Background

The invention of multicellularity is considered to be one of the major evolutionary transitions [1], but is by no means a rare event. Multicellular forms appeared at least 11 times independently in most eukaryote divisions and in prokaryotes [2–4]. The better known lineages, such as animals, fungi, green plants and red- and brown macroalgae evolved a form of multicellularity where cells stay together after cell division. However, the other forms mostly start off with foraging single cells that achieve multicellularity by aggregation, usually when starved or otherwise stressed. The aggregate then transforms into a fruiting structure with an elevated mass of dormant spores or cysts. Thus far, the social amoeba Dictyostelium discoideum shows the most complex form of aggregative multicellularity with an intermediate migrating “slug” stage and cells specializing into spores and three somatic cell types that form a stalk and structures to support the stalk and spore mass. D. discoideum is also a popular model organism for studying cellular processes such as cell migration [5], cytokinesis, vesicle trafficking, host-pathogen interactions [6], cell differentiation and morphogenesis [7], as well as evolution of sociality and prey–predator relationships [8, 9]. We aim to identify the molecular changes that caused the transition to multicellularity and the acquisition of phenotypic complexity during dictyostelid evolution. In collaboration with Baldauf and coworkers, we initiated this research by constructing the first molecular phylogeny of the then 99 known Dictyostelia using SSUrDNA and α-tubulin sequence data [10]. Phylogenies derived from either gene subdivided Dictyostelia into four major groupings, but did not agree on the position of the root and of several group-intermediate taxa. The position of taxa at terminal branches was also poorly resolved. The latter was improved by sequencing of the less conserved internal transcribed spacer (ITS) regions between the ribosomal RNAs [11]. Further expansion of the phylogeny with 47 new taxa indicated that the intermediate species may actually represent additional minor groups, which were named the violaceum, polycephalum and polycarpum complexes [12].

The availability of the phylogeny allowed us and others to select group-representative taxa for genome sequencing in addition to the D. discoideum genome, which was already available [13]. These taxa are D. purpureum [14], like D. discoideum in group 4. D. lacteum in group 3 [15], Polyspondylium pallidum in group 2, clade 2B [16], Acytostelium subglobosum in group 2 clade 2A [17] and D. fasciculatum in group 1 [16]. The sequenced genomes enabled phylogenetic inference from multiple concatenated protein sequences and these efforts, using respectively 32 [18] and 213 protein sequences [19], both repositioned the root from a location between groups 1 and 2 to a location that separated the four groups into two branches, with branch I containing groups 1 and 2 and branch II, groups 3 and 4. Since no genome sequences for group-intermediate species were available, their positions remained unresolved.

Mapping of phenotypic characters onto the phylogeny revealed that groups 1, 2 and 3 share many common features, such as small clustered or branched fruiting bodies with stalks being formed by dedifferentiation of prespore cells, use of glorin as attractant for aggregation and retention of encystation, the survival strategy of unicellular amoebas [18, 20]. Group 4 taxa generally form large, solitary and unbranched fruiting bodies, and set aside a proportion of specialized cells to form the stalk. Group 4 taxa furthermore use cAMP as attractant, have lost encystation and acquired a migrating slug stage, while many taxa in the group acquired one or two novel cell types. Appropiate understanding of the evolutionary history of these major innovations requires correct understanding of the relationship between groups 3 and 4 and the group-intermediate taxa. The SSU rDNA phylogeny subdivides group 2 into two clades, of which one (2A) uniformly lacks the cellular stalk. However, one taxon (A. ellipticum) that lacks stalk cells groups with the other clade (2B), suggesting that the cellular stalk was gained twice, which seems unlikely [10]. Because the intermediate species link the deepest nodes in the tree, their position is also crucial for understanding the earliest processes that triggered the transition from uni- to multicellularity in Dictyostelia.

Proper reconstruction of phenotypic evolution and its association with changes at the gene and genome level critically depends on a correct understanding of the positions of the group-intermediate taxa. We therefore sequenced the genomes of one taxon each of the violaceum, polycarpum and polycephalum complexes. Additionally, we sequenced the genome of A. ellipticum, with its problematic placement in clade 2B, and to better outline clade 2A, A. leptosomum, which is in the SSU rDNA phylogeny most distant from A. subglobosum, with an already sequenced genome. We also sequenced the genome of D. deminutivum, the earliest diverging species in group 1 with the most primitive multicellular features. We retrieved 47 orthologous proteins from all dictyostelid genomes and three sequenced genomes of unicellular Amoebozoa and inferred phylogenies from concatenated alignments of all proteins and various subsets by different methods. This yielded a very robust consensus phylogeny of Dictyostelia with well defined positions for the group-intermediate taxa.

Results

Genome sequencing of early diverging dictyostelid taxa

To improve the internal node structure of the dictyostelid phylogeny, the genomes of four group-intermediate species and two other early diverging Dictyostelium species were sequenced, using the Illumina platform. Despite careful washing and 4 h starvation to rid cells of their bacterial food source, the genome reads of two species, P. violaceum (Pvio) and A. leptosomum (Alep) consisted for respectively 78% and 39% of E.coli sequences (Additional file 1). The other genomes of D. deminutivum (Ddem), D. polycephalum (Dcep), A. ellipticum (Aell) and D. polycarpum (Dcar) contained from 0.5 to 19% E.coli reads. With all species being prepared for gDNA extraction in the same manner, it is not clear what caused this large variation in contamination with genomic DNA from their food source. Possibly, Pvio, strain P6 and Alep are “farmers”, that retain bacteria in their fruiting bodies for future cultivation [21].

We retrieved mostly complete genes for all of our selected test proteins from the Ddem, Dcep, Aell and Dcar genomes. However, BLAST searches of the Pvio genome returned many prokaryote hits, while for the Alep genome, the orthologous sequences were often incomplete. Fortunately, genome contig sequences for Pvio strain QSvi11 were made available in Genbank by BCM-HGSC (https://www.hgsc.bcm.edu/) under accession number AJWJ00000000, which proved to be of good quality. The retrieved Alep sequences still contributed to 45% of the concatenated aligned sequences, which was sufficient to place Alep at its expected position as sister to A. subglobosum (Asub).

Protein selection, alignment and phylogenetic inference

We selected 40 proteins from 14 primary metabolic pathways and 12 proteins with a range of roles in basic cell biology from the D. discoideum (Ddis) genome with a size >200 amino acids. This selection of proteins from a broad spectrum of metabolic and cellular roles

was intended to average out any effects of natural selection on gene polymorphisms in concatenated alignments. The setting of a minimum size to the selected proteins was meant to i. avoid underrepresentation of the phylogenetic signal from very small proteins and ii. to provide opportunities to select some proteins that individually reproduce the consensus phylogeny for a gene amplification approach for classification of a broad range of taxa.

We used BLASTp to search for orthologs of the 52 proteins sequences in the annotated genomes of D. purpureum (Dpur), D. lacteum (Dlac), P. pallidum (Ppal), D. fasciculatum (Dfas), Asub, Acanthamoeba castellani (Acas), Physarum polycephalum (PhyP) and Entamoeba histolytica (Ehis) and we used tBLASTn to find orthologous genes in the newly sequenced Dcar, Dcep, Aell, Alep and Ddem genomes and the unannotated Pvio QSvi11 genome. For the latter genomes, gene models were manually predicted, assisted by the models of the orthologous genes in the annotated genomes. See the analysis pipeline in Fig. 1.

Fig. 1
figure 1

Bioinformatics pipeline. Chain of procedures for protein selection, cognate gene identification, gene model prediction and phylogenetic analysis

Full size image

The individual orthologous proteins were aligned and all sections that did not align unambiguously as well as large insertions in individual proteins were deleted. Five proteins which showed only small sections of consensus alignment were rejected from the set. The alignments of the remaining 47 proteins were concatenated fully, and into five sets of nine or ten proteins (see Additional file 2, sheet 3). All sets as well as alignments of the individual protein sequences were used for phylogenetic inference.

Bayesian inference of all 47 concatenated proteins using a mixed amino acid substitution model yielded a different topology from the previous SSU rDNA tree (Fig. 2a,b). Firstly, as in the published multi-protein phylogenies [18, 19], the root is no longer positioned between groups 1 and 2, but separates two branches containing groups 1 and 2 and groups 3 and 4. Additionally, the group-intermediate species Dcar changed position from the base of the clade comprising groups 3 and 4 to the base of group 2, while Aell which was previously sister to clade 2B is now branching at the base of the clade composed of 2A and 2B. Due to the repositioned root, groups 1 and 2 are now separate sister groups, while group 4 remains in a clade with group 3. We included the obligate parasite Ehis in the initial phylogeny, but as also found previously [18, 19], most Ehis genes are either absent or very divergent from those in other Amoebozoa. Because this could lead to long-branch attraction errors, Ehis was omitted from further analysis. This had no effect on tree topology (Fig. 2c).

Fig. 2
figure 2

Phylogenetic inference from 47 concatenated proteins. a Location of selected species in the previously inferred SSU rDNA phylogeny [10]. b–f Phylogenetic trees inferred by Bayesian inference (b–d), RAxML (e) or Phylobayes (f) from an alignment of 47 concatenated orthologous proteins that were identified in the species shown in a. Bayesian analyses were run for 100,000 generations with either a mixed amino-acid model (b,c) for the entire alignment, or with a partitioned alignment in which each protein was run under its optimal amino-acid substitution model (d). All analyses converged within 6000 generations (SD of split frequencies = 0). The RaxML analysis was run with 100 bootstrap replicates on an alignment partitioned as in d. Phylobayes MPI [22] was run over two chains under the CAT-GTR model (f). Trees were rooted using A. castellani as outgroup. The average GC content of the genomic DNAs encoding the 47 proteins is plotted onto the phylogeny in panel c. Posterior probabilities or bootstrap support for the nodes are shown. Abbreviated and full species names are colour-coded to reflect the taxon group to which the species belong

Full size image

Instead of running a mixed model on the entire concatenated alignment, we also partitioned the alignment to combine proteins with the same optimal amino-acid substitution model into one partition and then ran each partition under its optimal model. This returned the same phylogeny as the analysis with the mixed model (Fig. 2d). The use of maximum likelihood based tree inference on a similarly partitioned alignment also returned the same phylogenetic tree (Fig. 2e) as did inference by Phylobayes under the CAT-GTR model (Fig. 2f). In the latter model both the individual amino-acid substitution propensities and substitution rates are inferred from the underlying alignment [22]. The nodes in all trees have Bayesian posterior probabilities of 1.0 or 100% bootstrap support. The Bayesian analyses all converged to SD of split frequencies = 0 within 6000 generations. All these parameters indicate that the trees derived from the concatenated alignment are extremely robust.

In the course of gene model prediction, we noted extreme differences in the G/C content of the Dictyostelium genes. Plotting of the averaged G/C content of the 47 test genes onto the phylogeny showed that the branch I genes are more G/C rich than those in branch II, with Acytostelids being particularly G/C-rich and the two group 4 taxa being very A/T- rich (Fig. 2c).

Trees from protein subsets and individual proteins

We next assessed to what extent smaller subsets of the concatenated protein sequences returned robust consensual phylogenies. To retain the functional diversity of proteins also in the subset, we subdivided 47 protein set in five sets of nine or ten proteins by joining the fifth proteins in the gene list (Additional file 2) five times in staggered fashion. Figure 3 shows that sets 1 to 4 yielded trees with identical topologies to the tree inferred from all 47 proteins, further referred to as “the consensus topology”, and also with probabilities of 1.0 for all nodes. Only the tree derived from set 5 showed a single divergent node with a probability of 0.94.

Fig. 3
figure 3

Phylogenetic inference from subsets. The 47 protein set was subdivided in sets of nine or ten proteins by joining the fifth rows of five staggered columns of protein identifiers. The protein alignments were concatenated per set, and each alignment was subjected to Bayesian inference with a mixed amino-acid model. Only set 5 yielded a tree topology that was different (red branch) from the 47 protein consensus topology (top left)

Full size image

Trees were also inferred by Bayesian inference for all individual protein alignments (Additional file 3). Only two out of the 47 trees displayed the consensus topology (Additional file 2, sheet 4), while 12, 18, 14 and 1 tree(s) show 1, 2, 3 or 4 alternative node configurations respectively. The most common alternative topology was Dcar as outgroup to group 1 instead of group 2 (13 proteins) and permutations in the relative positions of Pvio and Dlac (nine proteins) (Additional file 2, sheet 5). Proteins that yielded trees with only one non-consensual node yielded consensus trees when concatenated with a second protein that yielded trees with a different error in four tested cases, while one protein, aco1, required two other proteins to produce a consensus tree (Additional file 4).

To gain insight into parameters that render some proteins more suitable than others for taxonomy, we plotted the number of non-consensual nodes per tree against the number of aligned positions or variable positions, and against the averaged node posterior probabilities. There was only a very weak negative correlation between the number of aligned positions and the number of non-consensual nodes, which became slightly stronger, but still statistically insignificant, when only the variable positions in the alignment were considered (Fig. 4a,b). Only 7% of the variance in the number of non-consensual nodes is explained by the variance in the number of variable positions in the alignment, which implies that the remaining 93% is dependent on some unknown intrinsic property of the protein sequence itself.

Fig. 4
figure 4

Correlations between protein alignment statistics and tree features. For trees inferred from alignments of orthologs for each of the individual 47 test proteins (Additional file 3), the number of non-consensual nodes in each tree was plotted against either the number of aligned (a) or variable (b) positions per alignment, or against the averaged posterior probabilities of all nodes in the tree (c). Averaged posterior probabilities were also plotted against the number of variable positions per alignment (d). Regression lines for all plots are shown with their equations and coefficients of determination. Correlations between the variables were determined by Spearman rank order and P-values are shown. All variables are listed in Additional File 2, sheet 4

Full size image

There was a significant negative correlation between the number of non-consensual nodes and the averaged node probabilities (Fig. 4c), as well as a significant positive correlation between the number of variable positions and the averaged node probabilities per tree (Fig. 4d). Altogether this indicates that while longer alignments with more variable positions produce better supported trees, these trees are not necessarily more accurate in assessing relationships between species. We could not detect significant differences between the number of non-consensual nodes per protein tree and participation of the protein in a specific metabolic pathway (Additional file 2, sheet 6), but this could be due to the small number of proteins per pathway.

Tests for alternative hypotheses

While the full set of 47 proteins and four out of five sets of nine or ten proteins produced identical consensus core phylogenies for the Dictyostelia with 100% statistical support, most trees derived from individual proteins and the tree derived from the 5th set of nine proteins produced a range of alternative topologies, as did the original SSU rDNA tree. We used the Approximately Unbiased (AU) test, implemented in the CONSEL software [23] to compare the confidence levels of the consensus tree topology and the most commonly encountered alternative topologies (Fig. 5, Additional file 5). Table 1 shows that alternative topologies such as the root position of the earlier SSU tree [10], monophyletic origins for Acytostelids and Polysphondylids and alternative positions for the group-intermediate species Dcep and Pvio are all strongly rejected. Only alternative placement of Dcar as outgroup to group 1 or branch I, instead of group 2 comes with a probability of 1:10,000 somewhat closer to the consensus topology with a probability of 1.0. We are therefore confident that the consensus is correct.

Fig. 5
figure 5

Alternative tree topologies. Schematic of repositioning of tree branches in the 47 protein consensus tree to yield the most commonly encountered alternative topologies found in either the earlier SSU rDNA tree of Dictyostelia [10], or the trees inferred from single proteins (Additional file 3). All alternative trees are shown in Additional file 5

Full size image
Table 1 Alternative topology tests
Full size table

Discussion

To understand the directionality of evolutionary processes, a correct understanding of the genetic relationships between species is essential. For Dictyostelia, the relationships between the four major groups were well defined previously from multi-locus phylogenies [18, 19], but the positions of early diverging species that occupy intermediate outgroup positions to the major groups were poorly resolved. Their position is however of crucial importance to identify the molecular changes that triggered the phenotypic diversification of the major groups. To reliably position the intermediate groups, we sequenced the genomes of six early diverging species, and used the information to identify around 50 well-conserved proteins for phylogenetic inference.

The final set of 47 proteins has only one or seven proteins in common with sets of 32 or 213 proteins, respectively, that were used previously to root the dictyostelid phylogeny [18, 19]. Nevertheless, phylogenies inferred from all three sets position the root between branch I, containing groups 1 + 2 and branch II, containing groups 3 + 4, indicating that in all these multi-locus phylogenies, the function of the individual selected proteins made little difference. However, in all three cases, but particularly in the 213 protein set [19], the functional diversity of the selected proteins was large.

Just like groups 1 and 2 in branch I, which are sister clades that evolved independently, groups 3 and 4 are sister clades in branch II, but with Pvio, one of six taxa in the "violaceum" complex [12] occupying a basal position to group 4. Group 4 taxa display major phenotypic innovations, such as two more cell types, cell-type proportioning, extensive slug migration and enhanced fruiting body robustness, which are not yet evident in the “violaceum” complex [18, 20]. To identify the molecular changes that caused these innovations, the genomes of taxa in the “violaceum” complex can now be considered as the baseline from which these changes occurred.

Dcep, one of four D. polycephalum isolates that make up the “polycephalum” complex is now the earliest diverging species in branch II. Although one might argue whether four isolates of the same species make up a “major division”, as suggested previously [12], we feel that the polycephalid isolates with their distinctive phenotype are likely to represent a group of cryptic species. Unlike all other non-group 4 species, polycephalids form very long actively migrating slugs, which unlike group 4 slugs do not show a pattern of prestalk and prespore cells [20]. Also unlike group 4 slugs, polycephalid slugs subdivide into many small fruiting structures with stalks that adhere tightly `along approximately their lower two-thirds. Similar to P. violaceum isolates, which are also grouped together as one species, due to their conspicuous purple branched fruiting bodies, the D. polycephalum isolates show considerable diversity in their SSU rDNA sequences [12, 24].

Dcar, member of the “polycarpum” complex [12], was previously located as sister to branch II in the SSU rDNA and α-tubulin phylogenies [10, 12]. The current 47 protein phylogeny confidently positions Dcar as outgroup to or earliest diverging species of group 2. With only two isolates of Dcar representing the “polycarpum” complex, its status as a “complex” is less convincing, particularly since it has no phenotypic features that set it apart from many species in groups 1–3. The Acytostelid Aell, previously positioned as outgroup to clade 2B with SSU rDNA, and as outgroup to group 2 with α-tubulin [10], is now firmly placed as outgroup to group 2. Group 2 consists of clade 2B, which consists of white Polysphondylids and other species with a cellular stalk, and clade 2A which contains only Acytostelids without cellular stalk. As before, this still presents a confusing history of stalk cell differentiation in Dictyostelia. Dcar, the outgroup to Aell and group 2 already have a cellular stalk, as do species in group 1 and branch II. This indicates that the cellular stalk evolved before the separation between branch I and II in taxa that are either extinct, or as yet undiscovered. The stalk was then lost twice, in Aell and in clade 2A, which likely involved changes in only one or a few genes. Interestingly, all Acytostelids, including Aell, have GC-rich genomes, but this is not the case for Dcar and Ppal (Fig. 2c). This suggests the possibility that an A/T to G/C non-synonymous mutation in a coding sequence, or an A/T to G/C conversion in a gene regulatory element may have caused the loss of the cellular stalk.The consensus phylogeny was also reproduced by four out of five subsets of nine or ten concatenated proteins from the 47-protein set, but only two proteins (rpa1 and smdA) out of the 47 set yielded the consensus tree on their own. The most common alternative topologies show Dcar as outgroup to group 1, or sister to the group 1 species Ddem (17 proteins), while the positions of Pvio and Dlac in branch II were reversed or joined into a single clade in nine proteins (Additional file 2, sheets 4 and 5). However, when these and other popular alternative topologies were individually constrained during inference of the 47 protein tree, their probability of the alternative topology being correct was at least 10,000 times lower than that of the consensus topology Table 1.

Table 2 Species and culture conditions
Full size table

The ability of specific proteins to reproduce the consensus phylogeny was only marginally correlated with the number of total or variable positions in the alignment, and did not appear to be correlated with protein function (involvement in a specific metabolic pathway). The reason why trees inferred from some proteins predict species relationships better than others is therefore unclear. The proteins that yielded trees with one erroneous node, mostly yielded consensus trees when combined with one or at most two proteins that yielded a different error. This implies that with a carefully selected set, 3–5 genes may be sufficient to generate robust species phylogenies. This allows the use of a gene amplification approach rather than whole genome sequencing as the means to correctly classify all known and newly to be discovered Dictyostelia. This is a low-cost alternative to whole genome sequencing, but also avoids problems with contamination of genome reads with bacterial DNA that rendered one of our six sequenced genomes completely and another partially useless. Since the genes used in this study are also present in representatives of the two major divisions Conosa and Lobosa of Amoebozoa (with Acas in Lobosa, and PhyP and Dictyostelia in Conosa) a similar approach could also assist to generate a robust phylogeny of Amoebozoa, to replace and/or complement the current very poorly resolved SSU rDNA phylogeny [25]. In future studies we will explore use of a multi-locus gene amplification approach for incorporation of most known Dictyostelia in the phylogeny.

Conclusions

Accurate phylogenies are of elementary importance for understanding the directionality of organismal evolution. The lack of resolution and topology errors of single gene phylogenies progressively disappear when more genes are incorporated in the underlying alignments. However, such a multi-locus approach requires completely sequenced genomes, which, particularly for unicellular eukaryotes, are only sparsely available and poorly representative of the genetic breadth of eukaryotes.

The major division of Amoebozoa gave rise to at least two independent inventions of multicellularity, of which the Dictyostelia represent the best studied and phenotypically most diverse group. Previous phylogenies based on SSU rDNA and proteins from existing genomes only partially resolved relationships between major and minor clades. To resolve these relationships, we sequenced the genomes of six early diverging dictyostelid taxa and annotated 47 functionally diverse deeply conserved proteins. These sequences combined with orthologous sequences from six dictyostelid and three unicellular amoebozoan sequenced genomes were used both concatenated and individually for phylogenetic inferences. The extremely robust phylogeny derived from concatenated sequences highlights monophyly of Dictyostelia, but polyphyletic origins of its three genera. Somatic cells were present in the last common ancestor of Dictyostelia, but were likely lost twice independently in group 2. Only two proteins individually reproduced this consensus phylogeny, but sets of two or three proteins, which individually yielded minor errors, again reproduced the core phylogeny. This emphasizes the inherent error-prone nature of single gene phylogenies, but also indicates that reliable classification of taxa can be achieved by sequencing just a few genes. This will be particularly useful for broader and deeper sampling of taxa in Dictyostelia, Amoebozoan and other major eukaryote divisions, and for classification of newly discovered species.

Methods

Species, culture and genomic DNA preparation

All species were grown in association with Escherichia coli 281 using the culture media and culture temperatures listed in Table 2. After amoebas had cleared the bacteria, they were harvested in 10 mM phosphate buffer pH 6.5, washed three times and starved shaken at 21 °C and 150 rpm for 4 h to allow clearance of bacteria.

To prepare genomic DNA, cells were lysed in 62.5 mM EDTA, 1% (w/v) SDS and 1% (v/v) β-mercaptoethanol in 50 mM Tris pH 7.2, and incubated at 80 °C for 10 min. Genomic DNA was purified by extraction with phenol:chloroform:isoamylalcohol 25:24:1 pH 8.0 using phase-lock columns and ethanol precipitation. After resuspension in 0.1 mM EDTA in 1 mM Tris, pH 8.0, DNA was incubated for 3 h with 0.5 μg/μl RNAse A. Integrity of DNA was tested visually on a 1% TAE agarose gel.

Genome sequencing and assembly

Six Illumina TruSeq DNA libraries were prepared from the six species listed in Table 2, using standard Illumina protocols followed by Pippin (Sage) size selection to a target insert size of 400 bp. The six libraries were barcoded and sequenced together on a single lane of an Illumina HiSeq 2500 platform using 150 base paired-end reads in rapid mode to yield approximately 120 M read pairs. Removal of adapter sequences and quality trimming of the reads was performed with fastq-mcf v1.1.2-537 and reads were assembled using the CLC-BIO assembler v4.2.0 (Qiagen) to yield the pre-filtering assemblies. The contigs were annotated using ncbi-blast + v2.2.28, which showed variable levels of contamination between genomes with Escherichia coli DNA. The reads were also aligned against the E. coli K12 genome, using bwa v0.7.5a [26], and a new assembly was prepared using all unaligned reads. The annotation of the contigs assembled from these E.coli filtered reads was again assessed using ncbi-blast + v2.2.28. The assembly statistics of the six genomes before and after filtering of E.coli sequences is listed in Additional file 1. Library preparation, sequencing and bioinformatic analyses of raw Illumina reads were carried out by Edinburgh Genomics at the University of Edinburgh (http://genomics.ed.ac.uk).

Gene selection, identification of orthologs and gene model prediction

Forty Dictyostelium discoideum proteins larger than 200 amino acids were initially selected from enzymes mediating 14 biochemical pathways of primary metabolism, retrieved from the KEGG database [27], which were conserved between D. discoideum, Polysphondylium pallidum and D. fasciculatum [16]. This set was supplemented with an additional twelve conserved proteins with a range of cell biological functions. Putative orthologs were retrieved by BLASTp from the published genomes of D. purpureum, D. lacteum, P. pallidum, Acytostelium subglobosum, D. fasciculatum, Physarum polycephalum and Entamoeba histolytica and Acanthamoeba castellani.

The newly sequenced genome contigs of D. polycarpum, D. polycephalum, D. deminutivum, A. ellipticum, A. leptosomum and the genome contig sequences of P. violaceum strain QSvi11, which were available in Genbank, were archived in Artemis [28] and converted into a BLAST database. This database was queried for the presence of the coding genes for each of the 52 proteins using tBLASTn. The contig and coordinate information of the hit was used to retrieve a genomic DNA fragment from the Artemis archive with sufficient flanking sequence to allow reconstruction of the complete gene model. The gene models were manually predicted using DNAMAN software (Lynnon Corp., Quebec, Canada), guided by existing gene models for the published orthologous Dictyostelium genes. Further corrections were made when protein alignments suggested the presence of indels or inappropriately assigned start and stop codons (see Fig. 1).

Phylogenetic inference

Protein sequence alignment and concatenation

All putative orthologs for each of the 52 proteins were aligned using M-coffee [29] version 10.00, using eight different alignment algorithms to generate a consensus multiple alignment. Sections with poor consensus alignment (<50%) and long insertions affecting few genes were deleted using Jalview v 2.8.2 [30]. Five proteins were rejected from further analysis due to insufficient consensus alignment. Inspection of the alignments and of trees inferred from individual alignments revealed that some species contributed paralogs or bacterial contaminants rather than orthologs to the alignment, which were replaced by orthologs, when available, or deleted. In alignments, paralogs stand out by being markedly divergent from genes in related species, while in trees paralogs end up in outgroup positions. The final alignments of the 47 proteins were concatenated using catfasta2phyml v1.0 (https://github.com/nylander/catfasta2phyml) yielding a total of 37,410 aligned amino acid positions. This alignment, shown in Additional file 6, was used in its entirety for phylogenetic inference, or divided into five subsets of nine or ten proteins.

Phylogenetic trees

Phylogenetic trees were inferred using MrBayes v3.2.2 [31] and RAxML version 8.1.20 [32] with either a mixed amino acid substitution model for the entire alignment, or with concatenated alignments partitioned into sections which contained genes with the same optimal amino acid substitution model. Each partition was then run under its optimal model. For all analysis, rate variation between sites was estimated by a gamma distribution with four rate categories and a proportion of invariable sites. Bayesian analyses were run for one to two million generations for alignments of individual genes and for 100,000 generations for concatenated alignments. The concatenated alignments already fully converged (SD of split frequencies = 0) at less than 10,000 generations. RaxML analyses were run with 100 bootstrap replicates. Trees were visualized using Figtree v.1.4.2 (http://tree.bio.ed.ac.uk/software/figtree/) and rooted at midpoint or on the outgroup of unicellular Amoebozoa.

Test of alternative topologies

Alternative tree topologies were investigated using the approximately unbiased (AU) test as implemented in the program CONSEL v0.20 [23]. Trees with constrained nodes, as shown in Additional file 5 were generated using RaxML, and the RaxML log likelihood values for the consensus tree topology and the alternative tree topologies were used as input for the AU test.