The ACE1 secondary metabolism gene cluster of M. grisea comprises 15 genes: ACE1 and SYN2 are PKS-NRPS hybrid genes; RAP1 and RAP2 code for enoyl reductases; CYP1-CYP4 for cytochrome P450 monoxygenases; ORFZ for an α/β-hydrolase; OXR1 and OXR2 for oxidoreductases; MFS1 codes for a transporter in the MFS superfamily; BC2 codes for a binuclear zinc finger transcription factor; OME1 codes for an O-methyl transferase; and ORF3 has no homology to known proteins (Collemare et al, unpublished results). To find gene clusters homologous to the ACE1 cluster in other fungal species, we used an algorithm that searched 26 fungal genomes for loci where at least three likely orthologs of genes from the ACE1 cluster were linked (see Materials and methods). This search identified nine similar clusters in seven fungal species from the subphylum Pezizomycotina: three Sordariomycetes (Chaetomium globosum, Fusarium oxysporum and F. verticillioides), one Dothideomycete (Stagonospora nodorum) and three Eurotiomycetes (Aspergillus clavatus, Coccidioides immitis and Uncinocarpus reesii) (Figure 1).

Two types of clusters related to the ACE1 cluster were identified: large clusters with eight or more genes are found in M. grisea, C. globosum and S. nodorum, whereas smaller clusters with three to six genes are found in the three Eurotiomycetes and in Fusarium species (Figure 1). C. globosum is unusual as its genome contains two large ACE1-like clusters, which we refer to as clusters 1 and 2. Similarly, the A. clavatus genome has two clusters as discussed below. Interestingly, a core set of three genes (homologs of ACE1, RAP1 and ORF3; boxed in Figure 1) is present in all eight species. The presence of this core suggests that the physical linkage between these three genes is ancient and can be inferred to have existed in the common ancestor of all the genomes considered in Figure 1. As well as the genes in the eight clusters shown in Figure 1, we also identified a small number of single homologs of genes from the M. grisea ACE1 cluster that are located at dispersed genomic locations in other species.

Gene-by-gene phylogenetic analyses were carried out to decipher the evolutionary history of the loci using homologs (even at dispersed locations) of genes from M. grisea ACE1 cluster (Figure 2 and Additional data file 1). The first trend evident from this phylogenetic analysis is that genes from clusters in Eurotiomycetes and Fusarium spp. are distant from those of the M. grisea, C. globosum and S. nodorum clusters. Indeed, genes in clusters from these last three species define clades supported by high bootstrap values (> 91%), to the exclusion of genes from Eurotiomycetes and Fusarium species (Figure 2a,b,e,f). Interestingly, genes from one of the two clusters in A. clavatus are more closely related to genes in the M. grisea ACE1 cluster than to those in ACE1-like clusters from other Eurotiomycetes (see below). In view of the gene contents of the clusters and their phylogenetic relationships, we refer to the large clusters in M. grisea, C. globosum, S. nodorum and the larger of the two clusters in A. clavatus as "ACE1 clusters", and to the smaller clusters in Eurotiomycetes and Fusarium spp. as "ACE1-like clusters". These two types of cluster have probably had a long history of independent evolution, although they certainly share a common ancestor.

We then focused on the origins of the duplicated genes in the M. grisea cluster. Phylogenetic trees show clearly that in M. grisea RAP2 is a paralog of RAP1, CYP3 is a paralog of CYP2, CYP4 is a paralog of CYP1, and SYN2 is a paralog of ACE1 (Figure 2a-d). Notably, in each of these pairs, one gene is located on the left-hand side of the M. grisea cluster and the other is on the right-hand side. Thus the M. grisea cluster appears to have undergone partial tandem duplication at some stage during its evolution, although the gene order is not conserved between the two parts. The presence of two ACE1 clusters in C. globosum is suggestive of a second block-duplication event in this species. However, for most genes present in both C. globosum ACE1 clusters, the copy from cluster 1 forms a clade with their M. grisea homologs. This close phylogenetic relationship is observed for ACE1, RAP1, ORFZ, OXR1, CYP1, and OXR2. The only exception to this pattern is M. grisea ORF3, which is marginally closer to the C. globosum cluster 2 gene, but with low bootstrap support (Figure 2 and Additional data file 1). This observation suggests that the duplication that gave rise to the current C. globosum clusters 1 and 2 occurred in a common ancestor of C. globosum and M. grisea, and that the corresponding cluster 2 in M. grisea was lost.

On the basis of this analysis, we divided the M. grisea cluster into two parts, A and B, so that each of the duplicated genes in M. grisea has one copy in part A and one in part B (Figure 1). Part A in M. grisea consists of nine genes, all of which have orthologs in one or both of the clusters in its closest relative C. globosum. The clusters in other species consist of homologs of genes from M. grisea part A, plus one gene from part B (ORF3; see Discussion). The order of the part A genes is not conserved among M. grisea, C. globosum and S. nodorum.

Surprisingly, this phylogenetic analysis shows that five of the six genes from part B of the M. grisea ACE1 cluster group with genes from the larger of the two clusters in A. clavatus, rather than with the genes in the more closely related (Sordariomycete) species C. globosum, or with their part A paralogs in M. grisea. Bootstrap values for grouping the M. grisea part B genes SYN2, RAP2, CYP4, CYP3 and ORF3 with their A. clavatus homologs are 98-100% (Figure 2a-e). The only gene from part B of the M. grisea cluster that does not group with A. clavatus is OME1 (panel e of Additional data file 1), but this is also the only gene whose detected homolog in A. clavatus (ACLA_002520) is not physically clustered with the others, which calls its orthology into question. The consistency of this phylogenetic result for part B genes, and its disagreement with the expected species relationships, are indicative of HGT between A. clavatus and part B of the M. grisea cluster. In contrast seven of the nine genes from part A of the M. grisea cluster, including ACE1 itself, lie at the expected phylogenetic position forming a clade with C. globosum (Figure 2 and Additional file 1; the two exceptions are CYP2, which is discordant but has a low bootstrap value of 66%, and MFS1, which cannot be analyzed because there is no homolog in the C. globosum clusters).

For the four panels in Figure 2 that include sequences from other Eurotiomycetes (C. immitis and U. reesii) as well as A. clavatus, we used the likelihood ratio test (LRT) to test whether the topologies shown (Figure 2a,b,e,f) have significantly higher likelihoods than alternative trees where the Eurotiomycetes were constrained to form a monophyletic group. In all four cases the topology shown in Figure 2 is significantly more likely than the tree expected if genes were inherited vertically (p < 0.001 for each).

To determine whether part B of the cluster was transferred from an M. grisea-like donor to an ancestor of A. clavatus, or vice versa, we examined phylogenetic trees constructed from those genes that have orthologs both in species that are close relatives of M. grisea and in species that are closer to A. clavatus. We would predict that if an ancestor of A. clavatus was the recipient of HGT, then the genes in its ACE1 cluster would not show the expected close relationship to other Eurotiomycete species such as C. immitis and U. reesii (Figure 1), and would instead form a clade with the donor lineage (represented by M. grisea). Conversely, if the direction of transfer was from an A. clavatus-like donor into the M. grisea lineage, we would expect the M. grisea part B genes not to form a monophyletic clade with the other Sordariomycete species C. globosum, and instead to group with A clavatus.

In the phylogenetic tree of ORF3 sequences, the shared A. clavatus-M. grisea branch lies within a clade that contains homologs from the two clusters in C. globosum, as well as the Dothideomycete S. nodorum (Figure 2e). The ORF3 orthologs from C. immitis and U. reesii clearly lie outside this clade with 95% bootstrap support. Similarly, the phylogenetic tree of RAP1 and RAP2 orthologs (Figure 2b) shows that the shared branch containing the A. clavatus gene and the part B M. grisea gene (RAP2) lies within a larger clade that includes the C. globosum and M. grisea part A (RAP1) orthologs. The homologs from C. immitis and U. reesii lie outside (91% bootstrap support). Likewise, the phylogenetic tree of the ACE1-SYN2 pair (Figure 2a) places the A. clavatus sequence within a Sordariomycete/Dothideomycete clade, distant from the other Eurotiomycetes (C. immitis and U. reesii). These topologies all indicate that an ancestor of M. grisea was the donor of the transferred part B genes, and an ancestor of A. clavatus was the recipient.

ORFZ is the only gene in the A. clavatus ACE1 cluster that does not have a homolog in part B of the M. grisea cluster. The origin of this gene in A. clavatus is not clear. Phylogenetic analysis (Figure 2f) indicates that A. clavatus ORFZ does not group with the C. immitis and U. reesii genes, and this conclusion is supported by the LRT. This result suggests a foreign origin for A. clavatus ORFZ, but the absence of a homolog in M. grisea part B makes it impossible to test whether this gene has a similar origin to its five neighboring genes in A. clavatus.

We conclude that there is phylogenetic support for the hypothesis that at least five of the six genes in the ACE1 cluster of A. clavatus originated by HGT, and that the most probable single donor is a Sordariomycete ancestor related to M. grisea.

A complete ACE1 cluster is present in only four of the 23 sequenced Pezizomycotina genomes (M. grisea, C. globosum, S. nodorum and A. clavatus). Such a sporadic distribution could be the result of either independent HGTs or frequent losses of the whole cluster in different lineages (Figure 3). We favor the latter explanation because - with the exception of A. clavatus - our phylogenetic trees of genes from the cluster have topologies that are in broad agreement with the expected species phylogeny 27. We suggest that an ACE1-like cluster consisting of at least three genes (homologous to ACE1, RAP1 and ORF3) existed in the common ancestor of Pezizomycotina, but this cluster has been lost in many lineages subsequently. The scheme in Figure 3 identifies four independent lineages (shown by dashed lines) in which all copies of the cluster have been lost. We cannot tell, with current data, whether genes such as OXR1 that are present in the ACE1 clusters of Sordariomycetes and Dothideomycetes but not in the ACE1-like clusters of Eurotiomycetes correspond to lineage-specific additions or losses.

Any tree showing apparent HGT of a gene can also be explained by an alternative scenario of gene duplications and losses. However, the situation reported here is rather different to typical cases of possible HGT of individual genes, because it involves multiple genes that are arranged as a large tandem duplication (in M. grisea). The fact that the A. clavatus ACE1 cluster forms a clade with the M. grisea part B genes (to the exclusion of the part A genes) means that the only alternative scenario to HGT is one where the part A/part B tandem duplication occurred right at the base of the tree in Figure 3. This scenario would then necessitate at least four events of precise loss of exactly one part of the tandemly duplicated set of genes: part B in C. globosum, part B in the ancestor of C. immitis and U. reesii, part B in S. nodorum, and part A in A. clavatus. Because of the precise nature of the deletion required (and choice of gene copy to delete), we do not regard this scenario as likely.

The discontinuous distribution of the ACE1 cluster among fungal species suggests that evolutionary constraints act to maintain this cluster only in few species. As M. grisea, S. nodorum and C. globosum are plant or animal pathogens, it is tempting to speculate that the ACE1 cluster is involved in the infection process of these three species. The metabolite produced by this biosynthetic pathway may be an important pathogenicity factor, but such a role remains to be determined. A. clavatus is different as it is not pathogenic. The presence of the ACE1 cluster in A. clavatus may arise from selection involving an unknown biological role of this metabolite in this fungus. Identifying the molecules made by these different clusters will be necessary to understand the role of the ACE1 cluster in fungal biology and could give clues about evolution of the ancestral biosynthetic pathway controlled by this cluster.

The ACE1 cluster has a complex history with multiple events of large-scale duplication and multiple losses. The scenario we infer is summarized in Figure 3. An ancient duplication produced the large ACE1 and smaller ACE1-like clusters. A second duplication event in an ancestral Sordariomycete gave rise to the two clusters (1 and 2) presently seen in C. globosum. This event occurred prior to the speciation between C. globosum and M. grisea, but M. grisea later lost its counterpart of cluster 2. Independently, cluster 1 underwent a tandem duplication event, generating parts A and B. This tandem duplication survived in M. grisea, but in C. globosum the addition (part B of cluster 1) was lost again. It might seem simpler to suggest that the part A/B tandem duplication was an event that occurred specifically in M. grisea after it diverged from C. globosum, but we know that this is incorrect because the part B genes from M. grisea form outgroups to a clade consisting of C. globosum and M. grisea part A genes. We can also be sure that the surviving duplications seen in M. grisea and C. globosum were separate events because of the topology of the phylogenetic trees: if the surviving genes were descended from the same duplication event we would expect that in the ACE1-SYN2 tree, for example, M. grisea ACE1 and SYN2 should each form a separate monophyletic group with one of the C. globosum genes, but that is not seen (Figure 2a). Instead we interpret the trees as indicative of two duplications of the whole cluster in a Sordariomycete ancestor of M. grisea and C. globosum, the first of which was non-tandem and the second of which was tandem. After this tandem duplication, the M. grisea lineage lost its ortholog of cluster 2 of C. globosum, and the C. globosum lineage lost its ortholog of part B of M. grisea (Figure 3). This pattern of frequent loss is consistent with the cluster's sporadic distribution in fungi.

ORF3 is unusual as it is inferred to have been present in the ancestor of all ACE1 and ACE1-like clusters, but in M. grisea it is not duplicated and it shows phylogenetic affinity to A. clavatus rather than to C. globosum or S. nodorum (Figure 2e). These properties suggest that a homolog of ORF3 was lost from part A of the M. grisea cluster, after the tandem duplication occurred. Furthermore, we speculate that the location of ORF3 on the boundary between parts A and B may indicate that the tandem duplication event visible in M. grisea involved a recombination between two copies of this gene.

Gene order and orientation is quite poorly conserved among the ACE1 clusters, as is typical of many secondary metabolism gene clusters 7828. This makes it all the more striking that the duplicated M. grisea genes each have one copy in the part A and one copy in part B. Because the tandem duplication that is evident in the M. grisea genome is not particularly recent (it predates the M. grisea/C. globosum speciation), we suggest that some form of selection has acted on gene order in the cluster, preventing intermixing of the two parts. In this context it is notable that recombination seems to be inhibited in the M. grisea ACE1 cluster, because it displays a low frequency of targeted gene replacement, even in a KU80 null mutant background where homologous recombination rates are increased (29; Collemare et al, unpublished results). The way that part A and part B genes of the ACE1 cluster are distributed among species may indicate that they are involved in the biosynthesis of different molecules. Alternatively, parts A and B of the ACE1 cluster may be each involved in the biosynthesis of independent polyketide precursors that are fused into a final complex molecule as observed for lovastatin 253031. The fact that all 15 genes in the M. grisea ACE1 cluster are co-expressed at a very specific stage of the infection process (Collemare et al, unpublished results) favors the hypothesis that both part A and part B genes are involved in same biosynthetic pathway. However, gene knockout experiments have shown that two part B genes (RAP2 and SYN2) are not essential for the avirulence function supported up to now only by the part A gene ACE1 (Collemare et al, unpublished results). These latter results suggest that part A and part B genes could be involved in the biosynthesis of two different molecules, with only one (ACE1, part A pathway) being recognized by resistant rice cultivars. However, these two hypotheses are both plausible, and await the biochemical characterization of the Ace1 metabolite.

Although the genomics era has uncovered evidence for widespread horizontal gene transfer among prokaryotes 3233, and from prokaryotes to eukaryotes 1734353637 or vice versa 3839, relatively few instances of horizontal gene transfer have been documented from one eukaryote to another 404142. Among fungi, the best documented is the transfer of a virulence gene from S. nodorum to Pyrenophora tritici-repens, which occurred only about 70 years ago 16. In that case, the transferred DNA fragment was about 11 kb in size but contained only one gene. In this study we showed that part B of the ACE1 cluster (30 kb in size, containing 5-6 genes) was likely horizontally transferred from a close ancestor of M. grisea (a Sordariomycete) into an ancestor of A. clavatus (a Eurotiomyete). The mechanism by which HGT might have occurred remains a matter of speculation, but could perhaps have involved hyphal fusion between species, or endocytosis. Our inference of HGT is valid only if the Sordariomycete and Eurotiomycete clades are monophyletic as shown in Figure 1, but their monophyly is supported by several molecular and systematic analyses 274344454647.

To our knowledge, our study and the recent work of Patron et al 18 are the first reported instances of HGT of groups of linked genes involved in the same pathway between eukaryotic species. In both cases these secondary metabolite clusters show a punctate (sporadic) distribution among other species, with an ancestral cluster apparently having been lost by more species than the number that retain it. This pattern of frequent losses of genes and their occasional reacquisition by HGT resembles the pattern of evolution of "dispensable pathway" genes in ascomycete yeasts 48. Hall and Dietrich 48 noted that genes whose products function in dispensable pathways are one of the few categories of genes in S. cerevisiae that are physically organized into gene clusters. They found that the pathway for biotin synthesis was lost in a yeast ancestor and then regained in the S. cerevisiae lineage by a combination of HGT from bacteria and gene duplication with neofunctionalization. One possible explanation for this strange pattern of evolution could be that an intermediate in the pathway is toxic 48, although there is no direct experimental evidence of this. If a pathway can confer a selective advantage in some circumstances but also involves the production of a toxic intermediate, there can be strong selection in favor of the pathway in some conditions and strong selection against it in others. The consequences of such a situation could include the formation of physical gene clusters (to reduce the chances of coding for only part of the pathway, or for strong repression of transcription mediated by chromatin remodelling), and occasional selection for re-gain of function by HGT. Further exploration of this hypothesis will require the discovery of more examples of similar sets of genes, and detailed characterization of the biochemical pathways involved.