One of the evolutionary innovations that may affect a protein fold is the so-called circular permutation [1, 2]. This term refers to a situation whereby the N-terminal part of one protein is homologous to the C-terminal part of another protein and vice versa. Circular permutations have been found in a wide range of proteins [3] and other examples are definitely to be discovered in the rapidly growing protein sequence databases. Some authors discriminate between a swap and a circular permutation [3, 4]. The swap is then any mutually altered order of homologous regions in proteins, while the circular permutation refers to a special case of a swap where swapped regions cover essentially the whole protein length. We shall use the term "circular permutation" to describe an altered order of N- and C-terminal segments of individual proteins domains, regardless of the actual size and domain composition of the whole proteins.

P-loop GTPases and related proteins form a vast superclass of globular α/β proteins with the most conserved and functionally important regions denoted as G1 to G5 motifs [5, 6]. One of the many distinct evolutionary lineages of the superclass referred to as YawG/YlqF [5] or YRG [7] family comprises proteins with a non-canonical order of the conserved motifs, which are arrayed in a circularly permuted fashion G4-G5-G1-G2-G3. However, crystal structures of several representatives [8, 9] (Kniewel et al., unpublished, PDB accession number 1PUJ) and biochemical studies [7, 10] have demonstrated a rather typical GTPase fold and the expected GTP-binding/GTPase function of these proteins.

A recent systematic search for circularly permuted GTPases has identified the YRG family as the only group with such a modification of the GTPase domain [11]. The authors argued that other possible circular permutations of the GTPase domain are unlikely owing to structural or functional constrains. For example, according to Anand and colleagues, the G3-G4-G5-G1-G2 permutation, which breaks the reverse turn between the strands β2 and β3, should impair the proper folding of the protein [11].

In the course of our systematic survey of the Ras superfamily, we noted that two large uncharacterised proteins from Dictyostelium discoideum, EAL68747.1 and EAL60755.1 (hereafter referred to as DdiCPRas1 and DdiCPRas2, respectively), contain a putative circularly permuted GTPase domain with the conserved motifs G1-G2 lying downstream of the G3-G4-G5 motifs (Fig. 1A). At the secondary structure level, the region corresponding to strand1-helix1-strand2 of a canonical GTPase domain (here exemplified by the human HRAS) directly follows the region equivalent to helix5. Based on sequence comparison with available GTPase sequences, the domain in DdiCPRas1 and DdiCPRas2 could be readily classified into the RAS family, which consists of mostly small proteins composed of a GTPase domain and a hypervariable C-terminal tail with a prenylated cysteine residue ensuring membrane localisation of the protein [12, 13]. Except the circular permutation, the domains in DdiCPRas1 and DdiCPRas2 are rather typical RAS family GTPase domains including, for instance, the "TIE" G2 motif (Fig. 1A) characteristic for the family. DdiCPRas1 and DdiCPRas2 thus define a new subtype of RAS family proteins.

Using MODELLER 9.1 [14] and the human HRAS structure (1ZW6) as a template, we build a 3-D model of the cpRAS domain of DdiCPRas1 (Fig. 1B and Additional file 1). Because of the circular permutation, the DdiCPRas1 sequence had to be reshuffled so that it could have been aligned with HRAS and the N- and C-termini of the model were created artificially by removing the bond between the strands 2 and 3. The template did not help in elucidating the conformation of the putative loop between the helix 5 and strand 1, so this loop is not included in the model. Furthermore, it is likely that the actual appearance and position of the helix5 and strand1 is slightly different from that indicated by the model because of the connection between them implicated by the primary structure of the cpRAS domain. Nevertheless, although experimental verification is required, the model and the presence of typical GTPase signature motifs suggest that DdiCPRas1 very likely retains a GTPase function despite the circular permutation. The GTPase function is also conserved in the previously described version of a circularly permuted GTPase domain of the YRG family [7, 10], which however differs from the cpRAS domain in lacking a connection between the strand3 and helix2 rather than between strands 2 and 3 (Fig. 1B).

In order to illuminate the evolutionary origin of the cpRAS domain, we used BLAST [15] to search genome databases for orthologs of the Dictyostelium CPRas1 and CPRas2 proteins (see Additional file 1 for details on sequence data sources). Interestingly, the cpRAS domain was found in other species, but the phylogenetic distribution of cpRAS is highly scattered, comprising a few metazoans (so far only the lancet Branchiostoma floridae, the sea urchin Strongylocentrotus purpuratus, and the sponge Amphimedon queenslandica), the choanoflagellate Monosiga brevicollis, the unicellular opisthokont Capsaspora owczarzaki, the amoebozoan Acanthamoeba castellanii, and the heterolobosean Naegleria gruberi (Fig. 1A, 2A). According to their domain architecture, these proteins fall into two groups (Fig. 2B).

The first comprises the Dictyostelium DdiCPRas1 and CPRas from other species and is represented by proteins of ≈ 840–970 amino acid residues characterised by a poorly conserved N-terminal region of 40–66 residues without any recognisable homology to other proteins, the cpRAS domain, a poorly conserved putative linker region, and a conserved C-terminal half that contains a domain annotated as "von Willebrand factor (vWF) type A (VWA)" domain in the SMART collection [16] or as "Sec23/Sec24 trunk" domain (PF04811) in the Pfam database [17]. In addition, inspection of a multiple alignment suggests that a type of zinc finger motif including four absolutely conserved cysteine residues may be present in most of these proteins (except the protein from Acanthamoeba) upstream the VWA domain, although it receives only high E-values or is not recognised at all by domain identification tools. The CPRas protein from Naegleria additionally harbours a RING finger motif fused to the C-terminus. Interestingly, proteins remarkably similar to the C-terminal half of this CPRas type including the potential zinc finger but lacking even vestiges of an N-terminal cpRAS domain can be found in several taxa, namely Dictyostelium, Entamoeba dispar, some metazoans (the mollusc Lottia gigantea, the annelid Capitella sp. I, Strongylocentrotus), and ciliates (Fig. 2B). Based on the level of mutual similarity, these proteins appear to be a group of their own, evolutionarily separate from the C-terminal half of CPRas proteins with the cpRAS-VWA domain arrangement (data not shown).

The large Dictyostelium protein DdiCPRas2 (3933 residues) shares with other cpRAS proteins only the circularly permuted GTPase domain corresponding to residues 2844–3013, while the rest differs completely (Fig. 2B). The N-terminal part of the DdiCPRas2 shows no clear homology to other proteins or domains in searches with BLASTP, SMART, Search Pfam, or CD-search, but two large blocks of the protein flanking N- and C-terminally the cpRAS domain, approximately residues 2140–2771 and 3158 to the very C-terminus, show significant BLASTP similarity (E-values of 3.10–32 and higher) to a group of bacterial uncharacterised proteins (in Fig. 2B exemplified by a protein from Anabaena variabilis) representing the family COG2373 in the CDD database [18]. The family is named "Large extracellular alpha-helical protein" [19], but the basis for this annotation is unclear and should be treated with caution. This family may actually be homologous to the eukaryotic alpha-2-macroglobulin family of serum proteins, because a series of domain typical for this family, specifically the A2M_N (PF01835), A2M_N_2 (PF07703), and A2M_like (cd02891) domains, can be recognised in COG2373 proteins using Search Pfam and/or CD-search. These domains appear to be conserved also in DdiCPRas2, although the A2M_N_2 domain is recognised with insignificant E-value only. The unique and unusual architecture of DdiCPRas2 among other proteins with the cpRAS domain might raise doubts about the accuracy of the respective nucleotide sequence or gene model. However, an incomplete sequence of an obviously orthologous gene with potentially the same domain architecture is present in a low-coverage assembly of the genome of another Dictyostelium species, D. purpureum (scaffold_491 [20]), confirming the authenticity of the DdiCPRas2-like type of CPRas proteins.

According to our phylogenetic analysis based on an extensive sampling of the RAS family (168 sequences), all cpRAS domains are of a monophyletic origin with the cpRAS domain of DdiCPRas2 nested within the cpRAS clade, although without support from bootstrap analysis (data not shown). Given also its much wider phylogenetic distribution, it seems that the cpRAS-VWA domain organisation evolutionarily precedes the DdiCPRas2-like domain arrangement. Interestingly, when the DdiCPRas2 sequence excluding the cpRAS domain is used as a BLASTP query against the NCBI nr database, the significant hits comprise sequences from diverse bacteria but only one eukaryotic entry – another COG2373-like protein (but lacking a GTPase domain) from Dictyostelium (EAL60755.1). We therefore suggest that the type of CPRas proteins represented by DdiCPRas2, specifically its C-terminal part, evolved by insertion of a copy of a cpRAS domain from a cpRAS-VWA protein into a COG2373-like protein introduced into the Dictyostelium lineage by horizontal gene transfer (HGT) from prokaryotes.

The YRG family is widespread in all Eubacteria, Archaebacteria, and Eukaryota [6, 7, 11], so the circular permutation that generated the GTPase domain of this family probably occurred before the last universal common ancestor (LUCA). By contrast, the cpRAS domain is apparently a much later eukaryote-specific derivative of the GTPase fold. The scattered phylogenetic distribution of the cpRAS domain might indicate its origin in a particular eukaryotic lineage followed by dissemination by HGT. However, our phylogenetic analysis of the full cpRAS-VWA protein sequences (see Fig. S1 in Additional file 1) does not provide any strong evidence for HGT among the distinct eukaryotic lineages and the results are compatible with the scenario whereby cpRAS-VWA occurred already in the common ancestor of Excavata, Amoebozoa, and Opisthokonta but was lost independently from at least eight lineages (Fig. 2A). Although sequencing of additional genomes is necessary to test this hypothesis, it is likely that most cases of the absence of the cpRAS domain will indeed prove as resulting from secondary losses. The cpRAS domain thus appears to provide further evidence for an unexpected complexity of ancestral eukaryotes and the importance of secondary reduction in the eukaryotic evolution (see, e.g., [21, 22]).

There appear to be no specific or well-defined functions defined for the domains that physically combine with the cpRAS domain (the VWA domain and the alpha-2-macroglobulin family domains), so the domain architecture of CPRas proteins provides little clues as to their possible cellular roles. Their highly scattered phylogenetic distribution in a disparate set of organisms indicates that they may be implicated in rather specialised processes. The CPRas proteins are thus definitely interesting candidates for experimental characterisation.

In summary, our findings correct the claim by Anand and colleagues [11] that the G4-G5-G1-G2-G3 arrangement of the conserved GTPase motifs is „the only possible circular permutation that can exist in nature" and further add to the structural variations of the GTPase domain. Future broader sampling of eukaryotic genomes will reveal more on the evolutionary origin of the cpRAS domain and functional studies will establish the cellular role of CPRas proteins.