Extended interaction networks with HCV protease NS3-4A substrates explain the lack of adaptive capability against protease inhibitors

inhibitor, ABSTRACT Inhibitors against the NS3-4A protease of hepatitis C virus have proven to be useful drugs in the treatment of HCV infection. Although variants have been identified with mutations that confer resistance to these inhibitors, the mutations do not restore replicative fitness and no secondary mutations that rescue fitness have been found. To gain insight into the molecular mechanisms underlying the lack of fitness compensation, we screened known resistance mutations in infectious HCV cell culture with different genomic backgrounds. We observed that the Q41R mutation of NS3-4A efficiently rescues the replicative fitness in cell culture for virus variants containing mutations at NS3-D168. To understand how the Q41R mutation rescues activity, we performed protease activity assays complemented by molecular dynamics simulations, which showed that protease-peptide interactions far outside the targeted peptide cleavage sites mediate substrate recognition by NS3-4A and support protease cleavage kinetics. These interactions shed new light on the mechanisms by which NS3-4A cleaves its substrates, viral polyproteins and a prime cellular antiviral adaptor protein, the mitochondrial antiviral signaling protein MAVS. Peptide binding is mediated by an extended hydrogen bond network in NS3-4A that was effectively optimized for protease-MAVS binding in D168-variants with rescued replicative fitness from NS3-Q41R. In the protease harboring NS3-Q41R, the N-terminal cleavage products of MAVS retained high affinity to the active site, rendering the protease susceptible for potential product inhibition. Our findings reveal delicately balanced protease-peptide interactions in viral replication and immune escape that likely restrict the protease adaptive capability and narrow the virus evolutionary space.


ABSTRACT
Inhibitors against the NS3-4A protease of hepatitis C virus have proven to be useful drugs in the treatment of HCV infection. Although variants have been identified with mutations that confer resistance to these inhibitors, the mutations do not restore replicative fitness and no secondary mutations that rescue fitness have been found. To gain insight into the molecular mechanisms underlying the lack of fitness compensation, we screened known resistance mutations in infectious HCV cell culture with different genomic backgrounds. We observed that the Q41R mutation of NS3-4A efficiently rescues the replicative fitness in cell culture for virus variants containing mutations at NS3-D168. To understand how the Q41R mutation rescues activity, we performed protease activity assays complemented by molecular dynamics simulations, which showed that proteasepeptide interactions far outside the targeted peptide cleavage sites mediate substrate recognition by NS3-4A and support protease cleavage kinetics. These interactions shed new light on the mechanisms by which NS3-4A cleaves its substrates, viral polyproteins and a prime cellular antiviral adaptor protein, the mitochondrial antiviral signaling protein MAVS. Peptide binding is mediated by an extended hydrogen bond network in NS3-4A that was effectively optimized for protease-MAVS binding in D168-variants with rescued replicative fitness from NS3-Q41R. In the protease harboring NS3-Q41R, the N-terminal cleavage products of MAVS retained high affinity to the active site, rendering the protease susceptible for potential product inhibition. Our findings reveal delicately balanced protease-peptide interactions in viral replication and immune escape that likely restrict the protease adaptive capability and narrow the virus evolutionary space.

INTRODUCTION
Viruses under treatment with specific antivirals undergo rapid adaptive selection to maximize their fit for escape from drug pressure. Thereby, increased replicative fitness can lead to decreased drug sensitivity of viral variants (1) and determine the probability with which resistant variants will persist (2). Based on a deterministic model (3)(4)(5)(6) hepatitis C virus (HCV) variants under selective pressure with direct-acting antiviral agents (DAA) are selected by their degree of resistance and capability to replicate and assemble infectious virus particles (7). Protease inhibitors (PI) against the serine protease NS3-4A of HCV are burdened with a particularly high risk for drug resistance development that inflicts considerable fitness costs (7,8). Although compensatory second-site mutations that rescue fitness deficits from drug resistance mutations are pivotal drivers of viral quasispecies evolution (3,9,10), they are virtually not observed upon PI treatment failure in the NS3-4A protease of HCV. While the impact of mutations on PI resistance has been well characterized, their role in replicative fitness of viral variants is less understood. The NS3-4A protease plays an essential role in the HCV replication cycle (11)(12)(13). In particular, NS3-4A is involved in RNA synthesis by proteolytically processing nonstructural proteins from the viral polyprotein. In addition, the capacity to cleave and inactivate mitochondrial antiviralsignaling protein (MAVS) (13) (also called Cardif, VISA, and IPS-1 (15)(16)(17)) in the RIG-Isignaling pathway is a cardinal feature of NS3-4A, by which HCV blocks induction of interferon-(IFN)-β. MAVS cleavage is a key mechanism in the selection of drug resistance mutations against macrocyclic PIs and is important for promoting viral persistence. Novel PI-based treatment regimens still appear to fail in some patients, specifically those with advanced liver disease or liver cirrhosis (18). Importantly, most of these difficult to treat patients show dominant resistant variants with D168A or D168V amino acid substitutions in NS3-4A at the time of virologic failure (18). We previously showed that the D168 mutant proteases are characterized by increased capacity to cleave MAVS and suppress IFN-β induction (19). The reconstitution of a tight network of electrostatic interactions between protease and the peptide substrate in those variants allows much stronger binding of MAVS than in other protease variants or wild type HCV (19).
In the present study, we unraveled a molecular mechanism in PI-resistant D168 variants that affect the cutting efficiency of peptide substrates by a second-site mutation in the viral protease, NS3-Q41R. By interference with critically balanced protease-peptide interactions, the second-site mutation rescues replicative fitness deficits of D168 variants in infectious cell culture. The finding that NS3-Q41R, that itself is reported to confer resistance to macrocyclic PIs (20), has a fitness-compensatory effect in D168 variants is surprising, given the very low prevalence of variants harboring Q41R in PI-experienced patients (<0.1%) (21). Our findings highlight a previously unknown molecular mechanism for MAVS that potentially restricts virus evolution under PI pressure, providing a new paradigm for variant selection from HCV-infected cells.

Rational
approach to characterize replicative fitness of PI-resistant variants. We aimed to identify PI resistance-associated amino acid substitutions in NS3-4A that lead to high replicative fitness and potentially confer a general drug resistance phenotype to the virus (1). Resistance mutations were identified by a review of published data from both cell culture-based and clinical studies of PIs (7).
Replicative fitness of D168 variants is enhanced by the second-site mutation NS3-Q41R. To characterize the impact of PIresistance mutations on RNA replication, we constructed two comprehensive panels, each with 23 resistance mutations in derivatives of pH77S (22) (Fig. 1, A and B). The two panels only differed by the cell culture adaptive mutation Q41R in their NS3-4A protease backbone (7) (Fig. 1B). We measured GLuc activity from NS3-Q41/GLuc2A RNA and NS3-Q41R/GLuc2A RNA in media collected at 24-h intervals after transfection. NS3-Q41R showed no loss of replication capacity compared to NS3-Q41, although the replication capacity of mutants in both backbones, in general, was more compromised in NS3-Q41R (Fig. 1C). Interestingly, NS3-Q41R served to promote the replication of the NS3-D168 variants A/E/H/V despite acting synergistically to impair the replication of other key PI-resistant variants, such as R155K. Compared to wild type (WT), the GLuc activity of D168A was compensated to approximately 91% by NS3-Q41R (161% increase compared to NS3-Q41; fold change (FC) 2.6) and that in D168E was boosted to approximately 141% by NS3-Q41R (44% increase compared to NS3-Q41; FC 1.4). For the clinically relevant D168A/E mutants with increased replicative fitness by NS3-Q41R, we further measured the ability to produce infectious viruses in NS3-Q41 and NS3-Q41R protease backbones using a sensitive GLuc assay (Fig. 2). Both mutants produced more infectious viruses in the NS3-Q41R backbone than in NS3-Q41, which corresponded with the respective results on their RNA replication capacity (Fig. 2). D168 mutants are signature mutations selected from pressure with macrocyclic PIs, such as simeprevir, asunaprevir, paritaprevir and glecaprevir (18,23). To assess the ability of a macrocyclic PI to inhibit the RNA replication of PI-resistant variants in NS3-Q41 and NS3-Q41R protease backbones, antiviral EC 50 values were determined from the concentration of simeprevir required to cause a 50% reduction in the secretion of GLuc by RNA-transfected cells. The NS3-Q41R backbone caused an increase in EC 50 for most PI-resistance mutations tested with on average higher-level PI resistance than the respective mutants in an NS3-Q41 genomic background. All D168 mutants tested in the NS3-Q41R backbone showed high-level PI resistance against simeprevir (FC 32-156 compared to the WT protease NS3-Q41); see Table 1 and Fig. S2 for details on the replicative fitness and drug resistance.
For further in vitro characterization, we selected the clinically most relevant PIresistance associated mutations D168A and D168E that showed rescued replicative fitness due to NS3-Q41R, but omitted D168H (clinically less important) and D168V (only weak fitness compensation). D168T was selected from the D168 mutants as a negative control without impact on replicative fitness in protease backbones ± Q41R. NS3-Q41R had no effect on the protease fold. To characterize the fitness-compensatory mechanism of NS3-Q41R in D168 variants, we expressed and purified the NS3-4A wild-type (WT) fusion and different mutant forms (see Experimental Procedures and Fig. S1). We first tested if D168A/E/T ± NS3-Q41R had an effect on the protease stability ( Fig. 3 and Table 2). The impact of mutants on protein folding was determined by a thermal shift assay, where protein denaturation at increasing temperatures was revealed by increased fluorescence (see Experimental Procedures).
The NS3-Q41 protease WT was stable to temperature with a half maximal denaturation, T m , at 54.9°C (Fig. 3). NS3-Q41R had no effect on the protein stability. A significant difference in protein stability was observed between D168E and D168A/T mutant proteases irrespective of the NS3-4A protease background ± Q41R (p < 0.0001) (Fig. 3). As we showed previously (19), the impact on protein fold in D168 mutants is likely related to weakening or stabilization of a local salt bridge and H-bond network in the outer ligandbinding site of the protease. In contrast to D168A, the D168E mutant can compensate the missing interactions of the D168 residue with R155 and R123 and stabilize the protein fold (19). Our present study confirms the modeling results published earlier (19). The melting temperatures reflect a stable protease fold in D168E versus destabilization in D168A (Fig.  3). Importantly, NS3-Q41R showed no difference to the NS3-Q41 WT background neither as a single mutant nor as a second-site mutation with D168A/E/T ( Fig. 3 and Table  2). Thus, we considered it unlikely that the increased replicative fitness in the NS3-Q41R background in D168A/E could be due to a modulation of protein stability in the NS3-4A protease.
NS3-Q41R can modulate the kinetic properties of the protease for viral polyprotein substrates and MAVS. Next, we analyzed the protease enzymatic activity using an in vitro fluorescence resonance energy transfer (FRET)-based assay that measures the cleavage of a viral polyprotein substrate. We tested the kinetic properties of NS3-Q41 WT and D168A/E/T ± NS3-Q41R mutants using a peptide substrate derived from the viral NS4A/4B polyprotein cleavage site ( Fig. 4A and Table 2) comprising an excited donor molecule and a quencher (see Experimental Procedures). In this assay, fluorescence is released when the protease recognizes and cleaves the FRET substrate. We determined the amount of substrate needed to obtain half of the protease maximum rate of reaction (K M ) and substrate turnover rates (k cat ). The protease with the NS3-Q41R backbone exhibited a lower K M than the NS3-Q41 WT (p = 0.08; FC 0.5). No difference was found in the substrate turnover k cat between the two genomic backgrounds. The D168A mutant in NS3-Q41 showed a significantly lower k cat (p = 0.0008; FC 1.1) and a slight decrease in K M (p = 0.24; FC 0.3) compared to D168-WT. A further decrease in K M (p = 0.13; FC 0.8) paralleled by a significant increase in k cat (p = 0.003; FC 0.7) for D168A was found in the NS3-Q41R backbone. While differences in K M were not significant for D168E ± NS3-Q41R, k cat was significantly higher for D168E in the NS3-Q41R versus NS3-Q41 backbone (p = 0.01; FC 0.2) (Fig. 4A). No such effect was observed for D168T ± NS3-Q41R (Fig. 4A), that was tested as a D168 variant not compensated for replicative fitness deficits by NS3-Q41R (Fig.  1C). Although K M and k cat varied in rather narrow ranges between 1.7 and 3.1 µM and 0.09 and 0.19, respectively, the kinetic constants reflect the fitness-compensatory effect of NS3-Q41R for D168A/E variants that we observed in infectious cell culture (Fig.  1C).
Notably, NS3-Q41R is not observed as a fitness-compensatory mutation in PI-failure patients with D168 variants (24). As we showed previously, efficient MAVS cleavage is a characteristic feature of D168 variants and a key escape mechanism under selective pressure with macrocyclic PIs (19). Hence, we reasoned that NS3-Q41R could have varied the kinetic properties of D168A variants towards MAVS. To characterize the effect of D168A and NS3-Q41R on the protease MAVS cleavage kinetics, we designed a novel FRETpeptide incorporating the MAVS cleavage site. As done for NS4A/4B, K M and k cat of D168A ± NS3-Q41R were determined for the MAVS peptide ( Fig. 4B). In contrast to NS4A/4B, we observed vast variations in the kinetic constants for MAVS. Compared to the NS3-Q41 protease WT, NS3-Q41R showed an approximately 5-fold increase in k cat (p = 0.008; FC 4.7), although the K M was significantly higher than in the WT (p = 0.02; FC 4.8). A highly significant and fundamental change with inverse kinetic constants was observed when NS3-Q41R was present as a second-site mutation in the D168A-protease (Fig. 4B). Here, D168A showed a more than 16-fold increase in k cat compared to D168A without NS3-Q41R (p = 0.0003; FC 16.5). In parallel, the Michaelis-Menten constant, K M , was significantly decreased (p < 0.05; FC 7.1) (Fig. 4B). Thus, there was large reversal in K M and k cat for D168A upon the addition of NS3-Q41R. Next, we calculated the ratio k cat /K M , the 'specificity constant', to compare the relative rates of NS3-4A protease cleavage on the competing substrates NS4A/4B and MAVS (Table 2). We found the specificity constant of the D168A-protease for MAVS more than 100-fold increased when harboring NS3-Q41R and 3.3-fold higher than for NS4A/4B. NS3-Q41R increases the ability of the protease and D168 variants to cleave MAVS. The above results using peptide substrates in vitro suggested a fitness advantage due to NS3-Q41R in D168 variants via increased MAVS cleavage capacity. To test this hypothesis in cells, we prepared expressed constructs for D168-WT and D168A/E/T mutants in protease genomic backgrounds ± NS3-Q41R. The expressed constructs were transfected into U-2 OS cells that express high levels of endogenous MAVS (19). Western blot was employed after transfection with NS3-4A expression constructs to detect fulllength and cleaved MAVS (Fig. 5A). Our data revealed that NS3-Q41 constructs harboring D168E/T had a comparable capacity to cleave endogenous MAVS compared to the NS3-Q41 protease WT. A notable exception is D168A in the NS3-Q41 protease that displayed a significantly higher potency to cleave MAVS than the NS3-Q41 protease WT, which is in line with our previous findings (19). Interestingly, the addition of NS3-Q41R increased the protease capacity to cleave MAVS for all D168 variants tested (Fig. 5A). Most notably, full-length MAVS was not visible already after 12 h of transfection in D168A + NS3-Q41R. We observed complete cleavage of MAVS after 16 h of transfection for D168A/E/T in the NS3-Q41R background. Remarkably, we observed no degradation of the cleavage products for D168A/E/T in NS3-Q41R after 12 h of transfection (Fig. 5A).
Next, we tested the capacity of MAVS in U-2 OS cells to induce IFN-β for D168-WT and D168A/E/T mutants in protease genomic backgrounds ± NS3-Q41R after transfection with an IFN-β-reporter plasmid (Fig. 5B). As expected, over-expression of MAVS resulted in strong activation of the IFN-β reporter. As previously shown (19), we find IFN-β activity significantly inhibited by D168A in the NS3-Q41 protease backbone. Moreover, we find that proteases harboring NS3-Q41R or NS3-Q41R + D168E significantly attenuated the IFN-β activity. The competence of the Q41R-D168A protease to interfere with MAVSinduced IFN-β-reporter activation did not differ significantly from the Q41 background. A potential explanation is that peptides from MAVS cleavage could play an inhibitory role at the protease active site. Such a model could explain that the Q41R-D168A protease with high k cat but low K M for MAVS was not superior in inhibiting IFN-β activity compared to the D168A mutant protease (Fig. 4B and  5B).
An extended hydrogen bond network around the protease active site plays a critical role in substrate recognition. We previously showed that the mutant D168A reconstitutes electrostatic interactions between the protease and MAVS facilitating stronger binding of the MAVS peptide to the D168A mutant than to the D168-WT protease (19). To characterize the molecular mechanisms of fitness compensation by NS3-Q41R as a second-site mutation in the D168A protease, we performed molecular dynamics (MD) simulations of the NS3-Q41 protease WT, as well as the Q41R-D168A mutant protease, bound to (i) the viral polyprotein junction NS4A/4B or (ii) MAVS (Fig. 6, left panel). For each simulation, we analyzed the H-bond network surrounding the protease active center (Fig. 6, right panel) and calculated peptideprotease interaction energies (Table 3).
A comparison of the simulated systems showed that the D168-WT protease with NS3-Q41 bound to NS4A/4B formed a wellbalanced and stable extended H-bond network (Fig. 6, A and B). The core of this network consisted of H-bonds within the protease catalytic triad D81-H57-S139. This core network was extended by H-bond contacts to Q41 and R155-D168-R123. The two mutants D168A and NS3-Q41R substantially disturbed this extended network, weakening its coupling to the protease active site (Fig. 6, C and D). Instead, R41 and R123 participated in more Hbonds (+ ~ 1.3) and π-π contacts (+42% of simulation time) to the NS4A/4B peptide substrate than in the WT protease. For the Q41R-D168A mutant protease, the interaction energy between protease and the NS4A/4B non-cleaved substrate increased by approximately 29% compared to the WT protease (Table 3).
A similar analysis of the H-bond network for the WT protease bound to MAVS identified a substantially weaker stabilization of the catalytic site (Fig. 6, E and F) than in the NS4A/4B-protease complex (Fig. 6, A and B). The destabilization was a direct consequence of polar MAVS residues competing with intraprotease H-bonds within the extended network. This changed considerably due to the mutants D168A and NS3-Q41R (Fig. 6, G and  H). Q41R participated in strong H-bonds (+0.98) with the protease active site while establishing π-π interactions (+ 81%) to MAVS at the same time, rigidifying the complex around the active site (Fig. 6H). In addition, D168A gained strong H-bonds between MAVS and R155 (+ 1.31) / R123 (+ 0.21). These changes in the bonding network leveraged the affinity of the N-terminal MAVS and NS4A/4B cleavage products to the Q41R-D168A protease (77% versus 9% gain in protease-peptide interaction energy respectively) ( Table 3). Summing up effects from protease mutations and substrate binding, NS3-Q41R and D168A showed substantial impact on the H-bond network at the protease active site. Both mutations split the H-bond network and freed protease residues for additional interactions with the substrates, explaining the observed decreases in K M values. Binding of the natural substrate NS4A/4B yielded a stable extended H-bond network in the WT protease, which was weakened by NS3-Q41R and D168A. In contrast, MAVS binding to the protease naturally disturbed this network in the WT protease. Here, the PI-resistance mutation D168A and its second-site mutation NS3-Q41R showed unique compensatory interactions in the protease bound to MAVS and stabilization of the protease active site surroundings. This aligns well with the k cat trends from our enzyme activity assays and Michaelis-Menten kinetics (Fig. 4).

DISCUSSION
Fitness compensation by a second-site mutation in the HCV protease NS3-4A, as we show here, has not been noted previously (21,24). We find NS3-Q41R to permit a fitness recovery for viral variants with resistanceassociated amino acid substitutions at D168 in infectious cell culture at high-level resistance against a macrocyclic PI (Fig. 1C and Table 1). Our findings emphasize the capability of NS3-Q41R to yield 'superfit' PI-resistant D168 variants. However, NS3-Q41R is not observed in patients though the selection of D168 variants in advanced-stage liver disease is responsible for the vast majority of PItreatment failures (18). We considered the underlying mechanisms important to understand structural constraints and adaptive capability in HCV NS3-4A.
To characterize the molecular mechanism of fitness compensation by NS3-Q41R, we used enzyme kinetics from purified protein and complementary molecular dynamics simulations. As previously shown, a characteristic feature of D168-protease mutants is the molecular adaptation of the protease active site for cleavage and inactivation of MAVS (19), a key cellular antiviral adaptor protein whose expression level directly determines antiviral immunity (14)(15)(16)(17). Strikingly, we find the D168A mutant protease boosted to 16-fold higher MAVS cleavage efficiency by NS3-Q41R, while the efficiency for the viral polyprotein substrate was only 2-fold increased (Fig. 4). This is important, given that blocking RIG-I signaling by targeting MAVS is a key strategy in several highly diverse classes of viruses to evade type I-interferon immune responses (25)(26)(27). Conversely, MAVS in primates evolved under strong positive selection, driven by antagonism from ancient 'paleovirus' infections (28). As a consequence of protease-peptide coevolution, viral proteases can be highly adapted for MAVS cleavage already in the wild-type protein (28). In the HCV protease NS3-4A, we find that NS3-Q41R had a strong impact on the catalytic activity and turnover in D168 mutant proteases, thereby differentially tuning the protease for viral polyprotein substrates and MAVS. Noteworthy, the changes in the protease kinetic constants for the viral polyprotein substrate NS4A/4B reflect the fitness-compensatory effect of NS3-Q41R that we observed in infectious cell culture. Of note, the Huh-7.5 cells that we were using here are defective for signaling via RIG-I and MAVS (29). Hence, the MAVS cleavage capacity of the protease mutants should not have made an impact on the replicative fitness in our cell culture model of HCV infection.
Mechanistically, we find a delicately balanced extended hydrogen bond network around the protease active site that was altered by D168 mutants as well as the second-site mutation NS3-Q41R. Governed by changes on the opposite ends of this network (Fig. 6), the protease active site environment is effectively optimized for MAVS binding by D168A and NS3-Q41R ( Fig. 6H and Table 3). In line with these findings, we show that NS3-Q41R in cell culture shifted the balance in proteasesubstrate recognition in the D168 variants towards MAVS (Fig. 5). The presence of hydrophilic and charged residues in the peptide substrate between the P4 and P3' sites is likely a key factor determining how D168 and Q41 mutations influence binding and catalysis. Therefore, based on our observations for NS4A/4B and MAVS, the binding of other cellular substrates of the NS3-4A protease might also be affected by mutation of these sites: Primarily hydrophobic substrates like the membrane-associated glutathione peroxidase GPx8, the E3 ubiquitin ligase component UVdamaged DNA-binding protein 1 (DDB1) and the cleavage site b within T-cell protein tyrosine phosphatase (TC-PTP), a phosphatase involved in growth factor signaling, could behave similarly to NS4A/4B. While hydrophilic residues of the TIR domaincontaining adaptor inducing interferon-β (TRIF) or the TC-PTP cleavage site a might interact with the extended network along the lines of MAVS (Fig. S3).
Notably, we find that the MAVS cleaved products did not get degraded fast and retained high affinity to D168 protease variants when harboring NS3-Q41R ( Fig. 5 and Table 3). Hence, we speculated that the cleaved products might compete with protease substrates at the active site, a mechanism that potentially could downregulate the protease catalytic efficiency. Adding to this notion, we calculated proteasepeptide interaction energy differences from the molecular dynamics simulations that in the Q41R-D168A mutant context showed a remarkable gain of 77% for the N-terminal MAVS cleavage product compared to only 9% for the viral polyprotein peptide (Table 3). In support of a direct impact of MAVS on the protease catalytic activity, preincubation of the purified protein with MAVS peptides showed more potent blockade of enzyme activity in a protease harboring NS3-Q41R than in the wild-type protease (Fig. 7).
It is difficult to compare directly the in vitro biochemistry with the cellular degradation of MAVS. There are several strong limitations for this comparison such as ambiguous concentrations of viral protease, polyprotein substrates and MAVS expression levels in cells. However, from the biochemistry data we clearly conclude that mutations in the NS3-4A protease can have vast differences in the cutting efficiency of peptide substrates. Our data suggest that protease variants already optimized for MAVS cleavage can be easily overfitted for this peptide substrate by secondsite mutations in NS3-4A. Such a mechanism could interfere with RNA synthesis and immune escape and hence prevent selection of fitness-compensatory mutations such as NS3-Q41R in patients. The mechanism could potentially explain previous observations from a chimpanzee infected with a cell-culture derived virus containing NS3-Q41R that initially was poorly fit but evolved to establish long-term persistent infection (30). Yi and coworkers observed an unusual 4-to 6-week delay before detectable viremia in this animal during which the level of intrahepatic innate immune responses increased substantially. They found that the cell-culture adapted H77S.2 strain had reverted R41 back to Q41 wild-type before viremia (30), whereas other chimpanzees infected by NS3-Q41 wild-type strains were viremic within a week (31,32).
Collectively, our data support a model in which balancing between MAVS cleavage capability and efficiency of polyprotein processing likely restricts the selection of drug-resistant HCV protease variants in vivo. Our data suggest that the negative selection of fitness-compensatory mutations in NS3-4A to some extent is a consequence of protease-MAVS overfitting, which is important because it can prevent fitness compensation and hence ultimately contribute to extinction of PIresistant HCV variants upon treatment failure (2)(3)(4)(5).

EXPERIMENTAL PROCEDURES
Cell culture studies. Central to our studies has been the use of an infectious molecular clone of genotype 1a H77 virus (22). The clone pH77S.2 was derived from pH77S and differs from pH77S.3 by the cell-culture adaptive mutation NS3-Q41R in the NS3-4A protease domain (Fig. 1B). When transfected into permissive Huh-7.5 cells (22), genome-length RNA transcribed from pH77S.2 and pH77S.3 replicates efficiently and produces a virus that is infectious in naïve cells (22). To monitor replication, the Gaussia luciferase (GLuc) sequence, fused at its C terminus to the footand-mouth disease virus (FMDV) 2A autoprotease, was inserted between p7 and NS2 (7).

Site-directed mutagenesis and transfection.
Primers for NS3-4A amplification were used with restriction sites for NheI and EcoRI and designed as follows: Viral fitness and antiviral resistance. Genomelength RNA was transcribed from the mutated pH77S.2, pH77S.3, pH77S.2/Gluc2A and pH77S.3/GLuc2A plasmids in vitro, and studies to assess antiviral resistance and viral fitness carried out as described previously (7).

MAVS cleavage and IFN-β promoter activation.
To assess MAVS cleavage in cells, we used U-2 OS human osteosarcoma cells (obtained from the American Type Culture Collection, Rockville, MD) and Western blotting. U-2 OS cells express high levels of endogenous MAVS and are highly permissive for transfection with NS3-4A expression constructs as previously described (19). IFN-β promoter activation assays were based on luciferase reporter assays as described previously (33).
Thermal shift assay. Stability of the protein fold of expressed WT and mutant NS3-4A was measured by fluorescence-based thermal shift (35) using Sypro Orange, a temperature-stable fluorophore that exhibit enhanced fluorescence upon interacting with unfolded proteins. The impact of mutations on protein unfolding patterns is characterized by fluorescence emission curves. Melting temperatures (T m ) from purified NS3-4A protease mutants are determined by fitting the sigmoidal melt curve to the Boltzmann equation.
Protease activity assay. The enzymatic activity and Michaelis-Menten kinetics of WT and mutant NS3-4A proteases were measured by a fluorescence-based protease assay (36,37). To measure peptide cleavage kinetics, a viral polyprotein substrate derived from the NS4A/4B cleavage site was purchased from Anaspec (SensoLyte 520 HCV, Anaspec). To determine enzyme kinetics for MAVS cleavage, we designed a novel FRET peptide incorporating the MAVS cleavage site, produced by Anaspec (Ac -EREVPCHRPS -NH 2 ).
MAVS product inhibition assay. We used an untagged MAVS peptide incorporating the MAVS cleavage site (Ac -EREVPCHRPS -NH 2 ) to determine the influence of MAVS cleavage products on the cleavage kinetics of viral polyprotein substrates. Small amounts of purified NS3-4A protease (NS3-Q41 WT or NS3-Q41R, 50 nM) were preincubated with MAVS peptide (50 µM) for 0, 10 or 20 min at room temperature to allow different time spans for cleavage product formation. After preincubation, 2 µM of the NS4A/4B FRET peptide (SensoLyte 520 HCV, Anaspec) was added and reaction velocity measured as described above. Noteworthy, MAVS and NS3-4A are shown to colocalize in infected cells (38). Hence, the "effective concentration" of MAVS can theoretically reach levels higher than 50 micromolar.

Statistical analyses.
Comparisons of measurements were analyzed by two-sided Ttests with a level of significance set at a minimum of p = 0.05.
Computational studies. Molecular systems were prepared from a crystal structure of the NS3-4A protease WT bound to the N-terminal cleavage product of MAVS (PDB-3RC5) (39). We modeled both full-length substrates, MAVS and NS4A/4B, prior to the peptide bond cleavage. The P'1-4 residues of MAVS, missing in the experimental structure, were rebuilt based on the crystal structure of bovine Chymotrypsin bound to Eglin C (PDB-1ACB) (40), as previously performed (41). Topologies and starting coordinates for molecular dynamics simulations of protease WT and mutants bound to MAVS and NS4A/4B substrates were prepared using the tleap program of the AmberTools15 (42), minimized in gas phase while restraining the protein backbone. The structures were then solvated and neutralized in a box of SPC/E water molecules (43) using a minimum distance of 14.0 Å between the boundary and the protein atoms. All calculations were performed with the Amber14/AmberTools15 suite of programs (41) using the ff03.r1 Amber force field (44,45). After minimization and adjustment of the box size to reach a target density of 1,000 kg/m 3 , the systems were gradually heated up to 300K in the NVT ensemble (45). Finally, production runs were performed for 250 ns in the NPT ensemble at 300K and 1 bar. Both heat-up and production runs used a time step of 1 fs, and SHAKE bond length constraints were applied to all bonds involving hydrogen atoms. Temperature and pressure were controlled with Langevin dynamics (collision frequency of 4.0 ps -1 ) and isotropic position scaling, respectively.
Analysis of the trajectories was performed on frames extracted from the last 100 ns of the NPT production simulations. Analysis and visualization of residue-interaction networks was carried out using our in-house interaction network plugin CONAN (Manuscript in preparation) for Cytoscape 3 (46). Hydrogen bonds present in each time frame were extracted from MD trajectories using cpptraj (46) with a maximum donor-acceptor distance of 3.5 Å and a minimum donor-hydrogenacceptor angle of 135°. Edge widths in the networks shown represent the average number of H-bonds during simulation time. Only edges above a minimum simulation presence of 10% are shown. Cation-π stacking interactions between arginine (Arg) and histidine (His) residues were analyzed by considering sidechain distances and relative orientations. The sidechain plane was defined for the Arg and His residues based on the heavy atoms of the side chain guanidinium group and imidazole ring, respectively. In this scheme, an interaction was counted as present if the distance between the center of mass of the two planes did not exceed 5.0 Å, and if the angle between the plane normal vectors was found in the ranges of 0-30° or 150-180°. Proteinsubstrate interaction energies were obtained from the "Delta G gas" term as calculated by MMPBSA.py (AmberTools17) (47)(48)(49). The error of uncorrelated samples σ e was estimated by an autocorrelation time analysis (48), using a weighted single exponential fit (weight factor 1 / (x+1)), where x is the integer distance between frames. To reduce noise in the fit, only autocorrelation values larger than 0.1 were used.

DATA AVAILABILITY
All data that support the findings of this study are available from the corresponding author upon request.

CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest with the contents of this article.