Edited by Ralph R. Isberg, Howard Hughes Medical Institute and Tufts University School of Medicine, Boston, MA, and approved March 21, 2014 (received for review October 2, 2013)
Our past understanding of pathogen evolution has been fragmented because of tendencies to study human clinical isolates. To understand the evolutionary trends of pathogenic bacteria though, we need the context of their nonpathogenic relatives. Our unique and detailed dataset allows description of the parallel evolution of two key human pathogens: the causative agents of plague and Yersinia diarrhea. The analysis reveals an emerging pattern where few virulence-related functions are found in all pathogenic lineages, representing key “foothold” moments that mark the emergence of these pathogens. Functional gene loss and metabolic streamlining are equally complementing the evolution of Yersinia across the pathogenic spectrum.
The genus Yersinia has been used as a model system to study pathogen evolution. Using whole-genome sequencing of all Yersinia species, we delineate the gene complement of the whole genus and define patterns of virulence evolution. Multiple distinct ecological specializations appear to have split pathogenic strains from environmental, nonpathogenic lineages. This split demonstrates that contrary to hypotheses that all pathogenic Yersinia species share a recent common pathogenic ancestor, they have evolved independently but followed parallel evolutionary paths in acquiring the same virulence determinants as well as becoming progressively more limited metabolically. Shared virulence determinants are limited to the virulence plasmid pYV and the attachment invasion locus ail. These acquisitions, together with genomic variations in metabolic pathways, have resulted in the parallel emergence of related pathogens displaying an increasingly specialized lifestyle with a spectrum of virulence potential, an emerging theme in the evolution of other important human pathogens.
Bacterial species are defined on the basis of phenotypic characteristics, such as cellular morphology and biochemical characteristics, as well as DNA-DNA hybridization and 16S rRNA comparison. Using high-throughput whole-genome approaches we can now move beyond classic methods and develop population frameworks to reconstruct accurate inter- and intraspecies relationships and gain insights into the complex patterns of gene flux that define different taxonomic groups.
Bacterial whole-genome sequencing has revealed enormous heterogeneity in gene content, even between members of the same species. From a bacterial perspective the acquisition of new genes provides the flexibility to adapt and exploit novel niches and opportunities. From a human perspective, integration of genes by bacteria has been directly linked to the emergence of new pathogenic clones, often from formerly harmless lineages (1, 2). In addition to gene gain, gene loss is also strongly associated with host restriction in acutely pathogenic bacterial species, such as Yersinia pestis and Salmonella enterica serovars, including Salmonella Typhi (3⇓–5), where gene loss can remove functions unnecessary in the new niche (6). These specialist pathogens show a much higher frequency of functional gene loss than closely related host generalist pathogens, such as Yersinia pseudotuberculosis (7).
Previous Yersinia genome studies (8, 9) have examined the evolution of pathogenicity by comparing strains from a selection of species or species subtypes within the genus, limiting our understanding of the evolutionary context of individual species. The majority of the Yersinia species are found in the environment and do not cause disease in mammals. Three species are known as human pathogens: the plague bacillus Y. pestis and the enteropathogens Yersinia enterocolitica and Y. pseudotuberculosis. Only Y. pestis and Y. pseudotuberculosis have been studied extensively in a phylogenetic context (7, 9, 10). In our study, we present a global analysis of diversity in multiple isolates representing all current species of the Yersinia genus to examine the evolution of bacterial pathogens in the context of the entire genus, encompassing both genomic features and metabolic signatures. It has been previously suggested that the pathogenic Y. pestis, Y. pseudotuberculosis, and Y. enterocolitica share a recent common ancestor to the exclusion of the nonpathogenic species (11⇓–13). Contrary to this, we show conclusively that the human pathogenic Yersinia lineages have evolved independently. Early ecological separation is likely to have split the most acutely pathogenic Yersinia strains from the environmental species and nonpathogenic Y. enterocolitica. In human pathogenic lineages this was then followed by a common pattern of acquisition of similar pathogenicity determinants, namely the virulence plasmid pYV and the invasin ail, followed by the acquisition of further determinants within lineages along with functional gene loss and reduced metabolic flexibility. This parallel evolution in distinct pathogenic Yersinia species may represent a paradigm of the emergence of important human pathogens from nonpathogenic bacterial species.
To investigate the evolution of pathogenicity within the genus Yersinia we selected a representative sample of strains composed of Y. pseudotuberculosis and Y. pestis, as well as the understudied Y. enterocolitica. In addition, a diverse set of 94 isolates representing all of the other 15 Yersinia species were also sequenced. In total, 241 Yersinia strains were analyzed, all of which had been typed by traditional biochemical and serological methods (Table S1).
Because high diversity within the population precluded the use of a read mapping-based approach, we used a set of core genes common to all strains to produce a genus-wide phylogeny. The resulting phylogenetic tree (Fig. 1 and Fig. S1) shows a wide diversity of clearly defined lineages within the genus, with clustering of isolates at the tips of long branches signifying very ancient common ancestry. To subdivide the population by patterns of shared sequence similarity we used the program Bayesian Analysis of Population Structure (BAPS) (14). BAPS resolved the genus into 14 species clusters (SC) (Fig. 1). Routine Yersinia species identification is largely based on limited biochemical data. Superimposing this information onto the phylogenetic tree revealed the lack of resolution provided by biochemical tests, and emphasized the need to use modern molecular methods in classifying bacterial species. Yersinia frederiksenii is known to be heterogeneous and Yersinia massiliensis was recently subspeciated from “atypical” Y. frederiksenii (15). As can be seen in Fig. 1, isolates described traditionally as Y. frederiksenii can be found in species complexes SC8, SC9, and SC14, with the latter also containing both Yersinia intermedia and Y. massiliensis isolates. On the other hand, the three species Yersinia aleksiciae, Yersinia bercovieri, and Yersinia mollaretii form a single, heterogeneous species complex SC13 (Fig. 1). In line with previous findings, the pathogenic lineages of Y. pestis and Y. pseudotuberculosis form a tight group (16), with the nonpathogenic Yersinia similis appearing basal to these species on the tree (17) (Fig. 1). The other human pathogenic species, Y. enterocolitica, forms a pair of distinct complexes in the tree, SC6 composed of the pathogenic biotype (BT) 1B and the nonpathogenic BT 1A, and SC7 composed of the other pathogenic BTs 2–5 (Fig. 1).
The phylogeny of the genus Yersinia and the virulence plasmid pYV. Maximum-likelihood phylogenetic tree of the genus Yersinia based on the concatenated sequence of 84 housekeeping genes. Current species assignments based on biochemical typing (36) (color circle borders) are contrasted with the species complexes (colored circles) as allocated by BAPS. The SC for the Y. frederiksenii type strain (8) is depicted by an asterisk. Arrows show the independent acquisition events of the virulence plasmid pYV. The pYV plasmid tree for pathogenic Y. enterocolitica and Y. pseudotuberculosis samples is shown.
The remaining species occupy intermediate branching positions between the Y. pseudotuberculosis and Y. enterocolitica SCs with the fish pathogen, Yersinia ruckeri, falling on a distinct divergent branch (SC2) (Fig. 1). This branch is also occupied by Yersinia entomophaga and Yersinia nurmii, two newly characterized species grouped within a single species cluster, SC3.
Numerous studies have defined important virulence determinants in Y. pseudotuberculosis/pestis and Y. enterocolitica (18⇓⇓⇓⇓–23). Analysis of these virulence genes has formed a central narrative in our understanding of the evolution of these pathogens. However, athough these determinants have been well characterized in human pathogenic lineages, their origin and distribution across other members of the genus remain unclear or incomplete.
From the phylogenetic distribution of known pathogenicity determinants (Fig. 2 and Dataset S1) it is evident that the distribution of virulence-related genes falls into three broad categories: (group I) genes represented in all lineages of the genus, (group II) genes selectively gained or lost, wholly or partially, by entire lineages, and (group III) those peculiar to single isolates.
Distribution of pathogenicity determinants across the genus. Three broad groups of pathogenicity determinants present in the genus are highlighted (see text and Dataset S1). Heatmap colors are based upon average amino acid identity either of single genes or as an average of the amino acid identity across the genes in the operon, as indicated. The percent identities were identified using BLAST searches of the assembled genomes. The corresponding species complexes (Fig. 1) and PGs (Fig. 3) are highlighted. Comparator sequences used: 1, Y. enterocolitica (YE8081); 2, Y. enterocolitica (YE212/02); 3, Y. pseudotuberculosis (IP32953); 4, Y. pestis (CO92); 5, Y. intermedia (ATCC29909). Gene names are given for pathogencity determinants, operons are labeled with their names. Abbreviations: Flg, flagella cluster; Ygt, Yersinia genus T3SS; 2-CS, two component system; eff, effectors; reg, regulator; app, apparatus; mem, membrane proteins; met (salv), methionine (salvage); YAPI, Yersinia adhesion pathogenicity island; T2SS, type 2 secretion system (general secretion pathway); fes/fep, siderophore operon; tcPAI, toxin complex pathogenicity island; HPI, high-pathogenicity island.
Virulence-related genes present in all lineages (group I) include the Flag-1 flagella biosynthesis cluster and the global virulence regulator RovA (Fig. 2), which controls the expression of multiple key virulence factors (23, 24). RovA has been suggested as a promiscuous ancestral regulon into which related as well as unique functions were incorporated multiple times in different lineages, as indicated by the little overlap of regulatory networks between Y. enterocolitica and Y. pseudotuberculosis (23).
Selectively acquired or lost genes in group II include a chromosomal type 3 secretion system (T3SS) recently described in Y. enterocolitica (25⇓–27). This system is most similar to the ancestral Salmonella pathogenicity island 2 (SPI-2) encoded T3SS and we propose to call it Yersinia genus T3SS (Ygt). T3SS have been associated with important roles in bacterial disease because they allow direct injection of T3SS effector proteins into eukaryotic cells. Although there is clear evidence for the presence of Ygt in all Yersinia lineages, it appears to be in the process of being lost in the highly pathogenic species (Fig. 2). It is also clear that the loss of Ygt is always coincident with the acquisition of an alternative T3SS: either chromosomal Ysa or the virulence plasmid, pYV, bearing the Yop T3SS (see below).
Group II also includes metabolic pathways including the cobalamin (vitamin B12) biosynthetic operon (cob), 1,2-propanediol utilization pathway (pdu), tetrathionate respiration genes (ttr), hydrogenase complexes (hyd4 and hyd2), and cellulose biosynthesis pathway (cel). Across the genus, these dispersed operons are absent from a distinct branch of the Yersinia phylogenetic tree encompassing SC1–3 (Y. pestis, Y. pseudotuberculosis, Y. similis, Y. ruckeri, Y. nurmii, and Y. entomophaga), except for ttr, which has been independently deleted in SC1 and -2, and is only maintained by SC3 (Figs. 1 and 2). The distinct distribution of these functions across the genus is likely to be evidence of early ecological specialization in Yersinia lineages. Salmonella, like Yersinia, synthesizes cobalamin only under anaerobiosis (28, 29) and is able to degrade 1,2-propanediol or ethanolamine by cobalamin-dependent enzymes using tetrathionate as a terminal electron acceptor, making the endogenous production of cobalamine essential for growth (30). Tetrathionate was previously thought to be important for environmental survival, but has now been shown to be produced in the vertebrate inflamed gut following the host’s response to Salmonella Typhimurium infection, whereby the SPI-1 and -2 encoded T3SS are essential for stimulating the inflammatory response (31, 32). This widely distributed metabolic capability of Yersinia SCs 4–14 could provide a competitive growth advantage similar to that observed in S. Typhimurium over the largely fermentative gut microbiota by using naturally occurring carbon sources that are not readily fermented.
Only two acquisitions were evident on the tree as delimiters between pathogenic lineages and their most closely related “environmental” or nonpathogenic sister taxa (Fig. 2, group III, and Dataset S1). These “foothold moments” are the acquisition of the chromosomal attachment and invasion locus ail (19) (Fig. 2) and the plasmid pYV. Plasmid pYV encodes the Ysc-Yop T3SS apparatus, Yop-secreted protein effectors and chaperones (33), and is required for optimal virulence in all human pathogenic Yersinia. Reconstructing the phylogenies of pYV (Fig. 1) showed far fewer mutation events than would be predicted from the genus tree, but that it segregated with the host lineages. Also, unlike any other pYV lineage, pYV from BT 1B strains (SC6) possess a distinct origin of replication (34, 35). The most parsimonious explanation for the isolated presence of highly similar plasmids at extremes of the core phylogenetic tree is that recent independent acquisition of distinct versions of pYV has occurred in both the Y. enterocolitica lineages SC6 and -7, as well as the Y. pseudotuberculosis/pestis lineage SC1, contrary to existing hypotheses (11⇓–13).
Traditionally the Y. enterocolitica species cluster has been divided into six biotypes BT 1A, BT 1B, and BT 2–5 (36), which are further subdivided into various serotypes (Fig. 3 and Table S1). Biotyping is based on a number of biochemical tests that assesses the metabolization or acid production of a panel of metabolites (Fig. 3). The biotypes show differing pathogenic potential in the mouse model, with BTs 1A, BT 1B, and BTs 2–5 being described as non-, high-, and low-pathogenic lineages, respectively (11). Examining the phylogenetic relationships of the species using SNPs called against the reference BT 1B strain 8081 (37) revealed six distinct lineages (Fig. 3). BT 1A and 1B form discrete clusters, whereas BTs 2–5 consist of four closely related but distinct lineages, confirmed by BAPS clustering (Fig. 3). For the low-pathogenic Y. enterocolitica lineages, the phylogeny is largely congruent with serotype (Fig. 3). In subsequent analyses the six Y. enterocolitica lineages will be referred to as phylogroups (PG) 1–6.
The phylogeny of Y. enterocolitica. Maximum-likelihood phylogeny for the species based on SNPs across the whole genome excluding laterally acquired elements and phages. Lineages are characterized by biotype, biological origin, country of origin, and serotype. The biochemical tests used for biotyping (36) of Y. enterocolitica are presented for each strain. Shaded boxes highlight results of genomic serotype analysis differing from in vitro results. Biotyping reactions are: 1, salicin acid production; 2, pyrazinamidase activity; 3, esculin hydrolysis; 4, lipase activity; 5, indole production; 6, xylose acid production; 7, trehalose acid production; 8, sorbose acid production; 9, nitrate reduction; 10, ornithine decarboxylase; 11, Voges-Proskauer test; 12, inositol acid production.
The extent of divergence within the Y. enterocolitica SC is striking. PGs 3–6 display tight clusters with short terminal branches compared with PGs 1 and 2. One possible explanation for the low genetic diversity within PGs 3–6 is that they are the product of one or more recent population bottlenecks or population expansions. Although there is no discernible clustering of PGs 3–6 according to host origin, the strong correlation between the phylogenetic signal and serotype in this group may suggest that their early evolution was dominated by ecological specialization.
It is evident from the results of the 12 classic biotyping reactions (Fig. 3) that metabolic flexibility differs between the diverse PGs 1 and 2 and the nondiverse PGs 3–6. This difference in metabolic flexibility does not appear to result from markedly different distributions of the functional genes possessed by each phylogroup representative (Fig. S2A), but it does correlate with the accumulation of pseudogenes and the expansion of insertion sequence elements (Fig. S2B and Table S2). The absolute numbers of pseudogenes varies significantly (χ2 test P value 5.5E-59) between the different Y. enterocolitica lineages with the highest number present in strains in PGs 3 and 6 and the lowest in PG 1 strains. However, there is no significant difference in the class distribution between phylogroups, suggesting a general loss of functions across all classes (Fig. S2B). In agreement with the observed link between genetic and metabolic decay, PG 6 strains, with their high numbers of pseudogenes, have not been isolated from livestock and are reported to be host restricted to hares (38, 39).
To determine whether the signal seen from biotyping extended to the whole metabolic capacity of the biotypes, we selected representative strains from each of the six Y. enterocolitica phylogroups and tested their ability to metabolize a wide range of different carbon, nitrogen, sulfur, and phosphate sources on a phenotypic microarray (Biolog). The resulting signal values correspond to the intensity of utilization for different nutrient sources (Fig. S3 and Dataset S2). Variability in metabolic flexibility is evident across the 380 metabolic tests, with PG 1 strains showing the broadest metabolic capacity and the PG 6 strains exhibiting the most extreme loss of capacity. Principal component analysis (Fig. S4) using the phenotypic signals showed a clear separation between PGs 1, 2–5, and 6 along the first component, accounting for ∼60% of the variance in our phenotypic data matrix. We interpret this component as capturing a gradient of metabolic restriction, from the broadly active environmental PG 1 strains to the host-restricted PG 6 strain. We believe that the lack of a specific metabolic signal together with the observation that most of the mutations in the PGs 3–6 are derived from single insertion/deletion events suggests that the genome decay seen in PGs 3–6 is the result of recent evolutionary events. This finding is also consistent with the short evolutionary distance observed in the branch leading to PGs 3–6 (Fig. 3).
Like the phylogenetic analyses, the phenotypic results emphasize the parallelism between the two pathogenic species clusters. It has been shown that combined plasmid gain and chromosomal gene loss was a predominant driving mechanism in the evolution of the highly specialized, lifestyle-restricted clone Y. pestis, which lacks the metabolic repertoire of the wider Y. pseudotuberculosis species cluster. The reduction of metabolic flexibility through gene loss from the ubiquitous plasmid-negative PG 1 strains to the lifestyle-restricted PG 6 is particularly striking, and appears to show a similar pattern to that observed in the gradient from Y. similis to Y. pseudotuberculosis and Y. pestis. This appears to be a case of parallel evolution of virulence within a bacterial genus.
The data presented here bring together our often fragmented view of bacterial pathogen evolution. Through whole-genome sequencing of over 200 genomes we have provided a robust framework to redefine the species clusters of this genus, mapping genetic traits across its full diversity and scoring the significance of pathogenicity genes based on their presence or absence. By looking at functions gained and lost from specific lineages it is clear that there are metabolic functions, such as the ability to make endogenous vitamin B12 or anaerobically respire tetrathionate that differentiate SC1 and -2 from all other Yersinia lineages (Fig. 1). The wide phylogenetic distribution of these functions suggests they are ancestral, important for growth in a range of niches, and have been lost rather than gained in specific lineages.
It has been previously subject to debate whether human pathogenic Yersinia shared a common pathogenic ancestor (11⇓–13). In fact, the pathogenic lineages occupy positions at diametrically opposite ends of the Yersinia genus tree. From studies looking at Y. enterocolitica pathophysiology we know that infections are characterized by inflammation of the gut and sometimes of the mesenteric lymph nodes. It is also clear that inflammation requires pYV, as without it there is no inflammation and pYV-cured Y. enterocolitica are rapidly eliminated from the gut (40). In Y. pestis, pYV has been shown to have an early anti-inflammatory effect preceding the inflammatory response of the host. Although not demonstrated in Y. enterocolitica, it may be speculated that a similar early anti-inflammatory effect is necessary for Y. enterocolitica to establish itself at the site of infection before the onset of the host inflammatory response (41), Either way, it is clear that pYV is a key factor allowing Y. enterocolitica to persist in the mammalian gut lumen long enough for the host to mount an inflammatory response. In the course of such an inflammatory response tetrathionate is produced, and can be used to gain a metabolic advantage over the resident gut microbiota (32). This finding may partly explain why the acquisition of pYV represented an apparent foothold moment in the evolution of these pathogens, allowing them to exploit a new niche.
The dispensability of metabolic functions in Y. pestis can be explained by adoption of a lifestyle bypassing the gut infection phase, and for Y. pseudotuberculosis similar losses are likely to be explained by its greater invasiveness and occupation of alternative niches away from the mammalian gut. Because even the nonpathogenic members of SC 1–3, such as Y. similis, have lost these functions, this is evidence of an early change in niche before the evolution of the other human pathogenic Yersinia species in SC1. This finding may also explain why the evolution of pathogens in SC 1 is markedly different from Y. enterocolitica SC 6–7 in terms of gene gain. Y. pestis is characterized by the acquisition of an array of mostly plasmid borne virulence functions, which apart from pYV are found nowhere else in the genus.
Despite the differences in metabolic pathways there is considerable symmetry between the two branches leading to the pathogenic Yersinia; both have closely related nonpathogenic relatives and have independently acquired pYV. In addition, species clusters within these lineages show signs of having passed through an evolutionary bottleneck. Although this process has been well described for Y. pestis, it is evident from these data that much of the diversity has been similarly removed in Y. enterocolitica PG 3–6 populations. Large-scale functional gene loss through pseudogene formation and insertion sequence element expansion is evident in both lineages generally, but especially in the most extreme host-restricted PG 6 strains and Y. pestis (3). Looking outside of the Yersiniae, these signatures have also been shown to occur in recently emerged pathogens, such as S. Typhi (4), where they too are considered to be indicative of a recent evolutionary bottleneck and a change in lifestyle or niche (42). The patterns of niche adaptation and metabolic streamlining seen in PG 3–6 strains, unlike Y. pestis, are likely to be the consequence of transition from environmental ubiquity to specialization in enteric infection of animals rather than adapting to a new lifestyle including an insect vector.
What is significant about Yersinia is that until now we have associated widespread functional gene loss with acutely pathogenic lineages. Although this is true of Y. pestis, it is not completely true for Y. enterocolitica, where the degraded genomes are found in PG 3–6 strains. These latter strains show limited pathogenicity in mouse models and, although not the most acute pathogens, are the most successful lineages in terms of disease causation. These series of events that appear as common evolutionary paths in the most distant Yersinia lineages, and across the Enterobacteriaceae, are markers of the emergence of lineages with an increasingly restricted lifestyle. These highly adapted organisms can cause a spectrum of disease and have emerged independently, on multiple occasions, out of an environmental background to become successful human and animal pathogens. Thus, these dramatic genetic changes we have demonstrated in the genus Yersinia form a paradigm rather than a unique combination of chance events, and appear to underlie the emergence of pathogenic enterobacteria.
The genomes of six Y. enterocolitica strains (YE12/03, YE56/03, YE53/03, YE212/02, YE149/02, YE3094/96) were sequenced and completed to an Improved High Quality Draft (43) standard using multiple sequencing technologies (ABI3730 automated sequencers 2× coverage, 454/Roche GS20/FLX 30× coverage, and Illumina GAII 60× coverage). End-sequenced PCR products were used to close gaps and scaffold large repeat regions. The assembly was corrected using ICORN (44).
Sequence for the worldwide collection was generated to a minimum of 20-fold coverage using the Illumina sequencing platform (GAII instruments; 6–12 samples per lane of sequencing, 200- to 300-bp fragments, 76-bp paired-end reads). This collection included 27 Y. pseudotuberculosis and 118 Y. enterocolitica strains, as well as further 78 isolates of environmental species, plus previously published genomes (Table S1). The genome accession numbers can be found in Table S1.
To determine the phylogenetic relationship of the genus Yersinia housekeeping genes present, all Yersinia sequenced were identified based on maximum of 25% SNP divergence between Y. pestis and Y. enterocolitica. The genes were extracted from de novo velvet assemblies (45) and a concatenated sequence of the chosen 84 genes encompassing 99,724 bases (Table S3) was used to reconstruct the phylogeny of the chromosome. The plasmid alignment was generated by mapping sequenced reads against pYVe8081 (46). See Dataset S3 for single copy core genes used for reconstruction of pangenome genus phylogeny of 50 representative isolates (Fig. S1).
The concatenated sequences for the chromosome and pYV plasmid were both used to construct an unrooted phylogeny using RAxML with a general time-reversible evolutionary model and γ-correction for among site rate variation. Species complexes were defined using the BAPS program (14) run for three independent iterations with upper population sizes of 25, 30, and 35, to obtain the optimal clustering of the population (47).
The Y. enterocolitica SNP analysis was based on whole-genome alignment produced by mapping using the YE8081 genome as reference as described previously (46). Mobile genetic elements were removed, and the resulting SNP alignment was analyzed using RAxML (see above). Hierarchical BAPS clustering was performed to define the population structure.
Virulence genes were identified using BLAST searches of the assembled genomes (Dataset S1). Heatmap colors are based upon average amino acid identity either of a gene, where a single gene is examined, or as average of the amino acid identity across the genes in the operon, as indicated.
Bacterial strains were cultured on Luria Bertani agar plates at 25 °C. Inoculation and preparation was as manufacturer instructions (Biolog). PM1, 2A (both carbon), PM3B (nitrogen), and PM4A (phosphorus, sulfur) were chosen. l-Cystine (homo-dimer of l-cysteine, 12.5 µM) was added as a reduced sulfur source to all plates, and sodium pyruvate (2M) was added to PM3B and 4A. Plates were incubated under aerobic conditions for 48 h at 28 °C in the OmniLog Incubator/Reader, taking readings every 15 min. Three biological replicates were conducted for each experiment. Data were exported and analyzed in R with datapoints being transformed in to signal values, as described previously (Dataset S2) (48). Each plate showed a clear bimodal distribution of signal values, and normal distributions were fitted to the two peaks to characterize metabolic activity (“off” and “on”). Log-odds scores were calculated for each data point showing how likely the “on” distribution was for each nutrient source. A conservative cut-off of four times more likely to be “on” than “off” was used for further analysis. The limma R/Bioconductor package was used to examine differential metabolism whereby all strains were compared with YE8081c, and the Benjamini-Hochberg corrected P values were used to determine statistical significance of differences controlling for a false-discovery rate of 10%.
We thank Martin Cormican (University College Galway) and Dr. Brenda Murphy (University College Dublin), as well as Peter Roggentin, Henry Derschum, and Herbert Nattermann for providing strains for sequencing. This study was funded in part by Wellcome Trust Sanger Institute Grant 098051; Nottingham Trent University Vice Chancellor scholarships (to S.R. and M.E.H.); Academy of Finland Grant 1114075 (to M. Skurnik); and Finnish Ministry of Agriculture and Forestry Grant 4850/501/2004 (to A.S.). M.A. was funded by Science Foundation of Ireland Grant 05/FE1/B882.
↵1S.R. and T.R.C. contributed equally to this work.
Author contributions: S.R., T.R.C., B.W.W., J.P., M.A., A.M., and N.R.T. designed research; S.R., T.R.C., L.B., D.W., T.F., S.R.H., M.F., M.E.H., N.K.P., J.C., M.M., M. Shubin, and N.R.T. performed research; T.M.F., M.D., T.R., C.S., C.B., L.M., J.M.R., R.L.-N., L.M.S., A.S., J.P.F., H.F., and H.C.S. were involved in isolate collection and typing; S.R., T.R.C., L.B., D.W., S.R.H., M.F., M.E.H., N.K.P., J.C., M.M., M. Shubin, and N.R.T. analyzed data; and S.R., T.R.C., L.B., S.R.H., T.M.F., M. Skurnik, M.B.P., B.W.W., J.P., E.C., M.A., A.M., and N.R.T. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
Data deposition: The sequences reported in this paper have been deposited in the European Nucleotide Archive (ENA study nos. PRJEB2116, PRJEB2117) GenBank and SRA numbers are given in Table S1.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1317161111/-/DCSupplemental.
Freely available online through the PNAS open access option.
