Two-component systems required for virulence in Pseudomonas aeruginosa

1Pseudomonas aeruginosa has a remarkably diverse ability to thrive in many different environments both outside and within a host. To be successful in these diverse situations, P. aeruginosa needs to sense its environment, decide upon an appropriate response and modify its behaviour accordingly to better suit prevailing conditions. Regulatory networks are key to this decision-making process. Pseudomonas aeruginosa has a large genome (6.3 Mb for the reference PAO1 strain), reflecting the diverse range of environments and hosts that it can inhabit, and just under 10% of its genes are dedicated to these regulatory networks (Stover et al.). Two-component systems (TCSs) comprising sensor kinases (SKs) and response regulators (RRs) (Stock, Robinson and Goudreau ) play a major role in these regulatory networks with P. aeruginosa having 64 SKs, 72 RRs and 3 Hpt proteins (Rodrigue et al.; Stover et al.).

2As an opportunist pathogen, being capable of both acute and chronic infection, P. aeruginosa has a multitude of virulence factors and antibiotic resistance determinants (Driscoll, Brody and Kollef; Gooderham and Hancock; Coggan and Wolfgang ). Well over 50% of the TCSs of P. aeruginosa have been linked to virulence, controlling either virulence-related behaviour or contributing towards in vivo fitness and colonisation ability. This number has grown considerably in recent years, primarily due to the successful application of whole-genome-based methodologies for identifying genes involved in virulence, such as Tn-Seq approaches using animal infection models, and the study of pathoadaptive mutations in isolates from cystic fibrosis (CF) patients (Table 1 major).

3TCSs are generally considered to work alone, sensing either a single stimulus or a narrow range of stimuli to control appropriate responses, being insulated from significant crosstalk (Laub and Goulian; Capra et al.), with relatively few exceptions (Willett and Crosson ). However, a recently emerging theme, in which tremendous progress has been made in the last few years, is the discovery that multikinase networks play leading roles in orchestrating the virulence of P. aeruginosa. Multikinase networks comprise multiple SKs that collaborate to form sophisticated networks capable of sensing and integrating multiple stimuli. In the following sections, we explore how these networks regulate virulence.

4The GacS network plays a leading role in governing the transition between acute and chronic modes of infection. It has emerged as a prime example of a multikinase network, where multiple SKs work together to detect and integrate several different signals to reach a balanced decision. The central kinase in this network, GacS, controls the phosphorylation of the RR, GacA (Fig. 1). Phosphorylated GacA activates the transcription of two non-coding RNAs, RsmY and RsmZ, and they bind and sequester the translational regulators, RsmA (Brencic et al.) and the more recently discovered RsmN (Morris Elizabeth et al.). Free RsmA and RsmN bind to certain mRNAs, promoting the degradation of transcripts involved in chronic virulence (e.g. relating to biofilm formation, T6SS and extracellular products such as pyocyanin and cyanide) while favouring those involved in acute infection (e.g. relating to T3SS and motility) (; Parkins, Ceri and Storey; Pessi et al.; Valverde et al.; Heurlier et al.; Burrowes et al.;; Brencic and Lory; Moscoso et al.; Morris Elizabeth et al.). In short, when GacS signalling is active, GacA will be phosphorylated and this will favour the chronic mode of infection.

5GacS is an unorthodox kinase (containing HisKA, HATPase, REC and Hpt domains) whose signalling activity is controlled through kinase–kinase interactions by three hybrid SKs: RetS, LadS and PA1611. RetS and LadS interact with GacS, with RetS inhibiting, and LadS activating, GacS signalling (Goodman et al.; Laskowski, Osborn and Kazmierczak; Laskowski and Kazmierczak; Ventre et al.). RetS downregulates GacS signalling by binding to GacS and reducing its ability to autophosphorylate (Goodman et al.), whereas LadS upregulates GacS signalling through a phosphorelay mechanism where phosphoryl groups are transferred from the REC domain of LadS to the Hpt domain of GacS (Chambonnier et al.). Unlike RetS and LadS, PA1611 does not interact with GacS; instead, PA1611 binds to RetS, which prevents it from inhibiting GacS (Kong et al.; Bhagirath et al.). The interaction of the four SKs allows for the integration of signals to modulate GacS phosphorylation levels and therefore, the output of the pathway. The signals that activate the various SKs are largely unidentified. However, GacS and RetS are controlled by molecules produced at high cell density and during the lysis of kin cells, respectively, although the identity of these molecules remains elusive (Heeb, Blumer and Haas; LeRoux et al.). Recently, it has been shown that LadS from P. aeruginosa is activated by calcium ions to upregulate chronic phenotypes (Broder, Jaeger and Jenal ).

6The importance of the GacS network has been demonstrated using infection models, with Tn-Seq studies finding that most components of the network are required in either acute and/or chronic virulence in mice (Turner et al.). Moreover, isolates from CF patients often have pathoadaptive mutations within GacS network components, indicating that fine-tuning the signalling of the network can facilitate long-term colonisation and bacterial survival (Cramer et al.; Marvig et al.). Interestingly, strain PA14, which was originally isolated from a burn wound, has a frameshift mutation in ladS. Relative to many other strains, PA14 shows enhanced acute virulence, which can, in part, be attributed to the mutation in ladS (Mikkelsen, McMullan and Filloux ). Another clinical isolate, CHA, has a deletion in gacS and exhibits enhanced acute virulence phenotypes (Sall et al.). These studies show the importance of this network in infection and how environmental pressures can reshape the virulence of P. aeruginosa by mutationally fine-tuning this network.

7Two of the SKs that form part of the core of the GacS network, RetS and PA1611 (described above), also interact with HptB and together form the HptB branch of the GacS network along with two further hybrid SKs, SagS and ErcS’ (Lin et al.; Hsu et al.). HptB is a histidine phosphotransfer protein (Hpt) that serves in a phosphorelay connecting RetS, PA1611, SagS and ErcS’ with an unusual output RR, HsbR (PA3346). HsbR has an N-terminal REC domain, a protein phosphatase 2C (PP2C)-like domain and a C-terminal ser/thr kinase domain (Hsu et al.; Bhuwan et al.). When phosphorylated, HsbR acts as a phosphatase to dephosphorylate the anti-anti sigma factor, HsbA (PA3347). Dephosphorylated HsbA (red arrow, Fig. 1) then sequesters the anti-sigma factor FlgM, which is otherwise found in a complex with the sigma factor, FliA. Free FliA promotes expression of the flagellar genes and therefore both swimming and swarming motility (Bhuwan et al.).

8When HptB is inactive (i.e. not phosphorylated or absent), the receiver domain of HsbR dephosphorylates, which causes the ser/thr kinase domain of HsbR to be more active than its phosphatase domain. Consequently, HsbR phosphorylates HsbA, preventing it from binding and sequestering FlgM. FlgM instead binds FliA and this leads to a decreased expression of the flagellar genes. Furthermore, phosphorylated HsbA (blue arrow, Fig. 1) is thought to bind to, and activate, the diguanylate cyclase HsbD, which leads to an increase in c-di-GMP and RsmY levels (Bordi et al.; Valentini et al.). How exactly HsbD modulates RsmY levels is not known, but it is known that the upregulation of rsmY expression in the ΔhptB mutant depends upon intact GacS/GacA signalling (Bordi et al.; Jean-Pierre, Tremblay and Deziel ).

9SagS is involved in the motile-sessile switch and resistance to antimicrobials (Petrova and Sauer; Petrova et al.), and as well as being one of the SKs that can phosphorylate HptB (Petrova and Sauer ), SagS has a HptB independent signalling route. SagS regulates both RsmY and RsmZ through distinct pathways; its regulation of RsmY is HptB dependent (Bordi et al.; Petrova and Sauer ), while its regulation of RsmZ is HptB independent and involves an interaction with another SK, BfiS. BfiS is required for the transition to irreversible attachment of cells during biofilm formation. The interaction between SagS and BfiS relies upon the conserved phosphorylation sites of these SKs (Petrova and Sauer, ). The cognate RR of BfiS, BfiR, activates expression of CafA (RNase G). CafA reduces the level of RsmZ, which is required for maturation and maintenance of biofilms (Petrova and Sauer ). The SagS/BfiS branch of the network, therefore, regulates the level of RsmZ post-transcriptionally, while the rest of the GacS network regulates both RsmY and RsmZ at the transcriptional level (Ventre et al.; Goodman et al.). RsmY and RsmZ levels can also be influenced by other regulators such as Anr/NarL, which downregulates both sRNAs under conditions of low oxygen, and the β-lactamase regulator, AmpR, which can upregulate RsmZ (O’Callaghan et al.; Balasubramanian, Kumari and Mathee ). It appears that levels of these sRNAs are tightly coordinated by multiple intersecting regulators to orchestrate the transition from acute to chronic virulence and the planktonic to biofilm mode of growth.

10Two major ways that the GacS network is known to affect c-di-GMP levels are, first, that RsmA controls the translation of the sadC mRNA, which encodes the diguanylate cyclase, SadC (Moscoso et al.), and second, the HptB branch of the GacS network regulates the HsbD diguanylate cyclase (Valentini et al.). Intriguingly, in addition to controlling c-di-GMP levels, the GacS network appears to respond to c-di-GMP levels. Overexpressing diguanylate cyclases can induce the T3SS (acute) to T6SS (chronic) switch, and this is dependent upon the regulatory RNAs, RsmY and RsmZ (Moscoso et al.). RsmY and RsmZ levels have also been shown to be elevated in strains overexpressing diguanylate cyclases (Frangipani et al.). It is therefore tempting to speculate that increased c-di-GMP levels activate signalling within the GacS network to help promote biofilm formation and the chronic mode of virulence. In line with this, it has recently been shown that the PilZ domain protein, HapZ, can bind to SagS and inhibit phosphotransfer to HptB, in a c-di-GMP-dependent manner (Xu et al.). In addition, it is possible that c-di-GMP affects signalling elsewhere in the network in yet to be determined ways.

11In summary, the GacS network is a complex multikinase network that plays a major role in deciding between acute and chronic modes of virulence, and between planktonic and biofilm modes of growth. The complexity of the network and the large number of different sensors is likely to reflect the importance of making the correct decision to the survival of the bacterium, and the need to evaluate numerous signals (e.g. Ca2+, kin-cell lysis, c-di-GMP plus several other as yet unidentified signals) in order to inform this decision.

12Surface adhesins, known as Cup fimbriae (chaperone/usher pili), are required for the initial attachment stage of biofilm formation. Pseudomonas aeruginosa has three different sets of archetypal Cup fimbriae genes in its core genome (cupA, cupB and cupC). The PA14 strain has an extra set of fimbriae genes, cupD, within the PAPI-I pathogenicity island. The cupB and cupC genes are controlled by the Roc network, while the cupD genes of PA14 are regulated by the Rcs/Pvr network (Kulasekara et al.; Rao et al.; Mikkelsen et al., ). In addition to regulating the CupB and CupC fimbriae, the Roc network also controls expression of the MexAB-OprM drug efflux pump (Sivaneson et al.).

13Like the GacS network, the Roc network is another good example of a multikinase network, and again c-di-GMP signalling is involved, but unlike the GacS network, which is built from kinase–kinase interactions, the Roc network is instead based upon SKs sharing the same RRs (Fig. 2A). This network comprises two SKs—RocS1 and RocS2, which are both unorthodox (having HisKA, HATPase, REC and Hpt domains)—that control at least three RRs: RocA1 (helix-turn-helix DNA binding output domain), RocR (EAL, c-di-GMP degrading, phosphodiesterase output domain) and RocA2 (helix-turn-helix DNA-binding output domain). Each of the two SKs is capable of interacting with each of the RRs. The RRs target different genes: RocA1 activates expression of the CupC fimbriae, RocA2 inhibits expression of the MexAB-OprM drug efflux pump, and RocR by reducing c-di-GMP levels reduces expression of both cupB and cupC fimbriae genes. There is good reason to believe that an additional RR is involved in this network as the two SKs, RocS1 and RocS2, promote expression of CupB fimbriae genes in a manner independent of any of the three known RRs (Kulasekara et al.; Rao et al.; Sivaneson et al.). Although the controlling stimuli are unknown for the Roc network, the cross-regulation within this network should allow multiple inputs to be evaluated and for these signals to be integrated.

14Roc network signalling promotes adhesion and therefore biofilm formation, while reducing expression of the MexAB-OprM antibiotic efflux pump. Initially, this seems counterintuitive, as biofilms are usually associated with increased antibiotic resistance. However, reduced expression of mexAB-oprM is seen in mature biofilms, and strains isolated from CF patients often show inactivation of this efflux pump despite having a high propensity for biofilm formation (De Kievit et al.; Vettoretti et al.). This suggests that the MexAB-OprB drug efflux pump is not involved in the antibiotic resistance of biofilms.

15The cupD cluster, found in strain PA14, is regulated by an orthologous system to the Roc network consisting of two SKs, RcsC (unorthodox) and PvrS (hybrid), and two RRs, RcsB and PvrR (Fig. 2B). Like the Roc system, RcsB has a HTH DNA-binding domain, while PvrR has an EAL output domain. Interestingly, in this system, PvrS appears to act as a kinase, while RcsC functions primarily as a phosphatase and also acts in an intermolecular phosphorelay connecting PvrS with the output RRs. In this phosphorelay, phosphoryl groups are passed from the REC domain of the hybrid SK, PvrS, to the Hpt domain of RcsC and from there onto the REC domains of the output RRs. This kinase–kinase phosphorelay mode of interaction is reminiscent of the GacS/LadS interaction in the GacS network and is likely to represent a conserved signalling route where the Hpt domain of an unorthodox kinase is used to connect hybrid kinases (that lack Hpt domains) with their output RRs (Mikkelsen et al.,; Chambonnier et al.).

16During infection, Pseudomonas aeruginosa needs to evade host defences such as cationic antimicrobial peptides, and to resist any antibiotic treatments that the patient may be receiving. One major way that this can be achieved is by inducing the aminoarabinose modification of the lipid A component of the lipopolysaccharide layer. This modification reduces the negative charge on the LPS, thereby limiting its electrostatic interaction with, and the subsequent uptake of, cationic antimicrobial peptides and cationic lipopeptide antibiotics (including polymyxins such as colistin, which are often used as last-resort antibiotics in CF patients). The genes required for the modification are encoded by the arnBCADTEF operon and it is regulated by a sensory network comprising at least five distinct TCSs each comprising a SK and a RR: PhoQP, PmrBA, ColSR, CprSR and ParSA (Macfarlane et al.; Macfarlane, Kwasnicka and Hancock; McPhee, Lewenza and Hancock; Moskowitz, Ernst and Miller; Gooderham and Hancock; Gooderham et al.; Fernández et al.,; Gutu et al.; Lee and Ko ).

17Unlike the GacS and Roc networks, there is no documented linkage at the phosphosignalling level between these TCSs, instead the output RRs of the separate TCSs converge upon the aminoarabinose modification genes (Fig. 3), as a common feature of each RR's unique wider regulon. The SKs, PhoQ and PmrB, are active when the Mg2+ concentration is low (McPhee et al.), while the SKs, CprS and ParS, are activated by different cationic antimicrobial peptides (Fernández et al.,; Muller, Plésiat and Jeannot ), and ColS is activated by Zn2+ (Nowicki et al.). Extracellular DNA is a significant component of the biofilm matrix and is often found at infection sites, and it appears to play an important physiological role in the PhoQP and PmrBA responses, as it sequesters cations and can reduce Mg2+ levels to the extent that PhoQ and PmrB signalling are activated, thereby promoting LPS modification and increasing resistance to host cationic peptides and polymyxins (; Gellatly et al.; Lewenza ).

18This regulatory network undergoes strong selective pressures in CF patients and adaptive mutations are frequently identified in isolates from CF patients, particularly those who have been treated with polymyxins. These mutations can be in any of the TCSs of this network although mutations affecting PhoQP and PmrBA are particularly common; generally, they lead to either increased or constitutive expression of the genes for the aminoarabinose modification, and are frequently accompanied by other mutations in non-TCS genes (such as those for LPS biogenesis and outer membrane protein assembly) that further boost resistance levels (Barrow and Kwon; Fernández et al.; Miller et al.; Gellatly et al.; Moskowitz et al.; Gutu et al.; Jochumsen et al.).

19One way that Pseudomonas aeruginosa responds to growth on surfaces is by activating the Wsp chemosensory system. This pathway controls the production of the secondary messenger, c-di-GMP, which promotes biofilm formation and decreases expression of the flagellar genes. Like the Chp chemosensory system (below), the Wsp chemosensory system forms a signal transduction system (Fig. 4) resembling the bacterial chemotaxis system (He and Bauer ). The Wsp pathway incorporates the cytoplasmic SK, WspE, which phosphorylates two RRs, the methylesterase, WspF, and the diguanylate cyclase, WspR (Bantinaki et al.). Surface growth is sensed by the membrane-bound WspA protein (a methyl-accepting-chemotaxis protein homologue), possibly via mechanical sensing of physical pressure resulting from surface association and cell–cell contact (O’Connor et al.). Contact sensing by WspA triggers autophosphorylation of WspE, which in turn phosphorylates and activates WspR and WspF. WspR-P catalyses the production of c-di-GMP through its GGDEF domain (Bantinaki et al.; De et al., ). When WspR is dephosphorylated, it is delocalised within the cytoplasm, but when phosphorylated, it aggregates to form cytoplasmic clusters (Guvener and Harwood ), where its diguanylate cyclase activity is increased (Huangyutitham et al.). WspF-P acts to reset the system by removing methyl groups from WspA (Hickman, Tifrea and Harwood; Bantinaki et al.). Deletion of wspF results in constitutive activation of WspR (WspR-P) due to overmethylation of WspA and produces a distinctive wrinkled, small colony phenotype with enhanced biofilm formation (Hickman, Tifrea and Harwood ).

20Activation of the Wsp pathway by surface sensing triggers an increase in c-di-GMP levels (Hickman, Tifrea and Harwood; O’Connor et al.; Ha and O’Toole ). The transcriptional regulator, FleQ, is the major target for the c-di-GMP produced by the Wsp pathway. FleQ promotes expression of the flagellar genes and downregulates biofilm-associated genes (e.g. pel encoding exopolysaccharide biosynthesis proteins). FleQ is inhibited by binding c-di-GMP, and therefore Wsp pathway activation leads to reduced expression of the flagellar genes and increased expression of biofilm-associated genes (Hickman, Tifrea and Harwood; Hickman and Harwood ).

21Consistent with its role in promoting biofilm formation, Tn-Seq data has shown that the Wsp pathway is required for chronic wound infections in mice (Turner et al.). Moreover, isolates from CF patients often show pathoadaptive mutations in the Wsp pathway (Marvig et al.); wspF mutations being particularly common with their distinctive phenotype of having a rugose appearance and enhanced biofilm formation (D’Argenio et al.; Hickman, Tifrea and Harwood; Smith et al.; Starkey et al.; Sousa and Pereira; Blanka et al.). This indicates that the Wsp pathway is under selective pressures to affect its signalling output during long-term infection, with constitutive activation being favourable for biofilm growth and chronic infection.

22The Wsp pathway and the Chp/FimS/AlgR network are distinct but have many similarities; both sense surface contact, both involve a chemosensory pathway, both use secondary messenger signalling and, like many other signalling networks, both contribute to biofilm formation. In that sense they can be considered to form a super network (O’Toole and Wong ). The Chp/FimS/AlgR network is itself an example of a multikinase network. It regulates production of two different secondary messengers, cAMP and c-di-GMP, to control virulence and biofilm formation (Fig. 5). The production and activity of type 4 pili (T4P) are also controlled by this network and, moreover, they play a central signalling role. T4P are major surface adhesins allowing adherence and invasion of host tissues (Hahn ). They are located at the cell poles and undergo repeated cycles of extension, adhesion and retraction to pull the cell forward in a process called twitching motility (Skerker and Berg; Mattick ). The extension and retraction of these pili are controlled by the Chp chemosensory pathway part of the Chp/FimS/AlgR network, which also controls levels of the secondary messenger, cyclic AMP (cAMP) (Darzins; Whitchurch et al.; Fulcher et al.). cAMP regulates many other cellular processes and genes, primarily via the transcription factor Vfr (virulence factor regulator) which upregulates many virulence genes, including those involved with quorum sensing, type 2 secretion, T3SS, the FimS/AlgR TCS and the T4P themselves (Albus et al.; Wolfgang et al.; Kanack et al.; Bertrand, West and Engel; Fulcher et al.).

23The Chp chemosensory pathway resembles, but is distinct from, the chemotaxis pathway regulating flagellar rotation. It uses a methyl-accepting-chemotaxis-protein (MCP) homologue, PilJ, to detect surface contact and chemoattractants such as phosphatidylethanolamine (Kearns, Robinson and Shimkets; Jansari et al.). Sensing of surface contact involves mechanosensing, where PilJ is thought to respond to tension generated within the pili, when the cell retracts pili that have adhered to surfaces (Persat et al.). The signal from PilJ is relayed via two adaptor proteins, PilI and ChpC, to an unorthodox SK, ChpA. ChpA is one of the most complex SKs found in any bacterial species, having nine potential phosphorylation sites; it has eight ‘Xpt’ domains, six of which are conventional Hpt domains and two that contain either serine or threonine in place of the usual phosphorylatable histidine, plus a receiver domain (ChpArec) (Whitchurch et al.; Leech and Mattick ). ChpA autophosphorylates on Hpt domains 4–6 and phosphotransfer occurs from Hpts 5 and 6 to ChpArec, but also, at a slower rate, to two standalone RRs: PilG and PilH. Reversible phosphotransfer can occur from ChpArec to Hpt 2–6; however, as yet, no phosphorylation has been observed on Hpt 1 or the remaining two ‘Xpt’ domains (Silversmith et al.). Hpt 2 and Hpt 3 serve as the main phosphodonors to the two output RRs, PilG and PilH (Hpt5 and 6 also contribute but at a much slower rate), that control the adenylate cyclase, CyaB (Wolfgang et al.; Fulcher et al.; Silversmith et al.), and the pilus extension (PilB) and retraction (PilT/U) ATPases (Bertrand, West and Engel ).

24The RR, PilG, localises to the cell poles along with the pili forming a complex with FimL and FimV; presumably this helps to keep its local concentration high, proximal to its kinase, ChpA (Inclan et al.). The details of how PilG and PilH regulate adenylate cyclase and the pilus ATPases are not known, although models have been proposed based on genetic studies, where PilG stimulates pilus extension (via PilB) and CyaB activity, as the ΔpilG mutant has reduced piliation and reduced cAMP levels, while PilH stimulates pilus retraction (via PilT/U) and inhibits CyaB activity, as the ΔpilF mutant has increased piliation and increased cAMP (Bertrand, West and Engel; Fulcher et al.). The role of PilH is controversial though, and instead it might function as a phosphate sink for PilG rather than directly regulating CyaB and PilT/U.

25The Chp chemosensory pathway associates with the FimS/AlgR TCS (also known as AlgZ/R) to form the Chp/FimS/AlgR multikinase network. This network is constructed differently from the other examples of multikinase network discussed; here, the two SKs, FimS and ChpA, do not interact directly but instead they interact with a common partner, the MCP homologue, PilJ. FimS is thought to be activated by surface contact, and an attractive model would be for the surface contact sensor, PilJ, to control FimS activity via their interaction (Luo et al.). The FimS/AlgR TCS is best known for its role in controlling the production of the exopolysaccharide, alginate, but it is also required for twitching motility as it regulates expression of the T4P, and is involved in multiple other pathways including hydrogen cyanide and rhamnolipid production, T3SS, the Rhl quorum-sensing system and biofilm formation (Whitchurch, Alm and Mattick; Okkotsu, Little and Schurr ).

26The role of cAMP as the initial secondary messenger in the Chp/FimS/AlgR network is well known, with the Chp chemosensory system producing cAMP in response to surface contact, which activates Vfr, leading to activation of the expression of many virulence genes including the FimS/AlgR TCS. However, recently, c-di-GMP has been implicated as a delayed secondary messenger from this network (Fig. 5) i.e. following activation by surface contact, cAMP is produced initially and then several hours later c-di-GMP is produced, correlating with the onset of biofilm formation (O’Toole and Wong ). Two diguanylate cyclases are involved, SadC (which is also controlled by the GacS network) and MucR, with one of the targets for the c-di-GMP that they produce being the c-di-GMP binding protein, Alg44, which stimulates alginate production (Hay, Remminghorst and Rehm; Schmidt et al.). MucR expression is stimulated by AlgR when the network senses surface contact (Kong et al.). Regulation of SadC is more complex; AlgR and Vfr together upregulate the fimU-pilVWXXY1Y2E operon, which is necessary for T4P biogenesis and function (Luo et al.). PilY1, encoded by this operon, is a cell-surface-associated protein that promotes the activity of SadC and downregulates swarming motility (Kuchma et al.). Crucially, PilY1 depends upon the T4P for export ensuring an ordered signalling cascade where pili are made first, before PilY1 is deployed and c-di-GMP production initiated (Luo et al.).

27TCSs play a major role in controlling Pseudomonas aeruginosa virulence, with over 50% of its TCSs implicated in controlling either virulence or virulence-related behaviours such as biofilm formation and antibiotic resistance (Table 1). A theme highlighted by the above examples is that during infection, P. aeruginosa makes extensive use of multikinase networks to detect and integrate multiple environmental signals, and to reach a balanced decision about the most appropriate response. There are a multitude of different architectures for these multikinase networks: Kinase–kinase interaction. Seen in the GacS network (Fig. 1) and the Rcs/Pvr network (Fig. 2B).Multiple SKs can share the same RR(s), as in the Roc network (Fig. 2A) and in the HptB branch of the GacS network (Fig. 1).Connector proteins can link the SKs e.g. in the Chp/FimS/AlgR network, the surface contact sensing MCP homologue, PilJ, interacts with two SKs, ChpA and FimS (Fig. 5).Regulatory convergence between TCSs, where otherwise separate TCSs control the expression of the same genes, as seen in the network controlling LPS modification (Fig. 3).Transcriptional control of one TCS by another TCS e.g. in the Chp/FimS/AlgR system, the expression of the FimS/AlgR TCS is induced by Vfr, which is activated by binding cAMP that is produced by CyaB due to signalling by the ChpA SK (Fig. 5).

28Kinase–kinase interaction. Seen in the GacS network (Fig. 1) and the Rcs/Pvr network (Fig. 2B).

29Multiple SKs can share the same RR(s), as in the Roc network (Fig. 2A) and in the HptB branch of the GacS network (Fig. 1).

30Connector proteins can link the SKs e.g. in the Chp/FimS/AlgR network, the surface contact sensing MCP homologue, PilJ, interacts with two SKs, ChpA and FimS (Fig. 5).

31Regulatory convergence between TCSs, where otherwise separate TCSs control the expression of the same genes, as seen in the network controlling LPS modification (Fig. 3).

32Transcriptional control of one TCS by another TCS e.g. in the Chp/FimS/AlgR system, the expression of the FimS/AlgR TCS is induced by Vfr, which is activated by binding cAMP that is produced by CyaB due to signalling by the ChpA SK (Fig. 5).

33A further finding is that these regulatory networks undergo considerable selective pressure within hosts, particularly during chronic infection and it is common to isolate mutant strains with pathoadaptive mutations in these networks e.g. showing enhanced biofilm formation, increased antibiotic resistance or reduced motility (Marvig et al.,; Jochumsen et al.; Winstanley, O’Brien and Brockhurst ). This shows that while the wild-type regulatory networks may be capable of efficiently orchestrating virulence across a broad range of conditions, there are circumstances where the networks can be genetically fine-tuned to optimise behaviour to better suit the prevailing conditions e.g. chronic infection in the CF lung, although this often comes at expense of the bacterium's ability to thrive in other conditions e.g. at causing acute infections (Smith et al.; Jeukens et al.).

34Another key theme illustrated by the above examples is the interplay between multikinase networks and secondary messenger systems, with several of the networks discussed modulating levels of c-di-GMP. This provides another level of signal integration and decision making as all of the signals from several, otherwise separate, networks can feed into these secondary messengers to control common outputs important for virulence such as biofilm formation and motility.

35Key priorities for the future advancement of our understanding of these multikinase networks that could facilitate the development of new ways of targeting these networks and tackling infection are as follows: The ligands controlling many of the TCSs discussed above remain unknown, and although some recent progress has been made in this area (e.g. Broder, Jaeger and Jenal ) we urgently need systematic high-throughput methods for ligand identification.Determining which kinases work together in multikinase networks is a key priority. It is likely that many of the SKs in Table 1 will feature in yet to be discovered multikinase networks. A combination of biochemical, bioinformatic and genetic methods need to be employed for systematic screening for potential interactions.Revealing the complex interfaces with other regulatory mechanisms i.e. secondary messenger signalling and one-component regulators, which frequently form integral parts of multikinase networks.Understanding how multikinase networks process and integrate signals to make decisions. This will require a concerted effort employing mathematical modelling alongside a detailed biochemical understanding of the regulators involved, how they respond to signal, and their interactions and expression patterns.

36The ligands controlling many of the TCSs discussed above remain unknown, and although some recent progress has been made in this area (e.g. Broder, Jaeger and Jenal ) we urgently need systematic high-throughput methods for ligand identification.

37Determining which kinases work together in multikinase networks is a key priority. It is likely that many of the SKs in Table 1 will feature in yet to be discovered multikinase networks. A combination of biochemical, bioinformatic and genetic methods need to be employed for systematic screening for potential interactions.

38Revealing the complex interfaces with other regulatory mechanisms i.e. secondary messenger signalling and one-component regulators, which frequently form integral parts of multikinase networks.

39Understanding how multikinase networks process and integrate signals to make decisions. This will require a concerted effort employing mathematical modelling alongside a detailed biochemical understanding of the regulators involved, how they respond to signal, and their interactions and expression patterns.