Different csrA Expression Levels in C versus K-12 E. coli Strains Affect Biofilm Formation and Impact the Regulatory Mechanism Presided by the CsrB and CsrC Small RNAs

Escherichia coli C is a strong biofilm producer in comparison to E. coli K-12 laboratory strains due to higher expression of the pgaABCD operon encoding the enzymes for the biosynthesis of the extracellular polysaccharide poly-β-1,6-N-acetylglucosamine (PNAG). The pgaABCD operon is negatively regulated at the post-transcriptional level by two factors, namely CsrA, a conserved RNA-binding protein controlling multiple pathways, and the RNA exonuclease polynucleotide phosphorylase (PNPase). In this work, we investigated the molecular bases of different PNAG production in C-1a and MG1655 strains taken as representative of E. coli C and K-12 strains, respectively. We found that pgaABCD operon expression is significantly lower in MG1655 than in C-1a; consistently, CsrA protein levels were much higher in MG1655. In contrast, we show that the negative effect exerted by PNPase on pgaABCD expression is much stronger in C-1a than in MG1655. The amount of CsrA and of the small RNAs CsrB, CsrC, and McaS sRNAs regulating CsrA activity is dramatically different in the two strains, whereas PNPase level is similar. Finally, the compensatory regulation acting between CsrB and CsrC in MG1655 does not occur in E. coli C. Our results suggest that PNPase preserves CsrA-dependent regulation by indirectly modulating csrA expression.


Introduction
The Csr/Rsm (carbon storage regulator/repressor of stationary phase metabolites) system controls key phenotypes ranging from carbon metabolism to virulence and biofilm formation in Gammaproteobacteria. Over the years, studies by different groups, and in particular by T. Romeo's lab, have clarified many aspects of the mechanism by which the protein CsrA (or its ortholog RsmA) controls this system in Escherichia coli and other bacteria, showing that CsrA and its orthologs act by modulating translation, decay and transcription elongation of a number of mRNAs [1,2]. CsrA homodimers bind GGA motifs located in a single-stranded loop of short hairpins usually present in multiple copies in the 5 untranslated regions (5 -UTR) of CsrA mRNA targets [3,4]. CsrA protein activity is negatively regulated by the non-coding small RNAs (sRNAs) CsrB and CsrC, which contain multiple CsrA binding sites (i.e., 18 in CsrB and 14 in CsrC) and antagonize CsrA activity by sequestering it [5,6]. Another sRNA, namely McaS, also binds CsrA and negatively regulates its activity [7]. Multiple positive and negative feedback loops strictly control the amounts of CsrA and its sRNA regulators [8,9]. CsrA indirectly activates csrB and csrC transcription [10,11] and stabilizes CsrB and CsrC by repressing the expression of control the amounts of CsrA and its sRNA regulators [8,9]. CsrA indirectly activates csrB and csrC transcription [10,11] and stabilizes CsrB and CsrC by repressing the expression of csrD, which promotes CsrB and CsrC RNase E-dependent degradation [12]. Coordinated regulation also exists between the expression level of CsrB and CsrC, as the absence of each of them elicits compensatory effects on the expression of the other one [6,12].
In E. coli K-12, the csrA gene can be transcribed by multiple promoters (Figure 1), among which the Eσ D -dependent P5 and the Eσ S -dependent P3 are the most active in exponential and stationary phase, respectively. CsrA indirectly activates transcription at P3 and negatively regulates the translation of the P3 mRNA. Conversely, the transcript starting at P5 does not seem to be subject to CsrA translation modulation [13]. The csrA promoter region is altered in E. coli C strains because of the insertion of the IS3 transposable element within the −35 region of the P4 promoter ( Figure 1). This insertion abolishes csrA autoregulation acting on the P3 transcript. Indeed, in E. coli C, transcripts starting from all csrA promoters but P5, which is located immediately downstream of the IS3 insertion site, are predicted to terminate within IS3, and thus csrA is transcribed exclusively from P5 [14]. , and P5 promoters (bent arrows) located 137, 62, and 53 bp upstream of the ATG start codon, respectively, are represented [13]. Red and blue arrowheads, insertion points of IS3 in C-1a and of the kan R cassette in the csrA::kan mutants [14,15]. The region between positions +8 and +37 downstream of the csrA stop codon (t) was identified as a putative termination site [16]. mRNAs originating from P3 (gray line), P4, and P5 (black lines) and terminating at +37 with respect to the csrA stop codon are reported with their predicted length.
The pgaABCD operon is one of the targets of CsrA in E. coli [17,18]. pgaABCD encodes the enzymes for the biosynthesis of poly-β-1,6-N-acetylglucosamine (PNAG), an exopolysaccharide with a major role as an extracellular matrix component in biofilms of both Gram-positive and Gram-negative bacteria [19][20][21]. The pgaABCD operon expression is positively regulated by the transcription activator NhaR and negatively regulated by CsrA, which binds multiple sites in its 5′-untranslated region (5′-UTR) and causes premature transcription termination and pgaA translation repression [18,[22][23][24]. Surprisingly, we observed that pgaABCD operon expression and PNAG production were increased in C-1a ΔcsrB and ΔcsrC mutants [25]. Since free CsrA (i.e., not sequestered by CsrB/CsrC) should be enhanced in these mutants, this observation is difficult to reconcile with the current model of pgaABCD negative regulation by CsrA [17].
Besides CsrA, another negative regulator of pgaABCD expression is the RNA exonuclease polynucleotide phosphorylase (PNPase) [26], which also acts at the posttranscriptional level. Mechanism of PNPase-dependent pgaABCD regulation is unclear, Figure 1. csrA locus and csrA mRNAs. The 186 bp long open reading frame (ORF; box) of the csrA gene and the transcription start sites for P3, P4, and P5 promoters (bent arrows) located 137, 62, and 53 bp upstream of the ATG start codon, respectively, are represented [13]. Red and blue arrowheads, insertion points of IS3 in C-1a and of the kan R cassette in the csrA::kan mutants [14,15]. The region between positions +8 and +37 downstream of the csrA stop codon (t) was identified as a putative termination site [16]. mRNAs originating from P3 (gray line), P4, and P5 (black lines) and terminating at +37 with respect to the csrA stop codon are reported with their predicted length.
The pgaABCD operon is one of the targets of CsrA in E. coli [17,18]. pgaABCD encodes the enzymes for the biosynthesis of poly-β-1,6-N-acetylglucosamine (PNAG), an exopolysaccharide with a major role as an extracellular matrix component in biofilms of both Gram-positive and Gram-negative bacteria [19][20][21]. The pgaABCD operon expression is positively regulated by the transcription activator NhaR and negatively regulated by CsrA, which binds multiple sites in its 5 -untranslated region (5 -UTR) and causes premature transcription termination and pgaA translation repression [18,[22][23][24]. Surprisingly, we observed that pgaABCD operon expression and PNAG production were increased in C-1a ∆csrB and ∆csrC mutants [25]. Since free CsrA (i.e., not sequestered by CsrB/CsrC) should be enhanced in these mutants, this observation is difficult to reconcile with the current model of pgaABCD negative regulation by CsrA [17].
Besides CsrA, another negative regulator of pgaABCD expression is the RNA exonuclease polynucleotide phosphorylase (PNPase) [26], which also acts at the post-transcriptional level. Mechanism of PNPase-dependent pgaABCD regulation is unclear, but in cis determinants of PNPase-dependent regulation, as well as for CsrA, lie in the pgaABCD 5 -UTR [25]. Consistent with the role of PNPase as a negative regulator of pgaABCD, the deletion of the pnp gene encoding PNPase in the E. coli K-12 strain MG1655 determines increased adhesion, a phenotype suppressed by the deletion of the pgaA gene [25]. The effect of the ∆pnp mutation is stronger in the E. coli C C-1a genetic background, as cultures in the minimal medium of a C-1a ∆pnp mutant undergo massive aggregation due to pgaABCD operon overexpression and PNAG hyperproduction, whereas MG1655 ∆pnp cultures do not visibly aggregate in the same conditions (see Figure 2A in the Results section) [25], suggesting that they may produce less PNAG. In agreement with this hypothesis, it was recently shown that E. coli C produces more robust biofilm than other E. coli strains, among which E. coli K-12 [14].  It should be mentioned that around the 20% of cultures obtained by inoculating C-1a ΔcsrA colonies in LD broth showed no/poor growth, whereas C-1a csrA::kan cultures did not show grow defects in LD but had erratic growth rate upon dilution in M9-LG (Supplementary Figure S1), suggesting genetic variability. Inconsistency in growth rate upon dilution in M9-LG was not observed with MG1655 csrA::kan ( Figure 2A) [44].

CsrA-Dependent Regulation of pgaABCD Operon Is More Stringent in E. coli K-12 Than in E. coli C
Increased PNAG production in C-1a with respect to MG1655 may be due to enhanced expression of pgaABCD biosynthetic operon in the former strain. Consistent with this hypothesis, the pgaA mRNA was about two-fold more abundant in C-1a than in MG1655 ( Figure 2C). This did not depend on the higher activity of the pgaABCD promoter in C-1a, as transcription efficiency from the pgaABCD promoter was comparable in the two strains (Supplementary Figure S2).
Since pgaABCD mRNA level is subject to post-transcriptional control by CsrA, we checked whether CsrA was responsible for the pgaABCD expression differential between C-1a and MG1655. We found that the amount of pgaA mRNA (taken as representative of the operon mRNA) in the csrA::kan mutants of the two strains was similar and enhanced with respect to the csrA + strains. pgaA mRNA level was further increased in the C-1a ΔcsrA with respect to that in C-1a (C csrA + ). qRT-PCR of RNA extracted from cultures grown at 37 • C in M9-LG up to OD 600 = 0.5 was performed as described in Materials and Methods. Bars represent the average with range of determinations on 3 (csrA + and ∆csrA strains) or 2 (csrA::kan strains) cultures. The significance of the difference between the average expression in C-1a and MG1655 (K-12 csrA + ) is reported in the inset and was estimated with a t-test.
In this work, we investigated the molecular bases of different PNAG production in C-1a and MG1655 as models of E. coli C and K-12 strains, respectively. We found that the pgaABCD operon is more tightly regulated by CsrA in MG1655 than in C-1a. Conversely, the negative effect exerted by PNPase on pgaABCD expression is much stronger in E. coli C than in E. coli K-12. The sRNAs regulating CsrA activity have different expression profiles in the two strains. We discuss the hypothesis that both different pgaABCD expression in C-1a vs. MG1655, as well as negative CsrB and CsrC effects on pgaABCD observed in E. coli C, may be the consequence of adaptation to different CsrA levels found in the two genetic backgrounds.

Luciferase Activity Assay
Bacterial cultures were grown at 37 • C in LD broth supplemented with ampicillin 50 µg/mL up to OD 600 = 0.5. A total of 5 mL of cultures were harvested by centrifugation 5 min at 4000× rpm and 4 • C, and the bacterial pellet was resuspended in 5 mL of M9-LG broth supplemented with ampicillin 50 µg/mL. Bacteria were grown 90 min at 37 • C, harvested by centrifugation 5 min at 4000× rpm and 4 • C and resuspended in PBS at OD 600 = 0.1. To measure luciferase activity, 5 µL of bacterial suspension were diluted in 500 µL of fresh PBS, and 20 µL of 1% decanal in ethanol was added. Luminescence was measured with a Stratec luminometer.

PNAG Detection
PNAG production was determined as described [25]. Bacteria were grown overnight in M9-LG at 37 • C. A total of 1.5 OD 600 were collected, and 1/30 of cell lysate (10 µL) was spotted onto a nitrocellulose filter using a Dot-blot apparatus (Bio-Rad, Hercules, CA, USA), incubated overnight at 4 • C with PNAG antibodies (a kind gift from G.B. Pier [41]) and revealed using ECL Western blotting reagent PDS Standard (Genespin, Milano, Italy).

Western Blotting
E. coli crude extracts were obtained as described previously [42]. Protein content was determined using Coomassie Plus protein assay reagent (Pierce, Thermo Scientific, Waltham, MA, USA). Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) was performed on 10% resolving gels containing 0.1% SDS. PageRuler Prestained Molecular-weight markers (Thermo Scientific, Waltham, MA, USA) were used as a size reference. For immunological detection of PNPase and CsrA, the gels were blotted onto a nitrocellulose (Hybond ECL, SIGMA, Saint Louis, MO, USA) sheet and incubated with polyclonal anti-PNPase [43] and anti-csrA antibodies (Biorbyt Ltd., Cambridge, UK), respectively. Since the anti-csrA antibody provided many strong unspecific signals, we preincubated a 1:500 antibody dilution with 0.15 mg/mL of ∆csrA extract for 2 h at 4 • C before using it for filter immunodecoration. Immunoreactive bands were revealed using the ECL Western blotting reagent PDS Standard (Genespin, Milano, Italy).

Statistical Analysis
Statistical tests were applied to compare the means of results obtained by analyzing at least three biological replicates for each group/condition. We used one-way analysis of variance (ANOVA) with Tukey's post-hoc test for comparison among means of three or more groups and an independent two-tailed t-test for comparison between two groups.

Results
3.1. E. coli C Produces More PNAG than E. coli K-12 The pnp gene deletion causes massive aggregation in E. coli C growing in M9-LG medium [25], whereas in E. coli K-12 has no apparent effect (Figure 2A). This difference can be explained by lower PNAG production in E. coli K-12 pnp + and ∆pnp strains with respect to their E. coli C counterparts [14] ( Figure 2B). In both genetic backgrounds, csrA mutations such as the csrA gene deletion [25] or a csrA hypomorphic allele encoding a partially active protein (i.e., csrA::kan allele [15,44]) enhanced PNAG production, as expected, and slowed the growth after about 3-4 generations in C-1a and 5-6 in MG1655 (Figure 2A; Supplementary Figure S1).
It should be mentioned that around the 20% of cultures obtained by inoculating C-1a ∆csrA colonies in LD broth showed no/poor growth, whereas C-1a csrA::kan cultures did not show grow defects in LD but had erratic growth rate upon dilution in M9-LG (Supplementary Figure S1), suggesting genetic variability. Inconsistency in growth rate upon dilution in M9-LG was not observed with MG1655 csrA::kan (Figure 2A) [44].

3.2.
CsrA-Dependent Regulation of pgaABCD Operon Is More Stringent in E. coli K-12 than in E. coli C Increased PNAG production in C-1a with respect to MG1655 may be due to enhanced expression of pgaABCD biosynthetic operon in the former strain. Consistent with this hypothesis, the pgaA mRNA was about two-fold more abundant in C-1a than in MG1655 ( Figure 2C). This did not depend on the higher activity of the pgaABCD promoter in C-1a, as transcription efficiency from the pgaABCD promoter was comparable in the two strains (Supplementary Figure S2).
Since pgaABCD mRNA level is subject to post-transcriptional control by CsrA, we checked whether CsrA was responsible for the pgaABCD expression differential between C-1a and MG1655. We found that the amount of pgaA mRNA (taken as representative of the operon mRNA) in the csrA::kan mutants of the two strains was similar and enhanced with respect to the csrA + strains. pgaA mRNA level was further increased in the C-1a ∆csrA mutant ( Figure 2C). Thus, in both genetic backgrounds, CsrA negatively regulates pgaA expression, and it appears to be responsible for the difference in pgaA expression between E. coli C and K-12. To strengthen this hypothesis, since CsrA down-regulates pgaA translation by interacting with its 5 -UTR, we assayed the expression of a pgaAlux translational fusion between the promoter region and 5 -UTR of pgaABCD and the luciferase gene in C-1a and MG1655 and in their respective ∆csrA derivatives. All strains had the ∆pgaC mutation in the chromosome to avoid auto-aggregation due to PNAG over-production and ∆csrA suppressor selection [7,25,45]. We found that luciferase activity was ten-fold higher in C-1a than in MG1655 ( Table 2). The ∆csrA mutation resulted in a ca. 26-fold relative induction in C-1a and in a staggering 420-fold relative induction in K-12, boosting luciferase activity to comparable levels in E. coli C and K-12 strains, and thus strongly supporting the hypothesis that pgaABCD expression is lower in MG1655 than in C-1a because CsrA-dependent repression is tighter in the former strain.  All strains carried ∆pgaC mutation and plasmid pLpga2 [25]. b t-test performed between determinations in each wt strain and its isogenic mutants. na, not applicable. c t-test performed between data relative to E. coli C and K-12 strains with the same mutations. ns, not significant.
3.3. PNPase-Dependent Regulation of pga Operon Is More Stringent in E. coli C than in E. coli K-12 We compared the contribution of PNPase to pgaABCD regulation in E. coli C and K-12 by exploiting the pgaA-lux fusion described above. Luciferase activity was enhanced around 14-fold-in C-1a ∆pnp, in agreement with previously published data [25], and 3-fold in MG1655 ∆pnp with respect to their pnp + counterparts. Thus, in the presence of CsrA, PNPase negative effect on pgaABCD expression is stronger in E. coli C than in K-12. The additional ∆pnp mutation in the ∆csrA strains enhanced luciferase activity in E. coli C. In E. coli K-12, an increment was also observed, but this result is less convincing because of the high variability associated with MG1655 ∆csrA derivatives in this assay (Table 2).
It was reported that CsrA regulates PNPase translation by binding to pnp mRNA 5 -UTR [46]. However, the PNPase level was the same between E. coli C and K-12 and also between csrA + and csrA mutants ( Figure 3). Thus, the higher impact of PNPase on pgaABCD regulation in E. coli C vs. K-12 does not depend on differences in pnp expression.
had the ΔpgaC mutation in the chromosome to avoid auto-aggregation due to PNAG overproduction and ΔcsrA suppressor selection [7,25,45]. We found that luciferase activity was ten-fold higher in C-1a than in MG1655 ( Table 2). The ΔcsrA mutation resulted in a ca. 26fold relative induction in C-1a and in a staggering 420-fold relative induction in K-12, boosting luciferase activity to comparable levels in E. coli C and K-12 strains, and thus strongly supporting the hypothesis that pgaABCD expression is lower in MG1655 than in C-1a because CsrA-dependent repression is tighter in the former strain.  [25]. b t-test performed between determinations in each wt strain and its isogenic mutants. na, not applicable. c t-test performed between data relative to E. coli C and K-12 strains with the same mutations. ns, not significant.

PNPase-Dependent Regulation of pga Operon Is More Stringent in E. coli C Than in E. coli K-12
We compared the contribution of PNPase to pgaABCD regulation in E. coli C and K-12 by exploiting the pgaA-lux fusion described above. Luciferase activity was enhanced around 14-fold-in C-1a Δpnp, in agreement with previously published data [25], and 3fold in MG1655 Δpnp with respect to their pnp + counterparts. Thus, in the presence of CsrA, PNPase negative effect on pgaABCD expression is stronger in E. coli C than in K-12. The additional Δpnp mutation in the ΔcsrA strains enhanced luciferase activity in E. coli C. In E. coli K-12, an increment was also observed, but this result is less convincing because of the high variability associated with MG1655 ΔcsrA derivatives in this assay (Table 2).
It was reported that CsrA regulates PNPase translation by binding to pnp mRNA 5′-UTR [46]. However, the PNPase level was the same between E. coli C and K-12 and also between csrA + and csrA mutants ( Figure 3). Thus, the higher impact of PNPase on pgaABCD regulation in E. coli C vs. K-12 does not depend on differences in pnp expression.

Expression Profile of the csrA Gene and of sRNAs Regulating CsrA Activity in E. coli C and K-12
We analyzed csrA gene expression in E. coli C and K-12 by Northern blotting. In MG1655, the main csrA signal corresponded to an RNA with an estimated length of 350-370 nt, compatible with an mRNA originating from the P3 promoter and terminating 30-50 nt downstream of the csrA stop codon (Figures 1 and 4A). Two mRNAs of similar length and migrating slightly faster than the 300 nt long RNA marker were also present. These species likely correspond to transcripts starting at P4/P5 and terminating where P3 mRNA also ends. In C-1a, the main signal corresponded to the putative P5 mRNA, together with faint bands corresponding to longer RNAs. The abundance of csrA mRNAs, considering the overall amount of P3, P4, and P5 mRNAs for MG1655 and P5 mRNA for C-1a, was around three-fold higher in E. coli K-12 than in E. coli C.

K-12
We analyzed csrA gene expression in E. coli C and K-12 by Northern blotting. In MG1655, the main csrA signal corresponded to an RNA with an estimated length of 350-370 nt, compatible with an mRNA originating from the P3 promoter and terminating 30-50 nt downstream of the csrA stop codon (Figures 1 and 4A). Two mRNAs of similar length and migrating slightly faster than the 300 nt long RNA marker were also present. These species likely correspond to transcripts starting at P4/P5 and terminating where P3 mRNA also ends. In C-1a, the main signal corresponded to the putative P5 mRNA, together with faint bands corresponding to longer RNAs. The abundance of csrA mRNAs, considering the overall amount of P3, P4, and P5 mRNAs for MG1655 and P5 mRNA for C-1a, was around three-fold higher in E. coli K-12 than in E. coli C.  These results were confirmed also in C-1a and MG1655 strains not containing the ∆pgaC mutation (Table 3; see also Figure 5A). No signal was detected in the ∆csrA strains, as expected. Table 3. Expression of csrA, csrB, csrC, and mcaS.

RNA Relative Amount a C-5691 MG1655
a Relative amount with respect to C-1a. Cultures were grown, and Northern blotting performed as described in Figure 4A legend. Northern blot signals were quantified with ImageQuant, and the values were normalized for those of the 5S rRNA and for the C-1a values. Average of the results of three independent experiments are shown with standard deviation. For McaS, data are the average of two independent determinations with range. b The sum of signals corresponding to P3, P4, P5 mRNAs ( Figure 4A) was considered for MG1655. P5 signal was considered for C-1a.
ImageQuant normalized for L4 signals are shown below the lanes. The value obtained in MG1655 was taken as a reference for comparison. na, not applicable.
These results were confirmed also in C-1a and MG1655 strains not containing the ΔpgaC mutation (Table 3; see also Figure 5A). No signal was detected in the ΔcsrA strains, as expected. a Relative amount with respect to C-1a. Cultures were grown, and Northern blotting performed as described in Figure 4A legend. Northern blot signals were quantified with ImageQuant, and the values were normalized for those of the 5S rRNA and for the C-1a values. Average of the results of three independent experiments are shown with standard deviation. For McaS, data are the average of two independent determinations with range. b The sum of signals corresponding to P3, P4, P5 mRNAs ( Figure 4A) was considered for MG1655. P5 signal was considered for C-1a. The ∆pnp mutation in the C-1a background decreased csrA mRNA abundance, whereas it did not significantly change the overall abundance of csrA transcripts when present in MG1655 ( Figure 4A). Consistent with the csrA transcription profile, western blotting analysis showed that the CsrA level was higher in MG1655 than in C-1a ( Figure 4B). Indeed, we could not detect CsrA in any tested E. coli C strains, with the paradoxical exception of the csrA::kan mutant, in which the signal corresponding to a possible chimeric protein slightly bigger than wildtype CsrA [10] was stronger than in MG1655 csrA::kan ( Figure 4B). C-1a csrA::kan was obtained by P1-mediated transduction from MG1655 csrA::kan, implying that the csrA promoter region in C-1a csrA::kan is in all probability deriving from the donor MG1655. Consistent with csrA locus transcription from the same promoter in the two mutant strains, csrA mRNAs with the same electrophoretic mobility, and thus presumably of the same length, are produced in C-1a and MG1655 csrA::kan mutants (Supplementary Figure S3A). Given the growth variability shown by C-1a csrA::kan mutants in M9-LG (Figure 2A and Supplementary Figure S1), it is possible that in (some) C-1a csrA::kan cultures, suppressor mutations may result in an increased amount of the CsrA-kan chimeric protein.
The amount of sRNAs CsrB, CsrC, and McaS were dramatically different between C-1a and MG1655. CsrC and especially CsrB were much more abundant, and McaS strongly reduced in MG1655 with respect to C-1a ( Figure 4A; Table 3).
McaS transcription was previously reported to be activated in low glucose [47,48]. We thus compared its levels in bacteria growing in either LD broth, in which glucose is scarce and quickly consumed by growing bacteria [49], or M9-LG, which contains 0.4% glucose. As expected, in MG1655, McaS was more abundant in the LD medium than in M9-LG. Conversely, the McaS amount was comparable in C-1a cultures growing in either media ( Figure 4C).
The ∆pnp mutation had similar effects on CsrB, CsrC, and McaS in E. coli C and K-12 strains. It decreased CsrC and, to a lesser extent, McaS sRNAs abundance. Concerning CsrB, a nearly identical RNA pattern was found in E. coli C and K-12 ∆pnp strains, with a strong reduction in the full-length RNA and accumulation of shorter species ( Figure 4A and Table 3). Hybridization with oligonucleotides complementary to either the CsrB 5 -or the 3end (Supplementary Figure S3B) confirmed that these RNAs are CsrB degradation products mainly shortened at the 3 -end as already found in E. coli K-12 and Salmonella [50,51].
In ∆csrA and ∆csrA ∆pnp mutants, all sRNAs, and in particular CsrB and CsrC, were less expressed. Indeed, faint CsrB and CsrC signals corresponding to full-length transcripts were visible only upon long exposition of the filters ( Figure 4A and data not shown). These results were consistent with previous evidence showing that CsrA positively regulates csrB and csrC expression [6,10]. As for McaS, in contrast with our data, its level was reported to be similar in MG1655 and in its isogenic csrA::kan mutant [7]. To further check whether the McaS amount is modulated by CsrA, we analyzed the effect of ectopic csrA expression from a plasmid on McaS production in C-1a and MG1655. As shown in Figure 4D, McaS was more abundant in strains with plasmid pCSRA, which carries the csrA gene under the ptac promoter, than in those with the empty vector, and its amount further increased upon induction of csrA transcription with IPTG. Thus, CsrA positively controls the McaS level.

Compensatory Regulation of CsrB and CsrC Does Not Occur in E. coli C
According to literature data, the amount of CsrB and CsrC increases in MG1655 mutants with either csrC or csrB null mutations, respectively, a mechanism that compensates the lack of either sRNA by increasing the amount of the other one [6]. Consistent with these observations, we found around a 2.5-fold increase in CsrC in the MG1655 ∆csrB with respect to the csrB + . On the contrary, the CsrB amount was unchanged in the presence or absence of the csrC gene ( Figure 5A), showing that compensatory regulation takes place only for the ∆csrB strain in our experimental conditions. To assess whether this regulation also occurs in E. coli C, we analyzed CsrB and CsrC levels in C-1a ∆csrC and ∆csrB mutants, respectively. Surprisingly, CsrB decreased to 0.2 ± 0.04 in the ∆csrC strain and CsrC to 0.01 ± 0.02 in the ∆csrB mutant compared with their levels in C-1a ( Figure 5B). Thus, in E. coli C, the absence of CsrB or CsrC negatively affects the expression of the other one. In double ∆csrB ∆pnp or ∆csrC ∆pnp mutants, the expression of csrC and csrB, respectively, was similar to that found in the single ∆pnp mutant ( Figure 5B).
The level of csrA mRNA was two-fold higher in C-1a than in csrB or csrC defective mutants ( Figure 5C, left panel). As for E. coli K-12, the csrA mRNA, and in particular, the P3 transcript, was reduced by about 20% in the ∆csrB mutant, whereas no effect was observed in the ∆csrC ( Figure 5C, right panel).

Ectopically Expressed RNase II Restores CsrB and CsrC Production in C-1a ∆pnp
We previously found that ectopic expression of the rnb gene encoding RNase II from a plasmid suppressed auto-aggregation in C-1a ∆pnp. The suppression was specifically elicited by RNase II, as overexpression of rnr encoding the other E. coli exonuclease, namely RNase R, did not prevent aggregation [25]. We looked at CsrB and CsrC levels in strains overexpressing the two exonucleases to evaluate whether they have a different impact on the expression of these sRNAs.
As shown in Figure 6, we found that ectopic expression of all exonucleases caused the complete disappearance of signals corresponding to CsrB and CsrC degradation products, but only PNPase and RNase II partially restored CsrB and CsrC full-length production, whereas RNase R did not.
in the presence or absence of the csrC gene ( Figure 5A), showing that compensatory regulation takes place only for the ΔcsrB strain in our experimental conditions. To assess whether this regulation also occurs in E. coli C, we analyzed CsrB and CsrC levels in C-1a ΔcsrC and ΔcsrB mutants, respectively. Surprisingly, CsrB decreased to 0.2 ± 0.04 in the ΔcsrC strain and CsrC to 0.01 ± 0.02 in the ΔcsrB mutant compared with their levels in C-1a ( Figure 5B). Thus, in E. coli C, the absence of CsrB or CsrC negatively affects the expression of the other one. In double ΔcsrB Δpnp or ΔcsrC Δpnp mutants, the expression of csrC and csrB, respectively, was similar to that found in the single Δpnp mutant ( Figure  5B).
The level of csrA mRNA was two-fold higher in C-1a than in csrB or csrC defective mutants ( Figure 5C, left panel). As for E. coli K-12, the csrA mRNA, and in particular, the P3 transcript, was reduced by about 20% in the ΔcsrB mutant, whereas no effect was observed in the ΔcsrC ( Figure 5C, right panel).

Ectopically Expressed RNase II Restores CsrB and CsrC Production in C-1a Δpnp
We previously found that ectopic expression of the rnb gene encoding RNase II from a plasmid suppressed auto-aggregation in C-1a Δpnp. The suppression was specifically elicited by RNase II, as overexpression of rnr encoding the other E. coli exonuclease, namely RNase R, did not prevent aggregation [25]. We looked at CsrB and CsrC levels in strains overexpressing the two exonucleases to evaluate whether they have a different impact on the expression of these sRNAs.
As shown in Figure 6, we found that ectopic expression of all exonucleases caused the complete disappearance of signals corresponding to CsrB and CsrC degradation products, but only PNPase and RNase II partially restored CsrB and CsrC full-length production, whereas RNase R did not.  RNA samples (10 µg) were loaded on 6% polyacrylamide-urea gel, blotted onto a nylon membrane, and hybridized with radiolabeled oligonucleotides specific for the CsrB (PL208) and CsrC (FG2568) sRNAs (indicated on the left of the panels with the respective gene name). 5S, 5S rRNA used as the gel loading control.

Discussion
In this work, we show that CsrA is expressed at a low level in E. coli C because of impaired transcription caused by the insertion of IS3 in the csrA P4 promoter [14]. Transcripts starting at upstream promoters end presumably within the transposon, and only P5 mRNA is produced. Transposons are major drivers in evolution [52,53]. In fact, by integrating at multiple positions within a genome, they may stimulate genome rearrangements through homologous recombination. Moreover, they can have a deep impact on gene expression by inserting within coding or regulatory regions, which in turn may deeply affect bacterial physiology. The insertion of the IS3 transposable element into the csrA promoter of E. coli C may be considered a textbook case in this respect, as by downregulating csrA expression, it determines increased pgaABCD expression that, consequently, stimulates auto-aggregation and biofilm formation.
Not only CsrA but also the molecular decoys modulating its activity, namely the sRNAs CsrB, CsrC, and McaS, are expressed at different levels in C-1a with respect to MG1655. In particular, CsrB and CsrC are more than fifty-and six-fold more abundant, respectively, in MG1655 than in C-1a, whereas McaS is less abundant. This is not due to differences in their genes as the sequences of the csrB, csrC, and mcaS loci, including the intergenic 200 pb regions upstream overlapping their promoters, are identical between C-1a and MG1655. The same also applies to uvrY and barA genes that are involved in CsrB and C regulation (data not shown). It seems likely that low CsrB and CsrC expression in C-1a may be a consequence of low CsrA levels. In fact, csrA deletion almost completely abolishes csrB and csrC expression in E. coli C (and K-12), suggesting that the indirect transcriptional activation of CsrB and CsrC by CsrA operating in E. coli K-12 [6,10] is maintained in E. coli C.
Concerning mcaS, we found that it is expressed at comparable levels in C-1a and MG1655 cultures growing in LD broth, which contains very little glucose [49]. In M9-LG, which contains a higher glucose concentration, mcaS expression drops in MG1655 while remaining high in C-1a. Thus, the expression profile of McaS is consistent with its reported regulation by glucose [47,48] in MG1655 but not in C-1a, further highlighting differences in sRNA expression between the two strains. We do not have an explanation for the high McaS level in C-1a in the presence of glucose, which may be due to transcription activation by factors different than CRP-cAMP, which controls catabolite repression in E. coli [54]. It should be mentioned that also in E. coli K-12, glucose-dependent mcaS regulation seems to be only partially dependent on CRP [48]. Post-transcriptional mechanisms modulating McaS stability can also play a role in determining its expression profiles in the two strains. For instance, the csgD mRNA has a negative effect on mcaS expression, most likely because it pairs with McaS, and this stimulates McaS (and csgD mRNA) degradation [48]. In E. coli C, an IS5/IS1182 transposase gene replaces the csgD promoter and the first ca. 30 nt of the long 5 -UTR of the gene. Albeit the transposase gene is transcribed in the same direction as csgD [14], the csgD expression profile and csgD mRNA abundance are in all probability different in C-1a and MG1655, and this, in turn, may affect McaS.
We found that mcaS is positively regulated by CsrA, as its expression decreases in both C-1a and MG1655 ∆csrA mutants, whereas it increases upon CsrA ectopic expression from a plasmid. This result is in contrast with previous findings showing that McaS levels were comparable between MG1655 and its isogenic csrA::kan mutant [7]. Such discrepancy may be due to the leakiness of the csrA::kan allele, which could only marginally affect mcaS expression with respect to the ∆csrA mutation, and/or to differences in the experimental conditions in which mcaS expression was measured.
In C-1a ∆csrB or ∆csrC mutants, the csrA mRNA from the P5 promoter decreases. The mechanism responsible for this drop is not straightforward, and, unfortunately, we could not directly assess whether the reduction in csrA mRNA caused a correspondent decrease in CsrA protein because CsrA level in the wildtype C-1a was under the detection limit of the anti-CsrA antibodies. However, if there would be a drop in the CsrA level as well, this may explain a puzzling result that we published some years ago, namely that in C-1a ∆csrB or ∆csrC mutants, the pgaABCD operon expression and PNAG production increased [25]. Moreover, given the indirect transcriptional activation of csrB and csrC by CsrA discussed before, low CsrA may prevent transcription also of csrC and csrB genes in ∆csrB and ∆csrC mutants, respectively. This could explain why neither gene is expressed when one of them is deleted, thus abolishing compensatory regulation.
The role of PNPase in pgaABCD regulation may also be connected to csrA expression modulation. Indeed, PNPase protects CsrB and CsrC from degradation ( Figure 6). In doing so, PNPase may indirectly contribute to preserving csrA transcription efficiency. Consistent with this hypothesis, only RNase II, which restores CsrB and CsrC production, and not RNase R, which does not, prevent auto-aggregation [25]. Moreover, this interpretation may explain why the PNPase effect is much stronger in E. coli C than in K-12. In fact, C-1a contains less CsrA than MG1655, and this could make the Csr regulatory system less robust toward fluctuations of CsrA concentration.
Park and colleagues [46] showed that in vitro, CsrA binds the pnp mRNA and represses its translation and that a translational fusion encompassing the pnp promoter and 5 -UTR fused with the lacZ gene is activated in a csrA::kan mutant. However, in our experimental conditions (i.e., exponential cultures in M9-LG at 37 • C), neither the leaky csrA::kan or the ∆csrA mutations affect PNPase level, which is remarkably similar also between E. coli C and K-12 strains. Further analyses are required to assess in which conditions CsrA-dependent pnp regulation actually occurs in E. coli.

Conflicts of Interest:
The authors declare no conflict of interest.