Module 4.6.1

Modified Nucleosides of Escherichia coli Ribosomal RNA†

JAMES OFENGAND AND MARK DEL CAMPO*

[SECTION EDITOR: GLENN BJÖRK]

Posted December 29, 2004

Department of Biochemistry and Molecular Biology, University of Miami School of Medicine, Miami, FL 33136

*Corresponding author. Phone: (305) 243-6962. E-mail: This e-mail address is being protected from spambots. You need JavaScript enabled to view it

†James Ofengand died on 6 December 2004. This chapter is dedicated to his memory.

INTRODUCTION

The modified nucleosides of RNA are chemically altered versions of the standard A, G, U, and C nucleosides. Sometimes the alterations are baroque, as in tRNA (see chapter by Björk, module 4.6.2), but in ribosomal RNA (rRNA), the modifications are confined to isomerization of U to pseudouridine (Ψ), addition of H across the 5,6 double bond of U to form dihydrouridine (hU), and addition of methyl groups to the purine and pyrimidine rings and to the 2'-hydroxyl of ribose. The various modified nucleosides found in Escherichia coli rRNA are shown in Fig. 1. These residues are found at precisely determined locations and are placed there by specific enzymes for reasons that are for the most part still elusive. So far, all modifications studied have been made at the polynucleotide level, that is, after prior synthesis of the unmodified RNA.

Fig. 1The modified nucleosides found in E. coli rRNA. Ψ, 5−ribosyluracil (pseudouridine); m³Ψ, 3-methylpseudouridine; m³U, 3-methyluridine; m⁵U, 5-methyluridine; hU, 5,6-dihydrouridine; m⁴C, N₄-methylcytidine; m⁵C, 5-methylcytidine; Nm, 2'-OCH₃-N (N = A, G, U, or C); m²A, 2-methyladenosine; m⁶A, N₆-methyladenosine; m⁶₂A, N₆-dimethyladenosine; m¹G, 1-methylguanosine; m²G, N₂-methylguanosine; m⁷G, 7-methylguanosine.

Source:

In this chapter, we will review the nature and location of the modified nucleosides of Escherichia coli rRNA, the enzymes that form them, and their known and/or putative functional role. Other recent surveys that cover various aspects of this subject are Ofengand and Fournier (90), Ofengand and Rudd (92), Charette and Gray (23), Ofengand et al. (91), Ofengand (88), Decatur and Fournier (34), and Ferré-D'Amaré (45). Earlier work has been summarized by Björk (13).

NUMBER AND LOCATION OF MODIFIED NUCLEOSIDES

The modified nucleosides are listed in Table 1 for Ψ and in Table 2 for the methylated nucleosides and hU. References for the Ψ site identification are listed in the legend of Table 1 and for the other modified nucleosides in column 2 of Table 2. These tables also summarize the available information on the Ψ synthases, methyltransferases, their genes, and their crystal structures. The enzymes will be discussed in more detail below. There is one Ψ and 10 methylated residues in the 16S rRNA. All the methylated residues of 16S rRNA have the methyl group on the purine or pyrimidine ring and, in addition, m⁴C1402 has a 2'-O-methyl ribose. Since there are two dimethyladenosine residues, there are, in total, 13 methyl groups in 16S rRNA. This disproportion between Ψ and methyl groups is not found in the 23S rRNA. There are 10 Ψ and 14 methylated residues, 3 of which are 2'-O-methylations. There is also a single hU and a partially modified C2501 whose structure is not yet known (5).

Table 1Pseudouridines in E. coli rRNA and the synthases that make them

Table 2Other modified nucleosides in E. coli rRNA and their enzymes

The locations of these modified residues in the secondary structure of 16S and 23S rRNA are shown in Fig. 2 and 3 and in the three-dimensional form existing in the ribosome in Fig. 4 and 5. It is evident from these figures that modified nucleosides cluster primarily around known functionally important regions of the rRNA. This was first noted for 16S rRNA by Brimacombe et al. (16) and for 23S rRNA by Smith et al. (103) and Brimacombe et al. (16). That Ψ and the methylated residues coclustered in 23S rRNA became evident when the number of Ψ sites known more than doubled (10). Coclustering is not limited to E. coli. The phenomenon occurs in yeast (9) and in human large subunit RNA (89). A more detailed correlation of sites of modification with functionally important regions of the ribosome can be found in Ofengand et al. (91) and Decatur and Fournier (34). In this context it is worth noting that Ψ and methyl groups contribute opposite molecular characteristics. Methylation, except for m⁷G formation, which adds a positive charge, increases local hydrophobicity, whereas Ψ formation introduces an additional hydrophilic H-bond donor from its N₁ position. At first glance, coclustering of these two opposite chemical groups appears counterproductive. However, it may be that the precise positioning of both hydrophobic and hydrophilic patches on rRNA, especially 23S rRNA, provides a specific molecular means of stabilizing ribosome structure in the vicinity of functionally important regions and/or of improving interaction with the ligands of ribosomes such as tRNA, mRNA, and protein synthesis factors. In this regard, it is noteworthy that of the 22 nucleosides of 23S rRNA identified as in contact with tRNA acceptor stem mimics (11), 15 (68%) are within three residues of a modified nucleotide, yet only one, Ψ2604, is itself modified.

Fig. 2Location of the modified nucleosides of 16S rRNA. The secondary structure is modified from the Comparative RNA Web Site (http://www.rna.icmb.utexas.edu/). Modified nucleoside positions are as indicated.

Source:

Fig. 3Location of the modified nucleosides of 23S rRNA. The secondary structure was obtained as in Fig. 2 and is shown with the modified nucleosides as indicated.

Source:

Fig. 4Location of modified nucleosides on the 30S ribosome. (A) View from the interface side of the ribosomal subunit. (B) View in A rotated by 90°. The three-dimensional structure of the 16S RNA of the 30S ribosome of T. thermophilus (112; Protein Data Bank [PDB] entry 1J5E) is shown with all proteins removed and the phosphodiester backbone represented by a continuous line. Ψ residues are shown as 6-Å- diameter red spheres. Methylated nucleosides are shown as 6-Å-diameter green spheres irrespective of whether they are purine or pyrimidine derived or base methylated versus ribose methylated. All spheres are centered on the glycosyl N of the base. The identifying numbers are listed in Tables 1 and 2.

Source:

Fig. 5Location of modified nucleosides on the 50S ribosome. (A) View from the interface side of the ribosomal subunit. (B) View in panel A rotated by 90°. The three-dimensional structure of the 23S RNA of the 50S ribosome of Deinococcus radiodurans (56; PDB entry 1NKW) is shown the same way as in Fig. 4. Ψ and methylated nucleosides are denoted as in Fig. 4. Dihydrouridine and the unidentified modified C2501 are shown as a light blue and yellow 6-Å sphere, respectively. All spheres are centered on the glycosyl N of the base. Residue 2057 (cyan sphere marked by black triangle), 2058 and 2059 (cyan spheres), 2062 (orange sphere), 2609 (orange sphere marked by black triangle), and 2611 (black sphere) are all located at the interface entrance to the exit tunnel (54, 99). These spheres are 3 Å in diameter to conserve space.

Source:

A closer examination of Fig. 5 shows that in E. coli the modifications cluster at the peptidyltransferase area, loosely defined as that part of the 50S that binds the amino acyl ends of tRNAs in the A, P, and E sites. In this location, they would be well positioned to assist in some still undefined way in the binding of the amino acyl stem part of the tRNAs to the 50S. They are not, however, found in the peptide exit tunnel, although several, notably m¹G745, Ψ746, and m⁵U747 (numbers 2, 22, and 23 in Fig. 5), Ψ2580 (number 9), and possibly also m²A2503 (number 36) cluster at the entrance to the tunnel. This is not surprising since hydrophobic and hydrophilic residues at specific sites on the tunnel walls would be expected to interfere with peptide chains bearing particular functional groups.

PSEUDOURIDINE SYNTHASES

Sequence Homology and Classification into Related Subgroups

There are seven Ψ synthases to make the 11 Ψ in rRNA (Table 1). There is disparity in numbers because RluC and RluD each make 3 Ψ. The converse, more than one synthase making the same Ψ, does not occur. This was shown by deleting the genes for each of the synthases in turn and observing a corresponding absence of specific Ψ in vivo. Moreover, all the Ψ sites could be accounted for by the set of deletions. If Ψ were made by two synthases, deletion of only one synthase gene would have had no effect. The genes have been identified as indicated in Table 1, and the Salmonella enterica serovar Typhimurium homologs are noted there also. E. coli has, in addition, four tRNA-specific Ψ synthases. The 11 tRNA and rRNA Ψ synthases of E. coli have little overall amino acid sequence homology but they do possess several conserved sequence motifs. Based on these motifs, 10 of the 11 Ψ synthases were classified into four related families, named for the first member of each class identified (51, 69, 92). All the rRNA Ψ synthases are to be found in only two of these families (Table 1). The other two families consist of a single member each, namely TruA and TruB, which make Ψ in tRNA. Another Ψ synthase for tRNA, TruC, is a member of the RluA family (37). The 11th Ψ synthase, TruD, is specific for tRNA and is not related in amino acid sequence to any of the other synthases. It makes up its own fifth family (65, 66). Although the original member of the RsuA family makes the Ψ in 16S rRNA, this is not a defining characteristic of this family. RluB, RluE, and RluF, all of which make Ψ in 23S rRNA, also belong to the RsuA family. The situation is similar for the RluA family. RluA, RluC, and RluD all make Ψ in 23S rRNA, whereas TruC makes Ψ in tRNA.

The initial sequence analysis referred to above indicated that the TruA family was much less related to the other three families of Ψ synthases and that the TruD family was seemingly unrelated. However, more recent alignments based on superpositioning of the known crystal structures of TruA, TruB, RsuA, RluD, and TruD have shown that all five Ψ synthase families are in fact structurally related and contain five conserved motifs (38, 66). The alignment of the seven E. coli rRNA Ψ synthases is shown in Fig. 6. First, note that some of the rRNA Ψ synthases consist of two domains. The N-terminal domain is similar to the RNA-binding domain of ribosomal protein S4, a small ~60-amino-acid modular domain that is found in many proteins that interact with RNA (6). Possession of this domain is not a requirement for rRNA Ψ synthases, however, since RluA and RluE lack it. The role of the S4 domain in the activity or specificity of S4-domain-containing Ψ synthases is unknown. The larger C-terminal domain is the Ψ synthase catalytic domain and contains the aforementioned five sequence motifs. Five conserved residues (highlighted in yellow) from four of the motifs are found at the active site in Ψ synthase structures (38; discussed further below). In addition, six other residues (highlighted in blue) from three of the motifs are identical.

Fig. 6Multiple sequence alignment of E. coli RluA and RsuA family rRNA Ψ synthases. Five of the synthases have an N-terminal domain similar to the RNA-binding domain of ribosomal protein S4 (grey bar). The five motifs described by Del Campo et al. (38) are indicated with different colored bars matching the colored ribbon segments in Fig. 8. The S4 domain secondary structure (above) is from the crystal structure of RsuA (102; PDB entry 1KSK). The Ψ synthase domain secondary structure (above) is from the crystal structures of RsuA and RluD (38; PDB entry 1QYU) by using those elements consistent with both structures. Conserved residues are shown in bold face with conserved active-site residues highlighted in yellow and other identical residues highlighted in light blue. Numbers to the left of each line denote the sequence position of the first residue shown. Dashes indicate gaps in sequence, and numbers in parentheses indicate the number of residues that are not shown in the alignment.

Source:

Structural Homology

Crystal structures have shown that the Ψ synthase domain is a conserved fold found only in all five families of Ψ synthases (66). The domain consists of a mixed α/β-fold where the length of the domain is made up of an extended, mostly antiparallel β-sheet. In Fig. 7, this conserved domain is illustrated by the overlap of the E. coli RluD (38, 81, 101) and RsuA (102) structures. In this domain, RluD and RsuA overlap in two helices, six strands, and several loops (root-mean-square deviation [RMSD] = 2.8 Å over 145 C_α atoms). The structure of the Ψ synthase domain of RluC, recently described by Mizutani et al. (81), overlaps with RsuA and RluD in a similar manner. However, its overlap with RluD is much more extensive than with RsuA as might be expected from RluA family members. There are also obvious differences between the structures. For instance, RluC and RluD have two inserts that branch off the Ψ synthase domain, termed the Tail domain (38). These differences are likely to contribute to the substrate specificity of each synthase. Note that RsuA, RluC, and RluD have N-terminal S4 domains (Fig. 6) connected to the Ψ synthase domain by a flexible linker. The S4 domain of RluC was absent in the crystal structure (81) and that of RluD was either disordered (38, 81) or absent (101).

Fig. 7Comparison of RluD and RsuA crystal structures. (A) Ribbon cartoon of RsuA (102; PDB entry 1KSK) in blue superposed onto RluD (38; PDB entry 1QYU) in orange using DALI (60). Nonoverlapping domains are labeled. (B) Superposed side chains (starting at the C_?) from the five conserved active-site residues of RluD and RsuA (highlighted in yellow in Fig. 6). The N, C, C_?, C_?, and O atoms of all five residues from RsuA were superposed onto the equivalent five residues of RluD using LSQMAN (68). Colors are the same as in panel A. The figure was generated with PyMOL (36).

Source:

Another characteristic of the Ψ synthase domain is the presence of a positively charged cleft midway through the length of the extended β-sheet and leading to the catalytic center. The catalytic center itself is defined by a small pocket at the base of the cleft and the presence of the five conserved residues from four of the five conserved sequence motifs (Fig. 6). These residues can be superimposed on each other in RluD and RsuA (Fig. 7B) and in the other Ψ synthase structures (38, 66). The Asp from motif II is the presumed catalytic residue (for reasons discussed below), the Arg/Lys from motif III makes a salt bridge with the Asp, the Tyr has been shown to provide stacking interactions with the substrate U in a TruB cocrystal structure with RNA (59, 93, 94), and the hydrophobic Ile/Val from motif III and the Leu from motif IIa probably function to hold the active site in a particular shape. Figure 8 shows the location of the motifs and conserved active site residues in the structure of RluD. The conservation of these motifs and the positioning of conserved residues in the active site strongly imply that all Ψ synthases use the same mechanism for catalysis.

Fig. 8Location of the five Ψ synthase motifs on the structure of RluD. Ribbon cartoon of RluD is light gray. Motifs I, II, IIa, III, and IIIa are the same colors as the bars in Fig. 6. The side chains of the five conserved active-site residues (highlighted yellow in Fig. 6) are shown in ball-and-stick representation and are the same color as the motif to which they belong. The figure was generated with PyMOL (36).

Source: Reprinted with permission from reference 38.

Enzyme Mechanism

The conversion of uridine to Ψ has no precedent in known metabolic reactions. Other enzymes are known to cleave the glycosyl bond but none carry out rotation of the base and rejoining to the ribose while still enzyme bound. The uracil intermediate does not appear to exist free since it does not exchange with added uracil. By analogy with thymidylate synthase, the reaction was originally proposed to occur by covalent adduct formation between a Cys SH residue of the synthase and the C₆ position of the uracil ring. However, substitution of all the Cys residues of TruA with retention of activity ruled out Cys as the attacking species (118). Thus, as an alternative, the γ-COOH of an aspartate that was totally conserved in all Ψ synthase sequences known at the time was proposed as the attacking nucleophile (63). In the RsuA and RluA families, which contain all 7 rRNA Ψ synthases, the conserved Asp is embedded in the conserved sequences GRLD and HRL/ID, respectively (63, 91). (See also Fig. 6, motif II.) Two alternative mechanisms were originally proposed, the one mentioned above where the γ-COOH of Asp makes a nucleophilic attack on C₆ of the uracil ring, and another in which the attack is on the C₁' of the ribose. Product analysis using 5-fluorouracil-substituted substrate appeared to prove the C₆ mechanism (49), but subsequent mechanistic considerations suggested that the C₁' alternative is still viable (82, 104). Structural studies on cocrystals of TruB with a 5-fluorouracil-substituted RNA stem-loop substrate, while verifying that the conserved Asp residue was positioned such that interaction with the substrate U residue could occur, have not served to definitively clarify the point of attack on the substrate uridine since no covalent intermediate could be obtained (59, 93, 94). On balance, however, considering both the biochemical and structural data, the C₆ mechanism is the more likely one. Regardless of the exact mechanism, the conserved Asp identified above has been experimentally shown to be essential in all 11 E. coliΨ synthases (references cited in Table 2 of reference 37) (65), strongly implying that all Ψ synthases use the same catalytic mechanism.

Substrate Recognition

The sites for Ψ in E. coli rRNA run the gamut from being in a single-stranded or loop region (Ψ746, m³Ψ1915, Ψ1917, Ψ2504), adjacent to a double-stranded stem (Ψ516), part of a loop-closing base pair (Ψ955, Ψ1911, Ψ2457, Ψ2580), or part of a base pair in a stem (Ψ2604, Ψ2605). Consequently, it should not be surprising that the way different synthases recognize their intended uridines also varies. Since no structure for an rRNA Ψ synthase in complex with its target RNA has yet been obtained, what follows is only what can be deduced by solution experiments. We will consider each synthase in turn. All the RNA specificity studies described below were performed in vitro.

RsuA (113).

Protein-free full-length (1–1542) or truncated (1–526 or 1–678) 16S RNA transcripts were not substrates, but an RNP particle assembled from the 1–678 transcript and ribosomal proteins was. However, 30S ribosomes assembled in vitro from full-length synthetic transcripts and ribosomal proteins (39) were at best only a poor substrate, while a miniribosome (110) made from the 1–526 fragment and ribosomal proteins was not active at all. These findings indicate either that the synthase requires a defined RNA structure for recognition that is presumably created by the bound proteins or that the synthase interacts at least in part directly with some of the proteins of the active particle. Of course, both of the above might occur together. Presumably, the active RNP substrate mimics a ribosome assembly intermediate that is the true substrate for Ψ516 formation. If that is so, only a small subset of U residues may be available to the synthase. For example, there are only three other U residues in the 30mer comprising the stem-loop where Ψ516 is found. 23S RNA or tRNA^Val transcripts were not substrates for this synthase.

RluA (115).

This synthase can use an in vitro synthesized 23S RNA transcript as substrate, but a fragment consisting of residues 1–847, the Ψ being at position 746, gave a faster rate of reaction than the full-length (1–2904) RNA, and reaction of both RNAs was faster in 1 mM EDTA than in 10 mM Mg²⁺. In all cases, the yield was one Ψ per RNA. These results imply that tertiary, and possibly also secondary, structure is not an important determinant for this enzyme. They also show that neither of the adjacent modified nucleosides, m¹G745 or m⁵U747, are required for Ψ formation. The synthase was unreactive with 16S RNA transcripts whether as 1–526 or 1–678 fragments or as full-length RNA, and unmodified 30S subunits were also inactive. Mature 50S subunits lacking Ψ746 have not been tested. This synthase also makes Ψ32 in tRNA^Phe but not in tRNA^Val transcripts. rluA deletion and plasmid rescue experiments showed clearly that RluA is the only protein capable of making both Ψ746 in 23S RNA and Ψ32 in tRNA (97). This "dual specificity" appears to be due to the sequence and structural similarity between the stem-loop of 23S RNA containing Ψ746 and the anticodon stem-loop of tRNAs with Ψ32. Both stem-loops are similar, although the 23S RNA loop is eight-membered, whereas the tRNA loops are seven-membered; the Ψ in 23S RNA is one residue away from the stem, whereas in the tRNAs it is adjacent to the stem. However, the most important feature is the sequence 3' to the Ψ, UGAAAA, which is the same in both 23S RNA and tRNA^Phe. tRNA^Val, which is unreactive, has a different sequence. The three other tRNAs in E. coli that normally have Ψ32, tRNA^Cys, tRNA^Leu4, and tRNA^Leu5, have an almost identical sequence with only a single base change from G to U or C (RNA^Leu4,5) or a change of the 5'-most A to C (tRNA^Cys). It appears, therefore, that the primary recognition determinant of RluA is the consensus sequence, UUN(A/C)AAA, where the 5'-U is modified to Ψ. It is not yet clear whether the stem-loop configuration is helpful or required or whether the entire sequence need be single stranded. It is noteworthy that while the sequence UUN(A/C)AAA occurs at one other site in E. coli 23S RNA with the putative Ψ site at position 1781, no Ψ is found at this site naturally (10) or after in vitro reaction (115).

RluB (37).

Preliminary in vitro experiments have shown that this synthase can react with free 23S RNA. It is not known whether a 50S ribosome can be a substrate or whether it is preferred. It is even possible that, similar to RsuA, some intermediate ribonucleoprotein particle will turn out to be the true substrate. RluB, along with RluF, is interesting from the standpoint of specificity. In vivo, both synthases recognize adjacent U residues and carry out the same enzymatic reaction, yet are specific for their respective sites. The details of how this specificity is maintained will be most interesting.

RluC.

The in vivo site specificity of this synthase was confirmed by in vitro reaction of overexpressed and affinity-purified RluC with a 23S RNA transcript. The plateau molar yield of Ψ was about 2.5 and all three Ψ, namely, Ψ955, Ψ2504, and Ψ2580, were identified by sequencing (J. Conrad and J. Ofengand, unpublished results). In preliminary experiments, a 41mer containing the stem-loop with U955 was also a substrate (N. Englund and J. Ofengand, unpublished results). Activity with a 50S subunit lacking these three Ψ has not been tested. Therefore, although reaction with free RNA occurs, it is possible that when it is assembled into a ribosome the reactivity is enhanced or, alternatively, that some or all of the Ψ sites in the particle are shielded from reaction. The true in vivo substrate is unknown. How one enzyme recognizes all three sites for Ψ is unclear. The primary sequences bear no relation to each other, the secondary structure of the sites in mature 23S RNA (Fig. 3) shows that the 955 site is far from the other two, and, in the tertiary structure in the ribosome (Fig. 5), all three sites are separated from each other. The only obvious characteristic is that each Ψ is followed by a G. While there are many UG sequences in 23S RNA, none is closer than 11 nucleotides to these three sites. Possibly, the binding of RluC to these three distinct loci is because of a so far unrecognized common structural parameter that allows interaction of only a very limited set of the UG sequences with the catalytic center. Thus, the required specificity would be obtained by a combination of spatial limitation and active-site specificity for a UG sequence.

RluD.

All three of the Ψ lost (Ψ1911, Ψ1915, and Ψ1917) when rluD is deleted (Table 1) can be specifically made in vitro on full-length 23S RNA transcripts when appropriate conditions are used (96, 114). However, at 2 mM Mg²⁺, the synthase loses some specificity and other sites become modified also (114). It is not known if 50S subunits only lacking these three Ψ would be a better substrate or even would be a substrate at all. Their location at the surface of the 50S at the 30S–50S interface (Fig. 5) suggests that they should be accessible. All three sites cluster in a stem-loop, helix 69, which contains no other U residues except for U1923, which is six residues away. On the 5' side, the closest U is 13 residues away from U1911. Therefore, it may be that RluD is specifically bound to the stem-loop structure and converts all U residues within a certain range. Moreover, if RluD has a recognition specificity for UA analogous to the putative UG specificity of RluC, the nearest other UA sequences would be at U1898 and U1926, which are far away. The isomerization of several nearby U residues by one enzyme has a precedent in TruA. It is able to recognize and isomerize U residues at positions 38, 39, and 40 in tRNA. However, TruA makes no more than two Ψ in any one tRNA with a maximum spacing of one residue, whereas with RluD three Ψ are formed and the spacing is 3 between Ψ1911 and Ψ1915. Since the structure of RluD is now known (38, 81, 101), a cocrystal structure of RluD and its substrate RNA may be available in the not too distant future to clarify how RluD recognizes its substrate.

RluE.

This synthase is able to form Ψ2457 in vitro on free 23S RNA (37). Reaction with 23S RNA fragments or with Ψ2457-free 50S subunits has not been tested. The ability of RluE to react with 23S RNA does not preclude that a faster reaction with 50S might occur. The in vivo substrate remains unknown.

RluF.

Limited information is available on the substrate specificity of this synthase. Preliminary experiments have shown that RluF can use both 23S RNA and 50S subunits isolated from rluB- and rluF-deleted cells as substrates with equal efficiency at 1 mM Mg²⁺, and formed Ψ only at 2604. At 12 mM Mg²⁺, the reaction with both substrates was partially inhibited (Y. Kaya and J. Ofengand, unpublished results). 23S RNA fragments have not been tested. The ability of RluF to distinguish between U2604 and U2605 in vitro and in vivo is striking and parallels the equivalent specificity of RluB. How this is managed will be a fascinating question for the future.

METHYLTRANSFERASES

Ten methyltransferases (MTs) are needed to make all the methylated nucleosides in 16S RNA, and 14 are needed for 23S RNA, assuming, as is likely, that a separate enzyme is responsible for each site of methylation except for the dimethyl A residues (Table 2). Genes for only 8 of the 23 MTs have been identified, and structures for only 6 have been determined. Clearly, there is still much to be done.

Sequence and Structural Homology

Crystal and NMR structures of experimentally verified MTs have demonstrated that five structurally distinct folds or classes have evolved convergently to bind the cofactor S-adenosyl-l-methionine (SAM) and perform the methyltransfer reaction (reviewed in reference 100). Consequently, there is no significant sequence conservation across all structural classes. Since all known rRNA MTs belong to either class I or class IV, our discussion will be limited to these classes.

Class I contains the majority of all known MTs encompassing small molecule, protein, DNA, and RNA MTs. Consisting as it does of such a large number and variety of MTs, class I MTs lack any overall sequence similarity except for three or four short motifs, and even these are poorly conserved (24, 64). The motifs map to the SAM-binding pocket (100). A higher degree of sequence conservation is observed in any given subfamily of class I MTs that group according to the type of methylation and type of substrate. Seven of the eight known E. coli rRNA MTs are class I members; they have been grouped into six subfamilies. RsmC (m²G1207) belongs to the RNA m²G MT family (19, 20); RsmB (m⁵C967) belongs to the RNA m⁵C MT family (98); RsmA (m⁶₂A1518,1519) belongs to the Erm family of m⁶A MT (107); RlmA^I (m¹G745), previously named RrmA, belongs to the RlmA^I/II family of RNA m¹G MT (76); both RlmC (alias RumB, m⁵U747) and RlmD (alias RumA, m⁵U1939) belong to the RNA m⁵U MT family (1, 51); and RlmE (alias RrmJ, Um2552) belongs to the RrmJ/fibrillarin family of 2'-O-MT (43).

Class I MTs contain the archetype MT fold, which consists of alternating α-helices and β-strands that produce a seven-stranded, parallel except for one strand, β-sheet flanked on both sides by α-helices. Five of the seven E. coli class I rRNA MT crystal structures have been solved. They are RsmB (m⁵C) (47), RsmA (m⁶₂A) (87), RlmA^I (m¹G) (31), RlmD (m⁵U) (74), and RlmE (Um) (18). No structure is yet available for E. coli RsmC (m²G), but its structure likely belongs to class I because a Methanococcus jannaschii homolog (MJ0882) of RsmC (20) has a class I MT fold (61). Similarly, RlmC (m⁵U) is predicted to have a class I structure because it belongs to the same family of RNA m⁵U MT family as RlmD. In the E. coli crystal structures, RlmA^I appears to be a dimer, whereas RsmB, RsmA, RlmD, and RlmE are monomers. Of the five structures, only three contain bound SAM (those for RsmB, RlmA^I, and RlmE). Although the SAM-binding site is in a cleft in a similar location in the three MTs, the orientation/conformation of SAM in each structure is different, typical of MT catalyzing three different kinds of methylation, namely, aromatic carbon methylation, aromatic nitrogen methylation, and aliphatic oxygen methylation, respectively.

Even though all five of the class I rRNA MTs share the basic MT fold, only RlmE is a single-domain structure consisting entirely of an MT fold. The other four MTs have additional domains that are presumably used for substrate recognition/specificity. RsmB has two putative RNA-binding domains: an N-terminal domain similar to the structure of the transcriptional antiterminator NusB and a central domain, termed the N1 domain, similar to several RNA-binding proteins (47). RsmA has a small (~50 amino acids) C-terminal α-helical domain of unknown function (87). RlmA^I has a small (~35 amino acids) N-terminal β-sheet with a Cys₃His Zn finger that binds a single Zn²⁺, which may be involved in rRNA binding (31). RlmD has an N-terminal TRAM domain composed entirely of β-sheet that is predicted to bind RNA (74). In addition, RlmD has a central domain with a [Fe₄S₄] cluster of unknown function (74).

All known class IV MTs are members of the trefoil knot superfamily of MT (SPOUT in the literature, referred to in this work as TREFOIL) identified by sequence similarity (4). All TREFOIL members share two C-terminal motifs (4) that map to the SAM-binding pocket (3, 42, 75, 85). The TREFOIL superfamily consists of the trmH family of RNA Gm MT, the trmD family of RNA m¹G MT, and several other families of putative MT (4). E. coli RlmB (Gm2251) belongs to the trmH family (70) and hence is an rRNA TREFOIL MT. The crystal structure of RlmB conforms to the class IV/TREFOIL MT fold consisting of a six-stranded, parallel β-sheet flanked by α-helices and an unusual C-terminal trefoil knot structure (80). Additionally, RlmB is a homodimeric MT consistent with seven other class IV/TREFOIL MT crystal structures (3, 42, 46, 75, 84, 85, 117). One of these structures, the Thermus thermophilus MT structure referred to as RrmA (RlmA^I, class I) (84), is actually a homolog of E. coli RlmB. Although the SAM cofactor was not present in the RlmB structure, SAM is probably bound in a pocket created by the trefoil knot as observed in other class IV/TREFOIL MT structures (3, 42, 75, 85). RlmB also has an N-terminal domain that is most similar to ribosomal proteins L7/L12 from Haloarcula marismortui and L30 from Saccharomyces cerevisiae and is, therefore, a putative rRNA-binding domain (80).

Enzyme Mechanism

The eight known E. coli rRNA MTs transfer a methyl group from a bound SAM cofactor to the target RNA base or sugar. The precise enzymatic mechanism of methyl transfer has not been determined experimentally for any of these MTs, and no cocrystal structures of these MTs with their RNA substrates are currently available. However, mechanisms have been proposed for four of the MTs. The RNA m⁵C MT, RsmB, and the RNA m⁵U MT, RlmC, and RlmD are thought to have a common reaction mechanism that uses the thiol of a conserved Cys residue to attack the 6 position of the pyrimidine ring, thus activating the 5 position for methyl transfer. This mechanism was worked out in detail for E. coli RumT (67), an RNA m⁵U family member that modifies tRNA, and by extension it should be valid for the rRNA-specific family members RlmC and RlmD. The mechanism is also consistent with biochemical (77) and structural (47) evidence for RsmB. How these MTs discriminate between the C and U bases is not understood.

A mechanism for 2'-O-ribose methylation by RlmE (Um) has been proposed based on the finding that the first three of the four conserved active-site residues (Lys38, Asp124, Lys164, and Glu199) are required for catalytic activity (53). In this proposed mechanism, one of the Lys residues deprotonates the 2'-OH, activating the oxygen for nucleophilic attack on the methyl group of SAM while the remaining catalytic residues are involved in stabilization. Further work is necessary to validate this mechanism.

Predicted Methyltransferase Genes

Putative class I MTs have been identified by using the SAM-binding motifs (see above) to search genome databases (64). Since all MTs bind SAM as a cofactor, using the SAM-binding motifs has the disadvantage of not being specific for rRNA MT. Alternatively, one can identify the homologs of known rRNA MTs of any class and thereby identify putative rRNA MTs. Although this approach has the advantage of being specific for an rRNA MT, rRNA and tRNA MTs are quite similar in MT subfamilies and are difficult to distinguish from each other. Consequently, while the following discussion suggests roles as rRNA MTs, the gene products could equally well be tRNA MTs. The gene products of ybiN and ygjO belong to the RNA m²G MT family (19, 20) and could be responsible for any of the remaining four m²Gs (numbers 13, 19, 25, and 32 in Table 2) in rRNA. The gene product of yebU belongs to the m⁵C MT family (98) and may be responsible for either or both of the two remaining m⁵Cs (numbers 17 and 28 in Table 2) in rRNA. The gene product of ygdE belongs to the RrmJ/fibrillarin family of RNA 2'-O-MT (43) and it may be responsible for one of the two remaining Cms (numbers 16 and 34 in Table 2) in rRNA. Finally, the gene products of lasT, ybeA, yibK, yfhQ, and yfiF belong to the TREFOIL family of Gm and m¹G MT (4). Since there are no Gm or m¹G residues in rRNA whose enzymes have not been identified, these gene products could also be responsible for the two remaining Cms (numbers 16 and 34 in Table 2) or other methylations in rRNA whose MTs are unknown. The remaining eight unidentified rRNA MTs may be in lists of putative class I MTs found by SAM-binding motif searches (64) or they may be a new, as of yet undiscovered, type of MT.

Substrate Recognition

As was the case for Ψ, sites of methylation are found in both single- and double-stranded regions. No structure for an MT in complex with its substrate exists yet, so that, except as noted below, all the results come from in vitro solution experiments.

RsmB (m⁵C967) (105).

This enzyme is able to react with free 16S RNA transcripts yielding one m⁵C967 per RNA. 30S ribosomes reconstituted from 16S RNA transcripts and protein (39) were not substrates. 30S isolated from an rsmB-minus strain, which would have all the modified nucleosides except for m⁵C967, was not tested. However, it is unlikely that the 30S is a substrate because the addition of the secondary 30S ribosomal protein S19 to an RNP of 16S RNA and S7 was sufficient to block activity (111). S7 is the primary binding protein to the 3' major domain of 16S RNA, which becomes the head of the 30S, and is a prerequisite to S19 binding (27). Presumably, S19 would still block RsmB in a 30S subunit as well. Free 23S RNA was not a substrate for RsmB, nor was 16S or 23S RNA isolated from ribosomes and thus containing the normal complement of modified bases, showing that no unnatural sites were recognized. RsmB does not appear to require a stable RNA tertiary structure because it has the same activity in the presence or absence of Mg²⁺. Many of these results were confirmed by Gu et al. (48), who showed in addition that a tRNA^Phe transcript was not a substrate and that two small stem-loop 16S RNA fragments (residues 960–975 and 927–982) could be partially methylated, 0.21 and 0.32 mol/mol of RNA, respectively. The larger 927–982 fragment was successfully docked into the active-site cleft of the RsmB X-ray structure (47).

RsmC (m²G1207) (106).

The specificity of this MT is the opposite of RsmB in that it is active with 30S reconstituted from 16S RNA transcripts and ribosomal protein but does not react with the RNA transcript alone. In contrast to RsmB, Mg²⁺ is important for RsmC. Full methylation was obtained at 0.9 mM Mg²⁺, whereas at 6 mM where 30S would be expected to be more stable, only about 0.3 methyl groups per 30S could be incorporated. EDTA was inactivating also. These results indicate a substrate specificity similar to that for RsuA and imply that the true substrate is an assembly intermediate ribonucleoprotein on the way to 30S formation. It is not known whether RsmC has any activity with 23S RNA, 50S ribosomes, or tRNA.

RsmA (m⁶₂A1518,1519).

More familiarly known as KsgA, it was the first rRNA MT to be characterized. This came about because E. coli develops resistance to the antibiotic kasugamycin by mutating the gene for this MT, thus preventing the methylation of A1518 and A1519. Now known as RsmA, this MT is responsible for formation of all four methyl groups on A1518 and A1519 of 16S RNA. 30S particles from the kasugamycin-resistant strain can be methylated in vitro at Mg²⁺ concentrations from 2 to 10 mM, but 70S ribosomes isolated from the same strain are not methylatable (95). Free 16S RNA or tRNA from the resistant strain are not substrates (58). Methylation appears to occur at a very late stage of 30S assembly since 27S particles, containing additional residues at the 3' and 5' ends of the RNA, can be isolated from E. coli without any m⁶₂A (57). Methylation of each adenosine is independent of methylation of the other since 16S RNA with G, U, or C in place of A at one site does not perturb methylation of A at the other site (28). In line with this, the putative active site in the recent crystal structure of RsmA appears to be able to accommodate only one adenosine at a time (87).

RlmA^I (m¹G745).

This MT can react in vitro with free 23S RNA transcripts but not with 50S or 70S ribosomes from a RlmA^I-deficient strain (55). Thus, even when G745 is not methylated, it cannot react once assembled into a ribosome. Fragment transcripts are also reactive in vitro. A 92mer consisting of helices 33, 34, and 35 can be stoichiometrically methylated, and trimming off half of helix 33 has only a small effect. All three helices are needed, however, since removing either helix 33 or 34 leads to low or no activity and footprinting shows protection of residues in all three helices (55). Since fragment transcripts are substrates with affinity-purified MT, there is no requirement for the adjacent modified Ψ746 or m⁵U747.

The recent crystal structure of RlmA^I (31) has provided a persuasive model of how the substrate G745 may be recognized. Two monomers form a dimer creating a W-shaped cleft that fits a model of helix 35 with parts of helices 33 and 34 contacting Zn-fingers at either side of the cleft entrance. This model is consistent with what is known from biochemical studies and places G745 at the bottom of the cleft where the methyl donor SAM is bound.

RlmC (Alias RumB, m⁵U747).

Mass spectrometry showed that inactivation of the ybjF gene caused the loss of a methyl group from the trinucleotide U746U747G748. Since there is no naturally occurring methyl at either U746 that is Ψ or G748, the methyl group was assigned to U747 and the MT was assigned to ybjF, renamed rlmC (alias rumB). Although RlmC could be expressed by several methods, no MT activity in vitro could be obtained despite testing of extracts from a variety of purification methods on either 23S RNA isolated from ribosomes of the rlmC-disrupted strain or transcripts of a 694–767 23S RNA fragment (79). Therefore, no substrate-specificity information is available. The possibility that an RNP particle or even an intact 50S might be the true substrate was apparently not considered.

RlmD (Alias RumA, m⁵U1939).

This MT reacts well with a transcript of either full-length 23S RNA or a fragment consisting of residues 1930 to 1969 to yield about 0.8 mol of methyl per mole of added RNA (1). The rate with the fragment was only slightly diminished from that with the full-length RNA. Only U1939 is a substrate since a U1939C mutant 23S RNA was unreactive. This also means that this MT can completely discriminate between U and C at the active site. 16S RNA or yeast tRNA^Phe transcripts were completely unreactive. The Mg²⁺ requirement was not tested. Surprisingly, the purified protein contains an Fe₄S₄ cluster, which has no known function in this enzyme (74).

RlmB (Gm2251).

Although protein overexpression was done for structure determination purposes (80), no biochemical experiments have been done. Consequently, no information is available on the specificity of this MT.

RlmE (Alias RrmJ, Um2552).

This MT is specific for 2'-O-methylation of U2552 (21), a universally conserved residue in the A site of the ribosomal 50S subunit. Reaction occurs with intact 50S or 70S but not with 30S, free 23S, or 16S RNA, all obtained from a rlmE mutant strain (18, 21). There is no reaction with wild-type 50S showing that the reaction is highly specific. Up to 0.8 mol of methyl can be bound per mol of ribosomes (21). Mg²⁺ is needed but this is probably to keep the ribosomes intact rather than for the methylation reaction. A 40S particle, obtained by separation of ribosomal subunits of the rlmE mutant strain at 1 mM Mg²⁺, could not be methylated in vitro.

FUNCTION OF MODIFIED NUCLEOSIDES

Pseudouridine and/or Pseudouridine Synthases

General Considerations.

The case for an important function of Ψ in ribosomes rests on several facts. First, all sequenced organisms, including those rooted early in the evolutionary tree, have putative genes for ribosomal Ψ synthases (88, 91). If Ψ were not important, evolutionary selection should have led to their loss. Second, two distinct systems for making ribosomal Ψ exist, the specific synthases, as described here for E. coli, with both the specificity site and catalytic site in the same protein, and the snoRNP particle found in eukaryotes and probably also in archaea with catalysis associated with one of its proteins but the specificity due to the RNA component (35). Retention of such elaborate systems seems wasteful if Ψ is of little consequence. Third, Ψ residues are not distributed at random but are found at specific sites, often recurring from species to species, and often in the vicinity of functionally important elements of the ribosome.

Ψ is found only in those RNA molecules whose tertiary structure is integral to their function, for example, rRNA, tRNA, tmRNA, snRNA, and snoRNA, and the very act of making Ψ creates an H-bond donor where there was none before. Thus, the main function of Ψ may be to act as a structural stabilizer. When necessary, Ψ may be added to reduce flexibility by increasing RNA-RNA contacts or in some cases by improving RNA-protein interaction. The N-1 proton of Ψ could act as a hydrogen-bond donor intramolecular cross-strut to stabilize RNA structure at critical junctures and to interact with proteins. In the case of rRNA, this could occur during or after assembly of the ribosome. Ψ could also participate in intermolecular interactions with functionally important ligands of the ribosome, such as tRNA and mRNA, or even protein synthesis factors. Stabilization could also occur by H bonding via a water molecule to its own or 5'-adjacent residue's 5'-phosphate (7) and as a result of the increase in the 3'-endo conformation of the ribose moiety of Ψ (32, 33, 40). The occurrence of variable numbers of Ψ at slightly different positions in the secondary structure of different rRNAs (89) would fit such a structural role for Ψ since, to achieve the same three-dimensional result with different sequences, each rRNA could require a somewhat different placement of Ψ. Perhaps Ψ should be viewed as a molecular glue, stabilizing and/or reinforcing necessary RNA conformations that of themselves would be insufficiently rigid. Such a "glue" might be more important near the site of peptide bond formation and tRNA binding, accounting for the preponderance of Ψ in these locations (Fig. 4 and 5). This could explain the large variation in number of Ψ among different species, the failure to find exact correspondence in the sites for Ψ among different species, and the difficulty in detecting a functional effect upon single Ψ removal (see below).

Last, note that no consumption of energy is needed for Ψ formation once the synthase is made. This is in sharp contrast to methylation, which consumes one SAM for each methyl group donated. Furthermore, Ψ formation is a simple isomerization, unlike methylation, which sequesters the methyl group that could otherwise generate energy by oxidation to CO₂. Thus, Ψ synthesis is much less energy demanding of the cell than methylation.

Experimental Studies.

In E. coli, the approach used to study Ψ function has been to delete the synthase genes, resulting in the absence of the Ψ dependent on the corresponding synthases. Since the synthase is also absent, the results have the potential for ambiguity should the synthase have a second function. Nevertheless, this approach is preferable to mutating the precursor U in the RNA to C because the phenotype of such a mutation is unpredictable. Growth has been used as a preliminary screen. Deletion of six of the seven rRNA Ψ synthase genes had no effect on growth over a range of temperatures in rich medium; and likewise, the three deletion mutants tested in minimal medium showed no growth effect (Table 3).

Table 3Effect of deletion of E. coli rRNA pseudouridine synthases on cell growth

The only synthase whose absence caused a major decrease in growth rate under normal conditions was RluD, the synthase that makes Ψ1911, Ψ1915, and Ψ1917 in the helix 69 stem-loop. This growth defect was the result of a major failure in assembly and possibly stability of the large ribosomal subunit (91; N. Gutgsell and J. Ofengand, unpublished results). The defect could be prevented by supplying the rluD structural gene in trans. Therefore, the RluD synthase and/or its product Ψ appear to be directly involved in 50S subunit assembly/stability. The three Ψ made by RluD are highly conserved, and Ψ1917 is so far universally conserved. In fact, it is the only Ψ position to be totally conserved among all organisms tested (88). In the 70S tRNA structure (116), the loop of helix 69 containing the Ψ supports the anticodon arm of A-site tRNA near its juncture with the amino acid arm. The middle of helix 69 does the same thing for P-site tRNA. Unfortunately, the resolution in this work was not sufficient to reveal any specific roles for the Ψ residues. On the other hand, a recent study of the location of the ribosome release factor RRF on the 70S ribosome has shown that m³Ψ1915 makes a direct contact with amino acid residues E122 and V126 of RRF (2). It may therefore turn out that the primary role for Ψ in helix 69 is in ribosome release rather than in the more conventional aspects of protein synthesis.

To test whether multiple Ψ deletions would have a synergistic effect, a quadruple mutant was constructed lacking rluB, rluC, rluE, and rluF and thus lacking six Ψ, five of which are in and around the peptidyltransferase center. As shown in Table 3, at 37°C in rich medium, the cells grew normally in exponential phase.

Overall, these results, with the exception of those for RluD, show that the role of Ψ in ribosomes is subtle. However, so far, only the simplest of conditions have been tested. Various types of stress conditions need to be investigated now that a complete set of Ψ synthase gene deletions is available.

Methylated Nucleosides and/or Their Methyltransferases

As with Ψ, the approach taken to understanding function has been to first identify the gene for a specific MT, then to inactivate it by deletion or disruption, and finally to look for a functional effect. The same potential for ambiguity exists here also since gene inactivation results in the loss of the MT and the methylated product of the MT, and of course the approach is limited to MTs whose genes are known. There are eight such MTs (nine methylation sites) known so far (Table 2). No strain with an inactive rsmC (m²G1207) (106) or rlmD (alias rumA, m⁵U1939) (1) has been constructed, so no functional information is available for these two cases. Three other strains with inactive MT, ΔrsmB (m⁵C967) (48), ΔrlmC (alias ΔrumB, m⁵U747) (79), and ΔrlmB (Gm2251) (78) showed no growth rate defect, and, in the latter two cases, did not even show a growth defect in competition with wild type. ΔrsmB was not tested for competition. Lack of RsmA (m⁶₂A1518,1519) activity, despite blocking the addition of four methyl groups as two m⁶₂A residues, had only small effects on translation, although the cell was rendered resistant to the antibiotic kasugamycin (109).

In only two of the five cases tested did inactivation of an MT have a strong effect. Inactivation of rlmA^I (m¹G745) reduced the growth rate in rich medium at 37°C to 60% of wild type, caused a 20% reduction in polypeptide elongation rate, and increased the amount of dissociated ribosomal subunits to 1.7 times wild type (50). These effects disappeared when the gene was reintroduced on a plasmid. The effects are due to the absence of RlmA^I rather than to the lack of methylation since mutation of G745 to C, U, or A had no growth effect (75a). Hence, a second function of RlmA^I is required for proper cell growth.

RlmE (alias RrmJ, Um2552) is a highly inducible heat-shock protein that is the sole MT for U2552 methylation. Inactivation of rlmE had an even stronger effect on cell growth, ribosomes, and in vitro protein synthesis than did inactivation of rlmA^I. Inactivation caused a 50% reduction in growth rate in rich or poor media at 37°C or 42°C (22), or no growth on McConkey agar and a 2.5- to 3-fold reduction in growth rate in liquid LB medium (18, 53), all of which were reversed upon plasmid rescue with a wild-type copy of rlmE. The ΔrlmE strain was fourfold more sensitive to lincomycin and had a rate of polypeptide elongation that was only 25% to 30% that of wild type (22). The mutant strain also had a long delay in adjustment to heat shock (18). All these effects are probably explicable in terms of the ribosome defects of the mutant strain. One-third to one-half of the 50S subunits are defective in that they give rise to a 40S particle at 1 mM Mg²⁺, a condition under which wild-type 50S subunits are still stable (18). The 40S particles were subsequently found to be deficient in seven late assembly proteins (53). This effect likely accounts for the marked increase in subunits at 10 mM Mg²⁺, where normally only 70S ribosomes are found (18, 22). The reason behind the protein deficiency in the 40S particles is not known but, as noted above, although 50S from the mutant strain can be methylated in vitro, the 40S cannot.

These results are surprising because similar inactivation of rlmB, the MT that makes Gm2251 in the P loop of 23S RNA, had no detectable functional effect (see above). Clearly, the defects cannot be attributed to a generalized effect of 2'-O-methylation or even to being part of a tRNA-binding site on the ribosome (U2552 is part of the A loop) but rather must reflect a specialized function of U2552 and/or of its MT.

Are the effects due to the loss of the RlmE protein or to the absence of Um2552? Treatment of an S30 extract of the mutant with SAM and purified RlmE doubled the rate of polypeptide elongation up to two-thirds of the wild-type rate (22). At first glance, this result implies that the methyl group was responsible. However, in an S30 extract many changes to ribosomes could occur involving other molecules working in concert with RlmE that do not involve methylation. A more conclusive experiment would have been to treat purified ribosomes with purified RlmE and SAM, and then assay peptide bond formation, but this was not done.

Another approach to the same question involved making point mutations in RlmE to see if the growth/ribosome defects could be separated from the methylation of Um2552 (53). Unfortunately, these authors used the in vitro methylation activity of mutant RlmE as a substitute for direct measurement of Um2552. Even so, the results showed that some amino acid substitutions, for example, D83A and E199A, had widely different growth/ribosome effects but similar in vitro methylating activity. Thus, the k_cat/K_m for methylation by the D83A mutant possessing a mutant growth/ribosome phenotype was actually 1.5 times greater than for the E199A mutant with an almost wild-type phenotype. These results suggest, in contrast to the conclusions of the authors, that the RlmE protein has a role in ribosome assembly or stability that is independent of its methylating ability. They do not, however, rule out an additional role of the methyl group, although this still has to be shown by more direct in vitro studies.

Dihydrouridine

No synthase for hU2449 in 23S RNA has been identified (Table 2). Three synthases for hU in tRNA have recently been described but no information is available on whether any of them can also make hU2449 (12). This is currently under investigation (C. Ornek and J. Ofengand, unpublished results). Regarding a function for hU, mutation of the universally conserved U2449 to C had no detectable effect on cell growth, heat shock or cold shock survival, translational fidelity, or antibiotic sensitivity (86). This result should strongly indicate that hU is not needed for any of these reactions. However, the lack of effect of the U2449C mutation is at odds with the universal absence of C at position 2449 in all sequenced rRNAs, suggesting that U2449, and perhaps hU, is involved in some other aspect of ribosome assembly, structure, or function. It is not known when during the course of 50S subunit synthesis hU is made, nor whether the substrate is free 23S RNA or 50S ribosomes.

In contrast to this lack of information, there have been studies on the physical properties of hU in an RNA chain. A careful NMR study by Dalluge et al. (30) has shown that an important role of hU in RNA is to increase local flexibility. It does this by favoring the inherently more flexible C2'-endo sugar conformation over the C3'-endo form. It also causes complete destacking of bases in its vicinity. Thus, hU facilitates tertiary interaction while at the same time favoring loop formation. The location of hU just two residues away from the postulated site of peptide transfer from P-site tRNA to A-site tRNA, namely A2451, suggests that hU may be there to provide flexibility in vivo. A role in providing needed RNA flexibility is supported by the finding that psychrophilic bacteria (those needing low temperature to grow) have increased levels of hU (29), presumably to compensate for the decrease in thermal motion at their optimum growth temperature.

CONCLUDING REMARKS

Summary

There are 15 different kinds of modified nucleosides in E. coli ribosomal RNA (Fig. 1, Tables 1 and 2). Three are only in 16S RNA, eight are only in 23S RNA, and four are in both RNAs. 16S RNA has one Ψ and 10 methylated nucleosides, containing a total of 13 methyl groups (Fig. 2). They are localized on the 30S ribosome at or near the interface with the 50S subunit and in the vicinity of the binding sites for mRNA and the anticodon arms of tRNA (Fig. 4). 23S RNA has 10 Ψ, 14 methylated nucleosides with a total of 14 methyl groups, one dihydrouridine, and one unidentified modified C residue (Fig. 3). They are localized on the 50S ribosome in the peptidyltransferase area, loosely defined as that part of the 50S that binds the amino acyl ends of tRNAs in the A, P, and E sites (Fig. 5).

Each of the enzymes that make the modified nucleosides is specific both for the chemistry of modification and for the site in the RNA chain, with three exceptions. RsmA makes two adjacent m⁶₂A in 16S RNA, RluD makes three Ψ within a seven-residue loop, and RluC makes three Ψ that are widely separated in space. The ability of RsmA and RluD to make multiple modifications separated by only a few residues has a precedent in the reaction of TruA, which makes Ψ at residues 38–40 in tRNA, but the specificity rules for RluC remain mysterious. The 7 known Ψ synthases are sufficient to make the 11 known Ψ. It is improbable that any more Ψ exist since the entire 16S RNA has been examined (8) and the 3' 21 nucleotides of the 23S RNA that remained unexamined by Ψ sequencing (9) had been previously sequenced by classical methods (15). It is equally unlikely that any other Ψ synthases exist since there are no additional genes with sequence homology to the existing synthase genes. If one without any sequence homology did exist, it would of necessity only duplicate the activity of a known synthase. Crystal structures exist for three rRNA Ψ synthases. They all share a common protein-folding domain that includes the active-site aspartate, whose γ-COOH group is believed to be the key element in catalysis. The structural conservation of this domain is striking and is shared by the three crystal structures for tRNA Ψ synthases. The various enzymes have unique domains added to the conserved domain at the N or C terminus that likely provides the substrate-specificity determinants. Biochemical studies indicate that the modes of substrate recognition are idiosyncratic for each Ψ synthase since no common mode of recognition has been detected in studies of the seven synthases. Eight of the 24 expected MTs have been identified, and six crystal structures have been determined. Seven of the MTs and five of the structures are class I MTs with the appropriate protein fold plus unique appendages as described above for the Ψ synthases. The remaining MT, RlmB, has the class IV trefoil knot fold. All eight MTs use SAM as methyl donor, adding it to C, N, or O atoms, and as for Ψ synthases, the eight MTs use diverse modes of substrate recognition.

All sequenced eubacterial organisms have genes for putative Ψ synthases, and a complex guide RNA system for ribosomal RNA Ψ synthesis exists in eukaryotes and archaea. Thus, Ψ is an ancient invention common to all organisms and, since it has survived to this day, should be of value to the cell. Methylation of rRNA is also ubiquitous. Despite this, the function of modified nucleosides in rRNA remains unclear. The method of choice for assessing function has been to inactivate, by deletion or disruption, the gene for the Ψ synthase or MT in question and then look for an in vivo effect. Of the seven Ψ synthase and five MT genes tested, inactivation of only three, rluD, rlmA^I, and rlmE, showed strong effects on cell growth, reducing it to 16%, 60%, and 50% of wild type, respectively, and causing ribosome defects. It is not clear if the effect in the rluD and rlmE cases was due to the loss of the modified nucleoside or to the loss of the enzyme. In the rlmA^I case, it was clearly due to the lack of the MT and not to the lack of the modified nucleoside. Surprisingly, loss of the nine other enzymes singly and in certain combinations had no apparent effect.

What Is Still To Be Learned?

As noted above, the evidence is strong that all the Ψ have been discovered. This is probably true also for the methylated residues. The 16S RNA has been well analyzed by conventional methods, and a complete survey of 23S RNA by reverse transcription following carboxymethyl cellulose or hydrazine treatment only revealed one new modified residue, now confirmed as a modified C2501. The method detects any modification interfering with base pairing and also detects m⁵U. Nevertheless, modifications that do not interfere with reverse transcription and, in particular, that are labile to the isolation technique used, could still exist. If any do exist, it is more likely that they will be discovered as a byproduct of some other line of research rather than by a directed search.

Regarding the enzymes, 16 MTs have still to be discovered and characterized to account for the known methylated residues, assuming that each methyl group requires its own enzyme. The enzyme for hU2449 has not been identified, and the enzyme(s) for C2501 modification have still to be discovered. Some of these studies are currently in progress in the authors' laboratory.

Future Directions

The functions of modified nucleosides in rRNA still need to be understood since lack of most of them, singly or in certain combinations, has no discernable effect. What is being missed? Possibly one needs to look beyond standard laboratory conditions and into stress situations that may be deleterious in the absence of modification. Possible functional roles for the modifying enzymes in addition to making modified nucleosides also need to be examined. Good evidence for such an additional role was recently obtained for RlmA^I (75a) and some less clear-cut evidence exists for RluD (Gutgsell and Ofengand, unpublished results) and RlmE (53) as well. Clearly, one cannot look for additional roles unless the lack of enzyme causes a physiological effect, so the search for less obvious effects should be intensified. One role for the modifying enzymes might be as an RNA chaperone to assist in correct folding. In those cases where deletion of a modification enzyme gene results in a slow growth phenotype, analysis of fast-growing second-site suppressor mutations (pseudorevertants) should be pursued. For example, pseudorevertants of the ΔrluD strain were readily obtained (96). Identification of the second site mutation should be of great assistance in unraveling the mechanism behind the effect of the original mutation if the second site gene has a known or predicted function.

ACKNOWLEDGMENTS

Preparation of this review was supported in part by National Institutes of Health Grant GM58879 (J.O.), National Science Foundation Grant MCB0315684 (J.O.), and National Institutes of Health fellowship GM66374 (M.D.)