A regulatory network involving Rpo, Gac and Rsm for nitrogen-fixing biofilm formation by Pseudomonas stutzeri

Biofilm and nitrogen fixation are two competitive strategies used by many plant-associated bacteria; however, the mechanisms underlying the formation of nitrogen-fixing biofilms remain largely unknown. Here, we examined the roles of multiple signalling systems in the regulation of biofilm formation by root-associated diazotrophic P. stutzeri A1501. Physiological analysis, construction of mutant strains and microscale thermophoresis experiments showed that RpoN is a regulatory hub coupling nitrogen fixation and biofilm formation by directly activating the transcription of pslA, a major gene involved in the synthesis of the Psl exopolysaccharide component of the biofilm matrix and nifA, the transcriptional activator of nif gene expression. Genetic complementation studies and determination of the copy number of transcripts by droplet digital PCR confirmed that the regulatory ncRNA RsmZ serves as a signal amplifier to trigger biofilm formation by sequestering the translational repressor protein RsmA away from pslA and sadC mRNAs, the latter of which encodes a diguanylate cyclase that synthesises c-di-GMP. Moreover, RpoS exerts a braking effect on biofilm formation by transcriptionally downregulating RsmZ expression, while RpoS expression is repressed posttranscriptionally by RsmA. These findings provide mechanistic insights into how the Rpo/Gac/Rsm regulatory networks fine-tune nitrogen-fixing biofilm formation in response to the availability of nutrients.


INTRODUCTION
The term 'biofilm' can be defined as a community of microbes adhering to biotic or abiotic surfaces that is protected from environmental stresses by a self-produced extracellular matrix 1,2 . The extracellular matrix, often referred to as extracellular polymeric substances, is composed of exopolysaccharides, proteins and extracellular DNA present in various concentrations depending on the bacterial species 3,4 . The biofilm state provides potential advantages over the planktonic state, including increased resistance to antimicrobial agents, protection from environmental stresses, and improved adaptation to nutrient deprivation 5 . Numerous investigations in recent decades have demonstrated that bacterial biofilm formation is a sequential process governed by complex regulatory networks that differ from one bacterial species to another 1,6 . It is now well accepted that microbial biofilms are the most widely distributed and predominant mode of life on Earth, influencing our lives tremendously in both positive and negative ways [6][7][8][9] .
In general, as established in the model bacterium Pseudomonas aeruginosa, biofilm development usually begins with attachment to a surface, followed by microcolony formation and production of the extracellular matrix responsible for the biofilm architecture [10][11][12][13][14] . Biofilm formation has been studied intensively in the genus Pseudomonas, with an emphasis on genetic elements and molecular mechanisms; Gac/Rsm, c-di-GMP signalling and quorum-sensing (QS) pathways were reported as the main mechanisms leading to biofilm formation 15,16 . The Gac/Rsm signalling pathway involves the GacS/GacA two-component regulatory system, the RNA-binding protein RmsA, and its cognate regulatory non-coding RNAs (ncRNAs) 17,18 . The GacS/GacA two-component system activates the transcription of one or several genes for Rsm ncRNAs, which contain multiple GGA motifs in exposed stem loops of their predicted secondary structures 19 . The GGA motifs allow Rsm ncRNAs to bind the RNA-binding proteins that act as global posttranscriptional repressors, e.g., CsrA (in Escherichia coli) and RsmA (in P. aeruginosa), controlling important cellular processes, such as secondary metabolism (e.g., metabolism of pyocyanine or the QS signal N-butyryl-homoserine lactone in P. aeruginosa), motility, and biofilm formation 17,20 . RsmA specifically recognises and binds to conserved GGA motifs in the 5′-untranslated region (5′-UTR) of target mRNAs, thereby preventing ribosome access and protein translation 17,21 . RsmA controls biofilm formation through direct repression of various target genes, such as pslA (involved in the synthesis of the exopolysaccharide Psl) and sadC (involved in c-di-GMP synthesis) 22,23 . As a key biofilm regulatory molecule, the second messenger c-di-GMP is synthesised by diguanylate cyclases (DGCs) that bear a GGDEF domain and is degraded by phosphodiesterases (PDEs) that harbour EAL or HD-GYP domains. P. aeruginosa encodes several DGCs and PDEs; for example, WspR/SadC/RoeA (DGC) and RocR/BifA (PDE), are absent in the P. stutzeri A1501 genome, except for SadC and BifA, which modulate the level of cdi-GMP and influence 'surface-associated behaviours' by controlling polysaccharide syntheses 16,[24][25][26][27] . The P. aeruginosa biofilm matrix contains several polysaccharide components, including alginate, pellicle (Pel) and Psl exopolysaccharides 28 . It has been shown that pslA is the first gene in the psl operon, which comprises 15 cotranscribed genes that are involved in the synthesis of Psl 29 . Although current data relating to the roles of Psl are limited, Psl is a critical component of the P. aeruginosa 1 biofilm matrix, which functions as a scaffold, holding biofilm cells together to initiate biofilm development 30 . In addition, evidence demonstrates that biofilm formation is controlled positively by RpoN but negatively by RpoS, suggesting global antagonism between RpoN and RpoS, although there are contradictory reports [31][32][33][34][35] .
Microbial biofilms are common on plant surfaces and have been associated with phytopathogenic infections and colonisation by nitrogen-fixing rhizobacteria 36,37 . Because of dynamically fluctuating conditions in the rhizosphere, the ability of diazotrophic bacteria to form nitrogen-fixing biofilms may confer many ecological advantages and thereby facilitate their physiological and metabolic adaptation to successfully survive in the rhizosphere, a nitrogen-limited environment. An early study compared biofilm formation by a nitrogen-fixing strain of Klebsiella pneumoniae with that of two other members of Enterobacteriaceae, Salmonella enteritidis and E. coli, and showed that the nitrogen-fixing strain formed the densest and most metabolically active biofilms 38 . Many nitrogen-fixing bacteria, such as those of the genera Rhizobium, Gluconacetobacter and Azospirillum, produce biofilms containing various exopolysaccharides [39][40][41][42] . For instance, Sinorhizobium meliloti produces two symbiosispromoting exopolysaccharides, succinoglycan and galactoglucan, which function in host specificity and participate in early stages of a host plant infection, biofilm formation, and, most importantly, protection from environmental stresses [43][44][45] . Azospirillum cells are also capable of forming biofilms on both abiotic surfaces and in association with host plants 46 . Previous studies have demonstrated that two response regulator proteins, TyrR and FlcA, were found to be involved in the transcriptional regulation of biofilm formation by A. brasilense Sp7 via the production of capsular polysaccharides 42,47 .
The root-associated bacterium P. stutzeri A1501 is a rare example of a Pseudomonas strain with nitrogen fixation ability 48 . P. stutzeri A1501 can survive in the soil, colonise the root surface, and endophytically invade the root tissues of host plants. During evolution, A1501 acquired a nitrogen fixation island with a nifspecific regulatory system from a diazotrophic common ancestor 48 . Similar to many other Pseudomonas species, the nitrogen regulatory cascade in A1501 comprises the AmtB-GlnK-NtrBC-RpoN global nitrogen regulation proteins and a set of regulatory ncRNAs that control the expression of nif genes and the consequent optimal nitrogen fixation in response to nutrient stress [49][50][51][52] . Comparative genomics analysis showed that A1501 does not possess the well-known QS systems and does not produce alginate, but it contains genes possibly involved in cellulose biosynthesis and an incomplete psl operon 4,48 . It was previously shown that a nonpolar mutation of the fleQ gene, encoding FleQ (the main regulator of flagella synthesis), impaired motility and root colonisation but enhanced biofilm formation by P. stutzeri A1501 53 . Additionally, Wang et al. investigated the effect of physiological conditions on the formation and architecture of nitrogen-fixing biofilms by P. stutzeri A1501 41 . However, the composition of the polysaccharide matrix remains unknown. To date, studies on biofilm formation by nitrogen-fixing rhizobacteria have focused on ecological, physiological and architectural analyses. Despite its importance to microbial adaptation and survival, there is surprisingly little information about the genetics of nitrogen-fixing biofilm formation.
In this work, physiological conditions leading to nitrogen-fixing biofilm formation by the root-associated bacterium P. stutzeri A1501 were further investigated. We found that conditions favouring biofilm formation differ between diazotrophic and non-diazotrophic P. stutzeri strains, although both strains contain the same set of regulatory genes involved in biofilm formation in other systems. Thus, we systematically characterised genetic elements and molecular mechanisms involved in nitrogen-fixing biofilm formation. Genome-wide identification of putative genes involved in biofilm formation and mutant construction led to the identification of a complex regulatory circuitry involving the alternative sigma factors RpoN and RpoS and the Gac/Rsm regulators, and to the proposal of a model that integrates multiple levels of positive and negative regulation.

RESULTS
Effect of carbon and nitrogen sources on biofilm formation and biofilm-based nitrogenase activity It was previously shown that when lactate was the sole carbon source, P. stutzeri A1501 tended to form biofilms rather than maintain a planktonic state under nitrogen-deficient conditions 41 . To further examine this behaviour, the ability of A1501 to form mature biofilms and fix nitrogen was assayed 48 h after inoculation using carbon sources other than lactate and different concentrations of NH 4 Cl. Among the carbon substrates tested at 50 mM, lactate was the best for both biofilm formation and nitrogen fixation (Fig. 1a, b). The ability of A1501 to form biofilms gradually decreased with increasing NH 4 + concentration ( Fig. 1c) but was enhanced with increasing lactate concentration (Fig. 1d). In addition,~35% of the maximum nitrogenase activity was observed in planktonic growth at a low lactate concentration (1.0 mM), but very low nitrogenase activity was detected in biofilm growth (Fig. 1e), indicating that biofilm cells were incapable of fixing nitrogen unless supplied with an adequately available carbon source. These results indicated that nitrogen-fixing biofilm growth requires a sufficient supply of carbon sources, as both biofilm formation and nitrogen fixation are energetically expensive and highly regulated processes 16,54 .
In general, the expression of nitrogenase genes can be completely inhibited by small extracellular concentrations of NH 4 Cl 49,54 . The nitrogenase activity of mature biofilms incubated without NH 4 + was determined after the addition of different concentrations of NH 4 Cl (0.1 to 0.5 mM). As shown in Fig. 1f, the detectable nitrogenase activity of the planktonic cells was limited in response to the addition of 0.1 to 0.2 mM NH 4 + . The nitrogenase activity of biofilm cells was much higher than that of planktonic cells at low NH 4 + concentrations (0.1 to 0.2 mM NH 4 + ). In addition, 0.5 mM NH 4 + caused a small reduction (31%) in nitrogenase activity of biofilm cells but almost total loss of that of planktonic cells. This result is in good agreement with the gene expression data obtained from planktonic and biofilm cells treated with different concentrations of NH 4 + ( Supplementary Fig. 1). A1501 biofilms were previously reported to fix nitrogen in the presence of oxygen, suggesting that the matrix could be a barrier for oxygen diffusion 41 . It also appears that the matrix could limit the inhibitory effect of NH 4 + on enzyme synthesis and activity.

Functional identification of genetic elements governing biofilm formation in A1501
The A1501 genome contains a set of nine genes (namely, gacA, rsmA, rsmY, rsmZ, sadC, bifA, pslA, rpoN, and rpoS (Fig. 2a)), which are present in other Pseudomonas species and encode proteins and ncRNAs known to play roles in biofilm formation 16 (Fig. 2b). As a working hypothesis, we suggest that 1501 may use regulatory mechanisms such as the c-di-GMP signalling and Gac/Rsm pathways common to other Pseudomonas species for biofilm formation. To gain insights into the potential roles of the nine genes in nitrogen-fixing biofilm formation, we monitored the gene transcription levels under nitrogen fixation conditions. In nitrogen-fixing biofilm cells, the expression of rsmZ showed the most dramatic increase (>120-fold), followed by that of rpoS (~5.0fold) compared with their expression in planktonic state cells (Fig.  2c). In particular, the relative expression of most of these genes was superior in nitrogen-fixing biofilms than in planktonic biofilms (Fig. 2c). Indeed, the expression of genes specific for nitrogen fixation that were used as controls, such as nifH and nifA, was increased by~35-fold and~4.0-fold, respectively, in nitrogenfixing biofilm cells compared to non-nitrogen-fixing biofilm cells (Fig. 2d). To establish whether these genes play a role in biofilm formation, we constructed a set of mutant strains and corresponding strains containing complementing plasmids or overexpressing the gene. We found that mutations in most of the selected genes affect biofilm formation by either decreasing or increasing biofilm formation (Table 1). These effects were not due to differences in planktonic growth, as the corresponding mutant strains displayed similar growth as wild-type (WT) A1501 in minimal medium K containing 20 mM NH 4 + and 50 mM lactate (Table 1). It should be noted that mutation of the Gac/Rsm pathway genes, such as gacA, rsmZ and RsmY, resulted in a partial but not total loss of nitrogen-fixing biofilm production (Table 1), suggesting the involvement of additional regulatory pathways. This assumption was further confirmed by measuring the biofilm phenotypes of the strains lacking either rpoN or pslA. As shown in Supplementary Table 1, an rpoN mutation caused an almost total loss of biofilm production under nitrogen fixation conditions; a similar phenotype was observed in a mutant lacking the pslA gene, indicating that both RpoN and PslA are essential for nitrogen-fixing biofilm formation in A1501. Interestingly, the nine genes cited above (shown in Fig. 2b) are present in P. stutzeri ATCC17588, a non-diazotrophic strain isolated from a clinical specimen. However, when we examined the effect of nitrogen availability on biofilm formation by this strain, we found that its physiological conditions favouring biofilm formation differ from those found in A1501. Under nitrogen-sufficient conditions, the ability of this strain to form biofilms gradually increased with increasing NH 4 + concentration, in contrast to the diazotrophic strain A1501, which tended to form biofilms under nitrogendeficient conditions (Fig. 1g). A reasonable explanation for this finding is that the mechanisms underlying biofilm formation differ between diazotrophic and non-diazotrophic P. stutzeri strains.
RpoN governs nitrogen-fixing biofilm formation via transcriptional activation of pslA and nifA Mutation of rpoN resulted in a dramatic decrease in biofilm formation (Fig. 3a), consistent with the observation that the rpoN mutant was severely impaired in exopolysaccharide production (Fig. 3b). Furthermore, rpoN mutation led to a total loss of nitrogenase activity (Fig. 3c), suggesting that rpoN has a major role by controlling both the nitrogen fixation ability and the biofilm polysaccharides. Furthermore, qRT-PCR analysis provided additional evidence showing that the rpoN mutation affected the expression of GacA/Rsm pathway genes to different extents (Fig.  3d). In nitrogen-fixing biofilm cells, the rpoN mutation led to a significant decrease in the expression of rsmA but an increase in activity under planktonic growth (with shaking at 220 r.p.m. and 0.5% oxygen) and biofilm growth conditions (without shaking in air). f Effect of NH 4 + concentration on the nitrogenase activity of the planktonic and biofilm cells in minimal medium K containing 50 mM lactate. g Effect of NH 4 + concentration on biofilm formation by diazotrophic and non-diazotrophic P. stutzeri strains in minimal medium K containing 50 mM lactate. Unlike A1501, the non-diazotrophic strain ATCC17588 favoured biofilm formation under nitrogen-sufficient conditions. Each error bar indicates the standard deviation of three independent experiments. Asterisks indicate statistical significance by one-way ANOVA with LSD multiple-comparison test: *p < 0.05; **p < 0.01. the expression of gacA and rsmZ, suggesting that RpoN can exert a negative effect on the expression of GacA/Rsm pathway genes. A similar phenomenon has also been reported in the P. aeruginosa rpoN mutant, where the expression of the gacA gene was significantly increased 55 . Moreover, pslA is the only gene among the nine genes studied (listed in Fig. 2b) to have an RpoN box-like element upstream of its transcription start site, suggesting RpoNdependent expression, which is also the case for the nifLA promoter ( Supplementary Fig. 2a, b). Indeed, it was determined by DNase I footprinting assays. As shown in Fig. 3e, f, RpoN protects a 22 bp DNA region (CGACGGCACGCGGTTTGCAAAA) of the nifLA promoter and a 27 bp DNA region (CCGGAGAGGCACGGTCGGAG-CAGGAGT) of the pslA promoter. Two regions overlap with the putative RpoN-binding site located at positions −12 to −24 from the transcription start. Taken together, these data suggest that the expression of both nifA and pslA genes is dependent on RpoN.
Psl is the major exopolysaccharide of the P. aeruginosa biofilm matrix encoded by the psl gene cluster, but a disruption of the first pslA gene of the cluster resulted in severe attenuation of Psl production 56 . The pslA gene found in A1501 encodes a UDPglucose lipid carrier sharing 64% amino acid sequence identity with that of P. aeruginosa PAO1 (Fig. 2b). Carbohydrate monomer composition analysis of the polysaccharides from the A1501 culture showed that they were composed of glucose, mannose, galactose, ribose and rhamnose (Table 2), while those of the pslA mutant culture were composed mainly of glucose. Furthermore, Fig. 2 Systematic investigation of the P. stutzeri A1501 genes possibly involved in biofilm formation. a Localisation of the gene clusters on a linear map of the chromosome. The 49 kb nif gene island (red box) is located on the chromosome (from PST1302 to PST1359). b Functional description. Identities (%) shared between the amino acid sequences of orthologous proteins from Pseudomonas aeruginosa PAO1 and P. stutzeri ATCC17588 are presented. c Relative expression levels of the selected genes in planktonic cells on medium K containing lactate and ammonium versus nitrogen-fixing biofilm cells on medium K containing lactate. d Relative expression levels of the selected genes in biofilm cells on medium K containing lactate and ammonium versus nitrogen-fixing biofilm cells on medium K containing lactate. Each error bar indicates the standard deviation of three independent experiments. Asterisks indicate statistical significance by one-way ANOVA with LSD multiple-comparison test: *p < 0.05; **p < 0.01. the total amount of exopolysaccharides from the pslA mutant culture was 33% of the amount observed for the WT strain, whereas the exopolysaccharide-producing ability of the pslAoverexpressing strain was~2.0-fold greater than that of the WT strain ( Fig. 3b), indicating that PslA is a key player in the production of Psl-like exopolysaccharides. Most interestingly, the pslA mutant completely failed to form biofilms under nitrogendeficient conditions but produced 64% of the biofilm that WT produced under nitrogen-rich conditions (Table 1). These results, together with the observation that pslA expression was significantly upregulated in nitrogen-fixing biofilm growth (Fig. 2d), favour an important role of pslA in Psl-like exopolysaccharide production and consequently in nitrogen-fixing biofilm formation. In addition, mutation of the nifA gene, encoding an activator of all nif genes, led to a complete loss of nitrogenase activity but had no effect on biofilm formation (Table 1). Our results indicate that RpoN-driven positive regulation at the transcriptional level is one of the key mechanisms used in diazotrophic P. stutzeri to govern nitrogen-fixing biofilm formation.
RsmA posttranscriptionally represses biofilm formation by binding pslA and sadC mRNAs The presence of two RNA-binding proteins belonging to the CsrA family appears to be common in pseudomonads, e.g., RsmA and RsmE of P. fluorescens or RsmA and RsmF of P. aeruginosa 57 ; however, only one gene encoding RsmA was found in the A1501 genome. In the biofilm model bacterium P. aeruginosa, RsmA exerts a negative effect on biofilm formation through posttranscriptional repression of the sadC and pslA genes 22,23 . As shown in Fig. 2b, the P. stutzeri rsmA, pslA, and sadC gene products share a high identity with orthologous proteins in P. aeruginosa. Analysis of mutant strains showed that deletions of each of the three genes affected the biofilm-forming ability of P. stutzeri A1501 to different extents ( Table 1). The deletion of rsmA significantly enhanced Planktonic growth for 12 h in minimal medium K amended with 50 mM lactate and 20 mM NH 4 Cl. All experiments were performed in three biological replicates, and the mean values (±standard deviation) are shown. Asterisks indicate statistical significance when compared to the wild-type control by one-way ANOVA with LSD multiple-comparison test: *p < 0.05; **p < 0.01. biofilm formation compared with WT, whereas overexpression of this gene reduced biofilm formation (Fig. 4a), suggesting that RsmA may negatively regulate biofilm formation in A1501. Interestingly, both deletion and overexpression strains were more affected in biofilm formation under nitrogen-rich conditions than under nitrogen-deficient conditions ( Table 1). As noted before, biofilm formation was strongly decreased in the pslA mutant under N limitation, which was correlated with a strong decrease in exopolysaccharide content (Fig. 3b). In addition, the sadC mutant displayed a moderate decrease in biofilm production (Fig. 4a), and its intracellular c-di-GMP level was reduced by approximately 50% compared with that of WT (Fig. 4b). In contrast, the intracellular c-di-GMP level was significantly increased in a mutant lacking bifA, which encodes a c-di-GMP-degrading phosphodiesterase but was decreased by bifA overexpression (Fig. 4b). Similar effects of bifA mutation and overexpression on biofilm formation were observed (Table 1).
P. stutzeri A1501 RsmA is predicted to be a protein of 61 amino acids, sharing~99% sequence identity with the P. aeruginosa PAO1 RsmA (Fig. 2b). Indeed, this high similarity is reflected at the level of the L and R regions in A1501 and P. aeruginosa RmsA as well as other Rsm homologue proteins, including Escherichia coli CsrA (Csr/Rsm family), and has been reported to be involved in mRNA binding 58 . Furthermore, the 5′-UTR regions of both pslA  (Fig. 4g, h). Our results indicate that the influence on biofilm formation observed for RsmA takes place through direct repression of pslA and sadC at the posttranscriptional level.
The RsmY and RsmZ ncRNAs competitively sequester RsmA away from its mRNA targets Among the ncRNAs identified in the A1501 genome, two sharing identity with RsmY and RsmZ are predicted to have multiple GGA motifs located on the single-stranded outer stem loops (Supplementary Fig. 3), which is characteristic of Rsm ncRNAs able to bind the RsmA protein 19 . As expected from MST experiments, both RsmY and RsmZ bind to RsmA, exhibiting Kd values of 55 ± 6 and 320 ± 51 nM, respectively (Fig. 4i, j), and binding is abolished if the GGA motif is mutated (Fig. 4k, l). Furthermore, single deletion of either rsmY or rsmZ caused a limited reduction in biofilm production (Fig. 4m), whereas deletion of both rsmY and rsmZ resulted in a significantly decreased biofilm biomass (~76%), which was similar to the effect (~79%) observed in a gacA mutant ( Fig. 4m and Supplementary Table 1). Consistent with the report that both Rsm ncRNAs are known to be under the control of GacA in P. aeruginosa 17 , the expression of both rsmY and rsmZ was strongly decreased in the A1501 gacA mutant (Fig. 4n). Moreover, both rsm genes possess a sequence corresponding to a conserved GacA-binding site in their promoters ( Supplementary Fig. 3), suggesting GacA-dependent activation of the two genes.
The transcription pattern of rsmZ greatly differed from that of rsmY. For example, RsmZ expression in biofilm cells was upregulated more than 120-fold over that in the planktonic state, in contrast to RsmY, whose expression was downregulated 2.0fold (Fig. 2c). Furthermore, the expression level of RsmZ with different carbon sources showed no significant differences, whereas the RsmY level increased ∼6.8-fold in biofilm cells grown on glucose compared to those grown on lactate, suggesting a significant induction by glucose (Fig. 4o). Overexpression of either rsmY or rsmZ led to increased biofilm production to different extents, while single mutations of rsmY and rsmZ resulted in decreased and increased levels of nitrogenase activities, respectively (Supplementary Table 1). These results suggest that RsmY and RsmZ have overlapping and different functions in A1501.
Using a highly precise and absolute nucleic acid quantification technique 59 , termed droplet digital PCR (ddPCR), we further assessed the absolute copy number of the RsmA/Z/Y pool during nitrogen-fixing biofilm development. At the time of inoculation, RsmZ was present at up to~55,000 copies per ng total RNA, much higher than RsmY and RsmA (Table 3). Most strikingly, the expression of RsmZ was upregulated rapidly (up to~150,000) at the early stage of biofilm formation and then downregulated remarkably during biofilm maturation, suggesting that RsmZ functions as a potent trigger for the initiation of biofilm formation. In contrast, RsmY expression showed only a small upregulation. This suggests that in A1501, RsmZ but not RsmY antagonizes the posttranscriptional repression exerted by RsmA during biofilm development.

RpoS negatively regulates biofilm formation via RsmZ under nitrogen-sufficient conditions
The involvement of the stress and stationary-phase sigma factor RpoS in biofilm formation led to conflicting data in different bacteria 31,32 . In P. stutzeri A1501, the rpoS gene is immediately downstream of rsmZ (Fig. 2a), and it was shown to be involved in the regulation of two ncRNAs, nfiS and nfiR, specifically induced under nitrogen fixation conditions; therefore, the expression of both genes was downregulated in an rpoS mutant. In the present work, we found that the rpoS mutant exhibited increased biofilm production compared to that of A1501 under nitrogen-sufficient conditions (Table 1), consistent with results previously described for the P. aeruginosa rpoS mutant 32 but different from the E. coli rpoS mutant showing decreased biofilm production 31 . Furthermore, ddPCR was used to measure the absolute transcription levels of the Gac/Rsm pathway genes during biofilm formation. As shown in Supplementary Table 4, the levels of GacA, RsmA, and RsmY showed no significant differences between the WT and rpoS mutant strains during biofilm development, whereas the expression of RsmZ in the rpoS mutant increased~5.0-fold at the mature stage of biofilm development relative to that of the WT strain. This effect is probably indirect, as no RpoS-binding site was identified in the promoter region of the rsmZ gene ( Supplementary Fig. 3b). These results, together with the fact that the level of gacA mRNA did not vary significantly in the rpoS mutant (Table 4), suggest that RpoS exerts an inhibitory effect on RsmZ expression in a GacAindependent manner. The ddPCR results also showed that the mRNA level of rpoS was very low in early-stage biofilm cells but was enhanced~7.0-fold in mature-stage biofilm cells (Fig. 5a). In addition, the rsmA mutation remarkably increased the expression of rpoS in early-stage biofilm cells, indicating a negative effect of RsmA on rpoS expression (Fig. 5a). Further analysis revealed the conserved RsmA-binding GGA motif in the 5′-UTR of rpoS mRNA, implying a direct interaction between rpoS mRNA and RsmA (Fig.  5b). This possibility was checked by MST measurements, which showed that RsmA directly binds to the 5′-UTR containing the GGA motif and the ribosome-binding site of rpoS mRNA (Fig. 5c,  d). These findings suggest a novel repression circuitry that finetunes biofilm development by modulating the timing and intensity of both RsmZ and RpoS expression.

DISCUSSION
Numerous studies have established that regulatory circuits governing the transition from planktonic to biofilm lifestyles are very complex and differ between Pseudomonas species, although common regulatory mechanisms such as the c-di-GMP signalling and Gac/Rsm pathways exist. On the other hand, the available literature on the regulatory mechanisms underlying biofilm formation by nitrogen-fixing bacteria is still very scarce. Here, we aim to fill this knowledge gap by elucidating the complex mechanisms for fine-tuning nitrogen-fixing biofilm formation. In view of the data reported, we propose that multiple signalling systems regulate nitrogen-fixing biofilm formation in the rhizosphere bacterium P. stutzeri A1501 (as depicted in Fig. 6), including the well-studied Gac/Rsm pathway at the posttranscriptional level, RpoN-driven positive regulation at the transcriptional level, and a RpoS-mediated repression circuit at both levels. The Gac/Rsm pathway is generally considered the main mechanism controlling biofilm formation in non-diazotrophic Pseudomonas 15,17 . Indeed, we have shown that deletions of each of the A1501 gac/rsm genes can positively or negatively affect biofilm formation, but rpoN, by controlling the transcription of nifA and plsA, is the only gene whose inactivation resulted in the poorest biofilm and a Nif-minus phenotype. These results suggest that RpoN-driven positive regulation at the transcriptional level is one of the key mechanisms underlying nitrogen-fixing biofilm formation, which may override the effect of the Gac/Rsm pathway in diazotrophic P.
stutzeri. An additional level of complexity is added to this regulatory system by the presence of two structurally and functionally similar ncRNAs, RsmY and RsmZ. The presence of multiple ncRNAs with structural similarity was reported in other systems, e.g., RsmX, RsmY, and RsmZ in P. fluorescens 60 and RsmW, RsmY, and RsmZ in P. aeruginosa 58 . These regulatory ncRNAs show similar secondary structures with numerous unpaired GGA motifs that act to sequester RsmA proteins from their targets, suggesting possible functional redundancy 20 . Since the effectiveness of ncRNA regulation is directly related to ncRNA abundance relative to their mRNA targets, this redundancy has been proposed to permit a more efficient and precise regulatory response by providing additional possibilities for integrating various signals into complex networks 18 . In the case of P. stutzeri A1501, a double rsmY/rsmZ mutation caused the same phenotypic effects on biofilm formation as those observed in the gacA mutant (Supplementary Table 1), suggesting that no additional Rsm ncRNAs participate in the activation of biofilm formation via the Gac/Rsm cascade in A1501. Moreover, the transcription rates of the rsmY and rsmZ genes in A1501 are clearly distinct; rsmZ is expressed at~100-fold higher levels than rsmY under biofilm growth conditions, and the level of RsmZ is very high during biofilm growth compared to planktonic growth. In addition, a quantitative assessment by ddPCR demonstrated that RsmZ showed a biofilm stagedependent pattern of expression with a significant increase during early stages of biofilm formation caused by transcriptional activation by GacA, which was followed by a decrease in mature biofilms. We thus propose that RsmZ rather than RsmY acts as a signal amplifier to trigger the phenotypic switch from the planktonic mode to the biofilm mode of growth. Although the biological role of RsmY is unclear at this stage, a very strong induction of rsmY expression by glucose was observed (Fig. 4o), implying that this ncRNA may be required for glucose-related metabolism in A1501. We also observed that single mutations of rsmY and rsmZ limit biofilm formation by A1501 but decrease and increase nitrogenase activities, respectively. These results suggest that both ncRNAs have overlapping functions in the regulation of biofilm formation but distinctive roles in the regulation of nitrogenase activity.
Our results from ddPCR experiments quantitatively show that RpoS is a mature stage-induced protein whose expression is downregulated by RsmA at the early stage of nitrogen-fixing biofilm formation. Similarly, Huertas-Rosales et al. identified rpoS as a target of Rsm proteins in RIP-seq experiments as an indication that RpoS regulation by Rsm proteins is direct 61 . In addition, RpoS was also found to be negatively regulated by RsmA in P. protegens Absolute transcription level is expressed as a number of copies per copies per ng total RNA, which was measured using droplet digital PCR. All experiments were performed in three biological replicates, and the mean values (± standard deviation) are shown. Asterisks indicate statistical significance when compared to the initial inoculation by one-way ANOVA with LSD multiple-comparison test: *p < 0.05; **p < 0.01. Absolute transcriptional level is expressed as a number of copies per copies per ng total RNA, which was measured using droplet digital PCR. All experiments were performed in three biological replicates, and the mean values (± standard deviation) are shown. Asterisks indicate statistical significance when compared to the wild-type control by one-way ANOVA with LSD multiple-comparison test: *p < 0.05; **p < 0.01. CHA0 62 . This led us to speculate that RpoS contributes to the significant reduction in RsmZ levels when RsmZ is not needed at a high level in mature biofilm cells, while RsmA posttranscriptionally decreases RpoS expression and prevents the repression of RsmZ exerted by RpoS when RsmZ is most needed at a high level in early-stage biofilm cells. In addition to RpoS, we also found that RpoN monitors global changes in gene expression that may lead to more complex effects on nitrogen-fixing biofilm formation. For example, rpoN mutation significantly increased the expression of rsmZ in nitrogen-fixing biofilm cells, implying that RpoN likely acts as a repressor in the regulation of the RsmZ level. At least part of this effect might be mediated by GacA, as described previously in P. aeruginosa 55 . This appears to contradict the enhanced expression of RsmZ in nitrogen-fixing biofilm cells. However, the stronger effect of the RsmZ mutation on biofilm formation under NH 4 + -rich conditions than under NH 4 + -deficient conditions suggests that additional repression of RpoN ensures accurate and economical but not consistently high expression of RsmZ since the Gac/Rsm pathway is not the dominant player in nitrogen-fixing biofilm formation.
The initiation of biofilm formation in P. aeruginosa has been correlated with high intracellular levels of c-di-GMP 16 . In general, high internal levels of c-di-GMP induce the production of adhesins and extracellular matrix components, which enable bacteria to form biofilms, whereas low c-di-GMP levels lead biofilm bacteria into dispersal to shift to a planktonic mode of growth 24 . The Gac/ Rsm cascade in P. aeruginosa is genetically linked to c-di-GMP through SadC, whose production is repressed by RsmA. We also observed a similar connection, but deletion of the sadC gene resulted in a strain that is partially defective in biofilm formation and c-di-GMP synthesis. This means that SadC likely contributes some but not all of the c-di-GMP under the conditions tested. Therefore, we can further infer that at least one other DGC in the A1501 genome can produce c-di-GMP. Indeed, the exact mechanism underlying c-di-GMP synthesis and biofilm formation in A1501 remains to be elucidated.
Phylogenetically close members of the Pseudomonas genus produce a wide diversity of exopolysaccharides, such as cellulose, Psl, and Pel 3 . The Psl polysaccharide, which is composed of mannose, glucose and rhamnose, was first described in P. aeruginosa 63 . Although research on Psl polysaccharides has been mostly conducted in P. aeruginosa, a number of psl gene clusters have been identified in several Pseudomonas strains 4 and recently the existence of a psl-like gene cluster has been reported in some environmental non-aeruginosa Pseudomonas species 64 . Furthermore, two P. fluorescens strains isolated from rotted bell pepper, were previously described to produce an exopolysaccharide composed of mannose, rhamnose, and glucose substituted with pyruvate and acetate 65 . Although the exact composition of the PlsA-dependent polysaccharide, tentatively referred to as the Psllike exopolysaccharide, is not yet established, the analysis of the glycosyl residues present in a pslA mutant suggests that the A1501 Psl-like exopolysaccharide contains mannose and galactose since both sugars were not found in the mutant ( Table 2). From this analysis, it can be deduced that A1501 Psl differs from P. aeruginosa Psl, which does not contain galactose 28 . As glucose is the main sugar produced by the pslA mutant, it is likely that A1501 produces cellulose, in agreement with the presence of a cluster of genes in its genome that are similar to the cellulose biosynthesis genes of P. putida KT2440 48 . In the most recent review, Herredia-Ponce et al. 63 stress the fact that the differences in polysaccharide composition depending on growth conditions may reflect better adaptation to specific environments due to the differential evolution that occurs in different niches.
In the case of P. aeruginosa, the Psl exopolysaccharide is known to be a key element at the early stage of biofilm formation and is regulated transcriptionally by RpoS 22,30 . Unlike what was observed in P. aeruginosa, we found that in A1501, the PlsA-dependent exopolysaccharide is essential for biofilm formation under nitrogen fixation conditions but not under nitrogen-sufficient conditions, in agreement with the fact that pslA transcription is RpoN-dependent. In addition to playing a major structural role in biofilms, Psl was further shown to have a signalling role in stimulating two DGCs, SiaD and SadC, to produce more of the intracellular second messenger molecule c-di-GMP 66 . A Pslmediated increase in c-di-GMP was observed to result in two-to threefold higher levels of pslA transcripts, ultimately increasing the production of Psl itself and forming a unique positive feedback regulatory circuit 66 . These observations led us to speculate that PslA may be a rate-limiting enzyme of Psl synthesis. To experimentally address this possibility, pLpslA was introduced into A1501, generating a strain overexpressing PslA. As predicted, this overexpression strain produces much more Psl than the wildtype strain (Fig. 3c).
Bacteria in biofilms are surrounded by an extracellular matrix, which can account for up to 90% of the biofilm biomass and create a microenvironment favourable for protecting cells against various stresses 3,67 . Biofilms may provide especially suitable conditions for nitrogen fixation, as this process is extremely sensitive to oxygen and rapidly inhibited by ammonia. An early study reported that the production of exopolysaccharides under N-limiting conditions may be a survival mechanism favouring the exclusion of oxygen and increasing nitrogenase activity 68 . In addition, biofilm formation enables A1501 to fix nitrogen under aerobic conditions by forming EPS-encased cysts to protect nitrogenase from oxygen 41 . In accordance with these previous results, we found that biofilm formation was enhanced under nitrogen-deficient and carbon-sufficient conditions, which favour nitrogen fixation. Interestingly, we also observed that biofilms displayed significant nitrogenase activity at a concentration of NH 4 + that completely abolished the nitrogenase activity of planktonic cells.
Nitrogen-fixing bacteria occur predominately in the rhizosphere, where carbon-rich root exudates can support the energy demands of the nitrogen fixation process, while microbial cell densities and microbial activities are the greatest, making nitrogen a key modulator of survival and competitiveness. The colonisation of the root rhizosphere is an essential step in the establishment of efficient nitrogen-fixing associations, and thus, understanding the mechanism of biofilm formation is of major interest. In the present work, we found that conditions favouring biofilm formation differ between the diazotrophic and non-diazotrophic P. stutzeri strains, although both strains contain the same set of Gac/Rsm and c-di-GMP signalling pathway genes, reflecting the differential evolution of their regulatory networks due to different physiologies and niches. We hypothesised that variations in biofilm phenotypes could be due to differences in transcriptional regulation. However, we found no significant differences in the putative promoter sequences of the genes listed in Fig. 2b between the two strains, suggesting that the mechanism that causes the biofilm phenotypes of the two strains to differ is much more complex than we initially believed. In addition, it is not surprising that with evolutionary optimisation in the rice rhizosphere, A1501 has evolved sophisticated regulatory networks to respond to multiple environmental cues and adapt to the environmental conditions of the rhizosphere. Of particular note is that RpoN, an alternative sigma factor typically associated with general nitrogen responses in bacteria, was found to act as a critical regulatory hub to activate the transcription of pslA and nifA, consequently forming a novel regulatory link between nitrogen fixation and biofilm formation. This regulation is probably more direct and efficient than the Gac/ Rsm regulatory cascades widely found in Pseudomonas, and is likely advantageous, especially when diazotrophs face competition from other species in nitrogen-limited environments, such as the rhizosphere. To our knowledge, this is the unique example of multiple regulatory networks governing the transition from the planktonic mode to the nitrogen-fixing biofilm mode, which may contribute to diazotrophic P. stutzeri being highly adaptable to nitrogen-poor environments and have implications for the control of biofilm-related interactions between diazotrophs and host plants. Our results provide a basis for understanding a regulatory mechanism including RpoN, RpoS, Gac and Rsm regulators that underlies nitrogen-fixing biofilm development and may be applicable to various diazotrophic species. As nitrogen-fixing bacteria are found ubiquitously in most ecosystems and widely used as biofertilizers worldwide, our systematic study of nitrogenfixing biofilms will be of both ecological and biotechnological importance. Fig. 6 Proposed regulatory model for the P. stutzeri RpoN/RpoS/ Gac/Rsm signal transduction systems controlling nitrogen-fixing biofilm formation at multiple levels in response to the availability of nutrients. In this model, RpoN plays a central role and may be considered a 'hub' to bridge nitrogen fixation and biofilm formation by activating the transcription of the pslA and nifA genes under nitrogen-deficient and carbon-sufficient conditions. RsmZ was upregulated rapidly at the early stage of biofilm formation and then downregulated remarkably during biofilm maturation, thereby acting as a potent trigger for the initiation of biofilm formation. During biofilm development, RpoS exerts a braking effect on biofilm formation by transcriptionally downregulating RsmZ expression at the mature stage; this effect is restrained by RsmA at the early stage, thereby resulting in a novel repression circuit. Additionally, RpoN likely acts as a repressor of the Gac/Rsm pathway, markedly increasing the complexity of the regulatory circuitry. Arrows and T-shaped bars indicate positive and negative regulation, respectively. Broken lines indicate direct or indirect regulations for which evidence exists but that need to be studied in further detail. The black tail arrow indicates biochemical conversion reactions. The involvement of c-di-GMP in the biosynthesis of various polysaccharides has not been demonstrated experimentally, as marked by the punctuated tail arrow. For details, refer to the text.

METHODS Bacterial strains, plasmids and growth conditions
The bacterial strains and plasmids used in this study are listed in Supplementary Table 1. P. stutzeri A1501 and its derivatives were grown on LB medium or minimal medium K (containing 0.4 g l −1 KH 2 PO 4 , 0.1 g l −1 K 2 HPO 4 , 0.1 g l −1 NaCl, 0.2 g l −1 MgSO 4 ·7H 2 O, 0.01 g l −1 MnSO 4 ·H 2 O, 0.01 g l −1 Fe 2 (SO 4 ) 3 ·H 2 O, and 0.01 g l −1 Na 2 MoO 4 ·H 2 O, pH 6.8) supplemented by the desired carbon and nitrogen sources at concentrations indicated in the text. Unless stated otherwise, growth experiments were conducted using medium K containing NH 4 Cl (20 mM) and sodium lactate (50 mM) as the sole nitrogen and carbon sources at 30°C under vigorous shaking in a water-bath shaker. For measurements of biofilm formation, nitrogenase activity and gene expression, the concentrations of carbon substrates were adjusted to 50 mM. Antibiotics were used at the following concentrations: 50 μg ml −1 ampicillin (Amp), 50 μg ml −1 kanamycin (Km), 10 μg ml −1 tetracycline (Tc), 34 μg ml −1 chloromycetin (Cm), and 20 μg ml −1 gentamicin sulfate (Gm).

Constructions of mutants, complementing plasmids and overexpression strains
The plasmids and oligonucleotide primers used in this study are listed in Supplementary Tables 1 and 2, respectively. Strains and plasmids were constructed using conventional techniques. Nonpolar insertion mutant strains (e.g., rsmA) were generated by homologous suicide plasmid integration as described previously 50 . Appropriate oligonucleotide primers were designed to generate amplicons that were cloned into pK18mob as a vector 69 , and the resulting plasmids were introduced into A1501 by triparental mating using pRK2013 70 , generating the mutant strains. Correct recombination was confirmed by PCR followed by nucleotide sequencing of the amplicons obtained.
To generate nonpolar deletion mutant strains, amplification of DNA fragments located upstream and downstream of the target gene was performed using the appropriate primer sets upF/upR and downF/downR (Supplementary Table 2). Then, both amplicons and a DNA fragment containing the chosen resistance cassette gene were fused, and the resulting fragment was cloned into the pK18mobsacB vector, as depicted in Supplementary Fig. 4. The resulting plasmid was then introduced into A1501 by triparental mating as described above, and double recombination was selected on the basis of sucrose resistance. Correct recombination was validated by PCR and sequencing using primers testF and testR.
The complemented and overexpression strains were constructed using the broad host plasmid pLAFR3. A DNA fragment containing a WT gene (e.g., gacA) with its promoter and terminator was amplified from genomic DNA of A1501 and cloned into pLAFR3. The resulting complementing plasmid was then introduced into the WT or mutant strain by triparental mating, generating overexpression and complemented strains, respectively (Supplementary Table 1). The gene expression levels of the overexpression strains were confirmed to be higher than those of the WT using qRT-PCR.
The mutant with both rsmZ and rsmY deleted was constructed using ΔrsmZ as the starting strain. Briefly, a 1154 bp fragment containing the Gm resistance cassette located between the upstream and downstream DNA fragments of rsmY was generated by overlap extension PCR, doubledigested with BamHI/HindIII, and then cloned into the BamHI/HindIII site of pK18mobsacB. The resulting plasmid, pK18rsmY, was introduced into the genome of ΔrsmZ by triparental mating and double recombination was selected on the basis of sucrose resistance. Correct recombination in the resulting ΔrsmZ ΔrsmY double mutant was checked by PCR using the primers M-rsmY(up)-F and M-rsmY(down)-R (Supplementary Table 2), followed by nucleotide sequencing of the obtained PCR products. The resulting double deletion mutant (Supplementary Table 1) was used for further study.

Biofilm formation assays
Surface-adhered biofilm formation was assayed using the crystal violet (CV) method and performed in 96-well microtiter plates. Strains used for biofilm experiments were grown overnight in LB at 30°C. Cultures were centrifuged and diluted to a final OD 600 of 0.2 in fresh minimal medium K containing different carbon sources with or without 6 mM NH 4 Cl. Two hundred μl of each culture was aliquoted into separate wells in a 96-well PVC plate. Microtiter plates were carefully wrapped using parafilm and placed in a 30°C incubator without agitation for 12 or 48 h. In this study, the so-called early-and mature-stage biofilms were defined as biofilms formed 12 and 48 h after inoculation, respectively. At the indicated points, nonadhered planktonic cells were removed using a multichannel pipette without disturbing the biofilm area, and individual wells were washed twice with 160 μl of sterile double-distilled H 2 O. Then, 160 μl of 0.1% CV solution in ethanol was added to each well for 10 min and washed four times with 200 μl of ddH 2 O. Photos were taken, and the cell-associated CV was solubilized with 30% acetic acid and quantified by measuring the OD 560 of the resulting solution using a spectrophotometer (Thermo Scientific).

Nitrogenase activity assays
Nitrogenase activity was determined according to a previous protocol with modifications 71 . To examine the nitrogenase activity of cells grown planktonically, cells from an overnight culture in LB medium were centrifuged and resuspended in a 60 ml flask containing 10 ml of minimal N-free and lactate-containing medium at an OD 600 of 0.1. The suspension was incubated for 4 h at 30°C with vigorous shaking under an argon atmosphere containing 0.5% oxygen, and then 10% acetylene was added. Gas samples (0.25 ml) were taken at regular intervals (4, 6, 8, and 10 h) to determine the amount of ethylene produced. Samples were analysed on a polydivinylbenzene porous bead GDX-502 column using an SP-2100 gas chromatograph fitted with a flame ionisation detector (Beijing Beifen-Ruili Analytical Instrument Co., Ltd.). The ethylene content in the gas samples was determined by reference to an ethylene standard.
To determine the biofilm-based nitrogenase activity, strains used were grown overnight in LB at 30°C. Cultures were centrifuged and diluted to a final OD 600 of 0.2 in a 60 ml flask containing 10 ml of minimal NH 4 + -free and lactate-containing medium. The suspension was incubated for 48 h at 30°C under static conditions in air, and then 10% acetylene was added. Gas samples (0.25 ml) were taken at regular intervals (4, 6, 8, and 10 h) to determine the amount of ethylene produced. A 4 h incubation time was chosen for qRT-PCR or ddPCR assays of biofilm-related gene expression.
To examine the effect of NH 4 + on nitrogenase activity, both 10% acetylene and ammonium at different concentrations were added to 60 ml flasks containing suspensions of either biofilm or planktonic cells, and then gas samples (0.25 ml) were taken at regular intervals (4, 6, 8, and 10 h) to determine the amount of ethylene produced using the same method as described above. The nitrogenase activity was expressed as nmol ethylene min −1 mg −1 protein. Protein concentrations were determined using the Bio-Rad protein assay reagent kit (Bradford, Bio-Rad).

RNA isolation and qRT-PCR assays
Total RNA was isolated with an innuPREP RNA Mini Kit (Analytik Jena) according to the manufacturer's instructions. For quantification of gene expression, total RNA was reverse transcribed using random primers and the High Capacity cDNA Transcription Kit (Applied Biosystems) according to the manufacturer's instructions. PCR was carried out with Power SYBR Green PCR Master Mix on an ABI Prism 7500 Sequence Detection System (Applied Biosystems) according to the manufacturer's recommendations. The 16S rRNA gene was used as the endogenous reference control, and relative gene expression was determined using the comparative threshold cycle 2 −ΔΔCT method. Data were analysed using ABI PRISM 7500 Sequence Detection System Software (Applied Biosystems). Primers were designed based on the full genome sequence of P. stutzeri A1501, and they are listed in Supplementary Table 2.
Absolute quantification of RNA copy number by droplet digital PCR (ddPCR) Total RNA isolation and reverse transcription were performed as described above for qRT-PCR. Quantification by ddPCR was carried out in 20 μl reactions containing 10 μl of QX200 ddPCR EvaGreen SuperMix, 250 nM each commercial probe, 900 nM specific commercial primers, and 1 μl of cDNA according to the manufacturer's recommendations. A negative control contained sterile double-distilled water only. Emulsified 1 nl reaction droplets were generated using a QX100 droplet generator (Bio-Rad) and a droplet generator DG8 cartridge (Bio-Rad) containing 20 μl of reaction mixture and 70 μl of ddPCR droplet generation oil (Bio-Rad) per well. Thirty-five μl of the generated droplet emulsions was transferred to 96-well PCR plates that were then heat-sealed using foil sheets. Target DNA amplification was performed by thermal cycling of the droplet emulsions as follows: initial denaturation at 95°C for 10 min; 40 cycles of 94°C for 30 s and 60°C for 1 min; and then 98°C for 10 min. The fluorescence of each thermal cycled droplet was measured using a QX100 droplet reader (Bio-Rad). Data were analysed using QuantaSoft software (Bio-Rad) after setting a threshold using the fluorescence of negative controls.

5′ Rapid amplification of cDNA ends to determine transcriptional start sites
The transcriptional start site of the six target genes (pslA, nifLA, rpoS, sadC, rsmZ, and rsmY) was determined using the rapid amplification of cDNA ends (5′ RACE) method (Invitrogen) following the manufacturer's instructions. Briefly, the first-strand cDNA was synthesised using the primer GSP1, which was specific for the target gene sequence. The purified cDNA was tailed with dCTP by terminal deoxynucleotidyl transferase. PCR amplification was performed using the sequence-specific primer GSP2 and the anchor primer AAP. Primers GSP1 and GSP2, specific for the target gene tested here, are listed in Supplementary Table 2. The 5′ RACE products were cloned into the pGEM-T Easy vector (Promega) and sequenced to map the 5′ end of the transcript.

Expression and purification of RsmA for microscale thermophoresis (MST) measurements
The RsmA protein was expressed and purified using the IMPACT TM (Intein Mediated Purification with an Affinity Chitin-binding Tag) system according to the manufacturer's instructions (New England Biolabs). To this end, a fragment of rsmA was amplified by PCR using the pTWIN1-rsmA-F/R primers (Supplementary Table 2). The PCR product was digested with NdeI and EcoRI and ligated into the protein expression vector pTWIN1, which had been digested with the same enzymes. The resulting plasmid (named pTWIN1-RsmA) was introduced into the E. coli BL21 (DE3) strain. Overproduction of the RsmA-intein fusion was induced by the addition of 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) to cells grown to mid-log phase (OD 600 of 0.6). Cleavage of the RsmA-intein fusion was induced by equilibrating the chitin beads in buffer B3 (1 mM EDTA, 40 mM DTT, 20 mM Tris-HCl, 500 mM NaCl, pH 8.5) overnight. The untagged RsmA was eluted from the column and dialysed against 20 mM Tris-HCl and 500 mM NaCl, pH 8.0. The purity of the protein was as high as 90%, as judged by SDS-Tris-glycine PAGE. RsmA was quantitated using the Bio-Rad protein assay reagent kit (Quick Start Bradford, Bio-Rad) and stored at −80°C for further use.

MST measurements to determine interactions between the RsmA protein and its target RNAs
MST experiments were performed as previously described 72 . DNA templates carrying rsmY, rsmZ, sadC, pslA, and rpoS bearing point mutations within the GGA sequences were amplified using the WT or mutagenic primers (Supplementary Table 2). The following transcripts were synthesised from PCR-generated templates by GenePharma using a MAXIscript kit (Thermo Fisher): the full-length WT or mutated ncRNAs (RsmY and RsmZ) and the WT or mutated oligonucleotides containing the 5′-UTR and the first 50 nucleotides of the coding sequences of sadC, pslA, and rpoS mRNAs. The RsmA protein was labelled with NT-647-NHS dye using the Monolith Labelling Kit RED-NHS (no. MO-L011, NanoTemper Technologies) according to the manufacturer's instructions. For the measurements, the concentration of the labelled RsmA protein was kept constant (20 nM), while the concentrations of non-labelled ssRNA oligonucleotides varied from 0.3 nM to 10 μM. The binding reactions were carried out in MST buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 10 mM MgCl 2 ) supplemented with 0.1% Tween. The reactants were initially incubated at 37°C for 30 min to enable ssRNA binding with RsmA. The samples were then loaded onto NT.115 standard capillaries (no. MO-K002, NanoTemper Technologies). The measurements were carried out at 25°C with 40% excitation power and medium MST power. Data analyses were performed with NanoTemper Analysis software (NanoTemper Technologies).

Purification of His-tagged RpoN for DNase I footprinting assays
The pET-28a expression system (Novagen) was used to produce Cterminally His-tagged RpoN within host E. coli BL21 (DE3) cells. A fragment of rpoN was amplified by PCR using the pET28a-rpoN-F/R primers (Supplementary Table 2). The PCR product was digested with NdeI and HindIII and ligated into the protein expression vector pET28a, which had been digested with the same enzymes. The resulting plasmid (pET28a-RpoN) was introduced into the E. coli BL21 (DE3) strain. An overnight culture of BL21 (DE3) harbouring the expression plasmid was used to inoculate LB medium containing the appropriate antibiotics. This cell culture was incubated with shaking at 37°C until the OD 600 was 0.6-0.9, at which point production of His-tagged RpoN was induced by the addition of IPTG to a final concentration of 0.1 mM. His-tagged RpoN was purified using a Ni-NTA Fast Start Kit (Qiagen, Venlo, Netherlands) according to the manufacturer's instructions. The purity of the protein was as high as 90%, as judged by SDS-Tris-glycine PAGE. His-tagged RpoN was quantitated using the Bio-Rad protein assay reagent kit (Quick Start™ Bradford, Bio-Rad) and stored at −80°C for further use.

DNase I footprinting assays
DNase I footprinting assays were performed by Tolo Biotech according to a method previously described 73 . The DNA probe was prepared by PCRamplifying a 296-bp nifA promoter region using the primers FP-nifA-F/FP-nifA-R and a 279-bp pslA promoter region using the primers FP-pslA-F/FP-pslA-R (Supplementary Table 2). For each assay, 300 ng probes were incubated with 0.2 μg purified protein RpoN in a total volume of 40 µl. After incubation for 30 min at 30°C, 10 µl solution containing~0.015 units DNase I (Promega) and 100 nmol freshly prepared CaCl 2 was added and further incubation was performed at 37°C for 1 min. The reaction was stopped by adding 140 µl DNase I stop solution (200 mM unbuffered sodium acetate, 30 mM EDTA and 0.15% SDS). Samples were first extracted with phenol/chloroform, and then precipitated with ethanol. Pellets were dissolved in 30 µl MiniQ water. The preparation of the DNA ladder, electrophoresis and data analysis were the same as previously described 73 , except that the GeneScan-LIZ600 size standard (Applied Biosystems) was used.

Exopolysaccharide isolation and carbohydrate monomer composition analysis
Bacterial aerated cultures were grown in minimal medium K containing 50 mM lactate and 6.0 mM NH 4 Cl. Then, the cultures were centrifuged. The supernatants were collected and concentrated to 50 ml with a CentriVap concentrator (Labconco, USA). Exopolysaccharides were isolated from culture supernatants by the addition of two volumes of chilled absolute ethanol, and then proteins were removed from the exopolysaccharides by protease hydrolysis. After further precipitation by cold absolute ethanol and lyophilization, purified exopolysaccharides were obtained. The exopolysaccharide content of samples was monitored quantitatively by using the phenol-sulfuric acid method and was further normalised to the total cell protein remaining after extraction, as determined using the Bio-Rad protein assay reagent kit (Quick Start™ Bradford, Bio-Rad). Final concentrations were expressed as mg exopolysaccharide per g bacterial protein.
Carbohydrate monomer composition analysis was conducted on the exopolysaccharide samples from the A1501 and ΔpslA mutant strains at Beijing Ketian Technology Co., Ltd. Briefly, an exopolysaccharide sample (2.0 mg) was dissolved in 5 ml of trifluoroacetic acid (2 M) and subsequently hydrolysed at 120°C for 4 h. After repeated rotary evaporations to completely remove the trifluoroacetic acid, the sample was dissolved in 2 ml of deionized water. The hydrolysate (100 µl) was placed in a separate tube with 10 µl of deuterium-labelled succinic acid (1.5 mg ml −1 ) as the internal standard and then lyophilised. Methyl glycosides were prepared from the dry sample by suspension in 50 µl of methoxyammonium hydrochloride/pyridine solution (20 mg ml −1 ) at 40°C for 80 min. The resulting sample was then per-O-trimethylsilylated with N-methyl-N-(trimethylsilyl) trifluoroacetamide (80 µl) in a water-bath pot at 40°C for 80 min. The sample was centrifuged at 12,000 r.p.m. for 5 min. The supernatant fraction was filtered through a 0.22 µm vacuum filter and collected for glycosyl composition analysis. Samples were analysed using a gas chromatograph coupled to a mass selective detector (7890 A/5975 C MSD; Agilent Technologies, Inc.) equipped with an HP-5 (30 m by 0.32 mm, 0.25-µm film thickness; SGE Analytical Science) capillary column. The injector and detector temperatures were 250°C and 240°C, respectively. The column pressure was kept at 0.10 MPa, with a 1.0 ml/min carrier gas (N 2 ) flow rate. The chemical compounds were identified using a mass spectral library (NIST 08) and Agilent GC-MS Workstation software. Data from three biological replicates were analysed for each strain.