Spatial Organization and Dynamics of Rnase E and Ribosomes in Caulobacter Crescentus

Spatial organization and dynamics of RNase E and ribosomes in Caulobacter crescentus

Camille A. Bayasa, Jiarui Wanga,b, Marissa K. Leea, Jared M. Schraderb,1, Lucy Shapirob,c, and W. E. Moernera,2

aDepartment of Chemistry, Stanford University, Stanford, CA 94305; bDepartment of Developmental Biology, Stanford University School of Medicine, Stanford, CA 94305; and cChan Zuckerberg Biohub, San Francisco, CA 94158

Contributed by W. E. Moerner, March 5, 2018 (sent for review December 15, 2017; reviewed by Zemer Gitai and James C. Weisshaar) We report the dynamic spatial organization of Caulobacter cres- components of the degradosome or the ribosome that are on centus RNase E (RNA degradosome) and ribosomal protein L1 (ri- length scales of less than ∼200 nm. bosome) using 3D single-particle tracking and superresolution We used a combination of live-cell single-particle tracking microscopy. RNase E formed clusters along the central axis of the (SPT) and fixed-cell superresolution (SR) microscopy to study the cell, while weak clusters of ribosomal protein L1 were deployed dynamics and spatial distribution of eYFP-labeled (13) RNase E throughout the cytoplasm. These results contrast with RNase E and ribosomal protein L1 in Caulobacter on subdiffraction limit and ribosome distribution in Escherichia coli, where RNase E coloc- length scales. SPT and SR provide improved resolution down to alizes with the cytoplasmic membrane and ribosomes accumulate in ∼20–50 nm using fluorescent proteins (FPs) (14, 15). Moreover, Cau- polar nucleoid-free zones. For both RNase E and ribosomes in both SPT and SR are single-molecule (SM) methods, allowing us lobacter , we observed a decrease in confinement and clustering to investigate the heterogeneity in protein behavior and distribu- upon transcription inhibition and subsequent depletion of nascent tion across many different cells. To avoid artifacts from projecting RNA, suggesting that RNA substrate availability for processing, deg- SM localizations onto two dimensions, we have used the double- radation, and translation facilitates confinement and clustering. Im- helix point spread function (DH-PSF), which encodes 3D infor- portantly, RNase E cluster positions correlated with the subcellular mation in each SM image (16). location of chromosomal loci of two highly transcribed rRNA genes, Using these methods, we observed that the number of RNase suggesting that RNase E’s function in rRNA processing occurs at the site of rRNA synthesis. Thus, components of the RNA degradosome E clusters changed as a function of the cell cycle, while the size of and ribosome assembly are spatiotemporally organized in Caulo- the clusters remained constant. However, ribosomes formed only bacter, with chromosomal readout serving as the template for weak clusters, which were deployed throughout the cell, as we α this organization. found to be the case in another -bacterium, Sinorhizobium meliloti. For both RNase E and ribosomes, we observed a decrease ribosomes | RNA degradation | RNA processing | superresolution in confinement and clustering when transcription was inhibited, microscopy | single-molecule tracking likely due to depletion of RNA substrates for degradation, processing, and translation. We found that the localization of RNase E n bacteria, the RNA degradosome mediates the majority of clusters correlates with the subcellular chromosomal position of the ImRNA turnover and rRNA and tRNA maturation (1). The RNA degradosome assembles on the C-terminal scaffold region Significance of the RNase E endoribonuclease (2). RNase E in Escherichia coli and RNase Y, the RNase E homolog in Bacillus subtilis, both Regulated gene expression involves accurate subcellular posi- associate with the cell membrane through a membrane-binding tioning and timing of RNA synthesis, degradation, processing, helix (3–6). One proposed rationale behind the observed mem- and translation. RNase E is an endoribonuclease that forms the brane association is to physically separate transcription within scaffold of the RNA degradosome in bacteria, responsible for the nucleoid from RNA degradation, thus providing an inherent the majority of mRNA turnover and RNA processing. We used time delay between transcript synthesis and the onset of tran- 3D superresolution microscopy and single-particle tracking to script decay, avoiding a futile cycle. Caulobacter crescentus (Cau- quantify the spatial distribution and dynamics of RNase E and lobacter) is a Gram-negative α-proteobacterium widely studied as ribosomes in the asymmetrically dividing bacterium Caulo- a model system for asymmetric cell division. Unlike E. coli or B. bacter crescentus. Our results show that active transcription subtilis, Caulobacter contains a pole-tethered chromosome that and RNA substrate availability facilitate confinement and fills the cytoplasm (7–9). Surprisingly, Caulobacter RNase E does clustering of both RNase E and ribosomes. RNase E clusters not have a membrane-targeting sequence and is not membrane colocalized with the subcellular position of the two Caulo- associated (2, 10). Furthermore, diffraction-limited (DL) images bacter rRNA gene chromosomal loci, indicating that RNA pro- of RNase E exhibited a patchy localization throughout the cell, cessing can be spatially organized in a bacterium according to with RNase E directly or indirectly associating with DNA (10). its transcriptional profile. Moreover, the copy number of RNase E is regulated as a function of Author contributions: C.A.B., J.W., M.K.L., J.M.S., L.S., and W.E.M. designed research; the Caulobacter cell cycle. The copy number reaches maxima during C.A.B., J.W., and M.K.L. performed research; C.A.B., J.W., M.K.L., and J.M.S. contributed the swarmer to stalked cell transition and before cell division (2). new reagents/analytic tools; C.A.B., J.W., M.K.L., and W.E.M. analyzed data; and C.A.B., Fluorescently labeled ribosomal proteins S2 and L1, proxies J.W., L.S., and W.E.M. wrote the paper. for ribosomes in E. coli and B. subtilis, respectively, were found Reviewers: Z.G., Princeton University; and J.C.W., University of Wisconsin–Madison. enriched at the cell poles, spatially excluded from the nucleoid The authors declare no conflict of interest. on the hundreds of nanometer scale (11, 12). Similar fluores- Published under the PNAS license. cence imaging studies in Caulobacter revealed no apparent sep- 1Present address: Department of Biological Sciences, Wayne State University, Detroit, aration of ribosomes and the nucleoid, but did suggest that MI 48202. ribosome diffusion throughout the cell was spatially confined 2To whom correspondence should be addressed. Email: [email protected]. (10). However, because the Caulobacter cell is only ∼500 nm by This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. ∼3.5 μm in size, the DL resolution of conventional fluorescence 1073/pnas.1721648115/-/DCSupplemental. microscopy would obscure any organization or motion of the Published online April 2, 2018.

E3712–E3721 | PNAS | vol. 115 | no. 16 www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Downloaded by guest on September 24, 2021 highly expressed rRNA genes that are positioned at two distinct and 0.1944 ± 0.0010 μm2/s (mobile). For rif-treated cells, the PNAS PLUS sites on the chromosome, consistent with processing occurring at diffusion coefficient for confined subtrajectories is 0.0060 ± the site of rRNA synthesis. Thus, cotranscriptional activities exhibit 0.0001 μm2/s and for mobile subtrajectories is 0.1270 ± dynamic subcellular organization that is dependent on the site of 0.0072 μm2/s. To determine whether the confined subtrajectories chromosomal readout. are stationary or not, the apparent diffusion coefficient of stationary RNase E-eYFP molecules was calculated using the Results ensemble-averaged MSD of trajectories from fixed cells. Stationary, Diffusion of RNase E and Ribosomal L1 Proteins Is Influenced by fixed particles were found to have an apparent diffusion coefficient Transcriptional Activity. of 0.0022 ± 0.0015 μm2/s (error is SEM from 143 trajectories). RNase E. The endoribonuclease RNase E forms the core and Using this comparison, we determined that the observed diffusion scaffold of the Caulobacter RNA degradosome. Other compo- coefficient for WT confined subtrajectories is consistent with a nents include the exoribonuclease polynucleotide phosphorylase, stationary particle, whereas the diffusion coefficientforrifconfined a DEAD-box RNA helicase RhlB, the Krebs cycle enzyme aco- subtrajectories is larger than would be expected for a stationary nitase, and the exoribonuclease RNase D (2, 17). To study the particle given our localization precision (32/33/49 nm in x/y/z). motional dynamics of RNase E, we performed 3D SPT in a mixed The difference in diffusion between WT and rif-treated cells is population of live Caulobacter cells using a strain expressing not an artifact due to a potential rif-induced chromosome struc- RNase E-eYFP as the only copy of RNase E. We performed an ture disruption (22). We performed the same 3D SPT of RNase E initial photobleaching and waited for stochastic recovery and molecules in a ΔHU2 background with significantly reduced nu- blinking of eYFP SMs (18) (Fig. 1A, Left). We then obtained 3D cleoid compaction (23). Results show similar confinement as that trajectories of individual molecules by joining DH-PSF localiza- of WT cells (SI Appendix,Fig.S6), arguing against the possibility tions over consecutive frames (Materials and Methods) (Fig. 1A, of a difference in chromosome packaging between WT and rif- Right). In contrast to E. coli in which RNase E diffuses on the treated cells contributing to the difference in RNase E diffusion. membrane (3), RNase E molecules in Caulobacter (average copy Ribosomes. To obtain information about ribosome dynamics and number is 3,100 molecules per cell as determined for a mixed cell translation, we performed 3D SPT of ribosomes through imaging population; SI Appendix, Supplementary Methods)werefound eYFP-labeled ribosomal protein L1 in a mixed population of throughout the cytoplasm of the cell, diffusing limited distances cells. We found that L1 (average copy number is 16,300 mole- (Fig. 1 B and D) and exhibiting the possibility of confined diffusion. cules per cell as determined for a mixed cell population; SI

To study the effect of RNA substrate availability on RNase E Appendix, Supplementary Methods) diffuses much more freely BIOPHYSICS AND

dynamics, we treated the cells with rifampicin (rif), an antibiotic throughout the cytoplasm than RNase E (Fig. 2 A and C and SI COMPUTATIONAL BIOLOGY that blocks transcription and effectively depletes cells of RNA Appendix, Fig. S7, for additional examples). In contrast to the (19). The ensemble diffusion coefficient for wild-type (WT) RN- case for RNase E, multiple populations of diffusing molecules ase E was extracted by analysis of the ensemble-averaged mean- were present by visual inspection, and we chose to differentiate squared displacement (MSD) (Materials and Methods)(SI Ap- the various populations using another approach. We calculated pendix, Fig. S1A). The average diffusion coefficient for WT cells the squared displacements of the trajectories using a time lag of was found to be 0.0337 ± 0.0011 μm2/s (error is SEM as de- one frame (25 ms), then computed the empirical cumulative termined from 500 bootstrapped samples of individual tracks). distribution function (CDF) of these squared displacements. The The diffusion coefficient increased upon treatment with rif empirical CDF was then fit to a standard two-population model (0.0407 ± 0.0004 μm2/s) (Fig. 1C). There is a heterogeneity of (Materials and Methods) (Fig. 2 D and E and SI Appendix, Fig. diffusion coefficients across different trajectories (SI Appendix, S8). We found easily observable diffusion: in WT cells, 89% of Fig. S2) likely due to sampling a mixed population of cells in the population was found to have a diffusion coefficient of various developmental stages, stochastic gene expression between 0.039 ± 0.001 μm2/s (larger than would be expected for a sta- individual cells, and RNase E molecules not all acting on RNA (SI tionary particle) (error is SEM as determined from 500 boot- Appendix,Fig.S3, shows additional example cells). strapped samples of individual squared displacements), while As the change in the ensemble diffusion coefficient was not 11% of the population was found to have a much larger diffusion drastic, we analyzed individual trajectories of WT and rif-treated coefficient of 3.23 ± 0.12 μm2/s. We thus hypothesize that our cells to quantify their degrees of confinement. Each trajectory was slow-diffusing population (89% of trajectories measured) rep- divided into subtrajectories of five points (Materials and Methods). resents ribosomes involved in active translation slowed down by The maximum displacement from the first point of each sub- protein and mRNA tethers. The faster-diffusing population trajectory was then calculated (20, 21) (Materials and Methods) (11% of total trajectories) are likely unincorporated L1 subunits. (Fig. 1E). Displacements were compared with simulated Brownian When cells were treated with rif (Fig. 2 B and C), 95% of the motion inside a cell volume using ensemble WT and rif diffusion L1 population was found to have a diffusion coefficient of 0.215 ± coefficients (SI Appendix,Fig.S4) to control for the possibility that 0.002 μm2/s, compared with untreated cells where 89% of the trajectories appear confined by chance or due to the cell shape. A population had a diffusion coefficient of 0.039 ± 0.001 μm2/s (Fig. threshold was set at the Brownian 5th percentile, such that only 2E). In rif-treated cells, 5% of the L1 population was found to 5% of the simulated Brownian subtrajectories were confined. have a diffusion coefficient of 5.67 ± 0.39 μm2/s (Fig. 2E). Both Subtrajectories with displacements less than this value were clas- populations in rif-treated cells have diffusion coefficients that sified as confined. To include the possibility of an intermediate were greater than observed for WT cells. We did not detect the degree of confinement, another threshold was arbitrarily set at the actively translating population (89%, D = 0.039 ± 0.001 μm2/s) Brownian 50th percentile, and subtrajectories with displacements observed in WT cells. The 95% L1 population in the rif-treated greater than this value were classified as clearly mobile. All other cells could be attributed to fully assembled ribosomes that are not subtrajectories (between 5th and 50th percentile) were classified tethered to an mRNA, whereas the 5% fraction are unincorporated as intermediate. Upon rif treatment, the fraction of confined L1 subunits. The reason for the faster movement of the un- molecules decreased from 50.1% in WT subtrajectories to 34.2%, incorporated subunits in the rif-treated cells compared with WT cells while the fraction of mobile subtrajectories increased from 15.6% is unclear and may be due to a change in overall cell viscosity. in WT to 22.8% (Fig. 1 F–H). Ensemble-averaged MSDs for the confined and mobile WT RNase E and Ribosomal L1 Proteins Are Differentially Clustered in subtrajectories were analyzed (SI Appendix, Fig. S5), yielding Caulobacter. As our SPT data have shown, RNase E and ribo- drastically different D values of 0.0025 ± 0.0001 μm2/s (confined) somes both display dynamic behaviors in the cell. To study their

Bayas et al. PNAS | vol. 115 | no. 16 | E3713 Downloaded by guest on September 24, 2021 A

white light diffraction-limited time (25 ms/frame)

B WT C + rifampicin

1 3

2 4

D 1 3 4 2 y z x

EF 1 1

0.8 0.8 **** 0.6 WT - data 0.6 WT - simulation 0.4 WT - confinement threshold 0.4 rif - data **** 0.2 rif - simulation 0.2 rif - confinement threshold Fraction of subtrajectories Fraction of subtrajectories 0 0 0 250 500 750 1000 WT rif Displacement from first point of subtrajectory (nm) confined intermediate mobile

G WT H + rifampicin

1 3

2 4

Fig. 1. SPT of RNase E molecules measures decreased confinement when transcription is inhibited by rif. (A, Left) Representative white-light (WL) and diffraction-limited (DL) images of a live Caulobacter cell expressing RNase E-eYFP. The DL image displayed was taken during the initial bleach down of eYFP. (A, Right) Example images of SMs in subsequent frames used to produce a SM trajectory. The DH-PSF experimental raw data (Top row) matches well to the fit to two 2D Gaussians done through easy-DHPSF (Bottom row). The midpoint of the two lobes provides xy position, while the angle between the two lobes encodes axial position. (B) Two representative WT cells with their RNase E trajectories. (C) Two representative rif-treated cells and trajectories. Each trajectory is plotted as a different color. Only trajectories of at least four steps (five frames or 125 ms) are shown and analyzed. The cell outline is indicated by the dashed brown line. (D) Three-dimensional perspective views of the trajectories in B and C.(E) CDF of displacements used for confinement analysis. Thresholds for classifying a subtrajectory as “confined” or “mobile” were set each for WT and rif-treated cells, and were chosen to be the 5th percentile of the Brownian simulations for confined (pink and light blue dashed lines) and 50th percentile for mobile. All other subtrajectories are defined as “intermediate.” (F) Results from confinement analysis. (G and H)SamecellsasinB and C but with the subtrajectories color-coded according to their assigned confinement type. Analysis was performed on 2,817 trajectories from 302 WT cells and 2,709 trajectories from 351 rif-treated cells. [Scale bars: 500 nm (A and D); 1,000 nm (B, C, G,andH).]

spatial distribution without the confounding effect of motion, we obtain SR images with sufficient sampling is longer than the turned to SR imaging of fixed cells. The imaging time required to persistence time of individual mRNAs (24, 25). Accordingly, the

E3714 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al. Downloaded by guest on September 24, 2021 PNAS PLUS AB

DE BIOPHYSICS AND COMPUTATIONAL BIOLOGY

Fig. 2. SPT of single ribosomal L1 molecules reveals faster diffusion in cells in which RNA has been depleted, as well as free and ribosome-bound L1. (A) Example WT cells and their trajectories. (B) Example rif-treated cells and their trajectories. Each trajectory is plotted as a different color. Only trajectories of at least four steps (five frames or 125 ms) are shown and analyzed. (C) Three-dimensional perspective views of the trajectories in A and B.(D, Top) Semilog plot of the CDF of squared displacements. (D, Bottom) Residuals of the fit to the CDF of squared displacements. (E) Summary of the two diffusing populations in WT and rif-treated cells. Errors are SDs from 500 bootstrapped samples. [Scale bars: 1,000 nm (A and B); 500 nm (C).] Analysis was performed on 11,249 trajectories from 298 WT cells and 12,757 trajectories from 326 rif-treated cells.

cells were fixed with 1% formaldehyde to capture global RNase transcription was inhibited by rif, RNase E clusters dissipated and E or ribosome locations at a specific time window. Because the the observed clustering shifted down toward complete spatial photophysical properties of eYFP (stochastic and reversible re- randomness (Fig. 3F and Movie S2). To compare WT and rif- covery from a long-lived dark state) can cause artifacts that mimic treated cells, we calculated a single clustering metric from the protein clustering (26), we removed localizations that were likely integral between the two curves (Fig. 3I) for each cell that cap- due to oversampling (Materials and Methods). tures the difference between the observed clustering in the data RNase E. We first performed 3D two-color SR imaging of RNase and the artifactual small clustering in the complete spatial ran- E and the cell surface to confirm that Caulobacter RNase E is not domness simulations arising from the cell shape (Materials and sequestered to the cell membrane (10). In sharp contrast to Methods)(Fig.3J). The rif-treated values were significantly lower RNase E sequestration to the inner membrane in E. coli (4, 5), than those for the WT cells, which captures the decreased clus- RNase E in Caulobacter was observed to cluster along the central tering of molecules in rif-treated cells (P < 0.0001). axis of the cell and nucleoid (Fig. 3 A–D), consistent with pre- Ribosomes. We acquired SR images from a mixed population of vious DL imaging studies showing the cytoplasmic localization of fixed Caulobacter cells expressing ribosomal protein L1-eYFP (Fig. RNase E (10). RNase E formed distinct clusters in the cytoplasm 3G, SI Appendix,Fig.S11,andMovie S3 show additional example (Fig. 3E and Movie S1), although we observed various degrees of cells). Ribosomes were observed throughout the cytoplasm, con- clustering across the mixed population due to stochastic sam- sistent with DL imaging (10) and cryo-electron microscopy (29). pling as well as cell-to-cell variability in gene expression (SI This differs from the ribosome/nucleoid segregation observed in E. Appendix, Fig. S10). coli and B. subtilis (30). Ribosomes in fixed Caulobacter cells ap- RNase E in fixed Caulobacter cells was more clustered than pear weakly clustered, yet slightly more clustered than what would would be expected from complete spatial randomness, indicating be expected from complete spatial randomness (Fig. 3G, Bottom that the observed clustering is not due to chance or the cell shape row). Performing the same clustering analysis as we did with (27) (Materials and Methods). Positive deviations from the gray RNase E, we observed a heterogeneous distribution of clustering curve (Fig. 3E, Bottom row) indicate clustering. The clustering was between different cells (Fig. 3K). Nevertheless, similar to RNase E, heterogeneous across the mixed cell population, likely arising ribosome clustering exhibited a clear relationship with transcrip- from both intrinsic and extrinsic noise (24, 28) (Fig. 3J). When tion, as treating the cells with rif significantly shifted the observed

Bayas et al. PNAS | vol. 115 | no. 16 | E3715 Downloaded by guest on September 24, 2021 A BC D Median distance to Median distance 300 nm yz slice RNaseE-eYFP central axis = 109 nm to central axis = 272 nm rhodamine lactam surface label 60 250

50 200 40 150 30 Counts Counts 100 20 50 10

0 0 0 100 200 300 400 500 0 100 200 300 400 500 Distance to central axis (nm) Distance to central axis (nm) 200 nm

EFGHRNase E WT RNase E rif ribosomal L1 WT ribosomal L1 rif

150 data 150 data 150 data 150 data CSR CSR CSR CSR 100 100 100 100

50 50 50 50 H(r) (nm) H(r) (nm) H(r) (nm) 0 H(r) (nm) 0 0 0

-50 -50 -50 -50 0 100 200 0 100 200 0 100 200 0 100 200 radius (nm) radius (nm) radius (nm) radius (nm)

IJ K RNase E ribosomal L1 0.3 0.3 **** WT **** WT H data rif rif 0.2 0.2 ∫ H - H data sim

H(r) (nm) 0.1 0.1 H Normalized Counts sim Normalized counts 0 0 radius (nm) 0 0.5 1 1.5 2 2.5 3 02468 ∫ H - H 4 ∫ H - H 3 data sim ×10 data sim ×10

Fig. 3. Spatial clustering of RNase E and ribosomes decreases when RNA is depleted. (A) Example fixed cell expressing RNase E-eYFP (yellow) with its cell surface labeled with rhodamine spirolactam (magenta). (B) Distribution of distances of RNase E molecules to the cell’scentralaxis.(C)Distributionofdistancesof rhodamine lactam molecules to the cell’s central axis. (D)Ayz projection of a 300-nm slice perpendicular to the cell axis. (E) SR reconstruction of a fixed WT Caulobacter cell expressing RNase E-eYFP, with calculation of the H(r) clustering metric (below) for the data as well as for CSR (complete spatial randomness). (F) SR reconstruction of a fixed rif-treated RNase E cell. (G) SR reconstruction of a fixed WT Caulobacter cell expressing ribosomal protein L1-eYFP. (H)SRre- construction of a fixed rif-treated L1 cell. In the H function curves plotted in E–H, the lighter gray and lighter yellow (E and F)/lighter blue (G and H) show the 95% confidence intervals. In the reconstructions, each localization is plotted as a 3D Gaussian with a σ equivalent to the average xy localization precision of the localizations in each cell (29/30 nm in xy for RNase E and 24/25 nm in xy for ribosomes). An average of 120 molecules and 720 molecules are plotted for RNase E and ribosomes, respectively. Cells are on 1-μmgrids.(I) The degree of clustering quantity plotted in F and G were calculated from the area between the data curve and CSR. (J) Distributions for RNase E WT and rif-treated cells showing clustering relative to CSR. (K) Distributions for L1 WT and rif-treated cells showing clustering relative to CSR. Analysis was performed on 218 WT and 135 rif-treated RNase E cells, and on 195 WT and 239 rif-treated ribosome cells.

clustering toward complete spatial randomness (Fig. 3 H and K control, we also imaged an E. coli strain containing ribosomal and Movie S4)(P < 0.0001). protein S2-eYFP (30). Similar to Caulobacter, ribosomes in S. To investigate whether the cytoplasmic ribosomal distribution is meliloti occupied the entire cell volume and colocalized with the unique to Caulobacter,weimagedribosomalL1-eYFPinanother nucleoid, whereas the γ-proteobacterium E. coli maintained segre- α-proteobacterium, S. meliloti, together with the SYTOX orange gation of the DNA and polar nucleoid-excluded ribosomes (SI Ap- DNA intercalating dye using epifluorescence microscopy. As a pendix,Fig.S12). Notably, S. meliloti ribosomes contained a slightly

E3716 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al. Downloaded by guest on September 24, 2021 more patchy localization pattern than Caulobacter, but generally both swarmer and stalked cells. The increase in the number of PNAS PLUS showed a similar interpenetrating nucleoid–ribosome configuration clusters in the cell during the swarmer to stalked transition is similar to that observed with Caulobacter. consistent with biochemical data showing increased overall RNA abundance during this time in the cell cycle (31). This increase in Number of Clusters of RNase E Changes as a Function of the Cell Cycle. clusters may also be attributed to the initiation of DNA repli- Ribosome profiling (31) and Western blot analysis (2) have cation that occurs during this time period leading to an increase shown that RNase E levels in Caulobacter vary as a function of in overall transcription. the cell cycle, with maximum levels at the swarmer to stalked transition and just before cell division. To determine whether RNase E Clusters Colocalize with Highly Transcribed Genes. Our ob- RNase E clusters fluctuate as a function of the cell cycle, we servation that the number of RNase E clusters decreased upon a performed two-color 3D imaging of RNase E-eYFP and decrease in RNA substrate availability after treatment with rif PAmCherry-PopZ (to mark the cell poles). We imaged syn- suggested that RNase E clusters might form near regions of high chronized cells (Materials and Methods) fixed with 1% formal- transcriptional activity. Chromosomal loci along Caulobacter’s dehyde to capture snapshots of the RNase E clusters at distinct single circular chromosome occupy specific subcellular locations time points along the cell cycle (Fig. 4 A–E). Our 3D recon- (32). Using this distinct organization of the chromosome, we structions show RNase E clusters along the cell axis throughout were able to ask whether RNase E clusters correspond to DNA the cell cycle and PopZ foci at the cell poles. loci that are highly transcribed. The genes encoding rRNA are We then analyzed the sizes of RNase E clusters using the the most highly transcribed loci, comprising 95–97% of the density-based algorithm DBSCAN (Materials and Methods) (Fig. prokaryotic cell’s total RNA mass (33). There are two near- 4F). There was no change in the median number of molecules identical copies of rDNA in Caulobacter (Fig. 5A) that occupy detected in each RNase E cluster as cells progressed through the spatially distinct sites on the chromosome. Each of these loci cell cycle. However, we did find an increase in the number of encode 16S, 23S, and 5S rRNA, as well as two tRNAs. These RNase E clusters as the swarmer cells differentiated into stalked RNAs are transcribed as a single transcript that is subsequently cells (Fig. 4G). During the swarmer to stalked cell transition, the processed by RNases, including RNase E, to generate the mature median number of clusters in the cell doubled from two clusters rRNA species (1, 2, 34). to four clusters in keeping with the previous observation that the To determine whether the observed RNase E clusters are abundance of RNase E varies during the cell cycle with a max- positioned near the DNA loci of rRNAs, we labeled the two imum at the swarmer to stalked cell differentiation (2). The rDNA operons (rDNA1 and rDNA2) with a tetO/tetR labeling BIOPHYSICS AND histogram for the final 125-min population (Fig. 4G, Right) system (Fig. 5B) and visualized the positions of RNase E in a COMPUTATIONAL BIOLOGY shows the existence of two populations, since this cell type has mixed population of cells with respect to the two rRNA gene loci

A 5 min B 35 min C 65 min D 95 min E 125 min

RNaseE-eYFP PAmCherry-PopZ

F Ncells = 74 Ncells = 75 Ncells = 73 Ncells = 52 Ncells = 33 0.6 0.4 0.4 0.4 0.6 median = 12 median = 12 median = 13 median = 13 median = 13 0.3 0.3 0.3 0.4 0.4 0.2 0.2 0.2 0.2 0.2 0.1 0.1 0.1 0 0 0 0 0

Normalized counts 050100050100050100050100050100 Number of molecules per RNaseE cluster G 0.4 0.3 0.3 0.3 0.2 median = 2 median = 4 median = 4 median = 4 median 0.3 0.15 0.2 0.2 0.2 = 4 0.2 0.1 0.1 0.1 0.1 0.1 0.05 0 0 0 0 0

Normalized counts 05100510051005100510 Number of RNaseE clusters per cell

Fig. 4. Cell cycle-dependent spatial distribution of RNase E-eYFP in fixed Caulobacter cells. (A–E) SR reconstructions of fixed Caulobacter cells expressing RNase E-eYFP and PAmCherry-PopZ at (A) 5 min, (B)35min,(C)65min,(D) 95 min, and (E) 125 min after synchrony. The total cell cycle is 150 min. Cells are on 1-μm grids. (F) Distributions of the number of molecules detected in each RNase E cluster throughout the cell cycle. (G) Distributions of the number of RNase E clusters per cell throughout the cell cycle. We find a constant number of molecules in each cluster for all cell stages, but a doubling in the number of clusters per cell during the swarmer to stalk transition.

Bayas et al. PNAS | vol. 115 | no. 16 | E3717 Downloaded by guest on September 24, 2021 WL DNA label RNaseE C A 2.02 Mbp Control

2.0 Mbp Ter rDNA1 rDNA1 79% 2.87 Mbp rDNA2 Caulobacter crescentus rDNA2 Control 4.01 Mbp D

0 Mbp Ori

3.80 Mbp rDNA1 rDNA2 77% B E TetR-eYFP

gene TetO TetO TetO ...... 85x

Control 27%

Fig. 5. Colocalization of rRNA loci and CCNA_01879 loci with RNase E. (A) A cartoon of the labeled DNA loci. Approximate locations were calculated using the Caulobacter swarmer 1D chromosome map as described previously (32). (B) A schematic of the DNA-labeling scheme. (C, Left) A white-light (WL) image of live Caulobacter cells expressing RNase E-PAmCherry and tetR-eYFP/85x-tetO on rDNA1 locus. (C, Middle) A diffraction-limited (DL) image of the tetR-eYFP labeling the rDNA1 locus. (C, Right) A DL image of RNase E-PAmCherry. In total, 79.2% of rDNA1 locus (cyan) colocalizes with an RNase E cluster (yellow). (D) Caulobacter cells with rDNA2 labeled. In total, 76.9% of rDNA2 locus (red) colocalizes with an RNase E cluster (yellow). (E) Caulobacter cell with CCNA_01879 labeled. In total, 27.1% of the control CCNA_01879 locus (green) colocalizes with an RNase E cluster. (Scale bars: 500 nm.) Analysis was performed on 506 rDNA1 cells, 664 rDNA2 cells, and 321 control cells.

by DL imaging. The positions of the rDNA gene loci were found are excluded from the nucleoid, while similar studies in Caulo- by SR imaging, and these positions were plotted against the bacter revealed no apparent separation of ribosomes and the nu- RNase E DL images (Fig. 5 C–E)(Materials and Methods). We cleoid (10). However, due to the small size of Caulobacter,theDL found that the two rDNA loci each colocalized with an RNase E resolution of standard fluorescence microscopy would obscure all cluster. Specifically, 79.2% of the first rDNA locus and 76.9% of sub-DL motion or organization. Therefore, in this study, we used the second rDNA locus colocalized with an RNase E cluster 3D SPT and SR microscopy to probe and quantify the spatial (additional examples are in SI Appendix, Figs. S13–S15). These organization of RNase E and ribosomes at high resolution, thereby results suggest that RNase E processing of rRNA transcripts providing insight into the dynamic behavior of these complexes. occurs at the site of rRNA synthesis. Within each cell, we esti- Given that RNase E clusters were found in the cytoplasm in a mated that the fractions of RNase E that colocalized with each pattern overlapping that of the nucleoid, there is unlikely to be a of the two rDNA loci are 22 ± 9% (error is SD) and 24 ± 9% gross physical separation of the RNA production and degrada- (error is SD), respectively, suggesting that approximately one- tion machinery. We found the subcellular position of RNase E half of RNase E molecules are involved in rRNA processing clusters correlated with sites of rRNA synthesis, suggesting that (SI Appendix, Supplementary Methods). cleavage occurs where RNA substrate is produced. This obser- To confirm that the RNase E/rDNA correlations are valid, we vation is consistent with evidence that the rRNA transcript is also imaged a lowly transcribed gene, CCNA_01879, that is in the processed cotranscriptionally (36), and previous studies that bottom 0.5% of RNA-seq analysis (31) (Fig. 5A). In contrast to have shown RNase E associating with DNA (10). The increased rDNA, we found a lack of correlation of an RNase E cluster with local concentration of RNase E clusters near highly transcribed this poorly transcribed locus. Only 27.1% of CCNA_01879 loci genes allows a large number of substrates to be quickly degraded colocalized with an RNase E cluster. There was still some or processed. RNase E clusters might reflect the formation of a colocalization, possibly because within any given cell, a few large number of RNA degradosome complexes, which could RNase E clusters are present, which coincidentally can colocalize efficiently facilitate RNA cleavage. The size of these clusters did with our DNA labels. These results support the hypothesis that not change during cell cycle progression, suggesting that there RNase E clusters are preferentially localized to highly tran- might exist an optimum number of molecules per cluster. The scribed DNA loci where they immediately process RNA, most observed confined subtrajectories are likely actively processing likely cotranscriptionally. RNAs in large RNA degradosome assemblies, possibly acting cotranscriptionally on transcripts still tethered to the chromo- Discussion some. Mobile subtrajectories could belong to RNase E mole- RNase E is the core of the RNA degradosome complex. It has cules that have not yet bound an RNA substrate. Moreover, the been shown to be membrane associated in E. coli and B. subtilis decrease in the fraction of confined RNase E molecules and (3–6), but deployed throughout the nucleoid in Caulobacter (10). lower degree of confinement in cells in which RNA has been Previous studies of ribosome localization in E. coli (30, 35) and depleted highlights RNase E’s relationship with substrate B. subtilis (12) revealed enrichment at the cell poles where they availability.

E3718 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al. Downloaded by guest on September 24, 2021 Our SPT experiments in WT Caulobacter show two pop- were prepared using a Percoll density gradient (44), allowing the cells a 5- PNAS PLUS ulations of ribosomal L1: an 89% slow species (D = 0.039 ± min recovery in M2G at 28 °C before moving on to the next step. rif was μ 0.001 μm2/s) and an 11% fast species (D = 3.23 ± 0.12 μm2/s) added at 50 g/mL final concentration [from 10 mg/mL stock solutions in (Fig. 2E). We attribute the slow 89% to ribosomes actively en- DMSO (Fisher)] and incubated with the cells for 30 min with shaking. The cells were washed by centrifugation [3 min, 8,000 rpm, 4 °C (Eppendorf gaged in translation, and this population includes both 5430R, Eppendorf FA-45-24-11-HS rotor)] and resuspension in clean M2G. chromosome-tethered polysomes and actively translating single Cells were fixed with the addition of 1% formaldehyde (Lot 091161; Fisher ribosomes. The diffusion coefficient we found for the actively Scientific), followed by a 10-min incubation at room temperature and 30- translating ribosomes is consistent with reported diffusion coef- min incubation on ice, and finally washed three times with clean M2G. ficients of E. coli ribosomes engaged in active translation (30, Before imaging, the cells were resuspended in ∼20–50 μL of minimal me- 35). The fast 11% are likely free, unincorporated L1 subunits— dium, producing a concentrated cell suspension. To this suspension, ∼1nMof their observed diffusion coefficient is consistent with the diffu- fiducial markers were added (540/560 carboxylate-modified FluoSpheres, – μ sion of free FPs in bacteria (35, 37, 38). In contrast to our results 100-nm diameter; Molecular Probes), and then 1 2 L of this mixture was in which we found observable diffusion of actively translating deposited onto an agarose pad [composed of 1.5% (wt/wt) of low melting point agarose (Invitrogen) in M2G] and mounted onto an argon plasma- ribosomes, a previous fluorescence recovery after photobleaching etched glass slide (Fisher) and imaged immediately. (FRAP) study of Caulobacter ribosomes observed essentially sta- For imaging the tetO/tetR-labeled strains, the cells were incubated with = μ 2 tionary actively translating ribosomes (D 0.0011 m /s) (39). 0.03% xylose final concentration [from 30% (wt/vol) stock solutions in water] This discrepancy could be attributed to the slower temporal res- for 30 min before washing once with M2G and resuspending in M2G (∼20–50 olution and lower spatial resolution of FRAP. As opposed to SPT, μL), and imaged immediately on an agarose pad as described above. the DL resolution of FRAP is unable to quantify motions on scales smaller than ∼200 nm. SM Imaging for SR Microscopy and SPT. Both SPT and SR microscopy rely on Upon rif treatment, we observed two populations of ribosomal localizing single molecules with high precision; SPT is based upon localizing ’ L1: a 95% species (D = 0.215 ± 0.002 μm2/s) and a 5% species the same molecule over time to gain information about this molecule s (D = 5.67 ± 0.39 μm2/s). The diffusion coefficient we observed in dynamics. In contrast, SR microscopy localizes many different SMs over time and then builds up a reconstruction to gain information about the static WT cells that we attributed to actively translating ribosomes was structure of interest. Since the imaging time required to acquire enough SM not observed in rif-treated cells, likely due to RNA depletion. localizations for SR imaging can be long (approximately minutes), the Instead, the 95% L1 species in the rif population had a 5.5-fold structure of interest must be stationary on the timescale of imaging. faster diffusion coefficient compared with the slow (D = 0.039 ± Cells immobilized on agarose pads were imaged on a home-built epi- 0.001 μm2/s) 89% observed in WT cells. This suggests that the fluorescence microscope (Nikon Diaphot 200) with additional optics inserted BIOPHYSICS AND 95% L1 species in the rif-treated cells may be fully assembled before the 512 × 512-pixel Si EMCCD camera (Andor Ixon DU-897), which COMPUTATIONAL BIOLOGY ribosomes that have not yet bound an mRNA. Importantly, the enables two-color, 3D SR imaging described previously (45). Fluorescence emission from the sample was collected through a high-N.A. oil-immersion 5% fast species in rif-treated cells has a diffusion coefficient × close to the 11% fast species in WT cells, and we thus attribute objective (Olympus UPlanSapo 100 /1.40 N.A.), filtered by a dual-pass dichroic mirror (zt440/514/561rpc; Chroma), a 560-nm dichroic beamsplitter this population to unincorporated L1 subunits. (Semrock FF560-FDi01), a 514-nm long-pass filter (Semrock LP02-514RE), a The weak clustering of ribosomes we observed in our SR 561-nm notch filter (Semrock NF03-561E), and a bandpass filter (FF01-532/ measurements shows that there may be small compartments pre- 610; Semrock). Two-dimensional white-light transmission images of the cells sent within the cell that are able to confine ribosomes and form a were recorded before imaging. The eYFP was pumped with 514-nm excita- long-range diffusion barrier. We observed these weakly clustered tion (Coherent Sapphire 514–100 CW, ∼0.5–1 kW/cm2). A 405-nm laser (Obis, ribosomes throughout the cytoplasm, occupying the same regions ∼0.1–10 W/cm2) was used to photoactivate and a 561-nm laser (Coherent 2 as the nucleoid. This result is consistent with previous DL imaging Sapphire 561–100 CW, ∼0.5 kW/cm ) was used to excite fluorescence from of ribosomes and the observation that transcription and translation PAmCherry. Laser intensity was determined by measuring the power at the stage with a power meter (Newport 1918-C) and estimating the FWHM spot occur throughout the nucleoid (10). Importantly, the ribosome and size by fitting the average fluorescence from the sample to a 2D Gaussian by nucleoid association we observed is not unique to Caulobacter.We nonlinear least squares using a custom MATLAB (MathWorks) program. In- found a similar pattern in S. meliloti, suggesting that this ribosome/ tegration times were 25 ms for SPT data and 50 ms for SR data. nucleoid organization may be a feature shared by α-proteobacteria. These results contrast with studies from E. coli and B. subtilis.For Data Analysis Procedure. instance, ribosomes in B. subtilis are localized at regions distinct Fitting 3D SM data. Both 3D SR and SPT data were fit using a modified version from the nucleoid due to actively transcribed regions of the nu- of Easy-DHPSF (46), which is freely available at https://sourceforge.net/pro- cleoid being exposed to the cell periphery (12, 40), and our control jects/easy-dhpsf/. Calibration scans were created by axially scanning a ∼ μ ∼ μ experiments in the γ-proteobacterium E. coli recapitulated the nanohole array (NHA) (47) filled with 25 Lof 50 M Alexa 568 (Life Technologies) and Atto 520 (Sigma) dyes using a piezoelectric objective previously observed ribosome and nucleoid segregation (30, 35). scanner (C-focus; Mad City Labs). These calibration scans were used to relate These results suggest that the differing Caulobacter and E. coli ri- angular lobe orientation to axial position, and as templates to locate can- bosome organizations may be due to features unique to each or- didate SMs in the data. The NHA allowed us to account for field-dependent ganism such as chromosome organization or cytoplasmic viscosity, errors arising from the differing rotation rates, and hence apparent axial and future studies are needed to elucidate these mechanisms. position, of the DH-PSF as a function of lateral position. The background Taken together, our results show that even without the complex photons (∼16–22 photons/pixel) were estimated using a median filter in intracellular compartments of eukaryotic systems (41), bacteria MATLAB (46). On average, we detected 1,327 ± 449 photons per 25-ms spatially organize RNA processing, degradation, and translation frame for the eYFP-labeled RNase E and ribosomes. Localization precision was estimated with an empirical formula derived from repeatedly localizing within the nucleoid, and that this organization is governed by the single beads under variable background conditions (45). Images included in transcriptional profile of the chromosome. this paper have typical average localization precisions of 32 ± 8 nm (error is SD), 33 ± 9 nm, and 49 ± 13 nm in x, y, and z. Systematic errors from sample Materials and Methods drift resulting from mechanical and thermal fluctuations were corrected by Preparation of Cell Strains for Imaging. Caulobacter cells were grown over- adding ∼1–10 nM concentration of bright 560/580 Fluospheres (Life Tech- night from frozen culture in 5 mL of PYE growth medium (42) with shaking nologies) to the sample and using these as fiducial beads. High-frequency in a 28 °C water bath. The day before imaging, cultures were diluted noise in the drift is removed with wavelet filtering in MATLAB (45, 46). 1:1,000 into the defined minimal media M2G (43). After growing to midlog Two-channel registration. Arrays of control point positions were generated by phase, 1-mL aliquots of cells were washed once with M2G by centrifugation axially scanning the NHA (47) filled with ∼25 μLof∼50 μM Alexa 568 (Life [3 min at 8,000 rpm, 4 °C (Eppendorf 5430R, Eppendorf FA-45-24-11-HS ro- Technologies) and Atto 520 (Sigma) dyes using a piezoelectric objective tor)], and the pellet was resuspended into clean M2G. Synchronized cells scanner (C-focus; Mad City Labs) and imaging the NHA simultaneously in the

Bayas et al. PNAS | vol. 115 | no. 16 | E3719 Downloaded by guest on September 24, 2021 two channels using both the 514- and 561-nm lasers. The red channel was ! À Á Xn −r2 transformed to the green channel using the same algorithm described P r2, τ = 1 − x exp , [1] i r 2 previously (45). Target registration errors ranged from 5 to 15 nm, and fi- i=1 i ducial registration errors ranged from 4 to 12 nm. Removing oversampling. Oversampling, caused by including multiple locali- Δt Xd r 2 = 2dD τ − + 2 σ 2, [2] zations of the same protein from FP labels that emit for more than one frame, i i 3 j j=1 creates spurious clustering. Before analyzing clustering in fixed cells, over-

sampling was removed with a radial and temporal filter. Localizations within Xn a set radial and time region are assumed to belong to the same molecule and xi = 1. [3] are combined into a single localization located at the average position. A i=1 reasonable radial distance was determined by simulating SR data containing oversampling on the size scale of the localization precision. The threshold Fitting with three populations yields the same result as fitting with only two radius (as a function of the mean localization precision) was varied until a populations (SI Appendix, Fig. S9). grouping algorithm in MATLAB recovered the ground truth number of We performed a validation of finding two populations of Brownian dif- simulated single molecules (thus estimating the grouping of the oversampled fusers using this method by simulating particles with two different diffusion × coefficients, extracting the parameters in Eq. 1, and comparing the simulated clusters). The optimizedpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi spatial threshold was found to be 1.5 mean lo- 2 2 2 data from the fit. The percent discrepancies for the slow vs. fast fractions are calization precision ( σx + σy + σz ). The temporal threshold was de- 5% vs. 3%; the percent discrepancies for the slow vs. fast D values are 33% termined individually for each cell dataset by calculating the probability of vs. 12%. Detailed results can be found in SI Appendix,Fig.S16. finding another localization within the spatial threshold as a function of Clustering analysis by spatial point statistics. To quantify the degree of clustering time. The probability typically followed a single-exponential decay. The time in our SR images, we used spatial point statistics. Complete spatial ran- constant of the exponential was used as the temporal threshold. domness (CSR) was simulated within each cell volume using an identical Diffusion analysis. Trajectories were constructed by connecting localizations number of points with a custom MATLAB (MathWorks) program. The sim- from consecutive frames. Only trajectories of at least four steps (five frames or ulated points were then thinned to be the same number of localizations as 125 ms) were used for MSD analyses. The maximum allowed xyz displace- the data, then a random “kick” of localization precision error (pulled from a ments of RNase E-eYFP and L1-eYFP molecules over 25 ms were set to Gaussian distribution of the mean localization precision of the data) in all 1,000 nm. For both RNase E-eYFP and ribosomal L1-eYFP, we detect a mean three dimensions to each simulated point. In all cases, 100 simulations were trajectory length of seven frames (175 ms). Diffusion coefficients were cal- performed, as this balanced computational time with statistical information. culated from ensemble-averaged 3D MSDs of individual RNase E molecules using the first two time lags to avoid long-time effects of confinement and The uncertainty in the data from the localization precision was also non-Brownian motion. The reported errors are the SEMs of diffusion coef- bounded by simulation. Each detected localization had its own localization ficients as determined by bootstrapping (50% of the trajectories were precision error. A cloud of points (100 simulations) with shape dictated by the sampled 500 times). Error bars in the MSD curves are SEMs. individual localization precision were simulated around each localization. The Simulating trajectories inside cell-shaped object. Brownian simulations were same statistical metrics were calculated on these simulated points, and the performed inside a cylindrical cell of radius 250 nm and length 3.5 μm with 95% confidence bounds were used to report the error in the data from the two hemispherical caps; finite sampling from a camera and motion blur localization precision. ’ were taken into account by averaging 10 positions for every 25-ms frame. Global clustering is quantified using Ripley s K-function (27) and its nor- ’ Each 3D position was then given an xyz “kick” from a Gaussian distribution malized forms (52). As previously described, Ripley s K-function represents to simulate localization precision. the average probability of finding localizations a distance r from the typical Confinement analysis. To test for confinement, we divided each trajectory [of at localization (53). For CSR, K(r) scales with the volume and is described by Eq. d least four steps (five frames or 125 ms) in length] into subtrajectories of four 4, where bd r is the volume of the unit sphere in d dimensions. Deviations steps (five frames or 125 ms). For each subtrajectory, the maximum dis- from this shape indicate clustering or repulsion; however, visualizing devi- placement from the first point of that subtrajectory was calculated (20, 21). We ations from a straight or flat line is simpler than for an exponential curve. performed simulations of Brownian diffusers inside a cylindrical cell volume The normalized K-function is given by L(r) (Eq. 5), which can further be with hemispherical caps to determine whether a subtrajectory was confined normalized into the H(r) function, such that CSR is a flat horizontal line at or not. The effect of confinement, such as in the case of cytoplasmic molecules H = 0(Eq.7) (54). To be able to analyze clustering across many cells, the diffusing in a cell, results in a smaller value for the apparent, observed dif- quantity shown in Eq. 7 was calculated for each cell: fusion coefficient than in the case in which the molecule is not confined by ð Þ = d cell walls. To ensure that we were making a valid comparison of diffusion K r bd r , [4] coefficients in our confinement analyses (48), we performed Brownian simu- pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi d lations of a cytoplasmic molecule inside a cylindrical cell with hemispherical LðrÞ = KðrÞ=bd , [5] caps. We used an empirical diffusion coefficient that resulted in an apparent diffusion coefficient approximately equivalent to the ensemble-averaged HðrÞ = LðrÞ − r, [6] measured diffusion coefficients for WT and rif-treated cells (SI Appendix, Z Fig. S4). A threshold was set such that only 5% of the simulated Brownian HdataðrÞ − HCSRðrÞ. [7] subtrajectories are confined. Subtrajectories with displacements less than this r value (118 nm for WT and 123 nm for rif-treated, plotted as vertical dotted lines in Fig. 1E) were classified as confined. From the cumulative distributions Clustering analysis by density threshold. For the 3D two-color RNase E/PopZ shown in Fig. 1E, this criterion would imply that ∼50% of the WT sub- datasets, we use a density-based clustering algorithm, density-based spatial trajectories were mobile. However, we found that many molecules in this clustering of applications with noise (DBSCAN) (55), which detects clusters group did not diffuse very far, as suggested by fixed snapshots of the spatial according to the local density of localizations within a certain search radius. distribution (Fig. 3E). To include the possibility of an intermediate degree of For the RNase E clusters, we used the parameters « = 70 nm and minPts = 6; confinement, another threshold was arbitrarily set at the Brownian 50th for the PopZ clusters, we used the parameters « = 200 nm and minPts = 7. percentile, and subtrajectories with displacements greater than this value These thresholds were chosen empirically. (198 nm for WT and 208 nm for rif-treated) were classified as clearly mobile. In the RNase E-eYFP/PAmCherry-PopZ cells, the clustering results for RNase All other subtrajectories were classified as intermediate. E did not give an absolute count of the number of molecules in a cluster, but Fitting squared displacement CDFs to multiple populations. To extract the dif- rather gave the relative amounts of number of molecules detected. This is due ferent populations of ribosomal L1 proteins, we fit the empirical CDF of the to the photophysical properties of eYFP. Imaging SMs of eYFP requires an – 2 squared displacements of our trajectories to Eq. 1 (49 51), where r are the initial bleach down of the sample, then waiting for the stochastic recovery τ experimental squared displacements, is the time lag, and xi is the weight of and blinking of this FP. The ideal situation is one in which each copy of eYFP population i.Eq.2 (in which d is the number of dimensions, Di is the diffusion recovers. However, in practice, only about 5–15% of the molecules recover Δ σ coefficient of population i, t is the camera exposure time, j is the locali- and are detected. zation σ in the jth dimension) is substituted into Eq. 1 to perform the fit. The Superlocalization of labeled loci and colocalization. Fifty frames (100-ms expo- fitting is constrained by Eq. 3 to ensure that all weights sum to 1. The fitting sure) were summed for tetR-eYFP and RNase E-PAmCherry. TetR-eYFP loci 2 is performed in MATLAB, minimizing the objective function ðPdata − PfitÞ were superlocalized by fitting 2D Gaussians (which well-approximates the PSF using the function fmincon: of the microscope) in ThunderSTORM (ImageJ). These localizations were then

E3720 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al. Downloaded by guest on September 24, 2021 transformed on to the RNase E-PAmCherry images using the mapping function ACKNOWLEDGMENTS. We thank the laboratory of James Weisshaar at the PNAS PLUS generated with the following method. The array of 500 control points that University of Wisconsin–Madison for sending the E. coli S2 ribosome strain, Melanie Barnett and Jillynne Quinn in the laboratory of Sharon Long (De- were detectable in both channels simultaneously and evenly distributed across partment of Biology, Stanford University) for help creating the ribosome S. the whole field of view was localized by fitting a 2D Gaussian in Thunder- meliloti strain, Thomas H. Mann for help with constructing the RNase E- STORM. These localizations were used to generate a mapping function be- eYFP/PAmCherry-PopZ strain, and Alex von Diezmann for the Brownian dif- tween the two channels using local weighted mean transformation in custom fusion simulation code. This work was supported in part by National Institute MATLAB scripts to take into account field dependence of the transformation. of General Medical Sciences Grants R01-GM086196 (to W.E.M. and L.S.), R35-GM118067 (to W.E.M.), and R35-GM118071 (to L.S.). L.S. is a Chan Percentage of DNA loci colocalizing with an RNase E cluster was determined Zuckerberg Biohub Investigator, J.W. is a Stanford Center for Molecular by visual inspection of the transformed locus with respect to the RNase Analysis and Design Fellow, and J.M.S. was an NIH Postdoctoral Fellow EDLimage. (F32-GM100732).

1. Mackie GA (2013) RNase E: At the interface of bacterial RNA processing and decay. 29. Bowman GR, et al. (2010) Caulobacter PopZ forms a polar subdomain dictating se- Nat Rev Microbiol 11:45–57. quential changes in pole composition and function. Mol Microbiol 76:173–189. 2. Hardwick SW, Chan VSY, Broadhurst RW, Luisi BF (2011) An RNA degradosome as- 30. Bakshi S, Siryaporn A, Goulian M, Weisshaar JC (2012) Superresolution imaging of sembly in Caulobacter crescentus. Nucleic Acids Res 39:1449–1459. ribosomes and RNA polymerase in live Escherichia coli cells. Mol Microbiol 85:21–38. 3. Strahl H, et al. (2015) Membrane recognition and dynamics of the RNA degradosome. 31. Schrader JM, et al. (2016) Dynamic translation regulation in Caulobacter cell cycle PLoS Genet 11:e1004961. control. Proc Natl Acad Sci USA 113:E6859–E6867. 4. Moffitt JR, Pandey S, Boettiger AN, Wang S, Zhuang X (2016) Spatial organization 32. Viollier PH, et al. (2004) Rapid and sequential movement of individual chromosomal shapes the turnover of a bacterial transcriptome. eLife 5:e13065. loci to specific subcellular locations during bacterial DNA replication. Proc Natl Acad 5. Fei J, et al. (2015) RNA biochemistry. Determination of in vivo target search kinetics of Sci USA 101:9257–9262. regulatory noncoding RNA. Science 347:1371–1374. 33. Rosenow C, Saxena RM, Durst M, Gingeras TR (2001) Prokaryotic RNA preparation 6. Cascante-Estepa N, Gunka K, Stülke J (2016) Localization of components of the RNA- methods useful for high density array analysis: Comparison of two approaches. degrading machine in Bacillus subtilis. Front Microbiol 7:1492. Nucleic Acids Res 29:E112. 7. Ptacin JL, et al. (2010) A spindle-like apparatus guides bacterial chromosome segre- 34. Feingold J, Bellofatto V, Shapiro L, Amemiya K (1985) Organization and nucleotide gation. Nat Cell Biol 12:791–798. sequence analysis of an rRNA and tRNA gene cluster from Caulobacter crescentus. 8. Lee SF, Thompson MA, Schwartz MA, Shapiro L, Moerner WE (2011) Super-resolution J Bacteriol 163:155–166. imaging of the nucleoid-associated protein HU in Caulobacter crescentus. Biophys J 35. Sanamrad A, et al. (2014) Single-particle tracking reveals that free ribosomal subunits 100:L31–L33. are not excluded from the Escherichia coli nucleoid. Proc Natl Acad Sci USA 111: 9. Ptacin JL, et al. (2014) Bacterial scaffold directs pole-specific centromere segregation. 11413–11418. Proc Natl Acad Sci USA 111:E2046–E2055. 36. Apirion D, Gegenheimer P (1981) Processing of bacterial RNA. FEBS Lett 125:1–9. 10. Montero Llopis P, et al. (2010) Spatial organization of the flow of genetic information 37. Bakshi S, Bratton BP, Weisshaar JC (2011) Subdiffraction-limit study of Kaede diffu- in bacteria. Nature 466:77–81. sion and spatial distribution in live Escherichia coli. Biophys J 101:2535–2544. BIOPHYSICS AND

11. Bakshi S, Choi H, Weisshaar JC (2015) The spatial biology of transcription and trans- 38. English BP, et al. (2011) Single-molecule investigations of the stringent response COMPUTATIONAL BIOLOGY lation in rapidly growing Escherichia coli. Front Microbiol 6:636. machinery in living bacterial cells. Proc Natl Acad Sci USA 108:E365–E373. 12. Lewis PJ, Thaker SD, Errington J (2000) Compartmentalization of transcription and 39. Montero Llopis P, Sliusarenko O, Heinritz J, Jacobs-Wagner C (2012) In vivo bio- translation in Bacillus subtilis. EMBO J 19:710–718. chemistry in bacterial cells using FRAP: Insight into the translation cycle. Biophys J 13. Dickson RM, Cubitt AB, Tsien RY, Moerner WE (1997) On/off blinking and switching 103:1848–1859. behaviour of single molecules of green fluorescent protein. Nature 388:355–358. 40. Mascarenhas J, Weber MHW, Graumann PL (2001) Specific polar localization of ri- 14. Tuson HH, Biteen JS (2015) Unveiling the inner workings of live bacteria using super- bosomes in Bacillus subtilis depends on active transcription. EMBO Rep 2:685–689. resolution microscopy. Anal Chem 87:42–63. 41. Berry J, Weber SC, Vaidya N, Haataja M, Brangwynne CP (2015) RNA transcription 15. Gahlmann A, Moerner WE (2014) Exploring bacterial cell biology with single-molecule modulates phase transition-driven nuclear body assembly. Proc Natl Acad Sci USA 112: tracking and super-resolution imaging. Nat Rev Microbiol 12:9–22. E5237–E5245. 16. von Diezmann A, Shechtman Y, Moerner WE (2017) Three-dimensional localization of 42. Poindexter JS (1964) Biological properties and classification of the Caulobacter group. single molecules for super-resolution imaging and single-particle tracking. Chem Rev Bacteriol Rev 28:231–295. 117:7244–7275. 43. Ely B (1991) Genetics of Caulobacter crescentus. Methods Enzymol 204:372–384. 17. Voss JE, Luisi BF, Hardwick SW (2014) Molecular recognition of RhlB and RNase D in 44. Tsai JW, Alley MRK (2001) Proteolysis of the Caulobacter McpA chemoreceptor is cell the Caulobacter crescentus RNA degradosome. Nucleic Acids Res 42:13294–13305. cycle regulated by a ClpX-dependent pathway. J Bacteriol 183:5001–5007. 18. Biteen JS, et al. (2008) Super-resolution imaging in live Caulobacter crescentus cells 45. Gahlmann A, et al. (2013) Quantitative multicolor subdiffraction imaging of bacterial using photoswitchable EYFP. Nat Methods 5:947–949. protein ultrastructures in three dimensions. Nano Lett 13:987–993. 19. Campbell EA, et al. (2001) Structural mechanism for rifampicin inhibition of bacterial 46. Lew MD, von Diezmann ARS, Moerner WE (2013) Easy-DHPSF open-source software RNA polymerase. Cell 104:901–912. for three-dimensional localization of single molecules with precision beyond the 20. Milenkovic L, et al. (2015) Single-molecule imaging of Hedgehog pathway protein optical diffraction limit. Protoc Exch, 026. smoothened in primary cilia reveals binding events regulated by Patched1. Proc Natl 47. von Diezmann A, Lee MY, Lew MD, Moerner WE (2015) Correcting field-dependent Acad Sci USA 112:8320–8325. aberrations with nanoscale accuracy in three-dimensional single-molecule localiza- 21. Thompson MA, Casolari JM, Badieirostami M, Brown PO, Moerner WE (2010) Three- tion microscopy. Optica 2:985–993. dimensional tracking of single mRNA particles in Saccharomyces cerevisiae using a 48. Deich J, Judd EM, McAdams HH, Moerner WE (2004) Visualization of the movement double-helix point spread function. Proc Natl Acad Sci USA 107:17864–17871. of single histidine kinase molecules in live Caulobacter cells. Proc Natl Acad Sci USA 22. Le TB, Imakaev MV, Mirny LA, Laub MT (2013) High-resolution mapping of the spatial 101:15921–15926. organization of a bacterial chromosome. Science 342:731–734. 49. Schütz GJ, Schindler H, Schmidt T (1997) Single-molecule microscopy on model 23. Hong SH, McAdams HH (2011) Compaction and transport properties of newly repli- membranes reveals anomalous diffusion. Biophys J 73:1073–1080. cated Caulobacter crescentus DNA. Mol Microbiol 82:1349–1358. 50. Savin T, Doyle PS (2005) Static and dynamic errors in particle tracking microrheology. 24. Taniguchi Y, et al. (2010) Quantifying E. coli proteome and transcriptome with single- Biophys J 88:623–638. molecule sensitivity in single cells. Science 329:533–538. 51. Michalet X (2010) Mean square displacement analysis of single-particle trajectories 25. Bernstein JA, Khodursky AB, Lin PH, Lin-Chao S, Cohen SN (2002) Global analysis of with localization error: Brownian motion in an isotropic medium. Phys Rev E Stat mRNA decay and abundance in Escherichia coli at single-gene resolution using two- Nonlin Soft Matter Phys 82:041914. color fluorescent DNA microarrays. Proc Natl Acad Sci USA 99:9697–9702. 52. Besag J (1977) Comments on Ripley’s paper. J R Stat Soc B 39:193–195. 26. Durisic N, Laparra-Cuervo L, Sandoval-Álvarez A, Borbely JS, Lakadamyali M (2014) 53. Kiskowski MA, Hancock JF, Kenworthy AK (2009) On the use of Ripley’s K-function Single-molecule evaluation of fluorescent protein photoactivation efficiency using an and its derivatives to analyze domain size. Biophys J 97:1095–1103. in vivo nanotemplate. Nat Methods 11:156–162. 54. Ehrlich M, et al. (2004) Endocytosis by random initiation and stabilization of clathrin- 27. Ripley BD (1977) Modelling spatial patterns. J R Stat Soc Series B Stat Methodol 39: coated pits. Cell 118:591–605. 172–192. 55. Ester M, Kriegel H-P, Jörg S, Xu X (1996) A density-based algorithm for discovering 28. Elowitz MB, Levine AJ, Siggia ED, Swain PS (2002) Stochastic gene expression in a clusters in large spatial databases with noise. KDD-96 Proceedings (AAAI, Palo Alto, single cell. Science 297:1183–1186. CA), pp 226–231.

Bayas et al. PNAS | vol. 115 | no. 16 | E3721 Downloaded by guest on September 24, 2021