Background

Horizontal or lateral gene transfers (HGT and LGT) are important mechanisms of genome evolution that significantly contribute to the development of adaptive traits [1]. Although once considered a process of limited effect outside prokaryotes [2, 3], we now know that HGT and LGT have occurred in all major eukaryotic lineages (reviewed in [4]), including protozoans, plants, animals and fungi [5–7]. In fungi, HGT-driven gene innovation was shown to have resulted in refined repertoires of secreted and transporter proteins and increased metabolic capacities [8]. A survey of sixty fungal genomes detected hundreds of genes horizontally acquired from bacteria [9]. But the list of donors of fungal genetic material also includes plants [10], microbial eukaryotes [11, 12], and - most frequently - other fungi [13]. We will use the term LGT to describe the latter events.

Fungal secondary metabolites have a long history of positive (pharmaceuticals) and negative (toxins) impacts on mankind. Polyketides (PKS) make up a major group of them, most of which are formed by only a few, frequently not closely related species [14, 15]. The origin of PKS diversity has been explained as the result of gene duplication, HGT, LGT, recombination and domain shuffling [16]. However, most of these data have been obtained only for bacteria. Kroken et al. [17] postulated that the observed diversity in fungal PKS’s may not have resulted from HGT or LGT, but rather be due to birth-and-death evolution. However, increased sampling of genomic data from diverse taxonomic groups later provided evidence for the origin of several fungal PKS by HGT from bacteria, and also - in a few cases – by LGT from other fungi and plants [8, 18–22]. In almost all of these cases, translocation involved the whole secondary metabolite clusters (i.e. the PKS and the adjacently located genes encoding modifying enzymes, gene regulators and transporters) - rather than individual genes. To the best of our knowledge, the only exception is the demonstration of reacquisition of biotin prototrophy in Saccharomyces cerevisiae by stepwise HGT from bacterial donors [23].

Trichoderma is a genus of mycotrophic ascomycetes. Baker et al. [24] have recently compared the polyketide synthase (PKS) inventory of three Trichoderma species (T. reesei, T. virens and T. atroviride) and showed that two polyketide synthase encoding genes - pks10, pks11 - were unique to T. reesei. Pks10 and pks11 are located head-to-head in the center of chromosome 5 [25] and were shown to be responsible for the synthesis of sorbicillinoids [26]. These are complex cyclic polyketides, some of which have been shown to exhibit cytostatic and neuroprotective effects [27]. Sorbicillinoids are produced by T. reesei ([28, 29]; named T. longibrachiatum by the authors), but also some other fungal species belonging to the Sordariomycetes (e.g. Verticillium, Acremonium, Paecilomyces; for review see [27]) and the Eurotiomycete Penicillium notatum [30]. In support of this, a putative sorbicillinoid synthesizing cluster similar to the T. reesei cluster, is present in P. rubens [26, 31]. Moreover, the P. rubens orthologue of T. reesei pks11 (pks13) was shown to be essential for sorbicillinoid biosynthesis [32].

This limited, yet taxonomically widespread occurrence of sorbicillinoid biosynthesis in fungi led us to hypothesize that their evolution occurred by other mechanisms than vertical transfer. The goal of this study was to evaluate the evolutionary history of sorbicillin biosynthesis in T. reesei and other fungi. Here we show that this PKS cluster indeed originated from LGT, but in contrast to other reported cases [8, 18–22] it was not transferred as a whole cluster but formed by separate transfers of the individual genes from different donor species. The first almost complete cluster occurred in A. chrysogenum, from where it was transferred to P. rubens. In contrast, its further shaping in the Hypocreales occurred mainly via birth-and-death evolution and survived only in a few species including one of the most recent lineages of Trichoderma, the Longibrachiatum section.

Results

Identification of homologues of the sorbicillinoid biosynthetic clusters in Ascomycetes

To identify gene clusters potentially involved in sorbicillinoid biosynthesis in fungi, we first searched the National Center for Biotechnology Information (NCBI) protein database with the two PKS10- and PKS11- encoded proteins of T. reesei, which represent a non-reducing and a reducing PKS respectively, by bidirectional BLASTP (see Methods, (Additional file 1: Figure S1). Genes encoding proteins with highest similarity to both PKS10 and PKS11 were identified from the plant pathogenic fungus Colletotrichum graminicola (Sordariomycetes, Glomerellales), the opportunistic cephalosporin C-producer Acremonium chrysogenum (Sordariomycetes, Hypocreales), the “rice false smut” causing pathogen Ustilaginoidea virens (Sordariomycetes, Hypocreales) and Penicillium rubens (Eurotiomycetes, Eurotiales). Genes encoding proteins with still high similarity to PKS10 and PKS11 were also found in several other fungi (Eurotiomycetes and Sordariomycetes for PKS11, and – in addition – Dothidiomycetes for PKS10), but only the four species named above contained both of them.

Baker et al. [24] reported that PKS11 and PKS10 are unique to T. reesei, based on the absence in other Trichoderma species for which genome sequences were available at that date. Since the genomes of eight more Trichoderma spp. (i.e. T. harzianum, T. asperellum, T. hamatum, T. gamsii, T. longibrachiatum, T. citrinoviride and T. parareesei) are now available ([32–36]; http://genome.jgi.doe.gov/programs/fungi/index.jsf), we also screened them for the presence of pks10 and pks11 orthologs. The two genes were only found in T. longibrachiatum, T. citrinoviride and T. parareesei, which all are – as T. reesei – members of the Longibrachiatum Section of Trichoderma [37].

The PKS10 and PKS11 orthologs that were retrieved by BLASTP and by screening of the Trichoderma genomes shows that they form a significantly supported clade that contained all those species in which genomes both PKSs were present. C. graminicola occurred at a basal position in this clade (Fig. 1). To indicate that these two genes are part of the sorbicillin biosynthetic cluster, we will – in agreement with [26] - further name them sor1 (=pks11) and sor2 (=pks10) throughout the manuscript.

Fig. 1
figure 1

Phylogenetic analysis of SOR1/SorA (T. reesei PKS11) and SOR2/SorB (T. reesei PKS10) proteins by PhyML. Numbers at the nodes indicate the boostrap (1000 replica) support. Species printed in red belong to Hypocreales, brown to Glomerellales, black Pseudoeurotiaceae (all Sordariomycetes), blue Eurotiomycetes, green Dothidiomycetes. Accession numbers for proteins are given in Additional file 9: Table S1

Full size image

The sorbicillinoid biosynthetic gene clusters in T. reesei and P. rubens comprise 8 and 6 genes, located on chromosomes 5 and 1, respectively ([25, 26], http://www.ncbi.nlm.nih.gov/protein/CAP95405) (Fig. 2). sor3/sorC and sor4/sorD encode binuclear Zn2Cys6 transcription factors, of which sor4 is essential for the biosynthesis of sorbicillinoids in T. reesei [26]; sor5/sorE encodes a FAD-dependent monooxygenase responsible for the oxidative de-aromatisation of sorbicillin and dihydrosorbicillin to sorbicillinol and dihydrosorbicillinol, respectively [38]; and sor6/sorF encodes a transporter of the major facilitator superfamily (MFS).

Fig. 2
figure 2

Architecture of the sorbicillinoid biosynthesis gene cluster in those fungi that possess SOR1/SorA and SOR2/SorB orthologues. Red: PKS, blue/dark blue: Zn2Cys6 transcription factor; green, FAD monooxygenase; orange, MSF transporter; black, short chain reductase/dehydrogenase; white, FAD-dependent oxidase. The dotted cluster in A. chrysogenum specifies an unknown protein. Accession numbers for all proteins shown are given in Additional file 9: Table S1

Full size image

We consequently analysed whether the other species that contain sor1 and sor2 would indeed also contain the other 4 or 6 genes that are present in the P. rubens and T. reesei cluster, respectively, and have them organized in a genomic cluster (Fig. 2): C. graminicola contained no homologues of any of them, but A. chrysogenum, U. virens and the other 3 Trichoderma spp. contained sor3, sor4, sor5 and sor6 (i.e. the genes encoding the two transcription factors, the MFS transporter the FAD-dependent monooxygenase, respectively). All of them were located in immediate vicinity of sor1 and sor2, although U. virens sor5 is located a few genes farther apart than in the other fungi.

The cluster in T. reesei contained two further genes – sor7 and sor8 - that were absent from most other fungi: sor7, encoding a short-chain dehydrogenase/reductase, for which a P. rubens ortholog (CAP92704.1) - is present in the genome but not located in the vicinity of the sorbicillinoid gene cluster In A. chrysogenum, another gene of unknown function is found at the position of sor7. sor8 encodes an FAD-dependent oxidase (Fig. 2), of which an ortholog is present in A. chrysogenum and U. virens, but not located in the vicinity of the sorbicillinoid cluster, and absent from the P. rubens genome (Fig. 2).

We also looked at possible synteny of the 5′ and 3′ flanking regions of the cluster: while there was considerable synteny between the four Trichoderma spp., no synteny was found between Trichoderma and the other fungi possessing the sorbicillinoid biosynthesis cluster.

We tested the null hypothesis that the phylogenetic history of SOR3 – SOR8 was consistent with a vertical transfer within fungi by implementing phylogenetic analyses (Additional file 2: Figure S2 A-F). This shows that only SOR5 forms a strongly supported clade containing – except for C. graminicola - all species that also contain the SOR1 and SOR2 proteins. SOR3, SOR4 and SOR6 are distributed in several clades, which – even after collapsing branches with poor bootstrap support (<75%) are not concordant with the established Ascomycota phylogeny (cf. [39]). On the other hand, SOR7 and SOR8 display a phylogeny that strongly resembles the Ascomycota phylogeny (see below; cf. Additional file 3: Figure S3). They are also present in Trichoderma spp. which lack the sorbicillinoid biosynthetic cluster.

Evolution of the sorbicillinoid gene cluster in filamentous fungi

The above described discordance between the phylogeny of SOR1-SOR6 homologues and the Ascomycota phylogeny suggested that they may have arisen by LGT from different ancestors. To test this hypothesis, we applied three complementary approaches: the bipartition dissimilarity test implemented in T-REX [40], which identifies HGT/LGT events by quantifying the proximity between two phylogenetic trees using a refinement of the Robinson and Foulds distance [41, 42]; the reconciliation of each gene tree to the fungal species phylogeny, thereby assigning costs to gene duplications, HGT/LGT, gene loss, and incomplete lineage sorting, as implemented in Notung [43]; and the Jane software tool that uses a polynomial time dynamic programming algorithm in conjunction with a genetic algorithm to find solutions pairs of trees [44]. We accept proof for HGT/LGT only for those cases where (i) at least two of these programs provided consistent results that were not rejected by the third, and (ii) where the protein tree topology was contradictory to the Ascomycota phylogeny and could not be more parsimoniously reconciled using a combination of differential gene duplications (GD) and gene loss.

The evolution displayed by the results from this analysis (Additional file 4: Figure S4, Additional file 5: Table S2) are summarized in Table 1: evidence for LGT was obtained for A. chrysogenum (SOR4), Trichoderma and U. virens (SOR2, SOR3), and P. rubens (SorA, SorC, SorE and SorF). Interestingly, three of the genes of P. rubens (SorA, SorC and SorF) were obtained from A. chrysogenum, whereas SOR4 of A. chrysogenum was obtained from P. rubens (SorD), indicating frequent LGTs between these two species. In Trichoderma and U. virens, only SOR2 appears to have been obtained by LGT from C. graminicola.

Table 1 Summary of LGT events detected by T-Rex, Notung and Jane
Full size table

With the exception of SOR2, neither U. virens nor the four Trichoderma spp. appear to have received any of the other cluster genes by LGT. No LTG events could be inferred for sor7 and sor8 (neither by Notung, T-REX nor Jane) which is in agreement with the observation that these genes occupied positions concordant with Ascomycota phylogeny (vide supra).

Trichoderma SOR1 and SOR2 evolved by purifying selection

The Longibrachiatum Section of Trichoderma is one of the most recent branches in Trichoderma evolution [45]. The fact that we were unable to identify the sor genes in less evolutionary derived species of the genus but could not verify LGT as the mechanism of origin of the sorbicillioid gene cluster in the Longibrachiatum clade was thus unexpected. The alternative hypothesis to explain the absence of these genes in other species is that these genes have been lost. To test this hypothesis, we reconstructed the evolution of the eight SOR proteins by Count [46] and Gloome [47]. The results provided consistent evidence for loss of the respective genes in other infrageneric clades of Trichoderma and in those Hypocreaceae species that are close to Trichoderma but also lack them (Additional file 6: Figure S5). Interestingly, the ratio of the pairwise amino acid differences between SOR1/SorA and SOR2/SorB and the housekeeping genes used to construct the Ascomycota tree (see Methods), was significantly higher in the four Trichoderma spp. than in U. virens, P. rubens or A. chrysogenum (Table 2). This would be typical for LGT, as was found for SOR2. However, since sor1 has not been obtained by LGT, it may as well be due to a higher rate of evolution of these two genes in Trichoderma section Longibrachiatum. Determination of the Ka/Ks ratio for the Trichoderma sor1 and sor2 genes yielded values around 0.1, suggesting the operation of strong purifying selection.

Table 2 Pairwise amino acid distance of the four housekeeping genes (HKG). SOR1 and SOR2 between Trichoderma citrinoviride and other fungi
Full size table

Sorbicillinoid cluster gene expression in T. reesei

Many PKS synthesizing clusters in fungi are silenced [48]. We therefore used available oligonucleotide microarray data of T. reesei growing on glucose, glycerol, lactose or cellulose as carbon sources in submerged culture, or on glucose on agar plates to test whether the sor genes are indeed expressed. In fact the eight sor genes in the Trichoderma cluster are expressed at high levels under conditions of rapid growth (glucose, glycerol), whereas lower expression was detected on lactose which allows only slow vegetative growth (Fig. 3a). Most sor genes had only a low level of expression during asexual sporulation (Fig. 3a).

Fig. 3
figure 3

Heat plot of the expression of sor1 - sor8 in T. reesei during a growth on different carbon sources and onset of sporulation; Glc - glucose; Gly - glycerol, Lac - lactose, b during cultivation of a lae1 knock-out and a lae1 overexpressing strain on lactose, and c plotting of the ratio between the parent strain to the lae1 knock-out (white bars) and the lae1 overexpressing strain (black bars), respectively

Full size image

The protein methyltransferase LaeA is a major regulator of secondary metabolism in Eurotiomycetes and some Sordariomycetes [48]. Its T. reesei orthologue LAE1 regulates some but not all PKS genes [49]. As shown in Fig. 3c, a lae1 knock-out mutant shows significantly decreased expression of sor1 and sor2, and interestingly also of sor7 and sor8. No increased expression was observed for these four genes in a strain overexpressing lae1 under a constitutive promoter. However, the genes encoding one of the two transcription factors (sor3), the MFS transporter (sor6) and the FAD monooxygenase (sor5) were significantly upregulated by lae1 overexpression.

Discussion

Although HGT and LGT occur in the majority of cases by transfer of single genes only (for review see [9, 50], the transfer of multiple genes or gene clusters has also been shown [11, 19, 50–52], particularly for genes encoding proteins for secondary metabolite synthesis [18, 20, 53–59]. In contrast, our data show that the fungal sorbicillin biosynthesis cluster evolved by complementing symplesiomorphous genes by LGT from other fungal donors, which in P. rubens occurred in at least two steps. Based on the species phylogeny and the LGT events found, A. chrysogenum is the most ancient known taxon that contains an almost complete cluster that misses only the transcription factor sor4, and which it obtained by LGT from P. rubens. The more recent Hypocreales (U. virens and the four Trichoderma spp. of section Longibrachiatum) regained one of the PKSs (sor2) from C. graminicola, implying that this gene was lost in one of the Hypocreales that are more recent than A. chrysogenum.

In contrast, the cluster is missing in the Eurotiales with the exception of P. rubens, at least with respect to species whose genome sequence is available. The Eurotiales only contain orthologs of the PKS SorB and the transcription factor SorD, which lends to speculate that these two genes are involved in the synthesis of another polyketide. It is interesting to note that three missing genes (sorA, sorC and sorF) were obtained from A. chrysogenum, to which P. rubens transferred its sorD, indicating a history of a frequent gene exchange between these two fungi. We cannot say, however, whether the LGT from A. chrysogenum to P. rubens occurred in one or several steps. It is also interesting that sorE (encoding the FAD monooxygenase crucial for sorbicillin formation [38]) had not been transferred to P. rubens but to an unknown Eurotiales ancestor from P. fici. Thus sorE must have been present in P. rubens before LGT of sorA, sorC and sorF.

The absence of sor3 – sor6 from C. graminicola could be due to gene loss. An alternative hypothesis, however, would be that SOR1 and SOR2 are synthesizing a different polyketide than sorbicillin in this fungus. Indeed, an annotation of the genes flanking the C. graminicola sor1/sor2 locus revealed an adjacent oxidoreductase gene whose encoded protein exhibited 83% amino acid similarity to an oxidoreductase CtnB involved in citrinin biosynthesis in Monascus aurantiacus (Eurotiales) [60], and a putative aldehyde dehydrogenase (Additional file 7: Figure S6). We therefore assume that the resulting polyketide synthesized by Colletotrichum is (or was) probably not a sorbicillinoid. Sorbicillinoid were thus in fact first produced in A. chrysogenum or closely related but as yet unknown ancestor.

Despite of the occurrence of the sorbicillinoid gene cluster in only U. virens and Trichoderma spp. from section Longibrachiatum, we found (with the exception of sor2) no evidence for their origin by further LGT events. Instead, our data show that the cluster evolved by vertical transfer, and has been lost by the operation of massive birth-and-death evolution [61] within the Hypocreales. In fact, a scenario of gene duplications followed by gene loss has earlier been suggested for the evolution of fungal non-reducing polyketide synthases [17], and claimed to be responsible for the todays patchy distribution of distantly related secondary metabolites.

Yet our finding that the sorbicillinoid cluster only survived in Trichoderma species belonging to section Longibrachiatum is interesting. It is consistent with the formation of the characteristic yellow pigment secreted by these species [62], because sorbicillinoids have a characteristic yellow-orange color [63]. Species from this section are known to have smaller genomes than other Trichoderma spp. and represent one of the youngest phylogenetic clades of the genus [38, 45]. The fact that the sorbicillinoid gene cluster has been maintained in these species but not in other Trichoderma spp. suggests that the respective products are of selective importance to fungi from this section. This is also supported by our findings of strong purifying selection acting on sor1 and sor2. Unfortunately, the function of sorbicillinoids is not known yet: although some sorbicillinoids were reported to inhibit the growth of tumour cells [27, 64], they usually display only low inhibitory activity against bacteria and fungi [27]. Their role as components of antagonism against other organisms is unlikely. Rather, their ecological importance may reside in their high antioxidant and radical scavenging activity [27, 65]. Our findings of high expression of the sor genes in T. reesei under conditions of fast growth, but not during sporulation, supports a role of sorbicillinoids in vegetative growth, which is corroborated by finding them in high concentrations during submerged growth of T. reesei (Additional file 8: Table S3). Interestingly, sor1 and sor2 are also strongly upregulated upon confrontation of T. reesei with plant pathogenic Thanatephorus spp./Rhizoctonia solani (Cantharellales, Basidiomycota) [66]. At a first glance, this contradicts the above conclusion that sorbicillinoids are not involved in antagonism. However, the protection against radicals formed by reactive oxygen species is an important defence reaction of fungi, plants and higher eukaryotes when confronted by other organisms [67–70]. It will be intriguing to find out whether the sorbicillinoids indeed play such a role and - if so – why their biosynthesis was just maintained in only a small group of fungal species.

Conclusions

Tracking the evolution of secondary metabolite synthesizing gene clusters by LGT or HGT have so far in most cases been restricted to the detection of transfer of the whole clusters between two fungi. Our findings show how a fungal secondary metabolite cluster was assembled by individual genes from different fungi by LGT before it became subject to birth-and-death evolution in selected lineages.

Methods

Identification of sorbicillinoid biosynthetic genes in fungi

The eight proteins of the sorbicillinoid biosynthetic cluster in T. reesei were used in a preliminary sequence similarity search by BLASTP of the NCBI database. One hundred best hits were collected. In addition, we searched the genomes of T. longibrachiatum, T. citrinoviride, T. asperellum, T. hamatum and T. parareesei for homologues to SOR1 - SOR8. Since the latter two are not available in a public database, we prepared a local BLAST databases for these two fungi. All these sequences were then aligned by CLUSTALW [71], and subjected to phylogenetic analysis with PhyML 3.0 using the Dayhoff model and 1000 boostrap replica [72]. The topology of the resulting tree was analysed, and all proteins that formed clades not related to that comprising the T. reesei proteins were removed. The resulting collection of sequences was re-aligned with MUSCLE [73] and CLUSTALW [71] and edited by GBLOCKS [74] to identify potential differences in the phylogenetic reconstruction due to the use of different methods. Individual trees were reconstructed with the individual edited protein alignments with PhyML 3.0 using 1000 bootstrap repetitions, and their topology concordance confirmed. The four alignments were then concatenated, and Bayesian analysis performed with TOPALI v2.5 [75], using the WAG model, gamma substitution and 100,000 generations.

Ascomycota tree reconstruction

To reconstruct the reference Ascomycota phylogeny containing all fungi putatively involved in the LGT events described in this paper, the amino acids inferred from four nuclear genes, which were shown to be good phylogenetic markers for fungal species trees reconstruction (i.e. histone acetyltransferase subunit of the RNA polymerase II holoenzyme, FG533; NAD-dependent glutamate dehydrogenase, FG570; translation initiation factor eIF-5, FG832; and Tsr1p, a protein required for processing of 20S pre-rRNA, MS277) were retrieved from FunyBase [76] (http://genome.jouy.inra.fr/funybase). Proteins from species not contained in FunyBase were retrieved by BLASTP search of the GenBank (http://www.ncbi.nlm.nih.gov/genbank/), the Joint Genome Institute (http://genome.jgi-psf.org/programs/fungi/index.jsf?projectList), EnsemblFungi (http://fungi.ensembl.org/index.html) and Broad Institute (http://www.broadinstitute.org/) databases (all databases accessed 28-12-2015). Their alignment, and analysis by PhyML 3.0 and Bayesian analysis were essentially performed as described above.

Inferring HGT/LGT events

To test for the occurrence of HGT, three approaches were used: first, the bipartition dissimilarity test implemented in T-REX [40], which quantifies the proximity between two phylogenetic trees using a refinement of the Robinson and Foulds (RF) distance, was used by applying midpoint rooting and HGT identification by iteration. Second, a gene tree-species phylogeny reconciliation was performed in Notung, using its duplication, transfer, loss and ILS aware parsimony-based algorithm [43]. To this end, gene tree nodes with less than 0.90 SH-like local support were collapsed, and the resulting tree rooted and its polytomies resolved against the bifurcating species phylogeny. This resolved gene tree was then reconciled to the multifurcating, consensus species phylogeny using a duplication cost of 1.5, loss cost of 1 and ILS cost of 0, and the option to prune taxa not present in the gene tree enabled. Third, Jane version 4, a software tool for cophylogeny reconstruction problems that attributes costs to cospeciation, duplication, host switch, and sorting was used [44]. For our analyses we employed default cost settings, and the population size was set 50-fold the number of generations.

Gene gain and loss analysis

Gene gain and loss was tested by two methods: (i) Count [46], which can perform ancestral genome reconstruction by posterior probabilities in a phylogenetic birth-and-death model. Rates were optimized using a gain–loss–duplication model, with default parameters and allowing different gain–loss and duplication–loss rates for different branches, and one hundred rounds of optimization. (ii) Gloome [47], which enables accurate inference of gain and loss events by a stochastic mapping approach, using a variable gain and loss ratio.

Analysis of selection pressure by Ka/Ks ratio

Tajima’s D statistic [77] was determined with DNASp 5.0 [78], using a sliding-window approach.

Transcriptome analysis

We used transcriptome data from our own earlier studies. These included: cultivation of T. reesei QM 9414 on D-glucose, glycerol, lactose, and wheat straw in batch cultures [79, 80], during induction of asexual sporulation [81], at the onset of confrontation with the basidiomycete Thanatephorus spp./Rhizoctonia solani [82], and during growth on lactose in lae1 knock-out and lae1-overexpressing strains [51]. All transcriptome data were obtained by oligonucleotide array hybridization, using a high-density oligonucleotide microarray (Roche-NimbleGen, Inc., Madison, WI) with 60-mer probes representing the 9123 genes of T. reesei. Values were normalized by quantile normalization [83] and the RMA algorithm [84]. After elimination of transcripts that exhibited an SD >20% of the mean value within replicates, false discovery rates [85] were used to assess the significance of values. All transcriptome data and the related protocols are available at the GEO web site (http://www.ncbi.nlm.nih.gov/geo) under the accession numbers given in the cited papers.