Taxa from four fungal phyla with diverse reproductive structures are sampled in this study (Figure 1). For each taxon, we sequenced at least 3.1 kb of RPB1 (A-G) and 2.7 kb of RPB2 (3–11) (Figure 4). When overlapping multiple PCR products were sequenced from individual taxa, there was no evidence for more than one copy of RPB1 or RPB2 in most of the fungal species examined. However, two RPB1 genes with slightly different sequences were found in Allomyces macrogynus, and two similar RPB2 copies were detected in Glomus mosseae, Neocallimastix frontalis, Allomyces macrogynus and Chytriomyces hyalinus. Two RPB2 sequences also were recovered by a search of the genomic sequence of Rhizopus oryzae 36. For those species with paralogous gene copies, the percentage difference in nucleotide sequence ranged from 2.2 to 9.8, while the corresponding amino acid sequences had intraspecies variation of only 0.2 to 3.8%, since most of the nucleotide differences between them were synonymous substitutions. In both the RPB1 and RPB2 phylogenetic analyses, paralogs from the same species grouped closely together (Figure 2); therefore, only a single RPB1 and a single RPB2 gene were chosen for each of these six species in making up the combined data set (Figure 3).

Coding regions of the RPB1 and RPB2 genes of the major fungal taxa fall into a very distinct pattern as regards base composition. For three of the phyla, the range was similar, being from 39 to 55% G+C for Ascomycota, 48 to 57% for Basidiomycota and 37 to 54% for Zygomycota (Figure 3). In the Chytridiomycota the spread was much greater, ranging from a low of 33% G+C in Neocallimastix frontalis to a high of 66% in members of Blastocladiales (Figure 3). Previous studies have shown that phylogenetic analyses made by standard maximum likelihood or distance-based algorithms using nucleotide sequences give gene trees that differ significantly from one another when the GC content varies greatly between taxa 37. For this and other reasons, we chose to use encoded protein sequences for phylogenetic analysis in this study.

In phylogenetic analyses with the predicted amino acid sequences of RPB1 and RPB2, the majority of positions were informative based on the parsimony criterion (Table 1). The single most parsimonious tree obtained from RPB1, RPB2, or the combined dataset was in each case largely congruent with that from Bayesian analysis (Figures 2 and 3). The RPB1 and RPB2 phylogenies were also generally congruent (Figure 2), differing mainly as regards the branching within Taphrinomycotina and Chytridiomycota. In neither of these cases was this aspect of the topology highly supported (Figure 2).

The phylogeny from parsimony analysis of the combined RPB1 and RPB2 dataset was highly resolved, as was that from Bayesian inference (Figure 3). Our criterion for statistically significant support of an individual node requires that the parsimony bootstrap value be >70% and the Bayesian posterior probability at least 0.95 38. Both methods of analysis found that monophyletic Ascomycota, Basidiomycota, and Zygomycota all were strongly supported (Figure 3). The most parsimonious tree and the tree from Bayesian inference were highly congruent; both trees included Glomales, Zygomycetes and Trichomycetes in the Zygomycota clade (Table 2, Figure 3). In the SH test, the phylogeny obtained by Bayesian analysis had the highest lnL score among the topologies tested, while for the most parsimonious tree lnL was lower, but not significantly so (Table 2, Figure 3). As regards the monophyly of Zygomycota, the less likely alternative hypothesis of Glomales being sister to Dikaryomycota and the Zygomycota being paraphyletic was not rejected (Table 2, Figure 3).

Ascomycota as a sister group to Basidiomycota had strong support (Figure 3). Basal to the clade of Ascomycota-Basidiomycota (Dikaryomycota) was monophyletic Zygomycota, a group with highly non-uniform morphology (Figure 3). Support for the node separating Chytridiomycota from the clade of Zygomycota-Dikaryomycota was remarkably strong (84% bootstrap, 100% Bayesian, Figure 3).

All phylogenies based upon RPB1 and RPB2 indicated that Chytridiomycota was paraphyletic and that it was the basal taxon of Kingdom Fungi (Figures 2 and 3). Within the Chytridiomycota were two well-supported lineages: Blastocladiales, which was sister to the Zygomycota-Dikaryomycota clade (Figure 3) and Chytridiales, the most basal clade of the fungi (Figure 3). Monoblepharis was sister either to the Blastocladiales clade in parsimony analysis or to the Chytridiales in Bayesian analysis (Figure 3), while Neocallimastix invariably occupied a position between the clades of Blastocladiales and Chytridiales. Although phylogenies based upon 18S rDNA sequences have found both Zygomycota and Chytridiomycota to be polyphyletic (Figure 2A), this alternative hypothesis is rejected in our SH test based on the combined data set of RPB genes (Table 2). Kingdom Fungi was strongly supported as a monophyletic group (Figure 3). Microsporidia were sister to the fungi in all phylogenies based upon RPB1 and RPB2 sequences (Figures 2 and 3). All alternative positions for the Microsporidia were rejected (Table 2), except for the possibility that Microsporidia cluster with the most basal members of Chytridiomycota (not including Blastocladiales).

For each taxon sampled, we analyzed 1.8 kb (NS1-NS8) or more of aligned 18S rDNA. In the case of Coelomomyces stegomyiae the 18S rDNA is 2346 bp long and has many insertions. The 18S rDNA alignment contains regions of strongly conserved sequence interspersed with shorter regions prone to higher rates of substitution. Nuclear 18S rDNA exhibited a larger rate heterogeneity than do RPB1 and RPB2, as indicated by the higher proportion of invariable sites in the 18S rDNA (32% invariable sites, Table 1), and the relatively short interior branches and longer terminal branches in the 18S rDNA phylogeny (Figure 2A).

The 18S rDNA phylogenies inferred by Bayesian or Parsimony analyses were closely similar (Figure 2A). Twenty-two most parsimonious trees were obtained in the Parsimony analyses. In the 18S rDNA phylogeny (Figure 2A), there was significant support for monophyly of both Ascomycota and Basidiomycota and for a sister relationship between them. These results were consistent with the RPB1 and RPB2 phylogenies. The main conflict between the 18S rDNA and RPB1 + RPB2 phylogenies concerned the relationships between members of Zygomycota and Chytridiomycota (Table 3). In the former, both Zygomycota and Chytridiomycota were polyphyletic, with members of Zygomycota falling into three clades. One of these, the Glomales, was sister to Dikaryomycota, another included Mucor of Zygomycetes and members of Trichomycetes, while in the third clade Basidiobolus of Zygomycetes was clustered with the members of Chytridiomycota (Figure 2A). Chytridiomycota was split between two clades, one being the Blastocladeales; the other including all other members of Chytridiomycota and Basidiobolus of Zygomycetes (Figure 2A). Neither of the above phylogenetic relationships in the 18S rDNA phylogeny was supported statistically (Figure 2A).

Pairwise partition-homogeneity tests among three data sets indicated that 18S rDNA was neither congruent with the RPB1 nor the RPB2 data set, while the RPB1 and RPB2 data sets were congruent with one another.

We compared the major clades between phylogenies based on 18S rRNA and RPB genes and their respective bootstrap and Bayesian support values for the nodes subtending them (Table 3). In our RPB1 and RPB2 phylogenies, Chytridiomycota was shown to be the basal taxon in the Fungi and was separated from other fungi with high support (Figure 3, 100% Bayesian and 84% bootstrap). Therefore, the possibility that Zygomycetes were part of the earliest diverging fungal lineage as suggested by the 18S rDNA phylogeny was not considered further.

Determining the phylogenetic relationship between Chytridiomycota and Zygomycota is a central issue in fungal evolutionary biology and a controversial one. It relates both to the question of which phylum is basal in Kingdom Fungi and to the mode of evolution of the flagellum, the lack of which differentiates most fungi from metazoans and flagellate protists. Members of Chytridiomycota are considered to be the basal lineage of fungi since, like metazoans, they use glycogen to store energy and have flagellated spores (zoospores) but, like other fungi, they have chitinous cell walls, flattened mitochondrial cristae, and use the AAA lysine synthesis pathway 3940.

The RPB1 and RPB2 phylogenies show Chytridiomycota to be the basal taxon in the Fungi (Figures 2 and 3). Chytridiomycota is paraphyletic in the combined RPB1 and RPB2 phylogeny and consists of two major lineages; one being the Blastocladiales, the other including Chytridiales-Spizellomycetales-Monoblepharidales with Neocallimasticales as an outlier (Figure 3). These results are in accord with traditional morphological and ultrastructural studies of Chytridiomycota. A cladistic analysis of thallus morphology and ultrastructure showed three clades within the Chytridiomycota: Blastocladiales, Neocallimasticales, and Spizellomycetales-Chytridiales-Monoblepharidales 41. A consideration of both the molecular and morphological evidence identifies the lineage containing the Monoblepharidales, Chytridiales and their close relatives as the most basal one in Kingdom Fungi.

Members of Chytridiomycota can exist in either unicellular or filamentous form. They are the only fungi that typically reproduce by forming flagellated zoospores as part of the life cycle. Chytridiomycota are considered to be aquatic because their unwalled and flagellated zoospores require water for dispersal although some of them have lost their flagella and produce amoebae instead of zoospores (i.e. Amebochytrium).

An assemblage containing all Ascomycota, Basidiomycota, and Zygomycota is a well-supported clade in these studies (Figure 3). All taxa within it lack flagella, contrasting with the Chytridiomycota which are mainly flagellated. Thus the flagellum was lost as a single event, coinciding with the loss of 9+2 microtubule centrioles, the appearance of the spindle pole body (SPB) as microtubule organizing center for mitotic and meiotic nuclear division and the shift from aqueous to terrestrial growth habit.

With the loss of motile cells, alternative methods of gamete release and dissemination evolved in fungi. Both mitotic and meiotic spores act to facilitate long-distance dispersal and resistance to adverse environmental conditions. Sporangiospores and zygospores formed internally were retained in most Zygomycota 42. Novel mechanisms for conidium and meiospore formation and ballistosporic discharge have evolved in the Ascomycota and Basidiomycota.

The unifying characteristics of Zygomycota are: mostly coenocytic hyphae (lacking regular septation), formation of highly resistant zygotes by the fusion of gametangia, and the absence of flagellated cells and centrioles 5. The Zygomycota consist of two classes, the Trichomycetes and Zygomycetes5. All members of Trichomycetes are obligately associated with living arthropods 43. Traditionally Zygomycetes consist of many orders including the Glomales, asexually reproducing soil fungi with very large spores which form endomycorrhizae with many vascular plants 274445. Previously published molecular phylogenies based on rDNA, β-tubulin, and EF-1α found Zygomycota to be non-monophyletic 920212324252729344647.

The phylogeny inferred in this paper from RPB1 and RPB2 sequence data supported a monophyletic Zygomycota that encompassed Zygomycetes, Trichomycetes, and Glomales (Table 3 and Figure 3). A shared derived trait uniting the Zygomycota is the production of asexual nonflagellate mitospores in sporangia (Figure 3). Our SH tests show that a monophyletic Zygomycota including Glomales has higher likelihood than a clade of Glomales with Ascomycota and Basidiomycota, however, the latter hypothesis is not rejected (Table 2).

Microsporidia are widespread, obligate intracellular parasites of animals 48. The view that Microsporidia belonged to an ancient group, the Archaezooans, that predated mitochondrial endosymbiosis was based upon the small size of their ribosomes and their lack of typical eukaryotic cytoplasmic organelles 49. Early phylogenetic studies, using sequences of microsporidial 18S rDNA 50, EF-1α and EF-2 5152 also seemed consistent with a very early divergence of this taxon. Further analyses, however, attributed this conclusion to high rates of sequence divergence in microsporidial 18S rRNA and EF-1α gene sequences and consequent long branch attraction, leading to an incorrect assignment of phylogenetic position 535455.

Relatedness of the Microsporidia to Fungi was suggested by subsequent phylogenetic studies using the genes RPB1, TBP, vacuolar ATPase subunit A, Valyl-tRNA synthetase, nuclear 28S rDNA, and two genes encoding tubulins535556575859606162. Specialized features of the microsporidial RPB1 and EF-1α genes link microsporidia to the "crown" eukaryote group 5363 and a relationship to fungi is also suggested by the presence of chitin in the spore during the reproduction of Microsporidia. Strong evidence against the Archaezooan hypothesis came from the finding, in the genome of Encephalitozoon cuniculli, of genes that clearly are of mitochondrial origin 585962, implying that a secondary loss of mitochondria occurred during the evolution of Microsporidia.

While the preceding studies suggested a close relationship between Microsporidia and Fungi, they left open two possibilities: either that Microsporidia are a taxon within Kingdom Fungi or that Fungi and Microsporidia are sister to one another. These remain open because, in the initial study, no representatives were included from the two basal taxa of Fungi, the Chytridiomycota and Zygomycota 535556575859606162. A subsequent phylogenetic study, based on α- and β-tubulin sequences did include representatives of the basal fungal groups Zygomycota and Chytridiomycota. The phylogeny inferred from these data placed Microsporidia within Zygomycetes, in a position close to Entomophthorales and Zoopagales 3435. Because of the role that tubulin proteins play in determining cell shape, there is a strong possibility that in this instance, the affinity between Microsporidia and Zygomycetes results from common responses to developmental or environmental challenges and not from common descent. Between different groups of organisms, the rate of evolutionary change of tubulin genes is highly variable. For example, in members of Chytridiomycota, tubulin gene sequences are relatively conserved, while both in Zygomycetes and Microsporidia these genes diverged rapidly 343555. A likely consequence of this rate difference is long-branch attraction between Zygomycete and Microsporidial sequences.

It has been proposed by Cavalier-Smith 48 that Microsporidia evolved from harpellalean fungi (Trichomycetes). Three taxa of Harpellales (Trichomycetes) and other Zygomycota are included in this study; SH tests rejected the hypothesis of a harpellalean or zygomycete origin of Microsporidia (Table 2).

The results we have obtained show definitively that Microsporidia occupy a phylogenetic position outside Kingdom Fungi. Phylogenies inferred from RPB1, RPB2, and their combined sequences all show that Microsporidia are the sister taxon to fungi (Figures 2 and 3). In contrast to all of the other phylogenetic associations previously suggested for Microsporidia 343564, this one receives excellent statistical support from the data. The relevant node has 100% Bayesian posterior probability and 99% bootstrap support in Figure 3. Statistical tests of various alternative derivations of Microsporidia (SH test, Table 2) rejected not only the hypothesis of an origin of Microsporidia within Zygomycota, but also rejected a close relationship either to Dikaryomycota or to Metazoa.

Our results suggest that the loss of flagella and their 9+2 microtubule structure in the members of Microsporidia is a separate event from the loss event that led to Zygomycota, Basidiomycota and Ascomycota. We envision the common ancestor of Fungi, Microsporidia and Metazoans as a unicellular, flagellated heterotroph, based upon comparisons between basal fungi, Microsporidia, basal Metazoans and associated groups such as Choanoflagellates. Our results suggest that loss of the flagellum happened at only one point in the evolution of fungi enabling the derived fungal phyla to adapt from the aquatic environment to a terrestrial one.

Fifty-eight taxa were used, including four Microsporidia, nine members of Chytridiomycota, eight zygomycetes, nine basidiomycetes, and thirty-two ascomycetes. As outgroups, ten taxa of animals, plants and protists were included in the analyses. We sampled four of the five orders of the Chytridiomycota as well as four orders representing all the classes of Zygomycota 5. The sources of fungal strains and GenBank accession numbers for nuclear 18S rDNA, RPB1 and RPB2 gene sequences are listed in Supplementary Table [see Additional file 1]. The diverse reproductive structures of some representatives from the major fungal lineages are shown in Figure 1.

The methods for fungal culture, DNA isolation, PCR amplification, cloning, and DNA sequencing have been described 65. Primers NS1 through NS8 were used in this study to amplify nuclear 18S rDNA 66. The set of general oligonucleotide primers used in this study for amplifying regions A to G of RPB1 and regions 3 to 11 of RPB2 genes are listed in Figure 4. All pairs of primers were used in the initial PCR amplification. For RPB1, Dfor and Frev usually work well for most fungi. Species specific primers were designed in the D-F region and paired with Afor and Grev for subsequent amplifications for some fungi (Figure 4). For RPB2, regions 5–7 and 7-11a were usually amplified easily by PCR. Species specific primers were designed in the region of 5–7 and paired with 3bF, 3.1F-adaptor, 3.2F-adaptor and 11aR for subsequent amplifications for some fungi (Figure 4). For certain fungi, nested PCR with the primer of adaptor sequence and species-specific primer were performed to amplify the 3–5 region of RPB2.

The nucleotide sequences of nuclear 18S rDNA and the amino acid sequences of RPB1 and RPB2 translated from DNA sequences were aligned using Clustal X 67, with subsequent visual adjustment, resulting in 2388, 1119 and 944 aligned positions, respectively, including gaps. The regions that could not be aligned reliably were removed leaving a total of 1739 positions for 18S rDNA, 1061 positions for RPB1 and 901 positions for RPB2 to use in phylogenetic analyses. Bayesian inference and maximum parsimony phylogenetic analyses were carried out for 18S rDNA, RPB1, and RPB2 singly, and for the combined protein sequences. To detect topological incongruence among data partitions of 18S rDNA, RPB1 and RPB2, pairwise partition-homogeneity tests among three data sets were conducted using PAUP* 4.0b10.68 with parsimony criterion and heuristic search (10,000 replications).

The model selection approach, the Akaike Information Criterion (AIC), was used to estimate the best-fit model for Bayesian methods using MODELTEST 3.76970. A general time reversible (GTR) model including a proportion of invariant sites and a gamma distribution parameter was selected as the best-fit model for the 18S rDNA data set. For each dataset, Bayesian phylogenetic analyses with Markov chain Monte Carlo (MCMC) sampling was conducted using MrBayes v3.1 71. Six independent MCMC runs were carried out using the GTR model for the nucleotide substitution and JTT model for amino acid substitution. In addition, the proportion of invariable sites and a gamma distribution parameter to allow for rate heterogeneity among sites (six categories) and uniform prior probabilities and tree topologies were implemented in the analyses. The six runs included one run with 3 × 10⁶ generations, two runs with 2 × 10⁶ generations and three runs with 1 × 10⁶ generations to ensure a sufficient number of generations and sampling of the same posterior probability landscape. Each run started with random trees for each of four simultaneous chains, resulted in concordant joint posterior probability distributions for the topology. The sampling was done every 100^th generation for each run. The samples before the convergence of the Markov chain were discarded for each run. The remaining samples from each run were combined into a single file with a total of 97830, 98095, 98876, and 98970 phylogenetic trees for the data sets of 18S rDNA, RPB1, RPB2 and combined RPB1 and RPB2, respectively. These were then imported into PAUP* 4.0b10 to compute the 50% majority rule consensus tree. The percentages for the branches in the consensus tree represent the Bayesian posterior probabilities which are the rough equivalent of a maximum likelihood search with bootstrapping71. The resulting consensus tree from Bayesian analyses for each dataset was evaluated for branch lengths using TREE-PUZZLE 5.0 by maximum likelihood algorithm with a gamma-distribution, the GTR model for nucleotide substitution and the JTT model for amino acid substitution72.

Parsimony analyses were conducted using PAUP* 4.0b10.68 with equal weights for 18S rDNA sequences and a weighted step matrix converted from the JTT matrix 7374 for RPB1, RPB2 and their combined data set. Gaps were scored as missing. The heuristic search using the random addition of taxon option was performed with 1000 replicates to increase the chance of finding all of the most parsimonious trees. To evaluate the strength of the phylogenetic conclusions, 500 parsimony bootstrap replicates were performed using the heuristic search with the random addition of taxon option (10 times per replicate).

In order to test alternative phylogenetic hypotheses, analyses were conducted using PAUP to construct the most parsimonious trees under the constrained conditions, and the resulting trees were evaluated together with the most parsimonious tree and the tree of Bayesian inference in TREE-PUZZLE 5.2 using the SH test based on the combined data set of RPB1 and RPB2 7275. TREE-PUZZLE 5.2 incorporates the JTT matrix for amino acid substitutions and gamma-distributed rates to allow for rate heterogeneity among sites. The gamma distribution parameter α of 0.33 was estimated from our combined dataset of RPB1 and RPB2. Eight rate categories were used in the analyses.