X-Ray Crystallography Reveals a Large Conformational Change during Guanyl Transfer by mRNA Capping Enzymes

Kjell Håkansson¹, Aidan J. Doherty¹, Stewart Shuman², and Dale B. Wigley^1
1 Laboratory of Molecular Biophysics, University of Oxford, Oxford, OX1 3QU, United Kingdom
² Molecular Biology Program, Memorial Sloan-Kettering Institute, New York, New York 10021

Corresponding author: Dale B. Wigley, 01865 275381 (phone), 01865 275182 (fax), wigley@biop.ox.ac.uk.

We have solved the crystal structure of an mRNA capping enzyme at 2.5 Å resolution. The enzyme comprises two domains with a deep, but narrow, cleft between them. The two molecules in the crystallographic asymmetric unit adopt very different conformations; both contain a bound GTP, but one protein molecule is in an open conformation while the other is in a closed conformation. Only in the closed conformation is the enzyme able to bind manganese ions and undergo catalysis within the crystals to yield the covalent guanylated enzyme intermediate. These structures provide direct evidence for a mechanism that involves a significant conformational change in the enzyme during catalysis.

Eukaryotic mRNA is capped at the 5' end, by a 7-methyl-GMP moiety (see Shuman, 1995 for a review). This cap plays an important role in RNA synthesis and function and may also help to protect the mRNA from nucleolytic digestion. Since the capping of mRNA has been shown to be an essential activity in yeast (Shibagaki et al., 1992 ) and a number of eukaryotic viruses encode capping enzymes (Shuman, 1995 ), specific inhibition of this activity may have some potential in antiviral therapy.

The capping reaction involves three separate processes that are catalyzed by three different enzyme activities. The first step involves the removal of the 5'-ter-minal phosphate from the triphosphate tail by an RNA triphosphatase. The reaction is usually followed by guanylation and then methylation, although the order of these reactions can apparently be reversed in some viruses (Ahola and Kaariainen, 1995 ). The capping enzyme (GTP-RNA guanylyltransferase), which is responsible for the guanylation reaction, utilizes GTP in an unusual reaction involving a covalent enzyme-GMP adduct, from which the GMP moiety is transferred to the mRNA (Shuman and Hurwitz, 1981 ). Consequently, the GMP cap is attached to the mRNA via a 5'-5' triphosphate linkage. It has been shown that the GMP becomes attached to the enzyme via a lysine side chain that is part of a conserved active site motif (KxDG). Furthermore, this active site motif is one of a number of motifs that are conserved across a wide family of nucleotidyltransferases that includes ATP-dependent DNA ligases (Shuman et al., 1994 ). Extensive investigation using site-specific mutagenesis has revealed that the roles of these motifs are conserved in capping enzymes and ligases (Heaphy et al., 1987 ; Cong and Shuman, 1993 Cong and Shuman, 1995 ; Fresco and Buratowski, 1994 ; Schwer and Shuman, 1994 ; Shuman et al., 1994 ; Shuman and Ru, 1995 ), suggesting a common fold for this nucleotidyltransferase superfamily. Further evidence for this proposal came with the recent crystal structure determination of T7 DNA ligase, which revealed that all but one of these motifs are clustered around the ATP-binding site and indicated roles for residues within these motifs (Subramanya et al., 1996 ). On the other hand, there is little sequence homology between capping enzymes and ligases outside of these motifs. In order to investigate the structural similarity between capping enzymes and ligases, we have decided to solve the crystal structure of a capping enzyme from the Chlorella virus PBCV-1 (Li et al., 1995 ; Ho et al., 1996 ). This enzyme is the smallest capping enzyme identified to date, and preliminary examination of the recombinant enzyme, expressed in E. coli, has confirmed that it is active in transferring a GMP cap onto the end of mRNA that terminates with a 5'-diphosphate tail (Ho et al., 1996 ). Since the substrate specificities of DNA ligases and capping enzymes are very different, this structural information should help us also to understand the basis for substrate recognition in these enzymes.

Although the sequence of reactions that occurs during catalysis by capping enzymes is known, much less is known about the molecular details of these processes. Mutational analysis of capping enzymes and ligases has indicated likely roles for some residues (Heaphy et al., 1987 ; Cong and Shuman, 1993 Cong and Shuman, 1995 ; Fresco and Buratowski, 1994 ; Shuman et al., 1994 ; Shuman and Ru, 1995 ), which have been confirmed or clarified by the crystal structure of T7 DNA ligase (Subramanya et al., 1996 ). However, in spite of this information, much of the detail of the reaction mechanism of these nucleotidyltransferases remains unknown. To address this problem, we have determined the crystal structures of two forms of the GTP complex of the PBCV-1 capping enzyme and the GMP-enzyme adduct, all at 2.5 Å resolution.

Structure of the Protein
Details of the structure determination and refinement are presented in Table 1. Residues 11-327 of the 330 amino acids of the protein were defined in our final model. The enzyme comprises two domains with a cleft between them. Domain 1 consists of residues 11-236 and 319-327, while domain 2 comprises residues 238-317 (Figure 1). A large part of domain 1 has a topology similar to that of a group of ATP-binding proteins that includes DNA ligase, although the three-dimensional structure is rather different. Furthermore, the capping enzyme contains a 50 residue extension at the N terminus that is not conserved in ligases, which forms an additional two-stranded ß sheet and an helix. Domain 1 contains the GTP-binding site and the active site lysine to which the GMP becomes attached during catalysis. A series of conserved motifs, which have been proposed previously to indicate a conservation of structure across a wide family of nucleotidyltransferases that includes capping enzymes and ligases (Shuman et al., 1994 ), are located around the GTP-binding site in domain 1 and across the cleft between the two domains. These motifs are located in similar positions in the ligase structure (Subramanya et al., 1996 ). Domain 2 comprises a greek key motif followed by an additional ß strand and an helix. The last few residues of the protein (319-327) return across the cleft between the domains to form an amphipathic helix that packs against domain 1.

Figure 1. The Fold of PBCV-1 Capping Enzyme Stereo diagram illustrating the overall fold of the capping enzyme. Residue numbers are included to assist in following the path of the polypeptide chain. This figure was prepared using the program PREPPI. View larger version (38K): [in this window] [in new window]

The crystal structure of DNA ligase revealed that the enzyme comprised two domains with a cleft between them. Calculations of the electrostatic surface potential (Nicholls and Honig, 1991 ) suggest that the DNA-binding site is situated in this cleft, adjacent to the ATP-binding site. In addition to consideration of the structural data, evidence for the DNA-binding site in ligase resulted from limited proteolysis experiments that produced two fragments (Doherty et al., 1996 ). One of these fragments, which retained the cleft between the domains, was able not only to bind to DNA but also to compete effectively with the intact enzyme for ligation sites. The capping enzyme also comprises two domains with a cleft between them, although this cleft is much narrower in the capping enzyme than in DNA ligase. This difference may reflect the difference in polynucleotide substrate specificity of the two enzymes, in which the narrower cleft of the capping enzyme is suitable for single-stranded RNA, while the wider cleft in DNA ligase is required to provide access for the double-stranded DNA substrate. Domain 2 of the capping enzyme has a fold that differs significantly from that of domain 2 of DNA ligase, although both are derivatives of a greek key motif. The details of the differences between the structures of domain 2 in each case are also likely to be related to the substrate specificity of the two enzymes.

The positions of the base and the ribose of the bound GTP are similar to those of ATP bound at the active site of ligase and involve many similar contacts with residues from the conserved motifs. However, the specificity of the capping enzyme for GTP over ATP seems to be determined by residues outside of these motifs (Figure 2). The 6-oxo group of the guanine ring is involved in hydrogen bonds with the side chain of Lys-188 and van der Waals interactions with the side chain of Trp-190, while the 2-amino group is involved in a rather long hydrogen bond to the main chain carbonyl of Pro-59. None of these interactions would be possible with ATP. The specificity of DNA ligase for ATP over GTP was explained by interactions between the 6-amino group of the adenine ring and a glutamate residue (Glu-32), that is conserved in ligase sequences, and with the main chain carbonyl of Ile-33.

Figure 2. Structural Basis for Specificity of the Capping Enzyme for GTP Interactions between the guanine base of the bound GTP and the enzyme that give rise to the specificity of the enzyme for GTP rather than ATP. Van der Waals interactions are represented in green and hydrogen bonds in magenta. Distances are given in Ångströms after the residue labels. View larger version (28K): [in this window] [in new window]

Surprisingly, the two molecules of the capping enzyme in the crystallographic asymmetric unit adopt very different conformations (Figure 3). While one of the enzyme molecules has a deep, but narrow, cleft between the two domains, in the other molecule, this cleft is closed off as a result of a rigid body movement of domain 2. This movement results in a maximum displacement of 13 Å at the tip of the domain. This dramatic difference is important for the catalytic mechanism of the enzyme, and we shall refer to each of these conformations as "open" or "closed."

Figure 3. Conformational Changes in mRNA Capping Enzyme Diagram showing the difference in conformation between the two capping enzyme molecules in the crystallographic asymmetric unit. The open molecule is shown in red and the closed molecule is shown in white. The molecules were fitted by a least-squares procedure using C positions for residues in domain 1 (residues 11-236) in each case. This figure was prepared using the program RIBBONS (Carson, 1991 ). View larger version (56K): [in this window] [in new window]

GTP-Binding Sites of the Two Molecules
Although electron density for the bound nucleotide cofactor was evident at the active sites of both molecules, even the initial MIR, solvent-flattened electron density maps revealed that they were different. In light of these differences, we were careful to exclude these ligands from the refinement until the final stages, to prevent model bias. The difference electron density in the final stages of refinement revealed that the active sites of both enzyme molecules contained GTP (Figure 4). While the density for the bound GTP was strong for the closed molecule, it was poorer in the open molecule, with the density for the phosphate being particularly weak. The conformation of the triphosphate tail of each of the GTP molecules was different (Figure 4). We believe that this observation is important for a consideration of the enzyme mechanism (see below). Furthermore, the unexpected discovery of two conformational states of the GTP complex in the crystal provides direct evidence for a conformational change in the enzyme during nucleotide binding and catalysis.

Figure 4. The GTP-Binding Sites in Each Molecule Residues from domain 1 are labeled in yellow and those from domain 2 in white. (A) Electron density at the active site of the open molecule. Lys-82 and Lys-234 are the only residues in hydrogen-bonding contact with the triphosphate chain of the GTP. (B) Electron density at the active site of the closed molecule. In this complex, additional hydrogen-bonded interactions are formed between the GTP and residues from domain 2 as well as Arg-106, Arg-228, and Lys-236 from domain 1, which are not present in the open form. (C) Electron density at the active site of the closed molecule after treatment with Mn2+ ions. In this complex, the phosphate is interacting with Lys-234, Lys-236, and the manganese ion. In A-C, the difference electron density (Fo-Fc) was calculated at a stage in the refinement prior to inclusion of GTP in the model and is contoured at 3. (D) Comparison of the bound nucleotide in all three forms of the enzyme. Superposition was performed as described in Figure 3. View larger version (40K): [in this window] [in new window]

The GTP is bound in the active site of both enzyme molecules, with the guanosine base in a syn conformation buried in a hydrophobic pocket between Phe-146 (parallel stacking) and Ile-216. The ribose hydroxyls make hydrogen bonds with Glu-131 and Arg-87. O6 of the guanine base is hydrogen bonded to Lys-188 and makes a van der Waals contact with the edge of the aromatic ring of Trp-190. The triphosphate chain of the bound GTP makes very few contacts with the protein in the open form; the phosphate is hydrogen bonded to the active site lysine Lys-82 and to Lys-234 (Figure 4). Upon domain closure, the phosphate moves away from Lys-234 and hydrogen bonds to Lys-236. As a result, the distance between the active site lysine and the phosphorus decreases from 4.0 to 3.2 Å. What is perhaps more important is that the conformation of the entire triphosphate moiety becomes different from that observed in the open form, with the ß and phosphates folded away from the active site lysine. The amino group of the lysine and the leaving group are on opposite sides of the phosphorus, which is suitable for in-line attack of the phosphorus by the lysine. The ß and phosphates are within hydrogen bonding distance of not only Arg-106, Arg-228, and Lys-234 of the N-terminal domain, but also Asp-244, Arg-295, Lys-298, and Asn-302 of the C-terminal domain. All of the oxygens of the phosphate are bound to positively charged residues in what appears to be a strong anion-binding site.

Soaking of the crystals in harvest solution lacking GTP resulted in rapid diffusion of the GTP from the crystals, such that after two hours, the electron density for GTP could not be observed in either of the two molecules (data not shown). This confirms that the nucleotide moiety was not attached covalently in either molecule.

Formation of the GMP Adduct in the Crystals
The surprising discovery that there were two conformations of the protein in the crystals led us to investigate the biochemical reasons for these differences. A close examination of the electron density around the bound GTP molecules revealed that there was no candidate for a bound magnesium ion in either protein molecule, even though 5 mM magnesium chloride was present during the crystallization. We attribute this to the high salt concentration required for crystallization. To overcome this effect, we soaked the crystals in harvest solution containing a high concentration (100 mM) of manganese chloride for four hours. We chose the more electron dense manganese rather than magnesium to assist in the identification of the bound cation in the final difference electron density maps. Manganese has been shown to be as effective as magnesium as cofactor for the Chlorella capping enzyme (Ho et al., 1996 ).

After soaking, we examined the electron density of the active sites of each of the two molecules. The bound GTP in the open molecule retained the same conformation as observed in the original structure. Furthermore, there was no evidence of a bound manganese ion in this site. However, the electron density at the active site of the closed molecule had undergone significant changes (Figure 4). It was immediately evident that there was a large peak of difference electron density adjacent to the phosphate of the nucleotide that we interpret as a bound manganese ion. In addition, although the electron density for the ß and phosphates had disappeared, there was now continuous density between the phosphate and the active site lysine residue. There were also subtle changes in conformation of the protein side chains around the active site. All of the oxygens of the Lys-GMP phosphate are interacting with charged residues; one with Lys-234 and the other two with Lys-236 and the manganese ion, respectively. This latter interaction is the only close contact involving the manganese ion, although there are other interactions that appear to be mediated by water molecules. Electron density that we interpret as a sulphate ion (based upon the environment and the height of the density peak) is now found in the position occupied previously by the phosphate. These observations are consistent with turnover of the enzyme to cleave the bound GTP and produce the covalent guanylated enzyme intermediate. Since this change had only taken place in the closed conformation of the enzyme, we conclude that only this form is catalytically competent. We presume that this is due to an inability of the open form of the enzyme to bind the divalent cation that is required for cleavage of the GTP and to the differences in the conformation of the triphosphate moiety. The direct consequence of this observation is that upon binding of GTP, the enzyme undergoes a conformational change to produce the active form of the enzyme, which is then able to bind manganese (or magnesium) ions and promote guanylation.

Mechanistic Implications
Examination of the structures of the two different conformations of the capping enzyme reveals that while only the open form seems wide enough to accomodate the RNA substrate, only the closed form could be guanylated in the presence of excess GTP and Mn²⁺ ions. Therefore, since both reactions are catalyzed by the enzyme, both forms appear to be required in order to complete GMP transfer from GTP to RNA. The conformational differences between the ligands in the open and closed forms show how GTP is positioned optimally for reaction with Lys-82 only in the closed form of the protein. A similar electrophilic activation of the phosphate of ATP in the active site of tyrosyl tRNA synthetase has been suggested from model building (Leatherbarrow et al., 1985 ). In the case of the capping enzyme, the nucleoside moiety is fixed in a pocket on domain 1, and residues from both domains are responsible for activation of the nucleotide through interaction with the phosphate chain. Strong interactions with the ß and phosphates also facilitate cleavage of the bond to the leaving group as the phosphate approaches the active site lysine. While this conformational activation requires domain closure and the assistance of domain 2, it is plausible that a similar phosphate conformation could be achieved in the analogous capping reaction through interaction between RNA and enzyme.

What other evidence is there for this large conformational change? One obvious consequence of the domain closure is that a number of residues in the C-terminal domain are brought into the proximity of the GTP-binding site. Mutational analysis of an aspartate residue (Asp-257) in the Saccharomyces cerevisiae capping enzyme (Shuman et al., 1994 ) has shown that this residue is required for enzyme activity in vivo. Mutation of the equivalent residue (Asp-400) in vaccinia virus capping enzyme resulted in an enzyme that was impaired in its ability to form the GMP adduct in vitro (Cong and Shuman, 1995 ). These would be surprising results if one were only to consider the open form of the enzyme. However, the structure of the closed form of the capping enzyme complexed with GTP indicates a role for this residue as a result of the domain closure. The aspartate residue that is equivalent (Asp-244 in PBCV-1 capping enzyme) moves from a position remote from the active site, in the open conformation, to be close to the ß phosphate of the GTP in the closed conformation. Furthermore, the closed form of the capping enzyme reveals a role for the conserved sequence motif VI, which has been shown to be important for activity (Shuman, 1996 ). Two residues from this motif (Arg-295 and K-298) contact the triphosphate tail of the GTP in the closed GTP complex.

The structures presented here suggest that the magnesium-binding site can only be formed after domain closure. If a magnesium ion were bound at this site, then conversion of GTP to enzyme-GMP adduct would have taken place. Even though magnesium ions were present in the crystallization medium, the high salt concentration appears to have prevented the binding of magnesium ions to the enzyme-GTP complex, allowing us to observe the GTP-enzyme complex trapped in the crystal prior to cleavage. Once the complex is challenged with a high enough concentration of manganese ions, the complex is able to undergo catalysis. Comparison of the molecular surface of the open and closed forms of the enzyme (Figure 5) reveals that after domain closure, most of the GTP-binding site is blocked off from the solvent, with the exception of an opening that could provide access for the incoming cation. Hence, the closing of the protein both ensures that the GTP binds in a conformation that is ready for nucleophilic attack by Lys-82 and creates the binding site for the incoming cation. However, after formation of the GMP adduct, the enzyme must open up to provide access for the incoming mRNA substrate, since this site is blocked off in the closed form of the enzyme.

Figure 5. Electrostatic Surface Potential of the Open and Closed Forms Regions of positive potential are shown in blue and negative potential in red. The GTP is shown in stick representation. This figure was generated using the program GRASP (Nicholls and Honig, 1991 ). (A) Open molecule from the side. (B) Open molecule from above. (C) Closed molecule from the side. (D) Closed molecule from above. The inset shows a close-up of the region around the phosphate group of the GTP (shown in red and yellow). View larger version (64K): [in this window] [in new window]

There is some evidence that other members of this nucleotidyltransferase family, such as DNA ligase, undergo similar conformational changes during catalysis. The structural similarity between the capping enzymes and ATP-dependent DNA ligases was predicted from sequence alignments (Shuman et al., 1994 ; Shuman and Schwer, 1995 ) and supported by data from mutant enzymes (Heaphy et al., 1987 ; Cong and Shuman, 1993 Cong and Shuman, 1995 ; Fresco and Buratowski, 1994 ; Shuman et al., 1994 ; Shuman and Ru, 1995 ). We have shown that the structures of DNA ligase (Subramanya et al., 1996 ) and this mRNA capping enzyme are indeed similar, particularly for residues in the region of the NTP-binding site. It is likely, therefore, that the two enzymes will also share aspects of their mechanisms. The conservation of a series of motifs that line the nucleotide-binding sites and also cross the cleft at the domain interface was explained in large part by the ligase crystal structure (Subramanya et al., 1996 ). More recent work (A. J. D. and D. B. W., unpublished data) has revealed that although the entire ATP-binding site in DNA ligase is apparently contained within domain 1, and although domain 1 is able to catalyze the adenylation reaction, the latter reaction is stimulated strongly by domain 2. Furthermore, the two domains comigrate partially on gel filtration, indicating a strong, but transient, interaction between them that would not be predicted from the crystal structure. A simple explanation for these observations is that there is an interaction between these domains that promotes adenylation. This interaction is not evident in the crystal structure of the apo enzyme or in the complex with ATP produced by soaking the crystals in ATP and magnesium. We noted that although the conditions used to prepare this complex should have resulted in adenylation of the enzyme, there was no evidence of covalent attachment to the active site lysine or of a bound magnesium ion (Subramanya et al., 1996 ). Furthermore, prolonged soaking (several days) eventually resulted in cracking of the crystals. These observations are consistent with adenylation requiring a conformational change that is prevented by molecular contacts in the crystal. It has been shown also that T7 ligase has increased resistance to proteolytic digestion in the presence of ATP (Doherty et al., 1996 ), also suggestive of a conformational change. The structures presented here provide clear evidence for that conformational change in the capping enzyme, which we predict to be similar for DNA ligases and other related nucleotidyltransferases.

It is likely that the structures of cellular RNA guanylyltransferases will resemble that of the PBCV-1 enzyme. An alignment of the sequence of the PBCV-1 protein with the sequences of the capping enzymes encoded by S. cerevisiae (Shibagaki et al., 1992 ), Schizosaccharomyces pombe (Shuman et al., 1994 ), and Candida albicans (Yamada-Okabe et al., 1996 ) reveals conservation of a central core region of approximately 270 residues (residues 56-325 in the Chlorella virus enzyme; S. S., unpublished data). In the structures described above, almost every amino acid side chain of the Chlorella virus capping enzyme that makes contacts with the phosphates and the sugar of GTP is conserved in the cellular counterparts. These include key residues within the six motifs that define this nucleotidyltransferase superfamily, as well as residues outside of these motifs (e.g., Arg-106 and Asn-302). Residue Lys-188, which contacts the bound GTP, is conserved in the S. cerevisiae and C. albicans proteins.

We can now propose a general mechanism for phosphoryl transfer in this family of nucleotidyltransferases (Figure 6 and Figure 7), based upon four different crystal structures. In the first step of the reaction, GTP (or ATP in the case of ligase) binds to the open form of the enzyme. The binding of GTP promotes closure of the domains. This closure is stabilized by interactions between the bound nucleotide and residues on domain 2. The domain closure is followed by phosphoryl transfer from GTP to produce the GMP adduct of the enzyme. The cleavage of GTP breaks the linkage between residues in domain 2 and the nucleotide and thus destabilizes the closed form of the enzyme, which is then able to open up again to release pyrophosphate and expose the RNA-binding site, allowing RNA to bind to the enzyme ready for guanylation.

Figure 6. Proposed Mechanism of Nucleotidyl Transfer by Capping Enzymes and Ligases (A) In this scheme, molecules in the open conformation are shown in red, while those in the closed conformation are depicted in white. The side chain of the active site lysine is shown in yellow, and the nucleotide moiety (and anything to which it is attached) is shown in green. (a) apo enzyme; (b) enzyme-GTP complex, open conformation; (c) enzyme-GTP complex, closed conformation; (d) GMP adduct, closed conformation; (e) GMP adduct, open conformation; (f) GMP adduct-mRNA complex (mRNA shown in pale blue); and (g) enzyme-capped mRNA complex. Structures of (a)-(d) are presented in this paper, while those of (e)-(g) are modeled from these data. View larger version (44K): [in this window] [in new window]

Figure 7. A Schematic Diagram for Nucleotidyl Transfer by Capping Enzymes and Ligases The reaction pathway is the same as in Figure 6, but the roles of some of the most important residues for catalysis are shown. View larger version (38K): [in this window] [in new window]

The structures of the capping enzyme presented here confirm the similarity between the NTP-binding domains of mRNA-capping enzymes and ATP-dependent DNA ligases, indicating a conservation of function for the conserved motifs in both structures. The difference in conformation observed for the two GTP complexes provides direct evidence for a significant conformational change during nucleotidyl transfer in the capping enzymes, which is likely to be similar in other members of this enzyme family. However, it is evident that this is not the whole story and that there are still many unanswered questions about nucleotidyl transfer from the enzymes to their RNA or DNA substrates. Additional structural information, particularly about a complex of a capping enzyme or a ligase complexed with its nucleic acid substrate, should shed further light on this step in the enzyme mechanism.

Protein was prepared and crystallized as described previously (Doherty et al., 1996 ). The crystals belong to the space group C222₁, with unit cell dimensions a = 93.3 Å, b = 214.9 Å, c = 105.8 Å. All data were collected from crystals that were flash frozen in rayon loops at 100 K. X-ray data were collected on a MarResearch image plate detector, either at the Synchrotron Radiation Source (Daresbury) or using a rotating anode source. Integrated intensities were calculated with the program DENZO (Otwinowski, 1993 ). The mercury positions in the thimerosal data set were determined by manual inspection of difference Patterson syntheses and confirmed using the program SHELX (Sheldrick, 1992 ). The heavy atom sites in the selenomethionine data set were determined by difference Fourier calculations. It was evident that there were two sets of heavy atom positions, each corresponding to a single molecule. A 2.7 Å map was calculated using the MIR phases. The CCP4 program suite (Collaborative Computing Project No., 4, 1994 ) was used unless stated otherwise. The initial map was solvent flattened and an initial model was built into the solvent-flattened map using the graphics program "O" (Jones et al., 1991 ). Preliminary attempts to improve the MIR maps by averaging the density for the two noncrystallographically related molecules resulted in an improvement in the density of domain 1 but did not improve the density of domain 2. Inspection of the electron density revealed significant differences between the two molecules, corresponding to a rigid body movement of the smaller domain 2 with respect to domain 1. The positions of the domains were refined using rigid body refinement of a partial model in XPLOR (Brünger et al., 1989 ). The phases were improved further by averaging two masks independently, one corresponding to each domain, followed by solvent flattening in DM. The model was initially refined using positional refinement in XPLOR with some restraint of the noncrystallographic symmetry, followed by simulated annealing refinement, during which the restraints were released in light of the differences between the molecules. The refinement was completed using positional refinement in XPLOR without symmetry restraints, but with restrained temperature factors. Manual rebuilding was carried out between the refinement cycles. The crystallographic free R (Brünger, 1992 ) was monitored at each stage to prevent model bias. The nucleotide cofactors were omitted until the final stages of the refinement, also to prevent model bias. Statistics on the final model are presented in Table 1. When domains 1 of each molecule in the final model are superimposed (as in Figure 3), the rms difference in C positions is 0.44 Å for domain 1 but 8.5 Å for domain 2. However, when domains 2 alone from each molecule are superimposed, the rms deviation for C positions is only 1.4 Å. These data indicate that the domain movement is largely a consequence of a rigid body rotation between the two conformations of the protein rather than an alteration of the secondary structure.

The GMP adduct was prepared by harvesting crystals as described previously (Doherty et al., 1996 ) and then soaking them in harvest solution containing 100 mM manganese chloride and 5 mM GTP for 4 hr. Crystals were then flash frozen for data collection as described above. Refinement of the protein was carried out in a similar manner to that used for the GTP complex, using the coordinates for the protein part of the GTP complex as our initial model. The same subset of data was used for the calculation of free R factor, to prevent bias toward our initial model. The final stages of the refinement allowed the positioning of the solvent molecules and the nucleotide moieties. Statistics on the final model are presented in Table 1.

Correspondence regarding this paper should be addressed to D. B. W. We wish to thank Synchrotron Radiation Source (Daresbury) for access to synchrotron radiation facilities. We also thank H. Subramanya for advice on density averaging, S. Islam and M. Sternberg for providing their graphics program PREPPI, K. Ho for preparing the capping enzyme overexpressing cell line, and D. Barford for helpful discussions. This project was funded by the Wellcome Trust and the Medical Research Council.

Received February 14, 1997; revised March 26, 1997.

Ahola, T., and Kaariainen, L. (1995). Reaction in alphavirus mRNA capping: formation of a covalent complex of nonstructural protein nsP1 with 7-methyl-GMP. Proc. Natl. Acad. Sci. USA 92, 507-511.[Medline]

Brünger, A.T. (1992). Free R value: a novel statistical quantity for assessing the accuracy of crystal structures. Nature 355, 472-474.

Brünger, A.T., Karplus, M., and Petsko, G.A. (1989). Crystallographic refinement by simulated annealing: application to crambin. Acta Cryst. A45, 50-61.

Carson, M. (1991). Ribbons 2.0. J. Appl. Cryst. 24, 958-961.

Collaborative Computing Project No. 4 (1994). The CCP4 suite: programs for protein crystallography. Acta Cryst. D50, 760-763.

Cong, P., and Shuman, S. (1993). Covalent catalysis in nucleotidyl transfer. A KTDG motif essential for enzyme-GMP complex formation by mRNA capping enzyme is conserved at the active sites of RNA and DNA ligases. J. Biol. Chem. 268, 7256-7260.[Medline]

Cong, P., and Shuman, S. (1995). Mutational analysis of messenger-RNA capping enzyme identifies amino acids involved in GTP-binding, enzyme-guanylate formation and GMP transfer to DNA. Mol. Cell. Biol. 15, 6222-6231.[Medline]

Doherty, A.J., Ashford, S.R., Subramanya, H.S., and Wigley, D.B. (1996). Characterisation of proteolytic fragments of bacteriophage T7 DNA ligase. Nucl. Acids Res. 24, 2281-2287.

Doherty, A.J., Håkansson, H., Ho, C.K., Shuman, S., and Wigley, D.B. (1997). Crystallisation of the RNA guanyltransferase of Chlorella virus PBCV-1. Acta Cryst. D, in press.,.

Fresco, L.D., and Buratowski, S. (1994). Active site of the mRNA-capping enzyme guanylyltransferase from Saccharomyces cerevisiae: similarity to the nucleotidyl attachment motif of DNA and RNA ligases. Proc. Natl. Acad. Sci. USA 91, 6624-6628.[Medline]

Heaphy, S., Singh, M., and Gait, M.J. (1987). Effect of single amino acid changes in the region of the adenylation site of T4 RNA ligase. Biochemistry 26, 1688-1696.[Medline]

Ho, C.K., van Etten, J.L., and Shuman, S. (1996). Expression and characterization of an RNA capping enzyme encoded by Chlorella virus PBCV-1. J. Virol. 70, 6658-6664.[Medline]

Jones, T.A., Zou, J-Y., Cowan, S.W., and Kjeldgaard, M. (1991). Improved methods for building protein models in electron density maps and the location of errors in these models. Acta Cryst. A47, 110-119.

Laskowski, R.A., MacArthur, M.W., Moss, D.S., and Thornton, J.M. (1993). PROCHECK: a program to check the stereochemical quality of protein structures. J. Appl. Cryst. 26, 283-291.

Leatherbarrow, R.J., Fersht, A.R., and Winter, G. (1985). Transition state stabilisation in the mechanism of tyrosyl transfer RNA synthetase revealed by protein engineering. Proc. Natl. Acad. Sci. USA 82, 7840-7844.[Medline]

Li, Y., Lu, Z.Q., Burbank, P.E., Kutish, G.F., Rock, D.L., and Vanetten, J.L. (1995). Analysis of 43 KB of the Chlorella virus PBCV-1 330KB genomeMap positions 45-88. Virology 212, 134-150.[Medline]

Nicholls, A., and Honig, B.J. (1991). A rapid finite-difference algorithm, utilizing successive over-relaxation to solve the Poisson-Boltzmann equation. J. Comput. Chem. 12, 435-445.

Otwinowski, Z. (1993). Oscillation data reduction program. In Data Collection and Processing, S.E.R.C. Daresbury DL/SC1/R34, pp. 56-62.

Schwer, B., and Shuman, S. (1994). Mutational analysis of yeast mRNA capping enzyme. Proc. Natl. Acad. Sci. USA 91, 4328-4332.[Medline]

Sheldrick, G. (1992). In Crystallographic Computing 5, D. Moras, A.D. Podjarny, and J.C. Thierry, eds. (Oxford, U.K.: Oxford University Press), pp. 145-157.

Shibagaki, Y., Itoh, N., Yamada, H., Nagata, S., and Mizumoto, K. (1992). mRNA capping enzyme: isolation and characterization of the gene encoding mRNA guanylyltransferase subunit from Saccharomyces cerevisiae. J. Biol. Chem. 267, 9521-9528.[Medline]

Shuman, S. (1995). Capping enzyme in eukaryotic mRNA synthesis. Prog. Nucl. Acids Res. Mol. Biol. 50, 101-129.

Shuman, S. (1996). Closing the gap on DNA ligase. Structure 4, 653-656.[Medline]

Shuman, S., and Hurwitz, J. (1981). Mechanism of mRNA capping by vaccinia virus guanylyltransferase: characterization of an enzyme-guanylate intermediate. Proc. Natl. Acad. Sci. USA 78, 187-191.[Medline]

Shuman, S., Liu, Y., and Schwer, B. (1994). Covalent catalysis in nucleotidyl transfer reactions: essential motifs in Saccharomyces cerevisiae RNA capping enzyme are conserved in Schizosaccharomyces pombe and viral capping enzymes and among polynucleotide ligases. Proc. Natl. Acad. Sci. USA 91, 12046-12050.[Medline]

Shuman, S., and Ru, X.-M. (1995). Mutational analysis of vaccinia DNA ligase defines residues essential for covalent catalysis. Virology 211, 73-83.[Medline]

Shuman, S., and Schwer, B. (1995). RNA capping enzyme and DNA ligase: a superfamily of covalent nucleotide transferases. Mol. Microbiol. 17, 405-410.[Medline]

Subramanya, H.S., Doherty, A.J., Ashford, S.R., and Wigley, D.B. (1996). Crystal structure of an ATP-dependent DNA ligase from bacteriophage T7. Cell 85, 607-614.[Summary/Full Text]

Yamada-Okabe, T., Shimmi, O., Doi, R., Mizumoto, K., Arisawa, M., and Yamada-Okabe, H. (1996). Isolation of the mRNA-capping enzyme and ferric-reductase-related genes from Candida albicans. Microbiology 141, 2515-2523.