De novo synthesis of RNA by the dengue virus RNA-dependent RNA polymerase exhibits temperature dependence at the initiation but not elongation phase.

Replication of positive strand flaviviruses is mediated by the viral RNA-dependent RNA polymerases (RdRP). To study replication of dengue virus (DEN), a flavivirus family member, an in vitro RdRP assay was established using cytoplasmic extracts of DEN-infected mosquito cells and viral subgenomic RNA templates containing 5'- and 3'-terminal regions (TRs). Evidence supported that an interaction between the TRs containing conserved stem-loop, cyclization motifs, and pseudoknot structural elements is required for RNA synthesis. Two RNA products, a template size and a hairpin, twice that of the template, were formed. To isolate the function of the viral RdRP (NS5) from that of other host or viral factors present in the cytoplasmic extracts, the NS5 protein was expressed and purified from Escherichia coli. In this study, we show that the purified NS5 alone is sufficient for the synthesis of the two products and that the template-length RNA is the product of de novo initiation. Furthermore, the incubation temperature during initiation, but not elongation phase of RNA synthesis modulates the relative amounts of the hairpin and de novo RNA products. A model is proposed that a specific conformation of the viral polymerase and/or structure at the 3' end of the template RNA is required for de novo initiation.

encodes a single polyprotein precursor, arranged in the order, C-prM-E-NS1-NS2A-NS2B-NS3-NS4A-NS4B-NS5 (for a review, see Ref. 5). This precursor is processed in the endoplasmic reticulum by a combination of the signal peptidase and the viral serine protease to generate three structural proteins of the virion, C, prM, and E (6 -8) and at least seven nonstructural (NS) proteins.
NS3, the second largest protein encoded by the virus, contains a serine catalytic triad within the N-terminal 180 amino acids, and it requires NS2B for protease activity (9 -18). The crystal structures of the protease domain alone and in complex with an inhibitor have been reported (19,20). However, the function of other nonstructural proteins in viral replication is poorly understood.
NS3 also contains conserved motifs found in several NTPase/ RNA helicases (21)(22)(23). According to the current model, replication is initiated by the viral RNA-dependent RNA polymerase (RdRP) by synthesis of minus (Ϫ) strand to form a doublestranded RNA intermediate, which then serves as a template for genomic positive strand (24 -27). The viral NTPase and RNA helicase activities were reported for a recombinant NS3 lacking the protease domain, expressed and purified from Escherichia coli (28). These activities may be important for replication of positive (ϩ) strand from the double-stranded intermediate by an energy-dependent unwinding step. In addition to protease and RNA helicase activities, flavivirus NS3 protein possesses an RNA 5Ј-triphosphatase activity that can hydrolyze the ␥-phosphate of RNA (29). 2 This activity is the first of the three enzymatic activities required for 5Ј-capping, the other two being guanylyltransferase and 5Ј-RNA methyltransferase (30,31).
NS5, the largest of the DEN2 viral proteins, contains the conserved motifs found in several RNA-dependent RNA polymerases encoded by positive strand RNA viruses (32,33). Additionally, NS5 contains conserved motifs found in RNA 5Ј-methyltransferases (34,35), but has not been directly demonstrated for any flavivirus NS5. In DEN2-infected cells, NS3 and NS5 exist as a stable complex, suggesting that viral replication and 5Ј-capping are closely linked (36). The NTPase activity of dengue virus type 1 NS3 was stimulated by NS5 (37), and this observation is consistent with the role of these proteins as a complex in viral replication.
The in vitro replication systems developed to study positive strand RNA replication have revealed that viral replicases mostly are membrane-bound complexes containing the viral RdRP as well as other cellular and viral proteins. These in vitro RdRP assays utilize exogenous RNA templates and either crude or purified components of viral replicases that catalyze specific synthesis of RNA (38 -45). Viral replicases recognize specific elements contained within the 5Ј-and 3Ј-terminal regions of the viral genomes for initiation of viral replication (46 -58). In flavivirus genomes, two conserved sequences within the 3Ј-untranslated region (3Ј-UTR) as well as stem-loop structures within the 3Ј-and 5Ј-UTRs of flavivirus genomes are thought to be important for viral RNA replication (50, 59 -63). Within the 3Ј-conserved sequence of the dengue virus genome, 94 nucleotides (nt) from the 3Ј-terminus, there is a 9-nt motif that is complementary to a conserved motif within the 5Ј terminal region (5Ј-TR), located 133 nucleotides from the 5Ј-terminus. These motifs are separated by 10.5 kilobases in the full-length viral RNA. It was proposed that these motifs could bring the two ends of the genome together through base pairing interactions and play a role in viral replication. Hence, these motifs are referred to as "cyclization" (CYC) motifs (63).
Previously, we described an in vitro RdRP assay system derived from whole cell lysates isolated from DEN2-infected mosquito (C6/36) or monkey (LLC-MK-2) cells. These active viral replicase complexes can utilize exogenous RNA templates that contained the 5Ј-TR and 3Ј-UTR of DEN2 RNA with the internal coding sequences deleted (hereafter referred as "subgenomic RNA") and synthesize (Ϫ) strand RNA (64). There were two products formed in the in vitro RdRP assay; the first was a labeled template with the same size as the input RNA (770 nt), and the second was twice that of the template (ϳ1540 nt). Using this system, we showed that there is an interaction between the two terminal regions of the viral RNA which is required for RNA synthesis (64). Subsequently, a physical interaction between the two RNAs was also demonstrated using the psoralen/UV cross-linking method, and the two CYC motifs played an essential role for both physical interaction and for RNA synthesis. These results suggested that there is cross-talk between the two terminal regions through the conserved sequence elements in the viral RNA that is required for (Ϫ) strand RNA synthesis (65).
Mutational analysis revealed that RNA synthesis at the 3Ј-UTR of the subgenomic RNA template requires the 5Ј-UTR and the highly conserved 5Ј-CYC motif, which is complementary to the 3Ј-CYC motif within the 3Ј-UTR. Also required is the 3Ј stem-loop structure that includes a predicted pseudoknot structure. Furthermore, it was shown that the complementarity between the two CYC motifs rather than the actual sequences was important for RNA synthesis (64,65). A recent study using a Kunjin viral RNA replicon cell line also revealed that the complementarity rather than actual sequence of CYC motifs are required for replication of Kunjin viral replicon RNA in vivo (66).
Previous studies on flavivirus replication intermediates using Kunjin virus, dengue virus, and West Nile virus revealed that, in flavivirus-infected cells, three RNA species were detected: a genomic RNA of 40 -44 S, a double-stranded RNaseresistant replicative form of 20 -22 S, and a partially RNasesensitive replicative intermediate of 20 -28 S RNA species (24,26,67). The analysis of virus-specific RNAs isolated from Kunjin virus-infected cells in completely denaturing formaldehydeagarose gels (similar to the one used for analysis of in vitro RdRP products in our studies) did not reveal the presence of any species larger than genome size RNA (24). These results suggest that the copy-back mechanism of RNA synthesis resulting in dimer species is not detectable in infected cells under the experimental conditions and might be unique to the in vitro systems using either infected cell lysates or purified polymerase (in this study). Moreover, for replication of positive strand RNA viral genomes in general, de novo initiation is considered to be the key mechanism. The 3Ј-terminal elongation event resulting in dimer species, on the other hand, would result in loss of genome sequences. Because template size and hairpin RNA products were both formed in our in vitro RdRP assays, several questions remained to be addressed: How are these two species related? Are those products of specific structures of the template and/or different enzyme conformations? Is the template size product produced by de novo initiation of RNA synthesis or is it the byproduct of the hairpin RNA by nucleolytic attack at the single-stranded loop region?
To address these questions, in this study, we expressed fulllength NS5 with an N-terminal histidine tag in E. coli. We purified the protein to Ͼ90% in a soluble form. The purified protein is active in the synthesis of (Ϫ) strand RNA from positive strand subgenomic RNA templates but not from an RNA containing only the 3Ј-UTR or from nonspecific RNA templates to any significant extent. For optimal RNA synthesis, 5Ј-TR and 3Ј-UTR are both required as well as the wild type or complementary mutant CYC motifs. These results prove that the NS5 protein alone exhibits specificity for the DEN2 viral subgenomic RNA and is able to initiate (Ϫ) strand RNA synthesis de novo without the requirement of other viral or host cofactors. At the same time, the purified RdRP also synthesizes a hairpin RNA product by 3Ј end elongation. Interestingly, the ratio between de novo and hairpin products formed was directly dependent on the incubation temperature of the RdRP reaction during the initiation phase, but not during the elongation phase of viral RNA synthesis. Temperature is known to influence the structure of nucleic acid or protein. Based on the recently reported structural differences between polioviral and hepatitis C virus (HCV) RdRPs (68), we propose a model in which DEN2 RdRP enzyme assumes two conformations that are in equilibrium. According to this model, a temperature-dependent shift of the equilibrium could determine how the enzyme recognizes the conformation of the template RNA that ultimately results in RNA synthesis by de novo initiation or by 3Ј end elongation.

EXPERIMENTAL PROCEDURES
His Tag NS5 Expression Construct-A BamHI restriction site was engineered into our previously described pLZ-5 plasmid (69), which contains the coding sequence for NS5. PCR was carried out using the primers 5Ј-CGCGGATCCTCGGAACTGGCAACATAGGAGAGA-3Ј and 5Ј-CGACACAGCGTGATGGTCCG-3Ј (nt 7570 -7591 and nt 7716 -7735, respectively, in the DEN2 genome (Ref. 4)) and pLZ-5 as the template, producing a 175-base pair fragment. The PCR fragment contained the DEN2-encoded NruI site at the 3Ј end and an engineered in-frame BamHI site at the 5Ј end. The PCR product was then blunt end-ligated into pLZ-5, which had been previously digested with NruI. This intermediate plasmid was subjected to BamHI digestion, and the fragment was cloned into pQE-32 at the BamHI site to give rise to pMHA-77-3 plasmid. The plasmid contains full-length DEN2 NS5 with an N-terminal 6-histidine tag under control of the lac promoter.
Purification of DEN2 NS5/RdRP from E. coli-E. coli (XL1-Blue) cells (1-liter culture), transformed with pMHA-77-3 plasmid were grown in LB media containing 100 g/ml ampicillin and 0.5% glucose (w/v) at 37°C until A 600 nm reached 0.55. Bacteria were then centrifuged at 5,000 ϫ g in a Beckman HS-4 rotor at 4°C for 20 min. The pellet was resuspended in LB media containing 1 mM isopropyl-␤-D-thiogalactopyranoside and ampicillin, and incubated for 10 h at 18°C. Bacteria were pelleted and lysed by French press in 80 ml of lysis buffer (50 mM NaH 2 PO 4 , pH 7.0, 300 mM NaCl, 10% glycerol, 1% Nonidet P-40, and 1ϫ Complete protease inhibitor mixture without EDTA purchased from Roche Molecular Biochemicals (Mannheim, Germany). Lysate was then incubated with 1.5 ml of Talon resin (CLONTECH, Palo Alto, CA) at 4°C for 1 h. The resin was batch-washed five times, with 12 ml of buffer A containing 50 mM NaH 2 PO 4 , pH 7.0, 300 mM NaCl, 10% glycerol. After transfer of Talon resin to a disposable Bio-Rad column, nonspecific proteins were removed by washing with 40 ml of the buffer A containing 15 mM imidazole (pH 7.1), followed with 25 ml of 20 mM imidazole in the same buffer. Proteins were eluted from the Talon resin with buffer A containing 0.5 M imidazole. NS5-containing fractions were pooled, concentrated by Centricon-30 (Millipore, Bedford, MA) to ϳ500 l, and applied to a G-75 Sephadex (Sigma) column. Proteins were eluted from the column in 0.5-ml fractions at 5 ml/h. NS5 was eluted between fractions 21 and 30. Fractions were pooled and dialyzed against 50 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1 mM dithiothreitol, 5 mM MgCl 2 , and 40% glycerol. Purified NS5 protein was aliquoted and stored at Ϫ20°C.
Preparation of RNA Templates-Contruction of the plasmid encoding DEN2 subgenomic RNA has been described (64). RNA templates were prepared by T7-RNA polymerase (Promega)-catalyzed in vitro transcription of linearized plasmids (pTM1 cut with BamHI or KpnI), or PCR products produced from the plasmid templates as described previously (64). RNA was quantified by spectrophotometry and the integrity was verified by denaturing urea-polyacrylamide gel electrophoresis, followed by staining with acridine orange.
RdRP Assay-The standard reaction mixture (50 l) contained 50 mM Tris-HCl, pH 8.0, 50 mM NaCl, 5 mM MgCl 2, template RNA (1 g), 500 M each of ATP, GTP, and UTP, 10 M unlabeled CTP, and 10 Ci of [␣-32 P]CTP along with 200 ng of purified NS5 except when indicated. The reaction was carried out by incubation at 30°C for 1 h and terminated by acid phenol/chloroform extraction, followed by ethanol precipitation after the addition of yeast tRNA (5 g) as a carrier. The RNA pellet was collected by centrifugation, and the pellet was dried. RNA was resuspended in 50 l of nuclease-free H 2 O and passed through a Bio-Rad P-30 column to remove unincorporated nucleotides. Flowthrough fraction was precipitated with ethanol. RNA was analyzed by formaldehyde-agarose gel electrophoresis and visualized by autoradiography (64). The labeled bands were excised from dried gels and quantified by liquid scintillation counting and also analyzed by densitometry utilizing the NIH program Scion. RdRP reactions at different temperatures were carried out using a gradient thermocycler (Tgradient, Biometra, Göttingen, Germany). The reactions were terminated and analyzed as described above.
Sodium Periodate Treatment of RNA Transcripts-The 3Ј-hydroxyl of RNA transcripts was blocked by sodium periodate as described previously (64). Reactions were phenol-extracted and ethanol-precipitated as described above.
Heparin-trap Experiments to Distinguish Early Initiation Phase Versus Elongation Phase of RNA Synthesis-To isolate the initiation and elongation phases of RNA synthesis by NS5, we altered previously described protocols for use with our system (70 -73). Subgenomic RNA 770 nt was incubated at 30°C for 10 min with 500 M each of three nucleotides (ATP, GTP, and CTP), 10 Ci of [␣-32 P]CTP, and 200 ng of purified NS5 to allow partial synthesis of RNA (based on the template sequence to be six nucleotides). Heparin (50 ng) was then added (2 l of 25 ng/l) and incubated for 5 min before the addition of UTP (500 M). Reactions were then allowed to complete elongation synthesis for 60 min. To study the influence of varying the temperature of incubation on initiation phase of RNA synthesis, reactions were incubated at different temperatures with the enzyme and three nucleotides, but were moved to a constant temperature (30.8°C) after the addition of heparin and the fourth nucleotide. To study the effect of variations on the elongation phase of RNA synthesis, reactions as described above were all kept at 30.8°C until after the addition of heparin; UTP was then added, and incubations were continued at differing temperatures for 60 min.
Determination of Polarity of the RdRP Products by RT-PCR Using Strand-specific Probes-To determine the polarity of the RdRP products, strand-specific oligodeoxynucleotide primers of negative and positive polarity were synthesized and used for reverse transcriptasecatalyzed cDNA synthesis, followed by PCR (RT-PCR) amplification of cDNAs. Because there was an excess of positive stranded subgenomic RNA template used for the RdRP reaction and this interfered with annealing of the DNA primers, the RdRP reaction mixtures were subjected to RNase A digestion to remove all single-stranded RNA. The portion of the positive strand RNA template annealed to the newly synthesized negative strand would be resistant to RNase A digestion. Samples of RNA product after RNase A digestion were mixed with 20 mol of either positive (5Ј-AGCTGTACGATGGCGTAG-3Ј) or negative strand (5Ј-CTACGCCATCGTACAGCT-3Ј) primer and were denatured in a 11-l solution at 95°C for 2 min, followed by a slow cooling to 65°C. Reactions were then incubated on ice for 2 min. The reverse transcriptase reactions were carried out using a kit (Superscript II, Life Technologies, Inc.) according to manufacturer's instructions. Briefly, the reaction mixtures (8 l) contained 2.5 mM dNTPs, 1ϫ First Strand Buffer, 50 mM dithiothreitol, and 10 units of RNasin inhibitor (Promega). Reactions were incubated for 2 min at 42°C, before addition of 2 units (1 l) of Superscript II reverse transcriptase. Reactions were incubated for 50 min at 42°C and stopped by incubation at 70°C for 15 min. Dilutions of this mixture were then used for standard PCR with DNA primers. For negative strand visualization, 5Ј-AGAACCTGTTG-ATTCAACAGCACC-3Ј and 5Ј-AGCTGTACGATGGCGTAG-3Ј were utilized. For positive strand visualization, 5Ј-ACCGCGTGTCGACTGT-ACAACAGCTGA-3Ј and 5Ј-CTACGCCATCGTACAGCT-3Ј were used. The PCR products were analyzed by agarose gel electrophoresis and ethidium bromide staining and photographed using a Kodak digital camera.

RESULTS
Purification of Full-length DEN2 NS5-Initial attempts to express full-length NS5 were unsuccessful because of toxicity of NS5 to E. coli; the bacteria were able to grow only to an A 600 of about 0.35. We made use of previous observations that growth in glucose maintains a low level of cAMP and cAMP activator protein (also known as cAMP receptor protein), allowing the repression of lac promoter, resulting in a tight regulation of T7 RNA polymerase and NS5 expression. Prior to induction of the lac promoter with isopropyl-␤-D-thiogalactopyranoside, glucose was removed. The expression of full-length NS5 protein was significantly improved under these conditions; however, the protein was primarily located in the insoluble pellet fraction (data not shown). To maximize the amount of soluble NS5 and reduce cleavage of the full-length protein, expression was then performed at 18°C for 16 h. This resulted in ϳ80% full-length NS5 in the soluble fraction with cleavage products of NS5 partitioning primarily in the insoluble fraction.
Eluate of 6ϫHis-tagged NS5 from the metal affinity (Talon ® ) matrix contained the full-length NS5 as well as additional polypeptides in the size range of ϳ46 -28 kDa (Fig. 1A, fractions 6 -10). To separate the full-length NS5 from these contaminating proteins, gel filtration chromatography was employed. As shown in Fig. 1B, this gel filtration step yielded NS5 protein of over 90% purity (Fig. 1B, lanes 3-8) and all subsequent experiments were carried out using this protein.
Purified NS5 Is a True RNA-dependent RNA Polymerase and Is Active in RNA Synthesis on Subgenomic Dengue Viral RNA Templates-We then sought to determine whether the purified recombinant NS5 was enzymatically active in RNA synthesis using the subgenomic RNA 770 nt template (64). This RNA contains the entire 3Ј-UTR 454 nt and 5Ј-TR 230 nt of the DEN2 genome, including the stem-loop structures and 3Ј-and 5Ј-CYC motifs. The RdRP assay was carried out as described under "Experimental Procedures" except that initial assays were conducted in the presence of RNasin (RNase inhibitor; 0.2 unit/l) and actinomycin D (16 M) to inhibit any potential contaminating RNases or bacterial DNA-dependent RNA polymerase, respectively. The purified NS5 contains no detectable RNase activity, which was determined by incubation of the RNA template with enzyme for various time periods followed by denaturing polyacrylamide gel electrophoresis analysis. There was also no difference in polymerase activity in the presence or absence of actinomycin D (data not shown). Therefore, these two reagents were removed from the subsequent assays. Additionally, a DNA template containing a T7 polymerase promoter was not active with purified NS5, verifying that there is no detectable contaminating bacterial polymerase (Fig. 2, lane 3). The reaction was incubated at 30°C for 1 h and analyzed by a completely denaturing formaldehyde-agarose gel electrophoresis followed by autoradiography.
The results of RdRP activity assay indicate that RNA synthesis was dependent on the addition of subgenomic RNA 770 nt template and the NS5 protein in a complete system, but not when the NS5 was omitted or heat-inactivated (Fig. 2, lanes 1,  2, 4, and 5). Two products were formed in the complete system, one having the size of the input template (770 nt), and the second having a mobility consistent with twice the size of the template (ϳ1540 nt). The formation of these two products required all four nucleotides. When labeled CTP alone was present at the same concentration as in the complete reaction, no product could be detected (lane 6), suggesting that the purified enzyme is free of any terminal transferase as a contaminant. However, when reactions contained only CTP, UTP, and ATP, an approximately template length product was produced in significantly lower amounts compared with those formed when all four nucleotides were present (lane 8.) Lanes 6 -9 needed a 10-day exposure of the gel for the same experiment shown in panel A, which was visualized in an overnight exposure; therefore, these are shown separately in panel B. The products of similar size were also formed when either ATP or UTP alone were omitted from the reaction mixture and but these were even lower in abundance than when GTP alone was omitted (lanes 7 and 9 versus lane 8). The sensitivity of the product to RNase A under high salt conditions indicated that the labeled product is a single-stranded RNA (lanes 11 and 12). This result suggested that the template RNA was labeled in the reaction because of terminal addition of nucleotides to the 3Ј end by RdRP. Taken together, these results indicate that the purified NS5 is specific for RNA, acting as a true RNA-dependent RNA polymerase.
Template Specificity and RdRP Activity on DEN2 Subgenomic RNA Template for RNA Synthesis in Vitro Requires Both 5Ј-and 3Ј-Terminal Regions-To determine whether the purified NS5 specifically recognizes dengue viral RNA templates or is able to utilize any nonspecific RNA, we tested different RNAs as templates. Nonspecific RNA templates were produced by digestion of the pTM1 vector with either BamHI or KpnI to yield linearized plasmid templates, which were then used for T7 RNA polymerase-catalyzed in vitro transcription. The 3Ј-UTR RNA 454 nt was also tested as a template for the purified RdRP, as it alone was inactive for RNA synthesis in our previous in vitro system that utilized crude cytoplasmic extracts from DEN2-infected cells (64). The results of RdRP assays using the purified NS5 and different RNA templates shown in Fig. 3 indicated that only the subgenomic RNA 770 nt was an active template for RNA synthesis (Fig. 3, lane 1). The 3Ј-UTR RNA was inactive, similar to the result obtained using the infected cell lysate system (64). In addition, neither of the RNAs produced by in vitro transcription of linearized pTM1 plasmid (BamHI or KpnI) served as an efficient template (Fig.  3, lanes 3 and 4). These results indicate that purified NS5 alone exhibits high template specificity for RNA synthesis in vitro and its template specificity parallels that of crude cytoplasmic extracts from DEN2-infected cells that contain a complex of other host and viral protein(s) such as NS3 (64).
We further verified the template specificity of the purified RdRP in an assay system that required interaction between two RNAs containing the terminal regions of the viral genome for RNA synthesis by the viral replicase complex in the cytoplasmic extract (64). In that assay, the 3Ј-UTR 454 nt RNA alone was inactive for RNA synthesis but was activated by the addition of the 5Ј-TR 230 nt RNA. The two conserved CYC motifs that are complementary to each other were both required for the template activity of 3Ј-UTR 450 nt RNA as well as for physical interaction between the two terminal regions as determined by the psoralen/UV cross-linking method. Mutation of either the 3Ј-CYC or the 5Ј-CYC motif abrogated the ability of the 3Ј-UTR to be activated for physical interaction or RNA synthesis by the 5Ј-TR RNA, but was fully restored when the two mutant motifs were complementary to each other (64,65). We verified these results in this study using the purified polymerase. The results of RdRP assays using purified polymerase shown in Fig. 4 indicate that both CYC motifs are important in transactivation assays for RNA synthesis as mutation of either motif significantly interfered with RNA synthesis (Fig. 4B, lanes 4 and 5  versus lane 3); however, when both motifs were mutated such that they are complementary to each other, the RNA synthesis was restored (Fig. 4B, lane 6). In these assays, the 5Ј-TR RNA alone was active in the RdRP assay and addition of 3Ј-UTR RNA suppressed this template activity of the 5Ј-TR RNA, concomitantly activating RNA synthesis at the 3Ј-UTR RNA. Thus, the template specificity of the purified enzyme mimicked that of the replicase complex from the cytoplasmic extracts of infected cells as reported previously (64).
The purified enzyme was also analyzed for its specificity of recognition of subgenomic RNA templates in which both terminal regions of the viral genome are in the same RNA molecules as opposed to in two separate (5Ј-TR 230 nt and 3Ј-UTR 454 nt ) RNAs as described above. Our previous study using the cell lysate system indicated that mutation of the 5Ј-CYC motif in the subgenomic RNA still severely affected RNA synthesis in vitro, but the mutation of the 3Ј-CYC motif was tolerated as the activity was only reduced by 2-fold. The results shown in Fig.  4A (lanes 2-4) also support other data that the purified polymerase exhibited similar template specificities as the viral replicase from the cell lysate system.
Characterization of the RdRP Products-Our previous assays utilizing crude cytoplasmic extracts from DEN2-infected mosquito (C6/36) cells also produced two products on a denaturing formaldehyde-agarose gel system: a template-sized RNA (1ϫ) and a hairpin RNA, twice the size of the template (2ϫ) (64,65). The dimer-sized species are possibly formed by short additions of nucleotides, either by RdRP or the host terminal transferase to the 3Ј end of the template RNA, which are then extended by RdRP. Alternatively, the dimer species could also have resulted by 3Ј-elongation of a "folded-back" structure of the RNA template. Such an intrinsic structure in the viral genome was previously suggested to be the source of cDNA species that were isolated in which sequences complementary to 3Ј-internal region were covalently linked to the 3Ј-terminal sequences by the reverse transcriptase (63).
In addition to the dimer species, we also detected templatesized (1ϫ) RNA product using the cytoplasmic extract as the source of RdRP. The result that this template size product was resistant to RNase A digestion supported the conclusion that the labeled (Ϫ) strand RNA was annealed to the unlabeled input (ϩ) strand RNA template. This conclusion was further supported by the results of RNase H mapping using strandspecific oligodeoxynucleotides (65). However, the question of whether the 1ϫ product was synthesized by de novo initiation of RNA synthesis or was produced by digestion of the hairpin RNA product by a nuclease(s) present in the cytoplasmic extract was not resolved.
In this study, we sought to address this question using the purified polymerase as the enzyme was devoid of any nuclease contamination. When the assays were performed with the purified RdRP and all four NTPs, both hairpin and template size products were formed similar to our previous observation using the cell lysate system (64), suggesting that the 1ϫ product was more likely to be formed because of de novo initiation of RNA synthesis than from the hairpin product.
Therefore, to determine whether the 2ϫ product was in fact a hairpin RNA formed by 3Ј end elongation of the template RNA, we subjected the products to RNase A digestion under high and low salt conditions. Under high salt conditions, RNase A will digest only single-stranded RNA, whereas doublestranded RNAs will be resistant. The single-stranded loop region of a hairpin product, however, will be sensitive to RNase A digestion, and will migrate as a 1ϫ product on a completely denaturing gel (see Fig. 5A). Under low salt conditions, both double-stranded and single-stranded RNAs are sensitive to RNase A digestion. When the products were subjected to RNase A analysis, high salt conditions produced predominantly 1ϫ product, when related to the same reaction without addition of RNase A (Fig. 5B, lane 2 versus lane 1). Under low salt conditions, all RNA was digested (Fig. 5B, lane 3). This result establishes that the slower migrating band on the completely denaturing gel is in fact a hairpin RNA product, produced from 3Ј end-elongation of the input template. Further proof that (Ϫ) strand was synthesized de novo and by 3Ј-elongation was obtained by reverse transcriptase-mediated cDNA synthesis using strand-specific probes followed by PCR (RT-PCR) (see below).
Evidence for de Novo Initiation of Minus Strand Synthesis by the Purified RdRP-We were interested to know whether the template length product was produced through de novo initiation of the 3Ј-UTR. First we sought to determine whether a sodium periodate-treated subgenomic RNA 770 nt having a blocked 3Ј hydroxyl group can still serve as the template for RNA synthesis. This template should only form the 1ϫ product, and no hairpin product would be possible, as the 3Ј end elongation would be blocked.
The periodate-treated template RNA 770 nt produced only the 1ϫ product, and no hairpin product could be detected (Fig. 5B,  lane 4), which suggested that periodate oxidation of the 3Јterminal ribose moiety went to completion because the presence of any unmodified RNA template would have resulted in formation of hairpin RNA. This experiment also serves as an evidence against the possibility that the 1ϫ product is formed from a hairpin intermediate by a structure-dependent "ribozyme-like" nuclease, because no hairpin product was formed in this reaction. We sought to determine whether this product is in fact the minus strand formed by de novo initiation at the 3Ј end of the subgenomic RNA template. We analyzed the product by RNase A digestion. As shown in Fig. 5B (lane 5), the product was RNase A-resistant under high salt, whereas it was completely sensitive under low salt conditions (lane 6). As diagrammed in Fig. 5A, resistance of the product to RNase A digestion proves that it is a double-stranded RNA. This verifies that the radioactive RNA visualized on the gel is not simply a terminally radiolabeled input template RNA, as this would be degraded by RNase A. Next, we used strand-specific oligodeoxynucleotides as primers that could anneal only to specific regions of either (ϩ) or (Ϫ) strand polarity in the doublestranded RNA product for reverse transcriptase reaction. This step was followed by PCR amplification, and the cDNA products were analyzed by agarose gel electrophoresis and stained by ethidium bromide. The results shown in Fig. 5C indicate that cDNA products of expected sizes were formed only from the template size de novo product formed in the RdRP reaction using the sodium periodate-treated (Fig. 5C, lanes 5 and 6) and untreated samples (lanes 3 and 4); these results verify that (Ϫ) strand formed in the RdRP reaction is annealed to the (ϩ) strand template RNA. No cDNA products were synthesized from the (ϩ) strand template RNA alone with DNA primers of (ϩ) polarity, serving as the control (lanes 1 and 2). These results, taken together, indicate that the purified RdRP is able to initiate primer-independent de novo synthesis of (Ϫ) strand RNA in the absence of other viral or host factors.
Temperature Dependence of de Novo RNA Synthesis by the Purified NS5/RdRP-The evidence gathered thus far using our in vitro RdRP assay systems suggests that the structure of the 3Ј-UTR RNA dictates its template activity. The structure is modulated by interaction between the two terminal regions of the viral genome. The CYC motifs are required for this interaction, although they are not sufficient for initiation of RNA FIG. 4. Requirements for cyclization motifs for RNA synthesis from subgenomic RNA. A, subgenomic RNAs containing wild-type (box) and mutant (circle) CYC motifs were used for RdRP reactions and the products were analyzed by formaldehyde-agarose gel electrophoresis followed by autoradiography. Lanes 1-4, wild type CYC motifs, mutant 5Ј-CYC motif, mutant 3Ј-CYC motif, and mutant 5Ј-and 3Ј-CYC motifs, respectively. B, 5Ј-TR and 3Ј-UTRs containing wild type (lanes 1-3) or mutant CYC motifs (lanes 4 -6) were used either individually or added together as templates for the RdRP. In lane 7, a template size product was loaded as a size marker (ϳ770 nt).
synthesis. The conserved secondary structures at the 5Ј-UTR and the 3Ј-stem-loop region that includes the tertiary pseudoknot structure are also important for RNA synthesis in vitro (64,65). Several studies on RNA folding reveal that varying the temperature could influence RNA secondary structure as well as tertiary interactions such as pseudoknot structures (see, for example, Refs. 74 -76). To study the effect of temperature on the template efficiency and the products formed in the in vitro RdRP assay, we incubated the enzyme with subgenomic RNA template at different temperatures in the range between 20 and 40°C for 90 min, as described under "Experimental Procedures." Products of RdRP reactions were analyzed by completely denaturing formaldehyde gel electrophoresis and autoradiography. The results shown in Fig. 6A indicate that the 1ϫ de novo initiation product was predominantly formed at lower temperatures, whereas at higher temperatures a predominance of hairpin product was produced. We quantitated the amount of each product by excising the corresponding radioactive band from the gel followed by scintillation counting. The autoradiograms of labeled RNA were also quantitated by densitometry using the NIH software Scion (Scion Corp). The experiments were repeated three times, and the quantitative data were plotted. The results shown in Fig. 6B indicate that there is a good correlation between the formation of the hairpin product and an increase in the temperature of incubation of the RdRP reaction.
Temperature Dependence of de Novo RNA Synthesis Is at the Stage of Initiation and Not Elongation-With the knowledge FIG. 5. Characterization of RdRP products. A, schematic description of RNase A digestion of RdRP products. B, lanes 2 and 3, product from RdRP reaction using the subgenomic RNA template was incubated with RNase A in high or low salt conditions. Lane 1, no RNase A added. Subgenomic RNA was subjected to sodium periodate treatment to block the 3Ј-OH group and was used as a template for the RdRP reaction. Lane 4, no RNase A added; only template-length product was formed. Lanes 5 and 6, aliquots of reaction mixture was subjected to RNase A digestion under high and low salt, respectively. C, purified RNA from RdRp reactions was subjected to RT-PCR as follows. Reactions were first digested with RNase A to remove single stranded RNA. Samples were then subjected to RT-PCR utilizing primers of either (ϩ) polarity (lanes 1, 3, and 5) or (Ϫ) polarity (lanes 2, 4, and 6). The product from reverse transcription was then utilized for PCR with the corresponding primers, as described in "Experimental Procedures." Products utilized for RT-PCR were input subgenomic 770nt template before RdRp reaction (lanes 1 and 2), which was not digested with RNase A, products from RdRp reactions with subgenomic 770nt RNA (lanes 3 and 4), or periodate treated subgenomic 770nt RNA (lanes 5 and 6). Products were fractionated by agarose gel electrophoresis and visualized by ethidium bromide staining. Expected products were 280 nt for negative strand RNA and 340 nt for positive strand RNA. that de novo minus strand RNA synthesis by RdRP is dependent on temperature, we set forth to determine at which step, initiation or elongation, temperature plays a role in RNA synthesis. Heparin, which binds to free RNA polymerase and inhibits its activity, has been used previously to study promoter occupancy by the enzyme (71,(77)(78)(79). It is also known that heparin inhibits initiation of transcription but not elongation by the RNA polymerase already bound to the newly initiated RNA (77,80). We used a strategy of partial incorporation of nucleotides to allow limited synthesis of nascent chain after initiation (70 -73, 79). In this way, we could examine the temperature dependence of initiation events by the RdRP, independently from the elongation step, by addition of the missing nucleotide and heparin to inhibit new initiation events. By incubating the template RNA with the viral polymerase and three nucleotides (ATP, CTP, and GTP), based on the sequence at the 3Ј end, the enzyme was expected to add 6 nucleotides and pause; then, heparin and, 5 min later, UTP were added. This stable complex was able to carry out elongation of RNA synthesis, which was resistant to the presence of heparin in the reaction mixture. We were able to achieve a single round of RNA synthesis from this protocol, because the viral polymerase is inactivated by heparin once it releases the template, such as in the case of abortive transcription or runoff from the template. By segregating the binding and initiation phases from the elongation phase, we were able to study the effect of temperature during these two processes.
We varied the temperature between 20 and 40°C during initiation by incubating the template and the RdRP together with ATP, CTP, and GTP. After the addition of heparin, reactions were moved to a constant 30.8°C temperature, and then UTP was added. This effectively varied the temperature during the initiation phase of the RdRP activity, but kept the elongation phase constant at 30.8°C. The products were analyzed by formaldehyde-agarose gel electrophoresis and quantitated as described under "Experimental Procedures." The results of the temperature gradient for initiation events (Fig. 7, A and B) resemble the temperature dependence for the formation of the de novo 1ϫ product seen earlier (Fig. 6, A and  B). The amount of de novo 1ϫ product at lower temperatures (20 -24°C) was over twice that of the hairpin product. The formation of the two products was approximately equal at 30°C.
To prove that the initiation event and not the elongation step FIG. 6. De novo synthesis of RNA by the RdRP is temperaturedependent. A, Purified RdRP was incubated with subgenomic RNA for 90 min at varying temperatures as described under "Experimental Procedures." Products were resolved by formaldehyde-agarose gel electrophoresis and subjected to autoradiography. B, bands were excised from dried gel and subjected to liquid scintillation counting and verified by densitometry using an NIH software (Scion) as described under "Experimental Procedures." The average ratio of de novo (1ϫ) to 3Ј end elongation (2ϫ) products from three separate experiments were plotted as shown.
FIG. 7. Temperature dependence of de novo synthesis of RNA is at the initiation but not elongation phase. A, purified NS5 was incubated with subgenomic RNA and three nucleotides (A, G, C and [␣-32 P]CTP for 10 min at varying temperatures. 50 ng of heparin was added and incubated for an additional 5 min before addition of UTP. At this time, temperatures for all reactions were held constant at 30.8°C. Products were analyzed by formaldehyde-agarose gel electrophoresis followed by autoradiography. B, labeled bands were excised, quantitated by liquid scintillation counting, and verified by densitometry. The average ratio of 1ϫ:2ϫ products from four experiments was plotted as shown. is dependent on temperature, the initiation step was carried out at a constant temperature (30.8°C) in the presence of three nucleotides, followed by the addition of heparin as described above. The elongation reaction was carried out by the addition of the fourth nucleotide (UTP), and incubation was continued at different temperatures between 24 and 40°C.
The results shown in Fig. 8 (A and B) indicate that there is no difference in the relative amounts of de novo product and the hairpin product under these conditions. Moreover, above 27°C, the total amount of the two products produced was not dependent on the elongation temperature. This result suggests that 60 min of incubation is sufficient for the enzyme to complete elongation even at lower temperatures. The results of Fig. 8A also show that, at 23.6°C, the enzyme was unable to catalyze elongation as fully as at temperatures between 27°C and 38°C. The total incorporation of labeled nucleotides into RNA was reduced in the experimental lanes in both Figs. 7 and 8 because of the presence of heparin, which allowed only a single round of synthesis; in contrast, the reaction carried out in the absence of heparin was loaded in the control lanes (Figs. 7 (lane  1) and 8 (lane 6)), and the total incorporation of labeled nucleotides was much higher in the absence of heparin because multiple rounds of synthesis were permitted within the time period of incubation. Therefore, our results, taken together, indicate that temperature of the initiation step of the RdRP reaction directly influences the ratio of the products formed by de novo synthesis versus 3Ј end elongation from the dengue viral subgenomic RNA template. To our knowledge, this is the first report describing the effect of temperature on RNA synthesis by de novo initiation versus 3Ј end elongation by a viral RdRP.

DISCUSSION
Expression and Purification of DEN2 Viral RdRP from E. coli-Previously, we described an in vitro RNA-dependent RNA polymerase assay system using cytoplasmic extracts from DEN2-infected mosquito (C6/36) and monkey kidney (LLC-MK2) cells (64). To isolate the function of DEN2 NS5 from that of other viral and cellular proteins present in our previous assay system and study the biochemical properties of the viral RdRP, we set out to purify NS5 from E. coli. Purified viral polymerase enabled us to accurately assess the enzymatic functions of NS5 in the absence of any intervening host or viral proteins as well as endogenous viral replicative intermediate RNA species that were present in the previously characterized system (64).
Template Specificity of the Purified RdRP-The purified NS5 is a true RdRP, and it requires all four NTPs. It exhibited selective template specificity, as it is active optimally only on dengue viral-derived subgenomic RNA templates containing wild type CYC motifs but not on RNA containing only the 3Ј-UTR and a significantly reduced activity on the subgenomic RNA containing the mutant 5Ј-CYC motif, or on other nonspecific RNA templates. These results suggest that the enzyme recognizes some specific secondary and/or tertiary structure at the 3Ј-terminal region of RNA template for initiation of RNA synthesis.
Previous studies on other viral polymerases demonstrated initiation of RNA synthesis was primer-dependent and the enzymes acted on RNA templates without any specificity. For example, poliovirus, encephalomyocarditis virus, rhinovirus, brome mosaic virus, or dengue virus RdRPs all required a primer, which can be either the 3Ј end of the template that is then extended by a fold-back mechanism or an exogenous RNA primer (81)(82)(83)(84)(85)(86). In poliovirus replication, the genome-linked VPg is first uridylylated by the 3D polymerase, which requires polioviral RNA; the uridylylated VPg then serves as a primer for initiation of negative strand RNA synthesis on the polyadenylated polioviral RNA (87). An RNA hairpin within the coding region of poliovirus 2C protein is the site specifically used for uridylylation of VPg, and this hairpin structure, which is well conserved in Enterovirus family of Picornaviridae, is required for replication of the viral genome (88,89). An internal RNA structural element has also been identified that is required for rhinovirus14 replication (90). It is possible that 5Ј-TR RNA of dengue viral genome might also contain a cisacting replication element that includes the 5Ј-CYC motif. Mutation of this motif has been shown to affect RNA synthesis in vitro (Ref. 64 and this study) and for another flavivirus, Kunjin virus replicon, in vivo (66).
The HCV RdRP (NS5B protein), expressed and purified from E. coli or insect cells, has been characterized in great detail. The enzyme also exhibited a primer-template-dependent RNA synthesis on homopolymeric or heteropolymeric HCV-derived "D" RNA templates without any specificity (91)(92)(93)(94). In the absence of a primer, the polymerase catalyzed 3Ј end elongation to produce a hairpin RNA (91,(95)(96)(97). However, template size product was formed only in the presence of a primer complementary to its 3Ј-terminus of an RNA template, which FIG. 8. Variation of temperature during elongation phase has no effect on de novo versus 3 end "fold-back" synthesis of RNA. A, purified RdRP was incubated with subgenomic (770 nt) RNA, three nucleotides (A, G, C), and [␣-32 P]CTP for 10 min at 30.8°C. 50 ng of heparin was added to the reaction mixtures and incubated for an additional 5 min before addition of UTP. At the time of UTP addition, reaction mixtures were moved to varying temperatures. Products were visualized by formaldehyde-agarose gel electrophoresis followed by autoradiography. B, labeled bands were excised and quantitated by liquid scintillation counting and verified by densitometry. The average ratio of 1ϫ:2ϫ products from four experiments was plotted as shown.
was blocked at the 3Ј end to prevent "copy-back" synthesis (96).
De Novo Initiation of RNA Synthesis by the Purified RdRP-In contrast to the properties of these picornaviral RdRPs, previous studies on Q-␤ RdRP (98) and some plant viral enzymes (43, 45, 70, 99 -101), and more recent work on purified HCV NS5B (102)(103)(104), bovine viral diarrhea virus RdRP (105), and Kunjin RdRP (106) have shown that these enzymes can initiate RNA synthesis de novo. The HCV NS5B RdRP was able to utilize the HCV-derived "D" RNA and yielded both template size and hairpin products (103). After blocking the 3Ј-OH of the template by incorporation of the chain terminator, cordycepin, the formation of the hairpin product was blocked without affecting the formation of the template size product. From these results the authors concluded that both de novo and the copyback mechanism of RNA synthesis were responsible for the formation of these two products.
In this study, we show that dengue viral RdRP, purified from E. coli, catalyzes both de novo initiation of RNA synthesis and 3Ј end elongation from the viral template RNA. The enzyme exhibits high template specificity and is able to utilize, in the absence of any other viral or host factors, the dengue viralderived subgenomic RNA template containing either wild type or complementary mutant CYC motifs for optimal RNA synthesis. Substitution mutations in either of the CYC motifs reduced the template activity, but more severely with the mutation of the 5Ј-CYC motif. The mutation of the 3Ј-CYC motif was tolerated and the activity was reduced only by about 2-fold. Because the complementary mutations in both CYC motifs restored the near-wild type level of activity, this result argues against the role of 5Ј-CYC motif in a structure-specific initiation event similar to the requirement of an internal RNA hairpin for uridylylation of VPg primer as in the case of poliovirus RNA (88); on the other hand, these motifs could play a role in bringing the two ends of the dengue viral RNA genome together and the resulting RNA conformation could influence the tem-plate recognition by the viral polymerase for de novo initiation of RNA synthesis at the 3Ј end. We have shown previously that physical interaction facilitated by wild type CYC motifs per se is not sufficient and other motifs such as 5Ј-and 3Ј-stem loop structures within the UTR regions are also required for RNA synthesis (65). Analysis of the secondary structures of the four subgenomic RNA templates by using the software described by Zuker et al. (version 3.0) (107) revealed that, although the overall predictive structures of the wild type, complementary double CYC mutant, and the 3Ј CYC mutant are similar, the 5Ј-CYC mutant has a strikingly different structure (65). Experimental verification of the differences in the structures of the subgenomic RNA by enzymatic and/or chemical probing methods could reveal an insight regarding differences in their template efficiencies in both cell lysate and purified RdRP assay systems.
Interestingly, when RNA synthesis was carried out at various temperatures of incubation in the presence of all four NTPs, total incorporation of nucleotides into both template size and dimeric RNA species was much greater at higher temperatures than at lower temperatures (Fig. 6A, lanes 3 and 4). This increased synthesis of RNA by the enzyme at higher temperatures is perhaps attributable to the removal of pauses on the template RNA for the enzyme, as the incubation temperature is known to influence the secondary and tertiary structures of RNA (74 -76). The optimum temperature for RNA synthesis by the enzyme was also found to be between 29.1 and 30.9°C, above which it falls off (Figs. 6A and 7A).
When RNA synthesis by partial incorporation was carried in the presence of only three NTPs (ATP, GTP, and CTP) at various temperatures (20 -40°C), the enzyme pauses after limited synthesis of RNA, which is either in the form of a short primer or as the 3Ј-extension of the template RNA by a few nucleotides. The ratio between these two forms of short RNA products is shown to depend on the initial temperature of FIG. 9. Model for temperature-dependent switch between two conformational states of dengue viral RdRP. A model is proposed in which the viral polymerase exists in two conformational states: open and closed forms, which are in equilibrium. The equilibrium is shifted toward an open form at higher temperatures in which the enzyme can bind to a fold-back structure at the 3Ј end of RNA template and carry out 3Ј-elongation giving rise to a dimeric RNA. At lower temperatures, the binding to the fold-back structure is less efficient because the enzyme exists predominantly in closed conformation. This form of enzyme binds more efficiently to single-stranded region at the 3Ј-terminus of the template RNA and initiates de novo synthesis of RNA to yield a template size double-stranded RNA product. Further details of this model are described in the text.
incubation. For example, when the initial temperature was 27.2°C and then shifted to 30.8°C after the heparin and UTP were added, the template-sized (1ϫ) product was 2-fold over the dimer species compared with the ratio of products formed when the temperature of both initiation and subsequent elongation were kept at 30.8°C (Fig. 7A, lane 3 versus lane 4).
These results seem to suggest that conformation of the enzyme may play a key role in the template recognition by the enzyme and RNA products formed. This notion is supported by a recent report, which has shed much light on the structurefunction aspect of RNA polymerases (68). Although the active site of 3D pol exists in an open conformation, HCV RdRP contains a single ␤-hairpin consisting of 12 amino acid residues occluding the active site (68,108,109). A comparison of the crystal structures from the poliovirus 3D pol and HCV RdRP revealed an important difference at the active site of the two enzymes. HCV RdRP has an extra ␤-hairpin consisting of 12 amino acid residues in the thumb subdomain, which is absent in the poliovirus 3D pol . The template requirements for 3D pol and NS5B are quite different. Although the 3D pol is active on double-stranded templates, HCV NS5B is unable to bind double-stranded RNA templates productively (68,110). When this ␤-hairpin was shortened by 8 nucleotides (4 nucleotides on either side of the turn), NS5B was able to initiate synthesis on a double-stranded RNA as efficiently as the poliovirus 3D pol . It has also been demonstrated that HCV NS5B produces dinucleotide primers through abortive initiation that are able to effectively prime RNA synthesis (111). The ␤-turn prevents the RNA from sliding through the RNA active site "hole" and may aid in the formation of the dinucleotide primers through steric hindrance effects on abortive initiation.
If the dengue virus RdRP structurally resembles the HCV enzyme in having an active site occlusion by a ␤-hairpin, then temperature may strongly influence its conformation. The dengue viral RdRP may exist in two conformational states, which are in equilibrium (Fig. 9). The active site of the enzyme itself is likely to exist in an equilibrium between a more rigid "closed" conformation at lower temperatures and a more mobile "open" conformation at higher temperatures. The template RNA may also assume either a "fold-back" hairpin or single-stranded RNA region at the 3Ј-terminal sequences which leads to 3Ј end elongation or de novo initiation of RNA synthesis, respectively. Moreover, regardless of temperature, the viral RNA templates may be in an equilibrium between these two forms. At higher temperatures, the equilibrium is shifted to predominantly an open conformation of the enzyme and a very few abortive primers are produced; the enzyme, on the other hand, binds to a 3Ј-fold-back region of RNA and synthesizes a predominance of dimeric RNA product by 3Ј-elongation. However, at lower temperatures, the closed conformation of the enzyme may predominate and in this state it may be unable to bind to the fold-back structure of viral template RNA; the enzyme may therefore initiate de novo synthesis on a single-stranded region of RNA. It is possible that the conformational states of the RdRP and/or the viral template RNA may be stabilized by other viral and/or host proteins and by association with the endoplasmic reticulum membranes in the infected cells where viral replication has been localized (5,25,112). Our in vitro RdRP assay using the purified NS5 described in this study is likely to be useful for identification of other factors involved in viral replication.