The rat α-subunit of RGGTase is very similar to the corresponding α-subunit of FTase, containing 15 α-helices arranged in a crescent-shaped, double layered right-handed superhelix, enveloping the β-subunit 1112. Its structural architecture, together with statistically significant sequence similarity with the tetratricopeptide repeat (TPR) motif indicates that it belongs of the TPR superfamily 15. An individual TPR is a pair of anti-parallel α-helices, with consensus residues mediating the packing of these helices 15. In the rat RGGTase α-subunit these 7 TPR are helices 2 to 15. The crystal structure of the Rat RGGTase revealed two additional domains, a C-terminal Leucine Rich Repeat domain (LRR), and an Ig-like domain inserted between helices 11 and 12, i.e. between TPR 5 and 6. This is the exact same loop where in the FTase α-subunit there is a short 3₁₀ helix between α-helices 11 and 12 11.

We investigated the complete and partial genome sequences of 43 organisms, covering the crown eukaryotic groups discicristates, excavates, alveolata, heterokonts, plants, amoebozoa and the opisthokonts (metazoa, fungi and choanoflagelates) 28. In all the organisms with a complete genome sequence available, we detect a putative RGGTase α-subunit (Figure S1). Thus we can conclude that the last eukaryotic common ancestor already possessed this subunit. FTase α-subunits were also found in the majority of the organisms studied (Figure S1), suggesting that the duplication that gave rise to the two distinct α-subunits preceded the split of the eukaryotic crown groups.

Detailed phylogenetic reconstruction of the evolutionary relationships between α-subunits based solely on the TPRs shows that the α-subunits of each enzyme are monophyletic (Figure S1). It is then most likely that there was one and only one gene duplication event that created the two paralogous subunits, and that this happened at the base of the eukaryotic tree. Furthermore, the fact that the secondary structure composed of 15 repeating helices with TPRs is conserved suggests that the ancestral of prenyltransferases already had this configuration.

We find that the secondary structure of this subunit varies in evolution, displaying multiple domain insertions in different organisms, which we characterize below.

Rat RGGTase a-subunit displays an inserted globular domain between helices 11 and 12 12. The function of this domain is unclear, but it is clearly not involved in contacts with REP nor the Rab substrates 29. This globular domain is a β sandwich composed of eight strands in two sheets. It is a domain broadly related to the Ig fold, and in the two papers describing the structure of the complex it was termed an Ig-like domain. This is consistent with the CATH hierarchy, a fully automated classification of protein structures 30. Here we will consider instead the SCOP classification of evolutionary relationships between proteins structures, as it complements automated classification with manual curation 2131. SCOP 1.71 classifies this domain as an independent superfamily, included in the C2 domain-like fold. From now on we will refer to this domain as the C2 domain-like.

The full SCOP hierarchy for this domain is shown in additional file 1 (Figure S2). This fold includes superfamilies such as the C2 domain. This is a domain that is found in multiple eukaryotic proteins and is involved in signaling, vesicular transport, modification of lipids, among other functions 32. C2 domains usually regulate their respective protein function by establishing Ca²⁺-dependent and Ca²⁺-independent phospholipids complexes. One class of C2 domains can bind Ca²⁺ without binding phospholipids 33. Recently, the C2 domain of PKCd was shown to mediate protein-protein interactions by binding directly to phosphotyrosine peptides in a sequence-specific manner 34. It is unclear if the RabGGTase C2-lke domain displays any of these functions. Since the role of Rab isoprenylation is to allow hydrophilic Rab proteins to associated with cellular membranes, it is plausible to think that such modification should occur in proximity to those membranes. If this C2 domain-like is a phospholipid-binding domain, then it could play a role in bringing the prenylation reaction next to membranes. Ultra-structural studies could be used to test this hypothesis.

It was previously observed that this domain was not present in the yeast orthologue of α-subunit RGGTase, but that it would be present in worm also 15. We analyzed in detail the sequences between helices 11 and 12 of RGGTase, where the Rat C2-like domain is found 12, and also where the FTase a-subunit displays an inserted 3₁₀ helix 11. We found that only a restricted number of branches on the eukaryotic tree display insertions in this region (Figure 2). These include metazoa, plants and alveolata. All other branches of the tree have a predicted secondary structure similar to that of Bet2 in S. cerevisae, which does no display any insertion between the two TPRs.

Next we investigated whether the insert we found on the three distinct branches of the tree of life are similar, and hence likely to represent one single insertion event in the ancestor of all RGGTases that was subsequently lost in other branches. Or instead whether they are distinct domains, resulting from independent insertion or expansion events. Our results favor the second hypothesis (Figure 2). We find that the inserted domains are similar within taxonomical groups, but different beyond recognition across taxa. By difference beyond recognition we mean BLAST sequence searches 35, Pfam 36 and Superfamily 3738 domain assignments and secondary structure predictions 3940. Thus, in Metazoa, all Deuterostomes have a clearly defined C2-like domain, but the insects and the nematodes have a distinct insertion conserved solely in their taxonomical group. In the plants, angiosperms have a conserved domain in the same region of the α-subunit, but different beyond recognition from any other insert of the orthologous α-subunits. The same is true for alveolata. The size of the insertion is also conserved with taxonomic group but not across groups. For example, whereas deuterostomes have an insert of near 125 amino residues long, plants have a larger one, nearing 200 residues (Figure S3). Our results are thus compatible at least with up to five distinct insertion and/or expansion events in the same position of the α-subunit of RGGTase.

One exciting possibility is that all these insertions and or expansions represent the same function accomplished by different sequences, as this would expand our repertoire of sequence-function relationships.

Multiple sequence insertions/expansions in the same site suggest that this site is capable of accommodating structural variations more easily than others. Thus there seems to be a structural constraint in place. The fact that the paralogous α-subunit of FTase contains an inserted region in between these helices, in the form of a 3₁₀ helix adds support to this hypothesis. However, if the function of these distinct insertions and/or expansions is the same, then this could be the only place where this function is possible, and thus the recurrent use of the same site reflect functional rather than structural constraint. In the absence of information regarding the role played by these insertions/expansions it is not possible to resolve this question.

At the C-terminus of rat RGGTase α-subunit there is a Leucine Rich Repeat domain (Figure 1). This domain in not present in FTase (Figure 1). LRRs belong to the SCOP fold of the same name, which groups proteins forming a right-handed β-α superhelix 2131. It is formed by three superfamilies, the RNI-like, Outer arm dynein light chain and the L domain-like which includes the RGGTase LRR domain. LRRs are involved in a variety of biological processes, both in eukaryotes and prokaryotes. Their common role is the establishment of complexes with other proteins 4142.

We now investigate whether this LRR domain is a feature of all RGGTase α-subunits, which would suggest that it was present in the ancestral eukaryote, or if it is a recent acquisition restricted to a specific taxon or set of taxa. It is clear from the results shown in Figure 2 that the LRR domain is not universal, as we can only detect it in some animals, in angiosperms and in alveolata. This phylogenetic profile is consistent with two evolutionary scenarios – independent domain fusions or a single domain fusion at the base of the eukaryotic tree followed by a specific domain loss in multiple branches of the eukaryotic tree. Both scenarios seem equally unlikely, so we investigated this further using phylogenetic reconstruction based on the LRR domain sequences only. Our hypotheses is that there is enough phylogenetic signal in these sequences to solve this puzzle. In Figure 3 we show three phylogenies of the LRRs, including sequences from other superfamilies as reference, reconstructed by three distinct methods. Plant and vertebrate LRR are consistently monophyletic, suggesting a common origin. In contrast, the LRR of Ciona, of nematodes and of alveolata segregate with different reference sequences. This varies according to the method used to reconstruct phylogeny. It is thus impossible at this stage to resolve the question regarding the common or independent origin of the LRR sequences in the α-subunit. However, the recent observation that convergent evolution of domain architectures is very rare, with an estimated frequency of 0.4% to 4% 43, is more supportive of the first scenario.

The β-subunit of rat RGGTase is a α-α barrel composed of 12 α helices. It is very similar to the α-α barrel in the β-subunit of FTase 12. The β-subunits of prenyltransferases are more conserved than the α subunits, but identification and classification of RGGTase β-subunit was simple using a combination of BLAST searches of sequences databases followed by phylogenetic analysis (Figure S4).

We discussed above that the α-subunit display multiple sequence insertions in different species. Below we will show that in some species REP also display inserted sequences. The β-subunit in contrast appears to have an invariable domain architecture throughout evolution. We investigated its structural relatives in order to gain insight whether this is due to structural constraints. In order to do so, we investigated the SCOP hierarchical classification of protein structures. Prenyl transferase's β-subunits belong to the "α-α toroid" fold (SCOP: 48207), Terpenoid cyclases/Protein prenyltransferases superfamily (SCOP: 48239). This fold consists of multihelical proteins displaying up to seven alpha-hairpins arranged in a closed circular array. Therpene synthases are classified into the same superfamily. These proteins are characterized by two α-α domains in the same peptide chain; the first is an α6-α6 barrel of two concentric rings, whereas the second is a barrel with 10 α helices, one 3₁₀ helix and at least two β-strands inserted in between helical elements 44. It appears then that this superfamily can have varying number of helical elements as well as accommodate extra structural elements. It is then plausible that the invariable nature of the domain architecture of β-subunits is due to functional rather than structural constraints.

In animals and yeast it is clear that RGGTase is composed of independent α and β-subunits. In contrast GGTase I and FTase have distinct β-subunits, but share a α-subunit. Thus the GGTase I and FTase are related by a single gene duplication. In contrast, RGGTase is separated from the other prenyltrasnsferases by two gene duplications. Although the number of duplication steps that separates the different enzymes is clear, the order of duplication is not. In other words, we don't know which subunits emerged first and which resulted from these by duplication. We find that all organisms with a complete genome sequence that we investigated display a β-subunit of RGGTase (Figure 2). Considering we also always find a RGGTase α-subunit in the same organisms, it follows that the ancestor of all eukaryotes already had a distinct and fully separated RGGTase.

Rab escort proteins (REP) belong to the same protein family as Rab GDP dissociation inhibitors (RabGDI). They are both classified in the SCOP hierarchy 31 as FAD/NAD(P)-binding domain fold and superfamily, which suggests a common ancestry. Their structure comprises two domains: domain I include the Rab binding platform, whereas domain II in REP mediates binding to the alpha-subunit of RGGTase 2945. REP and RabGDI share conserved regions, termed SCRs (sequence conserved regions), which are highlighted in Figure 4 as brown boxes. Multiple sequence alignments of REPs and RabGDI reveal that mammalian REPs display an insertion between domain I and domain II, absent in RabGDI, which maps roughly to a sequence region delimited by the conserved regions SCR1B and SCR2 1345. This insert appears to be larger in vertebrate sequences than in S. cerevisae 13. The function of this insert is unclear, particularly at the light of the recent structure of REP1 in complex with RGGT, which shows that it is not involved in contacts with the RGGTase subunits nor with Rab substrates 29.

We identified REPs in all species studied here by searching GeneBank and the other genomic databases (see methods) with known REP sequences. We used phylogenetic reconstruction to classify the obtained sequences into the REP and RabGDI subfamilies (Figure 4B). We identified at least one distinct REP sequence in all species for which a complete genome sequence was available. Mammals have two paralogous REPS (REP-1 or Choroideremia and REP-2 or CHM-like). X. laevis also displays two paralogous REPs, but our phylogenetic analysis suggests that this is an independent and species-specific duplication (Figure 4b). It is interesting to note that in A. thaliana, the duplication of both enzyme subunits was not accompanied by REP duplication.

The most striking observation is that few branches of the tree of life are characterized by the presence of a larger insert region between SCR1B and SCR2 in REPs. Their size indicates that few independent taxa display inserts that are larger than those of yeast MRS6 (yeast REP) and RabGDIs (Figure 4C). At the sequence level there is no real conservation – the inserts are similar within taxa but very different across taxa. Systematic database searches using solely the insert regions from several specie can only find closely related REP sequences. For example, a BLAST search using the insert region of rat or human REP1 will find only deuterostomes REP inserts, but not plant inserts, and vice versa. They are also not similar to any other protein other than REPs, which means that we cannot use this approach to define hypothetical functions for this region.

In the crystal structure of rat REP1, this sequence insert corresponds to a region in the crystal with no clear electron density 29. Such regions are typically labeled natively unfolded 4647. Natively unfolded proteins are involved in a variety of cellular functions, namely transcriptional and translational regulation, signaling and regulation of the self-assembly of large multi-subunit complexes such as the ribosome and the bacterial flagellum 4748. They are also expected to be involved in a variety of human diseases 49. Although they can perform their function in the unfolded state, the majority of unfolded regions undergo a process termed induced-folding, in which upon binding to their physiological partners they undergo a transition to a structured form 50. We tested whether the different inserts were also natively unfolded, using the predictive algorithm Globplot2 51. We observed that angiosperm insert sequences are predicted to include a disordered, region, so are all vertebrate sequences and nematodes. In contrast, insects never show a predicted disordered region within the insert region (Figure 4C and S5). Thus, the presence of disordered regions is not restricted to the larger inserts of vertebrates, nor does it seem to correlate with insert size.

In most species studied, we find a single copy of the α- and the β-subunit. In contrast we found two RGGTase α- and β-subunits in A. thaliana. We have previously observed that duplication of whole protein complexes is frequent 16. We argued for the prevalence of stepwise duplications leading to complete duplication of subunits of complexes, like what is observed with the adaptin tetrameric complexes 52. The duplication of the two subunits of RGGTase in A. thaliana however appears to have occurred simultaneously, as a result of a whole genome duplication (WGD) estimated to have happened around 38 million years ago 53. The paralogous pairs localize to distinct chromosomes, to duplicated segments that were mapped to that WGD (α-subunit: At4g2424490 (chr.4) and At5g41820 (chr.5); β-subunit: At3g12070 (chr.3) and At5g12210 (chr.5)).

The functional relevance of maintaining these two copies of RGGTase is unclear. It is clear from all the phylogenetic trees that the subunits are very closely related (e.g. Figure S1), suggesting some selective pressure. A. thaliana may require high levels of this enzyme and concerted gene duplication is one way of boosting the levels of a given gene product 54. Alternatively, the large number and diversity of Rab GTPases in A. thaliana 5556 may require specific regulation of Rab prenylation. This could be achieved by having multiple copies of the enzyme subunits under differential regulation or displaying distinct substrate specificities.

Animals also display an expansion of the Rab family. They don't have multiple copies of the enzyme subunits; they have instead two paralogous REPs that appear to form complexes with distinct substrates and have distinct specificities to the RGGTase 57. This lends some support to the second hypothesis. It would be thus interesting to investigate whether there is specificity in the pairing of the A. thaliana subunits, and whether the different versions of the enzyme have distinct substrate specificities. In contrast, some animals display two or more copies of the β-subunit for a single α-subunit (Figure 2). Since substrate-binding specificity is not defined by the β-subunit in RabGGTase, it seems plausible that expression levels are at the root of these duplications. Amoebozoa like D. discoideum, which represents another example of independent expansions of the Rab family, have in excess of 50 Rab proteins (unpublished observations). D. discoideum only has one copy of each component (Figure 2), indicating that large Rab families are compatible with a single copy of each RabGGTase component.

The duplication of components of RabGGTase appear thus to be restricted to specific branches of the tree of life (Figure 2). Our analysis also indicates that it is also restricted to RabGGTase. We did not observe duplications of FTase nor GGTase I subunits. This is despite the fact that their substrates show expansions akin to those of the Rab family. For example, the Ras and Rho families expanded from 3 and 6 members respectively in S. cerevisiae, to 22 and 34, respectively, in H. sapiens 58. It is possible that dosage balance may place a barrier to duplications of subunits 1859.