Evolution of Chemical Diversity in Echinocandin Lipopeptide Antifungal Metabolites

Abstract

The echinocandins are a class of antifungal drugs that includes caspofungin, micafungin, and anidulafungin. Gene clusters encoding most of the structural complexity of the echinocandins provided a framework for hypotheses about the evolutionary history and chemical logic of echinocandin biosynthesis. Gene orthologs among echinocandin-producing fungi were identified. Pathway genes, including the nonribosomal peptide synthetases (NRPSs), were analyzed phylogenetically to address the hypothesis that these pathways represent descent from a common ancestor. The clusters share cooperative gene contents and linkages among the different strains. Individual pathway genes analyzed in the context of similar genes formed unique echinocandin-exclusive phylogenetic lineages. The echinocandin NRPSs, along with the NRPS from the inp gene cluster in Aspergillus nidulans and its orthologs, comprise a novel lineage among fungal NRPSs. NRPS adenylation domains from different species exhibited a one-to-one correspondence between modules and amino acid specificity that is consistent with models of tandem duplication and subfunctionalization. Pathway gene trees and Ascomycota phylogenies are congruent and consistent with the hypothesis that the echinocandin gene clusters have a common origin. The disjunct Eurotiomycete-Leotiomycete distribution appears to be consistent with a scenario of vertical descent accompanied by incomplete lineage sorting and loss of the clusters from most lineages of the Ascomycota. We present evidence for a single evolutionary origin of the echinocandin family of gene clusters and a progression of structural diversification in two fungal classes that diverged approximately 290 to 390 million years ago. Lineage-specific gene cluster evolution driven by selection of new chemotypes contributed to diversification of the molecular functionalities.

INTRODUCTION

Microbial secondary metabolites are usually biosynthesized by enzymes encoded by gene megacomplexes that can exceed 100 kb. Besides the core catalytic enzymes that assemble monomeric units, these complexes may include genes involved in precursor biosynthesis, regulation, resistance, and transport. Mapping the phylogenetic distributions of biosynthetic pathway genes across fungal genomes has provided new insights into the origins of these gene clusters and how they evolved to elaborate the chemical diversity so characteristic of higher fungi, while correlation of these patterns with life histories has offered clues about their natural functions (1–4).

The echinocandin lipopeptide family (Table 1 and Fig. 1 and 2) has been investigated for 4 decades and forms the molecular platform for a successful class of cell wall-active antifungal drugs (8–12). Their mode of action is the noncompetitive inhibition of the β-1,3-glucan synthase complex; thus, they interrupt synthesis of a primary cell wall structural polymer. Two of the variable features of the echinocandin family, acyl side chain methylation and homotyrosine sulfation (10, 11), critically shaped the development paths to the final drugs. Palmitoyl or linoleoyl linear side chains, such as those of echinocandin B and FR901379, were significantly more hemolytic than the 10,12-methyl myristoyl branched side chain of pneumocandin B₀ and necessitated strategies for side chain deacylation and synthetic replacement. The discovery and development of sulfated echinocandins at Astellas (11, 13) opened up new patent space and provided a starting molecule approximately 10-fold more water soluble than either of the echinocandin B or pneumocandin B₀ drug precursors.

TABLE 1.

Principal echinocandins, the organisms that produce them, acyl side chains, amino acids in positions 1 to 6, and accession numbers for genomes or gene clusters

Primary product of echinocandin pathways	Strain analyzed	Family classification	Acyl side chain	Amino acid in position:						Genome or gene cluster accession no.
				1	2	3	4	5	6
Echinocandin B	Aspergillus pachycristatusa NRRL 11440 (ATCC 58397)	Aspergillaceae	Linoleic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,4S-Dihydroxy-l-homo-Tyr	l-Thr	3S-Hydroxy-4S-methyl-l-Pro	JX421684
Echinocandin B	Aspergillus nidulans var. echinulatus NRRL 3860	Aspergillaceae	Linoleic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,4S-Dihydroxy-l-homo-Tyr	l-Thr	3S-Hydroxy-4S-methyl-l-Pro	AB720074
Mulundocandin	Aspergillus mulundensisb DSMZ 5745	Aspergillaceae	12-Methylmyristic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,4S-Dihydroxy-l-homo-Tyr	l-Ser	3S-Hydroxy-4S-methyl-l-Pro	KP742486
Aculeacin A	Aspergillus aculeatus ATCC 16872 (NRRL 5094)	Aspergillaceae	Palmitic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,4S-Dihydroxy-l-homo-Tyr	l-Thr	3S-Hydroxy-4S-methyl-l-Pro	JGI ATCC 16872 v1.1
Pneumocandin A0	Glarea lozoyensis ATCC 20868	Helotiaceae	10,12-Dimethylmyristic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,4S-Dihydroxy-l-homo-Tyr	3R-Hydroxy-l-Gln	3S-Hydroxy-4S-methyl-l-Pro	ALVE00000000
FR901379 (WF11899A)	Coleophoma empetri FERM BP 6252	Dermataceae	Palmitic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,7-Dihydroxy-l-homo-Tyr-7-O-sulfate	3R-Hydroxy-l-Gln	3S-Hydroxy-4S-methyl-l-Pro	AB723722, AB720725
FR209602	Coleophoma crateriformis FERM BP 5796	Dermataceae	Palmitic acid	4R,5R-Dihydroxy-l-Orn	l-Ser	4R-Hydroxy-l-Pro	3S,4S,7-Trihydroxy-l-homo-Tyr-7-O-sulfate	3R-Hydroxy-l-Gln	3S-Hydroxy-4S-methyl-l-Pro	AB720076
FR190293	Phialophora cf. hyalinac FERM BP 5553	Heloltiales, family unknowna	10,12-Dimethylmyristic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,7-Dihydroxy-l-homo-Tyr-7-O-sulfate	3R-Hydroxy-l-Gln	4R,5R-Dihydroxy-l-Orn	AB720726
FR227673d	Chalara sp.	Helotiales, family unknown	12,14-Dimethylpalmitic acid	4R,5R-Dihydroxy-l-Orn	l-Thr	4R-Hydroxy-l-Pro	3S,7-Dihydroxy-l-homo-Tyr-7-O-sulfate	3R-Hydroxy-l-Gln	4R,5R-Dihydroxy-l-Orn

^aPreviously incorrectly designated A. nidulans var. roseus and A. rugulosus (5).

^bPreviously misidentified as a variety of A. sydowii (6). Phylogenetic and morphological analyses indicate that the strain represents an undescribed species of Aspergillus sect. Nidulantes.

^cPreviously misidentified as Pochonia parasitica (= Tolypocladium parasiticum) (7). The familial and generic placement of this fungus will require revision because the type species of Phialophora belongs to the Chaetothyriales. The Latin abbreviation cf. (confer) signifies possible identity, or a significant resemblance, between a specimen at hand and a known species or taxon.

^dNot studied; included here for comparative purposes. See reference 7.

FIG 1.

Schematic representation of steps in echinocandin biosynthesis as exemplified by proposed pneumocandin biosynthetic pathway in Glarea lozoyensis (18). KS, ketosynthase domain; AT, acyltransferase domain; DH, dehydratase domain; MT, methyltransferase; ER, enoylreductase domain; KR, ketoreductase domain; ACP, acyl carrier protein; A, adenylation domain; T, thiolation domain; C, condensation domain; CT, terminal condensation domain. All reactions have undergone some degree of functional analysis, except for GlOXY3 (?), which has been inferred because its gene is present only in Leotiomycete clusters encoding echinocandins with hydroxy-glutamine in their peptide core. The numbering of the adenylation domains of GLNRPS4 follows the order of peptide assembly.

FIG 2.

Gene cluster schematic illustrating comparative organization, gene order, and direction of transcription of the principal echinocandin gene clusters. Gene functions are color coded. Genes in black are of unknown function or believed not to participate in the biosynthesis. Principal known pathway products are drawn to the right.

Until recently, echinocandin biosynthesis had been inferred from the patterns of incorporation of eight different biosynthetic precursor molecules into the hexapeptide core and acyl side chain of the pneumocandins and subsequent decorations with hydroxyl groups (14, 15). Recently, biosynthetic schemes have been proposed for echinocandin B (16, 17) and for the pneumocandins (18) (Fig. 1). The echinocandin biosynthetic gene cluster (ecd; Fig. 2) from Aspergillus pachycristatus (previously identified as A. rugulosus [5]) spans 12 genes (ecdA to ecdL) (16, 17). The pneumocandin gene cluster from Glarea lozoyensis (Fig. 1 and 2) spans about 66 kb (from GLP450-1 to GLHYP). Both gene clusters have been functionally characterized to a large degree, and the corresponding enzymes have equivalent functions. In both, maturation requires a lipoinitiation step, peptide chain elongation and cyclization, and a cascade of hydroxylation reactions (17–20) resulting in six (echinocandin) or seven (pneumocandin) hydroxyl groups on five amino acid residues of the cyclic hexapeptide core. Additional differences and similarities between the two pathways have been detailed previously (19–22). The genomic sequencing of representative strains of most of the echinocandin variants (Fig. 2 and Table 1), including the parent wild-type strains of the fungi employed for production of the antifungal drugs caspofungin, micafungin, and anidulafungin, has enabled characterization of the basic genomic architecture of the echinocandin pathways (21). Thus, a global perspective of how pathway genes interact to produce different forms of lipoinitiation, biosynthesis of nonproteinogenic amino acids, amino acid selection and hydroxylation, and sulfation of the core peptide is now evident.

Few nonribosomal peptide synthetases (NRPSs) are fully conserved across fungi of the Ascomycota (23, 24). The high diversity of domain structures and the patchy distribution of equivalent adenylation (A) domains complicate ortholog recognition among fungal species. Consequently, the commonalities of orthologs between the pneumocandin and echinocandin B pathways stand out against this background despite rearrangements in their gene order. DNA sequences for additional echinocandin-type gene clusters have been extracted from sequenced genomes of echinocandin-producing fungi (Table 1). These gene clusters include those for echinocandin B from A. nidulans var. echinulatus (NRRL 3860); aculeacins (25, 26) from A. aculeatus (ATCC 16872); a sulfated homotyrosine echinocandin variant, FR901379 (WF11899A) from Coleophoma empetri (27), which is the starting molecule for micafungin; FR209602, a sulfated echinocandin from C. crateriformis (28); FR190293, a sulfated echinocandin from a fungus previously misidentified as Pochonia parasitica (= Tolypocladium parasiticum) (7); and mulundocandin, an echinocandin with a 12-methyl myristoyl side chain from a fungus of Aspergillus section (sect.) Nidulantes, previously misidentified as a variant of A. sydowii (6, 29, 30). Access to the gene clusters responsible for almost the entire range of echinocandin complexity (Fig. 1) affords a framework to develop and test hypotheses about gene recruitments or losses, about the evolutionary history of the echinocandin-pneumocandin gene clusters, and about the significance of these potent cell wall-modifying metabolites to the fungi that produce them.

Multiple evolutionary processes have caused phylogenetic discontinuities and variations in secondary-metabolite pathway organization (31, 32). Several studies have shown that the A, thiolation (T), and condensation (C) modules from a single family of NRPS metabolites often form monophyletic lineages; thus, tandem duplication of modules has been proposed as a mechanism by which multimodular synthetases arise (23, 24). Moreover, other mechanisms, including gene conversion, recruitment, neofunctionalization and recombination, and pathway fusions, could contribute to the radiation and evolution of new chemotypes from conserved gene cluster architectures (33). The functional similarities among the different echinocandin chemistries in combination with the conserved architecture of their physically linked core genes (Fig. 1) suggested the hypothesis that the echinocandin NRPS genes may have arisen only once, possibly through tandem gene duplication and recruitment (21). Subsequently, selection pressures and pathway reorganizations acting on a beneficial suite of pathway elements likely have driven further functional chemical evolution.

In this report, we have sought to (i) clarify the phylogenetic relationships among echinocandin-producing fungi and identify orthologs in pathway genes across these fungi; (ii) analyze the structural variation of catalytic domains encoded by the core NRPSs, the four-gene homotyrosine biosynthetic cassette (hty), and auxiliary genes by phylogenetic analysis; (iii) characterize key positions in NRPS A domains that may determine amino acid substrate specificity; (iv) infer the chemical logic of structure diversity from comparative gene cluster content analyses; and (v) address the hypothesis that these pathways shared a single ancestral pathway. We demonstrated that the echinocandin gene clusters comprise novel lineage of fungal NRPSs and have an evolutionary pattern consistent with vertical inheritance. Their unusual disjunct Leotiomycete-Eurotiomycete distribution is likely the result of incomplete lineage sorting (34) and multiple losses of the gene cluster driven by strong purifying selection. Knowledge of these pathways will contribute to genome mining for fungal secondary metabolites by increasing accurate prediction of protein function from sequences and will provide insights into how gene cluster evolution has shaped secondary metabolism in the Fungi.

MATERIALS AND METHODS

Sampling of strains, genes, and gene clusters.

Strains of historically significant echinocandin-producing fungi were obtained from the NRRL, ATCC, and DSMZ and the NITE Patent Microorganisms Depositary (Table 1).

For A. aculeatus ATCC 16872, A. rugulosus NRRL 8113, A. pachycristatus NRRL 11440, A. mulundensis DSMZ 5745, C. crateriformis BP 5796, C. empetri BP 6252, G. lozoyensis ATCC 20868, and Phialophora cf. hyalina BP 5553, the genomic DNA was extracted as described previously (22) from cultures grown on YM agar (1% malt extract, 0.2% yeast extract, 2% agar) for 2 weeks. One-Taq DNA polymerase (New England BioLabs, MA, USA) was used for PCRs to amplify six gene regions: the 18S rRNA gene (primers NS1 and NS8), the internal transcribed spacer (ITS) RNA gene region and the 28S rRNA gene (primers ITS1 and LR7), the β-tubulin gene (primers Bt2a and Bt2b), the translation elongation factor 1-α (EF1-α) gene (primers 983F and 2218R), and the RNA polymerase II second-largest-subunit (RPB2) gene (primers fRPB2-5F and fRPB2-7cR). The PCR primers can be found at http://www.aftol.org/primers.php. The genomic sequence of G. lozoyensis (ALVE00000000) is available from the NCBI. The genomic sequence of A. aculeatus 16872 is available from the JGI. Nucleotide sequences for the echinocandin gene clusters and the flanking regions for the fungi producing FR901379, FR209602, and FR190293 and echinocandin B from A. nidulans var. echinulatus were downloaded from NCBI (Table 1).

The GenBank accession numbers of the echinocandin biosynthetic gene clusters from A. nidulans var. echinulatus, C. crateriformis BP 5796, C. empetri BP 6252, and P. cf. hyalina BP 5553 (a Phialophora species that looks like Phialophora hyalina) were not fully annotated except for the NRPS-encoding gene. The AUGUSTUS (http://bioinf.uni-greifswald.de/augustus/) program was used to annotate the pathway genes by reference to the annotated genome of A. fumigatus. Scaled illustrations of the echinocandin gene clusters and their degree of homology were prepared with Easyfig (35).

Genome sequencing.

In addition to the gene clusters that were available from public databases (Table 1), the genomes of A. mulundensis DSMZ 5745 and P. cf. hyalina BP 5553 were sequenced to obtain additional data. Lyophilized mycelia collected from liquid cultures were ground in liquid nitrogen, and genomic DNA was isolated by using the CTAB protocol (http://1000.fungalgenomes.org/home/wp-content/uploads/2013/02/genomicDNAProtocol-AK0511.pdf). A 180-bp-insertion library and a 5-kb-mated-pair library were constructed for Illumina sequencing, and sequences were determined on an llumina Hiseq2000 V4 sequencing platform. The Illumina sequencing reads were assembled using the current version of Velvet 1.2 (36). Ab initio gene predictions from the genome assembly were made with Augustus (37). Predicted genes were annotated by BLAST searches against UniProt databases (http://www.uniprot.org/). Protein domains were predicted using InterProScan against various domain libraries (superfamily, Pfam, ProDom, and SMART) (http://www.ebi.ac.uk/interpro/). Annotations were also assigned by homology searches against the Gene Ontology (GO), eukaryotic clusters of orthologous groups (KOG), and KEGG (http://www.genome.jp/kegg/) databases.

Recognition of orthologs.

Genes from echinocandin biosynthetic clusters were considered orthologous when they fulfilled the following criteria: (i) genes were colocalized in at least two different species; (ii) the genes were the best reciprocal BLASTN and TBLASTX hits with an E value of less than 10⁻⁸; and (iii) the genes showed 40% or greater amino acid similarities and at least 80% of the amino acids could be aligned to the reference sequence (i.e., G. lozoyensis). Most pathway genes in A. pachycristatus and G. lozoyensis have been functionally characterized; orthologs have equivalent functions (16, 17, 20, 22).

To determine whether the degree of divergence of echinocandin cluster genes from genes of a Leotiomycete and a Eurotiomycete species pair was equal to or less than that of the divergence in a set of highly conserved housekeeping genes in the same strains, we compared the divergence amino acid sequences from pairs of orthologous echinocandin genes to those of orthologous genes involved in glycolysis and gluconeogenesis from the same pair of strains. The predicted protein sequences of 78 genes from A. aculeatus ATCC 16872 and 71 genes from G. lozoyensis classified as enzymes of the glycolysis and gluconeogenesis by Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology-based annotation (http://www.genome.jp/kegg/ko.html) were downloaded from the Joint Genome Institute's Mycocosm database (http://genome.jgi.doe.gov/programs/fungi/index.jsf). The 71 protein sequences from G. lozoyensis were searched with BLASTP against 78 protein sequences of A. aculeatus. In cases with multiple hits, the highest scoring BLAST hit with coverage of at least >75% and an E value of at least <1.0-e15 was selected as the best ortholog for each gene. The percent amino acid identity score for each ortholog pair was recorded for the 71 hits, and the mean percent identity determined was compared to the mean percent amino acid identity of the 10 echinocandin gene orthologs by means of a two-tailed unpaired t test.

Prediction of adenylation domain specificity.

Because the A-domain structure determines the specific amino acids incorporated during module elongation, a critical step in predicting the A domain is identification of the essential binding-pocket residues that correlate with the amino acid structure. Key positions of the A-domain binding pockets were determined with NRPSpredictor2 (38).

Phylogenetic analysis of strains.

To establish the phylogenetic affinities of the fungi producing FR901379, FR209602, and FR190293, the ITS region and the D1-D2 region of the ribosomal DNA (rDNA) 28S gene were resampled from previous phylogenetic studies of the Leotiomycetes (12, 39–42) and aligned with ClustalW implemented in MEGA 6.0. The best-fit nucleotide substitution models were determined for each alignment based on the lowest Bayesian information criterion scores. Positions containing gaps and missing data were eliminated. Models for each alignment were applied to construct phylogenies using the maximum likelihood (ML) method with MEGA 6.0.

Phylogenetic hypothesis testing.

Phylogenetic trees for all the enzymes in the echinocandin pathway were explored in the context of equivalent functional enzymes from across the Fungi. Proteins encoded by G. lozoyensis and A. pachycristatus echinocandin clusters were used as a query in BLAST searches against NCBI and JGI databases. Each set of proteins was aligned by using M-Coffee (43), and the resulting alignment was manually curated. We inferred the phylogenies by using the ML method in RAxML BlackBox (44) under a WAG plus GAMMA model. Bootstrap supports were calculated using the default options in RAxML BlackBox with 100 replicates per run. For brevity, pruned subtrees are illustrated as described in Results (see Fig. 6). For comparative phylogenetic views of the functional enzyme types across the Fungi, the full trees are illustrated in the supplemental material.

FIG 6.

Abbreviated trees for maximum likelihood phylogenies inferred from genes in the echinocandin biosynthetic gene cluster. Each of the echinocandin biosynthetic-related genes forms a monophyletic lineage. (A) Phylogeny of acyl-AMP and AMP-dependent ligases. (B) Phylogeny of ABC transporters. (C) Phylogeny of 2-isopropylmalate synthase. (D) Phylogeny of d-amino acid aminotransferases. (E) Phylogeny of isopropylmalate dehydrogenases. (F) Phylogeny of aconitases. (G and H) Phylogeny of cytochrome P450s. (I to L) Phylogeny of nonheme iron, α-ketoglutarate-dependent oxygenases. Numbers at nodes are likelihood bootstrap support values. Dotted lines are aids to connect branch tips to strain name. Full phylogenies for each of the functional protein families listed above are illustrated in Fig. S8 to S15 in the supplemental material.

To test whether the phylogenies of the echinocandin pathway genes were congruent with the current classification of the Ascomycota, we built a reference tree from a combined six-gene dataset, including the DNA fragments of the small-subunit (SSU) (18S) rRNA gene, the large-subunit (28S) rRNA gene, the ITS RNA gene region, the β-tubulin gene, the EF1-α gene, and the RPB2 gene that corresponds to the echinocandin-producing fungi and a selection of genome-sequenced Ascomycota. The concatenated gene sequences for each species were aligned with ClustalW implemented in MEGA 6.0 and were analyzed with RAxML BlackBox by the ML method using a GTR-plus-GAMMA model (44). Ten pathway enzymes were found to be common among the seven echinocandin-producing fungi. The amino acid sequences of the 10 enzymes (NRPS, acyl AMP-dependent ligase, ABC transporter, oxygenase 1, oxygenase 2, oxygenase 4, P450-2, isopropylmalate dehydrogenase, 2-isopropylmalate synthase, and aconitase) in the G. lozoyensis echinocandin pathway were used as queries to search for similar sequences in fungal genomes in the GenBank nonredundant database. The selected genomes are listed in Table S2 in the supplemental material. For NRPSs, the cutoff values were less than 80% coverage and at least 24% identity. To select sequences from the other enzyme families, the cutoff values were at least 80% coverage and at least 40% identity. Among the hits, only the sequences with the highest maximum score were selected from each genome. No sequences met the cutoff for oxygenase 1 and P450-2. The presence of these enzymes was mapped onto the Ascomycota reference phylogeny (see Fig. 7A).

FIG 7.

Maximum likelihood phylogenies of fungal species and the enzymes of the echinocandin pathway. (A) Phylogenetic reconstruction of the echinocandin-producing fungi and other Ascomycetes using maximum likelihood analysis of a six-gene dataset consisting of DNA fragments of the 18S rRNA gene, the 28S rRNA gene, the ITS RNA region, the β-tubulin gene, the translation elongation factor 1-α gene, and the RNA polymerase II subunit 2 gene. The enzymes common to all seven echinocandin-producing fungi are color coded and mapped at the right; saturated colors represent the enzymes in the echinocandin pathway that were consistently grouped in the same clade. Homology to NRPS enzymes was based on ≥80% coverage and ≥24% identity. Homology to other enzymes was based on ≥80% coverage and ≥40% identity. Faded colors represent enzymes with substantial sequence similarity that fall outside the echinocandin gene clades. (B) Consensus phylogenetic pattern for gene tree for 10 enzymes in the echinocandin pathway (left) and the established phylogenetic tree extracted from the data in panel A. Trees are rooted at the midpoint. Numbers at nodes are likelihood bootstrap support values. The 10-cluster enzyme tree did not significantly conflict with the species tree (Shimodaira-Hasegawa test, P = 1; weighted Shimodaira-Hasegawa test, P = 1).

In a second analysis (see Fig. 7B), the concatenated phylogenetic marker gene sequences for the seven echinocandin-producing fungi were aligned separately with ClustalW implemented in MEGA 6.0 and analyzed with MEGA 6.0 by the ML algorithm using a TN93+G+I model. To construct a phylogenetic tree for the enzymes in the echinocandin pathway, a combined 10-cluster enzyme dataset, including NRPS, AMP-dependent coenzyme A (CoA) ligase, ABC transporter, oxygenase 1, oxygenase 2, oxygenase 4, P450-2, isopropylmalate dehydrogenase, 2-isopropylmalate synthase, and aconitase, was used. The concatenated amino acid sequences of these enzymes were aligned with MUSCLE and analyzed with MEGA 6.0 using the ML algorithm an a JTT+F+G model with or without a constraint from the reference fungal phylogeny of the seven echinocandin-producing fungi. Alternative hypotheses based on the tree topologies assessed under the null hypothesis that all topologies were equally good explanations of the data were tested with the Shimodaira-Hasegawa test (45) and with a weighted Shimodaira-Hasegawa test as implemented in TREEFINDER (46).

RESULTS

Phylogenetic affinities of the echinocandin-producing fungi.

The phylogenetic positions of G. lozoyensis (12, 22), A. pachycristatus NRRL 11440 (5), and A. aculeatus ATCC 16872 (47) have been confirmed previously. Because strains of Aspergillus sect. Nidulantes are morphologically distinct, we have accepted that the first reported echinocandin B-producing fungus identified as Aspergillus nidulans var. echinulatus (48) (Table 1) is a bona fide member of the Nidulantes group (48). Aspergillus rugulosus NRRL 3860 (Table 1) was sampled to obtain a set of phylogenetic marker sequences from a known echinocandin B producer. The fungus producing mulundocandin was originally designated A. sydowii var. mulundensis (6). Subsequent analysis of its ITS region rDNA indicated that it belonged to Aspergillus sect. Nidulantes (47). Examination of its morphology and comparison of its phylogenetic marker sequences extracted from the genome sequence to those of other species of Aspergillus sect. Nidulantes (5) indicated that it is likely an undescribed species (Jens Frisvad, personal communication). Until it can be formerly named, we will simply refer to it as “A. mulundensis.”

The phylogenetic affinities of the fungi producing FR901379, FR209602, and FR190293, discovered at the Fujisawa Pharmaceutical Co., Ltd. (now Astellas Pharma, Inc.), including the strain (i.e., BP 6252) responsible for producing the starting molecule for micafungin, have not yet been established with phylogenetic marker sequences. Importantly, the identification of strain BP 5553 producing FR190293 as Pochonia parasitica (7) would signify that echinocandin-type gene clusters were present in the Clavicipitaceae of the Sordariomycetes (49) and that the echinocandin gene cluster was dispersed among three major lineages of the Ascomycota. To confirm or reject this possibility, phylogenetic reconstructions and a morphological analysis tested whether strain BP 5553 was correctly identified. Strain BP 5553 exhibited a series of morphological features (see Fig. S2 in the supplemental material) that were uncharacteristic of the genera Pochonia or Tolypocladium (49, 50). Strain BP 6252 grew as a slowly growing melanized mycelium and formed open sporodochial conidiomata with cylindrical, hyaline conidia that were consistent with its identification as Coleophoma empetri (39). Likewise, strain BP 5796, identified as C. crateriformis, grew as a slowly growing melanized mycelium that was very similar to BP 6252, but this strain did not sporulate under the tested conditions.

Homology sequence searches with the ribosomal SSU, ITS, and large-subunit sequences from BP 5553, BP 6252, and BP 5796 consistently retrieved sequences of fungi from the Helotiales. Data from previous phylogenetic studies on the Leotiomycetes were resampled to build alignments of the partial 28S and ITS rDNA genes. ML trees constructed from ITS and 28S region alignments clearly placed BP 6252 and BP 5796 in a robustly supported clade with other strains of Coleophoma, including C. empetri, C. proteae, and C. eucalyptorum of the Helotiales and Dermataceae (see Fig. S3 and S4 in the supplemental material). BP 5553 was not a member of the Clavicipitaceae and was consistently placed in a well-supported clade with Phialophora hyalina, a soil fungus described from Germany (51) (see Fig. S3 and S4). Phialophora hyalina is not congeneric with true Phialophora species of the Chaetothyriales; therefore, since the appropriate generic disposition of P. hyalina has not been established (40), we refer to BP 5553 as P. cf. hyalina. In summary, we concluded that all echinocandin-producing fungi reported to date either belong to the Aspergillaceae of the Eurotiomycetes or to the Helotiales of the Leotiomycetes. To our knowledge, no other secondary-metabolite family exhibits such a disjunct Eurotiomycete-Leotiomycete pattern.

Structure and gene order conservation of the echinocandin gene clusters.

The functions determined for the predicted enzymes in the echinocandin pathways (16, 17, 22) were consistent with the presence or absence of specific genes in each strain and their corresponding pathway products (Fig. 1 and 2). Genes in pathways lacking experimentally determined roles could be linked to hypothesized biosynthetic steps based on the functions predicted from their sequences and inferred by their presence or absence in the gene clusters of G. lozoyensis and A. pachycristatus where most pathway genes have been functionally characterized. For example, clusters that incorporate a highly reducing polyketide synthase (PKS) gene (G. lozoyensis and P. cf. hyalina) encode echinocandins with a dimethyl acyl side chain (Fig. 1) (7, 22). Other gene changes responsible for structural differences in products can be inferred, such as in FR209602 from C. crateriformis (Fig. 2), where the single hydroxylation of the homotyrosine C4 is likely due to the loss of a cytochrome P450 gene orthologous to htyF and GLP450-1 (Fig. 1 and 2). Similarly, the absence of an ortholog of GLOXY3, which encodes a non-heme iron, α-ketoglutarate-dependent oxygenase in all Aspergillus echinocandin gene clusters, suggested that the orthologs of GLOXY3 hydroxylate the C3 of hydroxyl-glutamine of the core's fifth position in the strains of the Leotiomycetes (Table 1 and Fig. 1 and 2). Finally, orthologs of GLHYP and ecdJ were in all echinocandin gene clusters (Fig. 2). Their consistent presence suggested an important but still unknown function. In summary, the minimal elements needed to biosynthesize an antifungal echinocandin would include a six-module NRPS, an acyl-AMP ligase, enzymes HtyA to -D of the hty pathway to synthesize l-homotyrosine, and the four oxygenases equivalent to GLOXY2, GLoF (= GLOXY2), EcdG, EcdK, and EcdH, which hydroxylate the third residue l-proline, l-ornithine, l-homotyrosine, and the leucine precursor of the sixth 4-methyl-l-proline residue (17, 19, 20).

The most compact cluster architecture would correspond to that of aculeacin, where the hty genes sit immediately upstream of, and are separated by, two unknown genes and a cytochrome P450 from the core NRPS gene (Fig. 2). About 34.5 kb upstream of the aculeacin NRPS sits a PKS gene with a predicted domain structure of KS-AT-DH-MT-KR-TD, but it bears little resemblance to GLPKS4, and alignment of its predicted KS domain with those from known PKSs indicates that it corresponds to a fungal PKS-NRPS hybrid gene (Fig. 2). In contrast, the most elaborate pathway is assumed to be that of FR190293 (Fig. 2, P. cf. hyalina) that recruits a highly reducing PKS for the side chain synthesis and where the homotyrosine residue is sulfated. Predicted protein sequence similarities of the core NRPSs ranged from nearly 100% similarity between the two most similar clusters, the two Coleophoma species, and the two echinocandin B NRPSs from the different species of the A. nidulans complex, while the most dissimilar NRPSs were those of A. aculeatus versus A. pachycristatus and A. nidulans var. echinulatus versus G. lozoyensis (71% similarity) (Fig. 3; see also Fig. S1 and Table S1 in the supplemental material). Using the gene cluster from G. lozoyensis as a reference point, a significant difference was evident between the Leotiomycete and Eurotiomycete gene clusters across all pairwise comparisons of the protein and nucleotide sequences of G. lozoyensis and those of the other fungi (Fig. 3; see also Fig. S1 and Table S1).

FIG 3.

Comparison of microsynteny and sequence similarity among echinocandin gene clusters. The gene functions are color coded, and the hty gene module is colored in yellow-orange; other flanking genes are in black. Note that the hty gene module from A. pachycristatus is excluded from the analysis because its physical location is uncertain. The gene cluster from G. lozoyensis was arbitrarily designated the reference sequence. Genes and intergenic regions are drawn to approximate relative scales. The intensity of the red scale bar indicates the degree of nucleotide similarity.

Analyses of pathway sequence similarities, gene content, and microsynteny also indicated that the Leotiomycete and Eurotiomycete gene clusters differed significantly. Two features found in only some Leotiomycete pathways are a dedicated highly reducing PKS for biosynthesis of a methylated, branched acyl side (G. lozoyensis and P. cf. hyalina) and a homotyrosine residue that is sulfated by an as-yet-unknown mechanism (in Coleophoma spp. and P. cf. hyalina). To date, the enzymes responsible for the sulfation reaction have remained unknown. An aryl sulfotransferase gene encoding an enzyme that might sulfate the homotyrosine residue is not evident in any gene cluster. Nonetheless, bioinformatic analysis of the genome of P. cf. hyalina BP5553 identified a putative aryl sulfotransferase gene (g5850) at a genetic locus physically unlinked with its pneumocandin pathway. The predicted protein from this gene has significant sequence similarity with many other putative fungal and bacterial aryl sulfotransferases and exhibits a domain structure similar to those of bacterial aryl sulfotransferases, suggesting that it might be mechanistically related and perhaps employs an alternative sulfate donor instead of PAPS (3′-phosphoadenosine-5′-phosphosulfate) (see Fig. S5 in the supplemental material). No ortholog was found in the genome of G. lozoyensis ATCC 20868, which synthesizes pneumocandins with an unsulfated homotyrosine (Fig. 1 and 2). Therefore, we hypothesize that this enzyme and/or accessory oxygenases may mediate stepwise hydroxylation and sulfation of the pneumocandin homotyrosine residue (52, 53). The 12-methyl myristoyl side chain of mulundocandin raised a question as to whether it could be synthesized by a PKS as in the pneumocandins. BLAST searches of the A. mulundensis DSMZ 5745 genome with nucleotide or predicted amino acid sequences of GLPKS4 retrieved no significant hits, suggesting that the A. mulundensis gene cluster lacked an accessory PKS. The microsynteny of the mulundocandin gene cluster was essentially identical to that of A. nidulans var. echinulatus (NRRL 3860). Thus, the branched fatty acid side chain might result from monomethyl branched-chain fatty acid biosynthesis that incorporates 3-methylbutryl CoA as a starter unit instead of the usual acetate acyl primer or it might be the product of a highly reducing PKS elsewhere in the genome.

Examination of the genomic regions of ∼3 to ∼20 kb of DNA flanking each side of the echinocandin clusters indicated that the echinocandin-flanking regions of the different species differed in all cases, except for the two Coleophoma species, and, therefore, that the clusters are located in different genomic contexts (Fig. 1). Gene cluster microsynteny results were identical in the two Coleophoma species and in A. mulundensis and A. nidulans var. echinulatus (Fig. 1 and 2). They also shared similar genome contexts in that the downstream regions of the two Coleophoma species were essentially identical except for the cytochrome P450-encoding gene associated with C4 hydroxylation of l-homotyrosine that is missing in the C. crateriformis cluster. However, the region upstream of the echinocandin cluster of C. empetri was occupied by a large intergenic region while the corresponding position in C. crateriformis was occupied by hypothetical proteins.

Evolution of the echinocandin NRPS.

We tested whether echinocandin NRPSs were descended from bacterial NRPSs or from the common ancestors of fungal NRPSs. BLAST and HMMER (http://hmmer.janelia.org) searches across the NCBI and the JGI databases (54) performed with GLNRPS4 and EcdA retrieved only homologs of echinocandin NRPSs and other fungal NRPSs and no bacterial NRPSs (Table 1 and Fig. 3; see also Fig. S6 in the supplemental material). BLAST searches with individual A domains of EcdA and GLNRPS4 also indicated that an uncharacterized dipeptide NRPS, ANID_03496 (InpB) from A. nidulans (31, 32), and its orthologs Aspsy1 12236 in A. sydowii, Aspve1 13511 in A. versicolor, and g10398 in A. mulundensis were related. More-targeted BLAST searches performed with individual catalytic domains with the A, C, and T domains of the NRPSs retrieved only other Ascomycota NRPS domains. These searches, plus the fact that each of the echinocandin NRPSs has several introns, suggested that the echinocandin NRPSs, like other large fungal NRPSs, likely originated by module duplication during the radiation of the NRPSs in the Ascomycota and were not descendants of bacterial NRPSs (24).

The issue of whether the narrow and peculiar disjunct distributions of echinocandin gene clusters could be the result of incomplete inheritance or widespread inheritance followed by multiple gene losses in multiple Ascomycota lineages was examined by the same bioinformatic approach. In the absence of strong positive-selection pressure, strains may lose the gene cluster because of its energetic costs or the intrinsic toxicity of echinocandins (55). Under either of these hypotheses, we might have expected to encounter evidence of cluster degradation or to find at least a few cluster fragments or pseudogenes interspersed across the genomes of major secondary-metabolite-producing sister lineages of the Sordariomycetes, Dothideomycetes, and Lecanoromycetes or in other orders of the Leotiomycetes and Eurotiomycetes beyond the Helotiales and Eurotiales. However, in contrast to what has been observed in some fungal secondary-metabolite pathways (2, 56–58), this was not the case. BLAST searches using each of the shared pathway genes always retrieved the corresponding genes of other echinocandin-producing species as the most likely hits. BLAST searches with the P450s, acyl-AMP ligases, oxygenases, PKSs, ABC transporters, or NRPSs yielded no evidence that close orthologs of these genes had been incorporated into other pathways, although numerous possible paralogs of these genes were evident. The highest-scoring BLAST hits were always orthologs from echinocandin pathways, with a significant gap with respect to sequence similarity, coverage, and identity to the next nearest BLAST hits. We detected neither pseudogenes nor partial pathway fragments in other fungal genomes, except in two cases. The first case was that of the elements of the inp pathway of A. nidulans (59, 60), which includes NRPS ANID_03496 (InpB) and AMP-binding protein ANID_03490 (InpC) and their close relatives mentioned above and the possibly related acyl-CoA ligase of the emericellamide pathway (see below). However, in both cases, these genes were found in strains of the same section, sect. Nidulantes of Aspergillus, or, in the case of A. mulundensis, all these related genes were found within the same genome. Therefore, the closest reciprocal relatives of the echinocandin NRPS and acyl-AMP ligase genes were always found in echinocandin-producing fungi in the Eurotiales and Helotiales or, in the case of the inp pathway genes, in various strains of Aspergillus sect. Nidulantes. The second case was the four-gene hty cassette. As previously highlighted during the elucidation of the echinocandin B pathway (16), each of the genes of htyA to htyD (htyA-D) was highly similar to a gene in a four-gene cassette from the Alternaria alternata AM (Alternaria mali)-toxin pathway (the amount gene cluster) that is thought to encode synthesis of the α-amino-4-phenyl-valeric acid monomer of AM toxin (61). The highly similar gene sets that encode biosynthetically similar aromatic amino acids suggest an independent origin of the hty gene cluster, which then was recruited into different NRPS gene clusters in different fungal lineages.

In order to infer phylogenies from the NRPS modular structures, the deduced amino acid sequences for all the echinocandin A domains were aligned with predicted A-domain sequences from throughout the Ascomycota by resampling and updating a previously published dataset of fungal NRPS A domains designated the Euascomycete clade synthetase (EAS) subfamily (24). A ML tree of these domains (Fig. 4) reproduced many of topological features evident in this prior analysis of the EAS NRPSs. Inspection of the ML tree indicated that all the echinocandin A domains and the A domains of the predicted dipeptides NRPS ANID_03496 (InpB) and its orthologs Aspsy1 12236, Aspve1 13511, and g398 from A. mulundensis formed a distinctive clade (57% bootstrap value). The echinocandin A-domain clade clearly nested within the EAS subfamily of NRPSs (24) (Fig. 4). The same previous analysis (24) detected that the ANID_03496 NRPS (InpB) of A. nidulans formed a weakly supported single terminal branch and was a sister group to the A domains of the emericellamide NRPS (eas; ANID_2545) (62). Thus, the echinocandin NRPSs along with InpB and its orthologs formed a novel lineage within the EAS subfamily of fungal NRPSs (Fig. 4 and 5) (21).

FIG 4.

Phylogenetic analysis of Ascomycota NRPS A domains. Maximum likelihood phylogeny was inferred from A domains of the Euascomycete NRPS subfamily (24), including the echinocandin A domains; the blue region indicates the echinocandin A domain clade. The tree is rooted with fungal L-δ-(α-aminoadipoyl)-l-cysteinyl–d-valine (ACV) synthetases. Numbers at nodes are likelihood bootstrap support values. A. jesenskae, Alternaria jesenskae; B. fuckeliana, Botryotinia fuckeliana; C. carbonum, Cochliobolus carbonum; C. purpurea, Claviceps purpurea; E. festucae, Epichloë festucae; F. oxysporum, Fusarium oxysporum; M. oryzae, Magnaporthe oryzae; M. anisopliae, Metarhizium anisopliae; T. stipitatus, Talaromyces stipitatus.

FIG 5.

Expanded view of the echinocandin A domain clade from Fig. 4. The corresponding amino acids (AA) that are activated by each subclade of A domains are indicated to the right. Numbers at nodes are likelihood bootstrap support values.

The echinocandin A domains formed a monophyletic clade, and the six individual A domains exhibited a highly resolved intraclade relationship (Fig. 5). Remarkably, the clade's distal topology was resolved into six subclades, each corresponding to one of the six amino acid positions in the echinocandin nucleus (Fig. 5). This topology was supported by strong bootstrap values and indicated that the A domains for each amino acid position from each distantly related fungus were more similar to each other than to the individual A domains from a single given echinocandin-producing strain. This pattern of a one-to-one correspondence between modules and the amino acid specificity of different species is consistent with models of tandem duplication and subfunctionalization prior to divergence from a hypothetical ancestor, as has been hypothesized to occur during the evolution of cyclosporine and peptaibol NRPSs (24, 63). Furthermore, within each of these A-domain clades, the branchings of terminal leaves were congruent with the phylogenetic species trees. The relationships among the echinocandin NRPS genes were further tested by constructing ML trees of concatenated T and C domains (see Fig. S7 in the supplemental material). The ML phylogeny inferred from this dataset again yielded a monophyletic tree with essentially the same topology as the A-domain tree, with the exception that the T and C domains of InpB fell outside the echinocandin clade. These analyses strongly supported the hypothesis that present-day echinocandin NRPSs are descendants of a common ancestral peptide synthetase (Fig. 4 and 5; see also Fig. S7).

Prediction of A-domain amino acid binding specificity.

We predicted key positions in A-domain binding pockets of the echinocandin and InpB (ANID_03496) NRPSs with NRPSPredictor2 (38) (Table 2). We found that the sequence signatures were constant for module 2 (l-threonine; DAQTAVAIHK), module 3 (4R-hydroxy-l-proline; DVSSATTVCK), and module 6 (3S-Hydroxy-4S-methyl-l-proline; DNTMITAMSK). In module 1, even though 4R,5R-dihydroxy-l-ornithine was an invariable core amino acid, positions 239, 299, and 331 were distinct for the echinocandin B NRPSs. A similar pattern was repeated in module 4, where valine was substituted in position 330 in echinocandin B NRPSs (Table 2). In module 2, a substitution of isoleucine for alanine in position 239 resulted in a substitution of serine in the core peptide produced by C. crateriformis (Table 2). In module 5, key positions varied according to amino acid selection (3R-hydroxy-l-glutamine [DAQNIASINK], l-threonine [DAQTIVAIHK], or l-serine [DVQTIVAIHK]). A correlation between the sequence signatures and the overall A-domain phylogeny was not apparent (Fig. 5). Other amino acids in the vicinity of the signature positions may possibly be critical for amino acid selection. Furthermore, the variability of positions 239, 299, and 331 in the A1 domain indicated the other seven positions may play a role in binding 4R,5R-dihydroxy-l-ornithine. The signature positions of the A2 domain of C. crateriformis were only marginally similar to those of the A2 domain of GliP of A. fumigatus, also responsible for l-serine incorporation (64). This analysis suggested that the amino acid specificity of echinocandin A domains may be due to a unique compilation of nonribosomal codes which would be consistent with monophyly of their NRPS domains and accompanying pathway genes. The signature positions of the A2 domain of InpB (ANID_03496), in the other members of the echinocandin clade, differed from those of echinocandin A domain 5 by only a single amino acid. Echinocandin A domain 5 is responsible for incorporation of 3R-hydroxy-l-glutamine, which could imply that the InpB A2 domain might incorporate l-glutamine into the as-yet-unknown product of the inp gene cluster.

TABLE 2.

Key positions of the A-domain binding pocket of echinocandins and InpB (ANID_03496) NRPS identified by NRPSPredictor2

^aShaded columns represent variable sites for each A domain.

^bData represent amino acid specificity.

Data mining and phylogenetic analysis of other genes.

To test whether the gene cluster evolved as an integral functional unit or whether the gene clusters had assembled independently multiple times from alternative gene sources, we analyzed the other pathway enzymes, including the acyl-AMP ligase, the cytochrome P450s, the oxygenases, the ABC transporter, and the hty pathway genes that were common to all or most of the echinocandin gene clusters (Fig. 6; see also S8 to S15 in the supplemental material). In the scenario of inheritance of the gene clusters as an integral unit, each of these genes from the echinocandin-producing fungi would form a monophyletic clade among other fungal genes from the same functional family. Alternatively, if new pathway genes were recruited multiple times during convergent evolution to achieve equivalent reactions, then we would expect to see these genes dispersed among the gene family trees. The predicted protein sequences of these genes were aligned with other protein sequences from the same functional families. ML trees were constructed for each functional gene type, i.e., genes encoding AMP-dependent ligases, ABC transporters, cytochrome P450s, oxygenases, and the individual proteins of the hty pathway (Fig. 6; see also Fig. S8 to S15).

In each case, the pathway's two cytochrome P450s, the four oxygenases, the ABC transporter, and individual hty pathway genes formed robustly supported (98% to 100% bootstrap values) monophyletic lineages (Fig. 6; see also Fig. S8 to S15 in the supplemental material) and were exclusive of genes from other fungal secondary-metabolite pathways, except for the echinocandin acyl-AMP ligases and the enzymes of the hty pathway. Phylogenetic analysis of the echinocandin acyl-AMP ligases (Fig. 5; see also Fig. S10) showed that the putative AMP-binding enzyme, InpC (ANID_03490), encoded by the A. nidulans NRPS inp gene cluster (59, 60), was a closely related sister lineage and formed a monophyletic lineage with the echinocandin acyl-AMP ligases (98% bootstrap value). The InpC gene clade included orthologs from A. mulundensis, A. sydowii, and A. versicolor. This analysis also demonstrated that emericellamide acyl-CoA ligase and other AMP-binding enzymes from Pestalotiopsis fici and some Fusarium species possibly represent sister lineages (see Fig. S10). Likewise, each gene of the hty biosynthetic cassette formed an echinocandin-exclusive subclade (100% bootstrap values) with a larger highly supported clade of similar genes which included orthologous genes from the AM toxin pathway and, in some cases, similar individual genes from Pseudocercospora fijiensis, Zymoseptoria tritici, and Colletotrichum higginsianum (see Fig. S12 to S15).

The monophyletic echinocandin gene lineages exhibited two different topological patterns (Fig. 6), although the internal divergence between genes of the Aspergillaceae and those of the Helotiales was evident in all but one case, indicating significant divergence since the formation of the progenitor pathway. These nine individual gene phylogenies (Fig. 6; see also Fig. S8 to S15 in the supplemental material) were consistent with the hypothesis of a monophyletic origin of the echinocandin gene cluster and their inheritance as a cohesive functional unit. Also, in each case, we used BLAST searches as well as gene classifications to assemble gene sequences to construct the trees, but each of these lineages was exclusive to the echinocandin pathways, without any evidence of phylogenies contaminated with derived pseudogenes or neofunctionalized paralogs. These results strongly argue for vertical inheritance from a common progenitor. The inability to find similar genes or gene cluster fragments is suggestive of a pattern of incomplete lineage sorting (34) and of cluster loss relative to other major clades of the Ascomycota. Almost all of these monophyletic gene trees were congruent with the established species trees of the Ascomycota, i.e., a major separation into Leotiomycete and Eurotiomycete clades with a sorting of individual strains into their expected relative phylogenetic positions (Fig. 4 to 7; see also Fig. S5 to S12). In no case did we detect that pathway genes of one gene lineage were derived from the other. This fact, along with the consistency of these monophyletic gene trees, argues against the possibility of horizontal gene transfer between the Leotiomycetes and Eurotiomycetes.

To further consider the unlikely possibility of horizontal gene transfer, we compared the divergences of echinocandin gene cluster genes between a Leotiomycete and a Eurotiomycete echinocandin producer to the divergence between highly conserved housekeeping genes in the same strains. In the horizontal-gene-transfer scenario, one might expect the sequence similarity of echinocandin pathway genes to be greater than that of conserved primary metabolism genes, which would reflect a gene cluster transfer after the diverging speciation events. To test this idea, we compared the divergence in pairs of orthologous echinocandin genes to the divergence of gene orthologs involved in glycolysis and gluconeogenesis from the same pair of strains. Strains of A. aculeatus and G. lozoyensis were selected for this comparison because the predicted protein models of both genomes have been classified by KEGG orthology annotation and were available for download. A total of 71 gene orthologs corresponding to the glycolysis and gluconeogenesis pathways were recognized. Their mean protein identity level was 53.5% ± 17.1%, while the mean protein identity level between the 10 echinocandin pathway genes was 60.9% ± 6.5%; the means were not significantly different (T = 1.3266, P = 0.188, 79 df). Comparable levels of gene divergences between echinocandin biosynthetic genes and genes involved in essential carbohydrate metabolism in two fungi in the two echinocandin lineages indicated that the origin of the echinocandin gene cluster in a hypothetical progenitor was as ancient as the separation of the Eurotiomycete and Leotiomycete-Sordariomycete lineages.

To visualize the extent of gene cluster dispersion and loss during the evolution and descent of the echinocandin gene clusters among Ascomycota lineages, we estimated a reference phylogeny of Pezizomycotina based on a set of rDNA and conserved gene marker sequences from genome-sequenced strains and mapped the 10 different pathway proteins and their possible orthologs and paralogs onto the tree (see Fig. 7A). As predicted by the BLAST searches and reconstructions of individual genes described above, echinocandin pathway orthologs mapped only onto those strains that were known echinocandin producers or, in the case of Alternaria alternata, that had a gene cassette orthologous to the hty gene cassette. We then removed the non-echinocandin-producing species from the phylogenetic reference alignment and built a new ML tree and compared this tree to a protein sequence ML tree built from the 10 coevolved echinocandin pathway enzymes (see Fig. 7B). The two trees were essentially mirror images with respect to branch nodes, and both showed a deep coalescence that preceded the speciation events. Furthermore, a comparative topology test indicated that the data from the 10-cluster enzyme tree did not significantly conflict with the data from the species tree (Shimodaira-Hasegawa test, P = 1; weighted Shimodaira-Hasegawa test, P = 1).

DISCUSSION

The echinocandin gene clusters identified so far are known to be distributed in two relatively narrow phylogenetic regions in distantly related fungal lineages, Aspergillaceae (Eurotiomycetes) and Helotiales (Leotiomycetes), but are common among strains of these lineages. For example, echinocandin B production is common among strains of various species of the A. nidulans complex (65, 66), and many strains of Pezicula species produce pneumocandins and variants, e.g., cryptocandin and sporiofungin (67–69). Since detection of echinocandins has been bioactivity driven and since multiple productive isolates are normally discarded during antibiotic screening, it is not yet possible to determine how echinocandin gene clusters are distributed at the population level. We did not consider alternative hypotheses of rare and cryptic distribution of the pathway in other Ascomycota because the narrow phylogenetic distribution does not seem to be an artifact caused by undersampling. The perceived distribution is based on three lines of evidence. (i) Discovery of cell wall-active antifungal agents has been a holy grail for infectious disease therapies in various pharmaceutical companies. Fungi in the main classes of culturable secondary-metabolite-producing Ascomycota (Dothideomycetes, Eurotiomycetes, Sordariomycetes, and Leotiomycetes) have been targeted and intensively screened with highly sensitive assays for fungal cell wall-inhibiting natural products for at least 35 years (66, 69–73); echinocandins can be detected at very low concentrations in crude fermentation extracts because of their potency and distinctive chemical and biological signals (65, 74). Detection of echinocandins has so far been limited to strains in the Eurotiomycetes, family Aspergillaceae (Aspergillus sect. Nidulantes and A. aculeatus), Leotiomycetes, families Helotiaceae (G. lozoyensis, Chalara species) and Dermataceae (Pezicula and Coleophoma species), and P. cf. hyalina (as T. parasiticum). (ii) Sequence searches across the still relatively sparse samplings of fungal genomes have yet to retrieve echinocandin-like NRPSs beyond the above-mentioned types of fungi (Fig. 2). (iii) Although homotyrosine is characteristic of certain cyanobacterial secondary metabolites, e.g., the spumigins, lyngbyastatin, microcystins, cyanopeptolins, and anabaenopeptins (75, 76), it is unknown in other fungal NRPs (77, 78), and a dedicated homotyrosine pathway in fungi seems to be unique to echinocandins, as far as we know. However, the orthologous four-gene cassette from the amount gene cluster appears to have originated from a common ancestor. Nonetheless, not all fungi of the Ascomycota have been exhaustively screened for cell wall-active agents, and the results of fungal genome sequencing are hardly comprehensive; therefore, other species of echinocandin-producing fungi and new echinocandin variants may await discovery.

Various hypotheses have been formulated to explain the vast microbial chemical repertoire and propose a reciprocal progression of adaptation and counteradaptation between microbes and their symbionts, vectors, parasites, hosts, and predators that is shaped by mutual selection (79). How and why would biosynthetic pathways in fungi from two major evolutionary lineages, Eurotiomycetes and Leotiomycetes, that diverged approximately 290 to 390 million years ago (80) come to have such similar complex molecular scaffolds? Phylogenetic analysis rejected independent convergent evolution as an explanation. The monophyly of all the pathway genes, including the unique fungal homotyrosine biosynthetic pathway, indicates a common origin of the gene clusters. Furthermore, the phylogenies of the individual cluster genes mirror the phylogeny of the producing strains. Housekeeping genes and the echinocandin genes exhibit about the same levels of divergence between strains of G. lozoyensis and A. aculeatus. Therefore, evidence of horizontal gene transfer is weak. Although claims of horizontal gene transfer of secondary-metabolite gene clusters are becoming increasingly common (81), these claims usually are based on bioinformatic inferences and often result from analysis of gene clusters that have not been functionally characterized. Experimental evidence and mechanistic studies for such claims are still lacking. Other studies of disjunct distributions of fungal secondary-metabolite gene clusters have explained these dispersed distributions by normal genomic processes, including gene duplications and losses (4, 82). Therefore, the phylogenetic pattern underlying echinocandin distribution exhibits hallmarks of incomplete lineage sorting and multiple pathway losses driven by strong selection pressure (34), including environmental factors and interactions with other organisms that have acted on individuals to maintain or degrade the gene cluster (Fig. 8).

FIG 8.

Hypothetical scenario for evolution of echinocandin gene clusters. On the right side, the ancestral elements shown include the hty pathway, an acyl-AMP ligase that is related to the acyl-AMP ligase of the inp and eas (emericellamide) pathways, and the bimodular NRPS inpB (ANID_03496) and its orthologs, which are highly related to echinocandin A domains. Note that the hty gene is physically separated from the core NRPS in A. pachycristatus. The amount gene cluster in Alternaria alternata has a four-gene cassette that shares a common ancestor with the hty pathway. In the Leotiomycetes, a highly reducing PKS for branched side chain biosynthesis was gained in G. lozoyensis and P. cf. hyalina. Other pathway variations that evolved in the Leotiomycete lineage include a mutation in A domain 5 resulting in incorporation of glutamine versus threonine or serine in position 5 in the aspergilli, incorporation of an oxygenase for glutamine hydroxylation, acquisition of a highly reducing PKS, and a biotransformation of homotyrosine by a hypothetical aryl sulfotransferase in some species of the Leotiomycetes.

The echinocandin gene clusters share a conserved architecture and chemical logic that closely tracks variations in chemical structures of pathway products. Gene linkages that are conserved across the echinocandin gene clusters, e.g., the hty subgene cluster, and the combinations of coevolved P450s, oxygenases, anchor NRPSs, and the acyl-AMP ligase genes responsible for the formation of core acylated hexapeptide are universal features (Fig. 2). The pathways can be categorized as those that have recruited a highly reducing PKS to synthesize a pathway-specific branched, acyl side chain and those that apparently activate and incorporate cytosolic fatty acids as acyl side chains (Fig. 2). Furthermore, on the basis of sequence similarities of the core NRPSs and other key pathway genes (acyl-AMP ligase, P450s, oxygenases, and hty pathway genes), the Eurotialean and Leotialean gene clusters appear to have diverged along their respective phylogenetic lineages (Fig. 6). Analyzed in the context of similar fungal genes, each of the individual pathway genes formed a nearly unique and often echinocandin-exclusive phylogenetic lineage (Fig. 5 to 7). This monophyletic pattern for all of the physically linked gene cluster elements suggests that, after the formation of an ancestral pathway, the gene cluster radiated among the descendants in each of the two lineages and the genes continued to evolve as autonomous functional units. We tested the statistical likelihood of a chance contiguous arrangement of these genes in fungi as divergent as Aspergillus spp., Coleophoma spp., and P. cf. hyalina and found it to be nil (Table 2). The monophyly of the pathway genes at first may seem discordant with the established phylogeny of the Ascomycota; however, the branch nodes of the echinocandin clades do not conflict with the species phylogeny, thus supporting the hypothesis that the genes had already formed a cooperative cluster in an ancestral species and have subsequently undergone vertical descent.

It was not possible to infer whether the Eurotiomycete gene cluster might be more similar than the Leotiomycete gene cluster to an extinct gene cluster progenitor. However, elements of the anchor NRPS and the related key catalytic enzyme, i.e., inp NRPS ANID_03496 (Fig. 5) and the AMP-dependent CoA ligase from the emericellamide gene cluster (Fig. 6; see also Fig. S10 in the supplemental material), are present in aspergilli, including A. sydowii and A. versicolor, and all occur together in A. mulundensis; all of those species share a common ancestor with A. nidulans, suggesting the possibility that other building blocks of the ancestral pathway are retained in the Aspergillaceae. The products encoded by inp remain unknown (59), but they eventually may be key to understanding the origin of the ecd gene cluster. We also speculate that the apparent physical separation of the hty gene cluster from the core NRPSs, oxygenases, and P450s observed in A. pachycristatus NRRL 11440 (16) could be evidence that significant genomic rearrangements captured a preexisting homotyrosine biosynthesis gene module that was eventually consolidated into the gene cluster as in A. nidulans var. echinulatus NRRL 3860 and A. mulundensis DSMZ 5745 (Fig. 2). That scenario would also be consistent with the independent incorporation of a hty ancestor into an ancestor of the AM toxin pathway in the Dothideomycete lineage.

It would appear that an ancestral echinocandin pathway evolved by interaction of an AMP-dependent CoA ligase and a six-module NRPS gene followed by cooperation with genes of the hty pathway (Fig. 8). It is also tempting to speculate that some of the tailoring enzymes with similar functions, e.g., the pair of P450s that hydroxylate the C4 of homotyrosine and ornithine and the oxygenases responsible for the C3 homotyrosine and the glutamine (Fig. 1), originated by gene duplication and subfunctionalization. Subsequent recruitment of tailoring enzymes by extinct ancestor genomes further shaped structural conformation and influenced product specificity and solubility. The presence of close gene relatives in related aspergilli, even in the same genome as in A. mulundensis, suggests that the ancestral fungi harboring the first echinocandin gene clusters were more similar to fungi of this genus, possibly in common ancestors of sections Nidulantes and Nigri (83). The ancestral pathway could have been similar to that of echinocandin B or aculeacin B (Fig. 1 and 7). In the Leotiomycete lineage, a simple mutation(s) in A domain 5 (Table 2) resulted in selectivity for glutamine over serine/threonine. The most primitive predicted cluster configuration in a Leotiomycete would result in a molecule like cryptocandin (67), with an unhydroxylated glutamine and palmitoyl side chain. Recruitment of oxygenase gene GLOXY3, the gene encoding glutamine hydroxylation, or its possible duplication of and subfunctionalization from GLOXY1, is apparently restricted to the Leotiomycetes. Coincidental introduction of an echinocandin pathway into some Leotiomycete lineages with enzymatic capacity for homotyrosine sulfation would result in sulfated forms with an unbranched side chain, e.g., FR209602 and FR901379. Also, the recruitment of a highly reducing PKS with a methyl transfer domain seems be a derived feature because dimethyl-myristoyl and dimethyl-palmitoyl side chains (Fig. 1 and 7) are restricted to the Leotiomycete-type echinocandins.

Interestingly, no regulatory gene(s) has been recognized among those in the set of pathway genes. A transcription factor ecdB gene in the echinocandin B is unrelated to the C2H2 zinc finger gene upstream of the pneumocandin gene cluster (16, 22) and was not observed in the vicinity of the other Aspergillus gene clusters. Whether this means that regulatory genes might be elsewhere in the genome or, as suggested by the facile production of echinocandins, whether the pathway is constitutive remains to be investigated. Introgression of a gene set where the proteins are functional and stoichiometrically balanced does not require efficient regulation as part of a larger complex and would offer immediate functionality (i.e., antibiotic production). Nonetheless, this introgression could impose substantial up-front costs for competitive fitness that might be incurred before recruitment of the self-resistance mechanisms, regulators, and auxiliary genes that eventually would enable efficient stoichiometric equilibrium within the cellular network. Such a scenario may have contributed to the limited distribution of the echinocandin gene clusters.

Recruitment, integration, and modularization of multiple biosynthetic steps offer multiple advantages to the fungus, including on-demand coordinated biosynthesis of complex precursors. For example, homotyrosine is essential for bioactivity, and recruitment of the hty pathway adds antifungal functionality to the peptide. Physical integration of the hty cassette into the gene cluster would guarantee coordinated biosynthesis. Likewise, recruitment of a dedicated highly reducing PKS would reduce dependence on the fatty acid pool and would coordinate biosynthesis in an energetically and stoichiometrically efficient manner. The sulfation step may simply represent a coincidental biotransformation reaction absent in the ancestral echinocandin pathway. In vitro sulfation of aromatic phenols has been occasionally reported in the Fungi (84–87). Although the mechanism of the pneumocandin homotyrosine sulfation remains to be elucidated, the putative aryl sulfotransferase identified in P. cf. hyalina BP 5553 might mediate a stepwise oxidation at the meta-position on the homotyrosine residue followed by conjugation with activated sulfate donor, e.g., PAPS (3′-phosphoadenosine 5′-phosphosulfate) or an alternative donor (87, 88). Hydrophilic sulfate conjugates are known to facilitate excretion. Sulfation increases the solubility of the echinocandin product (13) and therefore might increase the effective range of action of echinocandins outside the fungal mycelium.

The hypotheses regarding the maintenance and inheritance of complex antibiotic mega-synthases assume that these metabolites are of adaptive value to the fungi that produce them and that the organisms have mechanisms to cope with self-toxicity (55, 79). Echinocandin-producing fungi differ in their ecologies (8, 12, 66) and range from soil and litter saprobes (Aspergillus species and possibly G. lozoyensis and P. cf. hyalina) to endophytes and weak pathogens of woody plants (Pezicula and Coleophoma species). Echinocandin metabolites target the fungal cell wall, specifically inhibit β-glucan synthesis, and are inherently antifungal in vitro. Fermentations of all these strains were potently active in antifungal screens; therefore, the pathways are operative. In our experience, their biosynthesis is not tightly regulated in the laboratory (12, 65, 89, 90). The fungi also need mechanisms to manage their own toxic metabolites by exporting them from the site of synthesis, by excreting them through the plasma membrane, by transporting them into an internal compartmentalization, or by mechanistic resistance. An obvious resistance mechanism is not apparent in the echinocandin gene clusters other than the possible involvement of the ABC transporter that is present in all clusters (Fig. 2 and 6B). The function of this transporter is unknown, and, at least in G. lozoyensis fermentations, most of in vitro pneumocandin production is mycelium bound and is not abundant in the culture broth. Plausibly, the genes encoding synthesis of such potent cell wall-deforming compounds could be easily lost because of the high metabolic costs and debilitating self-toxicity. Whether these metabolites play a role in the life histories of these fungi is as yet unclear. However, the ability to produce null pathway mutants (16, 18, 22) opens up intriguing possibilities for experimentation in fungal microcosms.

In conclusion, our phylogenetic analyses offer a unified model (Fig. 8) for the classification of echinocandins as either Eurotiomycete- and Leotiomycete-type echinocandins. Nucleotide composition, conserved microsynteny, phylogenetic analyses, and correlations with product structural diversity portray a speculative scenario (Fig. 8) where a prototypical echinocandin pathway arose in a progenitor to the ancestors of Aspergillus spp. and the Leotiomycetes. The predecessor clusters evolved by cooperation followed by physical linkage between an acyl-AMP ligase and a six-module NRPS, possibly a product of internal module duplication and mutation from elements similar to those of the inp gene cluster, accompanied by fusion with an ancestral hty biosynthetic gene cluster and recruitment of accessory oxidases. To date, the closest relatives of the echinocandin gene clusters to have been identified have been the inp gene cluster of A. nidulans and related aspergilli, while an ancestral hty gene cluster has descended independently into the Dothideomycete lineage and has been incorporated as a building block of the amount gene cluster. Separation of the Leotiomycete and aspergillus lineages correlates with mutations at the fifth A domain and substitution of glutamine for threonine or serine, resulting in either a molecule such as cryptocandin or an aculeacin. Additionally, an oxygenase gene for glutamine hydroxylation, a PKS gene for autonomous side chain biosynthesis, and homotyrosine sulfation were gained during the evolution of the Leotiomycete-type echinocandins. The latter two functions are responsible for improved pharmacological properties that led to selection of Leotiomycete variants pneumocandin B₀ and FR901379 as candidates for development of caspofungin and micafungin, respectively. Phylogenetic analysis indicates that the echinocandin gene cluster has evolved by direct descent without paralogs or horizontal gene transfer.