In Vitro RNA Synthesis from Exogenous Dengue Viral RNA Templates Requires Long Range Interactions between 5 * - and 3 * -Terminal Regions That Influence RNA Structure*

Viral replicases of many positive-strand RNA viruses are membrane-bound complexes of cellular and viral proteins that include viral RNA-dependent RNA polymerase (RdRP). The in vitro RdRP assay system that utilizes cytoplasmic extracts from dengue viral-infected cells and exogenous RNA templates was developed to understand the mechanism of viral replication in vivo . Our results indicated that in vitro RNA synthesis at the 3 * -untranslated region (UTR) required the presence of the 5 * -terminal region (TR) and the two cyclization (CYC) motifs suggesting a functional interaction between the TRs. In this study, using a psoralen-UV cross-linking method and an in vitro RdRP assay, we analyzed structural determinants for physical and functional interactions. Exogenous RNA templates that were used in the assays contained deletion mutations in the 5 * -TR and substitution mutations in the 3 * -stem-loop structure including those that would disrupt the predicted pseudoknot structure. Our results indicate that there is physical interaction between the 5 * -TR and 3 * -UTR that requires only the CYC motifs. RNA synthesis at the 3 * UTR, however, requires long range interactions involving the 5 * -UTR, CYC motifs, and the 3 * -stem-loop region that includes the tertiary pseudoknot structure. Dengue viruses type 1–4 are members of the flavivirus consisting the promoter sequence upstream 30 min. The reaction was terminated by phenol:CHCl 3 extraction followed by ethanol precipitation. RNA products were analyzed by denaturing formaldehyde/agarose gel electrophoresis (52). The se- quences of oligodeoxynucleotides were selected from a 19-nt region (between 231 and 250 nt from the 3 9 -end of 3 9 -UTR) for annealing. RNase H digestion would generate two fragments, 530 and 230 nt in length.

Dengue viruses type 1-4 are members of the flavivirus family of positive-strand RNA viruses. The diseases caused by dengue viruses range from a simple form of dengue fever to a more complex form of dengue hemorrhagic fever/shock syndrome, which exhibits considerable morbidity and mortality, especially among children in the tropical and subtropical regions of the world (for reviews see Refs. [1][2][3]. It is currently estimated that about 40% of the world's population living in these areas is at risk for dengue viral diseases, and about 5% of about one million dengue hemorrhagic/dengue shock syndrome cases are fatal (1). Dengue virus type 2 (DEN2) 1 is the most prevalent serotype identified over a 10-year period in epidemiological studies. DEN2 has a single-stranded RNA genome consisting of 10,723 nt (in New Guinea-C strain) (4) which encodes a single polyprotein arranged in the order, NH 2 -C-prM-E-NS1-NS2A-NS2B-NS3-NS4A-NS4B-NS5-COOH. The RNA genome contains a type I cap at the 5Ј-end but lacks a poly(A) tail at the 3Ј-end (5). The polyprotein precursor is processed in the endoplasmic reticulum, where the action of the cellular signal peptidase results in three structural proteins, the components of the virion, the capsid (C), precursor membrane protein (prM), and the envelope (E) protein (6 -8). The remainder of the polyprotein codes for at least seven nonstructural proteins that are generated by the combination of both the cellular signal peptidase and the virally encoded, twocomponent serine protease, NS2B/NS3.
The NS3 protein of flaviviruses has a serine protease catalytic triad within the N-terminal region of 180 amino acid residues (9, 10) that requires NS2B for protease activity (11)(12)(13)(14)(15)(16)(17)(18)(19)(20). The conserved motifs found in the NTP-binding proteins and the DEXH family of RNA helicases were also identified within a region from amino acid residue 160 to the C terminus of the NS3 protein (Refs. 10, 21, and 22 and for a review see Ref. 23). The RNA helicase activity of NS3 is thought to be required in flaviviral replication and is involved in unwinding of a putative double-stranded RNA replicative intermediate (5,24). In addition, the West Nile viral and dengue viral NS3 proteins exhibit 5Ј-RNA triphosphatase activity (25). 2 The largest of the flaviviral proteins, NS5, contains conserved motifs found in many viral RNA-dependent RNA polymerases (RdRP) (26,27), suggesting a role for NS5 in viral RNA replication. In addition to the RdRP motifs, NS5 has conserved motifs found in RNA methyltransferases (28,29) at the Nterminal region, although this activity has not been experimentally demonstrated for any flaviviral NS5. 5Ј-RNA triphosphatase, guanylyltransferase, and RNA methyltransferase are the three enzyme activities required for 5Ј-capping (30,31). Since NS3 and NS5 exist as a complex in dengue virus-infected cells (32), they are likely to be the components of not only the viral replicase but also the enzyme complex involved in 5Ј-capping.
Positive-strand viral RNA genomes contain recognition motifs for specific initiation of minus-and plus-strand RNA synthesis. These promoter elements are usually contained within the 5Ј-and 3Ј-terminal regions (TR), especially in the conserved stem-loop (SL) structures that are shown to be important for viral RNA replication in many (ϩ)-strand RNA viruses (33)(34)(35)(36)(37)(38)(39)(40)(41)(42)(43)(44)(45). In flavivirus RNAs, two conserved sequences (CS1 and CS2) within the 3Ј-untranslated region (UTR) and conserved SL structures and cyclization (5Ј-CYC and 3Ј-CYC) motifs within the 3Ј-and 5Ј-TRs have been identified by sequence and secondary structure prediction analyses (46 -49). Recent studies suggested a possible role for the 3Ј-SL structure in flavivirus viral RNA replication (45). The 3Ј-UTR of West Nile virus is known to interact with the translation elongation factor, eF-1␣ (50) although the biological significance of this interaction is not clear. In addition, the 3Ј-UTR of the Japanese encephalitis viral RNA was found to interact with NS3 and NS5, the putative components of the viral replicase (51) that exist as a heteromeric complex in infected cells (32,51).
To study the mechanism of viral replication in molecular detail, we reported the development of an in vitro assay system that utilized exogenous subgenomic RNA templates containing these conserved elements at the 5Ј-and 3Ј-TRs. A subgenomic RNA containing the 5Ј-and 3Ј-UTRs and the two CYC motifs was active as a template for RNA synthesis in vitro. RNA synthesis at the 3Ј-end of the 3Ј-UTR in the subgenomic RNA gave rise to two labeled products. The first was a doublestranded RNA that was resistant to RNase A under high ionic strength and had a mobility identical in size to the template RNA. The second product was twice the size of the input RNA template. This product was converted to a template size RNA species upon digestion with RNase A under high ionic strength. These results suggested that the 2ϫ product had a hairpin-like structure with an RNase A-sensitive single-stranded region and that this product was formed by a mechanism involving a snapback of the 3Ј-end of the template and elongation by the viral RdRP. Moreover, when the 3Ј-UTR RNA alone was used as a template, it did not form a 2ϫ hairpin product in the in vitro RdRP assay with infected C6/36 cell extracts. However, addition of 5Ј-TR 230nt RNA in trans activated RNA synthesis from the 3Ј-UTR resulting in the formation of this 2ϫ product of the 3Ј-UTR in the RdRP assay. Furthermore, it was shown that the CYC motifs are important for RNA synthesis in vitro because if the motif in the template 3Ј-UTR RNA or the activator 5Ј-TR 230nt RNA was mutated, RNA synthesis was significantly reduced but was restored when the mutant motifs were complementary to each other (52).
The results of our previous study supported the conclusion that a functional interaction between the 5Ј-TR and 3Ј-UTR is required for RNA synthesis at the 3Ј-end of the positivestranded template. In this study, we sought to determine whether there is any physical interaction between the 5Ј-TR and the 3Ј-UTR. Moreover, the 5Ј-TR RNA that we had used in our previous study contained the 5Ј-UTR as well as the 5Ј-CYC motif. Therefore, it remained to be established whether the 5Ј-UTR was required or only the 5Ј-CYC motif alone was sufficient for physical interaction with the 3Ј-UTR as well as for RNA synthesis. It also remained to be established whether the conserved 3Ј-SL structure of the 3Ј-UTR played any role in RNA synthesis in vitro. By using a psoralen-UV cross-linking method, we now show that there is physical interaction between the two ends of the dengue viral RNA that is dependent on the two cyclization motifs. However, our results also indicate that the interaction between the two ends alone is not sufficient for RNA synthesis in vitro at the 3Ј-UTR and that the 5Ј-UTR is also required for this process. For analyzing the importance of 3Ј-SL structure in RNA synthesis, we used the mutants that had previously been characterized for their replication efficiencies in vivo in the context of full-length infectious dengue viral RNAs (45). Some of these 3Ј-SL mutants were constructed by nucleotide substitutions that either main-tained or disrupted base pairings within the dengue viral 3Ј-SL; others were constructed by substituting different regions of the dengue viral 3Ј-SL with corresponding regions of West Nile virus 3Ј-SL structures. We engineered these 3Ј-SL mutations into the subgenomic RNAs. Analysis of these mutant subgenomic RNA templates in the in vitro RdRP assays revealed that the secondary structure as well as tertiary pseudoknot structure (53) of the 3Ј-SL region were required for efficient RNA synthesis. In general, the 3Ј-SL mutations that interfered with the infectivity of the genomic RNAs in vivo also had similar effects on the template efficiencies in the in vitro RdRP assays. A few 3Ј-SL mutations that were defective in viability in vivo or exhibited "host range" phenotype (45) did not affect template efficiency in vitro suggesting that the defect in vivo could be at a step other than the initial negative-strand RNA synthesis.

Preparation of DEN2-infected Cell Lysates
Aedes albopictus cells (C6/36 cells) were cultured as described (54,55). LLC-MK2 cells were cultured at 37°C in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Cells were infected by DEN2 (New Guinea C-strain) for 3 days using ϳ5 plaqueforming units/cell. Cell lysates from infected C6/36 or LLC-MK2 cells were prepared as described (52). The viral proteins, NS3 and NS5, were detected in the cell lysates by Western blotting using rabbit polyclonal antibodies, followed by a chemiluminescence detection system (ECL system from Amersham Pharmacia Biotech) (16,56).

Plasmid Constructs
Plasmids Containing cDNAs Encoding Subgenomic RNAs with Mutant CYC Motifs-The preparation of pSY-2 plasmid containing the wild type 5Ј-CYC and 3Ј-CYC motifs, TCAATATGC and GCATATTGA, respectively, and the PCR fragment containing the 5Ј-TR 230nt -mutCYC or the 3Ј-UTR 373nt mutCYC motif was as described (52). The PCR fragments were phosphorylated by polynucleotide kinase and cloned into pSP64 vector (Promega, Madison, WI) that was digested with SmaI and PvuII and dephosphorylated at the blunt ends to prevent religation. The pSY-1 plasmid (52) was digested with two sets of double digestions; for cloning the 5Ј-CYC motif, the plasmid was digested with EcoRI ϩ SalI, and for cloning the 3Ј-CYC motif, the plasmid was digested with NcoI ϩ PflM1. For construction of subgenomic RNA containing the 5Ј-CYCmut motif, the pSP64 intermediate plasmid was digested with EcoRI and SalI, and the 200-nt fragment containing the 5Ј-CYC was isolated; from the EcoRI ϩ SalI digest of pSY1 plasmid, the Sal fragment (950 nt) containing the wild type 3Ј-CYC motif and the EcoRI-SalI vector fragment (2.9 kb) were also isolated and ligated with the 200-nt fragment to yield the pSY-1-5Ј-CYCmut plasmid. Similarly, the NcoI-PflM1 fragment (225 nt) containing the 3Ј-CYCmut motif was isolated from the pSP64 intermediate plasmid clone and was ligated to the two fragments derived from the NcoI ϩ PflM1 digest of the pSY-1 plasmid, NcoI fragment (430 nt) and the vector fragment (3.3 kb). In both ligation reactions, the vector fragments of 2.9 and 3.3 kb were dephosphorylated prior to ligation. The pSY-2 plasmid derivatives (containing the cDNA coding for the 770-nt-long subgenomic RNA) were obtained from the pSY-1 plasmid constructs by digesting the respective plasmids with XmnI and Bsu36I, followed by treatment with the Escherichia coli Klenow fragment and ligation of the blunt ends. The subgenomic RNA construct containing both 5Ј-and 3Ј-CYCmut motifs was constructed by digesting the respective pSY-2 construct containing a single mutant motif with AlwN1 and NcoI. The 1.2-kb fragment containing the 5Ј-CYCmut motif was ligated to the vector fragment generated from the pSY-2 plasmid containing the 3Ј-CYCmut motif to yield the pSY-2-5Ј-3Ј-CYCmut plasmid.
Plasmid Constructs Containing cDNAs Encoding 3Ј-SL Mutant Subgenomic RNAs-In a previous study, several substitution mutations in the 3Ј-SL structure of DEN2 were generated by repositioning of the DEN2 sequences within the 3Ј-stem either without disrupting base pairing (D2-SL(a) and -(b) mutants) or with disruption of base pairing within the top half of the long stem (D2-SL(c) mutant) as well as by replacing different segments of 3Ј-SL with corresponding regions from West Nile virus (WN) 3Ј-SL. The effects of these mutations in the context of DEN2-infectious clones were analyzed by their replication efficiencies and infectivities in C6/36 and LLC-MK2 cells (45). To analyze the effects of these 3Ј-SL mutations on the exogenous RNA tem-plate efficiencies in the in vitro RdRP assays, plasmids encoding the 3Ј-SL mutant 3Ј-UTR RNAs and subgenomic RNAs containing the 3Ј-SL mutations were constructed by PCR. The 3Ј-SL mutants of the 3Ј-UTR were first constructed by using the 5Ј-primer consisting of the T7 promoter sequence (underlined) upstream of the 3Ј-UTR start site, 5Ј-TAATACGACTCACTATAGAAGGCAAAACTAACATGAAAC-3Ј, and a specific 3Ј-primer depending on the mutant. Four different 3Ј-end primers used for PCR were as follows: D2 3Ј, 5Ј-AGAACCTGTTGATT-CAA-3Ј; WN 3Ј, 5Ј-AGATCCTGT GTTCTC-3Ј; mutE 3Ј, 5Ј-AGAACCC-TGTGTTCTC-3Ј; and mutF 3Ј, 5Ј-AGATCCGTTGATTCAA-3Ј. To amplify the wild type D2-SL, D2/WN-SL(mutA), -SL(mutD), D2-SL(a), -SL(b), -SL(c), and D2-SL(C71G) mutants, the 5Ј-primer and D2 3Јprimer were used. For the D2/WN-SL, D2/WN-SL(mutB), and -SL-(mutC) mutants, the WN 3Ј-primer was used. To amplify the D2/WN-SL(mutE) and -SL(mutF) mutants, the mutE 3Ј-and mutF 3Ј-primers were used, respectively. PCR was for 30 cycles in the following order: 94°C for 1.5 min, 40°C for 1.5 min, and 72°C for 2 min. To increase the efficiency of in vitro transcription, a second PCR was carried out with a different 5Ј-primer containing the T7 promoter (underlined), 5Ј-TTAC-GAATTCGAGCTCGCCCTAATACGACTCACTATAG-3Ј, and the same 3Ј-primer that was used for the first PCR of each mutant 3Ј-SL. The sequences upstream of the T7 promoter were found to increase the efficiency of in vitro transcription probably by increasing the stability of T7 RNA polymerase interaction with the promoter.
The amplified PCR products were purified from agarose gel by Qiagen gel extraction kit. Plasmids encoding the subgenomic RNAs containing 3Ј-SL mutations were constructed by cloning the PCR products as follows. The 5Ј-end of PCR products was phosphorylated by T4 polynucleotide kinase and subcloned into the SmaI-digested and dephosphorylated pSP64 vector. Clones that contained the 3Ј-UTR inserts upstream of the SP6 promoter were selected (series pUPT7-X clones where "X" is a specific 3Ј-SL mutation). Each intermediate clone was cleaved with ApaI and XbaI, and the ϳ200-nt fragment was cloned into the ApaI/XbaI-digested pSY-2 plasmid (52). The regions of the 3Ј-SL structure replacements in the subgenomic RNA constructs were confirmed by DNA sequencing.
Plasmid Constructs Containing the Pseudoknot Mutation in the 3Ј-SL-In order to disrupt the potential pseudoknot structure in WN-3-SL structure (WN-3ЈSL(C73G)), a site-directed mutagenesis was performed by PCR. Two PCR products were generated with overlapping sequences using two sets of primers. PCR1 was obtained using the 5Ј-primer (T7-3ЈUTR), 5Ј-TAATACGACTCACTATAGAAGGCAAAAC-TAACATGAAAC-3Ј that corresponds to the T7 promoter (underlined) followed by the first 21 nt of 3Ј-UTR, and 3Ј-primer (C73G(A)), 5Ј-TTGTGCAGAGCACAAGATCTCCTAG-3Ј. PCR2 was produced using the 5Ј-primer (C73G(B)), 5Ј-CTAGGAGATCTTGTGCTCTGCACAA-3Ј and 3Ј-WN primer, 5Ј-AGATCCTGTGTTCTC-3Ј. The underlined complementary nucleotides in bold letters in C73G(A/B) primers represent the mutated pseudoknot nucleotide C-G. The two products, PCR1 and PCR2, were mixed and used for a third PCR in the presence of the primer set T7-3Ј-UTR and 3Ј-WN primers. The final PCR product was purified and used as template for the second PCR as described above. The PCR product was treated with T4 polynucleotide kinase. This PCR fragment was first subcloned into pSP64-SmaI/PvuII cut plasmid. The intermediate plasmid was digested by ApaI/XbaI and subcloned into the ApaI/XbaI-digested pSY-2 plasmid to yield the plasmid encoding the subgenomic RNA containing the pseudoknot mutation. The region of the mutation was confirmed by DNA sequencing.

Preparation of DNA Fragments for Use as Templates in the in
Vitro RNA Transcription by T7 RNA Polymerase PCR Products Encoding the Subgenomic RNA 770nt (WT and CYC Mutants)-To generate the subgenomic RNA 770nt WT as well as the CYC motif mutants without any additional 3Ј-terminal nucleotides, PCR was performed using a plasmid construct encoding a specific subgenomic RNA as a template and two primers as follows: the 5Јprimer containing the EcoRI site, 5Ј-AGCTATGACCATGATTAC-GAATTC-3Ј, that corresponds to the sequences upstream of the T7 promoter, and the pSP64 vector sequence in each plasmid and the 3Ј-primer, 5Ј-AGAACCTGTTGATTCAACAGCACC-3Ј, that anneals with the 3Ј-end region of the 3Ј-UTR. PCR was performed for 35 cycles of each step as follows: 94°C for 1.5 min, 66°C for 1.5 min, and 72°C for 2 min.
PCR Product Encoding the Subgenomic RNA 674nt (without 5Ј-UTR)-The DNA template for preparation of the subgenomic RNA 674nt that lacks the 5Ј-UTR (96 nt) but is otherwise identical to the subgenomic RNA 770nt was prepared by PCR using pSY-2 as the template and the two primers as follows: the 5Ј-primer, 5Ј-TAATACGACTCACTATAGA-TGAATAACCAACGAAAAAAGGCG-3Ј, containing the T7 promoter (underlined), and the region beginning with the downstream capsid coding sequence. The 3Ј-primer contains the complementary sequence of the 3Ј-end of the 3Ј-UTR. A second PCR was performed using the 5Ј-primer described under "Plasmid Constructs" and the same 3Јprimer as mentioned above.
PCR Product Encoding the 5Ј-TR 230nt RNA-PCR was carried out with the pSY-2 as a template and two primers: the 5Ј-primer, same as under "Preparation of DNA Fragments for Use as Templates in the in Vitro RNA Transcription by T7 RNA Polymerase," and the 3Ј-primer, 5Ј-TGAGGTCCTCGTCCTG-3Ј (the underlined sequence is in the vicinity of Bsu36I site).
PCR Fragment for the 5Ј-TR 130nt RNA Transcript-To generate 5Ј-TR 130nt RNA without the 5Ј-UTR, PCR was performed as described for the production of the 5Ј-TR 230nt PCR fragment except that the 5Ј-primer used for making the subgenomic RNA 674nt (see under "PCR Product Encoding the Subgenomic RNA 674nt (without 5Ј-UTR)") was employed.
Production of PCR Products Containing Mutations in the 3Ј-SL Region-To generate the subgenomic RNA templates with various mutations in the 3Ј-SL region, the corresponding plasmid construct described above was used as a template for PCR using the same 5Ј-primer as used for the subgenomic RNA 770nt described above along with an appropriate 3Ј-primer. Again four different 3Ј-end primers (see "Plasmid Constructs Containing cDNAs Encoding 3Ј-SL Mutant Subgenomic RNAs") were used for PCR. All PCR products were purified from an agarose gel using the Qiagen gel extraction kit.

Preparation of in Vitro RNA Transcripts
In vitro transcription was performed at 37°C for 3 h in a 100-l reaction volume as described by the manufacturer (Promega, Madison, WI) (see also Ref. 52 except 40 units of T7 RNA polymerase was used). The in vitro transcripts were quantitated spectrophotometrically, and their integrity was verified by electrophoresis on 3.5% polyacrylamide, 7 M urea gel followed by staining with acridine orange.

In Vitro RdRP Assay with Exogenous RNA Transcripts
The RdRP assays using the DEN2-infected A. albopictus (C6/36) or the LLC-MK2 cell extracts and the various exogenous RNA templates were carried out as described (52). In the RdRP assays, where the various 5Ј-TR RNAs were added in trans to the reaction mixtures containing the 3Ј-UTR 454nt RNA template (3 g), the RNAs were added in equimolar amounts. The template efficiencies of the various 3Ј-UTR 454nt RNAs containing different 3Ј-SL mutations were assayed by incubating 3Ј-UTR 454nt RNAs (3 g each) with 1.2 g of the 5Ј-TR 180nt RNA. The reactions were terminated by phenol extraction (phenol: chloroform, 5:1, pH 4.5), followed by chloroform extraction (chloroform: isoamyl alcohol, 24:1), and the unincorporated radiolabeled CTP was removed by passage through a G-30 Spin Column (Bio-Rad). After ethanol precipitation, the RdRP assay products were analyzed on a 1.5% agarose-formaldehyde gel followed by autoradiography (52). In the RdRP assays where the subgenomic RNA 770nt containing the various 3Ј-SL mutations was used as templates, 3 g of each of the RNAs were used per assay.

Analysis of the Polarity of the RdRP Product by RNase A Digestion followed by RNase H Mapping
The RdRP reaction was first carried out under standard conditions at 30°C for 1.5 h using the infected C6/36 extracts, and the RNA products were purified by phenol:CHCl 3 extraction and ethanol precipitation. RNA pellet was then resuspended in 20 l of H 2 O and purified by using the G-30 Spin Column as described above. An aliquot of RNA was digested with RNase A under high ionic strength conditions as described (52) except in a 50-l reaction mixture containing 300 ng of RNase A. The RNase A-digested sample was phenol:CHCl 3 -extracted and precipitated with ethanol. This step, expected to remove the singlestranded loop region of the hairpin, was found to be necessary before annealing with the oligodeoxynucleotides followed by RNase H digestion. Aliquots of RNA sample were resuspended in RNase H reaction mixture (29 l) containing 50 mM Tris-HCl, pH 8, 0.1 M NaCl, 2 mM MgCl 2 , 1 mM dithiothreitol, 6.6 M oligodeoxynucleotide of either positive-or negative-strand polarity, and formamide (30% final concentration). The reaction mixture was heated at 95°C for 2 min and slowly cooled down to 65°C by turning off the heating source. It was then chilled in ice, and 3 units of RNase H was added and incubated at 37°C for 30 min. The reaction was terminated by phenol:CHCl 3 extraction followed by ethanol precipitation. RNA products were analyzed by denaturing formaldehyde/agarose gel electrophoresis (52). The sequences of oligodeoxynucleotides were selected from a 19-nt region (between 231 and 250 nt from the 3Ј-end of 3Ј-UTR) for annealing. RNase H digestion would generate two fragments, 530 and 230 nt in length.

RNA:RNA Interaction Analysis
Preparation of a Radiolabeled RNA Template-The various 5Ј-TR RNAs were radiolabeled during in vitro transcription. DNA template (1 g) was mixed with 20 l of 1ϫ reaction buffer for in vitro transcription as mentioned above, except containing 12.5 M CTP, 30 Ci of [␣-32 P]CTP (800 Ci/mmol), and 10 units of T7 RNA polymerase. The reaction mixture was incubated at 37°C for 1 h. Then, DNase I (1 unit) was added, and the reaction was incubated at 37°C for 15 min to digest the DNA template. Labeled RNA was purified by phenol:chloroform extraction and by using a Micro-spin P-30 column (Bio-Rad) followed by ethanol precipitation. The radiolabeled RNA transcript was run on a polyacrylamide (4%), 7 M urea gel and eluted from the gel as described (52). The radioactivity of the eluted RNA was measured by a scintillation counter.
Psoralen/UV Cross-linking-To study the physical interaction between the 5Ј-TR RNA and 3Ј-UTR RNA, the psoralen/UV cross-linking method was used (57). Briefly, the radiolabeled 5Ј-TR RNA (0.3 ϫ 10 4 cpm) and 5 g of the unlabeled 3Ј-UTR RNA were denatured at 90°C for 2 min and chilled quickly on ice. Then, 25 l of annealing buffer (50 mM sodium cacodylate, 300 mM KCl) was added, and the mixture was incubated at 70°C for 20 min. The samples were supplemented with 5 mM MgCl 2 , and were incubated at 37°C for an additional 15 min. The samples were kept on ice until 2 l of AMT-psoralen (4Ј-aminomethyl-4Ј,5Ј,8Ј-trimethylpsoralen; 1 mg/ml in RNase-free water; Sigma) was added, followed by further incubation on ice for 15 min in the dark. The samples were transferred to a 96-well flat bottom plate on ice. The plate was exposed to UV irradiation (365 nm) at a distance of 2-3 cm for 30 min. The samples were transferred back to microcentrifuge tubes, followed by phenol extraction and ethanol precipitation. The precipitated and cross-linked RNA samples were analyzed on a formaldehyde (2.2 M)-agarose (1.5%) gel. The gel was dried under vacuum and was subjected to autoradiography.

Secondary Structure Modeling of RNA
The predicted secondary structures of the mutant subgenomic RNAs were analyzed by using the RNA structure algorithm developed by Zuker et al. (58). Among the predicted structures for each RNA, the most stable forms based on ⌬G values were chosen for comparisons.

Analysis of the Polarity of the RdRP Product by RNase H
Digestion-In a previous study (52), we reported the first in vitro RdRP assay that utilizes exogenous subgenomic DEN2 RNA templates and membrane-bound cytoplasmic extracts from the DEN2-infected mosquito (C6/36) cells. By using a subgenomic RNA template that contained wild type 5Ј-and 3Ј-UTRs and CYC motifs in this assay system, we found that two RNA products were formed as follows: one had the same mobility and size as the input RNA template, and the second product had a faster mobility on a partially denaturing polyacrylamide, 7 M urea gel but a mobility that was consistent with a size twice that of the template RNA on a completely denaturing formaldehyde-agarose gel. The RNA product that was twice the size of the input RNA (2ϫ) was converted to a product with the same mobility as the template RNA (1ϫ) upon digestion with RNase A. These results suggested that 2ϫ product was a double-stranded hairpin RNA formed via a snapback mechanism by the 3Ј-end elongation of the input RNA template. The 1ϫ RNA species was also resistant to RNase A suggesting that this double-stranded RNA product was formed either by de novo initiation of RNA synthesis by RdRP at the 3Ј-end of the template RNA or it was formed by a cleavage of the 2ϫ hairpin product by a nuclease present in the cytoplasmic extract at the single-stranded loop region. Since the predominant RNA product that was formed in the in vitro RdRP assay was the 2ϫ hairpin product, the structural requirements of the template for the formation of the 2ϫ product were analyzed in this study.
First, the polarity of the radiolabeled RNA formed in the RdRP reaction carried out using the subgenomic RNA 770nt template of (ϩ)-polarity was determined by RNase H mapping. The oligodeoxynucleotides were designed to anneal with the region 231-250 nt from the 3Ј-end of the genomic RNA so that RNase H digestion would be expected to yield fragments of 520 and 230 nt in length. However, denaturation of the covalently linked double-stranded RNA hairpin followed by annealing with an oligodeoxynucleotide of either polarity followed by RNase H digestion did not produce any discrete RNA products suggesting that complete denaturation of the highly structured RNA was not attained under these conditions. Therefore, the 2ϫ RNA product formed in the standard RdRP reaction was first digested with RNase A under high salt conditions in order to remove the single-stranded loop region that connects the unlabeled input RNA with the newly synthesized complementary RNA strand. RNase A treatment would produce a doublestrand RNA not covalently linked which would facilitate denaturation and annealing with the oligodeoxynucleotides (Fig. 1,  panel A). Inclusion of 30% formamide in the annealing reaction was also found to be necessary for optimum annealing of RNA-DNA hybrids and that under these conditions RNase H activity was not affected.
The RdRP product (Fig. 1, panel B, lane 1) was digested by RNase A under high salt conditions. Equal aliquots were digested with RNase H after annealing with oligodeoxynucleotide of either (ϩ)-or (Ϫ)-polarity or in the absence of any oligodeoxynucleotide as a negative control. The RNA product without the addition of any oligodeoxynucleotide was not susceptible to RNase H (Fig. 1, panel B, lane 2). However, the RNA product upon digestion with RNase H gave rise to essentially two distinct fragments of 520 and 230 nt after annealing with oligodeoxynucleotide of (ϩ)-polarity (panel B, lane 3) but not with (Ϫ)- oligodeoxynucleotide (panel B, lane 4). These results indicated that the labeled strand from the RdRP reaction was of (Ϫ)-polarity, which is complementary to the input (ϩ)-strand RNA template. The radiolabeled (ϩ)-RNA annealed with the (Ϫ)-oligodeoxynucleotide served as a positive control for RNase H digestion (data not shown).
5Ј-TR RNA and 3Ј-UTR RNA Physically Interact via the Cyclization Motif-In a previous study, we showed that in vitro RNA synthesis by the cytoplasmic extract from dengue virusinfected cells that required specifically the 5Ј-and 3Ј-TRs of the dengue virus RNA be present in the same template RNA molecule (in cis) or that the 5Ј-TR 230nt RNA be added in trans to the 3Ј-UTR 373nt RNA. However, the 3Ј-UTR 373nt RNA alone was not active to form the 2ϫ RNA product. The results of that study suggested that there is a functional interaction between the 5Ј-and 3Ј-TRs that conferred to the inactive 3Ј-UTR 373nt RNA the ability to serve as an efficient template for RNA synthesis resulting in the formation of the 2ϫ product in vitro. Similar results were also obtained when the full-length 3Ј-UTR 454nt and the 5Ј-TR 230nt RNA were both present in the RdRP assay (data not shown).
In this study, we investigated whether there is any physical interaction between the 5Ј-TR and 3Ј-UTR RNA by using the psoralen-UV cross-linking method. Psoralen intercalates between base pairs, and upon UV treatment it cross-links opposite strands especially at pyrimidine bases. The psoralen-UV method has been used extensively to study RNA-RNA and RNA-protein interactions (57, 59 -64). To determine whether there is a physical interaction between the two ends of DEN2 RNA, the 5Ј-TR 230nt RNA was radiolabeled with 32 P in the in vitro transcription reaction and purified from the polyacrylamide 7 M urea gel. It was then mixed with unlabeled 3Ј-TR 454nt RNA and treated with psoralen, followed by UV irradiation. The cross-linked products were analyzed by formaldehyde-agarose gel electrophoresis and autoradiography. The results shown in Fig. 2, panel A indicate that the formation of cross-linked products was observed only when the 5Ј-TR 230nt and the 3Ј-UTR 454nt RNAs containing the wild type or mutant CYC motifs, which were complementary to each other, were present in the psoralen/UV cross-linking reaction (lanes 3 and 10). The cross-linked products were significantly reduced under conditions in which one of the CYC motifs was mutated which would disrupt complementarity between the two CYC motifs (lanes 5 and 8). No cross-linked products were formed in control reactions omitting the psoralen (lanes 2, 4, and 9) or reactions containing only the 5Ј-TR 230nt RNA but treated with psoralen/UV (lanes 1 and 6). In the reaction mixtures containing only the labeled 5Ј-TR RNAs, treatment with psoralen/UV resulted in loss of radioactivity probably due to degradation, and the loss was greater for 5Ј-TR 230nt RNA than for 5Ј-TR 130nt RNA (lanes 1 and 6, panel A versus B). These results taken together indicated that there was a physical interaction between the 5Ј-TR and 3Ј-UTRs in the viral subgenomic RNA and that the CYC motifs played an important role in facilitating this interaction. Moreover, this interaction still occurred even if CYC motifs were mutated as long as the complementarity between them was maintained.
Next, we investigated whether a shorter RNA from the 5Ј-TR containing the CYC motifs but without the 5Ј-UTR could still physically interact with the 3Ј-UTR 454nt RNA. The results of the psoralen/UV cross-linking experiment shown in Fig. 2,  panel B (lane 3) indicated that the 5Ј-TR 130nt RNA containing the 5Ј-CYC motif alone was sufficient for interaction with the 3Ј-UTR 454nt RNA. Again, substitution mutations in the 5Ј-CYC motif did not abolish interaction as long as the complementary mutations were introduced into the 3Ј-CYC motif (lane 10). In control experiments, it was found that omission of either 3Ј-UTR 454nt RNA (lanes 1 and 6) or psoralen/UV treatment (lanes 2, 4, 7, and 9) did not give any cross-linked products. Substitution mutations of either of the CYC motifs in the two RNAs significantly reduced the ability to form cross-linked products although they did not abolish cross-linking completely (see lanes 5 and 8). In both experiments shown in Fig. 2, panels A and B, there were two major species of cross-linked products formed suggesting that psoralen cross-linking occurred at more than one site. The results presented thus far indicated that the FIG. 2. Analysis of RNA/RNA interaction. Physical interaction between two RNAs was analyzed by the psoralen/UV cross-linking method which is described under "Experimental Procedures." The 3Ј-UTR 454nt RNA (thick dotted line) which had either wild type (W) or mutant (M) 3Ј-CYC motif was unlabeled. The 5Ј-TR 230nt RNA is indicated by a thick solid line with a solid rectangular box denoting the 5Ј-UTR. Thick rectangular vertical bars in 3Ј-UTR and 5Ј-TR RNAs denote CYC motifs. An X indicates the mutant CYC motif. An asterisk denotes a radiolabeled RNA used for the cross-linking experiment. Panel A, interaction between *5Ј-TR 230nt RNA and 3Ј-UTR 454nt RNA. Lanes 1 and 6, *5Ј-TR 230nt wt and 5Ј-TR 230nt /cycMUT, respectively, after psoralen/UV treatment. Lanes 2 and 3, *5Ј-TR 230nt /wt ϩ 3Ј-UTR 454nt /wt without and with psoralen/UV treatment, respectively. Lanes 4 and 5, *5Ј-TR 230nt /wt ϩ 3Ј-UTR 454nt /cycMUT without and with psoralen/UV treatment, respectively. Lanes 7 and 8, *5Ј-TR 230nt /cyc-MUT ϩ 3Ј-UTR 454nt /wt without and with psoralen/UV treatment, respectively. Lanes 9 and 10, *5Ј-TR 230nt /cycMUT ϩ 3Ј-UTR 454nt /cycMUT without and with psoralen/UV treatment, respectively. Although the same amount of labeled 5Ј-TR RNA was used for each experiment, treatment of labeled 5Ј-TR RNA alone with psoralen/UV seemed to result in loss of labeled RNA which might be attributable to degradation (lanes 1 and 6; see also panel B). Panel B, interaction between 5Ј-TR 130nt RNA and 3Ј-UTR 454nt . The experiments were carried out as described in panel A except that labeled 5Ј-TR 130nt RNA (*) was used. The order of samples loaded was the same as in panel A.
interaction between the 5Ј-and 3Ј-TR was facilitated by the CYC motifs, and that the 130-nt region containing the CYC motif alone in the absence of the 5Ј-UTR was sufficient for physical interaction with the 3Ј-UTR 454nt RNA. The 5Ј-UTR 96nt RNA alone, lacking the 5Ј-CYC motif, did not interact with the 3Ј-UTR 454nt RNA to form the psoralen cross-linked product (data not shown).
5Ј-UTR Is Crucial for RdRP to Initiate RNA Synthesis-Next, we sought to determine whether this 5Ј-TR 130nt RNA can transactivate the 3Ј-UTR 454nt RNA in the in vitro RdRP assay. To address this question, in addition to the 5Ј-TR 230nt RNA, we used three different RNA templates in the in vitro assays as follows: the 5Ј-TR 180nt RNA containing 5Ј-UTR and the CYC motif but shorter at the 3Ј-end, the 5Ј-TR 130nt RNA lacking the 5Ј-UTR region, and the 5Ј-UTR 96nt lacking the CYC motif. The 3Ј-UTR 454nt RNA alone was inactive in the formation of the 2ϫ product in the RdRP assay as reported earlier (Fig. 3, lane 1). However, a labeled RNA species of the same size as the input template RNA (1ϫ) was formed (lane 1). The 3Ј-UTR 454nt RNA alone in the absence of 5Ј-TR was radiolabeled with varied efficiency in the RdRP assays with different infected C6/36 extracts or with uninfected cell lysates (data not shown; see also Ref. 52). The formation of this labeled (1ϫ) RNA species is likely due to addition of a few nucleotides to the 3Ј-end of the template strand by the host terminal nucleotidyltransferase (a membrane-bound enzyme), and this 1ϫ product was sensitive to RNase A digestion under high ionic strength conditions; however, the 1ϫ RNA product formed from the 3Ј-UTR RNA template in the presence of 5Ј-TR 230nt RNA or the subgenomic RNA template was resistant to RNase A (52). Furthermore, the results shown in Fig. 3 indicate that the 5Ј-TR 230nt and the 5Ј-TR 180nt RNA could both transactivate the 3Ј-UTR 454nt RNA (lanes 7 and 9), but the 5Ј-TR 130nt and the 5Ј-UTR 96nt RNAs were essentially inactive in conferring RNA synthesis at the 3Ј-end of the 3Ј-UTR 454nt RNA to form the 2ϫ product (lanes 3 and 5). In control RdRP assays containing only 5Ј-TR RNA species as templates, the labeled 1 and 2ϫ products of the 5Ј-TR RNAs (lanes 2, 4, 6, and 8) were also formed, and the former was more predominant than the latter species. This observation was consistent with our previous study (52) which also reported that blocking the 3Ј-end of the 5Ј-TR 230nt RNA abolished its template activity but not its ability to transactivate the 3Ј-UTR 454nt RNA template for 3Ј-end elongation. However, the presence of 3Ј-UTR 454nt and 5Ј-TR RNAs together in the RdRP assay resulted in significant suppression of RNA synthesis at the 5Ј-TR RNA templates for reasons unknown (lanes 2 and 4 versus 3 and 5). Our results shown in Fig. 3, taken together, indicated that although the 5Ј-TR RNA 130nt containing the CYC motif alone was sufficient for physical interaction with the 3Ј-UTR 454nt (Fig. 2, panel B), it was not sufficient for activation of RNA synthesis at the 3Ј-end of 3Ј-UTR 454nt RNA template, and both the 5Ј-UTR and the 5Ј-CYC motifs are required.
In the previous study, we demonstrated a functional interaction between the 5Ј-TR and the 3Ј-UTR for RNA synthesis at the 3Ј-end when both regions are in the same RNA template such as the subgenomic RNA 770nt that includes the CYC motifs. To confirm further the role of 5Ј-UTR and CYC motifs, four mutant subgenomic RNA 770nt templates were used in the RdRP assays as follows: subgenomic RNA 674nt with a deletion of the 5Ј-UTR, two subgenomic RNAs 770nt containing the same substitution mutations as described before within either the 5Ј-CYC or the 3Ј-CYC motif which would disrupt base pairings, and the subgenomic RNA 770nt in which both CYC motifs were mutated such that the complementarity was restored (Fig. 4).
The template efficiencies of these mutant subgenomic RNAs were analyzed using the in vitro RdRP assay. The results indicated that the subgenomic RNA 674nt (without the 5Ј-UTR) had significantly reduced template activity for RNA synthesis (Fig. 4, panel A). These results support the conclusion reached from the transactivation assays that the 5Ј-UTR is necessary for the viral RdRP to initiate RNA synthesis at the 3Ј-end of the RNA template (Fig. 3). Next, the contribution of the two CYC motifs for RNA synthesis at the 3Ј-end of subgenomic RNA templates was examined. The subgenomic RNA 770nt , containing either of the CYC motifs mutated or both mutated, were used as templates in RdRP assays. The results indicate that the double mutant subgenomic RNA containing complementary mutant CYC motifs was an equally efficient template as the wild type RNA for RNA synthesis (Fig. 4, panel B, lanes 1 and 4). Furthermore, the subgenomic RNA with the mutated 5Ј-CYC motif showed a dramatic reduction in template efficiency compared with the wild type (lane 2). These results are consistent with the conclusion reached in the previous study based on the transactivation assays (52). However, the subgenomic RNA with the 3Ј-CYC mutation showed a reduced template activity (about 60% of the wild type) (lane 3 versus lane 1). But this result was contrary to the results obtained in the transactivation assays in that the 5Ј-TR with wild type CYC motif was essentially inactive in promoting RNA synthesis from the 3Ј-UTR RNA containing the mutant 3Ј-CYC motif (52). This experiment was repeated four times, and similar results were obtained. One

FIG. 3. Transactivation activities of various 5-TR RNAs.
In order to map the minimum region of 5Ј-TR RNA which is necessary for transactivation of the 3Ј-UTR 454nt for in vitro RNA synthesis, three truncated mutant 5Ј-TR RNAs were constructed. The RdRP products were analyzed on a formaldehyde-agarose gel (1.8%), followed by autoradiography. Lane 1, 3Ј-UTR 454nt RNA alone. Lanes 2, 4, 6, and 8, 5Ј-UTR 96nt , 5Ј-TR 130nt , 5Ј-TR 180nt , and 5Ј-TR 230nt RNAs alone, respectively. Lanes 3, 5, 7, and 9, 3Ј-UTR 454nt RNA in the presence of the 5Ј-UTR 96nt , 5Ј-TR 130nt , 5Ј-TR 180nt , and 5Ј-TR 230nt RNA, respectively. The labeled 1 and 2ϫ products produced from the unlabeled 3Ј-UTR 454nt RNA are indicated by arrows. The unlabeled 5Ј-TR RNAs also served as templates in the in vitro RdRP assay, and the 1 and 2ϫ products are indicated as closed and open dots, respectively. In lane 7, 1ϫ of 3Ј-UTR 454nt and the 2ϫ of 5Ј-TR 180nt were separated, whereas in lane 9, the 3Ј-UTR 454nt RNA band was not well resolved from the 2ϫ product of 5Ј-TR 230nt RNA. The presence of 3Ј-UTR RNA significantly inhibited the template activity of the 5Ј-TRs (see lanes 3, 5, 7, and 9). possible explanation for this difference is that the interaction between the 5Ј-TR containing the wild type 5Ј-CYC motif with the 3Ј-UTR RNA containing the mutant 3Ј-CYC motif is intramolecular and may be more favorable as the two ends are in the same molecule compared with the scenario in which the interaction between the two terminal regions of RNAs is intermolecular as is the case with 5Ј-TR 230nt /wtCYC and the 3Ј-UTR 373nt /mutCYC used in the transactivation assays.
Importance of 3Ј-Stem-loop Structure in the 3Ј-UTR for RNA Synthesis in Vitro by RdRP-The 3Ј-most 90 -100 nucleotides of the flavivirus RNA genomes have been predicted to form a stable stem-loop (SL) structure (Fig. 5, panels A and B) (40,45,46,48,53,(65)(66)(67)(68). Although the primary sequences of the region are less homologous, the 3Ј-SL structures are highly conserved throughout the flavivirus family. This implies that the conserved 3Ј-SL structure might have functional importance for viral replication.
Recently, Zeng et al. (45) reported the functional role of the 3Ј-SL structure in DEN2 viral replication and infectivity. They constructed several substitution mutations within the DEN2 3Ј-SL either by rearrangement of base pairings that would disrupt or maintain secondary structure or by replacing all or part of the 3Ј-SL of the DEN2 genome with the corresponding regions of the West Nile 3Ј-SL, without disrupting the overall secondary structure. These substitution mutations were cloned into the full-length DEN2 infectious clone (diagrammatically shown in Fig. 5, panel C). In that study, monkey kidney (LLC-MK2) cells were electroporated with the genome length RNA transcript from each mutant clone, and virus production and the spread of infection was followed over time by immunofluorescence assay (45). The ability of the viable mutant viruses to replicate in the mosquito (C6/36) cells and monkey kidney cells (LLC-MK2) was also examined. From the analysis of infectivity of these RNA transcripts, it was concluded that the stem structure of the top half and the nucleotide sequence of the 11 base-paired region in the uppermost portion of the bottom half of the 3Ј-SL are important for efficient viral replication (45). Furthermore, a point mutation that would disrupt the predicted pseudoknot structure within the 3Ј-SL (53) in the context of infectious clone was lethal for viral replication and infectivity in vivo 3 (C71G; see Fig. 5, panel A). Four of the first 6 nt of the WN small loop region also had the potential to form a pseudoknot structure with nucleotides 71-74 in the long stem (53), suggesting an interesting possibility that disruption of this tertiary structure interaction could account for the loss of replication potential of the chimeric mutants in this part of the viral genome. 3 L. Zeng, B. Falgout, and L. Markoff, personal communication.

FIG. 4. In vitro
RdRP assay with various subgenomic mutant RNA templates. RNA templates were produced by in vitro transcription as described under "Experimental Procedures." The integrity of RNAs were verified by formaldehyde, 1.8% agarose gel electrophoresis and visualized by staining with acridine orange. Panel A, analysis of the template efficiency of the subgenomic RNA 674nt (without the 5Ј-UTR). Lane 1, subgenomic RNA 770nt WT. Lane 2, subgenomic RNA 674nt containing the deletion of 5Ј-UTR. Panel B, analysis of the template efficiency of the subgenomic RNA 770nt WT, 5Јcyc MUT, 3ЈcycMUT, and 5Ј3Јdouble cycMUT (lanes 1-4, respectively). To study the effects of these substitutions within the DEN2 3Ј-SL structures on their template efficiencies in the in vitro RdRP assay system, we subcloned these 3Ј-SL mutations into the plasmid coding for the subgenomic RNA. RNAs from these constructs were produced by in vitro transcription and were used as templates in the in vitro RdRP assays using DEN2infected mosquito (C6/36) as well as monkey kidney (LLC-MK2) cell lysates. The rationale for using these two cell lysates was to examine whether there are any differences in host restriction for RNA synthesis in vitro as was found previously with one mutant (mutF) in the in vivo infectivity assays (45).
As shown in Fig. 6, the 3Ј-SL mutant subgenomic RNAs showed template efficiency at different levels. Subgenomic RNAs with the D2/WN-SL(mutA), D2-SL(a), D2-SL(b), D2/WN-SL, D2/WN-SL(mutB), and D2/WN-SL(mutF) mutations exhibited template efficiency either close to or only slightly reduced compared with the wild type template in the in vitro RdRP assays that utilized either C6/36 or LLC-MK2 cell extracts for the source of viral replicase (Fig. 6, panels B and C, lanes 1, 3 , 4, 7, 8, and 11 versus 12). Of these mutants, mutagenesis of the top half of the DEN2-3Ј-SL either completely or partially, as in D2/WN-SL(mutA), D2-SL(a), and D2-SL(b) (see Fig. 5, panel C), had only modest effects on virus growth compared with the wild type DEN2 genome (45). The in vitro template efficiencies also support this finding (Fig. 6, panels B and C, lanes 1, 3 and  4). In contrast, D2/WN-SL and D2/WN-SL (mutB) behaved like the wild type in the in vitro assays, whereas they were severely affected for growth in vivo (45). Moreover, the mutants D2/WN-SL(mutD) and D2/WN-SL(mutF) exhibited similar template activities with the C6/36 extract which were closer to that of the wild type template (Fig. 6, panel B, lanes 2 and 11 versus  lane 12). On the other hand, with the LLC-MK2 cell extract, the -(mutD) was less active than -(mutF), but both mutants were less active than the wild type template in the in vitro assay (Fig. 6, panel C, lane 2 versus 11; compare lanes 2 and 11 with lane 12). The -(mutF) exhibited a "host-range" phenotype in vivo in that it was severely restricted for replication in C6/36 cells but grew like the wild type in LLC-MK2 cells (45). In contrast, the results of the in vitro RdRP assays showed little difference in template efficiencies with the extracts from LLC-MK2-or C6/36-infected cells.
Template efficiency of the subgenomic RNAs containing D2/ WN-SL(mutC) and -(mutE) mutations exhibited a reduced template efficiency for RNA synthesis using either the LLC-MK2or the C6/36-infected cell extract (Fig. 6, panels B and C, lanes  9 and 10). Consistent with these results, these mutants exhibited either no DEN-positive antigens (not viable) or only 10% of the cells were positive for viral antigen (barely viable) as visualized by immunofluorescence of DEN2-transfected LLC-MK2 cells (45). In -(mutC) and -(mutE) mutants, the mutations were introduced at the bottom half of the 3Ј-stem (Fig. 5, panel C). In addition, the mutant subgenomic RNAs containing the D2-SL(c) and D2-SL(C71G) mutations lost their template activities dramatically when either of the two infected cell extracts was used in the in vitro assays (Fig. 6, panels B and C, lanes 5 and 6) which are consistent with their lethal phenotypes in vivo. In the D2-SL(c) mutant, the lethal phenotype was attributed to the disruption of the top half of the 3Ј-SL (45). The loss of template activity of the D2-SL(C71G) mutant is particularly interesting because this subgenomic RNA contains a mutation that would disrupt the predicted pseudoknot structure based on the comparison to the homologous structure in the West Nile viral RNA 3Ј-SL (53) (see Fig. 5, panels A and B). When the mutant genomic RNA was electroporated into LLC-MK2 cells, the mutation reverted to wild type, and only the latter grew out of the transfected cells suggesting that this pseudoknot mutant was lethal in vivo. 4 In order to prove that the tertiary interaction involving a pseudoknot structure in the 3Ј-SL is important for template efficiency for RNA synthesis in vitro, we engineered the C73G mutation to disrupt the pseudoknot base pairing in the D2/WN-SL subgenomic RNA mutant in which the entire D2-SL was replaced with WN-SL. These subgenomic RNA templates containing the wild type or mutant pseudoknot base pairs within 3Ј-D2-SL or 3Ј-WN-SL were analyzed for RNA synthesis in vitro. The results from Fig. 7 indicate that the templates harboring a single base pair change that would disrupt the pseudoknot structure within either of the two 3Ј-SLs significantly affected the template efficiency for RNA synthesis in vitro. Based on these results, the strong template activity of D2/WN-SL(mutB) with either of the two extracts could be explained by the possibility that the pseudoknot structure of the West Nile 3Ј-SL is not affected in this mutant. These results presented in this study, taken together, support the conclusion that RNA synthesis in vitro at the 3Ј-end of the 3Ј-UTR is not only dependent on the conserved CYC motifs and the 5Ј-UTR but also, more importantly, on the determinants of the conserved 3Ј-SL structures of the subgenomic RNA template that are influenced by the conserved elements at the 5Ј-TR of the viral genome.

DISCUSSION
Functional Interaction between the 5Ј-TR and 3Ј-UTR Required for RNA Synthesis in Vitro-In this study, we have further established that the newly synthesized RNA from the subgenomic RNA template is of (Ϫ)-polarity. Furthermore, the 4  results of analyses of the structural requirements for RNA synthesis indicate that both 5Ј-UTR and 3Ј-SL structures contribute to template efficiency in addition to the conserved CYC motifs. The results of psoralen-UV cross-linking experiments further establish that there is physical interaction between the 5Ј-and 3Ј-TRs that require complementarity between the CYC motifs.
A number of studies have employed computer algorithms that are capable of analyzing predictive secondary structures as a guide or as a prelude for chemical and/or enzymatic probing to understand RNA-RNA interactions. Therefore, we sought to compare the predictive secondary structures of the subgenomic RNAs having wild type and mutant CYC motifs using the software described by Zuker et al. (58) (version 3.0). Among several structures that were produced from each RNA molecule, the most stable structures were selected for comparison and correlation with its in vitro template efficiency (Fig. 8). As shown in Fig. 8, the subgenomic RNAs containing the CYCwt motifs and the complementary, double-mutant CYC (5Ј3ЈCYCmut) motifs seem to maintain a similar predicted secondary structure (panels A and D). Interestingly, the subgenomic RNA/3ЈCYCmut also seems to maintain the same overall predicted secondary structures of the above two subgenomic RNAs (Fig. 8, panel C), although there is no predicted base pairing between the wild type 5Ј-CYC and the mutated 3Ј-CYC motifs. On the other hand, the predicted structures of the subgenomic RNA/5ЈCYCmut and the subgenomic RNA 674nt (without 5Ј-UTR) are quite different from that of the wild type (panels E and B versus panel A). Thus the predictive analyses of RNA secondary structure suggest that mutations in the conserved 3Ј-CYC motif are tolerated, and the template efficiency in RNA synthesis is not affected because the overall RNA structure is maintained. Moreover, the overall conformation of RNA influenced by the interaction of the 5Ј-and 3Јterminal SL structures seems more important for RNA synthesis than the CYC motifs which seem only to facilitate this interaction. Further work using chemical and/or enzymatic RNA structure probing methods and by site-directed mutagenesis is required to understand the RNA structure-template activities of these subgenomic RNA molecules using our in vitro RdRP assay.
The important role of the 3Ј-SL structure of the viral RNA genome in dengue viral replication in vivo was analyzed in an earlier study by Zeng et al. (45). The chimeric virus D2/WN-SL, which had the 96-nt sequence of the WN 3Ј-SL as a substitute for the 93-nt DEN2 3ЈSL, showed a phenotype severely defective but not lethal in monkey kidney LLC-MK2 cells (defined as "sublethal" in Ref. 45). It was suggested that the possible mechanisms involving RNA/RNA or protein/RNA interaction might have been affected which could have slowed down the replication of the transfected genome compared with that of the wild type (45).
There was a good correlation between the template efficiency of these 3Ј-SL mutants in the in vitro assays and the in vivo viabilities in the context of full-length infectious clones (45) for lethal mutations, D2-SL(c), D2/WN-SL(mutC), and D2-SL(C71G), as well as for the sublethal mutant, D2/WN-SL-(mutE). Moreover, this correlation could also be extended for the mutants D2-SL(a) and D2-SL(b) and the D2/WN-SL(mutA) in which substitutions were made within the upper half of the 3Ј-SL (Fig. 5, panel C). The good accord between the in vivo and the in vitro results with regard to the latter three mutants support the previous conclusion that the secondary structure rather than DEN2-specific nucleotide sequence of the top half of the 3Ј-SL is a major determinant for viral replication.
Notable exceptions to this good correlation are the two mutants, D2/WN-SL(mutB) and the D2/WN-SL (mutB and WN in Fig. 6, panel A), which exhibited significant template activity in vitro, but were lethal or sublethal, respectively, in vivo (45). It was previously reported that the 3Ј-SL structures of the DEN and WN viral genome contain a pseudoknot structure between the "G" in the loop of the small SL and the "C" in the bulge of the bottom half of the 3Ј-SL although the length of the basepaired regions involved in the 3Ј-SL structures of the DEN2 and WN RNA genomes are different (53) (illustrated in Fig. 5, panels A and B). Our study clearly indicates that the template efficiency of the pseudoknot mutant D2-SL(C71G) with a single nucleotide substitution was significantly reduced, suggesting that the pseudoknot structure within the 3Ј-SL could play an important role in RNA synthesis in vitro. This result also suggests that perhaps the near-wild type template activities of the mutants, D2/WN-SL(mutB) and the D2/WN-SL, might be due to the possibility that the potential pseudoknot structure of the West Nile 3Ј-SL could be preserved in these mutants. This hypothesis was confirmed by analysis of the C73G mutation in the D2/WN-SL RNA, which would be expected to disrupt this pseudoknot structure, and the resultant mutant subgenomic RNA had significantly reduced template efficiency for RNA synthesis in vitro. The same explanation could be considered for the mutant subgenomic RNA with D2/WN-SL(mutF); the regions containing the DEN2 sequences are presumably involved in the pseudoknot structure although the bottommost half of the 3Ј-SL is from the West Nile RNA. Thus, this -(mutF) subgenomic RNA template had significant retention of template efficiency in vitro. There is good correlation between the template efficiency in vitro and the in vivo viability of the mutant D2/WN-SL(mutD) which replicated well in both LLC-MK2 and C6/36 cells (45). However, the template efficiency of this -(mutD) mutant RNA was greater in the extracts from the infected C6/36 cells compared with the LLC-MK2 cells. Although our in vitro results are consistent with the notion that tertiary pseudoknot interaction might play an important role for RNA synthesis, they do not explain why mutants, D2/WN-SL(mutB), D2/WN-SL, and D2/WN-SL(mutF), that are active as templates in vitro are defective in vivo. One possible explanation for the disparity is that other steps in the viral life cycle that contribute to the overall viability such as defects in reinitiation of positive-strand RNA synthesis from the doublestranded RNA intermediate or in viral assembly might be affected in vivo. All of the 3Ј-SL mutant RNAs were also able to interact physically with the 5Ј-TR 180nt RNA after psoralen/UV treatment (data not shown), suggesting that substitutions within the 3Ј-SL do not affect the ability of the 5Ј-TR RNA to interact with the 3Ј-UTR. These results indicated that although both 5Ј-TR and 3Ј-UTR RNAs could interact with each other, possibly mediated by the wild type CYC motifs that are upstream of the 3Ј-SL, the mutations at the 3Ј-SL domains of the 3Ј-UTR RNAs per se must have contributed to their differences in template efficiency for RNA synthesis in vitro. These results also emphasize that the overall conformation of the 3Ј-SL is more important for RNA synthesis than the CYC motifs alone.
Possible Role of Cross-talk between the 5Ј-and 3Ј-Terminal Regions of the Viral RNA Genome in RNA Synthesis-There is increasing evidence from studies of other positive single-strand RNA viruses that the 3Ј-end of the viral genome has to maintain a certain secondary and tertiary structure for RNA synthesis (69 -74). The 3Ј-UTR of brome mosaic virus (BMV) RNA-3 is predicted to have five pseudoknots at the upstream end of the 3Ј-end tRNA-like structure (70). The deletion mutants of RNA-3 within these regions replicated poorly, yielding no detectable RNA-3 or RNA-4 progeny. This result suggested that the pseudoknot regions of the RNA-3 at the 3Ј-UTR contributed significantly to the overall replication of the BMV genome. Deiman et al. (72) also reported that the disruption of the stable pseudoknot structure at the 3Ј-end of the Turnip yellow mosaic virus genome gave rise to a drop in transcription efficiency to about 50%, indicating that the stable pseudoknot structure is crucial for initiation of negative-strand RNA synthesis.
Our results suggest that the cross-talk between the 5Ј-TR and 3Ј-UTR of the viral genome might define a precise structure at the 3Ј-end involving tertiary pseudoknot interaction such that the initiation of negative-strand RNA synthesis could be carried out by the viral replicase complex. In fact, possible cross-talk between specific regions of the viral genome of RNA viruses has been suggested to participate in the viral replication of Q␤ virus (75), BMV (76), rhinovirus type 14 (77), poliovirus (78), hepatitis C virus (79,80), and tobacco etch virus (81).
Proteins interacting with the flavivirus terminal regions of the RNA genome remain to be studied. The results of this study indicate that using our in vitro RdRP assay and the wild type and mutant subgenomic RNA templates, it is possible to study the higher order RNA structural interactions and identify the viral and cellular factors that contribute to the stability and functional long range interactions between the ends of the viral genome.