On the Spatial Organization of mRNA, Plasmids, and Ribosomes in a Bacterial Host Overexpressing Membrane Proteins

By using fluorescence imaging, we provide a time-resolved single-cell view on coupled defects in transcription, translation, and growth during expression of heterologous membrane proteins in Lactococcus lactis. Transcripts encoding poorly produced membrane proteins accumulate in mRNA-dense bodies at the cell poles, whereas transcripts of a well-expressed homologous membrane protein show membrane-proximal localization in a translation-dependent fashion. The presence of the aberrant polar mRNA foci correlates with cessation of cell division, which is restored once these bodies are cleared. In addition, activation of the heat-shock response and a loss of nucleoid-occluded ribosomes are observed. We show that the presence of a native-like N-terminal domain is key to SRP-dependent membrane localization and successful production of membrane proteins. The work presented gives new insights and detailed understanding of aberrant membrane protein biogenesis, which can be used for strategies to optimize membrane protein production.


Introduction
Integral membrane proteins are key elements in essential cellular processes such as signal transmission and membrane transport. Elucidation of their molecular structure is important since their malfunctioning lies at the heart of a great number of diseases and because they constitute important drug targets [1]. Cell-based expression systems are widely employed to produce convenient amounts of recombinant membrane protein for downstream biochemical characterization. In this respect, bacterial species are considered valuable production hosts, next to yeast and insect cells [2][3][4]. A significant bottleneck in the drug discovery pipeline, however, is that overexpression of membrane proteins in bacterial hosts such as Escherichia coli and Lactococcus lactis for structural and drug-screening studies remains a significant challenge. Therefore, ample research has aimed to elucidate the physiological responses of these bacteria to overexpression of membrane proteins by means of, for instance, transcriptomic and proteomic analyses in order to identify cellular constraints hampering production [5][6][7]. Although the 'omics' data give a good representation of the average state of the cells in a culture, they do not record problems in membrane protein production at the level of individual cells. Impediments might occur anywhere along the process, from transcription to translation to the insertion of correctly folded proteins in the cytoplasmic membrane. Functional synthesis of most polytopic membrane proteins is dependent on co-translational translocation realized by the signal recognition particle (SRP) pathway, which is responsible for the timely delivery of elongating, membrane protein-producing ribosomes to the membrane [8]. With the rise of techniques to peek inside single cells, the spatiotemporal background of processes that underlie problematic synthesis can now be identified and examined in great detail.
Recently, it was shown in E. coli that mRNA species encoding SRP-dependent proteins (mostly integral membrane proteins) display translation-dependent membrane localization, including polycistronic mRNAs containing ORFs for both soluble and integral membrane proteins [9][10][11]. All other mRNAs, including those of the SecB-dependent periplasmic and outer membrane proteins, were maintained throughout the cytoplasm [9]. The SRP-dependent pathway is conserved among all kingdoms of life, but the subset of molecular components deviate between species [12]. The notion that mRNAs of SRP-dependent proteins have a defined spatial organization in bacteria is extremely important in this context. Since codon choice and tRNA-availability differ between cells of prokaryotic or eukaryotic origin, the signal that is responsible for SRP interaction may be unavailable when trying to express a nonoptimized gene in an evolutionarily distinct host. It is therefore tempting to speculate that mRNAs, and the heterologous membrane proteins they encode, might not be properly targeted to the membrane due to a lack of the correct bacterial ZIP-code for balanced SRP-dependent membrane delivery.
To address the impact and effect of mRNA targeting on heterologous production of membrane proteins, we investigated the spatial organization of membrane protein-encoding mRNAs in L. lactis at the single cell level. In contrast to mRNAs of well-expressed membrane proteins that localize at the membrane in agreement with previous studies, our results indicate that poorly expressed membrane proteins experience obstacles during translation, which eventually leads to accumulation of mRNA transcripts at the cell poles. These immobile mRNA clusters are spatially distinct from the location of protein aggregation seeds and from the site where the mRNAs are transcribed. They are also formed when translation is obstructed. Once their disassembly is initiated, a synchronized restoration of translation and growth takes place. Our findings add an empirical understanding of why heterologous membrane proteins are poorly expressed in bacterial hosts, and provide a time-resolved single-cell view on coupled defects in transcription, translation, and growth during aberrant synthesis.

Results
Transcripts of poorly expressing membrane proteins accumulate in immobile polar foci Differential localization of bacterial mRNAs of SRP-dependent proteins has thus far only been shown to exist in E. coli [9][10][11]. Therefore, we first tested, by widefield fluorescence microscopy, whether mRNAs specifying either integral membrane proteins or cytoplasmic proteins adopt different subcellular sites in the Gram-positive bacterium L. lactis. L. lactis bcaP mRNA was chosen as a model encoding a well-expressed endogenous amino acid transporter [7,13]. L. lactis codY mRNA, coding for a central regulator of nitrogen metabolism [13], and gfp mRNA were utilized as representatives specifying cytoplasmic proteins. Fluorescence in situ hybridization (FISH) was adapted in order to be able to image transcripts with the same probe (ms2) by appending the transcripts with repetitive MS2 binding sites (12bs) (For validation of the method see S1 Text) [14,15]. The MS2 binding sites were chosen because these also allow for live cell imaging of the transcripts, as discussed below. Since we were specifically interested in the fate of various membrane-protein-mRNAs upon their overproduction, the nisin-induction system for overexpression (NICE) was used to produce recombinant mRNA and cognate proteins [16]. The successful production of STREPII-tagged BcaP and CodY proteins was confirmed by immunoblotting (S1A Fig). After 1 hr of induction with 5 ng/ml nisin, starting at mid-exponential phase, codY 12bs and gfp 12bs mRNAs were mainly detected in the cytoplasm outside the area occupied by the chromosome (Fig 1A and S1B Fig), while the bcaP 12bs transcripts appeared as fluorescent patches along the inside of the cytoplasmic membrane ( Fig 1B). We realize that pinpointing the location of overexpressed transcripts does not reflect the native situation. However, the production of recombinant membrane proteins traditionally uses multi-copy number expression vectors. Hence, we wanted to study the temporal and subcellular factors of membrane protein biogenesis in a malfunctioning scenario to understand (over) expression bottlenecks.
Next, we employed FISH to examine the localization of mRNAs encoding two poorly expressed heterologous membrane proteins: A sucrose transporter from Solanum tuberosum (SUT1) and the human γ-secretase component, Presenilin 1 missing exon 9 (PS1Δ9). Both proteins are only produced to levels less than 10% of that of BcaP [7,17]. In contrast to bcaP 12bs , SUT1 12bs and PS1Δ9 12bs transcripts appeared as single bright fluorescent foci in one or both poles in the majority of cells (Fig 1C and 1D). We ruled out that the polar mRNA foci were methodological artifacts by visualizing untagged PS1Δ9 and bcaP mRNA with gene-specific FISH probes (S1 Text). In order to get a population-wide view of the subcellular distribution of the overexpressed transcripts, >600 randomly chosen cells expressing either bcaP 12bs , SUT1 12bs , bcaP 12bs , or codY 12bs were further examined. Fluorescent spots inside the cells, as well as spot coordinates relative to the cell contour, were obtained using a Gaussian fitting algorithm in the ImageJ ISBatch plug-in [18]. The peak coordinates were normalized, and all spots were jointly projected in one half of a model cell (dimensions: 800×900 nm, based on the average length and width of L. lactis cells grown in GCDM Ã ; S1C Fig). Location maps of subcellular fluorescence and intensity profiles thereof enabled further discriminating mRNA encoding CodY, BcaP, PS1Δ9, and SUT1 (Fig 1A-1E). bcaP 12bs mRNA is localized along the Left panels: Localization of codY 12bs , bcaP 12bs , PS1Δ9 12bs or SUT1 12bs transcripts in L. lactis NZ9000 cells. Center panels: Corresponding phase contrast images. Scale bar = 2 μm. Colored panels: Location maps constructed from fluorescent TAMRA spots observed in 1185, 877, 765, and 661 cells, respectively, highlighting the preferential localization of each overexpressed mRNA in one half of a model cell. Thick cytoplasmic membrane, codY 12bs mRNA molecules tend to accumulate at the periphery of the chromosome, while SUT1 12bs and PS1Δ9 12bs mRNA foci are mostly confined to the cell poles. Approximately half of the imaged cells expressing PS1Δ9 or SUT1 contained densely packed, polar mRNA clusters ( Fig 1F) marked by spatially confined, yet bright fluorescence (S1D Fig). This type of mRNA-dense spots was far less pronounced or absent in cells expressing bcaP 12bs or codY 12bs , respectively (S1D Fig).
In a previous study, no clear differences in the population-averaged levels of transcripts were observed when producing different membrane proteins in L. lactis [17]. FISH analysis revealed that the population averages of bcaP 12bs and PS1Δ9 12bs transcript levels were roughly similar, but that the distribution of mRNA levels per cell differed quite considerably between the two populations ( Fig 1G). After induction, an increase in bcaP 12bs transcript levels was evident when compared to non-induced cells, but this was less so in cells expressing PS1Δ9 12bs mRNA ( Fig 1G). Instead, 50% of the latter cells contained very little, or no PS1Δ9 12bs mRNA. In cells with detectable levels of PS1Δ9 12bs mRNA, 90% had polar mRNA clusters (Fig 1G). A small fraction of these cells contained extremely high levels of PS1Δ9 12bs mRNA; cells with high levels of overexpressed mRNA were absent in the bcaP 12bs -expressing population. In the remaining~10% of cells, the PS1Δ9 12bs mRNA localized in a manner similar to that of overexpressed bcaP 12bs transcripts. Higher intracellular levels of PS1Δ9 12bs mRNA correlated with a greater disposition for a cell to carry polar mRNA clusters ( Fig 1H). This suggests that there is an expression window for this particular transcript above which cluster formation prevails (mRNA level >10 AU).
Live cell imaging reveals spatiotemporal aberration of heterologousmembrane-protein mRNA and a decline in transcription FISH only provides a static view on mRNA localization. We implemented the MS2 system to label mRNAs in living cells to examine mRNA dynamics (For validation of the method see S1 Text) [11,19,20]. In this approach, transcripts extended with MS2 binding sites are specifically recognized by MS2 phage coat proteins fused to GFP. To detect the MS2 binding site-extended transcripts, L. lactis LG010 was created in which the MS2-GFP protein (MG4: optimized for use in L. lactis; see S1 Text) can be expressed from the chromosomal pseudo10 locus [21] using the inducible nisA promoter. Importantly, GFP in this study refers to a monomeric superfolder GFP optimized for L. lactis ([22]; see Materials and Methods). The spatial distribution of the different MS2-GFP-tagged mRNAs was similar to that observed with FISH (S2 Fig). Timelapse movies revealed that many bcaP 12bs mRNA foci were relatively slow-moving clusters that seemed to be associated with the membrane (S1 Movie). codY 12bs mRNAs diffused as highly dynamic moieties; slow-moving foci similar to those present in bcaP 12bs -expressing cells were transparent lines: Cell boundaries including the portion occupied by cell wall and membrane as approximated using BcaP-GFP expressing cells (See S1C Fig). Thin transparent lines: Boundaries of chromosomal areas as approximated using DAPI staining in living cells (See S1C Fig). Scale bars depict the relative density of each mRNA species. (E) Intensity profiles drawn along the center of the X-axis of the location maps ((A-D), colored panels) of cells expressing bcaP 12bs (solid black line), codY 12bs (dotted black line), PS1Δ9 12bs (solid red line) or SUT1 12bs (dotted red line). (F) The percentage of cells with bcaP 12bs , PS1Δ9 12bs , or SUT1 12bs polar mRNA clusters. Error bars: Standard errors. (G) The population-wide distribution of bcaP 12bs or PS1Δ9 12bs mRNA content obtained from single-cell measurements depicted in box plots and histograms (lower panels). Red bars (right panel): The fraction of cells with polar PS1Δ9 12b mRNA clusters. Blue lines: Gamma distributions fitted to expression data. Fluor = fluorescence. (H) The percentage of cells containing polar bcaP 12bs (black) or PS1Δ9 12bs (red) mRNA clusters plotted as a function of intracellular TAMRA levels (corresponding to bcaP 12bs or PS1Δ9 12bs mRNA levels). absent (S2 Movie). PS1Δ9 12bs and SUT1 12bs mRNA formed static and long-lived clusters (>1 hr) at the cell pole (S3 and S4 Movies).
Growth of, and mRNA localization in PS1Δ9 12bs -expressing cells were followed over time using MS2-GFP (MG4). One-third of the cells did not grow and polar foci remained visible for up to 4 hrs (S5 Movie and Fig 2A and 2B). In another population tertile, polar foci redistributed into multiple smaller foci synchronously with a switch from prolonged (~3 hrs) growth arrest to cell expansion with disturbed division (S6 Movie and Fig 2A and 2B). The last fraction contained less intense fluorescent foci, an indication that P nisA -induced expression had been relatively low in these cells, that disappeared within 1 hr, after which growth continued as normal (S7 Movie and Fig 2A and 2B). Hence, only mild expression at the initiation of the time-lapse experiment led to microcolony formation after removal of polar mRNA clusters (Fig 2C). We identified a clear sequence of post-transfer PS1Δ9 12bs mRNA patterns (S5 and S6 Movies and Fig 2D). First, mRNA formed crescent patches at the poles. Then, a redistribution to spots with Gaussian distributions occurred. These spots eventually fell apart into multiple smaller foci and ultimately disappeared. Under identical conditions, bcaP 12bs -expressing cells did not cease to grow, and P nisA -driven expression continued for 60-90 min after transfer from liquid medium to microscopy slides lacking the inducer nisin (S8 Movie and Fig 2E).
Using MS2-GFP, we consistently observed that there are at least two deviant subpopulations of PS1Δ9 12bs -expressing cells: Non-growing cells with mRNA clusters and dividing cells without any visible PS1Δ9 12bs mRNA. Presumably, the latter cells grew because they were either not (yet) induced or already had terminated P nisA -controlled transcription. P nisA -driven co-expression of MS2-GFP with PS1Δ9, but not with BcaP, decreased without affecting the population-wide growth ( Fig 2F). Hence, either transcription from at least the nisA promoter or translation in general is abrogated in L. lactis cells producing recalcitrant membrane proteins. In any case, both scenarios create a negative feedback that contributes to reduced production yields.

Polar mRNA clusters are spatially unrelated to protein aggregation seeds
To examine if polar mRNA clusters correspond to sites of protein aggregation, BcaP and PS1Δ9 were fused to GFP. Even though GFP fused to the extra-cytoplasmic PS1Δ9 C-terminus likely interferes with its membrane insertion, information can still be gained on completion of translation, protein targeting, and protein-mRNA co-localization. The cells were induced to express BcaP-GFP or PS1Δ9-GFP for 1 hr, after which the level and subcellular localization of each of the GFP-tagged proteins and their respective mRNAs were examined using FISH. Cells with polar bcaP-gfp 12bs mRNA clusters (<1% of the population) contained the highest amount of bcaP-gfp 12bs transcripts, and held relatively high levels of BcaP-GFP ( Fig 3A). The production window for bcaP 12bs , defined as the range of mRNA levels that do not lead to mRNA clusters, was significantly wider than that of PS1Δ9 12bs mRNA ( Fig 3B). PS1Δ9-GFP was clearly detected albeit at levels at least an order of magnitude lower than BcaP-GFP ( Fig  3A-3C). No reports on inclusion bodies in L. lactis exist, but misfolded membrane proteins are known to co-fractionate with the membrane proteome [23]. In agreement, we occasionally observed membrane-proximal fluorescent protein aggregates, even in BcaP-GFP-expressing cells (S3A Fig). The protein aggregates did not typically co-occur with polar clusters of bcaPgfp 12bs mRNA. When present in the same cell, they did not co-localize (S3A Fig). No signs of PS1Δ9-GFP membrane insertion were detected. Rather, PS1Δ9-GFP was present as aggregation seeds [24] at the membrane ( Fig 3C and S3B Fig). Of the larger PS1Δ9-GFP aggregates, the majority (90%) did not co-localize with, but was often found in the proximity of polar mRNA foci (Fig 3C).
We attempted to indirectly detect (aggregates of) truncated or incompletely translated fusion proteins by localizing the molecular chaperone DnaK. DnaK associates with protein aggregates and inclusion bodies in E. coli [24,25]. Levels of DnaK-GFP were significantly elevated 15 min after initiation of PS1Δ9 12bs expression ( Fig 3D). The cellular distribution of LG010 cells was induced following a standard regime after which the cells were transferred to a microscopy slide carrying a thin layer of 1.5% agarose dissolved in GCDM* lacking nisin, for time-lapse fluorescence microscopy analysis. Shown are the portions of 100 tracked cells that contain polar PS1Δ9 12bs mRNA clusters that do not divide after transfer (ND), that grow very poorly and adopt swollen and deviating cell shapes (-), that grow at a moderate rate (+/-) and that grow at a normal rate (+). (B) Histogram displaying the time it takes for 100 individual cells to remove MS2-GFPtagged PS1Δ9 12bs mRNA clusters from their poles, as visualized with time-lapse microscopy. (C) Box plots displaying the initial MS2-GFP expression levels, which is directly correlated to the mRNA level, in PS1Δ9 12bsexpressing cells that either resumed growth (+) or remained in a non-growing state (-). A Students t-test was performed to test for significance (p<0.005). (D) Schematic representation of the post-transfer sequence of patterns adopted by MS2-GFP-tagged PS1Δ9 12bs mRNA (colored red). Plasmids predominantly co-localize with the nucleoid, not with mRNAdense polar clusters We entertained the possibility that pNZ8048 plasmids are predominantly present in the cell pole(s) like other multi-copy number plasmids, which might seed the formation of foci for mRNAs of poorly expressed membrane proteins due to high local transcription [26][27][28]. Plasmid pNZ8048 was visualized using a Gram-positive ParB/parS system (M. Kjos, J. Siebring, J. W. Veening, personal communication) and DNA FISH. A parS site was introduced in pNZ8048, pLG-BcaP and pLG-PS1Δ9, and L. lactis LG045a was constructed, a strain that constitutively produces low levels of ParB-GFP (the parS binding protein fused to GFP). The L. lactis NZ9000 genome does not contain a ParA homolog [29], allowing visualization of the subcellular location of parS-labeled pNZ8048 derivatives (Fig 4 and S4 Fig).  Unlike some high-copy plasmids in other bacterial species [26][27][28], pNZ8048 primarily populates the area occupied by the chromosome in L. lactis (S4 Fig). Covisualization of the plasmids and the full-length transcripts they specify shows that they are spatially separated (Fig 4A, 4B and 4D). Since our plasmid imaging approach does not have the time-resolution required to monitor rapid spatiotemporal changes, we cannot rule out the possibility that transertion, the coupled transcription, translation and translocation of membrane proteins, does not occur here [30]. Adding nisin to induce expression of bcaP 12bs or PS1Δ9 12bs transcripts did not severely affect the location of pNZ8048-derived plasmids, but location maps of plasmids show that induction of PS1Δ9 caused the area occupied by plasmids to slightly shrink compared to control cells (Fig 4C and 4D).
Because polar clustering of PS1Δ9 12bs transcripts cannot be explained by the spatial distribution of pNZ8048, we reasoned that downstream processes, including targeting of mRNA for degradation or complications in synthesis of PS1Δ9 might be involved. The shrinkage of the area occupied by the plasmids (and nucleoid) suggests that transertion of PS1Δ9 is inhibited under these conditions.

The relationship between mRNA cluster formation, translation activity, and mRNA degradation
What is the effect of translation on the localization and degradation of the studied mRNAs? Are polar mRNA clusters still available for translation or could they correspond to bacterial analogs of dedicated mRNA degradation hubs that are found in eukaryotic cells during stress? In E. coli, treatment with translation-inhibiting kasugamycin results in a loss of membranelocalized mRNAs [9]. We observed the same with overexpressed MS2-GFP-visualized bcaP 12bs mRNAs when L. lactis cells were treated with translation-inhibiting antibiotics erythromycin (Ery) and chloramphenicol (Cm) (Fig 5A). The bulky MS2-labeled mRNAs did not enter the nucleoid region due to chromosome condensation induced by Ery/Cm treatment [31][32][33] (S5A Fig). In contrast to bcaP 12bs , PS1Δ9 12bs mRNA remained visible in the poles after antibiotics treatment. These structures apparently do not readily release transcripts upon inhibition of translation. To further investigate the effect of translation on mRNA localization, we mutated the aaggagg ribosomal binding site (RBS) on gfp 12bs transcripts by changing it to aattcgg (S5B Fig), and employed the same strategy to disrupt the RBS on bcaP 12bs and PS1Δ9 12bs . Using FISH, we observed that a high percentage of the cells expressing RBS-less bcaP 12bs or PS1Δ9 12bs mRNA contained transcript clusters (Fig 5B and 5C). This suggests that the punctuate foci correspond to mRNAs that are blocked from translation. Closer examination (Fig 5D) of the population-wide localization of the RBS-less mRNAs revealed that they cluster more closely to the nucleoid (and thus also to the expression plasmids), while the RBS-containing PS1Δ9 12bs mRNA clusters are predominantly present at the cell poles (see also Fig 1C).
The localization of ribosomes in L. lactis is similar to that in other bacteria: The major fraction of the S2-eYFP protein-and therefore ribosomes-localizes outside the nucleoid [32,34,35] (Fig 5E and S5C Fig). Time-lapse movies similar to those described above reveal that non-growing cells expressing PS1Δ9 undergo a reorganization of S2-eYFP localization from typically nucleoid-occluded to highly homogeneously distributed (Fig 5E). Since 70S ribosomes are not able to freely diffuse into the nucleoid [32,34,35], a homogenous distribution indicates a loss of 70S ribosomes, and thus of translation. Non-induced cells transferred to agarose slides lacking nutrients showed a similar phenotype, suggesting that the loss of 70S ribosomes overlaps with entry into a stationary-like growth phase (S5C Fig). Surprisingly, S2-eYFP is excluded from the poles in PS1Δ9-expressing cells (Fig 5F), which indirectly suggests that polar PS1Δ9 12bs mRNA clusters are not accessible for ribosomes. The finding that Ery/Cm treatment does not lead to a release of mRNA from polar clusters strengthens this supposition. Under similar conditions, in about the same time required for cells to get rid of polar MS2-labeled PS1Δ9 12bs clusters (Fig 2B), S2-eYFP localization and growth were restored to wild-type ( Fig 5G).
Determining the lifetimes of mRNAs might help understanding the cause of the polar mRNA accumulation observed here. We treated cells expressing different transcripts and cognate proteins with rifampicin to stop translation, and used FISH to follow the level of each mRNA at several time points after rifampicin addition. We used FISH instead of Northern hybridization to obtain a population-wide quantifiable view on mRNA degradation. Thus, we were able to distinguish subpopulations of cells with significantly altered mRNA degradation parameters (see below). Considerable amounts of PS1Δ9 12bs (t ½ = 10.9 min) and SUT1 12bs (t ½ = 19.7) transcripts were still present after a 32-min treatment with rifampicin, in contrast to codY 12bs (t ½ = 7.1 min) and bcaP 12bs (t ½ = 6.2 min) (Fig 5H and S5D Fig). The degradation rate of codY 12bs and bcaP 12bs were in agreement with those reported previously (S5E Fig) [36], which shows that mRNA degradation rates in L. lactis are generally lower than those measured in other fast-growing bacterial species such as E. coli and B. subtilis [37][38][39]. Despite considerable deviations between experimental replicates, the half-lives of PS1Δ9 12bs and SUT1 12bs transcripts did not differ substantially from the estimated median half-life of all mRNAs in L. lactis (t ½ = 10.9 min; S5E Fig). Separation of cells into populations that did or did not contain polar mRNA clusters, and recalculation of their half-lives, revealed that, in cells without polar mRNA Localization during Membrane Protein Production mRNA clusters, PS1Δ9 12bs and SUT1 12bs mRNA is degraded at much higher rates than similar mRNA trapped inside polar mRNA clusters (PS1Δ9 12bs : t ½ = 4.6 min and t ½ (cluster) = 16.1 min; SUT1 12bs : t ½ = 6.2 min and t ½ (cluster) = 22.4 min; S5F Fig). Degradation of overexpressed PS1Δ9 12bs or bcaP 12bs mRNA lacking an RBS takes place at a higher rate (t ½ =~2.5 and~1.7 min, respectively) than the original transcripts (t ½ =~10.9 and~6.2 min, respectively; S5G Fig). The fact that the RBS-less PS1Δ9 12bs and bcaP 12bs mRNAs yielded nearly identical degradation curves indicates that the mRNA half-life of at least these two mRNA species is mainly influenced by ribosome occupancy in L. lactis. These results also imply that the observed slow degradation of PS1Δ9 mRNA does not originate from intrinsic signals in the alien mRNA sequence, but depends on binding of ribosomes and/or downstream processes. Importantly, the positioning of the major ribonuclease in Gram-positive bacteria, membrane-tethered RNase Y [40], remains unaffected by PS1Δ9 12bs or bcaP 12bs overexpression (Fig 5I).
The presence of an endogenous N-terminal domain is key to SRPdependent membrane localization Could a defect in proper targeting of mRNA-ribosome complexes to the membrane play a role the mislocalization of the recombinant mRNAs? It is well known that SRP-dependent targeting is important for co-translational translocation, and that recognition of a membrane protein emerging from the ribosomes by SRP is dependent on signals located on the N-terminal domain of SRP-dependent proteins [9,41]. Optimization strategies by which slow codons are incorporated at specific sites might stimulate timely SRP recognition [42], and fusion tags can improve the guiding or initial insertion of heterologous RNC-ribosome-mRNA complexes to the membrane [43][44][45]. To test if the N-terminal domain of PS1Δ9 (PS1Δ9 N ) is less adequate for membrane-targeting in L. lactis than that of BcaP (BcaP N ), we swapped both regions, including the first two transmembrane domains, and re-examined their mRNA localization (Fig 6A and 6B). Indeed, a high number of bcaP N -PS1Δ9 12bs transcripts was now localized at the membrane instead of in polar clusters (Fig 6B and 6C). Vice versa, PS1Δ9 N -bcaP 12bs mRNAs were mainly found in polar clusters (Fig 6B and 6C). Overexpression of the corresponding proteins was monitored using quantitative blotting, in which the membrane protein fractions of L. lactis cells were collected and directly spotted onto nitrocellulose membranes. In agreement, the replacement of the N-terminal domain of PS1Δ9 with that of BcaP dramatically elevated production yields, while the presence of the N-terminal PS1Δ9 domain lowered BcaP production (Fig 6E and 6F).
The finding that polar mRNA accumulation of PS1Δ9 N -bcaP 12bs had decreased while the corresponding protein production had improved prompted us to design a general optimization strategy for improved production of recalcitrant membrane proteins. The same N-terminal domain of BcaP was now fused to the N-terminus of PS1Δ9 via a flexible linker that included a TEV protease cleavage site (Fig 6D). We coined this new fusion partner BcaP leading segment (BLS). To test the effectiveness of BLS on PS1Δ9 production, the various L. lactis NZ9000 strains were grown in two types of media: GCDM Ã and GCDM Ã supplemented with extra peptides through the addition of casitone. The latter has been shown to positively affect the production of membrane proteins [46]. Induction was performed according to the standard regime, after which the quantity of original and chimeric membrane proteins in the membrane fraction was examined as described above. In both types of media an increased signal of BLS-PS1Δ9 over PS1Δ9 was observed, indicating that production of PS1Δ9 had improved (Fig 6E and 6F). Moreover, mRNA of BLS-PS1Δ9 was more often observed at the membrane than clustered in the cell poles (Fig 6G). Although follow-up studies are required to address the quality of overexpressed BLS-PS1Δ9, we conclude that the BLS-tag has great potential as a fusion partner to elevate the levels of recalcitrant (heterologous) membrane proteins in L. lactis by guiding their insertion in the cytoplasmic membrane.
Membrane protein mRNAs of the AT-rich L. lactis contain an increased U-density over other mRNAs Thus far, we were able to show a subset of cellular consequences of polar mRNA clusters, and excluded the localization of plasmids, protein aggregates, the chaperone DnaK, and the main mRNA Localization during Membrane Protein Production constituent of the RNA degradosome, RNase Y, as a direct cause for cluster formation. What then, besides ribosome exclusion, could drive the predisposition of mRNAs of recalcitrant membrane proteins at the poles? Codons for hydrophobic amino acid residues that constitute the transmembrane domains of membrane proteins have a strong uracil (U) bias [47]. This is more pronounced in bacteria such as E. coli than in higher, eukaryotic cells through differences in codon choice. The increased U-density of membrane protein mRNA was suggested to function as a post-transcriptional zip code for membrane targeting. Since L. lactis is an AT-rich organism, the general U-density of transcripts is intrinsically higher than that of E. coli. If the U-density of transcripts is of evolutionary importance, we would expect to see an adapted, higher U-bias in the preferred codons specifying hydrophobic amino acid residues in L. lactis. This is indeed the case for hydrophobic amino acid residues in L. lactis, whereas this bias is less pronounced in E. coli (Fig 7A). Therefore, despite the AT-rich genomic sequence of L. lactis, the U-density of membrane protein-encoding mRNA is significantly higher than that of mRNAs of cytosolic proteins, but also than that of membrane protein-encoding mRNAs of E. coli (Fig 7B). This suggests that in AT-rich organisms the deployment of mostly relatively Urich codons for hydrophobic amino acid residues has been selected for, through which the relative difference between the general U-density of mRNAs encoding membrane or cytoplasmic proteins is maintained.

Discussion
L. lactis has emerged as a suitable host for membrane protein production, although expression of heterologous membrane proteins is still suboptimal. To gain insights into potential obstacles, we examined the fate of transcripts of well and poorly expressing membrane proteins, and the relationship between mRNA localization, transcription, translation, and degradation. In E. coli, transcripts for many integral membrane proteins are present in SRP-dependent cotranslational translocation complexes during the major part of their lifetime [9]. Transcripts encoding the homologous transporter BcaP localized to the membrane in L. lactis while disruption of their translation led to loss of membrane-proximal mRNA foci, suggesting that a similar transertion system operates in this bacterium. Transcripts of poorly expressed membrane proteins, H. sapiens PS1Δ9 and S. tuberosum SUT1, formed distinctive polar clusters in L. lactis. Importantly, neither the expression vector pNZ8048 nor protein aggregates co-localized at the poles. Although high-copy number plasmids have been localized predominantly at the cell poles of bacteria [26][27][28], pNZ8048-derivatives randomly distributed in the cells, with a predisposition in the area where the chromosomal DNA is also situated. The ColE1-derived plasmid pBluescript, with~50 copies per cell, does not adopt any specific subcellular location in E. coli [48]. Both this and our study show that high-copy number plasmids do not necessarily cluster at the poles.
The occurrence of polar mRNA clusters was translation-dependent, positively correlated to the level of heterologous transcripts, and was always accompanied by severe growth defects, loss of nucleoid-occluded ribosomes, and activation of the heat-shock response. Furthermore, PS1Δ9-expressing cells displayed confined plasmid distribution and delayed degradation of polar PS1Δ9 12bs mRNA clusters. The disappearance of polar mRNA foci was a prerequisite for cells to be able to grow again, and was accompanied by a loss of nisin-driven transcription but also by relocalization and increase of ribosomal protein S2-eYFP. The sudden release of many PS1Δ9 12bs transcripts from polar foci, however, severely impacted growth, often leading to defects in cell division. All of these observations are an indication that cells with polar mRNA clusters adopt an altered cellular state. Previous ensemble studies of L. lactis overproducing membrane proteins suggested that the stringent response might be involved [17]. High alarmone ((p)ppGpp) levels in E. coli and B. subtilis lead to a shift from transcription of rRNA genes to transcription of genes involved in nutrient generation, preparing the cell for survival in a nutrient-poor environment [49,50]. In Staphylococcus aureus, (p)ppGpp inhibits ribosome assembly by binding GTPases involved in the maturation of ribosomal subunits [51]. It is therefore tempting to speculate that the loss of active ribosomes in L. lactis is triggered by the stringent response due to nutrient limitation or expression of the heterologous protein PS1Δ9.
The presence of polar mRNA clusters correlated to cells with a strong induction, as evidenced by both the level of overexpressed mRNAs and co-expressed MS2-GFP. As with the E. coli Walker strains, in which T7 RNA polymerase-driven transcription is reduced compared to the parent strain [52,53], the synthesis of less mRNA might improve the production of recalcitrant membrane proteins in L. lactis. Lowering the inducer concentration would be a straightforward solution albeit that nisin A-driven expression triggers a heterogeneous response at doses below 0.5 ng ml -1 [54]. Interestingly, mutations in the sensor histidine kinase NisK led to a 2-to 8-fold improvement in membrane protein production due to an increase in leaky expression in L. lactis DML1-3 strains [23]. We can now assume that this background transcription causes a steady state production of low levels of recombinant mRNA that do not lead to mRNA clustering or proteotoxic stress, prior to induction of gene expression.
The question remains what exactly causes the mRNA cluster formation. Below, three scenarios are discussed.

Availability for ribosomes
Overexpressed mRNAs encoding soluble proteins were mainly present in the nucleoid-free space, whereas RBS-less transcripts accumulated at the boundaries of the chromosomal area instead of the poles. A model on the spatial organization of intracellular mRNA localization in bacteria predicts that a great portion of transcripts reside in the nucleoid-free space and that a cyclic pattern of ribosome flow occurs, in which ribosomes that have finished translation in the nucleoid-free space pick up new transcripts from the nucleoid [55]. Transcripts might thus not readily diffuse away from their site of production without effective binding by ribosomes, resulting in the cluster formation at the nucleoid periphery observed here. Importantly, the adaxial clusters of untranslatable bcaP 12bs or PS1Δ9 12bs transcripts resembled polar mRNA clusters in terms of density, while S2-eYFP was excluded from the poles whenever PS1Δ9 was produced. Abrogated translation of PS1Δ9 12bs transcripts might therefore be the cause of their polar clustering.

Uracil-density
In agreement with a previous study in E. coli [47], the uracil density of membrane proteinencoding mRNA in the AT-rich L. lactis is significantly higher than that of mRNAs of cytosolic proteins. The fact that, through codon bias, this uracil-richness has been preserved in bacteria strengthens the idea that it might play a role in membrane protein biogenesis. Recently, two cold-shock proteins, CspE and CspC, were shown to interact with uracil-rich transcripts and the inner membrane of E. coli [56]. This suggests that these CSPs not only stabilize transcripts during cold shock, but also deliver specific U-rich cargos to the cytoplasmic membrane. Remarkably, gene expression of most CSPs in L. lactis was downregulated upon PS1Δ9 expression but not upon BcaP-expression [7,17]. Because bcaP mRNA (0.36) and other endogenous membrane protein transcripts contain a higher U-density than the heterologous PS1Δ9 (0.29) and SUT1 (0.31) mRNA, it is tempting to hypothesize that the homologous bcaP 12bs mRNA is recognized (better) by L. lactis homologs of E. coli CspE and CspC, such as CspE, CspB, and CspD2. This is supported by the notion that even RBS-less PS1Δ9 12bs transcripts are still more prone to accumulate in dense clusters than RBS-less bcaP 12bs transcripts, the latter of which might be pulled away from the cell center, resulting in fewer mRNA clusters. If CSPs exert similar functions in L. lactis and E. coli, the decline in these membrane-targeting factors observed during PS1Δ9-expression would lead to a loss of membrane-proximal mRNA localization, and possibly to polar predisposition instead.

SRP recognition
When SRP recognition and delivery fails, synthesis of membrane proteins continues in the hydrophilic cytoplasm, leading to protein misfolding and aggregation [12]. A consensus SRPspecific signal peptide on integral membrane proteins has never been identified, but recent data suggest that translational pausing at mRNA elements 30-40 codons downstream of the codons that specify the SRP binding site triggers SRP-binding [41,57]. Neither the gene encoding PS1Δ9 nor that for SUT1 was codon-optimized for L. lactis. Due to differences in codon choice and tRNA-availability, cis-acting signals that promote SRP-binding might be absent when trying to express a non-optimized gene in an evolutionarily distinct host [41,57]. The idea that the timely targeting of ribosomes to the membrane by SRP constitutes a bottleneck during membrane protein overproduction has been explored in E. coli [58,59]. Competition between binding to the ribosome-nascent chain of trigger factor (TF), a folding chaperone for cytoplasmic proteins, and SRP influences efficient targeting of membrane proteins [59]. An increase in correctly inserted membrane proteins was established by decreasing the amount of TF while at the same time elevating the SRP level. Some N-terminal fusion proteins are known to improve membrane protein production, putatively through promoting SRP-dependent targeting and/or membrane insertion of the overexpressed protein [43,[60][61][62]. We show here that these adaptations could also be responsible for guiding ribosomes carrying alien mRNA to the membrane: Interchanging the N-terminal domains of PS1Δ9 and BcaP resulted in a reversal in mRNA localization patterns and improved production of PS1Δ9. We delivered a proof-of-principle by introducing the BcaP N-terminal domain as a new L. lactis-specific fusion partner, BLS, which improves the heterologous production of recalcitrant membrane proteins such as PS1Δ9.

Bacterial strains, media and culture conditions
Bacterial strains and plasmids used in this study are listed in S1 Table. For cloning purposes, Lactococcus lactis NZ9000 was grown as a standing culture at 30˚C in Difco M17 medium (BD, Franklin Lakes, NJ, USA) containing 0.5% (w/v) glucose (GM17). When required, chloramphenicol and/or erythromycin were added at a final concentration of 5 μg ml -1 or 3 μg ml -1 , respectively. For all other analyses, L. lactis was grown in chemically defined medium (CDM; pH 6.8) with 0.5% glucose [63] lacking riboflavin (GCDM Ã ), in which it reaches a growth rate of~0.55 h -1 . When stated, 1.5% (w/v) pancreatic digest of casein (Bacto Casitone; BD Biosciences, Franklin Lakes, NJ, USA) was added to GCDM Ã Chemically defined SA medium with 0.5% (w/v) glucose and 20 μg ml -1 5-fluoro-orotic acid (5-FOA; Sigma-Aldrich, St. Louis, MO, USA) as a sole pyrimidine source was used for the generation of chromosomal knock-ins, as described previously [64]. Escherichia coli DH5α (Life Technologies, Gaithersburg, MD, USA) was used for cloning purposes; it was grown aerobically at 37˚C in LB medium (Formedium, Norfolk, UK) with, when required, erythromycin or ampicillin at a final concentration of 150 or 100 μg ml -1 , respectively. All antibiotics were obtained from Sigma-Aldrich.

Recombinant DNA techniques and oligonucleotides
Standard molecular cloning techniques were performed essentially as described [65]. Chromosomal DNA from L. lactis was isolated using the Wizard Genomic DNA Purification Kit (Promega Life Sciences, Madison, WI, USA). Plasmids and PCR products were isolated and purified using the High Pure Plasmid Isolation and PCR Purification kit (Roche Applied Science, Mannheim, Germany) and the NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel, Düren, Germany) according to the manufacturers' instructions. Enzymes were purchased from Fermentas (Thermo Fisher Scientific Inc, Waltham, MA, USA) and New England Biolabs (Ipswich, MA, USA). PCR reactions were performed with Phusion or DreamTaq polymerase (both from Fermentas) according to the manufacturer's protocol. In case DNA fragments were prepared for uracil-excision DNA-based engineering, PfuX7 polymerase was employed (Morten HH Nørholm, DTU, Denmark) [66]. The obtained PCR fragments were mixed and treated with the USER enzyme mixture (New England Biolabs), yielding 9-12 nt-long overhangs annealing to complementary overhangs. No ligation was required, and USER-treated mixtures were directly used to transform L. lactis or E. coli. Oligonucleotides employed in this study are listed in S2 Table and were purchased from Biolegio BV (Nijmegen, The Netherlands). Competent E. coli cells were transformed using heat-shock, while electrocompetent L. lactis cells were transformed using electroporation with a Bio-Rad Gene Pulser (Bio-Rad Laboratories, Richmond, CA). All nucleotide sequencing was performed at Macrogen Europe (Amsterdam, The Netherlands).

Plasmid and strain construction
Nucleotide sequences of the primers and their role in the construction work described below are presented in S1 Text and S2 Table. Pertinent regions of all plasmids were sequenced to confirm their proper nucleotide sequences. All plasmids and strains that were used or constructed during this study are listed in S1 Table. Protein analysis, western blotting, and immunodetection L. lactis NZ9000 carrying pLG-BcaP or pLG-CodY was inoculated 1:100 in chemically-defined medium containing 0.5% (w/v) glucose (GCDM; (18)) and grown as standing culture at 30˚C to an OD 600 of 0.4, and induced with 5 ng ml -1 of nisin A (Sigma-Aldrich). Cells were harvested 2 hrs post-induction by centrifugation at 9,000×g for 15 min at 4˚C, washed with icecold 100 mM potassium phosphate (KPi) pH 7.0 and resuspended to OD 600 of 1 in ice-cold 100 mM KPi. For lysis, glass beads were added to 1 ml of the cell suspension, and cells were broken by 3×35 sec incubations in a mini-bead beater (Biospec Products, Bartlesville, OK) with 1-min intervals of cooling at 4˚C. Cell debris was removed by centrifugation at 7650×g for 15 min (4˚C). To detect overexpressed StrepII-tagged BcaP, membrane fractions were collected by high-speed centrifugation (267,000×g for 30 min at 4˚C) in an Optima TLX Ultracentrifuge (Beckman Coulter, Brea, CA, USA). For immunodetection of StrepII-tagged BcaP, membranes were washed with 10 mM KPi buffer containing 10% glycerol, resuspended in 2×SDS loading buffer (1M Tris-HCl, pH 6.8, 10% SDS, 20% (v/v) glycerol, 25% (v/v) β-mercaptoethanol, 0.05% bromophenol blue), boiled for 5 min, subjected to SDS-(12.5%) PAGE. For immunodetection of StrepII-tagged CodY, cleared cell lysates were mixed with 2×SDS loading buffer, boiled for 5 min, and separated by SDS-(12.5%) PAGE. Proteins were transferred by semi-dry blotting to a polyvinylidene difluoride membrane (Roche Applied Sciences). For dot-blotting, strains were grown and fractionated as described above. Membrane and cytoplasmic proteins were directly spotted onto a nitrocellulose membrane (0.45 μm, Bio-Rad). Immunodetection of transferred or spotted StrepII-tagged proteins was performed as follows. Membranes were washed with 1×TBS-T (20 mM Tris, 150mM NaCl, and 0.1% Tween 20, pH 7.5) blocked with 5% bovine serum albumin in TBS-T, blocked with 0.001% avidin in 1×TBS-T for 20 min, washed with 1×TBS-T, after which the proteins were detected with Strep-MAB-Classic (IBA, Göttingen, Germany), using Amersham anti-mouse polyclonal antibodies from sheep, conjugated to horseradish peroxidase and ECL detection (both from GE Healthcare Europe GmbH, Eindhoven, The Netherlands) according to manufacturers' instructions.

Preparation of cells for in vivo fluorescence microscopy imaging or RNA FISH
Generally, L. lactis was grown overnight in GM17 with appropriate antibiotics prior to microscopy analysis. Overnight cultures were washed in sterile 1×PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , pH 6.8) and inoculated 1:100 in GCDM Ã , to limit background fluorescence, and antibiotics when required. Cultures were grown to mid-exponential phase (equivalent to OD 600 of~0.4) and induced with nisin A at a final concentration of 5 ng ml -1 . Depending on the experiment, cells were incubated for another 0 to 90 min.
For in vivo fluorescence microscopy imaging, 0.4 μl of a culture was transferred onto a microscopy slide that was prepared as follows: Multitest-slides (MP Biomedical, Santa Ana, CA, USA) were cleaned using 5M KOH and MilliQ, after which they were coated with 1.5% agarose (Sigma-Aldrich) dissolved in fresh GCDM Ã .
To stop translation, both chloramphenicol and erythromycin were added to an induced culture, each to a final concentration of 200 μg ml -1 . Cultures were then incubated for another 30 min, after which cells were examined using fluorescence microscopy (see below).
To determine the degradation rate of overexpressed transcripts, the mRNAs were produced for 1 hr in the presence of 5 ng ml -1 of nisin A. Then rifampicin was added to a final concentration of 200 μg ml -1 to inhibit transcription. Cells were fixed (see below) at different time points after rifampicin addition (0, 1, 2, 4, 8, 16, 32, and 64 min) and prepared essentially as described in the following paragraph. This procedure was repeated at least twice for each mRNA on separate days. The final mRNA half-lives were determined as further described in Data analysis.
For RNA FISH, performed essentially as described [14] and adapted for E. coli cells [15], the cells were fixed by directly adding formaldehyde to the cultures to a final concentration of 3.7%. After incubation for 30 min at RT, the treated cells were washed twice with dyethylpyrocarbonate (DEPC)-treated 1×PBS, resuspended in 1×PBS containing 10 mg ml -1 lysozyme (Sigma-Aldrich), incubated for 30 min at 37˚C, washed twice with 1×PBS and permeabilized in 70% ethanol. Permeabilized cells were washed twice with RNA FISH wash buffer (40% formaline and 2×SSC (150 mM NaCl, 15 mM sodium citrate, pH 7.0) in DEPC-treated MilliQ), resuspended in RNA FISH hybridization buffer (10% (w/v) PEG6000 (Sigma-Aldrich), 2 mM vanadyl ribonucleoside complex (VRC, 200 mM; New England Biolabs), 1 mg ml -1 yeast RNA, 40% formaline, and 2×SSC in DEPC-treated MilliQ) containing 1 mM of each FISH probe and incubated for 16 hrs in a water bath at 30˚C. Prior to imaging, excess probe was removed by two washings with 200 μl RNA FISH wash buffer. Finally, the cells were resuspended in 2×SSC and transferred to a microscopy slide coated with 1.5% agarose in 1×PBS.
For DNA FISH, the cells were treated similar to preparation for RNA FISH, except for the following changes. An RNase (Roche) step (50 μg ml -1 , 1 hr, and 37˚C) was incorporated after permeabilization with ethanol and cells were washed twice with 1×PBS. Prior to 16-hrs probe hybridization step, cells were heated to 78˚C for 10 min in hybridization buffer, after which the TAMRA-labeled ms2 probe was added to a final concentration of 1 mM.
The oligonucleotide probes used are summarized in S3 Table. All probes were purchased from Biolegio BV and were covalently linked at both ends to carboxytetramethylrhodamine (TAMRA) dyes. The ms2 FISH probe was designed to specifically hybridize to each of the 12 MS2 binding sites fused to the 3'-ends of overexpressed transcripts, leading to a theoretical 24-fold amplification of the signal. The three bcaP-FISH or PS1Δ9-FISH probes were designed to anneal next to each other and to target the middle of the protein-coding region of the transcripts.

Fluorescence microscopy
All micrographs were obtained with a DeltaVision Elite inverted epifluorescence microscope (Applied Precision, GE Healthcare, Issaquah, WA, USA) equipped with a stage holder, a climate chamber, a seven-color combined set InsightSSI Solid-State Illumination module, and an sCMOS camera (PCO AG, Kelheim, Germany). A 100× phase-contrast objective (NA 1.4, oilimmersion, DV) was used for image capturing, in combination with SoftWorX 3.6.0 software (Applied Precision) to control the microscope setup and to perform single-time-point or timelapse imaging of cells. The following standard fluorescence filter sets were used for visualization: GFP, excitation at 475/28 nm and emisson at 525/48 nm; TAMRA/TRITC, excitation at 542/27 nm and emission at 597/45 nm; DAPI, excitation at 390/18 nm and emission at 435/48 nm (all using a polychroic beam splitter suitable for DAPI, FITC, TRITC, Cy5); eYFP, excitation at 438/24 nm, emission at 548/48, using a polychroic beam splitter suitable for CFP/eYFP/ mCherry. TAMRA images were captured with 1 sec exposure time, S2-eYFP with 5 sec exposure time, and GFP fusions with 0.5 to 1 sec exposure time. If stated in the figure legends, deconvolution was performed using SoftWorX 3.6.0 software (GE Healthcare Bio-Sciences, Pittsburgh, PA, USA).
Fluorescence measurements using a plate reader L. lactis was grown overnight in GM17 with appropriate antibiotics. Cells were collected by centrifugation, washed in sterile 1×PBS, inoculated 1:100 in GCDM Ã with 5 μg ml -1 chloramphenicol, and divided as quintuplos in wells of a transparent 96-wells microtitre plate. Fluorescence was measured with the following equipment and settings: Infinite F200 plate reader (Tecan Group Ltd., Männedorf, Switzerland) with I-control 1.10.4.0 software (Tecan Group Ltd.), GFP filter set (excitation at 485 nm and emission at 535 nm). GFP signals were collected as top readings with a gain setting of 60. GFP fluorescence was divided by the corresponding OD 600 measurements to correct for cell density effects. The cultures were grown at 30˚C to mid-exponential phase (equivalent to an OD 600 of~0.4) and induced with nisin A at a final concentration of 5 ng ml -1 . L. lactis LG029(pNZ8048) was used as an empty vector control to obtain fold changes in expression of DnaK-GFP.
Spot detection was performed using a Gaussian fit algorithm available in the ImageJ plug-in ISBatch [18], collecting the coordinates of the spots relative to the image file as well as the Gaussian fitting parameters of each spot. Spots corresponding to intracellular signals were obtained using the tracked cell meshes, and those lying outside the meshes were discarded. The remaining spot coordinates were transformed and normalized to intracellular coordinates to fit in the dimension of one half of a coccoid model cell. To construct the model cell, the average L. lactis cell outline and nucleoid size were extracted from image files of live DAPIstained L. lactis NZ9000 cells using cell and nucleoid tracking (S1C Fig). The membrane-width inside the cell outline was estimated using images of cells expressing the fluorescent membrane protein BcaP-GFP. The resulting demi-coccoid form had a dimension of 800×900 nm, representing the average width×0.5(length) of an L. lactis cell grown in GCDM Ã . Remaining spots were conjointly projected into this model cell, and the resulting spot occupancy was plotted as density maps using MATLAB. The Gaussian fitting parameters obtained from the ISBatch plug-in were used to calculate the intensity (I) of fluorescent spots (I = πw 2 h, where w = fullwidth at half-maximum (FWHM); h = height or maximum of Gaussian fit parameter).
To obtain mRNA degradation curves, FISH was employed to obtain total TAMRA signals per cell for each time point after rifampicin addition as described in Preparation of cells for in vivo fluorescence microscopy imaging or RNA FISH. TAMRA signals minus background were extracted using MicrobeTracker. Bootstrapping (5×500 randomly picked cells per population) was performed to examine the accuracy of the sampling distribution. Single exponential decay curves described by N(t) = N 0 e −λt were fitted to the averages of normalized median values obtained from independent experiments using MATLAB. Transcript half-lives were calculated from the decay coefficients (λ) using t1 2 ¼ lnð2Þ=l. The ObjectJ plug-in for ImageJ was employed to track cell lengths and S2-eYFP distribution along the Y-axes of the cells (https://sils.fnwi.uva.nl/bcb/objectj). All box-and-whisker plots (box plots) are represented according to the Tukey method for plotting whiskers.
To the determine the U-density of transcript groups representing cytoplasmic proteins or membrane proteins, the nucleotide sequences of genes encoding all cytoplasmic proteins and membrane proteins were identified and extracted from the genome sequences of L. lactis MG1363 and E. coli K12 using tools available at https://genome2d.molgenrug.nl. The number of thymines was divided by the total number of nucleotides of each gene to obtain the uracil density per gene. The codon usage indexes for both organism were obtained from http:// genomes.urv.es/CAIcal.

S1 Fig. Verification of protein production of 12bs-labeled transcripts and analysis of mRNA distribution inside fluorescent foci. (A)
Analysis of overexpressed StrepII-tagged CodY (30.2 kDa) in cell-free extracts and of StrepII-tagged BcaP (50.6 kDa) in membrane fractions of L. lactis NZ9000 by Western hybridization using anti-StrepII antibodies. BcaP is known to migrate faster than expected on the basis of its molecular mass and typically forms a band at a position where proteins of around 37.5 kDa would migrate [13]. Control (-), emptyvector control. LG010 cells expressing MS2-GFP, with or with Cm/Ery treatment. Treated cells display a partial exclusion of MS2-GFP from the nucleoid region. This is not the case in the non-treated cells. Scale bar is 2 μm. (B) The fluorescence of L. lactis NZ9000 cells overexpressing gfp 12bs transcripts with or without ribosomal binding sites (RBS). (C) Snap shots of time-lapse microscopy following the subcellular S2-eYFP distribution in L. lactis(rpsB::rpsB-eYFP) grown in nutrition-rich or nutrition-limiting conditions. Cells were grown to mid-exponential phase in GCDM Ã , after which samples were transferred to a microscope slide with an agarose patch either dissolved in GCDM Ã or 1×PBS supplemented with 0.5% glucose (GPBS). (D) The relative level of overexpressed codY 12bs (black), bcaP 12bs (red), PS1Δ9 12bs (grey), and SUT1 12bs (blue) transcripts as a function of time after rifampicin addition (at t = 0) was monitored using FISH. The data was fitted to single exponential decay, of which the curves are shown as solid lines. Error bars depict standard deviations obtained from bootstrapped datasets. (E) The dataset of [36] was used to determine the mRNA degradation rates of lactococcal transcripts. The log values of median half-lives, log t1 2 , of a total of 994, 787, and 996 transcripts at growth rate (μ) of 0.11, 0.51, and 0.80 h −1 , respectively, were extracted and plotted as a function of the respective growth rates (open black circles). A best fit to the data (dashed black line) followed from log t1 2 ¼ À 2:3m 2 þ 0:6m þ 2:7. The closed black circle indicates the median log t1 2 at μ = 0.55h −1 , corresponding to a median t1 2 of 10.9 min. A similar strategy was employed to estimate the t1 2 values of codY (open red circle) and bcaP (open blue circle) mRNA at μ = 0.55h −1 , resulting in log t1 2 codY ¼ À 2:0m 2 þ 0:1m þ 2:4 (dotted red line) and log t1 2 bcaP ¼ À 5:2m 2 þ 3:3m þ 2:2 (dotted blue line). The lifetime of codY 12bs mRNA (t 1/2 = 7.1 min) was in good agreement with that previously reported for codY mRNA (t 1/2 = 7.4 min for μ = 0.55 h -1 ). This corresponds to a t1 2 for codY or bcap mRNA of 6.4 min or 11.5 min, respectively, at μ = 0.55h −1 . Breakdown of overexpressed bcaP 12bs mRNA (t1 2 ¼ 6:2 min) was at least not slower than that of endogenous bcaP mRNA. Note that media composition differs substantially between our measurements and those performed by Dressaire  Cells from a nisin-induced culture were transferred to a microscope slide with an agarose patch in growth medium without additional inducer. Scale bar represents 2 μm. (AVI) S7 Movie. Time-lapse movie of L. lactis cells with MS2-GFP-labeled PS1Δ9 12bs mRNA clusters that were removed from the poles, after which cell division was restored. Cells from a nisin-induced culture were transferred to a microscope slide with an agarose patch in growth medium without additional inducer nisin. Scale bar represents 2 μm. (AVI) S8 Movie. Time-lapse movie of L. lactis cells in which MS2-GFP-labeled bcaP 12bs mRNA foci disappeared. Cells from a nisin-induced culture were transferred to a microscope slide with an agarose patch in growth medium without additional inducer. Scale bar represents 2 μm. (AVI) S1