The A. adeninivorans strain LS3 was selected because of its established biotechnological use 20 . Both mitochondrial and nuclear genomes were sequenced using the Sanger and 454 pyrosequencing approaches with different shotgun, plasmid and BAC libraries (Additional file 1). The circular mapping mitochondrial genome has a final size of 31,662 bp. It encodes 24 tRNA genes, 15 protein-coding genes including the seven NADH: ubiquinone dehydrogenase subunits of complex I, the genes encoding the RNA component of RNase P and the two subunits of the mitochondrial ribosomal RNA, as expected from the phylogenetic position of this species. All of these genes are transcribed from the same DNA strand except for the tRNA-Cys gene (Additional file 2).

After directed finishing phases, the 11.8 Mb final assembly of the nuclear genome resulted in four contigs corresponding to the four chromosomes Arad1A, Arad1B, Arad1C and Arad1D, of 1,659,397, 2,016,785, 3,827,910, and 4,300,524 nt, respectively, as predicted from previous pulsed-field gel electrophoresis analyses 21 (Figure 1). K. pastoris and Y. lipolytica have four and six chromosomes respectively, while an average of eight and sixteen chromosomes is observed in protoploid and post-whole genome duplication species 22 . This may suggest a whole genome duplication event during early hemiascomycete evolution although there is presently no other evidence to support this hypothesis 23 . The four contigs contain no internal gaps and lack only terminal repeats at the telomeres. There is a single rDNA cluster, located approximately 75 kb upstream of the chromosome D right subtelomere. Based on 454 read counts, there are about 35 to 40 tandem repeats at this locus of the 18S, 5.8S and 26S rRNA genes, the latter housing a 411-bp group-IC, self-splicing intron 24 . We have left two copies of the rDNA units in Arad1D flanked by two artificial gaps of 874 “N”. As in Y. lipolytica, the 5S rRNA genes are not included in the rDNA repeat, but occur as 30 copies dispersed throughout the genome (Table 1). A set of 46 tDNAs encoded by 147 tRNA genes was identified and confirmed that A. adeninivorans follows the regular eukaryotic-type sparing rules to read CTY Leu and CGA Arg codons 25 . Forty seven genes encoding snRNAs or snoRNAs were identified: the small nuclear RNAs (U1, U2, U4, U5 and U6), the RNA components of the RNase P, and of the signal recognition particle, as well as 14 H/ACA and 33 C/D snoRNAs (Additional file 3). Additionally, three thiamine pyrophosphate (TPP) riboswitch sequence candidates were found in the 5′ region from homologs of S. cerevisiae THI4 (YGR144W), UGA4 (YDL210W) and DUR3 (YHL016C), namely ARAD1R43560g, ARAD1D08074g and ARAD1B12386g; they show a remarkable conservation of known structural domains and sequence motifs 26 . A single transposable element was identified on chromosome B (Taa3, ARAD1B13860t) that belongs to the Gypsy superfamily of Long Terminal Repeat (LTR) retrotransposons with the two gag and pol open reading frames separated by a minus 1 frameshift as seen in the homologous element Tyl6 of Y. lipolytica 27 . The single copy of Taa3 was found 13 bp upstream of a tRNA gene, suggesting a possible specificity of insertion 28 . Only three relics of solo LTRs were identified in the genome, which implies that Taa3 has low activity.

CDS, coding DNA sequence; G, guanine, C, cytosine; ncRNA, non coding RNA.

Putative centromeres were identified within one region per chromosome with a conspicuous G + C (Guanine + cytosine) bias defining approximately 6 kb G + C troughs, with a G + C content of 31 to 33% as compared to 48% for the whole genome. Like those of Y. lipolytica, they share features of both regional centromeres found in yeasts of the CTG group, and of point centromeres characteristic of Saccharomycetaceae 29 (Additional file 4).

A total of 6,116 protein-coding genes and 33 pseudogenes were identified. This is slightly less than reported for Y. lipolytica or Debaryomyces hansenii, but significantly more than for the Saccharomycetaceae species (Table 2). The frequency of pseudogenes is one of the lowest reported in hemiascomycetes, while gene density is one of the highest.

snRNA, small nuclear ribonucleic acid; snoRNA, small nucleolar ribonucleic acid; CDS, coding DNA sequence; G + C, guanine and cytosine; aa, amino acids; i-genes, intron-containing genes; ND, not-determined. Data from S. cerevisiae, Lachancea thermotolerans, D. hansenii and Y. lipolytica were taken from 30 ; data on intron-containing genes from 31 .

A total of 4,815 (78.7%) genes were assigned to gene ontology (GO) terms: 3,853 genes to molecular functions, 2,626 to cellular components and 3,308 to biological processes. In the biological processes, the largest fraction of genes, 1,351 (22.1%), was assigned to metabolism, while in the molecular functions the largest category was represented by genes encoding catalytic activities. The GO slim categories of A. adeninivorans are presented in Additional file 5. InterPro domains were detected in 5,147 (84.2%) proteins corresponding to 459 distinct Pfam domains. A secretion signal peptide of type I or type II was predicted in 957 (15.6%) gene products, including Atan1p and Alip1p that were previously characterized experimentally by N-terminal sequencing and mass spectrometry (MS) analysis 32 33 . Transmembrane helices were found in 1,271 (20.8%) proteins. An Enzyme Commission (EC) number was assigned to 676 (11.1%) genes. We assigned 884 (14.4%) genes to 98 metabolic pathways present at the Kyoto Encyclopedia of Genes and Genomes (KEGG) with the highest number of genes related to purine metabolism. Blast2GO BLASTx alignments using the NCBI NRPEP database confirmed that the closest matches to A. adeninivorans genes were very often found in Y. lipolytica (Additional file 5).

Spliceosomal introns are more frequent than in Saccharomycetaceae or in Debaryomycetaceae, but in the same range as reported for Y. lipolytica (914 versus 1119, Table 2). In this latter species, introns are characterized by a very short distance between the 3′ splicing site and the branch point, but have in contrast retained the ancestral consensus hemiascomycete 5′ splicing site (GTAAGT). Finally, multi-intronic genes tend to be more frequent in A. adeninivorans than in Y. lipolytica (21.5% vs. 11.5%). For additional information, see Additional file 6 and Genosplicing 31 .

A phylogenetic tree was reconstructed for each A. adeninivorans protein-coding gene, the so-called phylome, and used to identify orthology and paralogy relationships among related species 34 . This comprehensive collection of evolutionary histories is publicly available at PhylomeDB 35 . Species phylogenies were computed on a set of concatenated orthologs and using a super tree approach combining all individual gene phylogenies. The two methods gave the same topology, in which A. adeninivorans groups with Y. lipolytica (Additional file 7), although the two species have greatly diverged. For instance, our analyses identified 2,520 A. adeninivorans proteins that lack an ortholog in Y. lipolytica, 591 of which do not even have a homolog in that species. For 121 proteins we could only detect homologs in Pezizomycotina genomes (Additional file 8). Horizontal gene transfer between prokaryotes and fungi was detected using a published pipeline 36 , which pinpointed six candidates with putative enzymatic function that are likely to have been transferred from prokaryotes to Arxula (Additional file 8). Few genes of bacterial origin have been reported in Saccharomycotina so far, but most of them encode metabolic enzymes with important physiological roles that may facilitate host adaptation to biotope variations (see 36 37 for large-scale trans-kingdom transfer in fungi).

The number of conserved gene blocks between A. adeninivorans and other genomes ranged from 300 with S. cerevisiae to >800 with Y. lipolytica, and was roughly proportional to the mean percentage of protein similarity, as is expected when species have greatly diverged. Indeed, in the comparison between Y. lipolytica and A. adeninivorans, 92% of the blocks contained less than four genes, showing that there is no large-scale conservation of synteny (Additional file 7).

The gene trees in the phylome were scanned to detect and date duplication events 38 . With an average of 0.253 duplications per gene in the specific lineage leading to Arxula, this genome does not seem to contain a large amount of duplications. This is nevertheless greater than the 0.015 value found in the common ancestor of Y. lipolytica and A. adeninivorans (Additional file 9). Most Arxula-specific expansions are not very large (between three and nine sequences) and correspond to peptidases, transporters, dehydrogenases and some proteins related to nitrogen metabolism (Additional file 9). One expansion, however, contains over 100 members of unknown function and no homologs in any database, which is to our knowledge the largest gene family described in yeast (Figure 2 and Additional file 9).

There are fewer transporters in A. adeninivorans than in for example D. hansenii or Y. lipolytica, but some have undergone strong amplification. Remarkably, sugar transporters appear overrepresented in this species: there are 60 members of the Sugar Porter family, which is three times as many as in Kluyveromyces lactis or K. pastoris, and 1.8 times more than in S. cerevisiae (Additional file 10). These include 15 glycerol: H⁺ symporters, paralogs of the S. cerevisiae singleton STL1, compared to eight in the osmotolerant yeast D. hansenii, which may reflect the salt tolerance of A. adeninivorans. The ability to use various carbon sources is highlighted by the abundance of high affinity glucose: H⁺ symporters (10 members), maltose: H⁺ symporters (10 members), lactose permeases (four members versus one in K. lactis and D. hansenii), allantoate permeases (six members), and of facilitators for the uptake of xylose (six members), quinate (four members), fructose (four members) and myo-inositol (three members). Surprisingly, there are few glucose uniporters (two members, versus eighteen and four in S. cerevisiae and D. hansenii, respectively) and few sugar sensors. High-affinity nicotinic acid transporters (six members), polyamine transporters (15 members) and nitrate/nitrite permeases (three members) are also amplified (Additional file 10).

About 10% of the duplicated genes (213/2285) are organized in tandem gene arrays (TGAs), mostly as arrays of two genes. These arrays are sometimes entirely duplicated on the same or on different chromosome(s), a situation that so far remains unusual. The mechanism involved has given rise to the largest protein family in yeasts as mentioned above. BLASTn searches indicated that coding and intergenic regions of duplicated TGAs are highly conserved at the nucleotide level, suggesting propagation of ancestral tandems by segmental duplication at ectopic positions (Figure 2 and Additional file 11).

A. adeninivorans LS3 is only known to reproduce asexually 20 , yet a MAT locus was identified on chromosome D as is the case in many asexual species 40 . The region around the mating type locus is conserved between Y. lipolytica and A. adeninivorans, while it is rearranged in basal species such as Lipomyces starkeyi, filamentous fungi, and in species that emerged later, such as K. pastoris or K. lactis (Figure 3). The MAT locus encodes a homolog of the transcriptional factor Matα1 present in other yeast species (ARAD1D19294g, MTAL1), with a canonical DNA binding domain and a C-terminal extension partially conserved in Y. lipolytica, but absent from other species (Additional file 12). There is no Matα2 coding sequence (MTAL2) contrary to the situation reported in other heterothallic yeast species such as S. cerevisiae and Y. lipolytica. The presence of only MTAL1 at the MATalpha locus is, however, found in several sexually competent filamentous fungi and yeasts such as Aspergillus nidulans, Clavispora lusitaniae, Meyerozyma guillermondii, Scheffersomyces stipitis or D. hansenii 40 . Whether A. adeninivorans is asexual or not is still an open question. Either A. adeninivorans is truly asexual and the loss of MTAL2 may be the cause, or alternatively, A. adeninivorans is sexual but strains of the opposite mating type have not yet been identified, thus preventing successful mating.

A search for genes involved in mating, meiosis and sporulation in S. cerevisiae identified the presence of most genes conserved in the sexual species D. hansenii, K. pastoris and Y. lipolytica (Additional file 13). For example, out of 368 genes tested, 292 were conserved in Y. lipolytica and 288 in A. adeninivorans. Candidates for the mating pheromones MFa and MFα and of their cognate receptors as well as for the signaling cascade were identified, confirming that A. adeninivorans is either still sexually active or has lost this ability only recently (Additional file 14).

A. adeninivorans is described as having a wide substrate spectrum that includes the assimilation of many nitrogenous and aromatic compounds such as nitrate and nitrite, purines, tannins and benzoic acid derivatives 13 41 42 . The ability to degrade purine compounds is reported in all kingdoms and can occur either aerobically or anaerobically in separate pathways. In the aerobic pathway, the critical step in the degradation of purine bases is the oxidation of hypoxanthine and xanthine to uric acid, catalyzed by xanthine oxidase and/or dehydrogenase. The various purine-degradative pathways are unique and differ from other metabolic pathways because they may serve quite different purposes, depending on the organism or tissue. While some organisms degrade the naturally occurring purines to CO₂ and ammonia, others contain only some of the steps of the purine degradation pathways, resulting in partial degradation of purines or certain intermediary catabolites 43 .

Purine catabolism is a characteristic feature of A. adeninivorans 13 . The purine nucleosides (adenosine, inosine, xanthosine and guanosine) are transported across the membrane and into the cytoplasm by a purine permease. They are then converted to adenine, hypoxanthine, xanthine and guanine, further degraded to uric acid and, after transport into the peroxisomes, to urea. All corresponding genes of this pathway are localized on different chromosomes and are induced by adenine and other pathway intermediates 4 44 . Interestingly, an adenosine deaminase, needed to transform adenosine to inosine in animals and human, is absent (Figure 4). This pathway allows A. adeninivorans to use all of these purine derivatives as nitrogen and carbon sources 4 44 .

As in Ogatea (Hansenula) polymorpha 45 and in Kuraishia capsulata 18 , a cluster of genes encoding a nitrate transporter, a nitrate reductase and a nitrite reductase has been previously identified in A. adeninivorans 46 . Genome data indicate that nitrate transporter encoding genes form a three member family, two of which are part of the nitrate cluster.

Microarrays were designed based on the complete genome data of A. adeninivorans to analyze gene expression changes before and after a shift from yeast minimal medium (YMM) + 2% glucose with NaNO₃ to YMM medium with adenine as the nitrogen source. A significant down regulation of the genes involved in nitrate metabolism was observed two hours after the shift. Key components of the purine degradation pathway on the other hand, clearly showed an increased activity (Figure 4). This provides further insight into the regulation of purine degradation by A. adeninivorans and emphasizes the possibility of using transcriptomic approaches to identify candidate genes for new biotechnological applications. Arxula specificities of the purine degradation pathway include the regulation of the respective genes. Activity tests, qRT-PCR experiments and microarray assays with xanthine dehydrogenase inducers demonstrated strong gene inducibility when cells were cultured on hypoxanthine and adenine and a lower level of induction with uric acid as the sole nitrogen source. However, enzyme induction by purines stops after supplementing the medium with NH₄ ⁺ or NO₃ ^- as nitrogen sources, which is in contrast to the situation in N. crassa where the enzyme is induced in the presence of NO₃ ^-, but not with NH₄ ⁺. It is known that in A. nidulans, NH₄ ⁺ inactivates the GATA factor AreA, which is responsible for expression of the urate-xanthine transporter 47 . It is not clear which mechanism triggers the repression with NH₄ ⁺ and NO₃ ^- in A. adeninivorans 4 .

Tannin, a plant polyphenol molecule, is widely distributed in the plant kingdom where it protects plants against attack by parasites and herbivores. It inhibits the activity of enzymes by binding and precipitation and is to a greater or lesser extent recalcitrant to biodegradation 48 . While tannins are growth inhibitors for most microorganisms, a few bacteria, fungi and yeast such as D. hansenii, Mycotorula japonica or Candida sp. are capable of exploiting tannins as a carbon and/or energy source for growth 49 50 51 . A. adeninivorans is one of these yeasts that use tannic acid and gallic acid as carbon sources 52 . Genes encoding tannases (ATAN1 - ARAD1A06094g, ATAN2 - ARAD1A19822g), gallate decarboxylase (AGDC - ARAD1C45804g) and catechol 1,2 dioxygenase (ACDO - ARAD1D18458g) have been identified and His-tagged recombinant enzymes and corresponding gene mutants were used to confirm the activity of these enzymes (data not shown). This demonstrated that the tannic acid catabolism pathway enables this yeast to assimilate tannic acid and other hydroxylated derivatives of benzoic acid by non-oxidative decarboxylation. All suitable derivatives require an hydroxide group at the m or p position of the carboxylic acid (Additional file 15). Interestingly, A. adeninivorans is thus the first eukaryote known to synthesize two tannases, one extracellular (Atan1p) 32 and one cell-wall localized (Atan2p - data not shown) which permits effective degradation of extracellular tannic acid. Both enzymes are able to remove gallic acid from both condensed and hydrolysable tannins. Substrate specificity, biochemical parameters (temperature optimum 35 to 40°C, pH optimum at ca. 6.0) and nearly complete extracellular localization (≥97%) distinguish Atan1p as an important industrial enzyme. First, transgenic tannase producer strains were constructed with a constitutively expressed ATAN1 module integrated into a chromosome. In fed-batch fermentation experiments, the transgenic strain produced 51,900 U/L of tannase activity after 42 h with a dry cell weight of 162 g/L 1 .

Another uncommon substrate used by this yeast is n-butanol. The n-butanol degradation pathway has not previously been reported to exist in eukaryotes. Genome mining suggests that n-butanol is oxidized to butyraldehyde by an alcohol dehydrogenase (Aadh1p, AADH1 - ARAD1B16786g) that has a high substrate specificity, and then to butyric acid by two aldehyde dehydrogenases (Aald2p, AALD2 - ARAD1B17094g; Aald5p, AALD5 - ARAD1C17776g). The last steps involve an acyl-CoA ligase, a cytoplasmic acyl-CoA carnitine acyltransferase and a peroxisomal acyl-CoA carnitine acyltransferase for butyryl-carnitine synthesis via a butyryl-CoA intermediate that is transported from the cytoplasm to peroxisomes or mitochondria for ß-oxidation. A special feature of this pathway is that the synthesis of butyryl-CoA from butyric aldehyde is a one-way reaction since the aldehyde dehydrogenase and acyl-CoA ligase steps are not reversible (Figure 5).

The genome of A. adeninivorans LS3 was sequenced independently by the Genoscope (Evry, France) using the capillary Sanger technology and by IPK (Gatersleben, Germany) using the 454 Roche methodology (GS-FLX Titanium version).

For the Sanger technology, a shotgun sequencing strategy using three different clone libraries and capillary Sanger sequencing was used to obtain a 12× coverage of the complete genome. For two of three libraries, genomic DNA was fragmented by mechanical shearing and 3 kb (A) or 10 kb (B) inserts were respectively cloned into pcdna2.1 (Invitrogen, Saint Aubin, France) and pCNS (pSU18 derived) plasmid vectors. In addition, a large insert (25 kb) BAC library (C) was constructed from Sau3A partial digest and cloned into pBeloBAC11. Vector DNAs were purified and end-sequenced (124,032 reads (A), 61,440 reads (B), 5,376 reads (C)) using dye-terminator chemistry on ABI3730 sequencers. The reads were assembled using the whole genome shotgun assembler ARACHNE and the chromosome sequences were individually reassembled using the Phred/Phrap/Consed software package. For the finishing step, we used primer walking of clones, PCR amplifications and in vitro transposition technology [Template Generation System™ II Kit (Fisher Scientific, Illkirsh, France) or Hypermu < Kan-1 > (Tebu-Bio, Le Perray-en-Yvelines, France), corresponding to 814, 33 and 17,975 reads, respectively. The final assembly consisted of four scaffolds larger than 1 Mb, hereafter referred to as chromosomes, and nine shorter contigs of a size ranging from 4 to 120 kb, including four mitochondrial scaffolds. Four of the remaining contigs were later incorporated at chromosome ends in the final assembly using data obtained from the 454 assembly. The mitochondrial genome sequence was assembled as a circular map using Sanger and 454 contigs and manually validated using single reads obtained with the Sanger technology. A. adeninivorans genome sequence data have been deposited at EMBL under the accession number PRJEB4557 [EMBL:PRJEB4557].

The shotgun library of A. adeninivorans for sequencing on Roche 454 GS FLX Titanium sequencer was prepared using 5 μg of genomic DNA. Based on random cleavage of the genomic DNA with subsequent removal of small fragments with Agentcourt AMPure SPRI beads (Beckman Coulter, Krefeld, Germany), the resulting single stranded DNA (ssDNA) library showed a fragment distribution between 300 and 1000 bp. The optimal amount of ssDNA library input for the emulsion (emPCR) was determined empirically through 4 small-scale titrations with one, two, four and eight copies per bead (cpb). Finally, one cpb was used for the large-scale experiment. One individual emulsion PCR (two cups, one full emPCR-Kit LV (Roche Applied Science, Mannheim, Germany) was carried out to generate 5.7 million DNA-carrying beads for two-region sized 70 × 75 PicoTiterPlates (Roche Applied Science, Mannheim, Germany) and each region was loaded with 2 million DNA-carrying beads. Two read sets were thus generated totaling 1,074,025 reads. This resulted in 542.3 Mb of sequence data (45-fold genome sequence coverage) with an average read length of 505 bp. Assembly was performed using the Newbler software (v2.3) within the Roche 454 suite package, MIRA multi-pass DNA sequence data assembler/mapper (v3.0.2) and CLC Bio assembler. To allow comparisons between the assemblies of different assembly programs, singletons and contigs shorter than 100 bp were discarded before subsequent analysis. Standard metrics describing the assembly, such as the total bases used in a assembled contigs, the amount of contigs longer than 300 bp, 500 bp, 1 kb, 2 kb and 5 kb, number of base pairs in the largest contig and N50 contig length (the smallest contig size in which half the assembly is represented) were used to compare the assembly programs. The highest number of contigs was produced by MIRA but only 161 contigs were longer than one kilobase pair. While Newbler and CLC Bio assemblers constructed longer contigs, however the longest contig was generated by MIRA (Additional file 1).

To assess the quality of the final assembled genome, the 454 reads were mapped onto the chromosomes using the Burrows-Wheeler Alignment tool BWA 53 . Two statistics were extracted from the mappings using Samtools 54 : the percentage of reads that mapped on the assembly and the percentage of reads that mapped to each chromosomes. The quality of the final assembled genome was estimated using the dot-plot analysis which was performed using Nucmer software (NUCleotide MUMmer v3.1 55 ). The dot-plot alignment was generated by comparison of all assembled chromosomes and contigs.

Mapping of 454 reads was used to estimate the gene copy number by computing the number of tags mapping to unique regions of the genome. For this purpose, sequences of 21 A. adeninivorans genes, deposited in GenBank (NCBI), were used in BLASTn searches together with the set of all 454 reads using three BLAST e-value cutoff = e-10, e-50 and e-100 to improve search specificity. The analysis of the gene copy number was performed using the formula: GeneCopyNumber = (Number of BLASTn hits * Average read length)/(Gene length * 454 sequence coverage).

Non-protein coding and protein-coding gene models were predicted according to Louis et al. 56 . All translations of models longer than 80 codons were compared against the proteomes of Y. lipolytica and S. cerevisiae as well as Uniprot-Fungi using BLASTp. In addition, the gene models were compared to position-specific scoring matrix (PSSM) representative of Genolevures protein families 57 with PSI-tBLASTn (Position-Specific Iterated BLAST). Pre-annotated gene models were then examined for validation in the framework of the Génolevures proprietary Magus annotation system by a community of curators, in three phases: (i) curation of models with PSI-tBlastn hits, as possible new members of protein families, for homogeneity of annotation, (ii) curation of other models, (iii) final finishing through contig walk by a single curator in charge. At any phase, curators could add or modify gene models.

Circos 58 was used for illustration of nuclear genome data such as: chromosome structure (position of centromeres, tRNA and rRNA genes), density of genes across the genome, content of repeat DNA, 454 reads mapped to chromosomes, syntenic blocks between A. adeninivorans and genomes of Y. lipolytica, K. pastoris and S. cerevisiae.

Functional annotation of genes according to the GO terms, EC numbers and the KEGG pathway were performed for each A. adeninivorans CDS using the Blast2GO software suite. Protein domains were detected by InterProScan with various databases (BlastProDom, FPrintScan, HMM-PIR, HMM-Pfam, HMM-Smart, HMM-Tigr, PatternScan, SuperFamily, HMM-Panther and Gene3D) through the European Bioinformatics Institute Web Services. Signal peptide and transmembrane helices were predicted by SignalP v.3.0′s neural network and hidden Markov model tools 59 and TMHMM, respectively.

The classification of A. adeninivorans protein sequences into protein families was performed along two procedures. First, protein sequences were tentatively incorporated into protein families defined in the previous round of Génolevures genome annotation using PSI-BLAST with relaxation factors based on family dispersion 30 . Second, the sequences rejected by the procedure were pooled with the sequences of the nine species already present in the Génolevures database which are not members of any protein family and a clustering with OrthoMCL 60 was applied to define new families.

The phylome, a complete collection of phylogenetic trees for each gene in A. adeninivorans, was reconstructed. Seventeen additional species were included in the phylome: three Pezizomycotina species and fourteen Saccharomycotina species. The phylome was reconstructed using a previously described pipeline 35 . Briefly, for each gene encoded in A. adeninivorans, a BLAST search was performed against the proteome database containing the 18 proteomes. Results were filtered according to e-values < 1e-05 and minimal overlaps with hit sequences at 50% of the query length. A maximum of 150 matches were accepted for each A. adeninivorans protein. Multiple-sequence alignments were performed in forward and reverse orders 61 , using three programs: Muscle 39 v3.8.31, MAFFT v6.814b 62 and DIALIGN-TX 63 . The six resulting alignments were then combined using M-COFFEE (T-Coffee v8.80) 64 and trimmed using trimAl 65 v1.3 (consistency cutoff: 0.1667; gap score cutoff: 0.1). Model selection for phylogenetic tree reconstruction was performed by reconstructing neighbour joining trees using BioNJ 66 with different models (JTT, WAG, MtREV, VT, LG, Blosum62, CpREV and DCMut) and then the two best models according to the AIC criterion 67 were chosen. The selected models were used to reconstruct maximum likelihood trees using phyML 68 . In all cases, a discrete gamma-distribution model with four rate categories plus invariant positions was used, the gamma parameter and the fraction of invariant positions were estimated from the data. A total of 4,992 trees were reconstructed. Trees and alignments are stored in the database phylomeDB with the PhyID code 178.

The trees reconstructed in the phylome were analyzed using ETE 69 v2.0. Orthology and paralogy relationships between the sequences were established using the species overlap algorithm from ETE v2.0. The algorithm scans the trees from seed to the root and at each node it establishes a duplication or a speciation node depending on the overlap between the species located at each side of the node. If there are common species, the node is assumed to be a duplication node, otherwise it is considered a speciation node. Once duplications were detected, they were mapped onto the species tree. It was assumed that the duplication occurred at the common ancestor of the species derived from the duplication node. The duplication rate at each node was calculated by dividing the duplications mapped at a given node by the number of trees that have an outgroup to the node. Species-specific expansions were also detected by selecting those duplication nodes that only contained sequences from A. adeninivorans. Groups of expanded proteins that overlapped in more than 20% of their sequences were fused into a single-gene expansion.

Gene transfers from prokaryotes to A. adeninivorans were detected using a previously published pipeline 36 . Briefly, a BLAST search was performed for each protein encoded in A. adeninivorans against a database that contained 102 completely-sequenced fungi (downloaded from their respective databases), 95 other eukaryotes and 1,395 prokaryotes (downloaded from KEGG as of June 2011). Only genes present in more than 30 prokaryotes, less than 10 fungi and no other eukaryotes were considered to be putative transfers.

The species tree was reconstructed by concatenating 253 genes that were found in all the genomes included in the phylome database and that were exclusively one-to-one orthologs. The genes were concatenated and the tree was reconstructed using RaxML 70 . A second tree was reconstructed using a super-tree approach as implemented in Duptree 71 , this algorithm tries to find the species tree that minimizes the number of duplication events that occurred in a set of gene trees. In this case the 4,992 trees reconstructed in the phylome were used.

Based on 6,025 annotated chromosomal sequences and 36 putative mitochondrial genes oligos were designed using Agilent Technologies eArray software (design number 035454). Depending on the sequence length of the genes up to ten 60-mers per gene were created resulting in a total of 56,312 A. adeninivorans specific oligos. The microarray was produced by Agilent Technologies (Böblingen, Germany) in 8×60k format.

Overnight cultures of A. adeninivorans LS3 in YMM with nitrate were shifted to YMM containing 4 mM adenine as the sole nitrogen source and YMM with nitrate as a control, respectively. After 2 h of shaking at 30°C and 180 rpm cells were harvested and total RNA was isolated. Probe labeling and microarray hybridization (duplicates) were executed according to the manufacturer’s instructions (Agilent Technologies “One-Color Microarray-Based Gene Expression Analysis”; v6.5; Böblingen, Germany).

Analysis of microarray data was performed with the R package limma 72 . Raw expression values were background corrected (method “normexp”) and normalized between arrays (method “quantile”). Differentially expressed genes were detected by fitting a linear model to log2-transformed data by an empirical Bayes method 73 . The Bonferroni method was used to correct for multiple testing.

A. adeninivorans genome sequence data have been deposited at EMBL under the accession number PRJEB4557 [EMBL:PRJEB4557]. The raw data of 454 reads have been deposited at EMBL/ENA database under the accession number ERP001774 [EMBL: ERP001774].