Structure and functionality in flavivirus NS-proteins: Perspectives for drug design

3.1.1.1. Functional aspects of the NS3 protease

The 375-kDa flaviviral polyprotein precursor is processed by host proteases and a virus-encoded protease activity localized at the N-terminal domain of NS3. Whereas the cleavage at the junctions C-prM, prM-E, E-NS1, NS4A-NS4B (Speight et al., 1988, Nowak et al., 1989), and likely also NS1-NS2A (Falgout and Markoff, 1995), is performed by the host signal peptidase located within the lumen of the ER, the remaining peptide bonds between NS2A-NS2B, NS2B-NS3, NS3-NS4A and NS4B-NS5 are cleaved by the virus encoded NS3Pro. Cleavage at the NS2B/NS3 site is performed in cis, but is not necessary for protease activity (Leung et al., 2001, Bera et al., 2007). Thus, the protease activity of NS3 is essential for viral replication and its inhibition has to be considered as a valuable intervention strategy for the treatment of flaviviral infections.

The activity of NS3Pro is strongly dependent on the association of a 40-amino acid region of NS2B acting as a cofactor for NS3Pro resulting in the formation of a heterodimeric complex. NS2B is a small protein (∼14 kDa) with a central hydrophilic part (residues 49–89) involved in binding to NS3, thereby fulfilling a chaperone-like role in stabilizing the latter protein, and two terminal hydrophobic regions responsible for membrane association of the NS2B/NS3 complex (Fig. 1) (Chambers et al., 1991, Chambers et al., 1993, Falgout et al., 1991, Falgout et al., 1993, Lindenbach and Rice, 2003, Lescar et al., 2008). The co-localization of NS2B and NS3 in convoluted membranes suggests these as the location for polyprotein processing by NS2B/NS3Pro, whereas Golgi-derived vesicle packets (the compartment presumably involved in RNA replication by NS3 and NS5) lack the presence of NS2B. Accordingly, the relevance of NS2B for non-proteolytic NS3 activities, such as helicase, nucleoside triphosphatase and 5′RNA phosphatase activities located in the C-terminal two-third of NS3, is yet unclear.

Even though the minimum requirements for proteolytic activity comprise the NS3 residues 1–160 (in WNV) or 1–167 in DENV2 (Li et al., 1999, Leung et al., 2001), the maximum activity concerning WNV NS3Pro, for an optimized fusion construct containing 44 NS2B residues covalently connected via a G4SG3 linker to the NS3Pro domain, has been observed for the N-terminal 1–184 residues (Chappell et al., 2007). Interestingly, a comparative analysis of the proteolytic activity of the full-length NS3 protein (1–618) fused to the optimized NS2B-G4SG3-linker region showed only marginal influence of the larger C-terminal domain on the NS3Pro kinetic parameters (Chappell et al., 2007). In contrast, studies of WNV full-length NS2B/NS3 and full-length NS3 exhibited different catalytic activities with respect to the unwinding of DNA and RNA: whereas full-length NS3 is capable to unwind both DNA and RNA templates, full-length NS2B/NS3 unwinds only RNA templates, and DNA unwinding is severely repressed (Chernov et al., 2008). Accordingly, the NS2B/NS3Pro part restricts substrate specificity of the C-terminal NS3Hel domain, however, in the absence of NS2B, the NS3 protein might dissociate from membranes and interfere with host DNA after translocation into the host-cell nucleus (Chernov et al., 2008).

3.1.1.2. Three-dimensional structures determined for NS3 protease

In 1989, sequence and structural comparison studies of the flavi- and pestiviral genomes suggested the presence of a serine protease related to the trypsin family, comprising a His, Asp, Ser catalytic triad (Bazan and Fletterick, 1989, Gorbalenya et al., 1989a). One year later, this prediction was verified for YFV NS3 by mutagenesis and characterization of segments of the NS3 gene (Chambers et al., 1990). The first crystal structure of a flavivirus NS3Pro was described in 1999 (Murthy et al., 1999) for DENV2. This crystal structure served as a template for homology modeling studies and interpretation of biochemical data (Nall et al., 2004, Zhou et al., 2006). Table 2 provides an overview of the crystal structures from flavivirus NS3Pro currently available. The binding mode of a peptidic mung-bean Bowman-Birk inhibitor in complex with DENV2 NS3Pro has been reported subsequently (see below). Even though these structures helped to explain various biochemical observations, such as the redundant nature of interactions formed by Arg and Lys residues in the S1 substrate-recognition sub-site, they were substantially different from the more relevant picture represented by the recently described crystal structures of flaviviral NS3Pro in complex with its cofactor NS2B (Fig. 2a and b). Accordingly, recent structure determination attempts were focused on the crystallization of fusion proteins containing the hydrophilic part of NS2B and the NS3Pro domain, linked via a glycine-rich linker (Erbel et al., 2006, Aleshin et al., 2007, Robin et al., 2009). So far, only crystal structures of flavivirus NS3Pro from DENV and WNV have been described (Table 2).

In its activated state, the flavivirus NS3Pro consists of the N-terminal domain of the full-length NS3 protein and its NS2B cofactor. The hydrophilic region of NS2B strongly interacts with NS3Pro, whereas both N- and C-terminal moieties of NS2B form two hydrophobic helices putatively acting as membrane anchors (Fig. 1). NS2B/NS3Pro adopts a chymotrypsin-like fold comprising two b-barrels, each formed by six β-strands, embedding the protease catalytic triad (His51, Asp75, Ser135) in the cleft between the two β-barrels (Fig. 2a). The presence or absence of the NS2B cofactor substantially influences the NS3Pro structure with respect to the extension and location of secondary structure elements (Erbel et al., 2006). Notably, in the cofactor-free DENV2 NS3Pro structure, the secondary structure elements are either shorter or even absent relative to DENV NS2B/NS3Pro. In the latter protein, the hydrophilic region of NS2B forms a link between the two b-barrels and contributes an anti-parallel β-strand to each of the b-barrels. The arrangement of the catalytic triad of the NS2B-bound NS3Pro suggests an exhaustive H-bonded network between the catalytic residues, in particular, a single-donor–double-acceptor (three-center) hydrogen bond between His51 and Asp75, whereas the structures lacking NS2B in the free or inhibitor-bound state exhibit an interaction geometry where only one weak H-bond between His51 and Asp75 is observed (Murthy et al., 1999, Murthy et al., 2000). These structural differences and the less constrained framework in absence of NS2B will presumably be related to the low proteolytic activity described for the non-cofactor-bound NS3Pro (Falgout et al., 1991, Falgout et al., 1993). While cleavage of small substrates by DENV2 NS3Pro is virtually not affected by the presence or absence of NS2B, degradation of larger substrates is strongly stimulated by presence of NS2B (Yusof et al., 2000).

Sharing a sequence identity of 50%, the overall fold observed for the NS2B/NS3Pro from WNV and DENV2 is very similar, with only subtle deviations in length and location of secondary structure elements (Aleshin et al., 2007). Interestingly, WNV NS2B/NS3Pro crystal structures presently available suggest conformational plasticity of the NS2B peptide: whereas in those protease structures hosting a small-molecule inhibitor in the active site, NS2B forms a belt around NS3Pro by contributing one β-strand to the N-terminal and two β-strands as β-hairpin motif to the C-terminal b-barrels, in the unbound state, the latter β-hairpin does not contribute to the C-terminal β-barrel (Fig. 2c). Instead, while the N-terminal NS2B fragment (residues 52-58) remains associated with the N-terminal b-barrel, the C-terminal residues form a short helical segment and a short β-strand interacting with strand B2a of NS3Pro, but the following hairpin motif points into the solvent and interacts with symmetry-related NS3Pro molecules. A similar fold for NS2B is observed in the inhibitor-free DENV2 NS2B/NS3Pro, with a disordered region corresponding to the b-hairpin in WNV NS2B/NS3Pro. The reasons for this unexpected NS2B plasticity are not completely understood (Erbel et al., 2006, Aleshin et al., 2007, Chappell et al., 2008). Nevertheless, the fold adopted by NS2B appears relevant for structure-based ligand design of inhibitors of WNV NS3Pro, as the NS2B b-hairpin tip in complexed NS2B/NS3Pro partly contributes to the formation of the S2 as well as the S3 pockets and may thereby directly interact with the bound ligand (see below).

3.1.1.3. Flavivirus NS3 protease complexes with inhibitors

In order to analyze the substrate preference of the proteases and to establish the basis for structure-based drug lead design, various contributions analyzed the interaction of peptide-like ligands with the protein active site (see Table 2). The first complex structure of NS3Pro lacking the NS2B cofactor described by Murthy et al. (2000) revealed the interaction of the DENV2 NS3Pro binding pocket with a polypeptide-type Bowman-Birk inhibitor isolated from mung beans, this being the only structure of a complex of DENV2 NS3 available to date. Despite the absence of NS2B, the structure allows general conclusions about the properties and ligand preference of the NS3Pro substrate-recognition pockets. The bivalent inhibitor possesses a lysine-head and an arginine-head, both occupying the substrate binding pockets of two different NS3Pro molecules simultaneously (Murthy et al., 2000). Both basic residues occupy the S1 pocket while establishing different interactions. The NS3Pro molecule hosting the inhibitor lysine head adopts virtual identical side-chain conformations as observed in the inhibitor-free NS3Pro. However, the second NS3Pro molecule exhibits strong conformational changes, particularly in the region Val126-Gly136, to adopt a binding-competent conformation (Fig. 3). The complex structure shows that Asp129 points either to the solvent (in the P1-Lys-bound molecule), or interacts with the basic residue (with P1-Arg bound) of the ligand, but the latter is involved in further charge-assisted hydrogen-bonds in a fashion obviously compensating the mutational loss of the interaction to Asp129 (Murthy et al., 2000). Additionally, a comparison of both NS3Pro molecules of the DENV2 NS3Pro-inhibitor complex reveals remarkable plasticity of active site residues (Fig. 3).

The flexible behavior of DENV2 NS3Pro is not observed in the three available WNV NS3Pro-ligand complexes, all containing the NS2B cofactor. One of them hosts aprotinin as the inhibitory ligand, the other two are complexed with a peptide-type substrate analogue covalently bound to the catalytic triad residue Ser135. Whereas the overall structure of the three protein/inhibitor complexes is very similar, in contrast to the peptidic inhibitor described by Erbel et al. (2006), aprotinin occupies all the major specificity pockets of the NS3Pro (S2-S2′). Additionally, it induces a catalytically competent conformation with a fully structured oxyanion hole established by the main-chain nitrogens of Gly133 and Ser135 (Aleshin et al., 2007). In the ligand-free state, the peptide bond between Thr132 and Gly133 is flipped, thereby forming a helical 310 conformation for residues 131–135. A superposition of the two conformations observed for the aprotinin-bound and ligand-free states is shown in Fig. 4. The main chain of aprotinin residues 13–19 forms antiparallel β-sheet interactions with the strands E2B and B1 of WNV NS2B-NS3Pro. In contrast, the substrate analogue inhibitor described by Erbel et al. adopts a loop conformation supported by favourable cation–p interactions between the P1-Arg residue and the inhibitor benzoyl cap (Fig. 2c). The protein–ligand complexes provide structural evidence for the strong preference for ligands (substrate or inhibitor) comprising basic residues at the P1 and P2 sites. Next to other interactions, the properties of the S1 pocket are governed by the salt-bridge between Asp129 and P1-Arg, as well as by Tyr161 contributing a face-to-face π-cation stacking with the P1 residue. Interactions with the P2 moiety are mainly contributed by the tip of the b-hairpin formed by the NS2B cofactor whose backbone oxygen atoms of Asp82 and Gly83 and the Od1 atom of Asn84 act as acceptors of charge-assisted H-bonds donated by the hosted P2-Arg residue. The latter forms an additional H-bond to Asn152 Oδ1. A P2-Lys moiety is capable of mimicking two of these interactions (H-bonds accepted by the backbone oxygen of Gly83 and by Asn152 Oδ1), and additionally establishes a hydrogen-bond to one of the carboxylate oxygens of Asp75. However, replacement of the P2-basic residue by alanine leads to a total loss of binding (Erbel et al., 2006). The S1′ pocket is lined on one side by the catalytic His51 as well as by Gly37, providing only sufficient space to accommodate small P1′ side chains such as Gly, Ser, or Thr (Aleshin et al., 2007). The different properties and plasticity behavior of the DENV2 and WNV specificity pockets could be exploited to design substrates with selectivity for only one of the flavivirus NS3Pro. Whereas probing the cleavage activity revealed a strict substrate specificity of WNV NS2B-NS3Pro, in agreement with the described mobility of the Val126-Gly136 segment (Fig. 3), the DENV2 NS2B-NS3Pro was less selective and tolerated well the presence of a number of amino acid types at either the P1′ or the P2′ site (Shiryaev et al., 2007).

Very recently, Robin et al. (2009) described a crystal structure of WNV NS2B-NS3Pro in complex with a substrate-based tripeptide inhibitor capped at its N-terminus by a naphthoyl moiety and at its C-terminal end by an aldehyde. The latter acts as an electrophilic warhead for covalent inhibition. Interestingly, in one of the two NS3Pro molecules present in the asymmetric unit, the catalytic His51 side chain adopts a split conformation. One conformer hydrogen-bonds to the aldehyde oxygen directing it for a nucleophilic attack by the catalytic Ser from the re side, whereas the other His conformer, inconsistent with a catalytic triad, points away from the reaction center enabling the oxyanion hole to direct the nucleophilic attack from the side. These observations suggested a role for the ligand to stabilize the His in its catalytically competent conformation.

Proteases related to the occurrence of pathobiochemical processes have raised the interest of biochemists and drug designers for many years (Mittl and Grutter, 2006). Benefiting from the knowledge thereby generated, current efforts to develop flavivirus NS3Pro inhibitors suitable for clinical use are indeed promising. Due to the increasing prevalence of infections caused by pathogenic flaviviruses such as WNV, different types of DENV, and SLEV, development of anti-flaviviral drugs is of utmost importance (Ghosh and Basu, 2008). Even though lessons from the treatment of Human Immunodeficiency Virus (HIV) and HCV infections show in a dramatic way the development of escape mutations conferring resistance to viral proteases upon single therapy with only one inhibitor (Manns et al., 2007), the protease inhibitors developed do contribute to an efficient combination therapy. Since NS2B-NS3Pro is obviously not only responsible for processing of the viral polyprotein, but also appears to contribute to further pathogenic processes such as induction of membraneous structures, neurovirulence and cleavage of host cell proteins (see above), inhibition of the proteolytic activity is a promising antiviral strategy.

3.1.2.1. Functional aspects of the NS3 helicase domain

It is well understood that RNA synthesis by the viral replication machinery requires unwinding of the RNA secondary structure in the template RNAs. The NS3Hel domain is held to assist in initiation of (−)ssRNA synthesis by unwinding the RNA secondary structure in the 3′ UTR (Takegami et al., 1995). The key role of helicase activity in viral replication has been demonstrated through site-directed mutagenesis (Grassmann et al., 1999, Matusan et al., 2001). Crystal structures of the flavivirus C-terminal NS3RTPase/Hel domain have been solved for YFV (Wu et al., 2005), DENV (Xu et al., 2005, Luo et al., 2008b) and JEV (Yamashita et al., 2008). In the context of the VIZIER Project, three new structures have been obtained for MVEV (Mancini et al., 2007); 1.9 Å resolution), KUNV, an Australian variant of WNV (Mastrangelo et al., 2007b; 3.1 Å) and KOKV (Speroni et al., 2008; 2.1 Å). In particular, the KOKV NS3Hel domain features high thermostability and good propensity to crystallize, making this an attractive model system for structural and biochemical analysis of inhibitor binding.

3.1.2.2. Three-dimensional structures determined for the NS3 helicase domain

The flaviviral NS3Hel tertiary structure is characterized by three domains, each of about 130–150 amino acids (Fig. 5). The first two domains (domains I and II) are structurally similar, displaying an α/β open sheet topology (Rossman fold), composed of six β-strands (topology β1–β6–β5–β2–β4–β3), surrounded by four and three α-helices, respectively. Domain III is mainly composed of five approximately parallel α-helices and two antiparallel β-strands (Fig. 5). In domain II, a β-hairpin (residues 426–450) protrudes from the core domain projecting towards domain III and is held to be a critical element in unwinding activity (Luo et al., 2008b). The first two domains, likely originated by gene duplication (Caruthers and McKay, 2002), host seven characteristic sequence motifs of the NS3Hel superfamily 2 (Gorbalenya et al., 1989b, Cordin et al., 2006), associated with nucleic-acid binding and NTP hydrolysis (Caruthers and McKay, 2002). In particular, the conserved motifs I (GAGKTRR) and II (DEAH), also known as Walker A (ATP phosphate-binding loop, or ‘P-loop’) and Walker B motifs (Mg²⁺-binding; Walker et al., 1982), are located in the amino-terminal region (domain I) where they interact with the NTP substrate and Mg²⁺, respectively (Xu et al., 2005). In addition, two conserved Arg residues in domain II (Arg458 and Arg461: “arginine fingers” in motif VI) are involved in ATPase and RTPase activities (Sampath et al., 2006), and in the structural rearrangement that results in RNA unwinding, following ATP hydrolysis. A similar role in coupling the ATPase and RNA unwinding activities is played by the residues of the Ia motif, as gathered from mutagenesis studies on the KOKV NS3Hel (Speroni et al., 2008). The central region of the NS3Hel, where the three domains contact each other, hosts a cleft held to be involved in ssRNA binding during the helicase activity (Xu et al., 2005, Luo et al., 2008b).

Differences between the YFV, DENV, KUNV, KOKV, JEV and MVEV structures were found to be confined primarily to the relative orientation/distance of domain II to domains I and III, suggesting that movement of domain II can affect nucleic-acid translocation in an ATP-dependent mode according to the ‘inchworm’ model (Mancini et al., 2004, Mastrangelo et al., 2007b). Such overall structural rearrangement was recently confirmed by a detailed structural study on DENV4 NS3Hel, which describes the structures of various complexes with ATP analogues and ssRNA of 12–13 nucleotides (Luo et al., 2008b) (Fig. 5). In particular, upon ssRNA binding domain III rotates about 11° away from domain I with the simultaneous narrowing the of cleft between domains I and II (12° rotation). The overall movement can be described as an opening of both the ssRNA access site, located between α-helices α2 (domain II) and α9 (domain III; as showed in normal mode analysis; Mastrangelo et al., 2007b), and the ssRNA exit site, by the repositioning of a loop (disordered in many crystal structures) located between strands β3 and β4 in domain I.

3.1.2.3. Characterization of helicase activity

RNA unwinding activities are assessed using a partially dsRNA molecule consisting of a 14 base 3′ single-stranded tail followed by a 16 base-pair dsRNA region (Wu et al., 2005). Generally, NS3Hel containing longer linker regions show higher activity than those with short linkers. The DENV4 NS3Hel (NS3 178-end) used by Luo et al. (2008b) was truncated close to boundaries shown previously for DENV, WNV and YFV to yield inactive or significantly impaired domains with respect to ATPase and helicase activities when compared to constructs with N-terminally extended linker regions (Li et al., 1999, Wu et al., 2005, Xu et al., 2005). In contrast, the MVEV NS3Hel construct includes a significantly longer linker suggesting that the observed reduction in activity for the DENV4 NS3Hel domain (Luo et al., 2008b) was due to the short linker. The reason why truncation of the linker region can have a detrimental effect on the activity of the C-terminal domain of NS3 remains unclear; it may have a functional role (Matusan et al., 2001), or cause structural artefacts as observed for KUNV NS3Hel (aa186-619) which forms a dimer (Mastrangelo et al., 2007b).

The VIZIER Project has substantially enhanced our knowledge on flaviviral NS3Hel, providing the bases for the structure-based design and development of specific antiviral molecules targeting this essential class of enzymes. The remarkable similarities in the Hel/ATPase catalytic regions indicate that it might be possible to develop compounds with a broad spectrum of activities – i.e. which are able to act on different flaviviral enzymes – and/or lead molecules that can be targeted to a specific viral enzyme through minimal ad hoc chemical modifications. Medicinal chemistry studies on protein kinases have shown that the most effective inhibitors are conformationally based; they exert their inhibitory effect through an allosteric alteration of the equilibrium among different protein conformations (Vajpai et al., 2008). Likewise, future drug-design studies on flaviviral NS3Hel will benefit from our improved understanding of the role of the various fingerprint residues and of the conformational changes that underlie the coupling between ATP hydrolysis and RNA unwinding activity.

3.1.3.1. Functional aspects of the NS3 protein

Understanding the biologically relevant functional properties of NS3 is complicated by the fact that in infected cells NS3 acts anchored to or in close proximity of membranes (Lindenbach and Rice, 2003), whereas most in vitro characterization has been done in solution. Furthermore, as the virus progresses towards maturation, different protein–protein and protein–RNA interactions occur which demarcate specific points in its life cycle, although the details remain unclear. Polyprotein processing and replication occur in distinct, membrane-bound compartments (convoluted membranes and vesicle packets, respectively), and in each compartment, NS3 engages with different proteins (Mackenzie, 2005). An intriguing finding has been the apparent absence of NS2B, the essential cofactor of the NS3Pro, in vesicle packets (Westaway et al., 1997, Mackenzie, 2005) suggesting that in the transition from polyprotein processing to replication, the NS3Pro becomes inactive. Finally, the structure and dynamics of the polyprotein as it emerges following translation remain largely unexplored and little information exists on interactions between NS3 and other parts of the polyprotein, which might be important for priming NS3Pro for its first and subsequent cleavage activities.

3.1.3.2. Three-dimensional structures determined for the NS3 protein

Two full-length NS3 structures have been solved by X-ray crystallography: those of DENV4 NS3 (Luo et al., 2008a) and of MVEV NS3 (Assenberg et al., submitted for publication) (Fig. 6). Both structures were solved in the presence of a fragment of protein NS2B, the essential co-factor and activator for NS3Pro, by producing a single polypeptide chain where this region was linked via a flexible tether to NS3. One difference between the two studies is that for DENV4 only part of the NS2B activating region was coupled to NS3 (18 amino acids of NS2B, DENV4 NS2B₁₈NS3), whereas for MVEV the full activating region was included (45 residues of NS2B, MVEV NS2B₄₅NS3).

In both the DENV4 NS3 (Luo et al., 2008a) and MVEV NS3 (Assenberg et al, submitted for publication) structures the NS2B-NS3 molecule consists of two separate globular folds linked by a short linker, an arrangement consistent with SAXS data for full-length KUNV NS3 in solution (Mastrangelo et al., 2007b). The individual domains are very similar between the two molecules (r.m.s.d. of 1.5 and 1.6 Å for the NS3Hel and NS3Pro, respectively) and to the structures of isolated domains. Yet the relative orientations of the NS3Pro and NS3Hel domains are dramatically different between MVEV NS2B₄₅NS3 and DENV4 NS2B₁₈NS3. When superimposing the NS3Hel domains of the two structures, a rotation of ∼180° and translation of 17 Å are required to align the NS3Pro domains. The 13-residue “linker” between the NS3Pro and NS3Hel domains (residues 169–181) is ordered in DENV4 NS2B₁₈NS3 but partially disordered in MVEV NS2B₄₅NS3. Even though the buried surface area between the NS3Pro and the NS3Hel in DENV2 NS2B₁₈NS3 is only 568 Å², the NS3Pro domain and the linker loop engage in possibly significant interactions with subdomains 1 and 2 of NS3Hel. Specifically, the linker interacts with the catalytic P-loop of the NS3Hel, which assumes the distinctive apo conformation seen in the NS3Hel domain crystal structure in the absence of bound nucleotides. These interactions are not seen in the MVEV NS2B₄₅NS3 structure where the buried surface area is only 30 Å²; however, the P-loop is found in the same apo conformation. Thus, a possible role for the linker loop could be to stabilize the apo conformation of the P-loop, in line with recent studies suggesting a functional role for the linker (truncating this acidic linker can have a substantial effect on the activity of isolated NS3Hel domains; Li et al., 1999, Wu et al., 2005, Mastrangelo et al., 2007b).

A striking observation relating to the influence of the NS3Pro domain on the NS3Hel domain emerges from a comparison of the two structures. In DENV4 NS2B₁₈NS3, the interactions between NS3Pro and NS3Hel are such that motions of domain 2 of the NS3Hel, known to be important for helicase activity, would be constrained. In the MVEV NS2B₄₅NS3 structure in contrast, such interactions are weak and therefore there are no constraints on the motility of the NS3Hel domain. This would suggest that the NS3Pro domain might repress helicase activity and that such activity might be regulated by switching between various transient configurations, such as those observed in the two structures. This conclusion contrasts that of Luo et al. (2008a), who saw an increased affinity for ATP when the activity of the DENV4 NS2B₁₈NS3 protein was compared to that of the isolated NS3Hel domain. Although this enhancement was explained by a positive contribution of the NS3Pro domain to the electrostatic potential of the NS3Hel nucleotide binding pocket raising helicase activity, an alternative explanation is that the isolated NS3Hel domain chosen for comparison was ‘hobbled’ by truncation of the linker. Indeed the truncation used has been shown to significantly reduce helicase activity of DENV2 and other flavivirus NS3Hel domains (Li et al., 1999, Wu et al., 2005). The latter interpretation is supported by biochemical analysis of the helicase activity of MVEV NS2B₄₅NS3, which showed no significant difference between the activity of full-length MVEV NS2B₄₅NS3 and that of the isolated NS3Hel domain (using a more appropriate linker than in the DENV4 study). Finally, NS2B is not found in vesicle packets and therefore not part of the replication complex (Westaway et al., 1997, Mackenzie, 2005), posing further questions over the role of the NS3Pro domain in regulating helicase activity since in the absence of NS2B the NS3Pro domain partially unfolds.

3.1.3.3. General structural properties of the NS3 protein

The structures of full-length NS3 raise two important questions: (1) do these structures represent two distinct and stable conformations of NS3 possibly adopted at different stages of the flavivirus life cycle or are they merely snap-shots of a highly dynamical interconversion process and (2) given the segregated nature of the two catalytic domains, what is the functional significance of this arrangement?

Our analysis suggests that the relative domain organization is probably highly dynamic, given the linker flexibility (disordered in the MVEV structure) and the small buried surface areas between the two domains in both structures. Further, the configurations are in principle inter convertable via simple rotations around the linker loop, and linker flexibility is probably paramount to the NS3 activities and its ability to associate with other proteins and RNA. On the other hand, specific configurations may be stabilized during a particular stage of the virus life cycle. To gain further insights into these issues, we modeled the MVEV and DENV NS2B-NS3 structures in the presence of a membrane (Assenberg et al., submitted for publication). Previous studies have shown that when in complex with NS2B, and in particular when fully activated, NS3Pro sits in a rather tight membrane-anchored sling (Clum et al., 1997, Robin et al., 2009). Fig. 7 shows a model of MVEV and DENV4 NS2B-NS3, with their NS3Pro domains superimposed and associated to the membrane as inferred from the anchoring of the published NS2B-NS3Pro structures (Erbel et al., 2006, Robin et al., 2009). In this model, the DENV4 NS2B₁₈NS3 NS3Hel is positioned near the membrane with the active site orientated towards the membrane and with little space to accommodate RNA, whilst in the MVEV structure the NS3Hel domain is positioned away from the membrane, with the active site facing the cytoplasm. Thus, the DENV4 NS2B-NS3 structure appears incompatible with RTPase/helicase activity. Although this could be taken as an argument that the DENV4 conformation may not be physiologically relevant, the strength of conformational constraints imposed by the cellular environment is difficult to assess. Prior to cleavage of the NS3-NS4A junction, NS3 is also anchored to the membrane at its C-terminus via the membrane-bound NS4A. However, although there are 50 residues separating the NS3 C-terminus and the first trans-membrane α-helix of NS4A (Miller et al., 2007), the structure of the first 50 residues of NS4A remains unknown. Thus, the presence of NS4A may limit the ability of NS3 to change its conformation in vivo. This leads to the interesting possibility that NS3 may adopt this conformation during polyprotein processing where helicase activity is probably not wanted. Formation of the replication complex, where the helicase activity is presumably needed, would release NS2B, inactivating the NS3Pro domain. In this view, the MVEV NS2B-NS3 conformation is likely to be relevant later in the virus life cycle, during the assembly and functioning of the replication complex. The regulation of the activities of NS3 by an environment-dependent re-configuration of the molecule offers a simple temporal and spatial control mechanism, coupling activities appropriately with the virus life cycle. This model provides answers to both of the questions posed in the previous paragraph.

In vivo, the situation is probably complicated by the modulation of the structure and function of NS3 by additional binding partners. Thus, the activity of NS3 may be affected by interactions within the polyprotein (Zhang and Padmanabhan, 1993), and NS3 binds to free NS5 (Johansson et al., 2001, Yon et al., 2005), NS4A (Mackenzie et al., 1998) and NS4B (Umareddy et al., 2006) as well as viral RNA. In particular, it has been suggested that the C-terminal domain of NS3 binds NS5 (Liu et al., 2002, Wu et al., 2005) during the formation of the replication complex. Unfortunately, the details of these interactions remain poorly understood. Clearly, further studies are required to test the functional significance of the two conformations in vivo, as well as the influence of the interactions between NS3 and other viral proteins, RNA, and lipids on the conformation of NS3.

The structures of full-length NS3 reveal that the molecule can assume two radically different configurations, defined by the relative positioning of the NS3Pro and NS3Hel via a flexible inter-domain linker. We suggest that these may be important in its interactions with other proteins and RNA and, possibly, in modulating the switch to helicase and triphosphatase activities during replication.

3.2.1.1. Functional background aspects of the NS5 methyltransferase domain

The flavivirus RNA is decorated with a conserved type-1 cap (N7meGpppA2′Ome-RNA) at its 5′-end, a unique structure consisting of an inverted guanosine linked to the first transcribed RNA nucleotide by a 5′–5′ triphosphate bridge. Viral MTases are involved in the mRNA capping process, transferring a methyl group from the cofactor S-adenosyl-l-methionine (AdoMet) onto the N7 atom of the cap guanine and onto the 2′OH group of the ribose moiety of the first RNA nucleotide. In the genus Flavivirus, both (guanine-N7)-methyltransferase (N7MTase) and (nucleoside-2′-O-)-methyltransferase (2′OMTase) activities have been associated with the N-terminal domain of the viral NS5 protein (Egloff et al., 2002, Ray et al., 2006, Zhou et al., 2007).

3.2.1.2. Three-dimensional structures determined for the NS5 methyltransferase domain

Crystal structures of the NS5MTase domain have been reported for different mosquito-borne flaviviruses, such as DENV (Egloff et al., 2002), WNV (Zhou et al., 2007) and YFV (Geiss et al., 2009). In the context of the VIZIER Project new high-resolution structures of NS5MTases have been obtained for MEAV, a TBV (Mastrangelo et al., 2007a), for the MBV MVEV (Assenberg et al., 2007) and WESSV (Bollati et al., 2009b), and for two NKVs: YOKV (Bollati et al., 2009a) and MODV (Jansson et al., 2009) (Table 3). Moreover, structures of DENV NS5MTase, MVEV NS5MTase and WESSV NS5MTase in complex with GTP or several cap analogues, GpppA/G and N7meGpppA/G (Egloff et al., 2002, Egloff et al., 2007, Assenberg et al., 2007, Bollati et al., 2009b) have been reported, shedding light on the substrate-binding mode during methylation and on the enzyme mechanism of action (Table 3).

The flaviviral NS5MTase domain consists of a 33-kDa protein comprising residues 1–260/270 of the N-terminus of the NS5 protein. It is characterized by an overall globular fold consisting of a core domain (residues 59–224) flanked by an N-terminal region (residues 1–58), and a C-terminal region (residues 225–265) (Fig. 8). The core domain comprises a seven-stranded β-sheet surrounded by four α-helices and two 310 helices, closely resembling the topology observed in the catalytic domain of other AdoMet-dependent MTases (Fauman et al., 1999, Bugl et al., 2000, Egloff et al., 2002). The N-terminal segment comprises a helix-turn-helix motif followed by a β-strand and an α-helix. The C-terminal region consists of an α-helix and two β-strands (Fig. 8). The core subdomain hosts the active site and the cofactor binding site (Ingrosso et al., 1989, Egloff et al., 2002, Egloff et al., 2007, Assenberg et al., 2007, Mastrangelo et al., 2007a, Zhou et al., 2007, Bollati et al., 2009a, Bollati et al., 2009b). In all the structures a cofactor molecule, in some cases the co-product AdoHcy, originated from E. coli and co-purified with the enzyme, is bound in this binding site, where it is stabilized by a network of hydrogen bonds and van der Waals contacts involving several residues – Ser56, Gly86, Trp87, Thr104, Leu105, His110, Asn131, Val132, Asp146, and Ile147 (residue numbering refers to WESSV NS5MTase) – and a series of interactions that are well conserved within the flaviviruses NS5MTases (Fauman et al., 1999, Egloff et al., 2002, Egloff et al., 2007, Assenberg et al., 2007, Mastrangelo et al., 2007a, Zhou et al., 2007, Bollati et al., 2009a, Bollati et al., 2009b).

The known flaviviral NS5MTases show a large degree of structural homology (r.m.s.d. < 1 Å), which reflects the high amino acid sequence conservation (between 30% and 90%). Comparison of the NS5MTases representative of each of the three flaviviruses branches (the NKV, YOKV NS5MTase; the TBFV, MEAV NS5MTase, and the MBFV, DENV NS5MTase) shows that most differences in the structures are caused by surface-loop flexibility and amino acid variability, displayed in four regions of the enzyme: (1) helix-loop-helix motif in the N-terminal domain, involved in the binding of the substrate (residues Gly21-Lys22-Thr23 in YOKV NS5MTase, substituted with Gly-Lys-Ser in DENV NS5MTase, and Thr-Lys-Glu in MEAV NS5MTase); (2) α3-αX loop (Asn47-Ile53; insertion of one amino acid in MEAV NS5MTase); (3) αD-β5 loop (Leu172-Thr178; deletion of two amino acids in DENV NS5MTase); and (4) α4-β8 loop (Leu246-Thr252; deletion of two amino acids in MEAV NS5MTase) (highlighted in Fig. 8) (Bollati et al., 2009a).

3.2.1.3. Structures of protein/ligand complexes

In the context of the VIZIER Project, structures of different NS5MTases in complex with cofactor and several capped substrate analogues have been solved (Egloff et al., 2002, Egloff et al., 2007, Assenberg et al., 2007, Bollati et al., 2009b). All the complexes show that the cap guanine moiety (methylated or not) binds to a site next to the N-terminal helix-turn-helix motif, called the high affinity binding site (HBS) formed by residues Lys13, Leu16, Asn17, Leu19, Phe24, Ser150, Arg197, Arg213, Ser215. In particular, Phe24 plays an essential role in driving the cap binding by stacking against the cap guanine base during the interaction (Egloff et al., 2002, Egloff et al., 2007, Assenberg et al., 2007, Mastrangelo et al., 2007a, Bollati et al., 2009b, Milani et al., 2009). The first nucleotide after the cap is often disordered, reflecting a lack of strong interactions with the enzyme. This is in line with other studies showing that a minimal number of nucleotides in capped RNA are required for substrate binding and methylation (Ray et al., 2006, Egloff et al., 2007, Dong et al., 2008, Selisko et al., in press). A unique conformation of the cap analogue is adopted by GpppA in complex with DENV NS5MTase. The guanine is bound in the HBS and GpppA displays an overall hairpin-like shape in which the two bases stack against each other in a way that allows to accommodate a longer capped RNA because the 3′-position of the A ribose is oriented towards a positively charged surface region (Egloff et al., 2007). It was proposed that this conformation might mimic the reaction product of a putative flavivirus GTase activity: pppG + ppAGN leading to GpppAGN + pyrophosphate. Such a guanylyltransfer reaction would be the only occasion where the GTP-binding site of DENV NS5MTase needs to accommodate a non-methylated GTP.

Since flaviviral NS5MTases display both N7MTase and 2′OMTase activities (Ray et al., 2006, Egloff et al., 2007, Zhou et al., 2007, Dong et al., 2008), in order for capped RNA to be methylated at the two different sites, it must adopt two distinct binding modes relative to the enzyme active site. To date structural details on catalytically relevant states for both N7 or 2′O methyl transfer are missing. Nevertheless, it has been speculated that when the cap is bound in the HBS, the 2′OH group of the ribose moiety of the nucleotide following the cap is the closest atom to the AdoMet exchangeable methyl group (Egloff et al., 2007, Mastrangelo et al., 2007a, Bollati et al., 2009b). For this reason, the HBS is assumed to host the cap during 2′O methylation. Moreover, structural considerations suggest the existence of a secondary, putative low-affinity binding site (LBS) located in a positively charged region close to the AdoMet binding site which could be involved in binding the capped RNA substrate (Egloff et al., 2007, Mastrangelo et al., 2007a, Dong et al., 2008, Bollati et al., 2009a, Bollati et al., 2009b).

In this context, two different models of the mechanism of action of the flaviviral NS5MTase have been proposed. Model 1 (Egloff et al., 2007, Mastrangelo et al., 2007a, Zhou et al., 2007, Dong et al., 2008, Bollati et al., 2009a, Bollati et al., 2009b, Milani et al., 2009) suggests that when the capped RNA substrate binds in the LBS, it undergoes methyl transfer in position N7 of the guanine with the cap guanine fixed in the active site. Afterwards, the NS5MTase slides on the RNA chain positioning the cap in the HBS and locating the first RNA adenine in the active site for 2′O methylation. In this binding mode, the rest of the RNA chain is still stabilized by the interaction with the residues of the LBS, but shifted by one nucleotide. The analysis of conserved residues within flaviviral NS5MTases shows that there is an almost continuous and conserved region extending away from the active site towards the back of the protein—residues Tyr89, Pro113, Gly120, Trp121, Asn122, Leu123, Ile124, Phe126, Lys127, Asp131, Gly263, Thr264 and Arg265 on the side of the NS5MTase, and residues Ala60, Trp64, Leu207, Val208, Arg209, Pro211, Met220 and Arg244 on the back (numbering referring to YOKV NS5MTase (Bollati et al., 2009a). This region may play a role in stabilization of the rest of the RNA chain following the cap (no crystal structure available so far) (Bollati et al., 2009a). In order to predict the interactions between the protein and the RNA during N7 methyl transfer, Milani et al. presented a model of a short capped RNA (GpppAGUp) bound to the LBS of WESSV NS5MTase (Milani et al., 2009). The model produced after 9 ns of Molecular Dynamic simulation shows that the cap guanine was located close to AdoMet and to residues Glu149 and Arg213; other residues found to interact with capped RNA were: Arg37, Arg41, Leu44, Ser56, Arg57, Arg84, Glu149, Lys112, Ser150, Arg160, Ser215. In line with this model, it has been recently shown that point mutations of a good part of these amino acids (Arg37, Arg57, Arg84, Trp87, Glu149, Arg213, and Tyr220) dramatically inhibit N7 methyl transfer activity in WNV NS5MTase (Dong et al., 2008).

Based on the structure of MVEV NS5MTase in complex with GpppA, which reveals a crystallographic dimer with two GpppA molecules bound per monomer, Assenberg et al. (2007) proposed Model 2. In this structure, the first cap analogue binds in the HBS with the adenosine base approaching the AdoMet binding pocket. The second analogue binds to the LBS and stacks against the first cap analogue, with the guanosine base stacking against the adenosine base of the first analogue (Fig. 9) (Assenberg et al., 2007). Although this indicates that the LBS can bind cap analogues, supporting in this way Model 1, the authors note that in the MVEV co-crystal structure the N7 and 2′O guanosine group of the two analogue bound to the dimer are too far away (>10 and >8 Å, respectively) from the AdoMet methyl leaving group to be catalytically relevant. This casts some doubt over the catalytic relevance of the structure and indeed the significance of the LBS in controlling N7MTase activity. Further analysis of the crystallographic dimer show that two LBS regions form a nearly continuous positively charged groove, with the four GpppA molecules interacting in a manner resembling that of a continuous strand of RNA. Based on this, Model 2 proposes that the dimerization of two NS5MTase monomers generates a large positively charged RNA binding cleft that facilitates binding and translocation of the capped RNA. As the RNA moves into the catalytic site of the first monomer, N7 methylation occurs, followed by RNA translocation via the LBS to the HBS of the second monomer where 2′O methylation occurs. In this model the LBS acts merely as an RNA-binding domain, in contrast to Model 1 discussed above.

3.2.1.4. Functional characterization

The first evidence of enzymatic activity from a flavivirus NS5MTase domain was demonstrated with DENV NS5MTase using short capped RNA substrates N7me ± GpppAC5 (Egloff et al., 2002). Such short substrates, even starting with GpppAG (the first two nucleotides strictly conserved in the 5′-end of the flavivirus genome), support specifically the 2′OMTase activity (Kroschewski et al., 2008, Lim et al., 2008). In the context of the VIZIER Project, a protocol was set-up to produce pure N7me ± GpppACn substrates of varying chain lengths (n = 1–7) in high amounts (Peyrane et al., 2007). The substrates were used to prove 2′OMTase activity in a variety of flavivirus NS5MTase domains (Mastrangelo et al., 2007a, Bollati et al., 2009a, Bollati et al., 2009b) and set-up inhibition assays (Luzhkov et al., 2007, Milani et al., 2009, Selisko et al., in press). Interestingly, it was found that DENV 2′OMTase activity and binding increases with the substrate chain length until they reach a plateau at n = 5. N7meGpppAC5 substrates were used to determine kinetic parameters of the 2′OMTase activity of DENV NS5MTase (Selisko et al., in press). On the other hand, in order to measure N7MTase activity of flavivirus NS5MTase domains, a substrate of 74 nucleotides is needed, which contains a conserved stem-loop structure (Ray et al., 2006, Dong et al., 2007). In the context of VIZIER, capped subgenomic RNA of DENV was produced and used to demonstrate N7MTase activity of DENV and WESSV NS5MTases (Milani et al., 2009). Moreover, in accordance with the hypothesis that the flavivirus NS5MTase domain contains the GTase activity, the WESSV NS5MTase domain was shown to covalently bind GMP at Lys28 (Bollati et al., 2009b). Following the classic scheme of cap formation (Furuichi and Shatkin, 2000) to complete the guanylyltransfer, GMP would be transferred to newly synthesized ppRNA (see introduction). To date, however, there are no available data fully proving the transfer of GMP on a ppRNA substrate catalyzed by a flavivirus NS5MTase domain.

3.2.2.1. General structural properties of the polymerase domain

The structure of the two flavivirus NS5RdRp domains have been recently analyzed and reviewed in (Malet et al., 2008). They adopt a typical RdRp right-hand structure comprising three subdomains: fingers, palm and thumb (Fig. 10). The fingers subdomain of the short KUNV NS5RdRp construct was partially disordered, whereas the long constructs started with the ordered N-terminal helix (277–287 and 275–285 for KUNV NS5RdRp and DENV NS5RdRp, respectively). Both flavivirus NS5RdRp domains display a closed conformation, where fingers and thumb subdomains are connected. This is a characteristic of RdRps and in particular of primer-independent (de novo) RdRps (reviewed in Ferrer-Orta et al., 2005). A structural element named the priming loop provides the initiation platform. It belongs to the thumb subdomain and points towards the active site in the palm. The active site is located at the intersection of two tunnels. Other de novo RdRps solved in complex with ssRNA template, NTPs and/or dsRNA product (Butcher et al., 2001, Tao et al., 2002, O’Farrell et al., 2003) suggest the following scenario shown in Fig. 10. The first tunnel, located between the fingers and the thumb, should allow the ssRNA template to access the active site. The second tunnel, roughly perpendicular to the first, stretches across the entire protein. The incoming NTP should arrive from the back of the tunnel and, after the polymerization has started, the nascent dsRNA should go out through the front of this tunnel. However, as for other de novo RdRps, a conformational change is necessary to avoid a steric clash with the priming loop and allow neo-synthesized RNA to exit.

As expected, the most closely related structures are those of de novo RdRps from members of the Flaviviridae family, namely HCV NS5RdRp (Ago et al., 1999, Bressanelli et al., 1999, Lesburg et al., 1999) and bovine viral diarrhea virus (BVDV) NS5RdRp (Choi et al., 2004). They show sequence identities as low as 11–21%, depending on the subdomain considered (Malet et al., 2007). Special features of flavivirus NS5RdRp structures are as follows (Malet et al., 2008): (1) The fingers subdomain was captured in a pre-initiation conformation since motif F, normally comprising the NTP-binding sites does not form the upper part of the NTP tunnel but is perpendicular to its normal position and partially disordered (Fig. 11). Additionally, the fingers subdomain presents a loop, named G-loop, which protrudes towards the active site (see Fig. 11). It was given the name G-loop because it harbours RdRp motif G in primer-dependent RdRps (Gorbalenya et al., 2002, Ferrer-Orta et al., 2004). This loop may play a regulatory role similar to the C-terminal in other RdRps (Adachi et al., 2002, Leveque et al., 2003, Ng and Parra, 2004). In summary, a concerted conformational change of motif F and G-loop of the trapped pre-initiation conformation is expected before flavivirus NS5RdRp initiate RNA synthesis. (2) The priming loop is provided by the thumb domain as observed for other de novo RdRps (bacteriophage phi6, HCV and BVDV RdRps (Ago et al., 1999, Bressanelli et al., 1999, Lesburg et al., 1999, Butcher et al., 2001, Choi et al., 2004) but does not contain any secondary structure. Two aromatic residues Trp795 or His798 (DENV NS5RdRp), may act as initiation platform stacking with the priming nucleotide. (3) Two Zn ions were found, one in the fingers and one in the thumb subdomain. The latter is localized at a supposed hinge position between the thumb and the palm subdomains. It might play a role in the regulation of the conformational change between initiation and elongation state of flavivirus NS5RdRps. (4) Two NLSs are present at the surface of the fingers subdomain. They comprise the first fingertip forming the connection between finger and thumb domain (see Fig. 11), run across the back of the NS5RdRp domain and end in an α-helix, which comes to the front and forms the interface between the bottom of the fingers subdomain and the palm.

The NLS of DENV2 NS5 has been shown to be functional and the transport of NS5 to the nucleus vital for virus replication (Pryor et al., 2007). Especially two lysine residues at the beginning of the α-helix seem to be important. In contrast, KUNV NS5 does not localize to the nucleus. Subtle differences in NLS geometry and charge distribution may be responsible for distinct behavior towards nuclear import in closely related viruses, but this is not yet deducible from the structures. It has not been shown if NS5 of DENV3 localizes to the nucleus (Malet et al., 2008).

3.2.2.2. Three-dimensional structures determined for the flaviviral polymerase domain

Two flavivirus NS5RdRp domain structures are known so far, KUNV NS5RdRp and DENV3 NS5RdRp (Malet et al., 2007, Yap et al., 2007). The structures are very similar with an r.m.s.d. of 1.9, 0.8 and 1.0 Å (Cα atoms of matched residues) for the fingers, palm and thumb subdomains, respectively. The overall r.m.s.d. is expected to be high because of a domain rotation based on the hinge region between the thumb and the palm subdomains near the Zn atom (see above). The rotation of the fingers subdomain by 8° leaves the DENV3 NS5RdRp structure more open in comparison to KUNV NS5RdRp. Another consequence is that the active site of KUNV NS5RdRp is more tightly closed by the priming loop than the active site of DENV3 NS5RdRp. KUNV NS5RdRp may need a wider opening movement upon the transition to elongation mode, which could explain why the transition from initiation to elongation seems to be kinetically more limiting for KUNV NS5RdRp compared to DENV2 NS5RdRp (Selisko et al., 2006).

Both flavivirus NS5RdRp structures were obtained with a Mg²⁺ ion in a non-catalytic position near the active center. The role of the non-catalytic ion is not known. It was proposed that it might play a role in the de novo initiation mechanism facilitating the movement of the nascent dsRNA after formation of the first dinucleotide out of the active site (Butcher et al., 2001). DENV3 NS5RdRp crystals were soaked with the nucleotide analog 3′dGTP and the complex structure solved at 2.6 Å resolution. Only the triphosphate moiety of 3′dGTP is visible. Nevertheless, its position near the priming loop and in particular next to residue Trp795 lead to the proposal that it represents the priming nucleotide and Trp795 acts as the initiation platform (Yap et al., 2007). This is consistent with a model of the de novo initiation complex of KUNV NS5RdRp, which assigned the same Trp, conserved in all flavivirus NS5 proteins, as initiation platform (Malet et al., 2007).

3.2.2.3. Characterization of polymerase activity

Within the VIZIER Project, a comparison of steady-state enzymatic activity parameters of both full-length NS5 and NS5RdRp domain of DENV2 and KUNV on a homopolymeric template poly(rC) suggested that the NS5MTase domain does not influence de novo RdRp activity (Selisko et al., 2006, Zhang et al., 2008), although others did not support fully this view (Yap et al., 2007). The NS5RdRp domain alone was used for screening processes and characterization of inhibitors of the flavivirus NS5RdRp activity. Furthermore, it has been shown by atomic force microcopy that a NS5RdRp domain of DENV2 with the same boundaries (272–900) binds to the circularized DENV2 RNA genome and that de novo RNA synthesis of the negative strand is enhanced by the presence of a promoter element, a large stem-loop structure, named SLA, present at the 5′-end of the genome (Filomatori et al., 2006). The authors demonstrated the physical interaction of the NS5RdRp domain with SLA. They proposed a novel mechanism for −ssRNA synthesis in which the flavivirus NS5RdRp is recruited by and specifically binds SLA at the 5′-end of the genome. It then reaches the site of initiation at the 3′-end recruited to the 5′-end via long-range RNA–RNA interactions.

The VIZIER Project contributed decisively to a precise identification of the flaviviral NS5RdRp domain and to its subsequent structural characterization (Selisko et al., 2006, Malet et al., 2007, Yap et al., 2007). Our contributions will greatly facilitate the exploration of the flavivirus NS5RdRp as a drug target (Rawlinson et al., 2006, Malet et al., 2008), hopefully leading to the discovery and design of drugs against flaviviruses.