Horizontal transfers (HTs) are a major force of evolution in prokaryotes 1 2 3 4 5 . The average amount of DNA transferred in prokaryote genomes varies from 0 to 17% according to different studies 4 6 7 8 . The transferred genes remaining in the genome either increase fitness or allow the colonization of new environments 2 3 9 10 11 . However, the extent of HT in eukaryotes is less known though they were proposed to play a role as important as for prokaryotes 12 13 14 15 16 17 18 19 . In fact, most of the documented cases concern insertions of viruses (especially retroviruses) into eukaryote genomes 20 21 22 23 and exchanges between symbiont, parasite 18 24 or organelle genomes 25 26 and their host genome. At last, as conjugation between distant species is unlikely by meiosis, a possibility of transfer between eukaryotes was evoked by gene introgression following hybridization between closely related species 27 .

For the former examples, the biological mechanisms are understood, demonstrated or are hypotheses with strong support. However the mechanisms involved in DNA exchanges between distant species are mostly unknown, either between eukaryotes or to explain the numerous reports of HTs between prokaryotes and eukaryotes. Among the mechanisms acting in prokaryotes, transformation by free DNA is possible for eukaryotes but is less efficient than it is for prokaryotes 28 . Transduction was hypothesized but its efficiency differs as a function of species families from possible to unlikely by lack of vectors 29 . Also, alternative mechanisms were suggested, like phagocytosis or by means of bacterial type IV secretion systems that could promote the transfer of DNA from prokaryotes to eukaryotes 13 30 . Thus, while HT results are observed, the underlying mechanisms are yet to be discovered.

The HT detection methods generally used in eukaryotes are based on gene homology. The determination method depends on the number of homologs of the target and of its phylogenetic distribution. In the case of the detection by Blast of only a few homologs for a gene of interest, an alignment analysis showing more homology with genes/proteins of distant species than to a closer one indicates a horizontal transfer event for this gene. A typical example of such detection is to find a close prokaryotic homolog to a eukaryotic gene 31 32 33 34 . A more reliable method can be used in the case of numerous homologs and their broad distribution in the evolution tree. In this latter case, a phylogenetic analysis is performed and incongruence in the phylogenetic tree leads to a similar conclusion 35 36 . In each of these cases, the study concerns only a peculiar gene or a small group of genes 37 38 39 40 41 42 . Indeed, due to the restrictions exposed above, a fair number of genes cannot be analyzed this way: ORFans of course but also genes with only one or a small number of homologs leading to an inconclusive situation. Moreover, due to the patchiness of eukaryote sequences in Genbank, it is difficult to assess horizontal transfer between eukaryotes species 43 , while it is easier to assess transfers between prokaryote and eukaryote species 14 . However, many newly sequenced genomes were analyzed for horizontally transferred genes (HGT) and in some cases a large number of HGTs were detected by phylogenetic analyzes (for instance 587 genes - 5.6% of all genes - were found of bacterial origin in the diatom P. tricornutum 44 ). Therefore, while a high number of genes could be found of alien origin, these studies, as discussed above, could not be qualified as whole genome studies.

In order to analyze whole genomes and to cope with the difficulties discussed previously, the so-called parametric methods were designed. They are based either on the whole set of genes of a species or on variations of the composition characteristics of the genomic sequence itself. Methods using gene information are based on differences in codon usage between highly expressed, lowly expressed and alien genes 45 46 47 48 . However, none of the methods based on codon usage can be applied to eukaryote genomes as gene regulation is different from prokaryotes and no tool has been designed to cope with this fact 47 48 .

The other methods are based on the variations of base composition detected by different order Markov models along a genome: the so-called genomic signature 49 50 . This genomic signature was demonstrated to be species-specific and quite similar all along the genome 50 51 52 53 54 . This species-specificity was used to detect horizontally transferred DNA by analyzing a genome and searching for regions exhibiting a different signature than the majority of the genome 4 6 7 8 55 56 57 58 59 60 61 62 63 64 65 . These methods use only the information contained in the analyzed genome and when applied to the whole genome sequence they allow the detection of atypical regions containing no annotated genes 6 61 .

Phylogenetic and parametric methods, while detecting common genes, diverge in certain cases. It was proposed that these two types of methods addressed different types of HGTs 66 67 . It was proposed that combining signature and gene based methods increased either specificity or sensitivity of HT detection 33 58 68 .

In general, when compared to prokaryotic genomes, eukaryote genomes are larger and more complex due to the presence of non-coding sequences, low complexity regions, isochores and fragmented genes. Therefore, most of the parametric methods used for prokaryotes are either inefficient or not suitable to eukaryotic genomes. Likewise, methods based on variations of the G+C composition work poorly due to the intrinsic variations of base composition in eukaryote genomes 69 . For these reasons, no genome-wide study of horizontal transfers in an eukaryotic genome using parametric methods was published. However, some eukaryotic genomes present characteristics close to prokaryotic ones and allow attempting the use of parametric methods on them. For instance, it has been shown that variation of short oligonucleotide usage is moderate in some fungi genomes and that parametric methods based on this type of criterion could be applied to them 50 70 71 . Moreover, HTs seem to play an important role in the evolution of fungi 29 72 73 74 75 . Therefore, we chose to analyze the extent of horizontal transfers in the genome of Aspergillus fumigatus 76 77 78 . A. fumigatus is a pathogenic fungus causing a wide range of diseases including mycotoxicosis, systemic diseases and allergic reactions. The mortality rate is high in infected patients, especially in immuno-compromised ones. Here we propose to use a simple and tested method based on short oligonucleotide usage 6 to evaluate the amount of HTs in the genome of Aspergillus fumigatus.

We found that HTs in fungi are not negligible, accounting for 1 Mb, representing about 3% of the genome and that donor species belong mainly to 3 classes, bacteria, fungi and viruses.

The Aspergillus fumigatus Af293 genome (Genbank NC_007194 - NC_007201) 78 has a size of 29.4 Mb and is composed of 8 chromosomes. Its base composition is balanced: G+C% = 49.8%.

We used a method based on the variations of tetranucleotide frequencies along a sequence. The method was already described and tested on prokaryotic genomes and the principles are recalled hereafter 6 . The specificities of eukaryotic genomes implied a pretreatment and in a first step, we removed from the genome all the centromeric and telomeric low complexity regions which exhibited an atypical signature and did not correspond to transferred DNA. The genome was subsequently analyzed by a 5 kb sliding window, with a step of 500 bp. The signature of each window was calculated and the Euclidian distance of every window signature to the whole genome signature was assessed. Then, the window signatures were clustered by a k-means algorithm and a partition based on the distance distributions per class and the average distances of the classes to the genome was performed. In a previous work with prokaryotic genomes, less than 8 classes (average 4 6 ) were required to take into account the intrinsic genome variation and the atypical signatures. Due to an increased intrinsic variation of base composition in eukaryote genomes, the number of classes was raised to 20 for the A. fumigatus genome. The partition separated the k-means classes into one group exhibiting rather homogeneous signatures whose distance to the whole genome signature was small (90% of the windows) and one group of heterogeneous classes with a large distance to the genome signature (10% of windows). Thus, we considered that the first group of classes represents the host genome and calculated the average signature of this host genome. The Euclidian distances of all the window signatures were recalculated with regards to this new host signature. Afterwards, taking into account only the windows of this putative host genome, we established a threshold equal to the 99% percentile of the Euclidian distance to the host genome. All the windows whose signature exhibited a distance above this threshold were considered atypical and potentially corresponding to foreign DNA. We chose a high threshold in order to favor specificity rather than sensitivity.

All genes included in the atypical regions were analyzed: we investigated their functions and compared them to Genbank by BlastP (E-value ≤ 10^-10 and coverage ≥ 80%) in order to identify the closest homologous sequences if any. For atypical regions containing no annotated coding sequences, a BlastX analysis (E-value ≤ 10^-1) was done in order to identify remnants of coding sequences and a BlastN (E-value ≤ 10^-1) to find homology at the DNA level.

Protein sequences from the Blast analysis were aligned by ClustalW 79 . The trees (neighbor joining algorithm 80 ) were bootstrapped (1000 trees) and the consensus trees calculated with the Philip package 81 . Species trees were inferred by retaining only one homolog per species (the best strain or the best homolog, the less significant paralogs were discarded).

We have derived and updated Genstyle, a database of species signatures 82 . Our database contains about 65,000 signatures of species strains, organelles, viruses and plasmids. It was composed as following: for each entry, all non redundant sequences longer or equal to 1 kb were gathered from Genbank then concatenated for signature calculation. We calculated the signature of each atypical region and searched the database for the closest signatures in terms of Euclidian distance. As genomic signatures are species-specific 6 50 52 53 54 83 84 85 , the species with the closest signatures could be considered as potential donors of the atypical regions only if the distances obtained were below the average threshold used for HT detection (241 AU) 6 .

In a first step, we checked that as already shown for other eukaryotes 71 all chromosomes of A. fumigatus presented a similar signature and intrinsic variability. The concatenated sequence of the 8 chromosomes was then used to establish the threshold. The study of the signature variations along the genome allowed for the distinguishing of 189 distinct atypical regions (Figure 1, Additional file 1). They represented 3.1% of the total genome (908 kb, Table 1). The average size of the atypical regions was 4.5 kb, ranging from 500 bp to 52.5 kb. In general, the atypical regions were spread along all chromosomes indicating no chromosome preference for foreign DNA insertions (Figure 1, Table 2).

Though all chromosomes contained atypical regions some seemed to exhibit a particular distribution like a sub-telomeric trend on chromosome 4 or an under-representation on the short arm of chromosome 2. We also denoted that in some cases, atypical regions were physically clustered as it can be seen at position 2.3 Mb of chromosome 6 (c6r14-c6r23, representing 53 kb of atypical sequences out of 107 kb of genomic DNA) (Figure 1, Table 2).

The 189 atypical regions detected can be divided into two groups: those containing annotated genes (134) and those with no coding features (55). A total of 214 annotated genes are encoded in the atypical regions. We checked by BlastP if new homologs were sequenced since the genome analysis 76 78 . Most of these genes exhibited homologous counterparts (Additional files 1 and 2) with the exception of ORFans. The ORFans can be divided in 2 classes: 16 genes from A. fumigatus have no homologs at all in GenBank and 5 have a homolog only in N. fischieri a very close neighbor of A. fumigatus.

The functions of 81 transferred genes are unknown. Considering the other 133 genes, a function is inferred for 39 of them and a putative one for the 94 others (Additional file 3). The majority of them (91; 68%) belong to central and intermediate metabolism. We detected few genes involved in virulence 78 86 among the horizontally transferred genes although this means of virulence spreading was already demonstrated for pathogenic fungi 16 72 87 88 . We detected a few genes proposed to play a role in pathogenicity: 1 lipase, 4 peptide transporters 89 , 5 genes of gliotoxin synthesis involved in virulence 90 91 and two genes coding for allergenic proteins. Also, we observed a high number of mobile elements detected in the atypical regions. Alongside the 214 genes, we found 129 transposons belonging to 5 families: Copia, Gypsy, hAT, Line and DDE1. In some cases, these transposons are clustered in a single region (Additional file 1, see c2p24, c4p18 or c6p2 for instance). We checked the signatures of mobile elements and found that they exhibited a signature close to that of the host genome and so were not the cause of the detection of the region but more likely markers of the transfer events 92 .

Fifty-five atypical regions lacked annotated genes. Despite this, a BlastX and BlastN analyses allowed to propose the presence of gene relics in 24 (47%) of these regions (Table 3). Besides some rRNA genes (regions c4r5 and c4r6), supposedly not transferred but detected by the method, and transposons, we found pseudogenes of nuclear or mitochondrial origin and plasmid parts. Figure 2 shows an example of such a region containing both transposons and a pseudogene. The large numbers of transposons contained in these regions (Table 3) supports their status of horizontally transferred regions 92 . It is interesting to notice that 3 annotated genes and a pseudogene are of mitochondrial origin, indicating HTs between mitochondrial and nuclear genomes.

It is possible from the BlastP analysis to get an indication of the donor species except for the ORFans (Table 4, Additional file 2) or genes/proteins with few homologs. The majority of the homologs detected originated only from fungal species (56%). It is to be noted that 16 genes are specific of A. fumigatus (no homolog in other fungal species). All the other genes had homologs in at least one or the other Aspergillus sp. or Neosartorya fischeri (a very close relative of A. fumigatus 91 ). This supports the view that most of the transfer events occurred before the Aspergillus speciation. For instance out of the 120 genes exhibiting homologs mainly in fungi, 18 (15% not taking into account the ORFans) had homologs only in Aspergillus sp. or in N. fischeri. However the patchiness of the Aspergillus species represented by the different genes suggests numerous rearrangements and gene losses in these species. Another point is that some genes had homologs only in N. fischeri (5, Table 4) confirming the very close relationship between A. fumigatus and N. fischeri. From the BlastP analysis, it can be noted that 19% exhibited homologs in other domains of life; for instance, 26 genes had homologs exclusively in prokaryotes out of the fungi homologs (Table 4). We also detected 19 homologs exclusively in other eukaryotic kingdoms (Table 4). From this analysis it is possible, not only to confirm the transferred status of the genes but also to propose in peculiar cases a source of these genes. The criterion for a confident result are a very high conservation (very low E-Value), a coverage over 90% and an alternation of fungi species with those from other domains or kingdoms. For instance, gene AFUA_7G06140 possibly originates from Amoebozoa species, gene AFUA_1G11310 from Metazoa species and genes AFUA_1G01660, AFUA_6G09600 and AFUA_6G09660 among others would be of Prokaryotic origin. Other genes exhibit a more complex perturbed evolutionary history like genes AFUA_1G05200 and AFUA_4G14130 originating from other Eukaryotic kingdoms and some exhibit a very complex history mixing Eukaryotic and Prokaryotic origins like genes AFUA_1G06810, AFUA_1G10110, AFUA_2G00720, AFUA_4G07710 or AFUA_5G10120 for instance. To confirm the transferred status and research an origin when the homologs were all from fungi origin or when the origin was more difficult to ascertain, phylogenetic trees were inferred (examples of phylogenetic trees are shown in Figure 3 and Additional file 4)). These phylogenetic protein trees exhibited large incongruencies as compared to their respective SSU rRNA trees. This confirmed a perturbed evolutionary history and supported the transferred status of these genes.

It is difficult to assess the species of origin of the transferred genes from the Blast or the phylogenetic analyses due to the bias in homologous sequenced genes. Another way to propose a species of origin for an HT was to benefit from the species-specificity of the genomic signature. If the horizontally transferred regions have kept the characteristics of their species of origin, then by comparing their genomic signature to a homemade database of species signatures, we can obtain indications about their origin. We compared the signature of the 189 atypical regions to the database 53 82 and for 117 of them plausible donor species could be assigned (samples of region and of their closest neighbor signatures are shown in Figure 4). Due to possible miss-assignments caused either by the representativeness of the database or to the amelioration of the transferred sequences 93 , only broad categories of donors are presented. Figure 5 presents the distribution of these donor species as a function of their origin. Three major groups of donors are identified: bacteria (40%), fungi (25%) and viruses (22%). Among the bacteria species two groups are over-represented: Proteobacteria and Actinobacteria. An important point is the very small number of exchanges between fungi and non-fungal eukaryotes detected either from the BlastP or the signature analyses.