Spatial organization and dynamics of RNase E andribosomes in Caulobacter crescentusCamille 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-centus RNase E (RNA degradosome) and ribosomal protein L1 (ri-bosome) using 3D single-particle tracking and superresolutionmicroscopy. RNase E formed clusters along the central axis of thecell, while weak clusters of ribosomal protein L1 were deployedthroughout the cytoplasm. These results contrast with RNase Eand ribosome distribution in Escherichia coli, where RNase E coloc-alizes with the cytoplasmic membrane and ribosomes accumulate inpolar nucleoid-free zones. For both RNase E and ribosomes in Cau-lobacter, we observed a decrease in confinement and clusteringupon transcription inhibition and subsequent depletion of nascentRNA, suggesting that RNA substrate availability for processing, deg-radation, and translation facilitates confinement and clustering. Im-portantly, RNase E cluster positions correlated with the subcellularlocation of chromosomal loci of two highly transcribed rRNA genes,suggesting that RNase E’s function in rRNA processing occurs at thesite of rRNA synthesis. Thus, components of the RNA degradosomeand ribosome assembly are spatiotemporally organized in Caulo-bacter, with chromosomal readout serving as the template forthis organization.

ribosomes | RNA degradation | RNA processing | superresolutionmicroscopy | single-molecule tracking

In bacteria, the RNA degradosome mediates the majority ofmRNA turnover and rRNA and tRNA maturation (1). The

RNA degradosome assembles on the C-terminal scaffold regionof the RNase E endoribonuclease (2). RNase E in Escherichiacoli and RNase Y, the RNase E homolog in Bacillus subtilis, bothassociate with the cell membrane through a membrane-bindinghelix (3–6). One proposed rationale behind the observed mem-brane association is to physically separate transcription withinthe nucleoid from RNA degradation, thus providing an inherenttime delay between transcript synthesis and the onset of tran-script decay, avoiding a futile cycle. Caulobacter crescentus (Cau-lobacter) is a Gram-negative α-proteobacterium widely studied asa model system for asymmetric cell division. Unlike E. coli or B.subtilis, Caulobacter contains a pole-tethered chromosome thatfills the cytoplasm (7–9). Surprisingly, Caulobacter RNase E doesnot have a membrane-targeting sequence and is not membraneassociated (2, 10). Furthermore, diffraction-limited (DL) imagesof RNase E exhibited a patchy localization throughout the cell,with RNase E directly or indirectly associating with DNA (10).Moreover, the copy number of RNase E is regulated as a function ofthe Caulobacter cell cycle. The copy number reaches maxima duringthe swarmer to stalked cell transition and before cell division (2).Fluorescently labeled ribosomal proteins S2 and L1, proxies

for ribosomes in E. coli and B. subtilis, respectively, were foundenriched at the cell poles, spatially excluded from the nucleoidon the hundreds of nanometer scale (11, 12). Similar fluores-cence imaging studies in Caulobacter revealed no apparent sep-aration of ribosomes and the nucleoid, but did suggest thatribosome diffusion throughout the cell was spatially confined(10). However, because the Caulobacter cell is only ∼500 nm by∼3.5 μm in size, the DL resolution of conventional fluorescencemicroscopy would obscure any organization or motion of the

components of the degradosome or the ribosome that are onlength scales of less than ∼200 nm.We used a combination of live-cell single-particle tracking

(SPT) and fixed-cell superresolution (SR) microscopy to study thedynamics and spatial distribution of eYFP-labeled (13) RNase Eand ribosomal protein L1 in Caulobacter on subdiffraction limitlength scales. SPT and SR provide improved resolution down to∼20–50 nm using fluorescent proteins (FPs) (14, 15). Moreover,both SPT and SR are single-molecule (SM) methods, allowing usto investigate the heterogeneity in protein behavior and distribu-tion across many different cells. To avoid artifacts from projectingSM localizations onto two dimensions, we have used the double-helix point spread function (DH-PSF), which encodes 3D infor-mation in each SM image (16).Using these methods, we observed that the number of RNase

E clusters changed as a function of the cell cycle, while the size ofthe clusters remained constant. However, ribosomes formed onlyweak clusters, which were deployed throughout the cell, as wefound to be the case in another α-bacterium, Sinorhizobiummeliloti. For both RNase E and ribosomes, we observed a decreasein confinement and clustering when transcription was inhibited,likely due to depletion of RNA substrates for degradation, pro-cessing, and translation. We found that the localization of RNase Eclusters correlates with the subcellular chromosomal position of the

Significance

Regulated gene expression involves accurate subcellular posi-tioning and timing of RNA synthesis, degradation, processing,and translation. RNase E is an endoribonuclease that forms thescaffold of the RNA degradosome in bacteria, responsible forthe majority of mRNA turnover and RNA processing. We used3D superresolution microscopy and single-particle tracking toquantify the spatial distribution and dynamics of RNase E andribosomes in the asymmetrically dividing bacterium Caulo-bacter crescentus. Our results show that active transcriptionand RNA substrate availability facilitate confinement andclustering of both RNase E and ribosomes. RNase E clusterscolocalized with the subcellular position of the two Caulo-bacter rRNA gene chromosomal loci, indicating that RNA pro-cessing can be spatially organized in a bacterium according toits transcriptional profile.

Author contributions: C.A.B., J.W., M.K.L., J.M.S., L.S., and W.E.M. designed research;C.A.B., J.W., and M.K.L. performed research; C.A.B., J.W., M.K.L., and J.M.S. contributednew reagents/analytic tools; C.A.B., J.W., M.K.L., and W.E.M. analyzed data; and C.A.B.,J.W., L.S., and W.E.M. wrote the paper.

Reviewers: Z.G., Princeton University; and J.C.W., University of Wisconsin–Madison.

The authors declare no conflict of interest.

Published under the PNAS license.1Present address: Department of Biological Sciences, Wayne State University, Detroit,MI 48202.

2To whom correspondence should be addressed. Email: wmoerner@stanford.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1721648115/-/DCSupplemental.

Published online April 2, 2018.

E3712–E3721 | PNAS | vol. 115 | no. 16 www.pnas.org/cgi/doi/10.1073/pnas.1721648115

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

highly expressed rRNA genes that are positioned at two distinctsites on the chromosome, consistent with processing occurring atthe site of rRNA synthesis. Thus, cotranscriptional activities exhibitdynamic subcellular organization that is dependent on the site ofchromosomal readout.

ResultsDiffusion of RNase E and Ribosomal L1 Proteins Is Influenced byTranscriptional Activity.RNase E. The endoribonuclease RNase E forms the core andscaffold of the Caulobacter RNA degradosome. Other compo-nents include the exoribonuclease polynucleotide phosphorylase,a DEAD-box RNA helicase RhlB, the Krebs cycle enzyme aco-nitase, and the exoribonuclease RNase D (2, 17). To study themotional dynamics of RNase E, we performed 3D SPT in a mixedpopulation of live Caulobacter cells using a strain expressingRNase E-eYFP as the only copy of RNase E. We performed aninitial photobleaching and waited for stochastic recovery andblinking of eYFP SMs (18) (Fig. 1A, Left). We then obtained 3Dtrajectories of individual molecules by joining DH-PSF localiza-tions over consecutive frames (Materials and Methods) (Fig. 1A,Right). In contrast to E. coli in which RNase E diffuses on themembrane (3), RNase E molecules in Caulobacter (average copynumber is 3,100 molecules per cell as determined for a mixed cellpopulation; SI Appendix, Supplementary Methods) were foundthroughout the cytoplasm of the cell, diffusing limited distances(Fig. 1 B andD) and exhibiting the possibility of confined diffusion.To study the effect of RNA substrate availability on RNase E

dynamics, we treated the cells with rifampicin (rif), an antibioticthat blocks transcription and effectively depletes cells of RNA(19). The ensemble diffusion coefficient for wild-type (WT) RN-ase E was extracted by analysis of the ensemble-averaged mean-squared displacement (MSD) (Materials and Methods) (SI Ap-pendix, Fig. S1A). The average diffusion coefficient for WT cellswas found to be 0.0337 ± 0.0011 μm2/s (error is SEM as de-termined from 500 bootstrapped samples of individual tracks).The diffusion coefficient increased upon treatment with rif(0.0407 ± 0.0004 μm2/s) (Fig. 1C). There is a heterogeneity ofdiffusion coefficients across different trajectories (SI Appendix,Fig. S2) likely due to sampling a mixed population of cells invarious developmental stages, stochastic gene expression betweenindividual cells, and RNase E molecules not all acting on RNA (SIAppendix, Fig. S3, shows additional example cells).As the change in the ensemble diffusion coefficient was not

drastic, we analyzed individual trajectories of WT and rif-treatedcells to quantify their degrees of confinement. Each trajectory wasdivided into subtrajectories of five points (Materials and Methods).The maximum displacement from the first point of each sub-trajectory was then calculated (20, 21) (Materials and Methods)(Fig. 1E). Displacements were compared with simulated Brownianmotion inside a cell volume using ensemble WT and rif diffusioncoefficients (SI Appendix, Fig. S4) to control for the possibility thattrajectories appear confined by chance or due to the cell shape. Athreshold was set at the Brownian 5th percentile, such that only5% of the simulated Brownian subtrajectories were confined.Subtrajectories with displacements less than this value were clas-sified as confined. To include the possibility of an intermediatedegree of confinement, another threshold was arbitrarily set at theBrownian 50th percentile, and subtrajectories with displacementsgreater than this value were classified as clearly mobile. All othersubtrajectories (between 5th and 50th percentile) were classifiedas intermediate. Upon rif treatment, the fraction of confinedmolecules decreased from 50.1% in WT subtrajectories to 34.2%,while the fraction of mobile subtrajectories increased from 15.6%in WT to 22.8% (Fig. 1 F–H).Ensemble-averaged MSDs for the confined and mobile WT

subtrajectories were analyzed (SI Appendix, Fig. S5), yieldingdrastically different D values of 0.0025 ± 0.0001 μm2/s (confined)

and 0.1944 ± 0.0010 μm2/s (mobile). For rif-treated cells, thediffusion coefficient for confined subtrajectories is 0.0060 ±0.0001 μm2/s and for mobile subtrajectories is 0.1270 ±0.0072 μm2/s. To determine whether the confined subtrajectoriesare stationary or not, the apparent diffusion coefficient of sta-tionary RNase E-eYFP molecules was calculated using theensemble-averaged MSD of trajectories from fixed cells. Stationary,fixed particles were found to have an apparent diffusion coefficientof 0.0022 ± 0.0015 μm2/s (error is SEM from 143 trajectories).Using this comparison, we determined that the observed diffusioncoefficient for WT confined subtrajectories is consistent with astationary particle, whereas the diffusion coefficient for rif confinedsubtrajectories is larger than would be expected for a stationaryparticle given our localization precision (32/33/49 nm in x/y/z).The difference in diffusion between WT and rif-treated cells is

not an artifact due to a potential rif-induced chromosome struc-ture disruption (22). We performed the same 3D SPT of RNase Emolecules in a ΔHU2 background with significantly reduced nu-cleoid compaction (23). Results show similar confinement as thatof WT cells (SI Appendix, Fig. S6), arguing against the possibilityof a difference in chromosome packaging between WT and rif-treated cells contributing to the difference in RNase E diffusion.Ribosomes. To obtain information about ribosome dynamics andtranslation, we performed 3D SPT of ribosomes through imagingeYFP-labeled ribosomal protein L1 in a mixed population ofcells. We found that L1 (average copy number is 16,300 mole-cules per cell as determined for a mixed cell population; SIAppendix, Supplementary Methods) diffuses much more freelythroughout the cytoplasm than RNase E (Fig. 2 A and C and SIAppendix, Fig. S7, for additional examples). In contrast to thecase for RNase E, multiple populations of diffusing moleculeswere present by visual inspection, and we chose to differentiatethe various populations using another approach. We calculatedthe squared displacements of the trajectories using a time lag ofone frame (25 ms), then computed the empirical cumulativedistribution function (CDF) of these squared displacements. Theempirical CDF was then fit to a standard two-population model(Materials and Methods) (Fig. 2 D and E and SI Appendix, Fig.S8). We found easily observable diffusion: in WT cells, 89% ofthe population was found to have a diffusion coefficient of0.039 ± 0.001 μm2/s (larger than would be expected for a sta-tionary particle) (error is SEM as determined from 500 boot-strapped samples of individual squared displacements), while11% of the population was found to have a much larger diffusioncoefficient of 3.23 ± 0.12 μm2/s. We thus hypothesize that ourslow-diffusing population (89% of trajectories measured) rep-resents ribosomes involved in active translation slowed down byprotein and mRNA tethers. The faster-diffusing population(11% of total trajectories) are likely unincorporated L1 subunits.When cells were treated with rif (Fig. 2 B and C), 95% of the

L1 population was found to have a diffusion coefficient of 0.215 ±0.002 μm2/s, compared with untreated cells where 89% of thepopulation had a diffusion coefficient of 0.039 ± 0.001 μm2/s (Fig.2E). In rif-treated cells, 5% of the L1 population was found tohave a diffusion coefficient of 5.67 ± 0.39 μm2/s (Fig. 2E). Bothpopulations in rif-treated cells have diffusion coefficients thatwere greater than observed for WT cells. We did not detect theactively translating population (89%, D = 0.039 ± 0.001 μm2/s)observed in WT cells. The 95% L1 population in the rif-treatedcells could be attributed to fully assembled ribosomes that are nottethered to an mRNA, whereas the 5% fraction are unincorporatedL1 subunits. The reason for the faster movement of the un-incorporated subunits in the rif-treated cells compared withWT cellsis unclear and may be due to a change in overall cell viscosity.

RNase E and Ribosomal L1 Proteins Are Differentially Clustered inCaulobacter. As our SPT data have shown, RNase E and ribo-somes both display dynamic behaviors in the cell. To study their

Bayas et al. PNAS | vol. 115 | no. 16 | E3713

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

spatial distribution without the confounding effect of motion, weturned to SR imaging of fixed cells. The imaging time required to

obtain SR images with sufficient sampling is longer than thepersistence time of individual mRNAs (24, 25). Accordingly, the

xy z

D

E F

B WT C + rifampicin

H + rifampicinG WT

1

2

3

4

12

3 4

1

2

3

4

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

Displacement from firstpoint of subtrajectory (nm)

0 250 500 750 1000

Frac

tion

of s

ubtra

ject

orie

s

0

0.2

0.4

0.6

0.8

1

WT rif

Frac

tion

of s

ubtra

ject

orie

s

0

0.2

0.4

0.6

0.8

1

WT - dataWT - simulationWT - confinement thresholdrif - datarif - simulationrif - confinement threshold

confinedintermediatemobile

****

****

A

Fig. 1. SPT of RNase E molecules measures decreased confinement when transcription is inhibited by rif. (A, Left) Representative white-light (WL) anddiffraction-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 two2D Gaussians done through easy-DHPSF (Bottom row). The midpoint of the two lobes provides xy position, while the angle between the two lobes encodes axialposition. (B) Two representative WT cells with their RNase E trajectories. (C) Two representative rif-treated cells and trajectories. Each trajectory is plotted as adifferent 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 sub-trajectory 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 forconfined (pink and light blue dashed lines) and 50th percentile for mobile. All other subtrajectories are defined as “intermediate.” (F) Results from confinementanalysis. (G and H) Same cells as in B and C but with the subtrajectories color-coded according to their assigned confinement type. Analysis was performed on2,817 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, and H).]

E3714 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al.

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

cells were fixed with 1% formaldehyde to capture global RNaseE or ribosome locations at a specific time window. Because thephotophysical properties of eYFP (stochastic and reversible re-covery from a long-lived dark state) can cause artifacts that mimicprotein clustering (26), we removed localizations that were likelydue to oversampling (Materials and Methods).RNase E. We first performed 3D two-color SR imaging of RNaseE and the cell surface to confirm that Caulobacter RNase E is notsequestered to the cell membrane (10). In sharp contrast toRNase E sequestration to the inner membrane in E. coli (4, 5),RNase E in Caulobacter was observed to cluster along the centralaxis of the cell and nucleoid (Fig. 3 A–D), consistent with pre-vious DL imaging studies showing the cytoplasmic localization ofRNase E (10). RNase E formed distinct clusters in the cytoplasm(Fig. 3E and Movie S1), although we observed various degrees ofclustering across the mixed population due to stochastic sam-pling as well as cell-to-cell variability in gene expression (SIAppendix, Fig. S10).RNase E in fixed Caulobacter cells was more clustered than

would be expected from complete spatial randomness, indicatingthat the observed clustering is not due to chance or the cell shape(27) (Materials and Methods). Positive deviations from the graycurve (Fig. 3E, Bottom row) indicate clustering. The clustering washeterogeneous across the mixed cell population, likely arisingfrom both intrinsic and extrinsic noise (24, 28) (Fig. 3J). When

transcription was inhibited by rif, RNase E clusters dissipated andthe observed clustering shifted down toward complete spatialrandomness (Fig. 3F and Movie S2). To compare WT and rif-treated cells, we calculated a single clustering metric from theintegral between the two curves (Fig. 3I) for each cell that cap-tures the difference between the observed clustering in the dataand the artifactual small clustering in the complete spatial ran-domness simulations arising from the cell shape (Materials andMethods) (Fig. 3J). The rif-treated values were significantly lowerthan those for the WT cells, which captures the decreased clus-tering of molecules in rif-treated cells (P < 0.0001).Ribosomes. We acquired SR images from a mixed population offixed Caulobacter cells expressing ribosomal protein L1-eYFP (Fig.3G, SI Appendix, Fig. S11, and Movie S3 show additional examplecells). Ribosomes were observed throughout the cytoplasm, con-sistent with DL imaging (10) and cryo-electron microscopy (29).This differs from the ribosome/nucleoid segregation observed in E.coli and B. subtilis (30). Ribosomes in fixed Caulobacter cells ap-pear weakly clustered, yet slightly more clustered than what wouldbe expected from complete spatial randomness (Fig. 3G, Bottomrow). Performing the same clustering analysis as we did withRNase E, we observed a heterogeneous distribution of clusteringbetween different cells (Fig. 3K). Nevertheless, similar to RNase E,ribosome clustering exhibited a clear relationship with transcrip-tion, as treating the cells with rif significantly shifted the observed

A B

C

D E

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 atleast 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 plotof the CDF of squared displacements. (D, Bottom) Residuals of the fit to the CDF of squared displacements. (E) Summary of the two diffusing populations inWT 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 on11,249 trajectories from 298 WT cells and 12,757 trajectories from 326 rif-treated cells.

Bayas et al. PNAS | vol. 115 | no. 16 | E3715

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

clustering toward complete spatial randomness (Fig. 3 H and Kand Movie S4) (P < 0.0001).To investigate whether the cytoplasmic ribosomal distribution is

unique to Caulobacter, we imaged ribosomal L1-eYFP in anotherα-proteobacterium, S. meliloti, together with the SYTOX orangeDNA intercalating dye using epifluorescence microscopy. As a

control, we also imaged an E. coli strain containing ribosomalprotein S2-eYFP (30). Similar to Caulobacter, ribosomes in S.meliloti occupied the entire cell volume and colocalized with thenucleoid, whereas the γ-proteobacterium E. coli maintained segre-gation of the DNA and polar nucleoid-excluded ribosomes (SI Ap-pendix, Fig. S12). Notably, S. meliloti ribosomes contained a slightly

∫ Hdata - H sim × 1040 0.5 1 1.5 2 2.5 3

Nor

mal

ized

cou

nts

0

0.1

0.2

0.3RNase E

WTrif

WTrif

****

0 2 4 6 8

Nor

mal

ized

Cou

nts

0

0.1

0.2

0.3ribosomal L1

radius (nm)

H(r

) (nm

)

Hdata

Hsim

∫ Hdata - H sim

****

× 103

I J K

E F G HRNase E WT RNase E rif ribosomal L1 rifribosomal L1 WT

radius (nm)0 100 200

H(r

) (nm

)

-50

0

50

100

150

radius (nm)0 100 200

H(r

) (nm

)

-50

0

50

100

150

radius (nm)0 100 200

H(r

) (nm

)

-50

0

50

100

150

radius (nm)0 100 200

H(r

) (nm

)

-50

0

50

100

150dataCSR

dataCSR

dataCSR

dataCSR

∫ Hdata - H sim

A

RNaseE-eYFPRNaseE-eYFPrhodamine lactam surface labelrhodamine lactam surface label

0 100 200 300 400 500

Cou

nts

0

10

20

30

40

50

60

Median distance to central axis = 109 nm

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

Cou

nts

0

50

100

150

200

250

Median distance to central axis = 272 nm 300 nm yz slice

Distance to central axis (nm)

B C D

200 nm

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 surfacelabeled with rhodamine spirolactam (magenta). (B) Distribution of distances of RNase E molecules to the cell’s central axis. (C) Distribution of distances ofrhodamine lactam molecules to the cell’s central axis. (D) A yz projection of a 300-nm slice perpendicular to the cell axis. (E) SR reconstruction of a fixed WTCaulobacter 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) SR re-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 lo-calizations 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 andribosomes, respectively. Cells are on 1-μm grids. (I) The degree of clustering quantity plotted in F and Gwere calculated from the area between the data curve andCSR. (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 clusteringrelative 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.

E3716 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al.

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

more patchy localization pattern than Caulobacter, but generallyshowed a similar interpenetrating nucleoid–ribosome configurationsimilar to that observed with Caulobacter.

Number of Clusters of RNase E Changes as a Function of the Cell Cycle.Ribosome profiling (31) and Western blot analysis (2) haveshown that RNase E levels in Caulobacter vary as a function ofthe cell cycle, with maximum levels at the swarmer to stalkedtransition and just before cell division. To determine whetherRNase E clusters fluctuate as a function of the cell cycle, weperformed two-color 3D imaging of RNase E-eYFP andPAmCherry-PopZ (to mark the cell poles). We imaged syn-chronized cells (Materials and Methods) fixed with 1% formal-dehyde to capture snapshots of the RNase E clusters at distincttime points along the cell cycle (Fig. 4 A–E). Our 3D recon-structions show RNase E clusters along the cell axis throughoutthe cell cycle and PopZ foci at the cell poles.We then analyzed the sizes of RNase E clusters using the

density-based algorithm DBSCAN (Materials and Methods) (Fig.4F). There was no change in the median number of moleculesdetected in each RNase E cluster as cells progressed through thecell cycle. However, we did find an increase in the number ofRNase E clusters as the swarmer cells differentiated into stalkedcells (Fig. 4G). During the swarmer to stalked cell transition, themedian number of clusters in the cell doubled from two clustersto four clusters in keeping with the previous observation that theabundance of RNase E varies during the cell cycle with a max-imum at the swarmer to stalked cell differentiation (2). Thehistogram for the final 125-min population (Fig. 4G, Right)shows the existence of two populations, since this cell type has

both swarmer and stalked cells. The increase in the number ofclusters in the cell during the swarmer to stalked transition isconsistent with biochemical data showing increased overall RNAabundance during this time in the cell cycle (31). This increase inclusters may also be attributed to the initiation of DNA repli-cation that occurs during this time period leading to an increasein overall transcription.

RNase E Clusters Colocalize with Highly Transcribed Genes. Our ob-servation that the number of RNase E clusters decreased upon adecrease in RNA substrate availability after treatment with rifsuggested that RNase E clusters might form near regions of hightranscriptional activity. Chromosomal loci along Caulobacter’ssingle circular chromosome occupy specific subcellular locations(32). Using this distinct organization of the chromosome, wewere able to ask whether RNase E clusters correspond to DNAloci that are highly transcribed. The genes encoding rRNA arethe most highly transcribed loci, comprising 95–97% of theprokaryotic cell’s total RNA mass (33). There are two near-identical copies of rDNA in Caulobacter (Fig. 5A) that occupyspatially distinct sites on the chromosome. Each of these lociencode 16S, 23S, and 5S rRNA, as well as two tRNAs. TheseRNAs are transcribed as a single transcript that is subsequentlyprocessed by RNases, including RNase E, to generate the maturerRNA species (1, 2, 34).To determine whether the observed RNase E clusters are

positioned near the DNA loci of rRNAs, we labeled the tworDNA operons (rDNA1 and rDNA2) with a tetO/tetR labelingsystem (Fig. 5B) and visualized the positions of RNase E in amixed population of cells with respect to the two rRNA gene loci

Ncells = 74 Ncells = 75 Ncells = 73 Ncells = 52 Ncells = 33

0 50 1000

0.2

0.4

0.6

0 50 1000

0.1

0.2

0.3

0.4

0 50 1000

0.1

0.2

0.3

0.4

0 50 1000

0.1

0.2

0.3

0.4

0 50 1000

0.2

0.4

0.6median = 13median = 13median = 13median = 12median = 12

Number of molecules per RNaseE cluster

Nor

mal

ized

cou

nts

F

Nor

mal

ized

cou

nts

0 5 100

0.1

0.2

0.3

0.4

0 5 100

0.1

0.2

0.3

0 5 100

0.1

0.2

0.3

0 5 100

0.1

0.2

0.3

0 5 100

0.05

0.1

0.15

0.2median

= 4median = 4median = 4median = 4median = 2

Number of RNaseE clusters per cell

G

RNaseE-eYFP PAmCherry-PopZ

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

Fig. 4. Cell cycle-dependent spatial distribution of RNase E-eYFP in fixed Caulobacter cells. (A–E) SR reconstructions of fixed Caulobacter cells expressingRNase E-eYFP and PAmCherry-PopZ at (A) 5 min, (B) 35 min, (C) 65 min, (D) 95 min, and (E) 125 min after synchrony. The total cell cycle is 150 min. Cells are on1-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 Eclusters 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 clustersper cell during the swarmer to stalk transition.

Bayas et al. PNAS | vol. 115 | no. 16 | E3717

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

by DL imaging. The positions of the rDNA gene loci were foundby SR imaging, and these positions were plotted against theRNase E DL images (Fig. 5 C–E) (Materials and Methods). Wefound that the two rDNA loci each colocalized with an RNase Ecluster. Specifically, 79.2% of the first rDNA locus and 76.9% ofthe second rDNA locus colocalized with an RNase E cluster(additional examples are in SI Appendix, Figs. S13–S15). Theseresults suggest that RNase E processing of rRNA transcriptsoccurs at the site of rRNA synthesis. Within each cell, we esti-mated that the fractions of RNase E that colocalized with eachof the two rDNA loci are 22 ± 9% (error is SD) and 24 ± 9%(error is SD), respectively, suggesting that approximately one-half of RNase E molecules are involved in rRNA processing(SI Appendix, Supplementary Methods).To confirm that the RNase E/rDNA correlations are valid, we

also imaged a lowly transcribed gene, CCNA_01879, that is in thebottom 0.5% of RNA-seq analysis (31) (Fig. 5A). In contrast torDNA, we found a lack of correlation of an RNase E cluster withthis poorly transcribed locus. Only 27.1% of CCNA_01879 locicolocalized with an RNase E cluster. There was still somecolocalization, possibly because within any given cell, a fewRNase E clusters are present, which coincidentally can colocalizewith our DNA labels. These results support the hypothesis thatRNase E clusters are preferentially localized to highly tran-scribed DNA loci where they immediately process RNA, mostlikely cotranscriptionally.

DiscussionRNase E is the core of the RNA degradosome complex. It hasbeen shown to be membrane associated in E. coli and B. subtilis(3–6), but deployed throughout the nucleoid in Caulobacter (10).Previous studies of ribosome localization in E. coli (30, 35) andB. subtilis (12) revealed enrichment at the cell poles where they

are excluded from the nucleoid, while similar studies in Caulo-bacter revealed no apparent separation of ribosomes and the nu-cleoid (10). However, due to the small size of Caulobacter, the DLresolution of standard fluorescence microscopy would obscure allsub-DL motion or organization. Therefore, in this study, we used3D SPT and SR microscopy to probe and quantify the spatialorganization of RNase E and ribosomes at high resolution, therebyproviding insight into the dynamic behavior of these complexes.Given that RNase E clusters were found in the cytoplasm in a

pattern overlapping that of the nucleoid, there is unlikely to be agross physical separation of the RNA production and degrada-tion machinery. We found the subcellular position of RNase Eclusters correlated with sites of rRNA synthesis, suggesting thatcleavage occurs where RNA substrate is produced. This obser-vation is consistent with evidence that the rRNA transcript isprocessed cotranscriptionally (36), and previous studies thathave shown RNase E associating with DNA (10). The increasedlocal concentration of RNase E clusters near highly transcribedgenes allows a large number of substrates to be quickly degradedor processed. RNase E clusters might reflect the formation of alarge number of RNA degradosome complexes, which couldefficiently facilitate RNA cleavage. The size of these clusters didnot change during cell cycle progression, suggesting that theremight exist an optimum number of molecules per cluster. Theobserved confined subtrajectories are likely actively processingRNAs in large RNA degradosome assemblies, possibly actingcotranscriptionally on transcripts still tethered to the chromo-some. Mobile subtrajectories could belong to RNase E mole-cules that have not yet bound an RNA substrate. Moreover, thedecrease in the fraction of confined RNase E molecules andlower degree of confinement in cells in which RNA has beendepleted highlights RNase E’s relationship with substrateavailability.

rDNA1rDNA2Control

A

B

C

D

WL DNA label

Caulobacter crescentus 4.01 Mbp

0 MbpOri

2.0 MbpTer

2.87 MbprDNA2

3.80 MbprDNA1

2.02 MbpControl

RNaseE

E

TetO TetO TetO ...... 85x

TetR-eYFP

gene

rDNA1

rDNA2

Control

79%

77%

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 usingthe Caulobacter swarmer 1D chromosome map as described previously (32). (B) A schematic of the DNA-labeling scheme. (C, Left) A white-light (WL) image oflive Caulobacter cells expressing RNase E-PAmCherry and tetR-eYFP/85x-tetO on rDNA1 locus. (C, Middle) A diffraction-limited (DL) image of the tetR-eYFPlabeling 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 withCCNA_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 per-formed on 506 rDNA1 cells, 664 rDNA2 cells, and 321 control cells.

E3718 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al.

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

Our SPT experiments in WT Caulobacter show two pop-ulations of ribosomal L1: an 89% slow species (D = 0.039 ±0.001 μm2/s) and an 11% fast species (D = 3.23 ± 0.12 μm2/s)(Fig. 2E). We attribute the slow 89% to ribosomes actively en-gaged in translation, and this population includes bothchromosome-tethered polysomes and actively translating singleribosomes. The diffusion coefficient we found for the activelytranslating ribosomes is consistent with reported diffusion coef-ficients of E. coli ribosomes engaged in active translation (30,35). The fast 11% are likely free, unincorporated L1 subunits—their observed diffusion coefficient is consistent with the diffu-sion of free FPs in bacteria (35, 37, 38). In contrast to our resultsin which we found observable diffusion of actively translatingribosomes, a previous fluorescence recovery after photobleaching(FRAP) study of Caulobacter ribosomes observed essentially sta-tionary actively translating ribosomes (D = 0.0011 μm2/s) (39).This discrepancy could be attributed to the slower temporal res-olution and lower spatial resolution of FRAP. As opposed to SPT,the DL resolution of FRAP is unable to quantify motions onscales smaller than ∼200 nm.Upon rif treatment, we observed two populations of ribosomal

L1: a 95% species (D = 0.215 ± 0.002 μm2/s) and a 5% species(D = 5.67 ± 0.39 μm2/s). The diffusion coefficient we observed inWT cells that we attributed to actively translating ribosomes wasnot observed in rif-treated cells, likely due to RNA depletion.Instead, the 95% L1 species in the rif population had a 5.5-foldfaster diffusion coefficient compared with the slow (D = 0.039 ±0.001 μm2/s) 89% observed in WT cells. This suggests that the95% L1 species in the rif-treated cells may be fully assembledribosomes that have not yet bound an mRNA. Importantly, the5% fast species in rif-treated cells has a diffusion coefficientclose to the 11% fast species in WT cells, and we thus attributethis population to unincorporated L1 subunits.The weak clustering of ribosomes we observed in our SR

measurements shows that there may be small compartments pre-sent within the cell that are able to confine ribosomes and form along-range diffusion barrier. We observed these weakly clusteredribosomes throughout the cytoplasm, occupying the same regionsas the nucleoid. This result is consistent with previous DL imagingof ribosomes and the observation that transcription and translationoccur throughout the nucleoid (10). Importantly, the ribosome andnucleoid association we observed is not unique to Caulobacter. Wefound a similar pattern in S. meliloti, suggesting that this ribosome/nucleoid organization may be a feature shared by α-proteobacteria.These results contrast with studies from E. coli and B. subtilis. Forinstance, ribosomes in B. subtilis are localized at regions distinctfrom the nucleoid due to actively transcribed regions of the nu-cleoid being exposed to the cell periphery (12, 40), and our controlexperiments in the γ-proteobacterium E. coli recapitulated thepreviously observed ribosome and nucleoid segregation (30, 35).These results suggest that the differing Caulobacter and E. coli ri-bosome organizations may be due to features unique to each or-ganism such as chromosome organization or cytoplasmic viscosity,and future studies are needed to elucidate these mechanisms.Taken together, our results show that even without the complex

intracellular compartments of eukaryotic systems (41), bacteriaspatially organize RNA processing, degradation, and translationwithin the nucleoid, and that this organization is governed by thetranscriptional profile of the chromosome.

Materials and MethodsPreparation of Cell Strains for Imaging. Caulobacter cells were grown over-night from frozen culture in 5 mL of PYE growth medium (42) with shakingin a 28 °C water bath. The day before imaging, cultures were diluted1:1,000 into the defined minimal media M2G (43). After growing to midlogphase, 1-mL aliquots of cells were washed once with M2G by centrifugation[3 min at 8,000 rpm, 4 °C (Eppendorf 5430R, Eppendorf FA-45-24-11-HS ro-tor)], and the pellet was resuspended into clean M2G. Synchronized cells

were prepared using a Percoll density gradient (44), allowing the cells a 5-min recovery in M2G at 28 °C before moving on to the next step. rif wasadded at 50 μg/mL final concentration [from 10 mg/mL stock solutions inDMSO (Fisher)] and incubated with the cells for 30 min with shaking. Thecells were washed by centrifugation [3 min, 8,000 rpm, 4 °C (Eppendorf5430R, Eppendorf FA-45-24-11-HS rotor)] and resuspension in clean M2G.Cells were fixed with the addition of 1% formaldehyde (Lot 091161; FisherScientific), followed by a 10-min incubation at room temperature and 30-min incubation on ice, and finally washed three times with clean M2G.

Before imaging, the cells were resuspended in ∼20–50 μL of minimal me-dium, producing a concentrated cell suspension. To this suspension, ∼1 nM offiducial markers were added (540/560 carboxylate-modified FluoSpheres,100-nm diameter; Molecular Probes), and then 1–2 μL of this mixture wasdeposited onto an agarose pad [composed of 1.5% (wt/wt) of low meltingpoint agarose (Invitrogen) in M2G] and mounted onto an argon plasma-etched glass slide (Fisher) and imaged immediately.

For imaging the tetO/tetR-labeled strains, the cells were incubated with0.03% xylose final concentration [from 30% (wt/vol) stock solutions in water]for 30 min before washing once with M2G and resuspending in M2G (∼20–50μL), and imaged immediately on an agarose pad as described above.

SM Imaging for SR Microscopy and SPT. Both SPT and SR microscopy rely onlocalizing single molecules with high precision; SPT is based upon localizingthe same molecule over time to gain information about this molecule’sdynamics. In contrast, SR microscopy localizes many different SMs over timeand then builds up a reconstruction to gain information about the staticstructure of interest. Since the imaging time required to acquire enough SMlocalizations for SR imaging can be long (approximately minutes), thestructure of interest must be stationary on the timescale of imaging.

Cells immobilized on agarose pads were imaged on a home-built epi-fluorescence microscope (Nikon Diaphot 200) with additional optics insertedbefore the 512 × 512-pixel Si EMCCD camera (Andor Ixon DU-897), whichenables two-color, 3D SR imaging described previously (45). Fluorescenceemission from the sample was collected through a high-N.A. oil-immersionobjective (Olympus UPlanSapo 100×/1.40 N.A.), filtered by a dual-pass di-chroic mirror (zt440/514/561rpc; Chroma), a 560-nm dichroic beamsplitter(Semrock FF560-FDi01), a 514-nm long-pass filter (Semrock LP02-514RE), a561-nm notch filter (Semrock NF03-561E), and a bandpass filter (FF01-532/610; Semrock). Two-dimensional white-light transmission images of the cellswere recorded before imaging. The eYFP was pumped with 514-nm excita-tion (Coherent Sapphire 514–100 CW, ∼0.5–1 kW/cm2). A 405-nm laser (Obis,∼0.1–10 W/cm2) was used to photoactivate and a 561-nm laser (CoherentSapphire 561–100 CW, ∼0.5 kW/cm2) was used to excite fluorescence fromPAmCherry. Laser intensity was determined by measuring the power at thestage with a power meter (Newport 1918-C) and estimating the FWHM spotsize by fitting the average fluorescence from the sample to a 2D Gaussian bynonlinear least squares using a custom MATLAB (MathWorks) program. In-tegration times were 25 ms for SPT data and 50 ms for SR data.

Data Analysis Procedure.Fitting 3D SM data. Both 3D SR and SPT data were fit using a modified versionof Easy-DHPSF (46), which is freely available at https://sourceforge.net/pro-jects/easy-dhpsf/. Calibration scans were created by axially scanning ananohole array (NHA) (47) filled with ∼25 μL of ∼50 μM Alexa 568 (LifeTechnologies) and Atto 520 (Sigma) dyes using a piezoelectric objectivescanner (C-focus; Mad City Labs). These calibration scans were used to relateangular lobe orientation to axial position, and as templates to locate can-didate SMs in the data. The NHA allowed us to account for field-dependenterrors arising from the differing rotation rates, and hence apparent axialposition, of the DH-PSF as a function of lateral position. The backgroundphotons (∼16–22 photons/pixel) were estimated using a median filter inMATLAB (46). On average, we detected 1,327 ± 449 photons per 25-msframe for the eYFP-labeled RNase E and ribosomes. Localization precisionwas estimated with an empirical formula derived from repeatedly localizingsingle beads under variable background conditions (45). Images included inthis paper have typical average localization precisions of 32 ± 8 nm (error isSD), 33 ± 9 nm, and 49 ± 13 nm in x, y, and z. Systematic errors from sampledrift resulting from mechanical and thermal fluctuations were corrected byadding ∼1–10 nM concentration of bright 560/580 Fluospheres (Life Tech-nologies) to the sample and using these as fiducial beads. High-frequencynoise in the drift is removed with wavelet filtering in MATLAB (45, 46).Two-channel registration. Arrays of control point positions were generated byaxially scanning the NHA (47) filled with ∼25 μL of ∼50 μM Alexa 568 (LifeTechnologies) and Atto 520 (Sigma) dyes using a piezoelectric objectivescanner (C-focus; Mad City Labs) and imaging the NHA simultaneously in the

Bayas et al. PNAS | vol. 115 | no. 16 | E3719

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

two channels using both the 514- and 561-nm lasers. The red channel wastransformed to the green channel using the same algorithm describedpreviously (45). Target registration errors ranged from 5 to 15 nm, and fi-ducial registration errors ranged from 4 to 12 nm.Removing oversampling. Oversampling, caused by including multiple locali-zations of the same protein from FP labels that emit for more than one frame,creates spurious clustering. Before analyzing clustering in fixed cells, over-sampling was removed with a radial and temporal filter. Localizations withina set radial and time region are assumed to belong to the same molecule andare combined into a single localization located at the average position. Areasonable radial distance was determined by simulating SR data containingoversampling on the size scale of the localization precision. The thresholdradius (as a function of the mean localization precision) was varied until agrouping algorithm in MATLAB recovered the ground truth number ofsimulated single molecules (thus estimating the grouping of the oversampledclusters). The optimized spatial threshold was found to be 1.5 × mean lo-

calization precision (ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiσx2 + σy2 + σz2

p). The temporal threshold was de-

termined individually for each cell dataset by calculating the probability offinding another localization within the spatial threshold as a function oftime. The probability typically followed a single-exponential decay. The timeconstant of the exponential was used as the temporal threshold.Diffusion analysis. Trajectories were constructed by connecting localizationsfrom consecutive frames. Only trajectories of at least four steps (five frames or125 ms) were used for MSD analyses. The maximum allowed xyz displace-ments of RNase E-eYFP and L1-eYFP molecules over 25 ms were set to1,000 nm. For both RNase E-eYFP and ribosomal L1-eYFP, we detect a meantrajectory length of seven frames (175 ms). Diffusion coefficients were cal-culated from ensemble-averaged 3D MSDs of individual RNase E moleculesusing the first two time lags to avoid long-time effects of confinement andnon-Brownian motion. The reported errors are the SEMs of diffusion coef-ficients as determined by bootstrapping (50% of the trajectories weresampled 500 times). Error bars in the MSD curves are SEMs.Simulating trajectories inside cell-shaped object. Brownian simulations wereperformed inside a cylindrical cell of radius 250 nm and length 3.5 μm withtwo hemispherical caps; finite sampling from a camera and motion blurwere taken into account by averaging 10 positions for every 25-ms frame.Each 3D position was then given an xyz “kick” from a Gaussian distributionto simulate localization precision.Confinement analysis. To test for confinement, we divided each trajectory [of atleast four steps (five frames or 125 ms) in length] into subtrajectories of foursteps (five frames or 125 ms). For each subtrajectory, the maximum dis-placement from the first point of that subtrajectory was calculated (20, 21).Weperformed simulations of Brownian diffusers inside a cylindrical cell volumewith hemispherical caps to determine whether a subtrajectory was confinedor not. The effect of confinement, such as in the case of cytoplasmic moleculesdiffusing in a cell, results in a smaller value for the apparent, observed dif-fusion coefficient than in the case in which the molecule is not confined bycell walls. To ensure that we were making a valid comparison of diffusioncoefficients in our confinement analyses (48), we performed Brownian simu-lations of a cytoplasmic molecule inside a cylindrical cell with hemisphericalcaps. We used an empirical diffusion coefficient that resulted in an apparentdiffusion coefficient approximately equivalent to the ensemble-averagedmeasured diffusion coefficients for WT and rif-treated cells (SI Appendix,Fig. S4). A threshold was set such that only 5% of the simulated Browniansubtrajectories are confined. Subtrajectories with displacements less than thisvalue (118 nm for WT and 123 nm for rif-treated, plotted as vertical dottedlines in Fig. 1E) were classified as confined. From the cumulative distributionsshown in Fig. 1E, this criterion would imply that ∼50% of the WT sub-trajectories were mobile. However, we found that many molecules in thisgroup did not diffuse very far, as suggested by fixed snapshots of the spatialdistribution (Fig. 3E). To include the possibility of an intermediate degree ofconfinement, another threshold was arbitrarily set at the Brownian 50thpercentile, and subtrajectories with displacements greater than this value(198 nm for WT and 208 nm for rif-treated) were classified as clearly mobile.All other subtrajectories were classified as intermediate.Fitting squared displacement CDFs to multiple populations. To extract the dif-ferent populations of ribosomal L1 proteins, we fit the empirical CDF of thesquared displacements of our trajectories to Eq. 1 (49–51), where r2 are theexperimental squared displacements, τ is the time lag, and xi is the weight ofpopulation i. Eq. 2 (in which d is the number of dimensions, Di is the diffusioncoefficient of population i, Δt is the camera exposure time, σj is the locali-zation σ in the jth dimension) is substituted into Eq. 1 to perform the fit. Thefitting is constrained by Eq. 3 to ensure that all weights sum to 1. The fitting

is performed in MATLAB, minimizing the objective function ðPdata − PfitÞ2using the function fmincon:

P�r2, τ

�= 1−

Xni=1

xi exp

−r2

ri2

!, [1]

ri2 = 2dDi

�τ−

Δt3

�+ 2Xdj=1

σj2, [2]

Xni=1

xi = 1. [3]

Fitting with three populations yields the same result as fitting with only twopopulations (SI Appendix, Fig. S9).

We performed a validation of finding two populations of Brownian dif-fusers using this method by simulating particles with two different diffusioncoefficients, extracting the parameters in Eq. 1, and comparing the simulateddata from the fit. The percent discrepancies for the slow vs. fast fractions are5% vs. 3%; the percent discrepancies for the slow vs. fast D values are 33%vs. 12%. Detailed results can be found in SI Appendix, Fig. S16.Clustering analysis by spatial point statistics. To quantify the degree of clusteringin our SR images, we used spatial point statistics. Complete spatial ran-domness (CSR) was simulated within each cell volume using an identicalnumber of points with a custom MATLAB (MathWorks) program. The sim-ulated points were then thinned to be the same number of localizations asthe data, then a random “kick” of localization precision error (pulled from aGaussian distribution of the mean localization precision of the data) in allthree dimensions to each simulated point. In all cases, 100 simulations wereperformed, as this balanced computational time with statistical information.

The uncertainty in the data from the localization precision was alsobounded by simulation. Each detected localization had its own localizationprecision error. A cloud of points (100 simulations) with shape dictated by theindividual localization precisionwere simulated around each localization. Thesame statistical metrics were calculated on these simulated points, and the95% confidence bounds were used to report the error in the data from thelocalization precision.

Global clustering is quantified using Ripley’s K-function (27) and its nor-malized forms (52). As previously described, Ripley’s K-function representsthe average probability of finding localizations a distance r from the typicallocalization (53). For CSR, K(r) scales with the volume and is described by Eq.4, where bdrd is the volume of the unit sphere in d dimensions. Deviationsfrom this shape indicate clustering or repulsion; however, visualizing devi-ations from a straight or flat line is simpler than for an exponential curve.The normalized K-function is given by L(r) (Eq. 5), which can further benormalized into the H(r) function, such that CSR is a flat horizontal line atH= 0 (Eq. 7) (54). To be able to analyze clustering across many cells, thequantity shown in Eq. 7 was calculated for each cell:

KðrÞ=bdrd , [4]

LðrÞ=ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiKðrÞ=bd

dp

, [5]

HðrÞ= LðrÞ− r, [6]

Zr

HdataðrÞ−HCSRðrÞ. [7]

Clustering analysis by density threshold. For the 3D two-color RNase E/PopZdatasets, we use a density-based clustering algorithm, density-based spatialclustering of applications with noise (DBSCAN) (55), which detects clustersaccording to the local density of localizations within a certain search radius.For the RNase E clusters, we used the parameters « = 70 nm and minPts = 6;for the PopZ clusters, we used the parameters « = 200 nm and minPts = 7.These thresholds were chosen empirically.

In the RNase E-eYFP/PAmCherry-PopZ cells, the clustering results for RNaseE did not give an absolute count of the number of molecules in a cluster, butrather gave the relative amounts of number ofmolecules detected. This is dueto the photophysical properties of eYFP. Imaging SMs of eYFP requires aninitial bleach down of the sample, then waiting for the stochastic recoveryand blinking of this FP. The ideal situation is one in which each copy of eYFPrecovers. However, in practice, only about 5–15% of the molecules recoverand are detected.Superlocalization of labeled loci and colocalization. Fifty frames (100-ms expo-sure) were summed for tetR-eYFP and RNase E-PAmCherry. TetR-eYFP lociwere superlocalized by fitting 2DGaussians (which well-approximates the PSFof the microscope) in ThunderSTORM (ImageJ). These localizations were then

E3720 | www.pnas.org/cgi/doi/10.1073/pnas.1721648115 Bayas et al.

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0

transformed on to the RNase E-PAmCherry images using themapping functiongenerated with the following method. The array of 500 control points thatwere detectable in both channels simultaneously and evenly distributed acrossthe whole field of view was localized by fitting a 2D Gaussian in Thunder-STORM. These localizations were used to generate a mapping function be-tween the two channels using local weighted mean transformation in customMATLAB scripts to take into account field dependence of the transformation.Percentage of DNA loci colocalizing with an RNase E cluster was determinedby visual inspection of the transformed locus with respect to the RNaseE DL image.

ACKNOWLEDGMENTS. We thank the laboratory of James Weisshaar at theUniversity of Wisconsin–Madison for sending the E. coli S2 ribosome strain,Melanie Barnett and Jillynne Quinn in the laboratory of Sharon Long (De-partment of Biology, Stanford University) for help creating the ribosome S.meliloti strain, Thomas H. Mann for help with constructing the RNase E-eYFP/PAmCherry-PopZ strain, and Alex von Diezmann for the Brownian dif-fusion simulation code. This work was supported in part by National Instituteof 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 ChanZuckerberg Biohub Investigator, J.W. is a Stanford Center for MolecularAnalysis and Design Fellow, and J.M.S. was an NIH Postdoctoral Fellow(F32-GM100732).

1. Mackie GA (2013) RNase E: At the interface of bacterial RNA processing and decay.Nat Rev Microbiol 11:45–57.

2. Hardwick SW, Chan VSY, Broadhurst RW, Luisi BF (2011) An RNA degradosome as-sembly in Caulobacter crescentus. Nucleic Acids Res 39:1449–1459.

3. Strahl H, et al. (2015) Membrane recognition and dynamics of the RNA degradosome.PLoS Genet 11:e1004961.

4. Moffitt JR, Pandey S, Boettiger AN, Wang S, Zhuang X (2016) Spatial organizationshapes the turnover of a bacterial transcriptome. eLife 5:e13065.

5. Fei J, et al. (2015) RNA biochemistry. Determination of in vivo target search kinetics ofregulatory noncoding RNA. Science 347:1371–1374.

6. Cascante-Estepa N, Gunka K, Stülke J (2016) Localization of components of the RNA-degrading machine in Bacillus subtilis. Front Microbiol 7:1492.

7. Ptacin JL, et al. (2010) A spindle-like apparatus guides bacterial chromosome segre-gation. Nat Cell Biol 12:791–798.

8. Lee SF, Thompson MA, Schwartz MA, Shapiro L, Moerner WE (2011) Super-resolutionimaging of the nucleoid-associated protein HU in Caulobacter crescentus. Biophys J100:L31–L33.

9. Ptacin JL, et al. (2014) Bacterial scaffold directs pole-specific centromere segregation.Proc Natl Acad Sci USA 111:E2046–E2055.

10. Montero Llopis P, et al. (2010) Spatial organization of the flow of genetic informationin bacteria. Nature 466:77–81.

11. Bakshi S, Choi H, Weisshaar JC (2015) The spatial biology of transcription and trans-lation in rapidly growing Escherichia coli. Front Microbiol 6:636.

12. Lewis PJ, Thaker SD, Errington J (2000) Compartmentalization of transcription andtranslation in Bacillus subtilis. EMBO J 19:710–718.

13. Dickson RM, Cubitt AB, Tsien RY, Moerner WE (1997) On/off blinking and switchingbehaviour of single molecules of green fluorescent protein. Nature 388:355–358.

14. Tuson HH, Biteen JS (2015) Unveiling the inner workings of live bacteria using super-resolution microscopy. Anal Chem 87:42–63.

15. Gahlmann A, Moerner WE (2014) Exploring bacterial cell biology with single-moleculetracking and super-resolution imaging. Nat Rev Microbiol 12:9–22.

16. von Diezmann A, Shechtman Y, Moerner WE (2017) Three-dimensional localization ofsingle molecules for super-resolution imaging and single-particle tracking. Chem Rev117:7244–7275.

17. Voss JE, Luisi BF, Hardwick SW (2014) Molecular recognition of RhlB and RNase D inthe Caulobacter crescentus RNA degradosome. Nucleic Acids Res 42:13294–13305.

18. Biteen JS, et al. (2008) Super-resolution imaging in live Caulobacter crescentus cellsusing photoswitchable EYFP. Nat Methods 5:947–949.

19. Campbell EA, et al. (2001) Structural mechanism for rifampicin inhibition of bacterialRNA polymerase. Cell 104:901–912.

20. Milenkovic L, et al. (2015) Single-molecule imaging of Hedgehog pathway proteinsmoothened in primary cilia reveals binding events regulated by Patched1. Proc NatlAcad Sci USA 112:8320–8325.

21. Thompson MA, Casolari JM, Badieirostami M, Brown PO, Moerner WE (2010) Three-dimensional tracking of single mRNA particles in Saccharomyces cerevisiae using adouble-helix point spread function. Proc Natl Acad Sci USA 107:17864–17871.

22. Le TB, Imakaev MV, Mirny LA, Laub MT (2013) High-resolution mapping of the spatialorganization of a bacterial chromosome. Science 342:731–734.

23. Hong SH, McAdams HH (2011) Compaction and transport properties of newly repli-cated Caulobacter crescentus DNA. Mol Microbiol 82:1349–1358.

24. Taniguchi Y, et al. (2010) Quantifying E. coli proteome and transcriptome with single-molecule sensitivity in single cells. Science 329:533–538.

25. Bernstein JA, Khodursky AB, Lin PH, Lin-Chao S, Cohen SN (2002) Global analysis ofmRNA decay and abundance in Escherichia coli at single-gene resolution using two-color fluorescent DNA microarrays. Proc Natl Acad Sci USA 99:9697–9702.

26. Durisic N, Laparra-Cuervo L, Sandoval-Álvarez A, Borbely JS, Lakadamyali M (2014)Single-molecule evaluation of fluorescent protein photoactivation efficiency using anin vivo nanotemplate. Nat Methods 11:156–162.

27. Ripley BD (1977) Modelling spatial patterns. J R Stat Soc Series B Stat Methodol 39:172–192.

28. Elowitz MB, Levine AJ, Siggia ED, Swain PS (2002) Stochastic gene expression in asingle cell. Science 297:1183–1186.

29. Bowman GR, et al. (2010) Caulobacter PopZ forms a polar subdomain dictating se-quential changes in pole composition and function. Mol Microbiol 76:173–189.

30. Bakshi S, Siryaporn A, Goulian M, Weisshaar JC (2012) Superresolution imaging ofribosomes and RNA polymerase in live Escherichia coli cells. Mol Microbiol 85:21–38.

31. Schrader JM, et al. (2016) Dynamic translation regulation in Caulobacter cell cyclecontrol. Proc Natl Acad Sci USA 113:E6859–E6867.

32. Viollier PH, et al. (2004) Rapid and sequential movement of individual chromosomalloci to specific subcellular locations during bacterial DNA replication. Proc Natl AcadSci USA 101:9257–9262.

33. Rosenow C, Saxena RM, Durst M, Gingeras TR (2001) Prokaryotic RNA preparationmethods useful for high density array analysis: Comparison of two approaches.Nucleic Acids Res 29:E112.

34. Feingold J, Bellofatto V, Shapiro L, Amemiya K (1985) Organization and nucleotidesequence analysis of an rRNA and tRNA gene cluster from Caulobacter crescentus.J Bacteriol 163:155–166.

35. Sanamrad A, et al. (2014) Single-particle tracking reveals that free ribosomal subunitsare not excluded from the Escherichia coli nucleoid. Proc Natl Acad Sci USA 111:11413–11418.

36. Apirion D, Gegenheimer P (1981) Processing of bacterial RNA. FEBS Lett 125:1–9.37. Bakshi S, Bratton BP, Weisshaar JC (2011) Subdiffraction-limit study of Kaede diffu-

sion and spatial distribution in live Escherichia coli. Biophys J 101:2535–2544.38. English BP, et al. (2011) Single-molecule investigations of the stringent response

machinery in living bacterial cells. Proc Natl Acad Sci USA 108:E365–E373.39. Montero Llopis P, Sliusarenko O, Heinritz J, Jacobs-Wagner C (2012) In vivo bio-

chemistry in bacterial cells using FRAP: Insight into the translation cycle. Biophys J103:1848–1859.

40. Mascarenhas J, Weber MHW, Graumann PL (2001) Specific polar localization of ri-bosomes in Bacillus subtilis depends on active transcription. EMBO Rep 2:685–689.

41. Berry J, Weber SC, Vaidya N, Haataja M, Brangwynne CP (2015) RNA transcriptionmodulates phase transition-driven nuclear body assembly. Proc Natl Acad Sci USA 112:E5237–E5245.

42. Poindexter JS (1964) Biological properties and classification of the Caulobacter group.Bacteriol Rev 28:231–295.

43. Ely B (1991) Genetics of Caulobacter crescentus. Methods Enzymol 204:372–384.44. Tsai JW, Alley MRK (2001) Proteolysis of the Caulobacter McpA chemoreceptor is cell

cycle regulated by a ClpX-dependent pathway. J Bacteriol 183:5001–5007.45. Gahlmann A, et al. (2013) Quantitative multicolor subdiffraction imaging of bacterial

protein ultrastructures in three dimensions. Nano Lett 13:987–993.46. Lew MD, von Diezmann ARS, Moerner WE (2013) Easy-DHPSF open-source software

for three-dimensional localization of single molecules with precision beyond theoptical diffraction limit. Protoc Exch, 026.

47. von Diezmann A, Lee MY, Lew MD, Moerner WE (2015) Correcting field-dependentaberrations with nanoscale accuracy in three-dimensional single-molecule localiza-tion microscopy. Optica 2:985–993.

48. Deich J, Judd EM, McAdams HH, Moerner WE (2004) Visualization of the movementof single histidine kinase molecules in live Caulobacter cells. Proc Natl Acad Sci USA101:15921–15926.

49. Schütz GJ, Schindler H, Schmidt T (1997) Single-molecule microscopy on modelmembranes reveals anomalous diffusion. Biophys J 73:1073–1080.

50. Savin T, Doyle PS (2005) Static and dynamic errors in particle tracking microrheology.Biophys J 88:623–638.

51. Michalet X (2010) Mean square displacement analysis of single-particle trajectorieswith localization error: Brownian motion in an isotropic medium. Phys Rev E StatNonlin Soft Matter Phys 82:041914.

52. Besag J (1977) Comments on Ripley’s paper. J R Stat Soc B 39:193–195.53. Kiskowski MA, Hancock JF, Kenworthy AK (2009) On the use of Ripley’s K-function

and its derivatives to analyze domain size. Biophys J 97:1095–1103.54. Ehrlich M, et al. (2004) Endocytosis by random initiation and stabilization of clathrin-

coated pits. Cell 118:591–605.55. Ester M, Kriegel H-P, Jörg S, Xu X (1996) A density-based algorithm for discovering

clusters in large spatial databases with noise. KDD-96 Proceedings (AAAI, Palo Alto,CA), pp 226–231.

Bayas et al. PNAS | vol. 115 | no. 16 | E3721

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

PNASPL

US

Dow

nloa

ded

by g

uest

on

Apr

il 5,

202

0