The Pseudomonas aeruginosa quorum-sensing (QS) systems, Las and Rhl, control the production of several virulence factors and other proteins, which are important to sustain adverse conditions. A comparative transcriptome analysis of a rpoS– and a rpoS–hfq – strain indicated that the Sm-like RNA-binding protein Hfq affects approximately 5% of the P. aeruginosa O1 transcripts. Among these transcripts 72 were identified to be QS regulated. Expression studies revealed that Hfq does not control the master regulators of the Las system, LasR and LasI. Upon entry into stationary phase, Hfq exerted a moderate stimulatory effect on translation of the rhlR gene and on the qscR gene, encoding a LasR/RhlR homologue. However, Hfq considerably stimulated translation of the rhlI gene, encoding the synthetase of the autoinducer N-Butyryl-homoserine lactone (C4-HSL). Correspondingly, the C4-HSL levels were reduced in a hfq– strain. To elucidate the stimulatory effect of Hfq on rhlI expression we asked whether Hfq affects the stability of the regulatory RNAs RsmY and RsmZ, which have been implicated in sequestration of the translational repressor RsmA, which in turn is known to negatively regulate RhlI synthesis. We demonstrate that Hfq binds to and stabilizes the regulatory RNA RsmY, which is further shown to bind to the regulatory protein RsmA. A model for the Hfq regulatory network is presented, wherein an alleviation of the negative effect of RsmA accounts for the observed stimulation of rhlI expression by Hfq. The model is corroborated by the observation that a rsmY– mutant mimics the hfq– phenotype with regard to rhlI expression.
Pseudomonas aeruginosa O1 (PAO1) is an opportunistic pathogen, which causes serious infections in immunocompromised hosts (Van Delden and Iglewski, 1998). Most of the extracellular virulence factors (e.g. exotoxin A, elastase, alkaline protease A, phospholipase C and rhamnolipid) produced by PAO1 are controlled by quorum sensing (QS) (Van Delden and Iglewski, 1998). QS is based on small diffusible molecules, which reach a threshold level at high cell density. They bind to and activate LuxR-like regulators. This activation results in transcription of target genes as well as in an enhanced synthesis of the diffusible molecules themselves. Therefore, the term autoinducer was coined (Fuqua et al., 1994). PAO1 possesses two main QS systems, the LasR/I and RhlR/I system, whereby LasR/RhlR are transcriptional regulators and LasI/RhlI are synthetases for N-(3-oxododecanoyl)-homoserine lactone (3O-C12-HSL) and N-Butyryl-homoserine lactone (C4-HSL) respectively (van Delden and Iglewski, 1998; see Fig. 9A). The two systems are organized in a hierarchical manner and the LasR/I system positively regulates the RhlR/I system (Latifi et al., 1996; Pesci et al., 1997). In addition, many other factors including GacA/GacS, MvfR, PQS, QscR, RpoN, RpoS, RsaL, RsmA and Vfr have been shown to control QS (Albus et al., 1997; Reimmann et al., 1997; de Kievit et al., 1999; Cao et al., 2001; Chugani et al., 2001; Heurlier et al., 2003; Schuster et al., 2004).
Pessi et al. (2001) identified in PAO1 a homologue of the Escherichia coli CsrA protein (Romeo, 1998) termed RsmA (regulator of secondary metabolites). The CsrA protein regulates cell surface properties and carbon metabolism in E. coli by acting as a translational repressor on several mRNAs (Baker et al., 2002; Dubey et al., 2003). The function of CsrA can be antagonized by two small RNAs (Romeo, 1998; Weilbacher et al., 2003), which bind to and thereby inactivate CsrA. Similarly, the PAO1 global regulatory protein RsmA has been shown to repress a number of target mRNAs including lasI and rhlI, encoding the enzymes synthesizing the autoinducers of both QS systems and the hcn mRNA, encoding the cyanide biosynthesis genes (Pessi et al., 2001). Two small RNAs, RsmY (Valverde et al., 2003; Heeb et al., 2004) and RsmZ (Heurlier et al., 2004), have been recognized in PAO1. Both display repeated GGA motifs, which are typical traits for RNAs binding to the CsrA/RsmA class of proteins, and recent work has provided evidence that RsmZ RNA titrates the PAO1 RsmA protein (Heurlier et al., 2004).
In this study, we examined the effect of Hfq on global gene expression by microarray analysis. A number of genes were identified that were QS and Hfq controlled. To test whether Hfq affects expression of the QS master regulator genes at the post-transcriptional level, translational lacZ fusions to lasR, lasI, rhlR, rhlI and qscR were employed. We observed a strong stimulatory effect of Hfq on translation of rhlI during late logarithmic growth and at high cell densities. Furthermore, our data revealed that Hfq stabilizes the small regulatory RNA RsmY. In addition, we show that RsmY binds to both, the translational regulator RsmA and Hfq. As RsmA was shown to be a negative regulator of rhlI expression (Pessi et al., 2001), these data suggest that Hfq indirectly affects QS via regulation of the RsmA/RsmY system.
Hfq-dependent alterations of the transcriptome
Hfq mediates post-transcriptional regulation in E. coli, which can result in inactivation of mRNAs (Tsui et al., 1997; Vytvytska et al., 2000; Masséet al., 2003; Afonyushkin et al., 2005; Udekwu et al., 2005). We therefore anticipated a difference in the transcript abundance in a PAO1hfq– strain when compared with the isogenic parental strain. When compared with late logarithmic growth [optical density at 600 nm (OD600) = 1.0] the abundance of Hfq in PAO1 increased about twofold (Fig. 1) upon entry into stationary phase (OD600 of 2.0). Therefore, total RNA was isolated from the PAO1 strains upon growth to an OD600 of 2.5 with the reasoning that larger differences in the transcript levels might be observed. The microarray analysis revealed that approximately 15% of the PAO1 genes showed a twofold or greater difference in transcript abundance in the PAO1hfq– strain when compared with that in PAO1 (see Supplementary material, Tables S1 and S2).
As Hfq exerts a stimulatory effect on rpoS expression in PAO1 (Sonnleitner et al., 2003), the transcriptome analysis was likewise performed with a PAO1rpoS– strain and a PAO1hfq–rpoS – strain with the rationale that this could disclose σS-independent targets of Hfq. We found 283 genes showing a twofold or greater difference that were affected by Hfq; 197 genes and 86 genes were up- and downregulated, respectively, in the hfq–rpoS– background (see Supplementary material, Tables S3 and S4). The genes were compiled according to the Pseudomonas genome project (http://www.pseudomonas.com). The three over-represented functional classes of genes found to be upregulated in the hfq–rpoS– genetic background are involved in carbon compound catabolism, related to phages, transposons, plasmids, or relevant for transport of small molecules (Fig. 2, left). On the other hand, genes encoding secreted factors (toxins, enzymes, alginate), involved in fatty acid and phospholipid metabolism, in cell wall, lipopolysaccharide (LPS) and capsule synthesis, were downregulated in the hfq–rpoS– background (Fig. 2, right). In particular, 9 quorum sensing-controlled virulence genes, PA1250 (aprI), PA1871 (lasA), PA2300 (chiC), PA3478 (rhlB), PA3479 (rhlA) and PA5161-PA5164 (rmlBDAC) (see Supplementary material, Table S4), were downregulated in the hfq–rpoS– mutant, which can be reconciled with the recently reported attenuation in virulence of a PAO1hfq– strain (Sonnleitner et al., 2003).
Hfq and QS regulation
Recent studies revealed that more than 300 genes of PAO1 are QS controlled (Schuster et al., 2003; Wagner et al., 2003). The transcript abundance of 72 of these genes was affected by Hfq. When compared with the PAO1rpoS– strain 46 genes were downregulated in the hfq –rpoS– double mutant and 26 were upregulated (see Supplementary material, Tables S3 and S4). Seventeen genes out of these 26 are also upregulated by QS and all transcripts but one (PA4359), which was downregulated in the hfq–rpoS– backgrounds, are stimulated by QS (see Supplementary material, Tables S3 and S4). One explanation for these findings could be that Hfq affects a negative as well as a positive regulator(s) of QS. However, the microarray data predicted that the majority of the QS regulatory genes were only marginally affected by Hfq (less than a twofold change). Only the transcript abundance of both rhlR (−5-fold change in the hfq –rpoS– strain and −2.3-fold change in the hfq– strain) and qscR (5.5-fold change in the hfq–rpoS– strain and 4.4-fold change in the hfq– strain) encoding the transcriptional regulators RhlR and QscR, respectively, was altered in the hfq– backgrounds (see Supplementary material, Tables S1–S4).
Hfq stimulates rhlI translation
In E. coli Hfq is known to mediate gene expression at the post-transcriptional level (Valentin-Hansen et al., 2004). As Hfq influenced the transcript abundance of many QS-regulated genes and, as mentioned above, the expression of at least two QS regulators was altered in the absence of Hfq, we asked whether Hfq affects translation of the genes encoding the master regulators of the LasR/I and RhlR/I QS systems during different phases of growth. To test this possibility, translational fusions of lasR, lasI, rhlR and rhlI to the lacZ reporter gene (Table 1) were employed and the synthesis of the respective fusion protein was monitored at different cell densities in PAO1hfq– and in the isogenic parental strain. To assess Hfq-dependent effects on translation the β-galactosidase activities were normalized to the respective mRNA concentrations, which in turn were normalized to 5S ribosomal RNA.
QscR is known to repress lasI as well as the phz (phenanzine) and hcn (hydrogen cyanide) gene clusters (Chugani et al., 2001). The transcriptome analysis showed that Hfq exerts a negative effect on the qscR gene. To test whether Hfq mediates this effect on qscR at the post-transcriptional level, the relative translational efficiency of a qscR–lacZ fusion was determined at different phases of growth. Between an OD600 of 1 and 2, the relative translational efficiency of the qscR–lacZ gene was higher in the hfq+ than in the hfq– strain (Fig. 3A). After an OD600 of 1.5, the expression of the qscR–lacZ fusion increased somewhat in the hfq – mutant, but remained below the level observed in the wild-type strain in stationary phase (OD600 = 2.5; Fig. 3A), which was at variance with the higher abundance of the qscR transcript in the absence of Hfq (see Supplementary material, Table S3).
As shown in Fig. 3B, at an OD600 of 1.5 the relative translational efficiency of the lasR–lacZ mRNA was approximately 40% decreased in the hfq – strain when compared with that obtained with PAO1, whereas it was hardly affected at lower and higher cell densities. Translation of the lasI–lacZ fusion was slightly elevated at OD600 between 1 and 1.5 in the hfq– strain when compared with the wild-type strain (Fig. 3C). However, at OD600 values higher than 2, only marginal differences in the relative translation levels were observed in either strain (Fig. 3C). Previous results have shown that the alkaline protease is regulated by the Las system (Gambello et al., 1993; Schuster et al., 2003). As another means to assess whether Hfq affects the LasR/LasI system we compared the alkaline protease activity in the PAO1hfq– mutant and in the parental strain at an OD600 of 2.5. The protease activity in the PAO1 wild-type strain was 4.57 ± 0.13 U ml−1 and in the PAO1hfq– mutant 4.65 ± 0.06 U ml−1. Taken together, these assays suggested that in stationary-phase Hfq has no major influence on translation of the lasR/lasI genes.
In agreement with the transcriptome analysis, the relative translation of the rhlR–lacZ gene was somewhat lower in the hfq– mutant during all phases of growth (Fig. 3D). The most striking effect of Hfq on translation of the QS master regulator genes was observed on rhlI–lacZ expression. While the relative translational efficiency of this fusion gene was slightly elevated during logarithmic growth in the hfq– strain, it decreased approximately sixfold at an OD600 of 2.5 in the mutant strain when compared with the wild type (Fig. 3E). In addition, we tested whether the decrease in rhlI translation in stationary phase is paralleled by a lower C4-HSL concentration in PAO1hfq – when compared with the wild-type strain. The concentration of the autoinducer released by PAO1 and the PAO1hfq– strain in the culture supernatant at an OD600 of 2.5 was determined by using the indirect method described by Pearson et al. (1997). The C4-HSL levels in the wild-type strain and hfq– strain were 3.06 ± 0.01 µM and 1.42 ± 0.19 µM, respectively, again reflecting a reduced RhlI synthesis in the absence of Hfq.
The PAO1 pqsH gene, which was downregulated −4.6-fold and −7.6-fold in the hfq– mutant and in the hfq–rpoS – double mutant (see Supplementary material, Tables S2 and S4), respectively, encodes an enzyme required for quinolone signal (PQS) synthesis (Gallagher et al., 2002). PQS was shown to positively regulate rhlI at the transcriptional level (McKnight et al., 2000; Gallagher et al., 2002). The microarray data revealed an approximately −1.5-fold change in the rhlI transcript level in the hfq– strains when compared with the wild-type strains (not shown), which was also reflected in a similar reduction in the β-galactosidase activity obtained with a transcriptional rhlI–lacZ fusion gene in PAO1hfq– when compared with the wild-type strain (E. Sonnleitner, unpubl. result). These experiments are paralleled by the observed approximately 1.2-fold reduction of rhlI transcription in a PAO1pqsH– mutant strain (Gallagher et al., 2002). However, as the β-galactosidase activities obtained with the rhlI–lacZ translational fusion in the hfq+ and hfq– backgrounds were normalized to the respective mRNA concentration (Fig. 3E), we expected that the Hfq-dependent strong stimulation of RhlI synthesis in stationary phase is primarily exerted at the translational level rather than at the level of transcription.
Hfq interferes with the RsmA/RsmY/RsmZ control system: Hfq affects hcnA, a target of RsmA regulation without altering rsmA translation
Several experiments including in vitro translation and toeprinting assays did not reveal a direct effect of purified PAO1 Hfq protein on rhlI mRNA translation (not shown). We therefore asked whether the Hfq could exert its positive effect on rhlI translation via the RsmA/RsmY/RsmZ control system. The post-transcriptional regulator RsmA, the concentration of which increases with higher cell densities, has been shown to repress translation of rhlI and lasI mRNA (Pessi et al., 2001). Two small RNAs, RsmY and RsmZ, the synthesis of which likewise increases at higher cell densities, have the potential to bind to and to antagonize the function of this protein (Valverde et al., 2003; Heeb et al., 2004; Heurlier et al., 2004). In E. coli a number of sRNAs associate with Hfq (Zhang et al., 2003) and some seem to require Hfq for stability control (Sledjeski et al., 2001; Moll et al., 2003). Therefore, we next examined whether the stimulatory effect of Hfq on rhlI translation at high cell densities could be explained by stabilization of RsmY and/or RsmZ, and thereby sequestration of RsmA, which in turn would relieve the negative effect of RsmA on translation of the rhlI mRNA.
Pessi et al. (2001) have shown that RsmA represses translation of the hcnA gene encoding a function required for cyanide biosynthesis. As a means to test the validity of our hypothesis we first asked whether Hfq controls hcnA expression. The translational hcnA–lacZ fusion construct used in this study contained the natural ribosome and RsmA binding sites but the authentic QS-dependent hcn promoter was replaced by the constitutively expressed tac promoter (Pessi et al., 2001). In late stationary phase (OD600 > 2.0) the relative translational efficiency of hcnA increased considerably in the wild-type strain when compared with that in the hfq– mutant (Fig. 4A), indicating that Hfq may indeed affect the RsmA/RsmY/RsmZ control circuit.
To exclude that Hfq affects the synthesis of RsmA, we tested whether a rsmA–lacZ translational fusion was differentially expressed in the wild-type strain and in the hfq– strain. As shown in Fig. 4B, there was no significant difference in the expression of the rsmA–lacZ mRNA in the wild-type strain and in the hfq– strain, demonstrating that Hfq does not affect rsmA expression in a direct manner. These experiments prompted us to test whether Hfq interacts with and affects the stability of the RsmY/RsmZ transcripts.
Hfq affects the abundance and stability of RsmY
To examine whether Hfq affects the stability of the RsmY and RsmZ RNAs we determined the steady-state levels of RsmY and RsmZ at an OD600 of 2.5 in both, the hfq– and the wild-type strain. When compared with PAO1 the steady-state level of RsmY was reduced by about 50% in the hfq– strain, whereas the steady-state level of RsmZ was hardly affected in either strain (data not shown). We therefore tested whether Hfq affects the stability of RsmY and RsmZ. At an OD600 of 2.5 rifampicin was added to the wild-type strain and to the hfq – mutant, and aliquots were withdrawn for total RNA isolation at different times thereafter. As shown in Fig. 5A and C, the half-life of RsmY was about 60 and 20 min in the wild-type strain and in the hfq– mutant respectively. In contrast, the stability of RsmZ was not significantly influenced by Hfq (Fig. 5B and C).
RsmY RNA binds specifically to RsmA and Hfq
As a physical interaction of RsmY RNA with RsmA protein has as yet not been demonstrated for PAO1, we next used band shift assays and the purified components of the Rsm system as well as purified PAO1 Hfq protein to examine whether RsmY binds specifically to both regulatory proteins. The RsmY RNA migrated as a single band in a denaturing gel (not shown) from which it was purified. In the native gel, the RsmY RNA resolved into two bands, which suggested the presence of two conformers (Fig. 6, lane 1: F1 and F2). Hfq was added in increasing molar ratios to labelled RsmY RNA. When Hfq was added in a fivefold molar excess (Hfq hexamer) over the RNA different band shifts were observed (Fig. 6, lane 2). The Kd values were 22 ± 10 nM (when 50% of RNA was present in band B1) and 72 ± 18 nM (when 50% of RNA was present in band B2) respectively. Complexes with a higher molecular weight appeared when Hfq was added in higher molar ratios to RsmY RNA (Fig. 6, lanes 3–5), which indicated binding of several Hfq hexamers to one RsmY molecule. The competition experiments indicated specific binding of Hfq to RsmY. Unlabelled RsmY RNA competed with the Hfq–RsmY complex (Fig. 6, lanes 6–9), whereas the non-specific competitor polyC did not, even at a concentration of 200 ng, which would correspond to an approximately 60-fold molar excess of a 200 nt long polyC RNA (Fig. 6, lanes 10–12).
Incubation of 5 nM 5′-end-labelled RsmY RNA with various concentrations of RsmA yielded three RsmA–RsmY complexes (B1–B3; Fig. 7A) and one additional band occurred in the competition assay (B4, Fig. 7B, lane 3). As free RsmY RNA showed two different conformations (Fig. 7A and B, lane 1), the supershifted forms could result either from binding of RsmA to this second conformation of RsmY (F2) and/or from binding of several RsmA molecules to one molecule of RsmY RNA. Similar multiple supershifts have been observed for RsmY in Pseudomonas fluorescens, and were interpreted to correspond to higher-molecular-weight species with several molecules of RsmA bound to one RsmY RNA molecule (Valverde et al., 2004). At least seven GGA motifs, which have been recently shown to be important for RsmA binding in P. fluorescens (Valverde et al., 2004), are present in the P. aeruginosa RsmY sequence (Fig. 7C). The bands corresponding to free RsmY RNA were completely shifted between 100 and 200 nM of RsmA (Fig. 7A, lanes 5 and 6), which was higher than observed in P. fluorescens (33–66 nM RsmA; Valverde et al., 2004). In addition, our binding assays indicated that some of the RsmY–RsmA complexes are preferentially formed: only the complex represented by band B2 shows the presence of more than 50% of bound RNA (Kd = 55 ± 7 nM) (Fig. 7A, lanes 4–9).
The competition assay demonstrated that RsmY binds specifically to RsmA (Fig. 7B, lanes 3–5). A molar ratio of 1:40:5 of labelled RsmY to RsmA protein to unlabelled RsmY RNA resulted in a downshift (Fig. 7B, lane 3), which occurred quantitatively when 200 nM competitor RNA was added (Fig. 7B, lane 5). RsmZ seemed to be a slightly better competitor (Fig. 7B, lanes 6–8), which is consistent with previous results in P. fluorescens (Valverde et al., 2003). The non-specific competitor polyC was not able to compete with RsmA–RsmY complexes (Fig. 7B, lanes 9–11).
A rsmY mutant mimics the rhlI phenotype of the hfq– mutant
The experiments described above lent support to the hypothesis that Hfq stimulates rhlI translation indirectly through stabilization of RsmY. This would increase the degree of inactivation of RsmA, which in turn would explain the enhanced expression of rhlI in stationary phase in the wild-type strain when compared with the hfq – strain. To further substantiate this hypothesis a PAO1rsmY deletion mutant was constructed (see Experimental procedures and Supplementary material, Fig. S1), and expression of the rhlI–lacZ fusion gene was monitored in the wild-type strain, in the PAO1hfq– and in the PAO1ΔrsmY strain. The expression of the rhlI–lacZ fusion gene was reduced in the PAO1ΔrsmY mutant when compared with the wild-type strain (Fig. 8), which mimicked the hfq– phenotype with regard to rhlI expression. Hence, this experiment corroborated our hypothesis that Hfq stimulates rhlI expression through stabilization of RsmY RNA. It should be noted that transcription of the rhlI–lacZ gene in plasmid pME3846 (Table 1) is driven by the authentic rhlI promoter, which is positively regulated by PQS (see above). As the β-galactosidase values shown in Fig. 8 were not normalized to the corresponding mRNA levels, the somewhat reduced expression of the rhlI–lacZ gene in the PAO1hfq– strain when compared with the PAO1ΔrsmY strain most likely reflects the reduced transcription of the pqsH gene in the absence of Hfq (see Supplementary material, Tables S2 and S4).
Hfq, a global regulator in P. aeruginosa
The transcriptome analysis indicated that Hfq is involved in the regulation of several different pathways. Hfq affected housekeeping genes containing Fe-S clusters, the alcohol dehydrogenase metabolism, pyochelin and pyocin production, degradation of aromatic compounds as well as ABC transport systems. Moreover, Hfq exerted an effect on QS-controlled genes, and thus on the regulation of several virulence factors (see Supplementary material, Table S4), which in turn explains the attenuated virulence of a PAO1hfq– mutant (Sonnleitner et al., 2003).
In E. coli Hfq not only regulates gene expression at the level of individual genes but can govern gene expression at a superior level by modulating the expression of global transcriptional regulators, such as Fur (Večerek et al., 2003), RpoS (Muffler et al., 1996) or H-NS (Lease and Belfort, 2000). Hfq can modulate the expression of such regulatory functions by facilitating the interaction of sRNAs with their target mRNA (Lenz et al., 2004). Thus, it is difficult to distinguish between direct and indirect effects of Hfq. As the majority of genes identified in the microarray study are upregulated in the hfq – background, it appears that much of the regulation by Hfq in P. aeruginosa is negative, which is in agreement with previous studies in E. coli (Večerek et al., 2003). On the other hand, it remains to be shown whether downregulation in the hfq – background, i.e. positive regulation by Hfq, is exerted via expression modulation of transcriptional regulators, the stability and/or function of sRNA(s) or both. For instance, two sRNAs (PrrF1 and PrrF2) involved in P. aeruginosa iron metabolism have a destabilizing effect on their target mRNAs. As the master regulator of iron metabolism, Fur, controls the transcription of PrrF1 and PrrF2, Fur has indirectly a positive effect on the target genes of the sRNAs (Wilderman et al., 2004).
Effect of Hfq on QS
Hfq exerted only slight effects on lasR, lasI, and a moderate effect on rhlR translation at high cell densities. In contrast, the presence of Hfq was required for stimulation of rhlI translation at high cell densities. We have recently shown that the synthesis of elastase encoded by the lasB gene was approximately sixfold reduced in a PAO1hfq– mutant. The expression of the lasB gene is positively regulated by the PQS as well as by the Rhl system (Fig. 9A; Latifi et al., 1996; Pearson et al., 1997; McKnight et al., 2000). The observed reduction in elastase production in the PAO1hfq – strain therefore appears to reflect the reduced abundance of the pqsH transcript (see Supplementary material, Table S2) involved in PQS synthesis as well as the reduced translation of the rhlI gene.
Neither the transcriptome analysis nor the studies employing lasR/lasI translational fusion revealed a significant effect of Hfq on the Las system. In addition, the expression of the Las-regulated alkaline protease gene aprA (Gambello et al., 1993; Schuster et al., 2003) was not affected by Hfq. LasR/LasI are also known as positive regulators of lasB expression (Gambello and Iglewski, 1991). Therefore, the residual levels of elastase observed in the PAO1hfq– strain (Sonnleitner et al., 2003) seem to result from the LasI/LasR-directed expression of lasB.
QscR is a LuxR-like regulator similar to LasR and RhlR (Chugani et al., 2001). The translation of qscR was higher in the wild-type strain than in the hfq– mutant during the examined growth phases, which would contradict the result from the transcriptome profile. At this point we can only speculate that Hfq is somehow involved in positive translational regulation of qscR mRNA being it via a sRNA-mediated mechanism or by governing the synthesis of a positive regulator. It was previously shown that QscR represses the QS-controlled gene clusters phz (phenanzine) (Chugani et al., 2001) during exponential growth. The positive control of qscR by Hfq can be reconciled with our previous finding that Hfq negatively controls pyocyanin production (Sonnleitner et al., 2003).
RhlI regulation by Hfq is mediated by the RsmA/RsmY control system
Based on the observations: (i) that Hfq did not directly affect the expression of the rsmA gene (Fig. 4), (ii) that Hfq binds specifically to and stabilizes RsmY RNA (Figs 5 and 6) and (iii) that the rsmY deletion mutant mimicked the hfq – phenotype with regard to downregulation of rhlI (Fig. 8), we suggest the following model for the observed Hfq-dependent effects on QS and QS-regulated genes (Fig. 9A and B). The synthesis of RsmA (Pessi et al., 2001), RsmZ (Heurlier et al., 2004), RsmY (Valverde et al., 2003) as well as of Hfq (Fig. 1) increases upon entry into stationary phase. At this point RsmA is expected to exert a larger negative effect on rhlI than on lasI expression (Pessi et al., 2001). As Hfq stabilizes RsmY, which in turn inactivates RsmA, an alleviation of the negative effect of RsmA on rhlI would explain the observed stimulation of rhlI translation by Hfq in stationary phase (Fig. 9B, left). It is worth noting that Hfq and RsmA can bind concurrently to RsmY (T. Sorger-Domenigg, unpubl. results). RsmA binding to P. fluorescens RsmY RNA requires GGA motifs (Valverde et al., 2004), whereas binding of Hfq depends on a single-stranded A/U-rich region preceded or followed by a stem-loop structure (Moll et al., 2003). Binding of Hfq to single-stranded regions could therefore explain the protection of RsmY RNA by Hfq from a single-strand specific RNA endonulease(s). On the other hand, the reduced expression of rhlI in the PAO1hfq– mutant can be reconciled with a lack of inactivation of RsmA by RsmY RNA, resulting from a reduced stability of the sRNA in the absence of Hfq (Fig. 9B, right).
In summary, these studies suggest that Hfq affects QS and QS-controlled genes in at least three ways: an unspecified positive regulation of the QS repressor QscR and of the pqsH gene, and by RsmY-mediated indirect positive regulation of the QS regulator RhlI (Fig. 9A).
Bacterial strains, plasmids and growth conditions
The strains and plasmids used in this study are listed in Table 1. PAO1 cultures were grown at 37°C in Luria–Bertani (LB) medium (Miller, 1972), supplemented with the appropriate antibiotics. Antibiotics were added to the following final concentrations: 50 µg ml−1 gentamicin, 125 µg ml−1 tetracycline, 200 µg ml−1 spectinomycin/streptomycin and 300 µg ml−1 carbenicillin.
Detection of Hfq by immunoblotting
Strain PAO1 was grown in LB medium at 37°C. Samples were taken at different optical densities (OD600 = 1.0, 2.0 and 2.5). As a negative control, the PAO1hfq– strain was used. Equal amounts of total protein of either sample were separated on a 12% SDS-polyacrylamide gel, and then transferred to nitrocellulose membranes (Schleicher and Schuell) by electroblotting. The blots were blocked with 5% dry milk in TBS (8 g l−1 NaCl, 0.2 g l−1 KCl and 3 g l−1 Tris-base in water, pH 7.5), and then probed with anti-Hfq antibodies raised against purified P. aeruginosa Hfq protein (Pineda-Antibody-Service, Berlin). The antibody–antigen complex was visualized with goat anti-rabbit immunoglobulin alkaline-phosphatase-conjugated antibody (Sigma Immuno Chemicals) using NBT (Nitroblue-tetrazolium-chloride; BIOMOL) and BCIP (5-Bromo-4-chloro-3-indolyl phosphate toluidine salt; BIOMOL) in alkaline phosphatase-buffer (10 mM NaCl, 5 mM MgCl2, 100 mM Tris/HCl, pH 9.5) as a chromogenic substrate. The quantification was performed with ImageQuant software (Molecular Dynamics; Version 5.1).
Expression profiling experiment
Total RNA was isolated from strains PAO1, PAO1hfq–, PAO1rpoS– and PAO1rpoS–hfq– grown to an OD600 of 2.5. A total of 1 × 109 cells were mixed with RNA Protect Bacteria reagent (Qiagen) and processed as recommended by the manufacturer. RNA purification, cDNA synthesis, cDNA fragmentation and 3′-end labelling was performed as previously described (Schuster et al., 2003). Fragmented and labelled cDNA was hybridized to GeneChip arrays (Affymetrix) by overnight incubation at 50°C. Washing, staining and scanning of the microarrays were performed with the Affymetrix fluidic station. All experiments were performed in duplicate. Affymetrix Microarray Software suite (MAS) (version 5.0) was used to determine the transcript levels and differences in transcript levels when different samples were compared. Affymetrix scaling was used to normalize the data from different arrays. For comparison analysis, the log2 ratio for absolute transcript signals obtained from a given pair of arrays was calculated by using MAS. A statistical algorithm of the software was also assigned for each transcript pair when the level of the respective transcript was significantly altered compared with the level of the baseline sample. To assess the effects of Hfq on transcript abundance we compared in the first experiment the transcriptomes of PAO1 with that of a PAO1hfq – strain. To reveal RpoS-independent target genes of Hfq, the transcriptional profile of the PAO1hfq –rpoS– double mutant was compared with that of the isogenic PAO1rpoS – strain. The fold change was considered to be significant, when the pair PAO1 versus PAO1hfq– or PAO1hfq –rpoS– versus PAO1rpoS– revealed at least a twofold increase or decrease.
Construction of plasmids pME-lasR–lacZ, pME-rhlR–lacZ and pME-qscR–lacZ
For construction of translational fusions between lasR, rhlR and qscR, respectively, and the lacZ gene, the EcoRI–PstI fragment of plasmid pME3859, containing the rsmA fragment, was replaced by PCR-amplified lasR′, rhlR′ and qscR′ fragments. Chromosomal DNA of PAO1 was used as a template. The lasR′ fragment comprised 344 bp of the 5′-untranslated region (5′UTR) and the first six lasR codons. It was amplified using the primer pair V25 and W25 (see Supplementary material, Table S5). The PCR with primers B26 and Z27 (see Supplementary material, Table S5) resulted in the rhlR′ fragment containing 278 bp of the 5′UTR and the first six rhlR codons. The qscR′ fragment (primers S27 and R27: see Supplementary material, Table S5) contained 741 bp of the 5′UTR and the first seven qscR codons. The insertion of these DNA fragments into EcoRI- and PstI-cleaved pME3859 resulted in plasmids pME-lasR–lacZ, pME-rhlR–lacZ and pME-qscR–lacZ respectively.
Construction of pET-rsmA
The RsmA production plasmid was constructed as follows. The rsmA-coding region was amplified by PCR using the forward primer Y28 (5′-TTT TTC ATA TGC TGA TTC TGA CTC GTC G-3′) containing a NdeI restriction site and the reverse primer Z28 (5′-TTT TTT CTC GAG ATG GTT TGG CTC TTG ATC-3′) containing a XhoI site (the restriction sites are highlighted in bold in either sequence). Then, the rsmA fragment was fused in frame to six histidine codons by cloning rsmA into the corresponding sites of the pET22b vector under control of a T7 promoter resulting in plasmid pET-rsmA. The pET-rsmA-encoded RsmA protein contains a C-terminal His6-tag.
The strains PAO1 and PAO1hfq– were transformed with the reporter constructs pME3843, pME3846, pME3853, pME3859, pME-lasR–lacZ, pME-rhlR–lacZ or pME-qscR–lacZ. Samples for β-galactosidase measurements together with samples for total RNA isolation were withdrawn at different times during growth. To assess Hfq-specific effects on translation and to account for possible variables such as mRNA levels and/or mRNA stability or plasmid copy number in the PAO1 and PAO1hfq– strains, the β-galactosidase activities were normalized to the respective mRNA concentration, which in turn was normalized to 5S ribosomal RNA. Total RNA was purified by the hot phenol method (Lin-Chao and Bremer, 1986). The mRNA concentrations of the respective lacZ fusions were determined by primer extension with AMV reverse transcriptase (Promega) using 4 µg of total RNA and the 5′-end 32P-labelled lacZ-specific oligonucleotide F6 (5′-GGT TTC CCA GTC ACG AC-3′) (0.5 pmol). The concentration of the 5S rRNA was likewise determined by primer extension using the 5′-end 32P-labelled oligonucleotide I26 (5′-CCC CAC ACT ACC ATC GGC GAT GCG TCG-3′). The signals were visualized by a Phosphorimager (Molecular Dynamics) and quantified by ImageQuant software (Molecular Dynamics). The relative β-galactosidase values were determined by normalization of the respective β-galactosidase values to the concentrations of the corresponding lasR–lacZ, lasI–lacZ, rhlR–lacZ, rhlI–lacZ, qscR–lacZ and hcnA–lacZ mRNAs, which in turn were normalized in either case to the 5S rRNA levels. As there was no significant difference in the rsmA–lacZ mRNA levels between the PAO1 and the PAO1hfq – strain the β-galactosidase read out obtained with the rsmA–lacZ fusion gene (see Fig. 4B) was not normalized to the corresponding mRNA level(s). The respective β-galactosidase activities obtained with the rhlI–lacZ fusion gene in PAO1, PAO1hfq– and PAOΔrsmY were not normalized to mRNA levels (see Fig. 8). All experiments were performed in duplicate.
Concentrations of C4-HSL were measured with a bioassay as previously described (Pearson et al., 1995; 1997). In short, C4-HSL was extracted with acidified ethyl acetate from supernatants of PAO1 and PAO1hfq–, which were grown in LB medium to an OD600 of 2.5. For detection of C4-HSL, an overnight culture of E. coli strain DH5α bearing plasmid pECP61.5 (Table 1) was diluted to an OD600 of 0.05 in LB medium and at an OD600 of 0.2 IPTG (1 mM final concentration) was added. At an OD600 of 0.5, 0.5 ml aliquots of this culture were added to dry ethyl acetate extracts of the PAO1 strains or to standards. Growth was continued for 3 h and then β-galactosidase was assayed with a GALACTO-Light Kit (Applied Biosystems). Synthetic C4-HSL was used to generate standard curves.
Determination of alkaline protease activity
PAO1 and PAO1hfq– were grown at 37°C in LB medium, and harvested by centrifugation at 5000 g for 5 min at an OD600 of 2.5. Protease activity was determined as described by Howe and Iglewski (1984). Briefly, 500 µl of culture supernatant was added to 1.5 ml of buffer (20 mM Tris/HCl, pH 8.0, 1 mM CaCl2) and 50 mg of hide azure blue powder (Sigma), and incubated at 37°C for 1 h with constant rotation. Then, the tubes were chilled on ice for 10 min and centrifuged at 13 000 g for 5 min and the absorbance was determined at 595 nm. The protease activity was determined as units per millilitre (U ml−1), whereby one unit is equivalent to an increase of one A595 unit per hour at 37°C.
Determination of RsmY and RsmZ stability and abundance
The steady-state level and the stability of RsmY and RsmZ were determined in strains PAO1 and PAO1hfq–. At an OD600 of 2.5, rifampicin (final concentration 500 µg ml−1) was added and 4 ml aliquots were withdrawn at 0, 10, 20, 30, 40 and 60 min thereafter for isolation of total RNA. Total RNA was purified by the hot phenol method (Lin-Chao and Bremer, 1986). The RsmY and RsmZ concentrations were determined by primer extension with AMV reverse transcriptase (Promega) using 2 µg of total RNA and the 5′-end-labelled oligonucleotides T26 (5′-GAT TTC CTG AGT TTC C-3′) for RsmY and S26 (5′-CCC TTC CCC GAT CC-3′) for RsmZ. The 5S rRNA was used as an internal control and was detected by primer extension with primer I26 as described above.
Purification of PAO1 Hfq protein
Escherichia coli strain BL21(DE3) harbouring the plasmid pHFQPA (Table 1) was incubated at 37°C in LB medium. At an OD600 of 0.6, Hfq synthesis was induced by addition of IPTG (1 mM final concentration) and the cells were harvested after 3 h. The cells were resuspended in lysis buffer (50 mM Tris/HCl, pH 8.0, 250 mM MgCl2, 1.5 M NaCl, 1 mM EDTA, 5 mM DTT and 0.5 mM PMSF), and then lysed using a French press (SimAminco). The lysate was heated for 20 min at 85°C, and then centrifuged at 18 000 r.p.m. for 40 min. After addition of ammonium sulphate (1.5 M final concentration), the lysate was loaded on a Butyl-sepharose column (Sigma). The column was washed with buffer (1.5 M NaCl, 1.5 M ammonium sulphate and 50 mM Tris/HCl, pH 8.0). Then, the Hfq protein was eluted with 50 mM Tris/HCl, pH 8.0, and dialysed against and stored in VD buffer (10 mM Tris, pH 7.4, 60 mM NH4Cl, 6 mM β-Mercaptoethanol and 2 mM Mg-acetate).
Purification of PAO1 RsmA protein
Plasmid pET-rsmA was transformed into E. coli strain BL21 (DE3). At an OD600 of 0.5, RsmA synthesis was induced by addition of IPTG (1 mM final concentration) and the cells were harvested after 3 h. The cells were resuspended in lysis buffer (50 mM NaH2PO4, pH 8.0, 300 mM NaCl, 10 mM imidazole and 0.5 mM PMSF) and then lysed using a French press (SimAminco). The lysate was loaded on a Ni-NTA agarose column (Qiagen). The column was washed twice with four volumes of buffer (50 mM NaH2PO4, pH 8.0; 300 mM NaCl) containing 20 mM imidazole and four times with one volume of buffer containing 250 mM imidazole. Then, the RsmA protein was eluted with buffer containing 500 mM imidazole. The purified RsmA protein was dialysed against and stored in VD buffer (see above) containing 2 mM Mg-acetate.
RsmY and RsmZ RNA were in vitro transcribed with T7 RNA polymerase (Fermentas) from PCR templates. The rsmY PCR fragment was amplified using the forward primer U26 5′-CCG TCT AGA CGT AAT ACG ACT CAC TAT AGT CAG GAC ATT GCG CAG GAA GC-3′ and the reverse primer V26 5′-AAA CTG CAG GAG CGA CGC GGT TTT CCT CGG GC-3′. For the rsmZ PCR template the forward primer W26 5′-CCG TCT AGA CGT AAT ACG ACT CAC TAT AGC GTA CAG GGA ACA CGC-3′ and the reverse primer X26 5′-AAA CTG CAG GGC GAC GAG TAA AAC GGC AGG C-3′ were used. Both forward primers, U26 and W26, contained a T7Φ10 promoter sequence (underlined). After in vitro transcription the sRNAs were resolved on a 6% polyacrylamide-8 M urea gel and purified from a single band using standard procedures. The RNA concentration was determined by measuring the A260.
Gel mobility shift assays
Gel-purified RsmY RNA was 5′-end labelled with [γ-32P]-ATP (Amersham), and re-purified on a 6% polyacrylamide-8 M urea gel. Labelled RNA (0.05 pmol) was incubated without or with increasing amounts of purified Hfq hexamer (Hfq6) protein (as indicated in the legends to Figs 6 and 7) in a 10 µl reaction in VD buffer (see above) for 30 min at 4°C. RsmZ and/or RsmY RNA were used as specific competitors and polyC RNA (Amersham) as a non-specific competitor. As the exact length of the polyC RNA is not specified, we used concentrations (1, 10 and 200 ng), which would correspond to a 0.3-, 3- and 60-fold molar excess of a polyC-200-mer respectively. Immediately before loading, the samples were mixed with 4 µl of loading dye (25% glycerol, xylencyanol, bromphenol blue) and loaded on a native 4% polyacrylamide gel. Electrophoresis was performed in TAE buffer at 160 V. The radioactively labelled bands were visualized with a PhosphorImager (Molecular Dynamics).
Construction of a P. aeruginosa rsmY deletion mutant
The rsmY upstream sequence (nucleotides −589 to −36 with regard to the transcriptional start site of the rsmY transcript (Valverde et al., 2003) were amplified by PCR using chromosomal DNA of PAO1 together with the forward primer F30 (5′-TTTTTTTTGTCGACGCCAGGGCCAACGGGGTG-3′), containing a SalI site (bold), and the reverse primer G30 (5′-TTTTTTTTGGATCCGCAGCCAAAACCACCGCCG-3′) containing a BamHI site (bold). The PCR fragment was cloned into the corresponding sites of pSUP202 (Table 1), resulting in pSUPrsmYup. The rsmY downstream sequence together with the rho-independent terminator sequence of rsmY (nucleotide +94 to +589 with regard to the transcriptional start site of the rsmY transcript) was amplified by PCR using chromosomal DNA of PAO1 together with the forward primer J30 (5′-TTTTTTTTCTGCAGCTGCAAAACCCCGCCC-3′), containing a PstI site (bold), and the reverse primer K30 (5′-TTTTTTTTAAGCTTCCTGCCGCCGCGCCAGCC-3′) containing a HindIII site (bold). This fragment was then cloned into the corresponding sites of pUC19, resulting in pUCrsmYdwn. The gentamicin cassette (aacC1) was amplified by PCR using pUCP24 (Table 1) as a template together with the forward primer H30 (5′-TTTTTTTTGGATCCGAATTGACAT AAGCCTGTTCGG-3′), containing a BamHI site (bold), and the reverse primer I30 (5′-TTTTTTTTCTGCAGGAATTGGC CGCGGCGTTGTGAC-3′) containing a PstI site (bold). The aacC1 cassette was cloned into the corresponding sites of pUCrsmYdwn, resulting in pGMrsmYdwn. Then, plasmid pGMrsmYdwn was cleaved with BamHI and HindIII, and the resulting fragment containing the aacC1 cassette and the rsmY downstream region was transferred into the corresponding site of plasmid pSUPrsmYup, resulting in pSUPΔrsmY, bearing mobilization genes and the inactivated rsmY gene. This plasmid was transformed into the E. coli strain S17-1 and transferred by conjugation to PAO1. The following homologous recombination resulted in the chromosomal deletion of rsmY in PAO1. To ensure that the double cross-over had occurred we scored for gentamicin resistance and a carbenicillin-sensitive phenotype of the PAO1ΔrsmY mutant.
The rsmY deletion was verified by Southern blot analysis. Chromosomal DNAs from the putative PAO1ΔrsmY mutants was digested with EcoRI, the fragments were separated on an agarose gel, and blotted onto nitrocellulose. The blot was probed with a biotinylated PCR fragment of the rsmY downstream region obtained with primer pair J30/K30 (see above). The fragment was biotinylated using the Neblot Phototope Kit ® (New England Biolabs). The hybridization fragments were visualized using the Phototope-Star ® Chemiluminescent Detection Kit (New England Biolabs). The deletion of the rsmY gene and insertion of the aacC1 cassette resulted in a signal corresponding to a 1419 bp fragment, whereas wild-type DNA resulted in a signal corresponding to a 707 bp fragment. In addition, a 998 bp fragment was detected in either genome, which results from another EcoRI restriction site downstream of rsmY (see Supplementary material, Fig. S1A).
We are grateful to Drs J.S. Suh, M. Garber and D. Haas for providing strains and plasmids, and the University of Iowa DNA Facility for array processing. This work was supported within the framework of the Special Research Program on ‘Modulators of RNA Fate and Function’ by Grant F1715 from the Austrian Science Fund to U.B.