Allosteric inhibition of the NS2B-NS3 protease from dengue virus

INTRODUCTION

Dengue virus (DENV) is the causative agent in dengue fever, dengue hemorrhagic disease, and dengue shock syndrome. All four serotypes of dengue virus are transmitted to humans by the Aedes aegypti and Aedes albopictus mosquitoes (1–6). Globally nearly 2.5 billion people are at risk of dengue virus infection and over 100 million infections are reported annually (6). Dengue virus infection also causes 22,000 deaths each year in areas where it is endemic (7). Despite intensive biomedical studies, no vaccine nor drug has been approved to date (1, 8–10).

Dengue virus contains a positive-strand RNA genome that can be directly translated into a single polyprotein chain by host-cell translation machinery. The polyprotein precursor comprises three structural proteins: capsid (C), membrane (M), envelope (E), and five non-structural proteins NS1–5. In the polyprotein precursor the proteins are arranged as NH₂-C-prME-NS1-NS2A-NS2B-NS3-NS4A-NS4B-NS5-COOH(11–13). Maturation of the precursor is essential for the viral life cycle and requires post-translational proteolytic processing by the dengue virus protease (NS2B-NS3pro) to liberate individual viral proteins from the polyprotein chain. NS2B-NS3pro, a trypsin-like serine protease, performs the post-translational proteolytic processing of precursor (14, 15). NS2B-NS3pro cleaves the viral protein at the NS2A/NS2B, NS2B/NS3, NS3/NS4A and NS4B/NS5 junctions as well as in the capsid protein (16, 17). Full processing releases the viral protease (NS2B-NS3pro), helicase (NS3) (18–20), methyltransferase (NS5), and RNA-dependent RNA polymerase (NS5) (21). The other non-structural proteins such as NS1 and NS2 are involved in the viral replication complex while NS4 is an integral membrane protein that plays a role in induction of membrane alteration (22–24).

It is clear that the flavivirus proteases including NS2B-NS3pro are essential for viral replication and infectivity, as shown by a number of previous studies. For example, Yellow fever virus genetically modified to contain only inactive NS2B-NS3pro is unable to infect target cells (25). Similarly, treatment of cells with a peptide that inhibits NS2B-NS3pro decreases dengue virus infection by 80% (26). In addition, inhibition of viral proteases is a well-established route for preventing viral infection in clinical settings. A large number of HIV protease inhibitors are clinically used to treat HIV infection and AIDS (for review see(27). Two hepatitis C virus protease inhibitors recently approved for therapeutic use have likewise become the standard of care (28). Thus one of the most prominent attempted strategies for dengue virus therapy has been development of small compounds that target the active site NS2B-NS3pro (29–33).

NS2B-NS3pro has a catalytic triad (His51 – Asp75 – Ser135) in the active site. NS2B-NS3pro has a somewhat altered specificity relative to other trypsin-like serine proteases. The fact that NS2B-NS3pro recognizes only sites that contain two cationic residues, whereas trypsin recognizes sites containing a single cationic residue, has necessitated the development of new classes of inhibitors for targeting the active site (34). The structural boundaries originally annotated within the polyprotein suggested that the NS3 domain encoded the functional protease. Later studies showed that expression of NS3 alone did not lead to production of an active protease, however including a portion of NS2B with NS3 leads full proteolytic activity (16). The crystal structure of NS2B-NS3pro from dengue virus type 2 shows that the catalytic triad is located between two β- sheets, each formed by six β-strands (Fig. 1).

One of the main hurdles in early drug development against this target has been that the NS2B-NS3pro active site is flat, so designing inhibitors by structure-based design has been challenging (35–37). Another difficulty in finding orthosteric drugs is that NS2B-NS3pro prefers positively charged amino acids in peptide and peptidomimetic substrates. Charged molecules typically display limited oral bioavailability, furthering the difficulty in design of peptidomimetics against NS2B-NS3pro that are effective inhibitors in vivo.

The work described here is based on our hypothesis that an alternative functional site, such as an allosteric site, might be identified and targeted for drug development. We define allosteric sites as those regions of the protein that, when bound to a small-molecule ligand, undergo a change in conformation or an alteration in the conformational equilibrium that impacts enzymatic function. Allosteric sites have previously been found to be important in proteases in general (for review see references 38 and 39), making it likely that allosteric sites in NS2B-NS3pro likewise could be exploited. In this work we have used both novel and known cysteinereactive chemical probes to identify sensitive regions of NS2B-NS3pro that may serve as functional allosteric sites.

RESULTS AND DISCUSSION

Most studies targeting NS2B-NS3pro by small-molecule inhibitors have focused on the active site, but unfortunately no drug that inhibits the enzyme via active site binding has been approved to date. In this work we reason that finding other sites — potentially allosteric sites that can be targeted for enzyme inhibition — may prove fruitful in drug development against dengue virus. In this study we have interrogated potential allosteric sites in NS2B-NS3pro that can be targeted by small-molecule compounds. Putative allosteric sites spaced across the protein surface were identified by visual inspection of the crystal structure of unliganded NS2B-NS3pro (PDB: 2FOM, Fig. 1). Eight individual cysteine residues were introduced near these cavities (Supplementary Fig. S1A–D) to provide reactive handles for cysteine-tethering-based covalent-compound binding. Wild-type (WT) NS2B-NS3pro does not have any intrinsic cysteine residues and therefore provides an ideal scaffold for this interrogation. The effects of single cysteine additions on enzyme activity were determined using a GRR-aminomethylcoumarin fluorogenic substrate (Table 1). The majority of the cysteine substitutions had no effect on catalytic activity; however, substitution of cysteine for Trp83, Thr111, or Leu115 fully abolished enzymatic function. Trp83 is a fully buried residue, so it is likely that formation of a cavity and destabilization of the folded form of NS2B-NS3pro results in loss of activity. Thr111 and Leu115 are mostly solvent exposed, so their loss of function is more surprising. Nevertheless, since most of the variants retained catalytic properties similar to the wild-type NS2B-NS3pro, we were able to undertake a cysteine scanning and tethering approach to identify potential allosteric sites.

Our approach to identify new allosteric sites on NS2B-NS3pro shows some similarities to previous versions of cysteine tethering that have lead to identification of allosteric sites (40–44), but is unique in that it requires only a small panel of cysteine-reactive probes (Fig. 2A). Original tethering libraries intended for ligand discovery were composed of a large number (1,200–40,000) of specific monophores (unique chemical entities or pharmacophores) each attached to a weakly electrophilic thiol moiety. Our panel for allosteric site discovery, is composed of monophores that we hypothesized would be broadly reactive, and are attached to highly reactive electrophiles. We included the biarylether monophore because in screening of large tethering libraries this monophore was broadly reactive, probably due to its hydrophobic nature and shape, which can be accommodated in many cavities and crevices on protein surfaces. In addition, we included more reactive electrophilic warheads including chloromethylketones and activated disulfides.

Each active cysteine variant was incubated with the panel of cysteine-reactive probes. To monitor the accessibility of the introduced thiols we used DTNB, a quantitative cysteine-reactive colorimetric reagent, which also functions as an inhibitor of A125C NS2B-NS3pro (Fig. 2). Most of the cysteine insertions resulted in shortening of the side chain at the insertion position, with the exception of A125C, which extends the side chain by addition of a thiol. Nevertheless, we found that all of the introduced cysteines were able to bind to DTNB and BAClMK (Table 1), suggesting that all are accessible and reactive. After incubation with the panel of probes, the remaining NS2B-NS3pro activity was measured. All of the introduced cysteines in active versions of NS2B-NS3pro are available, reactive, and are fully modified by the probes (Table 1). Two of the inactive variants (T111C and L115C) are fully susceptible to reactions with the introduced cysteine, but W83C only partially reacted. Although the introduced cysteines in the active versions of NS2B-NS3pro are capable of becoming fully modified, we found that only the A125C variant was inhibited by compound binding (Fig. 2B). A125C was fully inhibited by DTNB, Aldrithiol, or biarylchloromethylketone (BAClMK, Supplementary Fig. S2A), which form covalent adducts to Cys125, indicating that these chemical probes are effective inhibitors at this site. Inhibition by DTNB could be fully reversed by the addition of the reductant (β-mercaptoethanol), indicating that binding was cysteine dependant (Fig. 2C). A125C was only weakly inhibited by the non-cysteine-reactive biarylamine (Fig. 2B), suggesting that covalent binding is required for inhibition. A thiol-based biaryl compound (biarylthiol, Supplementary Fig. S2B) failed to inhibit A125C under these conditions, suggesting that a more reactive electrophile is required for reaction and inhibition. This also suggests that any interaction between the biaryl monophore and the protein is transient or weak. Covalent modification of A125C by DTNB or BAClMK was readily observed by mass spectrometry (Table 2), but no binding to WT NS2B-NS3pro was observed. This further demonstrates that both DTNB and BAClMK react with the introduced cysteine at A125C but do not react with the active site nucleophile, Ser135, or any other functional group present on the protein. This suggests that NS2B-NS3pro inhibition is not due to a random mechanism and that inhibition is mediated by compound binding specifically to the introduced cysteine at residue position 125.

To further probe the mechanism of inhibition at A125C, we measured the circular dichroism (CD) spectra to monitor large-scale changes in the structure upon cysteine introduction and compound tethering. Introduction of A125C had no pronounced effect on the CD spectra, suggesting that any changes in the structure of protein were minor (Fig. 3). This finding is also consistent with the observation that the catalytic properties of A125C were essentially unchanged from WT. Binding of DTNB also had no noticeable effect on the CD spectra, suggesting that the mechanism of DTNB-based inhibition was not through unfolding or disruption of the global NS2B-NS3pro structure.

Given that DTNB binding did not impact the overall structure, we explored whether compound binding would impact the stability of the protein. Thermal denaturation of NS2B-NS3pro as monitored by circular dichroism showed that WT, A125C, and DTNB-bound A125C all had T_ms of 50±1°C, suggesting that DTNB binding has no effect on stability. In addition, we monitored the impact of DTNB binding on other cysteine insertion variants. E19C, L31C, E86C, and N105C versions of NS2B-NS3pro all also showed similar T_ms to WT and A125C and changes in T_m of 1°C or less upon binding of DTNB (Supplementary Fig. S3), suggesting that the mode of interaction is similar between all of the variants. Thermal denaturation of NS2B-NS3pro results in an unusual, but not unprecedented (45), gain in CD signal; therefore, we also monitored stability using intrinsic tryptophan fluorescence. A125C showed stability indistinguishable from WT and DTNB binding also had no impact on stability (Fig. 4). The above findings suggest that the structures of A125C and the A125C complexes with ligands are similar to WT NS2B-NS3pro and that covalent compound binding resulting in the inhibited form is not altering the folding-unfolding equilibrium.

One of the most common sources of false positives in early drug discovery is formation of organic small-molecule colloidal aggregates or small molecule-protein co-aggregates, which can sometimes inhibit enzymes nonspecifically (46). The reasons for these phenomena are not well understood but a possible mechanism is that single enzyme molecules bind tightly and non-specifically to small compound aggregates, thereby leading to enzyme inactivation in protein:small-molecule co-aggregates. This kind of inhibition can be problematic, particularly with hydrophobic compounds such as the probes used here, and in our case could lead to spurious identification of allosteric sites. To investigate whether this type of mechanism is involved in inhibition of A125C, we subjected A125C to size exclusion chromatography, native gel electrophoresis, and dynamic light scattering (Fig. 5). WT, A125C, and DTNB-bound A125C all showed identical retention on a size-exclusion column, which was consistent with a monomer of 26 kDa (Fig. 5A). Likewise, WT, A125C, and DTNB-bound A125C all migrated at the same position on a native gel, suggesting that all three species are monomers of similar mobility (Fig. 5B). This also indicates that no intermolecular disulfides are forming that could be responsible for the observed loss of activity for DTNB-bound A125C. Finally, A125C had a comparable distribution of particle sizes by dynamic light scattering, further suggesting that the inhibited form of A125C is fully monomeric, soluble, and well behaved, even though it is fully inhibited (Fig. 5C). These analyses all suggest that no spurious or aggregation-mediated mechanisms of inhibition are responsible for the loss of activity conferred by DTNB and the other A125C inhibitors.

Together, these results suggest that targeting the 120s loop region, or potentially a cavity or crevice near the Ala125 site, with a small molecule inhibitor may be an effective way to achieve enzyme inhibition. To visualize the mechanisms of DTNB and BAClMK inhibition of NS2B-NS3pro, we sought to determine the crystal structures of A125C in complex with DTNB or BAClMK. We first determined the structures of unliganded WT and A125C (Supplementary Table S1) under pH 5.5 conditions, similar to those observed previously for WT NS2B-NS3pro from dengue virus serotype 2 (DENV2, (35)). Both WT and A125C had the same overall fold as the published structure of WT NS2B-NS3pro (PDB: 2FOM (35)) except that the 120s loop (residues 117–122, which are ~15Å away from the substrate-binding groove) was in a different conformation. Because two of the covalent inhibitors, BAClMK and DTNB, do not react to form a covalent adduct with A125C at pH 5.5 but do bind and robustly inhibit NS2B-NS3pro at pH 8.5, we sought alternative, neutral pH crystallization conditions. Extensive screening of unliganded NS2B-NS3pro failed to produce any new crystallization conditions. Numerous screens of NS2B-NS3pro covalently modified with BAClMK or DTNB also failed to produce crystals under any conditions, so we developed a method for transferring the unliganded A125C pH 5.5 crystals to pH 8.5 without negatively impacting diffraction (Supplementary Table S1). Crystals subsequently soaked with BAClMK failed to diffract (Supplementary Table S1). Fortunately, we were able to soak DTNB into NS2B-NS3pro crystals, yielding viable crystals that diffracted to 1.74 Å resolution (Supplementary Table S1). Analysis of this structure in detail suggested that conformational changes in the 120s loop could be important.

In the absence of substrate, DENV2 NS2B-NS3pro can adopt what is termed the open (inactive) conformation, wherein the NS2B region (Fig. 6A) sits at the base of the protein below the active site (Fig. 6B). No crystal structures of NS2B-NS3pro from DENV2 have been reported with substrate or active-site inhibitors bound, however the molecular details of the closed (active) conformation are available based on homology to the structure of NS2B-NS3pro from dengue virus serotype 3 (DENV3), which is 68% identical to that of DENV2 (Fig. 6C). A structure of the closed form of DENV3 NS2B-NS3pro was determined in the presence of a substrate-like active-site inhibitor (Nle-Gly-Lys-Lys-aldehyde). In the presence of active-site inhibitor, the NS2B region undergoes a large conformational rearrangement such that residues 73–88 sit above the active site near the 120s loop (Fig. 6D). It appears that this conformational rearrangement is required for activity and is essential for substrate binding and catalytic turn over. Recent extensive NMR analysis suggests that in solution NS2B-NS3pro favors the closed conformation in the presence and absence of substrate (47). Thus, flexibility of the NS2B region of the protease is critical for function.

The 120s loop also appears to be quite flexible; we and others have observed a number of conformations of this loop in unliganded versions of WT and A125C NS2B-NS3pro (Fig. 6E). The conformation of the 120s loop in the DTNB-soaked structure is incompatible with the active (closed) conformation of the NS2B region. Superposition of the DTNB-soaked structure with the active conformation of NS2B-NS3pro from DENV3 shows a steric clash between the 120s loop and the active conformation of NS2B (Fig. 6F). While we were not able to observe any density for DTNB attached to the cysteine residue at position 125 (Cys125), we are certain that the reaction between Cys125 and DTNB occurred in the crystals due to the appearance of a yellow color resulting from the DTNB reaction. In contrast, the same reaction did not occur with mother liquor alone. The presence of large (30Å diameter) solvent channels in the NS2B-NS3pro crystals (Supplementary Fig. S4) appear to provide ample access of DTNB to the A125C residues throughout the crystals. To thoroughly explore the possibility of weak density or multiple conformations for DTNB we used placement based on Glide (48, 49) with density-guided and force-field-based refinement. While the overall model statistics improve with the PrimeX refinement (Supplementary Table S1), no models could be generated that place DTNB in areas of significant electron density. As such, we hypothesize that DTNB may have incompletely reacted, dissociated after freezing or during synchrotron radiation or is dramatically disordered.

Although we were unable to observe any electron density for DTNB attached to the cysteine at position 125 (Supplementary Fig. S5), the presence of DTNB in the crystal does have a notable effect on the structure of A125C we observe. In addition to the altered conformation of the 120s loop, residues Thr77–Met84 are ordered in the DTNB-soaked structure — these residues were disordered in all structures to date of DENV2 NS2B-NS3pro including the high-resolution structures 2FOM (1.5 Å) and the WT pH 5.5 structure reported here (1.46 Å). We assume that the NS2B region of NS2B-NS3pro is in equilibrium between the closed (active) and open (inactive) conformations; it appears that the presence of DTNB may stabilize the open (inactive) conformation. Only in the presence of DTNB can we observe density for the Thr77-Met88 stretch, which is ordered by crystal contacts between Thr76 and Asn119 in the 120s loop (Supplementary Fig. S6). Thus, one model of DTNB-based inhibition is that this steric clash between the flexible 120s loop and the active conformation of the NS2B region is responsible for the inhibitory activity of DTNB.

A second model of inhibition by DTNB and BAClMK also involves NS2B reorganization. Modeling covalent binding of either DTNB (Fig. 6G) or BAClMK (Fig. 6H) to C125, we see that a likely conformation for the inhibitor is protruding toward the 120s loop in a position that would sterically clash with the active conformation of the NS2B region. This steric clash would prevent the NS2B region from adopting the closed (active) conformation thereby shifting the ensemble toward the open (inactive) conformation we observe in the DTNB-soaked structure. We favor this model for explaining the inhibition of NS2B-NS3pro activity with three structurally divergent compounds, all of which could utilize the same mechanism. Due to the covalent nature of the interactions between the probes and NS2B-NS3pro variants, we expect that if any other interrogated sites were equally susceptible to allosteric inhibition, our approach would have also recognized those sites. Given that only A125C was inhibited by any of the compounds, we posit that this region is one of the most susceptible regions of the protein for allosteric inhibition. We predict that small molecules that bind between the 120s loop and the 150s loop (residues 153–164, which sit below the S3 subsite) will serve as effective inhibitors of NS2B-NS3 protease activity (Supplementary Fig. S7).

CONCLUSIONS

The approach we developed in this work allowed us to identify the Ala125 allosteric site in NS2B-NS3pro and is promising for future applications due to its simplicity and feasibility. In this study, interrogating just eight single-site cysteine substitutions with a panel of just five chemical probes, three of which are commercially available, we were able to identify a productive allosteric site that was susceptible to full inhibition by three diverse molecules. The approach we describe is also generally applicable for allosteric site identification in a wide range of protein targets. This cysteine-based approach is particularly well suited to target proteins like NS2B-NS3pro that lack native cysteines. The lack of cysteines need not be a requirement for success, provided that at least some of the probes have no effect on the function of the native target protein or as is often the case, removal of reactive surface cysteines has no impact on function.

Prior to this study several pieces of evidence have suggested that the dengue virus protease family should be susceptible to allosteric inhibition. The 3F10 antibody disrupts the interaction of NS2B with NS3 in vitro, suggesting that allosteric inhibition of NS2B-NS3pro should be possible (50). Antibody-based inhibition of proteases has been used for half a century, and some of inhibitory antibodies utilize allosteric mechanisms (for review see reference 51). For example, another trypsin-like protease, hepatocyte growth-factor activator, can be allosterically inhibited by the Fab75 antibody through binding an active-site adjacent loop (residues 99–102) (52). Although the homologous region is a short turn in NS2B-NS3pro (residues 72–74) and thus not useful as an allosteric epitope, we can envision that a similar inhibitory mechanism could be used by an NS2B-NS3pro 120s- or 150s-specific antibody. Finally, compounds derived from in silico docking experiments that function as non-competitive inhibitors have also been reported (53). While we are unable to ascertain whether an innate biological role of the Ala125 allosteric site exists, it is our aim to continue to assess the native role and exploit this site chemically with specific, non-covalent ligands or antibodies.

The prevalence of diseases caused by flaviviruses is rapidly expanding, raising interest in the proteases across this family as drug targets. A similar conformational change in the NS2B region is observed for the protease from West Nile virus, which shares 46.8% identity with DENV2 NS2B-NS3pro, suggesting that the mechanism of activation may be conserved across the family. In the unliganded state, the NS2B region of NS2B-NS3pro from West Nile virus is in the open conformation. When substrate binds, the NS2B region moves above the 120s loop (Supplementary Fig. S7) adopting the closed conformation. Given the similarity in the mechanism of activation, we predict that the protease from West Nile virus would also be allosterically inhibited in a similar manner at the region of the 120s and 150s loops. In fact, uncompetitive/allosteric inhibitors of West Nile virus have been previously described, but no structural information on the site of binding has been reported. It is tempting to consider whether the Ala125 region could be the target of the pyrazole-ester-based West Nile NS2B-NS3 protease inhibitors (54, 55).

Allosteric inhibition of dengue virus protease is of potential interest due to the lack of clinically available inhibitors that target the NS2B-NS3pro active site. All small molecule binding sites, whether they be allosteric or otherwise, are composed of three components: a cavity (which may be induced by a small molecule), polar residues that contribute directional forces allowing specificity, and hydrophobic interactions that provide the enthalpy for binding. The region identified near Ala125 appears to possess these three essential elements. Ala125 sits between the 120s and 150s loops, which our work has shown to be very flexible. Thus it appears that the shape of this cavity is malleable, potentially accommodating a number of chemical entities in addition to the BAClMK and DTNB. This region also has a number of appropriately positioned polar residues (e.g. Asn119, Thr118, and Thr156) for providing specificity and hydrophobic residues (e.g. Phe116, Ile123, Val154, Val155, and Val162) for providing binding affinity (enthalpy). Our studies use a covalently linked version of NS2B-NS3pro that has been widely utilized and characterized (35). Recently Keller and coworkers have shown that unlinked NS2B-NS3pro exhibits slightly higher activity (up to 5 fold) and appears to prefer the closed conformation even prior to substrate binding (56). Nevertheless, we anticipate that the allosteric site identified at Ala125 will also be effective against the unlinked NS2B-NS3pro, but that remains to be tested. Development of non-covalent inhibitors that utilize the site we have identified is not trivial but prior work has demonstrated that covalent tethering hits can effectively be developed into non-covalent inhibitors (40, 57, 58). Thus we envision that both structure-guided rational approaches that take loop flexibility into account and empirical screening for inhibitors against NS2B-NS3pro may be fruitful routes to further exploitation of this allosteric site.

Drug resistance becomes an issue in any rapidly evolving system, including pathogenic human viruses such as dengue virus, so the development of allosteric inhibitors requires careful consideration. While combination treatments affecting two protein targets have been widely used in viral therapy, the combination of active-site and allosteric inhibitors against a single protein target has not been tested. We hypothesize that simultaneously targeting both the active and allosteric sites of NS2B-NS3pro should delay the onset of resistance over single-site inhibition. At the very least, the system described here will enable us to measure development of resistance in single-target combination therapy and directly assess whether active- or allosteric-site inhibitors are more prone to development of resistance in NS2B-NS3pro. In the best-case scenario, one might envision development of clinically useful allosteric inhibitors that target the allosteric site we have identified.

METHODS

Methods for all aspects of this work are contained in the Supporting Information.

Supplementary Material