Spatial organization of RNA polymerase and its relationship with transcription in Escherichia coli

Significance

In fast-growing Escherichia coli cells, RNA polymerase (RNAP) has been reported to form dense clusters, which were proposed to be active transcription centers for ribosomal RNA (rRNA) synthesis. Understanding how and why RNAP is spatially organized in bacterial cells could offer insights into the mechanism of transcription regulation in the native cellular environment. Using quantitative superresolution imaging coupled with genetic and biochemical analyses, we found that RNAP clusters were indeed active transcription centers engaged in rRNA synthesis under a rich media growth condition, but their formation was independent of rRNA transcription activity. Instead, the underlying nucleoid structure played a significant role in spatially organizing RNAP clusters. These results shed new light on the relationship between RNAP clusters and transcription activity.

Abstract

Recent studies have shown that RNA polymerase (RNAP) is organized into distinct clusters in Escherichia coli and Bacillus subtilis cells. Spatially organized molecular components in prokaryotic systems imply compartmentalization without the use of membranes, which may offer insights into unique functions and regulations. It has been proposed that the formation of RNAP clusters is driven by active ribosomal RNA (rRNA) transcription and that RNAP clusters function as factories for highly efficient transcription. In this work, we examined these hypotheses by investigating the spatial organization and transcription activity of RNAP in E. coli cells using quantitative superresolution imaging coupled with genetic and biochemical assays. We observed that RNAP formed distinct clusters that were engaged in active rRNA synthesis under a rich medium growth condition. Surprisingly, a large fraction of RNAP clusters persisted in the absence of high rRNA transcription activities or when the housekeeping σ⁷⁰ was sequestered, and was only significantly diminished when all RNA transcription was inhibited globally. In contrast, the cellular distribution of RNAP closely followed the morphology of the underlying nucleoid under all conditions tested irrespective of the corresponding transcription activity, and RNAP redistributed into dispersed, smaller clusters when the supercoiling state of the nucleoid was perturbed. These results suggest that RNAP was organized into active transcription centers under the rich medium growth condition; its spatial arrangement at the cellular level, however, was not dependent on rRNA synthesis activity and was likely organized by the underlying nucleoid.

Prokaryotes are traditionally viewed as bags of freely diffusing enzymes. This view is rapidly changing. New studies now document that bacterial cells possess a remarkable degree of spatial organization of cellular components and activities without the use of membranes, offering a level of functionality and regulation previously underappreciated (1–4). In both Escherichia coli and Bacillus subtilis cells grown in rich media, RNA polymerase (RNAP), the only enzyme responsible for all RNA transcription, was found to form dense foci instead of distributing homogenously within the cell (5, 6). Because the majority of cellular RNAP is dedicated to ribosomal RNA (rRNA) synthesis in fast-growing cells (7), a transcription factory model was proposed (8). This model suggests that dense RNAP foci are clusters of hundreds of RNAP molecules actively engaged in rRNA transcription and that their formation is driven by active rRNA synthesis in fast-growing cells under optimal growth conditions (such as LB, 37 °C) (5, 8, 9). This prokaryotic transcription factory model is reminiscent of the RNAP I transcription factory model in eukaryotic cells, in which RNAP I form concentrated, membrane-free condensates in the nucleolus for rRNA transcription (10, 11).

Understanding how and why RNAP is spatially organized in bacterial cells is important, as this information could provide new insights into the mechanisms of transcription regulation in a complex, heterogeneous cellular environment. For example, in eukaryotic cells, it was suggested that RNAP clusters might represent preformed transcription complexes that are “poised” ready for rapid transcription induction (12–15). In bacterial cells, such a role has not been demonstrated, but studies have shown that there are typically higher levels of RNAP association at promoter and promoter-like sequences than within coding sequences (16–21). However, partially due to technical limitations in dissecting the subcellular organizations of small bacterial cells, these possibilities remain unexamined. In particular, despite a number of recent studies that extensively investigated the distribution and characteristics of RNAP clusters in E. coli (22–24), whether RNAP clusters observed in fast-growing cells are indeed active in rRNA transcription, and whether RNAP clusters only form in the presence of active rRNA transcription, have not been directly examined. Previous studies have shown that treating cells with rifampicin (RIF), a global transcription inhibitor (25), largely abolished the appearance of RNAP foci (9, 24, 26). However, it remains unclear whether this change was due to diminished rRNA transcription activity, or the associated nucleoid structural changes under the condition of global transcription inhibition (27, 28).

In this study, we characterized the spatial organization and transcription activity of RNAP under different conditions using quantitative superresolution imaging in E. coli cells. We demonstrated that there was a rRNA transcription activity-independent spatial organization of RNAP in E. coli, and that the underlying nucleoid structure played important roles in organizing RNAP clusters.

Results

RNAP Forms Distinct Clusters in Cells Growing in Rich Defined Medium.

To investigate the spatial organization of RNAP in E. coli, we used a strain in which the chromosomal rpoC gene encoding for the β′ subunit of RNAP was replaced by a photoactivatable fluorescent protein gene fusion, rpoC–PAmCherry (23, 24, 29). We verified that the resulting RpoC–PAmCherry fusion protein was expressed in full-length (SI Appendix, Fig. S1A), was incorporated efficiently into the RNAP core enzyme complex (SI Appendix, Fig. S1B), and supported WT-like cell growth as the sole cellular source of the β′ subunit (SI Appendix, Fig. S1C). Therefore, the spatial distribution and dynamics of the RpoC–PAmCherry fusion protein should be representative of the native RNAP core or holoenzyme. In the text below, we refer to this fusion protein as RNAP–PAmCherry for simplicity.

Using RNAP–PAmCherry, we performed single-molecule localization-based superresolution imaging (30) on exponentially growing live cells in EZ rich defined medium (EZRDM) at room temperature (25 °C, cell doubling time = 73 ± 1 min, hereafter termed as the rich medium growth condition) (SI Appendix, Fig. S1C) with a measured 2D spatial resolution of ∼50 to 60 nm (SI Appendix, Fig. S2). We observed clustered distributions of RNAP–PAmCherry in individual cells (Fig. 1A). These clusters were distinct but less punctate compared to what were observed under faster growth conditions (EZRDM 37 °C and LB at 37 °C) (SI Appendix, Fig. S3A), consistent with what was reported previously (22–24). The averaged cellular distribution of all RNAP localizations displayed a 2-lobed pattern with a clear cleft in the middle (Fig. 1B), similar to that of the nucleoid imaged using 3D structured illumination superresolution microscopy (SIM) (SI Appendix, Fig. S4A). In contrast, free PAmCherry molecules and PAmCherry fused to a nonspecific DNA binding protein HU both exhibited a significantly more homogenous distribution in cells (SI Appendix, Fig. S5A). Using a truly monomeric mEos3.2-fused RNAP fusion protein (31), we further verified that the clustered distribution was not due to the weak dimerization property of PAmCherry (SI Appendix, Fig. S5 C and D). Additionally, we developed a stringent algorithm to eliminate false clusters caused by repeated localizations of same molecules due to the blinking of fluorophores (SI Appendix, Supplementary Materials and Methods) (32), and still observed a clustered RNAP distribution (SI Appendix, Fig. S5E). Note that all of the data used in this work were processed using the algorithm to eliminate repeated localizations. Therefore, we concluded that the clustered distribution of RNAP–PAmCherry reflected the property of RNAP and not the fusion fluorescent protein or imaging conditions.

Fig. 1.

Quantitative characterization of RNAP clusters in live E. coli cells. (A) Representative superresolution images of RNAP (RpoC–PAmCherry) in 3 cells under the rich medium growth condition. Cell outlines are indicated in yellow dashed lines. (Scale bar, 0.5 µm.) (B) Two-dimensional histogram of all RNAP localizations in a standard 3-μm × 1-μm cell under the rich medium growth condition. Because of the symmetry of the cell shape in both long and short axes, we calculated the absolute displacement of each RNAP localization to the center of the cell, normalized its long axis displacement to the standard cell length, and duplicated the quartile cell histogram along both the long and short axes to produce a full-sized 2D histogram of RNAP distribution. The bin size of the 2D histogram is 100 × 100 nm. The color bar indicates localization numbers used in each bin. A total of 564,615 localizations of 664 cells are used to construct the 2D histogram. (C) Identification and isolation of RNAP clusters using a tree-clustering algorithm. RNAP clusters identified in the 3 cells in A are shown as examples. (D) Two-dimensional histograms of RNAP localizations in clusters as plotted in B; a total of 39,438 localizations of 1,385 RNAP clusters are used. (E) Distribution of the number of RNAP clusters per cell (blue bars); PDF is probability density function. The mean is 2.13 ± 0.05 RNAP clusters per cell, μ ± SE, n = 664 cells. (F) Distribution of the fraction of clustered RNAP per cell. The mean is 0.16 ± 0.005, μ ± SE, n = 664 cells. (G) Distribution of fraction of RNAP localizations per cluster. The mean is 0.076 ± 0.001, μ ± SE, n = 1,385 clusters. (H) Distribution of the area of RNAP clusters. The mean for the radius is 129 ± 25 nm, μ ± SE, n = 1,385 clusters (assuming circularly shaped clusters). In all of the graphs (E–H), the blue curves are the experimentally measured distributions and the black curves are those calculated from simulated random distributions using the same number of RNAP localizations in the same cell volume for all of the cells. Error bars or shaded areas are SEs calculated from bootstrapping. The average value of each graph is also summarized in SI Appendix, Table S1.

To characterize RNAP clusters quantitatively, we performed a density-based threshold analysis to isolate individual RNAP clusters (Fig. 1C and SI Appendix, Supplementary Materials and Methods, and Tables S1 and S2) (33). The averaged cellular distribution of RNAP localizations inside clusters also showed a similar, nucleoid-like pattern (Fig. 1D), but was more toward the center of the nucleoid compared to that of all RNAP localizations (Fig. 1B). In cells under faster growth conditions (37 °C LB or EZRDM), RNAP clusters were even more inwardly distributed along the short axis of the cell (SI Appendix, Fig. S3 B–D). In contrast, when compared to the clusters of free PAmCherry and HU, RNAP clusters were located closer to the periphery of the cell (SI Appendix, Fig. S6). On average, we detected ∼2 to 3 dense RNAP clusters per cell (Fig. 1E). These clusters contained ∼16% of total detected cellular RNAP–PAmCherry molecules (Fig. 1F), corresponding to ∼350 RNAP molecules per cluster, given an average of ∼5,000 molecules of RNAP per cell (Fig. 1G and SI Appendix, Supplementary Materials and Methods) (34, 35). On average, these clusters occupied an area equivalent to that of a circle with a radius of ∼130 nm (Fig. 1H). These cluster properties were significantly different from that of free PAmCherry molecules, what would be expected from a completely random distribution pattern (Fig. 1 E–H, black curves, and SI Appendix, Figs. S5 and S6 and Table S2) or cluster properties from that of the nonspecific DNA binding HU (SI Appendix, Figs. S5 A and B and S6 and Table S2), therefore confirming the clustering of RNAP in E. coli cells under the rich medium growth condition.

RNAP Clusters Colocalize with Nascent rRNA Synthesis Sites in Cells under the Rich Medium Growth Condition.

Previous studies proposed that RNAP clusters are actively engaged in rRNA transcription but no direct evidence have been provided (8). To examine this hypothesis, we probed the colocalization of RNAP clusters with nascent or newly synthesized rRNAs. We used a highly efficient FISH probe conjugated with Alexa Fluor 488 or 647 (SI Appendix, Figs. S7 and S8A) to target the 5′ leader region of the 16S precursor rRNA (pre-rRNA) (Fig. 2A), which is absent from the mature 16S RNA inside the ribosome (36). The 5′ leader degraded rapidly with a half-life of ∼130 s after being processed (SI Appendix, Fig. S8B); therefore, the FISH probe only identifies newly synthesized pre-rRNA. Using 2-color superresolution imaging of pre-rRNA and RNAP–PAmCherry in fixed cells, we observed clear spot-like foci of pre-rRNA fluorescence signal with a spatial resolution of ∼40 nm (Fig. 2 B, Center, and SI Appendix, Fig. S2). On average we detected ∼4 pre-rRNA clusters per cell containing more than 60% of total cellular rRNA localizations (Fig. 2 C–F and SI Appendix, Tables S2 and S3). Furthermore, we observed qualitatively that RNAP–PAmCherry clusters predominately coincided with these pre-rRNA clusters (Fig. 2 B, Right). To quantify the extent of spatial colocalization, we calculated the fraction of RNAP clusters that had any molecule localizing to any molecule of a pre-rRNA cluster within a radius ranging from 50 to 250 nm (Fig. 2G, blue curve, and SI Appendix, Supplementary Material and Methods and Fig. S9), then compared the colocalization curve with the expected background level calculated by randomizing the positions of RNAP clusters in the same cells (Fig. 2G, black curve). We found that at all radii there were substantially higher fractions of RNAP clusters colocalizing with pre-rRNA clusters than that of the background level. For example, 83% ± 2% RNAP clusters (n = 404 RNAP clusters) had at least 1 pre-rRNA cluster within a radius of 50 nm (SI Appendix, Table S4). Given the significantly improved spatial resolution afforded by superresolution imaging, the high colocalization levels we observed between RNAP clusters and pre-rRNA clusters at a resolution limit (∼40 to 60 nm) comparable to the molecular size of RNAP molecules (∼20 nm) (37) suggested that the majority of RNAP clusters were active in rRNA synthesis under the rich medium growth condition.

Fig. 2.

RNAP clusters colocalized with nascent pre-rRNA clusters under the rich medium growth condition. (A) Schematics of pre-rRNA detection. The dye-labeled L1 probe binds to the 5′ leader sequence of 16S rRNA that is cleaved off from mature 16S rRNA and rapidly degrades. (B, Left) Ensemble pre-rRNA FISH images of cells (outlined in yellow) under the rich medium growth condition. (Scale bar, 0.5 μm.) (Center) Representative pre-rRNA FISH superresolution images of the 2 cells. (Right) Representative 2-color superresolution images of RNAP–PAmCherry (red) and pre-rRNA FISH (green) of the 2 cells in the Center. (C) Distribution of the number of pre-rRNA clusters per cell. The mean is 3.86 ± 0.09, μ ± SE, n = 288 cells. (D) Distribution of fraction of clustered pre-rRNA localizations per cell; PDF is probability density function. The mean is 0.63 ± 0.005, μ ± SE, n = 288 cells. (E) Distribution of fraction of pre-rRNA localizations per cluster. The mean is 0.16 ± 0.004, μ ± SE, n = 1,086 pre-rRNA clusters. (F) Distribution of the area of pre-rRNA clusters. The mean for the radius is 127 ± 22 nm, μ ± SE, n = 1,086 pre-rRNA clusters. The average value of each graph is summarized in SI Appendix, Table S3. (G) The fraction of RNAP clusters colocalizing with pre-rRNA clusters at different distances from 50 to 250 nm (blue curve). The black curve is the simulated colocalization fraction of RNAP clusters with pre-rRNA clusters when the spatial distribution of RNAP clusters was randomized in the same cells, and hence represent the basal level of colocalization due to chance. The plotted colocalization fraction is corrected for detection efficiency of pre-rRNA clusters (SI Appendix, Figs. S8A and S9), and all values are summarized in SI Appendix, Table S4. In all of the graphs, the error bars or shaded areas are SEs calculated from bootstrapping.

RNAP Forms Clusters in the Absence of High Levels of rRNA Synthesis.

Previous work has proposed that rRNA synthesis is the major driving force for the formation of RNAP clusters (8, 9, 26, 38). To test this hypothesis, we treated cells with serine hydroxamate (SHX) to perturb rRNA transcription and subsequently observed the spatial organization of RNAP. SHX binds to seryl-tRNA synthetase, induces the stringent response, and inhibits rRNA synthesis (39–41). We observed a dramatic reduction in total rRNA synthesis in SHX-treated cells as expected (∼3% compared to that of untreated cells) (SI Appendix, Fig. S10), but RNAP was still significantly clustered (Fig. 3A) compared to free PAmCherry (SI Appendix, Fig. S5). The number of RNAP clusters per cell decreased (∼1.9 clusters per cell) (Fig. 3 A and B and SI Appendix, Tables S1 and S2) and their sizes reduced (∼104 nm) (Fig. 3 C and D), but they contained a similar fraction of total detected RNAP molecules compared to those in untreated cells (Fig. 3E and SI Appendix, Tables S1 and S2). The averaged cellular localizations of all RNAP molecules in SHX-treated cells also exhibited a 2-lobed distribution, although the middle cleft was less distinct compared to WT cells (Fig. 3F). Note that the nucleoid morphology and volume in SHX-treated cells was not significantly different from that of the WT cells (SI Appendix, Fig. S4 A, B, and F). These results suggested that a high level of rRNA synthesis, as that in the rich medium growth condition, was not necessary for the formation of RNAP clusters.

Fig. 3.

Characterization of RNAP clusters in live E. coli cells treated with SHX (A–F), in a rrn deletion strain (Δ6rrn, G–L), in cells with an overexpression of AsiA (M–R), and in cells treated with the global transcription inhibitor rifampicin (S–X). (A, G, M, and S) Representative superresolution images of RNAP–PAmCherry. (Scale bars, 0.5 µm.) (B, H, N, and T) Distribution of the number of RNAP clusters per cell; PDF is probability density function. (C, I, O, and U) Distribution of the fraction of clustered RNAP per cell. (D, J, P, and V) Distribution of the area of RNAP clusters. (E, K, Q, and W) Distribution of the fraction of RNAP localizations per cluster. (F, L, R, and X) Two-dimensional histogram of all RNAP localizations in a standard 3-μm × 1-μm cell (Upper), and 2D histogram of only clustered RNAP localizations in a standard 3-μm × 1-μm cell (Lower). In B–E, H–K, N–Q, and T–W the blue bars/curves are those of the WT under the rich medium growth condition for comparison, and the black curves are those calculated from simulated random distributions using the same number of localizations in the same cell volume for all of the cells under each condition. All of the mean values of these graphs are summarized in SI Appendix, Table S1. In all of the graphs (B–E, H–K, N–Q, and T–W ), the error bars or shaded areas are SEs calculated from bootstrapping.

RNAP Forms Clusters in the Presence of only 1 rrn Operon per Chromosome.

One possibility to explain the presence of a significant level of RNAP clusters in SHX-treated cells was that RNAP clusters might remain associated with multiple rrn operons that may spatially colocalize with each other (42), despite the lack of high transcription activity from these operons. To examine this possibility, we used a Δ6rrn strain, in which 6 out of 7 rrn operons (except for rrnC) were removed from the chromosome (43). This strain also contained an additional plasmid ptRNA67 (44) to provide tRNA genes in trans (45). The Δ6rrn strain grew at a slower rate than WT cells under the same rich medium growth condition (cell doubling time = 91 ± 1 min) (SI Appendix, Fig. S11), and showed a significant reduction in total rRNA synthesis (∼28% of WT cells) (SI Appendix, Fig. S10). However, the cellular distribution of RNAP and the properties of RNAP clusters in the Δ6rrn strain were remarkably similar to those of SHX-treated cells (Fig. 3 G–L and SI Appendix, Tables S1 and S2), and remained highly colocalized to residual pre-rRNA clusters (SI Appendix, Fig. S12 and Table S4). Additionally, we found that the nucleoid morphology and the total nucleoid volume of these cells were comparable to WT cells (SI Appendix, Fig. S4 C and F). These results suggested that the formation of RNAP clusters did not require a high level of rRNA synthesis activity or the presence of multiple rrn operon coding regions.

RNAP Forms Clusters in σ⁷⁰-Sequestered Cells.

Next, to probe the possibility that other non-rRNA transcription activities may contribute to the formation of RNAP clusters under the conditions of significantly reduced rRNA synthesis, we inhibited transcription from all housekeeping σ⁷⁰ promoters by overexpressing a 10-kDa bacteriophage T4 anti-σ protein AsiA (46) from an arabinose-inducible plasmid. AsiA binds to σ⁷⁰ in the RNAP holoenzyme and prevents the holoenzyme to initiate transcription from σ⁷⁰ promoters, which constitute ∼75% of total E. coli promoters (47). In these cells, we observed significantly elongated cells after a 2-h arabinose induction (SI Appendix, Fig. S13), indicating the detrimental effect of shutting down σ⁷⁰ promoters (48). However, while a smaller fraction of RNAP formed smaller clusters in AsiA-overexpressing cells compared to WT cells (Fig. 3 M–R and SI Appendix, Tables S1 and S2), these clusters were significant compared to that of free PAmCherry or random distribution (SI Appendix, Fig. S5). The cellular distribution of all RNAP localizations appeared to expand and occupy a larger nucleoid volume compared to that in other conditions (Fig. 3R), likely indicating the presence of an increased fraction of free RNAP outside of the nucleoid. These results indicated that the formation of the remaining RNAP clusters did not require σ⁷⁰ promoter activities.

RNAP Clusters Are Significantly Reduced in RIF-Treated Cells.

In all of the perturbation experiments described above, there still existed low levels of transcription activity (for example, mRNA transcription in SHX-treated cells and alternative σ factor transcription in AsiA overexpressing cells). Therefore, it is possible that RNAP might still form clusters engaged in the remaining transcription. To test this possibility, we incubated cells with a global transcription inhibitor RIF (100 μg mL⁻¹, 2 h) (SI Appendix, Fig. S10). RIF caps the RNA exit channel on RNAP before the nascent RNA chain emerges, and hence blocks transcription initiation and traps RNAP in an abortive cycle on the promoter (25). We reasoned that if the majority of RNAP clusters we observed so far were indeed complexes involved in active transcription, these complexes would eventually finish transcription and run off when transcription initiation is inhibited, leading to the disappearance of RNAP clusters. Conversely, if RNAP clusters were long-lived complexes not involved in active transcription, they would persist despite the global inhibition of transcription. In RIF-treated cells, we observed that although the clustering of RNAP–PAmCherry was not completely eliminated compared to that of free PAmCherry control or the random distribution simulation, the extent of clustering was substantially reduced compared to all other conditions (Fig. 3 S–X and SI Appendix, Tables S1 and S2). The number and size of RNAP clusters decreased significantly (Fig. 3 T, V, and W), with more cells containing fewer clustered RNAP molecules than that in WT cells (Fig. 3U). These results could suggest that most RNAP clusters were active transcription complexes under the rich medium growth condition, and that global transcription activity was a major contributor to the formation of RNAP clusters. However, we could not exclude the possibility that the observed changes were due to the expansion of the nucleoid under the condition of global transcription inhibition, upon which bound RNAP clusters dispersed (49). Supporting this latter possibility was the observation that the cellular distribution of RNAP exhibited a homogenous, single-lobed pattern without a discernible central cleft (Fig. 3X), mimicking that of the nucleoid under the same condition (SI Appendix, Fig. S4D). Because both the global transcription activity and the nucleoid structure were significantly altered in cells treated with RIF, it is necessary to investigate RNAP’s distribution under conditions where the effect of the nucleoid structure could be isolated without interfering with the transcription activity of rRNA.

Inhibition of Gyrase Activity Leads to a Redistribution of RNAP Clusters and rRNA Synthesis Sites.

As we described above, the cellular distribution of RNAP closely mimicked that of the corresponding nucleoid structure visualized using 3D SIM imaging under all of the conditions tested (SI Appendix, Fig. S4). These observations suggested that the spatial organization of these clusters might reflect that of the underlying nucleoid organization. To isolate the effect of the nucleoid structure on RNAP distribution that is independent of transcription activity of rRNA, we decided to perturb the nucleoid organization by inhibiting gyrase activity.

The E. coli chromosome is highly compact and organized at different levels from topological domains to macrodomains (50–52). These organizations likely dictate the spatial arrangement of different DNA segments, upon which RNAP may preferentially bind and form clusters. Negative supercoiling is a major chromosome compacting factor, and it is only introduced by gyrase, a type II topoisomerase in E. coli (53). We thereby examined specifically the effect of DNA supercoiling on the spatial organization of RNAP.

We treated WT cells with a gyrase inhibitor novobiocin (NOV, 300 μg mL⁻¹ for 30 min) and performed 2-color superresolution imaging using pre-rRNA FISH and RNAP–PAmCherry (Fig. 4 A–D). NOV inhibits gyrase activity by abolishing ATP binding to the ATPase domain in the GyrB subunit (54, 55). We found that the average rRNA synthesis activity per cell was not significantly affected by the inhibitor, as the total intensity of pre-rRNA FISH signal remained similar to untreated cells (SI Appendix, Fig. S14A), even when high inhibitor concentrations and long-time treatment were used (SI Appendix, Fig. S14B). We further verified that the persistent rRNA synthesis during gyrase inhibition was not due to altered rRNA degradation kinetics in the presence of the inhibitor (SI Appendix, Fig. S14C). The minimal effect of gyrase inhibition on rRNA synthesis has been observed previously (56, 57), although conflicting results have been reported as well (58, 59). Interestingly, while the total pre-rRNA FISH signals remained unchanged under our experimental condition, we observed a greater number (on average 5.5 per cell compared to 3.9 in untreated cells) of less dense (52% of total cellular localizations compared to 63% in untreated cells) pre-rRNA clusters (Fig. 4 E–H and SI Appendix, Tables S2 and S3). RNAP clusters persisted in these gyrase-inhibited cells as well (Fig. 4I), remained highly colocalized with pre-rRNA clusters (Fig. 4M), but contained fewer RNAP molecules (Fig. 4 I–L and SI Appendix, Table S2). Interestingly, the cellular distributions of pre-rRNA and RNAP clusters in gyrase-inhibited cells exhibited a similarly, spatially dispersed pattern compared to that of untreated cells (Fig. 4 C and D), and the average positioning of RNAP clusters and pre-rRNA clusters moved ∼80 nm radially toward the nucleoid periphery (SI Appendix, Fig. S15). Note that the cellular distribution RNAP again mimicked that of the expanded nucleoid under the same condition (SI Appendix, Fig. S4 E and F). A different gyrase inhibitor, nalidixic acid (50 μg mL⁻¹ for 10 min), which acts on the GyrA subunit by stabilizing the DNA-cleaved complex, produced a similar effect on the cellular distribuctions of pre-rRNA and RNAP (SI Appendix, Fig. S16 A–J, and O) but less on the properties of RNAP clusters (SI Appendix, Fig. S16 K–N), likely due to the short time (10 min) used to treat cells in order to avoid double-stranded DNA breaks. The significant redistribution of RNAP and pre-rRNA clusters in the presence of altered nucleoid structure but unchanged rRNA transcription activity suggested that the characteristics and organization of the nucleoid, here in particular compaction by negative supercoiling, play a large role in the spatial distribution of RNAP clusters.

Fig. 4.

Inhibition of gyrase activity led to dispersed distributions of RNAP and pre-rRNA. (A) Ensemble fluorescence of pre-rRNA FISH signal in fixed, NOV-treated cells. Individual cells are outlined in yellow. (B) Representative superresolution images of pre-rRNA distribution in fixed, NOV-treated cells. (C) Two-dimensional histograms of all pre-rRNA localizations in a standard 3-μm × 1-μm fixed cell under the rich medium growth condition (Upper) and in cells treated with NOV (Lower). (D) Two-dimensional histograms of all RNAP localizations in a standard 3-μm × 1-μm fixed cell under the rich medium growth condition (Upper) and in cells treated with NOV (Lower). (E–L) Distributions of properties of pre-rRNA (E–H) and RNAP clusters (I–L) in NOV-treated cells; PDF is probability density function. (E and I) Distribution of the number of clusters per cell. (F and J) Distribution of fraction of clustered pre-rRNA (F) or RNAP (J) per cell. (G and K) Distribution of fraction of pre-rRNA (G) or RNAP (K) localizations per clusters. (H and L) Areas of clusters. (M) Fraction of RNAP clusters colocalizing with pre-rRNA clusters in NOV-treated cells. In all plots the WT rich medium growth conditions are plotted in blue for comparison; NOV-treated conditions are in dark red, and the background colocalization levels using simulated images are in black. All error bars or shaded areas are SE calculated using bootstrapping. All of the mean values of these graphs are summarized in SI Appendix, Tables S3 and S4.

Discussion

In this study, we investigated the prokaryotic transcription factory model in detail using a combination of quantitative superresolution imaging and perturbation analyses. Below we compare our results with previous work and discuss the implications of this work.

Spatial Organization of RNAP.

We observed that in E. coli cells grown in rich medium, RNAP was spatially organized into large, dense clusters occupying the same area as the nucleoid. These clusters had a radius of ∼130 nm (Fig. 1H), and could be made up of collections of multiple small RNAP clusters observed in a previous study (24). Given a total cellular level of RNAP at ∼5,000 molecules per cell (34, 35, 60) under a similar growth condition, and that majority (∼90%) of cellular RNAP remain bound on DNA (61, 62), we estimated that each RNAP cluster contained ∼350 molecules. The cellular distribution of all RNAP molecules exhibited a 2-lobed pattern with a clear cleft in the middle (Fig. 1B), mimicking that of 2 replicated and segregated nucleoids (SI Appendix, Fig. S4A). Compared to the distribution of all RNAP molecules, RNAP clusters were tighter and more concentrated toward the center of the 2 lobes with an average distance of ∼120 nm radially from the center of the cell (SI Appendix, Fig. S17). This observation is consistent with previous superresolution studies of the spatially separated ribosomes and RNAP in E. coli (22): ribosome at the nucleoid boundary but RNAP predominately at the center. We have shown that with respect to HU clusters under the same growth condition, RNAP clusters were positioned more peripherial within the nucleoid territory (SI Appendix, Fig. S6), in agreement with a previous study (24). In addition, under faster growth conditions (37 °C LB and 37 °C EZRDM), RNAP clusters also shifted their distribution even more toward the center of the cell along the short axis (SI Appendix, Fig. S3D), indicating that their distribution was sensitive to growth rate. In previous studies, the apparent spatial segregation between ribosome and RNAP has been used to question the coupling between transcription and translation in bacterial cells (22, 61). It is possible that periphery-localized small RNAP clusters, which may be undetectable in our stringent distance-based clustering algorithm, contained RNAP molecules actively engaged in mRNA transcription that is coupled to translation, while the larger, nucleoid center-localized RNAP clusters we observed were responsible for rRNA synthesis, which does not require translation.

Spatial Organization of pre-rRNA Clusters.

We used a pre-RNA FISH probe targeting the leader sequence of the 16S rRNA to detect rRNA transcription activity. Because newly synthesized pre-rRNAs are processed before they are incorporated into ribosomes, the pre-rRNA probe marks new rRNA synthesis sites. Compared to RNAP, pre-rRNAs formed similarly sized (∼130 nm in radius) but denser (containing > 60% of detected cellular pre-rRNAs) clusters. The overall cellular distribution of pre-rRNAs also exhibited a 2-lobed pattern, but relatively more concentrated at the center of the nucleoid compared to RNAP clusters in fixed cells (SI Appendix, Fig. S15). On average we observed ∼4 pre-rRNA clusters per cell or 2 per chromosome (Fig. 2C). Because 4 of the 7 rrn operons reside close to the replication origin oriC on the chromosome, it is possible that the 2 pre-rRNA clusters reflected 2 copies of replicated oriC for each chromosome, and thus most cells contained 2 copies of the chromosome with 4 copies of the oriC region, consistent with previous observations when the Ori region was labeled (63). Alternatively, it is also possible that the 2 pre-rRNA clusters reflected 2 groups of transcribing rrn operons on the same copy of the chromosome that are spatially distinguishable from each other under our resolutions. A recent study found that, except for rrnC, all of the rrn operons are within a spatial distance of ∼80 to 130 nm (median of distributions) to each other (42), but it remains unknown whether these rrn operons indeed co-occupy the same area in the nucleoid and whether they could be accommodated in 1 pre-rRNA cluster (radius of ∼130 nm). Because of the nearly identical pre-rRNA sequences of all of the rrn operons, we could not design a probe with high confidence to distinguish the transcription activity of individual rrn operons, and hence further investigation is required to address whether each pre-rRNA cluster reflects the transcription activity from an individual or a collection of rrn operons.

The Contribution of Transcription Activity to the Spatial Organization of RNAP.

We observed that under the rich medium growth condition, RNAP clusters highly colocalized with pre-rRNA FISH probe signals. This suggests that under the rich medium growth condition the majority of RNAP clusters were actively transcribing rRNAs. These results are the most direct demonstration of the activity of RNAP clusters for the rich medium growth condition. Previous studies have assumed but not validated that RNAP clusters are transcribing rRNAs in fast-growing cells (8, 23). Surprisingly, when we abolished rRNA transcription using SHX, reduced rRNA transcription to approximately one-third of WT cells using a mutant strain lacking 6 of the 7 rrn operons (Δ6rrn), or inhibited all σ⁷⁰-promoter activities by overexpressing AsiA, we found that there were still significant levels of RNAP clustering (Fig. 3 A–R). hence, the drastically different transcription activities but similar organizations of RNAP under the 3 different conditions suggested that rRNA transcription activity may not be the main driving force for the organization of RNAP clusters under our experimental conditions. Furthermore, because there was only 1 copy of the rrnC operon in the Δ6rrn strain, it is unlikely that multiple rrn operons were required for the formation of RNAP clusters, as previously proposed (5). The greatest perturbation to RNAP clusters was seen under the condition where global transcription initiation was inhibited by RIF (Fig. 3 S–X), although under this condition the nucleoid structure was also altered, and RNAP still exhibited a clustered distribution different from the random distribution. Note that our results did not automatically imply that these persisting RNAP clusters bind to the same chromosomal DNA sequences, have the same molecular compositions, or localize to the same cellular positions. Further biochemical investigations of these properties will be needed.

The Contribution of Nucleoid Structure to the Spatial Organization of RNAP.

In all of our experiments we observed that the cellular distribution of RNAP mimicked the shape of the underlying nucleoid irrespective of rRNA synthesis activity (SI Appendix, Fig. S4). We thereby investigated the contribution of the nucleoid structure on the spatial distribution of RNAP by inhibiting gyrase. Gyrase is the only type II topoisomerase in E. coli that introduces negative supercoiling into the chromosome, which is the major force in compacting the nucleoid (53, 64). In gyrase-inhibited cells via both nalidixic acid and NOV, we observed similar levels of pre-rRNA signal (SI Appendix, Fig. S14A) but saw a significant shift in the cellular positioning of RNAP and pre-rRNA clusters, both of which expanded ∼120 nm radially toward the nucleoid periphery (Fig. 4 C and D and SI Appendix, Figs. S15 and S16 E and F). Previous studies have documented that gyrase inhibition affects the expression of more than 300 mRNA genes that are sensitive to supercoiling (65, 66), but produced mixed results on the effect of rRNA transcription (57–59). The pre-rRNA FISH probe detects the 5′ leader sequence of 16S rRNA, hence the unchanged FISH signal in gyrase-inhibited cells only reflected unaltered rRNA transcription initiation. It is possible that rRNA elongation was inhibited due to accumulated positive supercoiling ahead of transcription in the absence of gyrase. In such a case, we should expect that after a long inhibition time, the rRNA transcription initiation rate would gradually decrease as accumulated positive supercoiling eventually inhibits transcription initiation, which was demonstrated previously for the production of mRNA in vitro (67). However, we observed similar pre-rRNA signal even after we incubated cells with high concentrations of NOV (300 to 1,200 μg mL⁻¹) for extended periods (90 to 150 min) (SI Appendix, Fig. S14B), demonstrating that the total rRNA transcription activity in these cells was minimally affected. Therefore, these experiments most likely suggest that the nucleoid structure contributes significantly to the spatial organization of RNAP. It is possible that a relaxed chromosome repositioned different DNA segments (upon which RNAP clusters form) to occupy a larger cell volume, as suggested by SIM imaging of the gyrase-inhibited nucleoid (SI Appendix, Fig. S4 E and F). Clearly, further studies, such as genetic and biochemical perturbations of nucleoid-organization factors, are required to investigate the effect of nucleoid structure on the spatial organization of RNAP.

In summary, our study demonstrates that there was an rRNA transcription activity-independent spatial organization of RNAP in E. coli and that the underlying nucleoid structure played important roles in organizing RNAP clusters. Further experiments investigating the molecular nature and function of RNAP clusters are required. In particular, we do not know whether these RNAP clusters we observed are associated with specific chromosomal DNA sequences, or whether they are self-promoting oligomeric complexes similar to liquid droplets observed in eukaryotic cells, which are mediated by multivalent interactions among proteins and nucleic acids (68, 69). Intriguingly, it has been shown that a small regulatory noncoding RNA 6S can sequester the σ⁷⁰ holoenzyme; these RNAP–RNA complexes may also contribute to the clusters we observe under conditions where transcription activity from σ⁷⁰ promoters is low, although currently there has been no report of a clustered 6S RNA distribution (70, 71). Furthermore, we do not know the biological significance of RNAP clusters. In eukaryotic cells, it was suggested that RNAP clusters might represent preformed transcription complexes that are “poised” ready for rapid transcription induction (12–15). In bacterial cells, such a role has not been demonstrated, but studies have shown that there are typically higher levels of RNAP association at promoter and promoter-like sequences compared to within coding sequences (16–21). Perhaps looking at the colocalization of RNAP with important transcription regulators [NusA (72, 73), NusB (74, 75), NusE (76), NusG (16, 77), SuhB (78), and so forth] that interact with RNAP under different conditions would help elucidate the molecular make-up and functional significance of RNAP clusters. Regardless, further investigations into the spatial organization of RNAP in small bacterial cells will certainly bring in new knowledge complementing in vitro biochemical and in vivo genetic studies of prokaryotic transcription.

Materials and Methods

All bacterial strains, plasmids, and cell growth conditions used in this study are described in SI Appendix, Supplementary Materials and Methods. Cell imaging procedures, data analysis methods used for all microscopy experiments, details on specific reagents, experimental protocols, and microscopy set-up can be found in SI Appendix, Supplementary Materials and Methods. All reported errors are SEMs unless stated otherwise.