Mutagenesis of the NS3 Protease of Dengue Virus Type 2

Abstract

The flavivirus protease is composed of two viral proteins, NS2B and NS3. The amino-terminal portion of NS3 contains sequence and structural motifs characteristic of bacterial and cellular trypsin-like proteases. We have undertaken a mutational analysis of the region of NS3 which contains the catalytic serine, five putative substrate binding residues, and several residues that are highly conserved among flavivirus proteases and among all serine proteases. In all, 46 single-amino-acid substitutions were created in a cloned NS2B-NS3 cDNA fragment of dengue virus type 2, and the effect of each mutation on the extent of self-cleavage of the NS2B-NS3 precursor at the NS2B-NS3 junction was assayed in vivo. Twelve mutations almost completely or completely inhibited protease activity, 9 significantly reduced it, 14 decreased cleavage, and 11 yielded wild-type levels of activity. Substitution of alanine at ultraconserved residues abolished NS3 protease activity. Cleavage was also inhibited by substituting some residues that are conserved among flavivirus NS3 proteins. Two (Y150 and G153) of the five putative substrate binding residues could not be replaced by alanine, and only Y150 and N152 could be replaced by a conservative change. The two other putative substrate binding residues, D129 and F130, were more freely substitutable. By analogy with the trypsin model, it was proposed that D129 is located at the bottom of the substrate binding pocket so as to directly interact with the basic amino acid at the substrate cleavage site. Interestingly, we found that significant cleavage activity was displayed by mutants in which D129 was replaced by E, S, or A and that low but detectable protease activity was exhibited by mutants in which D129 was replaced by K, R, or L. Contrary to the proposed model, these results indicate that D129 is not a major determinant of substrate binding and that its interaction with the substrate, if it occurs at all, is not essential. This mutagenesis study provided us with an array of mutations that alter the cleavage efficiency of the dengue virus protease. Mutations that decrease protease activity without abolishing it are candidates for introduction into the dengue virus infectious full-length cDNA clone with the aim of creating potentially attenuated virus stocks.

The Flaviviridae are a family of small, enveloped, positive-stranded RNA viruses. Yellow fever virus (YFV) and the dengue viruses are members of the Flavivirus genus of this family, which includes two additional genera, the hepatitis C viruses (HCV) and the pestiviruses. The flavivirus RNA genome of 10 to 11 kb is capped, lacks polyadenylation, and encodes a single open reading frame, which is translated into a polyprotein precursor. Mature viral products are generated via a series of co- and posttranslational proteolytic processing events (for a review, see reference 5). From the N to the C terminus of the viral polyprotein there are three structural and seven nonstructural (NS) proteins: C (core), prM (precursor to membrane), E (envelope), NS1, NS2A, NS2B, NS3, NS4A, NS4B, and NS5. At least two different host proteases and a viral protease are involved in this multistep protein maturation process (see references cited in reference 12). Cleavage at the C-prM, prM-E, E-NS1, and NS4A-NS4B junctions is performed by the endoplasmic reticulum enzyme signalase. NS1-NS2A processing may represent an aberrant signalase cleavage or may be effected by another host enzyme residing in the endoplasmic reticulum (12). The maturation of prM to the membrane structural protein, which occurs at a late stage in virion assembly, involves a host enzyme located in an acidified compartment of the exocytic pathway (33).

The virus-encoded protease is a complex of NS2B and NS3 (2, 7, 10). The viral protease is responsible for cleavage at a number of sites, including NS2A-NS2B, NS2B-NS3, NS3-NS4A, and NS4B-NS5; it also cleaves just upstream of the signal sequences at the C-prM and NS4A-NS4B junctions, within NS2A, and within NS3 itself (40; see references cited in reference 12). The high degree of specificity of the virus-encoded protease is well established, as is the conservation of cleavage sites among flaviviruses (for a review, see reference 5). Cleavage usually occurs following a pair of basic amino acids (Lys-Arg, Arg-Arg, or Arg-Lys) and before a small side chain amino acid (Gly, Ser, or Ala). The cleavage sequence at the NS2B-NS3 junction in the dengue viruses is somewhat different, as it contains the pair Gln-Arg (for a review, see reference 5). In YFV, it was shown that two residues upstream and one residue downstream of the cleavage site were the primary determinants of target site recognition (8, 20, 26).

Amino acid sequence alignments led Bazan and Fletterick (3) and Gorbalenya et al. (15) to propose that flavivirus NS3 is a serine protease with homology to trypsin-like enzymes. Bazan and Fletterick (3) aligned the amino acid sequences of the N-terminal portions of the NS3 proteins of several flaviviruses with the sequences of some cellular and bacterial serine proteases of the trypsin family. The alignment, coupled with available information on the serine protease crystal structures, showed that all of these proteins share highly conserved amino acid residues, including residues that are known to be involved in catalysis and substrate binding. The regions of homology fall into four clusters, called homology boxes 1, 2, 3, and 4. Boxes 1, 2, and 3 include, respectively, the active-center His, Asp, and Ser residues, which form the so-called catalytic triad of all serine proteases and which are found at positions in the viral proteins spatially similar to those in the trypsin-like enzymes. Homology boxes 3 and 4 contain residues that are involved in substrate binding and recognition. The alignment of Bazan and Fletterick also included the amino acid sequences of p80, the protease encoded by a related pestivirus (bovine viral diarrhea virus [BVDV]), and the cysteine proteases of several picornaviruses. Similarly, comparing a consensus of flavivirus sequences and a consensus of trypsin-like protease sequences, Gorbalenya et al. (15) reported that four homology stretches, which fall within the four boxes defined by Bazan and Fletterick, could be found around the catalytic triad residues and, in particular, around the nucleophilic Ser in the form of the GxSGxP sequence (where x is any amino acid).

In a number of flavivirus systems, the proposed catalytic residues were proven to be essential for protease activity. Site-directed mutagenesis experiments performed with YFV showed that replacement of the putative catalytic triad residues abolished protease activity in vitro, and when the changes were incorporated into the infectious full-length cDNA clone, virus was not recovered (6). The putative catalytic His was also replaced by Ala in West Nile virus (41), and the catalytic Ser was replaced by Thr in Murray Valley encephalitis virus (21); in both cases, no protease activity could be detected in vitro.

So far only one study (32) has attempted to investigate the determinants of flavivirus protease specificity. Chimeric NS2A-NS2B-NS3 constructs between dengue virus type 2 (DEN2) and YFV were assayed for cleavage in vitro. However, this study involved swapping of very large regions (all of NS2B, or NS3 fragments containing three or even all four of the homology boxes), thus changing numerous residues at a time, making it impossible to determine the roles of individual residues.

Here we present the results of a mutagenesis study of the flavivirus NS3, aimed at identifying key residues that determine substrate recognition and cleavage efficiency. Based on the proposed homology (3, 15) and on the published sequence alignment (3), we have specifically targeted the majority of the amino acids in or around the region defined by Bazan and Fletterick (3) as boxes 3 and 4. Starting from a pTM1-NS2B-NS3 wild-type (wt) construct, in which the NS2B-NS3 region of DEN2 is under the control of the T7 promoter, we have made 46 mutants with single-site substitutions in the NS3 protease domain. The effect of each mutation on protease activity was assessed by monitoring cleavage at the NS2B-NS3 junction in proteins expressed in vivo from wt and mutant constructs.

MATERIALS AND METHODS

Cells, bacterial strains, and viruses.

LLC-MK2 cells grown in Eagle’s minimal essential medium (EMEM) supplemented with 10% fetal bovine serum, glutamine, and gentamicin (EMEM10) were used for transfection experiments. All cloning experiments were performed with competent DH5αF′IQ cells (GIBCO BRL) except for site-directed mutagenesis, in which case CJ236 cells (Bio-Rad) were used to generate uracil-containing single-stranded DNA templates. Vaccinia virus recombinant vvTF7-3 (25), which expresses the T7 RNA polymerase, was originally obtained from Bernard Moss (National Institute of Allergy and Infectious Diseases, Bethesda, Md.).

NS2B-NS3 parental plasmid construction.

DEN2 New Guinea C strain (NGC) viral RNA was purified as described elsewhere (30). The DEN2 NS2B-NS3 sequence was first cloned in the pGEM-3Zf(−) vector (Promega). Viral RNA was used as a template in a reverse transcriptase (Stratascript; Stratagene) reaction primed with the oligonucleotide OLIGO RV1 to synthesize a cDNA encompassing the NS2B-NS3 region. Conditions for cDNA synthesis were as described elsewhere (30). OLIGO RV1 was designed with a BamHI site and a stop codon at its 5′ end followed by the complement of nucleotides (nt) 5076 (corresponding to amino acid [aa] 185 of the NS3 protein) to 5061 of the DEN2 RNA sequence. cDNA synthesis was followed by PCR using the same downstream OLIGO RV1 and OLIGO RV2 as the upstream primer. OLIGO RV2 contains a SacI site at its 5′ end and the sequence GCCATG to initiate translation, followed by DEN2 sequences from nt 4132 (the first nt of NS2B) to 4146. The 50-μl PCR mixture contained 0.5 μl of the reverse transcription reaction product, 0.2 μg of each primer, 350 μM each deoxynucleoside triphosphate (dNTP), and 2.6 U of Expand DNA polymerase in Buffer 1 (Boehringer Mannheim). After a 90-s denaturation at 94°C, amplification was performed for 30 cycles, with each cycle consisting of a 10-s denaturation step at 94°C, a 30-s annealing step at 60°C, and a 1-min polymerization step at 68°C. The 0.95-kb PCR product was digested with SacI and BamHI and cloned into similarly digested pGEM-3Zf(−) vector, creating pGEM-NS2B-NS3. The sequences of this and all other constructs were verified by dideoxynucleotide sequencing (Sequenase 2.0 kit; Amersham) of the entire PCR-amplified region. The sequence of pGEM-NS2B-NS3 differed from the published DEN2 NGC sequence (18) at five positions: nt 4428, C to T (silent); nt 4471, A to C (Ile to Leu), located in the C-terminal region of NS2B; nt 4829, G to A (Gly to Glu), located in the NS3 portion, between boxes 2 and 3; nt 4932, G to T (silent); and nt 4990, A to G (Arg to Gly), three residues downstream of box 4 in NS3.

To clone the NS2B-NS3 region in the pTM1 vector (25), a PCR was set up with OLIGO RV1 (downstream) and OLIGO RV17 (upstream) as primers and 10 ng of pGEM-NS2B-NS3 as the template. The conditions were identical to those described above. OLIGO RV17 carries a BspHI recognition sequence (TCATGA), in which the ATG functions as a translation initiator codon and the last A corresponds to the first nt of the NS2B region (nt 4132), followed by nt 4133 to 4142 of the DEN2 sequence. The amplified fragment was digested with BspHI and BamHI and cloned into NcoI- and BamHI-digested pTM1 to make pTM1-NS2B-NS3. This clone, which will hereafter be referred to as the parental construct, differs from the published DEN2 NGC sequence (18) at the same five positions as pGEM-NS2B-NS3, from which it was derived, and carries an extra mutation at nt 4256, where an A-to-G transition changes a Tyr to a Cys residue in the N-terminal region of NS2B.

Construction of NS2B- and NS3-expressing plasmids.

To create a pTM1 plasmid which expresses only NS2B, pGEM-NS2B-NS3 was used as the template in a PCR with OLIGO RV17 (upstream) and OLIGO RV19 (downstream) as primers, using Expand DNA polymerase, under the conditions described above for the pTM1-NS2B-NS3 parental plasmid. OLIGO RV19 carries, 5′ to 3′, the complement of nt 4536 to 4522 (the start of the NS3 region), a BamHI site, a stop codon, and the complement of nt 4521 (the end of the NS2B region) to 4507. The 0.4-kb amplified fragment was digested with BspHI and BamHI and cloned into NcoI- and BamHI-digested pTM1, creating pTM1-NS2B. This construct contains the same mutations in NS2B as the plasmid used as a template (i.e., at nt 4428 and 4471). Plasmid pTM1-NS3 was obtained similarly, using OLIGO RV18 and OLIGO RV1. OLIGO RV18 (upstream) carries, 5′ to 3′, 15 nt corresponding to nt 4505 to 4519 (the NS2B C-terminal region), an NcoI sequence (CCATGG) encompassing an ATG initiator codon followed by the first nt of the NS3 region, and then another 14 nt corresponding to nt 4522 to 4535 (the NS3 N-terminal region). The 0.6-kb PCR product was digested with NcoI and BamHI and cloned into NcoI- and BamHI-cut pTM1, creating pTM1-NS3. This plasmid contains the three mutations in NS3 present in pGEM-NS2B-NS3 plus an additional one at nt 4808, where an A-to-G transition results in a Gln-to-Arg change.

Nomenclature.

Mutants are designated by the wt aa, using the standard single-letter code, followed by the number of the relevant residue within the NS3 protein region involved (aa 1 to 185) and by the replacement aa. For example, D129E indicates that the Asp residue at position 129 in the NS3 protease was mutated to a Glu residue.

Construction of mutant plasmids.

Overall, 46 mutants were constructed, all targeting aa residues in or around boxes 3 and 4. Mutagenic oligonucleotides were designed with the following features: (i) the desired mutation; (ii) an additional silent mutation that creates a new restriction enzyme site for screening purposes (except in one clone); and (iii) a Bsu36I restriction site at the 5′ end, corresponding to the Bsu36I site located at nt 4921 in the DEN2 sequence. Each mutagenic oligonucleotide was used for PCR in combination with a wt oligonucleotide: either OLIGO RV18 (when the mutation site was located upstream of the Bsu36I site) or OLIGO RV1 (when the mutation site was located downstream of the Bsu36I site). The template was the pTM1-NS2B-NS3 parent plasmid (with the six nucleotide differences from the original DEN2 NGC sequence [see above]). The 50-μl PCR mixture contained 10 ng of template, 0.2 μg of each primer, 200 μM each dNTP, and 2.5 Units of Pfu polymerase (Stratagene). After 2 min 30 s of denaturation at 94°C, the reaction was subjected to 35 cycles, each consisting of a 10-s denaturation step at 94°C, a 30-s annealing step at 60°C, and a 1-min polymerization step at 72°C. The PCR product (420 or 160 nt) was digested either with NsiI (which cuts at nt 4695 in the DEN2 sequence) and Bsu36I or with Bsu36I and BamHI and then used to replace the homologous wt fragment in pTM1-NS2B-NS3. In the case of the R184A mutant, the mutation destroyed the BstBI site (nt 5068 in the DEN2 sequence), which is located far from the Bsu36I site. The design used was therefore slightly different: the mutagenic oligonucleotide had a BamHI site at its 5′ end and was used in PCR in combination with OLIGO RV18. The PCR product was digested with Bsu36I and BamHI and cloned into similarly digested pTM1-NS2B-NS3. Mutants were screened for the presence of the added restriction enzyme site (or for the loss of a site, in the case of R184A) and then verified by dideoxynucleotide sequencing.

Five of the mutations, which were located around the Bsu36I site, were obtained by site-directed mutagenesis by use of the Bio-Rad Mutagene kit. Briefly, a uracil-containing single-stranded DNA template was prepared in CJ236 cells from the parental pTM1-NS2B-NS3 plasmid that harbors an f1 origin of replication, after infection with the M13K07 helper phage. The template was hybridized to each of the mutagenic oligonucleotides OLIGO RV49 to OLIGO RV53, which are complementary to the DEN2 sequence. In OLIGO RV50 to OLIGO RV53, the encoded mutations destroyed the Bsu36I site. In the case of OLIGO RV49, the loss of that site was due to an additional change that did not affect the aa sequence. Template-oligonucleotide hybridization was followed by an in vitro polymerization and ligation reaction step, in accordance with the kit instructions, and the double-stranded product was used for transformation of DH5αF′IQ cells. Clones were screened for the loss of the Bsu36I site, and then the presence of the mutation was verified by sequencing.

NS2B-NS3 wt plasmid construction.

A pTM1-NS2B-NS3 construct containing a fully correct aa sequence was also obtained upon PCR amplification of the DEN2 infectious full-length clone (30) with OLIGO RV17 and OLIGO RV1. The 50-μl reaction mixture contained 10 ng of template, 0.2 μg of each primer, 200 μM each dNTP, and 2.5 U of Pfu polymerase. After 2 min 30 s of denaturation at 94°C, the reaction was subjected to 35 cycles of a 10-s denaturation step at 94°C, a 30-s annealing step at 60°C, and a 1-min polymerization step at 72°C. The amplified fragment was digested with BspHI and BamHI and ligated into NcoI- and BamHI-digested pTM1, creating wt pTM1-NS2B-NS3.

Construction of mutant plasmids in the wt background.

All 46 mutants were reconstructed to create plasmids carrying the relevant mutation in the wt rather than in the parental DEN2 NS2B-NS3 sequence background. Mutants D129E, V126A, R185A, S131P, K142A, and K143N were made by PCR as described above, except that the wt pTM1-NS2B-NS3 plasmid was used as the template and the PCR fragment was gel purified before restriction enzyme digestion. Cloning was done in wt pTM1-NS2B-NS3 also digested with the appropriate enzymes. The other 40 mutants were reconstructed by the following strategy: the NdeI fragment (nt 4755 to 5000 of the DEN2 sequence) from each mutant plasmid was amplified in a PCR under the conditions previously described for creating the mutant plasmids, except that 30 cycles were used rather than 35. The primers were OLIGO RV59 (an upstream primer carrying DEN2 sequences from nt 4746 to 4838) and OLIGO RV60 (a downstream primer carrying, 5′ to 3′, the complement of nt 5012 to 4981). To create mutants V154A and V155A, OLIGO RV60 was replaced by OLIGO RV61 (complementary to nt 5012 to 4983) and by OLIGO RV62 (complementary to nt 5012 to 4986), respectively. OLIGO RV59 corrects the G-to-A transition at nt 4829, and OLIGO RV60 to OLIGO RV62 correct the A-to-G transition at nt 4990 which is present in the parental NS2B-NS3 sequence background. After gel purification and NdeI digestion, each PCR fragment carrying the relevant mutation in box 3 or 4 was ligated to NdeI-digested, dephosphorylated, and gel-purified wt pTM1-NS2B-NS3. Positive clones were screened for the presence of the relevant mutation and the orientation of the NdeI fragment and fully sequenced over the entire PCR fragment either with a Sequenase 2.0 kit or by automated sequencing, using the ABI PRISM dye terminator cycle sequencing system with AmpliTaq DNA polymerase FS (Perkin-Elmer Applied Biosystems) and a Perkin-Elmer ABI Prism 377 DNA sequencer. With this scheme, all of the point mutations in the NS3 mutant plasmids that are inherent to the parental pTM1-NS2B-NS3 construct are corrected except the silent mutations at nt 4428 and 4932.

Other plasmids.

Parent and wt pTM1-NS2B-NS3 differ from each other at four positions: nt 4256, 4471, 4829, and 4990. Chimeric constructs, designated a to f, which carry various combinations of the four differences (see Fig. 2A) were generated as follows. The AvrII-NsiI, NsiI-Bsu36I, and Bsu36I-BamHI fragments were gel purified from parental and wt pTM1-NS2B-NS3 and recloned into similarly digested wt or parental pTM1-NS2B-NS3.

FIG. 2.

(A) Diagram of the parental pTM1-NS2B-NS3 plasmid (p) showing the position of the point mutations yielding the four aa changes: Y to C (nt 4256) and I to L (nt 4471) in the NS2B protein and G103E (nt 4829) and R157G (nt 4990) in the NS3 protein. The two silent mutations at nt 4428 and 4932 have been omitted. The diagram also outlines the structures of the wt basic plasmid and of the chimeric plasmids, a to f. The aa present at each relevant position in those plasmids is indicated. Restriction enzyme sites used to engineer all the plasmids by fragment swapping are shown. (B) In vivo cleavage of parental (p) and wt pTM1-NS2B-NS3 and of their chimeric derivatives, a to f. Analysis was as described in the legend to Fig. 1, except that 16- by 18-cm gels were used for SDS-PAGE. v, pTM1 vector alone; M, molecular mass markers (see the legend to Fig. 1).

In vivo expression assay.

LLC-MK2 cells were grown to confluency in six-well dishes and infected with vvTF7-3 at a multiplicity of infection of 5 PFU/cell in 0.6 ml of EMEM without fetal calf serum (EMEM0) for 90 min at 37°C. DNA (5 μg) was incubated with 30 μl of Lipofectin reagent (GIBCO BRL) in 1 ml of EMEM0 for 30 min at room temperature. The virus inoculum was removed, DNA-Lipofectin complexes were added to cells, and the cells were then incubated for 5 h 30 min at 37°C. Cells were then fed with EMEM10 overnight. The next day, cells were starved in Dulbecco’s modified Eagle medium lacking methionine for 30 min at 37°C. Cells were then labeled for 90 min at 37°C with 0.6 ml of the same medium containing 120 μCi of a mixture of [³⁵S]methionine and [³⁵S]cysteine (1,000 Ci/mmol; Amersham). The labeling medium was then removed, and cells were lysed in 0.6 ml of RIPA buffer (1% sodium deoxycholate, 1% Nonidet P-40, 0.1% sodium dodecyl sulfate [SDS], 0.1 M Tris hydrochloride [pH 7.5], 0.15 M NaCl). Viscous material was removed by centrifugation, and 150 μl of clarified cell extract was used for subsequent immunoprecipitation with 15 μl of mouse hyperimmune ascitic fluid (HMAF; American Type Culture Collection) against DEN2-NGC. Incubation was for 90 min on ice. Pansorbin cells (75 μl; Calbiochem) were added, and incubation was continued for 30 min on ice. Immunoprecipitates were collected by centrifugation and washed twice with RIPA buffer supplemented with 2% SDS and 1% bovine serum albumin. Finally, immunoprecipitates were resuspended in 2× Tricine sample buffer (0.1 M Tris hydrochloride, 4% SDS, 2% β-mercaptoethanol, 20% glycerol, 0.008% bromophenol blue), vortexed extensively, heated at 90°C for 5 min, centrifuged, and loaded on 15% polyacrylamide–SDS gels (16 by 18 cm) employing a Tricine-based buffer system (35). Electrophoresis was at 120 V for 12 h. Subsequently, gels were fixed in 40% methanol–10% acetic acid and then fluorographed. When minigels (Bio-Rad) were used, electrophoresis was at 170 V for 40 min. Minigels were fixed and treated with Enlightning (NEN) before being exposed to X-ray film.

In vitro transcription and translation.

Plasmids (4 μg) were linearized at the unique XhoI site located in the multiple cloning region of the pTM1 vector, downstream of the BamHI site. In vitro transcription with T7 RNA polymerase (Promega) was performed as described previously (12). Approximately 1 μg of RNA was obtained per microliter of reaction mixture, as estimated by ethidium bromide staining and comparison with known amounts of an RNA marker.

In vitro translations were carried out as described previously (24), except that reaction mixture volumes were 10 μl and incubations were for 90 min at 25°C.

RESULTS

Expression of parental NS2B-NS3.

The NS2B-NS3 region of DEN2 NGC was cloned in the pTM1 vector (25), under the control of a T7 promoter and the encephalomyocarditis virus leader sequence, which permits cap-independent translation. Previous work has shown that a complex consisting of 185 aa of the NS3 protein and NS2B is sufficient for the occurrence of autocatalytic cleavage at the NS2B-NS3 junction in vitro and in vivo (6, 7, 10, 31, 41). The NS2B-NS3 construct was designed to generate a protease target site as well as an active protease whose self-cleavage could be assayed in vitro and in vivo. For in vitro analysis, an RNA transcript was first synthesized by T7 RNA polymerase with the linearized plasmid as the template. The RNA was then used to program a rabbit reticulocyte lysate in the presence of [³⁵S]methionine. In vivo, the plasmid could be transfected into cells which had been previously infected with a vaccinia virus recombinant (vvTF7-3) expressing the T7 RNA polymerase. Following labeling, cell extracts were immunoprecipitated with anti-DEN2 HMAF. For the in vitro and the in vivo analyses, labeled proteins were visualized by SDS-polyacrylamide gel electrophoresis (PAGE) and fluorography. The pTM1 vector was chosen because the initial pGEM-NS2B-NS3 construct yielded inefficient expression in vitro and in vivo.

Figure 1 shows that in vivo expression of parental pTM1-NS2B-NS3 yielded three major bands upon immunoprecipitation by anti-DEN2 HMAF, which were identified as the NS2B-NS3 precursor (expected molecular mass, 34 kDa) and the NS3 (expected molecular mass, 20 kDa) and NS2B (expected molecular mass, 14 kDa) cleavage products. The last two bands comigrated with the more slowly migrating product expressed by pTM1-NS3 and the product expressed by pTM1-NS2B, respectively. Expression of pTM1-NS3 also yielded a faster-migrating band of approximately the size of NS2B. This could be either a degradation product of NS3 or the result of NS3 self-cleavage occurring in the absence of cofactor at either a typical or an unusual site. Cleavage of the parent construct was incomplete both in vitro (data not shown) and in vivo (Fig. 1); less than half of the NS2B-NS3 precursor was processed into NS2B and NS3 products. These results are in contrast to those of other studies of flavivirus protease activity (6, 10, 31, 42), in which cleavage occurred efficiently in vitro and in vivo at the NS2B-NS3 junction.

FIG. 1.

In vivo cleavage of DEN2 NS2B-NS3. The NS2B-NS3 parental construct and the NS3 and NS2B control constructs were expressed in LLC-MK2 cells by using the vvTF7-3 vaccinia virus recombinant-pTM1 expression system. Expression of individual plasmids and cleavage of the NS2B-NS3 precursor (34 kDa) into NS3 (20-kDa) and NS2B (14-kDa) products was analyzed by immunoprecipitation of [³⁵S]methionine-labeled proteins with anti-DEN2 HMAF. DEN2-specific labeled proteins were separated by SDS-PAGE on a 15% polyacrylamide–Tricine minigel. Autoradiograms and annotations made by MacDraw Pr. were scanned into Adobe Photoshop with a Dicomed scanner. M, ¹⁴C-methylated protein molecular mass markers (molecular mass, 14.3 to 200 kDa) from Amersham.

Expression of wt NS2B-NS3.

The parental NS2B-NS3 construct differs from the DEN2 published sequence at six positions (see Materials and Methods). These differences were possibly introduced by the initial reverse transcription reaction and/or by PCR with the Expand DNA polymerase. As these nt differences may be responsible for the diminished self-cleavage observed upon expression of parental NS2B-NS3, a wt version of the basic construct was made by PCR using the DEN2 infectious full-length clone as the template (30) and Pfu polymerase. The aa sequence of the latter construct was the same as that of the wt while retaining the mutations at nt 4428 and 4932, which are silent mutations present in the infectious clone (30). Figure 2A schematically shows the aa differences between the parent and wt constructs. Two of those differences are unlikely to affect cleavage: the I-to-L conservative mutation at nt 4471 and the Y-to-C mutation at nt 4256, which is located in the N terminus of NS2B, a region that, at least in DEN4, is known to be dispensable for protease activity (11). The remaining differences are the G103E change (nt 4829) and the R157G change (nt 4990). Figure 2B shows the results of in vivo expression of wt and parental pTM1-NS2B-NS3. In contrast to the parental construct, wt NS2B-NS3 was almost totally converted into NS2B and NS3 products, as expected.

To specifically assess the contribution of each mutation to the difference in the protease activities of the parent and wt constructs, six additional chimeric plasmids, a to f, were generated by swapping relevant restriction fragments (Fig. 2A). Protease activity was analyzed in vivo (Fig. 2B). Plasmid a, which carries the two downstream mutations (at nt 4829 and 4990) in a wt background, exhibited a low level of protease activity, although cleavage was slightly more efficient than in the parent. Plasmid b behaved like the wt, in spite of the presence of the two upstream mutations (at nt 4256 and 4471). This indicates that, as expected, the two NS2B mutations were not responsible for the difference in the protease activities of the parent and wt constructs, although they may have had an additive effect when present in combination with the downstream mutations (compare the parent and plasmid a lanes). Plasmid c, carrying all mutations but the one at nt 4829, displayed an extent of cleavage similar to that of the wt. Plasmid d, which carries solely the G103E mutation (nt 4829), exhibited a low level of protease activity similar to those of plasmid a. Plasmid e, which carries wt R157 (nt 4990) in a parental background, cleaved poorly, at a level comparable to that of plasmid a. Finally, plasmid f, carrying the G157 mutation in a wt background, cleaved similarly to the wt. Taken together, these data suggest that the difference between the parent and the wt is largely accounted for by the presence of the G103E change.

Mutagenesis rationale.

Our mutagenesis strategy was based on the work of Bazan and Fletterick (3). Figure 3 shows the alignment at residues 120 to 167 (DEN2 NS3 aa numbering) of the NS3 proteins of numerous flaviviruses and of HCV. For all of these viruses, it is known that the protease domain of NS3 maps to the first 180 to 185 aa. Moreover, residues 120 to 167 should include the region defined as boxes 3 and 4. This region contains the catalytic S as well as residues known to be part, both structurally and functionally, of the substrate binding pocket of trypsin-like enzymes. The boundaries of these boxes have been drawn in Fig. 3 according to the model of Bazan and Fletterick. The main features of boxes 3 and 4 indicated are the catalytic S at position 135, embedded in the GxSGxP motif (15), and five residues that Bazan and Fletterick identified as “contributing to substrate binding.” In DEN2 these are D129 and F130 in box 3 and Y150, N152, and G153 in box 4. Using the MACAW alignment program (37), we found a 39-aa-long sequence (aa 120 to 158) displaying significant homology within the aligned viral sequences. This homology stretch is shown in capital letters in Fig. 3. Although larger, this sequence overlaps with boxes 3 and 4.

FIG. 3.

Alignment of amino acid sequences of flavivirus NS3 proteases in the region of boxes 3 and 4. The 39-aa stretch of significant homology detected by the MACAW alignment program is shown in capital letters. Dashes indicate gaps in the alignment. The boundaries of boxes 3 and 4 shown are according to the model of Bazan and Fletterick (3); the five putative substrate binding residues are denoted by dots, and the catalytic S135 residue is highlighted by an asterisk. DEN2 NS3 (DEN2 aa#) and chymotrypsin (CT aa#) aa numbering are both indicated. Sequences are as follows: DEN1 (14), DEN2 (18), DEN3 (27), DEN4 (23), Japanese encephalitis virus (JE) (1), Kunjin virus (KUN) (9), tick-borne encephalitis virus (TBE) (28), West Nile virus (WN) (4), YFV (34), and HCV (17).

We targeted 20 positions in boxes 3 and 4 for mutagenesis, producing a total of 42 mutants. The original aa was replaced either with A as a neutral replacement (A is a neutral, nonpolar aa that is neither bulky nor a “helix breaker”), with an aa similar in nature to the original one, or with an aa that is present in that position in other members of the serine protease family. In addition to these 42 constructs, four mutants with mutations that are technically located outside of boxes 3 and 4 were made: V126A, G144A, G144P, and R184A. V126 is located outside of box 3 in the original alignment of Bazan and Fletterick but is inside the 39-aa homology stretch that is shown in Fig. 3. G144 is located between boxes 3 and 4; interestingly, G144 is conserved in all flavivirus and HCV. Finally, R184 is located outside of box 4. V126A, G144A, and R184A were routinely used as controls, since residues outside of the boxes were predicted to be nonessential.

The locations of all 46 mutations and a summary of their effects on NS2B-NS3 autocleavage are shown in Fig. 4. The code used to designate the mutation effect is as follows: underlined lowercase letters indicate barely detectable or total lack of activity, nonunderlined lowercase letters indicate very low activity, nonunderlined uppercase letters indicate decreased protease activity compared to that of the wt, and underlined uppercase letters indicate protease activity similar to that of the wt. The proteolytic activities of wt and mutant constructs were qualitatively assessed. Each mutant was independently assayed in vivo at least two times. Figure 4 reports the average result from all experiments performed on each mutant. Typical results from in vivo experiments using the mutants in the wt background are shown in Fig. 5 to 7. Note that in vitro and in vivo analyses of all 46 mutants in the parental background were also performed (data not shown); in vitro data were found in general to corroborate in vivo data.

FIG. 4.

Summary of results of site-directed mutagenesis of DEN2 NS3. The aa sequence of DEN2 NS3 (residues 126 to 184) in the protease region encompassing boxes 3 and 4, as defined in the model of Bazan and Fletterick, is shown (DEN2 aa#). The numbering of the corresponding residues in chymotrypsin is indicated (CT aa#). Ultraconserved residues among serine proteases are enclosed in small boxes. The five putative substrate binding residues are highlighted by dots, and the catalytic S135 residue is indicated by an asterisk. Forty-six mutants were made in a NS2B-NS3 construct; all have single-aa substitutions (indicated by the one-letter code) in or around boxes 3 and 4. The cleavage activity of each mutant at the NS2B-NS3 junction was assayed in vivo, and the results were classified as follows: wt levels of cleavage (underlined capital letters), decreased activity (nonunderlined capital letters), very low activity (nonunderlined lowercase letters), or barely detectable to undetectable activity (underlined lowercase letters).

FIG. 5.

In vivo analysis of cleavage activity of NS3 mutants with mutations in box 3, residues 126 to 134. Expression and analysis of the protease activity of each construct were as described in the legend to Fig. 2B. Amino acid changes for each relevant residue of the NS3 protein are designated by the single-letter code. wt, wt NS2B-NS3 construct; V126A, G144A, and R184A, control mutants (expected not to alter protease activity); M, molecular mass markers (as described in the legend to Fig. 1).

FIG. 7.

In vivo analysis of cleavage activity of NS3 mutants with mutations in box 4. Analysis of protease activity of each construct was performed as described in the legend to Fig. 5. M, molecular mass markers (see the legend to Fig. 1).

Mutagenesis of box 3.

Figure 5 shows the extent of in vivo cleavage of NS2B-NS3 precursors carrying one mutation between residues 129 and 134. D129, one of the five putative substrate binding residues, was changed to E, S (present in this position in chymotrypsin and in the HCV protease), A, K, R, or L. D129E, D129S, and D129A had decreased protease activity compared to the wt; processing appeared to be moderately efficient, with roughly one-half of the NS2B-NS3 precursor cleaved into products. The D129 K, R, and L substitutions resulted in a greater decrease in protease activity; while there is still good evidence of NS2B and NS3 products, cleavage efficiency was poor. D129L exhibits the lowest activity of this set of mutants. F130 was mutated to Y (present in other flavivirus proteases and in that of HCV), A, and S (present in trypsin and in chymotrypsin). This position, another of the five putative substrate binding residues, was more sensitive than D129 to modification. The conservative Y substitution decreased protease activity, in a manner similar to D129E, but the A and S substitutions reduced protease activity even further (product bands were visible in the original autoradiogram). NS2B-NS3 cleavage was unaffected in the S131P mutant (P is present in this position in most of the flavivirus proteases) and in S131C (C is present at this position in various cellular serine proteases). On the other hand, autoprocessing was strongly reduced by substitution of A for G133, which is one of the ultraconserved residues, part of the GxSGxP motif. The neighboring T134 residue, although not part of that motif, is conserved among flavivirus proteases (a conserved D is present at this position in cellular serine proteases). The T134A substitution only slightly affected protease activity, but the T134D substitution almost totally inhibited it; no evidence of cleavage products was observed in one experiment, and a faint NS2B band could be barely detected in another. Control mutants G144A and R184A behaved as expected, consistently yielding wt levels of cleavage. Conversely, the V126A control mutant showed decreased proteolytic activity.

Figure 6 shows the effect of mutations in residues 135 to 144 of the NS3 protease domain. The catalytic S at position 135, as expected, could not tolerate substitution with A and served as a negative control; no NS3 and NS2B cleavage products were detected. The S135C change did result in a slight residual activity, since traces of product bands were barely detected in the original autoradiograms. G136 is an ultraconserved residue among all serine proteases and is part of the GxSGxP motif. Mutation G136A strongly decreased the activity of the viral protease. At position 139, I is conserved among flavivirus proteases (the related HCV protease has L in this position) and a hydrophobic residue is conserved in cellular serine proteases. When I139 was changed to either A or L in DEN2 NS3, cleavage efficiency was only partly affected. The next position, I140, is occupied by hydrophobic aa in viral and cellular proteases. I140A behaved similarly to I139A and I139L, while I140L maintained wt levels of protease activity. Replacing D141, K142, or K143 either with A or with an aa close in nature to the original one did not affect autoproteolytic activity; the extent of cleavage was similar to that of the wt. Finally, G144, which separates boxes 3 and 4, behaved like the wt construct when replaced by A, but cleavage was reduced when G144 was changed to P, an aa known to disrupt protein secondary structure.

FIG. 6.

In vivo analysis of cleavage activity of NS3 mutants with mutations in box 3, residues 135 to 144. The panels on the left and right are from two independent in vivo experiments. Analysis of protease activity of each construct was performed as described in the legend to Fig. 5. M, molecular mass markers (see the legend to Fig. 1).

Mutagenesis of box 4.

Figure 7 shows the extent of in vivo cleavage of NS2B-NS3 precursors carrying one mutation between residues 148 and 155. G148 and L149, two ultraconserved residues among serine proteases, could not be replaced by A without loss of protease activity. Residue 149 could be replaced by I (present in cellular serine proteases), although some decrease in activity resulted, but not by the positively charged R (present in BVDV p80). In the latter case, protease activity was completely destroyed. When Y150, the third substrate binding residue according to the model, was changed to F, the mutant NS2B-NS3 construct showed decreased proteolytic activity, but substitution of A, V (present in cellular proteases and in BVDV p80), or H (present in picornavirus cysteine proteases) totally eliminated cleavage. When the next residue, G151, was mutated to A, the autocatalytic activity of NS2B-NS3 was abrogated. N152, the fourth substrate binding residue, could be replaced by A or Q with partial loss of proteolytic activity. When the ultraconserved G153, the fifth substrate binding residue, was changed to A or V, total loss of protease activity occurred. Finally, cleavage was reduced when either V154 or V155, located at the right border of box 4, was replaced by A.

DISCUSSION

With the immediate goal of identifying mutations that can alter the cleavage efficiency of the DEN2 protease, we mutagenized the NS3 protease domain in boxes 3 and 4, the region proposed by Bazan and Fletterick (3) to contain five substrate binding residues, the catalytic S, and some residues that are ultraconserved among all serine proteases. Of the 46 mutations, 12 almost completely or completely inhibited protease activity, 9 significantly reduced it, 14 decreased cleavage, and 11 yielded wt or nearly wt levels of activity (Fig. 4).

Control mutants R184A and G144A functioned as expected, but V126A and G144P, which exhibited reduced cleavage, did not. Contrary to the model (3), our results point to a role for V126 and G144 in DEN2 NS3 protease activity. This interpretation is strengthened by the fact that V126 is present in all flaviviruses and that G144 is conserved throughout the alignment (Fig. 3). It is worth noting that the definition of the four homology boxes by Bazan and Fletterick was based on the sequence alignment and structural information of trypsin-like enzymes and was extended to flavivirus proteases on the assumption that the three-dimensional structure of viral proteases is similar to that of trypsin-like enzymes. This assumption may not completely hold, since viral proteases exhibit peculiar features that are not shared by cellular trypsin-like serine proteases, such as being covalently attached to a helicase domain (39), requiring a cofactor, and cleaving in cis as well as in trans. Hence, based on the aa conservation of V126 and G144 and on our mutagenesis data, we suggest that one homology box should encompass at least residues 126 to 158 (Fig. 3).

In spite of having the same catalytic triad and similar three-dimensional structures, each member of the serine protease family exhibits a unique substrate specificity that is primarily determined by the structural characteristics of the substrate binding pocket. In trypsin, a D residue (D189 in the conventional chymotrypsin numbering system) lies at the base of the substrate binding pocket and forms an electrostatic bond either directly or indirectly, via water molecules, with the K or R residue at the cleavage site (16). In chymotrypsin, which cleaves on the carboxy-terminal side of aromatic aa, D189 is replaced by S189, and this is thought to be the basis for the difference in specificity between trypsin and chymotrypsin. Interestingly, a mutant trypsin in which D189 is replaced by K loses its trypsin-like activity toward basic aa but does not acquire specificity for acidic substrates, because it appears that the positive charge of K is directed outside the substrate binding pocket (16). In DEN2 NS3, the homolog of trypsin D189 is D129, one of the five proposed substrate binding residues. Interestingly, D129 was the most freely substitutable residue in our mutagenesis study (Fig. 4); in particular, it could be replaced by E, S, or A and still retain significant activity, while K, R, or L substitution significantly reduced but did not eliminate the activity. These data raise three possibilities. First, a residue in DEN2 NS3 other than D129 may be in direct contact with the basic aa at the cleavage site. Consistent with this is the HCV protease, which cleaves four junctions in the viral polyprotein after C or T residues. Amino acid alignment indicates that the residue corresponding to D189 of trypsin (D129 in DEN2 NS3) is replaced by S in the HCV protease, as in chymotrypsin (Fig. 3). However, crystallography data (19, 22) showed that the aa sitting at the bottom of the substrate binding pocket of the HCV NS3 is actually F213 (chymotrypsin numbering; corresponding to Y150 in DEN2), not S189. The second possibility is that D129 is in direct contact with the basic aa at the cleavage site but when D129 is substituted, the new aa can still interact indirectly with the cleavage site via a molecule of water. In fact, charge stabilization does not necessarily involve ionic interactions; it may be achieved by a hydrophobic environment as well, as in the case of the acetylcholine-acetylcholinesterase interaction (38). It may, therefore, be possible for the basic aa at the cleavage site to be stabilized by nonpolar side chains. Third, it is possible that D129 is not the only residue involved in direct binding to the substrate, so that the loss of this one contact is not enough to abrogate cleavage. Given the nature of the flavivirus protease cleavage site, i.e., specificity at residues P2, P1, and P′1, in the conventional nomenclature of Schechter and Berger (36), it is likely that interactions at residues other than P1 will be important. A simplistic suggestion is that if recognition of one basic aa requires one D in the substrate binding pocket, recognition of two basic aa may require two D’s. The other D could even be located outside of boxes 3 and 4. In cellular serine proteases, the so-called “met loop” is positioned such that it can interact with residues of the substrate upstream of the cleavage site, i.e., P2 to P5. The met loop could be a good candidate for hosting the other D, if it is present in the flavivirus protease. Alternatively, in addition to its already suggested role of providing NS3 with a membrane attachment, the NS2B cofactor of the flavivirus protease may contribute additional points of contact between the protease and its substrate. Therefore, the other D could be provided by the domain of NS2B that is essential to protease activity. This domain has been identified as a 40-aa hydrophilic sequence which is conserved among flaviviruses (11). Interestingly, there are five conserved acidic residues in the DEN4 NS2B essential domain; mutagenesis of this domain is in progress (13).

Mutagenesis of the other four putative substrate binding residues confirmed the importance of these residues in cleavage by the DEN2 protease (Fig. 4). Protease activity in the F130Y mutant was decreased, and that in the F130A and F130S mutants was greatly decreased. At position 150, only the conservative Y150F mutation is tolerated, leading to a decrease in protease activity. A Y150F mutant in the context of a DEN2 construct was also tested by Preugschat et al. (32), and wt cleavage levels (at the NS2A-NS2B and NS2B-NS3 junctions) were observed. We also found that protease activity was virtually inactivated by replacement of Y150 with A, V, or H. Note that the HCV homolog of DEN2 Y150 is F213 (chymotrypsin numbering), and it is the residue located at the bottom of the HCV substrate binding pocket (19). Mutations N152A and N152V severely affected protease activity but were not as deleterious as G153A or G153V. Preugschat et al. (32) replaced N152 with K in the context of a DEN2 construct, and protease activity was lost. Trypsin and chymotrypsin, which differ in their substrate specificity at P1, have the same aa at these four positions (S190, V213, W215, and G216; chymotrypsin numbering). By analogy, this suggests that F130, Y150, N152, and G153 in the DEN2 protease might not be determinants of the specificity at P1. This could be addressed by examining NS2B-NS3 cleavage in constructs with double mutations targeted to the substrate binding residues as well as to the cleavage site residues. Clearly, the complexity lying behind the homology between flavivirus and cellular trypsin-like proteases will be uncovered only by determination of the three-dimensional structure of the flavivirus NS2B-NS3 protease.

Of paramount importance in substrate selection by cellular serine proteases are the two aa residing at positions 216 and 226 (chymotrypsin numbering), which form the portal to the substrate binding pocket (29). In trypsin and chymotrypsin these are both G residues. In contrast, elastase has the bulkier residues V216 and T226. Elastase exhibits a very broad substrate recognition and cleaves peptide bonds after small aa, because only small aa can be accommodated by the shallow entrance to the substrate binding pocket provided by V216 and T226. Flavivirus proteases have a G homolog of G216 (in DEN2 NS3, G153). Our mutagenesis data are consistent with G153 having a major role in DEN2 protease substrate binding, since it appears to be unsubstitutable (Fig. 4). Although we cannot discriminate between the homology due to the remarkable phylogenetic conservation among flavivirus aa sequences and the homology due to the functional conservation of these viral proteases, it is noteworthy that the position homologous to chymotrypsin residue 226 in all flavivirus proteases is occupied by S (in DEN2 NS3, S163), while that position in the HCV protease is occupied by V, as confirmed by structural data (22). G153 and S163 in flavivirus NS3 form a slightly bulkier entry to the substrate binding pocket than G216 and G226 in chymotrypsin. This may be related to the requirement for a small side chain residue at P′1 in flavivirus cleavage sites.

Residues in DEN2 NS3 that are ultraconserved among all serine proteases, including the catalytic S135 as well as G133, G136, G148, L149, and G153, are likely to play a structural role either in maintaining the overall three-dimensional structure or in shaping the substrate binding pocket. Thus, it is not surprising that these residues do not tolerate any changes (Fig. 4). On the other hand, there are residues that are very sensitive to mutation despite the fact that they are neither ultraconserved among serine proteases nor substrate binding residues according to the model. This is the case for T134 and G151. These positions are highly conserved among flavivirus NS3 sequences, and therefore the data suggest that they may contribute to selective substrate recognition and cleavage specificity of these proteases.

Originally, the 46 mutants were constructed and assayed in the parental NS2B-NS3 background. The demonstration of the deleterious effect of E103 on protease activity and the potential additive effect of the two upstream mutations in the parental construct (Fig. 2) prompted us to engineer all 46 mutants back into the wt background. Mutations in the parental or wt background generally resulted in the same effect on protease activity, with the exception of G133A, G136A, I139A, I140A, G144A, and N152A, which scored better in the wt background. This suggests that the phenotype observed for each of these six mutants in the parental construct was the result of a synergistic interaction between that mutation and the G103E change present in the parental plasmid.

The work presented here is an analysis of the determinants of substrate recognition and cleavage efficiency in a flavivirus-encoded protease. The mutagenesis data generated may be useful in the design of protease inhibitors as well as in the interpretation of crystallography data related to the dengue virus protease. Finally, the identification of mutations that differentially affect the activity of the DEN2 NS3 protease may form the basis for the development of potentially attenuated live viruses, since a DEN2 infectious full-length cDNA clone was recently constructed in this laboratory (30). It is now possible to select mutations in NS3 that lead to decreased cleavage in the NS2B-NS3 autocleavage system and to introduce them into the DEN2 full-length clone. Mutant viruses can then be used to infect mammalian and mosquito cells. Analysis of the phenotype of the recovered virus will serve to assess the outcome of those mutations in the context of the virus life cycle.

ADDENDUM IN PROOF

While this manuscript was being revised, Xu et al. (J. Xu, E. Mendez, P. R. Caron, C. Lin. M. A. Murcko, M. S. Collett, and C. M. Rice, J. Virol. 71:5312–5322, 1997) reported on the cleavage site specificity of the BVDV protease and of Koch and Bartenschlager (J. O. Koch and R. Bartenschlager, Virology 237:78–88, 1997) reported on a mutagenesis study of the HCV protease.