To identify all of the putative pathogenicity islands that distinguish PA14 from PAO1, we sequenced the PA14 genome and found that it contains a slightly larger chromosome (6.5 megabases [Mb] versus 6.3 Mb for PAO1; genome sequence and annotations are available at the Ausubel lab PA14 sequencing website 20 and have also been deposited in GenBank [GenBank: CP000438]). Consistent with previous observations that overall strain similarity is high, we found that approximately 91.7% of the PA14 genome is present in PAO1, and that 95.8% of the PAO1 genome is present in PA14 (Table 1).

The PA14 and PAO1 genomes are largely colinear with the exception of one major rearrangement that has previously been described: an inversion between two of four dispersed copies of a large ribosomal RNA cluster (Figure 17). We used long-range polymerase chain reaction (PCR) spanning the rRNA clusters to verify the inversion in the sequenced PA14 and PAO1 genomes; PCR products predicted to be indicative of one orientation or the other were only generated when either PAO1 or PA14 DNA was used as the template (Figure 2a,b). When 18 diverse P. aeruginosa strains were subjected to this same PCR analysis, all 18 of these isolates were the same or resembled PA14 genome structure with respect to the large inversion, whereas none of the 18 strains had or resembled the PAO1 genome structure (data not shown). Of note, additional PAO1 clones (distinct from the isolate that was sequenced) also appear to contain the same inversion found in PA14 7.

We also compared the GC skews of the PA14 and PAO1 genomes. GC skew is defined as the value of [G-C]/[G+C] where G and C represent the local base frequencies of G and C, respectively. In prokaryotic genomes there is a bias toward G over C on the leading strand of DNA synthesis 3. Therefore, when measured at regular intervals along the chromosome (1 kilobase [kb] segments) GC skew tends to have a positive value on the leading strand of DNA synthesis and a negative value on the lagging strand, resulting in polarity changes at the origin and terminus of replication. The putative position of the replicative terminus is therefore operationally defined as the peak of the cumulative GC skew and it typically resides opposite the origin of replication in bacterial genomes 3. For PA14, the peak GC skew was indeed mapped opposite the replication origin (at 49.2% of the genome; Figure 2a,c). In contrast, the position of the terminus in the sequenced PAO1 chromosome is shifted relative to the origin (at 38.8% of the genome). These observations argue that the PA14 chromosome with respect to this large inversion is more representative of the canonical or ancestral P. aeruginosa genome than that of the sequenced PAO1 isolate. This interpretation that is further supported by the finding that the inversion in the sequenced strain occurred sometime after 1990 (a PAO1 cosmid library originally described in that year was shown not to contain the inversion 21). However, the physiologic consequences of such an inversion and asymmetric replication cycle are not clear.

We annotated the PA14 genome and identified 5973 predicted open reading frames (ORFs; 322 more than in PAO1; Table 1). Summaries of all predicted PA14 genes and their distribution among different functional categories are presented in Additional data files 2 and 3, respectively. We performed a reciprocal NCBI-BLAST (basic local alignment search tool) search of each gene in either PA14 or PAO1 against the total collection of genes in the other strain (specifically, each PA14 gene was BLASTed against the complete set of PAO1 genes, and vice versa). The position of each best BLAST hit was plotted (with the PA14 genomic location on the x-axis, and the PAO1 coordinate on the y-axis; Figure 3), confirming that the two genomes are largely colinear, with the exception of the large inversion described above. Individual PA14 genes that had no counterpart in PAO1 are shown as individual (green) data points on the x-axis, and PAO1 genes absent in PA14 are represented as (pink) data points on the y-axis. Large clusters of these strain-specific genes often correlated with regions of the respective genomes in which the local GC content was lower than that of the total genome (Figure 3).

Using a combination of raw sequence and ORF-based global alignments of the two genomes, we compiled a list of gene clusters present in one strain but absent in the other (58 PA14 regions absent in PAO1 containing 478 genes, and 54 PAO1 regions absent in PA14 containing 234 genes; Table 1 and Additional data file 4). We refer to these as PA14-specific or PAO1-specific regions for the purposes of this discussion (recognizing that these genes may be present in other isolates of P. aeruginosa and are not strictly strain specific). Many of these gene clusters have hallmarks of horizontally acquired DNA, including direct repeats, insertion sequences, tRNA genes at their boundaries (data not shown), and/or anomalous GC contents (PA14-specific clusters have an average GC content of 59.6%, which is more than one standard deviation below the whole genome average of 66.3%; Table 1).

These strain-specific regions contain a high proportion of genes of unknown function as compared with the whole genome. Each PA14 (and PAO1) annotation has a confidence score associated with the described gene function 722. A class 1 designation refers to genes whose functions have been experimentally validated in P. aeruginosa, class 2 genes are highly similar to genes whose functions have been validated in other organisms, class 3 genes have hypothesized functions based on limited similarity to other genes or structural/functional domains, and class 4 genes are ORFs of unknown function. For the PA14-specific regions, 63% of the predicted ORFs have no known function, whereas only 28.9% of genes in the whole genome have class 4 annotations (Table 1). Therefore, gene identity alone could not indicate whether virulence-related genes were enriched in PA14-specific regions.

If a PA14 strain-specific region were functioning as a canonical pathogenicity island, then it should be more prevalent among other pathogenic isolates (and less prevalent among avirulent isolates). To test this hypothesis we needed to establish an objective measure of pathogenicity to compare different P. aeruginosa strains as well as to develop a high throughput method for determining the genomic content of the different strains. We used a model host infection system (the nematode Caenorhabditis elegans) to rank order the virulence of 18 diverse P. aeruginosa strains, using PA14 and PAO1 as reference strains. The 18 P. aeruginosa strains included 13 clinical isolates from a variety of infection types (CF lung infections, urinary tract infections, ocular infections, and blood isolates), one laboratory strain, and four environmental isolates. This same set of strains had been used in a previous study of P. aeruginosa strain diversity 9. We used a custom microarray-based system (described below) to determine relatedness of the 20 (18 plus PAO1 and PA14) P. aeruginosa strains.

In the C. elegans model pathogenicity system 13, an age-synchronized population of nematodes is fed the pathogen to be tested (instead of Escherichia coli, its traditional laboratory food source) and the longevity of the nematodes is determined. As shown previously, the longevity of C. elegans feeding on a particular P. aeruginosa strain correlates with virulence of the strain in mice 131518. The set of P. aeruginosa strains tested exhibited a full range of virulence phenotypes, including both the upper and lower limits that the assay system is capable of measuring (Figure 4a). PA14 (dark blue diamonds, second curve from the left) is extremely efficient at killing nematodes, whereas the less pathogenic PAO1 is intermediate (pink squares), killing more slowly than PA14 but more quickly than E. coli (negative control; yellow squares, second curve from the right).

The observation that PAO1 is more virulent than many other tested strains suggests that it has not become attenuated because of extensive passaging in the laboratory; rather, its virulence is probably that of a moderately pathogenic strain, with other strains more representative of truly 'avirulent' isolates. When comparing strains derived from the same type of infection, there was no consistent clustering with respect to their phenotype in C. elegans. For example, both the most and the least virulent strains tested were isolates from CF infections. Similarly, the five urinary tract infection strains exhibited a wide range of virulent to avirulent phenotypes. Importantly, two closely related environmental isolates (MSH3 and MSH10) were the fourth and fifth most virulent strains tested, indicating that nonclinically isolated strains can also have the potential to be infectious.

To test the hypothesis that the virulence of the 20 P. aeruginosa strains correlates with the presence of particular virulence islands, we performed a microarray-based analysis of the genomic content (a process described as genomotyping 23) of the P. aeruginosa strains. We arrayed 285 synthetic oligonucleotides (70 mers) corresponding to PA14 genes that are absent in PAO1 and 130 oligonucleotides corresponding to PAO1 genes that are absent in PA14, along with additional sequences serving as positive and negative controls (see Materials and methods, below, and Additional data files 5 and 6). We first used the array data to generate a hierarchic clustering dendrogram showing the relatedness of the 20 P. aeruginosa strains (Figure 4b). Next, we used the C. elegans virulence data in Figure 4a to rank order the 20 strains and then examine the microarray data to determine whether the presence or absence of PA14-specific (or PAO1-specific) genes correlated with virulence (also see Table 2 and Additional data file 6). Strains were ordered from the most virulent (rank order 1) to the least virulent (rank order 20), with strains that had equivalent virulence assigned a tied rank. When the virulence rank order of each strain was superimposed on the dendrogram describing the relatedness of each strain based on genomic content (Figure 4b), no strong correlation was observed with the relative virulence of each strain in C. elegans. There are small clusters of similarly avirulent or virulent strains; for example, strains 6077, U2504, JJ692, S54485, and X13273 grouped together and possess similar virulence rankings (10, 6, 12, 10, and 10, respectively). However, these small clusters were occasionally punctuated by exceptions: a cluster of weakly virulent or avirulent strains (CF5, E2, PAK and CF27, with rank orders of 20, 17, 13, and 18, respectively) also includes strain UDL, which is the third most pathogenic strain tested.

In a previous study, Wolfgang and coworkers 9 used Affymetrix GeneChips to survey these same 18 strains (with the exception of PA14) for the presence or absence of PAO1 genes and found no obvious pattern that would correlate genomic content with disease (or the type of infection from which the isolate was derived). Similarly, examination of the distribution of PA14-specific genes among the tested isolates did not reveal any obvious clustering of genomic content with respect to the source of the strain (see dendrogram in Figure 4b).

The experiments described above showed that there was no correlation between the PA14-specific sequences in general and virulence in the C. elegans killing assay. We therefore performed a functional analysis of PA14-specific ORFs, identifying genes that were specifically required for pathogenicity and subsequently assessing their distribution among the other strains. Our laboratory has constructed a genome-wide, nonredundant transposon insertion mutant library in PA14 2425. Using this library, we conducted a screen for mutants in PA14-specific genes that had reduced virulence in C. elegans. Nine genes were identified, which fell into six distinct clusters (Table 3). Three of these clusters are present in all of the other 19 P. aeruginosa strains but contain highly divergent sequences, including the O-antigen biosynthetic cluster (region PA14R38) and two groups of type 4 fimbrial biogenesis genes (regions PA14R77 and PA14R79). Oligonucleotides corresponding to the PA14-version of the O-antigen biosynthesis genes were included on the array and demonstrated that none of the other strains contain the PA14-versions of these genes; the genes for the two fimbrial synthesis clusters were not included on the array (because the sequences were not sufficiently divergent to meet our criteria for oligonucleotide design as outlined in Additional data file 1).

The remaining three clusters contain genes that are absent in PAO1. As indicated in Table 3, these three clusters contain five virulence genes, including one gene with putative transcriptional regulator activity and four ORFs of unknown function. Figure 4c summarizes the array data for these three regions. Each of the 20 strains tested are represented in columns arranged from left to right in order of decreasing pathogenicity (see Table 2 for rank order of virulence). The PA14 genes tested (shown in rows) are described as present (blue), absent (yellow), or indeterminate (red, indicating that the hybridization intensity was intermediate and a present or absent call could not be made with confidence; see Additional data file 1 for further details). The positions of the genes that result in an avirulent phenotype when mutated in PA14 are indicated on the right (also see Table 3). If a cluster of genes was required for or predictive of virulence in nematodes then we would expect to see a bias for present (blue) calls toward the left and absent (yellow) calls toward the right.

For these five virulence genes (and their associated three clusters), Spearman's rank correlation coefficients (relating their presence, absence, or indeterminate status with the virulence of the strain) were found to vary dramatically. The best correlation was found for two genes in region PA14R41 (correlation coefficients of 0.63). This is a cluster of eight genes known as the clone C-specific region common to clone C isolates (members of a clone family associated with CF infections 26). Intermediate to no correlations were observed for the remaining two regions, namely PA14R78 and PAR09, which had correlation coefficients of 0.44 and 0.02, respectively. Of note, PA14R78 is a previously described pathogenicity island (PAPI-1) shown to contain genes required for pathogenicity in plants and mammals 19, although the two mutations in genes of unknown function that we identified in this cluster (PA14_59010 and PA14_59070; positions 6 and 7 in Figure 4c) were not among those previously examined. Regardless of the magnitude of the correlation, each virulence gene had exceptions to the expected trend, being present in attenuated strains and/or absent in virulent strains. Taken together, the functionally defined PA14 genes required for C. elegans killing are neither required for nor necessarily predictive of another strain's ability to be pathogenic.

Shotgun sequencing of PA14 was performed using 65,800 plasmids containing 2-4 kb fragments of genomic PA14 DNA, resulting in over 10-fold coverage. Sequence reads were assembled using the Phred, Phrap, and Consed tools 373839. Details of finishing methods are described in Additional data file 1.

A long-range PCR-based method was used to assess whether particular P. aeruginosa genomes contain a PA14-like or a PAO1-like arrangement of genome sequence between the two most distant copies of a large ribosomal RNA cluster (Figures 1 and 2a,b; also see Stover and coworkers 7). Primer pairs used to amplify two products specific for the PA14 chromosomal arrangement are as follows: 807.LL_rev (CGAACTGGAGGAAGTCTTCG) + 2-5'_2 (CGAGGCTTTCGTCTATCCAG), and 837.LL (AACTGGTGGAGGGAGAAGGAT) + 803.RR.rev (TAGCCTTCAATTCCACCTGG). Primer combinations used to amplify two PAO1-specific products are as follows: 807.LL_rev + 803.RR.rev, and 837.LL + 2-5'_2. These diagnostic primer pairs were also used to survey 18 additional strains (Table 2) to determine whether the PA14 or the PAO1 orientation was more representative of the ancestral P. aeruginosa chromosome. Of the 18 strains tested, none generated a PAO1-specific amplification product. Using the primers for the PA14 arrangement, seven of the 18 strains yielded products with both primer pairs, nine of 18 had a strong PCR product with one of the two primer pairs and a weak product with the second primer pair, and two of 18 exhibited a weak product with one primer pair and no product with the second.

In determining GC content for PA14 and PAO1, standard deviations were determined using a continuous sliding window of 1 kb, and comparing the GC content for these regions with that of the whole genome. GC skew analysis was performed (using the formula GC skew = [n(G) - n(C)]/[n(G) + n(C)], where n [i] is the number of nucleotides, applied to 1 kb segments of each genome) to identify the peak cumulative GC skew as the likely position of the replication terminus 3. Global alignments of the PA14 and PAO1 genomes were performed using the MUMmer 3.0 software package 40 to identify strain-specific regions. Automated ORF predictions were made using a combination of the BLAST and Glimmer2 algorithms 4142. Each predicted ORF was assigned a PA14 LocusName, beginning with 'PA14_' followed by five numerals. ORFs were numbered starting with PA14_00010 (dnaA), increasing in increments of 10 to allow for future insertions of additional genes or functional RNAs. Further details of ORF prediction and annotation are provided in Additional data file 1.

70-mer oligonucleotides for microarrays were designed as described previously 43 for 285 PA14-specific sequences, 130 PAO1-specific sequences, 11 genes common to both strains (to serve as positive controls), and additional genes and negative controls (see Additional data file 1). Chromosomal DNA from PA14, PAO1 (the sequenced isolate), and 18 additional P. aeruginosa strains described previously 9 was digested with HaeIII and labeled with Cyanine-3 or Cyanine-5 using the MICROMAX ASAP labeling kit (part #MPS544001KT; PerkinElmer, Wellesley, MA, USA). Labeled samples were combined in random pairs and hybridized to the arrays. Observed intensities for replicate hybridizations for each strain were averaged and log2 ratios were computed to determine the presence/absence of each gene based on a cut-off determined independently for each sample (see Additional data file 1). For each gene, a Spearman's rank correlation coefficient was calculated to describe the relationship between the spectrum of present/absent calls and the determined rank order virulence in C. elegans (see below). Heirarchical clustering analysis of strain relationships was performed using Cluster 3.0 44 and Java Treeview 45. All microarray data are included in Additional data files 5 and 6 and have also been deposited in ArrayExpress (ArrayExpress E-MEXP-824).

The 20 P. aeruginosa strains analyzed by genomotyping were cultured overnight in Luria broth (LB) and assayed in the C. elegans killing system as described previously 13, with the exception that assays were performed using fer-15(b26)ts;fem-1(hc17) temperature-sensitive sterile mutant nematodes. The assays were carried out at the restrictive temperature (25°C) to prevent progeny formation and to allow the experiments to continue long enough to examine less pathogenic strains (for which progeny would typically overwhelm the assay plate by the end of the experiment). Bacterial strains are rank ordered from most to least virulent based on the position of the LT₅₀ (the time at which 50% lethality was observed) in Figure 4a. In cases in which the LT₅₀ was overlapping, rank orders were represented as a tied value for the strains in question. Strain CF27 is not included in the data shown in Figure 4a; however, its rank order in a similar experiment places it between strains E2 and S35004. The relative rank orders shown in Figure 4a are consistent with those observed in two additional experiments; although the absolute LT₅₀ values increased or decreased for given strains between experiments, the relative rank orders remained consistent.

To screen for avirulent PA14 mutants, we utilized a nonredundant transposon insertion library 25. All available mutants in PA14-specific genes (349 in total) were grown in 150 ml of LB media in 0.5 ml 96-well Masterblocks (part #786261; Greiner, Monroe, NC, USA). Cultures were agitated at 225 rpm and incubated for 16 hours at 37°C. A volume of 10 μl of each culture was spotted onto slow killing agar in 2 wells of a six-well plate 1318, allowed to grow at 37°C for 24 hours, and left at room temperature for an additional 24 hours. Five L4 stage N2 nematodes were transferred manually to each well, and the number and age of progeny were recorded after 4 days at 25°C. The 349 mutants in PA14-specific genes were screened on two separate occasions, and any that were highly attenuated in either screen or were attenuated in both screens were re-examined in a secondary screen as described using wild-type N2 nematodes 13. Six mutants were tested for growth in minimal (M63) media and all were found to have wild-type growth curves (Table 3). These included four of the five mutants in PA14 genes completely absent in PAO1 (examined in Figure 4c).