Inhibitor‐induced dimerization mediates lufotrelvir resistance in mutants of SARS‐CoV‐2 3C‐like protease

Abstract

The emergence of SARS‐CoV‐2 and other lethal coronaviruses has prompted extensive research into targeted antiviral treatments, particularly focusing on the viral 3C‐like protease (3CL^pro) due to its essential role for viral replication. However, the rise of drug resistance mutations poses threats to public health and underscores the need to predict resistance mutations and understand the mechanism of how these mutations confer resistance. The binding of inhibitor to 3CL^pro drives it from the monomeric to the active dimeric form, which can counterintuitively lead to enzyme activation rather than inhibition. Furthermore, we find this allosteric coupling between binding and dimerization is sensitive to mutation, leading to a new mechanism for drug resistance. Understanding the relationship between inhibitor binding and dimerization is important for resistant strain surveillance and development of robust antivirals. Herein, we present a systematic study of drug resistance mediated by inhibitor‐induced dimerization of 3CL^pro.

1. INTRODUCTION

Coronaviruses have long been recognized as human pathogens (Kahn & McIntosh, 2005). However, they were not considered a major public health threat until relatively recently, as they were primarily associated with mild upper respiratory tract infections, similar to the common cold, and the significance of these viruses was largely underestimated (Kapikian, 1975; Macnaughton, ¹⁹⁸²; Myint, ¹⁹⁹⁵; Peiris et al., ²⁰⁰³; Tyrrell & Bynoe, ¹⁹⁶⁵). The emergence of severe acute respiratory syndrome coronavirus 1 (SARS‐CoV‐1) in 2002 (Peiris et al., 2003), followed by Middle East respiratory syndrome coronavirus (MERS‐CoV in 2012) (Zhang et al., 2021), and severe acute respiratory syndrome coronavirus 2 (SARS‐CoV‐2) (the coronavirus disease 2019 (COVID‐19 pandemic) in 2019 (WHO, ²⁰²⁴), drastically changed the perception of coronaviruses, highlighting their potential to cause severe disease in humans and create global health emergencies. Since the beginning of the COVID‐19 pandemic, there have been numerous drug campaigns aiming to discover effective antivirals against SARS‐CoV‐2. These small molecule antivirals usually target various stages of the virus replication cycle.

3CL^pro (also known as main protease, M^pro) is regarded as an attractive drug target as it is responsible for proteolytic processing of the majority of viral polyprotein cleavage sites, releasing nonstructural proteins (nsps), including RNA‐dependent RNA polymerase (RdRp), which are essential for the viral lifecycle (Thiel et al., 2003; V'kovski et al., ²⁰²¹; Ziebuhr et al., ²⁰⁰⁰). 3CL^pro is a highly conserved, dimeric cysteine protease (Graziano et al., 2006; Hsu et al., ²⁰⁰⁵; Mody et al., ²⁰²¹), which is only active in the dimeric form (Chen, Zhang, et al., 2008; Ferreira et al., ²⁰²²; Graziano et al., ²⁰⁰⁶; Nashed et al., ²⁰²²; Zhang et al., ²⁰¹⁰). Currently, the Food and Drug Administration (FDA) has approved one antiviral targeting the 3CL^pro for the treatment of COVID‐19: the orally administered Paxlovid (ritonavir‐boosted nirmatrelvir) (U.S. Food & Drug Administration, ²⁰²³). Paxlovid has an excellent 88% efficacy against hospitalization or death in unvaccinated high‐risk adults (Hammond et al., 2022). In addition, lufotrelvir is another potent 3CL^pro inhibitor developed by Pfizer as a potential treatment for COVID‐19 (Figure 1) (Hoffman et al., 2020), which also demonstrates broad‐spectrum effectiveness against various 3CL^pro enzymes across the coronavirus genera (Boras et al., 2021).

FIGURE 1.

Representative structures of 3CL^pro inhibitors and their in vitro potency values (Rai et al., 2022; Vries et al., ²⁰²¹).

Drugs targeting viral replication run the risk of selecting for variants with mutations that allow viruses to evade the therapeutic effects, especially in the face of large virus populations and high mutation rates (Irwin et al., 2016). Thus, the ability to avoid drug resistance mutations is essential for developing treatments with long‐term effectiveness. Several studies have already investigated 3CL^pro mutations that confer resistance to nirmatrelvir (active ingredient of Paxlovid) using both computational methods (Havranek et al., 2023; Sasi et al., ²⁰²²) and in vitro serial viral passage experiments (Heilmann et al., 2023; Iketani et al., ²⁰²³; Jochmans et al., ²⁰²³; Zhou et al., ²⁰²²). Moreover, naturally occurring mutations that are resistant to nirmatrelvir have also been discovered (Hu et al., 2023; Moghadasi et al., ²⁰²³; Noske et al., ²⁰²³). On the other hand, there have only been a small number of studies of 3CL^pro resistance mutations against other potent inhibitors such as lufotrelvir (Hu et al., 2023; Jiang et al., ²⁰²³). Understanding the link between mutation and resistance is important for managing the emergence of resistance and offers valuable insights for developing next‐generation antivirals.

3CL^pro is only active as a homodimer, yet it naturally exists in equilibrium between its monomeric and dimeric forms in solution (Ferreira et al., 2022). Interestingly, the monomer/dimer equilibrium is affected by ligand binding: 3CL^pro from SARS‐CoV (Cheng et al., 2010; Li et al., ²⁰¹⁰), MERS‐CoV (Tomar et al., 2015), and SARS‐CoV‐2 (Jochmans et al., 2023; Nashed et al., ²⁰²²; Paciaroni et al., ²⁰²³) all exhibit ligand‐induced dimerization (LID), where the binding of a ligand to the monomer promotes dimerization and formation of the active dimer. It implies that allosteric communication occurs between the ligand binding site and the dimerization interface such that ligand binding increases the affinity of dimerization and vice versa. LID has been shown to be essential for the maturation of polyprotein and the production of mature SARS‐CoV 3CL^pro, and is proposed as a mechanism to regulate enzyme activity during virus replication (Li et al., 2010). Moreover, under conditions where the enzyme is primarily in the inactive monomeric form, LID can lead to the counter‐intuitive situation where enzyme activity initially increases with increasing concentrations of the inhibitor. Binding of inhibitor to one monomer promotes dimerization, with one subunit bound to the inhibitor and the other subunit unoccupied and active. As such, LID leads to distinctive dose–response curves, with increasing activity at low inhibitor concentrations, maximum activity at an intermediate inhibitor concentration, followed by complete inhibition at high inhibitor concentrations. Given that LID has been demonstrated as an enzyme activation mechanism that shifts inhibition to higher concentrations (Tomar et al., 2015), we propose that LID can serve as a potential mechanism for drug resistance. However, to date, analyses of LID and its characteristic dose–response curves have only been qualitative, as no physical chemical model has been developed. Even though the effect of different inhibitors on the tendency of dimerization, defined by the dissociation constant, K _D, has been quantified as a function of the inhibitor concentration (Silvestrini et al., 2021), a model that explains LID is still lacking and it still remains unclear whether LID actually attenuates inhibition as opposed to simply altering the shape of the dose–response curve. A quantitative, mechanistic description of LID is essential for understanding its potential role in drug resistance.

To better understand this mechanism, we focused on lufotrelvir and employed a combination of molecular dynamics (MD) simulations, site‐directed mutagenesis, and Förster resonance energy transfer (FRET) based activity assays. We developed a general, quantitative model of LID, which allowed us to separately extract the affinity of inhibitor binding, the strength of allosteric coupling between the ligand binding site and dimerization interface, and to calculate how modulating LID could mediate resistance. We demonstrated LID as a new and hitherto underappreciated mechanism that can be exploited by pathogens to evade inhibitor drugs.

2. RESULTS

2.1. Computational predictions of lufotrelvir resistance mutations

Possible sites of resistance mutations may be located by calculating the distances of enzyme/substrate and enzyme/inhibitor interactions on a per‐residue basis in crystal structures or simulations. Residues that contact the inhibitor substantially more than the substrate may be considered possible mutation sites, as mutations could preferentially disrupt interactions with the inhibitor while retaining catalytic activity (Friedman, 2022). In this work, MD simulations were performed on 3CL^pro/substrate (TSAVLQSGFRK) and 3CL^pro/lufotrelvir complexes following previously published protocols (Osorio et al., 2022; Yañez et al., ²⁰²¹). The resulting trajectories were clustered into representative structures, and the strengths of interactions between 3CL^pro residues and either the substrate or the inhibitor were quantified using non‐covalent interactions. These values were population‐averaged across the representative structures and are reported in parallel coordinate plots in Figure 2a, comparing non‐covalent interaction values to the substrate and to the inhibitor. This comprehensive analysis revealed distinct interaction patterns between the substrate and inhibitor binding modes. By comparing these interaction profiles, as visualized, we identified specific residues that exhibited stronger interactions with the inhibitor compared to the substrate. This differential binding pattern was particularly significant because it suggested potential sites where mutations could selectively disrupt inhibitor binding while preserving substrate recognition. Four residues near the active site produced higher non‐covalent interaction values with the inhibitor than with the substrate, namely Y54, Q189, A191, and S1 (from the other subunit of the dimer). However, for Q189, the side chain was oriented away from the inhibitor and toward the solvent. Thus, mutations at this site were unlikely to affect inhibitor/enzyme interactions, and this residue was not investigated further. We therefore decided to focus our attention on S1, Y54, and A191, as shown in Figure 2b,c.

FIGURE 2.

Non‐covalent interactions of 3CL^pro with the substrate and the inhibitor. (a) Interactions of the substrate compared with the inhibitor, where "*" marks residues with higher non‐covalent interaction values with the inhibitor than with the substrate. (b) Positions of the potential resistance‐inducing residues near the binding site of lufotrelvir (in magenta). (c) Positions of the potential resistance‐inducing residues in the overall dimeric structure of 3CL^pro (Protein Data Bank [PDB] code: 6XHM).

2.2. Experimental characterization of potential resistance mutations

For each of the three residues identified in the MD analysis above, we generated three to four single site mutants that modified the physical properties of the targeted residue, for instance by changing the bulkiness, polarity, or charge, while restricting modifications to mono‐ or di‐nucleotide substitutions. For S1, S1A (polar to nonpolar), S1I (increased bulkiness), and S1D (neutral to negatively charged) mutations were selected. For Y54, Y54A (reduced bulkiness), Y54D (to negatively charged), Y54F (no OH), and Y54K (positively charged) were chosen. For A191, A191E (nonpolar to negatively charged), A191I (increased bulkiness), and A191R (nonpolar to positively charged) were selected.

All mutants were expressed with high yields in Escherichia coli and purified using previously published procedures (Stille et al., 2022). Activity was initially tested at a fixed enzyme concentration of 15 nM in a FRET‐based enzyme assay (Figure 3a) (Grum‐Tokars et al., 2008). S1I, S1R, Y54F, and A191I mutants exhibited similar activities to the wild type (WT), while the S1A and A191E mutants were 1.5 and 1.9 times more active than the WT, respectively. On the other hand, Y54 mutants, except for Y54F, showed no enzyme activity. Y54 is a highly conserved residue among Coronaviridae and is important for maintaining the enzyme's catalytic activity and structural stability (Al Adem et al., 2023). Mutation of this residue likely caused the binding site to collapse, resulting in inactive enzymes. Additionally, the S1D mutation was also detrimental to the enzyme, resulting in only 5% activity in comparison to the WT.

FIGURE 3.

Enzyme activities and dose–response curves of the wild‐type and mutant 3CL^pro with lufotrelvir in FRET assay. (a) Relative enzyme activities of the wild‐type 3CL^pro compared to the mutants, with the activity of wild‐type 3CL^pro set to 100%. (b) Dose–response curves of the wild‐type 3CL^pro with the mutants in logarithmic scale. All assays were performed at a fixed enzyme concentration of 15 nM and a fixed substrate concentration of 12 μM. The activity of the enzyme in the absence of inhibitor is set as 100% for each individual enzyme. All measurements are in triplicate. The error bars represent standard deviations of the data points.

Dose–response curves were measured with the WT and all the active mutants (Figure 3b), and the inhibition constants (K _i) were determined for all of them except S1D (Table 1) by fitting to the Morrison equation with Equation (1) and (2), where [E], [I], and [S] refer to the molar concentrations of enzyme, inhibitor, and substrate, respectively, while K _m represents the Michaelis constant (Morrison, 1969). The values of K _m are measured and presented in the later part of the paper. The K _i value of the WT (0.25 ± 0.04 nM) is in excellent agreement with the literature value (0.27 ± 0.01 nM) (Hoffman et al., 2020). Most mutants had similar K _i values to that of the WT, except for the A191E and A191I mutations, which showed increased K _i values of 1.6 ± 0.1 and 1.2 ± 0.1 nM, respectively. We speculate that the close proximity of the 4‐methoxyindole moiety of lufotrelvir to A191 (Figure 2b) led to some steric repulsion when the alanine was replaced with a bulkier residue.

$$\%\text{Activity}=100\times \left(1-\frac{\left(\left[\mathrm{E}\right]+\left[\mathrm{I}\right]+K_{\mathrm{i}}^{\mathrm{app}}\right)-\sqrt{{\left(\left[\mathrm{E}\right]+\left[\mathrm{I}\right]+K_{\mathrm{i}}^{\mathrm{app}}\right)}^2-4\left[\mathrm{E}\right]\left[\mathrm{I}\right]}}{2\left[\mathrm{E}\right]}\right).$$ (1)

$$K_{\mathrm{i}}^{\mathrm{app}}=K_{\mathrm{i}}\left(1+\frac{\left[\mathrm{S}\right]}{K_{\mathrm{m}}}\right).$$ (2)

TABLE 1.

Measured K _i values of lufotrelvir with the WT and mutant 3CL^pro, with all values expressed as mean ± standard deviation (SD).

Enzyme	K i (nM)
WT	0.25 ± 0.04
S1A	0.41 ± 0.08
S1I	0.35 ± 0.09
S1R	0.60 ± 0.09
Y54F	0.45 ± 0.09
A191E	1.6 ± 0.1
A191I	1.2 ± 0.1
A191R	0.9 ± 0.2

2.3. Characterization of the S1D mutant

S1 is located in the 3CL^pro dimer interface, where it forms a hydrogen bond with E166 of the other subunit and stabilizes the dimer (Figure 2b) (Ferreira et al., 2022). We assumed that the low enzyme activity of the S1D mutant was due to charge repulsion with E166 that weakened the dimer interface and shifted the monomer‐dimer equilibrium toward the non‐active monomer. To test this, we measured the disassociation constant, K _D ^app, of the process D → 2M, for the mutant using a concentration‐dependant enzyme activity assay (Figure 4), which is based on a previously published FRET‐based enzyme assay (see Section 5 for details) (Wang et al., 2024). At very low enzyme concentrations, the enzyme was mostly monomeric and inactive. With increasing enzyme concentrations, the active dimer formed, and the activity increased, leading to upward concave curvature in a plot of the catalytic rate v as a function of enzyme concentration [E] (Figure 4). The data were fit as described previously (Wang et al., 2024), yielding K _D ^app = 2000 ± 400 nM. This is significantly higher than the reported K _D ^app values of WT SARS‐CoV‐2 3CL^pro (ranging from 16 to 140 nM) (El‐Baba et al., 2020) (Wralstad et al., 2023) (Wang et al., 2024), demonstrating that the S1D mutation indeed disrupted the dimer interface and resulted in a weakly binding dimer. Furthermore, the weakened dimer interface completely explains the decrease in activity observed for the S1D mutant, since it was only 2% in the active dimeric form under the conditions of the assay (15 nM), compared to the WT, which was 48% active.

FIGURE 4.

Concentration‐dependent activity assay of S1D showing the upward concave shape of the curve. All experiments were performed in triplicate. The error bars represent standard deviations of the data points.

In addition, the dose–response curve of S1D exhibited pronounced downward concave curvature with a peak at approximately 12.5 nM inhibitor, where the enzyme activity was nearly 250% that of the negative control without inhibitor (Figure 3b). The enzyme was only fully inhibited when the inhibitor concentration exceeded 1 μM. Such downward concave dose–response curves were previously observed for WT MERS‐CoV 3CL^pro (Tomar et al., 2015) and a mutant of SARS‐CoV‐2 3CL^pro (Nashed et al., 2022) and attributed to LID. However, to date, analyses on LID have only been qualitative, and the relationships between LID, inhibitor binding, and enzyme catalysis remain poorly understood. We have developed a quantitative model to describe and analyze this phenomenon, as illustrated in Scheme 1.

SCHEME 1.

Proposed model for ligand‐induced dimerization. The monomer, M, can dimerize to form the active dimer, D, or bind to the inhibitor, I, to form the complex MI, which is penalized by the factor f _c. MI can bind to another monomer to form DI, which is controlled by the dissociation constant K _D. D can also bind to the inhibitor to form the singly occupied DI, which is controlled by the inhibition constant K _i. DI can also bind to I to form DI₂, which is also governed by K _i. Both D and DI are assumed to be catalytically active, which catalyze the substrate, S, to the product, P.

LID occurs when there is positive allosteric coupling between ligand binding and dimerization. This manifests as weaker dimerization if both subunits are ligand‐free, compared to when one of the subunits is bound to ligand, as well as a weaker equilibrium constant for ligand binding in the monomeric state, compared to the dimeric state, as illustrated in Scheme 1. In our model, the strength of the coupling is given by f _c, such that the dissociation constant for dimerization is K _D f _c for D → 2M, K _D for DI → M + MI, and K _D/f _c for DI₂ → 2MI, where D, M, and I are the dimer, monomer, and inhibitor, respectively (Scheme 1). Similarly, the dissociation constant for ligand binding is K _i f _c for MI → M + I and K _i for DI → D + I. Note that both K _D and K _i must be modified by the same factor f _c in order for the dimerization/inhibitor‐binding thermodynamic cycle to be internally consistent. Values of f _c >1 mean that ligand binding increases the strength of dimerization and vice versa. Analysis of LID dose–response data using Scheme 1 is complicated by the fact that the concentrations of enzyme and inhibitor are of the same order of magnitude as the K _i and K _D equilibrium constants, and we are not aware of any analytical solution giving the concentration of active enzyme (D and DI) as a function of inhibitor concentration. We therefore developed a numerical approach for solving these equations, which allowed us to calculate dose–response curves exactly in terms of the dissociation constants for dimerization and inhibitor binding, K _D and K _i, the total concentrations of enzyme and inhibitor, and the strength of allosteric coupling between inhibitor binding and dimerization, f _c (see Section 5).

We applied this approach to the dose–response curves for the S1D mutant against lufotrelvir obtained at 15 and 60 nM enzyme concentrations, fitting both data sets simultaneously (Figure 5a). The downward concave dose–response curve was much more prominent for the concentration at 15 nM compared to 60 nM, since the enzyme initially populated the monomeric state to a greater extent at the lower concentration. The fitted curves agreed well with the experimental data for the S1D mutant (Figure 5a), which yielded values of K _i = 0.82 ± 0.04 nM, K _D = 290 ± 30 nM, and f _c = 21 ± 1 (Table 2). These results also indicated that the inhibition constant for the dimer (DI → D + I) of the S1D mutant (0.82 ± 0.04 nM) is about three‐fold higher than that of the WT (0.25 ± 0.04 nM). The dissociation constant for the fully ligand‐bound S1D dimer (DI₂ → 2MI, K _D/f _c = 13 ± 2 nM) was very close to that of the WT enzyme (17 ± 2 nM). However, the strong allosteric coupling meant that the inhibition constant for the monomer (MI → M + I) of the S1D mutant monomer (K _i f _c = 17 ± 1 nM) is about 60‐fold higher than that of the WT (0.25 ± 0.04 nM) and that the dissociation constant for the ligand‐free dimer (2M → D) of the S1D mutant (K _D f _c = 6000 ± 700 nM) is about 400‐fold higher than that of the ligand‐free WT (17 ± 2 nM). This dimerization constant was in good agreement with the value obtained from enzyme concentration‐dependent activity measurements described above (K _D ^app = 2000 ± 400 nM). Note that fitting the model in Scheme 1 to WT data against lufotrelvir obtained at a 5 nM enzyme concentration gave a value of f _c = 1.7 (Figure 9a and Table 3), indicating that the S1D mutation substantially enhances the allosteric coupling (represented by f _c) between ligand binding and dimerization.

FIGURE 5.

Fitting of the dose–response curves of the S1D mutant with lufotrelvir. (a) Dose–response curves of lufotrelvir with S1D 3CL^pro at 15 nM (green) and 60 nM (blue) enzyme concentration. The activity of the enzyme in the absence of inhibitor is set as 100% for each enzyme concentration. (b) Relative dose–response curves of lufotrelvir with S1D (green) 3CL^pro compared to the WT at 15 nM (black) enzyme concentration. The pink region represents the inhibition concentration where the activity of S1D surpasses that of the WT. The activity of the WT enzyme in the absence of inhibitor is set as 100%, and the activity of S1D in the absence of inhibitor is 5% that of the WT. All measurements are in triplicate. The error bars represent standard deviations of the data points.

TABLE 2.

Measured K _i, K _D, and f _c values of lufotrelvir with the S1D mutant and S1D + A191E mutant, by fitting to the ligand‐induced dimerization model, with all values expressed as mean ± SD.

Enzyme	K i (nM)	K D (nM)	f c	K i f c (nM)	K D f c (nM)	K D/f c (nM)
S1D	0.82 ± 0.04	290 ± 30	21 ± 1	17 ± 1	6000 ± 700	13 ± 2
S1D + A191E	6.4 ± 0.2	140 ± 10	16 ± 1	105 ± 7	2300 ± 300	9 ± 1

FIGURE 9.

Dose–response curve of lufotrelvir with the WT and G11S mutant. (a) Dose–response curves of lufotrelvir with WT 3CL^pro at 5 nM enzyme concentration. The solid blue line is from fitting our ligand‐induced dimerization (LID) model. (b) Dose–response curve of lufotrelvir with the G11S mutant at 15 nM enzyme concentration. The solid pink line is from fitting our LID model. The activity of the enzyme in the absence of inhibitor is set at 100%. All measurements are in triplicate. The error bars represent standard deviations of the data points.

TABLE 3.

Measured K _i, K _D, and f _c values of lufotrelvir, the WT and G11S 3CL^pro, with all values expressed as mean ± SD.

Enzyme	K i (nM)	f c	K D app (nM)
WT	0.3 ± 0.2 a	1.7 ± 0.7 a	17 ± 2 b
G11S	0.2 ± 0.1 a	1.8 ± 0.3 a	10,000 ± 2000 b

^a Ligand‐induced dimerization model.

^b Previous study (Wang et al., 2024).

An important question is whether the S1D mutation provides any benefit to the enzyme when challenged by lufotrelvir, given that the enhanced LID allostery results in destabilization of the active dimer and a roughly 95% reduction in activity relative to the WT at 15 nM. We compared the activities of the S1D and WT enzymes, with the data of S1D normalized to that of the WT (Figure 5b). In the absence of inhibitor, the relative activity of the WT was 100%; while that of S1D was only 5%. However, at all inhibitor concentrations greater than about 10 nM, the activity of S1D was much greater than that of the WT, by a factor up to an order of magnitude. Therefore, the S1D mutation, while detrimental to the activity of the enzyme at low inhibitor concentrations, leads to increased enzyme activity in the presence of moderate concentrations of lufotrelvir, in principle conferring resistance to the drug.

2.4. Testing of the S1D/A191E double mutant

We further investigated the potential resistance mutations S1D and A191E by constructing the double mutant (DM). The DM was also produced with high yield and showed a two‐fold increase in activity compared to the S1D mutant. Dose–response curves of the DM were measured at both 15 and 60 nM enzyme concentrations (Figure 6a) and were typical of LID with pronounced concave‐down shapes. Fitting the curves using Scheme 1 yielded K _i = 6.4 ± 0.2 nM, K _D = 140 ± 10 nM, and f _c = 16 ± 1 (Table 2). Thus, the affinity of lufotrelvir for the DM dimer (DI → D + I, 6.4 ± 0.2 nM) was diminished by 26‐fold compared to the WT (0.25 ± 0.04 nM), which is close to the product (21‐fold) of the 6.4‐fold reduction in affinity for A191E (1.6 ± 0.1 nM) and the 3.3‐fold reduction for S1D (0.82 ± 0.04 nM), indicative of nearly additive changes in binding energy for the two. Again, the dissociation constant of the fully inhibitor‐bound DM dimer (K _D/f _c ≈9 ± 1 nM) was close to that of the WT. However, the increase in the LID allosteric coupling (f _c) meant that the affinity of the inhibitor for the DM monomer (MI → M + I, K _i f _c = 105 ± 7 nM) was reduced about 400‐fold compared to the WT, and the dimerization in the absence of inhibitor was reduced about 130‐fold (K _D f _c = 2300 ± 300 nM). Comparing WT and DM dose–response curves produced a similar result to that of S1D; although the DM showed lower activity than the WT in the absence of inhibitor, it was much more active in the presence of ≥10 nM lufotrelvir.

FIGURE 6.

Fitting of the dose–response curves of the double mutant (DM) with lufotrelvir. (a) Dose–response curves of lufotrelvir with DM 3CL^pro at 15 nM (green) and 60 nM (blue) enzyme concentration. The activity of the enzyme in the absence of inhibitor is set at 100% for each enzyme concentration. (b) Relative dose–response curves of lufotrelvir with DM (pink) 3CL^pro compared to the WT (black) at 15 nM enzyme concentration. The pink region represents the inhibition concentration where the activity of DM surpasses that of the WT. The activity of the WT enzyme in the absence of inhibitor is set at 100%; the activity of DM in the absence of inhibitor is 10% that of the WT. All measurements are in triplicate. The error bars represent standard deviations of the data points.

2.5. Native nanoelectrospray ionization mass spectrometry

We employed nanoelectrospray ionization mass spectrometry (nanoESI‐MS) to validate our observations of perturbed monomer/dimer equilibria and LID in S1D and the DM. As shown in the native mass spectra in Figure 7a, the WT 3CL^pro produced peaks corresponding exclusively to the dimer, consistent with tight dimerization. In contrast, both the S1D and the DM produced peaks corresponding to both monomers and dimers (Figure 7b,c), consistent with weaker dimerization and the presence of both monomers and dimers in solution. The monomeric peaks were slightly weaker for the DM than for S1D, which agrees with the slightly tighter dissociation constant for the DM (K _D f _c = 6000 ± 700 nM and 2300 ± 300 nM for S1D and DM, respectively). It is worth noting that, at a protein concentration of 2 μM, the two mutants are expected to exist as approximately 70% monomer (S1D) and 50% monomer (DM), based on Equation 4 and K _D f _c values determined by fluorescence assay (Table 2). The monomer: dimer peak intensity ratios in the nanoESI‐MS spectra (Figure 7) are substantially less than those values; however, comparisons between monomer and dimer peak intensities must be made with caution, as the transmission efficiencies of the two forms through the mass spectrometer may be highly unequal. Differences in protein conformation and solvation can also lead to dramatically different efficiencies of ionization and MS peak intensities, even for otherwise very similar proteins (Kuprowski et al., 2007). Furthermore, there are multiple charged residues that are buried at the dimer interface (including R4, K12, E14, E166, and E290) but exposed in the monomer, which could contribute to these discrepancies (Ferreira et al., 2022; Kitova et al., ²⁰¹²; Zhang et al., ²⁰⁰³). Importantly, the addition of lufotrelvir eliminated the monomeric peaks for S1D and the DM and caused a shift to exclusively dimeric peaks, providing independent corroboration of ligand‐induced dimerization for these proteins.

FIGURE 7.

Native mass spectrometry analysis of wild‐type and mutant 3CL^pro enzymes. 3CL^pro enzymes were nanoelectrosprayed at 2 μM concentration either alone (top panels) or in the presence of 2 μM of lufotrelvir (bottom panels). The charge states and molecular weights (observed and calculated) are indicated. (a) The WT (purple). (b) S1D mutant (orange). (c) The double mutant (DM) (red). Note that due to the gentle desolvation provided by native ESI‐MS, the protein ions remain partially solvated, resulting in MW_obs that are significantly larger than MW_calc. The WT enzyme existed solely as a dimer at this concentration, whereas small amounts of monomer were detected for the S1D and DM enzymes. Upon lufotrelvir binding, all enzymes were detected only in the dimeric form. The insets show a closer view of the indicated ions to illustrate the binding of up to two equivalents of lufotrelvir.

2.6. Testing the mutants with nirmatrelvir

We subsequently measured the dose–response curves of nirmatrelvir, the active ingredient of Paxlovid, with the WT, S1D, A191E, and DM variants of 3CL^pro, as shown in Figure 8a. The data for the WT and A191E followed simple sigmoidal shapes and were fit to the Morrison Equation (Equations 1 and 2). The K _i of the A191E mutant (2.3 ± 0.1 nM) was very similar to that of the WT (1.7 ± 0.2 nM). This may be rationalized by the observation that, unlike lufotrelvir, nirmatrelvir is not in close proximity to residue A191 in the x‐ray crystal structure (PDB code: 7SI9). Thus, substitutions of A191 would be expected to have less effect on the affinity of nirmatrelvir. The dose–response curves of the S1D mutant and DM both exhibited strong peaks at intermediate inhibitor concentrations, indicative of LID, similarly to what was seen for lufotrelvir. The analyses of S1D (K _i = 1.4 ± 0.1 nM, K _D = 220 ± 50 nM, and f _c = 37 ± 2) and DM (K _i = 6.3 ± 0.4 nM, K _D = 100 ± 20 nM, and f _c = 20 ± 5) also yielded K _i values close to that of WT, as well as K _D and f _c values similar to those obtained for lufotrelvir (Table 4). Thus, the S1D and DM dimers bind nirmatrelvir similarly to the WT, but the LID mechanism is responsible for the 20‐ to 40‐fold reductions in the affinities of the S1D and DM monomers for nirmatrelvir, compared to dimers or the WT. At the same time, the fully nirmatrelvir‐bound S1D and DM dimerize with affinities (K _D/f _c) of 5–6 nM, which is similar to that of the WT, but the dimerization of the unbound monomers is roughly 900‐fold weaker. As a result, both S1D and DM are far more monomeric and less active than the WT at low concentrations of nirmatrelvir, but above about 20 nM of the drug, both S1D and DM show higher absolute activities, indicating that these mutations confer a degree of resistance to nirmatrelvir via the LID mechanism.

FIGURE 8.

Dose–response curves of nirmatrelvir with the WT, A191E, S1D and double mutant (DM). (a) Dose–response curves of nirmatrelvir with WT (black), A191E (pale blue), S1D (green), DM (pink) 3CL^pro at 15 nM enzyme concentration. The activity of the enzyme in the absence of inhibitor is set as 100% for each individual enzyme. (b) Relative dose‐response curves of nirmatrelvir with S1D (green) and DM (pink) 3CL^pro compared to the WT (black) at 15 nM enzyme concentration. The pink region represents the inhibition concentration where the activity of S1D surpasses that of the WT. The blue region represents the inhibition concentration where the activity of DM surpasses that of the WT. The activity of the WT enzyme in the absence of inhibitor is set as 100%; the activity of S1D and DM in the absence of inhibitor is 5% and 10% that of the WT, respectively. All measurements are in triplicate. The error bars represent standard deviations of the data points.

TABLE 4.

Measured K _i, K _D, and f _c values of nirmatrelvir for the WT, S1D, A191E, and DM 3CL^pro, with all values expressed as mean ± SD.

Enzyme	K i (nM)	K D (nM)	f c	K i f c (nM)	K D f c (nM)	K D/f c (nM)
WT	1.7 ± 0.2 a	‐	‐	‐	‐	‐
S1D	1.4 ± 0.1 b	220 ± 50 b	37 ± 2 b	53 ± 3 b	8000 ± 2000 b	6 ± 1 b
A191E	2.3 ± 0.1 a	‐	‐	‐	‐	‐
S1D + A191E	6.3 ± 0.4 b	100 ± 20 b	20 ± 5 b	130 ± 40 b	2000 ± 700 b	5 ± 2 b

^a Morrison equation.

^b Ligand‐induced dimerization model.

2.7. Effects of mutations on enzyme kinetics

We characterized the catalytic parameters of the WT, S1D, A191E, and DM 3CL^pro variants, measuring activity as a function of substrate concentration (Figure S5). Data for both the WT and A191E had a hyperbolic shape and were well fit by the Michaelis–Menten equation. The K _m of A191E (224 ± 29 μM) was slightly lower than that of the WT (360 ± 30 μM), while the catalytic constant, k _cat, was slightly higher (4.0 ± 0.3 vs. 2.1 ± 0.1 s^−s), leading to a roughly three‐fold increase in enzyme efficiency (k _cat/K _m = 1.8 ± 0.3 × 10⁴ M⁻¹ s⁻¹ and 5.8 ± 0.6 × 10³ M⁻¹ s⁻¹ for A191E and WT, respectively). On the other hand, both the S1D mutant and the DM could not be saturated even at the highest substrate concentration of 400 μM, implying a large increase in the apparent K _m. Interestingly, substrate‐induced dimerization, that is, LID, with a value of f _c on the order of what we observed experimentally for the inhibitors would produce a very similar increase in the apparent K _m as we observed here. The idea that the substrate induces dimerization of the mutants is supported by the observation that at close to 400 μM substrate, the S1D and DM variants are 50% as active as the WT, while the apparent dissociation constant of S1D, measured at 12 μM substrate, would predict that only less than 2% of the enzyme should be in the active, dimeric form. Nevertheless, we cannot rule out the possibility that these mutations also change the intrinsic substrate affinity of the dimeric forms. In either case, the LID experiments in this study were performed with substrate concentrations of only about 5% of the WT K _m, so that any substrate‐induced dimerization can be neglected in the analysis of inhibitor‐induced dimerization.

2.8. G11S mutant with disrupted dimer interface does not exhibit LID

In order to better understand the relationship between disruption of the dimer interface and LID, we studied SARS‐CoV‐2 3CL^pro G11S, a naturally occurring mutant with a highly destabilized dimer interface (K _D ^app of 10,000 ± 2000 nM) (Wang et al., 2024). G11S was largely monomeric (and inactive) at the enzyme concentrations used in this study, and any LID due to lufotrelvir would be expected to produce a large peak at an intermediate inhibitor concentration. However, dose–response curve for G11S did not exhibit LID, as shown in Figure 9b, and fitting to Scheme 1 yielded a low value for f _c = 1.8, similar to that of the WT and an order of magnitude less than we measured for S1D and DM. Thus, destabilization of dimerization and enhancement of LID are separate effects in 3CL^pro, such that a single mutation can disrupt the interface without greatly affecting coupling between the interface and the active site.

2.9. Implications of LID on drug resistance

Given that LID was demonstrated to be an enzyme activation mechanism (Tomar et al., 2015), it could act as a mechanism of drug resistance. Our model of the process has allowed us to investigate quantitatively how modulating the strengths of dimerization, inhibitor binding, and allosteric coupling can enhance enzyme activity in the presence of inhibitors. Figure 10a shows a series of curves simulated for a range of coupling strengths (f _c), while the inhibitor affinity of the dimer (K _i) and the strength of the M:MI interface (K _D) were held fixed. The simulated enzyme concentration was lower than the K _D, meaning that, in the absence of inhibitor, the enzyme was largely (≈85%) monomeric. With f _c = 1, there was no allosteric coupling between inhibitor binding and dimerization, and a standard sigmoidal dose–response curve was obtained. For increasing coupling strengths (increasing f _c), the simulated activity of the enzyme was decreasing, due to the increasing dissociation constant of the M:M interface (K _D f _c). At the same time, the curves increasingly showed the peaked profiles associated with LID, along with higher activity at high inhibitor concentrations, that is, resistance. Figure 10b shows a similar scenario, except in this case the K _D of the enzyme was 100‐fold tighter, meaning that the enzyme was initially only about 20% monomeric. In this case, increasing the value of f _c did not produce the peaked curves characteristic of LID, except at the highest values of f _c, and the activity of the enzyme was not enhanced at high inhibitor concentrations. In Figure 10c, the value of f _c was held fixed, and the value of K _D was increased. The dose–response curves became more peaked and LID‐like at larger values of K _D, where the enzyme was more monomeric. However, in this case, increasing the value of K _D did not enhance the enzyme's activity at any inhibitor concentration. Finally, in Figure 10d, the values of f _c and k _d were held fixed, and K _i was varied. Increasing the value of K _i did not change the shape of the dose–response curve but shifted it to higher inhibitor concentrations, as expected. In summary, increasing the LID coupling strength (f _c) alone is sufficient to produce inhibitor resistance. However, this is only true under conditions where the enzyme must dimerize to become active but is initially largely monomeric. No enhancement is seen when the enzyme is already mostly dimeric in the absence of inhibitor. Furthermore, simply destabilizing the dimer interface (increasing K _D) does not produce any inhibitor resistance. In contrast, reducing the inhibitor binding strength (increasing K _i) also produces resistance in a mechanism that is wholly distinct from altering the coupling strength.

FIGURE 10.

Simulations to show the effects of ligand‐induced dimerization on the dose–response curves under various conditions. All simulations were performed with [E]_t = 10 nM. The activity of the fully dimerized enzyme is set to 100%. (a) K _D = 100 nM, K _i = 50 nM, f _c starts from 1 (deep purple) and increases two‐fold till it reaches 128 (light purple). (b) K _D = 1 nM, K _i = 50 nM, f _c starts from 1 (deep purple) and increases two‐fold till it reaches 128 (light purple). (c) K _i = 50 nM, f _c = 32, K _D starts from 1 nM (deep blue) and increases two‐fold till it reaches 256 nM (light blue). (d) K _D = 100 nM, f _c = 32, K _i starts from 12.5 nM (deep red) and increases two‐fold till it reaches 800 nM (light red).

3. DISCUSSION

Enzyme activation via substrate‐ and inhibitor‐induced dimerization is a relatively common phenomenon (Atkinson et al., 2018; Cheng et al., ²⁰¹⁰; Datta et al., ²⁰¹³; Ivanov et al., ²⁰¹²; Louis et al., ²⁰¹¹; Patel et al., ²⁰¹⁶; Tomar et al., ²⁰¹⁵; Xie et al., ¹⁹⁹⁹) that has been linked to functional regulation, including sensing enzyme concentration (Datta et al., 2013). LID has been reported for 3CL^pro from MERS (Tomar et al., 2015) and SARS‐CoV‐2 (Paciaroni et al., 2023) and 3CL^pro mutants from SARS‐CoV (Cheng et al., 2010). In the case of SARS‐CoV‐2 3CL^pro, saturation with nirmatrelvir increased the strength of dimerization about seven‐fold (Paciaroni et al., 2023). This ratio corresponds to f _c ² and is similar to the value of f _c ² we measured here for lufotrelvir and the WT enzyme (2.9‐ to 3.2‐fold). The f _c ² values for S1D and DM are more than 400‐fold greater. Thus, allosteric pathways linking the active site to the dimer interface likely have evolved in many enzymes, including the 3CL^pro from several coronaviruses. Our findings suggest that simply shifting the energetic balance of this allosteric coupling can be enough to generate inhibitor‐resistant enzyme variants. It is worth noting that several inhibitor‐resistant variants of human immunodeficiency virus (HIV) protease exhibit pronounced LID (Louis et al., 2011; Xie et al., ¹⁹⁹⁹), raising the possibility that resistance has arisen at least partly via this mechanism.

In Scheme 1, this shift is described by an increase in f _c; however, the model is deliberately general and does not speculate as to the physical nature of an increase in coupling strength. Two of the most studied allosteric paradigms used to explain coupling are those due to Monod–Wyman–Changeux (MWC) (Monod et al., 1965) and Koshland–Nemethy–Filmer (KNF) (Koshland Jr. et al., 1966). In the MWC model, conformational changes are synchronized such that monomers are either in an active form, capable of both dimerization and ligand binding, or in an inactive one, capable of neither. In this mechanism, f _c is the equilibrium constant between free, monomeric, inactive, and active forms, f _c = [inactive]/[active] (Figure S7). Any mutation that shifts the conformational equilibrium toward the inactive form could potentially generate inhibitor resistance. In the KNF model, both free and bound monomers can dimerize, but there is an energetic penalty of RTln(f _c) for a monomer that is dimerized but not bound, and vice versa (Figure S8). Likewise, any mutation that enhances this penalty could generate inhibitor resistance.

Structural studies of 3CL^pro variants offer some clues into the nature of allosteric coupling in this enzyme. Several monomeric 3CL^pro mutants have an active site that is collapsed due to the formation of a short 3₁₀‐helix in the S139‐F140‐L141 loop, which may render the monomer incapable of binding substrate or inhibitor (Chen, Hu, et al., 2008; Li et al., ²⁰¹⁶; Shi et al., ²⁰⁰⁸). One of these enzymes is G11A 3CL^pro from SARS‐CoV, which is exclusively monomeric and catalytically inactive (Chen, Hu, et al., 2008). However, SARS‐CoV‐2 G11S 3CL^pro does not exhibit LID, even though it is also predominantly monomeric (Wang et al., 2024) and might be expected to show the same active site collapse as G11A. This suggests that allosteric coupling between the active site and dimer interface may be more complicated than simply forming this 3₁₀‐helix in the monomeric state. In addition, several studies of R298A SARS‐CoV 3CL^pro showed that, in the absence of bound ligand, domain III undergoes a 33° rotation that creates a steric clash between protomers and prohibits dimer formation (Shi et al., 2008; Wu et al., ²⁰¹³). In the presence of ligand, domain III readopts to the correct orientation and the enzyme dimerizes. We hypothesize that ligand‐induced reorientation of domain III may be the mechanism that links the active site and dimer interface and produces LID in SARS‐CoV‐2 3CL^pro WT, S1D, and DM. Interestingly, G11A SARS‐CoV 3CL^pro does not exhibit this rotation of domain III in the crystal structure, which might explain why the SARS‐CoV‐2 G11S 3CL^pro does not exhibit LID (Chen, Hu, et al., 2008).

In SARS‐CoV 3CL^pro, E166 acts as a link between the substrate binding site and the dimer interface, and the E166A mutation eliminates the substrate‐induced dimerization of the R298A mutant (Cheng et al., 2010). In MERS‐CoV 3CL^pro, the homologous E169A mutation shifted the equilibrium toward the monomer, even in the presence of substrate (Ho et al., 2015). In SARS‐CoV‐2 3CL^pro, it was proposed that nirmatrelvir stabilized the dimer via a hydrogen bond network involving the inhibitor and E166 and F140 of one monomer and S1 of the other monomer (Paciaroni et al., 2023). Similar interactions are observed between lufotrelvir and 3CL^pro (Figure 2b). This hydrogen‐binding network that includes S1 may help to explain why the S1D mutation has such a profound effect on the strength of LID in SARS‐CoV‐2 3CL^pro.

Most resistance mutations on 3CL^pro directly affect the binding of the inhibitor by changing K _i. These mutations occur near the binding pocket, including mutations on S144, M165, E166, H172, F170, Q192, among many others (FDA, ²⁰²³; Hu et al., ²⁰²³; Iketani et al., ²⁰²³; Ip et al., ²⁰²³). Most notably, the E166V mutation not only reduces binding affinity to nirmatrelvir but also increases the kinetic barrier required for forming the covalent complex, resulting in more than 7000‐fold increase in K _i and up to 255‐fold increase in EC₅₀ (FDA, ²⁰²³; Ramos‐Guzmán et al., ²⁰²³). Other resistance mutations such as T21I, L50F, P252L, and T304I are distal to the binding pocket. These mutations themselves do not confer high level of resistance alone. Instead, they are often part of a combination of mutations that lead to significant resistance. LID‐promoting mutations could potentially work in tandem with these and other mutations to produce substantial antiviral resistance. That said, the S1D and A191E mutations we identified in this study are not common in the global initiative on sharing all influenza data (GISAID) database of SARS‐CoV‐2 patient‐derived sequences (Khare et al., 2021). The most common substitution at S1 is G, making up to 65% of all mutations at this position, while no record of S1D mutation has been reported to date. Since the S1D mutation requires two nucleotide substitutions, a possible pathway to achieve this mutation would be through the S1G mutation to G1D mutation. It should also be noted that even though S1D mutation is detrimental to the enzyme activity, it does not necessarily render the virus incapable of replication, as it has been shown that the virus carrying the G11S mutant is viable and detected in patient sequencing samples (Wang et al., 2024), and that the WT MERS‐CoV 3CL^pro is a weakly binding dimer (K _D = 7.8 μM) that mainly exists in the inactive monomer form (Tomar et al., 2015). As for substitutions of A191, the most common mutation is A191V, making up to 90% of the mutations at this position. A191E, on the other hand, accounts for only 0.06% of mutations at this site.

4. CONCLUSION

Using a combination of MD simulations and enzymatic assays, we have identified two mutations in SARS‐CoV‐2 3CL^pro, S1D and A191E, that decrease the binding affinity of lufotrelvir, one of the most potent 3CL^pro inhibitors to date, and nirmatrelvir, the active component of Paxlovid. Combining these in an S1D/A191E DM led to roughly additive inhibitor resistance. Dose–response curves for S1D and DM showed peaks in activity at intermediate inhibitor concentrations, which are indicative of ligand‐induced dimerization. We developed a general, numerical model of LID that allowed us to quantify the effects of the mutations on this process and clarify how it relates to inhibitor resistance. We found that the S1D mutation increases the strength of coupling between binding in the active site and dimerization by roughly 12‐fold, compared to the WT, in addition to slightly increasing the value of K _i. In parallel, we also directly observed ligand‐induced dimerization in nanoESI‐MS, which demonstrated good agreement with our LID model predictions. In general, increasing the strength of LID leads to reduced inhibitor sensitivity at high concentrations, under conditions where the enzyme is initially mostly monomeric. Since LID is a relatively common enzyme regulatory mechanism, we postulate that LID‐enhancing mutations may be an important source of inhibitor resistance that is orthogonal to mutations that affect inhibitor binding directly.

5. MATERIALS AND METHODS

5.1. In silico preparation of the enzyme ligand complexes

The 3CL^pro systems in complex with the inhibitor PF‐00835231 (V2M) were prepared from the crystal structures 6XHM at 1.41 Å resolution. The 3CL^pro/substrate complex was prepared using the 11‐residue substrate (TSAVLQSGFRK) coordinates obtained from the SARS‐CoV 3CL^pro/substrate complex (2Q6G) and the SARS‐CoV2 3CL^pro apoprotein coordinates (6YB7) (Xue et al., 2008). Point mutations were generated with the tleap program, modifying the atom type in each case to match the parameter file. The protonation states of the ionizable residues at pH 7.0 were established with the H++ web interface based on calculations of the pK values of the ionizable groups (Gordon et al., 2005). The polypeptide chains are parameterized with Amber's ff19SB force field. Inhibitors were parameterized with the generalized amber force field and the restrained electrostatic potential (REsP) charges were recalculated by quantum mechanics in the Gaussian16 program using a B3LYP functional and a 6‐31g* basis (Frisch et al., 2016; Salomon‐Ferrer et al., ²⁰¹³). Finally, the systems were hydrated in a TIP3P water and ion model box to reach charge neutrality.

5.2. Molecular dynamics simulation

MD simulations were performed with AMBER20‐GPU using the following MD protocol (Osorio et al., 2022; Yañez et al., ²⁰²¹): (i) minimization and structural relaxation of water molecules with 2000 minimization steps and MD simulation with an isothermal‐isobaric (NPT) assembly (300 K) for 1000 ps using harmonic constraints of 10 kcal molÅ⁻² for protein and ligand; (ii) full structure minimization considering 6500 steps of conjugate gradient minimization; (iii) the minimized systems were progressively heated up to 300 K, with harmonic constraints of 10 kcal molÅ⁻² for carbonate skeleton and ligand for 0.5 ns; (iv) the system was then equilibrated for 0.5 ns maintaining the constraints and then for 5 ns without constraints at 300 K in a canonical assembly (NVT); (v) finally, 200 ns simulations were performed for each system (Purawat et al., 2017; Salomon‐Ferrer et al., ²⁰¹³). During the MD simulations, the equations of motion were integrated with a time step of 2 fs in the NPT assembly at a pressure of 1 atm. The SHAKE algorithm was applied to all hydrogen atoms, and the van der Waals limit was set to 12 Å (Forester & Smith, 1998). The temperature was maintained at 300 K, employing the Langevin thermostat method with a relaxation time of 1 ps. The Berendsen barostat was used to control the pressure at 1 atm. Long‐range electrostatic forces were accounted for using the particle‐mesh Ewald (PME) approach (Harvey & De Fabritiis, 2009). Data were collected every 1 ps during MD tests. Molecular visualization of the systems and MD trajectory analysis were carried out with the visual molecular dynamics (VMD) software package (Humphrey et al., 1996). To assess the reproducibility of the results, three replicates were performed using ig = 40, 100, and 1000, which defines the seed for random number generation and controls the initial assignment of velocities, as well as the thermostat and barostat settings in the simulation. Results are shown in Figures S1–S3.

5.2.1. Cluster analysis

This statistical methodology separates the data points into several groups that exhibit similar properties and differ from the other groups. To perform the clustering, the density‐based spatial method of applications with noise (DBSCAN) implemented in the CPPTRAJ tool was used (Schubert et al., 2017). This algorithm performs the separation by considering a cluster in the data space as a contiguous region of high point density, separated from other similar clusters by contiguous regions of low point density. Each analysis was performed with a cutoff distance of 1.5 Å based on the root‐mean‐square (rmsd) of the ligand's distinct hydrogen atoms and 5 points as a minimum for each cluster. According to this calculation, the representative structures of each simulation were obtained, in relation to the position of the ligand in the protein.

5.3. Enzyme production

Plasmids carrying WT and SARS‐CoV‐2 3CL^pro mutants were purchased from GenScript (NJ, USA). These proteases were produced in E. coli BL21 (DE3) according to previously published procedures (Stille et al., 2022; Zhang et al., ²⁰²⁰). The protein sequence was preceded by an N‐terminal glutathione S‐transferase  sequence followed by a 3CL^pro cleavage site (SAVLQ↓SGFRK). Autocleavage activity of 3CL^pro leaves its authentic N‐terminus starting with serine. The C‐terminal glutamine was followed by a GPHHHHHH 6‐His purification tag (Xue et al., 2007; Zhang et al., ²⁰²⁰). Proteins were purified using a 5 mL HisTrap FF column (GE Healthcare, IL, USA) on an ÄKTA Avant system (Cytiva), applying a gradient between buffer A (50 mM Tris–HCl pH 8.0, 150 mM NaCl, 20 mM imidazole) and buffer B (50 mM Tris–HCl pH 8.0, 150 mM NaCl, 500 mM imidazole). Eluted fractions were pooled, buffer exchanged (50 mM Tris–HCl pH 8.0, 150 mM NaCl), and concentrated using Amicon® Ultra centrifugal filters with a 10 kDa cutoff (Merck Millipore, MA, USA) to final concentrations of 1–5 mg mL⁻¹. Proteins were stored at −80°C until use.

5.4. Assay setup

The enzymatic activity of 3CL^pro was measured using the fluorescent substrate DABCYL‐KTSAVLQ↓SGFRKME‐EDANS from CanPeptide (Montréal, Québec, Canada) (Mesel‐Lemoine et al., 2012). The rate of enzymatic activity was determined at 25°C by following the increase in fluorescence (excitation at 360 nm and emission at 460 nm) due to peptide hydrolysis as a function of time. Assays were conducted in Nunc™ F96 MicroWell™ black plates (Thermo Scientific, Waltham, MA, USA) in assay buffer (20 mM 4‐(2‐hydroxyethyl)‐1‐piperazineethanesulfonic acid, 10 mM NaCl, 1 mM ethylenediaminetetraacectic acid, 1 mM dithiothreitol, and 0.1% bovine serum albumin, pH 7.5) with a final reaction volume of 50 μL. The resulting fluorescence was monitored using a BioTek Synergy™ H4 Hybrid Multi‐Mode Microplate Reader (Winooski, VT, USA). The rate of the reaction in relative fluorescent units per second (RFU s⁻¹) was determined by measuring the initial slope of the progress curves. All reactions were carried out in triplicate.

5.5. K _i measurement

Reactions were performed in 100 μL volume. The final concentration of the enzyme was 15 nM, and the final concentration of the substrate was 12 μM. Fifty‐six microliters of enzyme (diluted in the assay buffer) were incubated for 30 min at room temperature with 4 μL of compound (PF‐00835231 or PF‐07321332, Selleck Chemicals, TX, USA) diluted in dimethyl sulfoxide (DMSO) (Millipore, MA, USA). To initiate the reaction, 40 μL of substrate was added, and the reaction was monitored by following the fluorescence as a function of time. Controls were (i) negative control: no inhibitor dissolved in DMSO, (ii) positive control: 4 μL of 500 mM GC376 (BPS Bioscience) in DMSO, and (iii) blank control: no enzyme was added in the 56 μL assay buffer, no inhibitor dissolved in DMSO. The reactions ran for 2 h, and the linear initial slopes of the progress curves were used to calculate the reaction initial velocity in RFU in time (RFU s⁻¹). K _i values were fit to the Morrison equation using MATLAB (MathWorks, Inc., Natick, MA, USA).

5.6. Measurement of disassociation constant

The dependence of the enzymatic activity on the total enzyme concentration was determined using the FRET‐based assay described above with the previously described method (Wang et al., 2024). The final enzyme concentrations were varied over a range from 20 nM to 5.12 μM for S1D 3CL^pro. Reactions were initiated by the addition of the substrate (final concentration: 12 μM) to the wells containing enzyme in the assay buffer. Initial rates in RFU s⁻¹ were converted to μM s⁻¹ using the calibration curve. The rate of the reaction was plotted as a function of the enzyme concentration and fitted in MATLAB (MathWorks, Inc., Natick, MA, USA) to yield K _D using fminsearch as the fitting algorithm and Monte‐Carlo sampling for generating the standard deviations of the fitted parameters.

5.7. Michaelis–Menten kinetics

The rates of enzymatic activity were measured as a function of varying substrate concentrations at a fixed enzyme concentration. Reactions were initiated by adding the enzyme to the wells containing varying concentrations of substrate (final concentrations: from 3.125 to 200 μM). The final concentration of 3CL^pro was 400 nM. The rate of the reaction was plotted as a function of the substrate concentration and fitted to the Michaelis–Menten equation in MATLAB using fminsearch as the fitting algorithm.

5.8. Ligand‐induced dimerization

Reactions were performed in 100 μL volume. The final concentration of the enzyme was 15 nM or 60 nM, and the final concentration of the substrate was 12 μM. Fifty‐eight microliters of enzyme (diluted in the assay buffer) were incubated for 30 min at room temperature with 2 μL of compound diluted in DMSO (MilliporeSigma, MA, USA). To initiate the reaction, 40 μL of substrate was added, and the reaction was monitored by following the fluorescence as a function of time. Controls were (i) negative control: no inhibitor dissolved in DMSO and (ii) blank control: no enzyme was added in the 28 μL assay buffer, no inhibitor dissolved in DMSO. The reactions ran for 2 h, and the linear initial slopes of the progress curves were used to calculate the reaction initial velocity in RFU in time (RFU s⁻¹).

To fit the experimental data presented in Figure 3b, the percentage activity measured in the FRET assay can be expressed as:

$$\%\text{Activity}=100\frac{\left[\mathrm{D}\right]+0.5\left[\mathrm{DI}\right]}{{\left[\mathrm{D}\right]}_{\mathrm{CTR}}}.$$ (3)

The singly occupied dimer, DI, is only half as active as the free dimer, D, as DI only has one site available for catalysis and is statistically 50% less active than D. [D]_CTR is the concentration of the dimer in the control experiment without inhibitor, which is expressed as (Wang et al., 2024):

$${\left[\mathrm{D}\right]}_{\mathrm{CTR}}=\frac{4{\left[\mathrm{E}\right]}_{\mathrm{t}}+K_{\mathrm{D}}{f}_{\mathrm{c}}-\sqrt{8{\left[\mathrm{E}\right]}_{\mathrm{t}}{K}_{\mathrm{D}}{f}_{\mathrm{c}}+{\left(K_{\mathrm{D}}{f}_{\mathrm{c}}\right)}^2}}{8}.$$ (4)

Dimerization of a monomer and a monomer‐inhibitor complex (M + MI → DI) is controlled by the dissociation constant K _D, where:

$$K_{\mathrm{D}}=\frac{\left[\mathrm{M}\right]\left[\mathrm{MI}\right]}{\left[\mathrm{D}\right]}.$$ (5)

And dimerization of two monomers (2M → D) is defined as:

$$K_{\mathrm{D}}{f}_{\mathrm{c}}=\frac{{\left[\mathrm{M}\right]}^2}{\left[\mathrm{D}\right]}.$$ (6)

Similarly, the inhibition constant, K _i, between the dimer and the ligand (D + I → DI) or between the singly occupied dimer and the ligand (DI + I → DI₂) is defined as:

$$K_{\mathrm{i}}=\frac{\left[\mathrm{D}\right]{\left[\mathrm{I}\right]}_{\mathrm{f}}}{\left[\mathrm{DI}\right]}=\frac{\left[\mathrm{DI}\right]{\left[\mathrm{I}\right]}_{\mathrm{f}}}{\left[\mathrm{D}{\mathrm{I}}_2\right]}.$$ (7)

where [I]_f is the concentration of the free inhibitor in the assay. Binding of the monomer to the inhibitor (M + I → MI) is controlled by:

$$K_{\mathrm{i}}{f}_{\mathrm{c}}=\frac{\left[\mathrm{M}\right]{\left[\mathrm{I}\right]}_{\mathrm{f}}}{\left[\mathrm{MI}\right]}.$$ (8)

Consequently, the total enzyme concentration can be calculated in terms of K _i , K _D, f _c, [M], and [I]_f according to:

$$\begin{array}{rcl}{\left[\mathrm{E}\right]}_{\mathrm{t}}^{\text{calc}}& =& \left[\mathrm{M}\right]+\left[\mathrm{MI}\right]+2\left(\left[\mathrm{D}\right]+\left[\mathrm{DI}\right]+\left[{\mathrm{DI}}_2\right]\right)\\ & =& \left[\mathrm{M}\right]+\frac{\left[\mathrm{M}\right]{\left[\mathrm{I}\right]}_{\mathrm{f}}}{K_{\mathrm{i}}{f}_{\mathrm{c}}}+2\left(\frac{{\left[\mathrm{M}\right]}^2}{K_{\mathrm{D}}{f}_{\mathrm{c}}}+\frac{{\left[\mathrm{M}\right]}^2{\left[\mathrm{I}\right]}_{\mathrm{f}}}{K_{\mathrm{D}}{K}_{\mathrm{i}}{f}_{\mathrm{c}}}+\frac{{\left[\mathrm{M}\right]}^2{\left[\mathrm{I}\right]}_{\mathrm{f}}^2}{K_{\mathrm{D}}{K}_{\mathrm{i}}^2{f}_{\mathrm{c}}}\right).\end{array}$$ (9)

Similarly, the total inhibitor concentration can be expressed as:

$${\left[\mathrm{I}\right]}_{\mathrm{t}}^{\text{calc}}=\begin{array}{l}{\left[\mathrm{I}\right]}_{\mathrm{f}}+\left[\mathrm{MI}\right]+\left[\mathrm{DI}\right]+2\left[{\mathrm{DI}}_2\right]={\left[\mathrm{I}\right]}_{\mathrm{f}}+\frac{\left[\mathrm{M}\right]{\left[\mathrm{I}\right]}_{\mathrm{f}}}{K_{\mathrm{i}}{f}_{\mathrm{c}}}\\ +\frac{{\left[\mathrm{M}\right]}^2{\left[\mathrm{I}\right]}_{\mathrm{f}}}{K_{\mathrm{D}}{K}_{\mathrm{i}}{f}_{\mathrm{c}}}+2\frac{{\left[\mathrm{M}\right]}^2{\left[\mathrm{I}\right]}_{\mathrm{f}}^2}{K_{\mathrm{D}}{K}_{\mathrm{i}}^2{f}_{\mathrm{c}}}.\end{array}$$ (10)

For any given set of K _i, K _D, and f _c parameters, the values of [I]_f and [M] were varied in order to minimize the target function:

$$\mathrm{RSS}={\left({\left[\mathrm{E}\right]}_{\mathrm{t}}^{\text{calc}}-{\left[\mathrm{E}\right]}_{\mathrm{t}}^{\text{true}}\right)}^2+{\left({\left[\mathrm{I}\right]}_{\mathrm{t}}^{\text{calc}}-{\left[\mathrm{I}\right]}_{\mathrm{t}}^{\text{true}}\right)}^2$$ (11)

using the fminsearch function in MATLAB and standard deviations of the fitted parameters were generated by Monte‐Carlo sampling. The resulting values of [I]_f and [M] were then used to calculate the concentrations of D and DI and overall enzyme activity. The values of K _i, K _D, and f _c were then varied until the calculated enzyme activities matched the experimental dose–response curves (for the details of this numerical solution, please see the MATLAB published on GitHub at: https://github.com/GuanyuWang-McGill/Ligand_Induced_Dimerization).

5.9. nanoESI‐MS

Immediately prior to native MS analysis, 3CL^pro proteins were buffer exchanged into 200 mM ammonium acetate using Micro Bio‐Spin 6 columns (Bio‐Rad, CA, USA) and the pH was adjusted to 7.5 using ammonium hydroxide. Following buffer exchange, the concentrations of the protein samples were determined using UV absorption and were diluted to 2 μM in 200 mM ammonium acetate. Protein samples (5–10 μL) were loaded into gold‐coated, nanospray emitters that were prepared in‐house from borosilicate capillaries (1.0 mm OD × 0.78 mm ID × 100 mm L, Harvard Apparatus). The capillaries were pulled to have a 1 μm orifice (estimated by light microscopy) using a Sutter Instruments Model P‐2000 tip puller (parameters: heat = 350, filament = 4, velocity = 60, delay = 255, pull = 0). The loaded emitter was briefly centrifuged to move the fluid to the end of the tip and was then mounted onto the nanospray ESI source of a Synapt G2‐Si mass spectrometer (Waters). Data were acquired in positive ion and sensitivity modes over an m/z range of 100–8000 using a capillary voltage = 1.5 kV, cone voltage = 50 V, source offset = 50 V, source temperature = 50°C, trap collision energy = 4 V, trap gas (argon) flow = 2 mL min⁻¹, and trap direct current bias = 35 V. Mass spectra were smoothed (Savitsky‐Golay, smooth window = 10, number of smooths = 3) and deconvoluted (Maxent 1, resolution = 1.00, uniform Gaussian, width at half height = 25 Da) using Mass Lynx v4.2 software (Waters).

AUTHOR CONTRIBUTIONS

Guanyu Wang: Conceptualization; writing – original draft; methodology; formal analysis; data curation; visualization; writing – review and editing; resources; investigation. Felipe Venegas: Resources; investigation; writing – review and editing. Andres Rueda: Investigation; resources; writing – review and editing. Osvaldo Yañez: Investigation; formal analysis; data curation; writing – review and editing; visualization. Manuel I. Osorio: Investigation; formal analysis; data curation; writing – review and editing; visualization. Sibei Qin: Data curation; writing – review and editing. José Manuel Pérez‐Donoso: Project administration; writing – review and editing. Christopher J. Thibodeaux: Formal analysis; data curation; writing – review and editing; visualization. Nicolas Moitessier: Project administration; writing – review and editing; funding acquisition. Anthony K. Mittermaier: Funding acquisition; writing – review and editing; project administration; supervision.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

Data S1. Supporting Information.

Wang G, Venegas F, Rueda A, Yañez O, Osorio MI, Qin S, et al. Inhibitor‐induced dimerization mediates lufotrelvir resistance in mutants of SARS‐CoV‐2 3C‐like protease . Protein Science. 2025;34(9):e70275. 10.1002/pro.70275

DATA AVAILABILITY STATEMENT

The data that supports the findings of this study are available in the supplementary material of this article.