INTRODUCTION

HIV¹ protease is an attractive target for antiretroviral therapy (1, 2, 3, 4). This enzyme is required for maturation of the viral capsid structural proteins and enzymes necessary for viral replication (5, 6). It is a homodimeric aspartyl protease. Inhibition or impairment of HIV protease by mutation of the active site Asp to Asn or Ala results in noninfectious virus particles (5). This finding spurred efforts to design potent inhibitors for use in antiretroviral therapy. More than 200 crystal structures of co-complexes of small molecule inhibitors with HIV-1 protease have been solved (7). Compounds of diverse chemical structure have been designed to have high affinity for this enzyme, i.e. subnanomolar K_i values (8, 9). HIV protease inhibitors have been shown to inhibit the replication of the virus in cell culture (10), and several protease inhibitors have advanced to clinical testing in AIDS patients and HIV-infected individuals (11, 12, 13). One of these compounds is VX-478, a potent inhibitor of HIV protease with high oral bioavailability in humans.² The structure of the co-complex of VX-478 with HIV-1 protease has been solved at high atomic resolution (14).

In vitro selection of resistant virus in the presence of HIV protease inhibitors is useful in understanding resistance at the molecular level (15, 16, 17, 18, 19, 20, 21, 22). For example, the HIV-1 strain GB8 in CEM cells was grown at increasing concentrations of saquinavir (23). At passage 11, the IC₅₀ increased about 40-fold over the wild type, and 9 out of 10 proviral clones examined contained both G48V and L90M mutations in the protease. In addition, earlier heterogeneity in the wild type sequences at other sequence positions had resolved into a homogeneous pattern after multiple passages. For indinavir, passage six of HIV-1_HXB2-infected MT4 cells yielded a double (M46L/V82A) and a triple (V32I/M46L/V82A) mutant, whereas by passage eight a quadruple mutant (V32I/M46I/A71V/V82A) became the dominant protease mutant (22). Cross-resistance studies of the molecular clone with the quadruple mutation showed 0.2-, 6-, and 1-fold increases in IC₅₀ for saquinavir, indinavir, and VX-478, respectively.

Partaledis et al. (24) reported that passage of HIV-1 virus in CEM-SS, a transformed T-lymphocyte line, in the presence of increasing concentrations of the sulfonamide inhibitors VB-11328 (K_i < 0.1 nM) or VX-478 (K_i = 0.6 nM) also led to the sequential accumulation of mutations in the protease gene. The identity and order in which the mutations appeared was identical for both inhibitors (Fig. 1). Each of these mutations observed is at or near the enzyme active site, except for L10F (Fig. 2). Ile⁵⁰ and Ile⁸⁴ are at the center of the active site and are located on the same plane as the carboxyl groups of the active site aspartates Asp²⁵ and Asp²⁵. Residues 46 and 47 are located on the flap region, which closes upon binding of the substrate or the inhibitor. Mutation of these residues, therefore, has the potential to affect flap dynamics and hence ligand binding. Since HIV is known for its high mutation frequency, it is important to determine which of the mutations observed in the in vitro passage experiment are due to selection pressure. It is also important to understand the implications for viral processing of gag-pol polypeptide during maturation of the virus. Hence, biochemical characterization of these mutations is necessary in understanding the contribution of each mutation to inhibitor binding and catalysis of polypeptide processing.

25

25

50

84

We have expressed and purified these mutant enzymes singly and in combination. In order to ascertain the catalytic competence of these protease mutants, we designed peptide substrates based on the natural cleavage sequences in the gag-pol polyprotein and evaluated the efficiency of cleavage of these substrates. We have also determined the inhibition constants for VX-478, and the structurally divergent inhibitors saquinavir and indinavir (Fig. 3). The ``vitality'' value (25) for each of these inhibitors, a composite ratio of the inhibition constant (K_i) and catalytic efficiency (k_cat/K_m), provides the basis for understanding why the M46I/I47V/I50V triple mutant is stably selected by the virus in the presence of VX-478.

MATERIALS AND METHODS

HIV-1 protease (catalog number 1256) and the peptide substrates His-Lys-Ala-Arg-Val-Leu-(NO₂)Phe-Glu-Ala-Nle-Ser-NH₂ (substrate 1; catalog number H-9035), His-Lys-Ala-Arg-Val-Leu-Ala-Glu-Ala-Met-Ser-NH₂ (substrate 2; catalog number H-9650), Ser-Gln-Asn-Tyr-Pro-Ile-Val-OH (substrate 3; catalog number H-7235), and Ac-Thr-Leu-Asn-Phe-Pro-Ile-Ser-Pro-OH (substrate 4; catalog number CSU-193) were obtained from Bachem BioSciences (Philadelphia, PA) and were used without further purification. The inhibitors VX-478, indinavir, and saquinavir were synthesized at Vertex Pharmaceuticals Inc. (Cambridge, MA). All other reagents were purchased from Sigma.

The procedure for the construction, expression, and purification of site-specific mutants of HIV-1 protease has been published elsewhere (24). In brief, mutagenesis was performed on pT7-HIV-1 using uracil enrichment of single-stranded DNA by the modification of Kunkel et al. (26) with the reagents provided in the Bio-Rad Mutagene kit. NdeI-EcoRI fragments coding for the HIV-1 protease open reading frame were subcloned into a pSPC27 vector (S. P. Chambers, Vertex Pharmaceuticals Inc.) for expression using an isopropyl-1-thio--D-galactopyranoside-inducible Tac promoter. Escherichia coli RB791 (from R. Brent, Harvard University) was used as the host.

Purification of the HIV protease mutants was carried out as described in Partaledis et al. (24). The purification method included batch treatment of bacterial lysates with DEAE-53 followed by S-Sepharose (cation exchange) chromatography, ammonium sulfate precipitation, and gel permeation chromatography on Sephadex G-75. The concentration of purified proteases was obtained by active site titration using tight binding inhibitors with K_i  5 nM.

Enzyme assays were performed in 50 mM citrate buffer, pH 4.5, containing 1 mM dithiothreitol and 1 mM EDTA. Vials containing this buffer and various concentrations of substrate were preincubated at 37 °C for 10 min before initiating reactions with wild type or mutant HIV-1 protease. After a sufficient reaction time to hydrolyze about 15% of the substrate (15-30 min), trifluoroacetic acid was added to a final concentration of 5% to quench the reaction. The products were separated and quantified on a C18 reverse phase HPLC (column: Rainin, Microsorb-MW, 5 µm, 300 Å, 4.6 × 50 mm, catalog number 86-203-F5; gradient: binary, with water and acetonitrile, each containing 0.1% trifluoroacetic acid). The product peaks were monitored by UV absorbance at either 210 or 280 nm. Kinetic data were fitted to the Michaelis-Menten equation to evaluate the k_cat and K_m parameters.

Using the K_i and k_cat/K_m values obtained for inhibitor binding and substrate processing, respectively, for the mutant (mut) and wild type (wt) enzymes, the vitality values (25) were calculated according to Equation 1.

(Eq. 1)

Analysis of the interactions of the four substrates used in this study with the HIV-1 protease active site was based on our earlier models of the natural substrates.³ Modeling of the natural substrate peptides spanning from P₅ to the P₄ region was carried out using the published crystal structures of peptidyl inhibitor complexes such as MVT-101 (PDB code 4hvp) and JG-365 (PDB code 7hvp). The peptides with Phe-Pro and Tyr-Pro cleavage junctions were modeled using the enzyme coordinates described in the 7hvp crystal structure (27). All other peptides were modeled using the 4hvp enzyme structure (28). The carbonyl of the scissile peptide bond of these substrates was modeled to occupy the transition state carbinol of the inhibitors. Initially the substrate models were built to mimic the bound conformations of the inhibitors JG-365 or MVT-101. These models were energy-minimized (200 steps of steepest descent followed by 2000 steps of Adopted-Basis Newton Raphson method) using the CHARMm22 force field in the QUANTA molecular modeling program (29). All enzyme atoms and the ``flap water'' were kept fixed for the first 1200 steps of minimization. The enzyme and flap water atoms were allowed to move during the last 1000 steps of minimization.

RESULTS

We investigated the catalytic parameters of HIV protease using four synthetic peptide substrates representing three natural cleavage sequences in the gag-pol polypeptide. Substrate 1 is a non-natural version of substrate 2 representing cleavage junction p24-p16 (p16 further gets cleaved into X(p2), p7, p1, and p6 proteins) in the gag polypeptide (30). The difference between substrate 1 and 2 is that Ala at P₁ (notation of Schechter and Berger; Ref. 36) of substrate 2 was substituted with [pNO₂]Phe in 1, and Met at P₄ of 2 was changed to Nle in 1 for stability. Substrate 1 was designed to be used in the spectrophotometric assay of HIV-1 protease and has been used by other investigators to study effects of HIV protease mutations on catalysis (33). The change in extinction coefficient of this substrate ( ~ 1000 M¹ cm¹ at 300 nm) is not large enough to provide a sensitive assay for evaluating tight binding inhibitors using subnanomolar concentrations of enzyme. We chose, therefore, to use an HPLC assay that has the required sensitivity. Peptide substrate 1 has good aqueous solubility (>6 mM) and is an excellent substrate for HIV protease. Hence, this was the substrate of choice for the evaluation of inhibition constants and for initial study of the catalytic parameters for the wild type and mutants of HIV-1 protease.

The catalytic parameters for processing the substrates by the wild type HIV-1 protease are presented in Table I. Among the four substrates, substrate 1 has the best catalytic parameters, with k_cat, K_m, and k_cat/K_m parameters of 16.3 s¹, 17 µM, and 9.4 × 10⁵ M¹ s¹, respectively. An eight-amino acid peptide, generated from substrate 1 by deleting two amino acids from the N terminus and one from the C terminus and modifying (pNO₂)Phe to Phe at P₁, was cleaved with a similar catalytic efficiency (k_cat/K_m) (<2-fold decrease compared with substrate 1).⁴ On the other hand, substrate 2 is cleaved about 54-fold less efficiently than substrate 1. Most of this effect is attributable to the loss of hydrophobic interactions by changing from (pNO₂)Phe to Ala at P₁. Substrates 3 and 4 have poor affinity for the enzyme, as exemplified by their K_m parameters. As reported by others (31, 32), HIV protease substrates with cleavage sites flanked by proline and/or aromatic amino acids (substrates 3 and 4) tend to have high K_m values (mM). The k_cat/K_m parameters for these substrates are also lower by a factor of 1340 and 1880, respectively, than for substrate 1.

Table I. Catalytic parameters for the wild type HIV-1 protease-catalyzed hydrolysis of peptide substrates Cleavage site Substratea Km kcat kcat/Km µM s1 mM1 s1 p24-X 1 17 16.3 940 p24-X 2 69 1.2 17.5 p17-p24 3 740 8.6 11.6 PR-RT 4 1465 0.7 0.5 a  Substrate 1 is His-Lys-Ala-Arg-Val-Leu-[pNO2]Phe-Glu-Ala-Nle-Ser-NH2, substrate 2 is His-Lys-Ala-Arg-Val-Leu-Ala-Glu-Ala-Met-Ser-NH2, substrate 3 is Ser-Gln-Asn-Tyr-Pro-Ile-Val-OH, and substrate 4 is Ac-Thr-Leu-Asn-Phe-Pro-Ile-Ser-Pro-OH. Residues flanking the cleavage site are in boldface type.

The same set of substrates was used to evaluate the kinetic parameters for the mutants L10F, M46I, I47V, I50V, the double mutant L10F/I50V, and the triple mutant M46I/I47V/I50V. The catalytic parameters k_cat, K_m, and k_cat/K_m were evaluated and are presented in Table II. For the L10F and M46I mutants, the kinetic parameters were evaluated for substrate 1 only. All other mutants were characterized against each of the four substrates. Without exception, the catalytic efficiency for these mutants is less than that of the wild type enzyme. For substrates 1 and 2, the reduction in catalytic efficiency is modest (5-fold). For 5-fold decreases in k_cat/K_m the effects are due largely to altered K_m values. For substrates 3 and 4, there is a 25-fold decrease in the k_c/K_m values against I50V, one of the largest effects we observed. Although there are experimental uncertainties associated with the evaluation of such high K_m values for these substrates, it appears that both K_m and k_cat values are significantly affected to produce the observed 25-fold decrease in k_cat/K_m. The double mutant L10F/I50V processes the substrates as efficiently as the I50V single mutant. The catalytic efficiency for the cleavage of substrates 2, 3, and 4 by the triple mutant is higher than that for the I50V mutant by a factor of 1.3-2.0.

Table II. Catalytic parameters for wild type and mutants of HIV-1 protease-catalyzed hydrolysis of substrates Enzyme Substrate 1 Substrate 2 Substrate 3 Substrate 4  kcat/Km (mm1 s1) Wild type 940  ± 240 17.5  ± 1.9 11.6  ± 1.7 0.5  ± 0.1 L10F 425  ± 41 NDa ND ND M46I 530  ± 55 ND ND ND I47V 320  ± 25 20.7  ± 3.1 7.0  ± 0.9 0.34  ± 0.06 I50V 300  ± 37 3.0  ± 0.6 0.46  ± 0.09 0.02  ± 0.004 L10F/I50V 230  ± 15 2.8  ± 0.8 0.35  ± 0.36 0.015  ± 0.002 M46I/I47V/I50V 185  ± 42 3.85  ± 0.4 0.7  ± 0.14 0.037  ± 0.001 Km (µM) Wild type 17  ± 4 62  ± 7 740  ± 105 1465  ± 260 L10F 15  ± 1.4 ND ND ND M46I 34  ± 3 ND ND ND I47V 27  ± 2 212  ± 28 800  ± 90 ND I50V 68  ± 8 366  ± 70 4600  ± 780 5800  ± 1100 L10F/I50V 31  ± 2 590  ± 140 2000  ± 1660 ND M46I/I47V/I50V 54  ± 12 1600  ± 112 2070  ± 370 ND kcat (s1) Wild type 16.3  ± 1.1 1.2  ± 0.02 8.6  ± 0.3 0.7  ± 0.05 L10F 6.4  ± 0.14 ND ND ND M46I 18  ± 1 ND ND ND I47V 8.6  ± 0.3 4.4  ± 0.3 5.5  ± 0.34 ND I50V 20.4  ± 0.6 1.0  ± 0.06 1.4  ± 0.14 0.11  ± 0.012 L10F/I50V 7.2  ± 0.14 1.7  ± 0.24 0.77  ± 0.45 ND M46I/I47V/I50V 9.9  ± 0.6 6.8  ± 0.4 1.4  ± 0.11 ND a  ND, not determined.

The inhibition constants, K_i, for VX-478, indinavir, and saquinavir were evaluated against all of the mutants (Table III). The single mutations L10F, M46I, and I47V have little effect on the binding of any of these inhibitors. On the other hand, the I50V mutation, either alone or in combination with the above mutations, weakens the binding of these inhibitors significantly. The largest increases in K_i were observed for VX-478 binding to the single mutant I50V and the triple mutant M46I/I47V/I50V. Decreased affinities of 83- and 247-fold, respectively, were observed.

Table III. Inhibition constant, Ki, for selected clinical candidate protease inhibitor drugs against mutants of HIV-1 protease This table is reprinted from Ref. 24. Enzyme Ki VX-478 Indinavir Saquinavir nM (ratio)a Wild type 0.6  ± 0.05 (1) 1.0  ± 0.13 (1) 0.8  ± 0.3 (1) L10F 0.5  ± 0.1 (1) 4.0  ± 0.4 (4) 1.4  ± 0.6 (2) M46I 0.3  ± 0.05 (1) 4.0  ± 0.7 (4) 0.8  ± 0.03 (1) I47V 0.4  ± 0.03 (1) 3.3  ± 0.3 (3) 0.4  ± 0.03 (1) I50V 50  ± 5 (83) 10  ± 1.1 (10) 17  ± 5 (21) L10F/I50V 36  ± 4 (60) 35  ± 4 (35) 27  ± 5 (34) M46I/I47V/I50V 160  ± 16 (267) 29  ± 5 (29) 33  ± 12 (41) a  Ratio of Ki for mutant protease over that of the wild type value rounded to the nearest whole number. Ratios 20 are in boldface type.

Vitality, as defined in Equation 1, is an index of survivability of the mutant virus and its ability to replicate in the presence of the inhibitor. The higher the vitality value, the greater the advantage offered by the mutation. Since vitality is a function of k_cat/K_m, it is dependent on the nature of the substrate. Vitality values were calculated for each inhibitor against each mutant using substrates 1-4. The I50V mutant showed significantly larger vitality values for VX-478. The vitality values obtained for I50V and the triple mutant against VX-478, saquinavir, and indinavir were plotted as a function of substrates as in Fig. 4. The mutant enzymes have much higher vitality values for VX-478 than for saquinavir or indinavir, against which these mutant enzymes remain sensitive.

DISCUSSION

HIV protease plays a vital role in the post-translational processing of gag and gag-pol polypeptides into functional structural proteins and enzymes. The effect of certain mutations on the catalytic efficiency of the protease has been evaluated by systematically modifying the residues on the substrate (32, 33). Although this type of structure-activity relationship is useful in understanding which subsite interactions are affected most by the mutations, it is less helpful if one wishes to determine to what extent the actual cleavage sites on the gag-pol sequence would be affected. Catalytic efficiency data for authentic cleavage sites, we believe, are useful in correlating biochemical data on in vitro selected mutations in the protease with the viability of the virus.

We chose four substrates, representing three cleavage sequences on the gag-pol polypeptide, to study the effects of mutation of the protease on processing polypeptides. It is clear from Table I that the efficiency of cleavage varies for each processing site of these mutants. Among the substrates we examined, substrate 4, which represents the PRRT cleavage junction, is the most slowly processed substrate. It is likely that in vivo processing of PRRT junction is one of the rate-limiting steps for the production of mature proteins from the gag-pol polypeptide. Any mutation that decreases the catalytic efficiency for processing the PRRT cleavage site may have significant consequences for the replication kinetics of the virus.

HIV-1 protease variants with Phe¹⁰ are naturally occurring (34), but the wild-type virus used in the passage experiments in the presence of VX-478 (24) has Leu at this position. In the protease, Leu¹⁰ is located on the surface of the enzyme about 10 Å away from the active site aspartate residues. Its side chain is in contact with the side chains of Arg⁸, Glu²¹, Leu²³, and Val⁸². When Phe¹⁰ was replaced with Leu and energy-minimized, it was found that the Phe side chain makes an aromatic-charge interaction with the Arg⁸ side chain in addition to the hydrophobic interactions with the side chains of Leu²³ and Val⁸². These additional interactions are likely to shift the monomer-dimer equilibrium toward the catalytically inactive monomer. A moderate reduction of 2.5-fold in turnover number (k_cat) is in qualitative agreement with the model. For the L10F mutation to have any effect on binding, the ligand must have P₃/P₃ side chains that interact with Arg⁸/Arg⁸ side-chains. Even then, only a small effect is expected. The modest increases in K_i for saquinavir and indinavir, 2- and 4-fold, respectively, and the lack of effect on the K_i of VX-478 are consistent with the predicted interactions between the inhibitor and mutant.

Met⁴⁶ is located on the flap of the enzyme with the amino acid side chain exposed to solvent. It is not involved in any direct contact with the inhibitors bound at the active site. The M46I mutation causes no reduction in binding of VX-478 or saquinavir binding, but a modest 4-fold decrease is observed in indinavir binding. A recent molecular dynamics study by Collins et al. (35) showed that the M46I mutation stabilizes the closed form of the enzyme flap. It is not clear why such stabilization of the flap would have selective effects on binding of different inhibitors or substrates. We are investigating the effect of the M46I mutation on the dynamics of the flap in the closed form and also the interactions of the neighboring residues (Ile⁴⁷ and Gly⁴⁸) with the inhibitors and the substrates by molecular dynamics simulations. Ile⁴⁷ is also located on the flap and is part of the constellation of residues forming the S₂/S₂ pocket. The phenyl ring of the indanol group in indinavir at P₂ buries deep into the S₂ pocket, closer to Ile⁴⁷ than the other inhibitors, and makes a number of contacts with the C- methyl group of this side-chain compared with the smaller tetrahydrofuran and Asn side chains of VX-478 and saquinavir, respectively. It is understandable, therefore, that I47V affects indinavir binding and not VX-478 or saquinavir binding. There is a 3-fold decrease in k_cat/K_m for substrate 1, which may also result from the loss of interaction between Val at P₂ of the substrate and the Val⁴⁷ side chain of the mutant.

Ile⁵⁰ and Ile⁸⁴ are at the heart of the active site and are capable of interacting with the peptide substrate side chains P₂ through P₂. All three inhibitors show >10-fold increases in K_i against I50V (see Table III) and I84V (data not shown) single mutants. Therefore, it is not surprising that during in vitro selection against VX-478, the virus first generated an I84V mutation in the protease but at the end selected a I50V mutation, which offered more resistance to VX-478 binding (24- and 83-fold increase in K_i for I84V and I50V mutants, respectively). Molecular modeling and solvent-accessible surface area calculations³ suggest that the increase in K_i is due to the loss of hydrophobic interactions but not due to loss of any hydrogen bond interactions of the flap water. The C- methyl group of Ile⁵⁰ makes hydrophobic interactions with all of the carbon atoms of the P₂ phenyl ring of VX-478 (Fig. 5). On the other hand, in saquinavir and indinavir, only two of the three methyl groups of t-butyl at P₂ are in direct contact with the C- methyl group of Ile⁵⁰. On the nonprime side, the P₂ indanol group of indinavir is making a direct hydrophobic interaction with the entire side chain of Ile⁵⁰, not just with the terminal C- group. Therefore, absence of the C- methyl group of Ile⁵⁰/Ile⁵⁰ residues by the I50V mutation causes a greater loss in binding of VX-478 than for indinavir or saquinavir.

pocket with the P

ring of VX-478.

The I50V mutation also affects catalysis significantly. The catalytic efficiencies (k_cat/K_m) decrease by 3-5-fold for substrates 1 and 2 and by 25-fold for substrates 3 and 4. The substrate models show that Ile⁵⁰ and Ile⁵⁰ side chains are in contact with P₂, P₁, P₁, and P₂ side chains of the substrates, as in the case of inhibitors noted earlier. The degree of contact of the Ile⁵⁰/Ile⁵⁰ residues with these different substrates varies, however. Using a 5.0-Å distance cut-off, we found that the C- methyl groups of Ile⁵⁰ and Ile⁵⁰ side chains make the largest number of contacts (four) with the Ile side chains at P₂ of substrates 3 and 4 but only one contact with the P₂ Glu side chain of substrates 1 and 2 (see Fig. 6). On the nonprime side, they make two contacts with the Val side chain of substrates 1 and 2 and one with the Asn side-chain of substrates 3 and 4. The larger number of interactions of the P₂ Ile is due to the fact that Ile side chain is buried deeper into the hydrophobic part of the S₂ pocket, whereas the Glu side chain is swung away from the hydrophobic part but toward the solvent-exposed portion of the enzyme, making hydrogen bonds with the main chain and side chain of Asp³⁰. Therefore, a larger effect observed on the catalytic efficiency of substrates 3 and 4 compared with substrates 1 and 2 is consistent with the predicted interactions of these substrates with the Ile⁵⁰/Ile⁵⁰ side chains.

During in vitro passaging of HIV-1, a number of mutations appear sequentially under the selection pressure of increasing concentrations of HIV protease inhibitors. Which mutation, if any, imparts a selective advantage to the virus? We have used the model by Gulnik et al. (25), which derives a predictive parameter for the viability of the virus from the K_i and k_cat/K_m values. Vitality is defined as the ratio of -fold increase in K_i for the mutant over -fold decrease in k_cat/K_m values for the same mutant with respect to the wild type enzyme values (Equation 1). A high vitality ratio signifies that this set of mutations will adversely affect the inhibitor binding more than their effect on the catalytic efficiency of the protease and hence will be incorporated as a resistance mutation. We evaluated the vitality factors for the I50V single mutant and the M46I/I47V/I50V triple mutant for all four of the substrates against VX-478, indinavir, and saquinavir. The results are plotted in Fig. 4. The following observations can be made. 1) For a given mutant and inhibitor, the vitality value will vary depending on which substrate is used for k_cat/K_m measurements. It is important to identify the natural substrate that exhibits the smallest vitality, because that cleavage site may be the bottleneck for the viral replication for that (set of) mutation(s). If that vitality value is large (say >10), then the mutation has a high potential to be incorporated. 2) Both the I50V and the triple mutants have overall high vitality values against VX-478. Therefore, it is advantageous for the virus to take the I50V path for survival when exposed to VX-478, but not in the presence of indinavir or saquinavir. 3) The vitality factors are consistently higher for all substrates for the triple mutant than for the I50V single mutant. There is no increase in K_i for VX-478 and saquinavir either for the M46I or I47V single mutants. Against the triple mutant, however, we observe a 3- and 2-fold increase in the K_i values for VX-478 and saquinavir, respectively, compared with I50V. Moreover, the addition of the M46I and I47V mutations results in an increased catalytic efficiency for substrates 3 and 4 by a factor of 1.5-2.0. Clearly, M46I and I47V are compensatory mutations. Addition of the M46I and I47V mutations, therefore, to the existing I50V mutation increases the survival chance of mutant virus.