macrodomains [131].
The Hi-C technique further confirmed a hierarchical spatial organization of CIDs in the form of macrodomains [124]. In other words, CIDs of a macrodomain physically interacted with each other more frequently than with CIDs of a neighboring macrodomain or with genomic loci outside of that macrodomain (Fig 8B and 8C). The Hi-C data showed that the E. coli chromosome was partitioning into two distinct domains. The region surrounding ter formed an insulated domain that overlapped with the previously identified Ter macrodomain. DNA-DNA contacts in this domain occurred only in the range of up to~280 kb. The rest of the chromosome formed a single domain whose genomic loci exhibited contacts in the range of >280-kb. While most of the contacts in this domain were restricted to a maximum distance of~500 kb, there were two loose regions whose genomic loci formed contacts at even greater distances (up to~1 Mb). These loose regions corresponded to the previously identified flexible and less-structured regions (NS). The boundaries of the insulated domain encompassing ter and the two loose regions identified by the Hi-C method segmented the entire chromosome into six regions that correspond with the four macrodomains and two NS regions defined by recombination-based assays. Thus, the two approaches were in good agreement with one another.

Proteins that drive macrodomain formation
MatP. A search for protein(s) responsible for macrodomain formation led to the identification of Macrodomain Ter protein (MatP) [27]. MatP almost exclusively binds in the Ter macrodomain by recognizing a 13-bp motif called the macrodomain ter sequence (matS) ( Table 2). There are 23 matS sites present in the Ter macrodomain, on average there is one site every 35-kb. Further evidence of MatP binding in the Ter macrodomain comes from fluorescence imaging of MatP. Discrete MatP foci were observed that co-localized with Ter macrodomain DNA markers [27]. A strong enrichment of ChIP-Seq signal in the Ter macrodomain also corroborates the preferential binding of MatP to this macrodomain (Fig 9).
MatP condenses DNA in the Ter macrodomain because the lack of MatP increased the distance between two fluorescent DNA markers located 100-kb apart in the Ter macrodomain. Furthermore, MatP is a critical player in insulating the Ter macrodomain from the rest of the chromosome [124]. It promotes DNA-DNA contacts within the Ter macrodomain but prevents contacts between the DNA loci of Ter domain and those of flanking regions. How does MatP condense DNA and promote DNA-DNA contacts? The experimental results are conflicting. MatP can form a DNA loop between two matS sites in vitro and its DNA looping activity depends on MatP tetramerization. Tetramerization occurs via coiled-coil interactions between two MatP molecules bound to DNA [133]. One obvious model based on in vitro results is that MatP promotes DNA-DNA contacts in vivo by bridging matS sites (Fig 10A). However, although MatP connected distant sites in Hi-C studies, it did not specifically connect the matS sites [124]. Furthermore, a MatP mutant that was unable to form tetramers behaved like wild-type. These results argue against the matS bridging model for Ter organization, leaving the mechanism of MatP action elusive. One possibility is that MatP spreads to nearby DNA segments from its primary matS binding site and bridge distant sites via a mechanism that does not depend on the tetramerization.
MukBEF. MukB belongs to a family of ATPases called structural maintenance of chromosome proteins (SMCs), which participate in higher-order chromosome organization in eukaryotes [139]. Two MukB monomers associate via continuous antiparallel coiled-coil interaction forming a 100-nm long rigid rod. A flexible hinge region occurs in the middle of the rod [140,141]. Due to the flexibility of the hinge region, MukB adopts a characteristic Vshape of the SMC family (Fig 10B). The non-SMC subunits associating with MukB are MukE and MukF. The association closes the V formation, resulting in large ring-like structures ( Fig  10B). MukE and MukF are encoded together with MukB in the same operon in E. coli [142]. The deletion of either subunit results in the same phenotype suggesting that the MukBEF complex is the functional unit in vivo [137]. DNA binding activities of the complex reside in the MukB subunit, whereas MukE and MukF modulate MukB activity.
MukBEF complex, together with Topo IV, is required for decatenation and repositioning of newly replicated oriCs [132,[143][144][145][146] The role of MukBEF is not restricted during DNA replication [147]. It organizes and condenses DNA even in non-replicating cells. The recent highresolution chromosome conformation map of the MukB-depleted E. coli strain reveals that   [124]. How is MukB prevented from acting in the Ter macrodomain? MatP physically interacts with MukB, thus preventing MukB from localizing to the Ter macrodomain [132]. This is evident in the DNA binding of MatP and MukB in the Ter macrodomain. MatP DNA binding is enriched in the Ter macrodomain, whereas MukB DNA binding is reduced compared to the rest of the genome (Fig 9). Furthermore, in a strain already lacking MatP, the absence of MukB causes a reduction in DNA contacts throughout the chromosome, including the Ter macrodomain [124]. This result agrees with the view that MatP displaces MukB from the Ter domain.
How does the MukBEF complex function to organize the E. coli chromosome? According to the current view, SMC complexes organize chromosomes by extruding DNA loops [148]. SMC complexes translocate along DNA to extrude loops in a cis-manner (on the same DNA molecule), wherein the size of loops depends on the processivity of the complex. SMC complexes from different organisms differ in the mechanism of loop extrusion [148]. Single molecule fluorescence microscopy of MukBEF in E. coli suggests that the minimum functional unit in vivo is a dimer of dimers (Fig 10B) [137]. This unit is formed by joining of two ATP-bound MukBEF complexes through MukF-mediated dimerization. MukBEF localizes in the cell as 1-3 clusters that are elongated parallel to the long axis of the cell. Each cluster contains an average~8-10 dimers of dimers. According to the current model, the MukBEF extrudes DNA loops in a "rock-climbing" manner ( Fig 10C) [137,149]. A dimer of the dimers releases one segment of DNA and capture a new DNA segment without dissociating from the chromosome. Besides DNA looping, a link between negative supercoiling and in vivo MukBEF function together with the ability of the MukB subunit to constrain negative supercoils in vitro suggests that MukBEF organizes DNA by generating supercoils [150][151][152]. A full understanding of a molecular mechanism of MukBEF action in vivo warrants further investigations. Spatial organization of the nucleoid by NAPs and naRNAs. In addition to contributing to the chromosome compaction by bending, bridging, and looping DNA at a smaller scale (~1-kb), NAPs participate in DNA condensation and organization by promoting long-rang DNA-DNA contacts. Two NAPs, Fis and HU, emerged as the key players in promoting longrange DNA-DNA contacts that occur throughout the chromosome [124]. It remains to be studied how DNA organization activities of Fis and HU that are well understood at a smaller scale (~1-kb) results in the formation of long-range DNA-DNA interactions. Nonetheless, some of the HU-mediated DNA interactions require the presence of naRNA4 [84]. naRNA4 also participates in making long-range DNA contacts. HU catalyzes some of the contacts, not all, suggesting that RNA participates with other NAPs in forming DNA contacts. HU also appears to act together with MukB to promote long-range DNA-DNA interactions. This view is based on observations that the absence of either HU or MukB caused a reduction in the same DNA-DNA contacts [124]. It is unclear how MukB and HU potentially act together in dimeric form, the dimerization of MukF results in an elongated hexameric complex (MukE 2 F) 2 [134]. In the absence of ATP, the (MukE 2 F) 2 complex binds to the MukB head domains through the C-terminal domain of MukF to form a symmetric MukBEF complex (shown on the left). The stoichiometry of the symmetric complex is B 2 (E 2 F) 2 . The ATP binding between the MukB head domains forces the detachment of one MukF molecule and two MukE molecules [134,136]. As a result, an asymmetric MukBEF complex of the stoichiometry B 2 (E 2 F) 1 is formed. Since MukF readily dimerizes, the MukF dimerization can potentially join two ATP-bound asymmetric molecules resulting in the formation of a dimer of dimers with the stoichiometry of B 4 (E 2 F) 2 (shown on the right). The stoichiometry of the MukBEF complex in vivo is estimated to be B 4 (E 2 F) 2 suggesting that a dimer of dimers is the functional unit in vivo [137]. C. A model for loop extrusion by a MukBEF dimer of dimers. A dimer of dimer loads onto DNA (depicted as a grey line) through DNA binding domains of MukB. MukB has been shown to bind DNA via its hinge region and the top region of its head domain [134,138]. The translocation of the complex away from its loading site then extrudes DNA loops. The loops are extruded in a rock-climbing manner by the coordinated opening and closing of the MukBEF ring through the MukB head disengagement that occurs due to coordinated ATP hydrolysis in the two dimers [137]. Dark and light blue circles represent ATP binding and hydrolysis events respectively. MukE is not shown in the complex for simplicity. https://doi.org/10.1371/journal.pgen.1008456.g010 promoting DNA-DNA interactions. The two proteins may interact physically. Alternatively, while MukBEF extrudes large DNA loops, HU condenses and organizes those loops by mechanisms described in section 2.1.1.
Spatial organization of the nucleoid by functional relatedness of genes. There are reports that functionally-related genes of E. coli are physically together in 3-D space within the chromosome even though they are far apart by genetic distance. Spatial proximity of functionally-related genes not only makes the biological functions more compartmentalized and efficient but would also contribute to the folding and spatial organization of the nucleoid. A recent study using fluorescent markers for detection of specific DNA loci examined pairwise physical distances between the seven rRNA operons that are genetically separated from each other (by as much as two million bp). It reported that all of the operons, except rrnC, were in physical proximity [153,154]. Surprisingly, 3C-seq studies did not reveal the physical clustering of rrn operons [124], contradicting the results of the fluorescence-based study. Therefore, further investigation is required to resolve these contradicting observations. In another example, GalR forms an interaction network of GalR binding sites that are scattered across the chromosome [155]. GalR is a transcriptional regulator of the galactose regulon comprised of genes encoding enzymes for transport and metabolism of the sugar D-galactose [156]. GalR exists in only one to two foci in cells [155] and can self-assemble into large ordered structures [157]. Therefore, it appears that DNA-bound GalR multimerizes to form long-distance interactions.

Nucleoid global shape and structure
The nucleoid is a helical ellipsoid, radially confined in the cell. Conventional transmission electron microscopy (TEM) of chemically fixed E. coli cells portrayed the nucleoid as an irregularly shaped organelle. However, wide-field fluorescence imaging of live nucleoids in 3D revealed a discrete, ellipsoid shape (Fig 11) [4,[8][9][10]. The overlay of a phasecontrast image of the cell and the fluorescent image of the nucleoid showed a close  [8]. B. Cross-sectioning of the E. coli nucleoid visualized by HU-mCherry. Fluorescence intensity is taken as a proxy for DNA density and is represented by blue to red in increasing order. Data were taken from [9].
https://doi.org/10.1371/journal.pgen.1008456.g011 juxtaposition only in the radial dimension along its entire length of the nucleoid to the cell periphery. This finding indicates the radial confinement of the nucleoid [8]. A detailed examination of the 3D fluorescence image after cross-sectioning perpendicular to its long axis further revealed two global features of the nucleoid: curvature and longitudinal, high-density regions. Examining the chirality of the centerline of the nucleoid by connecting the center of intensity of each cross-section showed that the overall nucleoid shape is curved (Fig 11A) [158]. The fluorescence intensity distribution in the cross-sections revealed a density substructure, consisting of curved, high-density regions or bundles at the central core, and low-density regions at the periphery ( Fig 11B) [8,9]. One implication of the radial confinement is that it determines the curved shape of the nucleoid. According to one model, the nucleoid is forced to bend because it is confined into a cylindrical E. coli cell whose radius is smaller than its bendable length (persistence length) [8]. This model was supported by observations that removal of the cell wall or inhibition of cell wall synthesis increased the radius of the cell and resulted in a concomitant increase in the helical radius and a decrease in the helical pitch in the nucleoid [8].
Connections between nucleoid and cell membrane. An expansion force due to connections between DNA within the nucleoid and cell membrane (DNA-membrane connections) appears to function in opposition to condensation forces to maintain an optimal condensation level of the nucleoid. Cell-fractionation and electron microscopy studies first indicated the possibility of DNA-membrane connections [159,160]. There are now several known examples of DNA-membrane connections. Transertion is a mechanism of concurrent transcription, translation, and insertion of nascent membrane proteins that forms transient DNA-membrane contacts [161]. Transertion of two membrane proteins, LacY and TetA, has been demonstrated to cause the repositioning of chromosomal loci toward the membrane [162]. Another mechanism of nucleoid-membrane connections is through a direct contact between membraneanchored transcription regulators and their target sites in the chromosome. One example of such as transcription regulator in E. coli is CadC. CadC contains a periplasmic sensory domain and a cytoplasmic DNA binding domain. Sensing of an acidic environment by its periplasmic sensory domain stimulates the DNA binding activity of CadC, which then activates transcription of its target genes [163]. The membrane-localization of genes regulated by a membraneanchored transcription regulator is yet to be demonstrated. Nonetheless, activation of target genes in the chromosome by these regulators is expected to result in a nucleoid-membrane contact albeit it would be a dynamic contact. Besides these examples, the chromosome is also specifically anchored to the cell membrane through protein-protein interaction between DNA-bound proteins, e.g., SlmA and MatP, and the divisome [164,165].
Since membrane-protein encoding genes are distributed throughout the genome, dynamic DNA-membrane contacts through transertion can act as a nucleoid expansion force. This expansion force would function in opposition to condensation forces to maintain an optimal condensation level. The formation of highly condensed nucleoids upon the exposure of E. coli cells to chloramphenicol, which blocks translation, provides support for the expansion force of transient DNA-membrane contacts formed through transertion. The round shape of overlycondensed nucleoids after chloramphenicol treatment also suggests a role for transertionmediated DNA-membrane contacts in defining the ellipsoid shape of the nucleoid.

Growth-phase dependent nucleoid dynamics
The nucleoid reorganizes in stationary phase cells suggesting that the nucleoid structure is highly dynamic, determined by the physiological state of cells. A comparison of high-resolution contact maps of the nucleoid revealed that the long-range contacts in the Ter macrodomain increased in the stationary phase, compared to the growth phase [124]. Furthermore, CID boundaries in the stationary phase were different from those found in the growth phase. Finally, nucleoid morphology undergoes a massive transformation during prolonged stationary phase [166]; the nucleoid exhibits ordered, toroidal structures [167].
Growth-phase specific changes in nucleoid structure could be brought about by a change in levels of nucleoid-associated DNA architectural proteins (the NAPs and the Muk subunits), supercoiling, and transcription activity. The abundance of NAPs and the Muk subunits changes according to the bacterial growth cycle. Fis and the starvation-induced DNA binding protein Dps, another NAP, are almost exclusively present in the growth phase and stationary phase respectively. Fis levels rise upon entry into the exponential phase and then rapidly decline while cells are still in the exponential phase, reaching levels that are undetectable in the stationary phase [168]. While Fis levels start to decline, levels of Dps start to rise and reach a maximum in the stationary phase [16]. A dramatic transition in the nucleoid structure observed in the prolonged stationary phase has been mainly attributed to Dps. It forms DNA/ crystalline assemblies that act to protect the nucleoid from DNA damaging agents present during starvation [167].
HU, IHF, and H-NS are present in both the growth phase and the stationary phase [16]. However, their abundance changes significantly such that HU and Fis are the most abundant NAPs in the growth phase, whereas IHF and Dps become the most abundant NAPs in the stationary phase [16]. HUαα is the predominant form in the early exponential phase, whereas the heterodimeric form HUαß predominates in the stationary phase, with minor amounts of homodimers [169]. This transition has functional consequences regarding nucleoid structure because HUαα and HUαß appear to organize and condense DNA differently; both form filaments, but only HUαα can bring multiple DNA segments together (DNA bunching) to form a DNA network [40]. The copy number of MukB increases two-fold in the stationary phase [136,147]. An increase in the number of MukB molecules could have an influence on the processivity of the MukBEF complex as a DNA loop extruding factor resulting in larger or a greater number of the loops.
Supercoiling can act in a concerted manner with DNA architectural proteins to reorganize the nucleoid. The overall supercoiling level decreases in the stationary phase, and supercoiling exhibits a different pattern at the regional level [170]. Changes in supercoiling can alter the topological organization of the nucleoid. Furthermore, because a chromosomal region of high transcription activity forms a CID boundary, changes in transcription activity during different growth phases could alter the formation of CID boundaries, and thus the spatial organization of the nucleoid. It is possible that changes in CID boundaries observed in the stationary phase could be due to the high expression of a different set of genes in the stationary phase compared to the growth phase [124].

Nucleoid structure and gene expression
NAPs and gene expression. The E. coli chromosome structure and gene expression appear to influence each other reciprocally. On the one hand, a correlation of a CID boundary with high transcription activity indicates that chromosome organization is driven by transcription. On the other hand, the 3D structure of DNA within nucleoid at every scale may be linked to gene expression. First, it has been shown that a reorganization of the 3D architecture of the nucleoid in E. coli can dynamically modulate cellular transcription pattern [171]. A mutant of HUα made the nucleoid very much condensed by increased positive superhelicity of the chromosomal DNA. Consequently, many genes were repressed, and many quiescent genes were expressed. Besides, there are many specific cases in which protein-mediated local architectural changes (Fig 2) alter gene transcription. For example, the formation of rigid nucleoprotein filaments by H-NS blocks RNAP access to the promoter thus prevent gene transcription [172].
Through gene silencing, H-NS acts as a global repressor preferentially inhibiting transcription of horizontally transferred genes [20,46]. In another example, the specific binding of HU at the gal operon facilitates the formation of a DNA loop that keeps the gal operon repressed in the absence of the inducer [173]. The topologically distinct DNA micro-loop created by coherent bending of DNA by Fis at stable RNA promoters activates transcription. DNA bending by IHF differentially controls transcription from the two tandem promoters of the ilvGMEDA operon in E. coli [174,175]. It is noteworthy that specific topological changes by NAPs not only regulate gene transcription but are also involved in other processes such as DNA replication initiation, recombination, and transposition. In contrast to specific gene regulation, how higher-order chromosome structure and its dynamics influences gene expression globally at the molecular level remains to be worked out.
DNA supercoiling and gene expression. A two-way interconnectedness exists between DNA supercoiling and gene transcription [176]. Negative supercoiling of the promoter region can stimulate transcription by facilitating the promoter melting and by increasing the DNA binding affinity of a protein regulator. Stochastic bursts of transcription appear to be a general characteristic of highly expressed genes, and supercoiling levels of the DNA template contributes to transcriptional bursting [177]. According to the twin supercoiling domain model, transcription-induced supercoiling can influence transcription of other nearby genes through a supercoiling relay. One such example is the activation of the leu-500 promoter [176]. Supercoiling not only mediates gene-specific changes, but it also mediates large-scale changes in gene expression [178]. Global gene expression analysis in response to a loss of supercoiling identified 306 supercoiling-sensitive genes (SSGs) in E. coli [178]. SSGs are present throughout the chromosome and encode proteins with diverse functions. The topological organization of the nucleoid could allow independent expression of SSGs in different topological domains. A genome-scale map of unrestrained supercoiling showed that genomic regions have different steady-state supercoiling densities, indicating that the level of supercoiling differs in individual topological domains [170]. As a result, a change in supercoiling can result in domain-specific expression levels of SSGs, depending on the degree of supercoiling in each domain.
The effect of supercoiling on gene expression can be mediated by NAPs that directly or indirectly influence supercoiling. The effect of HU on gene expression appears to involve a change in supercoiling and perhaps a higher-order DNA organization [179]. A positive correlation between DNA gyrase binding and upregulation of the genes caused by the absence of HU suggests that changes in supercoiling are responsible for differential expression [47]. HU was also found to be responsible for a positional effect on gene expression by insulating transcriptional units by constraining transcription-induced supercoiling [180]. Point mutations in HUα dramatically changed the gene expression profile of E. coli, altering its morphology, physiology, and metabolism [171]. As a result, the mutant strain was more invasive of mammalian cells [181]. This dramatic effect was concomitant with nucleoid compaction and increased positive supercoiling. In contrast to the wild-type dimer, the mutant protein is an octamer that wraps DNA on its surface in a right-handed manner, restraining positive supercoils as opposed to wild-type HU [182]. These studies show that amino acid substitutions in HU can have a dramatic effect on nucleoid structure, and consequently, on gene expression, which in turn results in significant phenotypic changes.
Although HU appears to control gene expression by modulating supercoiling density, the exact molecular mechanism remains unknown. Since MukB and HU have emerged as critical players in long-range DNA interactions, it will be worthwhile to compare the effect of each of these two proteins on global gene expression. The impact of MukB on gene expression is yet to be analyzed.

Current state and future perspective
Although studies of isolated nucleoids began in the 70s, a highly resolved and complete structure of the bacterial nucleoid is still not available. It remains a significant challenge even at 10 kb resolution, to accurately describe the hierarchical organization of chromosomal DNA resulting in the formation of a functional nucleoid. A single nucleotide resolution is even more difficult. However, we have significantly improved our understanding of the structural properties of the nucleoid. In summary, chromosomal DNA is a supercoiled molecule that folds into plectonemic loops. The supercoiling density is created and maintained by two topoisomerases (DNA gyrase and Topo I) with opposing functions. The NAPs organize DNA through DNA bending, bridging and looping activities. This organization together with transcriptional activities leads to higher-order organization of DNA into topologically and spatially distinct chromosomal interaction domains (microdomains) that are further organized as large size macrodomains. NAPs and macrodomain-specific proteins MukBEF and MatP appear to collaborate in the spatial organization. Supercoiling and topological domain formation also contribute to the proper regulation of gene transcription. The inherent bending of DNA due to Brownian motion and plectonemic supercoiling and DNA organization activities of NAPs and MukBEF provide condensation forces to sufficiently reduce the volume of the chromosomal DNA to fit within the cell volume. This process imparts a hierarchical organization, and the condensed chromosomal DNA results in a functional nucleoid with a helical ellipsoid shape.
Advances in imaging technologies can pave the way for direct visualization of higher-order nucleoid structure in vivo at high resolution. Recent electron microscopy methods that use high-pressure freezing and cryo-sectioning to preserve native ultrastructure [183], use improved DNA detection techniques for a sharp contrast [184], and allow 3D visualization with the cross-sectioning of cells using ion beams followed by 3D reconstruction [185] hold promises in driving new understanding of the 3D arrangement of the entire chromosomal DNA at high resolution. Although many issues regarding nucleoid fine structure and function remain to be resolved, some of the most interesting are the following: i. Is there a growth/environment dependent defined 3D structure of the DNA within the nucleoid?
ii. How are environmental cues transmitted to alter nucleoid structure?
iii. What is the nature of the supercoiling diffusion barriers that segregate the nucleoid into independent topological domains?