The study was carried out in E. coli K-12 MAS1081 (MG1655 rph⁺ gatC⁺ glpR⁺). All strains used in the study are listed in Supplementary Table 1. Unless otherwise noted, all cultures were grown in MOPS minimal medium (Neidhardt et al., 1974) at 37°C shaking at 160 rpm and were grown exponentially for at least ten generations before start of the experiment to obtain balanced growth. Antibiotics were added as described for each experiment.

Culture aliquots were harvested into 1/4 vol ice-cold stop solution (95% ethanol, 5% phenol) (Bernstein et al., 2002). Subsequently RNA was purified using hot phenol and flash freezing in liquid nitrogen as in Fessler et al. (2020). Briefly: Stopped culture aliquots were centrifuged 2 min at 20.000 g and resuspended in 0.1 vol cold 0.3 M sucrose, 0.01 M NaOAc pH 4.5 followed by addition of 0.1 vol 2% SDS 0.01 M NaOAc pH 4.5. Phenol (saturated with water) was added to the liquid phase at a 1:1 ratio, the tubes were vortexed and incubated 3 min at 65°C. After freezing 15 sec in liquid N₂ and centrifugation at 20.000 g for 5 min, the water phase was transferred to new tubes and the phenol extraction was repeated. If the RNA was used in an enzymatic reaction after purification, a chloroform extraction step was included. The RNA was ethanol-precipitated, washed by 96% ethanol, air dried at room temperature and dissolved in 10 mM NaOHAc, 1 mM EDTA. For northern blots, RNA was mixed 1:1 with loading buffer (0.1 M NaOAc (pH 5.0), 8 M urea, 0.05% (w/v) bromophenol blue and 0.05% (w/v) xylene cyanol) and size-separated on denaturing 0.4 mm thick polyacrylamide gels using 1 × TBE buffer (90 mM Tris, 90 mM boric acid and 2 mM EDTA). The RNA was electroblotted onto Hybond-N membranes (GE Healthcare) (1.5 V/cm, 1.5 h) in 40 mM Tris-acetate (pH 8.1), 2 mM EDTA. After UV-crosslinking (0.12 J/cm²) the membranes were pre-hybridized (1 h, rotating at 42°C) in hybridization solution (0.9 M NaCl, 0.05 M NaH₂PO₄ (pH7.7), 5 mM EDTA, 5 × Denhardt’s solution (0.1% BSA, 0.1% Ficoll 400, 0.1% polyvinylpyrrolidone), 0.5% (w/v) SDS and 100 mg/ml sheared, denatured salmon sperm DNA). Probe hybridization was done by adding 30 pmol of oligo-DNA, 5′-end labeled with ³²P (overnight, rotating at 42°C). Subsequently, membranes were washed several times with 0.3 M NaCl, 30 mM sodium citrate, 0.1% SDS at 42°C. Radioactive signals were quantified on a PhosphorImager (Typhoon-GE Healthcare) using ImageQuant software as previously described (Sørensen, 2001; Stenum et al., 2017) and in case of very low signals (e.g., Figure 2C) the signal found in the estimated position of a band was used. Before re-probing, membranes were stripped by washing several times with 98°C, 15 mM NaCl, 1.5 mM sodium citrate, 0.1% SDS, until no more radioactive signal could be detected by a Geiger-Müller tube. Probe sequences used in this study are listed in Supplementary Table 4.

RNAs that bind Hfq were isolated using an E. coli strain, where the hfq allele was tagged with a biotinylation sequence (Hfq_bio). We chose this tag as it is not positively charged, to reduce the risk that the tag might unspecifically bind negatively charged RNA. Hfq_bio was biotinylated by the biotin ligase BirA, and biotin’s high-affinity binding to avidin was used for purifying Hfq_bio (Kay et al., 2009). To ensure full biotinylation of Hfq_bio we introduced the plasmid pBirA where birA is under the control of an IPTG-inducible promoter. We found that production of sufficient BirA was achieved without induction of transcription by IPTG.

Wildtype and Hfq_bio cells both containing the pBirA plasmid were grown exponentially in MOPS medium supplemented with; 15 μg/ml chloramphenicol, 0.2% glucose, 10 μg/ml uracil, 50 μM biotin, at 37°C for at least 10 generations. At OD₄₃₆ = 0.8 the cells were pelleted and washed in medium without biotin, re-suspended in 2 ml lysis buffer (50 mM Tris-HCl pH 7.5, 50 mM NaCl, 5% glycerol), lysed by sonication and centrifuged (20,000 g, 60 min). Total RNA was prepared from an aliquot of the cleared lysate, and the remaining lysate was transferred to fresh tubes containing 300 μl equilibrated SoftLinkTM Avidin Resin (Promega) and left overnight rotating at 4°C. Then, the resin was washed four times in lysis buffer as above and RNA was harvested by phenol extraction and ethanol precipitation. For total RNA and the Hfq-bound RNA, RNA corresponding to 0.05 and 10% of the total culture volume respectively (approximately 2,000 ng for Hfq-bound RNA from the Hfq_Bio strain and 700 ng for total RNA samples), was used for analysis by northern blotting.

Briefly, a [γ-³²P]-ATP end-labeled DNA oligo antisense to the TLR area of interest was hybridized to total RNA and single-stranded overhangs were removed by addition of S1 nuclease. The resulting fragments were visualized on a denaturing polyacrylamide gel. If the synthetic oligo extends beyond the end of the target RNA, the position of the end can be determined by the number of nucleotides removed from the DNA oligo. S1 nuclease was used to map the RrB transcript of a strain harboring the plasmid pTSS1. RNA was harvested by hot phenol 1 h after IPTG induction. One pmol of 5′-end [³²P] -labeled probe was hybridized to 30 μg of total RNA from the strain of interest. Hybridization was done in 50% formamide, 20 mM HEPES, 0.5 mM EDTA, 0.2 M NaCl, 0.05% (w/v) SDS and performed overnight in a thermocycler (68°C for 10 min, then the temperature was lowered to 54°C and decreased 1°C every 30 min until reaching 20°C). Digestion was performed by adding 300 μl 0.28 M NaCl, 50 mM NaOAc pH 4.6, 4.5 mM ZnSO₄ along with 300 U/ml S1 nuclease (Thermo Fisher Scientific) and incubating at RT for 60 min. Samples were phenol/CHCl₃ extracted, ethanol precipitated, size separated by electrophoresis on 7 M urea, 10% polyacrylamide sequencing gels and detected by autoradiography.

Circular RACE was used to map the isoform of RrB enriched on Hfq. Briefly, RNA purified from the Hfq_bio purification experiment was circularized using RNA ligase, reverse transcribed using random hexamer primers, PCR amplified twice using nested sets of RrA, B, F-specific primers, and subjected to deep sequencing. The site of circularization thus reveals both ends of the transcript. Mapping was carried out essentially as described by McGrath (2011) omitting the TAP treatment: 500 ng of RNA co-purified with Hfq was circularized in 1× buffer by adding T4 RNA ligase. Reverse transcription was carried out using Super Script III RT (Thermo Fischer) and primed by random hexamer oligos. The area of interest was amplified twice by PCR with two different sets of specific primers (cRACE rrB-2 1F + cRACE rrB-2 1R and cRACE rrB-2 2F + cRACE rrB-2 2R, see Supplementary Table 3). The PCR-library was sequenced on an Illumina Mi-seq by 300 bp paired-end sequencing. The resulting sequences were merged and subsequently listed by abundance. All reads that could not be merged were left out of the analysis.

Cultures were grown exponentially at 37°C, shaking at 160 rpm for at least 10 generations in MOPS medium supplemented with 0.2% glucose and 10 μg/ml uracil. At an OD₄₃₆ of 0.7, 3 × 15 ml culture aliquots were collected (time 0 min) and rifampicin was added to the remaining culture to a final concentration of 100 μg/ml. Aliquots of 15 ml were collected at 2.5, 5, 10, and 20 min post rifampicin treatment. Five percent spike-in culture overexpressing tRNA^selC was added to each sample aliquot as in Stenum et al. (2017). RNA was harvested using TRI-reagent (Sigma), as described by the manufacturer. The RNA was size-separated on polyacrylamide gels, northern blotted and probed as described above. RrB^short transcript levels were normalized to the tRNA^selC level for each sample.

The structure of RrB was investigated in vitro with and without Hfq.

Reactions (10 μl) containing 0.1 pmol of 5′-end [γ-³²P] ATP-labeled transcript and 50 nM unlabeled E. coli tRNA were incubated with different concentrations of hexameric Hfq (Hfq₆) (3, 1.5, and 0 μM) at 37°C for 100 min along with the relevant cleavage buffer. For Pb²⁺ cleavage: 1 × Structural Probing Buffer (Ambion AM2237), Pb²⁺ was added to a final concentration of 10 mM and samples incubated 1 min at 37°C. Control (C1) was without Hfq and Pb²⁺. RNaseIII cleavage: 1 × Short Cut MnCl₂ buffer (New England Biolabs) and 0.002 units Short Cut RNaseIII (New England Biolabs), samples were incubated 20 min at 37°C. Control (C2) was without Hfq and RNaseIII. Control T1: 1 × Structural Probing Buffer (Ambion AM2237), sample was incubated at 95°C for 1 min, transferred to 37°C for 1 min. After addition of 0.05 U RNase T1 (Ambion AM2237) the sample was incubated at 37°C for 5 min. OH ladder: 1 × Alkaline Hydrolysis Buffer (Ambion AM2237), sample was incubated at 95°C for 5 min. All samples were cooled by addition of 200 μl ice-cold H₂O and transferred to ice. The digested RNA was phenol extracted, ethanol precipitated, resuspended in 1 × loading buffer II (Ambion AM2237), and separated on an 8% polyacrylamide/urea/TBE gel. Radioactive signal from the dried gel was visualized on a PhosphorImager (Typhoon -GE Healthcare). Hfq protein used for all in vitro experiments was a kind gift from Anders Boysen.

Affinity purification of MS2-tagged RNAs was done either with in vivo expressed RNA or in vitro transcripts that were added to the cell lysate. The 115 bp RrB^short sequence, mapped as the most abundant variant pulled down with Hfq (Figure 3) was cloned into the plasmid pNS21 using PCR amplifying RrB^short with either NheI or XbaI restriction sites on the ends. Purified PCR fragments were restriction digested with either NheI or XbaI and ligated into the pNS21 plasmid cut with the corresponding restriction enzyme, resulting in RrB^short 3′-fused to MS2 (NheI digestion) or RrB^short 5′-fused to MS2 (XbaI digestion), respectively. The transcripts from the resulting plasmids are terminated by the vrrA terminator originally found in Vibrio species. Transcription from the plasmids is under control of the P_LlacO–1 promoter. As the strains used for affinity purification harbor only the chromosomal copy of lacI, expression from these plasmids should be constitutive. This was verified by northern blotting (Supplementary Figure 4). Cell lysates of the strain hfq-FLAG (JVS814) harboring the different MS2-aptamer expressing plasmids, were prepared by growing cells in LB with 100 μg/ml ampicillin, to an OD₆₀₀ of 1.0. Cell pellets corresponding to 50 OD₆₀₀ units were resuspended in 800 μl of buffer A (20 mM Tris-HCl pH 8.0, 150 mM KCl, 1 mM MgCl₂, 1 mM DTT) and lysed by addition of glass beads (0.1 mm), flash freezing in liquid N₂ and shaking at 30 Hz for 10 min. Lysates were cleared by centrifugation (30 min 16,000 g, 4°C). Lysates corresponding to 2 and 0.5 OD₆₀₀ units of cell culture were used to prepare RNA and protein, respectively. Before addition of lysate, the affinity chromatography columns (Bio-Spin #732-6008, Bio-Rad) were prepared by adding 100 μl amylose resin (New England Biolabs, #E8021S), washing three times with 2 ml buffer A and subsequently adding 200 pmol of MS2-MBP recombinant protein (a gift from the Jörg Vogel group). All steps of the affinity chromatography were done at 4°C. Following addition of the lysate, the columns were washed four times with 2 ml buffer A, and the RNA-protein complexes were eluted with 900 μl buffer A containing 12 mM maltose. RNA was purified using phenol-chloroform, followed by ethanol precipitation with the addition of 1 μl GlycoBlue (Thermo Fisher). Protein was harvested from the organic phase by acetone precipitation. RNA was subjected to next generation sequencing on the Illumina platform at the University of Würzburg. Proteins were detected and quantified by mass spec at the mass spec facility at University of Würzburg, see Supplementary Methods.

The web version of Invenire sRNA found at http://markov.math.umb.edu/inveniresrna/was used for the analysis. The dataset included the 85 sRNAs, 22 TLR sequences, and 2217 intergenic regions (<1,000 bp) annotated in the Ecogene database¹ on November 24, 2017. In addition, 578 sequence peaks identified experimentally as CsrA binding sites by CLIP-seq (Potts et al., 2017) were included in the analysis.

Binding of RNAs to CsrA was examined using electrophoretic mobility shift assay (EMSA) with recombinant CsrA-3xFLAG (Supplementary Tables 1, 2) and in vitro transcribed RNA. DNA templates for in vitro transcription were made by PCR using pRrB, pRrB-3GGA and pCrsB (see Supplementary Tables 1, 2) as template for RrB, RrB-3GGA and CsrB, respectively. All primers used in the study are listed in Supplementary Table 3. Transcription was performed overnight at room temperature using the MEGAscript T7 transcription kit (ThermoFisher). The resulting RNA was treated with TurboDNase (ThermoFischer) (0.1 unit/μl, 30 min at 37°C), gel-purified from denaturing polyacrylamide gels, dephosphorylated using calf intestinal alkaline phosphatase (30 min at 37°C) and 5′ radiolabeled using T4 polynucleotide kinase and [γ-³²P] ATP. After each step the RNA was phenol extracted once, chloroform extracted twice and ethanol precipitated. Binding reactions contained 100 mM KCl, 10 mM MgCl₂, 2 mM DTT, 7.5% glycerol, 0.1 U SUPERase-IN RNase inhibitor (ThermoFisher), 2 ng total yeast RNA, 120 pM labeled RNA and 0–1,600 nM CsrA-3xFLAG in a 10 μl reaction. Samples were incubated 10 min at 37°C and separated on 8% native polyacrylamide gels using 1x TBE as buffer. Radioactive signals were detected on a PhosphorImager (Typhoon -GE Healthcare). The CsrA-3xFLAG was purified from the strain CsrA-3xFLAG carrying the plasmid pBAD-RBS-csrA:3xFLAG as described by Jorgensen et al. (2013).

For λ Red recombineering, we used a MAS1081-derivative strain (MAS1080) harboring the λ RED prophage [λ cI₈₅₇ Δ(cro-bioA)] imported from strain HME68 (Sawitzke et al., 2007) by P1 transduction. Deletion mutations were constructed in this strain as described (Sawitzke et al., 2007). For each deletion, a cat-sacB cassette was first inserted at the desired genomic region by selecting for chloramphenicol resistance and confirming by PCR. The cassette was then replaced by a DNA fragment designed to yield deletion of 107 bp, 103 bp and 103 bp for rrnA, rrnB and rrnD, respectively, by counter-selection of the cassette by sucrose tolerance and confirmation by PCR and DNA sequencing. The deletions roughly match the annotated TLR sequences, including all GGA motifs found in RrA, B, F. For reasons outside the scope of this study, the DNA fragments were constructed so that the TLR sequence was replaced with a tRNA gene in two out of the three deletion sites (see Supplementary Table 1).

To obtain an E. coli mutant where the λ Red recombination enzymes had not been expressed, and thereby reduce the risk of undesired genome mutations, each mutated locus was then moved to an otherwise wildtype strain by P1 transduction first of each cat-sacB cassette and next of the locus carrying a deletion. P1 transduction was done as described by Miller (1972).

Biofilm was measured in microtiter plates using peg-lids (Nunc-TSP, cat. no. 445497) and crystal violet staining. Cultures were grown in 96-well flat-bottom microtiter plates. Ten microliter outgrown culture was used to inoculate each well containing 150 μl YT media (per liter: 8 g tryptone, 5 g yeast extract and 5 g NaCl) supplemented with 100 μl/ml ampicillin and 1 mM IPTG for strains carrying plasmids. The outer-most wells on the plates were not used, as the results from these were found to fluctuate more than average. Plates were incubated at 37°C for 48 h (no shaking) to ensure all cultures were completely outgrown. After incubation the pegs were washed once in wash buffer (25 mM Tris pH 7.5, 100 mM NaCl) and placed in 0.01% crystal violet for 15 min. Then, the pegs were washed three times in wash buffer and transferred to a fresh microtiter plate containing 180 μl 96% ethanol in each well. When the crystal violet was completely dissolved, the A₅₉₀ of each well was measured along with the OD₆₀₀ of each well of the growth plate. The optical densities from the growth plate were used to normalize the absorbance signals from the stained biofilm.

Mobility was measured on soft agar plates. One microliter outgrown culture was used to inoculate each plate by injection of 1 μl culture halfway into the agar in the center of the plate. Plates contained YT media (per liter: 8 g tryptone, 5 g yeast extract and 5 g NaCl) + 0.3% agar, and were supplemented with 100 μl/ml ampicillin and 1 mM IPTG for strains carrying plasmids. The plates were incubated approx. 16 h at 37°C before the spread of the bacteria was measured. Measurements were repeated several times with approximately 2 h in between to compensate for differences in the growth rates of different strains.

The glgC-gfp in-frame fusion was constructed by replacing the NsiI-NheI fragment of pXG10-SF (Supplementary Table 2) with the 5′UTR and first 30 nucleotides of the glgC coding sequence by restriction digestion and ligation. The pgaA-lacZ in-frame fusion was made by cloning the promoter of lacI along with a NheI restriction site into pGH253-kan (Supplementary Table 2) by replacing the EcoRI-BamHI fragment. Subsequently, the 5′UTR and first 30 nucleotides of the pgaA coding sequence was inserted using BamHI-NheI digestion. All primers used in the study are listed in Supplementary Table 3.

Beta-Galactosidase (β-gal) activity was measured from overnight cultures harboring the translational fusion pgaA-lacZ, grown in test tubes in YT media (per liter: 8 g tryptone, 5 g yeast extract and 5 g NaCl) supplemented with 15 μg/ml kanamycin. The β-gal activity was measured using the fluorescent substrate 4-methylumbelliferone b-D-galactopyranoside (MUG), as described (Li et al., 2018). OD₆₀₀ of the cultures was measured and used to normalize the corresponding β-gal values.

Fluorescence of GFP was measured from strains harboring the glgC-gfp translational fusion. Cultures were grown in MOPS minimal media supplemented with 0.2% glucose, 15 μg/ml chloramphenicol and 100 μg/ml ampicillin when needed, at 37°C shaking at 300 rpm in 96-well black microtiter plates with clear, flat bottoms (Costar) in a plate reader (Synergy H1, Biotek). OD₆₀₀ and GFP fluorescence (excitation 470 nm and emission 510 nm) was measured simultaneously every 10 min. The values for GFP were normalized to the corresponding values for OD₆₀₀. The normalized data was plotted against time and the area under the curve was calculated as a measure of relative GFP expression. The same amount of data points were used for each strain and only data where cultures were growing exponentially was used in the analysis.

Three out of seven ribosomal RNA (rRNA) operons from E. coli contain a sequence from the TLR-family downstream from the 5S gene (Supplementary Figure 1). These three sequences, named rrA, rrB and rrF respectively, are very similar in sequence (Figure 1). Furthermore, their location downstream of 5S, and their sequence, is highly conserved in many species of the Enterobacteriales, suggesting a specific role for these sequences and structures (Supplementary Figure 2).

We hypothesized that the sequences may function as sRNAs. To begin testing this hypothesis we first examined whether RNA molecules of defined sizes could be detected after processing of the primary rrn transcripts. Total RNA harvested from exponentially growing E. coli K-12 cultures was subjected to northern blot analysis. To detect RrA, RrB and RrF, here collectively referred to as RrA, B, F, we used a single probe, which is expected to detect all three sequences since they only differ at a few positions. As shown in Figure 2A (lane 1 and 2), a distinct band of 150–200 nt was observed, confirming that RrA, B, F can be detected after processing of the primary transcript. The size of the RrA, B, F RNA was estimated based on re-probing of the membrane for transcripts of known sizes (Figures 2A,B) and comparison to an RNA ladder (Figure 2C). Next, we asked whether the RrA, B, F RNAs interact with the RNA chaperone Hfq, like many well-characterized E. coli sRNAs (Bilusic et al., 2014). To enable pull-down of Hfq and analysis of co-purified RNAs, a biotinylation sequence (Beckett et al., 2008) was inserted in the chromosomal hfq gene. The C-terminally tagged Hfq protein (Hfq_bio) is functional, as shown by intact repression of an mRNA target by an Hfq-dependent sRNA in the tagged strain (Supplementary Figure 3). After affinity purification of Hfq_bio using a streptavidin resin, co-precipitated RNA was analyzed by northern blot analysis. Detection of the known Hfq-binding sRNAs OxyS and ChiX only in the co-precipitated RNA verified that Hfq-binding RNAs had been enriched in the Hfq pull-down assay (Figure 2B). Interestingly, the pull-down revealed that a shorter form of RrA, B, F (RrA, B, F^short) was highly enriched on Hfq in addition to the band detected in total RNA (Figure 2A, lane 4). Further, to verify the identity of the bands on our northern blots found by the RrA, B, F, probe, we made a blot containing RNA from a strain overexpressing RrB from pRrB, a strain deleted for rrA, B, F, and several samples from the wildtype strain harvested at different stages of growth (Figure 2C). The blot verifies the identity of the bands and shows that RrA, B, F are only detected during exponential growth, where the rRNA operons are actively transcribed. During stationary phase, the transcriptional activity of rRNA operons is absent or very low (Baracchini and Bremer, 1991) but the 5S, 16S and 23S RNAs are stable molecules. Therefore the 5S rRNA detected during stationary phase in Figure 2C was transcribed during growth and serves as a qualitative loading control.

The ends of the longer RNA species were mapped using S1 nuclease protection assay (Berk and Sharp, 1977). As this assay was not sensitive enough to detect chromosomally expressed RrA, B, F, we used a strain overexpressing the 5S gene rrfB along with rrB from a plasmid. The S1 protected fragments are displayed in Figures 3A–C. Several bands spaced with one nucleotide intervals are seen, which is a common observation when using S1 nuclease transcript mapping (Green and Roeder, 1980; Aiba et al., 1981; Brosius et al., 1982) and is probably due to “end-nibbling” (Shenk et al., 1975). Thus, we define the transcript ends based on the longest protected bands. For the 5′ end the probe was shortened by 13 nt (Figure 3A). For the 3′ end the two different probes were shortened by 7 nt (Figure 3B) and 22 nt (Figure 3C), respectively. They both predict the same end. The 5′end of RrB was detected 3 nt downstream of the mature 3′end of the 5S transcript (see Figures 1, 3A). This site has been described as the initial RNase E processing site of the pre-5S RNA (Roy et al., 1983; Li and Deutscher, 1995) suggesting that the 5′ end of RrB is generated by RNase E cleavage. The 3′end of RrB was detected 13 nt downstream of its annotated 3′end (Figure 1). This 3′end was verified using two different probes (Figures 3B,C). These results predict that the most abundant RrA, B, F transcript in total RNA has a length of 173 nt, which is in good agreement with the northern blots shown in Figures 2A,C. We also carried out S1 analysis with probes antisense to the area between the two observed ends to detect any alternative transcript ends, but could not detect any (data not shown, the probes are shown in Figure 3D).

To determine the ends of the Hfq-enriched shorter version of RrA, B, F (RrA, B, F^short) we used the circular RACE method (McGrath, 2011), coupled with next generation sequencing. As this method is PCR-based it requires substantially less input RNA than the S1 nuclease assay. Circular RACE yielded a variety of sequences mapping to rrA, B, F, of which the ten most abundant are shown in Figure 3E. The most abundant reads predict the Hfq-associated RrA, B, F^short transcript to be 113–115 nt in length. This is in accordance with our observations from northern blots (Figure 2).

To detect the RrA, B, F^short fragment in the total RNA fraction, we carried out another northern blot analysis using substantially more total RNA per lane than the one presented in Figure 2A. This blot showed three different species of the RrA, B, F RNA (Figure 4A): the 115 nt RrA, B, F^short, the 173 nt fragment, and a longer version. We suspect that the longer version represents a processing intermediate that includes the T2 terminator sequence. Thus, the RrA, B, F^short is not uniquely seen in the RNA fraction co-precipitating with Hfq but can also be detected in the total RNA fraction.

A short half-life of the RrA, B, F RNA could explain its low abundance relative to the rRNA transcripts expressed from the same operons and the absence of signal from stationary phase cells (Figure 2C). We measured the half-life of RrA, B, F^short by monitoring the levels of RrA, B, F^short upon transcription initiation blocking by addition of rifampicin to a culture in balanced growth (Figure 4). Indeed, we found that the half-life of RrA, B, F^short was ∼2.5 min, which is very short relative to the rRNA transcripts that are stable on the time scale of hours during exponential growth (Piir et al., 2011).

Taken together, we conclude that the RrA, B, F RNAs can be detected in exponentially growing cells as at least two transcripts of ∼173 and ∼115 nt that appear to interact with Hfq and that the shorter variant RrA, B, F^short has a half-life of about 2–3 min. The presence of RrA, B, F is therefore dependent on active rRNA transcription.

In order to assess the conservation of RrA, B, F between species we did a BLAST search² with rrB and the sequences surrounding the locus as input (see Supplementary Figure 2 and Supplementary Material). Both the sequence and the location of the rrB downstream of the gene encoding 5S was found to be conserved in at least 31 species from 12 genera, all belonging to the order of Enterobacteriales (Supplementary Figure 2). To evaluate the structural conservation of the RrA, B, F homologs we used the locARNA algorithm (Will et al., 2007, 2012; Raden et al., 2018) to predict a consensus structure (Figure 5). The consensus structure shows a high degree of structural conservation, particularly in the three stem loop structures named 2-4 in the figure, which could indicate a conserved function of the RNA.

We speculated that the enrichment of RrA, B, F^short in RNA that had co-precipitated with Hfq could be due to a role for Hfq in chaperoning the correct folding of RrA, B, F, as shown for other RNAs (Geissmann and Touati, 2004; Soper and Woodson, 2008; Bordeau and Felden, 2014; Hoekzema et al., 2019). In this case, we would expect to see differences in the RrA, B, F^short structure with and without Hfq. To experimentally investigate the structure of the 115 nt RrB^short RNA, we performed RNA structure probing in the presence or absence of Hfq (Figure 6). As shown in Figure 1B, RrA^short and RrB^short are identical in sequence, and differ by only two nucleotides from RrF^short. We therefore expect RrA, B, F^short to serve identical functions, and arbitrarily chose the 115 nt long RrB^short from the rrnB operon for these experiments. An in vitro transcript of the RNA was incubated with increasing concentrations of Hfq, followed by exposure to the RNA cleavage agent Pb²⁺ that primarily hydrolyzes single-stranded RNA (Ciesiołka et al., 1998), or the endoribonuclease RNase III that predominantly hydrolyzes double-stranded RNA with a strong preference for helices of sufficient length (Robertson, 1990). Fragmented RNA was size-separated on a polyacrylamide gel and visualized by autoradiography (Figure 6A) In general, only subtle changes in the fragment pattern were observed upon Hfq addition (Figure 6A). The most notable Hfq-induced change was found in the Pb²⁺-treated samples at the apparent single-stranded region at nt 63-64, 66-68. This sequence displayed reduced hydrolysis by Pb²⁺ in the presence of Hfq (Figure 6A). In contrast, nt 53-56 showed slightly enhanced Pb²⁺ cleavage, upon addition of Hfq. Finally, nt 48-50 displayed slightly enhanced RNase III cleavage after incubation with Hfq (Figure 6A). The change in cleavage pattern observed at nt 63-68 as a consequence of Hfq addition could be interpreted as direct binding of Hfq in the area, leading to protection from Pb²⁺-induced cleavage. Alternatively, the main outcome of interaction with Hfq may be a structural rearrangement that allows the region around nt 63-68 to engage in intramolecular base-pairing. The structure probing data do not allow us to distinguish between these two potential effects of Hfq. To pursue the hypothesis that Hfq may facilitate a structural rearrangement of the region, we used the algorithm Mfold (Zuker, 2003)³ to predict the most thermodynamically favorable RrB^short structure with (Figure 6B) and without (Figure 6C) the constraint of a single-stranded region from nt 63-68 (open red triangles). The predicted structure for RrB^short in the absence of the constraint (Figure 6C) forms stem loop #2 predicted by the consensus structure (Figure 5), whereas the predicted structure given the structural constraint of a single-stranded region from nt 63-68, does not (Figure 6B). The structure shown in Figure 6B is not the energetically most favorable one. This leads us to suggest that Hfq could be involved in folding of the RrA, B, F in order for the RNA to attain its most thermodynamically favorable structure, namely that shown in Figure 6C, which shows the consensus stem-loop structure (stem loop #2) predicted by locARNA.

In order to identify potential interaction partners of RrA, B, F besides Hfq, we used MS2 affinity purification coupled with either mass spectrometry or RNA sequencing (Said et al., 2009; Corcoran et al., 2012; Lalaouna et al., 2015). Here, an RNA of interest is expressed as a fusion with an MS2-aptamer sequence, to allow purification of in vivo formed RNA-RNA or RNA-protein complexes using immobilized MS2 protein (Bardwell and Wickens, 1990; Said et al., 2009). We constructed plasmid-borne versions of RrB^short, MS2-tagged at either the 5′-end or the 3′-end and affinity-purified the RNAs. Expression and purification of the tagged RNAs was verified by northern blotting (Supplementary Figure 4). RNAs co-purifying with RrB^short were identified by deep RNA sequencing (RNA-seq) and co-purifying proteins were identified by mass spectrometry. The results were compared to results for affinity purification of the MS2 RNA alone. While no specific enrichment was reproducibly detected in the RNA-seq analysis (data not shown), several proteins were specifically enriched in pull-downs with MS2-tagged RrB^short, the most strongly enriched protein being the RNA-binding post-transcriptional regulator CsrA (Table 1). Somewhat surprising, Hfq was not found among the enriched proteins. We verified that RrB^short binds Hfq in vitro by electrophoretic mobility shift assay (EMSA). The EMSA analysis showed that RrB^short bound Hfq in vitro but with substantially lower affinity than the established Hfq-binding sRNA OxyS (Supplementary Figure 5). We will return to this point in section “Discussion.”

The canonical CsrA binding motif identified both in S. typhimurium (Holmqvist et al., 2016) and E. coli (Dubey et al., 2005; Potts et al., 2017) contains a GGA sequence located in the loop of a stem-loop structure. Interestingly, the RrA, B, F^short sequences have five GGA sequences, two of which are predicted to be located in the loops of stem-loop structures by the mfold and LocaRNA structure prediction algorithms (Figures 5, 6). Intriguingly, the structure probing experiments shown in Figure 6A suggest that the potential CsrA binding motif in loop 2 is only formed after interaction with Hfq.

The machine-learning-based algorithm InvenireSRNA (Fakhry et al., 2017) is designed to predict sRNAs of the CsrB/C family. To gauge how RrA, B, F rank in this algorithm compared to other E. coli non-coding RNAs, we provided the algorithm with a total of 2902 sequences, including all annotated E. coli sRNA sequences, the 22 TLR sequences, all sequences identified as CsrA-binding peaks from CLIP-seq data (Potts et al., 2017), and finally all intergenic regions of E. coli shorter than 1,000 bp. The results show that CsrB and CsrC both score high as expected (Table 2), while McaS and GadY both have low probability scores, suggesting that the features that make the latter acceptable binding partners for CsrA are not picked up by the algorithm. In accordance with the MS2-purification and structure prediction, RrA^short, RrB^short and RrF^short are predicted to bind CsrA with high probability and, remarkably, rank among the 16 highest scoring sequences in the analysis (Table 2). Notably, the remaining 19 members of the TLR family located in tRNA operons all obtained a probability score of <0.01, suggesting RrA, B, F are unique among the TLR family in their affinity for CsrA.

To further validate the binding between RrB^short and CsrA we conducted EMSAs. In vitro transcribed RrB^short was incubated with increasing amounts of purified CsrA-3xFLAG and separated on non-denaturing gels. As seen in Figure 7A, an RrB^short-CsrA complex is observed from a concentration of 50 nM CsrA, and a complete shift is seen at 100–200 nM CsrA. At higher concentrations of CsrA, several complexes with higher molecular weights are visible. The number of different band sizes suggests that RrB^short has multiple binding sites for CsrA, which is expected based on our RNA structure predictions (Figures 5, 6). The binding relationship between CsrA and RrB^short is therefore complex but a first approximation of an overall K_D from our data in Figure 7A is around 100 nM. We repeated the binding experiment using a mutant version of RrB^short, where three GGA sequences have been changed to “UUU” (number 1, 2 and 5 on Figure 6C), which includes the two motifs found in loops of stem structures. By only mutating the three GGA motifs found in single stranded regions, we do not expect this to have any consequences for the secondary structure of the RNA. The affinity for CsrA is clearly lower for this mutant compared to the wildtype RrB^short as the mutant does not show any shift within the tested range of CsrA concentrations. Note that part of the RNA is degraded at the highest CsrA concentration. We attribute this to residual activity of the RNase that was added during CsrA purification, see materials and methods. To further probe the specificity of RrB^short to CsrA we did competition binding experiments where a preformed CsrA-RrB^short complex (CsrA at 200 nM, RrB^short at 120 pM) was challenged with increasing concentrations of the unlabeled competitors RrB^short, RrB^short-3GGA and CsrB (Figure 7B). CsrB is believed to have ∼18 CsrA binding sites (Liu et al., 1997), a K_D around 1 nM of binding to CsrA (Weilbacher et al., 2003) and thus we expect it to be a more efficient competitor than RrB^short. This is indeed what we observe, CsrB fully competes off the labeled RrB^short at a concentration of 3.13 nM, whereas a concentration of 50 nM is needed in the self-competition with RrB^short. RrB^short-3GGA show significantly less competition and is not able to fully compete the labeled RrB^short off CsrA within the range of concentrations tested.

With these binding assays we clearly show that RrB^short binds CsrA in vitro and that this interaction relies on the predicted binding sites located in single-stranded regions. The two remaining GGA motifs in the mutant RNA RrB^short-3GGA do not efficiently bind CsrA on their own. We suspect that these low affinity GGA sites may require nearby high affinity sites in order to bind CsrA, this type of binding has previously been described for the CsrA homolog RsmE (Duss et al., 2014).

In combination, our results from MS2 affinity purification, structure prediction, the InvenireSRNA algorithm, and the EMSAs strongly suggest that RrB^short, and likely RrA, B, F^short, specifically interact with the post-transcriptional regulator CsrA.

We next asked whether RrA, B, F might act by sequestering CsrA, similarly to the effects of CsrB, CsrC, GadY and McaS. To this end, we constructed an E. coli mutant deleted for the chromosomal copies of rrA, B, F (see Materials and Methods). Additionally, to overexpress rrB in a fashion that would presumably allow normal processing of RrB^short, we cloned the rrB sequence including the 3′-half of the adjacent 5S gene and both terminators (T1 and T2) into a modified pUC18 plasmid (pJFR1, see Supplementary Table 2) to make pRrB.

Putative effects of RrB expression on the activity of CsrA were investigated by monitoring several CsrA-regulated phenotypes: growth rate, biofilm formation, motility, and post-transcriptional repression of the CsrA-regulated genes pgaA (Wang et al., 2005) and glgC (Baker et al., 2002). As a positive control, we moved a previously characterized csrA::kan allele (Romeo et al., 1993) into our wildtype strain (which is otherwise isogenic to the ΔrrA, B, F strain). The csrA::kan allele encodes a truncated version of CsrA which is known to have strongly reduced CsrA activity (Romeo et al., 1993). The results of these analyses are presented in Figure 8. In general, if RrA, B, F regulate CsrA activity in a manner similar to CsrB, we expect the phenotype of overexpression of rrB to display a similar tendency to that of the csrA::kan allele. Neither deletion nor overexpression of the rrA, B, F significantly affected the growth rate in minimal glucose medium (Figure 8A). For comparison, E. coli harboring a truncated CsrA have been reported to grow at approximately half the growth rate of the wildtype strain when glucose is the carbon source (Morin et al., 2016). The mutants lacking or overexpressing the rrA, B, F were also tested for motility on soft agar plates. Neither overexpression nor deletion of the rrA, B, F affected motility (Figure 8B). In contrast, the csrA::kan strain was severely impaired for motility as previously described (Romeo et al., 1993), although we note that this phenotype in our hands was very variable between replicate experiments.

CsrA is a negative regulator of PGA (poly-β-1,6-N-acetyl-d-glucosamine)-mediated biofilm formation (Wang et al., 2005). In agreement with this, we found that the csrA::kan strain produces more biofilm than wildtype as measured by crystal violet staining (Figure 8C). The ΔrrA, B, F strain also formed significantly more biofilm than the wildtype. However, overexpression of RrB also showed somewhat increased biofilm formation, although the difference from the strain carrying the empty vector control was not statistically significant.

In connection with biofilm, translation of the pgaABCD mRNA, encoding the machinery for synthesis and transport of PGA, is known to be strongly repressed by CsrA (Wang et al., 2005). Indeed, we found > 20-fold higher activity of β-galactosidase from a pgaA-lacZ fusion in the presence of CsrB overexpression than in the wildtype strain (Figure 8D). A modest increase in β-galactosidase activity was also observed in the ΔrrA, B, F mutant. Again, overexpression of RrB resulted in a phenotype similar to that of the Δrr,AB,F mutant, pointing to a surprisingly similar effect on the pgaA-lacZ fusion of decreasing and increasing the RrA, B, F levels. There is good correlation between our data on biofilm formation and our data on the pgaA-lacZ fusion for all the strains (Figures 8C,D), as would be expected for CsrA-dependent regulation of biofilm (Wang et al., 2005).

Lastly, we tested the expression of GFP from a plasmid-borne translational glgC-gfp fusion during exponential growth, the growth phase where RrA, B, F accumulated (Figure 2C). The glgCAP operon encodes enzymes involved in glycogen metabolism, and is known to be repressed by CsrA (Baker et al., 2002). CsrB overexpression increases glgC-gfp expression substantially, while we observed slightly more GFP signal both upon deletion and overexpression of rrA, B, F (Figure 8E). In summary, we only observed minor phenotypic effects of changes in the expression of RrA, B, F. The RrA, B, F RNAs appear to modestly influence biofilm formation and expression from the CsrA-controlled pgaA and glgC mRNAs, but all three assays showed the same curious trend, namely that both deletion of rrA, B, F and overexpression of rrB resulted in a change consistent with modestly reduced CsrA activity. A plausible explanation for this unexpected similarity of phenotypes would be if RrB RNA expressed from the vector was processed into a form that exerted a different action on the tested properties than chromosomally expressed RrB. To address this hypothesis, we also tested expression of the glgC-gfp fusion in the presence of the pKK3535 plasmid that contains the entire rrnB operon, including rrB. Although the rRNA operon promoters are feedback-controlled and thus difficult to overexpress, this plasmid was expected to cause a relative increase in RrB levels, because the pKK3535 plasmid is responsible for approximately 50% of total rRNA synthesis even in the presence of the seven chromosomal rRNA operons (Steen et al., 1986). As shown in Supplementary Figure 5, overexpression of the rrnB operon resulted in increased glgC-gfp expression, and was thus consistent with reduced CsrA activity under this condition. In our hands, however, the strains containing pKK3535 had impaired growth rates (see Supplementary Figure 6A) and thus we did not pursue additional experiments with this plasmid.

The very modest effects of RrA, B, F on CsrA-controlled phenotypes suggest that if there is a physiological consequence of RrA, B, F interaction with CsrA then we might not have examined it under the proper growth conditions. In E. coli, rRNA expression is controlled by the second messenger ppGpp (Zhang and Bremer, 1995). A decrease in ppGpp levels leads to increased rRNA expression (as observed upon treatment with antibiotics specifically blocking translation (Kurland and Maaløe, 1962; Muto et al., 1975)). To test whether the ppGpp effect also applies to rrA, B, F, we induced translational arrest, either by chloramphenicol addition, or by expression of the toxin MazF, and investigated the level of RrA, B, F by northern blot analysis at several time points after induction. Chloramphenicol blocks translation by binding to ribosomes and inhibiting the peptidyl transferase activity (Das et al., 1966) while MazF leads to translational arrest by disrupting ribosome biogenesis and cleaving mRNAs (Culviner and Laub, 2018), thereby removing the template for translation. In both cases of translational arrest, we detect a strong increase in RrA, B, F transcript levels 20–80 min after the treatment (Figure 9). As an independent indicator of transcription from the rRNA operons, we also investigated the level of tRNA^Glu. This tRNA is exclusively expressed from genes located in four rRNA operons (rrnB, rrnC, rrnE, rrnG), so we expected its expression pattern to resemble that of RrA, B, F. Indeed, the level of tRNA^Glu also increases during translational arrest, albeit not nearly to the same extent as RrA, B, F (Figure 9).