New RoxS sRNA Targets Identified in Bacillus subtilis by Pulsed SILAC

ABSTRACT Non-coding RNAs (sRNA) play a key role in controlling gene expression in bacteria, typically by base-pairing with ribosome binding sites to block translation. The modification of ribosome traffic along the mRNA generally affects its stability. However, a few cases have been described in bacteria where sRNAs can affect translation without a major impact on mRNA stability. To identify new sRNA targets in Bacillus subtilis potentially belonging to this class of mRNAs, we used pulsed-SILAC (stable isotope labeling by amino acids in cell culture) to label newly synthesized proteins after short expression of the RoxS sRNA, the best characterized sRNA in this bacterium. RoxS sRNA was previously shown to interfere with the expression of genes involved in central metabolism, permitting control of the NAD+/NADH ratio in B. subtilis. In this study, we confirmed most of the known targets of RoxS, showing the efficiency of the method. We further expanded the number of mRNA targets encoding enzymes of the TCA cycle and identified new targets. One of these is YcsA, a tartrate dehydrogenase that uses NAD+ as co-factor, in excellent agreement with the proposed role of RoxS in management of NAD+/NADH ratio in Firmicutes. IMPORTANCE Non-coding RNAs (sRNA) play an important role in bacterial adaptation and virulence. The identification of the most complete set of targets for these regulatory RNAs is key to fully identifying the perimeter of its function(s). Most sRNAs modify both the translation (directly) and mRNA stability (indirectly) of their targets. However, sRNAs can also influence the translation efficiency of the target primarily, with little or no impact on mRNA stability. The characterization of these targets is challenging. We describe here the application of the pulsed SILAC method to identify such targets and obtain the most complete list of targets for a defined sRNA.

protein is not generally required for efficient base-pairing and regulation of most sRNAs characterized so far, with relatively few exceptions (2)(3)(4).
In most cases, sRNAs bind at or close to the ribosome binding site (RBS) and inhibit mRNA translation. This often indirectly results in mRNA destabilization because ribosomes no longer protect the transcript from attack by ribonucleases (RNases). Small RNAs can also bind outside of the RBS and directly provoke mRNA degradation, by changing RNA structure or by directly recruiting RNases (5,6). sRNAs can also positively regulate the expression of some targets by modifying the secondary structure of the 59 untranslated region (59-UTR) to release the RBS and increase translation efficiency, or change structure to impair RNase cleavages (reviewed in reference 7).
Very few cases have been described in bacteria where sRNAs act at the translational level without affecting mRNA stability (8). A recent study in Staphylococcus aureus identified two targets of the IsrR RNA, induced during iron starvation, that are primarily affected at the translational level (9). In all such cases studied thus far, the sRNA basepairs with the RBS to inhibit translation.
The favored technique for global study of the effects of sRNAs on translation has been to compare proteomes in wild-type (WT) versus a strain deleted for the sRNA of interest. The main drawback of this approach is that it only provides a picture of the proteome at equilibrium, with numerous potential indirect effects. This issue has been partially circumvented by inducing sRNA expression for short periods of time (5 to 15 min induction), allowing direct or indirect repressive effects on mRNA levels to be detected due to their short half-lives. However, the half-lives of proteins can be much longer than these induction times, making downregulation of translation much more difficult to measure and potentially missed entirely in cases where the indirect effect on mRNA stability is minimal.
In this work, we have combined RNA sequencing (RNAseq) with pulsed SILAC (stable isotope labeling by amino acids in cell culture), first used in eukaryotes to identify miRNA targets (10), with a goal of identifying new sRNA targets in B. subtilis that potentially include some targets with primarily translational effects. To benchmark the approach, we chose to overexpress the RoxS sRNA, one of the best characterized sRNAs in Gram-positive bacteria, for a short period of time. RoxS is conserved in both B. subtilis and S. aureus, and uses three single-stranded C-rich regions (CRR1-3) to regulate the expression of many genes involved in central metabolism. In B. subtilis, its main proposed function is to adjust temporary imbalances in the NAD1 to NADH ratio, by controlling key steps in fermentation pathways and the TCA cycle. Several direct targets of this sRNA have already been identified by the methods described above, allowing us to validate the performance of the pulsed SILAC experiment. In addition to identifying most of the known RoxS targets, we discovered several new mRNAs potentially regulated by this sRNA. Indeed, our data suggests that RoxS downregulates the expression of all but one enzyme of the TCA cycle, a far greater impact than previously appreciated. As anticipated, we also identified some potential targets that were primarily regulated at the translational level, with only minor effects on mRNA stability. The potential repercussions of this type of regulation are discussed.

RESULTS
Pulsed SILAC consists of stable isotopic labeling by amino acids of newly synthetized proteins to allow their detection and quantification by mass spectrometry (MS). We performed experiments in a lysine auxotrophic strain to maximize the efficiency of lysine isotope incorporation into new proteins. The strain lacked the native version of the roxS gene and expressed RoxS under the control of an arabinose-inducible promoter from the amyE locus. The control strain carried the same arabinose promoter inserted at the amyE locus with the RoxS sRNA replaced by a transcriptional terminator to avoid polar effects on downstream genes. To ensure appropriate expression of mRNA targets, both strains were cultivated in controlled medium (MD) containing malate, a carbon source we have previously shown strongly induces RoxS expression (11). All 20 amino acids in their light isotopic form were supplied to the medium, but lysine was limited to an empirically determined concentration that allowed it to be almost completely consumed by mid-exponential phase (OD 600 = 0.6), leading to a growth arrest unless further supplemented with lysine isotopes (Fig. S1). At this point, arabinose and medium-heavy lysine were added to the control strain, and arabinose and heavy lysine were added to the RoxS overexpressing strain, to simultaneously induce RoxS expression and label newly synthesized proteins. After 15-min induction, bacteria were harvested to extract total RNA and soluble proteins. RNA was used for RNAseq analysis and equal amounts of extracted proteins were mixed and subjected to quantitative proteomic analysis. MS-based quantitative data (medium to light, heavy to light, and medium to heavy ratios) determined at the peptide level for each lysine-containing peptide were further used to relatively quantify newly synthesized proteins in both conditions and determine those that were differently expressed upon RoxS induction.
The up-and downregulated genes identified by RNAseq and by MS analysis after 15 min induction of RoxS are presented in Fig. 1 and Tables 1 and 2. Only statistically significant alterations in expression (P-value #0.05) with a fold change ,0.7 and .1.4 for negatively and positively regulated targets respectively were selected. This threshold includes all previously confirmed targets of RoxS in B. subtilis. As developed below, most of the new potential targets identified fit perfectly well with the proposed function of RoxS in the regulation of the NAD1/NADH ratio via the modulation of carbon metabolism (11)(12)(13).
Genes whose expression is downregulated by RoxS. In most cases described in literature, bacterial sRNAs that affect translation also indirectly alter mRNA stability. In agreement with this, we identified 21 potential targets (20 downregulated and one upregulated) that were found in both the RNAseq and the pulsed SILAC data sets (Fig. 1). Among these candidates were numerous previously identified targets of RoxS, such as ppnkB (an inorganic polyphosphate/ATP-NAD kinase), ykhA (a putative acyl-coenzyme A thioesterase), acsA, (an acetyl-CoA synthetase), and the gene encoding the TCA enzyme succinate dehydrogenase, sucCD (9-11). Interestingly, we identified eight additional genes encoding enzymes of the TCA cycle, present on four different mRNAs (citZ-icd-mdh, citB, odhAB, sdhAB). All of these potential targets have at least one predicted base-pairing site for RoxS on the mRNA (Fig. S2). A gene ontology (GO) analysis of the data confirmed the over-representation of the TCA cycle genes (P-value of 1 Â 10 210 ), compared to other cellular functions (Fig. 2) (14). These results are in agreement with our previous data suggesting that RoxS regulates cell metabolism in order to limit NADH production (13). To confirm a representative example of these TCA cycle     Identification of sRNA targets by Pulsed SILAC Microbiology Spectrum targets at the level of mRNA stability, we measured the half-life of the citZ-icd-mdh mRNA in a WT and DroxS mutant strain (Fig. 3A). Two major species of the citZ mRNA were detected by Northern blotting. The larger (;4kb) band of corresponds to the tricistronic citZ-icd-mdh mRNA and the lower band (;1kb) to the mono-cistronic citZ transcript. Two degradation products were also visible (D1 and D2), with D2 disappearing in the DroxS background, suggesting it is stabilized by the presence of RoxS (Fig. 3A). The half-life of citZ-icd-mdh mRNA increased 3-fold in a DroxS strain (3.3 min in a WT strain vs 9.6 min in a DroxS mutant) (Fig. 3A). This result is in good agreement with the effect of RoxS measured by RNAseq (3-fold) ( Table 1), but less than the effect detected by SILAC, where the level of the CitZ protein was reduced 5-fold upon overexpression of RoxS ( Table 2), suggesting that translational control has a greater impact for this mRNA. The glycine decarboxylation pathway was also over-represented (P-value of 1.2 Â 10 26 ) in the GO analysis of the potential target genes identified by pulsed SILAC (Fig. 2). The level of the GcvT protein was reduced 1.5-fold after RoxS induction and the gcvT-PA-PB mRNA was reduced 1.3-fold, just below the threshold. The levels of second protein of the operon, GcvPA, were also reduced 1.3-fold. Several putative basepairing sites for RoxS can be predicted on the gcvT-PA-PB mRNA, including in the SD regions of gcvT (Fig. S2) and gcvPB, suggesting the operon is a bona fide target despite the weak effects. The stability of this mRNA increased 1.8-fold in the absence of RoxS (1.6 min in a WT strain versus 2.9 min in a DroxS strain; Fig. 3B). Thus, this operon belongs to the category of targets where the effect on translation likely has a similarmagnitude indirect effect on mRNA stability, despite its position in the Venn diagram (Fig. 1). Interestingly, the steady-state level of an ;300-nt transcript corresponding to the glycine riboswitch was also increased in the DroxS strain. However, the half-life (;1.3 min) was similar in both strains suggesting that RoxS may have an additional indirect impact on the transcription of this operon.
Among the downregulated targets that were only identified by pulsed SILAC (and thus potentially regulated primarily at the translational level) were lutB and lutC of the lutABC operon, ycsA and pycA, all with prospective binding sites for RoxS in their SD regions ( Fig. S3; Fig. 4), and four genes (amhX, clpQ, rpmE, and yjiC), all with potential binding sites for RoxS in their ORFs (Fig. S4). In the following paragraphs, we will FIG 2 Gene ontology (GO) analysis of genes whose expression is altered at mRNA and protein levels after 15 min of RoxS overexpression. Analysis was performed using the Comparative GO web application (14). (A) Table presenting the different biological functions over-represented in the data set of potential RoxS targets with a P-value #0.05 (B) Histogram representing the number of genes belonging to each biological function that is overrepresented in the downregulated (left graph) and upregulated (right graph) data sets. describe the verification of the weak effect of RoxS on the half-lives of two of these potential targets, the lutABC and ycsA mRNAs, and additional experiments performed on ycsA to validate the effect on translation and the predicted RoxS binding site.
The lutABC operon encodes lactate utilization enzymes, and interestingly is also regulated in B. subtilis by FsrA, an iron-induced sRNA with similar single-stranded Crich motifs to RoxS involved in the binding of G-rich Shine-Dalgarno (SD) sequences (15). While FsrA was proposed to base-pair with the SD sequence of lutA (15), IntaRNA (16) predicts that both RoxS and FsrA could potentially bind close to the ribosome binding site of each of the three genes of this operon (Fig. S3). The levels of the LutA, LutB, and LutC proteins were downregulated 1.7-fold, 1.8-fold, and 1.5-  (Fig. 3C). In agreement with the RNAseq data, its half-life was only weakly affected by the deletion of RoxS (4.9 min in a WT strain versus 6.4 min a DroxS strain; Fig. 3B), suggesting the effect of RoxS occurs primarily at the translational level. The regulation of the ycsA gene, encoding a putative NAD-dependent tartrate dehydrogenase fits well with the proposed function of RoxS in managing NAD1/NADH ratios. The YcsA protein was reduced 1.6-fold upon RoxS induction in the pulsed SILAC experiment in MD medium. To obtain additional evidence for the translational regulation of ycsA by RoxS, we fused the ycsA ORF in frame to GFP and determined its rate of fluorescence production in the presence or absence of RoxS (DroxS). We also introduced these GFP fusions in DroxS strains complemented with either a WT (RoxS1) or mutant version of RoxS in which three C-residues of CRR3 were mutated to three G's (RoxS CRR3/3G ). Indeed, RoxS is predicted to interact with the ribosome binding site of the ycsA mRNA primarily by base-pairing between four C residues of CRR3 and four Gresidues of the ycsA SD sequence (Fig. 4A). This experiment was done in LB because the ycsA mRNA was shown to be much more expressed in these conditions in a previous study by Nicolas et al. (17) (Fig. 4; Fig. S5). The rate of fluorescence accumulation during exponential phase increased approximately 1.8 and in a strain deleted for RoxS (DroxS) compared to WT, and 1.7-fold in the strain complemented with the RoxS CRR3/3G mutant compared to RoxS1, confirming the translational regulation of ycsA expression by RoxS. We also determined the effect of the DroxS deletion on the stability of the ycsA mRNA in LB. Similar to the translational effect, the ycsA mRNA half-life increased 2.2-fold in a DroxS strain compared to the WT strain (4.9 min in the WT strain versus 10.7 min in the DroxS mutant; Fig. 3D). Off note, the level of ycsA mRNA was lower at t = 0 in the DroxS strain suggesting an additional transcriptional effect of RoxS on this mRNA.
Although the RoxS CRR3/3G mutant sRNA was unable to repress the ycsA-GFP fusion as anticipated, the steady state levels of this mutant RNA were lower than the WT RoxS, suggesting that it is less stable (Fig. S5C). It was therefore difficult to conclude whether the lack of regulation in the RoxS CRR3/3G mutant was due to impaired base-pairing or to the lower levels of the sRNA. To directly confirm the interaction between the two RNAs, we performed an electromobility shift assay (EMSA) using an in vitro-transcribed fragment of ycsA containing the SD sequence and either RoxS CRR3/3G or a similar mutant in CRR1 (RoxS CRR1/3G ) as a control (Fig. 4C). The ycsA RNA formed a shifted species with WT RoxS and with the RoxS CRR1/3G mutant, but failed form a complex with the RoxS CRR3/3G mutant, suggesting that the two RNAs indeed interact directly. The introduction of the compensatory mutation in ycsA mRNA, where three G's of the SD sequence were replaced by three C's, abolished the interaction with WT RoxS and RoxS CRR1/3G , but partially restored the interaction with the RoxS CRR3/3G mutant, confirming the interaction site prediction.
Genes whose expression is upregulated by RoxS. Only one potential positively regulated RoxS target, ytsJ, was identified by both RNAseq and pulsed SILAC (Fig. 1). YtsJ is thought to be the primary malate dehydrogenase of four such enzymes identified in B. subtilis (18). In a previous study we showed that RoxS expression is induced by addition of malate to the growth medium and that RoxS positively regulates the expression of a second malate transporter (YflS) by protecting its mRNA from degradation by RNase J1 (11). Of note, the YflS protein was not detected by pulsed SILAC in this study, due to its localization in the membrane (see next section).
Previous studies have suggested that ytsJ is expressed from three different promoters, located in front of the upstream gene dnaE (PdnaE), in the middle of the dnaE ORF (P2) (17) or immediately upstream of ytsJ (P1) (Fig. 5A) (18). We confirmed the 59 ends of the P2-ytsJ and P1-ytsJ transcripts within the dnaE ORF by primer extension (Fig.  S6) and they fit well with the promoter sequences previously proposed (17,18). Interestingly, a binding site for RoxS was predicted by IntaRNA within the dnaE ORF, approximately 190 nt downstream of the P2 promoter (Fig. 5A). A second RoxS binding site was also predicted within the ytsJ coding sequence (CDS), with a lower energy score.
We first measured the half-lives of the various ytsJ transcripts in a WT versus a DroxS strain grown in the same medium as the RNAseq experiment (MD 1 malate) by Northern blotting using a riboprobe against the first half of the ytsJ coding sequence. Surprisingly, none of the three expected ytsJ primary transcripts, or an additional species of unknown origin, showed a measurable decrease in half-life (Fig. S7), despite the 1.4-fold increase in mRNA levels found by RNAseq (Table 1). Given that our previous studies have shown that RoxS regulation can depend on growth medium (11,12), one possible explanation was greater RoxS expression under the control of the arabinosedependent promoter in the RNAseq experiment compared to that naturally occurring in MD 1 malate. (9,10). Indeed, in rich medium (2xTY 1 malate), we were able to confirm the stabilizing effect of RoxS on ytsJ transcripts (Fig. 5B). The half-life of the predominant P2-ytsJ (;2.7 kb) transcript was reduced 2-fold in the DroxS strain compared to WT (5.3 versus ,2.5 min). A similar effect was observed with the transcript corresponding in size to P1-ytsJ (5.6 versus ,2.5 min), while no impact was seen on the bicistronic dnaE-ytsJ species (Fig. 5B). We next tested the effect of RoxS on the translation efficiency of ytsJ mRNA, by constructing two GFP fusions, expressed under the control of a Pspac promoter to remove complications of transcriptional effects from the native promoters. These fusions have a transcriptional start close to the P1 (Pspac(1)-ytsJ-gfp) or P2 promoter (Pspac(2)-ytsJgfp). The experiments were done in LB 1 malate to reduce background fluorescence from the 2xYT medium. Consistent with the pulsed SILAC data where the YtsJ protein expression was induced 1.5-fold by RoxS, the rate of fluorescence accumulation of the P2-YtsJ-GFP fusion was 3.5-fold lower in the DroxS strain compared to WT, confirming that RoxS increases YtsJ expression ( Fig. 5C; Fig. S5B). However, we were unable to detect an impact of RoxS on the P1-YtsJ-GFP fusion (Fig. 5D). There results confirm that RoxS positively regulates expression of Pspac(2)-ytsJ-gfp transcript. To confirm the predicted interaction with CRR3, we measured the rate of fluorescence accumulation in the DroxS strain complement with either WT (RoxS1) or mutant RoxS (RoxS CRR3/3G ). As anticipated, the rate of fluorescence accumulation of the P2-YtsJ-GFP in the strain expressing RoxS CRR3/3G was 1.7-fold lower than the WT strain and 2.5-fold lower than the complemented (RoxS1) strain. However, as mentioned previously, the steady state levels of the RoxS CRR3/3G variant are lower than WT RoxS, making it difficult to definitively confirm the interaction with CRR3 in this way (Fig. S5C). To conclusively determine whether RoxS can bind the P2-ytsJ transcript, we tested the ability of RoxS to bind to RNA fragments containing the predicted binding sites by gel shift assay in vitro (Fig. 5E). As expected, WT RoxS was able to bind to an RNA fragment within the dnaE ORF containing the binding site downstream of P2 and this interaction was abolished with the RoxS CRR3/3G mutant. A compensatory mutation in dnaE (dnaE m ) restored basepairing with the RoxS CRR3/3G mutant, confirming their ability to interact via the CRR3 region. We were unable to detect binding between RoxS and a fragment within the ytsJ ORF, containing the second potential binding site downstream of P1 (Fig. 5F), consistent with the lack of regulation seen with the P1-ytsJ-gfp fusion. Together, these results suggest that RoxS directly regulates ytsJ expression from the P2 transcript, but not that from P1. The explanation for the effect of RoxS on the half-life of the P1-ytsJ transcript is unclear.
The position of the RoxS binding site far upstream (approximately 1 kb) of the ytsJ start codon on the P2 transcript suggested that the primary effect might be on mRNA stability. To determine whether RoxS protects the P2-ytsJ mRNA against RNase J1 as   Fig. 6A and B). These experiments were done in rich medium (2XTY) where the P2-ytsJ transcript is best detectable. The steady-state levels and stability of P2-ytsJ (and indeed all ytsJ transcripts) were no longer sensitive to RoxS in the Drny mutant 13.9 min in both strains), suggesting that RoxS cannot stabilize ytsJ without prior action of RNase Y ( Fig. 6A and B). In the DrnjA mutant, up to seven different degradation intermediates were detected compared to the WT strain (D1 to D7; Fig. 6A). Two of these were impacted by the deletion of the roxS gene; D2 disappeared and was replaced by an intensified D3. This suggests that RoxS protects D2 from degradation by RNase J1. To clarify how RoxS might protect D2 from RNase J1, we performed a primer extension assay in the different RNase mutant strains to identify the 59 ends of the degradation products (Fig. S6). Because the signal detected from the native ytsJ gene was very low, we overexpressed a part of the 59 UTR of the P2-ytsJ transcript to increase the intensity of the bands (Fig. S6). The 59 end corresponding to D2 was highly intensified in a DrnjA mutant strain and was absent in strains lacking either RoxS or RNase Y, suggesting that D2 is generated by RNase Y cleavage and protected from RNase J1 degradation by RoxS. Interestingly, this 59 end is only a few nucleotides (3 to 4) upstream of the RoxS binding site, a compatible location to block the processive activity of RNase J1 (19), similarly to what we previously observed for the yflS mRNA (11).
The primer extension profile in the region of the second potential RoxS binding site in the ytsJ ORF was insensitive to the presence of RoxS in the different RNase mutant strains, reinforcing our conclusion that RoxS does not bind in the CDS of the ytsJ mRNA. Targets identified only by RNAseq. A number of potential targets were identified by RNAseq, but not by pulsed SILAC. Most of these mRNAs are predicted to encode membrane proteins that would not have been present in the soluble protein fraction Identification of sRNA targets by Pulsed SILAC Microbiology Spectrum analyzed in the pulsed SILAC experiment, such as the yfls mRNA, encoding a malate transporter, which has been previously shown to be protected against the 59-39 exoribonuclease J1 by RoxS pairing in B. subtilis (11). Secreted proteins are a second class of proteins absent of the soluble fraction analyzed by pulsed-SILAC. This is the case for the VPR protein encoded by the vpr mRNA identified by RNAseq. IntaRNA predicted RoxS binding site(s) in all cases but one (the cidAB operon), with energy scores of ,-5 kcal/mol (see Materials and Methods). Although the other interactions remain to be validated, it is interesting that almost 50% of the downregulated targets identified by RNAseq have a classical predicted binding site in their SD region. One upregulated target, yfiY mRNA, which encodes a siderophore-binding subunit of an ABC transporter, has a predicted binding site for RoxS at its extreme 59-end. The position of this binding site is similar to that described for yflS mRNA, where RoxS protects from the degradation by the 59-39 exoribonuclease J1 (11). These base-pairing predictions reinforce our belief that several of these membrane protein-encoding mRNAs are also potentially direct targets of RoxS (Fig. 1).

DISCUSSION
This work shows that pulsed SILAC is an efficient alternative to identify direct effects of sRNAs on the translation of target genes, in particular downregulatory effects that would normally be masked by protein stability. It is less cumbersome than ribosomeprofiling experiments and has the major advantage of only focusing on newly synthesized proteins, without the background noise of translation occurring under equilibrium conditions. Using this technique, we not only identified most of the known targets of RoxS, but we also identified new potential targets for this sRNA in B. subtilis. We observed a clear enrichment of downregulated genes involved in the TCA cycle, in agreement with the proposed function of RoxS to limit flux through this pathway to restrict NADH production in the presence of favorable carbon sources such as glucose or malate. We also identified the ytsJ mRNA as a new positively regulated target of RoxS. This gene encodes one of the four malate hydrogenases in B. subtilis, the only one that uses NADP1 instead of NAD1 as cofactor. The preferential upregulation of an enzyme using NADP1 instead of NAD1 as cofactor fits well with the need for the cell to control NADH production. The enrichment of downregulated genes involved in metabolic pathways directly connected to the TCA cycle such as l-glutamate biosynthesis, citrate metabolism, glycine decarboxylation, and gluconeogenesis also is also consistent with the proposed metabolic role of RoxS (Fig. 2). This study also allowed us to identify ycsA mRNA as a direct target of RoxS. This target is repressed approximately 2-fold by RoxS. The fact that ycsA mRNA encodes a putative tartrate dehydrogenase that uses NAD1 as cofactor is also in good agreement with the proposed function of RoxS.
In contrast to the ycsA mRNA, which has a classical binding site for RoxS in its SD region, several mRNAs that seem to be regulated primarily at their translational level by RoxS, have at least one putative binding site predicted in their coding sequence (CDS) instead of the SD region, which is rather unusual (Fig. S8). A few examples have been described in bacteria where the sRNA binds to the CDS well downstream (.5 codons) of the AUG. Some of these sRNAs impact mRNA stability with a likely indirect affect their translation (5,20). In other cases, it has been shown that sRNA binding within the CDS can prevent formation of a secondary structure that stimulates translation initiation, lowering the translation efficiency of the target mRNA (21). The presence of sRNA binding sites deep within the CDS raise the question of how base-pairing occurs on an actively translated mRNA. Although we were able to confirm that RoxS has no impact on the mRNA stabilities of two mRNAs from this class of potential targets with putative RoxS binding sites within their ORFs: yisK (involved in the control of cell division) and yjiC mRNA (encoding a potential macrolide glycosyltransferase), we were unable to see an impact of RoxS on translation of these genes using a GFP-yjiC fusion or an antibody raised against YisK protein (data not shown). Furthermore, although the bicistronic dnaE-ytsJ and P2-ytsJ transcripts contain the RoxS binding site localized in dnaE, only the stability of the P2-ytsJ mRNA was affected by RoxS (Fig. 5), suggesting that translation of dnaE competes with RoxS regulation on the longer mRNA. We therefore cannot rule out that some mRNAs of this class may be off-targets due to the overexpression of RoxS or that they are transitory effects that quickly return to equilibrium following a pulse of RoxS expression. That said, some targets, such as yukB, probably merit further investigation because nine potential binding sites for RoxS are predicted within its CDS. One could imagine that transitory slow-downs in translation promoted by sRNA binding throughout the CDS could impact the folding of sub-domains of proteins without a major impact on the protein yield.
Another potential advantage of this type of regulation could be its reversibility. Indeed, removal of the sRNA from its nondegraded targets could allow new rounds of translation on negatively regulated targets or reducing the translation of positively regulated mRNAs but in keeping a basal level of expression. This impact on translation without affecting the pool of the mRNA could be a way to limit the need of new rounds of transcription and quickly adapt to changing environment. A parallel can be drawn to the situation in eukaryotes, where translationally repressed RNAs are stored in P-bodies to be recycled when needed (22).

MATERIALS AND METHODS
Strains, plasmids, and oligonucleotides. Strains and plasmids used in this study are presented Table S1 and oligonucleotides in Table S2.
The lysA::erm mutation was transferred by transformation of SSB1002 with chromosomal DNA from the strain BKE23380 (23).
Plasmid pDG1662-PxsA (pl801) is a plasmid containing the arabinose inducible promoter of xsa gene, amplified using oligonucleotides CC2250 and CC2251. The PCR fragment was cloned between the BamHI and HindIII sites of pDG1662.
Plasmid pDG1662-PxsA-RoxS (pl805) was made by amplification of RoxS by PCR using oligos CC2285/CC2286 and the resulting fragment cloned between the SpeI and EcoRI restriction sites in pl801.
The pDG148-ycsA-GFP (pl862) plasmid was made by PCR amplification of ycsA from chromosomal DNA using oligonucleotides CC2546/CC2547 and gfp from pHM2-5'hbs-GFP plasmid (pl696) using oligonucleotides CC2548/CC572. The two overlapping PCR fragments containing ycsA and gfp were assembled in new PCR with oligo pair CC2546/CC572. The ycsA-GFP fragment was cloned in pHM2 between the EcoRI and BamHI restriction sites. Due to the low expression after insertion of this fusion at the amyE locus, this construction was reamplified with oligo pair CC2611/CC2612 to clone it in pDG148 between the EcoRI and PvuII restriction sites.
To overexpress the part of dnaE predicted to base-pair with RoxS, we inserted pHM2-Pspac con -dnaE at the amyE locus. This plasmid was constructed by amplification of a part of dnaE by PCR using oligo pair CC2703/CC2705. This fragment was cloned in pHM2-Pspac con between BamHI and SalI restriction sites.
The ytsJ-GFP translational fusions were made by amplification by PCR of a DNA fragment starting at the 11 promoter of the P2-ytsJ mRNA and covering the first 96 AA of the YtsJ CDS using oligo pair CC2815/CC2822. The 39 UTR of ytsJ with its transcriptional terminator was amplified in a PCR using oligo pair CC2823/CC2816. The GFP fragment was amplified using oligo pair CC2821/CC2824. The three overlapping fragments were amplified in a new PCR using oligo pairs CC2815/CC2816. The resulting product, which is a translational fusion of the GFP after the 96th aa of YtsJ CDS, was cloned in pDG148 between the ClaI and HindIII restriction sites (pl912). Because the level of the GFP was too low to be measurable, we place this fusion under the inducible promoter Pspac of the pDG148-Pspac plasmid. The fusion was reamplified from pl912 using oligo pair CC2984/CC2816. The transcriptional start site of this transcript is close to the 11 of the P2 promoter (Pspac(P2)-ytsJ-GFP). We also amplified a shorter version of this fusion to get a transcriptional start site close to the 11 from P1 promoter (Pspac(P1)-ytsJ-GFP) using oligo pair CC2704/CC2816. Both fusions, Pspac(P1)-ytsJ-GFP and Pspac(P2)-ytsJ-GFP were cloned into the HindIII restriction site of the pDG148-Pspac plasmid to give pl935 and pl934, respectively.
Pl982 was made by cloning a mutated version of RoxS in CRR3 region (RoxS CRR3/3G ) in pDG1662 between the EcoRV and SphI restriction sites. The mutation of RoxS was made by using the oligos pairs CC1377/CC2929 et CC1378/CC2928. The two overlapping fragments were amplified in a new PCR using oligo pairs CC1377/CC1378.
Medium. Strains were grown in MD medium (24), 2xYT or LB medium as indicated, all supplemented with 0.5% malate.
Sample preparation for RNAseq and pulsed SILAC experiments. Strains CCB1244 and CCB1243 were grown in 50-mL MD modified supplemented with 0.5% malate and a limiting concentration of lysine (85.5 mM) with shaking. Cultures were grown to OD 600 = 0.5 and split in two cultures of 20 mL. For the pulse-labeling, 1% arabinose and 1.7 mM medium 2 H 4 -lysine or heavy 13 C 8 -lysine was added to CCB1244 and CCB1243, respectively, for 15 min before harvesting the cultures (10 mL for RNAseq analysis and 10 mL for pulsed SILAC analysis). Experiments were performed in triplicate.
RNAseq analysis. Total RNA was isolated from 10 mL culture (pelleted and frozen) by the glass beads/phenol method described previously (25). RNA samples were treated with RQ DNase Promega (37°C for 20 min) to remove potential contaminating chromosomal DNA. rRNA was removed using the RiboZero kit (Illumina), and rRNA depletion and overall RNA quality was analyzed by Bioanalyser (Agilent). cDNA libraries were prepared using the Smarter Stranded RNA-Seq Kit (Clontech) with adapters for multiplexing, according to the manufacturer's instructions. cDNA concentration and quality were checked by Bioanalyser (Agilent). The six samples were normalized to 2 nM, multiplexed and denatured at a concentration of 1 nM using 0.1 N NaOH (5 min at room temperature) before dilution to 10 pM and loading on a HiSeq Rapid SE65. Reads were mapped by Bowtie 2 (26). The analysis was performed using the R software, Bioconductor (27) packages including DESeq2 (28,29) and the PF2tools package (version 1.5.3) developed at PF2 (Institut Pasteur). Normalization and differential analysis were carried out according to the DESeq2 model and package (version 1.20.0).
All RNAseq raw data files can be freely downloaded at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE229415 Pulsed-SILAC samples preparation. Proteins were extracted from 10 mL cultures. Cells were lysed twice with a French press, centrifuged to remove cell debris, and proteins were precipitated with 5 volumes of ice cold acetone. Protein extracts from the pulse-labeling experiments (arabinose-induced RoxS small RNA labeled with heavy 13 C 8 -lysine and control cultures labeled with medium 2 H 4 -lysine) were mixed in a 1:1 (wt/wt) ratio and 25 mg of proteins were loaded on a 12% acrylamide gel prior to separation according to their molecular weight by SDS-PAGE. The gel was fixed and stained with Coomassie brilliant blue R250. The gel lanes corresponding to pulsed-SILAC biological triplicates were cut alike into five bands and subjected to a manual in-gel digestion with modified porcine trypsin (Trypsin Gold, Promega). After destaining, bands were subjected to a 30-min reduction step at 56°C using 10 mM dithiotreitol in 50 mM ammonium bicarbonate (AMBIC) prior to a 1-h cysteine alkylation step at room temperature in the dark using 50 mM iodoacetamide in 50 mM AMBIC. After dehydration under vacuum, bands were re-swollen with 250 ng of trypsin in 50 mM AMBIC and proteins were digested overnight at 37°C. Supernatants were kept and peptides present in gel pieces were extracted with 1% (vol/vol) trifluoroacetic acid. Corresponding supernatants were pooled and dried in a vacuum concentrator. Peptide mixtures were solubilized in 25 mL of solvent A (0.1% (vol/vol) formic acid in 3% (vol/vol) acetonitrile) for mass spectrometry analysis.
Tandem mass spectrometry of SILAC-labeled protein extracts. Mass spectrometry analyses were performed on a Q-Exactive Plus hybrid quadripole-orbitrap mass spectrometer (Thermo Fisher Scientific, San José, CA, USA) coupled to an Easy 1000 reverse phase nano-flow LC system (Proxeon) using the Easy nanoelectrospray ion source (Thermo Fisher Scientific). Each peptide mixture was analyzed in duplicate. Five mL were loaded onto an Acclaim PepMap precolumn (75 mm Â 2 cm, 3 mm, 100 Å; Thermo Fisher Scientific) equilibrated in solvent A and separated at a constant flow rate of 250 nL/min on a PepMap RSLC C18 Easy-Spray column (75 mm Â 50 cm, 2 mm, 100 Å; Thermo Fisher Scientific) with a 90-min gradient (0% to 20% B solvent [0.1% (vol/vol) formic acid in acetonitrile] in 70 min and 20% to 37% B solvent in 20 min).
Data acquisition was performed in positive and data-dependent modes. Full scan MS spectra (mass range m/z 400 to 1,800) were acquired in centroïd with a resolution of 70,000 (at m/z 200) and MS/MS spectra were acquired in centroid mode at a resolution of 17,500 (at m/z 200). All other parameters were kept as described in (30) Data processing and SILAC quantification. Raw data were processed using the MaxQuant software package (http://www.maxquant.org, version 1.5.6.5) (31). Protein identifications and target decoy searches were performed using the Andromeda search engine and the UniprotKB database restricted to Bacillus subtilis taxonomy (release: 01/2019; 4,260 entries) in combination with the Maxquant contaminant database (number of contaminants: 245). The mass tolerance in MS and MS/MS was set to 10 ppm and 20 mDa, respectively. Methionine oxidation, protein N-terminal acetylation as well as heavy 13 C 8 -lysine and medium 2 H 4 -lysine were taken into consideration as variable modifications whereas cysteine carbamidomethylation was considered as fixed modification. Trypsin was selected as the cutting enzyme and a maximum of two missed cleavages were allowed. Proteins were validated if at least two unique peptides with a peptide FDR , 0.01 were identified. The setting "Match between runs" was also taken into consideration to increase the number of identified peptides. For quantification, we only used unique peptides with a minimum ratio count $ 1. Protein ratios were calculated as the median of all peptide ratios for the protein of interest and a normalization step was also included in the quantification process.
Statistical analysis of the pulsed-SILAC data set. Statistical analysis was done using Perseus software (https://maxquant.org/perseus, version 1.6.0.7) on normalized protein ratios. Proteins belonging to contaminant and decoy databases were filtered. For each biological replicate, the median ratios (M/L and H/L) of the two injected replicates were determined. Then, a Student's t test (threshold Pvalue , 0.05) was applied to detect statistically relevant variations between these two ratios. Moreover, we considered only (H/L)/(M/L) ratios having a fold change .1.4 for upregulated gene and ,0.7 for downregulated gene to validate newly synthetized proteins that are specific to the arabinose-induced RoxS sRNA condition.
IntaRNA prediction. The following parameters have been used for the intaRNA prediction. The output parameters were defined as follows: number of interactions per RNA pair: 5; suboptimal interaction overlap: can overlap in target; max absolute energy of an interaction: 0, max delta energy above mfe of an interaction: 100, no lonely base pairs and no GU helix ends. The seed parameters were: minimum of base pairs in seed: 7. The folding parameters were: temperature for energy computation: 37°C; folding window size: 150; max. base-pair distance: 100; folding window size: 150; base-pair distance 100; energy parameter set: Turner model 2004. For each prediction, the target sequence encompasses the proximal