The amino-acid sequences of E. coli RsmC and RsmD were used in PSI-BLAST database searches to identify orthologous proteins (hits reported in the 1^st iteration with e-values <10^-40 and sequence identity > 40%; see Materials and Methods for details). The resulting multiple sequence alignment (Figure 1) was used to predict the secondary structure (separately for both protein families) and to precisely identify boundaries of the C-terminal catalytic domain. The Rsm sequences were divided into the N-terminal (variable) and C-terminal (conserved) domains (hereafter dubbed NTD and CTD) of approximately equal length. The sequences of NTD and CTD of RsmC and RsmD were used in additional PSI-BLAST searches and also submitted to the structure prediction MetaServer http://bioinfo.pl. The CTD exhibited similarity to a large class of genuine and predicted N-methyltransferases. Both the RsmC and RsmD CTD sequences showed highest similarity to each other (e-values in the range of 10^-20–10^-19 in the 1^st iteration), but also to a family of putative MTases from Gram-positive Bacteria and Archaea, typified by the Mj0882 protein from M. jannaschii (e-value 10^-19 in the 2^nd iteration starting with RsmC, 10^-35 in the 3^rd iteration starting with RsmD). Interestingly, the crystal structure of Mj0882 has been solved as a part of a structural genomics program (1dus; deposited in the Protein Data Bank in January 2000, cited as: LW Hung, L Huang, R Kim, SH Kim: Crystal structure and functional analysis of a hypothetical protein, Mj0882, from Methanococcus jannaschii, to be published). However, neither the biochemical characterization of this protein or the analysis of its structure has been published and its function remains a mystery. More distant homologs with shorter alignments reported in the CTD-initiated search included the HemK/YfcB family of protenr:N⁵-glutamine MTases 1415 and various distinct RNA and DNA MTase families (e-values > 10^-16 after the RsmD-initiated search converged in the 4^th iteration). Only Mj0882 and its close homologs could be aligned to the CTD of RsmC or RsmD over their entire length if the default PSI-BLAST parameters were used. Reciprocal database searches initiated with Mj0882 yielded its presumed orthologs from Archaea and Gram-positive Bacteria (detected in the 1^st PSI-BLAST iteration with e-values < 10^-30 and sequence identity to the query > 40%) and CTD sequences of proteobacterial m²G MTases (RsmC detected in the 2^nd iteration, e-values ≥ 2*10^-26 and sequence identities ≤ 26%; RsmD detected in the 3^rd iteration with e values ≥ 2*10^-27 and sequence identities ≤ 24%). This result of reciprocal database searches strongly suggested an orthologous relationship 16 between the RsmC/RsmD and Mj0882 families. Further support for this prediction was obtained from the threading analysis. Among structurally characterized MTases, Mj0882 was reported as the definitely best structural template for the CTD of RsmC and RsmD (confident Pcons scores 5.78 and 6.09; see Table 1 and Table 2 for the summary of the threading analysis).

In contrast, querying PSI-BLAST with the NTD sequences did not reveal any significant similarities. The NTDs could only be aligned to each other, displaying a number of conserved residues, if the full-length RsmC or RsmD sequences were used in the PSI-BLAST search. This suggested that the NTD could represent a novel type of a variable domain fused to the common catalytic domain or a strongly diverged version of some known nucleic acid binding domain. Quite unexpectedly, all threading algorithms (see Methods) reported that the sequences of the NTD of RsmC and RsmD are compatible with the MTase fold. Moreover, the majority of servers reported the Mj0882 structure as the best template with significant scores (Table 3 and Table 4). In the rankings produced by the consensus server Pcons, the top 6 hits corresponded to the Mj0882 structure with the best score of 3.615 (RsmC NTD) and 3.805 (RsmD NTD). These results strongly suggested that the NTD and the CTD in bacterial m²G MTases are related to Mj0882 and to each other. Nevertheless, we found that the NTD lacked many residues that may be important for the enzymatic activity of Mj0882 and the CTD (see below). Hence, the NTDs could be catalytically inactive versions of the MTase domain. Therefore, the sequences of NTD were not included in subsequent studies of the m²G MTase AdoMet-binding site and the active site based on analysis of conserved residues.

The YbiN family of predicted N-MTases has also been subjected to sequence comparisons and threading analyses. These results confirmed the previously observed similarity to sequences of RsmC and RsmD CTD 11 as well as to the Mj0882 family. PSI-BLAST reported similarity of YbiN to RsmC CTD in the 3^rd iteration with the e-value of 8 * 10^-5, whereas Pcons ranking of the top 4 hits corresponded to the Mj0882 structure with the best score of 3.41 (Table 5). In the PSI-BLAST searches and in sequence profile comparisons using the FFAS algorithm 17, the YbiN family showed similarity to members of the HemK/YfcB family with scores comparable to that of the RsmC/RsmD/Mj0882 family (data not shown). It has been recently shown that the HemK family comprises protein:N⁵-glutamine MTases, suggesting that the (D/N/S)-P-P-(F/Y/W/H) motif is characteristic for a subset of MTases that methylate -NH₂ groups of various substrates, rather than only bases in nucleic acids 1415.

Although the predicted key catalytic motif of the YbiN family (NPPFH) resembles more that of bona fide guanine-N2 MTases RsmC and RsmD than of any other N-MTase family, it cannot be excluded that these proteins methylate other substrates than guanine in RNA. We hope that this analysis will facilitate the choice of representative targets for experimental characterization, which would allow more precise inference of biological function (target specificity) and relationships between various families of N-MTases.

Another piece of information that should shed light onto the broad picture of the N-MTase family is the previously unnoticed high similarity of the YbiN family to uncharacterized proteins from Eukaryota and Archaea (BLAST e-values 10^-27, and 4*^-04, respectively). Their close similarity (confirmed in reciprocal searches with similar e-values) indicates an orthologous relationship to bacterial YbiN proteins. This was confirmed in reciprocal searches initiated with the eukaryotic and archaeal members of the family. Interestingly, while the predicted eukaryotic YbiN orthologs possess all sequence features typical for the bacterial MTases, their archaeal counterparts lack the key Asn residue in motif IV, suggesting that they may be inactive, unless the catalytic side chain "migrated" to another position in the primary sequence (see below). Regardless of the possible loss of functional residues, the presence of common motifs in the YbiN and Mj0882 families as well as in the NTD and CTD sequences of proteobacterial m²G MTases was confirmed using the Gibbs sampling procedure 18. Motifs I, VI and VIII were detected with probability of occurring by chance < 10^-13. In addition, there is a good correlation of secondary structure elements independently predicted for all protein groups (Figure 1).

The orthologous/paralogous relationships inferred from the results of PSI-BLAST searches were confirmed by phylogenetic analysis using the neighbor-joining method of Saitou and Nei 19. The bacterial YbiN sequences and their archaeal and eukaryotic homologs grouped together with 100% bootstrap support and the RsmC/RsmD NTD, RsmC/RsmD CTD and Mj0882 families formed groups with bootstrap support > 50% (data not shown). Taken together, these findings strongly suggest that the two domains of the proteobacterial m²G MTases RsmC and RsmD, and the predicted MTases YbiN and Mj0882 evolved from the common ancestor. An alternative scenario can be envisaged, in which the pseudodimeric structure of RsmC and RsmD results not from duplication but a fusion of the ancestral CTD with another distantly related MTase that has now degenerated. Nevertheless, fold-recognition results suggest that among various MTases of known structure, the NTD clearly shows highest similarity to Mj0882 (the archaeal counterpart of the CTD) (Table 3 and Table 4). In addition, comparison of the sequence profile of the NTD sequences with profiles of other MTase families (unpublished data) using the FFAS algorithm 17 confirmed that the Mj0882/CTD family is the best match (data not shown), as reported in the course of fold recognition searches against the profiles comprising proteins from PDB. Moreover, the scenario involving intragenic duplication of the catalytic domain in the ancestor of the RsmC/RsmD lineage, which generated a functional pseudodimer, is supported by the dimeric structure of Mj0882 in the crystal (see below). Determination of the crystal structure of a member of the RsmC/RsmD lineage will provide insight into the relationship of the inactivated NTD to known MTase structures and the mutual orientation of the two domains.

The Mj0882 structure was solved in the absence of any ligands. Prompted by the identification of close relationships between the family of m²G MTases and Mj0882, we sought to identify the cofactor- and guanine-binding sites in the 1.8 A crystal structure of the latter protein based on comparison with structures of known N-MTases. Searching the Protein Data Bank 20 using VAST 21http://www.ncbi.nlm.nih.gov/cgi-bin/structure/mmdbsrv?UID=1DUS and DALI 22http://jura.ebi.ac.uk:8765/holm/qz?filename=/data/research/fssp//1dusA.fssp revealed that the Mj0882 structure is more similar to various MTases alkylating hydroxyl groups, than to other nucleic acid N-MTases. This is not surprising, since previous evolutionary analysis of MTase structures and sequences demonstrated that some nucleic acid N-MTase families do not group together in the superfamily tree. The indication is that the common motif (N/D/S)-(I/P)-P-(F/Y/W/H) evolved independently several times on the common framework 23. It will be interesting to determine, if m²G MTases share a relatively recent common ancestor with any of the m⁶A or m⁴C MTase families and if the similarities in their active site are synapomorphic or homoplastic.

The issue of divergent or convergent evolution notwithstanding, the "catalytic loop", corresponding to motif IV, assumes a strikingly similar conformation in all N-MTase structures solved to date (Figure 2). This suggests that the interactions between the target amino group and the side chain of the first residue and the carbonyl oxygen of the second residue of the tetrapeptide are highly similar. The same conformation is retained in the Mj0882 structure, consistent with the prediction that this protein belongs to the N-MTase family. Nevertheless, in the superimposed structures there is a substantial variation in the conformation of the cofactor AdoMet or its analogs, and in their relative orientation with respect to the catalytic site (Figure 2). This variation was noted previously in the course of crystallographic analyses, and was attributed to subtle structural differences between cofactor-binding pockets of various MTases that impose distinct conformation of the same ligand 1224 and to the fact that different ligands (AdoMet, AdoHcy, and their analogs) make different interactions and do not necessarily retain the common conformation when bound to the same MTase 2425.

In order to dock the AdoMet molecule to the 1dus coordinates, we sought to identify those MTase structures, which display greatest similarity to the Mj0882 structure in the cofactor-binding region. VAST and DALI searches confirmed that the region spanning residues 55 to 117 in the Mj0882 sequence is the most similar to MTases that modify hydroxyl groups (data not shown), as well as the entire protein structure (see above). The catechol O-MTase structure (COMT) 26(1vid) was reported as the best hit by VAST (Score 7.1, RMSD 1.3). It was used as a structural template, because it was the only well-scored structure of a MTase complexed with AdoMet and not with its non-reactive analogs or the reaction product AdoHcy. Accordingly, the AdoMet moiety was copied from COMT to the Mj0882 structure based on superposition of the 1vid and 1dus coordinates. The obtained Mj0882-AdoMet complex showed no severe atomic overlaps and the cofactor seemed to fit the groove on the protein surface very well. According to AutoDock 27 the energy of the interactions between the 1dus structure and AdoMet in the docked complex is very favorable (-15.03 kcal/mol) even though it is lower from that calculated for the template COMT-AdoMet complex (-22.04 kcal/mol) in the 1vid structure 26. From the docked model, in striking analogy to most MTase structures (reviewed in ref. 12), the following three crucial contacts can be predicted: i) D⁶¹ from motif I coordinates the methionine amino group of AdoMet via an ordered water molecule. Even though the corresponding acidic residue is conserved in nearly all MTases analyzed to date, this contact has been identified only recently in the high-resolution structure of the RrmJ MTase 28. ii) D⁸⁴ from motif II coordinates the ribose hydroxyl groups, iii) D¹¹³ from motif III coordinates the amino group of the adenine moiety. Non-polar interactions between the side-chains of I⁸⁵ and L¹¹⁴ and the adenine ring further contribute to the binding (Figure 3).

Docking of the target guanine to the Mj0882 structure (Figure 4) was guided by superposition of the "motif IV loop" of Mj0882 and M.TaqI, the only N-MTase co-crystallized with the nucleic acid substrate 29. Under the assumption that both enzymes bind their targets in the same plane, the N2 group of guanine could be aligned with the N6 group of the adenine amino only if the purine ring was rotated by 120 degrees (i.e. with the atoms C2 and N3 of guanine superimposed onto atoms C6 and N1 of adenine). According to AutoDock 27 the energy of the interactions between the Mj0882-AdoMet complex and the docked guanine is quite favorable and comparable with that calculated for the target adenine in the M.TaqI structure (1g38) (-6.14 kcal/mol and -7.59 kcal/mol, respectively). These values are much lower than the energy of interactions between the protein and the cofactor (see above), because the AdoMet-binding groove is very deep and hydrophobic, while the base-binding site is relatively shallow. An alternative orientation of the target guanine in complex with 1dus could be obtained if its atoms C2 and N1 were superimposed onto atoms C6 and N1 of adenine. However, the latter model resulted in steric clashes between the ribose moiety and I¹³² (data not shown). Even if the flexible ligand docking option of AutoDock was used, which alleviated steric clashes, energies for the guanine in the "alternative" position (a cluster of similar conformations) were significantly worse (data not shown). In the first docked model the guanine moiety fits the groove on the protein's surface surprisingly well; a residue was identified, whose side chain might coordinate the ribose hydroxyl groups, namely D⁴¹ (Figure 3). Analysis of the multiple sequence alignment (Figure 1) revealed that this residue is highly conserved in the Mj0882, RsmD, RsmC, and YbiN families, supporting its functional importance. In the "alternative" docked model, D⁴¹ is too remote from the target base to make direct contacts to the target nucleoside (data not shown). However, a substantial conformational change upon substrate binding cannot be excluded, and direct participation of D⁴¹ in catalysis as well as other modes of guanine binding can be envisaged. Identification of this conserved carboxylate as a potentially important residue will hopefully prompt site-directed mutagenesis experiments and experimental determination of its role in binding and/or catalysis.

Although the presented docking model of Mj0882 complexed with AdoMet and guanine should be regarded as preliminary, it provides a useful platform for correlating the structure with sequence conservation in the Mj0882/RsmD/RsmC/YbiN family. The predicted AdoMet- and ribose-binding Asp residues (see above), as well as other invariant and highly conserved surface-exposed residues, map in the vicinity of the docked ligands, strongly supporting the model (Figure 5). The N2 group of the docked guanine is hydrogen bonded to the side chain of N¹²⁹ and the carbonyl oxygen of P¹³⁰. This is similar to the contacts made by the N6 group of the target adenine in the co-crystal structure of M.TaqI 29, which served as a docking template. As proposed for M.TaqI, these interactions could increase electron density of N2, change its hybridization from sp² towards sp³, where the lone pair of N6 is no longer conjugated with the aromatic system, and thus contribute to its activation for nucleophilic attack on the activated methyl group of AdoMet. From the Mj0882 structure it is not obvious which residue may be responsible for eliminating the proton from the methylated N2 amine. The only conserved acidic residue in the vicinity of the active site corresponds to D⁴¹, and its localization is incompatible with the localization of the protons on N2 in the above-mentioned mechanism. If no massive conformational change of the m²G MTase occurs during RNA binding and catalysis, we propose that the proton is transferred directly to a bulk solvent molecule. In M.TaqI, the target base forms a face-to-face π-stacking interaction with Y¹⁰⁸ from the NPPY tetrapeptide and a hydrophobic interaction with V²¹ from motif X. In Mj0882 these contacts are reversed (i.e. in motif IV Y¹⁰⁸ is substituted by non-aromatic I and V²¹ is replaced by aromatic F³⁵). Interestingly, all members of families RsmC, RsmD, and YbiN possess aromatic residues at both positions (Figure 1). It has been hypothesized that the methyltransfer reaction in N-MTases proceeds via a positively charged intermediate, which could be stabilized by cation-Ti interactions 30. In m²G MTases, the electron-rich aromatic rings on both sides of the target could contribute to such stabilization.

The predicted guanine-binding mode implies that the target nucleotide is "flipped out" of the rRNA into the catalytic site, similarly to the mechanism discovered for DNA: m⁵C MTases 31 and extended to DNA N-MTases 2932. In bacterial 16S rRNA G966 and G1516 occur in extrahelical regions, which should facilitate their binding in the catalytic pocket of the enzyme, but G1207 is base-paired in the helix 34 stem 7. The m²G1207 MTase RsmC reacts with the 30S ribosomal particles but barely at all with 16S RNA, while the m²G966 MTase RsmD requires only the S7 and S19 proteins to methylate its target in the 16S RNA 10. It has been inferred that both these nucleotides are modified after the association of several ribosomal proteins with the 16S rRNA, which is likely to assume the native secondary structure, but before assembly of the 30S subunit is complete 10. The positively charged surface of Mj0882 in the area next to the docked guanine suggests charge complementarity to the negatively charged phosphate groups in rRNA (Figure 6). It will be interesting to study the base-flipping capabilities of RsmC and RsmD, since they may shed the light on the evolution of presumed mechanistic differences between the nucleic acid MTases that flip their target out of the double helix and those modifying extrahelical bases.

Analysis of the multiple sequence alignment (Figure 1) suggested that the NTD of RsmC and RsmD lack several conserved residues. However, only in connection with the docking model of Mj0882 it can be concluded, which of these residues may be involved in binding or catalysis. Specifically, all NTD sequences lack the guanine N2-coordinating Asn and the aromatic or aliphatic residue in motif IV. The NTD of all RsmC, with the exception of T. thermophilus, lacks the N-terminal extension, which includes the aromatic residue predicted to make the face-to-face π-stacking interactions with the target guanine (F³⁵ in Mj0882). This conserved aromatic residue is also missing from the longer N-terminus of RsmD and T. thermophilus RsmC. The NTD of RsmD retained the predicted ribose-binding acidic residue at the position corresponding to D⁴¹ in Mj0882, while in the NTD of RsmC this residue is often substituted, in most cases with Ser or Thr. All NTD sequences, with the possible exception of C. crescentus and T. thermophilus RsmC, lack the AdoMet-coordinating acidic residue in motif I, corresponding to D⁶¹ of Mj0882. The NTD of RsmD retained the Asp residue corresponding to D⁸⁴ in Mj0882, however in nearly all RsmC sequences this residue is substituted with various polar amino acids. A region corresponding to motif III was so diverged in the NTD of both RsmC and RsmD that it could not be aligned with confidence and the position corresponding to D¹¹³ in Mj0882 could not be identified unambiguously. It is noteworthy that the residue corresponding to D¹¹³ (motif III) is weakly conserved in the MTase superfamily, but the acidic residue corresponding to D⁸⁴ (motif II) is conserved in nearly all MTase families studied to date 12(JMB, unpublished data).

In summary, the NTD of RsmC and RsmD lacks all residues, which were predicted to be a part of the active site based on the docking model and which are conserved between Mj0882 and the CTD. The NTD of RsmC also lacks most of the acidic residues predicted to coordinate AdoMet and the ribose moiety of the target nucleoside. Nevertheless, some of these residues are retained in the NTD of RsmD and in the NTD of T. thermophilus RsmC. Hence, it is inferred that the NTD of RsmC and RsmD lost the m²G MTase activity and are probably also deficient in the ability to bind the cofactor. There are known precedents of proteins with apparent severe defects in conserved MTase motifs. One example includes the yeast Kar4p protein required for expression of karyogamy-specific genes during mating 33. Using fold recognition and homology modeling it was demonstrated that Kar4p is related to mRNA:m⁶A MTases and the "permuted" DNA:m⁶ and m⁴C MTases, although it lost its active site and has the AdoMet-binding site blocked by an insertion. It is quite probable that Kar4p exerts its function due to the retained ability to bind to RNA or DNA (JM Bujnicki and coworkers, manuscript submitted). Another example is the eukaryotic heterodimeric tRNA:m^lA58 MTase, in which one subunit binds AdoMet and presumably carries out the catalysis, while the other is indispensable for tRNA binding 2. It has been demonstrated that the tRNA-binding subunit is homologous to the "active" subunit, but lost the cofactor-binding and catalytic side-chains. We have also identified a single "active" tRNA:m^lA58 MTase gene in Bacteria and Archaea, suggesting that the ancestral prokaryotic form was a homodimer 34. This prediction has been recently confirmed by the crystal structure of the Rv2118c protein from Mycobacterium tuberculosis, which forms a dimer of tight homodimers 35.

Interestingly, according to the EBI Macromolecular Structure Database, the suggested biologically relevant molecule for the Mj0882 structure is a homodimer http://pdb-browsers.ebi.ac.uk/pdb-bin/macmol.pl?filename=1dus. The Mj0882 homodimer exhibits 5 interchain salt bridges within the complex and shows a loss of 2045.5 Ang**2 of solvent accessible surface area upon complex formation. However, in each subunit the active site is completely blocked by a loop from the other subunit. Nevertheless, it is tempting to speculate that the active form of Mj0882 is a homodimer that undergoes a conformational change to allow the target 16S rRNA to bind and that RsmC and RsmD MTases function as pseudodimers. If this is true, the relationship between Mj0882 and the RsmC/RsmD families would be similar to the relationships between prokaryotic and eukaryotic tRNA:m^lA58 MTases.

We also found that the archaeal members of the YbiN family lack the counterpart of the conserved N¹²⁹ in Mj0882. However, analysis of the multiple sequence alignment revealed that they retained all other predicted binding residues, with the possible exception of the poorly conserved D¹¹³ (Figure 1). This prompted us to build a provisional homology model of the PH0266 protein, which revealed that the "missing" Asp sidechain in motif IV is spatially replaced by a His residue from motif X (data not shown). The alignment (Figure 1) reveals that the His residue is conserved at this position in all archaeal YbiN homologs, whereas bacterial and eukaryotic YbiN proteins have Pro or Ala and the Mj0882/RsmC/RsmD lineage have Gly or Ala (G⁴³ in Mj0882). "Migration" of catalytic and cofactor binding residues have been documented in many cases (for a review see ref. 36), including RNA MTases that generate m⁵C in rRNA 37 and m⁷G in mRNA 38. Nonetheless, no active nucleic acid MTase have been reported bearing the His substitution for the catalytic residue in motif IV. We hope that this analysis will stimulate experimental characterization of archaeal YbiN homologs, which may eventually lead to validation of a novel version of the methyltransfer active site.

The position-specific, iterative (PSI-) version of BLAST 4142 was used to search the non-redundant (nr) version of current sequence database and the publicly available complete and incomplete genome sequences available via the websites of the National Center for Biotechnology Information (Bethesda, USA; http://www.ncbi.nlm.nih.gov and the Bioinformatics Laboratory of International Institute of Molecular and Cell Biology (Warsaw, Poland; http://www.iimcb.gov.pl). All searches were initiated with stringent profile inclusion expectation (e) values (<10^-20) to avoid "explosion" of hits to multiple paralogous MTase families and the cutoff was relaxed in subsequent iterations as the profile become balanced (i.e. included more orthologs of the query, which reduced the probability of assigning a significant score to a non-ortholog that exhibited a fortuitous similarity to the query, but not to the orthologs of the query). The stringent criteria for provisional assignment of orthologous relationship in PSI-BLAST searches were: the alignment spanning more than 80% of both the query and the reported hit, e-value in the 1^st iteration < 10^-30, sequence identity > 40%. In all searches initiated with RsmC, RsmD, and YbiN sequences, hits to the catalytic domain of HemK family members were reported before relaxation of the cutoff with e-values ranging from 10^-5 to 10^-18 (i.e. subsequently to the presumed members of the above-mentioned families). The choice of the relaxed cutoff was dictated by the assumption that the HemK family of protein MTases should be regarded as an outgroup and its members shall not be included in the profile. The multiple sequence alignment was retrieved from the PSI-BLAST, degapped, and used as a profile, to which all the full-length sequences were realigned using CLUSTALX 43.

Protein structure predictions were carried via the MetaServer interface 44http://bioinfo.pl/meta/ using publicly available online services for fold recognition: FFAS 17, 3DPSSM 45, BIOINBGU 46, GenThreader 47, SAM-T99 48, and FUGUE 49, and secondary structure prediction: JPRED2 50 and PSI-PRED 51. The results were processed by the Pcons consensus server 52, which produced a ranking of potentially best target-template alignments and evaluated the likelihood of the models to be correct.

Comparison of atomic coordinates with the protein structures from the Protein Data Bank 20 were carried out using VAST 21 and DALI 53. SWISS-PDB VIEWER 54 was used for all protein structure manipulation, calculation of conservation scores and electrostatic potential distribution, and to generate the figures. Provisional homology modeling was carried out using the SWISS-MODEL/PROMOD II server http://www.expasy.ch/swissmod/SWISS-MODEL.html54.