Transcription of Bacterial Chromatin

Bacterial chromatin consists of DNA, RNA, and proteins that condense to form the nucleoid

Bacterial genomic DNA, which is typically ~2–6 Mbp despite outliers ranging from 0.1 Mbp in an obligate symbiont to 14 Mbp in an omnivorous myxobacterium [38, 39], must be compacted dramatically to fit inside a cell let alone the smaller nucleoid region. Uncompacted, a 5 Mbp DNA would form a random coil with a volume nearly 10³ times larger than a typical cell volume, which is 0.4–3 femtoliters [8, 40]. Instead, the genome, associated proteins, and RNA compact into a central, membrane-free space called the nucleoid that takes up ~15–25% of the cell volume [8, 41]. The surrounding cell volume is occupied by cytoplasmic proteins and RNA, mostly in the form of ribosomes in an actively growing cell (Fig. 1; [6]). This ≥2,000-fold compaction (relative to an uncompacted, random coil) requires physical constraints in addition to the surrounding cytoplasmic and outer membranes to create the subcellular nucleoid. Both experiments and computational analyses suggest that chromatin organization is mediated by the combined effects of supercoiling [42], DNA-binding proteins [5], transcription [43], molecular crowding, electrostatics, and macromolecular demixing (heterogeneity) driven by weak differential chemical affinities that produce phase separation in the extreme form [44, 45]. Abundant cellular solutes, such as K⁺, Mg²⁺, and spermidine³⁺, aid nucleoid compaction [44], modulate protein–DNA–RNA interactions [46, 47], and alter RNAP activity directly [48]. Shifts in osmotic strength [49], solute concentrations [44], or temperature [50] can profoundly impact the nucleoid, but the underlying mechanisms of these effects are currently not well understood.

These different effects influence compaction via forces exerted in different dimensions (i.e., along the DNA axis or through space) and at different scales [10]. Supercoiling and protein binding events can act along the dimension of the DNA axis, whereas crowding, chemical interactions, electrostatic interactions, and phase separation operate in the three-dimensional (3D) environment of the entire cell. The complex and heterogeneous chromatin state generated by the balance of these different forces defines the substrate for transcription in living cells. Thus, understanding the types of chromatin structures that RNAP can encounter is crucial to understanding how transcription of chromatin occurs.

Many DNA-binding proteins organize DNA and are modulated by transcription

Compaction of DNA and constraint of DNA supercoiling in bacterial cells are achieved by diverse DNA-binding proteins that wrap, bend, or bridge the DNA [51]. Without these proteins, the chromosomal DNA itself would occupy a volume much larger than the cell, let alone the nucleoid, due to the intrinsic stiffness of DNA. Both DNA-binding proteins and solutes can aid DNA compaction and ensure cellular integrity [52, 53] by bending DNA or modulating DNA flexibility, respectively. In eukaryotes, DNA is compacted by the wrapping of the DNA around octa-histone nucleosomes, which constrain (−) supercoils, and further compacted by nucleosomal packing [2].

Bacteria possess abundant DNA-binding proteins that bind throughout the genome with variable specificity. Some of these proteins resemble histones in distribution and in some functions, but not in sequence or structure [54–59]. In E. coli, they include HU, IHF, Fis, H-NS, StpA, Dps, Lrp, CRP, MukBEF, and MatP, which differ greatly in abundance, effects, and extent of conservation [5, 10] (Table 1). Their DNA-binding modes include bending, wrapping, bridging, and some higher levels of compaction, illustrating the array of mechanisms of DNA organization in bacteria (Fig. 1; see “Organization of the nucleoid into domains” section below). Bridging brings distal dsDNA sequences together in 3D space, which creates a loop of non-bridged DNA. At least two modes of bridging dsDNA segments have been observed (Fig. 1): (i) by binding two DNA segments (e.g. H-NS [60]) or (ii) by encircling two DNA segments with a proteinaceous loop (e.g. structural maintenance of chromatin, SMC, proteins – MukBEF in Gram-negative bacteria [61]). Extensive wrapping, bridging (e.g., Dps [62] or StpA [63]), or constraint of plectonemes (e.g., hyperplectonemes stabilized by HU, H-NS, or Fis [64]) can also lead to higher-order DNA compaction. Due to the variability in abundance, structure, and DNA-binding mode, the precise roles of the most abundant bacterial chromatin proteins in organizing chromatin remain incompletely defined. Additionally, many of these proteins modulate gene expression, but the mechanisms of these effects remain incompletely characterized. For example, E. coli Lrp alters directly the expression of ~10% of genes and the expression of other genes indirectly possibly through different DNA-binding modes [65], with more studies needed to understand the mechanistic details. Conversely, transcription can influence binding of these proteins (e.g., HU and H-NS, discussed below). Although these proteins have been traditionally referred to as nucleoid-associated proteins, we refer to them here simply as DNA-binding proteins or chromatin proteins because no bright line distinguishes bacterial nucleoid-associated proteins, which structure the DNA and control gene expression globally, from site-specific DNA binding proteins like conventional TFs [5, 66]. Instead, a continuum of DNA-binding proteins exists in bacteria ranging from low-copy proteins that may bind only one site per genome (e.g., the E. coli MelR protein [67]) to highly abundant proteins that are present throughout the genome but nonetheless play key roles in transcription (e.g., HU [59]). Thus, chromatin or nucleoid-organizing protein might be a more meaningful descriptor than “nucleoid-associated.”

HU is the most abundant and highly conserved chromatin protein in growing bacteria, and is a key player both in structuring the nucleoid and in gene transcription (reviewed in [68]). HU exists as both heterodimers and homodimers of HUα and HUβ subunits, each an ~10 kDa DNA binding protein with little if any sequence-specificity. HU binds non-specifically to linear dsDNA with low affinity (μM range) as a dimer or a multimer [69, 70] but binds non-B-form DNA structures, such as DNA forks, sharp bends, kinks, or bulges, with higher affinity (nM range) [71, 72]. HU can also bind to RNA [73, 74]. HU plays multiple roles in condensing the nucleoid, including bending DNA [75, 76], wrapping DNA [69, 77], constraining (−) supercoils [78–80], putatively bridging DNA [69], and facilitating formation of RNA–DNA complexes [74, 81]. HU also stimulates topoisomerase I activity to remove excess (−) supercoiling [82]. This wide array of binding modes and dynamic interactions with DNA has left the precise roles of HU and underlying structures poorly defined. HU can constrain (−) supercoiling in plectonemes and recruit topoisomerase I upstream from ECs to reduce torsional stress (Fig. 1). HU exhibits a high intrinsic off-rate and can be displaced by 2 pN of force [83], suggesting HU would not impede transcribing RNAP. Further, similar to effects on eukaryotic nucleosomes [84], (+) supercoiling could help displace HU in front of RNAP if HU prefers to bind (−) supercoiled DNA [85]. Together, these properties of HU suggest that transcription is a key determinant of the genomic distribution of HU [80] and that HU helps modulate supercoiling throughout the genome (see “DNA topology mediates transcription-bacterial chromatin interplay”). Although incompletely substantiated, a mutant HU appears to bind (+) supercoiled DNA [79]. Further, the ability to constrain supercoils and the relative amounts in growing vs. non-growing cells varies among HU homo- and hetero-dimers [86, 87]. Possibly, some form of HU could constrain (+) supercoils in front of ECs in some conditions. Elucidation of the mechanistic details of HU-transcription interactions should be high priorities for studies of bacterial chromatin.

H-NS is an ~15 kDa basic protein containing a winged-helix DNA binding domain and two oligomerization interfaces that both helps organize the nucleoid and inhibits transcription. H-NS dimerizes in solution and is found in gram-negative bacteria, primarily γ-proteobacteria [88], with functional analogs in other bacteria (e.g., pseudomonal MvaT [89] and mycobacterial Lsr2 [90]). Multiple H-NS paralogs are often present in a single species, such as StpA in E. coli; H-NS and StpA form heterodimers [91]. Additionally, some enterobacteria contain Hha family proteins that associate with H-NS but do not bind DNA [92, 93]. Here, we refer to H-NS-like proteins and Hha family proteins as H-NS modulators because they likely modulate the structure and function of H-NS. H-NS binds to AT-rich DNA and forms nucleoprotein filaments that silence gene expression [94]. Two types of filaments have been observed in vitro: a linear (or “stiffened”) filament in which H-NS binds to one segment of DNA and a bridged filament in which H-NS binds two segments of DNA (Fig. 4a) [95, 96]. Silencing occurs when linear or bridged filaments block transcription initiation and when bridged, but not linear, filaments block elongation topologically by promoting backtrack pausing and ρ-dependent termination [97–99] (see “Occlusion” and “DNA topology” below). Both Hha and StpA enhance formation of bridged H-NS filaments in vitro [96, 100], but it remains unclear which conformation predominates in vivo.

Although H-NS dramatically affects transcription, an elongating RNAP can also remodel an H-NS or H-NS:StpA filament [101, 102]. Elongating RNAP may encounter an H-NS filament in a coding region [54] or downstream of an antisense promoter [103], especially when ρ-dependent termination is suppressed [102]. Upon encountering certain filaments, ECs may disrupt or rearrange the filament; if ECs disrupt a filament covering a promoter, then H-NS silencing can be relieved depending on the number of ECs transcribing into the filament [101]. It remains unclear which interacting filaments (e.g., on different sequences, containing Hha or StpA [92, 100], or linear versus bridged) and ECs (e.g., with coding versus noncoding nascent RNAs) yield disrupted filaments versus halted ECs. Experiments designed to probe in vivo H-NS filament characteristics and interactions with RNAP will shed light on how ECs might affect H-NS silencing of nearby promoters.

Organization of the nucleoid into domains depends on an interplay between transcription and DNA-binding proteins

The overall structure of the nucleoid and its separation in distinct spatial, topological, and interaction domains is largely dictated by transcription itself, with modulation by DNA-binding proteins and their interplay with transcription. The organizing role of transcription is revealed when transcription is blocked by the RNAP inhibitor rifampicin. Upon addition of rifampicin to E. coli cells, the nucleoid first compacts due to loss of the contribution to expansion from coupled transcription–translation, but then eventually expands due to entropically driven intermixing of inactive ribosome subunits with the chromosome [6, 104]. However, deletion of abundant DNA-binding proteins like HU also decompacts the nucleoid [5, 68], showing that chromatin proteins also aid nucleoid organization.

The bacterial nucleoid is organized into sets of nested domain structures [10, 11] whose physical nature and functional definitions remain incompletely elucidated (Figs. 1&2). Large “macrodomains” [105, 106] are proposed to organize the chromosome into spatially distinct sections of the nucleoid based primarily on constraints imposed by the origin and terminus of DNA replication [11]. Topologically isolated “supercoil” domains were originally defined as segments of chromosomal DNA relaxed by the introduction of single- or double-strand nicks [32, 42, 107, 108]. With the advent of high-throughput in vivo DNA interaction assays (Hi-C; see below), segments of DNA exhibiting greater proximity as captured by formaldehyde-induced protein-DNA crosslinks were defined as chromosome interaction domains (CIDs) in bacteria [43, 106, 109–111] and topologically associated domains (TADs) in eukaryotes (reviewed in [112, 113]). The eukaryotic TAD refers to proximity of chromosome segments in three-dimensional space [114] rather than the mathematical definition of topology (properties of objects that are preserved when deformed as in stretched, bent, or twisted, but not when broken or rejoined) that underlies the classic definition of DNA topology by its linking number. This different use of topology in TAD confuses a rigorous description of chromatin domains. Thus, we will refer to domains demarcated by barriers to supercoil diffusion and within which supercoiling is connected as supercoil domains (SDs), although topologically isolated domain (TID) might otherwise be a better acronym. As we will explain below, CIDs and SDs are interrelated but distinct, can be nested, and can be further subdivided into several different subtypes depending on what forms the boundaries of the domains, transcription complexes being among the most important of these boundaries.

Large macrodomains (~1 Mbp each) have been reported in bacteria based on multiple methods (Fig. 1). Based on recombination frequencies between two λ att sites and microscopy in E. coli [115], Boccard and co-workers found four segments of the genome with higher levels of self-interaction than between-macrodomain interaction: ori, ter, left, and right with two nonstructured regions flanking ori [105, 116]. Macrodomains have also been reported in B. subtilis [106], but not in Caulobacter [11]. Macrodomains are proposed to play functional roles in replication and other cellular processes [13]. Their properties may depend in part on active transcription, at least during rapid growth [7, 117]. Recently, the Boccard group reported Hi-C data consistent with a robustly distinct ter region, long-range interactions within ori, left, and right at multiple scales, and strong interaction barriers defined by highly transcribed operons [110]. The distinct ori and ter segments of the chromosome are well-defined; they bind distinct chromatin proteins (DnaA [118] and MatP [106, 110], respectively, and membrane–ori-attachment proteins in some bacteria [11]) and have been found in all bacteria studied in detail. MukBEF modulates the long-range interactions in ori, left, and right and is excluded by MatP from ter. The relationships between the macrodomains, the CIDs and SDs nested within them, and transcription remain to be characterized definitively.

However, the organization of chromatin into CIDs and SDs unquestionably depends on transcription (Fig. 2b; see also “DNA topology”) [107, 108, 119]. CIDs are defined by chromosome segments with higher levels of internal interactions using Hi-C assays [43, 106, 110, 115, 119]. Hi-C assays report proximity between regions of the genome in 3D space using deep-sequencing of formaldehyde crosslinked and then ligated DNA fragments [115]. CIDs (~30–400 kb in E. coli) are smaller than macrodomains but larger than SDs (~2–70 kb; ~10–20 kb on average). About 30 CIDs have been identified in each species (E. coli [110], Caulobacter [43], and B. subtilis [106]) and some are nested; however, CIDs in E. coli appear less distinct and have proven difficult to capture [120]. In bacteria tested to date, inhibition of transcription by rifampicin causes a dramatic loss of the CID boundaries, which are typically located near highly transcribed genes (i.e., regions of high EC density); thus, active transcription appears to govern CID organization [43, 106, 110, 111]. Transcription creates domain boundaries by limiting supercoil diffusion for the same reasons it creates topological stress (see ‘twin supercoiled domain model’ above): the EC is unable to rotate when proteins bind its nascent RNA. A higher density of ECs may create a stronger topological barrier [32, 37]. Further, characterization of CID boundaries suggests that extensive transcription of longer operons (or possibly long bridged H-NS filaments) may favor physical separation of the domains on either side [119], which would appear as a strong CID boundary in Hi-C experiments.

Growing E. coli and Salmonella contain ~400 SDs, which were identified by assays that require two sites to be topologically connected (e.g., γδ or Tn3 resolvase action [32, 107, 108]) or by assays of transcription of supercoiling-sensitive genes near DNA nicks [42] (Fig. 1). Because genomic DNA is negatively supercoiled on average [36], plectonemes can form and move throughout the genome by slithering (dynamic plectonemes; Fig. 1). These plectonemes can be stabilized or trapped either by ECs or by DNA-binding proteins that limit diffusion of supercoils, thereby creating SDs (Figs. 1&2). Because ECs can’t rotate, they will both generate supercoils during transcription (twin-supercoiled domain) and prevent diffusion of supercoils [37]. Interestingly, the distribution of H-NS binding sites is consistent with a bridged H-NS filament creating ~11 kb loops in vivo [121], which matches the size of SDs.

Although CIDs and SDs have been separately defined experimentally, they are related because they can both be bounded by ECs. Given this relationship, why aren’t all SDs observed in Hi-C experiments? First, SDs may be more dynamic than CIDs, so they may be obscured in genome-scale experiments by averaging over a population of cells. Second, the physical separation between SDs may be small, allowing interactions between adjacent SDs. In other words, a low density of ECs may create a topological barrier and define an SD boundary, but still allow the domains on either side to interact and appear as a single CID in a Hi-C experiment [119] (Fig. 2b). This difference may explain in part why CIDs are larger than SDs on average. Modeling studies also suggest that SDs (i.e., plectonemes) can be nested within CIDs [43, 122] (Fig. 1). Thus, even though ECs can create boundaries for either CIDs or SDs, the interactions within and between these domains can differ. Differences may also arise from the extent of protein-constrained supercoiling within domains (e.g., by HU or Fis) or from contributions to boundaries or intersegment interaction by DNA-bridging proteins like H-NS or SMC [123].

ECs and DNA-binding proteins both form topological barriers that can demarcate CIDs and SDs, but these barriers differ in an important way (Fig. 2). ECs create active boundaries that generate additional supercoiling in both directions, whereas DNA-binding proteins that prevent diffusion of supercoiling (e.g., bridged H-NS filaments) create static barriers. In this context, we define an active barrier as one that generates supercoils (e.g., a barrier composed of ECs) and a static barrier as one that blocks supercoil diffusion without itself generating supercoils (e.g., bridged H-NS filaments; Fig. 2b); all domain barriers must rearrange at least once per cell cycle, but the lifetimes of barriers in general are not well-characterized. Further, abundant proteins like H-NS, HU, and Fis can either constrain supercoils within CIDs or SDs to relieve torsional stress and stabilize domains [43, 106, 110] or form bridged complexes between domains (e.g., hyperplectonemes [64]) and thereby form a larger CID with nested SDs. Finally, topoisomerases also affect both CIDs and SDs by modulating supercoiling. Together, active (EC) and static (DNA-binding protein) boundaries throughout the genome have the potential to form domain boundaries in different combinations that will have different predicted effects on the domain properties (Fig. 2c). For example, greater (−) supercoiling created by active EC boundaries and constrained as plectonemes may increase DNA-DNA interactions detected by Hi-C and form a CID, whereas ECs generating (+) supercoils or static boundaries may lessen DNA–DNA interactions within a domain. Targeted experiments coupled with computer modeling [43, 109, 122] to probe supercoiling, protein binding, and DNA-DNA interactions at domains with defined active (e.g., a highly transcribed gene) and static boundaries (e.g., bridged H-NS filaments) are needed to test the nested CID–SD model and to improve understanding of the interplay between supercoiling, transcription, chromatin proteins, and bacterial nucleoid substructure.

Macromolecular demixing may contribute to nucleoid and chromatin substructuring in bacteria

Macromolecular demixing arises because solutions can sometimes increase entropy by partitioning into heterogeneous sub-volumes to increase favorable and decrease unfavorable molecular interactions [124]. The molecular environment of the bacterial cell is crowded (200–300 mg protein/mL, 100 mg RNA/mL, and 11–18 mg DNA/mL, plus high mM concentrations of many solutes including nucleic acid compactors like spermidine; [125, 126]). Water molecules interact with each other and with macromolecules differently in the crowded nucleoid than in dilute solution [46]. In the cellular environment, molecular interactions both create the nucleoid and lead to substructuring within it and in the surrounding cytoplasm. These membraneless substructures are variously called macromolecular condensates, droplets, granules, speckles, bodies, densities, or clusters that in the extreme constitute a physical phase separation. These substructures are dynamic and aid rapid regulation of intracellular reactions [127]. The formation, properties, and impacts of substructures on cellular processes, including transcription, are a rapidly growing focus in eukaryotic cell and molecular biology research [127–131].

In bacteria, the nucleoid itself is a consequence of these demixing interactions. Computational analyses suggest that the electrostatic repulsion between DNA and ribosomes is a key factor driving nucleoid formation with a surrounding rich in ribosomes and mRNAs [6, 44, 45, 132]. The nucleoid remains liquid; thus, structural rearrangements and diffusion occur throughout it [8, 41, 44, 45, 133]. For example, an increase in transcription creates more mRNA, which favors demixing and further compaction of the nucleoid; in turn, compaction can decrease transcription. This feedback loop reflects the dynamic nature of the nucleoid and the physical properties driving demixing [132], and illustrates another way that transcription organizes the nucleoid.

In bacteria, macromolecular condensates are also thought to form within the nucleoid in subregions of highly transcribed DNA (Fig. 1; e.g., rrn ribosomal RNA operons; [7, 117, 134–137]). These transcription-driven condensates may arise because the high levels of nascent RNAs and associated proteins (e.g., ribosomes on mRNAs or ribosomal proteins assembling on rRNAs) generate a demixed heterogeneity in the nucleoid [117]. Evidence exists for a nucleolus-like structure formed when six of the seven ribosomal RNA operons in E. coli cluster together [134], and clusters of RNAP at the rrn operons have been observed during fast growth conditions [7, 117, 135, 136]. Clustering of non-rrn genes has also been reported [109, 138–140]. These transcription-driven clusters may provide substructure that play a role in organizing the nucleoid in conjunction with CIDs and SDs, and may facilitate gene regulation by localizing extensive transcription to the periphery of the nucleoid (Fig. 1; [117]). Formation of these transcription clusters could also be facilitated by the supercoiling effects discussed above.

Evidence for other condensates in bacteria, including the FtsZ, SlmA, and DNA complex formed during cell division [141] and the α-proteobacterial RNA degradasome [142], suggests that macromolecular demixing may play roles in many cellular processes including gene expression and regulation. The understanding of nucleoid substructuring remains in its infancy relative to studies of macromolecular demixing in eukaryotes owing principally to the much smaller size of bacteria. New computational, microscopic, and HT-sequencing methods are needed to overcome the size barrier and provide a better understanding of the interplay between transcription, nucleoid substructuring, and macromolecular demixing (i.e., phase separation) in bacteria.

Fis and H-NS connect gene expression to environment signals

Fis silences genes in E. coli and related bacteria by occluding either TF or RNAP binding to the promoter region (reviewed in [14]). Many AT-rich, Fis-binding motifs are found throughout the genome in intragenic regions where promoter elements are frequent [57]. Fis is highly expressed during exponential phase and decreases to undetectable levels in stationary phase (Table 1); Fis regulation coordinates gene expression, especially of virulence genes, with growth phase. Recent work on Dickeya dadantii (a γ-proteobacterial plant pathogen closely related to E. coli) highlights this role of Fis in virulence regulation [145]. Virulence is activated in D. dadantii by expression of the major virulence regulator, pelD, when cell density is high (i.e., stationary phase for D. dadantii in a plant). pelD is silenced during a rapid growth phase by Fis, which both blocks RNAP binding to the promoter, and blocks its activator CRP (the catabolite response or cAMP receptor protein) from binding an adjacent site. Thus, Fis silencing keeps pelD off when cell density is low but allows activation of the virulence program when cells reach high density in coordination with a decline in Fis levels. H-NS also binds to the pelD promoter, creating a second mechanism to occlude RNAP during non-virulent conditions [146]. H-NS silencing is relieved by changes in DNA supercoiling at the promoter that destabilize the H-NS filament [147]. pelD provides one example of many gene regulatory circuits that respond to growth conditions using Fis [148] or H-NS [149] – a paradigm by which growth-dependent changes in chromatin coordinate changes in gene expression.

H-NS occlusion of RNAP occurs by formation of inhibitory nucleoprotein filaments across promoter regions. These H-NS filaments typically form on horizontally-acquired genes, pathogenic operons, and antisense transcripts, all of which are silenced by H-NS (Fig. 4b) [54, 94, 97, 150]. Environmental factors, most notably increased temperature [151] or changes in osmolarity [152], may contribute to relief of H-NS occlusion at these genes. For DNA from enteropathogenic operons, the H-NS filament forms in vitro at lower temperatures (~20 °C), consistent with stronger effects on RNAP binding and gene silencing at lower temperatures (Fig. 4b) [98, 151, 153]. At higher temperatures (~37 °C, e.g., the temperature encountered inside mammals), occlusion by H-NS is reduced in vitro [153] by either temperature-dependent changes in the DNA structure (bending or changing supercoiling [153, 154]) or in H-NS conformation (unfolding of H-NS dimerization domains [155]) that could prevent formation of an inhibitory filament [151, 153]. The structural details of this derepression remain unclear in part because conflicting evidence exists as to whether bridged or linear H-NS filaments are responsible for gene silencing in vivo. H-NS mutants deficient in linear filament formation in vitro do not silence genes in vivo [156], but silencing of some genes requires the H-NS modulators, Hha and StpA [55, 157, 158], known to stimulate bridging by H-NS in vitro [96, 100]. Despite the clear role of environmental factors and H-NS modulators in H-NS-mediated occlusion, the changes to H-NS and the DNA that mediate occlusion in vivo remain unclear and in need of structural elucidation.

Although most H-NS occlusion inhibits transcription initiation, in some cases it may aid gene expression by facilitating RNAP access to a specific promoter. For example, in the E. coli ehxCABD operon, which encodes genes allowing secretion of the virulence factor hemolysin, H-NS filaments occlude promoters and promoter-like sequences that are adjacent to the primary promoter; occlusion of these other promoters appears to direct RNAP to the primary promoter and thus ensure that the correct mRNA is produced [159]. H-NS may also increase transcription of σ^S-dependent promoters in stationary phase by occluding promoters for the housekeeping sigma factor, σ⁷⁰ [160]. This function of H-NS adds yet another layer to the complex repertoire of gene regulation by H-NS.

Counter-silencing of H-NS filaments

In addition to temperature-induced derepression, promoter occlusion by H-NS filaments can be relieved by binding of H-NS antagonists in a process known as “counter-silencing” (Fig. 4b). H-NS antagonists typically do not activate transcription when H-NS is absent, suggesting that they do not directly contact RNAP but simply prevent the formation of repressive H-NS filaments (although some antagonists can also recruit RNAP) [161]. Conventional TFs can act as H-NS antagonists even when binding outside the typical distance of TF action (−60 to +20 relative to a transcription start site; e.g., PhoP in Salmonella, LeuO in enterohemorrhagic E. coli, VirB in Shigella [162], and IHF in E. coli [161, 163]). The ubiquity of antagonists among bacteria that utilize H-NS suggests that counter-silencing is a widely used mechanism of transcription regulation [164, 165]. Counter-silencing may involve one of three non-exclusive events: (i) displacement of the H-NS DNA-binding domain; (ii) remodeling of the H-NS filament into a non-inhibitory conformation; or (iii) binding of an antagonist to the H-NS oligomerization domain. In Salmonella, remodeling of the H-NS filament bound at the curli operon, which encodes curli fimbriae required for virulence and cell survival, was recently reported [166]. In vitro DNaseI footprinting suggests that the antagonist, CsgD, bends the DNA and remodels the H-NS filament without preventing H-NS binding to DNA [166]. The remodeled filament allows RNAP binding and gene expression, but the conformation of the remodeled filament is unknown (Fig. 4b).

In enterohemorrhagic E. coli (EHEC), expression of the LEE operon, which encodes virulence factors required to create attaching and effacing lesions, is induced when the H-NS antagonists, Pch and Ler, remodel or prevent binding of an H-NS:StpA:Hha filament, respectively [55, 167]. In Vibrio, the H-NS antagonist ToxT, along with the DNA-binding protein IHF, relieve H-NS repression of ctxAB, which encodes a potent Vibrio toxin, by displacing H-NS binding at a high-affinity site [149, 162, 168].

Some H-NS antagonists appear to interact with the oligomerization domain of H-NS to derepress genes. Both a phage protein, gp5.5 [169], and an EHEC transcription factor, Aar [170], can bind the oligomerization domain of H-NS to disrupt multimerization, but not DNA binding. In enteropathogenic E. coli, the truncated H-NS protein (H-NST) can form heterodimers with full-length H-NS, which disrupts bridged filament formation [96] and results in derepression of genes [171]. Evidence also exists that H-NST antagonizes H-NS DNA binding and induces expression of the LEE operon [172], suggesting some H-NS antagonists act by more than one mechanism. It is unclear if antagonism by H-NS binding proteins is gene-specific or if it affects H-NS silencing globally. One possibility is that the effect of H-NS binding antagonists may depend on the stability of H-NS filaments, which is determined by DNA sequence.

Overall, counter-silencing likely depends on a combination of interactions between H-NS and its modulators, antagonists, the DNA path, RNAP, and physio-chemical conditions. Some evidence suggests that inhibitory filaments adopt a linear conformation [173] and that remodeling of this filament may reduce filament length or perturb multimerization (Fig. 4b), both of which could make the filament too weak to compete with RNAP for DNA binding. Further studies of filament structure in the context of counter-silencing should contribute to our broader understanding of H-NS filament structure–function.

Interplay between kinetics of RNAP and protein binding determines roadblock efficiency

As originally shown by Steege and co-workers [184, 185], a protein must bind DNA tightly to pose an effective roadblock to RNAP, although the kinetics of transcript elongation, specifically the sequence-dependent propensity for pausing and backtracking, also dictates roadblocking efficiency. The primary determinant of roadblocking efficiency is the dwell time of a specifically bound DNA-binding protein (assuming a protein level sufficient to occupy the site). To affect RNAP, a roadblock must remain on DNA considerably longer than the step time of RNAP (20–50 ms at non-pause sites in vivo; slower at pause sites) because RNAP can advance after this step time if the protein releases. This dwell time corresponds to relatively tight binding; assuming near-diffusion-limited association, even proteins with K_ds in the micromolar to nanomolar range, where many DNA-binding proteins lie [186, 187], have little effect on transcription because their dwell times are in the 50 ms to few sec range. Demonstrated roadblocks, which include non-cleaving EcoRI [185], noncleaving Cas9 [174], LacI [184, 188], and biotin-streptavidin [189] all bind with dwell times in the several minute to multi-hour range [190–193]. Thus, even though the bacterial chromosome may be littered with DNA-binding proteins, in many cases transcription may easily prevail. This includes bacterial chromatin proteins; HU binds with micromolar affinity [71] and even H-NS does not slow RNAP in vitro unless it forms topologically entangled, bridged filaments [98].

As a case in point, the most-studied roadblocking barrier [184, 194–197], LacI, binds its operator in physiological solutes within a long DNA with a K_d of ~10⁻¹² M and dwell time of ~10 min [191]. Translocation through the LacI roadblock can occur with an increase in cooperating RNAPs transcribing the DNA [182], weakening the LacI K_d by mutation, or reducing the LacI concentration [175], whereas formation of a DNA loop by tetrameric LacI increases roadblocking [176]. Overcoming a tightly bound LacI roadblock is difficult in poorly transcribed regions but may become possible in regions of high transcription (Fig. 5) [175].

A more recent model for roadblocking, CRISPR interference (CRISPRi), supports these foundational mechanisms. CRISPRi silences genes by targeting the catalytically dead Cas9 (dCas9) to a gene downstream of the promoter with a complementary single guide RNA (sgRNA) [174]. The efficiency of the CRISPRi roadblock is influenced by dCas9 binding affinity (K_ds in the low nM range; [198]), the length of the RNA–DNA hybrid (≥10 bp of complementarity is required for gene repression; [199]), and the level of transcription in subsaturating dCas9 conditions. Additionally, dCas9 binding to the non-template strand nearer to the promoter appears to increase roadblock efficiency [174], although conflicting evidence exists about the importance of these parameters [200].

We note that extrapolation of in vitro DNA-binding protein behaviors (e.g., roadblocking) to behaviors in the nucleoid is fraught with unknowns. Diffusion rates, and thus the off-rates and dwell times, of DNA binding proteins in vivo will be strongly impacted by molecular crowding. GFP, for instance, diffuses in the E. coli cytoplasm about ten times more slowly than in dilute solution [201], but diffusion rates and dwell times in the nucleoid are reported for very few DNA-binding proteins and are difficult to predict. Besides crowding, at least two factors complicate such predictions. First, the effective on- and off-rates are likely to be dominated by electrostatics [202], so even slower diffusion of a model protein may not reflect the behaviors of DNA-binding proteins. Second, both classical and single-molecule experiments for the limited cases studied (e.g., HU, LacI [203, 204], RNAP [117, 205]) suggest DNA-binding proteins move by a combination of 1D sliding and 3D hopping among non-specific DNA locations rather than by free diffusion while searching for a specific binding site. Current and nascent strategies to study protein movements in vivo using fluorophores should be able to assess specifically bound DNA-binding protein dwell times, including with and without active transcription (e.g., downstream of a tightly regulated promoter). Such studies are needed now to gain better insight into transcriptional roadblocking in vivo.

RNAP backtracking modulates roadblock strength

In addition to protein binding kinetics and the number of transcribing RNAPs, the propensity for RNAP to pause and backtrack at a roadblock influences the strength of the roadblock. Pause sequences can trigger RNAP to reverse translocate RNA and DNA (backtrack) [206]; conversely, backtracking by roadblocked RNAP frequently occurs [182]. As first shown using nucleosomes [207, 208], the precise locations of halting and even whether RNAP halts appreciably, depend on the locations of pause sequences relative to a protein roadblock. Additionally, the upstream RNA structures [197], R-loop formation [177, 209, 210], and Gre factors [196, 206] all can influence the partition between active and backtracked RNAP [207, 208]. At the GalR roadblock, for example, base-pairing in the RNA–DNA hybrid within the roadblocked EC is thermodynamically unstable relative to a hybrid formed in a backtracked complex; this difference favors EC backtracking at the roadblock [177]. Introducing more stable DNA base-pairing upstream from the EC prevents backtracking and thus increases transcription through the GalR roadblock. At other backtracked pauses induced by roadblocks, like LacI, the addition of Gre factors to shift RNAP toward active elongation also can increase transcription through the roadblock [196]. The synergy between pause site locations, backtracking propensity, and DNA-binding site roadblocks in vivo remains unmapped. Further, it is difficult to separate effects of steric occlusion of transcription by a protein roadblock from topological stress that may itself induce delay, halting, or backtracking of RNAP and be exacerbated by inability of positive supercoils generated in front of ECs to diffuse through the roadblock. We address this topological effect of chromatin on transcription in the next section.

Genome-wide, transcription–translation coupling between the pioneering ribosome and RNAP also decreases backtracking. Specific interactions between RNAP and the ribosome are thought to facilitate coupling [211], which may aid RNA–DNA translocation, inhibit backtracking, and facilitate RNAP movement past roadblocks (Fig. 5) [183]. Thus, coupling aids in expression of genes bound by putative roadblocking DNA-binding proteins. For example, lipopolysaccharide biosynthetic operons in enteric bacteria, normally repressed by H-NS, use RfaH-mediated transcription–translation coupling as a mechanism to transcribe efficiently through potential H-NS roadblocks [212].

The DNA repair protein called Mutation frequency decline (Mfd) can either positively or negatively affect roadblock strength by interacting with backtracked ECs (Fig. 5), in addition to its function in DNA repair and genome maintenance [210, 213–215]. Mfd can increase roadblock strength by targeting stalled ECs and stimulating dissociation of the EC [17], as observed at the LacI [216], B. subtilis CodY [178], and B. subtilis CcpA [180] roadblocks. Mfd thus aids gene silencing, but the removal of the stalled RNAP is also necessary to ensure genomic integrity during replication [210]. In contrast, recent single-molecule results show that Mfd can associate with and aid backtracked ECs [217, 218], suggesting that Mfd may aid in transcription through some roadblocks and thereby decrease roadblock strength in these cases. In either case, Mfd appears tuned to ensure a roadblocked RNAP does not block other DNA-dependent processes either by assisting RNAP or removing it from DNA.

This same set of dynamics and assisting factors likely operate when RNAP encounters chromatin protein roadblocks on DNA. Despite knowledge of these basic features of roadblocking, many important questions await further study: (i) which bacterial chromatin proteins are capable of roadblocking RNAP; (ii) what are the precise binding characteristics that determine whether DNA-binding proteins create transcriptional roadblocks in vivo; (iii) which roadblocks are passively displaced due to relatively fast off-rates and which might be affected by positive supercoiling in front of RNAP (see next section, “DNA topology mediates transcription–bacterial chromatin interplay”); and (iv) if RNAP physically engages chromatin roadblocks, what is the structure of the RNAP-roadblock complex (i.e., the analog to RNAPII-nucleosomes structures recently determined by cryo-electron microscopy (cryo-EM) [219, 220])? These promising areas for future research will benefit from synergistic applications of new genome-scale methods, improved cell imaging, advanced single-molecule strategies, and new structural approaches (e.g., cryo-EM) now becoming available.

Topological effects on initiation of transcription

Changes in supercoiling at a promoter can affect initiation by either altering alignment of the − 10 and −35 elements or making DNA melting easier or harder ((−) or (+) supercoiling, respectively). These effects have been documented in E. coli, where expression of ~300 genes changes in response to supercoiling introduced by transcription, gyrase, or topoisomerase I [42, 108] and are regulated by proteins, such as Fis or HU, that modulate or respond to supercoiling. Levels of supercoiling throughout the genome are set by a negative feedback loop between gyrase, which increases (−) supercoiling, and topoisomerase I, which relaxes (−) supercoiling because their respective promoters are conversely activated or inhibited by (−) supercoiling due to effects of twist on −35 and −10 element alignment [66, 221, 222].

Fis can directly or indirectly regulate gene expression via supercoiling effects. Fis directly represses the gyrase promoter by occlusion [223], which decreases (−) supercoiling throughout the genome. Fis thereby indirectly affects expression of many supercoiling-sensitive genes by altering supercoiling, which is one mechanism to coordinate gene regulation with cell growth (Fis levels correlate directly with growth rate) [66]. At other promoters sensitive to supercoiling, Fis may directly alter supercoiling at promoters, but the mechanism has not been established [224].

The role of DNA-binding proteins in coordinating DNA topology and gene regulation is also exemplified by dramatic changes induced by a HU mutant that constrains (+) supercoiling. Wild-type HU constrains (−) supercoils generated by transcription throughout the genome [80]; however, when HU is altered by mutation, it can preferentially constrain (+) supercoils [225, 226]. When expressed in E. coli K-12, such an HU mutant induces compaction of the nucleoid, a decrease in supercoiling throughout the genome (i.e., the genome becomes less negatively supercoiled), and extensive changes in the transcriptome, all of which result in radical changes in cell morphology, physiology, and metabolism [225, 226]. Assays using differentially supercoiled plasmids in vitro showed that a decrease in supercoiling was responsible for activating promoters, such as the hlyE promoter, which are normally repressed by (−) supercoiling (hlyE encodes hemolysin E) [226]. Therefore, the overall decrease in supercoiling caused by the alteration of HU to preferentially constrain (+) supercoils influences which promoters are most active, presumably by changes in both the thermodynamics of melting and the relative alignments of −35 and −10 promoter elements. The change in supercoiling also may displace other chromatin proteins, like H-NS, bound at the promoter [54, 227]. These results support a model in which the constraint of supercoiling by HU, and possibly other chromatin proteins, is a fundamental mechanism of gene regulation.

H-NS creates closed topological domains that decrease transcription elongation

In addition to occluding promoters, H-NS inhibits elongation by stimulating backtrack pausing and ρ-dependent termination [98]. The mechanism by which H-NS affects elongation appears to depend on topological inhibition rather than on direct roadblocking of RNAP, providing an excellent illustration of how chromatin proteins can affect transcription through topological constraint. Although H-NS can form either linear or bridged filaments (Fig. 4a), somewhat surprisingly, H-NS switches from inhibition of RNAP elongation when present at ~66 H-NS dimers/kb DNA to having little effect when present at ≥200 H-NS dimers/kb DNA (i.e., a high H-NS concentration is less inhibitory than a lower H-NS concentration; [98, 100]). This effect can be explained because lower H-NS-to-DNA ratios favor bridging and bridged but not the linear filaments, which form at higher H-NS-to-DNA ratios, inhibit RNAP elongation. Further, H-NS modifiers that stimulate bridging greatly increase inhibition of RNAP elongation even at high H-NS to DNA ratios [100]. The pattern with which bridged filaments slow RNAP also is remarkable; they enhance pausing by RNAP at a subset of pause sites, which turn out to be sites at which RNAP backtracks; in contrast, bridged H-NS has little effect on nonbacktracked pauses [98, 100]. These observations can be explained if bridged filaments trap RNAP in a topologically closed domain (Fig. 4c). If H-NS acted via direct roadblocking of RNAP, then the linear filament would be expected to slow elongation and, as described above, H-NS dwell times would be expected to be longer than measured values (k_off = 1.5 s⁻¹ [228]). However, linear filaments do not pose the same barrier to DNA rotation as bridged filaments, which are expected to prevent DNA rotation upstream and downstream of RNAP. As a result, the (+) supercoils generated in front of RNAP and (−) supercoils generated behind RNAP may not easily be relieved by writhe within a bridged filament; both supercoiling effects energetically favor backtracking by RNAP. H-NS filament-stimulated pausing was relieved by addition of GreB [98, 100], which suppresses backtrack pausing [229] and is known to increase transcription opposed by topologically generated torque [34]. This pause stimulation also appears to explain the ability of H-NS to stimulate ρ-dependent termination (because ρ dissociates ECs at pause sites). The topological mechanism of inhibition of transcription remains to be verified in vivo, but it can explain the high correlation between sites of ρ-dependent termination and H-NS binding in E. coli K-12.

Transmission of signals via DNA topology within domains

Transcription-generated supercoiling also can impact how genes transcribed near each other within a chromosomal domain (e.g., a SD) affect each other’s expression [230] (Fig. 2a). In these domains, transcription of one gene can change the supercoiling level at nearby promoters because the diffusion of supercoils is restricted to within the domain by the SD boundaries (Fig. 2). The magnitude of effects will depend on the level of transcription, its distance from the affected promoter, the supercoiling-sensitivity of the promoter, and its orientation relative the promoter (which determines whether (+) or (−) supercoiling propagates toward the promoter; Fig. 2a). Specifically, transcription in an SD will stimulate expression of upstream promoters via diffusion of (−) supercoils and will decrease expression of downstream promoters via diffusion of (+) supercoils. Analysis of gene expression data in E. coli and Streptococcus pneumoniae by Sobetzko [230] showed a correlation in expression of operons within ~10 kb regions, consistent with the average size of a SD [42]. Given the abundance of promoters that are sensitive to supercoiling levels, coordinating transcription through supercoiling diffusion is likely to be an important way to coordinate gene regulation within a topological domain that is independent of the action of TFs. Additionally, the similarity of orientation of genes among bacteria suggests that the organization of genes within domains is conserved, and thus that SDs may play key roles in gene regulation (reviewed in [30, 31]). A prediction from this model is that arbitrarily shifting genes among topological domains may have large effects on expression. Studies have confirmed this prediction [231–233]; moving genes around the chromosome alters their expression levels, suggesting the local environment, which includes the level of supercoiling at a particular location, affects expression.

Overall, the dynamic nature of supercoiling and constraints on supercoiling provide a key mechanism by which chromatin topology affects transcription. Although local levels of supercoiling vary significantly throughout a bacterial genome, the average supercoiling among different bacteria also differs, suggesting a connection between supercoiling and evolved patterns of gene expression [36].

Many mechanistic questions remain about how HU and H-NS can act through topological effects on transcription including: (i) how dynamic is HU binding throughout the genome, especially near active transcription; (ii) can factors, such as other proteins or post-translational modifications, influence the ability of HU to constrain (−) supercoils (or (+) supercoils); (iii) how do changes to HU binding affect gene expression; (iv) do bridged H-NS filaments trap RNAP in a topological domains in vivo; (v) do H-NS modulators bind preferentially to certain genes to target gene silencing by H-NS; and (vi) do HU or H-NS influence the size of transcriptionally-isolated domains? New single-molecule and genome-scale approaches will be needed to shed light on the mechanisms by which DNA topology, chromatin structure, and transcription affect each other in bacteria.

Dps condenses DNA without inhibiting transcription

Whatever the role of macromolecular demixing and phase separation proves to be in actively growing bacteria, recent results suggest it is likely to operate in stationary phase (non-growing) bacteria where the conserved, chromatin protein Dps is induced to high levels and compacts DNA into a compact structure strongly resembling a phase-separated droplet [254] (Fig. 1). Dps is thought to improve survival in slow growth by both sequestering iron and protecting the DNA [255, 256]. Dps induces strong DNA compaction through the oligomerization of Dps dodecamers [62]. Like other DNA compacting proteins, it had long been thought that Dps would block gene expression [257, 258]. However, compelling recent studies from Meyers and co-workers show that Dps has little effect on transcription either in vivo and in vitro [254]. The results are consistent with Dps forming a phase-separated DNA condensate that allows access by RNAP even though it excludes some other proteins, like restriction endonucleases, in vitro.

These findings raise two intriguing possibilities for the interplay between bacterial chromatin and transcription. First, DNA compaction does not always silence transcription. Many proteins, like H-NS or the oxidative response protein WhiB4 from Mycobacterium [259], that bridge or compact DNA will repress gene expression, whereas Dps can compact DNA dramatically without affecting gene expression. Thus, DNA compaction and gene silencing cannot be equated in general. Second, DNA-protein condensates in bacteria may indeed attract some proteins (e.g., RNAP) and exclude others (e.g., restriction endonucleases), consistent with the properties of eukaryotic condensates known to exclude some molecules but not others [127].

Interestingly, a recently described Pseudomonas transcription factor that activates gene expression in stationary phase cells binds RNAP and contains essential IDRs of unknown function [260, 261]. It seems possible such a factor might work in part by aiding partition of RNAP into Dps-generated condensates. Although further evidence for formation of a phase-separated liquid condensate by Dps is needed [262], its discovery is a major step forward in understanding the possible roles of macromolecular demixing in bacterial gene regulation. Dps or Dps-like proteins in other species, such as Mycobacterium [263], also can condense DNA, suggesting that transcription within a separated phase could be a widespread phenomenon in bacterial chromatin.

Nucleoid substructuring could co-localize active genes to aid expression

An interesting implication for a role of phase separation in regulating gene expression is its ability to coordinate co-expressed genes in a 3D cluster. This strategy might be beneficial for the cell because it keeps RNAP nearby all the promoters that need to be co-expressed to increase initiation events. In other words, the 3D search for promoters used by RNAP could be facilitated by segregation of RNAP and co-regulated genes into a condensate [264, 265]. The close 3D proximity of co-expressed genes could also facilitate the putative phase separation of those loci from the rest of the nucleoid during times of high transcription [117]. Additionally, effects on gene expression when genes are moved around the chromosome [231, 232] could be explained by a model in which genes shift among phase-separated domains in addition to shifting among SDs. Although data does not currently exist for such phenomena in bacteria, they are plausible given the growing evidence that eukaryotic cells use condensates to coordinate expression of genes regulated by super-enhancers [249]. Experiments to identify characteristics of condensates along with technologies to map gene loci in 3D space, like PAINT [266] and a CRISPR-based technique [267], can help determine if phase separation aids in gene regulation in bacteria.