Crystal structure of the catalytic domain of RluD, the only rRNA pseudouridine synthase required for normal growth of Escherichia coli

Abstract

Escherichia coli pseudouridine synthase RluD makes pseudouridines 1911, 1915, and 1917 in the loop of helix 69 in 23S RNA. These are the most highly conserved ribosomal pseudouridines known. Of 11 pseudouridine synthases in E. coli, only cells lacking RluD have severe growth defects and abnormal ribosomes. We have determined the 2.0 Å structure of the catalytic domain of RluD (residues 77–326), the first structure of an RluA family member. The catalytic domain folds into a mainly antiparallel β-sheet flanked by several loops and helices. A positively charged cleft that presumably binds RNA leads to the conserved Asp 139. The RluD N-terminal S4 domain, connected by a flexible linker, is disordered in our structure. RluD is very similar in both catalytic domain structure and active site arrangement to the pseudouridine synthases RsuA, TruB, and TruA. We identify five sequence motifs, two of which are novel, in the RluA, RsuA, TruB, and TruA families, uniting them as one superfamily. These results strongly suggest that four of the five families of pseudouridine synthases arose by divergent evolution. The RluD structure also provides insight into its multisite specificity.

INTRODUCTION

Cells devote time and metabolic energy to making posttranscriptional modifications to their RNAs. The most abundant single nucleoside modification is the isomerization of uridine (U) to pseudouridine (5-ribosyluracil, Ψ). To date, Ψ has been detected in tRNA, rRNA, tmRNA, snRNA, and snoRNA, but not, so far, in mRNA. The precise role of Ψ in these RNAs has remained elusive, despite the many hints to its physiological importance that have been uncovered. For example, most ribosomal Ψ cluster in the vicinity of the peptidyl transferase center of the ribosome, the heart of ribosome function (Ofengand et al. 2001). Also, a single Ψ in U2 snRNA appears to stabilize the extrahelical branch point adenosine used in RNA splicing (Newby and Greenbaum 2002).

Ψ synthases are the enzymes responsible for the U-to-Ψ conversion in a reaction that requires no cofactor or energy source. Amino acid sequence similarity has been used to group Ψ synthases into five families, each named for the first Escherichia coli family member to be characterized: TruA, TruB, RsuA, RluA, and TruD. Statistically significant sequence similarity is found between the RluA and RsuA families using the BLASTP algorithm (~20% identity over ~200 residues) and it can be extended to include only the TruB family using the PSI-BLAST algorithm (Ofengand and Rudd 2000). The sequence similarity between families is located primarily in short sequence motifs. The RsuA and RluA families share three motifs, the TruB family shares only motifs I and II, and only motif II is shared by the TruA family (Gustafsson et al. 1996; Koonin 1996; Huang et al. 1998b). Motif II contains an aspartate residue that is conserved among these four families and is essential for catalytic activity (Del Campo et al. 2001). Even the TruD family has an essential conserved aspartate, although it lacks a conserved motif II (Kaya and Ofengand 2003). In the past three years, the E. coli TruA (Foster et al. 2000), TruB (Hoang and Ferré-D’Amaré 2001), and RsuA (Sivaraman et al. 2002) crystal structures have shown that despite their sequence conservation being limited to three or less motifs, they share the same basic fold in their catalytic domain along with placement of residues in the active site—clearly showing Ψ synthases arose by divergent evolution (Mueller 2002). In addition, the cocrystal structure of TruB with a 22mer T loop RNA elegantly showed that Ψ synthases gain access to their substrate U by a base flipping mechanism that places the U in close proximity to the catalytic aspartate (Hoang and Ferré-D’Amaré 2001).

In E. coli, the entire collection of Ψ in both rRNA and tRNA and the 11 responsible Ψ synthases have been determined (Kaya and Ofengand 2003). Deletion of each of the synthases in turn has shown that RluD is by far the most physiologically important synthase. Cells lacking functional RluD have a growth rate 20% of wild type (Gutgsell et al. 2001) and their ribosome particles are abnormal (Ofengand et al. 2001). Hence, cells require RluD and/or its Ψ products for normal ribosome function. RluD is an RluA family member responsible for making Ψ 1911, 1915, and 1917 in 23S RNA (Fig. 1A,B,C ▶). These three Ψ are in the loop of helix 69 (H69) and are the most conserved ribosomal Ψ known (Ofengand 2002). In the 70S ribosome structure, H69 is the main feature of the B2a intersubunit bridge and makes contacts with both P- and A-site tRNAs (Yusupov et al. 2001; Bashan et al. 2003). Here we describe the structure of the catalytic domain of RluD, the first structure of any RluA family member. The structure provides clues toward understanding the multisite specificity of RluD and reveals new similarities among the first four Ψ synthase families.

FIGURE 1.

The substrate of RluD is helix 69 in 23S RNA. (A) A schematic of the secondary structure of the 3′ half of E. coli 23S RNA from the Comparative RNA Web site (http://www.rna.icmb.utexas.edu). H69 is indicated by an arrow. (B) Closer view of H69 showing the Ψ made by RluD. The asterisk indicates N3-methyl-Ψ. (C) H69 from the structure of the 50S ribosomal subunit of Deinococcus radiodurans, which was modeled with unmodified uridines (PDB entry 1NKW; Harms et al. 2001). The sequence of H69 in D. radiodurans is identical to that of E. coli. In this panel, uridines 1911, 1915, and 1917 are black and the remainder of H69 is gray. Panel C was generated with PyMOL (DeLano Scientific).

RESULTS

Overall structure of RluD

The 2.0 Å structure of the catalytic domain of RluD (R = 0.219, R_free = 0.232) consists of 250 amino acid residues (Phe 77 to Leu 326 at the C terminus) and 261 water molecules. The N-terminal domain consisting of a 23 residue affinity tag and residues M1 to R76 is disordered in our electron density maps and could not be modeled. The catalytic domain (Fig. 2A,B ▶) folds into an extended nine-strand β-sheet consisting of two antiparallel sheets (β1-β2-β3 and β11-β10-β4-β9-β8-β5) joined by a parallel interaction between one strand in each sheet (β3 and β10). The front face of the sheet contains secondary structure elements clustered into two groups: four helices (α1-3₁₀1-3₁₀2-α2) and several large loops in one group and one helix (α3), two short strands (β6-β7), and more loops in the other. The juxtaposition of these groups forms a deep, central cleft in RluD with average dimensions 25 Å long by 10 Å wide by 14 Å deep. The essential, catalytic aspartate residue (Asp 139) is located at the base of this cleft. Off to the far side of the main sheet is a subdomain, which we term the tail region, consisting of two helices (α4 and α5) and several loops.

FIGURE 2.

Model of the catalytic domain of RluD. (A) Ribbon cartoon of secondary structure elements shown in cyan, except for the linker to the N-terminal domain (residues 77–84 [L1]) in red and the tail region (residues 249–276 [L13–α4-L14] and 300–326 [L16–α5-L17]) in green. The side chain of catalytic Asp 139 (yellow) is shown in ball and stick representation. Note: The C_α of Asp 139 appears to be separated from L4 because the ribbon cartoon is smoothed. (B) A 90° turn of the view shown in A. Generated with PyMOL (DeLano Scientific).

RNA binding cleft and the active site

One or more residues from helices 3₁₀2, α2, and α4, strands β3, β4, β6, β7, β9, and β10, and loops L3, L4, L6, L9, L10, L12, L13, and L14 come together to make up the walls and floor of the central cleft in RluD. The walls of the cleft have a highly positive character contributed by residues Arg 131, Arg 137, Arg 165, Arg 169, Arg 191, Lys 206, and Arg 232 (Fig. 3A,C ▶). This positive cleft could be used to bind and position a largely negative substrate RNA, consistent with other Ψ synthase structures (Foster et al. 2000; Hoang and Ferré-D’Amaré 2001; Sivaraman et al. 2003). The cleft walls start out wide (~20 Å at the widest point) and taper inwards (~2.7 Å at the narrowest point) over many side chain and backbone protrusions, eventually leading to a roughly spherical, central cavity (~6.5 Å long, 6.5 Å wide, 5.5 Å deep). The cavity is formed largely by the side chains of Asp 139, Thr 142, Leu 228, Ile 236, Arg 237, and Leu 279 (Fig. 3C ▶). The presence of the catalytically essential Asp 139, as well as other conserved residues described in Ψ synthase active sites (see below), defines this cavity as the active site of RluD. The side chain of Asp 139 hydrogen bonds to two of eight water molecules that occupy the active site.

FIGURE 3.

Active site and crystal packing. (A) Molecular surface of RluD colored by local electrostatic potential (using GRASP; Nicholls et al. 1991): positive is blue and negative is red. This view is from above the positively charged putative RNA-binding cleft. Note the central, small, uncharged active site pocket flanked by the catalytic Asp 139 (asterisk). Arg 165 is marked by an arrow and loops L3 and L9 are indicated. (B) Same surface representation as in A turned 180° to view the negatively charged tail region (circled). (C) Stereo view of the electron density in the active site pocket from a 2F_o-F_c map contoured at 2.0 σ. Carbons are yellow, nitrogens are blue, and oxygens are red; lone oxygens represent water molecules. (D) Stereo view of the crystal packing of RluD molecules shown as C_α backbones. The cleft of the central molecule (red) contacts the tail region of a symmetry related molecule (green). The yellow sphere marks the position of the N-terminal residue Phe 77. (E) The interaction between the cleft of one RluD molecule and the tail region of a symmetry-related molecule. Molecular surface of RluD is shown in gray, residues involved in the contact are yellow, and residues from two loops (L3 and L9) with highest main chain B-factors are red. Residues 219 (blue), 262–269 (green), and 314–326 (cyan) of a symmetry mate are shown in stick representation. Note the position of Asp 139 (asterisk) and Arg 165 (arrow). All panels generated with PyMOL (DeLano Scientific).

Crystal contacts in the putative RNA binding cleft

Despite solving the structure of RluD in the absence of any RNA, the presumed RNA-binding cleft is not empty. In our crystals, symmetry-related RluD molecules pack such that the negatively charged tail region (Fig. 3B ▶) of one symmetry mate (in green) is bound in the positively charged cleft of another (Fig. 3D ▶, in red). The tail region of a symmetry mate occludes most of the cleft and completely occludes the catalytic pocket of RluD. This interaction involves 28 cleft residues and 17 tail region residues through numerous bridging water molecules, hydrogen bonds, and van der Waals contacts (Fig. 3E ▶). Approximately 14% of the total solvent-accessible surface area (SASA) of a single RluD molecule is buried in this crystal packing interaction: 7% of the cleft and 7% of the tail region. This is on the order of a protein–RNA interaction, considering that ~9% of the SASA of TruB is buried in contacts with a 22mer RNA substrate (Hoang and Ferré-D’Amaré 2001). However, this interaction is obviously a crystal packing artifact, because, in nature, the cleft is needed for binding RNA. It also renders this crystal form useless for binding small substrate nucleosides or RNAs.

Part of the cleft is actually closed off by this crystal packing interaction. The side chain of Arg 165 from loop L6 closes one side of the cleft (see arrow, Fig. 3A,E ▶) because it is making a salt bridge to Glu 322 in the tail region of a symmetry mate. In this position, Arg 165 also makes a single hydrogen bond to the carbonyl of Glu 229 in loop L12. In the absence of the salt bridge, the side chain of Arg 165 could be in another conformation, reopening one end of the cleft, and effectively making the entire cleft larger.

Four families of Ψ synthases are structural neighbors

A Dali (Holm and Sander 1993) search identified the Ψ synthases RsuA, TruB, and TruA as the closest structural neighbors to the catalytic domain of RluD. The resulting superposition of their C_α backbones and alignment of their sequences are shown in Figures 4A ▶ and 5 ▶, respectively. RsuA is the closest neighbor (Z score = 11.4; RMSD = 2.8 Å over 145 C_α atoms), followed by TruB (Z score = 8.5; RMSD = 3.0 Å over 142 C_α), and then TruA (Z score = 8.0; RMSD = 3.7 Å over 159 C_α). The superimposed structures clearly illustrate the core fold of the four Ψ synthase families, which includes the central beta-sheet arrangement, two helices, and two loops (β2, α1, L4, β3, β4, β8, β9, L12, α3, β10; see Fig. 5 ▶), and the conserved placement of the active site (Fig. 4A ▶). However, the superimposed structures also reveal that each synthase has its own additional features. A pseudouridine synthase and archaeosine tRNA guanine transglycosylase (PUA) RNA-binding domain has been appended to TruB at its C terminus (Aravind and Koonin 1999; Ferré-D’Amaré 2003), and an S4 RNA-binding domain has been added to RsuA at its N terminus (Aravind and Koonin 1999; Sivaraman et al. 2002). Interestingly, the tail region of RluD is a unique feature. Only TruA has a loop in this area, but it is used for dimerization contacts.

FIGURE 4.

The conserved core fold of four families of Ψ synthases. (A) Stereo view of the C_α backbones of RsuA (PDB entry 1KSK; red), TruB (PDB entry 1K8W; green), and a monomer of TruA (PDB entry 1DJO; gray) superimposed onto RluD (blue) by the Dali server (Holm and Sander 1993). Nonoverlapping domains are labeled. An asterisk marks the location of the catalytic site in all four enzymes. (B) Superimposed side chains (starting at the C_α) from five conserved active site residues in RluD, RsuA, TruB, and TruA. The N, C_α, C_β, C, and O atoms of all five residues from RsuA, TruB, and TruA were superimposed onto the equivalent five residues of RluD using LSQMAN (Kleywegt 1999). Colors are the same as in A. All panels generated with PyMOL (DeLano Scientific).

FIGURE 5.

Structural sequence alignment of RluD, RsuA, TruB, and TruA. The entire sequence of RluD (residues 76–326) is shown with every 10th residue marked underneath. Parts of RsuA, TruB, and TruA that were superimposed onto RluD by the Dali server (Holm and Sander 1993) are shown underneath. Gaps in sequences are noted by dashes. The number of residues left out because they do not align to RluD is shown in parentheses. Secondary structure elements are shown above their respective sequences, and are labeled for RluD. Positions of the classical Ψ synthase motifs I, II, and III and two new motifs IIa and IIIa are indicated. Conserved active site residues are shown boxed and highlighted in yellow; other conserved residues are highlighted light blue if they occur in all four sequences or orange if they occur in three sequences.

Five residues in the active sites of RluD, RsuA, TruB, and TruA are conserved. The catalytic aspartate, a basic arginine or lysine to make a salt bridge to the aspartate, and a tyrosine that provides a stacking interaction for the uracil base have all been previously noted (Hoang and Ferré-D’Amaré 2001). There are also two hydrophobic residues: a leucine and a second residue that can be an isoleucine or valine. These five residues appear to be completely conserved among these four families of Ψ synthases (M. Del Campo, unpubl. observations). In fact, the extended main chain atoms (N, C_α, C_β, C, and O) of these five residues from RsuA, TruB, and TruA superimpose onto RluD with RMSD’s of 0.51 Å, 0.67 Å, and 0.75 Å, respectively. The side chains are aligned as shown in Figure 4B ▶. Overall the side chains are similar with some conformational differences. Asp 102 of RsuA points away from the active site (Mueller 2002). Arg 181 of TruB points back into a position that cannot be adopted by the basic residues of RluD, RsuA, or TruA because of the presence of a histidine (His 234, His 184, and His 202, respectively) in that location (data not shown). Despite these differences, the close similarity of the active sites in size and composition implies that the base-flipping mechanism observed for TruB (Hoang and Ferré-D’Amaré 2001) is likely also used by RluD, RsuA, and TruA.

Motifs: something old, something new

It was suggested (Ofengand and Rudd 2000), but never shown, that motif III exists in the TruB family and motifs I and III exist in the TruA family. From the structural alignment in Figure 5 ▶, it is clear that TruB contains motif III and TruA contains both motifs I and III. Motif I in the TruA family is clearly different from the other three families, consistently having a YD in place of the KP. The five conserved amino acids in the active site (Fig. 4B ▶) are distributed among motifs II, IIa, III, and IIIa (Fig. 5 ▶). The conserved aspartate is in motif II, the arginine/lysine residue is in motif III adjacent to the isoleucine/valine residue, the tyrosine is found in a new motif we designate as IIa, and the newly recognized leucine occurs in a new motif IIIa. Motif IIa occurs after motif II in a β-strand (β4 in RluD) and generally follows the form “+xYx###” where + is a positive side chain, x is any residue, and # is a hydrophobic residue. Motif IIIa is located adjacent to motif III in a β-strand (β10 in RluD) and would technically be part of motif III if it were not for the insertion of a long loop following motif III in some Ψ synthase sequences, like RluD and TruA (Fig. 5 ▶). Motif IIIa differs between families, but it generally follows the form “Lxxxx#”. To determine whether all five motifs were conserved in other members of the RluA, RsuA, TruA, and TruB families, we examined sequences from each family obtained by BLAST searches (http://www.ncbi.nlm.nih.gov/BLAST) against databases of finished genomes at the National Center for Biotechnology Information (NCBI). All five motifs were found in 50 RluA members from 15 bacteria and 3 eukaryotes, 25 RsuA members from 15 bacteria and 1 eukaryote, 34 TruB members from 13 bacteria, 16 archaea (Yeast Cbf5-like), and 3 eukaryotes, and 37 TruA members from 16 bacteria, 12 archaea, and 3 eukaryotes (data not shown). There are no known RluA or RsuA family members in the archaea (Ofengand 2002). This preliminary analysis shows that motifs I, II, IIa, III, and IIIa are conserved in the RluA, RsuA, TruB, and TruA families. Thus, it appears the first four families of Ψ synthases to be characterized are more similar in sequence than originally described (Koonin 1996) and comprise one Ψ synthase superfamily.

Clues toward understanding the multisite specificity of RluD

RluD has been shown to make all three substrate Ψ on H69 of an in vitro transcript of 23S RNA. Thus, ribosomal proteins are not essential (Raychaudhuri et al. 1998; Wrzesinski et al. 2000). There are three ribosome structures available that include H69, so it would be reasonable to try to dock these structures into the putative RNA binding cleft of RluD. However, H69 is disordered in the 50S structure from Haloarcula marismortui (Ban et al. 2000) and in the 70S ribosome from Thermus thermophilus, only the backbone of this helix is discernable because of limited resolution (Yusupov et al. 2001). The only structure of H69 that is ordered and to sufficient resolution is from the 50S of another eubacterium Deinococcus radiodurans (Fig. 1C ▶; Harms et al. 2001). Attempts at docking this RNA into the cleft of RluD did not yield any orientations without steric clashes (data not shown) because the cleft, although big enough to accommodate this RNA in several ways, does not provide a complementary surface to the H69 structure. However, it did define some restraints. It is clear that uridines 1911, 1915, or 1917 cannot get into the active site cavity without base flipping from their positions in the stem-loop, and the active site can only accommodate one uridine at a time. Thus, more than one conformation of the RNA and/or RluD must exist to allow each uridine to flip into the active site. Two conformations of H69 have already been noted (Harms et al. 2001).

The N3-methyl-Ψ at 1915

In E. coli 23S RNA, the naturally occurring modification at uridine 1915 is N3-methyl-Ψ (Kowalak et al. 1996). This means an as yet unidentified E. coli methyltransferase also recognizes H69, and specifically modifies position 1915. Cells lacking RluD appear to be methylated on uridine 1915, so the methyltransferase would appear to not require Ψ for action (Raychaudhuri et al. 1998; Gutgsell et al. 2001). However, it is not known if RluD can convert N3-methyluridine to N3-methyl-Ψ or whether the methyltransferase can convert Ψ to N3-methyl-Ψ. Perhaps both reactions are possible. If RluD can convert N3-methyluridine to N3-methyl-Ψ then the active site pocket of RluD must be able to accommodate the uracil base plus the N3-methyl group. To test this, we modeled the structure of N3-methyluridine (Partridge and Pritchard 1995) into the active site pocket of RluD (data not shown). Although the active site pocket appears to be big enough to accommodate the N3-methyluracil base, no conclusion can be made regarding the orientation or placement of this modification in the absence of structural data on RNA substrate binding.

The N-terminal S4 domain

The S4 domain, named after the rRNA-binding domain of ribosomal protein S4, is a small, modular domain found in many proteins either known or predicted to bind RNA (Aravind and Koonin 1999; Staker et al. 2000). It is found at the N terminus of some RsuA and RluA family members that modify rRNA, and it has been shown that RluD residues Gln 18-Asn 68 fit the consensus for an S4 domain (Sivaraman et al. 2002). Hence, the N-terminal domain of RluD most likely folds into a similar structure. The S4 domain should be connected to the catalytic domain of RluD through the end of loop L1 (Fig. 2B ▶). Because extended loops are known to be substrates for proteases, as was noted for the linker between RsuA and its S4 domain (Fig. 4A ▶; Sivaraman et al. 2002), it is possible to think that the S4 domain of RluD in our crystals is not disordered but rather has been removed by a protease. Three lines of evidence argue that the domain is present but disordered. First, RluD crystals analyzed by gel electrophoresis showed migration consistent with full-length protein (data not shown). Second, there is electron density extending past F77, but it is too poor in quality to be certain in which direction the backbone is going. Third, there is more than adequate space around each molecule of RluD, in the asymmetric unit, for an S4 domain to be accommodated (Fig. 3D ▶). The current solvent content is on the high end, ~75% (average protein crystals are 40%–60 solvent), whereas the estimated solvent content for full-length RluD including the purification tag is 60% (Del Campo et al. 2003).

DISCUSSION

The structure of RluD further emphasizes that the RluA, RsuA, TruB, and TruA families of Ψ synthases arose by divergent evolution from a common ancestor. Currently, the only structures with overall significant similarity to RluD in the Protein Data Bank are RsuA, TruB, and TruA. It seems that during the course of evolution, the Ψ synthase core fold has been reserved for carrying out the uridine-to-Ψ conversion. In addition, we have identified even more sequence similarities in the core between these families. The assertion that motifs I, II, and III exist in all families (Ofengand and Rudd 2000) has now been shown by a structural alignment (Fig. 5 ▶). Two novel additional motifs, IIa and IIIa, are also shared among these families. The five shared motifs unite the RluA, RsuA, TruB, and TruA families into one superfamily. All five motifs come together in space to form the core of the Ψ synthase fold (Fig. 6 ▶). Whereas motif I appears to be solely structural in nature based on biochemical (Spedaliere et al. 2000) and structural (Mueller 2002) evidence, motifs II, IIa, III, and IIIa not only provide hydrophobic residues to the hydrophobic core of the Ψ synthase fold, but also contribute five conserved residues to the active site pocket (Fig. 6 ▶). Three of these residues, the catalytic aspartate (motif II), a tyrosine (motif IIa), and a leucine (motif IIIa), are almost invariant among these four Ψ synthase families (M. Del Campo, unpubl. observations). Although the tyrosine has recently gained attention for possible stacking interactions with the flipped out target uridine (Hoang and Ferré-D’Amaré 2001; Ferré-D’Amaré 2003), the leucine has not been noted previously. Of the remaining two conserved residues in the active site pocket, the basic arginine/lysine and hydrophobic isoleucine/valine (motif III) both have been noted previously (Hoang and Ferré-D’Amaré 2001; Sivaraman et al. 2002).

FIGURE 6.

Five Ψ synthase motifs form the core of the structure of RluD. Motifs I (residues 93–99), II (residues 136–148), IIa (residues 169–175), III (residues 224–243), and IIIa (residues 279–284) are colored violet, yellow, green, blue, and red, respectively (the same colors are used for motif labels in Fig. 5 ▶). The side chains of the five conserved active site residues (highlighted yellow in Fig. 5 ▶) are shown in ball and stick representation and are colored the same as the motif to which they belong. Ribbon cartoon shown at half normal width for clarity. Generated with PyMOL (DeLano Scientific).

The exact contribution of peripheral N-terminal S4 domains to the catalytic activity of Ψ synthases that modify rRNA has not yet been determined. Had the S4 domain of RluD been ordered in our structure, its exact function would still not be clear because it is at the end of a flexible linker and could adopt many conformations in the absence of bound substrate. Two (RluA and RluE) out of seven rRNA Ψ synthases in E. coli do not have an S4 domain, so it is not required for these enzymes to make Ψ. In fact, plasmid-borne RluC lacking its S4 domain is able to form all three of its target Ψ in a ΔrluC::kan mutant (S. Jean-Charles and J. Ofengand, unpubl. results). Whether this is true for RluD is currently being investigated.

The RluA family of Ψ synthases has traditionally been intriguing because of the multisite specificity of two of its members, RluC and RluD. There are many unanswered questions about how RluD is able to form Ψ1911, Ψ1915, and Ψ1917 in H69 of 23S RNA. Is there a sequential order to the modification or is it random? Does RluD bind H69 once and make all three Ψ or are there three separate binding events? In one study, RluD was capable of making Ψ1915 and Ψ1917, but not Ψ1911, on an in vitro transcript of 23S RNA, perhaps an indication that RluD can more easily access the uridines in the loop of H69 (Raychaudhuri et al. 1998). In other in vitro studies, RluD was capable of modifying all three sites plus additional sites in 23S RNA and even modifying some 16S RNA sites when the Mg²⁺ concentration was lowered (Huang et al. 1998a; Wrzesinski et al. 2000). Because higher order RNA structure depends on the presence of sufficient Mg²⁺, this may indicate that RluD can access certain uridine residues on partially unfolded RNA. This is consistent with our attempts at manually docking the one available structure of H69 (Fig. 1C ▶) into the RNA-binding cleft of RluD. Many initial orientations are possible, but for any of the three substrate uridines to get close to the pocket without any clashes, some of the RNA would have to be unfolded, presumably by RluD. In addition, to get each of three uridines into the active site pocket would require three unfolded conformations of H69, possibly three inherent conformations, as its structure is known to vary (Bashan et al. 2003). Indeed, recent cocrystal structures of RNA modifying enzymes with substrate RNAs illustrate various conformational rearrangements of the RNA in the bound state. For example, E. coli TruB flips three bases out of a 22mer T-loop RNA (Hoang and Ferré-D’Amaré 2001), prokaryotic tRNA guanine transglycosylase rearranges the loop of a 17mer RNA stem loop (Xie et al. 2003), and archaeal archaeosine tRNA guanine transglycosylase completely rearranges the D-arm of a bound tRNA (Ishitani et al. 2003).

It is likely that parts of the putative RNA-binding cleft of RluD also undergo a conformational change upon RNA binding. For instance, loop L9 (residues 192–198) is in a location equivalent to the thumb region of TruB that packs into the major groove of a bound T-loop RNA and is described as “closing” on the RNA (Hoang and Ferré-D’Amaré 2001). Because loop L9 has the highest main chain B-factors in the RluD structure (Fig. 3E ▶), we speculate that it may change conformation upon RNA binding and function in a similar fashion. Loop L3 (residues 105–112) also has high B-factors (Fig. 3E ▶) in the cleft and may move to accommodate RNA. The remainder of the cleft is involved in a crystal-packing interaction that may have altered the cleft in a nonphysiological way. The precise nature of the protein–RNA interactions between RluD and H69 will require cocrystal structures of RluD with bound substrate RNA.

MATERIALS AND METHODS

Protein expression, purification, crystallization, and data collection

SeMet-substituted RluD expression, purification, crystallization, and data collection were carried out as previously described (Del Campo et al. 2003). In brief, full-length RluD was overexpressed with a 23 amino acid N-terminal hexa-histidine tag. SeMet-substituted RluD was purified and crystallized using ethylene glycol as a precipitant by the hanging drop technique (Del Campo et al. 2003). These crystals are tetragonal (space group P4₃2₁2) with a single molecule of RluD in the asymmetric unit. A three-wavelength multiwavelength anomalous dispersion (MAD) data set was collected on a single flash-frozen crystal of SeMet RluD at the X12-C beamline, National Synchrotron Light Source (NSLS), Brookhaven National Laboratory (BNL).

Structure determination

Details of data collection and refinement are shown in Table 1 ▶. The diffraction data was processed using HKL (Otwinowksi and Minor 1997), and MAD phases were determined using SOLVE (Terwilliger and Berendzen 1999). Of the 14 expected selenium sites, SOLVE located nine using MAD data to 2.1 Å and generated an initial electron density map with an overall figure of merit of 0.56. None of the selenomethionines in the disordered N-terminal part of RluD could be located by SOLVE. The phases were improved to a figure of merit of 0.64 using density modification and solvent flattening in RESOLVE (Terwilliger 2000). The automated building feature in RESOLVE was used to position about 150 amino acids in the electron density, and the model was then extended by several rounds of manual building in O (Jones et al. 1991). Model refinement and phase combination between rounds of model building were done with CNS (Brünger et al. 1998).

TABLE 1.

Crystallographic data, phasing, and refinement statistics

Data collection and phasing
Space group	P43212
Unit cell dimensions
a = b (Å)	75.14
c (Å)	181.81
α = β = γ (°)	90
	Inflection	Peak	Remote
Wavelength (Å)	0.978795	0.978462	0.950037
Resolution (Å)	50.0–2.0	50.0–2.0	50.0–2.0
Completeness (%)a	95.1 (69.6)	88.9 (51.2)	84.7 (45.1)
Overall I/σ(I)	38.3 (9.2)	31.5 (5.8)	31.1 (4.3)
Rmerge (%)b	5.3 (13.3)	7.3 (18.2)	5.8 (18.0)
F.O.M.c	0.56
Refinement
Resolution range (Å)	40.0–2.0
Number of reflections	32009
R (%)d	21.9
Rfree (%)a	23.2
Cross-validated Luzatti error	0.26
R.m.s. deviations
Bond length (Å)	0.006
Bond angle (°)	1.3
Mean B-factor (Å2)	35.9
Number of protein atoms	2007
Number of solvent atoms	261

^aData in parentheses are for the highest-resolution shell (2.07– 2.00 Å).

^bR_merge = ∑(|I_j − 〈I〉|)/∑〈I〉, where I_j is the observed intensity of reflection j and 〈I〉 is the average intensity of multiple observations.

^cFigure of merit of the MAD solution in SOLVE (Terwilliger and Berendzen 1999).

^dR = ∑(|F_obs|−|F_calc|)/∑〈F_obs|, where F_obs and F_calc are the observed and calculated structure factor amplitudes, respectively.

^eR_free is the R value calculated for a test set of reflections, comprising a randomly selected 5% of the data not used during refinement.

Although a cross-validation set with 5% of the reflections was used for all manual model building and refinement, the initial automated build by RESOLVE included all reflections. This is likely the reason for the small spread between R and R_free (Table 1 ▶).

The N-terminal purification tag and 76 amino acids could not be modeled due to poor density. The final model contains residues 77–326, and 261 water molecules in the asymmetric unit, with 88.2% of the residues in the most favored region of the Ramachandran plot and no residues in the disallowed region. Model quality was monitored using PROCHECK (Laskowski et al. 1993) and the MOLPROBITY web tool (Lovell et al. 2003). A number of ordered water molecules fill the positively charged active site cavity, although some of them appear to adopt multiple positions based on positive F_o-F_c electron density maps. We attempted to model chloride and tris ions in a large blob of density in this region, but these gave poorer R factors and F_o-F_c maps; this region is modeled in our structure with a group of water molecules.

Coordinates

The atomic coordinates and structure factors have been deposited in the Protein Data Bank (accession code 1QYU).

ADDED IN PROOF

While this paper was in proof, a structure of TruB with and without substrate has appeared (Pan et al. 2003), revealing significant movements in TruB upon binding RNA that provide support for our prediction that RluD must undergo a conformational change to bind helix 69 of 23S rRNA. Also, a recent publication (Sivaraman et al. 2004) describes a structure of residues 75–326 of E. coli RluD that agrees well with the structure described here despite different crystal packing parameters.