Fluorescence microscopy studies have shown that RNase E is localized to the membrane in E. coli^10,11,16, but the percentage of molecules bound to the membrane has not been quantitatively examined in live cells. To analyze RNase E localization and dynamics at the single-molecule level, we fused RNase E with a photo-convertible fluorescent protein, mEos3.2²⁵ and imaged individual RNase E molecules over time in two dimensions (Fig. 1A). The location of fluorescent molecules was identified in each frame and linked to form tracks using the open-source software u-track²⁶ (Fig. 1A). The subcellular locations were also calculated relative to the cell boundaries identified from bright-field images using another open-source image analysis package Oufti²⁷ (Fig 1B). To combine data from many cells, molecular positions along the short and long axes of a cell were normalized to the cell width and cell length for xNorm and yNorm, respectively (Fig. 1C). Based on the yNorm, molecules in the cylindrical part of cells were selected, and their xNorm values were used to obtain an xNorm histogram. Hereinafter, we will focus on the xNorm histogram to compare the membrane enrichment of proteins.

The xNorm histogram of RNase E shows two peaks corresponding to each side edge of the membrane (Fig. 1D and Fig. S1), very similar to the xNorm histogram of LacY, obtained by imaging LacY-mEos3.2 using the same method. LacY is a membrane channel for lactose and is composed of 12 transmembrane segments²⁸. It is expected to be inserted into the inner membrane during translation^29-31, such that all imaged LacY is expected to be localized in the inner membrane³².

We further quantified the percentage of molecules bound to the membrane (membrane-binding percentage or MB%) from the xNorm histogram. To do this, we developed a mathematical model based on a 2D projection of molecules randomly distributed on the surface of or inside a cylinder. The model includes imaging effects that affect the shape of xNorm histograms: the localization error, limited focal depth of the quasi-TIRF illumination we used, and the location of the inner membrane relative to the cell boundary (Fig. S2). The model fitting was performed using the Markov-Chain Monte Carlo algorithm (MCMC). For validation, we applied the model to xNorm histograms of LacY and LacZ for complete membrane binding or complete cytoplasmic localization, respectively. The MB% of LacY was 99% with a 95% confidence interval of [96%, 100%], and LacZ showed MB% of 3.4% [0.2%, 7.3%] (Fig. 1E-F, Fig. S3A). Both MB% agree with the expectation for membrane and cytoplasmic proteins. For RNase E, we found MB% of 93% [91%, 96%], suggesting that most of the RNase E molecules are localized in the inner membrane (Fig. 1E).

The membrane localization of RNase E likely limits its interaction with mRNAs in the cytoplasm. We measured how fast RNase E moves in the membrane by analyzing the trajectories of individual RNase E-mEos3.2 imaged at 21.7 ms acquisition interval (Fig. 1A). We calculated the diffusion coefficient D by fitting the mean-squared displacement (MSD) of trajectories with MSD = 4Dτ + b, where τ is lag time and b is a combination of dynamic and static localization error³³ (Supplementary Information). We obtained D of 0.0184 ± 0.0002 μm²/s (mean ± SEM) for RNase E (Fig. 2A). As a comparison, RNase E’s D was larger than the low limit of D our microscope can measure from stationary, surface-immobilized mEos3.2, i.e. D = 0.0020 ± 0.0001 μm²/s (Fig. S3A, Supplementary Discussion). Also, RNase E’s D was comparable to that of its RNA substrates, i.e. mRNAs bound with ribosomes, estimated to be D = 0.015 μm²/s based on the diffusion of ribosomal protein L1 (Fig. S4C). We investigated the origin of RNase E’s slow mobility by testing the effect of (i) mRNA substrates, (ii) RNA degradosome components, and (iii) membrane attachment on the diffusion of RNase E.

To test the effect of mRNA substrates on the diffusion of RNase E, we treated cells with rifampicin, which blocks transcription initiation thus depleting cellular mRNAs. In rifampicin-treated cells, D of RNase E was 0.0270 ± 0.0003 μm²/s, a 1.47-times increase from untreated cells (Fig. 2B). We note that LacY also exhibited a statistically significant increase in D in rifampicin-treated cells with a 1.25-times increase from untreated cells (Fig. 2C). Although this increase is relatively small, it was unexpected because LacY is not an RNA-binding protein. The increase in D of LacY in rifampicin-treated cells likely results from the depletion of mRNAs near the membrane (e.g., the mRNAs undergoing transertion), which could otherwise hinder the diffusion of membrane proteins^34,35.

The fact that RNase E exhibits less than a two-fold increase in D upon rifampicin treatment is surprising. In the case of the ribosome, the diffusion of its large subunit L1 protein was 7 times faster in rifampicin-treated cells (Fig. 2D), consistent with previous reports^36,37. This big change can be explained by the fact that ribosomes form polysomes, such that the effective mass of L1 protein in non-treated cells would be 2 or more times larger than that in rifampicin-treated cells (where it remains as a free subunit). In the case of RNase E, it forms the RNA degradosome complex, whose mass can be from 450 kDa⁶ to 2.3 MDa depending on the partial or full occupancy of the canonically known component proteins (RhlB, PNPase, enolase)^2,38-41 (Supplementary Discussion). In either case, the RNase E complex is smaller than the 70S ribosome of ∼2.5 MDa⁴². If RNase E interacts with an mRNA with n ribosomes, the total mass of the RNA degradosome complex would increase by (n + 1) fold or more. Thus, a substantial increase in D of RNase E is expected upon mRNA depletion.

To explain the marginal increase in RNase E’s diffusion upon rifampicin treatment, we considered two possibilities; it is possible that only a small percentage of RNase E interacts with mRNAs and/or RNase E interacts with mRNA very briefly, unlike ribosomes spending an order of 10-100 s in a polysome state for translation elongation⁴³. When we increased the polysome pool of the cellular mRNA by treating cells with a translation elongation inhibitor chloramphenicol⁴⁴ or by overexpressing lacZ mRNA from a high-copy plasmid, D of RNase E remained unchanged from untreated cells (Fig. S4D). This result rules out the possibility that only a small percentage of RNase E interacts with mRNAs and weighs in favor of the scenario that RNase E interacts with mRNAs briefly.

Next, we investigated the effect of membrane association on the diffusion of RNase E by comparing D of the WT RNase E with a cytoplasmic RNase E mutant lacking the MTS segment. The WT RNase E and RNase E ΔMTS have a similar molecular mass since MTS is only 15 residues of 1000 residues in a monomer RNase E (Fig. 2E and Fig. S5). Therefore, the difference in D should be attributed to their subcellular localizations (Fig. 2F; membrane vs cytoplasm) instead of the mass. We found that the cytoplasmic RNase E diffuses ∼5.5 times faster than the membrane-bound, WT RNase E (Fig. 2G). A similar increase (∼5.3 times) was observed when we examined the CTD truncation mutants, membrane-bound RNase E (1-592) and cytoplasmic RNase E (1-529), which have similar mass but different localizations (Fig. 2F-G).

The CTD region is where the RNA degradosome proteins (enolase, RhlB, and PNPase) bind to create the large RNA degradosome complex (Fig. 2E). To test how these proteins affect the diffusion of RNase E, we measured the diffusion of RNase E in strains where rhlB or pnp was deleted. D increased 1.08 and 1.04 times for ΔrhlB and Δpnp, respectively. The increase was very minute (especially for Δpnp), considering that the mass of the RNA degradosome is expected to decrease by 15% and 40% for rhlB and pnp deletions, respectively (Fig. 2H-I). This finding possibly suggests complex stoichiometry in the RNA degradosome (see Discussions). Despite this ambiguity, a significant 4.04 times increase in D was observed in RNase E ΔCTD, or RNase E (1-592), where the CTD of RNase E was deleted to eliminate interactions with all RNA degradosome components while keeping the membrane localization (Fig. 2H). In this mutant, the mass of the RNase E complex is expected to decrease to 396 kDa (Fig. 2I). Altogether, these results suggest that the slow diffusion of RNase E is largely affected by the membrane localization and the formation of the massive RNA degradosome and less by interaction with mRNA substrates, unlike other RNA-interacting proteins, such as ribosomes.

Unlike RNase E in E. coli, RNase Y, a functional homolog of RNase E in B. subtilis, is localized to the membrane via a transmembrane domain. Any differences between a peripheral (like MTS of E. coli’s RNase E) and a transmembrane motif on membrane localization and mobility of RNase E, remain unclear. To address this question, we created RNase E mutants that contain a transmembrane domain in place of the MTS.

Before creating the mutants, we characterized individual short membrane-binding motifs by fusing them with mEos3.2. LacY in its native form has 12 transmembrane segments that are spatially arranged into two groups of 6 transmembrane segments²⁸. We were able to express the first two transmembrane segments (LacY(1-73) or LacY2), the first six transmembrane segments (LacY(1-192) or LacY6), and the original LacY (LacY12) fused to mEos3.2 from the IPTG-inducible promoter on the chromosome. We then imaged their membrane localization and diffusivity (Fig. 3A). The MTS segment and LacY-derived transmembrane segments showed a strong membrane enrichment (Fig. 3B). In terms of diffusion, LacY2 and LacY6 diffused faster than the MTS segment (Fig. 3C). This is remarkable because their molecular masses (and thus size) are expected to be larger than that of MTS (Fig. 3D). The Saffman-Delbruck diffusion model states that the diffusion coefficient of membrane proteins decreases logarithmically as the radius of the embedded part increases given the same membrane environment⁴⁵. Our data suggest that the two types of membrane-binding motifs (MTS and LacY2) experience different membrane environments, possibly due to the way they interact with the lipids.

To further explore protein-membrane interactions, we conducted all-atom molecular dynamics simulations of MTS and LacY2 interacting with the E. coli membrane using the NAMD software⁴⁶ (Fig. 3E; see Supplementary Information). In the simulations, protein motion was calculated for 1 μs. Although the D values were larger than the experimental values (possibly due to the absence of mEos3.2 in the model), MTS again diffused slower than LacY2 (Fig. 3F and Fig. S6A). By calculating membrane-protein interaction energies, we found that MTS-membrane interaction is more stable than LacY2-membrane interaction (Fig. S6B). These results suggest that the diffusion of MTS, located on the surface of the membrane, is slower than that of LacY2 due to stronger interaction with lipid head groups.

Since LacY2 and LacY6 showed a strong membrane enrichment similar to LacY12 (Fig. 3B), we replaced MTS in RNase E with LacY2, LacY6, and LacY12 in the presence or absence of CTD (Fig. 4A-B and Fig. S5). All the chimeric RNase E mutants were expressed from the native locus on the chromosome as the only copy of RNase E and had mEos3.2 fused at the C terminus for imaging. The strains did not show noticeable differences in growth rate compared to the WT strain (Table S3), suggesting that the RNase E mutants were functionally active.

xNorm histograms of ΔCTD mutants indicated membrane localization similar to LacY (Fig. 4C). However, mutants containing CTD showed noticeable cytoplasmic subpopulations when LacY2 and LacY6 were used in place of MTS (Fig. 4D). The mathematical model fitting of xNorm histograms suggested that MB% of RNase E-LacY2-CTD and RNase E-LacY6-CTD to be 69% [66%, 73%] and 86% [84%, 90%], respectively (Fig. 4F). We note that the non-perfect membrane localization was observed only in the mutants containing CTD and the same protein without CTD showed MB% of 100% (Fig. 4E-F, Fig. S7A-B). This observation suggests that CTD contributes to the unstable membrane binding of RNase E mutants. Such a difference between the presence and absence of CTD was not observed in the chimera based on LacY12 (Fig. 4E-F), possibly due to a stable membrane insertion of LacY12.

Membrane lipid composition is affected by the central metabolism⁴⁷ and hence, may be different when cells are grown in a different carbon source. To test the possibility that MB% of RNase E changes in different media (due to changes in the membrane property), we imaged RNase E from cells grown in M9 succinate (instead of M9 glycerol with casamino acids and thiamine, CAAT; see Table S3 for growth rate difference). We found that MB% of RNase E-LacY2-CTD reduced even lower to 43% [41%, 46%] (Fig. S7C).

This large reduction contrasted with the small MB% change observed in WT RNase E (from 93% to 92%) and LacY2 (from 99.6% to 100%) when cells were grown in M9 succinate, suggesting that RNase E-LacY2-CTD may exhibit a broad MB% range across different growth conditions (Fig. S7C-D).

Next, we examined whether the diffusion coefficients of chimeric RNase E mutants differed based on the type of membrane motifs they had. For example, MTS diffused slower than LacY2, despite being smaller in size (Fig. 3C-D). From this, we expected the chimeric RNase E with LacY2 to diffuse faster than the RNase E with MTS. Indeed, in the absence of CTD, we found that LacY2-based RNase E diffuses 1.33 times faster than MTS-based RNase E (Fig. 4G). In the presence of CTD, LacY2 and LacY6-based RNase E diffused 3.33 and 1.14 times faster than the MTS-based RNase E counterpart (i.e. the WT RNase E). In this case (with CTD), the large increase in diffusion is also affected by the presence of cytoplasmic population (∼31% for LacY2 and ∼14% for LacY6; Fig. 4H) which diffuses faster than the membrane-bound population (possibly about 5 times faster, based on Fig. 2G). Taken together, our data suggest that transmembrane motifs can facilitate the diffusion of RNase E and allow for dissociation from the membrane in the presence of CTD.

To check the functional consequence of cytoplasmic localization of RNase E, we measured lacZ mRNA degradation in various RNase E mutants presented in this study. Recently, we developed an assay measuring co-transcriptional and post-transcriptional mRNA degradation rates of lacZ by inducing its transcription for only 75 s to capture the degradation of nascent mRNA¹⁵. We found that the rate of co-transcriptional degradation (k_d1) of lacZ mRNA is about 10 times slower than that of post-transcriptional degradation (k_d2) in WT cells. We found that k_d1 increases about 3 fold in cells expressing RNase E ΔMTS, suggesting that cytoplasmic RNase E can freely diffuse and degrade nascent mRNAs in the cytoplasm, in contrast to the membrane-bound RNase E¹⁵. This result indicates that the chimeric RNase E that showed cytoplasmic subpopulation (Fig. 4F) may exhibit larger k_d1 than other RNase E variants.

We repeated this assay in a strain expressing the WT RNase E fused with mEos3.2. Note that this strain is different from the one we used for imaging because a monocistronic lacZ gene is needed to aid data interpretation¹⁵. The relative abundances of 5’ lacZ mRNA (Z5) before 3’ lacZ mRNA (Z3) level increases (between ∼100 and 210 s) were constant, confirming negligible co-transcriptional degradation of lacZ mRNA¹⁵ (Fig. 5A). However, when we examined lacZ mRNA degradation in cells expressing RNase E-LacY2-CTD (69 MB%), the Z5 levels at the same time points exhibited a downward trend (Fig. 5B). Its k_d1 was larger than any other RNase E mutants and close to what was observed in RNase E ΔMTS¹⁵ (Fig. 5C). RNase E-LacY6-CTD also showed some cytoplasmic population (86 MB%) but did not exhibit a significant increase in k_d1 (Fig. 5C). The result from RNase E-LacY2-CTD suggests that the cytoplasmic subpopulation of RNase E is functional and may affect the degradation pattern of mRNAs globally.

In terms of the post-transcriptional mRNA degradation rate (k_d2), we tested if the slow diffusion of RNase E in the membrane affects k_d2. In the presence of CTD, k_d2 did not vary much among different membrane motifs (MTS and LacY; Fig. 5D). For example, even though the LacY12 motif slows down the diffusion of RNase E, k_d2 was similar to that of WT RNase E (Fig. 5D). This might be due to differences in RNase E copy numbers. Specifically, the LacY12 version of RNase E might be more abundant than the WT version because RNase E expression level is affected by its ability to regulate its own transcript levels⁴⁸. The higher expression level might compensate for the slow diffusion to give a similar k_d2. However, when we examined the over-expression of WT RNase E, we observed that k_d1 and k_d2 remained unaffected (Fig. S8A). These results suggest that k_d2 is not limited by the copy number and diffusion of membrane-bound RNase E in the presence of CTD.

In the absence of CTD, k_d2 was, in general, lower than those measured from RNase E variants including CTD (Fig. 5D), indicating the importance of CTD in RNase E’s catalytic activity. Similar results have been reported earlier for RNase E based on MTS^15,22-24. Among different membrane motifs for RNase E ΔCTD, we noticed that the LacY2 version, which diffuses faster than the MTS version (Fig. 4G), does not necessarily yield higher k_d2 (Fig. 5D). Plus, LacY6 and LacY12 versions showed even lower k_d2 than MTS-based RNase E ΔCTD. Overall, the slow k_d2 coming from the absence of CTD could not be rescued by faster diffusion of RNase E, but it can be worsened by the large and slow membrane motifs. The presence of CTD appears to buffer the effect of membrane motifs on the catalytic activity of RNase E (k_d2).