Structural coupling between RNA polymerase composition and DNA supercoiling in coordinating transcription: a global role for the omega subunit?

In growing bacterial cells, the global reorganization of transcription is associated with alterations of RNA polymerase composition and the superhelical density of the DNA. However, the existence of any regulatory device coordinating these changes remains elusive. Here we show that in an exponentially growing Escherichia coli rpoZ mutant lacking the polymerase ω subunit, the impact of the Eσ(38) holoenzyme on transcription is enhanced in parallel with overall DNA relaxation. Conversely, overproduction of σ(70) in an rpoZ mutant increases both overall DNA supercoiling and the transcription of genes utilizing high negative superhelicity. We further show that transcription driven by the Eσ(38) and Eσ(70) holoenzymes from cognate promoters induces distinct superhelical densities of plasmid DNA in vivo. We thus demonstrate a tight coupling between polymerase holoenzyme composition and the supercoiling regimen of genomic transcription. Accordingly, we identify functional clusters of genes with distinct σ factor and supercoiling preferences arranging alternative transcription programs sustaining bacterial exponential growth. We propose that structural coupling between DNA topology and holoenzyme composition provides a basic regulatory device for coordinating genome-wide transcription during bacterial growth and adaptation. IMPORTANCE Understanding the mechanisms of coordinated gene expression is pivotal for developing knowledge-based approaches to manipulating bacterial physiology, which is a problem of central importance for applications of biotechnology and medicine. This study explores the relationships between variations in the composition of the transcription machinery and chromosomal DNA topology and suggests a tight interdependence of these two variables as the major coordinating principle of gene regulation. The proposed structural coupling between the transcription machinery and DNA topology has evolutionary implications and suggests a new methodology for studying concerted alterations of gene expression during normal and pathogenic growth both in bacteria and in higher organisms.

Introduction

During bacterial growth, changes in environmental conditions induce coordinated alterations of genomic expression. In the classical model organism Escherichia coli, such coordinated changes are especially conspicuous in genome-wide transcriptional responses to experimentally induced alterations of chromosomal DNA topology (1–3). Topological alterations of DNA are normally associated with growth transitions (4), but the regulatory device that coordinates the instant response of global transcription to variations in DNA supercoiling remains unspecified. The process of transcription by RNA polymerase (RNAP) is intimately linked to the overall structure and superhelicity of the DNA in the bacterial nucleoid context (5), and there is ample evidence that DNA supercoiling is under the control of a homeostatic network comprising DNA topoisomerases, abundant nucleoid-associated proteins (NAPs), and the components of the transcription machinery (6–18). Since regulation of supercoiling necessitates the sensing of topological transitions in DNA, the networked genes themselves explicitly respond to changes in overall superhelicity (2, 3, 12, 19–21).

The cross talk between NAPs and DNA topoisomerases has been implicated in partitioning of the constrained and free DNA supercoils, generating topological domains and spatial transcription patterns in the genome (1, 3, 18, 22–28). Therefore, mutations of NAPs can alter both the overall superhelicity and the transcriptional response to variations in supercoiling (3, 29–31). Accordingly, under conditions of DNA relaxation and high negative supercoiling, NAPs organize distinct chromosomal domains with coherently expressed genes (18, 32). This type of transcriptional regulation depends on the available supercoil energy and is therefore dubbed a continuous or “analog” type of control, as opposed to the “digital” control exerted by specific binding of dedicated transcription factors (TFs) targeting a few high-affinity DNA sites in the gene promoter regions (32). The major impact of analog control on cellular physiology is evident from observations that the transcription of both the genes mediating adaptation to stress and the genes of central metabolic pathways of assimilation and dissimilation is coordinated by their distinct responses to DNA superhelicity (3, 33–36). Perhaps unsurprisingly, the biosynthetic genes are preferentially transcribed under conditions of high negative superhelicity, whereas transcription of catabolic genes involved in energy production is activated under conditions of DNA relaxation (3). Importantly, the introduction of negative supercoils into the DNA by the E. coli DNA gyrase depends on the ATP/ADP ratio, such that the overall superhelicity correlates with the phosphorylation potential in the cell (37–39). The availability and utilization of supercoil energy will therefore vary with the physiological state, supporting the notion of tight coupling between cellular metabolism and chromatin architecture (3, 14, 18, 36, 40).

The E. coli RNAP, acting as a unique transcribing machine, requires its composition to be adapted during bacterial growth to genome-wide changes in chromatin architecture. Genomic transcription sustaining fast cellular growth is normally associated with high levels of negative superhelicity and is driven predominantly by RNAP containing the major σ70 subunit encoded by the rpoD gene, whereas overall DNA relaxation on the transition to stationary phase (4) is correlated with enhanced utilization of the alternative σ38 subunit encoded by the rpoS gene (41, 42). In fast-growing cells under conditions of high phosphorylation potential, the σ38 protein interacts with a phosphorylated form of the RssB protein and is subsequently degraded by the ClpXP protease (43), such that the maximal gyrase activity and σ38 availability appear anticorrelated. Accordingly, studies on isolated promoters showed that the Eσ70 and Eσ38 holoenzymes favor highly supercoiled and more relaxed DNA for transcription, respectively (44, 45). Whether the distinct topological preferences of the Eσ70 and Eσ38 holoenzymes play a role in determining the supercoiling regimen of genomic transcription remains unclear. Yet, this question is of central importance for understanding the coordinated transcriptional responses observed during bacterial growth transitions and adaptation. On the assumption that the major metabolic pathways are coordinated by their distinct responses to supercoiling (3, 36), a shift in RNAP holoenzyme composition could provide a straightforward mechanism for reorganizing metabolism by switching the supercoiling regimen of genomic transcription.

In this study, we addressed this question by investigating the influence of the compositional alterations of RNAP on the overall DNA superhelicity and the supercoiling preferences of transcription in growing E. coli cells. We identified distinct transcription programs utilizing alternative clusters of σ factor- and supercoiling-dependent genes to sustain bacterial exponential growth and propose a role for the ω subunit of RNAP in optimizing the transcription of highly negatively supercoiled DNA.

RESULTS

Deletion of ω alters the σ70 content of RNAP.

We explored the relationships between RNAP composition and DNA topology using the E. coli CF1943 wild-type strain and an isogenic CF2790 rpoZ mutant lacking the RNAP ω subunit, in which our previous studies revealed topological alterations of DNA (M.G. and G.M., data not shown). The doubling times during growth at 37°C in rich 2× YT medium were 25.7 min for the wild type and 26.5 min for the mutant. RNAP subunit composition was monitored by Western analyses of whole-cell extracts and purified holoenzyme preparations isolated from exponentially growing cells. In cellular extracts, no significant differences in σ38 and σ70 subunit content were observed (Fig. 1A and B). However, in purified RNAP preparations isolated from the rpoZ mutant, the σ70 subunit was noticeably depleted whereas the σ38 subunit appeared slightly enriched (Fig. 1C). Since analyses of both whole-cell extracts and purified preparations may not faithfully reflect the RNAP composition found in vivo, we fractionated cell extracts by size exclusion chromatography to quantify the relative amounts of the σ38 and σ70 subunits in eluted holoenzyme fractions. Determinations of σ38/σ70 ratios clearly demonstrated enrichment for the Eσ38 holoenzyme in the cellular extracts of rpoZ mutant cells (Fig. 1D and E). These results are in keeping with the previously proposed “latching” effect of the ω subunit on Eσ70 holoenzyme assembly (46, 47) and suggest a shift of the Eσ38/Eσ70 holoenzyme balance in the rpoZ mutant.

FIG 1

Deletion of ω leads to depletion of σ70 and enrichment for σ38 in the RNAP holoenzyme. (A) Western blot analyses of the crude cellular extracts of exponentially growing CF1943 wild-type (WT) and isogenic CF2790 rpoZ mutant cells using antibodies against the β′, σ70, σ38, and α subunits. (B) Western blot analyses of the crude cellular extracts of stationary-phase (overnight) cultures of CF1943 and CF2790 rpoZ mutant cells using antibodies against the β′, σ70, σ38, and α subunits. (C) Western blot analyses of purified RNAP preparations isolated from exponentially growing CF1943 wild-type and CF2790 rpoZ mutant cells using antibodies against the α, σ70, and σ38 subunits. (D) Fractionation of CF1943 wild-type and CF2790 rpoZ mutant whole-cell extracts (harvested at an OD of 1; see Fig. 2A) by size exclusion chromatography. Fractions containing the RNAP holoenzyme were subjected to Western analysis using antibodies against β, σ70, and σ38. (E) Quantification of the σ38/σ70 ratios in the RNAP holoenzyme fractions of wild-type and rpoZ mutant cells using Western blot assays as shown in panel D. The values represent averages from two independent fractionation experiments.

RNAP composition impacts overall DNA topology in vivo.

We next assessed the influence of rpoZ mutation on DNA topology using plasmid pACYC184 as a reporter of overall DNA superhelical density in wild-type and mutant cells, which grew at similar rates under our experimental conditions (Fig. 2A). Nevertheless, growing wild-type cells have a clear selective advantage over the rpoZ mutant (Fig. 2B). Since we failed to overproduce the RpoZ protein from an expression vector in the rpoZ mutant (data not shown), to counterbalance the increased Eσ38/Eσ70 ratio observed in the rpoZ mutant, we overproduced σ70 from the episomal rpoD gene (pTrcSC-RpoD; see Materials and Methods) and monitored its influence on the topology of plasmid pACYC184. High-resolution gel electrophoresis analyses of the topoisomer distributions of the plasmids isolated at intervals after the inoculation of cells into fresh medium demonstrated a global relaxation of pACYC184 DNA in rpoZ mutant cells (linking deficit [|ΔLk|] = 3 to 4), whereas overproduction of σ70 in this mutant increased the average negative superhelicity to levels comparable to wild-type levels during the mid- to late-exponential phase (|ΔLk| = 4 to 5; Fig. 2C and D). The same procedure did not affect the plasmid topology in wild-type cells (data not shown). We infer that an increased Eσ38/Eσ70 ratio is associated with global relaxation of DNA in the exponentially growing rpoZ mutant, whereas overproduction of σ70 restores the high levels of negative superhelicity.

FIG 2

Effect of changing RNAP composition on DNA topology. (A) Growth curves of CF1943 wild-type cells, CF2790 rpoZ mutant cells mock transfected with pTrcSC, and CF2790 rpoZ mutant cells with the pTrcSC-RpoD plasmid overproducing σ70 after inoculation into fresh 2× YT medium. (B) Growing CF1943 cells have a selective advantage over CF2790 rpoZ mutant cells. Wild-type (wt) and mutant (chloramphenicol-resistant) cells were combined at either a 1:1 (black squares) or a 9:1 (open squares) proportion and diluted 1:10,000 into fresh medium. After 12 h of growth, two equal aliquots were plated on a selective medium containing chloramphenicol and on a nonselective solid medium. This procedure was repeated several times, and after each dilution cycle (abscissa), the percentage of chloramphenicol-resistant survivors on the selector plates was scored. The ordinate shows the percentage of chloramphenicol-resistant colonies. (C) High-resolution agarose gel electrophoresis of plasmid pACYC184 isolated from exponentially growing wild-type and rpoZ mutant cells with and without overproduction of σ70. Tracks 1 to 3, 4 to 6, and 7 to 9 correspond to 2 h, 3 h (mid-log phase), and 4 h (late-log phase) after inoculation, respectively. The genetic backgrounds of the host and the σ70 overproducer are indicated above the tracks. More negatively supercoiled topoisomers migrate faster in this gel. (D) Digital scans of topoisomer distributions (centers indicated by arrows) of three plasmid populations isolated 4 h after inoculation as shown in tracks 7 to 9 of panel C. Black line, wild type; grey line, rpoZ mutant; dotted line, rpoZ mutant overproducing σ70. Differences in the equilibrium distribution of the topoisomers (/ΔLks/) between the wild type and the rpoZ mutant (|ΔLk| = 3 to 4) and between the rpoZ mutants with and without σ70 overproduction (|ΔLk| = 4 to 5) are indicated.

Deletion of ω alters the σ factor selectivity of promoter recognition.

To compare the effective impacts of the Eσ70 and Eσ38 holoenzymes in the wild-type and rpoZ mutant backgrounds, we employed pSynP21 and pSynP9 constructs carrying synthetic promoters selectively recognized by the Eσ70 and Eσ38 holoenzymes, respectively (48; see Materials and Methods). The strains were transformed with synthetic promoter constructs, and the amount of specific transcripts produced during bacterial growth was quantified by real-time PCR (see Materials and Methods). Comparisons of the Eσ70-dependent SynP21 promoter activities in wild-type and rpoZ mutant cells demonstrated that the transcription rates varied with growth, being slightly lower in the mutant at most time points (Fig. 3A, left panel). In contrast, the Eσ38-dependent SynP9 promoter was both activated earlier and transcribed at a significantly higher rate in the rpoZ mutant (Fig. 3A, right panel), fully consistent with an increased proportion of the Eσ38 holoenzyme observed in fractionated extracts of mutant cells (Fig. 1). A similar bias was observed when we compared the Eσ70-dependent fis and Eσ38-dependent dps promoter activities in wild-type and rpoZ mutant cells (see Fig. S1 in the supplemental material). Furthermore, when we compared the topologies of the pSynP21 and pSynP9 plasmids isolated from exponentially growing wild-type and rpoZ mutant cells, we found that the pSynP9 plasmid demonstrated lower negative superhelical density than pSynP21 in both strains (Fig. 3B and C). Since pSynP9 is preferentially transcribed by the Eσ38 holoenzyme, this finding strongly suggests that the global relaxation of DNA observed in the rpoZ mutant during exponential growth is due to an increased impact of the Eσ38 holoenzyme, whereas increased negative superhelicity on the overproduction of σ70 reflects an enhanced effective impact of Eσ70 overriding the DNA relaxation phenotype.

FIG 3

Deletion of ω enhances the Eσ38 selectivity of transcription in vivo. (A) Relative expression of pSynP21 (left panel) and pSynP9 (right panel) determined by real-time PCR at intervals after inoculation of wild-type (WT) and rpoZ mutant cells in fresh medium. (B) Digital scans of topoisomer distributions of pSynP21 (left panel) and pSynP9 (right panel) plasmid populations isolated from exponentially growing wild-type and rpoZ mutant cells. (C) High-resolution agarose gel electrophoresis of pSynP21 and pSynP9 plasmid preparations isolated from exponentially growing wild-type and rpoZ mutant cells. More negatively supercoiled topoisomers migrate faster in this gel. The direction of migration in the gel is indicated by the arrow.

The σ factor-dependent reorganization of transcription.

Since the DNA relaxation phenotype of the rpoZ mutant was apparently complemented by the overproduction of σ70, we asked whether this overproduction also induced a transcription program for exponential growth similar to that of the wild type. For this purpose, we used microarray analyses to compare the DNA transcript profiles of wild-type and rpoZ mutant cells on the one hand and rpoZ mutant cells with and without overproduction of σ70 on the other. Two separate sets of the differential transcripts thus obtained were together subjected to average-linkage cluster analysis (see Materials and Methods) identifying coherently expressed genes. The gene transcripts showing high log ratios (upregulated) in rpoZ mutant backgrounds from both experiments formed together four clusters (clusters 3 to 6 in Fig. 4), whereas the genes showing high log ratios in the σ70-complemented rpoZ mutant and in wild-type cells formed two clusters (respectively, clusters 1 and 2 in Fig. 4; for full clustered gene lists, see Table S1 in the supplemental material). In these six clusters, we analyzed the σ factor selectivity of the gene promoters using the RegulonDB database (49) and calculated Z scores to evaluate the significance of the observed distributions (Table 1). Interestingly, clusters 1 and 2 were not conspicuously enriched for σ70-dependent genes, perhaps because the annotation of the σ70 dependence of promoters is far from complete. However, in these two clusters, only a small number of σ38-dependent genes was found, suggesting that transcription is programmed predominantly by Eσ70 polymerase. In contrast, the four clusters comprising the transcripts upregulated in the rpoZ mutant background contained many σ38-dependent genes (especially clusters 5 and 6; Fig. 4; Table 1), fully consistent with the increased impact of Eσ38-programmed transcription observed in rpoZ mutant cells.

FIG 4

Cluster analysis of transcript profiles. Clusters shown as light and dark grey bars are numbered 1 to 6. Cluster 2 corresponds to genes upregulated in wild-type cells compared to rpoZ mutant cluster 4. Cluster 1 corresponds to genes upregulated in the rpoZ mutant overproducing σ70 from pTrcSC-RpoD (indicated by the dashed rectangle) compared to the rpoZ mutant mock transfected with pTrcSC (clusters 3, 5, and 6). The distributions of hyp and rel genes, gyrase binding sites, and σ factor dependences are shown as barcodes at the bottom. Coefficients of correlation between clusters and the main TFs impacting the clusters are indicated.

TABLE 1

Composition of the identified clusters of sigma factors and supercoiling sensitivity

Cluster	Size (no. of genes)	No. of ORFs (Z score)[a]
		hyp	rel	σ70	σ38	gyr
1	197	53 (2.50)	16 (−3.80)	49 (−0.33)	3 (−1.20)	123 (16.00)
2	74	21 (1.90)	13 (−0.24)	20 (0.23)	1 (−0.79)	51 (11.00)
3	105	21 (0.06)	24 (1.10)	23 (−0.93)	7 (2.30)	61 (11.00)
4	65	13 (−0.04)	12 (−0.04)	15 (−0.50)	0 (−1.40)	30 (5.80)
5	36	3 (−1.70)	10 (1.40)	9 (−0.12)	14 (13.00)	20 (5.80)
6	62	9 (−1.00)	19 (2.40)	14 (−0.61)	13 (8.40)	33 (7.30)

Within each cluster, the numbers of known σ38 and σ70 transcribed ORFs, the supercoiling sensitivities of the ORFs, and the gyrase binding sites within the ORFs were determined. Corresponding Z scores were calculated for each analyzed parameter by comparison of detected values with appropriate null models.

Supercoiling preferences of gene clusters.

Previous studies identified about 300 genes distinctly responding to DNA relaxation (2), whereas direct transcriptome comparisons of E. coli cells grown under conditions of DNA relaxation or hypernegative DNA supercoiling, including mutant strains lacking the NAPs, allowed an almost 10-fold increase in the number of identified supercoiling-sensitive genes (3, 36). The latter data set, successfully used in previous studies to link the supercoiling response to both genomic expression patterns and cellular metabolism (3, 32, 36), was employed to determine the supercoiling preferences of the six identified clusters. We found that clusters 1 and 2, respectively corresponding to the wild type and the σ70-complemented rpoZ mutant, were enriched for hypernegative (hyp) genes (Fig. 4; Table 1). Conversely, Eσ38-controlled clusters 3, 5, and 6 from the rpoZ mutant background were enriched for genes responding to DNA relaxation (rel). These findings are fully consistent with the background-dependent differences in the topology of isolated plasmids (Fig. 2). Since DNA gyrase is the major enzyme introducing negative supercoils into DNA, we have also analyzed the frequency distributions of gyrase binding sites using the available library of sequences identified in the E. coli genome (1). We found that, compared to clusters 5 and 6, cluster 1 was noticeably enriched for gyrase binding sites (Table 1). Taken together, these results suggest that, by and large, a combination of two variables—σ factor selectivity and the supercoiling dependence of genes—underlies the organization of clusters.

Properties of the impacting transcriptional regulators.

We next analyzed the transcriptional regulators constraining the effective transcriptional regulatory networks (32) in each cluster (see Table S2 in the supplemental material; Fig. 4). We found that although the gene clusters share the NAPs, each cluster is impacted by a unique combination of NAPs and dedicated TFs. Notably, the regulatory impacts of FIS, IHF, and CRP were observed in clusters with opposite supercoiling and σ factor preferences, whereas the impacts of H–NS and LRP were conspicuous only in Eσ38-controlled clusters 5 and 6. Furthermore, in clusters 1 and 2, associated with high negative superhelicity, we observed the impacts of several regulators (AgaR, CyrR, ExuR, MalT, NarL, NarP, and SrlR) which respond to conditions of high negative supercoiling or are repressed by relaxation of DNA; conversely, the regulators impacting clusters 3 to 6 associated with DNA relaxation in the rpoZ mutant (BolA, DnaA, Fur, GadE, GadW, GatR, IscR, LacI, NhaR, OxyR, and TrpR) were found to respond to DNA relaxation (2, 3; M.G. and G.M., unpublished data). Finally, when we analyzed the distributions of annotated Eσ70 and Eσ38 promoters among the genes of the impacting regulators themselves, we again found that the regulators of clusters 3 to 6 pooled together were enriched for σ38-dependent genes in comparison with regulators of pooled clusters 1 and 2, where we did not find any significant bias (see Table S3 in the supplemental material).

Couplon matrix analysis.

To verify the increased impact of the Eσ38 holoenzyme in the rpoZ mutant by an independent method, we used couplon matrix analysis, which enables the measurement of significant changes both in the regulons corresponding to individual regulators (be it a NAP or a σ factor) and in couplons—subsets of regulons jointly regulated by distinct combinations of σ factors and NAPs (40). Couplon analysis was carried out with the same two sets of differential transcripts obtained from comparisons of (i) wild-type and rpoZ mutant cells and (ii) rpoZ mutant cells with and without overproduction of σ70. In the first pool, the genes that were upregulated in the wild-type background (cluster 2) did not reveal any significant changes, whereas analysis of genes upregulated in the rpoZ mutant background (cluster 4) demonstrated changes in the σ38 regulon (Fig. 5A and B). Similar analyses of the second set of differential transcripts demonstrated significant changes in the CRP/σ70 couplon, as well as the CRP and Fnr regulons in the σ70-complemented rpoZ mutant background (cluster 1, Fig. 5C). In contrast, the genes upregulated in the rpoZ mutant background without σ70 overproduction (clusters 3, 5, and 6) revealed significant changes in all of the NAP/σ38 couplons, as well as in the entire σ38 regulon (Fig. 5D). Thus, the couplon matrix analysis also indicates an increased impact of the Eσ38 holoenzyme in the rpoZ mutant background in both sets of differential transcripts. Taken together, our observations suggest that in wild-type and rpoZ mutant cells, the compositionally different RNAP holoenzymes utilize distinct combinations of NAPs and supercoiling regimens for the assembly of alternative transcription programs sustaining exponential bacterial growth.

FIG 5

Couplon analyses of transcript profiles. The couplon matrix is organized by orthogonally intersecting regulons (derived from the RegulonDB E. coli K-12 transcriptional network) of the NAPs and RNAP σ factors indicated on the horizontal and vertical axes, respectively. Each square at an intersection thus defines a distinct couplon coordinated by two coupled regulators (40). In each panel, the last rows and columns indicated by plus signs represent the entire regulons. Rainbow colors indicate significantly increased (red) or decreased (blue) numbers of genes in the effective profile set compared to 10,000 equal-sized random input sets. Corresponding Z scores (values inside squares) are indicated. (A) Couplon matrix of upregulated genes in the wild-type background (cluster 2). (B) Couplon matrix of upregulated genes in the rpoZ mutant background (cluster 4). (C) Couplon matrix of upregulated genes in the rpoZ mutant background with σ70 overproduction from pTrcSC-RpoD (cluster 1). (D) Couplon matrix of upregulated genes in the rpoZ mutant background plus mock transfection with pTrcSC (i.e., without σ70 overproduction; clusters 3, 5, and 6).

DISCUSSION

In this study, we show that during the exponential growth of the rpoZ mutant lacking the RNAP ω subunit, Eσ38-programmed transcription is increased in concert with an overall reduction in negative superhelicity and a preferential utilization of genes associated with DNA relaxation. Conversely, overproduction of σ70 in this mutant complements the DNA relaxation phenotype and leads to the preferential transcription of genes associated with high negative superhelicity. While this suggests a tight coupling between holoenzyme composition and DNA topology, a noteworthy difficulty in assessing the impact of holoenzyme composition on DNA supercoiling arises from the peculiarity of the transcription process, which itself, independent of the molecular constitution and source of the holoenzyme, affects supercoil dynamics and the overall shape of the DNA (50, 51). It is assumed that the supercoil dynamics of the genomic DNA depend on the strength of transcription (26). Consistently, we observe a correlation between the rate of transcription (compare the relative expression levels on the ordinates of graphs in the right and left panels of Fig. 3A) and overall superhelical density of isolated pSynP21 and pSynP9 plasmids (Fig. 3B and C). However, we note that the transcription of plasmids pSynP21 and pSynP9 is preferentially driven by the Eσ70 and Eσ38 holoenzymes, respectively, and that only negligible topological change is associated with substantially increased pSynP9 construct transcription in the rpoZ mutant compared to that in the wild-type strain (compare Fig. 3A and B, right panels). We propose, therefore, that the molecular composition of the holoenzyme is the primary determinant of both the strength of transcription and the associated DNA superhelicity in vivo. Early studies suggested that mutations of RNAP can alter global DNA supercoiling (52, 53), and several lines of evidence are consistent with a direct impact of the molecular composition of RNAP on overall DNA topology. It was observed that changes in RNAP composition affect not only the specificity of promoter sequence recognition (54) but also the preference for the binding of upstream sequences (48). Furthermore, holoenzyme composition was found to determine the extent of the DNA wrapped in the initial complex (55). In addition, compositional changes in RNAP could modulate the interactions with secondary channel regulators, DNA topoisomerases, and DNA architectural factors (27, 56–58), thus transmitting the compositional change to DNA superhelicity. Other conceivable possibilities for composition-dependent effects of RNAP on DNA topology include altered propensities to generate R loops or form transcription foci (18, 59, 60).

A global role for ω?

Our data implicate ω in the optimization of the Eσ70-programmed transcription of highly negatively supercoiled DNA during exponential growth. How can loss of the RNAP ω subunit lead to reorganization of transcription? The rpoZ gene is transcribed during exponential growth (M.G. and G.M., unpublished), but it is not clear whether every polymerase molecule is associated with ω (61). Yet, our stoichiometry determinations (see Fig. S2 in the supplemental material) suggest that during exponential growth there is at least one copy of ω per RNAP molecule. Therefore, a simple hypothesis is that the phenotype of the rpoZ mutant is dependent on structural changes in the RNAP holoenzyme lacking the latching effect of ω (46, 47).

Since the phenotypic effects of the loss of ω can be suppressed by overproduction of σ70, it is conceivable that ω modulates the relative affinities of σ38 and σ70 for the core polymerase. The σ factor competition is subject to complex regulation involving the “alarmone” ppGpp (62), which represses the “stringent” promoters utilizing Eσ70 (e.g., stable RNA promoters) and activates the Eσ38-dependent promoters and also those Eσ70-dependent promoters involved in the response to stress (63). However, in the rpoZ mutant, the “stringent response” is not impaired (64) nor is stable RNA transcription diminished (65), although in the absence of ω, the altered sensitivity to ppGpp could be masked by the secondary channel regulator DksA (66). Notably, DksA is not only implicated in the balance of secondary channel regulators compensating for certain effects of ppGpp (67, 68) but also modulates the topological barriers to supercoil diffusion (27). Another relevant regulator would be the Crl protein facilitating Eσ38 holoenzyme assembly (69, 70), but we did not find changes in either dksA or crl expression in our transcriptome studies (see Table S1 in the supplemental material). The possibility of a direct effect of ω on σ factor competition thus remains an attractive hypothesis to be elucidated.

Alternative programs for exponential growth.

Our observation of alternative ways of organizing transcription programs sustaining fast exponential growth—one utilizing mainly Eσ70 and the other utilizing both the Eσ70 and Eσ38 holoenzymes—reveals a new flexibility of genetic regulation and is in keeping with the notion that σ38 is acting as a second vegetative σ factor under nonoptimal growth conditions (71). These alternative growth programs employ functional gene clusters with distinct σ factor and DNA topology preferences (Fig. 4). In Eσ70-dependent cluster 1, enriched for hyp genes, we also calculated the highest Z score for associated gyrase binding sites, whereas clusters 5 and 6 showed both increased Eσ38 selectivity and an increased preference for rel genes (Table 1). Clusters 3 and 4 appear less clearly defined. Perhaps unsurprisingly, clusters preferring different holoenzymes nevertheless share the impacting global regulators—the NAPs (Fig. 4; see Table S2 in the supplemental material). Yet, for each cluster, the constellations of regulators are different, suggesting substantial combinatorial flexibility in optimizing transcription. More specifically, the impact of H–NS and LRP, known as modulators of Eσ38 selectivity (53, 72), was detected in Eσ38-dependent clusters 5 and 6 (see Table S2), whereas the impact of FIS was observed in two clusters (2 and 5) with opposite σ factor-supercoiling couplings. This is consistent with buffering effects of FIS on Eσ70-dependent promoters requiring high negative superhelicity (e.g., stable RNA promoters) that are rescued by FIS on DNA relaxation and also with comodulation of gene expression by FIS and σ38 (73, 74). Similarly, the impact of CRP observed in distinct clusters 1 and 4 is consistent with both the proposed cooperation between σ38 and CRP (42) and the existence of two promoters enabling crp transcription under different supercoiling regimens (20).

While the overproduction of σ70 in wild-type cells has no effect on holoenzyme composition (75), in the exponentially growing rpoZ mutant, the directional switch of the transcription program is especially conspicuous when cells with and without σ70 overproduction are compared. The significant changes in couplons observed in the rpoZ mutant under these conditions (Fig. 5D) suggest cross talk between Eσ38 and the NAPs on the relaxation of DNA. In keeping with this observation, in the rpoB114 mutant containing the “stringent” RNAP holoenzyme with impaired sensitivity to ppGpp, increased production of IHF is associated with overall DNA relaxation (76; M.G. and G.M., unpublished). Importantly, despite overall DNA relaxation and increased Eσ38-programmed transcription, under our experimental conditions, wild-type and rpoZ mutant cells grew at comparable rates, such that a selective growth advantage of wild-type cells could be revealed only under conditions of direct competition (Fig. 2A and B). This suggests that sufficient Eσ70 is available in rpoZ mutant cells for stable RNA synthesis, as previously reported (65), and that reduced overall superhelicity per se does not necessarily preclude efficient growth, as inferred also from the “experimental evolution” study of E. coli populations (77). Indeed, the phenotype of the rpoZ mutant does not resemble the phenotype induced by σ70 underproduction (78). However, overproduction of σ70 in this mutant represses the uspB and uspF genes, as well as other stress-related genes normally activated on the transition to stationary phase (see Table S1 in the supplemental material). This is in keeping with the recently proposed Eσ70-dependent shift in balance toward cellular growth at the expense of maintenance (75). Nevertheless, despite similar growth rates, the transcription profiles of exponentially growing wild-type and σ70-complemented rpoZ mutant cells are substantially different, again consistent with remarkable flexibility of genetic regulation.

Coordination of genomic transcription.

In order to coordinate genomic transcription with physiological demands, the RNAP holoenzyme has to both integrate the entire information processed by the homeostatic network and organize the environment in which it operates. Importantly, the expression of transcription machinery components of the network, including the rpoD gene, is strongly dependent on DNA superhelicity (2, 3), whereas we show here that RNAP composition can, in turn, determine the supercoiling regimen of genomic transcription. We thus reveal a relationship of interdependence which is pivotal for coordinating genomic transcription, as alterations of DNA topology and transcription machinery can be reciprocally transmitted (40). The identification of clusters of alternative σ factor- and supercoiling-dependent genes suggests a simple mechanism for rearranging metabolism by coupling the different holoenzyme forms with distinct structural dynamics of DNA at the genome-wide level. What is the structural basis of this coupling? Observed differences in cognate promoter sequence organization (54, 79), in preferences for distal and proximal half-sites of the UP elements (48), and in the propensity to wrap DNA (55) implicate the Eσ70 and Eσ38 holoenzymes in the recognition of different supercoil structures. By selecting distinct DNA geometries, these holoenzymes could determine the relative expression of abundant NAPs constraining different superhelical densities and thus optimize the σ factor selectivity of transcription (17, 44, 45, 72).

We propose that interdependent alterations of transcription machinery composition and DNA topology represent a basic regulatory device coordinating genome-wide transcription during bacterial growth and adaptation (Fig. 6). This necessarily raises the question of how tight this interdependence is. For example, overproduction of σ38 does not induce Eσ38-dependent osmE transcription until relaxation of the template DNA (45). This observation, together with our data presented in Fig. 1, suggests that it is the effective concentration of a particular holoenzyme that matters and not the cellular concentration of a σ factor per se. We believe that an exploration of structural coupling between the transcription machinery and chromosomal DNA topology has far-reaching evolutionary implications, providing a new methodology for studying the coordinating principles of gene regulation during normal and pathogenic growth in both bacterial and eukaryotic cells (80).

FIG 6

Coordination of transcription by structural coupling between the transcription machinery and DNA topology. In this model, a shift in the holoenzyme σ factor composition determines both the preferred superhelicity for transcription and the subsets of NAPs and dedicated TFs cooperating in the assembly of the transcription program. The subsets of NAPs, in turn, optimize the utilization of DNA superhelicity (3, 36), fine-tune TF selection, and also act themselves as hubs in the transcriptional regulatory network. In this way, distinct combinations of regulators coordinate genomic transcription with metabolism, thus sustaining alternative growth programs A and B. Metabolism feeds back into DNA topology and RNAP composition by determining the ATP/ADP ratios and ppGpp concentration and thus affecting gyrase activity (37–39) and σ factor competition (62, 63), respectively. As we show in this study, lack of ω leads to an increased impact of Eσ38-driven transcription, but the altered program nevertheless can support exponential bacterial growth. The regulatory effects of the anti-sigma factors and degrading proteases are omitted for clarity. For further details, see the text. -ve sc, negative superhelicity.

MATERIALS AND METHODS

Bacterial strains and plasmids.

Isogenic E. coli K-12 strains CF1943 (wild type) and CF2790 (rpoZ mutant) (64) were used throughout this study. The σ70 protein was overproduced using pTrcSC-rpoD, a derivative of pTrc99A (81) harboring the rpoD gene and a pSC101 replication origin. Synthetic promoter constructs pSynp21 and pSynp9 (kind gift of Regine Hengge, Freie Universität Berlin), are variants of pSynp213 derived from the core Ptac promoter and translationally fused to the lacZ gene (48).

The competition experiment between CF1943 and chloramphenicol-resistant CF2790 rpoZ mutant cells for the determination of a selective growth advantage was carried out as described by Schneider et al. (82). Briefly, overnight cell suspensions of the two strains after mixing at different ratios and subsequent growth for 12 h were diluted 1 × 10−6 before plating and the percentages of survivors on chloramphenicol plates were scored. The final concentration of chloramphenicol in the selective plates was 25 µg/ml. The cultures were again diluted 1:10,000 in fresh 2× YT medium and allowed to grow for a further 12 h. This procedure was repeated seven times.

Purification of RNAP.

Bacterial cultures were grown under carefully controlled conditions in BIOSTAT Bplus (Sartorius AB, Göttingen, Germany). The holoenzymes were purified from exponentially growing cells by high-performance liquid chromatography (82, 83). In brief, crude cell extract was loaded onto a HiPrep 16/10 heparin FF column (GE Healthcare, Munich, Germany). The collected peak fraction was reloaded onto a Mono Q 5/50 GL column (GE Healthcare, Munich, Germany) and polished using an Amicon Ultra-4 100K centrifugal filter (Millipore, Schwalbach, Germany).

Western analyses.

Quantitative Western analyses (ECF kit; GE Healthcare, Munich, Germany) of crude cellular extracts and purified RNAP preparations were performed using mouse monoclonal antibodies raised against the β′, β, α, σ70, and σ38 subunits (Neoclone, Madison, WI). Quantifications were carried out using a PhosphorImager.

Analysis of plasmid transcription.

For quantifications of transcripts produced from the pSynp21 and pSynp9 plasmids, total RNA was extracted using the Qiagen RNeasy MinElute cleanup kit, followed by DNase I digestion using the Qiagen RNase-Free DNase set (Qiagen GmbH, Hilden, Germany). An additional DNase I treatment was performed using the Ambion RNase-free DNase I kit (Applied Biosystems, Foster City, CA). RNA concentrations were determined at 260 nm using a NanoDrop ND-1000 spectrophotometer, and transcripts were quantified by one-step real-time PCR using the QuantiTect SYBR green kit (Qiagen GmbH, Hilden, Germany) and an Mx3000P real-time cycler (Stratagene, La Jolla, CA).

The primers used for real-time PCR are as follows: pSyn_for, 5′ CCCTATTCAGCAATGCAACC 3′; pSyn_rev, 5′ GTAAAACGACGGGAGCAAGC 3′.

Fractionation of cellular extracts.

Wild-type and rpoZ mutant cells were grown in 2× YT at 37°C to an optical density (OD) at 600 nm of 1. The cells were harvested, washed with phosphate-buffered saline, and resuspended in 5 ml of equilibration buffer (10 mM Tris-HCl [pH 7.8], 0.1 mM dithiothreitol, 0.1 mM EDTA, 200 mM NaCl). Cells were disrupted by sonication in the presence of protease inhibitor mix HP (Serva, Heidelberg, Germany) and 10 U/ml Benzonase (Sigma-Aldrich Chemie GmbH, Steinheim, Germany). After clearance of whole-cell extracts by centrifugation at 10,000 × g for 1 h at 4°C, a total of 3.5 mg of protein was applied to a size exclusion column (Superose 6 10/300; GE Healthcare) using an ÄKTA system (GE Healthcare). Elution with equilibration buffer was performed at the recommended flow rate of 0.5 ml/min at 4°C. Fractions of 250 µl were acetone precipitated, analyzed by 12% SDS-PAGE, and transferred onto a polyvinylidene difluoride membrane for Western analyses. Experiments were performed twice with reproducible results.

High-resolution gel electrophoresis.

High-resolution gel electrophoresis of plasmids was carried out as described previously (21). The ImageQuant software was used to analyze topoisomer distributions and estimate the ΔLk value by the band-counting method (84).

DNA microarray analyses.

All of the strains used for transcript profiling were grown in 2× YT medium at 37°C. DNA microarray experiments were performed using OciChip E. coli K-12 V2 arrays according to the OciChip application guide (Ocimum Biosolutions, Hyderabad, India) as previously described (3). In brief, for each comparison, two biological replicates with two technical replicates were performed, resulting in a total of 8 hybridizations. Scanned array images were analyzed using the TM4 software package (85). Spot intensities were quantified, and the quality of each spot was verified by calculating a quality control (QC) score depending on the signal-to-noise ratio for every channel and calculating P values for each channel (as a result of a t test comparing the spot pixel set and the surrounding background pixel set) using the Spotfinder software from The Institute for Genomic Research (TIGR). Data were normalized by locally weighted linear regression (86). A one-class t test (87) was applied to obtain differentially expressed genes within each data set (significance level, α < 0.05).

Cluster analysis of the transcript profiles.

Ward’s linkage cluster analysis based on a Euclidean distance matrix was applied to gene expression data sets. Within each cluster, σ70 and σ38 dependencies, TF-target gene (TG) relationships, and supercoiling sensitivities of transcript profiles were determined. While σ70 and σS dependencies and TF-TG relationships were derived from the RegulonDB database (49), supercoiling sensitivities were determined as described previously (3). A publicly available chromatin immunoprecipitation-chip data set was used and analyzed as described in the DNA microarray section to investigate open reading frames (ORFs) that show significant gyrase binding (1; Gene Expression Omnibus accession number GSE1735). Within each cluster, Z scores for σ dependencies, TF-TG relationships, supercoiling sensitivities, and gyrase binding were calculated from the mean and standard deviation of 10,000 runs of the corresponding null model (i.e., random sampling of ncluster size genes).

Couplon analyses.

Couplon analyses were carried out using the coordination matrix and measuring couplon penetrance essentially as described in reference 40.

Microarray data accession number.

The microarray data reported here have been submitted to the Array Express database and assigned accession number E-MEXP-942.

Deletion of ω enhances the Eσ38 selectivity of transcription in vivo. (A) Relative expression of the Eσ70-dependent fis promoter cloned into plasmid pUC18 (pUCfis4) determined by real-time PCR at intervals after inoculation (abscissa) of wild-type and rpoZ mutant cells into fresh medium. All values were normalized to the activity measured in wild-type cells 1 h after inoculation, which was taken as 100% (black circle). (B) Relative expression of the Eσ38-dependent dps promoter cloned into plasmid pUC18 (pUCdps6). The values were normalized to the activity measured in wild-type cells 2 h after inoculation. The ordinate shows relative expression normalized by OD. Measurements were done using two biological replicates with two technical replicates each. The fis and dps promoters in the pUCfis4 and pUCdps6 constructs (kind gift of V. Gerganova, Jacobs University) each contain the respective minimal promoter regions plus about 300 bp of the upstream sequence. Both promoters were used as transcriptional fusions to the yfp gene, and the amount of transcripts was quantified by real-time PCR using yfp-specific primers. It is noteworthy that promoters located on multicopy plasmids do not faithfully reproduce the growth phase patterns of the chromosomal fis and dps gene expression (V. Gerganova, unpublished data; see also Fig. 2h in the paper by A. Zaslaver et al. [Nat. Methods 3:623-628, 2006]). However, the activity of the Eσ38-dependent dps promoter is, as predicted, increased in the rpoZ mutant background. (C) Growth curves of wild-type and rpoZ mutant cells transformed with plasmids pUCfis4 and pUCdps6. Download Figure S1, PDF file, 0.115 MB.

Figure S1, PDF file, 0.115 MB

(A) Quantification of the ω to σ stoichiometries in isolated RNAP preparations. Densitometric scans of RNAP holoenzyme preparations stained with colloidal Coomassie were used to determine stoichiometries. The calculation was done both with correction for the number of lysines contained in each protein (3 and 16 in ω and σ, respectively) and without correction. (B) Representative PAGE with the ω and σ subunits (verified by mass spectrometry) indicated. M, size marker. Download Figure S2, PDF file, 1.455 MB.

Figure S2, PDF file, 1.455 MB

Genes demonstrating significant changes in expression. Scanned array images were analyzed using the TM4 software package (86). Spot intensities were quantified, and the quality of each spot was verified by calculating a QC score depending on the signal-to-noise ratio for every channel and shape quality for each spot calculating P values for each channel as the result of a t test comparing the spot pixel set and the surrounding background pixel set using TIGR Spotfinder software. Data were normalized by locally weighted linear regression (87), and a one-class t test (88) was applied to replicated experiments to obtain differentially regulated genes with significant P values (<0.05) using the TIGR MIDAS software. For both experimental sets (set 1 compares the wild type with the rpoZ mutant, and set 2 compares the rpoZ mutant with and without σ70 overproduction), the mean log ratios, corresponding P values, cluster affiliations, regulating sigma factors, supercoiling sensitivities (SC), gyrase binding sites (GYR), and regulator genes are indicated. Note that some genes identified in both sets are identical.

Table S1, XLS file, 0.106 MB.

Identified transcriptional regulators constraining effective transcriptional regulatory networks (eTRNs) in six clusters. Within each cluster, the number of known TF-TG interactions was determined (enclosed in parentheses). For each cluster, the Z scores obtained by comparisons of detected TF-TG interactions with appropriate null models are shown and the significance of regulatory interactions is indicated.

Table S2, DOCX file, 0.024 MB.

Enrichment of σ factor modulation of the TFs in identified clusters. Within each cluster, the number of known TF targets for a given σ factor is shown (enclosed in parentheses). Using known σ-regulated TFs available from RegulonDB, appropriate null models were generated by pooling the TFs impacting all six clusters and calculating the Z scores shown and P values indicating the significance of modulating a given cluster by a particular σ factor.

Table S3, DOC file, 0.038 MB.