RNA polymerase redistribution supports growth in E. coli strains with a minimal number of rRNA operons

Abstract Bacterial transcription by RNA polymerase (RNAP) is spatially organized. RNAPs transcribing highly expressed genes locate in the nucleoid periphery, and form clusters in rich medium, with several studies linking RNAP clustering and transcription of rRNA (rrn). However, the nature of RNAP clusters and their association with rrn transcription remains unclear. Here we address these questions by using single-molecule tracking to monitor the subcellular distribution of mobile and immobile RNAP in strains with a heavily reduced number of chromosomal rrn operons (Δrrn strains). Strikingly, we find that the fraction of chromosome-associated RNAP (which is mainly engaged in transcription) is robust to deleting five or six of the seven chromosomal rrn operons. Spatial analysis in Δrrn strains showed substantial RNAP redistribution during moderate growth, with clustering increasing at cell endcaps, where the remaining rrn operons reside. These results support a model where RNAPs in Δrrn strains relocate to copies of the remaining rrn operons. In rich medium, Δrrn strains redistribute RNAP to minimize growth defects due to rrn deletions, with very high RNAP densities on rrn genes leading to genomic instability. Our study links RNAP clusters and rrn transcription, and offers insight into how bacteria maintain growth in the presence of only 1–2 rrn operons.


INTRODUCTION
Tr anscription, a centr al process in gene expression, is spatially organized in many organisms; this organization is thought to increase the efficiency for RNA synthesis ( 1 ) and help cells adapt to different growth environments, nutrients and types of stress. In eukaryotes, synthesis of rRNA by RN A pol ymerase (RN AP) I occurs in the nucleolus, a nuclear compartment ( 2 ); further, eukaryotic mRNA transcription occurs in spatially enriched foci called 'transcription factories' ( 3 ), which contain RNAP II clusters ( 4 ) with lifetimes correlated to the le v els of mRNA synthesis ( 5 ). Some viral transcription systems are also spatially organized; for example, RNAPs of poliovirus form planar arrays / lattices with hundreds of molecules ( 6 ).
Transcription has also been shown to be spatially organized in bacteria, where early studies using conventional fluor escence microscop y in fix ed cells showed that fluor escent deri vati v es of RN AP in Esc heric hia coli and Bacil-lus subtilis ( 7 ) form bright, diffraction-limited foci in rich medium, but not in minimal medium; these prokaryotic transcription foci have been likened to transcription factories ( 3 , 8 ). Subsequent studies using photo-activated localization microscopy (PALM), a super-resolution imaging method, provided further insight into RNAP spatial organization; using PALM on fixed cells under different growth conditions, it was shown that RNAPs form large clusters with ∼70 and > 100 molecules in rich medium, and smaller clusters with ∼35 molecules in minimal medium ( 9 ). Singlemolecule localization studies in li v e E. coli cells showed that RNAPs tend to co-localize with the nucleoid lobes, while being nearly absent from the ribosome-rich cell endcaps ( 10 , 11 ). Further li v e-cell wor k combining PALM and single-molecule tracking was able to distinguish between mobile RNAPs (i.e., RNAPs exploring the nucleoid for promoters) and immobile RNAPs, with the latter fraction including transcriptionally acti v e RNAPs that localized primarily at the nucleoid periphery; this study also provided the first observation of RNAP clustering in living bacteria ( 12 ).
Surprisingly, subsequent localization-based work ( 13 ) suggested that RNAP clustering remained significant e v en when transcription was suppressed, and only decreased substantiall y w hen all transcription was inhibited by rifampicin, leading to the proposal that the underlying nucleoid (rather than high transcription activity) controls the organization of these RN AP clusters. Recentl y, it was suggested that RNAPs in bacteria form 'biomolecular condensates' ( 14 ) via liquid-liquid phase separation (LLPS), a phenomenon seen in many organisms (15)(16)(17), including bacteria (18)(19)(20)(21); the condensates were shown to contain highdensity RNAP clusters in fast-growth conditions, and were mediated by pr otein-pr otein interactions, offering LLPS as an alternati v e mechanism that dri v es RNAP clustering ( 22 ).
A central point of debate in the spatial organization of transcription and the formation of transcription foci is the exact role of ribosomal rRNA operons (rRNA operons, rrn ). rRNA transcription (which involves 16S, 23S and 5S rRNA) accounts for ∼85% of all acti v e transcription in fast-growing cells ( 23 ); such high transcription le v els are essential for sustaining rapid synthesis of the ∼55 000 ribosomes ( 10 ) needed per daughter per cell cycle during rapid growth ( 24 ). Notably, rrn transcription is much less prevalent in minimal medium. In the genomic map of E. coli , most of the se v en rrn operons locate near the origin of replication ( oriC ), and all rrn operons orient in the same direction as DNA replication (Figure 1 A); this chromosomal location leads to increased gene dosage for rrn genes.
Transcription foci and RNAP clustering have been linked to rrn operons e v en during the first RNAP distribution studies ( 1 , 9 ), which raised the possibility that transcription foci involve multiple (perhaps all) rrn operons operating in close proximity and in a growth-dependent manner (since such foci were absent in minimal medium), supporting a 'bacterial nucleolus' model ( 2 , 24 ). Consistent with that model, Gaal et al . measured the pairwise distance of rrn operons in single cells and found that six out of the se v en rrn operons in E. coli are in close proximity in 3D space [ Figure 1 A; ( 25 )]. Endesfelder et al . also linked RNAP clus-tering to rrn operons, and suggested that clusters with 35-70 molecules r epr esent single rrn operons, and large clusters ( > 100 molecules; 50-300 nm in diameter) r epr esent superclustered multiple rrn operons ( 9 ). These indirect links were supported by Weng et al . ( 13 ), who directly showed that RNAP clusters were indeed co-localizing with sites of high rrn transcription in rich medium, while the formation of these clusters was independent of rrn transcription activity ( 13 ). The persistence of significant clustering despite the dramatic loss in rrn transcription was la ter a ttributed to LLPS ( 22 ).
Despite the progress in the understanding of spatial organization of transcription, there are still many open questions. W ha t is the link between RNAP clusters and rrn operons, if any, in minimal medium? To what extent does LLPS contribute to RNAP clustering forming on rrn operons? W ha t are the mechanisms that maintain the ability of cells to grow e v en when the number of chromosomal rrn operons is very small (1-2 copies)?
Here, we study the link between RNAP clusters and rrn operons by using single-molecule imaging and tracking ( 26 , 27 ) to obtain the RNAP spatial distribution and mobility in strains featuring deletions of most rrn operons ( Δrrn strains; Figure 1 B). We show that, remar kab ly, in strains with only one or two chromosomal rrn copies, bacterial cells maintain the same le v el ( ∼48%) of immobile RNAPs (which mainly reflects RNAPs engaged in transcription) during moderate growth rates; immobile RNAPs in Δrrn strains move close to cell endcaps, suggesting that RNAPs relocate to the remaining rrn operons, which have a pole-proximal location. During fast growth in rich medium, loss of most rrn operons leads to only a modest decrease of the immobile RNAP fraction, suggesting that RNAP redistributes to other rrn and nonrrn genes on the chromosome. RNAPs retained their clustering in the Δrrn strains, w hereas co-localization anal ysis showed a good correlation between RNAP clusters and rrn operons. Our work expands our understanding of how RNAP is organized and allocated between transcription activities and how bacteria regulate their transcription to adapt to variations of their chromosomal content and growth environment.

Bacterial strains
The rpoC:PAmCherry wild-type (WT) strain carrying PAm-Cherry fused to the ␤' subunit under the control of its nati v e promoter used was built as described previously ( 9 ). The Δrrn strains were obtained by P1 transduction of the rpoC:PAmCherry gene in the Δrrn strains. The deletion strains wer e acquir ed from CGSC ( E. coli Genetic Stock Center at Yale): SQ88 as 5rrn with rrn B and rrn C remaining; SQ110 as 6rrn with rrn E remaining; and SQ2158 as 7rrn supplemented with plasmid-borne rrn B [pK4-16, based on pSC101, see also ( 28 )].
Cell growth rate measurements were performed using OD 600 on a microplate reader (FLUOStar, BMG Labtech). Three separate measurements were carried out with individual blank media. The absorbance of OD 600 was measured e v ery 5 min for 16 h to generate the growth curves.
Nucleic Acids Research, 2023, Vol. 51, No. 15 8087 Cell pr epar ation f or imaging Strains wer e str eaked onto Luria-Bertani (LB) plates supplemented with r equir ed antibiotics for each strain. For the WT, we used 100 g / ml ampicillin; for 5 and 6, we used 100 g / ml ampicillin and 40 g / ml spectinomycin, respecti v ely; and for 7, we used 100 g / ml ampicillin, 40 g / ml spectinomycin and 50 g / ml kanamycin. Single colonies were inoculated into LB and grown at 37 • C and 220 rpm for a pr e-cultur e of 2 h, then diluted 1 / 250 into M9Glu medium (1 × M9 medium supplemented with CaCl 2 , MgSO 4 and 0.2% glucose but without any additional vitamins or amino acids) or RDM (rich defined medium; Teknova) and grown at 37 • C ov ernight. Ov ernight cultur es wer e diluted into fresh medium and grown for > 2 h at 37 • C until early exponential phase (OD 0.1-0.2 for M9Glu culture, or OD 0.2 for RDM culture). A 1.5 ml aliquot of cell culture was centrifuged down, concentrated to 30 l and immobilized on 1% low-fluorescence agarose (BioRad) pads (supplemented with r equir ed M9Glu or RDM to keep media consistent). After immobilizing the cells on agarose pads with fresh medium, we monitored the RNAP localization using singleparticle tracking by PALM at 22 • C and measured the apparent dif fusion coef ficient (D*) of RNAPs (see also Figure  1 C, D).
For fixed-cell co-localization experiments combining PALM with fluorescence in situ hybridization (FISH), 1.5 ml of culture of the WT or rrn strains carrying rpoC:PAmCherry were spun down and then resuspended into 1 ml of phospha te-buf fered saline (PBS). A 1 ml aliquot of 4% paraformaldehyde (PFA) was mixed 1:1 with the bacterial culture and incubated for 40 min with mild shaking on a nutator mixer at room temperature. After three washes with PBS, we added 500 l of absolute ethanol to permeabilize the cells, and washed them twice with PBS. We then immobilized 20 l of cells on chitosan ( 29 ) housed in a self-adhesi v e gasket. For the FISH studies, the pre-rRNA probes (5 M) carrying the sequence [Atto488]TGCCCA CA CAGA TTGTCTGA TAAA TTGTT AAA-GA GCA GTGCCGCTT CGCT ( 13 ) wer e incubated with the permeabilized cells and incubated for 5 min at room temperature, then washed three times with PBS. Cells were then imaged as discussed below.

PCR-based measurement of the ori:ter ratio
The ori:ter ratio, which provides a measure of the rate of DNA replica tion initia tion, was measured from genomic DNA extracted from strains with a different number of rrn operons and in different media. The procedure was performed as described (30)(31)(32) with minor modifications. Briefly, cells were grown overnight in either RDM or M9Glu at 37 • C. The following morning, cells were diluted 1:100 in 30 ml of the corresponding starting medium until OD 600 ∼0.2-0.25; 20 ml of the cultures were then spun down, and the pellets were frozen a t -80 • C . The pellets were treated with 5 mg / ml lysozyme and RNase, and the DNA was extracted using the phenol-chloroform method. The purified DNA was digested by EcoRI, and 100 ng was used in the polymerase chain reaction (PCR). The origin and terminus regions were amplified utilizing oligos in gidA and dcp genes, respecti v ely, using sequences found in ( 30 , 31 ).
The PCR was set up with Sso Advanced Uni v ersal SYBR Green Supermix (Biorad) in 20 l reactions and carried out on an ABI 7500 Fast Real-Time PCR system (Thermo Scientific) under the cycling conditions in ( 31 ) with the exception that the initial denaturation temperature was kept at 98 • C for 120 s. The ori:ter ratio was calculated using the comparati v e cy cle threshold (Ct) analysis method utilizing the 2 − Ct approach ( 33 ). The data r epr esent the mean and standard error of the mean (SEM) of three technical repeats.

Single-molecule imaging of living cells
A custom-built single-molecule tracking photo-activated localization microscope ( 12 ) was used for the imaging of single RNAP-PAmCherry molecules and the detection of diffraction-limited rrn foci. Cells mounted on 1% agarose pads were imaged under bright-field illumination to perform cell segmentation. Prior to PALM imaging, preactivated PAmCherry molecules were photobleached under continuous 561-nm excitation. Sparse photoactivation of the remaining population of PAmCherry was performed by continuous exposure to low intensity 405-nm excitation, such that the dataset consists of well separated single molecules. Under simultaneous excitation with a 561-nm laser, these photoactivated molecules fluoresce until permanently photobleached. Imaging was performed at a frame rate of 15 ms / frame for at least 30 000 frames, until the entire pool of RNAP-PAmCherry molecules had been imaged.

Two-colour co-localization assay
For FISH imaging of rrn foci with fixed cells, a 488-nm laser was used for 20 frames at 500-ms exposures. Brightfield imaging and pre-bleaching with a 561-nm laser were performed as for li v e cells. Excitation of RNAP-PAmCherry molecules was performed using a 561-nm laser for 90 000 frames at 15 ms / frame to capture the entire pool of RNAP molecules.

Image processing and data analysis
Li v e-cell data were processed following published procedures using custom-written MATLAB software ( 12 ) for localizing single RN AP-PAmCherry. Briefly, individual frames of the PALM video were processed to obtain the approximate positions of molecules. The precise location of each molecule was further refined by fitting a 2D elliptical Gaussian function to each of these candidate positions. The molecular trajectories were obtained by linking these localizations in successi v e fr ames. In r are cases for w hich m ultiple localizations ar e pr esent sim ultaneousl y within this search radius, localizations were linked to their closest counterpart in the following frame. If a localization is absent from a single frame of the trajectory (e.g. as a consequence of fluorophore blinking), the localizations on either side of the empty frame were connected.
A histogram of the apparent diffusion coefficient, D*, was compiled for each dataset by computing the mean squared displacement (MSD) of individual trajectories using at least four single-step distances. Histograms of the apparent diffusion coefficient were fitted with two-gamma distributions, with the value for the immobile species fixed at a value measured using an experimental control [0.08-0.10 m 2 / s, based on DN A pol ymerase I measurements, as in ( 12 )], and the value of the mobile species left unconstrained. This fitting routine allowed us to accommodate conditions where the presence of larger amounts of chromosomal DNA in the cell (e.g. as we move from M9Glu to RDM) leads to a D* decrease due to more pronounced RN AP non-specific DN A binding [w hich effecti v ely decreases RNAP mobility; ( 34 )]. Comparison of different strains in defined growth medium was done by collecting data in triplicate. Fixed cell data were processed using ra pidSTORM as previousl y ( 9 , 35 ) to localize single RNAP-PAmCherry molecules while removing any repeated localizations in flanking frames.

Clustering analysis of RNAP localizations
Clustering analysis of RNAP molecules localized by rapidSTORM was done using a MAT-LAB implementation of the DBSCAN algorithm [see ( 9 , 36 ) and Yarpiz page: Mostapha Kalami Heris, DBSCAN Clustering in MATLAB (URL: https://yarpiz.com/255/ypml110-dbscan-clustering ), Yarpiz, 2015]. Clusters of molecules were identified by constructing a coordinate list of the first localization from each trajectory. The DBSCAN algorithm operates on this coordinate list using two parameters, ε and MinPts, to categorize the localizations into three groups. Any localization in a region containing at least MinPts localizations, including itself, within the distance ε are classified as 'core' localiza tions. Localiza tions tha t do not meet these criteria, but fall within the distance ε of a core localization, are classified as 'directly reachable', while the remaining localizations are classified as 'noise'. Using these classified localizations, a cluster of molecules is then defined as any group of connected core and directly reachable localizations. From a Monte-Carlo simulation of localizations of RNAPs in M9Glu medium and based on pr evious measur ements in fix ed cells ( 9 ), we determined the appropriate parameters for reliable clustering to be ε = 20 nm and MinPts = 4.
While the DBSCAN algorithm identifies individual clusters present in a cell, we also require a global picture of clustering for each dataset that is also ideally independent of input parameters. For a parameter-free quantitati v e description of clustering, we ther efor e use the pair correlation function, which describes how the density of molecules varies as a function of distance from a r efer ence molecule. In the case of RNAP localizations, the pair correlation function is evaluated by computing the Euclidean distances between all pairs of molecules in a single cell, and binning the results into a histogram of e v enly spaced intervals. The segmented cell boundary is then used to generate a uniform (non-clustered) distribution of the same number of molecules throughout the cell volume. The distribution obtained from the experimental result is then normalized by this simulated distribution, to produce the pair correlation function. This process of normalization eliminates artefacts from the confining geometry of the cell. To avoid projection ef fects, the simula ted molecules were distributed in a 3D volume generated by rotating the segmented cell boundary around its long axis, and then projected into 2D by removing one of these dimensions. In the resulting pair correlation function, g(r), molecules distributed e v enly throughout the cell result in a flat distribution with g(r) = 1 for all values of r, whereas a population of molecules exhibiting clustering results in g(r) > 1 at short distances, and g(r) < 1 at long distances.

P air corr elation of RNAP clusters with rrn f oci
Analysis of rrn foci in fixed cells was performed by localizing both the rrn foci and RNAP-PAmCherry molecules by applying a bandpass filter and intensity threshold to identify molecules, followed by free elliptical Gaussian fitting to obtain high-precision localizations. The pair correlation g(r) function was computed for all RNAPrrn distances within the cell. According to ( 37 ), the nucleoid area covers ∼56% of the overall cell area in M9Glu medium; therefore, we shrink both the cell length and width to ∼75% to estimate the nucleoid area. A second uniform distribution was then generated and normalized using the same distribution as the experimental data to provide a visual guide for completely uncorrelated data, shown as the dashed line in the plots of the pair correlation function. Finally, the fraction of molecules found within 200 nm of rrn foci was calculated for both the experimental and uniform data.

Heatmap plotting
Cell boundaries were determined from brightfield images using the software microbeTracker to obtain the spatial location of individual molecules relati v e to the major and minor cell axes. These spatial locations were binned into a 2D histogram normalized by the cell length and width, to produce a heatma p w hich visualizes the spatial density of molecules. Heatmaps were produced for cells containing single and double nucleoids by a ppl ying a threshold for cell length (Supplementary Figure S2). Heatmaps were further subcategorized into mobile and immobile molecules by a threshold for diffusion coefficient. Finally, a heatmap illustra ting the dif ference between the immobile and mobile heatmaps was generated by subtracting each element of the mobile heatmap from the immobile hea tmap. Hea tmaps were additionally projected along their long and short axes to illustrate the RNAP distribution throughout the cell volume.

Simulations
The 3D simulations of RNAP molecule locations were performed using Monte-Carlo methods for the WT, 5 and 6 strains. RNAP molecules were categorized into four populations; mobile; bound to rrn operons at the nucleoid periphery; bound in small clusters throughout the nucleoid; and 'noise' found throughout the cell volume. A total of 1800, 1800 and 1200 molecules were distributed between these categories in the WT, 5 and 6, respecti v ely, consistent with experimental results in Supplementary Figure S9. A split of 48% immobile, 52% mobile molecules was simulated based on experimental data presented in Figure 1 . For the methods used to obtain the estimates for growth at 37 • C, see the Materials and Methods . a Number of RNAP molecules transcribing rrn operons in the different strains. b 100% occupancy is defined as the RNAP occupancy at the maximum growth rate (72 rRNAPs / rrn ). Note that this occupancy is substantially lower than the maximal physical occupancy (see also the Discussion).
A probability density function for nonrrn -associated small RNAP clusters was obtained by fitting cluster size distributions obtained experimentally to an exponential function of the form y = ae −bx . The simulated population of nonrrn small cluster sizes are then generated by sampling from the in verse transf orm of this exponential probability density function, The centres of nonrrn -associated RNAP clusters were distributed uniformly throughout the nucleoid volume, around w hich RN AP molecules were distributed isotropicall y with a Gaussian radial density profile. Any molecules generated outside of the modelled cell volume were regenerated until a complete distribution was obtained.
The mobile population distributed uniformly within the nucleoid was generated via rejection sampling within a prolate spheroid volume positioned with a 150 nm separation between the nucleoid and cell poles. The same code was used to position the centres of small (nonrrn ) clusters. A fraction of the mobile population was di v erted to a population of localization 'noise' that was generated uniformly across the entire cell volume via rejection sampling.
rrn operons were positioned along the pole-proximal periphery of each nucleoid. Simulations were performed with se v en operons / nucleoid in the WT, two in 5 and one in 6. The location of each cluster centre along the long axis was weighted by the genomic distance of each rrn operon relati v e to oriC , defining a ring of possible locations around the nucleoid periphery. Candidate locations around this ring were proposed until a position was obtained with a minimum separation of at least 70 nm from other rrn operons. RNAP molecules were then distributed around each of these rrn cluster centres isotropically as described above for small nonrrn clusters. The proportion of immobile molecules associated with rrn operons was simulated across the range of 30-80%, with the value of 60% most closely matching the experimental data in Figure 4 . These 3D simulations were then projected into 2D, and analysed by computing the pair correlation function g(r) for each cell. The process was then automated for 2000 cells to obtain the mean g(r) for the distribution.

Estimation of the copy number of rrn operons and rRNAPs for a given growth rate
The total number of RNAPs engaged with the rrn operons (N r ) for each strain was estimated by interpolation of the N r values from ( 38 ) [which were calculated using the expression N r = r r / c r , where r r and c r are the overall rate of rRNA synthesis and the rRNA elongation speed (85 nt / s) as measured by Bremer and Dennis ( 39 )] by fitting to a single exponential. The expected number of rrn operons in all strains for a gi v en growth rate was obtained as in Bremer and Dennis [equation 9 in table 5 of ( 39 )], by considering the growth rate of each strain and the location of each rrn on the map of the E. coli chromosome.

RNAPs remain heavily engaged with the chromosome despite deletion of most rrn operons
To clarify the relationship between RNAP clusters and rrn operons, we compared a well characterized E. coli strain carrying all se v en chromosomal rrn operons ('wild type', WT) with strains carrying a drastically reduced number of rrn operons; these rrn deletion strains ( Δrrn ) were originally de v eloped to study the link between rrn operon multiplicity and ribosome function ( 40 , 41 ).
Specifically, we studied a strain in which fiv e out of se v en operons were deleted, leaving only rrnB and rrnC on the chromosome ( 5, Figure 1 B); a strain in which six out of se v en operons were deleted, leaving only rrnE on the chromosome ( 6, Figure 1 B); and a strain in which all se v en chromosomal rrn operons were deleted, and instead supplemented by a low copy-number plasmid ( ∼5 copies per chromosome) containing a single rrnB operon ( 7, Figure  1 B). To enable tracking of single RNAP molecules in cells, all strains contained a fully functional C-terminal fusion of the ␤' subunit of RNAP with a photoacti vatab le mCherry (PAmCherry ( 9 , 12 ); see also the Materials and Methods).
To check the fitness of Δrrn strains relati v e to the WT, we monitored their growth in different media (Figure 1 B; Supplementary Figure S6); in general, the growth rates in the deletion strains correlated to the number of remaining copies of rrn operons, with 6 being the slowest growing. The WT str ain display ed a 47 min doubling time at 37 • C in minimal M9 medium supplemented with 0.2% glucose (M9Glu; see the Materials and Methods); in comparison, 7 grew marginally more slowly (50 min), 5 grew significantly more slowly (57 min) and 6 grew substantially more slowly (73 min, Table 1 ). The growth rates of Δrrn strains in rich medium followed a similar pattern; in RDM, the WT was the fastest growing (36 min), followed by 7, 5 and  Figure S6). In general, reduction in the number of rrn copies led to a small to moderate decrease in the growth rate, presumably by affecting the rate at which different strains produce ribosomes ( 24 , 42 ).
To follow the RNAP mobility in li v e cells, we performed single-particle tracking of RNAP molecules using PALM on surface-immobilized cells, as described [ ( 12 ) Figure S1A, top). The immobile fraction includes RNAPs bound to the bacterial chromosome for se v er al fr ames ( 12 ), w hich in turn correspond mainl y to RNAPs bound to promoters and transcribed genes; the immobile species includes any RNAPs found in condensates, since they have been suggested to possess very low mobility ( 22 ). On the other hand, the mobile fraction corresponds to RNAPs interacting non-specifically and transiently with the entire chromosome during their promoter search ( 34 ).
To assess the effect of rrn operon loss on RNAP mobility in M9Glu, we compared the D* distribution of the WT with those of Δrrn strains (Figure 1  To assess the effect of rrn operon loss on RNAP mobility in RDM, where the rrn operons should be much more heavily occupied by RNAPs than mRNA-coding genes [and where any RNAP condensates should be more visible, potentially accounting for ∼30% of the immobile fraction; ( 22 )], we performed similar RNAP mobility comparisons between WT and Δrrn strains in RDM. As we observed before ( 12 ), the immobile RNAP fraction in the WT was ∼63% (Figure 1 E, Figure S1B, C, right). In general, for all Δrrn strains grown in rich medium, the immobile RNAP fraction stays surprisingly at the same high le v el (54-56% on average).

DN A-bound RN APs r elocate to pole-pro ximal positions in Δrrn strains in M9Glu
To examine whether the deletion of most or all chromosomal rrn operons leads to any RNAP relocation within cells, and to gain insight regarding any redistribution between cellular RNAP pools, we examined the spatial distributions of mobile and immobile RNAPs in Δrrn strains via sorting single-molecule tracks using a D* threshold [ Figure 2 A, B; see also ( 12 ) and the Materials and Methods). To capture the a verage beha viour for cells of similar size, we pooled the normalized positions of RNAPs from individual cells within different size ranges, and genera ted spa tial hea tmaps for both mobile and immobile fractions (see the Materials and Methods).
As we observed previously ( 12 ), the RNAP spatial distribution in WT cells with two nucleoids in M9Glu showed that mobile RNAPs localize throughout the nucleoid, essentially highlighting the nucleoid location (Figure 2 B, left), w hereas immobile RN APs tended to localize at the nucleoid periphery [ Figure 2 B, middle; see also ( 12 )]; this redistribution towards the periphery for immobile molecules can be seen more clearly in the normalized difference heatmap between the two mobility fractions (Figure 2 B, right). Similar r esults wer e obtained for shorter cells, which carry only a single nucleoid (Supplementary Figure S2).
We then examined strain 5 to see how the deletion of fiv e rrn operons affects the RNAP spatial distribution. The mobile RNAPs in 5 had a spatial distribution nearly identical to that of the WT, i.e. spanning the entire nucleoid (Figur e 2 C, left), r eflecting the transient, non-specific interactions of this target-searching RNAP fraction with the nucleoid ( 34 ). In contrast, the spatial distribution of immobile RNAPs in 5 is substantially different from that of the WT, with immobile RNAPs becoming much more concentrated at the pole-proximal edges of the nucleoid (Figure 2 C, middle; see also the difference heatmap, Figure 2 C, right). For a clearer view of this RNAP relocation, we projected the heatmaps along the cell length (Figure 2 Figure S3).
To explain our spatial distributions of immobile RNAPs, we need to consider that they contain se v eral RNAP pools: RNAPs transcribing rrn operons (rRN APs), RN APs transcribing mRNAs (mRNAPs) and any condensateassociated RN APs (cRN APs). Removal of se v eral rrn oper ons fr om the chr omosome essentially releases many rRNAPs; since the immobile fraction for all three Δrrn strains does not change relati v e to the WT, the released rRN APs m ust join one or mor e of the thr ee main pools of immobile RNAPs. Since the Δrrn strains do not have a substantial growth defect, it is likely that to maintain sufficient rRNA synthesis to support ribosome biogenesis, many (perhaps all) of the released rRNAPs are captured by the remaining rrn operons (see also the Discussion). Our results also show that a large fraction of immobile RNAPs in the Δrrn strains engage with the entire nucleoid ( Figure  2 C-E, middle); we attribute this fraction mainly to nonrrnassociated immobile RNAPs.

Immobile RNAPs spread throughout the nucleoid in rich medium
Since growth conditions dramatically influence the RNAP spatial distribution ( 1 ), we examined the spatial distribution in Δrrn strains growing exponentially in RDM (Figure 3 ), a condition wherein cells need to accumulate high numbers of ribosomes [up to 70 000; see also ( 41 )] and thus r equir e high rrn expression ( 43 , 44 ). In RDM, WT cells divided every ∼36 min, a growth rate that corresponds on average to ∼2.7 chromosomes / cell, 3.7 replication origins per cell and a high copy of rrn genes [ ∼22 rrn / cell; ( 38 , 39 ); Table 1 ]. The RNAP distribution in the WT strain in RDM showed that, as in M9Glu, mobile RNAPs explor e the entir e nucleoid, whereas many immobile RNAPs appear in clusters distributed throughout the nucleoid (see the next section), with some enrichment at the nucleoid periphery ( Figure 3A In contrast to its profile in M9Glu, the 5 strain shows a profile similar to the WT for both mobile and immobile RNAPs, i.e. both populations are e v enly distributed along the long axis of cells, and much of the immobile population  Figure  S5). Notably, long cells feature a more polar and 'nucleoidexcluded' localization of immobile RNAPs in 7 relati v e to the mobile population (Figure 3 E), probably reflecting the localization of most plasmids.
To explain the spatial distributions of immobile RNAPs, we consider that, during fast growth conditions in rich medium, cells contain multiple copies of the chromosome ( 9 , 45 ) and multiple sets of rrn operons, with the ori -proximal location of rrn genes further increasing the number of rrn copies ( 46 ); for example, for 5, we expect the group of long cells to have ∼3 chromosomes and ∼8 rrn copies on average ( 39 ). As in M9Glu, removal of se v eral rrn oper ons fr om the chr omosome releases many rRNAPs; since the bound fraction for all three Δrrn strains in RDM decreases only by ∼6% relati v e to the WT, the released rRN APs m ust join one or mor e of the thr ee main pools (rRN APs, mRN APs and cRN APs). We reason that, to maintain sufficient rRNA synthesis to support ribosome biogenesis (albeit at reduced growth rates), most of the released rRNAPs in Δrrn strains are re-captured by the remaining rrn operons (see also the Discussion), whereas the remainder join the mRNAP pool.
Our interpretation above is consistent with the location of RNAP clusters in single cells (e.g. 5 cells in Figure  3 A), which roughly map to the expected location of the four replication origins for cells of this size and growth rate; no-tably, the remaining rrn operons in 5 are proximal to ori . Howe v er, since the group of long cells covers a range of lengths (3.5-4.5 m), and since the location of rrn operons varies for cells of different length, the average picture for the group of long cells is blurred and features fairly continuous distributions that do not reflect the localized nature of the clusters seen in single cells. On the other hand, 7 shows a clear profile of nucleoid exclusion for a large fraction of immobile RNAPs, suggesting that these r epr esent RNAPs transcribing rrn genes on nucleoid-excluded plasmids. These rrn -centred RNAP pools are in addition to any pools of cRNAPs, although it is unclear whether cRNAP pools nucleate on the rrn -centred RNAP pools, or exist in isolation.

RNAP clustering increases upon loss of most chromosomal rrn operons
The RNAP spatial distribution in Δrrn established that RNAPs in M9Glu relocate in pole-proximal regions, raising the possibility that relocation forms new clusters or enlarges smaller ones. To assess the le v el of RNAP clustering, we performed clustering analysis using the DBSCAN algorithm [ ( 36 )  To gain another perspecti v e to RNAP clustering, we performed pair correlation analysis of the RNAP localizations ( 12 ), wherein the distances between all pairs of individual molecules are analysed and compared with a random distribution; a pair correlation g(r) value of >> 1 for a range of intermolecular distances indicates significant clustering, whereas g(r) ∼1 indicates a non-clustered distribution. Notably, the pair correlation analysis is unaffected by differ ences in RNAP cop y numbers per strain, and r equir es no optimization in analysis parameters. We first examined RNAPs in fixed cells grown in M9Glu, and observed that RNAPs in the WT show only slight clustering at distances within ∼100 nm, whereas all three Δrrn strains showed much higher clustering within the ∼100 nm range (Figure  4 C), with 7 being the most cluster ed strain. P air corr ela-tion analysis on simulated data for the two strains retaining chromosomal rrn copies ( 5 and 6; see the Materials and Methods) also showed that the WT is expected to maintain the lowest le v el of RNAP clustering relati v e to 5 and 6 (Supplementary Figures S11 and S12), consistent with our experimental results.
We also performed pair correlation analysis in li v e cells in M9Glu. These experiments are complicated by any 3D motions of clustered RNAPs during the ∼8 min of imaging; such motions will reduce the pair correlation and spread it out to longer length scales; howe v er, any persistent clustering should still be visible. Since we can separate the mobile and immobile RNAP species, we performed pair correlation analysis of the two species separately. Since we do not anticipate mobile RNAPs to be clustered (apart from exploring the entire nucleoid; as such, they do not fill the entire cell), this analysis should offer clearer views of the clustering of immobile RNAPs. Indeed, mobile RNAPs for all strains do not cluster (Figure 4 D, dotted lines); in contrast, the immobile RNAPs of both WT and Δrrn strains ( Figure  4 D, solid lines) appear much more clustered than mobile RNAPs. Further, all strains show a similar le v el of clustering for immobile RNAPs. Taken together, our results indica te tha t in M9Glu, RNAPs become mor e cluster ed, consistent with the remaining rrn operons in Δrrn strains accommoda ting reloca ted RNAPs to compensa te the loss of many chromosomal rrn operons.
In rich medium, RNAPs in fixed cells of the WT are more clustered than in M9Glu, whereas, in contrast, RNAPs in all Δrrn strains show reduced clustering relati v e to their levels in M9Glu, and relati v e to the WT in RDM ( Figure 4E; see Supplementary Figure S10 for the number of RNAPs per cell in RDM). P air corr elation analysis on the immobile RNAP molecules in li v e cells shows similar differences between the WT and the Δrrn strains (Figure 4 F); further, the le v els of clustering in all strains e xceed significantly the clustering seen in M9Glu. This result reinforces our qualitati v e observations of clustering in the discussion of the spatial RNAP distribution in RDM; the absence of prominent peaks in the projection of immobile RNAP localizations in the Δrrn strains is not due to the presence of highly distributed immobile RNAPs, but rather to the presence of RNAP clusters with variable positions along the long cell axis (due to the lack of synchronization of the cells, and, in turn, due to variable positions of the remaining rrn operons).

The expected cellular rrn copy number in Δrrn strains can sustain the measured growth rates in M9Glu, but not in RDM
To evaluate whether the remaining rrn copies can accommodate the number of RNAPs needed to sustain the measured growth rate [which is proportional to both cellular ribosome content (hence to the rRNA cellular content) and peptide elonga tion ra te ( 47 )], we estima ted the copy number of rrn operons ( 39 ) and of rrn -associated RNAPs ( 38 ) for the growth rates of our strains (Table 1 ), and used them to estimate the average fractional occupancy of each rrn operon (Table 1 ).
Due to the ongoing process of DNA replication in bacterial cells, and due to the une v en (and highly ori -proximal) distribution of the rrn operons on the chromosome, we expect for the WT strain (doubling time T of ∼47 min in M9Glu) an average of ∼2 genome equivalents, ∼17.5 rrn operons per cell ( 44 ) and ∼400 rRNAPs ( 40 ), yielding an average of ∼23 RNAPs / rrn . Considering a maximum rrn occupancy of ∼72 RNAPs [using the average rrn occupancy at the maximal E. coli growth rate of 24 min; ( 38 )], the operons function at only ∼32% of their full capacity, hence being far from sa tura tion and having significant spare capacity. We note that the value of ∼72 RNAPs / rrn is the RNAP occupancy at maximal growth rate (set by cell physiology), and not the maximal RNAP occupancy dictated by the physical RNAP footprint on the DNA (see the Discussion).
Regarding 5 (T ∼57 min), we expect to have ∼240 RNAPs engaged in rrn transcription and ∼5 rrn operons per cell, leading to an estimate of ∼50 RNAPs per operon, and ∼70% of max occupancy; e v en if we use our experimental result of ∼3.2 rrn foci / cell (see the last Results section), and assume conservati v el y that a focus contains onl y one rrn operon, we recover an upper bound that does not exceed the rrn operon capacity. These estimates strongly suggest that the remaining rrn genes in 5 can accommodate the number of RNAPs r equir ed for the observed growth r ate. Similar ly, e v en for 6, the doubling time of 73 min can be maintained by an ∼85% occupancy of the remaining ∼2 rrnE copies per cell. In essence, the rrn transcription r equir ements for the growth rates of the deletion strains in M9Glu can be fulfilled by relocating RNAPs to the remaining rrn operons. Regarding 7, which has a growth rate similar to that of the WT, the presence of ∼10 copies of the rrn -containing plasmid [pK4-16, based on pSC101, see also ( 28 )] also provides enough rrn copies to sustain the r equir ed le v els of rRNA.
A mor e complex pictur e emerges for the Δrrn strains in rich medium. In RDM, the WT (T ∼36 min) has ∼22 rrn operons per cell and ∼800 rRNAPs, with an average of ∼37 RN APs / rrn , w hich corresponds to opera ting a t ∼50% full capacity. As in the case of M9Glu, since we estima te tha t 7 has many copies ( ∼15) of the rrn B-containing plasmid, it can maintain high rrn transcription le v els despite the loss of all chromosomal rrn ; indeed, 7 shows the smallest fitness cost (increase in doubling time) amongst the Δrrn strains.
Howe v er, the fact that 7 appears to show RNAP engagement that is closer to that of 5 and 6 (which have a lot fewer rrn copies) indicates that the RNAP association and / or transcription of rrn on the plasmid is not as effecti v e as on the chromosome.
Howe v er, in the case of 5 (T ∼46 min, ∼420 RNAPs), we expect ∼5.6 rrn per cell based on the measured growth rate, hence ∼75 RNAPs / rrn , which exceeds the maximum capacity by ∼5%; this means that 5 is at the limit of being able to sustain growth by fully loading all remaining rrn operons with RNAPs. This limit is substantially exceeded in 6, the slowest growing strain in RDM ( ∼51 min), which r equir es ∼320 rRNAPs to maintain its growth rate, corresponding to ∼130 RNAPs / rrn . This result strongly suggests that mechanisms other than simple RNAP relocation to the number of rrn copies expected purely on the basis of growth rate are needed to explain the ability of 6 (and, possibly, of 5) to sustain the observed growth rate in RDM . rrn transcription in rich medium for 6 is associated with reduced operon copy number per cell Our estimates of RNAP occupancy of rrn in RDM clearly showed tha t, a t least for the 6 strain, the cells cannot sustain the measured growth rate purely on the basis of the cellular number of rrn operons expected for the measured growth rate. To address this, we first tested the hypothesis that more rrn copies are generated due to increased replication initiation frequency.
To examine whether the replication initiation frequency is affected in 6 relati v e to the WT, we performed quantitati v e PCR (qPCR) measurements of the ori : ter ratio, which acts as a proxy for the rrnC : ter ratio. The results ( Figure  5 A) clearly showed that the 6 strain grown in RDM shows a significantly lower ori : ter ratio compared with the WT ( ∼2.6 versus ∼7), and established that increased replication initiation cannot explain the high degree of RNAP engagement with the chromosome. The cell length distribution also showed that 6 cells are significantly shorter than those for the equivalent WT strain ( ∼2.8 versus ∼3.6 m, respecti v ely; Supplementary Figure S13).
The profile was very different for 5, which instead showed a small increase in the ori:ter ratio (from ∼7 to ∼8.4, albeit not statisticall y significant), w hile having a mean cell length similar to the WT strain ( ∼3.8 versus ∼3.6 m; Supplementary Figure S2), and a slo wer gro wth rate than the WT. These results suggest some increase in replica tion initia tion in 5, but any ef fect is fairly modest.
We also examined the WT, 6 and 5 strains in M9Glu ( Figure 5 ), and found that they all had an ori:ter ratio of ∼1.5, suggesting that when the n ucleoid n umber is small, loss of multiple rrn operons in the deletion strains does not lead to any significant changes in replication initiation, consistent with our expectations.

Tw o-colour imaging r eveals that RNAP clusters correspond to rrn foci
To provide direct evidence for the link between RNAP clusters and rrn operons in the Δrrn strains, we performed twocolour co-localization assays by combining FISH imaging of rrn foci with single-molecule RNAP localization in 5. Similarly to pub lished wor k ( 13 ), which showed that RNAP clusters co-localize with nascent rRNA in WT cells grown in rich medium, we used a fluorescent FISH probe that targeted the 5 leader region of the 16S precursor rRNA (pre-rRN A), w hich is absent from mature rRN A and ribosomes (see the Materials and Methods). Signal from our FISH probe in fixed cells in M9Glu allows us to capture transcribing rrn foci, visualized as bright diffraction-limited spots in pole-proximal regions (Figure 6 A). When used in conjunction with PALM data, the relati v e co-localization identified acti v e rrn clusters in 5 and estimated their copy numbers in M9Glu and RDM to be ∼3.2 ± 0.1 and ∼6.8 ± 0.2, respecti v ely (Figure 6 D). To record the position of rrn foci, its centroid was determined by a Gaussian fitting and superimposed on RNAP localizations and RNAP clusters; we observed that most RNAP clusters (especially large clusters containing > 50 localizations; C > 50) locate within 200 nm from the rrn centroid (Figure 6 A), suggesting significant colocalization between RNAP clusters and rrn foci.
To quantify the degree of co-localization, we performed pair correlation analysis between the rrn loci and RNAP clusters (either using all clusters with N ≥ 4, or large C > 50 clusters). Our results show a high correlation between rrn foci and RNAP clusters, supporting their co-localization. Specifically, we found that ∼46% of all RNAP clusters, and ∼77% of RNAP large clusters localize within 200 nm of rrn foci, compared with ∼23% expected on the basis of simulated random RNAP localizations (Figure 6 B), which employs similar analysis algorithms to our pair correlation analysis in Figure 4 .
To visualize the distribution of RNAP clusters, we generated the heatmap of RNAP clusters from the normalized positions of clustered RNAPs in 5 grown in 9Glu (Figure 6 C). Our results clearly show that RNAP clusters are concentra ted a t pole-proximal r egions, a fact also r eflected in projections along the cell length axis. The projection of the normalized positions of rrn foci also display pole-proximal peaks, w hich highl y overla p with the peaks of RNAP clus- ters (Figure 6 C). These results clearly establish the physical proximity of the RNAP clusters and rrn foci, and further support the suggestion that RNAPs relocate to the remaining rrn operons to sustain high le v els of rrn transcription and largely maintain the growth rate achie v ed in the absence of any rrn deletions.

DISCUSSION
The spa tial organiza tion of RNAP in bacteria has been a long-standing question e v er since the first observations of transcription foci in cells grown in rich medium ( 1-3 , 8 , 48-50 ), and the linkage between transcription foci and rRNA synthesis has remained controversial ( 1 ). Here, we applied super-resolution imaging and single-molecule tracking on strains with a heavily reduced number of rrn operons to elucidate the relationship between RNAP spatial organization and rRNA synthesis, and study how cells redeploy their transcription machinery to sustain a healthy growth rate with only one or two chromosomal rrn operons. Notably, most bacterial species ( ∼80%) have 1-4 rrn copies in their genome, with ∼35% having just 1-2 copies ( 51 ); a large number of rrn copies enables the provision of high numbers of ribosomes per cell, which in turn allows bacteria harbouring a large rrn number to adapt more quickly to nutritional upshifts and switch to fast growth (and, in general, r espond mor e ra pidl y to changes in the nutrient availability) ( 51 ).

RNAPs maintain their chromosome engagement in Δrrn str ains by incr easing the loading of the r emaining rrn operons
Our RNAP mobility analysis showed that the fraction of immobile RNAPs, a proxy for the fraction of RNAPs engaged in transcription (plus any RNAPs involved in condensates), is surprisingly robust to the loss of fiv e and six chromosomal rrn copies, as well as to the loss of all chromosomal rrn copies when cells are supplemented by a low copy-number plasmid harbouring a single rrn operon. In M9Glu, a medium that supports a doubling time of 47 min in the WT, about half of all RNAPs were immobile both for the WT [as seen in ( 12 , 52 )] and for all Δrrn strains. Even in rich medium (RDM; supporting a doubling time of 36 min in the WT), heavy loss of rrn operons led to only a modest decrease in the immobile RNAP fraction (63% for the WT; 57-58% for the Δrrn strains).
The robustness of the RNAP immobile fraction to the loss of most chromosomal rrn genes raises the question of how RNAPs 'released' from the deleted rrn operons redistribute to other immobile fractions, and how this redistribution minimizes any growth rate defects in Δrrn strains. Under our growth conditions in the WT strain (Figure 7 , top), we expect that rrn promoters are not sa tura ted with RNAP; as a result, an increase in the concentration of available free RNAP should lead to increased rrn promoter activities ( 47 ). Our results support a scenario where the remaining rrn copies in the cell are more heavily loaded by RNAPs (Figure 7 , middle).
Such RNAP re-distribution had been observed in a study of a more limited rrn deletion in rich LB medium; specifically, deletion of four rrn operons ( 4) resulted in the remaining rrn operons accommodating ∼71 RNAPs, an increase from 53 RNAPs / rrn in the strain with all operons intact, with the increased occupancy being linked to increased transcription initiation and elongation ( 40 ). Notably, the gro wth slo w-do wn seen in 4 ( ∼24 min for WT versus ∼30 min for 4) is similar to that we see for 5 in RDM.
Our observations are consistent with the 'sa tura tion model' for the passi v e regulation of gene expression ( 47 , 53 ), and with studies showing that a 2-to 3-fold ov ere xpression of the RNAP-70 holoenzyme leads to a 2-fold increase of rrnB P1 transcription, and a large increase in the tendency to transcribe rrn genes versus mRNA genes ( 54 ). In general, our results clearly show that the rrn promoters are competing very effectively against mRNA promoters.

RNAP clusters do form on rrn operons
Our results clearly establish that RNAPs are more clustered in the Δrrn strains, and that RNAP clusters are in physical proximity to the rrn foci, adding further support to the proposal that RNAPs relocate to the remaining rrn operons to sustain high le v els of rrn transcription and largely maintain the growth rate achie v ed in the absence of any rrn deletions. Our results are consistent with results from Weng et al . ( 13 ), where it was shown that large RNAP clusters are maintained in a strain with a single rrn on the chromosome in rich medium, as well as during low le v els of transcription ( 13 ). The presence of clustering on rrn (rRNAPs) does not exclude the presence of other forms of clustering, such as condensates (cRNAP) or heavy transcription on mRNA genes. It demonstrates, howe v er, the ability of RNAP to redistribute and reprogram gene expression due to the cellular response to changes in the local environment or chromosome context ( 1 , 13 ).

The location of remaining rrn operons dictates the location of the clusters in M9Glu
Considering the genomic map (Figure 1 B), the remaining rrn operons in the deletion mutants are either near ori ( 40 ), which is situa ted a t pole-proximal regions along the cell long axis ( 45 ), or on plasmids known to localize pr efer entially at the polar endcaps; these positions are consistent with the new peak of localizations (along the long cell axis) that appears in all three Δrrn strains (Figure 2 D). Relocation of released immobile RNAPs in M9Glu in Δrrn to pole-proximal positions is thus dictated by the places of the remaining rrn operons.
The presence of a well-defined location for the relocated rRNAPs in the case of M9Glu makes it unlikely that the r eleased rRNAPs r elocate to transcribe mRNA, since such a relocation would have resulted in a much more e v enly distributed spatial profile for RNAP. Equally, our results are not consistent with rRNAP relocation to any RNAP- coli strain with all se v en chromosomal rrn copies carries a large number of rrn copies ( > 15 copies) per cell either in M9Glu or in rich medium, with the operons being far from sa tura tion (30-50% of the occupancy seen in cells growing at a maximal rate) in both media, and RNAP clusters forming on rrn spreading throughout the entire nucleoid. In contrast, in rrn deletion mutants lacking most rrn copies from the chromosome, the remaining rrn copies are much more occupied by RNAPs ( ∼70% in 5 in M9Glu), and RNAP clusters are more concentrated in the pole-proximal positions. More demanding growth conditions with regard to rRNA resources, such as in rich medium, maximize the RNAP occupancy on the remaining rrn operons, while they also increase the transcription of some mRNA genes that deal with the consequences of replication-transcription conflicts in heavily transcribed rrn operons, as manifested in the 6 strain grown in rich medium.
containing condensates not associated with transcription, since it is unlikely that such condensates will have the same spa tial pa tterns as those dictated by the locations of the remaining chromosomal rrn and the plasmid-borne rrn operons. In essence, our results suggest that it is the clustered rRNAPs that recruit cRNAPs (at least in the M9Glu case), and not the other way around.

Heavy loss of rrn operons in rich medium leads to RNAP redistribution to both rRNA and mRNA genes
Our measurements of the ori:ter ratio indicated that in 6 in rich medium, mechanisms additional to RNAP relocation to the remaining rrn operons are necessary, since the number of rrn operons is reduced by ∼3 fold. The reason for this decrease is likely to be high le v els of b locked DNA replication and high DNA damage due to transcriptionreplication conflicts in the remaining rrn operon, as shown in a strain similar to 6 ( 55 ); the same work also showed that such a strain was linked with induction of the SOS response (activation of ∼40 genes). Given our results and the work of Fleurier et al . ( 55 ), it is highly likely that 6 cells in RDM are in a stressed / DNA-damaged state, and, in this state, they re-distribute many RNAP molecules to nonrrn genes, in order to balance a moderate growth with the need to minimize DNA damage and repair their DNA.
Our initial estimates predicted that ∼320 RNAPs were needed to produce the ∼20 000 ribosomes needed to sustain the growth rate of 6 ( 39 ); this number of rRNAPs corresponds to ∼128 RNAPs per rrn (gi v en the e xpected number of rrn copies for the 6 growth rate; see Table 1 ). This number exceeds substantially the number of ∼72 RNAP / rrn observed at maximal growth and constrained by cell physiology. Howe v er, the maximal physical RNAP occupancy is considerably higher. The footprint of elongating RNAP on DN A is onl y 25-35 bp [as determined by DNase I footprint-ing ( 56 )]; further, the amount of DNA covered by RNAP in structures of elongation complexes is ∼30 bp ( 57 ). If we instead use a footprint of 40 bp per elongating RNAP, this will yield a maximal physical occupancy of 135 RNAPs / rrn , which would in principle be able to support the 6 growth rate in RDM. Howe v er, such a dense occupancy is clearly toxic for cells, since it leads to substantial DNA damage ( 55 ). This also means that the maximal growth rate RNAP occupancy is kept substantially lower than the physical one to avoid problems with genomic instability.
Notably, the profile in rich medium is very different for the 5 strain, which presumably avoids the high le v els of DNA damage seen in 6 by having two chromosomal rrn copies and benefiting from high gene dosage ( ori:ter ratio of 8.4 and 1.5 for 5 in RDM and 9Glu, respecti v ely). Since the rrn operons will be close to full occupancy in RDM, we also specula te tha t they will be associated with co-directional transcription-replication conflicts and increased replication restart, as has been reported previously for highly transcribed rrn operons in B. subtilis ( 58 ).

DA T A A V AILABILITY
Movies and images of cells as well as localization files for single molecules will be available upon request.