Sequencing of the type strains of the five Nakaseomyces species; Candida nivariensis CBS9983, Candida bracarensis CBS10154, Nakaseomyces (Kluyveromyces) delphensis CBS2170, Candida castellii CBS4332, and Nakaseomyces (Kluyveromyces) bacillisporus CBS7720, was performed at Genoscope (Evry, France), using a combination of Illumina and 454 technologies (see Methods). Genome data has been deposited at the EMBL.

Final assemblies showed a close correspondence between scaffolds and chromosomes (Additional file 1), and were annotated for coding and non-coding genes (see Methods). Flow cytometry results show that species are haploid, except N. bacillisporus which is diploid (Additional file 2). All species have an haploid genome size of 10 to 12 Mb. Chromosome numbers, as estimated by Pulsed-Field Gel Electrophoresis (not shown), range from eight in C. castellii, the smallest yet recorded number of chromosomes in post-WGD yeasts 16 , to fifteen in N. bacillisporus; with the ‘glabrata group’ exhibiting the least variation (10–13 chromosomes). Centromeres, similar in structure to S. cerevisiae’s, i.e. composed of three short “centromere defining elements”, CDEI, II and III 17 ; were identified in all species but not in all scaffolds. Telomeric repeats identical to those in C. glabrata and the putative telomerase RNA-component were found in the ‘glabrata group’.

Several ncRNA genes are known to be surprisingly large in C. glabrata, such as the RPR1 gene of RNAse P 18 and the above-mentioned TLC1 gene 19 . This tendency to exceptionally large ncRNAs seems to be general in the Nakaseomyces, such as the 1368 nt-long RPR1 gene in N. delphensis (only 369 nt-long in S. cerevisiae), and the 1937 nt-long U1 snRNA gene in C. castellii (568 nt-long in S. cerevisiae). Other structural genomic features are discussed in the supplementary results (see Additional file 3). None of the species contain detectable active transposons in their nuclear genome.

In order to clarify the phylogenetic relationships of the newly-sequenced Nakaseomyces and 17 other Saccharomycotina, we used two alternative phylogenomics approaches, namely i) a Maximum Likelihood (ML) analysis of a concatenated alignment of 603 protein families that have one-to-one orthologs in all the species considered, and ii) a super-tree approach based on the analysis of 4,965 individual gene trees, which finds the species topology that is most parsimonious in terms of implied duplication events in all the individual gene trees 20 . Both approaches yielded the same, highly resolved topology (Figure 1), which is largely congruent, for the shared species, with our current understanding of Saccharomycotina phylogeny 21 .

This topology supports the the existence of the Nakaseomyces genus, and defines two clear sub-groups separated by an ancient split. The first group comprises the highly divergent C. castellii and N. bacillisporus, whereas the second group (i.e. the ‘glabrata group’) includes the three pathogenic species and N. delphensis. The average protein identity between orthologs of species in the different sub-groups ranges from 51 to 53% (Additional file 4), which is similar to that of orthologs in the C. castellii/N. bacillisporus (53%) and C.glabrata/S.cerevisiae (54%) pairs, pointing to large levels of divergence. The ‘glabrata group’, is more compact and shows higher identity levels (77-88%). Notably, the two newly identified pathogens, C. nivariensis and C. bracarensis, are both closer to N. delphensis than to the most frequently isolated pathogen, C. glabrata (Figure 1). Thus the emerging picture for the appearance of pathogenesis is complex, with plausible alternative scenarios involving either gain of the ability to infect humans at the base of the sub-clade followed by loss of the trait in N. delphensis or three independent events of acquisition of pathogenicity. These possibilities will be further discussed below.

The degree of synteny, i.e. gene order conservation between species, was found to correlate with phylogenetic proximity, with the highest conservation occurring between C. bracarensis, C. nivariensis and N. delphensis (Additional file 5). Conservation of synteny of the Nakaseomyces in general, relative to S. cerevisiae is low. Detailed analysis of syntenic regions will be presented elsewhere (HD, TG, CF, in preparation).

As is the case for many fungal species described as asexual, genes involved in sexual reproduction are known to be conserved in C. glabrata 22 23 24 . In particular, although mating has never been reported in C. glabrata, the two additional MAT cassettes, HMRa and HMLalpha, and the HO gene are present in its genome, and haploid isolates of both mating types are found. In N. delphensis, also a mainly haploid species, these elements are also conserved 12 23 and mating-type switch may occur in culture 12 . For the remaining four species, three are considered asexual and mainly haploid, but the last one, N. bacillisporus, is considered to be diploid and homothallic 25 . As mentioned above, ploidy was confirmed by flow cytometry. All species contain a well-conserved homolog of the HO gene (Additional file 6), the most diverged encoded protein being the one from N. bacillisporus, which exhibits a C-terminal extension rich in proline and serine.

In all species, additional sequencing was needed in order to obtain the cassettes. In N. delphensis, N. bacillisporus and C. castellii, amplification of the MAT-like cassette yielded both a and alpha fragments. This raises the intriguing possibility that C. castellii switches mating types in culture, whilst having no described sexual cycle, or that it is sexual and goes through a diploid phase. This will need further experimental analysis. Figure 2 shows which cassettes are currently identified in these genomes. Genes within cassettes, a1, alpha1 and alpha2 are identified in all species, as is the Ho site, which can be recognised at the YZ junction. In cases where the three cassettes are identified, their configuration is apparently similar to C. glabrata’s: both HML- and MAT-like cassettes are on the same scaffold, while the HMR-like cassette is on a different one, except for N. delphensis, where the three cassettes are on the same scaffold. Thus, Nakaseomyces species follow the rule of conservation of HML and MAT on the same chromosome, noticed by Gordon et al. 26 .

C. nivariensis has two HMR-like cassettes, i.e. cassettes that contain type a information, and that are present in addition to the MAT locus, a situation already noted in other species 26 . Experimental testing of these cassettes will reveal what role, if any, these extra loci have, in organisms where sexual reproduction has not been characterized yet.

The total numbers of protein-coding genes range from 5400 to 5900, which is similar to what is found in C. glabrata 2 , and lower than in S. cerevisiae. Indeed, the number of true protein-coding genes in S. cerevisiae is estimated at around 5800, but this rises to around 6600 when dubious and other non-experimentally characterized ORFs are included, a figure more comparable to our predicted gene sets (SGD, 12 January 2012, http://www.yeastgenome.org, and Additional file 7). To assess the coverage of the predicted gene repertoires we tested for the presence of a set of 2,007 protein families previously found to be widespread in Saccharomycotina 21 . Our proteomes contained 99.4-100% of this core dataset, attesting for a high coverage in our procedures.

Intron-containing protein-coding genes are far fewer in C. glabrata than in S. cerevisiae (129 vs 287, 27 ). This paucity of introns is shared by all Nakaseomyces, which have intron counts lower than 230. Although experimental validation is needed, our data point to a remarkable loss of introns in the Nakaseomyces.

To accurately trace the evolution of the genetic repertoires across the Nakaseomyces, and confidently establish orthology and paralogy relationships, we performed an exhaustive phylogenomic analysis that included the reconstruction of ML phylogenies for every gene encoded in the Nakaseomyces genomes (i.e. the phylome). These were used to detect orthology and paralogy relationships 28 , and to map lineage-specific gene duplication 29 and gene loss events onto the species tree. Figures 3 and 4 and Additional file 8 summarize the main findings regarding the presence or absence of genes relevant to central processes. Lower gene numbers in the Nakaseomyces when compared to S. cerevisiae are reflected in the number of gene loss events indicated on the tree from Figure 3.

In particular, we have paid special attention to the retained copies of so-called ohnologs (i.e. gene duplicate pairs originated during the WGD event; 30 ). In S. cerevisiae, only 551 ohnologous pairs are left in the contemporary genome, as most such pairs have lost one member 31 . This is also the case in the Nakaseomyces genomes, as in all post-WGD species. The exact complement of ohnologs retained in duplicates varies across species and is likely to reflect particular physiological differences.

A noteworthy example comes from central carbon metabolism: it has been suggested that the six ohnologous gene pairs encoding glycolytic enzymes, such as ENO1/ENO2, and PYC1/PYC2, conserved in S. cerevisiae and C. glabrata, play an important role in the Crabtree effect, ie fermentation even in the presence of high glucose and oxygen 32 , a trait shared by all Nakaseomyces 33 34 . The genes involved are also in pairs in the ‘glabrata group’, but, in both C. castellii and N. bacillisporus, the situation is rather different, with single pyruvate kinase, enolase, and Glyceraldehyde-3-phosphate dehydrogenase genes (Figure 4, and Additional file 9). Furthermore, C. castellii has two hexokinase genes (HXK), as compared to four (two pairs of ohnologs) in the other species. Other features of central carbon metabolism in the Nakaseomyces are mentioned below and in Additional file 3.

Comparison of the proteins encoded in C. glabrata and S. cerevisiae’s genomes had revealed several features specific to the former. In particular, since there are fewer genes in C. glabrata than in S. cerevisiae, there was a special interest in specific gene loss. Indeed, there are at least four entire multigenic families which are absent in all Nakaseomyces or represented by a sole member in C. castellii and/or N. bacillisporus: the PHO, SNZ, SNO and PAU families. It is noteworthy that the paralogous PHO family of acid phosphatases has been lost, while the rest of the PHO pathway is conserved. Functional analysis in C. glabrata has shown that the PHO2 gene is present but not essential for regulation, and that the PHO4 gene is poorly conserved at the sequence level, but is functional 35 . Phosphate-starvation induced phosphatase activity in C. glabrata has been identified and is encoded by PMU1 homologs in tandem ( 36 , and see section on tandem arrays). The SNZ and SNO gene families are poorly characterized in S. cerevisiae, but their expression is known to be induced in stationary phase. Finally, the PAU family consists of at least 20 subtelomeric genes in S. cerevisiae, possibly involved in anaerobiosis 37 and similar to two other gene families, DAN and TIR, that encode cell-wall mannoproteins. There are no homologs to any of these genes in C. castellii and N. bacillisporus, while C. glabrata and C. nivariensis contain two copies of DAN-like genes, and the remaining species seem to contain only one homologous copy.

Also of particular interest was the loss of genes involved in de novo biosynthesis of nicotinic acid (BNA), which was hypothesized to result from the adaptation of C. glabrata to its human host 38 . Comparison to the newly sequenced Nakaseomyces has shown that the lack of BNA genes is common to them all (Additional file 8), regardless of their habitat, suggesting that the loss of this pathway is an ancient event. Notably, loss of BNA genes has also been observed in other, more distant species, such as Kluyveromyces lactis and other species, and seems to be a volatile trait 39 . In fact, all gene losses observed in C. glabrata relative to S. cerevisiae are shared by the ‘glabrata group’, whereas in the two other species gene absence is variable. A remarkable example concerns the genes necessary for allantoin catabolism. In S. cerevisiae, the subtelomeric DAL (

d

al

Other notable examples of gene gain/loss involve the translation machinery: both the number of ohnologous ribosomal protein (RP) gene pairs and the RP gene regulators, CRF1 and SFP1, vary between species, with no correlation between absence/presence of regulator genes and number of ohnologous RP gene pairs (Figure 4, Additional files 3 and 11).

Central carbon metabolism again provides examples of gene gain/loss events; all Nakaseomyces genomes contain only two copies of ADH genes, which, both by similarity and by conservation of synteny, correspond to orthologs of ADH1 and ADH3, the S. cerevisiae genes that encode, respectively, the cytoplasmic and the mitochondrial activities converting acetaldehyde into ethanol. ADH2, which in S. cerevisiae is specialized in the conversion of ethanol to acetaldehyde, has no ortholog in the Nakaseomyces. It is possible that bi-directional activities exist, or that alternative enzymes take over this conversion (i.e. co-option). For example, S. cerevisiae has additional alcohol dehydrogenase genes, in particular the family of aryl alcohol dehydrogenases encoded by seven subtelomeric AAD genes and the non-subtelomeric YPL088W gene 41 . C. glabrata possesses an array of three such genes in tandem, while other Nakaseomyces have several dispersed copies, except C. castellii which harbors a single such gene. Experimental analysis is needed to tell which enzymes catalyze which reactions in Nakaseomyces, and even in S. cerevisiae, in which enzymatic activities are still in the process of being characterized 42 . As for anaerobic growth, all Nakaseomyces have the ability to grow in micro-aerobiosis, as tested by standard laboratory methods (not shown). Two pairs of regulators are described as essential to anaerobiosis in S. cerevisiae; the ECM22/UPC2 pair and the SUT1/SUT2 pair. These genes are conserved in all Nakaseomyces except C. castellii. There are also pairs of genes that differ by their expression under aerobic and anaerobic growth, such as COX5A/COX5B and of CYC1/CYC7 in S. cerevisiae. In contrast, all Nakaseomyces retain a single member of each of these ohnologous pairs, raising the question of their regulation.

In C. glabrata, the EPA genes, a family of glycosyl-phosphatydylinositol (GPI)-anchored cell-wall protein (CWP) genes 43 , are the best characterized genes involved in adhesion to human epithelia 44 45 , an ability associated to virulence in diverse pathogens 46 . Notably, our search for homologs of C. glabrata EPA genes in the newly sequenced Nakaseomyces (see methods), revealed higher numbers of such genes in the three pathogenic species. More specifically, in the most prevalent pathogen, C. glabrata, the type strain harbors 18 members of this family. Seven new variants of EPA genes are found in a different C. glabrata strain, BG2 (data kindly provided by Brendan Cormack). The other two pathogenic Nakaseomyces, C. bracarensis and C. nivariensis have respectively, 12 and 9 members of the EPA family. In contrast, the non-pathogenic N. delphensis harbors a single copy. These differences cannot be attributed to differences in the quality of assemblies, which were similar in all newly sequenced species. Of the two remaining Nakaseomyces species, only C. castellii contains three homologs of the EPA genes, while N. bacillisporus presented only one distant homolog that clustered with PWP (PA-14 containing Wall Protein, adhesin gene) and adhesin-like protein genes in C. glabrata. These proteins, more similar to the floculin homologs in S. cerevisiae, are only distantly related to Epa proteins 47 48 . A closer inspection of the corresponding trees in the phylome and of a composite tree constructed with all members of the identified EPA-like members (Figure 5a), revealed that a significant fraction of the adhesin gene copies in C. glabrata emerged from lineage-specific duplications in this species, whereas many duplications in C. bracarensis and C. nivariensis are shared. This independent expansion of EPA-like genes in C. glabrata and the emerging pathogens supports the idea of an independent emergence of pathogenesis and may explain the important differences in prevalence across these species. Strikingly, the non-pathogenic member of the ‘glabrata group’, N. delphensis possesses a sole representative of this family, although the phylogenetic scenario implies that the common ancestor of this species and the two emergent pathogens would have had a higher number of copies. This scenario implies specific losses of EPA-like genes in N. delphensis, and a possible secondary loss of adhesion capabilities in this species. Interestingly, our search for adhesin-like proteins uncovered the presence of a group of four proteins in N. delphensis with poor similarity with the EPA-like family (Figure 5c). Whether this group represents highly divergent members -or even pseudogenes- of the EPA-like family in N. delphensis, remains to be further investigated. In any case, their high levels of sequence divergence, particularly at the N-terminal region known to harbor the ligand-binding domain in Epa 49 , suggest they cannot be functionally equivalent.

Other genes shown to be involved in virulence in C. albicans and/or C.glabrata, such as the phospholipase B gene, or the Super Oxide Dismutase genes (SOD genes), exhibit variable presence/absence in the Nakaseomyces, with no obvious correlation to pathogenicity of the species.

Tandem arrays are a special genomic arrangement of certain gene families that imply specific amplification mechanisms and positive selection. They can either be shared by all members of a species, such as the globin genes in mammals 52 , or display polymorphism in terms of number of copies, such as the CUP1 gene in S.cerevisiae, whose amplification is positively selected in copper-containing medium 53 . A few dozen tandems were identified in the Nakaseomyces genomes (HD, TG and CF, ms in preparation). The species with the largest tandem arrays is C. glabrata, with two arrays of eight genes each (Figure 6), the MNT3 array (shown to be variable in different strains, 54 ) and the YPS array 55 , and an array of five genes, coding for a protein of unknown function predicted to be involved in carbohydrate metabolism. None correspond to tandem arrays of more than two genes in other Nakaseomyces. Two other arrays of three genes are specific to C. glabrata, the aryl alcohol dehydrogenase gene array, and the PMU1 array. These findings indicate that most tandem arrays found in C. glabrata originated specifically in this lineage. This, together with the finding that some of these arrays are variable across strains, suggests that the amplification of these families may be driven by (directional) selection. Functions encoded by these families, are thus good candidates for finding possible physiological advantages underlying the success of C. glabrata as an opportunistic pathogen. In the case of the YPS cluster, the genes have been shown to be involved in virulence 55 . PMU1 encodes the phosphate-starvation inducible phosphatase activity, which has been hypothesized to be a specific adaptation of C. glabrata to its mammalian host 36 .

As mentioned above, the ‘glabrata group’ contains several species with the ability to infect humans. We have previously discussed changes that occurred within this lineage in terms of acquisition and loss of genes. We next set out to investigate whether we could identify signatures of possible positive selection in the form of genes with accelerated rates in the branch preceding the diversification of the ‘glabrata group’. For this we focused on 2,153 genes predicted as one-to-one orthologs shared by all Nakaseomyces species and S. cerevisiae. Using the species’ phylogeny, we used a likelihood ratio test (LRT) to compare two nested rate models (Additional file 12). For 991 of the 2153 genes (43%), a model including a different rate in the branch leading to the ‘glabrata group’ was favored, and in 35 of them there was evidence for accelerated rates, suggestive of positive selection (i.e. d _N /d _S rate >1, with an average of 3). In contrast, we did not find a significant rate acceleration in either the S. cerevisiae, N. bacillisporus or C. castelli lineages, nor within the ‘glabrata group’ itself, where the average d _N /d _S rate was 0.04 indicating high levels of purifying selection. These findings suggest that prior to the diversification of the ‘glabrata group’, there was an increase in the non-synonymous substitution rate in at least 35 genes (1.6% of the tested genes). Among these we did not find families that have been related to pathogenesis in C. albicans. Finally, we focused on the ‘glabrata group’ to identify genes accelerated in parallel in the three lineages leading to the three pathogenic species or exclusively in the C. glabrata lineage. We identified 94 proteins with evidence for different selective constrains in pathogenic species and N. delphensis, although this difference was mostly due to stronger purifying selection, rather than positive selection, in the pathogens. With respect to the C. glabrata specific branch, only four genes presented d _N /d _S >1. Overall, these results would suggest a burst of nonsynonymous substitution rates in a significant number of genes preceding the divergence of the ‘glabrata group’ and a subsequent stasis within the group itself. Nevertheless, genes that are positively selected in these lineages constitute good candidates for testing potential roles in virulence.

All strains are the type strains of the corresponding species 4 5 8 10 62 63 , obtained from the CBS collection. All media used were prepared as for S. cerevisiae: glucose and glycerol-based complete media, broth and solid. Cultures were performed at 30°C or 37°C, with agitation for broth cultures.

Anaerobic growth was tested by inoculating Sabouraud plates and using the Oxoid™ Anaerogen system 64 . Plates were examined after 120 h of incubation.

Petite mutants were obtained by exposing cells to Ethidium Bromide 65 . Petite mutants completely lacking mitochondrial (mt) DNA (ρ⁰ mutants) were used for the flow cytometry experiments so that mt DNA did not interfere with measures, since mt DNA content can be quite high in these species.

For the same reason, DNA for sequencing was prepared by standard zymolyase extraction followed by separation on a CsCl gradient with bis-benzimide 66 . The upper mt DNA band was discarded and the lower band used for sequencing. This procedure allowed the mt DNA sequence to be acquired nonetheless at an acceptable fold-coverage.

An aliquot of 4 mL from a fresh yeast culture at 1–2 10⁶ cells/mL is mixed with 9.2 mL of pure ethanol and incubated at 4C overnight. After centrifugation, cells are washed in 50 mM, pH7 Sodium Citrate and resuspended at 10⁸ cells/mL. An aliquot of 200 μL is treated with RNAse by adding 2 μL of a 100 mg/mL solution and incubating 2 hrs at 37°C. Half is then labeled by adding 400 μL of Propidium Iodide at 50 μg/mL in 50 mM Sodium Citrate, and incubating 20mn in the dark. The sample is then ready for the flow cytometer.

Sequencing was done by the Genoscope (Evry, France). Briefly, a whole genome shotgun (WGS) strategy associating different types of sequencing technologies was performed for each strain. An mean 24.8 genome equivalent (from 15.4 up to 41) was achieved using 454 GSFlx approach using a mixture of 8 kb mate pairs sequencing and single reads. Genome assembly was performed using 454 Newbler software. Subsequently, a mean 70 genome equivalent coverage for each strain was obtained using Illumina GAIIx technology 36 or 76 bp single reads, and these data were used to correct assembly errors 67

Probably because of the complete sequence identity between segments of the three MAT-like cassettes, most of these loci were absent from the automated assembly generated from the Illumina and 454 reads. Only three loci were present in the assemblies: one HMRa from C. bracarensis and two HMRa from C. nivariensis. We therefore searched for sequence gaps in regions of synteny with the cassettes from C. glabrata. In all cases except one, there was indeed a gap in the region where the cassette was expected to be, by synteny conservation. The HMR cassette from N. bacillisporus is still missing from the assembly. Fragments were amplified by PCR and sequenced by Sanger sequencing.

Annotation of protein-coding genes was performed with an in-house procedure, which consists of two phases: syntactical annotation (prediction and location of protein coding genes), followed by functional annotation of each element based on comparison with known sequences. The first phase calls upon 6 gene prediction algorithms: CONRAD 68 , AUGUSTUS 69 , GETORF 70 , SNAP 71 , GENEMARK 72 , GENEID 73 , using the same training set of gene sequences, which contains genes with and without introns, for those which needed a training step; the intron-containing genes being defined by comparison to intron-bearing genes of C. glabrata and S. cerevisiae. All predictions as well as tBLASTn alignments to proteomes of reference species and Uniprot, and PSI-tBLASTn alignments to PSSM representative of Génolevures protein families are integrated using JIGSAW 74 . Then, all gene models from the 6 prediction algorithms plus JIGSAW are put together and filtered to eliminate gene models having unrealistic introns. The overlap conflicts between elements and validation of gene models are solved by taking into account predicted gene models, other chromosomal elements already validated, and similarity regions, strands and frames. The resulting gene models are then submitted to functional annotation, based on a decision tree inspired by previous semi-automated annotation projects held by the Génolevures Consortium (which used BLASTp alignments to proteomes of reference species and Uniprot).

Identification of centromeres was done by searching with fuzznuc from the Emboss suite 70 , for sequences homologous to S. cerevisiae consensus centromere sequence, ie “[AG]TCA[TC][AG]TG[AC][TC]N(73,167)G[GT]N(7,15)TTCCGAA” 75 , and by manually checking for conservation of synteny in case of multiple hits within a single scaffold. Telomeric repeats were searched for by using the repeat motif from C. glabrata, CAGCACCCAGACCCCA, as blastn queries against the genomic sequences. This also revealed the putative template inside the TLC1 gene.

tRNA genes were identified by both cloverleaf structure detection 76 and tRNAscan-SE 77 . BLAST 78 and Infernal 79 searches were performed on each of the five Nakaseomyces genomes for other ncRNA genes; using annotated ncRNAs from the Genolevures database 80 as queries for BLAST, and the covariance models from RFam database 81 for the Infernal search. The hits of both searches were combined according to the respective ncRNA family and extended in order to detect contiguous hits corresponding to a same candidate. All hits from the same families were aligned and manually checked. Hits were accepted as candidates if: i) the sequence agrees with known structural features, guiding sequences (for snoRNAs) and conserved sequence motifs for homologous molecules and ii) known synteny was verified.

A phylome, the complete collection of phylogenetic trees for each gene in a genome, was reconstructed for each one of the six Nakaseomyces species (five newly sequenced and the reference C. glabrata genome). The phylomes include 16 other species: Saccharomyces cerevisiae, S. mikatae, S. (Naumovia) castellii, S. kluyveri, S. bayanus, Vanderwaltozyma polyspora, Lachancea thermotolerans, Ashbya gossypii, Candida dubliniensis, Kluyveromyces lactis, Candida albicans, Debaryomyces hansenii, Zygosaccharomyces rouxii, Yarrowia lipolytica, Pichia stipitis, Lachancea waltii. Phylomes were reconstructed using the pipeline described in 82 . In brief, for all genes in each Nakaseomyces genome, a Smith-Waterman search 83 was used to retrieve homologs using an e-value cut-off of <10^-5, and considering only sequences that aligned with a continuous region representing more than 50% of the query sequence.

Once the sets of homologous sequences were defined, phylogenetic trees were reconstructed as follows. Selected sequences were aligned using three different programs: MUSCLE v3.7 84 , MAFFT v6.712b 85 , and DIALIGN-TX 86 . Alignments were performed in forward and reverse direction (i.e. using the Head or Tail approach 87 ), and the six resulting alignments were combined using M-COFFEE 88 . The resulting combined alignment was subsequently trimmed with trimAl v1.3 89 using a consistency score cutoff of 0.1667 and a gap score cutoff of 0.9.

The selection of the evolutionary model best fitting each protein alignment was performed as follows: A phylogenetic tree was reconstructed using a Neighbour Joining (NJ) approach as implemented in BioNJ 90 ; The likelihood of this topology was computed, allowing branch-length optimisation, using seven different models (JTT, LG, WAG, Blosum62, MtREV, VT and Dayhoff), as implemented in PhyML v3.0 91 . The two evolutionary models best fitting the data were determined by comparing the likelihood of the used models according to the AIC criterion 92 ; Then, ML trees were derived using these two models. All trees and alignments have been deposited in PhylomeDB 82 and can be browsed on-line (http://www.phylomedb.org, phylome codes 78 to 83). Trees were scanned to i) define orthology and paralogy relationships using a phylogeny-based, species overlap approach 28 ; ii) detect and date duplication events 29 , including large expansions of gene families; and iii) transfer functional annotations from one-to-one orthologs in S. cerevisiae. Unless indicated otherwise, all operations with phylogenetic trees were performed using scripts implemented within the ETE package 50 .

To reconstruct a species phylogeny, alignments of 603 proteins that had a single ortholog in all species considered in the phylome were concatenated into a single trimmed alignment of 288,995 positions. A Maximum Likelihood tree was reconstructed using phyML using the same parameters indicated for the trees in the phylome and the LG model. Branch support values were computed using the aLRT approach (see above) and based on an analysis of 100 bootstrap repetitions. In addition, a super-tree was reconstructed from the 4,965 gene phylogenies contained in the N. delphensis phylome, using a tree parsimony approach as implemented in DupTree 20 . Both approaches yielded identical topologies.

Average levels of sequence identity between orthologs were computed as follows: Each orthologous pair was aligned with MUSCLE v3.7 and the level of sequence identity was measured with trimAl V1.3 89 as the number of identical residues over the length of the shortest protein.

In order to investigate how selective pressure varied along specific lineages in the phylogeny and whether positive selection was involved in the evolution and diversification of the Nakaseomyces, we used a subset of the original data. The subset analyzed included 7 species: the 6 Nakaseomyces and S. cerevisiae as outgroup. Following the same automated pipeline previously described to construct the Nakaseomyces phylome, we retrieved all the shared orthologous genes present in a single copy in all 7 genomes (i.e., one-to-one orthologs), we concatenated their respective alignments and estimated a phylogenetic tree using maximum likelihood. The resulting topology is consistent with that of the species tree in Figure 1. Using this phylogeny, we tested whether the rate of evolution of the one-to-one orthologs had accelerated along specific branches of the tree (i.e., affecting different species in the group), which would be consistent with either the relaxation of selective constraints or with the action of positive selection. We used the program codeml in the paml 4 package 93 to estimate the d _N /d _S rate ratio variation in each individual one-to-one ortholog, along particular branches in the tree. Subsequently, we compared pairs of nested models by means of a likelihood ratio test (LRT) where the degrees of freedom correspond to the difference in the number of parameters estimated in each model, and the distribution of the d _N /d _S rate is assumed to follow a chi square distribution. Values of d_N/d_S larger than 5 were filtered out.

We hypothesized an acceleration in gene evolution rate, and possible cases of positive selection, preceding the diversification of the ‘glabrata group’, that were perhaps involved in the capability of these species to become pathogenic. In LRT A (Additional file 12) we compare: i) A model that assumes an overall d _N /d _S rate ratio for all branches in the tree (omega 1) and a different rate for the branch that is ancestral to the ‘glabrata group’ (omega 2), where we hypothesize a rate acceleration; and ii) an alternative model that assumes one d _N /d _S ratio for the basal branches including S. cerevisiae, C. castellii and N. bacillisporus (omega 1), a different d _N /d _S rate in the branch ancestral to the ‘glabrata group’ (omega 2), and another d _N /d _S rate for the genus itself (omega 3). In this test, we were interested in verifying whether there was a significant increase in the d _N /d _S rate in the branch ancestral to the ‘glabrata group’ relative to the overall rate and whether this increase also occurred in the different species within the group.

To investigate whether pathogenic and non-pathogenic species were subjected to different selective pressure, we built two more LRTs to analyse another subset of species focusing on the ‘glabrata group’, we therefore excluded S. cerevisiae, and C. castellii, and used N. bacillisporus as the outgroup (Additional file 12). In LRT B we compared i) a model with two rates, one for the ‘glabrata group’, and one for the outgroup (N. bacillisporus); with ii) a model with four rates, one for the pathogenic species, one for the single non-pathogenic species in the subset (N. delphensis), one for the branch ancestral to the ‘glabrata group’, and one for the outgroup. In LRT C we compared i) a model with two rates, one for the ‘glabrata group’, and one for the outgroup (N. bacillisporus); with ii) a model with four rates, one for the the ‘glabrata group’, one for the branch ancestral to the ‘glabrata group’, one for the C. glabrata species itself and one for the outgroup.

Putative adhesins were detected in the newly sequenced Nakaseomyces using a similarity search based on the known EPA genes in C. glabrata and the FLO genes in S. cerevisiae. Hits were filtered using the same thresholds applied during phylome reconstruction. Additionally, all proteomes were scanned for the presence of the Pfam domain PF10528, which is related to adhesins in fungi. The search was performed using HMMER v3, and proteins containing this domain with an e-value below 1e-05 were added to the selection of adhesins. Proteins were then scanned for undetermined regions as their sub-telomeric location can cause problems in the assembly. Proteins containing more than 33% of undetermined regions were excluded from further analysis (this affected 8 proteins, of which 1 in C. castellii, 2 in N. delphensis, 4 in C. bracarensis, and 1 in C. nivariensis). Putative adhesins were then clustered using the TribeMCL 94 algorithm as implemented in scps (inflation = 1.5) 95 . Clusters were then used to infer phylogenetic trees. First the repetitive regions of each sequence were masked using SEG 96 . Alignments were then reconstructed using MUSCLE v3.7 84 followed by a maximum likelihood tree reconstruction as implemented in PhyML. The LG model was used along with four rate categories and invariant positions infered from the data.