Linking biochemical and cellular efficacy of MERS coronavirus main protease inhibitors

CompoundsthatbindtotheMiddleEastRespiratorySyndromeCoronavirus(MERS- CoV)mainprotease(MPro)oftenproducebiphasicconcentration-responsecurves(CRCs) in biochemical assays; low concentrations activate the enzyme and high concentrations inhibit it. This biphasic behavior complicates data analysis. Here, we compare three approaches to data analysis: fitting the Hill equation to the activation phase, fitting it 1 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint to the inhibition phase, and fitting an enzyme kinetics model that incorporates dimer- ization and ligand binding to the complete CRC. In the latter case, cellular efficacy is predicted by extrapolating the model to high enzyme concentrations. For compounds in our drug lead series, all three procedures yield inhibitory concentrations that are correlated with live-virus antiviral assays. The latter procedure provides the most ac- curate forecast of cellular efficacy rank. These data analysis procedures may be valuable for antiviral drug discovery against MERS-CoV MPro and other enzymes with similar kinetics. 1 Introduction The Middle East Respiratory Syndrome Coronavirus (MERS-CoV) is a serious threat to global health. The virus was first identified in Saudi Arabia in 20121 and has caused spo- radic outbreaks, predominantly in the Middle East, Africa, and South Asia. According to the WHO, no vaccine or antiviral treatment has been approved for MERS-CoV.2 The virus has evolved between 2015 and 20193 and further evolution could produce increased trans- missibility. Given this possibility and the alarmingly high fatality rate of 35%,4 MERS-CoV could lead to large-scale mortality. Many drug discovery efforts for coronaviruses have focused on identifying compounds that inhibit the main protease (MPro).5–11MPro is essential to the life cycle of coronaviruses. It is one of 16 non-structural proteins produced upon viral entry into host cells, forming part of the replicase-transcriptase complex responsible for genomic RNA replication and subgenomic mRNA synthesis.12,13 Inhibiting MPro disrupts the viral replication cycle,5,9,14 facilitating its clearance by the immune system. Indrugdiscoverycampaignsfocusingonenzymeinhibitors, concentrationresponsecurves (CRCs) that measure the progress of a catalyzed reaction as a function of inhibitor concen- tration can be a key part of the assay cascade. Improving potency of enzymatic inhibition is one of the most direct objectives of structure-based drug design. While it is possible to forego 2 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint an enzyme inhibition assay and direct test inhibitors in a cell-based antiviral activity assay, the former generally has fewer safety risks, is less expensive, and is less subject to biological variability. Moreover, cell-based assays can introduce confounding factors, such as membrane permeability and active efflux pumps, that can confuse structure-activity relationships. Unfortunately, MERS-CoV MPro inhibition assays often show biphasic behavior that complicates their interpretation. MPro is most active as a dimer,15 but analytical ultracen- trifugation shows that its dissociation constant (Kd) is 52 µM,16 weaker than MPro from SARS-CoV (6 µM)17 and SARS-CoV-2 (7 µM).18 Due to the high fraction of enzyme in the relatively inactive monomeric form, ligand-induced dimerization16,19 produces biphasic CRCs, also known as activation-inhibition CRCs, in biochemical assays performed at low enzyme concentrations.16 Ligand binding to one monomer can trigger dimerization, locking the catalytic site in an active conformation that stabilizes hydrogen bonding across the dimer interface to the N-terminal serine of the opposite subunit (c.f. Fig. 6 of Nguyen et al.20). If ligand concentrations are low, the other monomer is usually available to bind to substrate and produce product, leading to an overall increase in the catalytic rate. At high ligand concentrations, both monomers are occupied by ligand and enzyme catalysis decreases. For such biphasic curves, the traditional four-parameter Hill equation - which includes bottom response, top response, IC50, and Hill slope - does not fit the complete curve. Thus, it has been unclear how to fit models to these data and how to interpret model parameters for the evaluation of antiviral compounds targeting MERS-CoV MPro and other enzymes that produce biphasic CRCs. Here, weevaluatethreepossibleprocedurestointerpretingthesebiphasicCRCs. Oneisto ignore the activation phase and fit the Hill equation to the inhibition phase. This yields what we will refer to as theinhibition pIC50. In another, the inhibition phase is also extracted, but instead of fitting four parameters, the top response is set by the negative control (no inhibitor) and the three remaining parameters are estimated. This procedure essentially assumes that there is no ligand-induced dimerization. We refer to the pIC50 obtained from 3 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint this fit as thecontrol pIC50. A third procedure is based on fitting an enzyme kinetics model that incorporates both dimerization and ligand binding that we recently introduced.21 This model produces biphasic CRCs that can be fit to the entire curve without ignoring any data. Here, we develop a protocol for fitting the model to a large number of CRCs. (Binding free energies from this fitting procedure have been used as a benchmark in a blinded challenge for the computational chemistry community.22) After fitting the model, we predict CRCs at high enzyme concentrations (reflecting cellular conditions), yielding thedimer pIC50. The three procedures are evaluated based on the correlation between different pIC50s and pEC50s in a live-virus antiviral assay. 2 Methods 2.1 Assays CRCs were measured in both biochemical enzymatic activity and live-virus antiviral as- says, as reported in the AI-driven Structure-enabled Antiviral Platform (ASAP) Discovery Consortium (https://asapdiscovery.org/) protocols.io repository of experimental proto- cols.24 2.1.1 Biochemical enzyme activity Biochemical CRCs were obtained by the MERS-CoV MPro fluorescence dose response for antiviral testing protocol25 and variants with different concentrations of enzyme, substrate, and inhibitor. The protocol is similar to that described for SARS-CoV-2 MPro,23 but ap- plied to MERS-CoV MPro. Two categories of experiments were performed. In the first category, the enzyme concentration was fixed and the response was measured as a function of substrate concentration. Datasets from this category are referred to as enzyme-substrate (ES) datasets, comprising three datasets with enzyme concentrations at 25, 50, and 100 nM, respectively. Substrate concentrations were 50, 150, 350, 550, 750, 950, 1150, and 1350 nM. 4 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint For ES datasets, six replicates were measured at each substrate concentration. In the second category, both enzyme and substrate concentrations were fixed while inhibitor concentra- tions were varied. Datasets from this category are referred to as enzyme-substrate-inhibitor (ESI) datasets. For thirteen inhibitors, CRCs were measured under four conditions (ESI4c): 50 nM enzyme, 150 nM substrate; 100 nM enzyme, 50 nM substrate; 100 nM enzyme, 750 nM substrate; and 100 nM enzyme, 1350 nM substrate. Inhibitor concentrations were 50, 100, 194, 388, 776, 1552, 2488, 7463, 12440, 24880, 49750, and 99500 nM. For 85 inhibitors, CRCs were measured with 50 nM of enzyme and 550 nM of substrate (ESI1c), while the inhibitor concentrations were 0.888, 2, 4, 15, 50, 133, 460, 1227, 2488, 9950, 32340, and 99500 nM. In the ESI datasets, two replicates were measured at each inhibitor concentration. The ESI1c data were obtained by the reported protocol25 and ES and ESI4c experiments were performed analogously, but with different concentrations. Data are provided in Tables ES, ESI4c, and ESI1c of the Supplementary Information. Inhibitors were part of a drug discovery campaign for MPro inhibitors targeting both MERS CoV and SARS-CoV-2 conducted by ASAP. Compounds were synthesized by Enam- ine (Ukraine). Complete CRCs and ASAP identifiers are available in Figure S21 and an Excel spreadsheet in the Supplementary Information. Chemical structures have been deposited to ChEMBL 37, with public release anticipated in late Spring 2026. 2.1.2 Live-virus antiviral assay The Live-virus MERS-CoV Vero-TMPRSS2 with PgP Inhibitor Antiviral Screening Assay26 was performed at the Icahn School of Medicine at Mount Sinai. All assays were performed at Biosafety Level 3 (BSL-3) in the Emerging Pathogens Facility (EPF). Vero-TMPRSS2 cells were seeded in 96-well plates at 2,000 cells per well in 10% growth media supplemented with puromycin the day before the assay and incubated at 37°C and 5% CO2. Two hours before infection, cells were treated with 100µL of a 1 to 3 dilution series of antiviral hits in 2% infection media supplemented with PgP-inhibitor. Dilutions 5 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint were performed using a Tecan D300e (Tecan). Concentrations of antiviral hits were 50% higher than the target concentrations to account for infection volume. DMSO and uninfected controls were also included on each plate. Plates were then transferred to the BSL-3 and appropriate wells were infected with MERS-CoV/EMC/2012 at MOI 0.5 in 50 µL of 2% infection media supplemented with PgP-inhibitor, bringing the dilution series to the target concentrations. Plates were then incubated for 48 hours at 37C 5% CO2. 48 hours post infection, supernatants were removed from the wells and replaced with 100ul of 4% formalin and incubated for 15 minutes. Outer surfaces of the plates were decontaminated; platesweredoublebagged, removedfromthefacilityandlefttofumigatefor 48 hours. Plates were then immunostained using MERS-CoV nucleoprotein (NP) antibody (SinoBiological #40068-RP01) with a DAPI counterstain (Total Cells). Plates were analyzed using a Cytation1 (Biotec). Infectivity was measured by the accumulation of viral N protein (Infected Cells; 488nm). Percent infection was quantified as ((Infected Cells/Total Cells) - Background) * 100, with DMSO control readouts as 100% infection. Data was fit using nonlinear regression and IC50s for each experiment were determined using GraphPad Prism v10.0.0 (San Diego, CA). All data are included in the Supplementary Information. 2.2 Bayesian regression for biochemical enzyme activity We fit our enzyme kinetics model21 to the ES and ESI datasets. A schematic of the model, complete set of equations, and descriptions of numerical solutions to the equations are in- cluded in our previous publication. 6 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint 2.2.1 Parameters The objective of our Bayesian regression was to infer the following parameters, θ ≡ (∆Gd, ∆GS,M , ∆GS,D, ∆GS,DS, ∆GI,M , ∆GI,D , ∆GI,DI , ∆GS,DI , kcat,M S, kcat,DS , kcat,DSI , kcat,DSS , [E]t, αp, σc). (1) The enzyme kinetics model, which is based on a rapid equilibrium assumption, has thermo- dynamic (∆G) and kinetic (kcat) parameters as previously described.21 ∆G are the binding free energies of species including the MPro monomer (M), MPro dimer (D), substrate (S), inhibitor (I): ∆Gd is a free energy of dimerization;∆GS,M is the binding free energy of the substrate to the monomer;∆GS,D is the binding free energy of the substrate to the dimer; ∆GS,DS is the binding free energy of the substrate to the dimer-substrate complex;∆GI,M is the binding free energy of the inhibitor to the monomer;∆GI,D is the binding free en- ergy of the inhibitor to the dimer; ∆GI,DI is the binding free energy of the inhibitor to the dimer-inhibitor complex; and∆GS,DI is the binding free energy of the substrate to the dimer-inhibitor complex. kcat are enzyme velocities:kcat,M S is the velocity of the monomer- substrate complex; kcat,DS is the velocity of the dimer-substrate complex; andkcat,DSS is the velocity of the dimer bound to two substrates. Some thermodynamic and kinetics parameters were treated as global, the same for every dataset, and others local, dependent on the inhibitor. The global parameters were the binding free energy of dimerization∆Gd, binding free energies between the enzyme and the substrate ∆GS, and rate constantskcat,DS and kcat,DSS. The inhibitor-dependent parameters were the binding free energies of the inhibitor binding to the enzyme∆GI, the binding free energy of the substrate binding to the enzyme-inhibitor complex∆GS,DI, and the velocity of the enzyme-substrate-inhibitor complexkcat,DSI. In addition to the thermodynamic and kinetic parameters, our model also uses several localparameters: [E]t, αp, andσc. [E]t isthetrueenzymeconcentration. Trueconcentrations 7 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint may differ from the stated concentrations due to dilution errors or protein degradation. We used one parameter[E]t for each of the three stated monomer concentrations of 25, 50, and 100 nM; [E]25 is the true concentration for the solution with a stated concentration of 25 nM, and analogously for 50 and 100.αp is a scaling factor for all velocities on a given plate p. It accounts for differences in velocity calibration due to factors such as plate material or path length variation, instrument lamp intensity or detector sensitivity fluctuations, and sample variations such as pH or buffer evaporation. There was 1 plate for ES, 4 plates for ESI4c, and 45 plates for ESI1c.σc is the standard error of the velocity, indexed byc, and is assumed to be constant for all points in a CRC.σc was also used in our previous work.21 2.2.2 Likelihood For each CRC, the dataD ∈ {y1, y2, ..., yn} are initial velocities (v, M/min) of the enzymatic reaction. Initial velocities were calculated based on linear regression and normalization to ensure that the same rates are obtained for the same reaction conditions. The total fluorescence response R was assumed to the sum of the response of the fluorescent substrate and product. For each species, the fluorescence response is the concentration of the species (cs for substrate and cp for product) and its molar response (rs for substrate and rp for product), R = csrs + cprp, (2) which has the time derivative, dR dt = rs dcs dt + rp dcp dt . (3) Thus, as substrate is converted into product, the observed slope ism = (rp − rs)v, wherev is the initial velocity of the reaction. For each well, the slopem was determined by measuring the response from the biochemical assay every 2 minutes for 10 minutes after addition of sub- 8 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint strate and performing ordinary least squares linear regression (linalg.lstsq) as implemented in numpy.rs was determined by dividing the intercept by the initial substrate concentration. On the master plate with the ES dataset,rp was calibrated by measuring fluorescence after 21 hours, at which point the substrate was assumed to be completely converted to product. For plates with the ESI4c datasets,rp for each plate was determined by solving a system of linear equations such that at the same enzyme and substrate concentrations, the initial velocity of the plate and the master plate are equal. Velocities in the ESI1c dataset were normalized to a velocity interpolated from the ES and ESI4c datasets. After fitting the veloc- ities from the ES and ESI4c datasets by Bayesian regression, maximum a posteriori (MAP) parameters were used to estimate a reference velocityv0, the velocity at 50 nM enzyme and 550 nM substrate; this condition was measured on all of the plates of the ESI1c dataset. During exploratory data analysis, we identified outliers, many which we attributed to limited solubility at high compound concentrations. Before fitting the model, outliers were removed using az-score test.27 A pooled standard deviation was calculated for each CRC in the ESI datasets as, σ = 1 N − 1 X c X i (yc,i − ¯yc)2 , (4) where yc,i is a measured velocity at a given condition (enzyme, substrate, and inhibitor concentration) and ¯yc is the sample mean of velocities at the condition. Sums are over the measurements and conditions andN is the number of measurements in all conditions (six for ES datasets and two for ESI datasets.) Thez-score was calculated as, zc,i = yc,i − ¯yc σ . (5) An observationyc,i is considered an outlier if the absolute value of its correspondingzc,i score exceeds 2.5. The thresholds of -2.5 and 2.5 correspond to the 2.5th and 97.5th percentiles of the observations in the dataset, respectively. Any outliers present in each CRC were removed 9 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint before fitting. Measurements were assumed to follow a normal distribution centered around model- predicted values (scaled byαp), yn ∼ N (αpy∗ n(θ), σ2). The likelihood of the dataD is given by, p(D|θ) = 1 (2π)N/2σN exp " − 1 2σ2 NX n=1 (yn − αpy∗ n(θ))2 # (6) in which the measurementy∗ n(θ) is a function of all parameters inθ, except forαp. 2.2.3 Prior Assuming that the parameters are independent, the priorp(θ) is a product of the prior for all parameters, p(θ) =Q i p(θi). Based on the reported value ofKd (52 ± 5 µM)16 and the relationship betweenbinding free energyand dissociation constantthrough∆G = −RT ln K, the prior of ∆Gd would follow a normal distribution with a mean of -5.9 and a standard deviation of 0.06. However, to reduce the influence of this prior we chose a larger standard deviation, ∆Gd ∼ Normal(−5.9, 0.3) (kcal/mol). (7) Broad uniform priors were chosen for other binding free energies. The range of∆GS was based onKS between 1 nM and 1 M. The range of∆GI was based onKI between 1 pM and 10 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint 1 M. ∆GS,M ∼ Uniform(−12.4, 0.0) (kcal/mol) ∆GS,D ∼ Uniform(−12.4, 0.0) (kcal/mol) ∆GS,DS ∼ Uniform(−12.4, 0.0) (kcal/mol) ∆GI,M ∼ Uniform(−16.5, 0.0) (kcal/mol) ∆GI,D ∼ Uniform(−16.5, 0.0) (kcal/mol) ∆GI,DI ∼ Uniform(−16.5, 0.0) (kcal/mol) ∆GS,DI ∼ Uniform(−16.5, 0.0) (kcal/mol). (8) Broad uniform priors were also selected for the kinetic parameters. Based on the reported value of kcat (0.2 ± 0.02 min−1),16 we chose, kcat,DS ∼ Uniform(0.0, 5.0). kcat,DSS ∼ Uniform(0.0, 5.0). (9) (10) Due to the biphasic behavior observed in CRCs,16 we set a higher upper limit forkcat,DSI ∼ Uniform(0.0, 10.0). For prior ofαp, uniform distribution was used, αp ∼ Uniform(0.0, 2.0). (11) where p is the index for the plate. Because the uncertainty in concentration due to sample preparation in biochemical assays has been shown to be approximately 10%,28 we used a 11 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint log-normal prior with 10% uncertainty for the enzyme concentration, [E]t ∼ LN (µ = [E]s, σ = 0.1 ∗ [E]s), (12) where [E]s for s ∈ {25, 50, 100} nM is the stated value of the enzyme concentration and[E]t is the true value. The uninformative Jeffreys prior29 was used forσ of each CRC, as in our previous work.21 2.2.4 Sampling from the posterior As the complexity of the model and large amount of data made global fitting computationally prohibitive with our limited computing resources, we divided the fitting process into several steps, leveraging the posterior distribution from one step to limit the prior of the next.30 1. AsimplifiedenzymekineticsmodelwithoutinhibitorwasfittotheESdatasettoobtain ranges of ∆Gd, ∆GS,M, ∆GS,D, ∆GS,DS, kcat,DS, kcat,DSS, and [E]t. As we treated the ES plate as a reference,αp was set to 1. 2. The full enzyme kinetics model was fit to the ES dataset and curves fromeach inhibitor in the ESI4c dataset. The priors of∆Gd and ∆GS were defined based on the minimum and maximum values of these parameters observed in posteriors from step 1. 3. The full enzyme kinetics model was globally fit to the ES dataset and thefull ESI4c dataset. The priors of ∆Gd, ∆GS, and αp were defined based on the minimum and maximum values of these parameters observed in every posterior from step 2. 4. The full enzyme kinetics model was globally fit to the ES dataset, the full ESI4c dataset, and three selected curves from the ESI1c dataset. Priors were defined as in Step 4. The three curves were selected based on criteria described below. 5. When fitting to the ESI1c dataset, global parameters were fixed to the MAP of Step 3 or 4 and local parameters were sampled from the conditional probability. 12 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint The three curves in Step 4 were selected based on using the MAP from Step 3 in Step 5. The model from Step 5 did not fit to these three ESI1c curves. Therefore, we incorporated the data in Step 4 in order to obtain a MAP capable of fitting not only ES and ESI4c data, but also the selected ESI1c data. The No-U-Turn sampler (NUTS) was used to sample from posterior distributions.31 NUTS was run for 10,000 samples in steps 1 and 2. Because we observed that the posteriors were already converged by 1,000 samples in steps 1 and 2, we collected 1,000 equilibrated samples in steps 3 and 4. The equilibration time of all the parameters was detected using automated equilibration detection32 as implemented in pymbar v4.0.3.33,34 2.3 Estimating pIC50s and pIC90s Inhibitory concentrations were estimated by fitting data with the Hill equation, yi(Ci, Rb, Rt, pIC50, H) = Rb + Rt − Rb 1 + 10(pIC 50−pCi)∗H , (13) where Rb is the bottom response, Rt is the top response, pIC 50 is the negative base 10 logarithm of the half maximal inhibitory concentrationIC 50, and H is the Hill slope. As outlined at the end of the introduction, we estimated inhibitory concentrations based on three types of CRCs. For the inhibition pIC50, we first identified the concentration of inhibitor that yields the maximum response. The Hill equation was fit to data at this and higher concentrations. For the control pIC50, the Hill equation was fit to data at this and higher concentrations where the response is less than the negative control (Figure 1). For the dimer pIC50, we simulated CRCs using the dimer-only kinetic model and fit the Hill equation to the entire curve.21 Enzyme and substrate concentrations were drawn from lognormal distributions with a 10% uncertainty. The substrate concentration was selected to be saturating, 1000 times larger than the enzyme concentration. These concentrations were selected by dividing the ESI1c dataset (85 curves) into a training set (45 curves) and 13 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint a testing set (40 curves). Enzyme concentrations were optimized by minimizing the root mean square deviation between the dimer pIC50s and cellular pEC50s within the training set. This optimization was performed byscipy.optimize.minimize.35 Optimized concen- trations were then applied to estimate dimer pIC50/pIC90 values in the testing set. The reported mean and standard deviation are results from repeating this procedure 100 times. Velocitiesweresimulatedat50geometricallydistributedinhibitorconcentrationsbetween 1 pM to 1 mM and normalized to be between 0 and 100%. Figure 1: Representative inhibitionIC 50 and control IC 50. Hill equation parameters were estimated using maximum likelihood estimation. For the biochemical data, estimation was performed using our custom code.36 The EC50 is the half maximal effective concentration in cellular assays. Cellular pEC50s for antiviral assays were estimated by fitting the Hill equation to the data using CDD Vault (https://www. collaborativedrug.com/). Besides IC50, another commonly used metric for assessing the potency of a drug is the IC 90. This parameter denotes the concentration at which 90% of the enzyme is inhibited. Given the IC 50, the Hill slope H, and a specific percentage F ranging between 0 to 100, representing for the degree of enzymatic inhibition or cellular viability, theIC (F ) can be 14 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint calculated using the formula 14,37 IC (F ) =  F 100 − F 1/H IC 50. (14) With F set to 90, the formula simplifies to, pIC 90 = pIC 50 − log(9)/H. (15) Analogous formulae apply to pEC50 and pEC90 values for cellular assays. 2.4 Correlation analysis Correlation between the pIC50/pIC90 values obtained from different biochemical procedures and cellular pEC50/pEC90 were analyzed by a range of statistical measures, including the Pearson R,38 Spearman ρ,39 Kendall τ,40 root mean square deviation (RMSD), and adjusted RMSD (aRMSD).41 Unlike the Pearson R, which measures linear correlation, the Spearman ρ assesses the monotonic relationship between two variables by ranking data points and evaluating how well the ranks correspond, making it robust to nonlinearity. The Kendall τ evaluates whether pairs of data points move in the same or opposite directions. RMSD evaluates the absolute differences between predicted and observed values, RM SD = vuu t 1 N NX n=1 (xn − yn)2. (16) The adjusted RMSD (aRMSD) avoids the effect of systematic bias by normalizing RMSD usingthemeansofthedata, providingalocation-independentmeasureofpredictiveaccuracy, aRM SD = vuu t 1 N NX n=1 [xn − yn − (¯x − ¯y)]2. (17) For the dimer pIC50/pIC90, the ESI4c 85 dataset was randomly split into a training set 15 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (45 curves) and a test set (40 curves). Enzyme concentrations optimized using the training set were applied to estimate dimer pIC50/pIC90 values in the test set. This procedure was repeated 100 times and the mean and standard deviation of results are reported. 2.5 Code All code is freely available athttps://github.com/vanngocthuyla/kinetic_mpro/tree/ main/mers. 3 Results 3.1 Bayesian credible intervals were converged The convergence of sampling from Bayesian posteriors was evaluated based on the 5-th, 25-th, 50-th, 75-th and 95-th percentiles of the marginal probability of parameters. In representatives of all analyses (Figure S1, S4, S16, S20), these percentiles exhibited minimal changes as the number of samples increases. Estimated standard errors were negligible. This convergence indicates that the posterior distributions have been thoroughly sampled after a small number of samples from the posterior. 3.2 Global fitting increases the precision of parameter estimates Overall, we observed that Bayesian posterior probability distributions became narrower with the inclusion of additional data. In this section, we describe results from Steps 1 through 4. All Step 2 results are with a representative compound, ASAP-0000214. Marginal distributions of all enzyme-substrate binding free energies are unimodal and nearly independent (Figures 2, S2, and S6). ∆Gd has a relatively small highest density interval (HDI) that is unaffected by the amount of data analyzed. The other parameters are broader in Steps 1 and 2, with∆GS,M and ∆GS,DS approaching the upper limit of weak 16 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint affinity. All parameters are defined more precisely in Steps 3 and 4. For instance, the 95% HDI of∆GS,M in Steps 1 and 2 ranged between -7.5 and 0, whereas in Step 3 and 4, the HDI fell into a narrower range between -5.7 and -4.4. Between Steps 3 and 4, there is no significant difference in posterior distributions of these parameters. In Steps 1 and 2, two-dimensional marginal distributions suggest that pairs of enzyme-substrate binding free energies are nearly independent (Figures S3 and S7). In Steps 3 and 4, however,∆GS,M is negatively correlated with ∆GS,D (Pearson R = -0.75) and positively correlated with∆GS,DS (Pearson R = 0.94). ∆GS,D and ∆GS,DS are negatively correlated with each other (Pearson R = -0.76) (Figures S13, S14, S17, and S18). With additional data, the binding cooperativity of the substrate also becomes more clearly determined (Figure 3). In steps 1 and 2, the estimated difference between∆GS,DS and ∆GS,D is broad (step 1: median 6.5 and 95% HDI [1.7, 11.0] kcal/mol; step 2: median 6.6 and 95% HDI [2.2, 10.0] kcal/mol). The median is positive, suggesting negative cooperativity of substrate binding, with the caveat of low precision. However, with additional data, the posterior is much narrower and the median indicates positive cooperativity of substrate binding (step 3: median -0.99 and 95% HDI [-2.3, 0.57] kcal/mol; step 4 median -0.89; 95% HDI: [-1.9, 0.54] kcal/mol). Marginal distributions of all enzyme-inhibitor binding free energies are unimodal, and some show significant correlations (Figure S7). In Step 2,∆GI,DI reaches the lower bound of strong affinity, indicating that the second binding of the inhibitor to the dimer-inhibitor complex is highly favorable. Correlations can be summarized via Pearson correlation co- efficients. In Step 2, the heatmap of the estimated parameters indicates that∆GS,DS and ∆GI,M, which have broad posterior distributions, have no correlation with any other param- eters (Figure S8). On the other hand,∆GI,D displays a strong negative correlation with both ∆GI,DI and ∆GS,DI, but have no correlation with any other dissociation constants (Figures S8, S9a and S9b).∆GS,DI demonstrates a positive linear correlation with∆GI,DI, although the difference between them is small (∆GS,DI=∆GI,DI+1.83, Figure S9c). 17 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint Figure 2: 1D marginal distributions of free energies (kcal/mol). Estimates were based on 1,000 MCMC samples generated from the Bayesian posterior in Steps 1 (purple dotted line), 2 (green dashdot line), 3 (orange dashed line), and 4 (blue solid line). Red bars represent 95% HDIs from Step 4. The red triangle marks the median of that posterior. Figure 3: Differences in binding free energies of substrate (kcal/mol). Estimates were based on 1,000 MCMC samples generated from the Bayesian posterior in Steps 1 (purple dotted line), 2 (green dashdot line), 3 (orange dashed line), and 4 (blue solid line). Red bars represent 95% HDIs from Step 4. The red triangle marks the median of that posterior. 18 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint Bayesian posterior distributions of most rate constants are broad; data are insufficient to estimate these parameters accurately (Figure 4). For Steps 1 and 2, the marginal ofkcat,DSS has a peak at low rates and a heavy tail that spans the range of the prior. In Steps 3 and 4, the posterior of low rates is significantly decreased. The posterior ofkcat,DSS is flat in Steps 1 and 2, but sharply defined after Steps 3 and 4. For the representative ligand, posteriors of kcat,DSI are flat despite the inclusion of more data (Figure S10). Additionally, 2D joint marginal distributions show that there is no correlation between rate constants (Figure S11). Figure 4: 1D marginal distributions of rate constants (min−1). Estimates were based on 1,000 MCMC samples generated from the Bayesian posterior in Steps 1 to 4. The posterior distribution of kcat,DSS is shown zoomed in opposed to the full domain between 0 and 5 observed for Steps 1 and 2. Annotations are analogous to Figure 2. While individual rates are difficult to determine, the data are sufficient to estimate ratios of rate constants. The posterior distributions for kcat ratios are better defined compared to those for kcat themselves. In Step 4, the catalytic rate for the dimer bound to a single substrate (DS) is faster than the rate when bound to two substrates (DSS) (Figure S19). For the representative inhibitor, both are considerably lower than the rate observed for the dimer-substrate-inhibitor complex (DSI) (Figure S10). In Step 5, for the majority of ligands in the dataset, kcat,DSI is comparable to or larger thankcat,DS (Figure 5). However, a few ligands (ASAP-00008489-001, ASAP-00011343-001, ASAP-00011513-001, ASAP-00012331- 001, ASAP-00012335-001, ASAP-00013263-001, ASAP-00013299-001, ASAP-00013301-001, 19 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint ASAP-00013407-001, ASAP-00013412-001, ASAP-00014551-001, ASAP-00014717-001, ASAP- 00014750-001, ASAP-00014776-001, ASAP-00015517-001) clearly have lowkcat,DSI. Figure 5: Ratios of rate constants for all ligands from the ESI4c data set. Estimates were based on 1,000 MCMC samples generated from the Bayesian posterior. Red bars represent 95% HDIs, while green triangles mark the median of the posteriors. The vertical blue lines represented for the prior distributions. Enzyme concentration parameters are consistent with stated values except at the highest enzyme concentration (Figure 6). In Steps 1 and 2, there is still significant uncertainty in the enzyme concentrations; marginal distributions are broad. In Steps 3 and 4, the 95% HDI of enzyme concentration parameters included the stated concentrations of 25 and 50nM, but the median enzyme concentration parameter for 100nM is only 59.7nM. It is possible that at high concentration, the enzyme precipitates or forms inactive higher-order oligomers, reducing the concentration of active enzyme. 20 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint Figure 6: 1D marginal distributions of enzyme concentrations (nM). Estimates were estimated based on 1,000 MCMC samples generated from the Bayesian posterior. Annotations are analogous to Figure 2. 3.3 The enzyme kinetics model closely fits nearly all data In Step 4, the enzyme kinetics model is a close fit to nearly all of the data from the ES, ESI4c, and three ESI1c datasets (Figure 7). For some ligands, the velocity is higher than the model near the peak velocity and at the highest ligand concentration of 99.5µM. For high concentrations, ligands may have solubility limits that reduce the amount of ligand in solution. Similar results are achieved in Step 5 (Figure S21). A small subset of compounds (ASAP-0000219-001, ASAP-0000375-001, ASAP-0010712-001, ASAP-0013397-001, ASAP- 0013405-001, ASAP-0013407-001, ASAP-0013423-001) exhibit a significantly broader 95% posterior predictive interval of the velocity. 21 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint 10 7 10 6 Substrate (M) 0.0 0.5 1.0 1.5Rate (nM min 1) 10 7 10 6 10 5 ASAP-0000153-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000214-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000223-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000272-001 (M) 0 1 2 3Rate (nM min 1) 10 7 10 6 10 5 10 4 ASAP-0000577-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000654-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000733-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0000773-001 (M) 0 1 2 3Rate (nM min 1) 10 7 10 6 10 5 10 4 ASAP-0008374-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0008401-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0008420-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0008452-001 (M) 0 1 2 3Rate (nM min 1) 10 7 10 6 10 5 10 4 ASAP-0008502-001 (M) 0 1 2 3 10 7 10 6 10 5 10 4 ASAP-0008642-001 (M) 0.0 0.2 0.4 0.6 10 8 10 6 10 4 ASAP-0011124-001 (M) 0.0 0.2 0.4 0.6 Figure 7: Fit of the model to ES + all ESI4c + 3 ESI1c datasets. X axes are concentrations (M). Dots are observed velocities. Velocities predicted by the model are y∗ n(θM AP) (dashed line), whereθM AP is the MAP estimate, the mean of the posterior prediction (solid line), and the 95% posterior predictive interval (shaded region). Data are colored by plate. For the upper left plot: 100 nM (brown), 50 nM (red), and 25 nM (black) of the enzyme. For other plots: enzyme 100 nM, substrate 1350 nM (blue); enzyme 100 nM, substrate 750 nM (orange); enzyme 50 nM, substrate 150 nM (purple); enzyme 100 nM, substrate 50 nM (green); enzyme 50 nM, substrate 550 nM (pink, olive, gray). 3.4 MERS-CoV MPro undergoes substrate-induced dimerization We modeled substrate-induced dimerization by calculating the effect on substrate on the amount of enzyme in the monomeric versus the dimeric form. We computed the concentra- tions of monomeric ([M] and [MS]) and dimeric ([D], [DS], [DSS]) species given an initial 22 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint enzyme concentration of 0.3µM and substrate concentration of 600µM, similar to a previous study,42 using thermodynamic and kinetic parameters drawn from the posterior. Monomeric and dimeric oncentrations were used to compute anapparent binding affinity, ∆Gd,app ∼ −RT ln  ([M] + [M S])2 [D] + [DS] + [DSS]  . (18) The 95% HDI of∆Gd,app is [-12, -9.2] kcal/mol with a mean of -10 kcal/mol. In contrast, the 95% HDI of∆Gd is [-6.7,-5.4] kcal/mol with a mean of -5.9 kcal/mol. Thus, the model predicts strong substrate-induced dimerization under conditions similar to prior work.42 3.5 The MERS-CoV MPro dimer binds most ligands with positive cooperativity Mostinhibitorsexhibitedpositivecooperativitywiththeenzyme, asthefreeenergydifference between the second and first binding events was less than zero (Figure 8), except for ASAP- 00013249-001, ASAP-00013301-001, ASAP-00013412-001, ASAP-00013894-001, and ASAP- 00014900-001. 3.6 Biochemical and cellular potencies of ASAP MERS-CoV MPro inhibitors are highly correlated Correlations between inhibition, control, and dimer pIC50s are greater than 0.8 but there are weaker correlations with cellular pEC50s (Figure 9 and Table 1). The highest correlation is observed between inhibition and control pIC50s, with the Pearson R and Spearmanρ above 0.93. Furthermore, the three pIC50s derived from the same biochemical CRC exhibit a high level of correlation with each other. On the other hand, the coefficients between the cellular pEC50 and the other three pIC50s are less than 0.8, indicating a clear discrepancy between the biochemical and cellular assays. This discrepancy suggests that some factors in the cellular environment, such as cellular permeability and metabolic stability, may not be 23 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint Figure 8: Differences in binding free energies of inhibitors (kcal/mol). Estimates were based on 1,000 MCMC samples generated from the Bayesian posterior from Step 5. Red bars represent 95% HDIs, while green triangles mark the median of the posteriors. captured in the biochemical assay. Table 1: Correlation matrix of biochemical pIC50 and cellular pEC50 by Pearson R, Spearman ρ, and Kendallτ Pearson R Inhibition pIC50 Control pIC50 Dimer pIC50 Control pIC 50 0.955 ± 1.493E-2 Dimer pIC 50 0.888 ± 7.541E-2 0.878 ± 6.918E-2 Cellular pEC50 0.720 ± 6.576E-2 0.712 ± 6.678E-2 0.711 ± 6.152E-2 Spearman ρ Inhibition pIC50 Control pIC50 Dimer pIC50 Control pIC 50 0.937 ± 2.156E-2 Dimer pIC 50 0.925 ± 4.765E-2 0.915 ± 4.493E-2 Cellular pEC50 0.677 ± 6.061E-2 0.659 ± 6.792E-2 0.718 ± 5.166E-2 Kendall τ Inhibition pIC50 Control pIC50 Dimer pIC50 Control pIC 50 0.807 ± 3.352E-2 Dimer pIC 50 0.839 ± 4.543E-2 0.805 ± 4.485E-2 Cellular pEC50 0.501 ± 5.298E-2 0.492 ± 5.797E-2 0.530 ± 4.819E-2 The dimer pIC50 outperforms the inhibition and control pIC50 in forecasting the rank 24 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint Figure 9: Correlogram of inhibition, control, dimer pIC50 and cellular pEC50. Table 2: Correlation matrix of biochemical pIC50 and cellular pEC50 by RMSD and aRMSD RMSD Inhibition pIC50 Control pIC50 Dimer pIC50 Control pIC 50 0.684 ± 1.565E-2 Dimer pIC 50 0.560 ± 5.985E-2 0.337 ± 1.173E-1 Cellular pEC50 0.568 ± 3.833E-2 0.415 ± 2.503E-2 0.415 ± 7.359E-2 aRMSD Inhibition pIC50 Control pIC50 Dimer pIC50 Control pIC 50 0.132 ± 1.499E-2 Dimer pIC 50 0.261 ± 1.327E-1 0.267 ± 1.194E-1 Cellular pEC50 0.338 ± 2.350E-2 0.352 ± 2.434E-2 0.405 ± 7.405E-2 of the cellular pEC50, but has similar Pearson R and higher error (Table 1). Assuming that the statistical metrics are independent and follow a normal distribution, we evaluated 25 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint p-values from the two-sample comparison of coefficients with the null hypothesis that there is no difference in the means of coefficients, then applied the Bonferroni adjustment for the comparison of 4 procedures.43 With the threshold at 0.05, the adjusted p-values suggested stronger evidence to reject the null hypothesis for Spearmanρ and Kendall τ (Table S3). In other words, the correlation coefficients between dimer pIC50 and cellular pEC50 are higher than the coefficients between other biochemical pIC50s and cellular pEC50 when evaluated by Spearmanρ and Kendallτ. This indicates that, according to rank order, the dimer model may align more closely with cellular responses than the inhibition and control procedures. 3.7 The dimer pIC90 better ranks cellular pEC90 than the dimer pIC50 ranks cellular pEC50 As many of the CRCs from antiviral assays have a Hill slope distinct from one, we hypoth- esized that a high level of MPro inhibition is required to improve cell viability and that biochemical pIC90 would better predict cellular pEC90 than biochemical pIC50 predicts cel- lular pEC50. Overall, correlation and error metrics for pIC90/pEC90 (Figure S22, Table S1, Table S2, and S4) were higher than those for pIC50/pEC50 (Figure 9, Table 1, and Table 2) when evaluated by Spearmanρ and Kendall τ. 4 Discussion 4.1 Enzymekineticsexperimentsaresufficienttodetermineparam- eters of a complex model We have demonstrated that enzyme kinetics experiments with varying concentrations of enzyme, substrate, and multiple inhibitors are sufficient to precisely determine most pa- rameters of a complex enzyme kinetics model. When we recently introduced our enzyme kinetics model that incorporates both dimerization and ligand binding,21 we not only fit it 26 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint to biochemical enzyme kinetics but also to analytical ultracentrifugation data for two vari- ants of SARS-CoV-2 MPro: the wild-type enzyme and a mutant engineered to have a weaker dimerization affinity.44 Here we pursued an alternative strategy of fitting many CRCs for one enzyme. We have also addressed the computational challenge associated with globally fitting nu- merous CRCs. We leveraged the structure of Bayesian posteriors to break down the fitting process into multiple steps and incrementally incorporate additional information into the model. Even with uninformative priors (besides the dimerization affinity), we observed that as the amount of data are increased from Steps 1 to 4, the binding free energies and ratios of species-specific rates are estimated more precisely. The success of global fitting extends be- yond the estimation of binding affinities for the inhibitors present in the fitted datasets; the shared parameters derived from this process can also be leveraged to fit additional datasets that were not included in the original fitting procedure. 4.2 Parameters are within ranges reported for MERS-CoV MPro and distinct from other MPro variants Our estimated parameters are consistent with values reported for MERS-CoV MPro enzyme kinetics. The 95% HDI ofKd in our paper is consistent with reported values from different measurements: 7.8 ± 0.3 µM by enzymatic assay;16 52 ± 5 µM by analytical ultracentrifu- gation;16 and 7.7 ± 0.3 µM by analytical ultracentrifugation.42 Based on the assumption that only dimeric enzyme is catalytically active, Tomar et al.16 fit the observed rate as a function of the enzyme concentration to a model in which the apparent rate is proportional on the dimer concentration. They reported kcat of 0.2 ± 0.02 min−1, which is faster than kcat,DSS but slower thankcat,DS. Given that the apparentkcat should be linear combination of turnover numbers from both species, the reported intermediate value does not raise concerns. MERS-CoV belongs to a group of RNA viruses that have caused several human outbreaks over the past two decades, including the severe acute respiratory syndrome (SARS) CoV and 27 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint SARS-CoV-2.4,45 Based on phylogenetic analysis, MERS-CoV belongs toβ-CoV lineage C and is more closely related toTylonycteris bat CoV HKU4 andPipistrellus bat CoV HKU5, while SARS-CoV and SARS-CoV-2 are classified intoβ-CoV lineage B.46–48Our estimated Kd for MERS-CoV MPro is weaker than for human SARS-CoV (0.7± 0.02 µM)42 and SARS-CoV-2 (1.32 ± 0.2 µM),44 as well as the enzymes from closely related bat CoVs such as HKU4-CoV (0.1± 0.03 µM) and HKU5-CoV (0.06± 0.01 µM).42 The dimerization free energy is between what we reported for the SARS-CoV-2 MPro wild type (median: -8.9 kcal/mol; 95% HDI: [-9.8, -7.0] kcal/mol) and slightly stronger than the mutant (median: -4.2 kcal/mol; 95% HDI: [-4.6, -1.7] kcal/mol).21 Differences in dimerization affinities between MPro from different CoVs may be largely attributed to dimerization interfaces. In the case of SARS-CoV MPro, there are intermolec- ular polar interactions involving four amino acid pairs (Ser1-Glu166, Arg4-Glu290, Ser123- Arg298, and Ser139-Gln299).42 For SARS-CoV-2 MPro, in addition to three of the pairs (Ser1-Glu166, Arg4-Glu290, and Ser139-Gln299), hydrogen bonding between the side chains of two Ser10, along with a long-distance ionic interaction between Lys12 and Glu14, were reported to contribute to this process.49 In contrast, only two pairs are found in MERS-CoV MPro: Ser1-Glu169 and Ser142-Gln299.42 The reduced number of intermolecular interac- tions likely cause MERS-CoV MPro to have a weakerKd compared to the enzymes from the other human CoVs. On the other hand, the differences in theKd values between MERS- CoV MPro and its closely related HKU4-CoV and HKU5-CoV can be attributed to the non-conserved residues located in the N-terminal finger, the N-terminal helix, and domain III.16 Comparing other enzyme kinetics parameters for SARS-CoV-2 MPro from our previous analysis21 and MERS from our present analysis illustrates variations in how the enzyme can behave and interact with ligands. The binding free energy of substrate to the monomer ∆GS,M is comparable in MERS-CoV MPro (median: -5.1; 95% HDI: [-5.7, -4.4] kcal/mol) and SARS-CoV-2 MPro (median: -4.7; 95% HDI: [-4.8, -4.5] kcal/mol), but dimerization 28 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint has a much greater effect on substrate binding for MERS-CoV MPro. While the binding affinity of the substrate for the dimer∆GS,D is much lower than∆GS,M for MERS-CoV MPro (median: -9.0; 95% HDI: [-9.9, -8.4] kcal/mol), it is similar to∆GS,M for SARS-CoV- 2 MPro (median: -4.7; 95% HDI: [-6.0, -4.0] kcal/mol). There is also a contrast in binding cooperativity behavior; while the binding free energy of the substrate to the dimer-substrate complex ∆GS,DS for MERS-CoV MPro (median -10.0; 95% HDI: [-10.1, -9.3] kcal/mol) is lower than∆GS,D, indicating positive cooperativity,∆GS,DS for SARS-CoV-2 MPro (median -0.8; 95% HDI: [-2.7, 0.] kcal/mol) is higher than∆GS,D, indicating negative cooperativity. Relative rate constants also differ between MPro variants. For wild-type SARS-CoV-2 MPro, all of the rates -kcat,DS, kcat,DSS, and kcat,DSI - are comparable.21 For the mutant,kcat,DSI and kcat,DSS are comparable to each other and larger thankcat,DS. Here, we report that kcat,DSI and kcat,DS are comparable to each other and larger thankcat,DSS for MERS-CoV MPro. These comparisons suggest that coronaviruses could employ different strategies for the regulation of enzyme activity. Compared to SARS-CoV-2 MPro, MERS-CoV MPro requires more enzyme to dimerize and become active, but the dimer binds more tightly to the substrate. As a word of caution, differences in enzyme kinetics parameters may not be due the pro- teins themselves, but can be affected by assay conditions such as the choice of substrate. Our study used the substrate [5-FAM]-AVLQSGFR-[Lys(Dabcyl)]-K-amide. In contrast, Nashed et al.44 used Dabsyl-KTSAVLQ/SGFRKM-E(Edans)-NH2, a longer peptide that could have greater difficulty occupying both binding sites. Additional data and analyses would be re- quired to dissect whether parameter differences originate from the protein sequence or the assay conditions. 29 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint 4.3 FullCRCfittinganddimerpIC90calculationsarerecommended for drug discovery targeting MERS-CoV MPro and similar en- zymes We recommend interpreting CRCs of MERS-CoV enzyme inhibition by fitting an enzyme kinetics model and calculating dimer pIC90s. We compared three data analysis procedures using Pearson R, Spearmanρ, and Kendallτ correlation metrics based on two assumptions: the null hypothesis for multiple-sample comparisons that there are no differences in the means of the coefficients; and the correlation coefficients follow a normal distribution. While bounded metrics cannot strictly follow a normal distribution, these assumptions are a rea- sonable approximation. Based on these assumptions, all three data analysis procedures that we evaluated yielded inhibition constants that are correlated with cellular efficacy. While all three biochemical procedures have similar Pearson R, dimer pIC50/90 generally have a higher Spearman ρ and Kendall τ rank correlation with cellular pEC50/90. Additionally, the pIC90 was found to be more predictive than the pIC50. In the context of a drug dis- covery campaign, the major objective of performing biochemical assays is to prioritize a subset of compounds to test in more expensive and challenging antiviral assays. Rank order correlations are the most relevant metrics for prioritization decisions. Beyond achieving the highest rank correlations with cellular efficacy, fitting the enzyme kinetics model provides insight into the mechanism of specific inhibitors. MPro enzyme kinetics can be altered by inhibitor binding free energies to different species including the monomer, dimer, and dimer-ligand complexes. Determining these parameters can help eval- uate whether a compound promotes dimerization or binds cooperativity to the target. The parameters can also quantify whether inhibitor binding to one catalytic site increases activ- ity of the opposite site. Finally, determined parameters can be used to validate results from molecular simulations. Because molecular simulations are performed with specific species, inhibitor binding free energies to specific species can be directly compared to forecasts from 30 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint these calculations. Parameters from our fitting procedure have been used as benchmarks to validate an ASAP computational workflow Castellanos et al.50 and in a blind challenge for the computational chemistry community MacDermott-Opeskin et al.22. The main drawback of full CRC fitting is its relative complexity. Fortunately, we have made our code freely available at https://github.com/vanngocthuyla/kinetic_mpro/ tree/main/mers. In service of the ASAP drug discovery campaign targeting SARS-CoV- 2/MERS-CoV MPro, the procedure was integrated into an automated data analysis pipeline that presents results on CDD Vault (https://www.collaborativedrug.com/). Beyond MERS-CoV MPro, substrate/ligand-induced dimerization has been observed in multiple serine proteases.51,52 Biphasic CRCs have been reported for the cysteine protease caspase-1,53 and may be a factor in other enzymes. 5 Conclusion We have developed a multi-step statistical analysis procedure to fit an enzyme kinetics model that incorporates dimerization and ligand binding to many CRCs with multiple inhibitors from a drug discovery campaign. The analysis precisely determines many binding parameters and ratios of rate constants and quantifies substrate-induced dimerization and ligand binding cooperativity. For the leads in the ASAP drug discovery campaign targeting MERS-CoV and SARS-CoV-2, inhibition constants from multiple data analysis procedures are highly correlated with cellular efficacy. pIC90s estimated by simulating CRCs at high enzyme concentration are have higher rank correlation with cellular pEC90 than the other tested procedures. Code implementing the procedure is freely available and is recommended for the interpretationofbiphasicCRCsfromMERS-CoVMProandenzymeswithsimilarproperties. 31 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint 6 Disclosures JDC is a current member of the Scientific Advisory Board of OpenEye Scientific, and has equity in and serves as the Chief Executive Officer of Achira, Inc. which is engaged in the creation of open foundation simulation models for drug discovery. A complete history of all entities that have provided funding for the Chodera lab can be found athttp:// choderalab.org/funding. DDLM is a founder of Biagon Inc. Biagon is not doing work related to this paper.

ThisprojectwassupportedinpartbyNIAIDoftheNationalInstitutesofHealthunderaward no. U19AI171399 (JDC) and the NIH/NCI Cancer Center Support Grant P30 CA008748 (JDC). This work used the Jetstream2 cloud-based environment at Indiana University through allocation MCB150144 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Science Foundation or the National Institutes of Health. Supporting Information A vailable

(1) Zaki, A. M.; van Boheemen, S.; Bestebroer, T. M.; Osterhaus, A. D. M. E.; Fouchier, R. A. M. Isolation of a Novel Coronavirus from a Man with Pneumonia in Saudi Arabia. The New England Journal of Medicine2012, 367, 1814–1820. 32 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (2) WHO. https://www.who.int/health-topics/middle-east-respiratory-syndrome- coronavirus-mers. (3) AlBalwi, M. A.; Khan, A.; AlDrees, M.; Gk, U.; Manie, B.; Arabi, Y.; Alabdulka- reem, I.; AlJohani, S.; Alghoribi, M.; AlAskar, A.; AlAjlan, A.; Hajeer, A. Evolving Sequence Mutations in the Middle East Respiratory Syndrome Coronavirus (MERS- CoV). Journal of Infection and Public Health2020, 13, 1544–1550. (4) Bhadoria, P.; Gupta, G.; Agarwal, A. Viral Pandemics in the Past Two Decades: An Overview. Journal of Family Medicine and Primary Care2021, 10, 2745–2750. (5) Kumar, V.; Shin, J. S.; Shie, J.-J.; Ku, K. B.; Kim, C.; Go, Y. Y.; Huang, K.-F.; Kim, M.; Liang, P.-H. Identification and Evaluation of Potent Middle East Respiratory Syndrome Coronavirus (MERS-CoV) 3CLPro Inhibitors.Antiviral Research2017, 141, 101–106. (6) He, J.; Hu, L.; Huang, X.; Wang, C.; Zhang, Z.; Wang, Y.; Zhang, D.; Ye, W. Po- tential of Coronavirus 3C-like Protease Inhibitors for the Development of New Anti- SARS-CoV-2 Drugs: Insights from Structures of Protease and Inhibitors.International Journal of Antimicrobial Agents2020, 56, 106055. (7) Jin, Z. et al. Structure of Mpro from SARS-CoV-2 and Discovery of Its Inhibitors. Nature 2020, 582, 289–293. (8) Ma, C.; Sacco, M. D.; Hurst, B.; Townsend, J. A.; Hu, Y.; Szeto, T.; Zhang, X.; Tar- bet, B.; Marty, M. T.; Chen, Y.; Wang, J. Boceprevir, GC-376, and Calpain Inhibitors II, XII Inhibit SARS-CoV-2 Viral Replication by Targeting the Viral Main Protease. Cell Research 2020, 30, 678–692. (9) Roe, M. K.; Junod, N. A.; Young, A. R.; Beachboard, D. C.; Stobart, C. C. Target- ing Novel Structural and Functional Features of Coronavirus Protease Nsp5 (3CLpro, Mpro) in the Age of COVID-19.The Journal of General Virology2021, 102, 001558. 33 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (10) Lu, J.; Chen, S. A.; Khan, M. B.; Brassard, R.; Arutyunova, E.; Lamer, T.; Vuong, W.; Fischer, C.; Young, H. S.; Vederas, J. C.; Lemieux, M. J. Crystallization of Feline Coro- navirus Mpro With GC376 Reveals Mechanism of Inhibition.Frontiers in Chemistry 2022, 10, 852210. (11) Barkan, D. T. et al. Identification of Potent, Broad-Spectrum Coronavirus Main Pro- tease Inhibitors for Pandemic Preparedness.Journal of Medicinal Chemistry2024, (12) de Wit, E.; van Doremalen, N.; Falzarano, D.; Munster, V. J. SARS and MERS: Recent Insights into Emerging Coronaviruses.Nature Reviews. Microbiology2016, 14, 523–534. (13) Gorkhali, R.; Koirala, P.; Rijal, S.; Mainali, A.; Baral, A.; Bhattarai, H. K. Structure and Function of Major SARS-CoV-2 and SARS-CoV Proteins.Bioinformatics and Bi- ology Insights 2021, 15, 11779322211025876. (14) Lei, J.; Kusov, Y.; Hilgenfeld, R. Nsp3 of Coronaviruses: Structures and Functions of a Large Multi-Domain Protein.Antiviral Research2018, 149, 58–74. (15) Almutairi, G. O.; Malik, A.; Alonazi, M.; Khan, J. M.; Alhomida, A. S.; Khan, M. S.; Alenad, A. M.; Altwaijry, N.; Alafaleq, N. O. Expression, Purification, and Biophysi- cal Characterization of Recombinant MERS-CoV Main (Mpro) Protease.International Journal of Biological Macromolecules2022, 209, 984–990. (16) Tomar, S.; Johnston, M. L.; St John, S. E.; Osswald, H. L.; Nyalapatla, P. R.; Paul, L. N.; Ghosh, A. K.; Denison, M. R.; Mesecar, A. D. Ligand-Induced Dimerization of Middle East Respiratory Syndrome (MERS) Coronavirus Nsp5 Protease (3CLpro): IMPLICATIONS FOR Nsp5 REGULATION AND THE DEVELOPMENT OF AN- TIVIRALS. The Journal of Biological Chemistry2015, 290, 19403–19422. (17) Graziano, V.; McGrath, W. J.; Yang, L.; Mangel, W. F. SARS CoV Main Pro- 34 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint teinase: The Monomer-Dimer Equilibrium Dissociation Constant.Biochemistry 2006, 45, 14632–14641. (18) Silvestrini, L.; Belhaj, N.; Comez, L.; Gerelli, Y.; Lauria, A.; Libera, V.; Mariani, P.; Marzullo, P.; Ortore, M. G.; Palumbo Piccionello, A.; Petrillo, C.; Savini, L.; Pacia- roni, A.; Spinozzi, F. The Dimer-Monomer Equilibrium of SARS-CoV-2 Main Protease Is Affected by Small Molecule Inhibitors.Scientific Reports 2021, 11, 9283. (19) Paciaroni, A. et al. Stabilization of the Dimeric State of SARS-CoV-2 Main Protease by GC376 and Nirmatrelvir.International Journal of Molecular Sciences2023, 24, 6062. (20) Nguyen, H. H.; Tufts, J.; Minh, D. D. L. On Inactivation of the Coronavirus Main Protease. Journal of Chemical Information and Modeling2024, 64, 1644–1656. (21) La, V. N. T.; Kang, L.; Minh, D. Enzyme Kinetics Model for the Coronavirus Main Protease Including Dimerization and Ligand Binding.Biophysical Journal 2025, Ac- cepted. (22) MacDermott-Opeskin, H. et al. A Computational Community Blind Challenge on Pan- Coronavirus Drug Discovery Data. 2025. (23) Boby, M. L. et al. Open Science Discovery of Potent Noncovalent SARS-CoV-2 Main Protease Inhibitors.Science 2023, 382, eabo7201. (24) ASAP Discovery - Research Workspace on Protocols.Io. https://www.protocols.io/workspaces/asap-discovery. (25) Barr,H.; Lahav,N.MERS-CoVMproFluorescenceDoseResponseforAntiviralTesting V5. 2025. (26) L McGovern, B. Live Virus MERS-CoV - Vero-TMPRSS2 + Pgp Inhibitor- Antiviral Screening Assay V1. 2024. 35 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (27) Grubbs, F. E. Procedures for Detecting Outlying Observations in Samples.Technomet- rics 1969, 11, 1–21. (28) Kadian, N.; Raju, K. S. R.; Rashid, M.; Malik, M. Y.; Taneja, I.; Wahajuddin, M. Com- parative Assessment of Bioanalytical Method Validation Guidelines for Pharmaceutical Industry.Journal of Pharmaceutical and Biomedical Analysis2016, 126, 83–97. (29) Jeffreys, H. An Invariant Form for the Prior Probability in Estimation Problems.Pro- ceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences 1946, 186, 453–461. (30) Zuckerman, D.; Otten, L. Combining Datasets for Bayesian Inference. 2025. (31) Homan, M. D.; Gelman, A. The No-U-turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo.J. Mach. Learn. Res.2014, 15, 1593–1623. (32) Chodera, J. D. A Simple Method for Automated Equilibration Detection in Molecular Simulations. Journal of chemical theory and computation2016, 12, 1799–1805. (33) Shirts, M. R.; Chodera, J. D. Statistically Optimal Analysis of Samples from Multiple Equilibrium States.The Journal of Chemical Physics2008, 129, 124105. (34) Shirts, M. R.; Ferguson, A. L. Statistically Optimal Continuous Free Energy Surfaces from Biased Simulations and Multistate Reweighting.Journal of Chemical Theory and Computation 2020, 16, 4107–4125. (35) Virtanen, P. et al. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods 2020, 17, 261–272. (36) La, V. N. T.; Nicholson, S.; Haneef, A.; Kang, L.; Minh, D. D. L. Inclusion of Control Data in Fits to Concentration–Response Curves Improves Estimates of Half-Maximal Concentrations. Journal of Medicinal Chemistry2023, 66, 12751–12761. 36 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (37) Webb, N. E.; Montefiori, D. C.; Lee, B. Dose–Response Curve Slope Helps Predict Therapeutic Potency and Breadth of HIV Broadly Neutralizing Antibodies. Nature Communications 2015, 6, 8443. (38) Pearson, K. VII. Mathematical Contributions to the Theory of Evolution.—III. Re- gression, Heredity, and Panmixia.Philosophical Transactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character1896, 187, 253–318. (39) Spearman, C. The Proof and Measurement of Association between Two Things.The American Journal of Psychology1904, 15, 72–101. (40) Kendall, M. G. A NEW MEASURE OF RANK CORRELATION.Biometrika 1938, 30, 81–93. (41) Xie, B.; Nguyen, T. H.; Minh, D. D. L. Absolute Binding Free Energies between T4 Lysozyme and 141 Small Molecules: Calculations Based on Multiple Rigid Receptor Configurations. Journal of Chemical Theory and Computation2017, 13, 2930–2944. (42) Ho, B.-L.; Cheng, S.-C.; Shi, L.; Wang, T.-Y.; Ho, K.-I.; Chou, C.-Y. Critical Assess- ment of the Important Residues Involved in the Dimerization and Catalysis of MERS Coronavirus Main Protease.PLOS ONE 2015, 10, e0144865. (43) Bonferroni, C. Teoria statistica delle classi e calcolo delle probabilita.Pubblicazioni del R istituto superiore di scienze economiche e commericiali di firenze1936, 8, 3–62. (44) Nashed, N. T.; Aniana, A.; Ghirlando, R.; Chiliveri, S. C.; Louis, J. M. Modulation of the Monomer-Dimer Equilibrium and Catalytic Activity of SARS-CoV-2 Main Protease by a Transition-State Analog Inhibitor.Communications Biology 2022, 5, 1–9. (45) Mirza, A. Z.; Shamshad, H.; Osra, F. A.; Habeebullah, T. M.; Morad, M. An Overview 37 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint of Viruses Discovered over the Last Decades and Drug Development for the Current Pandemic. European Journal of Pharmacology2021, 890, 173746. (46) Woo, P. C. Y.; Huang, Y.; Lau, S. K. P.; Yuen, K.-Y. Coronavirus Genomics and Bioinformatics Analysis.Viruses 2010, 2, 1804–1820. (47) Lau, S. K. P.; Li, K. S. M.; Tsang, A. K. L.; Lam, C. S. F.; Ahmed, S.; Chen, H.; Chan, K.-H.; Woo, P. C. Y.; Yuen, K.-Y. Genetic Characterization of Betacoronavirus Lineage C Viruses in Bats Reveals Marked Sequence Divergence in the Spike Protein of Pipistrellus Bat Coronavirus HKU5 in Japanese Pipistrelle: Implications for the Origin oftheNovelMiddleEastRespiratorySyndromeCoronavirus. Journal of Virology2013, 87, 8638–8650. (48) Chen, Y.-J.; Huang, J. C.; Yang, C.-P.; Hsu, K.-F.; Liu, H.-F. A Comprehensive Phylo- genetic Analysis of SARS-CoV-2: Utilizing a Novel and Convenient In-House RT-PCR MethodforCharacterizationwithoutVirusCultureandBSL-3Facilities. Viruses 2023, 15, 1562. (49) Ferreira, J. C.; Fadl, S.; Rabeh, W. M. Key Dimer Interface Residues Impact the Catalytic Activity of 3CLpro, the Main Protease of SARS-CoV-2.Journal of Biological Chemistry 2022, 298, 102023. (50) Castellanos, M. A.; Payne, A. M.; Scheen, J.; MacDermott-Opeskin, H.; Pulido, I.; Balcomb, B. H.; Griffen, E. J.; Fearon, D.; Barr, H.; Lahav, N.; Cousins, D.; Stacey, J.; Robinson, R.; Lefker, B.; Chodera, J. D. A Structure-Based Computational Pipeline for Broad-Spectrum Antiviral Discovery.bioRxiv: The Preprint Server for Biology2025, 2025.07.29.667267. (51) Waxman, L.; Darke, P. L. The Herpesvirus Proteases as Targets for Antiviral Chemotherapy.Antiviral Chemistry & Chemotherapy2000, 11, 1–22. 38 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint (52) Shimba, N.; Nomura, A. M.; Marnett, A. B.; Craik, C. S. Herpesvirus Protease Inhibi- tion by Dimer Disruption.Journal of Virology2004, 78, 6657–6665. (53) Datta, D.; McClendon, C. L.; Jacobson, M. P.; Wells, J. A. Substrate and Inhibitor- induced Dimerization and Cooperativity in Caspase-1 but Not Caspase-3.The Journal of Biological Chemistry2013, 288, 9971–9981. 39 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted February 21, 2026. ; https://doi.org/10.64898/2026.02.20.707097doi: bioRxiv preprint