Live-cell superresolution microscopy reveals the organization of RNA polymerase in the bacterial nucleoid

Significance

Transcription is one of the most fundamental processes for life. In eukaryotic cells, transcriptional activity is regulated to a large degree by chromosome packaging. In bacteria, despite the absence of a nuclear envelope and many of the DNA-packaging proteins of eukaryotes, the chromosome is still highly condensed into a structured object, the nucleoid. The spatial organization of transcription within the nucleoid and the effect of transcription on DNA organization remain poorly understood. In this work, we characterize how RNA polymerase accesses transcription sites on DNA, and show that active transcription can cause spatial reorganization of the nucleoid, with movement of gene loci out of the bulk of DNA as levels of transcription increase.

Abstract

Despite the fundamental importance of transcription, a comprehensive analysis of RNA polymerase (RNAP) behavior and its role in the nucleoid organization in vivo is lacking. Here, we used superresolution microscopy to study the localization and dynamics of the transcription machinery and DNA in live bacterial cells, at both the single-molecule and the population level. We used photoactivated single-molecule tracking to discriminate between mobile RNAPs and RNAPs specifically bound to DNA, either on promoters or transcribed genes. Mobile RNAPs can explore the whole nucleoid while searching for promoters, and spend 85% of their search time in nonspecific interactions with DNA. On the other hand, the distribution of specifically bound RNAPs shows that low levels of transcription can occur throughout the nucleoid. Further, clustering analysis and 3D structured illumination microscopy (SIM) show that dense clusters of transcribing RNAPs form almost exclusively at the nucleoid periphery. Treatment with rifampicin shows that active transcription is necessary for maintaining this spatial organization. In faster growth conditions, the fraction of transcribing RNAPs increases, as well as their clustering. Under these conditions, we observed dramatic phase separation between the densest clusters of RNAPs and the densest regions of the nucleoid. These findings show that transcription can cause spatial reorganization of the nucleoid, with movement of gene loci out of the bulk of DNA as levels of transcription increase. This work provides a global view of the organization of RNA polymerase and transcription in living cells.

Cellular functions in bacteria are not compartmentalized as they are in eukaryotes, yet they are still highly organized, with specific subcellular regions associated with specific machineries. In Escherichia coli, many of these machineries operate on the bacterial nucleoid, a highly structured and dynamic object made primarily by the 4.6-Mbp circular chromosome that is compacted between 10³- and 10⁴-fold compared with the equivalent linear DNA and segregated from the cytoplasm (1). Transcription is one of the most abundant processes occurring on the nucleoid DNA; transcription is driven by 1,500–5,000 molecules of RNA polymerase (RNAP) per cell (2–4) and plays a crucial role in maintaining both global and local DNA organization. For example, blocking cellular transcription with the antibiotic rifampicin causes loss of global organization reflected by nucleoid expansion (5, 6). Transcription has also been shown to modify the local DNA topology by unwinding the DNA double helix (7). Despite this, little is known about the spatial distribution of active RNAPs in living bacterial cells and how they affect the organization of DNA and the nucleoid.

The nucleoid structure depends on growth conditions. Under slow growth conditions, where each cell contains between one and two chromosomes, the nucleoid lacks observable structures and appears diffuse, occupying most of the cellular volume. By contrast, under fast growth conditions, where the cells contain between four and eight chromosome equivalents, the nucleoid is highly structured and displays dramatic variation in local DNA density (8). Growth conditions also influence transcriptional activity on different genes; for example, the overall rate of synthesis of ribosomal RNA increases ∼40-fold in fast growth compared with slow growth conditions (9).This change in the level of expression is reflected in large changes in the spatial distribution of RNAP: Under slow growth conditions, RNAP appears to be fairly homogeneously distributed over the diffuse nucleoid (10), whereas under fast growth conditions, dense clusters of RNAPs emerge (4, 10). These dense clusters have been likened to transcription factories in eukaryotic cells, where a single site contains multiple RNAPs active on different genes (11, 12); however, they have only been reported in chemically fixed cells, and little is known about their presence, dynamics, or behavior in live E. coli.

Transcription and nucleoid structure are also interconnected through the coupling of transcription and translation. In bacteria, ribosomes can bind to nascent mRNA as soon as the ribosome binding site emerges from the transcribing RNAP (13). This raises an intriguing puzzle, because ribosomes and DNA are spatially separated in E. coli (3, 14, 15), whereas RNAP is strongly associated with DNA (3, 10). To reconcile this with cotranscriptional translation, it has been hypothesized that much of cellular transcription must happen at the surface of the nucleoid, and close to the pool of ribosomes (16). However, evidence for this remains weak, because studies under moderate growth conditions did not reveal any RNAP enrichment at the nucleoid periphery (3).

Here, we use a combination of live-cell superresolution microscopy techniques to provide a comprehensive analysis of the behavior of RNAP at both single-molecule and population levels in individual cells. By using photoactivation localization microscopy (PALM) (17) combined with single-particle tracking (18, 19) of individual RNAP molecules in live cells, we sort them into DNA-bound or mobile subpopulations (20). We show that mobile RNAPs are uniformly distributed across the nucleoid, suggesting that all DNA is probed through random nonspecific interactions during the promoter search. The distribution of specifically bound RNAPs shows that low levels of transcription can occur throughout the nucleoid. However, clustering analysis and 3D structured illumination microscopy (3D SIM) show that specifically bound RNAPs are more clustered than the mobile population, and the denser clusters form preferentially at the nucleoid surface, indicating that heavily transcribed genes tend to move out of the bulk of nucleoid DNA. Imaging cells growing in both rich and minimal media showed that clustering increases in rich media conditions and that segregation between RNAP and DNA also increased.

Single-Molecule Tracking Discriminates DNA-Bound and Mobile RNAPs

RNAP can be either specifically bound to DNA (while interacting with promoter regions or with transcribed genes during transcription elongation) or can diffuse through the nucleoid searching for promoters to initiate transcription. Because the movement of DNA loci is extremely slow compared with the diffusion on cytoplasmic proteins (20, 21), and the time to open a promoter and transcribe a gene [>20–100 s (9)] is at least 500-fold longer than nonspecific DNA interactions [30 ms (22)], we reasoned that individual RNAPs could be sorted into specifically bound molecules or mobile (diffusing and binding only transiently) based on their intracellular mobility (20, 23). To track RNAP molecules, we used an endogenous fusion of photoactivatable fluorescent protein PAmCherry (24) with the beta′ subunit of RNAP (4). We imaged molecules in live cells by photoactivating and localizing fluorophores (17), and joining localizations over multiple frames to obtain trajectories of individual molecules (Fig. 1A) (18). By photoactivating and tracking all available molecules in cells grown in minimal media, we estimated the mean RNAP copy number to be 2,710 ± 700 per cell, which increased with cell length as expected (Fig. S1). We note that this may be an underestimate of the true copy number of RNAP due to immature or nonphotoactivatable PAmCherry molecules (25).

Fig. 1.

Single-molecule tracking allows mobility-based categorizing of individual RNAPs as DNA-bound or mobile. (A) Example trajectories of individual RNAPs with specifically bound molecules colored red and mobile molecules colored blue. (Scale bar, 1 µm.) (B) The distribution of apparent diffusion coefficients, D*, for 69,900 RNAP molecules in live cells can be fitted with two diffusing species (Inset), giving a ratio of 48% bound to 52% diffusing. Using these values allows a D* threshold to be determined to categorize bound (red) and mobile (blue) molecules. (C) Representative example cells show the spatial distribution of categorized RNAP trajectories colored according to their D* value. (D) The distribution of D* values for 39800 RNAP molecules in cells after incubation with rifampicin. (E) Example cells after rifampicin treatment show fewer bound molecules. Mobile RNAPs explore most of the cell volume due to the expansion of the nucleoid.

Fig. S1.

RNAP copy number. The mean RNAP copy number per cell is 2,710 ± 700. Copy number calculated from taking long PALM movies (typically 40,000 frames) with low photoactivation levels until all proteins have photobleached.

To measure RNAP mobility, we calculated an apparent diffusion coefficient (D*) from the mean squared displacement (MSD) of trajectories of individual molecules (20). As expected, the D* distribution for the entire population of RNAP molecules (Fig. 1B) does not fit to an analytical expression for a single diffusing species (Fig. S2A). To establish the apparent motion of molecules specifically bound to DNA, we used E. coli DNA polymerase I (Pol1) as a control that shows distinct D* populations for DNA-bound and mobile molecules (SI Materials and Methods and Fig. S2B) (20). We then fitted the RNAP D* distribution with a two-species model with a DNA-bound population (constrained using the DNA-bound population in the Pol1 control) and a second, unconstrained D* species that corresponds to the population of mobile RNAP molecules. This analysis showed that ∼48% of RNAPs were bound and ∼52% were mobile (Fig. 1B, Inset); this result is in agreement with previous estimates from both fluorescence recovery after photobleaching (26) and single-molecule tracking (23). We then determined a D* threshold (0.16 µm²/s) which preserves the bound-to-mobile ratio, and allows sorting of individual trajectories as corresponding to bound or mobile RNAP molecules (SI Materials and Methods). Plotting the spatial distribution of the sorted molecules results in a 2D cellular map that shows the intracellular location of bound and mobile RNAPs in live cells (Fig. 1C).

Fig. S2.

Distributions of apparent diffusion coefficients, D*, for RNAP and control protein DNA polymerase 1 (Pol1). (A) The distribution of RNAP D* values fits poorly to a single diffusing species. (B) Pol1 was used a control to determine the D* value of DNA-bound molecules. When DNA methylation damage is induced by MMS (see ref. 20), the fraction of bound molecules increases dramatically and is clearly resolvable from the pool of mobile molecules (20). Fitting this distribution with two diffusing species allows us to determine the D* value of specifically bound molecules, with D*_bound = 0.11 µm²/s. (C) Using this D*_bound value to constrain one species and allowing a second unconstrained D* species fits well to the RNAP distribution, giving two populations of 48% bound and 52% mobile (D*_mobile = 0.36 µm²/s). (D) Fitting to the RNAP distribution after rifampicin incubation gives a threefold reduction in the fraction of bound molecules to 16%. The D* of mobile molecules increases 75% from D*_mobile = 0.357 µm²/s (95% confidence interval: 0.338–0.378 µm²/s) to D*_mobile = 0.636 µm²/s (0.609−0.658 µm²/s). (E) Wild-type pol1 had 3.6% bound molecules. (F) After rifampicin incubation, the bound fraction of Pol1 shows negligable change (3.6–3.5%), indicating that the reduction in bound population seen in RNAP is due to blocked transcription. The D* for mobile population of Pol1 also increases after rifampicin incubation [40% increase from D*_mobile 1.05 (1.02–1.07) µm²/s to D*_mobile = 1.37 (1.40–1.35) µm²/s], indicating that the effect on the mobile population is due to the global decompaction of the nucleoid caused by rifampicin.

To check whether most DNA-bound RNAPs are indeed actively transcribing, we treated the cells with the antibiotic rifampicin, which binds to RNAP and blocks transcription beyond a 3-nt RNA but does not affect promoter binding, open complex formation, or transcription by RNAPs already in transcription elongation (27). Incubation with rifampicin for 30 min is thus expected to cause most RNAPs to become mobile, with only promoter-bound RNAPs remaining bound to DNA, because any elongating RNAP will complete transcription and dissociate from the DNA. Consistent with these expectations, the rifampicin treatment led to a clear increase (from 52% to 84%) in the fraction of mobile RNAPs (Fig. 1 D and E). In contrast, the rifampicin treatment did not change the fraction of mobile DNA polymerase I (Pol1), confirming that the effect seen for RNAP is specific to blocking active transcription (Fig. S2 E and F). Given our estimates for the average RNAP copy number per cell, the bound fraction in the rifampicin-treated cells corresponds to ∼430 bound RNAPs per cell. This estimate is in good agreement with results from ChIP experiments, which showed that RNAP remains associated with ∼530 promoters per chromosome after rifampicin incubation (28).

The rifampicin treatment also increased the apparent diffusion of mobile RNAPs (D* increases from 0.357 µm²/s (95% confidence interval: 0.338–0.378 µm²/s) to 0.636 µm²/s (0.609−0.658 µm²/s) (Fig. S2 C and D), likely the result of global chromosome decompaction triggered by rifampicin treatment (Fig. S3) (6). Chromosome decompaction leads to less densely packed strands of DNA, and we therefore expect mobile RNAPs moving between more-distal DNA strands to show faster overall apparent diffusion. In agreement with this interpretation, our control protein Pol1 also showed increased mobility of its mobile population upon rifampicin treatment [D* increases from 1.05 (1.02–1.07) µm²/s to 1.37 (1.40–1.35) µm²/s; Fig. S2F].

Fig. S3.

Timelapse images of chromosome compaction and decompaction. Time lapse images showing the nucleoid stained by HU-mCherry (yellow) relative to the cell membrane stained with FM4-64 (red). At t = 0, cells are added to an LB agarose slide containing either no antibiotics (Top), 50 µg/mL rifampicin (Center), or 100 µg/mL of chloramphenicol (Bottom). Cells were imaged every 10 min for a total of 40 min.

SI Materials and Methods

Bacterial Strains.

PAmCherry and GFP fusion strains of RNAP, expressed from their native promoter in E. coli MG1655, were previously described in ref. 4. PAmCherry fusion strains of HUβ and H-NS, expressed from their native promoter in E. coli MG1655, were generated by replacing the genes by their PAmCherry fusion versions. Genes coding for PAmCherry1 and Kanamycin resistance cassette were amplified from the plasmid pROD85 (carrying an 11aa linker preceding PAmCherry and followed by frt-flanked kan), using primers with 40–50 nt overhangs homologous to the insertion site at the chromosome. DNA fragments were electroporated in AB1157 cells overexpressing λ-Red proteins from pKD46 (47). Correct insertion of the fragment into the chromosome was evaluated by PCR using primers flanking the insertion site. Fusions were moved to MG1655 background by P1 transduction. The growth rates in LB and M9 glycerol medium were the same for wild-type and fusion strains. DNA polymerase 1-PAmCherry in AB1157 background (described in ref. 20) was moved into MG1655 for this work by P1 transduction. Unconjugated PAmCherry was expressed from a pBad plasmid in wild-type MG1655 cells. RNAP−PAmCherry DnaC(ts) cells were constructed by P1 transduction of DnaC(ts) and Tetracycline resistance into RNAP−PAmCherry cells. Colonies were selected with both Tet resistance and temperature sensitivity when plated and grown overnight at 42 °C.

Cell Preparation.

Strains were streaked onto LB plates with 25 μg/mL ampicillin or 50 μg/mL kanamycin. Single colonies were inoculated into LB and grown at 37 °C for 4 h, then diluted into M9 or rich defined medium (EZRDM; Teknova) supplemented with 0.2% glucose and grown overnight at 37 °C. Overnight cultures were diluted and grown for >2 h at 37 °C to early exponential phase (OD < 0.2). Cells were centrifuged and immobilized on 1% low-fluorescence agarose (Biorad) pads. Glass coverslips (#1.5 thickness) were heated to 500 °C in a furnace for 1 h to remove any fluorescent background particles. Where indicated, cells were incubated with 50 μg/mL rifampicin or 100 μg/mL of chloramphenicol for 30 min, or 5 μg/mL norfloxacin for 10 min. For nucleoid imaging performed on the PALM microscope, where 405-nm excitation is reserved for photoactivation, SYTOX green was used at 500 nM (in live cells, following ref. 29). For nucleoid imaging using SIM cells were incubated with 4 μg/mL of 2 4’,6-diamidino-2-phenylindole (DAPI).

Microscopy.

Live-cell single-molecule-tracking PALM, rapid high-density PALM, and RNAP−GFP cluster tracking experiments were performed on a custom-built total internal reflection fluorescence microscope built around the Rapid Automated Modular Microscope System (ASI Imaging). Photoactivatable mCherry activation was controlled by a 405-nm laser and excitation with 561 nm. GFP excitation was provided by a 488-nm laser. All lasers provided by a multilaser engine (iChrome MLE; Toptica). At the fiber output, the laser beams were collimated and focused (100× oil immersion objective, NA 1.4; Olympus) onto the sample under an angle allowing for highly inclined thin illumination (48). Fluorescence emission was filtered by a dichroic mirror and notch filter (ZT405/488/561rpc and ZET405/488/561NF; Chroma). PAmCherry emission was projected onto an EMCCD camera (iXon Ultra, 512 × 512 pixels; Andor). The pixel size was 96 nm. Brightfield cell images were recorded with an LED source and condenser (ASI Imaging). Sample position and focus were controlled with a motorized piezo stage, a z-motor objective mount, and autofocus system (MS-2000, PZ-2000FT, CRISP; ASI Imaging).

The 3D SIM imaging was performed as in ref. 45, on a DeltaVision OMX V3 (Applied Precision/GE Healthcare) equipped with a Blaze SIM module, a 60×/1.42 oil UPlanSApo objective (Olympus), 405-nm and 488-nm diode lasers, and three sCMOS cameras (PCO). The 3D two-color stacks of RNAP–GFP and DAPI-stained DNA were obtained using sequential acquisition of 512 × 512 pixels image stack with 8–12 125-nm z sections. Each z section results from striped illumination pattern rotated to the three angles (−60°, 0°, +60°) and shifted in five phase steps. Acquisition settings were as follows: RNAP–GFP, 7−10 ms exposure with 488-nm laser (100% transmission) and DAPI, 20-ms exposure with 405-nm laser (100% transmission). The 3D SIM raw data were computationally reconstructed with SoftWoRx 6.0 (Applied Precision) using a Wiener filter setting of 0.002 and channel specifically measured optical transfer functions to generate a superresolution 3D image stacks. Images from the different channels were aligned using parameters obtained from calibration measurements with 0.12-µm-diameter TetraSpeck beads (Life Technologies). Imaris analysis software (Bitplane) was used to generate 3D rendering of fluorescent signal and volume surfacing corresponding to Movies S1−S4 and S6, and the 3D wireframe/surface renderings in Fig. 5.

Fig. 5.

Organization of transcription in rich media. (A) The fractions of bound RNAPs determined per cell for different media. In rich media (531 cells), a significantly larger fraction of RNAP is specifically bound compared with minimal media (883 cells). Controls after chemical fixation (322 cells) and after rifampicin incubation (204 cells) show concomitant increases and decreases in bound fraction, respectively. (B) Localizing dense RNAP foci in live cells (Inset) allows them to be tracked in time-lapse experiments (Movie S5). MSDs of RNAP foci (red) and labeled DNA loci (blue). (C) Representative cells imaged with fast acquisition PALM, showing all localizations (Top), and clustered RNAPs, with number of molecules in clusters indicated (Bottom). (D) Histogram of the cluster sizes (in number of RNAPs) for 218 cells in rich media and 298 cells in minimal media. (E) Example fields of view showing the DAPI-stained nucleoids of live cells imaged with 3D SIM. Surface contour plots highlight the nucleoid structure. (F) The 3D surface renderings of RNAP−GFP (red) and DNA (DAPI, blue) distributions in example cells grown in minimal (Left) and rich (Right) media.

PALM Localization and Tracking.

PALM data for single-molecule-tracking analysis were localized using custom-written MATLAB software (MathWorks). Fluorophore images were identified for localization by band-pass filtering and applying an intensity threshold to each frame of the movie. Candidate positions were used as initial guesses in a 2D elliptical Gaussian fit for high-precision localization. Free fit parameters were x position, y position, x width, y width, elliptical rotation angle, intensity, background. Positions were linked to a track if they appeared in consecutive frames within a window of 5 pixels (0.48 μm). In rare cases when multiple localizations fell within the tracking radius, tracks were linked such that the sum of step distances was minimized. We used a memory parameter of 1 frame to allow for transient disappearance of the fluorophore image within a track due to blinking or missed localization. Rapid high-density PALM imaging localization was performed using DAOStorm (38).

Measuring the Diffusion of RNAP.

An apparent diffusion coefficient, D*, was calculated from the mean-squared displacement (MSD) for each track with a minimum of four steps using D* = MSD/(4Δt). Shorter tracks were discarded for this analysis because of the higher uncertainty in D* value. We note that D* is an apparent diffusion coefficient that does not equal an accurate diffusion coefficient because of cell confinement, motion blurring, and localization error (33, 49). Correcting for these factors was done by simulations, detailed in Monte Carlo Diffusion Simulations.

Categorizing Individual RNAP Molecules.

To determine the diffusion coefficient D, we fit the probability density of D* values of individual molecules, $x$, with an analytical equation for D, using

$$f_D\ast \left(x;D,n\right)=\frac{{\left(\raisebox{1ex}{$n$}\!\left/ \!\raisebox{-1ex}{$D$}\right.\right)}^n{x}^{n-1}{e}^{\raisebox{1ex}{$-nx$}\!\left/ \!\raisebox{-1ex}{$D$}\right.}}{\left(n-1\right)!}$$

where n is the number of steps in the trajectory used to generate the D* values (i.e., the number of squared displacements averaged to generate the MSD from which the D* is calulated). We used trajectories of four steps long; tracks of less than four steps long are not used, and tracks longer than four steps are truncated at the fourth step. We therefore fitted the equation

$$f_D\ast \left(x;D\right)=\frac{{\left(\raisebox{1ex}{$4$}\!\left/ \!\raisebox{-1ex}{$D$}\right.\right)}^4{x}^3{e}^{\raisebox{1ex}{$-4x$}\!\left/ \!\raisebox{-1ex}{$D$}\right.}}{6}.$$

We reasoned that at least two RNAP species with different mobilities are present: mobile molecules diffusion and searching DNA, and those specifically bound to DNA. This was confirmed by the poor fit to the data from a fitting this single species (Fig. S2A). We therefore introduced a second species,

$$f_D\ast \left(x;D_1,D_2,A_1,A_2\right)=\frac{A_1{\left(\raisebox{1ex}{$4$}\!\left/ \!\raisebox{-1ex}{$D_1$}\right.\right)}^4{x}^3{e}^{\raisebox{1ex}{$-4x$}\!\left/ \!\raisebox{-1ex}{$D_1$}\right.}}{6}+\hspace{0.17em}\frac{A_2{\left(\raisebox{1ex}{$4$}\!\left/ \!\raisebox{-1ex}{$D_2$}\right.\right)}^4{x}^3{e}^{\raisebox{1ex}{$-4x$}\!\left/ \!\raisebox{-1ex}{$D_2$}\right.}}{6}$$

where $D_1$ and $D_2\hspace{0.17em}$ are the diffusion coefficients of the two different species, and $A_1$ and $A_2$ are the fraction of molecules found in each state, and $A_1+\hspace{0.17em}{A}_2=1$.

To establish the apparent diffusion of the DNA-bound species, we then turned to a well-charaterized control protein DNA polymerase 1 (Pol1), which shows clearly distinct D* populations for molecules specifically bound to DNA and those searching the chromosome for substrate (Fig. S2B). Inducing DNA methylation damage by incubating cells with methyl methanesulfonate (MMS) increases the fraction of specifically bound molecules (described in ref. 20), making it easier to resolve the distribution of bound molecules. Fitting this distribution to two diffusing species allows us to determine the D* value of specifically bound molecules, as D* = 0.11 µm²/s. This apparent motion of bound molecules is mainly due to localization error.

Using this D* value for DNA-bound molecules to constrain one species and allowing a second unconstrained species fits well to the data, giving two populations of 48% bound and 52% searching (D* = 0.36 µm²/s). Taking a theshold at D* = 0.16 µm²/s allows us to categorize individual trajectories as bound or diffusing with the same ratio preserved. We note that this ratio is in reasonable agreement with previous estimates of 54 ± 7% of RNAP molecules transcribing (23). Following previously published work with Pol1 (20), we also introduced a second threshold using the D* value calculated from the 2*Δt lag times. Molecules showing D* values below both thresholds were categorized as bound, whereas molecules with D* values above both thresholds were categorized as mobile; this sorting method gives an estimated categorization precision of ∼90%. To minimize the uncertainty when determining the D* for categorizing individual molecules, only trajectories with at least four steps were used, and all steps longer than four were included.

Pair Correlation and Clustering of RNAP.

Pair correlation and clustering analysis was performed in Matlab (Mathworks). For pair correlation comparison between mobile and bound RNAPs (Fig. 2D), the pair correlation function, g(r), was calculated for categorized RNAP trajectories inside segmented cells. Only the first localization in each trajectory was used in the analysis to avoid biasing the results due to the spatial proximity of localizations generated by the same molecule. The pairwise distances for molecules were calculated on a cell-by-cell basis. Because of the small size of E. coli, as the radius r of the pairwise distance increases, much of the area falls outside the cell. To account for this effect, we simulated the randomly distibuted points within the same segmented cell outline. For each segmented cell, the same number of molecules was simulated as the experimentally observed molecules, and their pairwise distances were calculated. This was repeated over all cells, and the histogram of pairwise distances was divided by the simulated random pairwise distances to give the empirical pair correlation function, g(r), with random distribution having a g(r) of 1. For pair correlation comparison between cells in different growth conditions (Fig. S7), the pair correlation function was calculated as above except all localizations were used, as fast acquisition crowded-field PALM does not allow for particle tracking.

Fig. 2.

Comparison of categorized RNAPs with the distribution of DNA. (A) An example cell shown in a brightfield image (Left, used for cell segmentation), and in a SYTOX fluorescence image (Right, showing the nucleoid DNA distribution); x axis, short cell axis; y axis, long cell axis. (Scale bar, 1 µm.) (B) The distribution of mobile molecule trajectories (blue lines/bars) closely matched the distribution of DNA (green line) as shown in histogram projections along the x and y cell axes. (C) The distribution of bound RNAPs in the same example cell is more heterogeneous, and does not closely follow the DNA distribution. (D) Pair correlation of bound molecules (from 256 cells) shows a more clustered distribution than mobile molecules. (E) The average x axis distribution of molecules is measured from many cells by taking the relative distance from the cell midline through the center of the nucleoid, with −1 and 1 representing the cell membrane. Short cells between 1.6 µm and 2.5 µm long were chosen as they have a single nucleoid, centrally located along the y axis. (F) Plot of the x axis distribution of HU molecules from 256 cells shows the average DNA density is highest at the center and lowest at the cell periphery, with 29% of molecules found in the exterior 50% of the cell width (yellow highlighted area), compared with the expected distribution from molecules evenly occupying the full cylindrical cell volume (39% expected in cell periphery; dashed gray line). The distribution of mobile RNAP molecules (from 256 cells) matched extremely well to the distribution of HU.

Fig. S7.

Diffusion and clustering of RNAP in cells grown in rich and minimal media. (A) Fitting two species to the distribution of RNAP D* values for cells grown in minimal media. (B) Two species fitting for cells grown in rich media shows a higher fraction of bound RNAPs (63%) compared with minimal media. (C) Pair correlation of bound RNAPs (from single-molecule tracking data) shows an increase in clustering for cells in rich media, verifying that that the observed increase in clustering is not only due to the increased fraction of transcribing RNAPs. (D) Pair correlation of all localizations from rapid, high-density PALM confirms that clustering increases with growth rate. The most dramatic increase in clustering occurs at distances of <150 nm, which indicates the characteristic size of clusters.

For clustering analysis, we used a density-based clustering algorithm: density-based spatial clustering of applications with noise algorithm (4, 35) which detects clusters based on the local density of points within a search radius. To avoid biases in clustering due to variations in density of molecules per cell, the cells were segmented and their density determined from the number of molecules divided by the segmented cell area. Cells below the mean density were not used, and molecules in cells above the mean density were discarded to reduce the density to the mean.

Intracellular Spatial Distribution of Bound and Mobile Molecules.

Analysis was performed in Matlab (Mathworks). Cells were segmented from brightfield images using MicrobeTracker, giving a cell outline and a cell midline (50), and the positions of molecule trajectories were determined relative to the cell midline, with the x axis defined as the cell short axis and the y axis defined as the cell long axis. The distances were normalized to 1 and −1 at the cell membrane (Fig. 2E). The 2D histograms of the distributions of categorized molecules shown in Fig. S5 were generated by binning cells into different lengths (1.6–2.5 µm and 2.6–3.5 µm long), with short cells having a single nucleoid located in the center of the y axis, and long cells having two nucleoids. Subtracting the distribution of mobile RNAPs from the distribution of bound RNAPs highlights the difference in spatial distribution between the two populations.

Fig. S5.

Normalized 2D histogram plots showing the average spatial distribution of categorized RNAP molecules from many cells. (A) Cells in minimal media binned by cell length with short cells (1.6–2.5 µm long) having a single centrally located nucleoid, and longer cells (2.6–3.5 µm long) having two clearly separate nucleoids. A difference plot showing the distribution of bound RNAPs from which the distribution of mobile RNAPs has been subtracted highlights the bound RNAPs located at the periphery of the nucleoid. (B) After rifampicin incubation, the nucleoid expands to fill almost all of the cell volume, and separation between nucleoids is lost in long cells. Bound RNAPs no longer have a noticeably broader distribution, which is highlighted by the difference plot. (C) Chloramphenicol incubation causes nucleoid compaction, but the broader distribution of bound RNAP remains.

To determine the probability distribution across the x axis (Figs. 2F, 3D, and 4, ii), short cells (1.6–2.5 µm long) were used, as these cells had a single nucleoid located in the center of the y axis. Only molecules falling within the central 30% of the y axis were used, to exclude molecules toward the cell poles, which may be at the exterior of the nucleoid along the y axis but centrally located along the x axis (as shown in Fig. S5A). The analytical probability distribution for a uniform distribution within a cylindrical volume, $f_{uni}$, is given by

$$f_{uni}\left(\rho \right)=\hspace{0.17em}\frac{2\sqrt{\text{max}(1-{\rho }^2,0)}}{\pi }$$

where $\rho \hspace{0.17em}$ is the relative x-axis position.

Fig. 3.

DNA-free diffusion of RNAP. (A) Example minimal-DNA cell (Inset); temperature-sensitive DnaC mutant cells are grown at a nonpermissive temperature to give long cells with a single centrally located chromosome. Diffusion of RNAP in minimal-DNA cells (blue columns, 97,900 molecules) is much faster than wild-type cells (gray columns, 69,900 molecules). (B) Tracking RNAPs only in the DNA-free cell endcaps (green bars, 2,400 molecules) allows the free 3D diffusion to be determined. (C) Simulated molecular tracks undergoing Brownian diffusion within a confined cell endcap volume. Analyzing the tracks using the same protocols as the experimental data gives an estimated accurate D value that best matches the experimental data. Scanning through D values from 1 µm²/s to 5 µm²/s, the best value was D = 2.6 µm²/s (black dashed line in B). (D) Plot of the x axis distribution of RNAP molecules from 72 DNA-free cell endcaps; 40% of these RNAP molecules are found in the exterior 50% of the cell width (yellow highlighted area); 39% of molecules evenly occupying the full cylindrical cell volume are expected to be found in the periphery (dashed gray line).

Fig. 4.

Transcribing RNAPs cluster at the nucleoid periphery. (A) (i) An example cell growing in minimal media with trajectories of mobile (blue) and bound (red) RNAPs shows the difference in location of transcribing RNAPs compared with the mobile population. (ii) Probability density distribution of categorized RNAPs in the nucleoid across the x axis for 256 cells. Mobile RNAPs (blue line) follow the distribution of the nucleoid (HU distribution; gray line), with 29 ± 0.7% located in the exterior 50% of the cell width (yellow highlighted area). Transcribing molecules have a significantly broader distribution (P < 0.001), with 41 ± 0.9% of molecules locating in the periphery. (iii) The 3D SIM images of live cells in minimal media show that the densest regions of RNAP (red) are located at the edge of the nucleoid (blue). Projections onto the z–x axis highlight that dense regions apparently located in the center of the cell in X−Y projections are in fact above or below the central bulk of the DNA. (B) (i) Example cells showing clustering of bound RNAP molecules, with clustered RNAPs (>6 molecules) colored purple and nonclustered RNAPs (single or pairs of RNAPs) colored green. (ii) x-axis distribution of clustered and nonclustered bound RNAPs in 120 cells shows that nonclustered RNAPs are distributed throughout the nucleoid, whereas dense clusters form at the periphery (71 ± 9%). (iii) SIM images confirm that, although the RNAP distribution frequently overlaps with the DNA distribution, the densest RNAP regions locate at the cell edge (see also Movie S2). (C) (i) An example cell after rifampicin incubation. (ii) After rifampicin incubation, the width distributions of mobile and bound RNAPs become almost identical (P > 0.05), with 31 ± 0.8% of mobile molecules and 31 ± 1.6% of bound molecules found in the cell periphery. (iii) SIM images show DNA and RNAP distributions fill most of the cell volume homogeneously. (D) (i) An example cell after chloramphenicol incubation. (ii) The width distribution shows that 46 ± 1.7% of bound RNAP is found in the periphery compared with 33 ± 0.4% of mobile molecules. (iii) SIM image showing a cell with a compacted nucleoid with RNAPs located at the edge of the bulk of DNA.

For Fig. 4B, ii, clustering was performed first, as previously descibed, then categorized clustered and unclustered molecules were analyzed as above.

Expected Mobility of Free RNAP.

Estimates of the diffusion coefficient of unconjugated fluorescent proteins diffusing in E. coli range between 7 µm²/s [eYFP (51)], 7.7 µm²/s [GFP (52)] 8 µm²/s, [Venus (34)], 10 µm²/s [Dendra2 (32)], and 7.2 µm²/s (PAmCherry, Fig. S4). Given that the diffusion coefficient is inversely proportional to the Stokes radius, and estimated Stokes radii for GFP vs. RNAP are 2.4 nm and 6.65 nm, respectively (53), we would therefore expect free RNAP to diffuse within the cytoplasm with D between 2.5 µm²/s and 3.6 µm²/s.

Fig. S4.

Distribution of unconjugated PAmCherry. (A) The distribution of D* values for unconjugated PAmCherry imaged at 1-ms exposure times. A single species fit gives a D* value of 7.2 µm²/s. (B) The probability density distribution of unconjugated PAmCherry across the x axis for 32 cells; 38 ± 2% are located in the exterior 50% of the cell width (yellow highlighted area), compared with 39% expected from an even distribution.

Monte Carlo Diffusion Simulations.

The apparent diffusion determined experimentally through particle tracking does not take into account 3D confinement in the bacterial cell, nor other effects such as localization error and motion blurring (20, 33). To determine the accurate D values from D* experimental data, we simulated Brownian motion within a cell endcap corresponding to the average size of those determined from experiments. For diffusion in DNA-free cell endcaps, the cell endcap was defined as a cylindrical volume of length 1.2 µm and 0.85-µm wide with hemispherical endcaps with a radius of 0.85 µm. For wild-type cells, we used a volume roughly equal to the nucleoid volume, defined as cylindrical volume of length 1.6 µm and 0.65-µm wide with hemispherical endcaps with a radius of 0.65 µm. Each 15-ms frame was split into 100 subframes, with Gaussian distributed displacements in each subframe. Each molecule trajectory was given a random starting time to mimic stochastic photoactivation, and a duration sampled from an exponential distribution with a mean time equal to our experimentally determined photobleaching lifetime (85 ms). The subframe distributions were then averaged to give a position for each frame, and a localization error sampled from a Gaussian distribution with sigma = 40 nm (determined from the distribution of localizations from a bound molecule) was added. The list of simulated localizations, with their corresponding frame numbers, was then analyzed using the same tracking algorithm with the same settings as used for the experimental data. The outputted tracks could then be analyzed in exactly the same way as experimental data.

Tracking RNAP Foci in Rich Media.

Imaging dense RNAP clusters in live cells has previously proved challenging with conventional fluorescence microscopes. By reducing background fluorescence using laser excitation at shallow illumination angles (48) and using a sensitive camera (9, 18), we clearly resolved bright foci, in GFP-labeled RNAP in live cells (Movie S5). To determine the mobility of these foci, we used 2-s exposures, and low excitation power, over a 2-min timescale. Foci were identified for localization by band-pass filtering and applying an intensity threshold to each frame of the movie. Candidate positions were used as initial guesses in a 2D elliptical Gaussian fit with free fit parameters: x position, y position, rotation angle, intensity, background. X width, Y width, and elliptical eccentricity were constrained to prevent fitting to the overall RNAP background. Tracking localizations, as described in Measuring the Diffusion of RNAP, and plotting the MSD show that these structures are fairly dynamic. To compare the mobility of these structures with DNA loci, we labeled the ori region of the chromosome using an array of TetO operator sites and chromosomally expressed TetR-YFP in a separate strain. TetR-YFP foci were localized, tracked, and analyzed as RNAP−GFP foci.

Bound and Mobile RNAPs Show Different Spatial Distributions

The locations of mobile and bound RNAPs were compared with the spatial distribution of DNA by staining the nucleoid DNA with the intercalating fluorescent dye, SYTOX green (Fig. 2A) (29). The spatial distribution of mobile RNAPs in the cell closely followed that of DNA, with only a few excursions of RNAP tracks into the surrounding cytoplasm, indicating that most promoter-searching RNAP molecules remain within the volume of the nucleoid (Fig. 2B). In contrast, bound RNAPs were much more heterogeneously distributed, and their distribution did not clearly match to that of DNA (Fig. 2C).

To quantify the spatial clustering of mobile and bound RNAPs, we calculated the pair correlation function, which gives the relative probability of finding a protein at a given distance away from another protein compared with a random distribution (30). The pair correlation function for molecules in >200 cells shows a clear clustering of bound RNAPs at distances shorter than 150 nm (Fig. 2D), as expected for multiple RNAPs simultaneously bound at highly transcribed genes. This clustering was absent in the case of mobile molecules, pointing to their independent, nonspecific RNAP interactions throughout the nucleoid.

As an additional verification that mobile RNAPs remain strongly associated with the nucleoid, we compared the spatial distribution of mobile RNAPs to that of nucleoid-associated heat unstable protein (HU), which reports on nucleoid DNA density (31). Plotting the probability density of HU molecules across the cell short axis (defined here as the x axis, Fig. 2E) through the center of the nucleoid (>100 cells) showed that the DNA density is lowest at the cell periphery. Defining the periphery as the exterior 50% of the cell width (yellow highlighted areas, Fig. 2F) gave only 29% of HU molecules in this region compared with 39% expected from a uniform distribution in a cylindrical cell volume (32). Performing the same analysis on mobile RNAP molecules showed that they matched extremely well to the distribution of HU, and also have 29% located in the cell periphery rather than uniformly filling the cell volume. These results confirm our SYTOX-based results and strongly suggest that mobile RNAPs explore the entire nucleoid, having access to even the very dense regions of the nucleoid DNA.

RNAP Spends 85% of Its Promoter Search Time Bound to Nonspecific DNA

We then focused on the process of promoter search, wherein RNAP must identify specific substrate DNA sequences in the midst of huge amounts of nonspecific chromosomal DNA (22). The high level of association between mobile RNAPs and DNA and the absence of excursions into the cell cytoplasm further suggest a high level of transient nonspecific RNAP interactions with DNA during the promoter search. Consistent with this, the apparent diffusion coefficient for mobile RNAPs in our experiments, D*_mobile, is an order of magnitude smaller than the expected free diffusion of RNAP, given its size (see SI Materials and Methods), indicating an interconversion between 3D diffusion and transient binding.

To determine the fraction of time that RNAP spends nonspecifically bound to DNA, two apparent diffusion coefficients need to be obtained: the D* of DNA-bound RNAP molecules (D*_bound) and the D* for intracellular RNAP diffusion in the absence of DNA (D*_free). Although obtaining D*_bound is straightforward (extracted using the control experiments with Pol1; Fig. 1B and Fig. S2), obtaining D*_free is more challenging, because virtually all RNAPs in wild-type E. coli are located within the nucleoid, complicating the study of freely diffusing RNAP molecules.

To characterize the intracellular diffusion of RNAP in the absence of the nucleoid, we used a temperature-sensitive DnaC mutation to generate a minimal-DNA E. coli strain with long DNA-free endcaps (Fig. 3A, Inset). At nonpermissive temperatures, these cells were unable to initiate DNA replication but kept elongating, yielding long cells containing a single chromosome. We also treated cells with norfloxacin (a DNA gyrase inhibitor) to enhance DNA compaction and increase the relative volume of DNA-free endcaps. In these minimal-DNA cells, there is a dramatic increase in the mobility of RNAP compared with wild-type cells (Fig. 3A). Tracking RNAPs molecules located only in the DNA-free cell ends allowed us to determine the D*_free distribution, which was centered at ∼1.1 µm²/s (Fig. 3B). The x axis distribution of RNAP molecules located in the DNA-free endcaps agreed well with the expected distribution for molecules evenly distributed through across the cell (Fig. 3D).

To correct for the effects of confinement due to the small cell volume, motion blurring, and localization error on the experimentally observed D*, and obtain accurate unbiased D values, we used Monte Carlo simulations of Brownian motion within an average-size cell endcap (Fig. 3C) (19, 20, 33). By scanning through D values and analyzing simulated tracks as we did for experimental tracks, we found that a diffusion coefficient D_free of 2.6 µm²/s best matched our data (dashed black line, Fig. 3B). This diffusion is within the theoretically expected range (2.5–3.6 µm²/s), given the size of RNAP (see SI Materials and Methods and Fig. S4), and noticeably faster than previous estimates [0.7 µm²/s (23)].

Performing similar simulations on nucleoid-confined molecules, we found that the D*_mobile of 0.36 µm²/s measured in wild-type cells matched a D_mobile of 0.4 µm²/s. We can then determine the fraction of time, x, that the interconverting species, D_mobile, spends binding nonspecifically to DNA using D_mobile = xD_bound + (1-x)D_free (34). Using our experimentally determined D values to solve this equation shows that RNAP spends 85% of its promoter search time nonspecifically bound to DNA. This high amount of nonspecific binding could explain the very high spatial correlation of mobile RNAP with DNA observed in Fig. 2.

Transcribing RNAPs Tend to Cluster at the Edge of the Nucleoid

Simple visual inspection of the spatial distribution of bound RNAPs showed that they were much more frequently located at the periphery of the nucleoid compared with the mobile population (Fig. 4A, i). To quantify this, we plotted the probability distribution of mobile and bound RNAPs across the cell short axis (x axis) through the center of the nucleoid, from >200 cells. Bound RNAPs showed a broader distribution than the nucleoid (shown by both the HU distribution and mobile RNAP distribution), with 41% of them located in the cell periphery (yellow highlighted areas, Fig. 4A, ii), compared with 29% for the nucleoid. This broader distribution is evident along the y axis as well as the x axis, with bound RNAPs also located at the periphery of the nucleoid toward the cell endcaps (Fig. S5).

The relative localization of RNAP and nucleoid DNA was further characterized using 3D SIM coimaging of RNAP−GFP and DAPI-stained DNA in live cells (Fig. 4A, iii and Movie S1). SIM produces a superresolved visualization of the nucleoid volume and edges, enabling better assessment of the relative position of the RNAP clusters in three dimensions. We observed that the overall RNAP signal overlaps with the nucleoid; however, the densest RNAP clusters are nearly always located at the nucleoid edge (Fig. 4B, iii and Movie S2). Projections onto the z–x axis highlight that dense RNAP clusters that appear to be located in the center of the cell in x−y projections are in fact located at the surface of the nucleoid, where the DNA density is lower.

We confirmed this observation by clustering analysis of DNA-bound RNAPs from PALM data sets. We used a density-based clustering algorithm (4, 35) to distinguish nonclustered RNAPs (single and pairs of RNAPs) from those in dense clusters (>6 molecules). The resulting plot shows that dense clusters of bound RNAPs are clearly located at the nucleoid periphery (71% in the periphery), whereas nonclustered RNAPs were located throughout the nucleoid (33% in the periphery) (Fig. 4B, ii). This shows that, although low levels of transcription can occur throughout the nucleoid, dense RNAP clusters (i.e., more highly transcribed genes) were located at the nucleoid periphery, exactly at the interface between the DNA and the ribosome-enriched cytoplasm. When we performed the same analysis on another nucleoid associated protein, the histone-like nucleoid-structuring (H-NS) protein, we saw the opposite trend, with dense clusters located in the nucleoid center (only 12% in the periphery), and small clusters essentially matching the nucleoid density (30% in the periphery; Fig. S6). This result establishes that the clustering of NAPs does not universally lead to exclusion from the nucleoid, and adds weight to the hypothesis that active transcription drives RNAP segregation from the nucleoid.

Fig. S6.

Distribution of clustered and nonclustered H-NS. Performing the same clustering and x-axis distribution analysis as performed on RNAP (Fig. 4B) with a control protein, H-NS, shows that dense H-NS clusters appear to form preferentially at the center of the cell x axis, with only 12% forming in the cell periphery. Nonclustered H-NS locate throughout the nucleoid with 30% located in the cell periphery, similar to the 29% of HU molecules found in the same region.

Blocking transcription with rifampicin abolished this spatial pattern, as the distribution of bound RNAPs (on promoters, but not transcribing) and mobile RNAPs became almost identical, with 31% of both distributions located in the cell periphery (Fig. 4C, ii). Consistent with this, SIM images show homogenous, overlapping RNAP and DNA distributions (Fig. 4C, ii and Movie S3). This demonstrates that active transcription is necessary for maintaining this characteristic spatial distribution of RNAP with respect to the nucleoid DNA.

We then examined the effect of active translation on the RNAP spatial profile, because cotranscriptional translation has also been proposed to play an important role in the nucleoid structure. According to the transertion model (36), transmembrane proteins produced by cotranscriptional translation indirectly attach transcribed genes to the cell membrane, promoting nucleoid decompaction and potentially helping chromosome segregation. To block translation, we used chloramphenicol (Fig. 4D), an antibiotic that blocks protein chain elongation by inhibiting the peptidyl transferase activity of the 50S ribosomal subunit. Consistent with previous reports (16), SIM imaging showed that chloramphenicol triggers a dramatic compaction of the nucleoid into a ring-like structure that locates at the center of the cell (Fig. 4D, iii); however, RNAP molecules still cluster at the nucleoid surface (Movie S4). These results are supported by PALM results, which show that the periphery bias of bound RNAPs is maintained; indeed, many cells showed distinct rings of bound RNAPs surrounding the nucleoid (Fig. 4D, i). On average, 46% and 33% of bound and mobile RNAPs, respectively, were located in the periphery (Fig. 3D, ii). These results suggest that transertion indeed promotes chromosome expansion but is not the main driving force for the global spatial localization of large RNAP clusters to the nucleoid surface.

Transcription in Rich Media Occurs in Dense Dynamic Clusters at the Nucleoid Periphery

As growth rate increases, transcription on most genes is reduced; however, a small number of genes, particularly stable RNAs (such as ribosomal RNAs), become very heavily transcribed (9, 10). When we performed tracking and sorting of RNAPs in cells grown in rich media, we found that a significantly higher fraction of RNAPs (63%) were bound, compared with minimal media conditions (48%; Fig. 5A and Fig. S7 A and B). This difference indicates a higher fraction of RNAPs engaged in transcription, despite the fact that cells in rich media have roughly half the number of active genes relative to minimal media (37).

In cells grown in rich media, and subsequently chemically fixed, RNAP has been shown to form dense clusters, which were not observed in minimal media (4, 10); however, little is known about their presence, dynamics, or behavior in live E. coli. The observation of these large RNAP clusters, reminiscent of “transcription factories” in eukaryotic cells (11), raised the question of whether these are static structures that stably anchor in cells and pull transcribed DNA through them, or whether RNAPs assemble in large numbers on active promoters and simply follow the motion of the segregating chromosome during transcription. To address this question, we measured the mobility of RNAP clusters using GFP-labeled RNAP in live cells growing in rich media (10). Time-lapse images over a 2-min timescale showed that these structures were dynamic (Movie S5). To compare their mobility with that of DNA loci, we labeled the DNA close to the origin of replication (Ori) on the chromosome by using an array of TetO operator sites and chromosomally expressed TetR-YFP. Computing the MSD for tracked RNAP clusters and DNA loci over the same timescale, we found that clusters have similar mobility to DNA (Fig. 5B). This indicates that these foci were moving along with the transcribed DNA region and are unlikely to be rigidly tethered to static cellular structures.

To characterize the size of these RNAP clusters in live cells, we performed rapid localization microscopy with high photoactivation (38). Typical PALM imaging takes several minutes per superresolved image, which would blur these RNAP clusters due to their movement over this timescale. Chemical fixation, on the other hand, distorts biological structures (39). To capture sharp superresolved PALM images in live cells, we localized RNAPs photoactivated at high density and analyzed using a crowded-field localization algorithm (38), achieving a 50-nm resolution at 15 s per image. These snapshots of RNAP localizations confirm that clustering is much more extensive under fast growth conditions compared with minimal media conditions (Fig. 5C). To quantify this difference, we calculated the pair correlation distribution of the localizations, which showed an increase in clustering at pairwise distances less than ∼150 nm for rich media compared with minimal media (Fig. S7 C and D).

To get an estimate of the number of RNAP molecules per cluster in cells grown in different media, we clustered our localizations using a density-based algorithm (4, 35). We determined the mean number of localizations per molecule (by particle tracking using the same acquisition protocol but lower photoactivation) and used this to convert localizations to numbers of molecules (Fig. 5D). Our results showed that the cluster sizes in live cells match broadly to previous estimates in fixed cells (4): We observed intermediate-size clusters in minimal media (∼50 molecules), and large clusters (up to 500 molecules) in rich media.

SIM imaging in live cells confirmed that the organization of the nucleoid also changes as growth rate increases. In minimal media, the nucleoid is relatively homogenous (Fig. 5E, Left), whereas in rich media, nucleoids are more heterogeneous, with distinct structures (Fig. 5E, Right). In rich media, the very dense RNAP clusters clearly form at the nucleoid edge, showing stronger segregation than in minimal media, with very little overlap between RNAP and DNA (Fig. 5F and Movie S6). Interestingly, the RNAP clusters form not just at the cell periphery but also in areas of low-density DNA throughout the cell volume (Fig. 5F, Right). Comparing rich and minimal media shows that increasing RNAP clustering correlates with increased bias of RNAP to the edge of the nucleoid. In rich media, this effect is more dramatic because not only is transcription on a few genes increased heavily (leading to increased clustering) but transcription is also reduced on most of the remaining genes, which could allow this DNA to be more tightly compacted.

Discussion

Using live-cell superresolution microscopy, single-molecule tracking, and diffusion simulations, we have shown how RNAP locates promoters distributed throughout the nucleoid, and how active transcription causes spatial reorganization of the DNA.

Mobile RNAPs and Promoter Search.

In minimal media conditions, approximately half of RNAPs are mobile and diffusing within the nucleoid, engaging in frequent nonspecific interactions with DNA as they search for promoters. Contrary to a recent study (40), we see that mobile RNAPs are not excluded from any part of the nucleoid. Indeed, mobile RNAPs show striking agreement with the DNA distribution, with the densest regions of DNA correlating with the densest concentrations of mobile RNAPs. This implies that searching RNAPs can find promoters located throughout the nucleoid (Fig. 6A).

Fig. 6.

Mechanisms of gene spatial organization. (A) RNAPs and free ribosome subunits can explore the entire nucleoid search for specific nucleic acid sequences. RNAP starts transcribing a gene within the nucleoid, and the first ribosome binds to the emerging mRNA. As the polyribosome grows and other RNAPs start transcribing, entropic forces favor movement away from the bulk of the nucleoid. (B) In the case of rRNA transcription, there is no coupled translation, but rRNA operons are extremely highly transcribed. Multiple RNAPs on the same gene may also drive movement of the gene toward the periphery of the nucleoid.

The fraction of specifically bound molecules, f_bound, can tell us about the approximate timescale for an average RNAP to find a promoter and engage in transcription, t_mobile, because f_bound = t_bound/(t_bound + t_mobile) (20). In minimal media, f_bound ≈ 0.5, hence the search time, t_mobile, is roughly the same as the average transcription event, on the order of 30–120 s (considering the approximately timescales for promoter opening, promoter escape, and transcriptional elongation, and the length of the average gene). During this search process, RNAP spends about 85% of its time nonspecifically bound to DNA, which is a similar percentage to that seen for the lac repressor (34). In rich media, f_bound is nearly 30% higher (0.63); therefore the average search time should be only ∼60% the duration of the average transcription event.

Bound RNAPs Found in Clusters Are Biased Toward the Nucleoid Periphery.

Transcribing RNAPs are more clustered than mobile molecules. Interestingly, nonclustered transcribing RNAPs are located throughout the nucleoid; however, the densest clusters are preferentially located at the edge of the nucleoid where the DNA density is lowest. It appears, therefore, that although low levels of transcription can occur on genes distributed throughout the nucleoid, the most heavily transcribed genes locate to the nucleoid periphery. Incubation with rifampicin shows that active transcription is necessary to maintain these genes at the periphery. These results provide compelling evidence that transcription itself is the driving force for locating active genes to the nucleoid edge, rather than a mechanism whereby regions of the chromosome are inaccessible to RNAP (40), leaving only genes at the periphery available to be transcribed.

Mechanism of RNAP Spatial Organization.

One mechanism that could cause highly transcribed genes to locate to the nucleoid periphery is cotranscriptional translation (12, 32). Ribosomes are strongly segregated from the E. coli nucleoid (3, 14, 15). Statistical models have suggested that this phase separation between the ribosome-containing cytoplasm and the DNA can be explained by simple entropic forces: The DNA polymer avoids the walls to maximize conformational entropy, and the polysomes (multiple 70S ribosomes on a single mRNA) occupy the empty space near the walls to maximize translational entropy (41). Although mRNA-bound 70S ribosomes are excluded from the nucleoid, it has recently been shown that free 50S and 30S subunits do have access to the nucleoid center (32, 42). These results indicate that both transcription and translation can begin in the center of the nucleoid (Fig. 6A). Transcription initiation and early elongation produce a nascent mRNA that contains a ribosome binding site, which nucleates the recruitment and assembly of one or several ribosomes when translation is initiated. It seems likely that the large DNA−RNAP−ribosome complexes are also entropically excluded from the bulk of the nucleoid and tend to migrate toward the nucleoid periphery, thus leading to the spatial separation of genes depending on their level of expression, with highly expressed genes enriched at the surface of the nucleoid. The inner volume of the nucleoid is consequently less metabolized, potentially allowing it to become more highly compacted. Consistent with this proposal, in minimal media, the nucleoid is less structured, showing less variation in local DNA density due to the fact that genes are expressed throughout the chromosome. In contrast, in rich media, fewer genes are expressed, thus allowing large parts of the chromosome to be inactive, giving rise to more structured and compacted nucleoids.

In rich media, most transcription occurs on the ribosomal RNA operons. Despite the fact that these transcripts are not translated, there appears to be even stronger segregation between RNAP clusters and nucleoid DNA in rich media compared with minimal media. Coupled transcription and translation is therefore not the only mechanism of locating RNAP clusters to the nucleoid edge. In rich media, the active genes are fewer but are more heavily transcribed; for example, ribosomal operons can have up to 80 transcribing RNAPs on them at any one time (4). The large number of RNAPs, which may also be coupled with associated ribosome assembly cofactors (43), are likely to exert similar entropic bias out of the nucleoid DNA as coupled RNAP−mRNA−ribosome complexes (Fig. 6B).

Our results also indicate that other potential mechanisms for creating the phase separation between RNAP and the nucleoid have significant shortcomings. For example, because MreB and RNAP copurify in cell extracts and interact in vitro (4), it has been proposed that RNAP clusters could be anchored to the cytoskeleton. Although our data do not rule out some interaction between RNAP and the cytoskeleton, we believe it is unlikely for this to be a primary mechanism for locating RNAP clusters to the nucleoid periphery, for two reasons. Firstly, tracking of RNAP clusters in rich media showed that they have very similar mobility to DNA, suggesting that they are not rigidly tethered to the cytoskeleton. Secondly, 3D SIM imaging showed that in rich media, where the nucleoid forms more heterogeneous structures, RNAP clusters form not just at the cell periphery but also in the central volume of the cell in low-density regions of the nucleoid that are distal from the cell membrane.

Transertion has also been proposed to play a role in nucleoid decompaction because blocking translation with chloramphenicol causes nucleoid compaction. Indeed, individual genes expressing membrane proteins have been shown to migrate to the membrane when expressed (44). However, we show that clusters of bound RNAPs still remain at the nucleoid edge after chloramphenicol incubation, indicating that transertion might play a role in the nucleoid conformation but is clearly not the driving force behind localization of large RNAP clusters to the nucleoid surface.

Taken together, our work shows that transcription can cause spatial reorganization of the nucleoid, with movement of gene loci out of the bulk of DNA as levels of transcription increase. The peripheral localization of transcription is reminiscent of that of DNA repair (45), supporting the view that DNA metabolism activities involving large machineries locate preferentially at the nucleoid periphery, where large protein complex formation is facilitated. In rich media conditions, the number of RNAP clusters observed per cell (10), and the number of RNAPs found in each cluster (4), has led to the hypothesis that these clusters represent RNAPs active on multiple heavily transcribed genes spatially located to the same site, analogous to transcription factories in eukaryotes (11, 46). Similarly, during DNA break repair, distant DNA loci are brought together at the periphery of the nucleoid (45). The localization of heavily transcribed genes outside the bulk of DNA may therefore also facilitate spatial clustering of genes located at distant sites on the chromosome.

Materials and Methods

Complete details of materials and methods are available in SI Materials and Methods. In brief, cells were grown in either M9 minimal or EZ Rich Defined Media (Teknova) supplemented with 0.2% glucose to OD of ∼0.2 and immobilized for imaging on 1% agarose pads. Where indicated, cells were incubated with 50 µg/mL rifampicin or 100 µg/mL chloramphenicol for 30 min. Live cell single-molecule-tracking PALM, rapid high-density PALM, and RNAP-GFP cluster tracking experiments were performed on a custom-built total internal reflection fluorescence (TIRF) microscope. PALM data for single-molecule-tracking analysis was localized using custom-written MATLAB software (MathWorks). Fluorophore images were localized to 40-nm precision by elliptical Gaussian fitting. Localizations within a radius of 0.48 μm in consecutive frames were linked into tracks. An apparent diffusion coefficient, D*, was calculated from the mean-squared displacement (MSD) for each track with a minimum of 4 steps. 3D-Structured illumination (SIM) imaging was performed on a DeltaVision OMX V3 (Applied Precision/GE Healthcare) equipped with a Blaze SIM module. Complete details of PALM and SIM microscopy are described in SI Materials and Methods. Data analysis and simulations were performed in MATLAB; full details are presented in SI Materials and Methods.