The CsrA-FliW network controls polar localization of the dual-function flagellin mRNA in Campylobacter jejuni

The widespread CsrA/RsmA protein regulators repress translation by binding GGA motifs in bacterial mRNAs. CsrA activity is primarily controlled through sequestration by multiple small regulatory RNAs. Here we investigate CsrA activity control in the absence of antagonizing small RNAs by examining the CsrA regulon in the human pathogen Campylobacter jejuni. We use genome-wide co-immunoprecipitation combined with RNA sequencing to show that CsrA primarily binds flagellar mRNAs and identify the major flagellin mRNA (flaA) as the main CsrA target. The flaA mRNA is translationally repressed by CsrA, but it can also titrate CsrA activity. Together with the main C. jejuni CsrA antagonist, the FliW protein, flaA mRNA controls CsrA-mediated post-transcriptional regulation of other flagellar genes. RNA-FISH reveals that flaA mRNA is expressed and localized at the poles of elongating cells. Polar flaA mRNA localization is translation dependent and is post-transcriptionally regulated by the CsrA-FliW network. Overall, our results suggest a role for CsrA-FliW in spatiotemporal control of flagella assembly and localization of a dual-function mRNA.

P ost-transcriptional control involves a complex interplay between mRNAs, small regulatory RNAs (sRNAs) and protein regulators. Although regulatory functions have typically been attributed to proteins or sRNAs, mRNAs have canonically been considered as targets of this regulation. However, regulatory functions have recently also been described for mRNAs that either encode sRNAs in their untranslated regions (UTRs) or act as sponges that sequester other regulatory factors [1][2][3][4] .
The widespread bacterial Csr/Rsm (Carbon storage regulator/ Regulator of secondary metabolism) regulatory network 5 is an ideal model system to study the complex post-transcriptional cross-talk between mRNAs, sRNAs and protein regulators. About 75% of all sequenced bacterial genomes encode a homologue of the central RNA-binding protein (RBP) of this system, CsrA (RsmA/E). CsrA is a pleiotropic regulator of global physiological phenomena in Gammaproteobacteria 5 and considered the most conserved post-transcriptional virulence regulator 6 . CsrA mainly acts by repression of translation initiation via binding to 5 0 regions of mRNAs 7 . The homodimeric CsrA binds GGA-rich motifs that are often located in hairpin loops and/or overlap the Shine-Dalgarno (SD) sequence 5 . In Gammaproteobacteria, CsrA activity is regulated through the CsrB/C and RsmX/Y/Z families of sRNAs 5,7 . These antagonizing sRNAs are often induced by environmental signals 6 and harbour multiple stem-loops with high-affinity GGA motifs that sequester CsrA/RsmA 8 . Despite the presence of CsrA, many bacteria lack homologues of these antagonizing sRNAs. Also, the global CsrA regulon and its general biological function outside the Gammaproteobacteria are unclear. In the Gram-positive Bacillus subtilis, the flagellar assembly protein FliW antagonizes CsrA via direct binding 9 . Although FliW homologues are relatively widespread 9 , proteinmediated regulation of CsrA has not yet been shown outside B. subtilis. Whether FliW can cooperate with RNA-mediated regulation of CsrA is also unknown.
Here we use co-immunoprecipitation (coIP) combined with RNA sequencing 17,18 (RIP-seq) to globally determine the direct RNA-binding partners of C. jejuni CsrA and investigate whether RNA-based regulation of CsrA occurs in the absence of canonical antagonizing sRNAs. Our genome-wide approach reveals many mRNAs of flagellar genes as potential CsrA targets and we demonstrate that flaA mRNA, encoding the major flagellin, has dual (coding and regulatory) function. As the most abundantly co-purified transcript, flaA mRNA is the main target of CsrA translational repression. In addition, the flaA leader can act as an mRNA-derived RNA antagonist of CsrA. Together with the main CsrA antagonist, the FliW protein, flaA mRNA titrates CsrA to regulate expression of other flagellar genes.
In addition, using confocal and super-resolution microscopy imaging, we show that flaA mRNA is expressed in elongating cells and localizes to the cell poles of the amphitrichous C. jejuni. In contrast to eukaryotes 19 , RNA localization is so far only poorly understood in prokaryotes. Bacterial mRNAs can remain localized close to their genomic site of transcription 20 or can migrate to places in the cell where their encoded products are required in a translation-independent manner involving cis-acting signals in the RNA itself 21 . Besides the mechanisms of bacterial RNA localization, even less is known about how this process may be regulated and which, if any, RBPs are involved.
Here we show, based on a variety of C. jejuni mutants that disrupt or maintain flaA translation, that polar flaA mRNA localization requires its translation. Furthermore, we demonstrate that FliW facilitates polar flagellin mRNA localization by antagonizing CsrA-mediated translational repression of flaA. The unexpected role of the CsrA-FliW system in spatial control of flagellin mRNA expression provides new insight into the role of RBPs in bacterial mRNA localization, a process only recently described in prokaryotes.

Results
Global RIP-seq reveals direct CsrA targets in C. jejuni.
To globally identify C. jejuni CsrA targets and any RNA regulators of CsrA activity, we applied a RIP-seq approach 17,18 . The csrA (Cj1103) gene was chromosomally 3xFLAG-tagged at its C-terminus in strains NCTC11168 and 81-176. CsrA-3xFLAG is constitutively expressed during growth in rich medium, and neither introduction of the FLAG-tag nor deletion of csrA affects C. jejuni growth under the examined conditions ( Supplementary  Fig. 1). We performed coIPs on mid-exponential-phase lysates of csrA-3xFLAG strains and, as control, their respective untagged wild-type (WT) strains ( Fig. 1a and Supplementary Fig. 2a). After conversion of co-purified RNAs into cDNA and deep sequencing, 93.2-95.8% of the 4.6-6.2 million sequenced reads for the individual libraries were mapped to the respective genomes (Supplementary Table 1). Most of the NCTC11168 control-coIP library reads mapped to presumably non-specifically pulled-down abundant classes of RNA (rRNA, tRNA and housekeeping RNAs; Fig. 1b and Supplementary Table 2). In contrast, a B36-fold and B5-fold enrichment for reads mapped to 5 0 UTRs or open reading frames (ORFs) of mRNAs, respectively, was observed in the CsrA-3xFLAG coIP library (Fig. 1b). No specific sRNA enrichment was detected. As the coIP of strain 81-176 showed similar enrichment patterns ( Supplementary Fig. 2b), we focused on strain NCTC11168.
C. jejuni CsrA primarily binds flagellar mRNAs. Functional enrichment analysis of the 154 top CsrA targets with 45-fold enrichment in the CsrA-3xFLAG-versus control-coIP (Supplementary Data 1) revealed an overrepresentation of mRNAs from the class 'Surface Structures', including flagellar genes ( Supplementary Fig. 3a,b). In fact, 90% of the reads mapping to the 45-fold-enriched CsrA targets belonged to flagella-or motility-related genes (Fig. 1c). The alternative sigma factors RpoN (s 54 ) and FliA (s 28 ) hierarchically control flagellar expression in Campylobacter 22 . Early genes are expressed from RpoD/s 70 -dependent promoters, whereas class 2 (middle) and class 3 (late) genes are RpoN-and FliA-dependent, respectively 22 . Most of the enriched transcripts belonged to either class 2 or class 3 (Table 1 and Supplementary Fig. 3c). The most abundantly co-purified transcript, with more than 300-fold enrichment, was flaA mRNA, encoding the major flagellin (Fig. 1c).
cDNA peaks reveal CsrA binds in diverse mRNA regions. Visual inspection of the cDNA read-patterns showed that numerous flagellar mRNAs, including flaA, flaG and flgI (encoding the major flagellin, a gene involved in flagellum formation, and a P-ring component, respectively) showed strong enrichment in their 5 0 UTRs ( Fig. 1d and Supplementary Fig. 4a). CsrA binding was also observed between two genes in polycistronic mRNAs, such as the Cj0310c-Cj0309c and Cj0805-dapA operons. Analysis of the potential CsrA-binding sites in an Escherichia coli green fluorescent protein (GFP) reporter-system, ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms11667 originally developed to study sRNA-mediated regulation 23 , revealed all of the tested 5 0 UTR targets (flaA, flaG, flgI, flaB, pseB and Cj1249) were highly upregulated (410-fold) in the absence of E. coli csrA as measured by western blot and FACS analyses ( Fig. 1d and Supplementary Figs 4 and 5). Reduced reporter fusion expression was restored by complementation of DcsrA with C. jejuni CsrA. Using an operon reporter, where the C-terminal part of the upstream gene is fused to FLAG-lacZ' and the N-terminal part of the downstream gene to GFP, we observed that both E. coli and C. jejuni CsrA can repress the downstream genes in polycistrons (Cj0310c-Cj0309c and Cj0805-dapA). Expression of the upstream genes was only slightly affected and they do not contain any strong internal transcriptional start sites that could lead to uncoupled transcription of the downstream genes 14 . As we observed that potential SD sequences right at the 3 0 end of the upstream genes are covered by CsrA target sites, CsrA probably interferes with ribosome binding and translation of the downstream genes and thereby might mediate discoordinate operon regulation.
Automated peak-detection reveals a CsrA-binding motif.
To automatically identify CsrA-binding regions and a binding motif from coIP cDNA enrichment patterns, we developed a peak-detection algorithm based on a sliding window approach (see the Methods for details). This approach predicted 328 potential CsrA-binding sites with 45-fold enrichment in the NCTC11168 coIP (Supplementary Data 2). As a control, peak

WT (control)
CsrA-3xFLAG CsrA-3xFLAG C u lt u r e L y s a t e S u p e r n a t a n t W a s h c o I P C u lt u r e L y s a t e S u p e r n a t a n t  Mapped RNA-seq reads for the control (black) and CsrA-3xFLAG coIP (blue) in strain NCTC11168. Grey arrows: ORFs; black arrows: transcriptional start sites (TSS). Examples of enrichment patterns in 5 0 UTRs (flaA) and between genes in a polycistron (Cj0310c-Cj0309c operon; encoding two paralogous efflux proteins). (Right) Western blot analysis using anti-FLAG and anti-GFP antibodies of reporter fusions to potential C. jejuni CsrA target genes in E. coli DpgaA, DpgaA/DcsrA and DpgaA/DcsrA þ pBAD-csrA Cj (complementation with C. jejuni CsrA-Strep under control of an arabinose-inducible pBAD promoter) strains. Putative CsrA targets from C. jejuni were fused in-frame (for example, 33 aa for flaA) to GFP or a FLAG-lacZ' tag ( Supplementary Fig. 4b). As deletion of csrA dramatically enhanced biofilm formation and led to poor growth in liquid culture in our E. coli strain, reporter experiments were performed in a DpgaA background. GroEL served as loading control. Protein samples corresponding to 0.1 OD 600 were loaded. Quantifications of reporter expression are given below the blots. (e) (Left) CsrA-binding motif predicted by MEME 24 (E-value ¼ 2.1E-11). (Right) Consensus secondary structure motif of C. jejuni CsrA-binding sites predicted by CMfinder 62 . detection was performed in reverse manner by scanning for enriched regions in the control-versus CsrA-3xFLAG-coIP. This analysis revealed only five peaks, without a common motif, indicating a high specificity of the peaks detected in the CsrA-3xFLAG-coIP. MEME 24 analysis of the 328 enriched sequences revealed a (C/A)A(A/T)GGA motif in 324/328 input sequences (Fig. 1e). Analysis of the 81-176 coIP led to a similar motif ( Supplementary Fig. 2c). To check if a similar motif can be found in non-enriched regions, we conducted the peak-detection in reverse manner using a cutoff of only 41-fold enrichment in the control-versus CsrA-3xFLAG-coIP. This revealed 448 'enriched' sites in the control library. Subsequent motif prediction did not yield any significant motifs, further supporting high specificity of the coIP approach. Consensusstructure motif screening of the enriched CsrA-coIP sequences revealed an AAGGA motif in a hairpin-structure loop in 276/328 input sequences (Fig. 1e). These C. jejuni sequence/structural motifs agree with binding sites of other CsrA homologues 25 .
Basal body components (classes 1 and 2) fliE (Cj0526c) flaA mRNA is translationally repressed by CsrA. The flagellar filament, consisting mainly of the FlaA flagellin, is among the last components produced during flagellum assembly. In our coIP, 77% of the reads from 45-fold enriched genes mapped to flaA, indicating it as the main CsrA target (Fig. 1c). Secondary-structure predictions revealed that the 45-nt-long flaA 5 0 UTR can fold into two stem-loops (SL1 and SL2), both of which harbour an ANGGA motif in their loops (Fig. 2a). The second ANGGA motif   Fig. 2b and Supplementary Fig. 8a).
In vitro gel-shift assays using recombinant C. jejuni CsrA-Strep and T7-transcribed, 5 0 -end radiolabelled flaA WT leader showed strong CsrA binding (K d ¼ B50 nM) with two defined shifts, indicating at least two CsrA-binding sites (Fig. 2c). In contrast, flaA leaders with GGA point-mutations in either SL1 (M1 and M2), SL2 (M3) or both SL1 and SL2 (M2/M3) showed four-to tenfold higher K d values (200-500 nM), confirming that the mutations reduced CsrA binding ( Fig. 2d and Supplementary  Fig. 9a). To map CsrA-binding sites on the flaA leader, we performed in-vitro footprinting assays with labelled flaA leader in the absence or presence of CsrA using enzymatic and chemical cleavage (RNase T1; single stranded G-residues and lead(II) acetate; single-stranded RNA). Cleavage patterns without CsrA confirmed the predicted flaA leader structure ( Fig. 2e and Supplementary Fig. 8b). A clear protection was observed at the SL1 and SL2 GGA motifs of the WT leader upon addition of increasing CsrA amounts, but not for a flaA M2/M3 mutant with disrupted binding motifs. The third GGA downstream of the start codon was not protected. Overall, our data suggest C. jejuni CsrA represses flaA translation by high-affinity binding to the two GGA-containing stem-loops SL1 and SL2 in the flaA leader.
The flagellar assembly factor FliW binds CsrA in C. jejuni. The constitutive expression of CsrA during routine culture (Supplementary Fig. 1) suggested modulation of its activity rather than its expression. Because homologues of the CsrB/C sRNAs are absent in C. jejuni, we hypothesized that other RNAs, or even proteins, might control CsrA activity in Campylobacter. One candidate (Cj1075, 129 aa) is a potential homologue of the flagellar assembly factor, FliW, which has a role in motility 26,27 but is otherwise uncharacterized. In B. subtilis, FliW binds CsrA and antagonizes CsrA-mediated translational repression of hag mRNA, encoding the major flagellin 9 . FliW can also bind Hag, which accumulates in the cytoplasm before flagellar hook completion. Hag thus sequesters FliW from CsrA, allowing CsrA to repress Hag synthesis. Upon completion of the hook, Hag is secreted, FliW is released and CsrA repression of flaA translation is relieved. Thus, this Hag-FliW-CsrA partner-switch mechanism ensures appropriate temporal flagellin synthesis. In Epsilonproteobacteria, fliW homologues are present, but, unlike Bacillus, are not encoded adjacent to csrA (Fig. 3a). To investigate whether FliW can interact with CsrA and FlaA in C. jejuni, we performed protein-protein coIP experiments using chromosomal C-terminal 3xFLAG-tag fusions as bait. The anticipated interaction partners were tagged with mCherry at their C-terminus to allow detection by western blotting. In a FliW-3xFLAG-coIP, CsrA-mCherry was successfully co-purified, indicating the two proteins can interact ( Supplementary  Fig. 10). Similarly, FliW-mCherry was co-purified in a FlaA-3xFLAG-coIP, indicating conserved interactions between all three proteins. As control, none of the proteins was co-purified in coIPs with strains that carry the mCherry-fusion proteins but not the FLAG-tagged proteins.
FliW antagonizes CsrA-mediated translational repression. To determine whether the FliW-CsrA interaction could antagonize CsrA function in Epsilonproteobacteria, we used FlaA protein levels as a read-out for CsrA activity (Fig. 3b). Whereas FlaA-3xFLAG was B3-fold upregulated in DcsrA, deletion of fliW led to B6-fold downregulation, consistent with further repression of flaA translation by additional CsrA released upon deletion of its protein antagonist (Fig. 3b). A DcsrA/DfliW double deletion confirmed that the observed downregulation was indeed mediated through CsrA, as FlaA-3xFLAG levels increased back to those in the DcsrA mutant. Despite strong reduction of FlaA-3xFLAG protein levels, a B2-fold higher flaA mRNA level was observed upon deletion of fliW, indicating additional effects of FliW on flaA expression ( Supplementary Fig. 11a). Thus, we constructed a transcriptional reporter composed of the unrelated Cj1321 5 0 UTR and its early coding region (Cj1321_mini) under the control of the flaA promoter. This reporter was, like the endogenous flaA mRNA, B2-fold upregulated in the DfliW mutant ( Supplementary Fig. 11b). As Cj1321 is independent of CsrA-mediated control, FliW seems to have a negative effect (direct or indirect) on flaA transcription.
To uncouple transcriptional control of flaA from its translational regulation, we replaced the s 28 -dependent flaA promoter in the FlaA-3xFLAG strain with a constitutive s 70 -dependent metK promoter. Upon deletion of csrA in this strain, a B3-fold increase in FlaA-3xFLAG level was observed, further confirming post-transcriptional regulation of FlaA-3xFLAG protein expression by CsrA (Fig. 3b). Like for the strain expressing FlaA-3xFLAG from its native promoter, FlaA-3xFLAG expressed from the metK promoter was strongly downregulated upon deletion of fliW and was restored to DcsrA levels in the DcsrA/DfliW double mutant. This further indicates FliW antagonizes CsrA-mediated translational repression of flaA in a promoter-independent manner. In addition, decreased flaA mRNA stability was observed upon fliW deletion in rifampicin stability assays. This is consistent with increased translational repression of flaA in the absence of fliW, despite overall higher steady-state flaA mRNA levels because of FliW-dependent increased transcription ( Supplementary Fig. 11c).
In line with strong downregulation of the FlaA protein upon fliW deletion, transmission electron microscopy revealed shorter flagella on DfliW bacteria compared with those of the WT strain ( Fig. 3c,d). In fact, the flagella of DfliW appeared similar to those of a DflaA mutant strain and of bacteria lacking s 28 (DfliA), required for flaA transcription. In contrast, the DcsrA and DcsrA/ DfliW strains expressed normal flagellar filaments. The short flagella of the DfliW strain are probably composed mainly of the minor flagellin FlaB, which is transcribed from an RpoN (s 54 )-dependent promoter. Upon deletion of both flagellin genes (DflaA/DflaB), the bacteria no longer had filaments but the hook structure was visible at the poles (black arrowheads, Fig. 3c). Furthermore, a DrpoN mutant strain had neither flagella nor hooks. Motility assays revealed that the DcsrA or DfliW strains showed a halo-radius reduction to 78% and 72% of WT, respectively (Fig. 3d). Likely due to its shorter flagella, DfliW also showed slower autoagglutination than WT, but greater than the non-motile DfliA and DrpoN mutants ( Supplementary  Fig. 12). Overall, these data suggest that, besides a mild effect on flaA transcription, FliW affects post-transcriptional control of FlaA, and therefore filament assembly and motility, in a CsrAdependent manner.
Because flaG and flaA are primarily transcribed from s 28 -dependent promoters 28 and are thus expressed at the same time, monitoring FlaG-3xFLAG might reveal the potential role of flaA 5 0 UTR as a CsrA antagonist. The chromosomal M1 flaA leader mutation (GGA -AAA in SL1, Fig. 2a), which leaves the coding region intact but abolishes CsrA binding (Fig. 2d), decreased FlaG-3xFLAG levels B3-fold ( Fig. 5 and Supplementary Fig. 13a). Upon introduction of DcsrA, FlaG-    3xFLAG expression was restored to WT levels, indicating decreased FlaG expression in the flaA-M1 mutant is dependent on CsrA, and suggesting that the flaA leader can also titrate CsrA.
Combining both DfliW and flaA-M1 led to a tenfold reduction in FlaG-3xFLAG levels, showing their cumulative effect in antagonizing CsrA. In line with this, the M1/DfliW/DcsrA triple mutant restored FlaG-3xFLAG levels back to WT levels ( Fig. 5 and Supplementary Fig. 13a). Growth curves showed that there was no major impact on growth of the individual mutations under the examined conditions ( Supplementary Fig. 13b).
Although the DfliW and M1/DfliW mutants showed a slightly increased growth rate compared with WT, this increase was less than a non-motile DfliA strain.
To further confirm the role of the flaA 5 0 UTR as a CsrA antagonist, a B250-nt long flaA_mini transcript comprising the flaA leader and first 17 codons followed by a stable ribosomal rrnB terminator was ectopically expressed from the native flaA promoter (Supplementary Fig. 14a). Expression of the flaA_mini transcript in a DfliW mutant, which has strong CsrA-mediated flaA translational repression, increased FlaA-3xFLAG levels around 2.6-fold ( Supplementary Fig. 14b). This indicates flaA_mini can bind and antagonize CsrA and partially relieve CsrA-mediated repression of flaA translation. A smaller, yet significant, complementation of the effect of a fliW deletion was also observed for FlaG-3xFLAG levels.
Next, the effect of the two antagonists on CsrA-mediated regulation of the RpoN-dependent genes flaB and flgI was evaluated. A similar, yet less pronounced effect compared with FlaG-3xFLAG, was observed for FlaB-3xFLAG upon single or double mutations of fliW and M1. In contrast, FlgI-3xFLAG levels were only significantly reduced upon fliW deletion ( Fig. 5 and Supplementary Fig. 13a). Overall, this reveals FliW as the major CsrA antagonist under the examined growth conditions that titrates, along with the flaA mRNA antagonist, CsrA from lower affinity flagellar targets such as flaG.
flaA mRNA localizes to the poles of elongating cells. As flaA mRNA can titrate CsrA activity, we wondered when flaA mRNA levels change to modulate CsrA activity. Expression of flaA mRNA appeared constitutive during growth ( Supplementary  Fig. 13c). However, in the amphitrichously flagellated C. jejuni, after every cell division, a new flagellum has to be synthesized at the new pole of each daughter cell. As bacteria in batch culture are not synchronized in cell cycle, differences in flaA mRNA expression might be obscured because of the population-based northern analysis. To monitor flaA mRNA expression in single bacteria, we performed RNA-FISH (fluorescence in situ hybridization) in fixed C. jejuni cells from exponential phase. Although the control RNA, 16S rRNA (Fig. 6a,  cells, flaA mRNA (Fig. 6a, red) was detected in only some of the cells. As a negative control, we also performed flaA mRNA FISH on a DfliA mutant strain (Fig. 7a), which showed no expression of flaA ( Supplementary Fig. 11a). Whereas 16S rRNA was equally distributed throughout the cell, flaA mRNA was specifically detected at the cell poles in B20% of WT cells (Fig. 6a,b). Quantification of cell length across the population showed that cells with localized flaA mRNA were significantly shorter than cells without flaA expression (Fig. 6b,c). Live-cell imaging of a non-motile C. jejuni strain (DfliA) over two or three division cycles showed regular patterns of an increase in cell length until cells divide at mid-cell, resulting in short daughter cells ( Supplementary Fig. 16). This indicates shorter cells likely correspond to cells that have divided and are elongating. Together, these data suggest differential expression of flaA mRNA during the cell cycle and accumulation in elongating cells at the required site of its encoded protein.
FliW impacts flaA mRNA localization via CsrA. To investigate whether CsrA-FliW impacts flaA mRNA localization, we next performed RNA-FISH in DfliW, DcsrA and DfliW/DcsrA mutant strains. Although csrA deletion had no effect on flaA localization, it was completely abolished in a DfliW mutant (Fig. 7a). Instead of a polar localization, flaA mRNA was now dispersed throughout the cell. The loss of flaA mRNA localization upon fliW deletion was not due to lower transcript abundance as its mRNA level is increased despite strong repression at the protein level ( Supplementary Fig. 11a). Strikingly, flaA mRNA localization was restored to the cell poles in the DfliW/DcsrA double mutant, showing CsrA affects localization of flaA mRNA. As a further confirmation of flaA mRNA localization, we performed superresolution imaging of flaA mRNA FISH in WT and mutant strains using direct stochastic optical reconstruction microscopy (dSTORM) 29 , which has only recently been applied for bacterial RNA localization 30 . dSTORM analysis fully supported and complemented the observations from confocal microscopy analysis (Fig. 7b and Supplementary Fig. 17). Overall, this suggests a model where flaA translation is required for polar localization: upon deletion of fliW, CsrA is released and in turn strongly represses flaA mRNA translation to impede its localization to the poles.
Polar flaA mRNA localization requires its translation. To support the translation-dependent model of flaA localization, we constructed several point mutants in the native flaA gene that either maintain or disrupt flaA translation (Fig. 8a). Mutation of the start codon of flaA (AUG -AAG (X1) or AUU (X2)) to abolish translation initiation resulted in dispersed flaA mRNA (Fig. 8b). In contrast, when the start codon was changed to an alternative start codon (AUG -GUG (X3)), flaA mRNA still localized to the cell poles, indicating translation of flaA mRNA is indeed required for polar localization. Mutation of the third flaA codon to a stop codon (UUU -UAG (X4)) also resulted in a completely dispersed flaA mRNA signal ( Fig. 8b and Supplementary Fig. 17) required for recruiting flaA mRNA to the cell poles. Overall, these data support a role of the FliW/CsrA post-transcriptional network in controlling translation-dependent polar flaA mRNA localization in C. jejuni.

Discussion
Using genome-wide RIP-seq, we have identified direct RNA targets of the translational regulator CsrA in a bacterium that lacks the canonical antagonizing sRNAs. Our study revealed the major flagellin mRNA is both the main CsrA target and a dualfunction mRNA, which can titrate CsrA activity together with the FliW protein, the main CsrA antagonist (Fig. 9). Compared with microarray-based transcriptome analyses of csrA loss-of-function strains 31,32 , which might reveal indirect effects or miss targets because of a lack of changes in target mRNA levels despite translational repression, a coIP approach facilitates the identification of direct targets and binding sites. Sanger sequencing of cDNAs from an RsmA-coIP identified six target mRNAs in P. aeruginosa 32 . RNA-seq of a CsrA-coIP in E. coli revealed 721 co-purified transcripts 33 , and in vivo ultraviolet crosslinking combined with RNA-seq (CLIP-seq) revealed 467 potential CsrA-binding sites in Salmonella typhimurium, including binding sites in many virulence mRNAs 34 . In our RIP-seq approach, we used untagged WT strains as a negative control to allow for elimination of nonspecifically bound transcripts. Our peak-detection tool confirmed the high specificity of this approach, as it detected an 'ANGGA' sequence in 324/328 targets, which resembles the CsrA consensus-motif determined by in-vitro selection 25 . Besides canonical binding to 5 0 UTRs or early codons 5,35 , our coIP also revealed CsrA binding within coding regions or between genes in polycistrons to mediate discoordinate operon regulation.
Our coIP approach revealed many mRNAs of flagellar genes as direct CsrA targets. The motility defect of DcsrA suggests that tight regulation of flagellar genes by CsrA, and especially of the major flagellin FlaA, is required for proper motility. Balancing CsrA activity through the antagonizing protein FliW also appears crucial for flagellar assembly, as we observed that a C. jejuni NCTC11168 DfliW mutant expresses short flagella, as also reported in other strains 27,36 , and is defective for autoagglutination and motility in both B. subtilis and C. jejuni 9,26 . Although CsrA impacts motility by directly controlling flagellin expression in C. jejuni, B. subtilis and Borrelia, the strong motility defect of an E. coli csrA mutant 37 is due to a requirement of CsrA for stabilization of the mRNA encoding the master regulator FlhDC 38 . The flagellum also plays an essential, multi-factorial role in C. jejuni colonization and pathogenesis, including secretion of Cia/Fed effectors 28,39 , and is required for proper cell division 40 . Future studies might reveal CsrA-affected phenotypes beyond motility.
Instead of CsrA-activity control by antagonizing sRNAs 5 , we demonstrated that the flaA mRNA itself can titrate CsrA. This represents a new mode of CsrA activity control by a target mRNA-derived antagonist. The flaA leader has higher affinity for CsrA compared to other flagellar targets. It has two GGA motifs in adjacent hexaloops, resembling high-affinity CANGGANGcontaining apical hexaloop structures targeted by CsrA/ RsmE 25,41 . The 21-nt spacing between the flaA GGA motifs is close to the 18-nt optimal intersite distance for binding of a CsrA dimer 42 . Whereas flaA mRNA probably only binds one CsrA dimer, multiple RsmE dimers are cooperatively assembled on RsmZ sRNA 8,41 . CsrA titration by a 5 0 UTR has recently been shown to mediate hierarchical control of fimbriae expression in Salmonella typhimurium 43 . The fimAICDHF mRNA leader, which in contrast to flaA mRNA is not itself a CsrA target, cooperates with the CsrB/C sRNAs to antagonize CsrA-mediated activation of plasmid-encoded fimbriae. Small RNAs other than CsrB/C can also sequester CsrA in addition to functioning as antisense RNAs 44 . Global approaches such as RIP-seq are ideally suited to identify additional antagonizing sRNAs or members of the emerging class of dual-function, cross-regulating mRNAs 2,3 .
Analysis of flaA mRNA expression in single bacteria using RNA-FISH showed that this transcript localizes to the poles of shorter, and presumably elongating, cells. As a new flagellum is synthesized after each cell division at the new pole of the amphitrichous C. jejuni, polar flaA mRNA localization might facilitate this process. This temporal and spatial modulation of flaA mRNA expression might also affect CsrA-mediated regulation of other flagellar genes through mediating varying levels of this CsrA RNA antagonist. Mutations that either abolish or maintain translation showed flaA translation is required for its polar localization. Bacterial mRNA localization has only recently been described and unlike eukaryotes the underlying mechanisms and regulation of this process are poorly understood 45,46 . Besides co-translational targeting of mRNAs to the required sites of their encoded products, translation-independent mechanisms of RNA localization have also been described 20,21 , including spatial expression according to chromosome organization. We observed that a flaA mRNA variant with a premature stopcodon mutation at the 101 st codon partially localizes, suggesting a role of the N-terminus in directing the nascent peptide along with the mRNA to the secretion apparatus. Little is known how flagellar substrates are selected for secretion, as they do not share a secretion-signal sequence or cleavable signal peptide. N-terminal domains are required for secretion of flagellar proteins in diverse bacteria, including C. jejuni 36 , and both 5 0 UTR and N-terminal peptide secretion signals have been shown to contribute to secretion efficiency 47 . In addition, flagellar chaperones play a role in regulating the coupling of translation to secretion of flagellar substrates 48 . In Yersinia, cis-encoded RNAlocalization elements in the early coding region are required for secretion of effector proteins by type III secretion systems 49 . Future studies will identify and clarify the role of elements, either in the protein N-terminus or the mRNA 5 0 UTR, as well as potential interaction partners that are crucial for directing the peptide and/or mRNA to the cell poles and secretion apparatus. Besides the requirement of flaA translation for localization, other factors such as the flaA genomic location or the transcriptional complex might also contribute to polar flaA mRNA localization.
Our study revealed an unexpected function for the CsrA-FliW network in spatial and temporal gene-expression control, and specifically FliW affects translation-dependent polar localization of the flagellin mRNA by antagonizing CsrA-mediated translational repression. The limited CsrA activity in WT cells under standard growth conditions, because of sequestration by the FliW protein antagonist, probably allows sufficient translation of flaA mRNA for its polar localization. Strong CsrA-mediated translation repression of flaA upon fliW deletion is probably responsible for the diffuse flaA localization in the DfliW mutant. CsrA binding might mediate storage of translationally inactive flaA mRNA until synthesis of FlaA is required or proper localization is achieved, similar to mRNP granules in eukaryotes 50 . Future studies will show whether other flagellar mRNAs also polarly localize and if the CsrA-FliW regulatory network also impacts their localization. CsrA-mediated regulation of mRNA localization might also occur in B. subtilis and B. burgdorferi, where CsrA overexpression represses the major flagellin 51-53 . An analogous system might have also evolved in the Alphaproteobacterium Caulobacter crescentus, which encodes two proteins with opposing activities on flagellin regulation, FlaF and FlbT, whereby FlbT post-transcriptionally regulates flagellin expression 54 .
Our identification of C. jejuni CsrA titration by FliW indicates that CsrA-activity control by a protein antagonist, a mechanism first identified in the Gram-positive B. subtilis 9 , is more widespread than previously appreciated. Besides the posttranscriptional effect of FliW on flaA and other flagellar genes by antagonizing CsrA, deletion of fliW directly or indirectly increases flaA transcription. Transcription of hag is also twofold upregulated in B. subtilis upon fliW deletion 9,55 . Although FliW appears to be the main CsrA antagonist, its synergistic interplay with the flaA mRNA antagonist affects other flagellar genes showed that RNA-based regulation can also impact CsrA activity in this type of Csr network. Gammaproteobacterial genomes encode CsrA 56 as well as the antagonizing sRNAs 5 and an anticorrelation between the presence of the CsrB/C sRNAs and FliW has been observed 57 . As the csrA gene is located next to a tRNA cluster in E. coli, this strongly suggests the pleiotropic function of CsrA in Gammaproteobacteria might have been horizontally acquired, followed by evolution of the antagonizing sRNAs. Thus, the conserved or possibly more ancient function of the CsrA-FliW system might be to mediate temporal and spatial control of proper flagellum assembly. During our conservation analysis we observed that certain non-flagellated Campylobacter species, such as C. hominis, C. gracilis and C. ureolyticus, lack csrA and fliW homologues, further supporting their conserved function in flagellar regulation. Further studies are required to unravel the full complexity of the CsrA-FliW regulatory network and its impact on RNA localization.

Methods
Bacterial strains, oligonucleotides and plasmids. All C. jejuni and E. coli strains used in this study are listed in Supplementary Table 3   Bacterial growth conditions. C. jejuni strains were routinely grown on Müller-Hinton agar plates or with shaking in Brucella broth (BB), both supplemented with 10 mg ml À 1 vancomycin, at 37°C under microaerobic (10% CO 2 , 5% O 2 ) conditions as described previously 14 . The agar was further supplemented with marker-selective antibiotics (20 mg ml À 1 chloramphenicol, 50 mg ml À 1 kanamycin , 20 mg ml À 1 gentamicin or 250 mg ml À 1 hygromycin B) where appropriate. E. coli strains were grown aerobically at 37°C in Luria-Bertani (LB) medium supplemented with appropriate antibiotics. For induction of arabinoseinducible pBAD promoter, 0.001% ( þ ) or 0.003% ( þ þ ) L-arabinose was added to LB media.
Construction of bacterial mutant strains. All C. jejuni mutant strains (deletion, chromosomal 3xFLAG-tagging, chromosomal point mutations) were constructed using double-crossover homologous recombination. Cloning strategies and the generation of constructs are described in detail in the Methods and Supplementary Methods. Oligonucleotides used to amplify regions of upstream/downstream homology and resistance cassettes for homologous recombination, as well as recipient strains and oligonucleotides for validation of mutant strains by colony PCR, are listed in Supplementary Table 6 for each generated strain. Introduction of PCR products with 500 bp homologous ends or genomic DNA with mutant constructs into C. jejuni was performed by electroporation or natural transformation, respectively, as described previously 14 .
Construction of 3xFLAG epitope-tagged proteins in C. jejuni. C. jejuni genes were chromosomally tagged at their C-terminus either by cloning of constructs for C-terminal epitope tagging on plasmids or by construction of 3xFLAG constructs by overlap PCR.
Tagging of proteins using PCR products amplified from plasmid constructs. The CsrA, FlaA, FlgI and FlaB proteins were fused to a 3xFLAG epitope at their C-termini by cloning regions encoding B500 bp of their C-terminal coding region (C-term) and B500 bp downstream of the stop codon (DN) into plasmid pGG1 to flank a 3xFLAG tag and aphA-3 Kan R cassette. Afterwards, the 3xFLAG-tag constructs were amplified by PCR and introduced into the chromosome of C. jejuni strains by electroporation and double-crossover homologous recombination. An example of this plasmid cloning strategy is described for csrA. Approximately 500 bp of the region downstream of csrA was amplified from genomic DNA (gDNA) with primers CSO-0173/-0174. These primers included XbaI and EcoRI sites, respectively. Following cleanup, the PCR product was digested with EcoRI and XbaI and ligated into a similarly digested pGG1 backbone, generated by inverse PCR with primers CSO-0074/-0075, to create pGD2-1. The plasmid was verified by colony PCR with primers JVO-0054/CSO-0173 and the sequence was verified using JVO-0054. Next, the backbone of this plasmid, including the csrA 'DN' region, was amplified by PCR with primers CSO-0073 (XhoI) and JVO-5142 (blunt). The C-terminal coding region of csrA (B500 bp) without the stop codon was amplified with primers CSO-0171/-0172 from NCTC11168 WT gDNA. The sense primer (CSO-0172) included an XhoI site, whereas the antisense primer (CSO-0171) contained a 5 0 -phosphate. Both the plasmid backbone with the 'DN' insert and the C-term insert were digested with XhoI and ligated to create plasmid pGD4-1. Integration of the PCR product was confirmed by colony PCR using primers CSO-0172/-0023 and the plasmid was validated by sequencing using CSO-0023. The entire integration cassette was then amplified with Phusion High-Fidelity DNA polymerase (NEB) using primers CSO-0172/-0173 and electroporated into C. jejuni and selected on kanamycin plates. Mutants were confirmed by colony PCR with primers CSO-0196/-0023 and western blot analysis with an anti-FLAG antibody. 3xFLAG tagging of proteins by overlap PCR. Construction of a C-terminal 3xFLAG translational fusion at its native locus was performed by overlap PCR for flaG as described in Supplementary Methods for gene deletions, but with the following modifications. The final overlap PCR product contained B500 bp of the C-terminal coding region of flaG minus the stop codon (C-term) and B500 bp downstream of flaG (DN) for homologous recombination. These regions flanked an in-frame 3xFLAG tag and stop codon followed by an aphA-3 Kan R cassette. For example, for tagging flaG, the 3xFLAG tag and Kan R cassette was amplified from plasmid pGG1 with primers JVO-5142 and HPK2. The 'C-term' region of flaG was amplified using primers CSO-1002/-1098, where CSO-1098 is antisense and contains region of complementarity at its 5 0 end to the 3xFLAG tag/JVO-5142, from NCTC11168 gDNA. The 'DN' region was amplified using primers CSO-1099/-1003, where CSO-1099 is sense to flaG DN and contains a region of complementarity to the 3 0 end of the Kan R cassette/primer HPK2. If the coding region of the target gene contained sequences required for expression of a downstream ORF (that is, SD sequence or codons), these sequences were included in the 'DN' amplicon. The three PCR products were then used for overlap PCR with primers CSO-1002/-1003, and the resulting amplicon was electroporated into C. jejuni, followed by selection of positive clones on kanamycin plates. Mutants Introducing chromosomal point mutations into the flaA leader. To introduce point mutations into the 5 0 UTR of flaA at the native locus, a 1,100-bp region around the flaA promoter was amplified using oligos CSO-0752/-0753. These primers introduced XhoI and XbaI sites, respectively, into the resulting PCR product. After XhoI and XbaI digestion, the product was then ligated into a similarly digested plasmid pJV752-1, resulting in plasmid pGD70-5. Plasmid pGD70-5 was checked by colony PCR using primers pZE-A/CSO-0753 and sequencing with pZE-A. Next, plasmid pGD70-5 was amplified by inverse PCR using primers CSO-0754/-0755, thereby introducing NdeI and BamHI restriction sites 40 nt upstream of the flaA transcriptional start site (TSS). An aac(3)-IV gentamicin resistance cassette with its own promoter and terminator was amplified using CSO-0483/-0576 and introduced into PCR-amplified pGD70-5 in the reverse orientation to flaA, just upstream of its promoter, using the NdeI/BamHI restriction sites, resulting in plasmid pGD76-1. Plasmid pGD76-1 was checked by colony PCR using primers CSO-0576/-0753 and sequencing with CSO-0753.
Point mutations were then introduced into the flaA 5 0 UTR by inverse PCR on pGD76-1 using complementary oligos harbouring the desired mutation, followed by DpnI digestion and transformation of the resulting purified PCR product into respectively, resulting in plasmids pGD77-1, pGD93-1, pGD114-2, pGD205-1, pGD204-1, pGD206-1, pGD207-1 and pGD208-1, respectively. For combination of the flaA M2 and M3 mutations, a similar mutagenesis approach was performed based on PCR amplification of the M2 plasmid pGD77-1 using oligonucleotides CSO-1116/-1117, resulting in pGD95-1 harbouring both the mutations. To introduce the flaA 5 0 UTR mutations into C. jejuni, a PCR product covering the homologous ends and the gentamicin resistance cassette was amplified from the respective WT (pGD76-1) or mutant plasmids using CSO-0752/-0850 and electroporated into C. jejuni as described above. To confirm introduction of point mutation in C. jejuni, colony PCR was performed using CSO-0576/-0753 and sequencing with CSO-0850.
Construction of E. coli mutants. The E. coli DpgaA and DpgaA DcsrA deletion strains were constructed in the TOP10 background using the l Red protocol 58 . Briefly, a kanamycin resistance gene, amplified from plasmid pKD4 using primers CSO-0652/-0653, was used to replace the entire pgaA ORF excluding the start and stop codon. The mutant strain was verified by colony PCR using the primer pairs CSO-0654/-0653 and CSO-0652/-0655. After verification, helper plasmid pCP20 containing FLP recombinase was introduced to remove the kanamycin resistance marker 58 . The helper plasmid, which is temperature-sensitive and carries an ampicillin resistance marker, was then cured by recovering colonies at 37°C and confirming ampicillin sensitivity, resulting in strain CSS-0556. Similarly, the ORF of the csrA gene excluding the start and stop codon was then replaced by the kanamycin resistance marker (amplified using CSO-0611/-0612) in the DpgaA strain resulting in strain CSS-0557, harbouring both pgaA and csrA deletions. The csrA deletion was verified by colony PCR using primer pairs CSO-0639/-0612 and CSO-0611/-0640.
RIP-seq of C. jejuni CsrA-3xFLAG. coIP combined with RNA-seq (RIP-seq) to identify direct RNA-binding partners of CsrA-3xFLAG in C. jejuni was performed as previously described 18,59 with minor modifications.
CoIP of RNA with CsrA-3xFLAG. CoIP of chromosomally epitope-tagged C. jejuni CsrA with an anti-FLAG antibody and Protein A-Sepharose beads was performed from lysates of C. jejuni NCTC11168 and 81-176 WT (control) and isogenic csrA-3xFLAG strains grown in 100 ml (50 ml Â 2 flasks) BB containing 10 mg ml À 1 vancomycin to mid-exponential phase (OD 600 ¼ 0.6) at 37°C as described previously for H. pylori 18 . Cells were harvested by centrifugation at 6,000g for 15 min at 4°C. Afterwards, cell pellets were resuspended in 1 ml Buffer A (20 mM Tris-HCl, pH 8.0, 150 mM KCl, 1 mM MgCl 2 , 1 mM dithiothreitol (DTT)) and subsequently centrifuged (3 min, 11,000g, 4°C). The pellets were shock-frozen in liquid nitrogen and stored at À 80°C. Frozen pellets were thawed on ice and resuspended in 0.8 ml Buffer A. An equal volume of glass beads was then added to the cell suspension. Cells were then lysed using a Retsch MM40 ball mill (30 s À 1 , 10 min) in pre-cooled blocks (4°C) and centrifuged for 2 min at 15,200g, 4°C. The supernatant was transferred to a new tube, and an additional 0.4 ml of Buffer A was added to the remaining un-lysed cells with beads. Lysis of the remaining cells was achieved by a second round of lysis at 30 s À 1 for 5 min. Centrifugation was repeated and this second supernatant was combined with the first one. The combined supernatant was centrifuged again for 30 min at 15,200g, 4°C for clarification and the resulting supernatant (lysate fraction) was transferred to a new tube. The lysate was incubated with 35 ml anti-FLAG antibody (Monoclonal ANTI-FLAG M2, Sigma, #F1804) for 30 min at 4°C on a rocker. Next, 75 ml of Protein A-Sepharose (Sigma, #P6649), prewashed with Buffer A, was added and the mixture was rocked for another 30 min at 4°C. After centrifugation at 15,200g for 1 min, the supernatant was removed. Pelleted beads were washed five times with 0.5 ml Buffer A. Finally, 500 ml Buffer A was added to the beads and RNA and proteins were separated by phenol-chloroform-isoamyl alcohol extraction and precipitated as described previously 18 . From each coIP, 700-1,000 ng of RNA was recovered. 100 ml of 1 Â protein loading buffer (62.5 mM Tris-HCl, pH 6.8, 100 mM DTT, 10% (v/v) glycerol, 2% (w/v) SDS, 0.01% (w/v) bromophenol blue) was added to the final protein sample precipitated along with beads. This sample was termed the coIP sample. For verification of a successful coIP, protein samples equivalent to 1.0 OD 600 of cells were obtained during different stages of the coIP (culture, lysate, supernatant, wash and coIP (beads)) for further western blot analysis. One hundred microlitres of 1 Â protein loading buffer was added to the protein samples and boiled for 8 min. Protein sample corresponding to an OD 600 of 0.1 or 0.15 (culture, lysate, supernatant and wash fraction) and 10 or 5 (for proteins precipitated from beads) were used for western blot analysis.
RIP-Seq cDNA library preparation. Residual gDNA was removed from the coIP RNA samples isolated from the control (WT) and CsrA-3xFLAG coIPs of the two strains C. jejuni NCTC11168 and 81-176 using DNase I treatment. cDNA libraries for Illumina sequencing were constructed by vertis Biotechnologie AG (http://www.vertis-biotech.com) in a strand-specific manner as described previously 14 . In brief, equal amounts of RNA samples were poly(A)-tailed using poly(A) polymerase. Then, 5 0 -triphosphates were removed using tobacco acid pyrophosphatase, and an RNA adapter was then ligated to the resulting 5 0 -monophosphate. First-strand cDNA was synthesized with an oligo(dT)-adapter primer using M-MLV reverse transcriptase. In a PCR-based amplification step, using a high-fidelity DNA polymerase, the cDNA concentration was increased to 20-30 ng ml À 1 . For all libraries, the Agencourt AMPure XP kit (Beckman Coulter Genomics) was used to purify the DNA, which was subsequently analysed by capillary electrophoresis.
A library-specific barcode for multiplex sequencing was included as part of a 3 0 -sequencing adapter. The following adapter sequences flank the cDNA inserts: The samples were sequenced on an Illumina HiSeq instrument with 100 cycles in single-read mode. The resulting read numbers are listed in Supplementary  Table 1.
Analysis of deep sequencing data. To assure high sequence quality, the Illumina reads in FASTQ format were trimmed with a cutoff phred score of 20 by the programme fastq_quality_trimmer from FASTX toolkit version 0.0.13. After trimming, poly(A)-tail sequences were removed and a size filtering step was applied in which sequences shorter than 12 nt were eliminated. The collections of remaining reads were mapped to the C. jejuni NCTC11168 (NCBI Acc.  Table 1. Coverage plots representing the numbers of mapped reads per nucleotide were generated. Reads that mapped to multiple locations contributed a fraction to the coverage value. For example, reads mapping to three positions contributed only one-third to the coverage values. Each graph was normalized to the number of reads that could be mapped from the respective library. To restore the original data range, each graph was then multiplied by the minimum number of mapped reads calculated over all libraries. The overlap of sequenced cDNA reads to annotations was assessed for each library by counting all reads overlapping selected annotations on the sense strand. These annotations consist of strain-specific NCBI gene annotations complemented with annotations of previously determined 5 0 UTRs and small RNAs 14 . Each read with a minimum overlap of 10 nt was counted with a value based on the number of locations where the read was mapped. If the read overlapped more than one annotation, the value was divided by the number of regions and counted separately for each region (for example, one-third for a read mapped to three locations).
Enrichment analysis of CsrA targets. Enrichment of transcripts in the CsrA-3xFLAG coIP versus control coIP libraries was determined based on mapped cDNA read counts for annotations provided in NC_002163.gff (NCBI) for NCTC11168 using GFOLD version 1.0.9 (ref. 61) but with manually defined normalization constants based on the number of reads that could be mapped to the respective libraries. For determination of genes enriched in the CsrA-3xFLAGtagged library, log2 fold changes (FCs) rather than GFOLD values were used. Similar analysis was done for strains 81-176 using annotations provided in NC_008787.gff (chromosome), NC_008770.gff (pVir plasmid) and NC_008790.gff (pTet plasmid).
Peak detection and CsrA-binding motif analyses. To automatically define CsrA-bound RNA regions or peaks from the CsrA-3xFLAG coIP data sets, an in-house tool 'sliding_window_peak_calling_script' was developed based on a sliding window approach. A detailed description of the tool will be described elsewhere. The script has been deposited at Zenodo (https://zenodo.org/record/ 49292) under DOI 10.5281/zenodo.49292 (http://dx.doi.org/10.5281/ zenodo.49292). The script is written in Python 3 and requires installation of the Python 3 packages numpy and scipy for execution.
In brief, the 'sliding_window_peak_calling_script' software uses normalized wiggle files of the CsrA-3xFLAG and control coIP libraries as input to determine sites showing a continuous enrichment of the CsrA-3xFLAG-tagged library compared with the control. The identification of enriched regions is based on four parameters: a minimum required fold change (FC) for the enrichment, a factor multiplied by the 90th percentile of the wiggle graph, which reflects the minimum required expression (MRE) in the tagged library, a window size in nt (WS), for which the previous two values are calculated in a sliding window approach, and a nucleotide step size (SS), which defines the steps in which the window is moved along the genomic axis. All consecutive windows that fulfill the enrichment requirements are assembled into a single peak region. The peak detection is performed separately for the forward and reverse strand of each replicon. For the CsrA-3xFLAG coIP data set, the following parameters were used: FC ¼ 5, MRE ¼ 3, WS ¼ 25 and SS ¼ 5.
For the prediction of consensus motifs based on the peak sequences, MEME 24 and CMfinder 0. Functional classes enrichment analysis. To check for overrepresentation of functional classes of CsrA-bound genes, we considered genes with at least fivefold enrichment in their 5 0 UTR and/or coding sequence in the CsrA-3xFLAG coIP library (versus control) as CsrA-bound and the remaining genes as unbound. We applied an existing functional classification 64 of genes from strain NCTC11168 to determine statistically enriched functional classes. Because a similar classification was not available for strain 81-176, a table with orthologue mappings between the two strains was downloaded from OrtholugeDB 65 and used to assign the NCTC11168 functional classes to their respective 81-176 counterparts. Genes in our annotation lists without an existing functional classification in NCTC11168 or without an orthologue match were assigned to class 5.I, defined as 'Unknown', in the original classification scheme. Genes encoded on the pVir and pTet plasmids of strain 81-176 were assigned to new pVir and pTet classes, respectively. Functional overrepresentation was analysed for each functional class via a two-sided Fisher's exact test followed by multiple-testing correction using the Benjamini-Hochberg method. An adjusted P-value of 0.05 was selected as significance threshold for functional overrepresentation.
Protein-protein coIP. The FliW and CsrA protein-protein coIP was performed exactly as described for the RIP-seq coIP protocol (see above) until the step where beads were washed five times with Buffer A. After washing, the beads were suspended in 200 ml of 1 Â protein loading buffer (62.5 mM Tris-HCl, pH 6.8, 100 mM DTT, 10% (v/v) glycerol, 2% (w/v) SDS, 0.01% (w/v) bromophenol blue) and boiled for 8 min. Lysate samples corresponding to an OD 600 of 0.05 and 2 (for proteins precipitated from beads) were used for western blot analysis.
Validation of CsrA targets with a GFP reporter system. Validation of CsrA targets was performed using a heterologous E. coli system previously developed for validation of sRNA-mRNA interactions 23 . Selected candidate C. jejuni CsrA target sequences from the coIP were cloned as translational fusions to GFP or FLAG in plasmids pXG-10 or pXG-30 as listed in Supplementary Tables 5 and 7. Levels of FLAG or GFP translational fusions were then determined by western blotting or FACS in E. coli DpgaA, DpgaA DcsrA and a DpgaA DcsrA strain harbouring plasmid pGD72-3 with C. jejuni CsrA-Strep under the control of an arabinoseinducible promoter.
Flow cytometric analysis. For FACS analysis of GFP reporter fluorescence in E. coli, cells corresponding to 1 OD 600 were collected from LB cultures in log phase and resuspended in 0.25 ml PBS. Cells were then fixed for 10 min with 0.25 ml of 4% paraformaldehyde, collected by centrifugation and washed twice with 0.5 ml PBS before final resuspension in 0.5 ml PBS. A 1/100 dilution of the fixed sample in PBS was used for measurement. Measurements (50 000 counts per sample) were performed on a BD FACSCalibur machine and analysed using FlowJo (V10).
Purification of C. jejuni CsrA. Recombinant, C-terminal Strep-tagged C. jejuni CsrA (Cj1103) was overexpressed and purified from E. coli TOP10 DpgaA/DcsrA using Strep-Tactin Sepharose (IBA GmbH, #2-1202-001). Primers and plasmids used for cloning are listed in Supplementary Tables 4 and 5 The csrA gene, including its SD sequence, was fused to a C-terminal Strep-tag in the arabinoseinducible plasmid pBAD/Myc-His A (Invitrogen) for overexpression and affinity purification. The csrA-coding region and SD were amplified from C. jejuni NCTC11168 genomic DNA using primers CSO-0746/-0747, and the pBAD/Myc-His A plasmid was amplified by inverse PCR with JVO-0900/-0901 as previously described 66 . CSO-0747 and JVO-0901 introduce an XbaI site to the insert and vector, respectively, whereas CSO-0746 has a 5 0 -phosphate to facilitate blunt-end ligation. XbaI-digested insert and vector were then ligated, resulting in pGD68-1. Plasmid pGD68-1 was checked by colony PCR using primers pBAD-FW/CSO-0747 and sequencing with pBAD-FW. A Strep-tag (WSHPQFEK) was then added at the C-terminus of csrA by inverse PCR using oligonucleotides CSO-0852/-0853, resulting in plasmid pGD72-3. Plasmid pGD72-3 was checked by sequencing with pBAD-FW. Plasmid pGD72-3 was then introduced into an E. coli TOP10 DpgaA/ DcsrA deletion strain resulting in strain CSS-0931. CSS-0931 was grown in 500 ml LB broth with 100 mg ml À 1 of ampicillin at 37°C and shaking at 220 r.p.m. to an OD 600 of 0.3, at which time L-arabinose was added to a final concentration of 0.01%. The culture was then incubated for an additional 8 h at 18°C. Cells were harvested by centrifugation at 7,000g for 30 min at 4°C. The pellet was resuspended in 5 ml of Buffer W (IBA GmbH, #2-1003-100). The rest of the protocol was followed as per the manufacturer's instructions using 1 ml Gravity flow Strep-Tactin Sepharose. After washing steps, the CsrA-Strep protein was finally eluted using Buffer E (IBA GmbH, #2-1000-025) in three successive steps (E1: 0.8 ml, E2: 1.4 ml and E3: 0.8 ml). The majority of CsrA-Strep was concentrated in the E2 fraction. Concentration was quantified using Roti-Quant (Carl ROTH, #K015.3), and the protein was stored at À 20°C in 50 ml aliquots.
RNA isolation. Bacteria were grown to the indicated growth phase and culture volume corresponding to a total amount of 4 OD 600 was harvested and mixed with 0.2 volumes of stop-mix (95% ethanol and 5% phenol, vol/vol). The samples were snap-frozen in liquid nitrogen and stored at À 80°C until RNA extraction. Frozen samples were thawed on ice and centrifuged at 4°C to collect cell pellets. Cell pellets were lysed by resuspension in 600 ml of a solution containing 0.5 mg ml À 1 lysozyme in TE buffer (pH 8.0) and 60 ml of 10% SDS. The samples were incubated for 1-2 min at 65°C to ensure lysis. Afterwards, total RNA was extracted using the hot-phenol method as described previously 13,14 .
Northern blot analysis. For northern blot analysis, 5-10 mg RNA sample was loaded per lane. After separation on 6% PAA gels containing 7 M urea, RNA was transferred to Hybond-XL membranes (GE-Healthcare) by electroblotting. After blotting, the RNA was ultraviolet cross-linked to the membrane and hybridized with g32P-ATP end-labelled DNA oligonucleotides (Supplementary Table 4).
Rifampicin RNA stability assays. To determine the stability of flaA mRNA in C. jejuni NCTC11168 WT, DcsrA, DfliW and DcsrA DfliW strains, cells were grown to an OD 600 of 0.45 (mid-log phase) and treated with rifampicin to a final concentration 500 mg ml À 1 . Samples were harvested for RNA isolation at indicated time points following rifampicin addition (0, 4, 8, 16 and 32 min) as described above. After RNA isolation, 10 mg of each RNA sample was used for northern blot analysis as detailed above.
In-vitro T7 transcription and RNA labelling. DNA templates containing the T7 promoter sequence were generated by PCR using oligos and DNA templates listed in Supplementary Table 8. T7 in-vitro transcription of RNAs was carried out using the MEGAscript T7 kit (Ambion) and sequences of the resulting T7 transcripts are listed in Supplementary Table 8. In vitro transcribed RNAs were quality checked and 5 0 end-labelled (g 32 P) as previously described 66,67 .
In vitro structure probing assays. In vitro structure probing of flaA WT and flaA M1/M2 leaders with RNase T1 and lead(II) acetate was performed as previously described 68 . For each reaction, 0.1 pmol of a labelled flaA leader variant was denatured for 1 min at 95°C and chilled on ice for 5 min. One microgram yeast tRNA as competitor and 10 Â RNA Structure Buffer was added (provided together with RNase T1, Ambion). Unlabelled recombinant C. jejuni CsrA protein was then added at 0-, 20-, 50-or 100-fold molar excess. After incubation for 15 min at 37°C, 2 ml RNase T1 (0.01 U ml À 1 ) or 2 ml freshly prepared lead(II)-acetate solution (25 mM) were added and reactions were incubated for 3 min or 90 s, respectively. As a control, B0.1 pmol labelled RNA with 100-fold excess CsrA was also prepared without nuclease/lead(II) treatment. The reactions were stopped by addition of 12 ml Gel loading buffer II (#AM8546G, Ambion). For RNase T1 ladders, B0.1 pmol labelled RNA was denatured in 1 Â Structure Buffer for 1 min at 95°C and afterwards incubated with 0.1 U ml À 1 RNase T1 for 5 min. The OH ladder was generated by incubation of B0.1 pmol labelled flaA WT leader RNA in 1 Â alkaline hydrolysis buffer (Ambion) for 5 min at 95°C. Ladders and samples were then separated on 10% (v/v) PAA/7M urea gels in 1 Â TBE buffer. Gels were dried, exposed to a screen and analysed using a PhosphorImager (FLA-3000 Series, Fuji).
Transmission electron microscopy. C. jejuni WT and mutant strains were grown for 14 h on MH plates supplemented with vancomycin (10 mg ml À 1 ). Cells were resuspended gently in PBS using a cotton swab and centrifuged at 5,000g for 5 min. The cell pellet was resuspended in 2% glutaraldehyde in 0.1 M cacodylate and incubated at 4°C overnight. The next day, samples were stained with 2% uranyl acetate and imaged using a Zeiss EM10 transmission electron microscope.
Motility assays. C. jejuni strains were inoculated from the appropriate selective MH agar plates into 20 ml BB containing 10 mg ml À 1 vancomycin and grown microaerobically with shaking at 37°C to an OD 600 of B0.5. Cells were harvested by centrifugation at 6,500g for 5 min and resuspended at an OD 600 of 0.5 in BB. For each strain, 0.5 ml of bacterial suspension was inoculated into motility soft-agar plates (MH broth þ 0.4% agar) poured the day before. Plates were incubated rightside-up for B24 h microaerobically at 37°C. Three measurements of each motility halo were made for each inoculation, which were averaged to give the mean swim distance for each strain on a plate. All strains were inoculated together on six replicate plates and the mean swim distance ± standard error on these plates was used to the compare motility of each strain.
Autoagglutination assay. Autoagglutination was determined as described previously 26 . Briefly, strains grown in liquid cultures for motility assays were resuspended in PBS, pH 7.4, to an OD 600 of 1.0. Two millilitres were placed into three replicate tubes and the OD 600 was measured. Tubes were incubated at 37°C microaerobically without shaking, and at indicated time points, 100 ml was carefully removed from the top of the suspension, diluted tenfold in PBS, and the OD 600 was measured. Measurements were normalized to the optical density of each strain at the zero time point.
Time-lapse microscopy to monitor cell division. C. jejuni DfliA mutant cells corresponding to an OD 600 of 0.5 were collected from BB culture in log phase by centrifugation and resuspended in 0.5 ml BB. The cells were further serially diluted 100-and 1,000-fold in BB. Five microlitres of the diluted samples were spotted on a BB-agarose (1%) plate. The plate was incubated under microaerophilic conditions at 37°C for 10 min. The agarose patch was excised and inverted onto a Petri dish with a glass bottom. Single cells were then monitored over time using several bright-field images in a fluorescence microscope (Leica DMI6000 B) maintained at 37°C under aerobic conditions. RNA FISH. RNA-FISH was performed as previously described 69 with some modifications. A total amount of cells corresponding to two OD 600 was collected from BB cultures in mid-log phase (OD 600 ¼ 0.4) and resuspended in 0.5 ml PBS. Cells were then fixed for 3 h with 0.5 ml 4% paraformaldehyde at room temperature, collected by centrifugation and washed twice with 0.5 ml PBS before final resuspension in 0.5 ml 70% ethanol. After 10 min, cells were collected by centrifugation and resuspended in 95% ethanol and incubated at room temperature for 1 h. Cells were again collected by centrifugation, completely dried in a laminar flow hood and then washed once with 2 Â SSC before final resuspension in 0.5 ml of 2 Â SSC containing 10% formamide. Fluorescently labelled DNA oligos (14 Cy5-labelled oligos to detect flaA mRNA and one FITC-labelled oligo specific for 16 S rRNA, Sigma, Supplementary Table 4) were then added at a concentration of 10 ng ml À 1 and incubated at 37°C overnight. The next day, cells were collected by centrifugation and washed three times for 1 h at 37°C with 0.5 ml of 2 Â SSC containing 10% formamide before final resuspension in 2 Â SSC (50-250 ml). Cells were then imaged in a Leica Confocal TCS SP5 II microscope using sequential scanning mode.
dSTORM was performed on a wide-field setup for localization microscopy 71 . An optically pumped semiconductor laser (Genesis MX STM-Series, Coherent) with a wavelength of 639 nm (maximum power of 1 W) was used for excitation of Cy5 and a diode laser (iBeam smart Family, TOPTICA Photonics) with a wavelength of 405 nm (maximum power of 120 mW) was used for reactivation of Cy5. Laser beams were cleaned-up by bandpass filters (Semrock/Chroma) and combined by appropriate dichroic mirrors (LaserMUX filters, Semrock). Afterwards they were focused onto the back focal plane of the high numerical oil-immersion objective (Olympus APON 60XO TIRF, numerical aperture 1.49), which is part of an inverted fluorescence microscope (Olympus IX71). To separate the excitation light from the fluorescence light, suitable dichroic beam splitters (Semrock) were placed into the light path before the laser beams enter the objective. Fluorescence light collected by the objective was filtered by appropriate detection filters (Semrock/Chroma) and was detected by an EMCCD camera with 512 Â 512 pixels (iXon Ultra 897, Andor Technology). The pixel size in the image was 129 nm px À 1 . Cy5 was excited with the 639-nm laser at a maximum intensity of 4.19 kW cm À 2 . During imaging, the 405-nm laser was switched on to keep up a suitable switching ratio. Its laser power was increased successively to a maximum intensity of 0.04 kW cm À 2 . For every image, 5,000-25,000 frames were taken with an integration time of 15 ms per frame. For every imaged area, additionally a bright-field image was taken to identify single bacteria. Data analysis was performed using rapidSTORM open source software 72 .
Statistical analysis. All data for western, northern blot or FISH analysis are presented as mean±s.e.m. Statistical analysis was carried out using Student's t-test. For statistical comparison of two groups, a two-tailed paired Student's t-test was used. A value of Po0.05 was considered significant and marked with an asterisk (*) as explained in the legends. For FISH analysis, fluorescence data curves from 10 cells from a single image were merged as a single averaged curve after cell length normalization. The data were acquired and normalized over cell length using ImageJ and subsequently the merged average curve was generated using Microsoft Excel.
Data availability. The raw, de-multiplexed reads as well as coverage files of the RIP-seq libraries have been deposited in the NCBI Gene Expression Omnibus 73 under the accession number GSE58419. The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files, or from the corresponding author upon request.