Phylogenomic analysis of the Eukaryotic Dataset with one of the currently most realistic phylogenetic models (category mixture model (CAT); 37) confirms the monophyly of major eukaryotic groups including Holozoa, Fungi, Amoebozoa, and Viridiplantae. Further, Amoebidium, Sphaeroforma plus Capsaspora form a monophyletic group, and Nuclearia is without a doubt the closest known sister-group to Fungi (100% BS; Figure 1). Also some higher-order relationships are recovered with significant support, such as opisthokonts and the two recently proposed supergroups JEH (jakobids, Euglenozoa plus Heterolobosea 20) and CAS (Cercozoa, Alveolata plus Stramenopila 203839), whereas monophyly of Plantae, Excavata and Chromalveolata is not found. Evidently, the taxon sampling of protists in our dataset is insufficient for (and not aimed at) resolving the phylogenetic relationships among these latter lineages, as it was meant to constitute only a strong and well sampled outgroup to opisthokonts.

Analysis of the Eukaryotic Dataset with maximum likelihood (ML) using RAxML 40 and the commonly used WAG+Γ model generated a similar tree topology (Figure 1 and additional file 1). Deep opisthokont divergences are predicted consistently and with significant support (BS > 98%), with Nuclearia clearly sister to Fungi (100% BS) and choanoflagellates the closest neighbor of animals. Amoebidium, Sphaeroforma plus Capsaspora form a monophyletic sister group to animals plus choanoflagellates, consistent with a previous analysis 5 but contradicting others 78. The reasons for this incongruence may be related to differences in data and taxon sampling. Our dataset contains 50 eukaryotic species with a close outgroup to Holozoa (i.e., including nucleariids together with fungal representatives), compared with a total of only 30 species in a more extensive previous analysis 7. In contrast to our analysis using Bayesian inference (BI), ML associates Malawimonadozoa with JEH (77% BS), a tendency noted and discussed previously 2041, and an issue to be addressed by better taxon sampling in this group (currently, data are available from only two species). Other minor differences between WAG versus CAT model analyses (yet without statistical support in favor of alternatives) are in relationships within Plantae and the placement of Haptophyceae.

We further investigated if the position of Nuclearia next to Fungi might be affected by potential phylogenetic artifacts, such as compositional sequence bias and/or LBA 3642. This is suspected because of the highly varying evolutionary rates both within Fungi and in protist outgroups, and the unusual result that better taxon sampling in Fungi reduces phylogenetic support for the Nuclearia position (18; see introduction). To do so, we first eliminated fast-evolving species from the dataset: S. cerevisiae, Blastocystis hominis, Cryptosporidium parvum, Sterkiella histriomuscorum, Diplonema papillatum and Leishmania major. The results from analyses using RAxML were essentially unchanged, both with respect to tree topology and BS values (additional file 2). To counteract sequence bias, we recoded the 20 amino acids into six groups as previously proposed 43. Again, phylogenetic analysis of this dataset using P4 44 generated essentially the same tree topology, with some support values decreased due to loss of information by recoding (additional file 3).

Finally, we evaluated the positioning of Nuclearia next to Fungi with the AU and weighted Shimodeira Hasegawa (wSH) likelihood tests 45. For this, we compared the topology presented in Figure 1 with competing tree topologies in which the two Nuclearia strains were moved as sistergroup to all major eukaryotic lineages, and all possible positions within Opisthokonta. The results of both tests confirm Nuclearia as the closest neighbor group of Fungi, with all alternative topologies rejected at a significance level of p = 0.002 (Table 1). Given the unequivocal support for Nuclearia as the fungal sistergroup, we propose the term 'Holomycota' to refer to the assemblage of Nucleariida plus Fungi.

Phylogenetic analyses of nuclear versus mitochondrial datasets are expected to come to similar conclusions, thus providing independent evidence for the given phylogenetic relationships. To this end, we sequenced and analyzed the complete mitochondrial DNA (mtDNA) of one of the N. simplex strains (a circular mapping DNA of 74 120 bp; see additional file 4). Note that growth of Nuclearia is complicated (the standard method calls for growth on Petri dishes with a bacterial lawn as food source), and that it is difficult to obtain sufficient cell material for mtDNA purification, explaining why we succeeded for only one of the two Nuclearia species.

The Nuclearia mtDNA contains a high number of introns (21 group I, and one group II), and mitochondrial protein genes appear to be translated with the standard translation code. These features are also widespread in Fungi. In contrast, Holozoa all use a mitochondrial UGA (tryptophan) codon reassignment, and contain no or only a few introns (with the notable exception of Placozoa, an enigmatic group of Metazoa 46).

Phylogenetic analysis of a dataset with 56 species and 13 of the ubiquitous, most conserved mtDNA-encoded proteins predicts the monophyly of Opisthokonta, Stramenopila, Holozoa and Fungi with confidence, and also recovers Nuclearia as the sister-group of Fungi, albeit with a moderate BS value of 85% (Figure 2). To verify if the limited support for Holomycota is expected (i.e., correlating with the number of available sequence positions in the respective datasets), we performed a variable length bootstrap (VLB) analysis. It compares the development of BS values with the number of sequence positions, for the nucleariid/fungal sister relationship. For this, we chose the 29 species shared between the two datasets (for the tree topology of the respective nuclear dataset see additional file 5). The results show that the development of BS values is similar for nuclear and mitochondrial data (Figure 3), and that the available mitochondrial dataset (as well as the above-cited nuclear phylogenies with five genes) is too small to resolve the phylogenetic position of nucleariids with high confidence. A better taxon sampling primarily in nucleariids will be imperative for improved phylogenetic resolution, motivating sequencing projects with new technologies, which are likely to provide mitochondrial as well as nuclear genome sequences - even with the limited amount of cellular material that is available for some taxa (e.g., 47).

Analyses of both the nuclear and mitochondrial datasets have been insufficient to assess with confidence, neither zygomycete mono/paraphyly, nor the phylogenetic position of Blastocladiomycota (Blastocladiales) (Figure 1, 2). For instance, a recent mitochondrial multi-gene phylogeny with the first complete Glomus mtDNA sequence groups Glomus and Mortierella, yet lacks significant statistical support 47. To re-address these questions, we have assembled a large dataset of nuclear-encoded genes from an extended, representative selection of fungal species, plus the two Nuclearia species as outgroup (i.e., the Fungal Dataset). The analyses show overall strong BS for the paraphyly of zygomycetes (Figure 4), i.e., the Entomophthoromycotina represent a significantly supported and completely independent fungal lineage. However, monophyletic Mucoromycotina including Mortierella as recently redefined 28 is not recovered (rendering the taxon Mucoromycotina paraphyletic), neither is the taxon Symbiomycota (Glomeromycota plus Dikarya; 29). Instead, there is moderate support to group Mucorales plus Dikarya (92% BS in BI) and Glomus as their next neighbor (85% BS in BI). Although the placement of Glomus relative to Mortierella differs between our BI and ML analyses (Figure 4), we assume that the result of the BI analysis with its superior evolutionary model is more reliable. In light of these results, taxonomic reordering based on stable phylogenetic resolution of the traditional zygomycetes will require phylogenomic analyses with a much improved taxon sampling. Currently, nuclear and mitochondrial genome data are available only for single species in the latter two taxa; i.e. Glomus intraradices and M. verticillata.

Rooting of the fungal tree with nucleariids confirms that the traditional chytridiomycetes are also paraphyletic, again assuming that the result of the BI analysis is correct (Figure 4). Confirmation of this result (justifying an elevation of Blastocladiomycota as a separate phylum; 28) is highly desirable, as genome-size datasets in Blastocladiomycota are limited to the two moderately distant species Blastocladiella emersonii and A. macrogynus. Similarly, in light of the significant support for a monophyletic Chytridiomycota plus Neocallimastigomycota (100% BS with BI; Figure 4), their division into separate taxonomic higher ranks should be reconsidered, but only after phylogenomic analysis with improved taxon sampling in both groups. Finally, our results motivate genome or EST sequencing in Rozella species (Rozellida), potential relatives of Microsporidia and close neighbors of Fungi. The availability of a largely improved taxon sampling in zygomycetes, chytrids and Rozellida will provide a solid basis for evaluating the proposed placements of Microsporidia - either within or as a sistergroup to Fungi - based on phylogenomic analyses.

The results presented here are consistent with previous notions on how Fungi came into being. For example it is thought that the first Fungi probably had branched chytrid-like rhizoids, which developed by enclosure of nucleariid-like filopodia (sometimes branched) into cell walls, during a nutritional shift from phagotrophy to saprotrophy, thus giving rise to fungal hyphae and rhizoids 7. However, the picture is more complicated as it is widely thought that the ancestral opisthokont also had a single posterior flagellum 48. This structure was lost during evolution of most but not all fungal lineages (e.g., 9254950), with a separate loss in the nucleariid sistergroup. In this sense, nucleariids are unlikely to represent a primitive developmental stage, but rather a secondary reduction resulting in a unicellular, amoeboid life style. Obviously, the clarification of the chain of events leading to the emergence of multicellularity in Fungi is by no means complete. These issues will only become clear with a much broader sampling of genomes from taxa near the animal-fungal divergence and the discovery of additional protist groups that are closely related to Fungi.

Two N. simplex (CCAP 1552/2 and 1552/4) cDNA libraries were constructed following recently published protocols 51. Cells were grown in liquid standing cultures in WCL medium http://megasun.bch.umontreal.ca/People/lang/FMGP/methods/wcl.html supplemented with 0.5 × Cerophyll, with E. coli cells as food, which were pre-grown on LB medium in Petri-dishes as food. Plasmids were purified using the QIAprep 96 Turbo Miniprep Kit (Qiagen), sequencing reactions were performed with the ABI Prism BigDye™ terminator version 3.0/3.1 (Perkin-Elmer, Wellesley, MA, USA) and sequenced on an MJ BaseStation (MJ Research, USA). Trace files were imported into the TBestDB database http://tbestdb.bcm.umontreal.ca/searches/login.php52 for automated processing, including assembly as well as automated gene annotation by AutoFact 5354.

N. simplex (CCAP 1552/2) was grown as described above. The harvested cells were disrupted by addition of SDS plus proteinase K, and mitochondrial DNA was purified following a whole cell lysate protocol 55 and sequenced from a random clone library 56. For mitochondrial genome assembly we used Phred, Phrap and Consed 5758; http://www.phrap.org/. Mitochondrial genes and introns were identified using automated procedures (MFannot, N. Beck and BFL unpublished; RNAweasel, 59), followed by manual curation of the results.

A previously published alignment of nuclear-encoded proteins 20 was used for adding the new Nuclearia cDNA sequences generated in our lab, plus extra sequences available from GenBank (a taxonomic broad dataset containing 50 eukaryotes will be referred to as the 'Eukaryotic Dataset'; another one containing 26 fungal species plus the two Nuclearia species as 'Fungal Dataset') using MUST 60 and FORTY (Denis Baurain and HP, unpublished). The number of species has been limited (to allow phylogenomic analyses within reasonable time frames), but only in well-sampled phylogenetic groups of undisputed phylogenetic affinity. Species that were not included are either fast-evolving and/or are incompletely sequenced. Other procedures for dataset construction, in particular the elimination of paralogous proteins, have been described previously 61. Within opisthokonts, major lineages had to be represented by at least two distant species, and the extremely fast-evolving Microsporidia were excluded, as these are known to introduce phylogenetic artifacts and an overall reduction of phylogenetic resolution (at an extreme leading to misplacement of species; e.g., 6263). Sampling within the protist outgroup of the Eukaryotic Dataset is also not comprehensive (Stramenopila, Alveolata, and Euglenozoa) and limited to slow-evolving representatives of major eukaryotic lineages. The final Eukaryotic Dataset contains 118 proteins (24 439 amino acid positions) and the Fungal Dataset 150 proteins (40 925 amino acid positions). Proteins included in the nuclear datasets are listed in additional file 6.

For a dataset of mitochondrial proteins, 13 ubiquitous genes (cox1, 2, 3, cob, atp6, 9, and nad1, 2, 3, 4, 4L, 5, 6) were selected. Muscle (64), Gblocks (65) and an application developed in-house (mams) were used for automatic protein alignment, removal of ambiguous regions and concatenation. The final dataset contains 56 taxa and 2 710 amino acid positions.

Phylogenetic analyses were performed at the amino acid (aa) level using methods that are known to be least sensitive to LBA artifacts (363766, and references therein). The concatenated protein datasets were analyzed either by Bayesian inference (BI, PhyloBayes 37) with the CAT+ Γ model and four discrete gamma categories, or by maximum likelihood (ML, RAxML 40 with the WAG+ Γ model and four discrete categories. BI analyses using the CAT model have been shown to be more reliable than ML, due to the application of a more realistic evolutionary model. ML analyses were essentially performed to identify differences in topology, pinpointing problematic parts of the tree for which addition of new data would be in order (i.e., preferentially genome sequences from slowly-evolving species, and those that are expected to break long internal branches at questionable tree topologies).

In case of BI and the Eukaryotic Dataset (values for the Fungal Dataset in brackets), chains were run for 3000 (1000) cycles, and the first 1500 (500) cycles were removed as burn-in corresponding to approximately 1,200,000 (400,000) generations. Convergence was controlled by running three independent chains, resulting in identical topologies. The reliability of internal branches for both, ML and BI analyses was evaluated based on 100 bootstrap replicates. For BI, we inferred a consensus tree from the posterior tree topologies of replicates.

Likelihood tests of competing tree topologies were also performed. The site-wise likelihood values were estimated using Tree-Puzzle 67 with the WAG+ Γ model, and p-values for each topology were calculated with CONSEL 45.

We compared the performance of nuclear and mitochondrial datasets in phylogenetic inference by Variable Length Bootstrap (VLB) analysis 68. Sequences of 29 common species were taken from the eukaryotic (24,439 aa positions) and mitochondrial (2,710 aa positions) datasets. From these, two respective series of sub-datasets were constructed by randomly choosing 400, 600, 800, 1 000 ... sequence positions. Phylogenetic inferences were then performed using RAxML with the WAG+Γ model and four discrete categories, after which the BS values for the grouping of nucleariids and Fungi were recorded.