Several strains of the human opportunistic pathogen Pseudomonas aeruginosa infect plants, nematodes and insects. Our laboratory has developed a multihost pathogenesis system based on the P. aeruginosa clinical isolate PA14, in which non-mammalian hosts are used to screen directly for virulence-attenuated mutants. The majority of PA14 mutants isolated using non-mammalian hosts also displayed reduced virulence in a burned mouse model. Surprisingly, only a few host-specific virulence factors were identified, and many of the P. aeruginosa mutants were attenuated in virulence in all the hosts. These studies illustrate the extensive conservation in the virulence mechanisms used by P. aeruginosa to infect evolutionarily diverged hosts, and validate the multihost method of screening for virulence factors relevant to mammalian pathogenesis. Through the use of genetically tractable hosts, the multihost pathogenesis model also provides tools for elucidating host responses and dissecting the fundamental molecular interactions that underlie bacterial pathogenesis.
The challenge of controlling bacterial pathogens has driven the development of increasingly sophisticated techniques that broaden our understanding of virulence mechanisms. Methods of identifying bacterial virulence factors have advanced from earlier microbe-centred approaches that involved screening for virulence factors based on altered biochemical phenotypes to those that take into account the critical role of the host in the disease process (reviewed by Strauss and Falkow, 1997). In this review, we focus on one such approach, that of using non-mammalian hosts to model bacterial pathogenesis, highlighting specific features of host–pathogen interactions revealed by this method.
Development of a multihost pathogenesis system
That some commonalties exist in the pathogenic mechanisms used by diverse bacteria has been established by many studies (reviewed by Finlay and Falkow, 1997). Further, recent investigations have uncovered many parallels in innate or non-adaptive immunity against pathogens in plant, invertebrate and mammalian hosts (Kopp and Medzhitov, 1999). Taken together, these results suggest that some of the virulence mechanisms of pathogens as well as the host defences against them are likely to have ancient evolutionary origins. A pathogenesis model in which both the pathogen and the host are genetically tractable would greatly facilitate our understanding of some of the universal mechanisms underlying host–pathogen interactions. Rahme et al. (1995) introduced such a system by showing that several strains of the human opportunistic pathogen Pseudomonas aeruginosa cause disease in both plants and animals. Using the clinical isolate UCBPP-PA14 (PA14), they showed that the same subset of virulence factors was required for the full virulence of this strain in both plant and animal hosts. These findings were expanded to show that P. aeruginosa was also a pathogen of nematodes (Mahajan-Miklos et al., 1999; Tan et al., 1999a,b) and insects (Jander et al., 2000; S. Mahajan-Miklos, G. Jander, G. Lau, L. Perkins, L. G. Rahme and F.M. Ausubel, unpublished data), leading to the development of a ‘multihost’ pathogenesis system in which plants, nematodes and insects were used as adjuncts to animal models for the identification and study of bacterial virulence factors. After the generation of libraries of PA14 using random transposon mutagenesis, individual clones were screened for reduced virulence in the various hosts. The relevance of the virulence factors identified using these screens to mammalian pathogenesis was examined using a mouse full-thickness burn model (Stevens et al., 1994). We discuss each host used in these analyses with respect to their advantages, potential caveats and the results that were obtained using them.
P. aeruginosa–plant interactions
The use of Arabidopsis thaliana as a host in the study of plant–pathogen interactions has allowed the dissection of disease response pathways that are activated in response to many bacterial and fungal pathogens (reviewed by Glazebrook, 1999). In addition, systematic screening has been used successfully to identify virulence determinants in plant pathogens (Willis et al., 1990; Rahme et al., 1991). Building on the wealth of information obtained from the analyses of plant pathogens, the A. thaliana model was also exploited for the study of the virulence mechanisms of P. aeruginosa (Rahme et al., 1997).
Most literature reports cite P. aeruginosa as a pathogen that infects injured, burned, immunodeficient or otherwise compromised humans and causes a wide range of infections, including the chronic lung infections in cystic fibrosis patients (Wood, 1976). However, P. aeruginosa is a ubiquitously distributed bacterium, found in association with water, soil and plants, and some strains have been described as plant pathogens (Burkholder, 1950; Cho et al., 1975). Such reports led our laboratory to search for and subsequently identify several strains of P. aeruginosa that were capable of infecting plants as well as mice (Rahme et al., 1995). P. aeruginosa strains caused varying degrees of chlorosis (yellowing) and tissue maceration 4 days after infiltration into A. thaliana leaves. Similar symptoms were observed when the P. aeruginosa strains were infiltrated into the mid-rib of lettuce leaf stems. Several strains, including PA14, a clinical isolate, and UCBPP-PA29, a plant isolate, caused severe soft-rot symptoms in some, but not all, of the A. thaliana ecotypes (wild varieties) tested, a result typical of a well-studied A. thaliana plant pathogen such as Pseudomonas syringae. Further, the degree of proliferation of strains PA29 and PA14 in A. thaliana leaves correlated with the severity of disease symptoms caused. Based on these results, the clinical isolate, strain PA14, was selected for further studies (Rahme et al., 1995). It is important to note that PA14 is not unique among P. aeruginosa strains in its ability to cause disease in both plants and mice. In addition to the examples cited above, the well-studied P. aeruginosa strain PAO1, which is known to be infectious in a variety of mouse models, also elicits disease symptoms in lettuce (Rahme et al., 1997).
Using the plant leaf infiltration assays, 2500 prototrophic TnphoA PA14 mutants were screened for attenuated pathogenicity, as evidenced by reduced disease symptoms in lettuce and at least a 10-fold reduction in growth in A. thaliana leaves compared with the wild-type PA14, leading to the identification of nine mutants (Table 1). Eight of these mutants showed a significant reduction in pathogenicity in a burned mouse model, demonstrating a surprising amount of functional conservation in the virulence mechanisms used by P. aeruginosa to infect plants and animals (Rahme et al., 1997).
Table 1. P. aeruginosa mutants identified by screening in plants and nematodesa.
b. Six-week-old male AKR/J inbred mice weighing between 20 and 30 g were injected with 5 × 105 bacterial cells, a dose at which 100% of mice die when injected with wild-type PA14. The number of mice that died of sepsis was monitored daily for 7 days.
NT, not tested.
Mutants identified in the plant leaf infiltration assay
Unknown, present in C. tepidum
Periplasmic disulphide bond-forming enzyme
Two overlapping ORFs
ORF2 contains helix–turn–helix motif
Mutants identified in the C. elegans fast-killing assay
Contains histidine kinase motif
Plant bacterial virulence factor
Mutants identified in the C. elegans slow-killing assay
Transcriptional regulator of multidrug transporter
Integral membrane protein
Transcriptional regulator of RpoN-dependent operons
Molecular analysis of the genes interrupted by the transposon insertions in the nine mutants identified using the plant model showed that only two of the mutated genes, gacA, which encodes a two-component response regulator (Reimmann et al., 1997), and dsbA, which encodes a periplasmic disulphide bond-forming enzyme (Bardwell et al., 1991), were previously known to be involved in pathogenesis. The successful isolation of new pathogenic loci, including those encoding proteins of unknown functions, demonstrates the utility of the plant model for the study of bacterial virulence (Table 1). Many intriguing mutants were identified using this screening method in plants, such as pho34B12, which is also attenuated in virulence in nematodes and mice (Table 1). Disruption of the pho34B12 gene affects a number of virulence-related factors, including those responsible for the production of pyocyanin, elastolytic and haemolytic activities (Rahme et al., 1997). Sequence analysis of the DNA flanking this insertion shows the existence of two completely overlapping open reading frames (ORFs) that are transcribed in opposite directions, one of which contains a DNA-binding motif found in the LysR family of regulators (Rahme et al., 1997; H. Cao and L.G. Rahme, unpublished data). Ongoing work is unravelling the complexity of this and other newly identified pathogenic loci with respect to their structure, regulation and function.
P. aeruginosa –Caenorhabditis elegans interactions
C. elegans are killed when fed on lawns of PA14 (Mahajan-Miklos et al., 1999; Tan et al., 1999a). As demonstrated for plants, additional P. aeruginosa strains, such as PA29 and PAO1, are also able to kill C. elegans, showing that PA14 is not unique in its pathogenic potential (Darby et al., 1999; Tan et al., 1999a). Interestingly, the mechanism by which P. aeruginosa kills C. elegans is determined by bacterial growth conditions. In low-salt medium, PA14 accumulates within the lumen of the intestine, killing worms over a 2–3 day time span through an infection-like mechanism (slow killing; Tan et al., 1999a). In contrast, PA14 grown in a rich and high-osmolarity medium kills worms through a toxin-mediated mechanism (fast killing; Mahajan-Miklos et al., 1999). A third mechanism of C. elegans killing has been demonstrated by Darby et al. (1999) using the P. aeruginosa strain PAO1. Worms exposed to PAO1 grown on brain–heart infusion (BHI) medium become paralyzed and die within 4 h, most probably because of the generation of one or more neurotoxins.
In addition to its genetic tractability, the attraction of C. elegans as a model lies in its short life span, ease of manipulation and the wealth of information available regarding its biology, behaviour and genome. Although little is known about how nematodes combat bacterial pathogens, C. elegans and P. aeruginosa inhabit the same natural environment, the soil. It is thus reasonable to hypothesize that these two organisms have developed an arsenal of weapons to combat each other, and that some of these strategies may have been conserved during evolution. Does the inherent simplicity of the C. elegans model limit its application to the complex process of infection in mammals? The results obtained so far indicate the contrary, as many of the mutants obtained using the C. elegans model also exhibited attenuated virulence in mice. At the time of writing, a total of 5500 mutagenized PA14 clones have been screened in the fast- or slow-killing assays, resulting in the identification of mutations in 13 genes that attenuate slow or fast killing. A significant number of these genes (10) were required for full pathogenicity in the burned mouse model (Table 1 and Fig. 1).
By screening 2200 PA14 mutants, mutations in eight genes were identified that resulted in the attenuation of C. elegans slow killing, as judged by both a delay in the killing process and greater numbers of surviving C. elegans progeny when compared with the wild-type PA14 strain. Only three of these mutations were in known virulence-associated genes (Table 1) (Tan et al., 1999b). As slow killing of worms by PA14 is most probably an infectious process, an important future contribution of this model will be the functional analysis of virulence factors at specific stages of interaction with the host, such as colonization, survival and replication, information that is difficult to ascertain using traditional animal models.
Screening 3300 TnphoA mutants in a second C. elegans model (fast killing) resulted in the identification of six mutants that exhibited attenuated virulence, as judged by a delay in the killing process. Analysis of some of these PA14 mutants led to the identification of phenazines, low-molecular-weight, pigmented toxins, as one of the mediators of fast killing (Mahajan-Miklos et al., 1999). In addition to revealing information about the mechanism of fast killing, these results are significant for two reasons. First, many of the phenazine mutants identified using this screen exhibited attenuated virulence in the A. thaliana and mouse models, demonstrating a possible role for these toxins in plant and mammalian pathogenesis. Secondly, phenazines were the only class of known effector or toxin molecules that were recovered using the plant or nematode screens, a point we address later in this review.
As C. elegans fast killing is toxin mediated, one limitation of this model is that it cannot be used for the study of the infectious process. However, the simplicity of the toxin-mediated model facilitates interactive genetics or the ability to combine pathogen and host mutants in order to dissect interactions between bacterial virulence factors and host defences against them. The C. elegans NL130 mutant strain, which contains mutations in two p-glycoprotein genes (pgp-1 and pgp-3;Broeks et al., 1995) was found to be extremely sensitive to fast killing. The function of the Pgps, proteins belonging to the ATP-binding multidrug transporter proteins (Higgins, 1995), was examined by combining the pgp mutant strain NL130 with the different bacterial mutants defective in fast killing. These experiments showed that the pgp mutant worms were sensitive to phenazines, as the sensitivity of mutant worms to fast killing was suppressed when combined with bacterial mutants that were defective in phenazine production, but not by bacterial mutants that generated wild-type levels of phenazines. This genetic analysis uncovered a role for the Pgps in host defence against P. aeruginosa, specifically that of extruding phenazines (Mahajan-Miklos et al., 1999).
In addition to using the wealth of existing C. elegans mutants, forward genetic screens could be used to identify host factors involved in interactions with pathogens. Such a screen has been conducted using the C. elegans–PAO1 pathogenesis model (Darby et al., 1999). PAO1 causes a rapid paralysis of C. elegans as a result of the production of one or more toxins that are under the control of the LasR and RhlR quorum-sensing systems. Screening for C. elegans mutants that were resistant to paralysis resulted in the identification of two mutants, both of which contained mutations in the egl-9 gene. Previously isolated alleles of egl-9, so named because they have an egg-laying defect, also conferred resistance to PAO1. Interestingly, the egl-9 mutants were not resistant to either fast or slow killing of C. elegans by PA14 (M.-W. Tan and F.M. Ausubel, unpublished data). Cloning and molecular characterization of the egl-9 gene predicted an 80 kDa protein that is homologous to a rat protein (SM-20), but does not contain homologies to other proteins of known function. Based on genetic and expression studies, the authors proposed two models for the role of the EGL-9 protein. First, EGL-9 could be part of a pathway that is activated by the PAO1 toxin, resulting in muscular contraction and paralysis. Secondly, the EGL-9 protein could be required for the transport or function of the PAO1 toxin. Although further characterization is required to reveal the function of EGL-9, this work demonstrates how the C. elegans model can be used effectively for the study of host defences and targets of bacterial toxins.
Studies such as these are just beginning to probe the C. elegans defence system, which remains largely uncharacterized. Although homologues of some of the components of the innate immune response pathway that are conserved in insects, plants and mice have been found in the C. elegans genome, their role in defence pathways is still unclear (Tan and Ausubel, 2000). Screens resulting in the identification of C. elegans mutants altered in their ability to defend against P. aeruginosa will provide a critical piece in the puzzle of nematode immunity and help to address the issue of whether similarities exist between this system and that of other organisms.
P. aeruginosa–insect interactions
In contrast to the little understood immune mechanisms in nematodes, there is an extensive body of literature regarding the mechanism by which insects defend themselves against pathogens. Although lacking in an adaptive B- and T-cell-based immune system, the innate defence system of insects parallels that of mammals. The similarities include the signalling pathways involving Toll receptor-like proteins and NF-κB transcription factors, and the humoral responses leading to the induction of antimicrobial peptides such as defensins (reviewed by Hoffmann et al., 1999). In addition, insects possess a circulatory system and have complex cellular responses, such as the phagocytosis of microbial invaders by the haemocytes (Mullett et al., 1993).
Prompted by these studies, the greater wax moth Galleria mellonella was developed as a model system to study P. aeruginosa virulence. One of the principal advantages of using this insect system is that the large size of G. mellonella larvae (250 mg) permits the injection of specific doses of bacteria and, thus, the calculation of an LD50, which is not possible using either the plant or the nematode models. Injection of mutant bacterial strains showed that there is a significant correlation between an increased LD50 in G. mellonella larvae and reduced lethality in mice, an indication that G. mellonella may be an excellent model for the identification and study of bacterial virulence factors relevant to mammalian pathogenesis (Jander et al., 2000).
Despite the many advantages of using G. mellonella for identifying bacterial factors, the inability to manipulate this organism genetically hinders the dissection of host responses. This obstacle could be overcome with the use of Drosophila melanogaster, which has served as a model for the study of insect immunity (Hoffmann et al., 1999). We have shown that P. aeruginosa effectively kills D. melanogaster adults (S. Mahajan-Miklos, L. Perkins and F. M. Ausubel, unpublished results) and also that many of the bacterial mutants identified using the plant and nematode models are also attenuated in virulence in D. melanogaster (G. Lau, L. Perkins and L. G. Rahme, unpublished results).
Applications of the alternative models to bacterial pathogenesis in mammals
Results obtained using the plant and nematode screens support the hypothesis that disparate eukaryotic hosts can be used to model P. aeruginosa pathogenesis in mammals. Thus far, the screens have led to the identification of 21 virulence-attenuated PA14 genes, of which 17 are also required for full pathogenicity in a burned mouse model (Fig. 1). It is important to note two points in this context. First, until now, only 8000 TnphoA mutants have been screened in either nematode or plant models, representing a mere 25–33% of the total number that should be screened to ensure a 95% probability of testing each P. aeruginosa gene in each of the assays. This suggests that the multihost screening approach could be expanded for the discovery of many additional factors involved in mammalian pathogenesis. Secondly, the relevance of these studies to mammalian pathogenesis has been determined solely on the basis of attenuation in a burned mouse model. However, P. aeruginosa causes a wide range of infections in humans, some of which have been replicated in animal models, including a chronic infection model in the lungs of mice containing mutations in the cystic fibrosis transmembrane conductor regulator (CFTR) gene (Heeckeren et al., 1997). Mutants obtained using multihost screening methods should also be tested in additional mouse models, and this could lead to the identification of subsets of virulence factors critical for the establishment of specific infections. For example, it is reasonable to hypothesize that mutants identified using the hyperosmolar and low-pH conditions of the C. elegans fast-killing assay might be important for virulence in the cystic fibrosis lung infection model.
Is multihost pathogenesis unique to P. aeruginosa
The requirement for specific bacterial–host interactions during infection limits the host range of many bacterial pathogens to a single or small number of species. Our studies showed that P. aeruginosa strain PA14 does not conform to these limitations. Although the multihost screens did result in the identification of some virulence factors required specifically in just one host, the majority of virulence mechanisms uncovered through this method showed a striking degree of conservation across the different hosts. Perhaps the successful identification of virulence factors relevant to mammalian pathogenesis reflects specific parameters used in the multihost screening methods. For example, in the case of the plant model, infiltration of bacteria directly into the plant intercellular spaces of leaves may circumvent some of the events required for colonization or other host-specific events necessary for the initiation of the infectious process. By screening for factors required downstream of these events, host-specific virulence factors could be bypassed, allowing the identification of conserved interactions. P. aeruginosa produces a number of proteases, such as elastase, LasA and alkaline protease, that are capable of degrading a broad range of host proteins. Thus, one concern is that the apparent conservation of virulence mechanisms may simply be a consequence of widespread tissue damage by degradative enzymes, which allows the replication and dissemination of bacteria once they are introduced into the host. This is unlikely to be the case for two reasons. First, as described below and shown in Fig. 1, several classes of virulence factors were shown to be required for pathogenicity in more than one host. Secondly, most of the mutants obtained using the plant and nematode screens were tested for levels of protease activity, and the levels were only found to be reduced in a few of the mutants (Rahme et al., 1997; Tan et al., 1999b).
The conservation of virulence mechanisms in diverse hosts raises the question of whether multihost pathogenesis is unique to P. aeruginosa, thereby limiting its utility. This would be in contrast to the widespread applications of a technique such as signature-tagged mutagenesis (STM), which has been widely applied to the study of diverse Gram-negative and Gram-positive bacteria (reviewed by Perry, 1999). Although the ability to infect multiple hosts is unlikely to be a universal trait of bacterial pathogens, it is not limited to P. aeruginosa. Other examples of human pathogens that can kill C. elegans include Gram-negative bacteria such as Serratia marcescens and Burkholderia cepacia (Tan and Ausubel, 2000). G. mellonella is sensitive to the mammalian fungal pathogens Fusarium oxysporum (Jander et al., 2000) and Aspergillus fumigatus, and to the mammalian bacterial pathogens Proteus vulgaris, Proteus mirabilis and Serratia marcescens (Chadwick, 1967). In addition to opportunistic pathogens, it has been demonstrated recently that the highly specialized invasive pathogen Salmonella typhimurium also kills C. elegans through an infectious process involving the proliferation and persistence of S. typhimurium in the C. elegans intestinal lumen (A. Aballay, P. Yorgey and F. M. Ausubel, submitted; C Kurz and J. Ewbank, personal communication).
Even before the discovery of additional pathogens that infect multiple hosts, results obtained from the PA14 multihost pathogenesis system could shed light on the pathogenic mechanisms used by diverse bacteria. Many of the virulence factors identified using the multihost screening method have homologues in other pathogenic bacteria. This group contains genes whose role in pathogenesis remains unclear. One such example is the hrpM gene, which was identified as a virulence factor using the C. elegans fast-killing screen and is also required for full pathogenicity in plants, mice (Mahajan-Miklos et al., 1999) and insects (Jander et al., 2000). This locus encodes an enzyme involved in the synthesis of membrane-derived oligosaccharides (MDOs), which have been found in the periplasm of many Gram-negative bacteria including animal pathogens such as Salmonella and Klebsiella, although little is known about how they function (Kennedy, 1996). Some of the functions of these molecules could be uncovered through the use of interactive genetics and the C. elegans model, similar to the approach described for the role of the phenazines.
Summation of results obtained using the multihost screens and comparisons with other screening methods
Sequence information obtained from the mutants that exhibited attenuated virulence in the burned mouse model showed that many classes of genes were identified using the multihost screens. These data have to be held to the rigorous standards required for any screens involving insertional mutagenesis, that of ruling out the effects of secondary mutations and polar effects on genes downstream of insertion sites, much of which is ongoing. The virulence-associated genes identified thus far encode proteins involved in transcriptional and post-transcriptional control, efflux systems, biosynthetic enzymes involved in phenazine production and proteins of unknown function (Table 1). As expected, genes involved in general cellular metabolism and nutrient uptake were absent, as auxotrophs were intentionally omitted from the screens. However, some anticipated classes of mutations were conspicuously under represented. First, no mutations were obtained within genes encoding the type III secretion system, which transports virulence factors in response to host cell contact. Curiously, this system is one of the best-characterized examples of the parallels that exist between plant and animal pathogens (reviewed by Hueck, 1998). A second class of mutations that was insufficiently recovered was in genes encoding effectors or toxins that interact directly with the host. Although missing classes of mutations could be an indication that the screen is not yet comprehensive, and some effectors may be overlooked as they are encoded by the genes lacking matches in the GenBank database, an alternative explanation is that effectors, unlike regulatory molecules that control their expression, are host specific, and those relevant to mammalian pathogenesis might be missed through these screens. Although it is too early to disprove this latter argument definitively, there are some indications that this may not be the case. First, at least one class of multihost effector molecules, the phenazines, was identified using the C. elegans fast-killing screen (Mahajan-Miklos et al., 1999). Secondly, mutations specifically generated in the toxA gene, encoding the known virulence factor exotoxin A, attenuated virulence in plants (Rahme et al., 1995), nematodes (Tan et al., 1999a) and mice, suggesting a conservation of the mechanism of action of this toxin in different hosts. Thirdly, recent studies have revealed that the similarities in the type III secretion system in plant and animal pathogens extend beyond the mechanism of secretion to the actual effectors that are transported (Hardt and Galan, 1997; Mills et al., 1997; Monack et al., 1997). In view of these findings, it is likely that genes encoding multihost effectors do exist but, given the multifaceted nature of P. aeruginosa pathogenesis, such mutations might be only marginally reduced in virulence, making them harder to recover. In addition to comprehensive screening, more sophisticated and sensitive screens may be needed to identify these effectors. Such screens are possible in the context of a host and a pathogen that are both amenable to genetic analysis.
In recent years, two additional genetic methods, those of in vivo expression technology (IVET) and signature-tagged mutagenesis (STM; reviewed by Strauss and Falkow, 1997), have been applied to the analysis of P. aeruginosa virulence (Wang et al., 1996a,b; Lehoux et al., 1999). Like the multihost screening approach, these methods allow the application of the powerful techniques of random mutagenesis and high-throughput screening for the identification of bacterial virulence factors in the context of actual host environments. Using STM, 45 P. aeruginosa mutants have been identified that exhibited attenuated virulence in a chronic lung infection model; these mutants await characterization (Lehoux et al., 1999). By applying the IVET selection system to a neutropenic mouse model, 22 P. aeruginosa loci that are specifically induced in vivo were identified, 15 of which shared sequence homology with reported GenBank sequences (Wang et al., 1996b). IVET was also used to identify P. aeruginosa loci that are inducible by respiratory mucus derived from cystic fibrosis patients, leading to the identification of three genes, all of which shared homology with known sequences (Wang et al., 1996a). Interestingly, there is no overlap between the genes of known sequence identified using the two IVET screens and the multihost screens. Although the limited numbers of P. aeruginosa mutants screened could explain this result, application of the techniques of IVET and STM to Yersinia enterocolitica also resulted in the identification of few overlapping loci (reviewed by Perry, 1999). These findings illustrate the complementary nature of different screening methods and the importance of applying multiple approaches to the study of a single pathogen. As none of the existing technologies has been used comprehensively to screen the genome of P. aeruginosa, additional comparisons between them are not very informative.
The method of using non-mammalian hosts advances the field of bacterial pathogenesis in several respects. Our studies illustrate that there is an extensive degree of conservation in the virulence mechanisms used by P. aeruginosa in diverse hosts, and that many of these fundamental mechanisms can be identified efficiently using non-mammalian hosts. This information could lay the foundation for the development of new antimicrobial agents against P. aeruginosa, whose high intrinsic resistance to antibiotics currently makes it difficult to treat. Whether the multihost screening approach can also be applied successfully to the identification of virulence factors in additional bacterial pathogens, including specialized human pathogens, is a significant question that remains to be answered. From an evolutionary perspective, use of the P. aeruginosa multihost system to uncover both conserved and host-specific virulence factors will be extremely informative. Finally, the use of a pathogenesis model in which both the pathogen and the host are genetically tractable provides us with tools to decipher the function of bacterial virulence factors as well as host responses against them, issues at the core of host–pathogen interactions.