Cyclic-di-GMP-mediated signalling within the σS network of Escherichia coli


*E-mail; Tel. (+49) 30 838 53119; Fax (+49) 30 838 53118.


Bis-(3′−5′)-cyclic-di-guanosine monophosphate (c-di-GMP) is a bacterial signalling molecule produced by diguanylate cyclases (DGC, carrying GGDEF domains) and degraded by specific phosphodiesterases (PDE, carrying EAL domains). Neither its full physiological impact nor its effector mechanisms are currently understood. Also, the existence of multiple GGDEF/EAL genes in the genomes of most species raises questions about output specificity and robustness of c-di-GMP signalling. Using microarray and gene fusion analyses, we demonstrate that at least five of the 29 GGDEF/EAL genes in Escherichia coli are not only stationary phase-induced under the control of the general stress response master regulator σS (RpoS), but also exhibit differential control by additional environmental and temporal signals. Two of the corresponding proteins, YdaM (GGDEF only) and YciR (GGDEF + EAL), which in vitro show DGC and PDE activity, respectively, play an antagonistic role in the expression of the biofilm-associated curli fimbriae. This control occurs at the level of transcription of the curli and cellulose regulator CsgD. Moreover, we show that H-NS positively affects curli expression by inversely controlling the expression of ydaM and yciR. Furthermore, we demonstrate a temporally fine-tuned GGDEF cascade in which YdaM controls the expression of another GGDEF protein, YaiC. By genome-wide microarray analysis, evidence is provided that YdaM and YciR strongly and nearly exclusively control CsgD-regulated genes. We conclude that specific GGDEF/EAL proteins have very distinct expression patterns, and when present in physiological amounts, can act in a highly precise, non-global and perhaps microcompartmented manner on a few or even a single specific target(s).


The cyclic nucleotide bis-(3′−5′)-cyclic-di-guanosine monophosphate (c-di-GMP) was recently recognized as a signalling molecule nearly ubiquitously used by bacteria (D'Argenio and Miller, 2004; Jenal, 2004; Römling et al., 2005). C-di-GMP is synthesized from two GTP molecules by diguanylate cyclases (DGC) and is degraded by c-di-GMP-specific phosphodiesterases (PDE), both of which occur in soluble or membrane-associated variants. DGC activity is associated with GGDEF (or GGEEF) domains (Chan et al., 2004; Paul et al., 2004; Ryjenkov et al., 2005), whereas c-di-GMP PDE activity has been shown for EAL domain proteins (Chang et al., 2001; Christen et al., 2005; Schmidt et al., 2005; Tamayo et al., 2005). These domains, which were named for their most obvious conserved amino acid sequence motifs, can occur either alone or in combination in a single polypeptide, and their activities seem to be controlled by various N-terminal sensory domains (including PAS, two-component receiver, GAF, BLUF and other domains) (Jenal, 2004; Römling et al., 2005). GGDEF and EAL domains occur in all branches of the bacterial kingdom, and belong to the largest domain families with poorly understood molecular and physiological function in the protein data bases (Galperin et al., 2001; Römling et al., 2005).

The very first reports of c-di-GMP associated this signalling molecule with the control of cellulose synthesis in Gluconacetobacter xylinus (Ross et al., 1987; Weinhouse et al., 1997; Tal et al., 1998). More recently, the Salmonella GGDEF protein AdrA (which corresponds to YaiC in Escherichia coli) was also shown to activate the biosynthesis of cellulose, which provides an extracellular matrix in so-called ‘rdar’ (or ‘wrinkled’ or ‘rugose’) colonies, which represent a highly structured biofilm morphotype visible on agar plates (Römling et al., 2000; Zogaj et al., 2001). In general, functions that contribute to biofilm formation, are positively regulated by c-di-GMP, while motility is downregulated by c-di-GMP, as can be shown conveniently by overproducing GGDEF or EAL proteins (Ausmees et al., 1999; D′Argenio et al., 2002; García et al., 2004; Kirillina et al., 2004; Simm et al., 2004; Tischler and Camilli, 2004; Lim et al., 2006). In Caulobacter crescentus, the c-di- GMP-producing GGDEF protein PleD is required for flagellum ejection and differentiation of a swarmer cell into a stalk cell during the cell cycle (Aldridge et al., 2003). Moreover, PleD is recruited to the site of action, i.e. the prospective stalk pole of the cell which points to the possibility that in some cases, c-di-GMP synthesis and action may be locally confined (Paul et al., 2004). For a number of pathogenic bacteria, evidence is also accumulating that GGDEF/EAL proteins are involved in the control of virulence genes (Brouillette et al., 2005; Hisert et al., 2005; Tischler and Camilli, 2005). In all these examples, however, the direct molecular target or effector component for the signalling molecule c-di-GMP has not been identified.

Another intriguing observation is that most bacteria have multiple GGDEF and EAL genes, sometimes up to several dozens (e.g. E. coli has 19 GGDEF and 17 EAL genes, with an overlap of seven genes, which encode both domains in a single polypeptide). This raises a number of interesting questions. In such a context, how can a single GGDEF or EAL protein significantly affect a certain c-di-GMP-dependent function at all? How can the cellular c-di-GMP level be a robustly controlled parameter in view of the normal intrinsic and extrinsic noise in the expression and activities of possibly dozens of DGC and c-di-GMP PDE? A solution to this problem may be sequestration. This could be temporal regulatory sequestration, i.e. expression and activity of GGDEF/EAL proteins should be tightly controlled, such that only a few of them are present and active at the same time. Sequestration could also be spatial, i.e. some GGDEF/EAL proteins (in particular proteins that contain both domains in the same polypeptide) may act locally, e.g. in the microcompartment of a larger complex. Seen from the ‘downstream perspective’, some target processes might be controlled by the cellular c-di-GMP level, whereas other target components may be locally affected by being part of a larger GGDEF/EAL complex.

The present study started with a genome-wide microarray analysis (Weber et al., 2005) that indicated that the expression of seven of the 29 GGDEF/EAL genes in E. coli significantly depends on σS (RpoS), the master regulator of the general stress response. Thus, these genes are over-represented in the σS regulon (which comprises about 10% of all genes in the E. coli genome), suggesting a physiological role for c-di-GMP during the general stress response and/or in post-exponential cells. Moreover, these co-regulated GGDEF/EAL proteins may establish a c-di-GMP signalling subnetwork within the σS network, which allows to address questions of specificity, target processes and potential microcompartmentation of c-di-GMP signalling within a relatively well-known physiological context. Here, we present specific and differential expression patterns of these σS-controlled GGDEF/EAL genes, that indicate an important role for c-di-GMP in stationary phase cells. We identify a target process specifically and antagonistically controlled by a distinct pair of σS-dependent GGDEF/EAL proteins. By microarray analysis, we show that these two GGDEF/EAL proteins precisely control the regulatory output of a single target. Moreover, the σS network even contains a GGDEF cascade, with one such protein controlling the expression of another one in a temporally regulated manner.


Expression patterns of σ S-dependent GGDEF/EAL genes

As demonstrated by genome-wide transcription analysis on microarrays, approximately 10% of the genes in the E. coli genome are under significant direct or indirect control of σS when assayed under various stress conditions. Of these genes, again about 10% encode regulatory or signal transducing proteins which indicates a complex cascade architecture of the σS network (Weber et al., 2005). Among this latter group we identified seven GGDEF/EAL genes (ratios of expression in rpoS+ and rpoS::Tn10 strains for these seven genes are given in Table 1, for all 29 GGDEF/EAL genes in Table S1). The GGDEF genes ydaM, yddV and yedQ as well as yciR, which encodes both GGDEF and an EAL domains, exhibited σS dependence under more than one stress condition, whereas the GGDEF gene yeaI and the EAL genes ycgG and ydiV showed σS control only under one specific condition. This may indicate that these genes are expressed only in a very specific context. None of the 29 GGDEF/EAL genes exhibited significant negative regulation by σS.

Table 1.  Ratios of expression of significantly σS-dependent GGDEF/EAL genes in rpoS+ and rpoS::Tn10 strains as determined by genome-wide microarray analysis.a
Geneb-NumberGGDEFEALExpression ratios (rpoS+/rpoS::Tn10)
LB/OD 4+0.3 M NaClpH 5
  • a. 

    For details of the procedure and statistical analysis, see Weber et al. (2005). Genes and ratios indicative of significant σS control are listed in boldface. A full list of all 29 GGDEF/EAL genes in E. coli is given in Table S1.

  • n.d., non-detectable, i.e. absolute fluorescence intensities on the microarrays were below reliable detection.

ycgGb1168 x1.362.461.55
ydaMb1341x 6.749.543.15
yddVb1490x 2.013.331.39
ydiVb1707 xn.d.1.043.47
yeaIb1785x 1.232.801.17
yedQb1956x 2.423.641.88

We decided to study the σS-dependent GGDEF/EAL genes and proteins for several reasons: (i) With these genes being under σS control, their respective target processes should be part of the relatively well-characterized general stress response (Hengge-Aronis, 2000), which should facilitate their identification, and (ii) with these genes being co-regulated and therefore their gene products being most likely coexpressed, they might constitute an excellent model system for studying signalling specificity and/or convergence in a c-di-GMP synthesizing and degrading network. As potential target processes can only be identified under conditions of expression of the corresponding regulatory or signal transducing factors, any analysis of the physiological function of a c-di-GMP network has to start from an expression analysis of the respective GGDEF and EAL genes. We therefore constructed lacZ reporter fusions in all the σS-dependent genes mentioned above (ydaM, yddV, yedQ, yciR, yeaI, ycgG, ydiV). In addition, we included the GGDEF gene yaiC in our analysis (whose expression is usually below detection on our microarrays), as yaiC as well as its Salmonella counterpart adrA are known to be under the control of a regulator, CsgD, which in turn is σS-regulated (Römling et al., 2000; Brombacher et al., 2003; 2006). These reporter constructs, which were integrated in single copy into the chromosome, reflect transcription patterns of the respective genes (for details, see Experimental procedures).

Expression of the reporter gene fusions was assayed in wild type and rpoS mutant backgrounds along the entire growth cycle. Cells were grown in rich Luria–Bertani (LB) medium as well as in minimal medium (with glucose limitation causing entry into stationary phase). In addition, these experiments were performed at 37°C as well as at 28°C as the biofilm-associated ‘rdar’ colony morphotype, which is known to be under c-di-GMP control (Römling, 2005), as well as the expression of curli fimbriae, which are known to be σS-controlled (Olsén et al., 1993) and to contribute to the ‘rdar’ morphotype (Römling, 2005) are stimulated at lower temperatures. Expression patterns observed (summarized in Figs 1 and 2, full data sets are shown in Fig. S1) can be summarized as follows:

Figure 1.

Stationary phase induction and σS dependence of various GGDEF/EAL genes during the growth cycle. Strain MC4100 (circles) and its rpoS::10 derivative (squares) carrying single copy lacZ fusions in ydaM, yddV, yedQ, yaiC and yciR as indicated in the figure were grown in LB at 28°C. OD578 (open symbols) and specific β-galactosidase activities (closed symbols) were determined.

Figure 2.

Stationary phase expression and σS dependence of various GGDEF/EAL genes in rich or minimal media at different temperatures. Strain MC4100 (white bars) and its rpoS::Tn10 derivative (hatched bars) carrying single copy lacZ fusions in ydaM, yddV, yedQ, yaiC and yciR were grown in LB and M9 minimal medium at 37°C and 28°C throughout the entire growth cycle. Only stationary phase β-galactosidase-specific activities are shown here (averages of two measurements each; full data sets for all phases of growth are shown in Fig. S1).

  • i. σS control was clearly confirmed for five of these eight GGDEF/EAL genes (ydaM, yddV, yedQ, yaiC, yciR; Figs 1 and 2 and Fig. S1); for the remaining genes, expression was not detectable (yeaI, ydiV) or extremely low (ycgG) under the conditions used (data not shown).
  • ii. The GGDEF genes ydaM, yddV, yedQ and yaiC as well as the GGDEF/EAL gene yciR exhibited stationary phase induction (Fig. 1 and Fig. S1); in rich LB medium, ydaM, yddV and yaiC were not at all or very weakly expressed in log phase, whereas yedQ and yciR exhibited measurable basal levels of expression already.
  • iii. yaiC, even though its overall expression level indicates that its gene product is a minor GGDEF protein, exhibits drastic temperature regulation; its stationary phase expression in LB is approximately 15-fold higher at 28°C than at 37°C (Figs 1 and 2 and Fig. S1); also ydaM and yddV show two to threefold higher stationary phase expression at 28°C than at 37°C (Fig. 2).
  • iv. Stationary phase induction (at 28°C) of yaiC exhibits a delay of approximately 2 h in comparison to induction of ydaM, yddV, yedQ and yciR (Fig. 1).
  • v. The major GGDEF genes ydaM and yddV exhibit similar stationary phase expression in rich medium, whereas in minimal medium, yddV becomes the dominantly expressed σS-dependent GGDEF gene (Fig. 2).

We conclude that besides their common σS-dependence, these five GGDEF/EAL genes exhibit distinct regulatory patterns as they differentially respond to entry into stationary phase and environmental parameters such as temperature and medium composition. Moreover, in comparison to many other genes (routinely assayed in our laboratory using similarly constructed lacZ reporter fusions), the overall expression levels of all these GGDEF/EAL genes are relatively low (e.g. compare to the fusion shown in Fig. 5). Together with the observation that it very much depends on the actual conditions, which GGDEF gene is the predominantly expressed one, this is consistent with the notion that only small but conditionally changing subsets of GGDEF/EAL proteins are expressed.

Figure 5.

Role of the GGDEF/EAL proteins YdaM and YciR and growth temperature in the expression of CsgD and the csgBAC operon. MC4100 and its derivatives carrying ydaM or yciR mutations were grown in LB medium to an OD578 of 4.5, RNA was prepared and CsgD transcripts were visualized by Northern blot analysis (A). Photographs in (A) show different gel blots (also with reaction times of the detection reaction and film exposure being different depending on respective signal strength), but on all gels (also in the following figures) the samples shown in the first three lanes were included as a reference, which allowed quantification of all samples relative to each other as shown as a percentage of wild-type csgD mRNA levels (in lane 1; quantification shown below the Northern blot). In parallel, CsgD protein levels were determined and quantified by immunoblot analysis (B). In addition, the expression of a single copy csgB::lacZ fusion was monitored in strain backgrounds otherwise identical to those used in A and B. OD578 (open symbols) and specific β-galactosidase activities (closed symbols) were determined as a function of growth temperature (C; circles: 28°C; squares: 37°C) or the allelic status of ydaM and yciR (D; circles: parental strain, i.e. ydaM+yciR+ diamonds: ydaM::cat; squares: yciR::kan).

Mutants with defects in specific σ S-dependent GGDEF/EAL genes are affected in biofilm-related phenotypes

Knowing their specific conditions of expression, we proceeded to search for mutant phenotypes for the σS-dependent GGDEF/EAL genes ydaM, yddV, yedQ, yaiC and yciR (for a detailed description of the knockout mutations, see Experimental procedures). As c-di-GMP has been implicated in biofilm formation, we tested Congo red binding on rich medium agar plates. Congo red binding is a complex phenotype that reflects various outer membrane and surface properties including the presence of adhesive structures such as curli fimbria which are involved in biofilm formation (Römling, 2005). We observed that Congo red binding on agar plates (Fig. 3A) was specifically reduced in the ydaM mutant, and slightly enhanced in the yciR mutant. Mutations in the three other GGDEF genes yddV, yedQ and yaiC had no effects. For comparison, we also included various regulatory mutants in the analysis: mutations in any of the components of the σS-MlrA-CsgD regulatory cascade that controls curli and cellulose synthesis, strongly reduced or eliminated Congo red binding.

Figure 3.

Phenotypes of mutants defective in various σS-controlled GGDEF/EAL genes.
A. Congo red binding after overnight growth at 28°C of derivatives of MC4100 carrying the mutations indicated in the figure.
B. Sedimentation after overnight growth in liquid LB at 28°C. Relevant genotypes: 1, parental (MC4100); 2, rpoS; 3, mlrA; 4, csgD; 5, yciR; 6, ydaM; 7, yddV; 8, yedQ; 9, yaiC; 10, flu; 11, yciR flu; 12, mlrA yciR.

In addition, we observed complete sedimentation of stationary phase cells of the yciR mutant, when overnight cultures were no longer shaken but maintained statically (Fig. 3B). By contrast, no aggregation at all was observed for the ydaM mutant. The yddV, yedQ and yaiC mutants showed intermediate behaviour like the wild type, i.e. partial sedimentation in a still turbid suspension (Fig. 3B). Strong autoaggregation in the yciR mutant is not due to overexpression of Ag43 [a surface adhesin encoded by the flu gene known to stimulate autoaggregation (Roux et al., 2005)] as the yciR phenotype was not reversed by the introduction of a knockout mutation in flu. Thus, autoaggregation observed in our experiments seems to be due to (over)expression of curli fimbriae, as it is completely abolished by any mutation in the curli regulatory cascade σS-MlrA-CsgD, and autoaggregation in such a mutant can not be rescued by introducing a second mutation in yciR (Fig. 3B).

Finally, this effect on curli expression was directly demonstrated by electron microscopy (Fig. 4). Typically curli-shaped fibres (as first described by Olsén et al., 1989) were overproduced by the yciR mutant, and were as absent in the ydaM mutant as in the σS-deficient strain known to be curli-negative.

Figure 4.

Electron micrographs of whole cells and curli fimbriae (best visible at larger magnification). Strains carrying the following mutations were grown overnight in liquid LB and prepared for electron microscopy as described in Experimental procedures: (A) parental (MC4100); (B) yciR; (C) rpoS; and (D) ydaM.

These data indicate that (i) specifically YdaM and YciR, but none of the other σS-dependent GGDEF proteins, control curli formation and curli-related phenotypes and (ii) YdaM and YciR exert this control antagonistically, suggesting that the GGDEF-only protein YdaM is likely to be a DGC, whereas YciR, which contains a GGDEF as well as a EAL domain, may act as a PDE.

The σ S-dependent GGDEF/EAL proteins YdaM and YciR inversely control curli biosynthesis by affecting transcription of the regulator CsgD

Regulation of the expression of curli fimbriae, which has been best studied in Salmonella, is strikingly complex (Olsén et al., 1993; Arnqvist et al., 1994; Hammar et al., 1995; Römling et al., 1998a; 2000; Brown et al., 2001; Prigent-Combaret et al., 2001; Gerstel et al., 2003; Jubelin et al., 2005). Curli fibres are produced in stationary phase when cells are grown below 30°C. Expression of the csgBAC operon, which encodes the curli structural proteins, is driven by a regulatory cascade consisting of σS, the MerR-like regulator MlrA and the CsgD regulatory protein. Mutants deficient in any of these positive regulators are devoid of curli (see also Fig. 3). The molecular mechanisms of action of MlrA (in csgD transcription) and CsgD (in csgBAC transcription) have not been clarified. Besides requiring σS and MlrA, csgD expression is also modulated by IHF and the two-component response regulators OmpR and CpxR. In addition, the histone-like protein H-NS represses curli synthesis, but seems to have an unexplained positive effect on the expression of the CsgD regulator. The regulatory gene csgD, which is followed by the csgEFG genes (encoding factors involved in secretion and assembly of curli subunits; Robinson et al., 2006), is transcribed divergently from the csgBAC operon, with approximately 750 bp making up the non-coding intergenic control region. This chromosomal arrangement and the major features of curli control are similar in Salmonella and E. coli (Olsén et al., 1993; Arnqvist et al., 1994; Römling et al., 1998b; Brombacher et al., 2003), although details in regulation might be different.

Thus, there are plenty of putative targets for YdaM/YciR-mediated c-di-GMP control of curli expression. Using a single copy translational mlrA::lacZ fusion, we found that expression of MlrA, i.e. the second factor in the σS-MlrA-CsgD transcription factor cascade, is not affected by mutations in ydaM or yciR (data not shown). As a next step, we tested the expression of CsgD, i.e. the point of the curli regulation cascade, where most modulating effects seem to converge. On Northern blots (Fig. 5A), high levels of csgD mRNA were found at 28°C (where curli fimbriae are expressed). The ydaM mutant exhibited strongly reduced csgD mRNA levels under these conditions, whereas the yciR mutant had elevated csgD mRNA levels. Reduced csgD mRNA levels in the ydaM mutant are due to reduced csgD transcription, as the ydaM mutant exhibited even slightly longer csgD mRNA half-life than the otherwise isogenic wild type (2.4 and 1.2 min respectively; data not shown). At 37°C (where no curli fimbriae are made) the cellular csgD mRNA content was low and not affected by mutations in ydaM or yciR. Very similar effects of growth temperature and the mutations in ydaM and yciR were observed at the CsgD protein level as assayed by immunoblot analysis (Fig. 5B), indicating that control by YdaM/YciR mainly or even exclusively affects csgD transcription (see also below).

In parallel, we monitored the expression of the target of CsgD regulation, i.e. the csgBAC curli operon, by using a single copy csgB::lacZ reporter fusion. The activity of this reporter seemed to strictly reflect csgD expression: csgB expression remained completely shut-off at 37°C, but exhibited clear stationary phase induction at 28°C (Fig. 5C) and this induction was reduced in the ydaM mutant and further stimulated in the yciR mutant (Fig. 5D). The mutations in the other σS-dependent GGDEF genes yddV, yedQ and yaiC did not alter csgB::lacZ expression (data not shown). Induction of csgB starts approximately 2 h after the induction of ydaM and yciR under the same conditions (compare to Fig. 1), and in fact parallels the induction of yaiC (also see Fig. 1).

From these data we conclude that (i) csgD transcription, which is repressed at 37°C, becomes derepressed or activated at 28°C, (ii) the GGDEF protein YdaM is required for this derepression/activation of csgD, (iii) the GGDEF/EAL proteins YdaM and YciR play an antagonistic role in this control (at 28°C), and (iv) YdaM and YciR affect csgD transcription as CsgD protein levels as well as csgBAC and therefore curli fimbriae expression simply follow this regulatory pattern of csgD transcription.

In order to further confirm that YdaM and YciR are not involved in a post-transcriptional control of csgD expression or in CsgD activity, we eliminated the natural transcriptional control of csgD by expressing it ectopically under the control of the arabinose-inducible pBAD promoter. In this construct, the described σS-dependent promoter of csgD (Römling et al., 1998a) was precisely replaced by pBAD, such that the csgD mRNA produced corresponds to the natural csgD mRNA (which exhibits an untranslated 5′-region of 147 nucleotides). In other words, transcriptional control is different, but whatever regulatory mechanism may act on this mRNA or the protein made thereof, should be conserved for the ectopic csgD construct. We observed that neither mutations in ydaM and yciR nor temperature had any effect on csgD transcript levels obtained from this construct, no matter whether cells were grown in the presence or absence of the inducer arabinose or at 28°C or 37°C (Fig. S2). Again, csgB expression immediately followed the csgD mRNA levels obtained (Fig. S2A) and no regulation by YdaM/YciR or temperature could be detected (Fig. S2B–D). Interestingly, in this strain with ectopic σS-independent csgD expression, csgB expression was completely σS-independent (data not shown), indicating that csgB is expressed from a vegetative promoter, and is indirectly σS-controlled via CsgD. Taken together, all these data strongly indicate that regulation of curli expression by the c-di-GMP-controlling YdaM/YciR system, as well as by a temperature signal occurs at the level of csgD transcription. Moreover, also the multiple inputs of σS into curli control converge in the control of csgD transcription.

Enzymatic activities of YdaM and YciR

The finding that the GGDEF-only protein YdaM positively regulates curli expression, whereas the GGDEF + EAL protein YciR has the opposite effect, suggested that YdaM is a c-di-GMP producing DGC, whereas YciR may act as a PDE. However, as GGDEF domains can also have regulatory instead of enzymatic function (Christen et al., 2005) and composite proteins containing GGDEF as well as EAL domains usually seem to have only one of the two possible enzymatic activities (Schmidt et al., 2005), putative enzymatic functions of GGDEF/EAL proteins have to be demonstrated directly.

We therefore tested purified YdaM and YciR proteins in vitro for DGC and PDE activities respectively. As a positive control for the DGC assay, an mutationally activated form of the C. crescentus PleD protein (PleD*) was used (Paul et al., 2004). Unlike PleD*, wild-type YdaM, which carries an N-terminal sensory region (containing a PAS domain) which senses an unknown signal, can be expected to produce basal levels of c-di-GMP at most. Nevertheless, in the standard DGC assay (Paul et al., 2004; Christen et al., 2005), we could observe c-di-GMP synthesis from GTP (Fig. 6A). C-di-GMP does not further accumulate after 10 min, indicating that YdaM exhibits product inhibition as described for PleD. Both proteins contain the I-site motif (RXXD located five amino acids upstream of the GGDEF motif), which represents an inhibitory allosteric binding site for c-di-GMP (Chan et al., 2004; Christen et al., 2006). As in vitro activity of YdaM is weak in the absence of its unknown sensory signal input, we confirmed DGC activity of YdaM also genetically. The two acidic residues in the GGDEF (or GGEEF) motif are crucial for GTP binding as well as catalysis of the DGC reaction (Chan et al., 2004). When these two acidic residues in YdaM (E334 and E335) were replaced by alanine, the mutant version of YdaM could no longer complement a ydaM knockout mutation, with csgB::lacZ expression used as a read-out of the system (Fig. 7). Thus, YdaM acts as a DGC also in vivo.

Figure 6.

Purified YdaM and YciR exhibit diguanylate cyclase and phosphodiesterase activities respectively.
A. Synthesis of c-di-GMP from GTP was assayed in vitro using purified YdaM and PleD* proteins and visualized by thin layer chromatography (Co, control sample without protein added; grey arrow head, intermediate reaction product observed with YdaM, and to a lesser extent also with PleD*, which may represent the incomplete reaction product pppGpG).
B. Cleavage of c-di-GMP (purified after synthesis from GTP by PleD*) was assayed in vitro with purified YciR protein (Co, no protein added). Spots representing the reaction products pGpG, GMP and phosphate were assigned according to (Tamayo et al., 2005). For both assays, reaction times are given in the figure and experimental details are described in Experimental procedures.

Figure 7.

The GGEEF motif in YdaM is essential for its activity in vivo. Derivatives of MC4100 carrying ydaM::cat and the single copy csgB::lacZ fusion and also containing either pBAD18-YdaM (diamonds, squares), pBAD18-YdaME334A + E335A (inverted triangles, triangles) or the pBAD18 vector (circles) were grown in LB/ampicillin medium at 28°C with 0.05% arabinose (diamonds, inverted triangles) or without arabinose (squares, triangles, circles) added at an OD578 of 3.0. OD578 (open symbols) and specific β-galactosidase activities (closed symbols) were determined.

For the PDE in vitro assay, radiolabelled c-di-GMP was prepared using the strongly active PleD* DGC (for details, see Experimental procedures), and was incubated with purified YciR (Fig. 6B). This resulted in rapid and quantitative disappearance of c-di-GMP and the appearance of a primary product, which seems to be pGpG [according to previously published studies that used the same assay (Christen et al., 2005; Tamayo et al., 2005)], and which in the longer run is further transformed, probably into GMP and phosphate (Fig. 6B). From these in vitro data we conclude that YdaM and YciR indeed exhibit DGC and PDE activities respectively.

The role of H-NS in the curli control network is connected to YdaM and YciR

The histone-like protein H-NS plays multiple and crucial roles in the σS network (Barth et al., 1995; Yamashino et al., 1995), and also affects curli control (Arnqvist et al., 1994; Gerstel et al., 2003). Here we observed that YdaM/YciR-mediated control is also connected to the action of the histone-like protein H-NS in csgD regulation: In the absence of H-NS, csgD transcription at 28°C is not only reduced (confirming a positive role of H-NS in csgD expression), but becomes nearly YdaM/YciR-independent (Fig. 8A, compare lanes 4–6). Correspondingly, csgD expression in the ydaM mutant is no longer affected by a mutation in hns (Fig. 8A, compare lanes 2 and 5). These results suggested that H-NS could act upstream of ydaM. Indeed we observed that during entry into stationary phase, i.e. when CsgD and curli fimbriae are expressed in the wild type, the expression of ydaM is reduced in the hns mutant (Fig. 8C), whereas yciR expression is stimulated (Fig. 8D). Thus, the positive effect of H-NS on csgD transcription can be explained by H-NS stimulating YdaM expression and therefore c-di-GMP synthesis, and at the same time downregulating YciR expression and therefore PDE activity. This positive H-NS effect is also transmitted onto the expression of the downstream target csgB (Fig. 8B). During log phase, however, the expression of ydaM is increased in the hns mutant, which is a typical behaviour of σS-dependent genes reflecting the strongly increased σS levels in hns mutants (Barth et al., 1995). The moderate log-phase activation of the csgB promoter in the hns mutant may also be due to the absence of direct binding of H-NS in the control region (Gerstel et al., 2003). The finding, however, that the hns mutant still exhibits clear stationary phase induction of csgB expression (Fig. 8B), shows that CsgD-mediated activation of the csgB promoter is not simply due to overcoming H-NS repression. In summary, these data clarify the input of H-NS into the emerging curli control network.

Figure 8.

Role of the histone-like protein H-NS in the curli control network. csgD transcript levels were determined by Northern blot analysis (A) as described in Fig. 5, with the exception that hns::Tn10 derivatives of MC4100 and its ydaM and yciR mutant variants were included in the analysis. The effects of H-NS on csgB (B), ydaM (C) and yciR (D) expression was assayed using MC4100 carrying single copy lacZ fusions in the respective genes (circles) as well as the corresponding hns::Tn10 derivatives (squares). Cells were grown in LB medium at 28°C and OD578 (open symbols) and specific β-galactosidase activities (closed symbols) were determined.

A GGDEF cascade: YdaM controls the expression of YaiC

The CsgD regulatory protein is known to control at least one more gene besides the csgBAC curli structural operon: in Salmonella this gene is called adrA (Römling et al., 2000), in E. coli it is yaiC (Brombacher et al., 2003). It encodes a GGDEF protein, which plays a role in cellulose biosynthesis in Salmonella (Römling et al., 2000), perhaps in a mechanism similar to the direct stimulation of cellulose synthase described for G. xylinus (Weinhouse et al., 1997). We have shown here (see Fig. 1) that yaiC exhibits a regulatory pattern very similar to that observed for csgB (see Figs 5C or 8B): both genes are only expressed at 28°C, both are completely σS-dependent, and both show parallel induction during entry into stationary phase which starts about 2 h later than that of ydaM and the other σS-controlled GGDEF/EAL genes – the only obvious difference between the two genes is their absolute expression level (approximately 50-fold lower for yaiC than for csgB; fusion activities can be compared as the single copy reporter fusions were constructed in exactly the same way). If yaiC is regulated by CsgD, one would predict that yaiC should also be under positive control by YdaM. Using the yaiC::lacZ reporter fusion, we observed that not only the regulatory csgD and mlrA mutations but also the ydaM mutation indeed reduced the expression of yaiC (Fig. 9). This cascade regulation is specific for yaiC, as the expression of the other GGDEF/EAL genes yddV, yedQ, yciR and ydaM itself was not affected by the mlrA, ydaM and csgD mutations (data not shown). These data indicate a GGDEF cascade within the σS network, in which an earlier expressed GGDEF protein, YdaM, controls the later and weaker, but precisely regulated expression of another one, YaiC.

Figure 9.

GGDEF cascade regulation within the σS network: YdaM controls the expression of yaiC. MC4100 containing a single copy yaiC::lacZ fusion (circles) as well as its derivatives carrying ydaM::cat (triangles), mlrA::kan (squares) and csgD::cat (diamonds) were grown in LB at 28°C, and OD578 (open symbols) and specific β-galactosidase activities (closed symbols) were determined.

Genome-wide microarray analyses of target genes: do the σ S-controlled GGDEF/EAL proteins have global regulatory effects?

GGDEF/EAL proteins have been reported to affect various phenotypes including biofilm-related functions, motility and virulence (Jenal, 2004; Römling et al., 2005). This suggested that c-di-GMP has a relatively global role in cellular regulation. In a very recent microarray transcriptome study, overproduction of the GGDEF protein YddV in E. coli resulted in altered expression of several global regulators (e.g. RcsA, SoxS and Fur) and at least parts of their respective regulons (Méndez-Ortiz et al., 2006). However, such global effects have been observed when GGDEF proteins are overproduced, i.e. when the cellular c-di-GMP pool was artificially elevated (which can be by two or more orders of magnitude as recently demonstrated; Kader et al., 2006).

We therefore asked whether the knockout mutations in ydaM and yciR also have global effects on gene expression under conditions where their gene products are expressed in natural amounts and are active in the wild type (inducing curli expression, i.e. in LB at 28°C during entry into stationary phase). Using microarray analysis, we compared genome-wide gene transcription patterns for these mutants and their parental strain. The microarray expression data clearly reproduced the antagonistic regulation of the csg genes by YdaM and YciR (Table 2). Expression of the genes in the csgBAC operon was between 10- and 30-fold reduced in the ydaM mutant and 2.5- to fivefold elevated in the yciR mutant; csgD and the following genes csgEFG were between 3.2- and 4.5-fold downregulated in the ydaM mutant, and about twofold upregulated in the yciR background (only csgG is listed for the yciR mutant in Table 2, as csgD, csgE and csgF remained just below the cut-off ratio of 0.5). As in the microarray experiments described in Table 1, yaiC expression again was too low to be reliably detected. Interestingly, only very few other genes besides the csg genes were affected by the ydaM mutation (Table 2): ymdA, which exhibits the same antagonistic regulation by YdaM and YciR, is the gene right downstream of csgBAC, and this regulatory pattern indicates that this gene of so far unknown function actually belongs to the csgBAC operon. ymfE is a gene of unknown function of a lysogenic defective prophage in the E. coli chromosome, and the effects of the ydaM mutation on the remaining two genes, rpsN (encoding a ribosomal protein) and hdeB (an acid resistance gene) are small and perhaps not even significant. From these results, we conclude that c-di-GMP controlled by YdaM and YciR does not play a global regulatory role, but has a precise, strong and nearly exclusive effect on CsgD-controlled genes suggesting that its only molecular target is csgD transcription.

Table 2.  Ratios of expression of genes differentially expressed in MC4100 and mutants defective in the GGDEF/EAL genes ydaM, yciR, yddV and yedQ as determined by genome-wide microarray analysis.a
Genebb-numberExpression ratios (wild type/mutant)
  • a. 

    Expression ratios are given for MC4100 versus its respective mutant derivative. Only genes with expression ratios above 2.0 or below 0.5 in both of two independent experiments are listed (ratios given are the average of the two experiments). For details of the procedure and statistical analysis, see Weber et al. (2005).

  • b. 

    Genes involved in curli biosynthesis are indicated in boldface. Other gene products: rpsN, ribosomal protein S14; hdeB, acid resistance protein; dppA, periplasmic binding protein for dipeptide transport; tnaA, tryptophanase; miaA, tRNA delta(2)-isopentenylpyrophosphate transferase, paaJ, acetyl-CoA acetyltransferase (lipid metabolism); glmM, phosphoglucosamine mutase (peptidoglycan and LPS synthesis). ygdI and ytfK encode proteins of unknown functions, ymfE seems to belong to a defective prophage.

  • c. 

    The yaiC gene, which exhibits weak but YdaM-dependent expression late during the growth cycle (Fig. 9), has not been reliably detected in any microarray experiment in this study. It is therefore suspected that the yaiC oligonucleotide spot on the commercially available microarrays used here is not optimal for hybridization.


In addition, we tested whether the knockout mutations in the other σS-dependent GGDEF genes expressed in parallel to ydaM and yciR, i.e. yddV and yedQ, have any effect on genome-wide transcription patterns. In both mutants, only few genes (Table 2) showed an apparent differential expression that came close to the cut-off ratio (> 2 or < 0.5) considered statistically significant (Weber et al., 2005). Thus, just as YdaM/YciR, also YddV and YedQ do not act as global regulatory factors under the conditions tested. Some of the apparent target genes actually belong to operons (e.g. dppA or paaJ), whose other member genes were not differentially expressed, suggesting that in these cases, the apparent control by YddV or YedQ is not significant. Based on these data, we conclude that, while the yddV and yedQ genes are expressed in parallel to ydaM and yciR (Fig. 1) under the conditions used for our transcriptome microarray analysis, the YddV and YedQ proteins do not act as global regulators of gene expression. Either the proteins are not present (due to post-transcriptional control) or they are inactive (because they may require specific signals sensed by their N-terminal domains) or they may affect protein activities rather than the transcription of genes. Taken together, this means that even coexpressed GGDEF/EAL genes can have non-overlapping functions which would be consistent with the concept of functional sequestration of c-di-GMP produced by different GGDEF proteins.


GGDEF/EAL proteins and a role for c-di-GMP in the σ S network

The present study started from the observation that seven of the 29 GGDEF/EAL genes in E. coli exhibited positive control by σS in genome-wide transcription studies on microarrays (Table 1). As shown by single copy lacZ reporter fusion analysis, five genes, i.e. the GGDEF genes ydaM, yddV, yedQ and yaiC, as well as yciR, which encodes a GGDEF + EAL protein, exhibit σS-dependent stationary phase induction, which is further and differentially modulated by medium composition and temperature (Figs 1 and 2 and Fig. S1). For several reasons, we consider it likely, that additional GGDEF/EAL genes may be σS-controlled: (i) for some genes, a σS-dependent promoter may be located further upstream and may not have been present in our reporter fusions (although depending on the specific genome context, these constructs contained up to several hundred base pairs upstream of the coding regions), (ii) in microarray analyses, differential expression of genes that possess a σS-dependent as well as a vegetative promoter often remains below the statistically significant ratio indicating σS control (Weber et al., 2005) and (iii) some σS-dependent promoters can require specific conditions for being expressed that were not met in our microarray and reporter fusion experiments (McLeod et al., 2000; Germer et al., 2001; Metzner et al., 2004). It should also be noted that none of the 29 GGDEF/EAL genes is under negative control of σS (Table S1). Taken together, this indicates that GGDEF/EAL genes are over-represented in the σS regulon or network (which consists of approximately 10% of the E. coli genes), and that the signalling molecule c-di-GMP, whose level is controlled by GGDEF and EAL domain proteins, respectively, plays a crucial role in stationary phase and stress physiology.

An important part of this physiology is obviously the control of biofilm-promoting features, most prominently the expression of the adhesive curli fimbriae. Here we show that a specific strongly σS-dependent cytoplasmic GGDEF protein, YdaM, is required for expression of the regulatory gene csgD, the directly following csgEFG genes (involved in curli subunit secretion and assembly), the CsgD-activated genes csgBAC (encoding curli structural proteins) and yaiC (encoding a GGDEF protein that activates cellulose synthesis) (Figs 5, 7 and 9, and Table 2). While the genetic organization of these genes as well as their control by environmental signals is similar in E. coli and Salmonella enterica serovar typhimurium (Römling, 2005), there is no YdaM homologue present in Salmonella (which possesses only 12 GGDEF proteins, all of which are membrane-bound; Römling et al., 2005). By contrast, a very recent report (published while this article was under revision) provided evidence that curli biosynthesis in Salmonella is under the combined control of two other GGDEF proteins, STM2123 and STM3388 (Kader et al., 2006), the latter of which has no counterpart in E. coli. This indicates that signal input into curli regulation via c-di-GMP is different in these two closely related organisms and that GGDEF/EAL networks exhibit rapid evolutionary adaptation to changing habitats.

The curli control network represents a module within the σS network which is expressed under very specific conditions, i.e. when E. coli cells enter stationary phase in rich medium at 28°C (but not at 37°C). This module exhibits complex feedforward regulation by σS, i.e. the target genes are under multiple direct and indirect control by σS (summarized in Fig. 10): σS activates the expression of MlrA (an essential positive regulator for csgD expression; Brown et al., 2001); in addition, σS is probably directly involved in csgD transcription (Römling et al., 1998a), and, as shown here, σS is required for the expression of YdaM and YciR (Figs 1 and 2, and Fig. S1), which provide an additional and essential signal for the expression of the csgD regulatory gene (Fig. 5). This complex feedforward arrangement, which converges in csgD expression (the downstream target csgBAC is expressed from a vegetative promoter activated by CsgD) probably not only increases the amplitude and precision of induction of the downstream target (feedforward loops can filter out stochastic noise in gene expression; Mangan and Alon, 2003), but also restricts target expression to very specific conditions by allowing additional signal integration in the indirectly operating cascade part(s) of the feedforward system. YdaM/YciR illustrate this latter role: In all our experiments temperature control of csgD and the target operon csgBAC always correlated with control by YdaM. Thus, temperature control of csgB was eliminated when csgD expression was uncoupled from regulation by YdaM/YciR by expressing it ectopically under pBAD promoter control (Fig. S2); and in the absence of YdaM, temperature control of csgD expression was strongly reduced (Fig. 5, lanes 2 and 5). YdaM may therefore also act as a thermosensor in curli control. ydaM expression is thermoregulated (Fig. 1), but its activity could also be controlled in response to reduced temperature.

Figure 10.

A model for c-di-GMP-mediated signalling by GGDEF/EAL proteins within the σS network. Control of curli and cellulose biosynthesis represents a hierarchically structured module within the σS-controlled general stress response network. During entry into stationary phase, σS is induced and exerts multiple activating effects in parallel (on the expression of the GGDEF/EAL proteins YdaM/YciR and of the serially arranged activators MlrA and CsgD). This establishes a multiple feedforward cascade system that converges onto the expression of the essential activator gene csgD. The csgB promoter is transcribed by vegetative polymerase and indirectly σS-activated via CsgD. Also the histone-like protein H-NS plays multiple roles by downregulating σS during log phase growth (by post-transcriptional control mechanisms that become relieved during entry into stationary phase). H-NS activates ydaM expression during entry into stationary phase and represses yciR expression throughout the entire growth cycle. As a result, H-NS has an indirect activating effect on csgBAC expression during entry into stationary phase that overcomes its direct repressive effect in growing cells. Not only the influence of σS, but also signal processing and integration via additional regulators converges in the control of transcription of the CsgD regulator (note that the arrangement of any arrows in the figure that illustrate this input from other regulators does not imply any molecular mechanisms). σS also activates the expression of at least two other GGDEF genes, yddV and yedQ, which control unknown targets. For further details, see Discussion.

Curli and cellulose regulation also features another prominent network motif, i.e. the regulatory cascade (Fig. 10). The σS-MlrA-CsgD-csgBAC cascade has been recognized a while ago (Römling et al., 2000; Brown et al., 2001). In addition, we have shown here a GGDEF cascade, in which an earlier expressed GGDEF protein, YdaM, positively controls the later expression of another GGDEF protein, YaiC (Fig. 9), which is known to activate cellulose biosynthesis (Bokranz et al., 2005). Taken together, the σS-YdaM-CsgD-YaiC-cellulose synthase cascade intersects with the curli control cascade in a complex feedforward network, with CsgD representing the major signal integration node (Fig. 10). In bacteria, such long cascades are not very frequent and are characteristic for developmental processes, which have to be triggered very precisely and over time have to integrate multiple temporal and sometimes spatial signals (Milo et al., 2002). It is still under debate whether biofilm formation represents a true developmental process (Lazazzera, 2005). It seems, however, that at least curli and cellulose control exhibits all the hallmarks of a complex developmental regulatory cascade system. Especially noteworthy is the precise timing of expression in this cascade: While σS itself and YdaM expression are activated relatively early during entry into stationary phase (at an OD578 of approximately 2.5; Fig. 1), the expression of the downstream target genes csgBAC and yaiC starts almost exactly 2 h later (at an OD578 of approximately 4; Figs 1 and 9). It is conceivable that YdaM via its sensory input domain may also act as the recipient of some checkpoint signal that determines the timing of the subsequent events in the curli control cascade.

The σ S-dependent GGDEF/EAL proteins YdaM and YciR play a specific and antagonistic role in the transcriptional regulation of the curli regulator CsgD

Our in vivo data demonstrate that the GGDEF protein YdaM plays a crucial positive role in curli expression, whereas YciR, which contains both GGDEF and EAL domains, acts as an antagonist in this regulatory process (Figs 3–5). These findings are also consistent with a previous report that overexpression of yciR(gcpE) or a knockout mutation in this gene in Salmonella eliminated or increased, respectively, cellulose synthesis and biofilm formation (assayed just phenotypically, not at the genetic level) (García et al., 2004). We also demonstrate that purified YdaM and YciR proteins exhibit DGC and PDE activities, respectively (Fig. 6), which allows the conclusion that these proteins affect the expression of curli by maintaining an adequate balance of c-di-GMP synthesis and degradation.

All our data indicate that the target of YdaM/YciR-controlled c-di-GMP is the transcriptional regulation of the CsgD regulator, which in turn controls the expression of the csgBAC and yaiC genes: (i) Northern blot analysis of cells grown under curli expression-promoting conditions (LB, 28°C) showed strongly reduced csgD transcript levels in the ydaM knockout mutation, while the yciR knockout mutant produced more csgD transcript (Fig. 5A), (ii) the process affected by YdaM is transcription of csgD, because the ydaM mutant exhibits an even somewhat longer half-life of csgD mRNA, (iii) with respect to regulation by YdaM/YciR and temperature, CsgD protein levels just follow csgD mRNA levels (Fig. 5B), (iv) replacing the csgD promoter by the pBAD promoter eliminated the control by YdaM/YciR no matter whether this ectopically expressed csgD mRNA was expressed at high or low levels (Fig. S2), and (v) in all these experiments, the expression of the target gene csgB (assayed as a single copy lacZ reporter fusion) simply reflected and, in dynamic induction experiments, immediately followed csgD expression without any further modulation by YdaM/YciR (Fig. 5B and Fig. S2). We conclude that csgD transcription is the major target of YdaM/YciR in curli control.

Consistent with this conclusion, it was reported very recently (Kader et al., 2006) that Salmonella mutants defective in the two GGDEF proteins STM2123 and STM3388 (neither protein corresponds to YdaM, which in fact does not occur in Salmonella, see above) also exhibit reduced csgD transcript levels. However, based on the finding that overproduction of the EAL protein YhjH downregulated CsgD protein levels, when csgD was strongly expressed from a heterologous promoter on a plasmid, it was suggested that c-di-GMP also plays a role in an unidentified post-transcriptional control mechanism of csgD. We consider it possible that in this experiment with Salmonella, depletion of the cellular c-di-GMP pool by overexpression of YhjH (which in physiological amounts is not involved in curli control) may produce indirect and non-physiological effects. It is also not excluded that some other GGDEF protein(s) may play a positive role in a post-trancriptional or activity control of CsgD (although all our data with E. coli indicate that csgBAC expression simply and immediately follows csgD expression). In conclusion, the target of c-di-GMP produced by the YdaM/YciR system in E. coli (Fig. 5) or by STM2133 and STM3388 in Salmonella (Kader et al., 2006) clearly is the transcription of csgD.

In terms of numbers of regulatory factors acting upon it, the csgD transcriptional control region is the most complex σS-dependent promoter region studied so far (Römling et al., 1998a; Brown et al., 2001; Prigent-Combaret et al., 2001; Gerstel et al., 2003). In principle, YdaM-generated c-di-GMP may modulate the action of any of these regulators (i.e. OmpR, IHF, MlrA, CpxR, H-NS) at the csgD promoter. While the role of H-NS has been further clarified in this study (Fig. 8, and see below), linking YdaM/YciR-controlled c-di-GMP to the action of any of the remaining or even yet unknown additional factors will require intensive further genetic and biochemical studies.

The role of the histone-like protein H-NS and its connection to YdaM/YciR in the curli control network

This study also clarifies the role of H-NS in the curli control network and shows a direct link to YdaM/YciR. In contrast to its much more general role as a transcriptional repressor or silencer (Dorman, 2004; Rimsky, 2004), H-NS was previously observed to activate csgD transcription by an unknown mechanism (Fig. 8 and Gerstel et al., 2003). Our finding that the reducing effects on csgD transcription of knockout mutations in ydaM and hns are not additive (rather, the hns mutation has no or almost no effect in a ydaM background and vice versa; Fig. 8), raised the possibility that H-NS may act via YdaM/YciR. Indeed, we found that ydaM expression is under positive control of H-NS, whereas yciR is repressed by H-NS (Fig. 8). Thus, the positive role of H-NS in csgD transcription is indirect, and low csgD expression in the hns mutant is due to a YdaM/YciR balance strongly shifted towards c-di-GMP degradation.

Interestingly, H-NS plays multiple roles in the curli control network, which also differ in the different phases of the growth cycle (Figs 8 and 10). In log phase cells, H-NS strongly downregulates σS by still elusive and probably indirect mechanisms that strongly interfere with rpoS mRNA translation and stimulate σS proteolysis (Barth et al., 1995; Yamashino et al., 1995). In addition, H-NS has a direct repressing effect on the very downstream target of curli control by binding to the promoter region of csgBAC[Fig. 8B; (Olsén et al., 1993; Arnqvist et al., 1994)]. During entry into stationary phase, however, the complex post-transcriptional inhibition of σS by H-NS is relieved (Hengge-Aronis, 2002), H-NS affects the YdaM/YciR balance in favour of c-di-GMP production (Fig. 8C and D), which allows CsgD expression and therefore strong induction of csgBAC (Fig. 8B). Thus, the role of σS and H-NS go precisely hand in hand at every level of the curli control cascade (Fig. 10).

Regulatory and functional sequestration of c-di-GMP production and action

Studies in which GGDEF or EAL proteins were overproduced indicated a global role for c-di-GMP in the control of biofilm formation (including curli and cellulose synthesis), motility and virulence (summarized by Jenal, 2004; Römling et al., 2005). Genome-wide transcription analysis on microarrays performed with an E. coli strain that overproduced the GGDEF protein YddV indeed showed altered expression of numerous genes (including various stress response genes) (Méndez-Ortiz et al., 2006). However, in GGDEF protein-overproducing strains, the actual cellular c-di-GMP levels can be orders of magnitudes higher than in wild-type strains under otherwise identical conditions (Kader et al., 2006). This is likely to eliminate specificity of c-di-GMP signalling. Moreover, overproduction of GGDEF proteins and therefore c-di-GMP can be detrimental or even toxic as it may deplete the GTP pool (Christen et al., 2005). Therefore stress response genes may be activated which would not show a specific response to c-di-GMP when present in normal concentrations. To circumvent these drawbacks, we have first studied the regulation of a set of GGDEF/EAL genes, and then searched for their individual downstream targets by genome-wide microarray analysis using strains that express physiological levels of these GGDEF/EAL proteins (in comparison to the respective knockout mutants) under conditions where the respective genes are naturally expressed.

We find that under such physiological conditions, the GGDEF/EAL proteins under study here do not act as global regulators. In particular, the antagonistically acting YdaM/YciR system significantly affects only csgD, the following csgEFG genes and the CsgD-regulated operon csgBAC(ymdA) and the yaiC gene (Table 2; the weakly expressed, but according to gene fusion analysis YdaM-controlled yaiC mRNA could not be reliably detected on the microarrays). Thus, this regulation is not only strong (expression ratios up to 30-fold on microarrays are exceptionally high) but also highly specific, and in fact suggests that YdaM/YciR affects a single target process, i.e. transcription from the csgD promoter. There are two potential explanations for this high specificity: either, this ‘isolation’ from other known c-di-GMP effects (which may occur at the same time) could be an indication that the YdaM/YciR system acts in a microcompartmented manner, perhaps in a complex which includes both proteins as well as a direct molecular c-di-GMP effector component and which may be located to the site of action, i.e. the csgD promoter region. Alternatively, if the YdaM/YciR system controls the cellular c-di-GMP level, csgD expression may be the only process under c-di-GMP control under the specific conditions used here (rich medium, 28°C, entry into stationary phase) – or other c-di-GMP-controlled processes that may occur in parallel, operate in a microcompartmented manner. Work is in progress to distinguish these possibilities. It should be noted that very recently Kader et al. (2006) also suggested a compartmentation of cellular c-di-GMP pools, based on their finding that the two GGDEF/EAL proteins, which in Salmonella activate csgD transcription, do not affect the cellular c-di-GMP level, whereas another GGDEF protein (AdrA), which contributes to the cellular c-di-GMP level does not affect csgD expression.

Our finding that the ydaM and yciR mutations affect the expression of csg genes only, also raises the question why some other genes that are believed to be under the control of CsgD (Brombacher et al., 2003; 2006), are not differentially expressed in these mutants. In contrast to these previous studies, which used CsgD expressed from a plasmid, a different strain background and cells grown in minimal medium, we also analysed the CsgD regulon under the same physiological conditions as used above by comparing wild type and csgD mutant strains on microarrays (C. Pesavento and R. Hengge, unpublished results). A few more genes, which showed hardly any overlap with the genes identified by (Brombacher et al., 2006), were observed to be differentially regulated, but with relatively small ratios only. These genes might require CsgD for expression, but in contrast to the expression of csgBAC, the reduced CsgD levels still present in the ydaM mutant (approximately 10% of the wild type, see Fig. 5) may be sufficient for their expression. Clearly, the CsgD regulon should be further characterized in future studies.

Also mutations in the GGDEF genes yddV and yedQ do not produce global alterations in gene expression when assayed under conditions of expression of these genes in wild-type strains (Table 2). This is especially noteworthy for YddV as overproduction of this protein does produce global effects on gene expression (Méndez-Ortiz et al., 2006). That only very few genes seem to be affected by the mutations in yddV and yedQ (and with low ratios only) suggests that either also these GGDEF proteins have an extremely restricted target spectrum, or, perhaps more likely, that despite their expression at the genetic level, the proteins are not present (due to post-transcriptional control) or inactive (due to the absence of specific signals sensed by their N-terminal input domains) or that c-di-GMP produced by these GGDEF proteins directly affects the activity of target components not involved in gene regulation.

Taken together with our data on the regulation of σS-dependent GGDEF/EAL genes, these genome-wide transcription data suggest regulatory and/or functional sequestration even of GGDEF/EAL proteins which are coexpressed at the genetic level. This may be due to post-transcriptional control of GGDEF/EAL protein levels or activity, but functional sequestration may also be due to microcompartmentation in complexes in which locally produced c-di-GMP may act on colocalized effectors. Of course, this concept should not be generalized – the activities of some GGDEF/EAL proteins may be confined to microcompartments, others however, may act via the cellular pool of c-di-GMP. In the future, this will have to be sorted out for the entire GGDEF/EAL network of a given organism. Current knowledge indicates that brute force overproduction of GGDEF/EAL proteins overruns these complex interrelationships by which high cellular specificity of c-di-GMP signalling is achieved. Rather, we will have to study the regulation, function and localization of specific GGDEF/EAL proteins under conditions of physiological expression.

Experimental procedures

Bacterial strains, plasmids and growth conditions

All strains used in this study are derivatives of the E. coli K-12 strain MC4100 (Casadaban, 1976). rpoS359::Tn10 (Lange and Hengge-Aronis, 1991), hns205::Tn10 (May et al., 1990), yddV::Tn5 (originally in strain FB22176 described in Kang et al., 2004) and the newly constructed mutant alleles ydaM::cat, yciR::kan, yedQ::cat, yaiC::kan, mlrA::kan, csgD::cat and flu::cat, which are all deletion-insertion mutations constructed by one-step inactivation according to (Datsenko and Wanner, 2000) using the primers listed in Table S1, were introduced by P1 transduction (Miller, 1972).

The plasmid for the ectopic expression of csgD under pBAD promoter control is a derivative of pBAD18 (Guzman et al., 1995) generated using oligonucleotides listed in Table S2. The csgD-carrying fragment inserted was chosen such that transcription under pBAD control starts in the same nucleotide region as for wild-type csgD expression under its own promoter control.

For complementation experiments, ydaM was cloned onto pBAD18 (Guzman et al., 1995) using oligonucleotides listed in Table S2. Replacement of E334 and E335 in the GGEEF motif of ydaM by alanine residues was achieved by a four-primer/two-step polymerase chain reaction (PCR) mutagenesis protocol previously described (Germer et al., 2001). The ‘internal’ mutagenic primers are listed in Table S2, as ‘external’ primers, the ones used for cloning ydaM into pBAD18 were used again.

Standard batch cultures were grown at 37°C or 28°C in a rotary shaker at 300 r.p.m. in LB or minimal medium M9 supplemented with 0.03% glucose (Miller, 1972) Antibiotics were added as recommended (Miller, 1972). Growth was monitored by measuring the optical density at 578 nm (OD578).

Construction of chromosomal lacZ fusions

In order to construct lacZ reporter fusions in the various GGDEF/EAL genes under study here, appropriate PCR fragments starting between 400 and 200 bp (depending on the chromosomal context of the specific genes) upstream of and ending a few nucleotides downstream of the translational start site of the respective gene (details are given in Table S1), were cloned into pJL28 (Lucht et al., 1994) using the restriction sites indicated in the respective primers (Table S1). These reporter constructs are translational fusions, but do not contain regions right downstream of the translational start site usually required in those cases where genes are subject to translational control involving complex mRNA secondary structures and small RNAs (Gottesman, 2005). Thus, these reporter fusions reflect transcription of the corresponding genes.

All reporter fusions were transferred to the att(λ) location of the chromosome of MC4100 via phage λRS45 or λRS74 as described (Simons et al., 1987). Single lysogeny was tested by a PCR approach (Powell et al., 1994).

Northern blot analysis

For RNA preparation, cells were grown in LB medium at 37°C or 28°C and harvested at an OD578 of 4.5 on stop solution (Bernstein et al., 2002) and immediately frozen in liquid nitrogen. Cell lysis and RNA isolation was according to (Tani et al., 2002), DNase I treatment and phenol/chloroform extraction were also previously described (Weber et al., 2005). RNA was run on standard denaturing formaldehyde agarose gels (Sambrook et al., 1989), blotted onto positively charged PVDF membranes and hybridized in Dig Easy Hyb Solution (Roche Diagnostics, Penzberg) with a csgD-specific digoxygenine-labelled probe obtained by PCR using the PCR Dig Labelling Mix (containing Dig-11-dUTP; Roche Diagnostics) and the primers 5′-GTTTAATGAAGTCCATAGTATTC-3′ and 5′-CGCCTGAGGTTATCGTTTGC-3′. For signal detection on the blots, anti-Dig-AP-conjugate and CDP Star Working Solution (Roche Diagnostics) were used. Times of reaction and subsequent exposure to Kodak BioMax Light films was dependent on signal strength. On all gels, the samples shown in the first three lanes in Fig. 5 were always included as a reference for densitometric quantification (with the MC4100 and the ydaM mutant sample used as references for strong and weak csgD mRNA signals respectively).

Microarray analysis

Cells for RNA preparation for microarray analysis were grown and harvested as described above for RNA used in Northern blot analyses. Cell lysis, RNA isolation, DNase I treatment and phenol/chloroform extraction were previously described (Weber et al., 2005). Escherichia coli K-12 microarray slides were obtained from MWG (Ebersberg, Germany). The arrays contain 4288 gene-specific 50 mer oligonucleotide probes representing the whole E. coli genome. The cDNA was labelled with Cy3/5-dCTP and hybridized to the microarray according to the manufacturers protocol. Fluorescence detection and image analysis was carried out as described before (Weber et al., 2005) using a GenePix 4100A (Axon) laser scanner. Each microarray experiment was repeated two times (biological replicates). Genes were considered differentially regulated when (i) signal-to-noise ratios exceeded a factor of three, (ii) the sum of median intensity counts was above 200 and (iii) relative RNA level differences (ratios) were at least twofold in both of the two independent experiments.

SDS-PAGE and immunoblot analysis

Sample preparation for SDS polyacrylamide electrophoresis (PAGE) and immunoblot analysis were as described before (Lange and Hengge-Aronis, 1994). Ten milligrams of cellular protein was applied per lane. A polyclonal serum against CsgD protein, and a goat anti-rabbit IgG alkaline-phosphatase conjugate (Sigma) were used for protein visualization in the presence of a chromogenic substrate (BCIP/NBT; Boehringer Mannheim).

β-Galactosidase assay

β-Galactosidase activity was assayed by use of o-nitrophenyl-β-d-galactopyranoside (ONPG) as a substrate and is reported as μmol of o-nitrophenol per min per mg of cellular protein (Miller, 1972).

Purification of YdaM and YciR and DGC and PDE assays

For protein purification yciR and ydaM were cloned into the expression vector pQE60 (which results in C-terminally His6-tagged gene products; Qiagen), using the primers listed in Table S2 and MC4100 chromosomal DNA as a template for PCR. MC4100 containing pQE60::ydaM or pQE60::yciR was grown in M9 minimal medium (0.5% glucose, 100 μg ml−1 ampicillin) at 30°C to an OD578 of 0.6. Specific protein expression was induced by adding IPTG to a final concentration of 0.04 mM and cultivation was continued overnight at 19°C. Cell lysis and native purification by Ni-NTA affinity chromatography were carried out according to standard protocols (Qiagen). Protein preparations were examined for purity by SDS-PAGE and finally dialysed against 250 mM NaCl, 25 mM Tris-Cl pH 8.0, 10 mM MgCl2 and 5 mM β-mercaptoethanol overnight at 4°C.

Diguanylate cyclase and PDE reactions were performed according to (Christen et al., 2005) with slight modifications. For measuring DGC activity His6-tagged YdaM (5 μM) in 19 μl of reaction buffer (250 mM NaCl, 25 mM Tris-Cl pH 8.0, 10 mM MgCl2, 5 mM β-mercaptoethanol) was incubated with 1 μl of [α-32P]-GTP (10 mCi ml−1, 3000 Ci mmol−1) at 30°C, 5 μl of samples were taken at regular time intervals and reactions were stopped by adding an equal volume of 0.5 M EDTA pH 8.0. A reaction with purified PleD*, a constitutively active DGC (Paul et al., 2004), was included as a positive control. Samples were blotted on Polygram CEL 300 PEI cellulose thin layer chromatography plates (Macherey-Nagel). Plates were developed in 1:1.5 (v/v) saturated NH4SO4 and 1.5 KH2PO4, pH 3.6, dried and exposed on a Phosphor Imaging Screen (Fuji). For measuring PDE activity radioactive c-di-GMP was generated by incubating 19 μl of PleD* (17 μM) with 1 μl of [α-32P]-GTP for 60 min at 30°C. After heat precipitation for 5 min at 95°C followed by centrifugation for 5 min at 14 000 r.p.m., the supernatant with radiolabelled c-di-GMP was isolated. His6-tagged YciR (10 μM) in 19 μl of reaction buffer was incubated with 1 μl of radiolabelled c-di-GMP at 30°C, samples were taken at regular time intervals and the reaction was stopped by adding an equal volume of 0.5 EDTA pH 8.0. Samples were analysed by thin layer chromatographie as described above.

Phenotypic assays

Congo red binding was tested on LB agar plates without salt supplemented with Congo red (40 μg ml−1) and Coomassie brilliant blue (20 μg ml−1) (Römling et al., 1998a) that were incubated for 20 h at 28°C.

For the sedimentation assay, cells were grown for 18–22 h at 28°C in 5 ml of cultures. Cultures were left statically for 5 min, before photographs were taken of the test tubes.

Electron microscopy

Escherichia coli cells were taken directly from culture medium without fixation and transferred to carbon-stabilized Pioloform (Wacker Chemie, Munich) films covering specimen support copper grids. The transfer occurred by cell diffusion in medium droplets and adsorption to adjoining pioloform film, followed by short washing of cells with H2O and final negative staining with 1% uranyl acetate. Electron micrographs were obtained with a Siemens TEM 101 at 80 kV. Negatives were scanned with an Epson 1680 pro Scanner at 1200 dpi.


We thank F. Blattner for the yddV mutant strain. The gift of the pleD* plasmid as well as valuable advice with the DGC and PDE assays by U. Jenal, R. Paul and M. Christen is gratefully acknowledged. We also thank T. Lamparter and N. Michael for advice with protein purification. Financial support was provided by the Deutsche Forschungsgemeinschaft (He 1556/11-3, and 13-1), and the Fonds der Chemischen Industrie.