Evolutionary connection between the catalytic subunits of DNA-dependent RNA polymerases and eukaryotic RNA-dependent RNA polymerases and the origin of RNA polymerases
- 19k Downloads
The eukaryotic RNA-dependent RNA polymerase (RDRP) is involved in the amplification of regulatory microRNAs during post-transcriptional gene silencing. This enzyme is highly conserved in most eukaryotes but is missing in archaea and bacteria. No evolutionary relationship between RDRP and other polymerases has been reported so far, hence the origin of this eukaryote-specific polymerase remains a mystery.
Using extensive sequence profile searches, we identified bacteriophage homologs of the eukaryotic RDRP. The comparison of the eukaryotic RDRP and their homologs from bacteriophages led to the delineation of the conserved portion of these enzymes, which is predicted to harbor the catalytic site. Further, detailed sequence comparison, aided by examination of the crystal structure of the DNA-dependent RNA polymerase (DDRP), showed that the RDRP and the β' subunit of DDRP (and its orthologs in archaea and eukaryotes) contain a conserved double-psi β-barrel (DPBB) domain. This DPBB domain contains the signature motif DbDGD (b is a bulky residue), which is conserved in all RDRPs and DDRPs and contributes to catalysis via a coordinated divalent cation. Apart from the DPBB domain, no similarity was detected between RDRP and DDRP, which leaves open two scenarios for the origin of RDRP: i) RDRP evolved at the onset of the evolution of eukaryotes via a duplication of the DDRP β' subunit followed by dramatic divergence that obliterated the sequence similarity outside the core catalytic domain and ii) the primordial RDRP, which consisted primarily of the DPBB domain, evolved from a common ancestor with the DDRP at a very early stage of evolution, during the RNA world era. The latter hypothesis implies that RDRP had been subsequently eliminated from cellular life forms and might have been reintroduced into the eukaryotic genomes through a bacteriophage. Sequence and structure analysis of the DDRP led to further insights into the evolution of RNA polymerases. In addition to the β' subunit, β subunit of DDRP also contains a DPBB domain, which is, however, distorted by large inserts and does not harbor a counterpart of the DbDGD motif. The DPBB domains of the two DDRP subunits together form the catalytic cleft, with the domain from the β' subunit supplying the metal-coordinating DbDGD motif and the one from the β subunit providing two lysine residues involved in catalysis. Given that the two DPBB domains of DDRP contribute completely different sets of active residues to the catalytic center, it is hypothesized that the ultimate ancestor of RNA polymerases functioned as a homodimer of a generic, RNA-binding DPBB domain. This ancestral protein probably did not have catalytic activity and served as a cofactor for a ribozyme RNA polymerase. Subsequent evolution of DDRP and RDRP involved accretion of distinct sets of additional domains. In the DDRPs, these included a RNA-binding Zn-ribbon, an AT-hook-like module and a sandwich-barrel hybrid motif (SBHM) domain. Further, lineage-specific accretion of SBHM domains and other, DDRP-specific domains is observed in bacterial DDRPs. In contrast, the orthologs of the β' subunit in archaea and eukaryotes contains a four-stranded α + β domain that is shared with the α-subunit of bacterial DDRP, eukaryotic DDRP subunit RBP11, translation factor eIF1 and type II topoisomerases. The additional domains of the RDRPs remain to be characterized.
Eukaryotic RNA-dependent RNA polymerases share the catalytic double-psi β-barrel domain, containing a signature metal-coordinating motif, with the universally conserved β' subunit of DNA-dependent RNA polymerases. Beyond this core catalytic domain, the two classes of RNA polymerases do not have common domains, suggesting early divergence from a common ancestor, with subsequent independent domain accretion. The β-subunit of DDRP contains another, highly diverged DPBB domain. The presence of two distinct DPBB domains in two subunits of DDRP is compatible with the hypothesis that the ultimate ancestor of RNA polymerases was a RNA-binding DPBB domain that had no catalytic activity but rather functioned as a homodimeric cofactor for a ribozyme polymerase.
KeywordsRoot Mean Square Deviation Core Catalytic Domain Catalytic Cleft RDRP Sequence Detectable Sequence Similarity
Polymerization of ribonucleoside triphosphates (NTPs), which is central to a variety of crucial biological processes, including transcription, primer synthesis during DNA replication, addition of polyA tails to mRNA, addition of CCA to tRNA, uridylation in RNA editing, replication of RNA viruses, amplification of RNA in eukaryotic post-transcriptional gene silencing (PTGS), and oligoadenylate synthesis during interferon signaling, is catalyzed by a broad variety of distinct enzymes [1, 2, 3, 4]. These polymerases belong to two major mechanistic categories, those that copy a nucleic acid template and those that are independent of a template. The DNA-dependent RNA polymerases (DDRPs) involved in the transcription of cellular and DNA viral genes, primases, and RNA-dependent RNA polymerases (RDRPs) of RNA viruses and cellular PTGS systems are template-dependent polymerases. Template-independent RNA polymerases (nucleotidyltransferases) include CCA-adding enzymes, polyA polymerases, uridylyl transferases and oligoA-synthetases. Some of the RNA polymerases function as single-subunit proteins, whereas others form large complexes that consist of multiple, distinct subunits; however, in all known cases, the basic catalytic activity maps to a single polypeptide containing a characteristic metal-chelating active site [5, 6, 7]. Typically, active sites of polymerases contain acidic or polar residues, which coordinate divalent metal cations, most often Mg2+. The metal cations direct a 5' nucleoside triphosphate to form a phosphoester bond with the 3' hydroxyl of the preceding nucleotide (the 3'-terminal nucleotide of a growing polynucleotide chain), with the elimination of pyrophosphate .
Despite the basic biochemical similarity among all RNA polymerases, sequence and structure comparisons indicate that these enzymes belong to at least five evolutionarily unrelated folds. The RDRPs of RNA viruses define one major lineage of nucleic acid polymerases, which additionally includes reverse transcriptases, archaeo-eukaryotic DNA polymerases, and nucleotide cyclases [8, 9, 10, 11, 12, 13]. The DNA-dependent RNA polymerase of certain bacteriophages, such as T7, and the archaeo-eukaryotic primase (also detected in some bacteria) are divergent derivatives of the same fold [11, 14]. The core catalytic domain of all these enzymes, the so-called "palm" domain, has an RNA-recognition motif (RRM)-like fold with strategically placed metal-coordinating residues, which form the active site [11, 15, 16]. In contrast, bacterial DnaG-type primases (also present in archaea and some eukaryotes) contain a polymerase domain of the Rossmann-like TOPRIM fold, which is shared with topoisomerases and OLD-family nucleases [17, 18, 19]. The recently solved structures of the DDRPs from yeast and the thermophilic bacterium Thermus thermophilus indicate that the β' subunit (according to the subunit nomenclature of Escherichia coli DDRP, which we hereinafter employ to designate all orthologs of the respective E. coli subunits) of these enzymes defines another distinct catalytic scaffold, which is unrelated to any of the above template-dependent RNA polymerases [20, 21, 22, 23, 24]. Additionally, the structural and evolutionary affinities of two other template-dependent RNA polymerases, namely RDRPs involved in PTGS [25, 26, 27] and primases of herpesviruses , remain obscure. In contrast to the template-dependent RNA polymerases, which have several distinct scaffolds of the catalytic domains, all template-independent RNA polymerases have the same fold of the principal catalytic domain and belong to the pol β superfamily of nucleotidyl transferases [29, 30].
Thus, RNA polymerase activity apparently has been independently "invented" on several occasions. Nevertheless, DDRP is one of the most conserved enzymes, which is represented, without exception, in all cellular life forms [31, 32]. The DDRP complex from most organisms consists of 5 to 15 polypeptides. Of these, four subunits, which correspond to the bacterial α, β, β' and ω, are universally present in all cellular DDRPs and constitute the conserved DDRP core . Orthologs of the β and β' subunits are also encoded in the genomes of several families of large eukaryotic DNA viruses [33, 34, 35]. Biochemical studies have shown that the catalytic site of DDRP resides in the β' subunit and contains three invariant aspartates that coordinate a Mg2+ cation [22, 23, 24, 36]. Recent structural analyses demonstrated that the core of DDRP is assembled around the β and β' subunits, which interact with each other to form a positively charged nucleic-acid-binding cleft. The α subunit further stabilizes this cleft, whereas the ω subunit interacts with the β' subunit . Given their ubiquitous presence in all cellular life forms and high level of sequence conservation, it seems evident that the core subunits of this complex RNA synthesis machine were already present in the last universal common ancestor (LUCA) of the extant cellular life forms and performed functions mechanistically similar to those of modern DDRP.
In contrast to the ubiquitous DDRP, cellular RDRPs that are involved in PTGS so far have been detected only in eukaryotes [1, 26]. The PTGS phenomenon covers a variety of complementarity-dependent silencing pathways, such as RNA interference (RNAi) in animals and slime mold, co-suppression (silencing of transgene and the corresponding endogenous genes) and virus gene resistance in plants, and quelling in fungi, all of which share a common mechanism of RNA turnover [28, 37, 38, 39, 40]. Essentially, double-stranded (ds) RNA, which is formed in these processes, triggers the activation of a sequence-specific RNA degradation system, which targets homologous RNAs [41, 42]. The dsRNA is broken down by the Dicer enzyme into 21–25 nucleotide (nt) fragments called small interfering RNA or siRNAs [43, 44, 45]. The siRNAs subsequently associate with a complex of proteins called RISC and target homologous RNA by serving as guides for multiple rounds of RNA cleavage [45, 46].
The RDRP is a component of the PTGS system that has been initially described in plants as a cellular RNA polymerase activity induced upon viral infection that synthesized antisense RNA in a primer-dependent or independent manner [26, 47, 48, 49, 50]. Subsequent genetic studies showed that mutations in RDRP genes impaired PTGS in a variety of systems [25, 38, 51, 52]. Experimental studies in C. elegans and in plants suggested that the RDRP is involved in production and amplification of dsRNA using the target RNA as template and siRNAs as primers or guides and resulting in amplification of the RNA silencing response [53, 54, 55, 56].
The RDRP is present in one or more copies in a wide range of eukaryotes, from early-branching parabasalids, such as Giardia, to multicellular forms, including fungi, plants and animals [57, 58]. This broad representation in eukaryotes, including organisms that are considered primitive, suggests that RDRP might have been encoded in the genome of the common ancestor of all modern eukaryotes. However, this gene has been subsequently lost in several eukaryotic lineages, such as the yeast Saccharomyces cerevisiae, insects and vertebrates. Although RDRPs display some diversity in domain architecture and the level of sequence conservation, they form a tight, well-conserved family of large proteins with no detectable prokaryotic or viral homologs.
Here, we investigate the evolutionary history of RDRPs and DDRPs in an attempt to unveil the origin of the RDRP. We identify the first homologs of the cellular RDRP outside the eukaryotic clade, in bacteriophages, and show that RDRPs and DDRPs share a homologous catalytic core, which comprises a six-stranded double-psi-β-barrel (DPBB) domain. Evidence is presented that the ultimate ancestor of RNA polymerases might have been a RNA-binding DPBB domain, which functioned as a cofactor for a polymerase ribozyme, and that subsequent evolution of DDRPs and RDRPs proceeded via accretion of various simple modules, such as the sandwich-barrel hybrid motif domain, around this conserved core.
Results and discussion
Bacteriophage homologs of the eukaryotic RDRPs and prediction of their active site
To investigate the evolutionary affinities of the RDRPs, we carefully examined the results of BLAST searches of the non-redundant (NR) protein sequence database (National Center for Biotechnology Information, NIH, Bethesda) for various RDRP sequences. These searches identified a large region of 700–800 residues as the conserved module shared by all RDRPs. This region was further examined by searches of NR using the PSI-BLAST program, which was iterated to convergence with a profile inclusion threshold of E = 0.01. These searches, e.g. the search initiated with the core RDRP module from Petunia (residues 37–775), readily retrieved RDRPs from plants, C. elegans, Dictyostelium discoideum, several fungi, and Giardia lamblia with statistically highly significant expectation (E) values. Interestingly, the second iteration of this search retrieved the YonO protein from the Bacillus subtilis phage Spβc2 with a significant E-value (7 × 10-3). YonO was also detected with significant E-values in searches initiated with other eukaryotic RDRP modules, e.g., that of Schizosaccharomyces pombe. Reciprocal searches started with the YonO protein sequence retrieved closely related proteins encoded by prophages in the genome of Clostridium acetobutylicum (CAC1139) and Clostridium perfringens (CPE1103). Hereinafter, we refer to these bacteriophage proteins as the YonO-like RDRP homologs (YRHs).
Structural and evolutionary relationship between the catalytic domains of the RDRPs and the DDRPs
Interestingly, the PSI-BLAST searches started with the sequences of the RDRP module of the RDRPs and the YRH proteins consistently retrieved the β' subunits of DDRPs, albeit at statistically not significant E-values. The high-scoring segment pairs (HSPs) detected in these searches aligned the highly conserved region of the RDRPs between the predicted strands 18 and 20, including the DbDGD motif, with the portion of the DDRP β' subunit sequence, which contains the metal-chelating active site, with a similar conserved motif, DxDGD. Additionally, a search of the NR database with a position-specific scoring matrix (PSSM) that included all unique RDRP and YRH sequences detected, with borderline E-values, a large protein from Corynebacterium glutamicum (Cgl1702), which also contained a DXDGD motif. Reciprocal searches showed that this protein contained two regions of statistically significant similarity to the DDRPs. The N-terminal region of similarity corresponded to the conserved portion of the β subunits and the C-terminal region corresponded to the core of the β' subunit and included the DXDGD motif (see discussion below). Despite the low statistical significance of the similarity between RDRPs and DDRPs, the presence of the same signature motif in the experimentally identified or predicted active sites of the two classes of enzymes prompted us to perform a detailed comparison of the catalytic domains of these two classes of RNA polymerases.
Among nucleic acid polymerases, a DxDxD motif is conserved only in the RDRPs, the DDRP β' subunits and the euryarchaea-specific DNA polymerase subunit II. However, in the latter polymerase, the motif has the signature DGDED, as opposed to the DxDGD pattern shared by the RDRPs and DDRPs. To investigate the distribution of this motif in the entire protein database, we conducted pattern searches using the GREF program of the SEALS package, with queries corresponding to the residue conservation profile in the predicted active sites of the RDRPs. All these queries contained the DxDGD motif with additional flanking conserved residues. The only proteins with evolutionarily conserved DxDGD motifs that were detected in these searches were the RDRPs, the β' subunit of DDRPs, the integrin calcium-binding module, and the EF-hand domain (data not shown). Among these domains, the motif was embedded in the context of conserved (predicted) β-strands only in the RDRPs and DDRPs. Moreover, in searches conducted with extended queries, such as DbDGDxhxh or DGDxhxh patterns (h is a hydrophobic residue), the RDRPs and the β' subunits of DDRPs were identified as the only protein families in which these motifs were conserved. We also ran database searches using the PHI-BLAST program that combines a regular BLAST search with a pattern search . In these searches, the active site motif was used as the pattern query, and various RDRP sequences as the sequence queries. All these searches retrieved the YonO proteins with significant E-values and some of the DDRP β' subunits with borderline E-values. For example, searches seeded with the RDRP (RrpB gene product) sequence from Dictyostelium (gi:14475571) and the pattern [GAS]-D- [FLYMQN]-D- [G]-D-X- [ACLIVMFY]-X- [ACLIVMFY] detected the rice DDRP catalytic subunit with an E-value of 0.048.
Examination of the recently solved structures of the DDRP showed that, although β' subunit is a large protein, the active site containing the DxDGD motif mapped to a small, compact β-barrel domain. A search of the PDB database using the DALI program , with this β-barrel from Thermus thermophilus β' subunit submitted as a query (pdb: 1iw7, chain D region: 625–749), retrieved the N-terminal domain of the CDC48-like AAA ATPases, formate dehydrogenases, aspartate decarboxylases and Barwin. The reported structural alignments span ~70 residues with significant Z-scores (8.8–6.6) and root mean square deviations (RMSD) between the C-α backbones in the range of 1.8–2.5 Å, which is characteristic of homologous domains. The region of structural similarity shared by the catalytic domain of the β' subunit of the DDRPs and the above proteins corresponds to a distinct globular fold known as the double-psi beta-barrel (DPBB) [62, 63, 64]. The SCOP database included the RNA polymerase subunits in the multidomain protein class but mentions the presence of a DPBB domain in each of the β and β' subunits (; see below).
The DPBB domains and early evolution of the catalytic domain of RNA polymerases
Although the cores of the two DDRP subunits are homologous and are likely to have evolved from a common ancestor (see discussion below), they have completely different sets of conserved residues and contribute distinct moieties to the catalytic cleft (Figs. 3, 4). Thus, these specific active residues apparently emerged after the divergence of the β and β' subunit DPBBs, whereas the ancestral DPBB homodimer probably did not have catalytic activity and merely bound a ribozyme that originally catalyzed the RNA polymerization. Subsequently, the two subunits diverged from each other, with one (β') acquiring a divalent cation-binding site in the insert between the two terminal strands of the DPBB, whereas the DPBB in the other (β) subunit acquired two basic residues that interacted with the nucleic acid. At this stage, which might be considered the emergence of modern-type RNA polymerases, the catalytic activity probably was taken over by the protein.
The basic DPBB fold consists of six β-strands and two variable regions, which are located after strand 2 and strand 5 and, at least in some case, adopt a helical conformation [62, 63, 64]. The domain can be further split into two symmetrical structural units, each with three core β-strands. These two units are combined into a barrel with a complex topology, with strand 5 wedged between strands 1 and 2 and strand 2 wedged between strands 4 and 5, yielding two 'psi' loop structures (Fig. 4). The psi-loops between strands 1 and 2 and between strands 4 and 5 often harbor the active site or substrate-interaction residues of the respective DPBBs. Sequence comparisons and identification of specific conserved motifs showed that the DPBB fold includes five major superfamilies. These superfamilies are typified by the formate dehydrogenase C-terminal (FDHC) domain, Barwin endoglucanase, aspartyl protease, and the DPBB domains from the β and β' subunits of DDRP. Two of these superfamilies show a restricted phyletic distribution. Barwin-like endoglucanases are present only in plants and fungi, whereas aspartyl proteases are found in eukaryotes, retroviruses and a minority of bacteria. The latter superfamily also has a circular permutation, which suggests secondary derivation. Thus, none of these superfamilies are likely to have been present in the last universal common ancestor (LUCA) of extant life forms. In contrast, the DPBBs detected in DDRP subunits are universally present in all modern life forms and hence are traceable to LUCA. The FDHC superfamily includes several distinct families. The N-terminal domain of the CDC48-like AAA ATPases is present largely in eukaryotes and archaea and the N-terminal domain of thymidine phosphorylases is archaea-specific. In contrast, the FDHC-like DPBB domain associated with oxidoreductases, such as formate dehydrogenase and molybdopterin-containing dehydrogenases, is widespread in archaea and all major lineages of free-living bacteria. Thus, at least one DPBB domain of the FHDC superfamily probably was present in LUCA.
The above analysis shows that LUCA probably encoded three distinct DPBB domains. Although the DPBB domain in the RNA polymerase β subunit is distorted by inserts and shows extreme sequence divergence, its role in the formation of the catalytic cleft and the spatial arrangement with respect to the more canonical DPBB of the β' subunit (Fig. 4) are compatible with the hypothesis that the two DPBB domains of DDRP shared a common ancestor to the exclusion of the other ancient, FDHC-like DPBB. A structure-based sequence alignment shows that the DPBB domains of the RNA polymerases share several conserved residues with other DPBB domains. These include 7 hydrophobic residues, two sets of symmetrically positioned small (mostly glycine) and polar (mostly aspartate) residues preceding strands 3 and 6, respectively, and at least three other small residues, two polar residues and two bulky residues (Fig. 3). Of these, the small residues preceding strands 3 and 6 contribute a positive φ angle, which results in turning of the C-α backbone to stabilize the β-barrel . These shared features notwithstanding, the ancient DPBBs have no conserved residues in their catalytic or substrate-binding sites. This is best compatible with the notion that the common ancestor of the DPBB domains was a generic binding domain devoid of high specificity or catalytic activity (see also discussion above).
The same three-stranded structural units that interlock to give rise to DPBB also form another type of β-barrel via terminal dimerization. This is the E longation factor-I somerase (EI)-barrel found in ribosomal proteins, such as L3, GTPase translation factors, ferredoxin reductase, and L-fucose isomerase [62, 66]. It seems likely that the EI-barrel and the DPBB evolved from ancient three-stranded ancestral units, which were stabilized via the formation of different barrel structures through homo-dimerization.
Ancient conserved domains and evolution of DNA-dependent RNA polymerases
While all bacteria, archaea and eukaryotes encode orthologs of at least four distinct subunits of DDRP, the divergent RNA polymerases of baculoviruses and yeast killer plasmids and the Corynebacterium Cgl1702 protein only contain counterparts to the β and β' subunits. The minimal RNA polymerase seems to be represented by the predicted catalytic core of Cgl1702, which appears to consist entirely of the β-type and β'-type DPBBs, in this case juxtaposed in the same polypeptide. These observations are not particularly surprising given that β and β' subunits form the catalytic cleft of the DDRP (Fig. 3), whereas the other subunits occupy peripheral positions in the complex. In an attempt to gain further insight into the early evolution of the catalytic core of RNA polymerases, we sought to identify additional globular domains in β and β' subunits.
Although the sequences of the Zn-ribbons of bacterial and archaeo-eukaryotic DDRP subunits have diverged considerably, their similar location in the orthologous polypeptides from different kingdoms suggests that β' subunit of LUCA already contained a Zn-ribbon. Based on the crystal structure, it has been proposed that the region containing the Zn ribbon may interact with the promoter during transcription initiation in bacteria . Hence, it is plausible that a nucleic-acid-binding Zn-ribbon had been recruited to the ancestral polymerase core, which consisted of the DPBB domains, and contributed to non-specific interactions with the template. Consistent with this, the Zn-ribbon had been subsequently reused in several basal archaeo-eukaryotic transcription factors, such as TFIIB, TFIIE and TFIIS [67, 68, 71, 72, 73].
A potential additional nucleic-acid-binding moiety that is conserved in all β'-subunits is the AT-hook like module (Fig. 6). The AT-hook is a simple, small protein module that consists of a flap-like structure with a positively charged surface that is inserted into the minor groove of DNA . The two AT-hook-like modules (residues 583–603 of 1IW7 chain D) in the DDRP β' subunit are inserted within the α-helix that is located immediately upstream of the DPBB domain (Fig. 6). Previously, AT-hooks have been detected primarily in eukaryotic chromatin-associated proteins, particularly transcription factors, where they function as accessory DNA-binding domains . The ubiquity of the AT-hook-like modules in the β' subunits of DDRPs suggests that they were recruited for nucleic acid recognition in LUCA if not earlier.
The core SBHM domain consists of a repeat of two three-stranded units with the characteristic "waist"-like structure occurring between the second and third strands of each repeat (Fig. 8). In the complete SBHM, strand 1 packs with strand 6 and the loops bounded by the two "waist"-like regions adopt an extended conformation resulting in the formation of a barrel. In the crystal structure of DDRP, the SBHM domain inserted into the β-subunit DPBB is in contact with the outward-projecting insert between strands 2 and 3 of the β'-subunit DPBB. Furthermore, the SBHM also contributes to the stabilization of the catalytic site by forming an interface between the two DPBBs (Fig. 4). The universal presence of the SBHM inserted in the DPBB domain of the DDRP β-subunit suggests that this domain was already present in the common ancestor of all extant DDRPs. The wide spread in different phylogenetic lineages of two other SBHM domains, namely those found in dihydrolipoamide acetyltransferase and the glycine-cleaving enzyme, is indicative of their presence in LUCA . The SBHM of the DDRPs is specifically related to the biotin/lipoate-binding SBHMs. The two share a conserved basic residue that is present immediately prior to strand 4. In all biotin/lipoate-binding SBHMs, this position is occupied by a lysine, which covalently binds organic radical ligands (Fig. 7), but there is no evidence for such a role in the DDRP β subunits. These observations suggest that the ancestral SBHM interacted with various ligands with a low specificity. From such a precursor, which existed prior to LUCA, the SBHM domain apparently diversified into two forms, one that covalently bound organic radicals, and another that specialized in non-covalent interactions with proteins or nucleic acids. This latter form was recruited into the RNA polymerase catalytic core where it provided additional surfaces for the interaction between the DPBB domains and formed part of the interface between the catalytic domains. Thus, it appears likely that the ancestral DDRP evolved from the coalescence of at least 3 distinct domains: i) the DPBB that probably originally bound RNA and, subsequently, segregated into the cores of the β' subunit, where it acquired the metal-coordinating active site, and of the β subunit, with the two lysines projecting into the catalytic cleft (Fig. 4), ii) the Zn-ribbon, which was probably involved in interactions with the template, and iii) the SBHM domain that stabilized the interactions between the DPBB domains forming the catalytic cleft (Fig. 6). Additionally, two AT-hook like modules, which probably had an accessory DNA-binding function, were also inserted into a conserved α-helix directly N-terminal of the DPBB domain in the β' subunit.
The remaining conserved regions in the β and β' subunits of the DDRPs do not show detectable relationship with any other ancient conserved domains. Examination of these structures shows that they are principally composed of large α-helical hairpins that form coiled coil structures or stretches with successive shorter α-helices. Some of the other conserved regions consist of β-meanders with unusually long β-strands. These regions probably emerged, respectively, through duplication and divergence of a simple α-helical unit or through stabilization of long loops by selection for hydrophobic residues, resulting in the formation of the core units of the β-meanders.
Proliferation of the Sandwich-Barrel Hybrid Motif and lineage-specific innovations of the large subunits of bacterial DDRP
Further visual examination and structural alignments revealed the presence of one additional SBHM domain in the β subunit (1iw7 chain C: region 592–659) and several repeats of this domain in the β' subunit (1iw7 chain D, regions: 163–195, 248–308, 311–368, 369–419 and 1270–1329) of the bacterial DDRPs (Fig. 6). A PSI-BLAST search started with the N-terminal SBHM from the Aquifex aeolicus DDRP β' subunit detected several SBHMs from proteins other than the β' subunit, such as acetylornithine deacetylase (E = 2 × 10-7, iteration 2), the C-terminal region of bacterial cytochrome F (E = 10-4, iteration 2), biotin carboxyl carrier domain of biotin transcarboxylases (E = 6 × 10-4, iteration 2) and the phosphotransferase system enzyme II (E = 10-3, iteration 2). This search also retrieved the bacteria-specific C-terminal SBHM domain of the β subunit (e.g. Thermotoga, E ~10-3) and, interestingly, also a SBHM in the N-terminal region of the NusG protein from Thermotoga maritima (E = 2 × 10-7, iteration 2). Reverse searches with the C-terminal region of cytochrome F (gi: 11467387, residues 150–321) retrieved the bacterial β' sequences in the second and subsequent iterations with statistically significant E-values. None of these additional SBHMs were detectable in the archaeal or eukaryotic DDRPs, either in sequence searches or in direct comparisons of the crystal structures. Thus there appears to have been a lineage-specific proliferation and dissemination of the SBHM domains in bacteria.
Bacterial β and β' subunits show major variability in the distribution of the SBHM domains (Fig. 6). All bacterial β subunits share one additional SBHM, which is located N-terminal of the DPBB domain. Similarly, an extra SBHM, located C-terminal of the DPBB, is conserved in bacterial β' subunits. In some cyanobacteria, the portion of the β' subunit containing the SBHM occurs as a separate polypeptide. In addition to the SBHMs that are conserved among all bacterial lineages, a cassette of up to four SBHMs are present N-terminal of the DPBB domain in the β' subunit of Thermus, Deinococcus and Thermotoga (Fig. 6). Similarly, the β' subunit of proteobacteria, Aquifex, spirochaetes, cyanobacteria and chlamydiae contains an additional set of SBHMs C-terminal of the DPBB domain, preceding the C-terminal SBHM that is shared by all bacteria (Fig. 6). The bacterial-specific SBHM domains present in the β and β' subunits are typically located on the periphery of the holoenzyme, which makes them accessible for interactions with other proteins. Two of these SBHMs (regions 163–195 and 369–419 in 1iw7 chain D) are involved in the interaction with the σ 70 subunit, which is critical for transcription initiation in bacteria . Thus, it appears likely that, originally, the SBHM domain of the β subunit participated in generic protein-protein interactions that allowed the ancestral DDRP to recruit accessory subunits. Subsequently, in the bacterial lineage, proliferation of SBHMs provided for several other, specific interactions involved in initiation and elongation.
Several of the SBHMs that are specific to the bacterial DDRPs show degradation of the "waist"-like loop of the first internal repeat. This is clearly a derived state because the SBHM originally evolved through the duplication of a single ancestral unit that contained an intact "waist"-like loop between strands 2 and 3. The form of the SBHM with a shortened loop in the first internal repeat is additionally seen in several bacterial-specific families of SBHMs. These include the SBHM of the phosphotransferase system enzyme II (glucose permease or PTS-EII) found in several lineages of free-living bacteria, the one at the C-termini of cytochrome F from cyanobacteria and chloroplasts, and the NlpD-like cell wall peptidases that are present in most bacteria. In addition to the specific structural relationship between some of the SBHMs of the DDRPs and those present in these bacterial proteins, they also show detectable sequence similarity. For example, a PSI-BLAST search initiated with the SBHM from NlpD retrieved the derived SBHMs of the bacterial RNA polymerase subunits (e.g. Streptococcus β subunit, E = 10-4, iteration 2) well before the classical SBHMs (eg. Mesorhizobium dihydrolipoamide acetyltransferase, E = 10-3, iteration 4). The β and β' subunits of the bacterial DDRP are the only proteins that contain both versions of the SBHM (Fig. 7). Hence, it appears likely that the form of the SBHM with the degraded loop in the first internal repeat initially evolved in the bacterial DDRP subunits. Subsequently, these domains appear to have been recruited for various interactions in other bacterial proteins.
The RNA polymerase α subunit-core-related domain and other elaborations of the archaeo-eukaryotic RNA polymerase catalytic subunits
The bacterial DDRPs have two α-subunits that form a homodimer at the "base" of the β-β' dimer and stabilize the catalytic-binding cleft. In the archaeo-eukaryotic lineage, there is a duplication of the α-subunit, resulting in two paralogous subunits, RBP3 and RBP11. In the eukaryotic DDRP complex, these subunits occupy a spatial position very similar to that of the α-subunit homodimer of the bacterial DDRPs. All these proteins share a common α + β core domain, with two α-helices, which form the dimer interface. In the SCOP database, the α-subunit core domains have been classified together with several other domains, including the ligand-binding domain of the bacterial arginine repressor, the dimerization cofactor of the transcription factor HNF1 (DCOH), and a domain specific to the bacterial aspartyl-tRNA synthetases. To investigate this domain further, we conducted structural similarity searches of the PDB database using the DALI program, with the α-subunits of the bacterial and eukaryotic DDRPs used as the queries. These searches showed that the core domain of the α-subunits were most similar (Z scores 5.4–7, with RMSDs of 2.2–3 Å over the aligned C-α residues) to the arginine repressor ligand-binding domain  and several domains that were not previously recognized as being related (Fig. 9B). These newly detected domains included the SUI1 domain, that is found in eIF1 and related proteins , the third globular domain of DNA gyrase/Topoisomerase II and, most interestingly, a distinct globular domain in the yeast DDRP β' subunit (PDB: 1k83; region 1143–1271). Reciprocal DALI searches with these domains readily recovered the α-subunits of DDRP as the best hits. For example, the domain from the yeast β' subunit hit the α-subunits of bacterial DDRPs with Z-scores of 5–7 and RMSD of 2.8–3 Å. Hereinafter we refer to these domains as the α-subunit core-related (ASCR) domains. The ASCR fold has a β2-α-β2-α topology and forms a two-layered structure with a 4-stranded β-sheet and characteristic strand order of 1-2-4-3 (Fig. 9B). The ASCR domains of bacterial α-subunits and eukaryotic RBP3 share a common insert domain between helix-1 and strand 3, which apparently has been lost in the RBP11 subunit of the eukaryotic DDRPs.
The ASCR domain in the yeast β' subunit is inserted into another α + β domain. Visual examination of the topology and structural searches of the PDB database using the DALI program showed that this domain it adopts an RNA Recognition Motif (RRM)-like structure with a β-α-β2-α-β topology forming a 4-stranded β-sheet. These two domains comprise a conserved module that is specifically shared by the archaeo-eukaryotic DDRPs to the exclusion of the bacterial β' subunits (Fig. 6). This two-domain module is inserted precisely in the same region as the C-terminal SBHMs seen in the bacterial lineage (Fig. 6). The closest relatives of the ASCR domain detected in the β' subunit are those present in the α-subunits (Fig. 9B). This observation suggests that the primary event in the divergence of the archaeo-eukaryotic β' subunit was the insertion of an RRM-like domain into the ancestral core. This was followed by the insertion of the ASCR domain, which probably evolved as a result of a duplication of the RBP3/11 subunits, into this RRM-like domain. Both the spatial arrangement of the ASCR domain in the β' subunits and its extreme sequence divergence with respect to the ASCR domains of the α-subunit suggests that it has acquired a new function. The RRM-like domain and the ASCR domain are positioned in the exterior entrance to the catalytic cleft and could interact with cofactors or with nucleic acid. There is no evidence that the ASCR domain in the DDRP α-subunits binds nucleic acids and studies on SUI1 protein and the arginine repressor ligand-binding domain implicate the ASCR domain in protein-protein interactions and small molecule binding [77, 79]. However, most of the ASCR domains are present in proteins that interact with nucleic acids (see above), raising the possibility of a role in non-specific nucleic acid binding.
Mirroring the insertion of the ASCR domain in the β' subunit, a distinct α + β globular domain, N-terminal to the DPBB domain (region 578–632 of 1K83, Chain B), is inserted in the β subunits of all archaeal and eukaryotic DDRPs (Fig. 6). This domain contains a 4-stranded sheet with a β2-α-β2 topology, with the strands in a 2-1-4-3 order. Visual examination of the topology and searches of the PDB database using the DALI program showed that this insert domain has the β-grasp fold [80, 81]. For example, in the DALI search, the Staphylococcus immunoglobulin-binding protein (IGBP; PDB code 2igd) was detected with a Z score of 5.2 and a RMSD of 2.7 over ~50 residues. This insert domain in the archaeo-eukaryotic β subunits resembles the version of the β-grasp domain that is present in the bacterial translation initiation factor IF3 and the IGBPs, in contrast to the distinct version found in ubiquitins, ferredoxins and ThiS. Specifically, the β-subunit insert domain shares a 4-stranded core with the former group and, like these proteins, lacks a long insert between the two terminal strands (data not shown). Given that this domain lies in the periphery of the holoenzyme, it is likely to participate in interactions with cofactors, which is compatible with the protein-protein interaction function identified in several proteins with the β-grasp fold [80, 81].
Thus, a common set of ancient conserved modules appears to have been repeatedly reused in the evolutionary diversification of the RNA polymerase subunits from their ancestral cores. The lineage-specific domain insertions in bacteria and in archaea-eukaryotes correlate with the fundamental differences between the basal transcription machinery of these two lineages [82, 83]. This suggests congruent evolution of the RNA polymerase catalytic subunits and the basal transcription factors .
Origin of the RDRPs
The evidence of an evolutionary relationship between the core catalytic domains of the RDRPs and the DDRPs has implications for the origin of the former group of proteins. Given that DDRP is universal, whereas RDRP is eukaryote-specific, a plausible hypothesis might be that RDRP evolved at the onset of eukaryotic evolution through extensive, rapid sequence divergence following a duplication of the β' subunit of the ancestral DDRP. However, despite extensive comparisons using PSSM and careful visual examination, we failed to detect any residual sequence similarity between DDRP and RDRP, outside the core DPBB domain containing the metal-binding active site. In particular, no traces of the Zn-ribbon and the AT-hook-like modules, which are conserved in all β' subunits (Fig. 6), were detected. All RDRP sequences contain two closely spaced lysines, which are conserved in the YRH proteins (Fig. 1) and might functionally correspond to the conserved lysines of the β-type DPBB. These residues are embedded in a secondary structure context that is enriched in predicted β-strands (Fig. 1); hence, in principle, this region might correspond to the β-type DPBB of the DDRPs. However, the respective region of the RDRP sequences shows no detectable similarity to the to this DPBB domain, suggesting that the presence of two functionally important lysines in both RDRP and DDRP might be convergent. In principle, radical divergence could have eroded all similarity between RDRP and DDRP beyond the DPBB domain. An alternative, perhaps more likely explanation is that, while the core catalytic domains of the two polymerases evolved from the same ancestral DPBB, they subsequently accreted distinct sets of peripheral modules around this core. The bacteriophage YRH proteins show no specific relationship with the RDRPs from any one eukaryotic lineage, arguing against a recent lateral transfer from eukaryotes to the phages. Taken together, all these observations point to a scenario for the origin of the RDRP, under which the RDRPs and the YRHs diverged from the DDRPs at a stage of evolution that succeeded the divergence of the DPBB domains of the β and β' subunits (Fig. 5) but preceded the emergence of eukaryotes and even LUCA. One class of substrates of the extant eukaryotic RDRPs are small RNAs, such as microRNAs and stRNAs [84, 85, 86, 87, 88, 89, 90], and analogous small RNAs have been identified in bacteria [91, 92, 93, 94]. Given that small RNAs are likely to have been abundant in the hypothetical ancient RNA world, RDRPs might have been involved in the replication of such RNAs since an early stage of evolution. Should that be the case, the YRH proteins could be late surviving derivatives of these ancient enzymes, which apparently have been shunned from cellular life forms during the evolution of prokaryotes. Eukaryotes might have acquired the YRH gene from a bacteriophage at an early stage of their evolution and retained the RDRPs thanks to the selective advantage conferred by the new level of gene expression regulation mediated, in part, by these enzymes.
The eukaryotic RDRP is part of a large ensemble of functionally linked proteins, which include the Dicer helicase-nuclease, PIWI family proteins, the Hen-2/Corymbosa-like RNA methylase, AlkB-related RNA demethylases, and Lin28-like RNA-binding proteins [1, 45, 46]. Since no closely related homologs of any of these proteins are encoded in the genomes of the phages that encode YRH proteins, the latter are unlikely to function in a biological context similar to that of the eukaryotic RDRPs. Nevertheless, the extended similarity between the YRH and RDRP sequences and the presence of potential nucleic-acid-binding modules, such as a C-terminal Zn-ribbon in the SpβC2 YonO protein, suggest that the YRH proteins also have a RNA polymerase activity. Possible functions of the YRH proteins include a regulatory activity via amplification of antisense RNAs or perhaps a more conventional role as a DNA-dependent RNA polymerase of the respective phages.
General discussion and conclusions
Here, we identified bacteriophage homologs of the eukaryotic RDRP, which is involved in post-transcriptional gene silencing. We also present evidence that RDRP and the ubiquitous β' subunit of DDRP contain a homologous, metal-coordinating, catalytic domain. This domain adopts the DPBB fold, in which the signature DbDGD motif, shared by RDRP and DDRP, is ensconced in the insert between the ultimate and penultimate β-strands. A second DPBB domain is present in the β subunit of the DDRPs. This version of the DPBB domain lacks the metal-coordinating motif but contributes two lysine residues, which are critical for substrate interactions in the catalytic cleft. The DPBB domain in the β subunit is distorted by large inserts and shares no detectable sequence similarity with the DPBB from the β' subunit. However, the DPBBs from the two DDRP subunits are spatially juxtaposed in the DDRP crystal, with the catalytic cleft located between them. This arrangement suggests that the primordial RNA polymerase was a head to tail homodimer of a DPBB domain; a duplication of this primordial DPBB domain probably gave rise to the two DPBBs of modern DDRPs. The presence of completely different conserved residues and the asymmetric head to tail arrangement of the monomers in the pair suggest that the ancestral DPBB dimer was not specialized, had no catalytic activity, and merely bound RNA. Originally, these DPBB domains might have functioned as protein cofactors that stabilized a ribozyme RNA polymerase and eventually displaced the ribozyme as they acquired key residues required for protein-based polymerase activity. This evolutionary scenario complements a similar model proposed for the evolution of the other major polymerase class, the palm-domain-containing RNA and DNA polymerases, whose primordial core apparently consisted of a RNA-binding domain of the RRM fold .
The β and β' subunits of DDRP additionally contain a Zn-ribbon and a SBHM domain respectively. The former is a widespread nucleic-acid-binding domain, whereas the latter functions in a variety of biochemical contexts as a small-molecule-binding and protein-protein interaction domain. Analysis of the conserved protein superfamilies within these folds and their phyletic distribution patterns suggests that diverse versions of these domains with distinct functions were already present in LUCA. Thus, several duplication-divergence events apparently preceded the emergence of the forms of these domains that are seen in the catalytic core of DDRPs (e.g., Fig. 5). The ancestral versions of these domains would not have the adaptations for specific roles possessed by their descendant forms, such as those in extant DDRP or RDRP, and probably functioned as generic RNA-binding and protein-protein interaction domains, with the specificity conferred by catalytic RNA molecules. Hence, considerable diversity of protein domains appears to have emerged prior to the "crystallization"  of a transcription machinery similar to the one that operates in modern cells and probably had been already in place in LUCA . Similar conclusions regarding the early stages of protein domain evolution have been reached previously as a result of evolutionary analysis of proteins involved in translation, such as aminoacyl-tRNA synthetases and GTPases: substantial diversification of protein domains had occurred prior to the "crystallization" of elaborate, modern-type translation machinery [66, 97].
Although the major events in the evolution of polymerases occurred at very early stages of evolution and LUCA already had an advanced DDRP resembling the extant forms, subsequent lineage-specific elaborations of considerable magnitude have occurred. The most notable of these is the proliferation and dissemination of SBHM and the BBM domains in the bacterial β and β' subunits. In the archaeo-eukaryotic lineage, another domain, apparently derived through duplication of the dimerization domain of the DDRP α-subunit, was inserted into the C-terminal region of the β' subunit. Additionally, the Zn-ribbon underwent duplication and elaboration in the β' subunit and was also recruited in various transcription factors. Similarly in the archaeo-eukaryotic β subunit, a minimal β-grasp domain was added to the conserved core.
While we provide evidence that the catalytic cores of the RDRPs, their bacteriophage homologs and the DDRPs evolved from a common ancestor, the provenance of the rest of the conserved RDRP module and, accordingly, the evolutionary scenario for RDRP remain less clear. No similarity to DDRPs was detected in the RDRP sequences beyond the core DPBB domain. One possibility is that the RDRP module has emerged at the onset of the evolution of eukaryotes through extreme sequence divergence following a duplication of the DDRP β' subunit. An alternative scenario holds that the RDRPs are ancient enzymes that diverged from the evolutionary precursor of DDRP at a very early, pre-LUCA stage of evolution, albeit after the duplication that led to the differentiation of the β and β' versions of the DPBB domain. In fact, it appears likely that the ancestral protein RNA polymerase, at the time of the differentiation of the β-type and β'-type DPBB domains, functioned as an RDRP because DNA probably had not evolved by that stage . The switch of the ancestral enzyme to the DDRP activity and the emergence of the evolutionary precursor of RDRP might be tentatively linked to one another and associated with the advent of DNA. This view of the evolutionary history of RDRPs is compatible with the presence of RDRP homologs in phages and with their role in replicating microRNAs (potential relics of the RNA world) during PTGS in eukaryotes. Subsequently, this enzyme apparently has been expunged from the cellular RNA synthesis systems and survived only in some parasitic elements, such as bacteriophages, through which it might have been reintroduced into the genome of an ancestral eukaryote. This scenario shows parallels to the probable evolutionary history of another major, unrelated class of polymerases, the RDRPs and reverse transcriptases containing the palm domain. Among extant biological entities, these polymerases are encoded largely by RNA viruses and retroid viruses, but a reverse transcriptase had been recruited by eukaryotes as the catalytic subunit of the telomerase [98, 99]. There seems to be a distinct possibility that both classes of RDRPs are vestiges of the ancient RNA world that have been largely displaced from cellular biochemical machinery at an early stage of evolution, because of the deleterious effects of RNA replication in DNA-based organisms, while surviving in selfish genetic elements.
Material and Methods
The non-redundant (NR) database of protein sequences (National Center for Biotechnology Information, NIH, Bethesda) was searched using the BLASTP program . Position-specific scoring matrix (PSSM) searches were conducted using the PSI-BLAST program, typically with a PSSM inclusion expectation (E) value threshold of 0.01, and were iterated until convergence [100, 101]. Prior to PSI-BLAST searches, query sequences were evaluated for compositional bias using the SEG program . If no such bias was detected, searches were run with the composition-based statistics turned off, in order to maximize sensitivity . Multiple alignments of protein sequence were constructed using the T_Coffee program , followed by manual correction based on the PSI-BLAST results. Identification and statistical evaluation of conserved motifs in multiple protein sequences were performed using the Gibbs sampling method as implemented in the MACAW program [105, 106]. Pattern searches were conducted using the GREF program of the SEALS package , and pattern-initiated BLAST searches were carried out using the PHI-BLAST program ?. Protein structure databases were searched for similar structures using the DALI program . Protein secondary structure was predicted using the PHD program implemented on the PredictProtein server with a multiple alignment submitted as the [109, 110]. Protein structures were visualized and manipulated using the Swiss-PDB viewer program  and the ribbon diagrams were made using MOLSCRIPT .
- 2.Nelson DL, Cox MM: InLehninger Principles of Biochemistry. Worth Publishers Inc 2000.Google Scholar
- 63.Mizuguchi K, Dhanaraj V, Blundell TL, Murzin AG: N-ethylmaleimide-sensitive fusion protein (NSF) and CDC48 confirmed as members of the double-psi beta-barrel aspartate decarboxylase/formate dehydrogenase family. Structure Fold Des 1999, 7: R215–216. 10.1016/S0969-2126(99)80169-5PubMedCrossRefGoogle Scholar
- 64.Coles M, Diercks T, Liermann J, Groger A, Rockel B, Baumeister W, Koretke KK, Lupas A, Peters J, Kessler H: The solution structure of VAT-N reveals a 'missing link' in the evolution of complex enzymes from a simple betaalphabetabeta element. Curr Biol 1999, 9: 1158–1168. 10.1016/S0960-9822(00)80017-2PubMedCrossRefGoogle Scholar
- 103.Schaffer AA, Aravind L, Madden TL, Shavirin S, Spouge JL, Wolf YI, Koonin EV, Altschul SF: Improving the accuracy of PSI-BLAST protein database searches with composition-based statistics and other refinements. Nucleic Acids Res 2001, 29: 2994–3005. 10.1093/nar/29.14.2994PubMedCentralPubMedCrossRefGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.