Regulatory Interplay between RNase III and Antisense RNAs in E. coli: the Case of AsflhD and FlhD, Component of the Master Regulator of Motility

The role of antisense RNAs (asRNAs) in the regulation of gene expression remains largely unexplored in bacteria. Here, we confirm that asRNAs can be part of layered regulatory networks, since some are found opposite to genes encoding global regulators. ABSTRACT In order to respond to ever-changing environmental cues, bacteria display resilient regulatory mechanisms controlling gene expression. At the post-transcriptional level, this is achieved by a combination of RNA-binding proteins, such as ribonucleases (RNases), and regulatory RNAs, including antisense RNAs (asRNAs). Bound to their complementary mRNA, asRNAs are primary targets for the double-strand-specific endoribonuclease, RNase III. Taking advantage of our own and previously published transcriptomic data sets obtained in strains inactivated for RNase III, we selected several candidate asRNAs and confirmed the existence of RNase III-sensitive asRNAs for crp, ompR, phoP, and flhD genes, all encoding global regulators of gene expression in Escherichia coli. Using FlhD, a component of the master regulator of motility (FlhD4C2), as our model, we demonstrate that the asRNA AsflhD, transcribed from the coding sequence of flhD, is involved in the fine-tuning of flhD expression and thus participates in the control of motility. IMPORTANCE The role of antisense RNAs (asRNAs) in the regulation of gene expression remains largely unexplored in bacteria. Here, we confirm that asRNAs can be part of layered regulatory networks, since some are found opposite to genes encoding global regulators. In particular, we show how an antisense RNA (AsflhD) to the flhD gene, encoding the transcription factor serving as the primary regulator of bacterial swimming motility (FlhD4C2), controls flhD expression, which in turn affects the expression of other genes of the motility cascade. The role of AsflhD highlights the importance of fine-tuning mechanisms mediated by asRNAs in the control of complex regulatory networks.

regulation, as in the case of the regulation of plasmid copy number and toxin-antitoxin systems (7,8).
The mechanisms of action of asRNAs are diverse. They can negatively regulate transcription by interference due to the collision of two converging RNA polymerases (RNAPs) or by attenuation due, in some cases, to the stabilization of a terminator structure in the mRNA upon binding of the asRNA (9,10). Moreover, despite their complementarity, the interaction of an asRNA with its target requires, in some cases, formation of an intermediate called "kissing complex" (7,11). These interactions can have negative or positive consequences on gene expression since they induce modifications to the RNA secondary structure and/or physically interfere with the activity of other regulators (12,13). In a surprisingly large number of cases, the mechanism by which a specific asRNA regulates its target remains unclear due, in part, to the impossibility of modifying the sequence of the asRNA independent of its target.
More recently, various genomic approaches have been used leading to the identification of hundreds to thousands of asRNAs and/or antisense transcription start sites (TSSs) in the transcriptome of E. coli. These approaches include genomic library overexpression (14), inhibition of Rho-dependent termination (15), mapping of transcriptional units (16), capture of double-stranded RNAs (17,18) and enrichment of primary (19)(20)(21) or small transcripts (22). It is interesting to note that comparison of some of these data sets revealed only a modest overlap (19), enforcing the idea that asRNAs are difficult to identify and may not be well conserved even between related bacterial species.
Taking advantage of available transcriptomic data sets and of our previous study, during which a tailored transcriptome sequencing (RNA-seq) analysis was performed (23), we compared transcriptomes of an rnc mutant to its isogenic wt strain. We identified and validated the expression of four new asRNAs complementary to the coding sequence of genes crp, ompR, phoP, and flhD. We then investigated in detail how AsflhD controls the expression of flhD, one gene of the master regulator of swimming motility, FlhD 4 C 2 , and consequently affects the process of motility.

RESULTS
Characterization of asRNAs upon RNase III inactivation. We previously performed a transcriptomic approach in E. coli K-12 (wt) and its rnc105 derivative strain (rnc) by tagging transcripts according to their 59-phosphorylation status, allowing us to distinguish between 59-triphosphate fragments (primary transcripts, TSS), 59-monophosphate fragments (processed transcripts, PSS), and internal fragments resulting from the fragmentation (INT) (23). From this data set, we sorted the antisense reads to open reading frames (ORFs) that were enriched upon RNase III inactivation and checked whether they have been detected in independent data sets (as described in Text S1) (14,16,17,19,21). Four asRNAs complementary to gene coding for major transcriptional regulators were selected, i.e., asRNAs to crp, ompR, phoP, and flhD. Their coordinates are indicated in Table 1, and their expression was confirmed by northern blotting (Fig. 1). Of note, the identification of RNase III processing sites in the wt strain was usually not obvious since cleavage by RNase III presumably provoked the subsequent rapid degradation of the asRNA.
The crp gene encodes the major regulator of carbon catabolite repression, and it was shown previously to be transcriptionally regulated by the recruitment of the RNAP to a divergent and overlapping TSS located 3 nt upstream from the crp TSS (24,25) encoding the divergently expressed gene yfhA. We observed antisense reads validating this previously characterized TSS in the wt strain. In addition, antisense reads complementary to the ORF and 59-untranslated region (59-UTR) of the crp mRNA accumulate in the rnc mutant ( Fig. 1A; Table 1), and we predicted an additional TSS to be located 112 nt downstream from the start codon of crp. Northern blot analysis revealed that , phoP (C), and flhD (D). RNAs extracted from wild-type (wt) strain N3433 and its rnc derivative IBPC633 were analyzed on agarose or denaturing acrylamide gels, and northern blots were probed by using pairs of complementary uniformly radiolabeled RNA probes to the same region of each target and a primer complementary to the 5S rRNA. To note, crp and AsompR were successively probed on the same membrane; thus, they share the same loading control. The membrane lanes shown for phoP and AsphoP (C) and AsflhD (D, left) correspond to the zero time points of the stability experiments presented in Fig. 2 (D and E) and Fig. S1A, respectively, which show the uncropped membranes. Transcriptome sequencing (RNA-seq) reads were aligned to the genome of reference (MG1655, GenBank identifier U00096.3) and visualized with Integrative Genomic Viewer (IGV) software, version 2.4.2. Antisense reads are oriented from left to right (59 to 39) independently of their genomic orientation. The fractions isolated during the RNA-seq analysis are color-coded: strands of the target gene in wt (dark blue) and in rnc mutant (light blue) and strands of the asRNA gene in wt (red) and in rnc mutant (orange). Reads corresponding to transcription start site (TSS), processing sites (PSS), and internal fragments (INT) are indicated. The scale for the absolute number of reads identified is indicated on the top right of each lane. Localization of open reading frames (ORFs) and known promoters (full bent arrows) and putative antisense promoters deduced from transcriptomic data sets (dashed bent orange arrows) are schematized on their respective loci, while the positions of the probes relative to the DNA sequence are indicated by a gray dashed box. this asRNA, here referred to as Ascrp, is stabilized in the rnc mutant as a fragment longer than 307 nt.
The ompR gene encodes the response regulator of a two-component system involved in cell wall homeostasis and response to low pH, EnvZ-OmpR (26)(27)(28)(29). From our and previous data sets, we predicted a TSS located 147 nt downstream from the start codon of ompR ( Fig. 1B; Table 1). Northern blotting with complementary probes corresponding to the 59-end of the ompR ORF confirmed the presence of the ompR-envZ dicistronic mRNA and revealed a long asRNA, here referred to as AsompR, of about 2,000 nt, which is likely to also encode the divergently expressed greB gene. In addition, smaller fragments (less than 500 nt) are detected in the rnc mutant for both ompR and AsompR (Fig. 1B), which could correspond to a stable duplex between ompR and AsompR cleaved by RNase III in the wt strain.
The phoP gene encodes the response regulator of the PhoQ-PhoP two-component system, involved in response to low magnesium and in cell wall homeostasis (30,31). From our and previous data sets, we predicted a TSS to be located 282 nt downstream from the phoP start codon ( Fig. 1C; Table 1). Northern blots hybridized with complementary probes corresponding to the 59-end of the phoP ORF confirmed the accumulation of two fragments (AsphoP) about 300 to 320 nt long in the rnc mutant (Fig. 1C).
The flhDC operon encodes the master regulator of motility, FlhD 4 C 2 (32). Our and previous data sets allowed us to predict a TSS to be located 22 nt downstream from the flhD start codon ( Fig. 1D; Table 1). Northern blot analysis with probes specific to the 59-UTR of flhD confirmed the accumulation of an asRNA to flhD in the rnc mutant, here referred to as AsflhD, as a major fragment of about 220 nt and a minor fragment of about 160 nt upon RNase III inactivation. At the same time, an increase in the amount of the full-length flhD mRNA and the stabilization of a flhD fragment with an approximate size of 220 nt were observed in the rnc mutant (Fig. 1D), which could correspond to a double-stranded RNA formed between flhD and AsflhD RNAs.
The four identified asRNAs fragments are likely to be processed by RNase III since they are only visible in the rnc strain. However, we cannot exclude the possibility that RNase III is involved indirectly at another stage in the regulation of their transcription. Crp, OmpR, PhoP, and FlhD are major regulators of gene expression involved in the control of large regulons (RegulonDB version 10.5 [33]), which are conserved among gammaproteobacteria (34). We wondered whether these asRNAs and RNase III have a functional regulatory role on their targets and thus affect cell physiology. We investigated asRNAs to phoP and flhD, two regulators whose expression is known to be tightly controlled at both the transcriptional and post-transcriptional levels (35,36).
Regulation of phoP and AsphoP by RNase III. Candidate consensus 210 and 235 sequences were predicted upstream of AsphoP TSS ( Fig. 2A). To validate this promoter, we constructed a P AsphoP -lacZ transcriptional fusion containing 150 nt before and 15 nt after the putative TSS of AsphoP, with the wt sequence (P AsphoP(wt) ) or mutations (P AsphoP (24) ) decreasing the agreement with the consensus in the predicted 235 and 210 boxes (Fig. 2B). The mutated AsphoP promoter (P AsphoP (24) ) strongly decreased the expression of P AsphoP -lacZ (20-fold), confirming it as an endogenous AsphoP promoter (Fig. 2C). Furthermore, inactivation of RNase III led to an increase (1.6-fold) of phoP mRNA stability (Fig. 2D) and revealed that AsphoP is much more stable than phoP mRNA in the rnc mutant (Fig. 2E). In summary, we confirmed that AsphoP is transcribed from the predicted promoter and that RNase III negatively affects the expression of phoP and AsphoP.
Sequence comparison with other bacterial species showed that although the region of the AsphoP promoter is moderately well conserved, there are several A to G substitutions in the 210 box at positions 29 and 212, suggesting that this promoter may be inactive in the compared genomes ( Fig. 2A). This, in turn, implies that, if AsphoP has any function, it could be limited to E. coli K-12 and has been counterselected in these other species or, more likely, represents a novel, evolving trait.
Physiological expression of AsflhD. Candidate 210 and 235 boxes were identified upstream from the putative TSS of AsflhD and sequence alignment of this region in other enterobacteria shows a good conservation of a promoter with an extended 210 59-TG box (37), suggesting that AsflhD expression is conserved among enterobacteria (Fig. 3A). To validate the predicted promoter, a P AsflhD -lacZ transcriptional fusion (P AsflhD(wt) ) was constructed containing 165 nt before and 15 nt after the putative TSS of AsflhD (Fig. 3B). This fusion showed a relatively low level of b-galactosidase activity (10MU; Fig. 3C). Its expression was strongly increased (34-fold) when the 210 motif was improved toward the RpoD consensus (P AsflhD (11) ), while mutating the 235 and 210 to less consensus sequences (P AsflhD (22) , P AsflhD (21) , and P AsflhD(23) ) decreased expression (3.7-to 8.7-fold), hence validating the predicted promoter (Fig. 3C). It should be noted that mutations were designed to be used in the endogenous flhD locus and chosen to minimally affect the coding sequence of flhD and to avoid introduction of rare codons. However, the P AsflhD(11) mutation produces an aspartate to asparagine change at position 12 of FlhD (D12N), which may affect FlhD function (see below). The low level of expression made us wonder whether AsflhD may be expressed using an alternative sigma factor. Heat shock increased P AsflhD -lacZ expression 1.8-fold after 15 min and 4.3-fold after 60 min (Fig. 3D). Comparison of P AsflhD with consensus sequences for the two heat-shock sigma factors, s H (RpoH) and s E (RpoE) (38,39), shows better correlation with the s E consensus than with s H (Fig. 3A), suggesting that RpoE could be involved in the transcription from the P AsflhD .
Hence, we examined whether P AsflhD is under the control of RpoE by using a strain deleted for rseA (anti-sigma factor inhibitor of RpoE), which leads to the strong induction of the RpoE regulon (40,41). Deletion of rseA increased the expression of the wt P AsflhD -lacZ fusion (Fig. 3E, 1.7-fold), comparable with the effect of the heat shock at . Nucleotides mutated to inactivate P AsphoP are shown in red, stars represent conserved nt relative to Eco, and the 235 and 210 motifs of AsphoP are highlighted in gray. (B) Genetic structure of the transcriptional P AsphoP -lacZ reporter fusion (P AsphoP(wt) in MG2118). Mutations repressing activity of the AsphoP promoter (P AsphoP (24) in MG2120) are in red. (C) Expression of b-galactosidase activity (given as Miller units [67]) was determined from the P AsphoP(wt) -lacZ and P AsphoP(-4) -lacZ fusions in the wt strain. The values are the means of three biological replicates for each strain, and the bars indicate the standard deviations. Statistical significance was determined by analysis of variance (ANOVA). ***, P # 0.001. (D and E) Total RNA was prepared from samples taken from the wt strain (N3433) and its rnc derivative (IBPC633) at different times after addition of rifampicin and was subjected to northern blot analysis. The membranes were probed successively for phoP or AsphoP and 5S. The decay rate of phoP mRNA was calculated as described in the Materials and Methods section. 46°C, known to induce the RpoE regulon (42). Thus, our results suggest that AsflhD is more expressed at high temperature and when the RpoE sigma factor is activated.
To further characterize AsflhD, the P AsflhD (22) , P AsflhD (21) , P AsflhD (23) , and P AsflhD(11) mutations were introduced at the endogenous flhD locus, and the expression of AsflhD was examined by northern blotting (Fig. 3F). No AsflhD was detected in the rnc derivatives of strains with the three mutations reducing the expression of AsflhD. Conversely, AsflhD overexpression from the P AsflhD(11) mutation led to the detection of a faint smear in the wt strain and to the much greater accumulation of AsflhD in the rnc mutant (4.6fold relative to the rnc mutant containing the P AsflhD(wt) ). In summary, we have identified the promoter of AsflhD and shown that mutations in the promoter of AsflhD can be used as tools to study the function of AsflhD at the flhD locus.
Processing of AsflhD. Circular RT-PCR (cRT-PCR) experiments confirmed that AsflhD is expressed in both the wt and rnc strains from P AsflhD with heterogeneous AsflhD 39-ends extending up to 345 nt in the rnc mutant (Fig. 4A). Surprisingly, no 220-nt-long RNA was detected in the mutant by cRT-PCR, while a 149-nt-long fragment was found several times exclusively in the wt strain, which might be an intermediate in the degradation of AsflhD, e.g., via RNase E (see below). Surprisingly, we did not detect 220-nt RNAs observed by northern blotting in the rnc strain. As this species is likely to be present as double-stranded RNA with the processed flhD mRNA (as described below), we suspect that it is less efficiently ligated during the initial step of the cRT-PCR, as previously reported (see Materials and Methods) (43). We investigated the degradation of AsflhD by The wt strain (N3433) and its derivatives, pnp, rnc, pnp-rnc, rne ts , and rnc-rne ts mutants (N3433-pnp, IBPC633, IBPC633-pnp, N3431, and IBPC637, respectively), were grown at 37°C until mid-log phase. Where indicated, the cells were grown at 30°C and submitted to a heat shock at 42°C for 15 min in order to inactivate RNase E in the strain carrying the thermosensitive rne ts allele. Total RNA was analyzed by northern blotting. The membrane was probed successively for AsflhD and 5S. AsflhD fragments are indicated by orange arrowheads.
other RNases. Inactivation of the major endonuclease RNase E allowed the detection of a 300-nt fragment (independently of the presence of RNase III), whereas the loss of the exoribonuclease polynucleotide phosphorylase (PNPase) had no effect on AsflhD degradation (Fig. 4B). Hence, RNase III and RNase E are independently involved in the rapid turnover of AsflhD.
To understand the role of RNase III in the degradation of flhD and AsflhD RNAs, we analyzed the effect of RNase III inactivation on the stability of both RNAs. In the rnc mutant, the major 220-nt-long AsflhD transcript and minor 160-nt-long transcript were highly stable, while both the amount and the stability of the full-length flhDC mRNA (here referred to as flhD FL ) increased 2-fold ( Fig. 5A and B). In addition, the 59-UTR probe used ( Fig. 1D and 5B, top; Table S1) detected a 220-nt-long 59-UTR flhD RNA fragment (here referred to as flhD p ), complementary to AsflhD, which is also highly stable in the rnc strain. The interaction of flhD p with AsflhD is expected to generate a double-stranded RNA (Fig. 4A), the degradation of which depends on RNase III. Supporting this hypothesis, we could not detect flhD p RNA in a strain with RNase III inactivated and in which the endogenous AsflhD expression was reduced (P AsflhD (22) ), while flhD p RNA accumulates when AsflhD expression was increased (P AsflhD (11) ; Fig. 6A; Fig. S1B). We further confirmed the interaction and cleavage by RNase III of AsflhD and flhD in vitro. A 308-nt-long flhD transcript corresponding to the 59-UTR and part of the ORF of the flhD mRNA and a 256-nt-long AsflhD asRNA were synthesized and labeled at their 59-ends. These two RNAs form a duplex when present in equimolar concentrations (Fig. S2A), which is completely degraded upon addition of RNase III (Fig. S2B).
Remarkably, under the same condition, RNase III cleaves the individual RNAs independently at two sites on AsflhD and at four major sites on flhD ( Fig. S2C and D). These cleavage sites are located within regions able to form a secondary structure on each molecule (44, 45) ( Fig. S2C and D). RNase III is thus able to cleave both AsflhD and flhD RNAs in vitro, at specific sites but is also able to drive the complete degradation of the Regulation of flhD expression by AsflhD. We next determined the effect of increased or decreased expression of AsflhD, in cis, on flhD expression by following flhD mRNA abundance and stability in strains carrying the endogenous P AsflhD mutations described above. While a decrease in flhD mRNA abundance results from both decreased AsflhD (1.5-fold in P AsflhD (22) , 1.6-fold in P AsflhD (21) , and 1.2-fold in P AsflhD (23) ) and increased AsflhD expression (1.8-fold in P AsflhD (11) ; Fig. 6A and B; Fig. S3A), the stability of flhD mRNA was not significantly affected in either the P AsflhD (22) or P AsflhD(11) mutant ( Fig. S1A to C).
A transcriptional/translational (P flhD -flhD-lacZ) and a translational flhD-lacZ reporter fusion under the control of a constitutive promoter (P tet -flhD-lacZ) encompassing the 59-UTR and the first 34 amino acids of FlhD (which includes P AsflhD ) were introduced at the lacZ chromosomal locus (Fig. 6C). Mutations in P AsflhD resulting in decreased (P AsflhD (22) ) and increased expression (P AsflhD (11) ) of AsflhD were also introduced into both fusions. Both mutations reduced expression of both fusions, but the effect of the AsflhD-overexpressing  (22) in ML73 and P AsflhD (11) in ML241) and their rnc mutant derivatives (ML65, ML75, and ML341, respectively) and subjected to northern blot analysis. The membrane was probed successively for flhD and for M1 RNA. (B) Average flhD mRNA abundance in the P AsflhD (22) (red) and P AsflhD(11) (green) mutants relative to the wt strain (gray), as shown in panel A, was calculated as the mean of five biological replicates. (C) Genetic structures of the P flhD -flhD-lacZ (ML219) and P tet -flhD-lacZ reporter fusions (ML233) and their derivatives containing the mutations leading to either decreased expression (P AsflhD (22) , ML221 and ML235, respectively, in red) or increased expression (P AsflhD (11) , ML226 and ML237, respectively, in green) of AsflhD. (D, E) Expression of P flhD -flhD-lacZ (D) and P tet -flhD-lacZ (E) reporter fusions (gray) and their derivatives are given as b-galactosidase activity. The values are means of 10 biological replicates for each strain, and the bars indicate standard deviations. Statistical significance was determined by ANOVA. **, P # 0.01; ***, P # 0.001; ****, P # 0.0001.
Antisense RNA Control of flhD Expression mBio mutation (P AsflhD (11) ) was greater (4.9-fold decrease of the fusion driven by the wt flhD promoter and 2.3-fold on the P tet -driven version) than the effect of the promoter-down mutation (1.4-to 1.2-fold in P AsflhD (22) ; Fig. 6D and E). These results, obtained during the exponential phase of growth, were confirmed by measurements in late exponential phase (A 600 % 1; Fig. S3B and C) when flhD expression, from the P flhD promoter, increases (as previously reported [46]). It can also be noted that AsflhD overexpression has a greater impact on flhD-lacZ expression from the native flhD promoter (Fig. 6D) than on flhD mRNA abundance (Fig. 6A) or flhD-lacZ expression from the P tet promoter (Fig. 6E). In brief, reducing or increasing the expression of AsflhD reduces flhD expression at the translational and mRNA levels, while not appreciably affecting the stability of the flhD mRNA. Hence, this suggests that a native intermediate level of AsflhD expression is required for optimal flhD expression.
AsflhD represses flhD expression in trans. Increasing the endogenous AsflhD expression leads to the repression of flhD expression. Hence, we next investigated the ability of AsflhD to repress the expression of flhD when expressed in trans. AsflhD was overexpressed from a plasmid, under the control of a P tac promoter inducible by isopropyl-b-D-thiogalactopyranoside (IPTG). The short (242 nt) AsflhD is transcribed from the 11 to the 1220 nucleotides (nt) relative to the TSS of AsflhD (i.e., 1220 to 11 relative to the TSS of flhD) with a rrnBT2 terminator to stabilize the transcript. Under inducing conditions, trans-overexpression of AsflhD is stronger than cis-overexpression (9-fold in the rnc mutant; Fig. 7A) and can be directly observed in the wt strain. In agreement with results presented in the previous section (Fig. 6B), the trans-overexpression of AsflhD decreased the abundance of flhD mRNA both in the wt strain and in the rnc mutant ( Fig. 7B and C). Furthermore, the short flhD p RNA, accumulating in the rnc mutant upon cis-overexpression of AsflhD, was even more abundant upon trans-overexpression of AsflhD (Fig. S4B). It is not observed with the 39-probe (Fig. 7B), confirming that it is derived from the region complementary to AsflhD. It is noteworthy that despite a higher expression in trans, plasmid-borne AsflhD is not as efficient at repressing the expression of flhD. This difference may be due to the presence of the rrnBT2 terminator and/or due to pervasive plasmid transcription, including transcripts antisense to the AsflhD insert, which can be observed when probing for the 59-UTR of flhD in a strain deleted for the endogenous flhD locus (Fig. S4A). Of note, a similar problem has been reported for other vectors (both in eukaryote and bacteria) in which spurious expression of multiple overlapping transcripts was detected (18,47). We confirmed these results on flhD mRNA levels by measuring the expression of the flhD-lacZ reporters under conditions of trans-overexpression of AsflhD and observed a small repression of flhD expression in both P flhD -flhD-lacZ and P tet -flhD-lacZ fusions (1.4 and 1.6-fold; Fig. 7D and E). It should be emphasized that these effects are independent of the flhD promoter (native P flhD or P tet ). A combination of cis-overexpression (endogenous to the reporter) and trans-overexpression (from the plasmid) of AsflhD had little additive effect on the final repression when flhD was expressed from its own promoter (Fig. S4C), consistent with AsflhD repressing flhD expression by a common mechanism when expressed in cis or in trans. In summary, we show that trans-overexpression of AsflhD can repress flhD expression at both mRNA and translational levels.
Control of the motility cascade by AsflhD. The flhDC operon encodes the FlhD 4 C 2 transcriptional regulator, main activator of the cascade of motility-related genes, which are divided into three classes (48). The flhDC operon encodes the only class I protein complex, FlhD 4 C 2 , which is essential for expression of class II genes, which in turn control class III genes. Thus, we next investigated the impact of changing AsflhD levels on representative class II and III genes. We selected the following class II genes: fliA that encodes FliA, the sigma factor for class III motility genes, and flgB that encodes FlgB, a component of the flagella proximal rod and the class III gene fliC gene, encoding the main component of flagella, FliC. The amounts of fliA, flgB, and fliC mRNAs are reduced (from 1.2-to 1.9-fold) when AsflhD expression is reduced (mutations P AsflhD(22), P AsflhD (21) , and P AsflhD (23) ) and strongly reduced (from 5-to 100-fold) upon cis-overexpression of AsflhD (P AsflhD (11) ; Fig. 8A to C; Fig. S5A to C). The effects were strongest for the class III gene fliC. Using a P fliC -lacZ transcriptional reporter fusion, we confirmed the reduced expression of fliC when AsflhD levels decreased (from 1.6-to 2.4-fold; Fig. 8D; Fig. S5D). Surprisingly but consistent with the northern blot ( Fig. 8C; Fig. S5C), cis-overexpression of AsflhD by the P AsflhD(11) mutation produced a very large decrease in fliC-lacZ expression (218-fold; Fig. 8D). It has previously been observed that a modest reduction of flhD transcription could lead to a strong repression of the motility cascade (49,50), and this could be the case for fliC, which depends upon the FlhD 4 C 2 -dependent class II sigma factor, FliA, for its expression. We had measured a reduction in flhD expression and mRNA and a greater repression of class II gene fliA, but the very strong repression of fliC-lacZ exerted by the cis-overexpression of AsflhD raised the question of whether the P AsflhD (11) mutation affected the activity of FlhD since it leads to the mutation of the 12th amino acid (FlhD D12N ). Thus, we cannot exclude that this change might affect FlhD activity. However, it was previously shown that the FlhD D12A mutation did not affect motility (51), and D12 is not in a region involved in contacting FlhC in the FlhD 4 C 2 complex (52). Moreover, several pieces of evidence show that the FlhD D12N (P AsflhD (11) ) protein is still active, in particular because the P AsflhD(11) mutant strain is still motile (see below).
To verify that the changes in AsflhD gene expression were reflected in bacterial behavior, we analyzed the effect of AsflhD on motility using low-agar plates and observed a reduction in the swimming speed when AsflhD expression was either reduced (1.3-fold in P AsflhD (22) ) or increased (5.7-fold in P AsflhD (11) ), while a strain deleted for the flhD locus (DflhD) showed no motility ( Fig. 8E and F). Swimming motility was measured in super optimal broth in the presence of magnesium (SOB1Mg). This medium (as opposed to LB used for other studies) was preferred to allow measurements within 12 h, since flhD expression and motility are increased (see Materials and Methods). In addition, we verified that expression of the P flhD -flhD-lacZ and P flic -lacZ reporter fusions were also repressed by the cis-overexpression of AsflhD (P AsflhD (11) ) in SOB1Mg (5.7-and 81-fold, respectively; Fig. S6). Importantly in SOB1Mg the expression of fliC-lacZ in the P AsflhD(11) mutant is 15-fold higher than in a strain deleted for flhD (Fig. S6B), confirming that the FlhD (D12N) protein is at least still partially functional.
We also attempted to confirm that AsflhD expressed in trans from the plasmid reduced fliC expression. As shown above, the trans-overexpression of AsflhD modestly represses the expression of flhD (Fig. 7B to E), while we found a small reduction of fliC expression at both mRNA (1.6-fold; Fig. S7A) and translational levels (2-fold; Fig. S7B) compared to much stronger effects on fliC by the P AsflhD (11) mutation. Expression of AsflhD in trans did, however, lead to a slight reduction of the swimming speed (1.2-fold; Fig. S7D).
It is possible that the effect of AsflhD could also be partially due to independent regulatory events on other targets within the motility cascade. However, a bioinformatics search (TargetRNA2 [53] and CopraRNA [54]) for possible direct trans targets of AsflhD found no candidates among genes from the motility cascade. Furthermore, as shown previously (48) and above (in SOB1Mg; Fig. S7C), fliC expression is dependent  (22) ) or increasing (P AsflhD (11) ) swimming motility compared to wt (MG1655-B) and loss of flhD. (F) Swimming motility speed, measured as described in the Materials and Methods section, are given for MG1655-B wt (gray) and its derivatives containing mutations reducing (P AsflhD (22) ; red) or increasing (P AsflhD (11) ; green) the expression of AsflhD. The values are means of seven (D) and three (F) biological replicates, and bars indicate standard deviations. Statistical significance was determined by ANOVA. **, P # 0.01; ****, P # 0.0001. n.s., not significant.
Antisense RNA Control of flhD Expression mBio upon flhD via the sigma factor fliA. trans-overexpression of AsflhD had no effect on the expression of fliC-lacZ in a flhD mutant strain (Fig. S7C). Thus, all results are congruent with the notion that AsflhD affects the expression of fliC and motility through the repression of flhD ( Fig. 8; Fig. S7D). In summary, both reduced and increased expression of AsflhD repress the expression of flhD, which in turn leads to repression of the whole cascade of motility and a reduction of the swimming speed. Transcriptional repression of flhD by AsflhD in vitro. We next investigated the mechanism of action of AsflhD. The overexpression of AsflhD represses the expression of flhD at the mRNA and translation level without affecting the stability of the flhD mRNA. Hence, we reasoned that AsflhD could be involved in the transcriptional repression of flhD. To test this hypothesis, we performed in vitro transcription experiments using a DNA template corresponding to the flhD gene from 76 nt before to 388 nt after the TSS of flhD, which allows the transcription of a 388-nt flhD RNA (flhD FL ) and of a 335-nt AsflhD RNA (AsflhD) (Fig. 9A). We compared the abundance of both transcripts synthesized from the latter DNA fragment to those generated from templates carrying the promoter mutations, leading to either decreased (P AsflhD (22), P AsflhD (21) , and P AsflhD(23) ) or increased expression (P AsflhD (11) ) of AsflhD. We confirmed that expression of AsflhD is strongly impaired when transcribed from the template carrying mutations reducing AsflhD repression (around 10-fold in P AsflhD (22) , P AsflhD (21) , and P AsflhD(23) ), and AsflhD expression increases from the template carrying the mutation enhancing the expression of AsflhD (2.7-fold in P AsflhD (11) ; Fig. 9B, left panel, orange bars). At the same time, AsflhD overexpression resulted in a decrease in the transcription of flhD RNA (1.7-fold in P AsflhD (11) ), while the loss of AsflhD led to an increase of the transcription of flhD RNA (1.1-fold in P AsflhD (22) , 1.2-fold in P AsflhD (21) , and 1.5-fold in P AsflhD (23) ; Fig. 9B, left panel, purple bars).
The transcription factor CAP promotes the transcription of flhD by binding to a sequence located 72 nt upstream from the TSS of flhD (55). As expected, the presence of cAMP/CAP increased the transcription of flhD, which was still reduced when expression of AsflhD was increased (1.9-fold) but only slightly increased when AsflhD expression was reduced from the down-mutations (maximum 1.3-fold for P AsflhD (23) ; Fig. 9B, right panel, purple bars). In vitro transcription assays were performed in a single round of elongation in the presence of heparin and with RNAP prebound to templates; hence, the observed effects are restricted to the elongation step and should be independent of the initiation of transcription.
We also investigated the effect of AsflhD on flhD transcription using a template in which the P tet promoter replaced the P flhD promoter (Fig. S8A). This template produces the same 388-nt flhD RNA but a shorter (260 nt) AsflhD RNA. Using these templates with the strong P tet promoter, we observed only a slight reduction of flhD mRNA accumulation from the template carrying the P AsflhD(11) up-or P AsflhD down-mutations (maximum 1.2-fold; Fig. S8B, purple bars). In summary, in vitro enhanced expression of AsflhD from the P AsflhD(11) fragment leads to the repression of transcription elongation of flhD from the P flhD in the presence or absence of CAP/AMPc, while reduction in AsflhD expression has only a modest positive effect on flhD expression.
Mutual repression of flhD and AsflhD transcription in trans. We also investigated the effect of including purified AsflhD or flhD RNA on the transcription of both flhD and AsflhD using the same linear DNA templates. The addition of increasing amounts of AsflhD led to a linear decrease of flhD (up to 1.9-fold in the presence of 120 nM AsflhD; Fig. 9C) without affecting the level of AsflhD expression. The reciprocal assay (addition of increasing concentrations of purified flhD RNA) led to a linear decrease of the amount of AsflhD synthesized (up to 2.8-fold in the presence of 120 nM flhD), while the amount of flhD was not affected. We performed the same assay with the template carrying the P tet promoter (Fig. S8A) and observed similar results (Fig. S8C).
In summary, AsflhD can repress the transcription elongation of flhD both in cis and in trans. Thus, we propose that AsflhD asRNA and flhD mRNA are involved in their mutual transcriptional attenuation in which the interaction of one molecule with the other  (Table S1) and corresponds to 276 to 1388 of the flhD transcript relative to its TSS, with a 40-nt extension carrying the rrnBT2 terminator (fragment length, 504 bp). This fragment carries the native flhD promoter (210 and 235 sites) and includes the cAMP/CAP site at 272 compared to the flhD TSS, at its upstream extremity. (B, C) In vitro transcription assays were performed on templates carrying wt, P AsflhD (11) , P AsflhD (22) , P AsflhD (21) , and P AsflhD(23) mutations as described in the supplemental Materials and Methods section (Text S1), with or without the addition of 100 nM CAP and 0.2 mM cAMP for 15 min at 37°C before addition of RNA polymerase (RNAP) (B) or with 100 nM CAP and 0.2 mM cAMP and the addition of in vitro purified AsflhD or (Continued on next page) Antisense RNA Control of flhD Expression mBio leads to a reduction in transcription at the level of transcription elongation (see discussion).

DISCUSSION
Regulatory RNA molecules are often part of complex genetic networks in bacteria. They correspond to a heterogeneous class of molecules that differ in gene organization, size, and function. Our goal was to detect, identify, and investigate the function of antisense transcripts in E. coli. In this work, we provide evidence that asRNAs can be important players in the expression of transcriptional factors despite their low level of expression. In particular, we have shown that changes in the level of the asRNA to flhD can affect the expression of its target and lead to defects in swimming motility as recapitulated in Fig. 10.
Conservation of AsflhD and AsphoP. Sequences of intergenic regions are usually less well conserved than their neighboring coding sequences, allowing the rapid evolution of regulatory signals. This characteristic is reflected in the presence and conservation of promoters for asRNA and sRNAs. For asRNA initiated within the coding sequence of their target, nucleotide changes within the coding region risk upsetting the function of the ORF and could be counterselected. AsflhD corresponds almost entirely to the 59-UTR of flhD but with the promoter located in the ORF, which is fairly well conserved in enterobacteria (Fig. 3A) (51). Thus, the conservation of the promoter of AsflhD could be the result of direct selection for FlhD activity or for the regulatory function of AsflhD (potentially) controlling the expression of FlhD in these bacteria. In S. enterica serovar Typhi, AsfD, a long asRNA complementary to flhDC and motA mRNAs, was observed as a 2,000-nt fragment by northern blotting in a wt strain (56). AsfD was implicated in the positive regulation of flhDC during stationary phase by an uncharacterized mechanism. However, this asRNA is likely to originate from a region located downstream from the flhDC locus that is not present in E. coli K-12 MG1655 (and no equivalent transcript was observed in our and independent transcriptomic data sets). To our knowledge, no transcriptomic data set is available in Salmonella species inactivated for RNase III. However, a low abundance asRNA to flhD, whose start corresponds to that of AsflhD and that was slightly enriched under nitric-oxide shock, has been previously detected (http://bioinf.gen.tcd.ie/cgi-bin/salcom.pl?_HL) (57). Its localization is consistent with the conservation of AsflhD in S. enterica but we cannot exclude that this transcript is processed from AsfD in S. enterica.  In the case of AsphoP, the promoter also located in the coding region of the phoP ORF is not conserved in other enterobacteria. The lack of conservation in other bacteria suggests that, in contrast to AsflhD, any function of AsphoP may be unique to E. coli, where it most likely arose.
Role of RNase III in the degradation of AsflhD and flhD. We showed that RNase III can cleave both AsflhD and flhD and degrade the AsflhD-flhD RNA duplex in vitro (Fig. S2B to D). In addition, RNase III affects the expression and stability of flhD independent of AsflhD expression in vivo ( Fig. S1; Fig. 6A). Hence, there appear to be two aspects to its action: first, the complete degradation of AsflhD-flhD RNA duplexes, and second, the destabilization of flhD mRNA by cleavages within its 59-UTR. Degradation of flhD/AsflhD duplexes appears to be a stoichiometric event, removing flhD proportionally to the level of AsflhD transcription, since AsflhD is never detected free when RNase III is active in the wt strain. The second mechanism presumably involves cleavages within the 59-UTR that likely modify the complex post-transcriptional regulation of flhD and may facilitate RNase E-mediated degradation.
Mechanism of regulation by AsflhD. Transcriptional repression by asRNAs in bacteria has been proposed to arise either from transcriptional interference (upon colliding convergent RNAP) or from transcriptional attenuation (upon binding of the asRNA to its complementary target) (58). For example, the asRNA RNAb promotes the premature termination of the operon fatDCBA-angRT in Vibrio anguillarum (59), while the asRNA anti-Q in Enterococcus faecalis is responsible for both transcriptional interference due to RNAPs collisions and attenuation by an uncharacterized mechanism (60). In this work, we show that both a decrease and an increase in AsflhD expression reduce the abundance of flhD mRNA and flhD translation in vivo (Fig. 6A to D) without significantly affecting the stability of the flhD mRNA (Fig. S1). Using an in vitro system, we reveal that AsflhD synthesized in situ or added exogenously can repress the transcription elongation of flhD ( Fig. 9; Fig. S8). Since exogenous AsflhD can repress the transcription of flhD in vitro to a similar extent as when it is synthesized in cis from its endogenous promoter (up to 2-fold repression; Fig. 9B and C; Fig. S8B to C), we propose that AsflhD represses the transcription elongation of flhD mainly via transcriptional attenuation. Furthermore, our experiments do not detect the accumulation of a shorter transcript in vitro upon addition of one or the other of the transcripts, suggesting that binding of AsflhD to flhD does not stabilize a terminator structure but could rather modify the stability of the elongating RNAP, leading to heterogenous 39-termini as observed for AsflhD in vivo by cRT-PCR (Fig. 4A).
The repression of flhD expression is weaker (1.4-fold compared to 4.9-fold) when AsflhD is expressed in trans compared to in cis in the P flhD -lacZ reporter fusion ( Fig. 6D and E and 7D and E). This suggests that AsflhD could also repress flhD by transcriptional interference when expressed from its own promoter. However, we cannot rule out that these variations are due to differences in the stoichiometry between flhD and AsflhD RNAs. In addition, the transcription of AsflhD from the same locus as flhD could lead to the increased local concentration of AsflhD in the vicinity of the nascent flhD transcript, thus enabling AsflhD to interfere with and terminate flhD transcription more efficiently.
Experiments performed in vitro provide evidence for the transcriptional attenuation of flhD expression upon overexpression of AsflhD. However, we also observed that a reduction in AsflhD expression leads to a decrease in flhD expression in vivo. This phenomenon was not observed in vitro, so it is likely that this second positive regulatory mechanism involves other factors, such as the numerous post-transcriptional regulators (RNA-binding chaperones and sRNAs) of flhD expression (35). Close to the translational start of flhD, binding of the McaS sRNA is required to expose the ribosome-binding site and activate translation (61), while the RNA-binding chaperone CsrA protects the 59-end of flhD (45). On the contrary, binding of the sRNAs, OxyS, ArcZ, OmrA, and OmrB represses translation (35). AsflhD binding could interfere with the binding of any of these sRNAs at their sites along the 59-UTR of the flhD mRNA. Remarkably, most post-transcriptional regulatory events on flhD were shown to have weak effects (i.e., often close to 2-fold repression [35] or activation [45,61]). Hence, deciphering the effect of each regulator and its interference with the regulation by AsflhD will be an interesting challenge for future studies.
Outlook. In this work, we have demonstrated the existence of asRNAs complementary to four major regulators of gene expression in E. coli. As we have shown for the asRNA AsflhD, it is likely that they affect both the expression of their direct target and the downstream control of the target's regulon. Regulatory RNAs are far from being fully understood in bacteria, and new mechanisms of action are likely to be discovered. Furthermore, as in the case of AsflhD, asRNAs demonstrate unexpected regulatory functions that raise the question as to how, when, and to what extent asRNAs participate in complex regulatory circuits.

MATERIALS AND METHODS
Bacterial strains and culture conditions. Strains and plasmids used in this work are listed in Table S1. Constructions and mutations were made by using primers given in Table S1 and are described in the supplemental Materials and Methods (Text S1). Strains were grown in LB medium at 37 or 30°C and shifted to 42 or 46°C for the heat-shock experiments, and samples were taken in the mid-log phase (A 600 % 0.4) or as indicated. Strains carrying the pCA24N control and pCA24N AsflhD (containing the 220 first nt of AsflhD relative to its TSS followed by the rrnB T2 terminator) were grown in the presence of chloramphenicol and induced by isopropyl b-D-1-thiogalactopyranoside (10 24 M).
Northern blotting and RNA methods. Total RNA was extracted using the hot-phenol procedure (62). Five mg of total RNA were electrophoresed either on 1% agarose with 1Â Tris-borate-EDTA (TBE) or 6% polyacrylamide gels (19/1) with 7 M urea and 1Â TBE for analysis by northern blotting (63,64) along with a RiboRuler High-Range marker (ThermoFisher) or radiolabeled MspI-digested pBR322 (NEB). The membranes were hybridized with cRNA probes. DNA templates for the synthesis of the RNA probes were obtained by PCR amplification using the pair of "m" and "T7" oligonucleotides (Table S1). The probes were synthesized by T7 RNAP with [a-32 P]UTP yielding uniformly labeled RNAs (65). The membranes were also probed with M1 RNA (or 5S) as loading control by using 59-end-labeled primers (Table S1). DNA templates for in vitro processing and in vitro transcription assay carrying a T7 promoter sequence were generated by PCR using primers Up-T7-flhD 1 308/Down-flhD and Up-T7-AsflhD/Down-AsflhD (Table S1). They allow the transcription of the first 308 nt of flhD and of the first 256 nt of AsflhD, respectively. RNAs were synthesized by T7 RNAP with [a-32 P]UTP as a tracer and were gel purified. Our previously published transcriptomic data set (available in the ArrayExpress database at EMBL-EBI under accession number E-MTAB-9507) (23) was used to compare the transcriptomes of the wild-type (N3433) and the RNase III-deficient strain (IBPC633) and sorted against independently analyzed data sets (described in Table 1) in which asRNAs and TSSs have been identified.
b-Galactosidase assays. Reporter fusions were constructed in the lacZ locus as described in the supplemental Materials and Methods (Text S1). In brief, the P AsphoP -lacZ fusion contains nt 2150 to 115 from the AsphoP TSS, the P AsflhD -lacZ fusion contains nt 2165 to 115 relative to the AsflhD TSS, the P flhD -lacZ fusion contains nt 2108 to 1300 from the flhD TSS, and the fusion P tet -flhD-lacZ contains the P tet promoter sequence followed by nt 11 to 1300 relative to the flhD TSS. The fliC-lacZ fusion was described previously (66). It carries nt 279 to 196 with respect to the fliC TSS. The cultures were initiated at A 600 = 0.02 and sampled at A 600 = 0.4 to 0.5. Samples (100 or 200 mL) were lysed in Z buffer (1 mL total). b-Galactosidase activity was assayed as described (67); the results are the means of at least three biological replicates as indicated in the legends. Since the lacZ mRNA was previously reported to be negatively regulated by RNase III through multiple cleavages within the lacZ mRNA ORF (68, 69), we have not attempted to compare lacZ reporter fusion expression between wt and rnc mutant.
Circular RT-PCR. Circular RT-PCR was performed with total RNA extracted from N3433 and IBPC633 treated with 59-polyphosphatase. After circularization with T4 RNA ligase 1 (Biolabs), mflhD2 was used to prime reverse transcription and mflhD6 and masflhD10 to generate PCR products (Table S1), which were cloned and analyzed as described (70). It should be noted that the efficiency of ligation by the RNA ligase 1 was previously shown to be affected by the presence of secondary structures, which could explain the exclusion of double-stranded RNAs (43).
RNA band-shift assay and in vitro processing by RNase III. AsflhD (256 nt) and flhD (308 nt) RNAs were synthesized as described in the section "Northern blotting and RNA methods." Transcript 59-end labeling, hybridization, RNase III digestion, and sample analysis were performed as described previously (70,71) and are also described in the supplemental Materials and Methods (Text S1).
In vitro transcription assay. Single-round in vitro transcription experiments were carried out on linear templates as described in the supplemental Materials and Methods (Text S1). AsflhD and flhD RNAs added in trans were synthesized as described in the section "Northern blotting and RNA methods." Motility assay. Stationary-phase bacterial cultures (wt, MG1655-B; P AsflhD (22) , ML73; and P AsflhD (11) , ML241), with or without the pCA24N control (Ctl) or the pCA24N AsflhD (As) plasmid, were inoculated (2 mL) on soft agar (0.2 g/liter) Super optimal broth motility plates (containing 2.4 g/liter MgSO 4 and 10 24 M IPTG for strains carrying pCA24N plasmids) at 37°C and pictures were taken using a Gel Doc (Bio-Rad) imager between the beginning and the end of the linear swimming motility period (from 5 to 8 h). Representative images of swimming motility are shown at 6 and 7 h. Super optimal broth was used because bacteria are more motile (presumably due to its lower NaCl concentration, which alleviates OmpR repression of flhD [72,73]), which allows accurate measurements within 12 h. Swimming speed was then calculated by comparing the increase of motility diameters over time.
Data availability. The RNA-seq data set comparing wt and rnc mutant (RNase III inactivation) is available in the ArrayExpress database at EMBL-EBI under accession number E-MTAB-9507.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.5 MB.