A Structural Basis for the Inhibition of the NS5 Dengue Virus mRNA 2′-O-Methyltransferase Domain by Ribavirin 5′-Triphosphate*

Ribavirin is one of the few nucleoside analogues currently used in the clinic to treat RNA virus infections, but its mechanism of action remains poorly understood at the molecular level. Here, we show that ribavirin 5′-triphosphate inhibits the activity of the dengue virus 2′-O-methyltransferase NS5 domain (NS5MTaseDV). Along with several other guanosine 5′-triphosphate analogues such as acyclovir, 5-ethynyl-1-β-d-ribofuranosylimidazole-4-carboxamide (EICAR), and a series of ribose-modified ribavirin analogues, ribavirin 5′-triphosphate competes with GTP to bind to NS5MTaseDV. A structural view of the binding of ribavirin 5′-triphosphate to this enzyme was obtained by determining the crystal structure of a ternary complex consisting of NS5MTaseDV, ribavirin 5′-triphosphate, and S-adenosyl-l-homocysteine at a resolution of 2.6 Å. These detailed atomic interactions provide the first structural insights into the inhibition of a viral enzyme by ribavirin 5′-triphosphate, as well as the basis for rational drug design of antiviral agents with improved specificity against the emerging flaviviruses.

The guanosine analogue ribavirin is a broad spectrum antiviral agent discovered almost 30 years ago (1). Since its discovery, many mechanisms of action have been proposed (reviewed in Refs. 2 and 3). Like most nucleoside analogues, ribavirin is phosphorylated by cellular kinases at its 5Ј-position upon entry into the cell. Ribavirin 5Ј-monophosphate is a potent inhibitor of the cellular enzyme inosine 5Ј-monophosphate dehydrogenase. This inhibition results in the depletion of the intracellular guanosine nucleotide pool, which feeds capping and polymerase enzymes from both viral and cellular origin. Consequently, the depressed guanosine nucleotide pool may exert an indirect antiviral effect, since viral enzymes would not compete advantageously for guanosine nucleotides with cellular enzymes. In addition, ribavirin nucleotides may have a viral target, such as RNA polymerization and RNA capping or induce lethal mu-tagenesis of viral genomes (4), accounting for the observed antiviral effect. Both direct and indirect mechanisms may thus contribute to the ribavirin mode of action. To date, ribavirin nucleotides have been crystallized with two cellular enzymes, namely inosine 5Ј-monophosphate dehydrogenase (5) and nucleoside diphosphate kinase (NDPK (6)) but not with any viral enzyme or protein.
The genus Flavivirus comprises important human pathogens such as West Nile, dengue, and yellow fever viruses, which are moderately sensitive to ribavirin (7)(8)(9). These mosquito-borne viruses are currently expanding their distribution throughout the world. The introduction of West Nile virus in North America may be an important milestone in the history of this virus, as exemplified by outbreaks in the New York area (10) followed by the gradual spread to 47 of the 49 continental states of the United States of America (www.cdc.gov/ncidod/dvbid/westnile/ index.htm). The Camargue area in France has re-witnessed West Nile viral infection of horses after 40 years, and the first human cases were reported in the French Riviera in October 2003. Likewise, dengue virus, an agent responsible for hemorrhagic fever, infects more than 50 million persons annually with an increasing incidence in tropical areas around the world.
The single-stranded RNA genome of flaviviruses is of positive polarity, and is capped with a cap 1 structure Me7 GpppA 2ЈOMe (11). The N-terminal domain of the dengue virus polymerase NS5 is a 2Ј-O-methyltransferase that is active on RNA cap structures (12). This enzyme, referred to as NS5MTase DV , is able to bind a GTP molecule that may mimic the RNA cap structure prior to methylation. Thus, it was of interest to determine whether guanosine analogues could bind to the GTP binding site of NS5MTase DV . If so, guanosine analogues may act as potential competitive inhibitors of RNA cap binding and aid in the rational design of inhibitors directed against flaviviruses.
In this report, we present biochemical evidence that ribavirin 5Ј-triphosphate (RTP) 1 inhibits the 2Ј-O-methyltransferase activity of dengue virus NS5MTase DV. We also show that a series of RTP analogues, as well as EICAR 5Ј-triphosphate (EICAR-TP) and acyclovir 5Ј-triphosphate (acyclovir-TP), compete with GTP for binding to NS5MTase Dv . In addition to show that RTP and GTP share a common binding site on NS5MTase DV , the crystal structure of the RTP-NS5MTase DV complex reveals a unique mode of binding for the ribavirin pseudobase that is consistent with a lack of discrimination of this antiviral molecule relative to GTP.

EXPERIMENTAL PROCEDURES
Enzymes and Reagents-The purification and crystallization of the NS5 capping domain of the dengue virus RNA-dependent RNA polymerase has been described (12). Synthesis and purification of RTP, ribavirin nucleotide analogues, and acyclovir 5Ј-TP to homogeneity, as determined by 1 H, 13 C, and 31 P NMR and HPLC, have been described (6). EICAR 5Ј-TP was a kind gift from P. Herdewijn (Leuven, Belgium). All other nonradioactive and 32 P-labeled nucleotides were of HPLC grade and purchased from Amersham Biosciences.
Inhibition of RNA 2Ј-O-Methyltransferase Activity by RTP-The methyltransferase assay was performed in 40 mM Tris-HCl, pH 7.1. 100 M S-adenosylmethionine with 10 Ci of Ado[methyl-3 H]Met (74 Ci/ mmol; Amersham Biosciences) were incubated with 2 g of purified NS5MTase DV , 35 l of RNA substrate, and various concentrations of ribavirin 5Ј-triphosphate (50, 100, 250, 300, 500 and 750 M) in a 50-l reaction mix (12). The RNA substrate mix was produced using purified T7 DNA primase in conjunction with the synthetic DNA oligonucleotide T 10 CTG 5 (10 M), 300 M CTP, and 200 M cap analogue to obtain capped AC 5 (GpppAC 5 and m7 GpppAC 5 ) as described (13). The methyltransfer reaction was incubated at 30°C and monitored from 30 min to 3 h. 8-l aliquots were spotted onto DEAE-81 filter paper (Whatman) and washed with 20 mM sodium formate, pH 8.0 to remove any remaining S-adenosylmethionine. The experiment without ribavirin 5Јtriphosphate served as a reference. Radioactively labeled RNA was quantified by liquid scintillation counting.
Determination of RTP Dissociation Constant for NS5MTase DV -The dissociation constant, K d , was determined using 58 M [ 32 P]GTP and increasing concentrations of RTP (0, 100, 200, 500, 800, and 1000 M) incubated with 2 g of NS5MTase DV . The bound radioactivity was quantitated after SDS-PAGE using photostimulatable plates and a FujiImager. Data were fit to a hyperbolic function from which the K d was determined.
GTP-binding Inhibition by RTP and Its Analogues-Two g of NS5MTase DV was incubated with RTP or RTP analogues and with [ 32 P]GTP (50 M, 10% volume of [ 32 P]GTP) in 50 mM Tris, pH 7.6, 5 mM dithiothreitol, and 5 mM MgCl 2 . The reaction mixture was irradiated with UV light for 3 min using a UV lamp (40 watts, at 254 nm) at a distance of 12 mm, boiled for 5 min, and subjected to denaturing gel electrophoresis. The gel was stained using Coomassie Blue dye, and radioactive products were visualized using photostimulatable plates and a FujiImager.
Structure Determination and Refinement-Crystals of NS5MTase DV were grown by vapor diffusion as described (12) and further soaked for 3 h in 2 mM RTP. A 2.6-Å data set was collected at 100 K using a charge-coupled device detector (ADSC Q4) at ID14-2 beamline, ESRF, Grenoble, France. Data were processed using DENZO (14), and intensities were merged with SCALA (15). Since the unit cell dimensions of the crystal were slightly different from those of the apo-form, the structure was solved by the molecular replacement program AmoRe, whereas polypeptide chain of the apo-form served as a model (PDB code 1L9K). A random set comprising 5% of the data was omitted from refinement for R free calculation (16). A round of simulated annealing refinement, as implemented using crystallography CNS software, was performed (17). As in the case of the apo-form, no density was observed for the first 6 and the last 30 residues. Models for one molecule of S-adenosyl-L-homocysteine (SAHC) and one molecule of ribavirin 5Јmonophosphate were built into the F o Ϫ F c and 2F o Ϫ F c SIGMAAweighted electron density. The model was further built and refined through alternating cycles using the programs TURBO-FRODO (18) and CNS (crystallography NMR software), respectively. During this process seven sulfate ions, which were already present in the native structure, and 66 additional water molecules were refined. Refinement statistics are given in Table I. Atomic coordinates have been deposited in the Protein Data Bank under code 1R6A.

Inhibition of the RNA 2Ј-O-Methyltransferase Activity of Dengue Virus NS5MTase DV Domain-The
NS5MTase DV domain is able to bind GTP (12). In addition, the crystal structure of a ternary complex consisting of NS5MTase DV , SAHC, and GTP has been described previously (12). In this complex, GTP mim-ics an RNA cap. Since NS5MTase DV acts as an RNA 2Ј-Omethyltransferase (12), we tested whether RTP could have a direct effect on this activity. Since ribavirin is a guanosine analogue and a weak inhibitor of flavivirus growth (7-9, 19, 20), the NS5MTase DV domain may be a target for RTP. In the RNA 2Ј-O-methyltransferase assay, a short capped RNA is incubated with NS5MTase DV , the radiolabeled methyl donor S-adenosyl-L-methionine, and various concentrations of RTP. The methylated RNA product is monitored over time using a filter paper binding assay (12). Fig. 1A shows the results of such an inhibition assay. RTP inhibits the RNA 2Ј-O-methyltransferase activity. Under the experimental conditions tested here, the inhibitory concentration required to yield a 50% inhibition (IC 50 ) is 101 Ϯ 20 M, leading to the conclusion that RTP is indeed an inhibitor of the RNA 2Ј-O-methyltransferase activity contained in the NS5MTase DV domain.
Mapping of the Guanosine Analogue Binding Site to the Dengue Virus NS5MTase DV Domain-The crystal structure of NS5MTase DV (12) indicates that this domain has two nucleoside/nucleotide binding pockets (i.e. one for SAHC and the other for GTP), which could account for the binding of RTP and subsequent inhibition of the RNA 2Ј-O-methyltransferase activity. Since RTP is a GTP analogue, a GTP-binding competition experiment was used to map the RTP binding site. When the protein is incubated with a fixed concentration of radiolabeled GTP and increasing concentrations of RTP, GTP binding is decreased (Fig. 1B). This indicates that RTP displaces GTP from NS5MTase DV . When the relative inhibition is plotted against RTP concentration, an apparent K d value of 55 Ϯ 9 M is calculated. RTP exhibits a NS5MTase DV binding affinity similar to that of GTP (K d ϭ 58 Ϯ 14 M (12)). We then tested the binding of other analogues to NS5MTase DV . Acyclovir 5Јtriphosphate is a guanosine analogue with an acyclic backbone replacing the ribose, and EICAR 5Ј-triphosphate is a 5-deaza 5-ethynyl ribavirin 5Ј-triphosphate. Both analogues bound to NS5MTase DV with K d values of 80 Ϯ 7 and 165 Ϯ 7 M, respectively (Fig. 1C). Finding that RTP has the best affinity of the three guanosine analogues and that the acyclic moiety of acyclovir 5Ј-TP is seemingly less important for binding than a modified base such as that of EICAR prompted us to assay other RTP analogues modified on their ribose moiety (6). A  series of preliminary experiments was performed to determine whether any of these analogues would bind better to NS5MTase DV than RTP. Fig. 2 shows the result of such experiments in which 2Ј-deoxy-, 2Ј,3Ј-dideoxy-, 3Ј-deoxy-, 2Ј,3Ј-epoxy-, and 2Ј3Ј-dideoxy-2Ј3Ј-didehydro-RTP were used to compete for GTP binding. None of these analogues exhibits a better affinity for NS5MTase DV than RTP. 3Ј-deoxy-RTP binds almost as efficiently as RTP, whereas 2Ј3Ј-dideoxy-RTP binds with the lowest relative affinity. We conclude that RTP and GTP compete for the NS5MTase DV binding site, and the structural basis for this competition was further investigated using x-ray crystallography.
Structural Basis for the Binding of RTP to the Dengue Virus NS5MTase DV Domain and for Its Inhibition-A crystal of NS5MTase DV was soaked for 3 h in 2 mM RTP under the conditions described previously (12). The structure was solved using molecular replacement techniques and revealed that (i) an RTP molecule was bound to the enzyme, and (ii) the RTP molecule was located at the previously identified GTP binding site. As for the GTP analogue GDPMP, the ␤and ␥-phosphates of RTP were present but not well defined in the electron density maps and were therefore not modeled. The ␣-phosphate and the ribose moiety of RTP make contact with NS5MTase DV in a manner similar to the GTP analogue (Fig. 3, A and B). However, interesting features are revealed when comparing the respective nucleo-or pseudobase orientation and the positioning of both compounds. The structural resemblance of ribavirin with guanosine originates from the spatial position of both the 1-and 6-position (Fig. 3, C and D). However in the NS5MTase DV -GTP analogue complex, neither of these positions interacts with the protein, and the NH 2 group located at the 2-position makes the only specific interactions of the nucleobase with the protein (12). The complex formation of NS5MTase DV with either the GTP analogue or RTP does not result in any conformational change of the protein. When superimposing the protein atoms of both complexes, it becomes obvious that ribavirin mimics the interaction established by the NH 2 group of the 2-position of guanine by a 30°rotation around an axis perpendicular to the nucleobase plane and passing through the C-1 position of the ribose (Fig. 3E). As a result, the NH 2 group of ribavirin is nearly superimposed (0.38 Å away) with the NH 2 group of the guanine and is therefore recognized in a similar fashion; the carbonyl group of RTP does not interact with any residue, and the NH 2 group forms a hydrogen bond to the same carbonyl groups of Leu-17, Asn-18, and Leu-20 as the 2-amino group of GDPMP. Thus, RTP makes the same atomic contacts as GTP, and this result is in agreement with the finding that RTP and GTP exhibit similar affinity constants (Fig. 1).
We conclude that RTP is a moderate inhibitor of the RNA 2Ј-O-methyltransferase activity carried by the NS5MTase DV domain. In light of our mapping and structural data, RTP acts as a GTP or RNA cap analogue, binds to the RNA cap binding site of the NS5MTase DV domain, and impedes RNA cap 2Ј-O-methylation. DISCUSSION In this paper, we have show that RTP inhibits the RNA 2Ј-O-methyltransferase activity of dengue virus NS5MTase DV domain. We further mapped the binding site of RTP binding to the GTP/RNA cap binding site using GTP competition experiments. The RTP binding site was then characterized at the atomic level using x-ray crystallography at 2.6 Å resolution.
The structure presented here is the first example of a viral enzyme complexed to a ribavirin nucleotide. Two cellular protein structures have been solved in complex with ribavirin nucleotides: the H122G mutant of nucleoside diphosphate kinase from Dyctyostelium discoideum (PDB code 1MN9 (6)) and the inosine-5Ј-monophosphate dehydrogenase from Tritrichomonas foetus (PDB code 1ME8 (5)). In the former complex, the binding of RTP is completely aspecific; the phosphate and ribose of RTP interact via hydrogen bonding with the protein, but the only ribavirin pseudobase-protein interaction is an aromatic stacking involving Phe-64, which could be achieved by any nucleobase (Fig. 3, B and E). The atomic contacts are different in the complex of RMP with inosine-5Јmonophosphate dehydrogenase, as the ribavirin pseudobase establishes three hydrogen bonds with the protein (Fig. 3F). In the ternary complex with coenzyme A (PDB code 1ME7), no structural rearrangements take place, as the CoA binds in a preformed binding site and the interactions of ribavirin with the protein are not modified. However, a comparison of the relative position of RMP and IMP in complex with inosine 5Ј-monophosphate dehydrogenase reveals that ribavirin superimposes with IMP perfectly, contrary to what is observed in the NS5MTase DV -RTP complex. This is because of the intrinsic nature of both nucleotides; with IMP, the NH 2 group of ribavirin could have been mimicked by the NH group in the 1-position, whereas with GMP, either the 1-or the 2-position could have been mimicked by the NH 2 group of ribavirin.
It is not known whether ribavirin targets NS5MTase DV in flavivirus-infected cells. It has been demonstrated that ribavirin significantly reduces the growth of several flaviviruses, with no information about the actual viral target (7-9, 19, 20). In the case of dengue serotypes 1, 2, and 3, West Nile, and yellow fever viruses, the reported IC 50 values are between 81 and 171 M (7,9,20). The apparent inhibition parameters observed in this report for RTP (IC 50 ϳ100 M) are in good agreement with these IC 50 values. Because ribavirin decreases the intracellular GTP concentration drastically, it potentiates the successful competition of RTP against GTP for NS5MTase DV binding. Indeed, the average intracellular GTP concentrations in various mammalian cell types have been measured and found to be ϳ0.48 mM (compiled in Ref. 21). Treatment of murine leukemia L1210 cells with 20 M ribavirin decreases the GTP concentration to a value equal to 12% of that of the untreated control, i.e. around 60 M (22). This latter value is precisely the same as that of the K D (GTP) for NS5MTase DV (58 Ϯ 14 M (12)). This observation suggests that with ribavirin treatment, which involves ribavirin concentrations of 600 -800 M in biological fluids (23), the down-regulated GTP concentration is not likely to saturate NS5MTase DV , leaving open the possibility that the binding of RTP to NS5MTase DV may be relevant to the known ribavirin antiflavivirus activity. This discussion may also be relevant to other viruses. Indeed, RTP has been reported to be an inhibitor of the (guanine-N7)-methyltransferase of vaccinia virus, as well as to lead to abnormal RNA cap formation in alphaviruses (reviewed in Ref. 24). To date, there is little evidence that ribavirin actually targets viral enzymes other than RNA polymerases, as in the case of hepaciviruses (3,25). The structural data presented here suggest a model for the inhibition of the RNA 2Ј-O-methyltransferase activity by RTP. RTP may compete with viral RNA cap structures to bind the NS5MTase Dv and prevent efficient RNA cap methylation.
Whether or not RTP causes flavivirus growth inhibition through the targeting of NS5MTase DV in vivo, our results still have implications in terms of drug design. First, screening of the Protein Data Bank revealed that no specific recognition of GTP with only the purine 2-position (as shown in Fig. 3, B and D) has been reported before that of NS5MTase DV (12). The great majority of cellular NTP-binding enzymes appear to contact at least two of the purine at position 1, 2, or 6 (12). Among over 60 protein-GTP or protein-RNA cap complexes examined, only two (PDB codes 1KHB and 1DOA) bind GTP using a single position (O-6 in this case) in addition to base-stacking contacts. In theory, since RTP and GTP have equivalent H-bond donors and acceptors at positions 1 and 6 but not position 2, RTP could bind to GTP-binding proteins using positions 1 and 6. To the best of our knowledge, there is no example of such binding in the Protein Data Bank involving a protein of human origin. The observation of an antiviral selectivity for ribavirin may indicate that ribavirin either targets viral proteins through this position 2 alone (amine of RTP) or binds both viral and cellular GTPbinding proteins through positions 1 and 6 (carbonyl and the NH of RTP) but has a more profound effect on viral targets, hence the observed selectivity. Second, the same series of RTP analogues has been used to inhibit HCV NS5B, a hepacivirus RNA-dependent RNA polymerase (6). Hepaciviruses do not have an RNA capping machinery, but their RNA-dependent RNA polymerase domain is ϳ10% identical (22% amino acid similarity) to that of flaviviruses. We note that the most important ribose groups for binding/inhibition are not similar between the hepatitis C virus NS5B polymerase and NS5MTase DV . For hepatitis C virus NS5B the 3Ј-hydroxyl is critical (6), whereas our present results indicate that the 2Јhydroxyl is most important in the case of NS5MTase DV . Therefore, if the most important group is also the 3Ј-hydroxyl for the flavivirus RNA polymerase, it seems difficult to design a simple ribose modification that would target both the RNA polymerase and the cap-binding enzyme of flaviviruses simultaneously. Third, only main-chain contacts are involved in the binding of RTP to NS5MTase DV . A mere substitution of an amino acid side chain cannot directly discriminate RTP relative to GTP in order to provide potential drug resistance.
Since RNA capping is essential for various viruses (18), the mechanism for ribavirin-mediated inhibition of RNA capping presented here may account in part for the antiviral activity of ribavirin against flaviviruses, but it does not exclude the inhibition of additional viral enzymatic activities. Our results would then provide a basis for rational drug design against human pathogens of viral origin, of which the emerging flaviviruses are a timely example.