Differentiating serine and cysteine protease mechanisms by new covalent QSAR descriptors

Reactivities of acrylamide warheads toward cysteine targets: a QM/ML approach to covalent inhibitor design

Covalent inhibition offers many advantages over non-covalent inhibition, but covalent warhead reactivity must be carefully balanced to maintain potency while avoiding unwanted side effects. While warhead reactivities are commonly measured with assays, a computational model to predict warhead reactivities could be useful for several aspects of the covalent inhibitor design process. Studies have shown correlations between covalent warhead reactivities and quantum mechanic (QM) properties that describe important aspects of the covalent reaction mechanism. However, the models from these studies are often linear regression equations and can have limitations associated with their usage. Applications of machine learning (ML) models to predict covalent warhead reactivities with QM descriptors are not extensively seen in the literature. This study uses QM descriptors, calculated at different levels of theory, to train ML models to predict reactivities of covalent acrylamide warheads. The QM/ML models are compared with linear regression models built upon the same QM descriptors and with ML models trained on structure-based features like Morgan fingerprints and RDKit descriptors. Experiments show that the QM/ML models outperform the linear regression models and the structure-based ML models, and literature test sets demonstrate the power of the QM/ML models to predict reactivities of unseen acrylamide warhead scaffolds. Ultimately, these QM/ML models are effective, computationally feasible tools that can expedite the design of new covalent inhibitors.

Exploring the Versatility of the Covalent Thiol-Alkyne Reaction with Substituted Propargyl Warheads: A Deciding Role for the Cysteine Protease

Terminal unactivated alkynes are nowadays considered the golden standard for cysteine-reactive warheads in activity-based probes (ABPs) targeting cysteine deubiquitinating enzymes (DUBs). In this work, we study the versatility of the thiol-alkyne addition reaction in more depth. Contrary to previous findings with UCHL3, we now show that covalent adduct formation can progress with substituents on the terminal or internal alkyne position. Strikingly, acceptance of alkyne substituents is strictly DUB-specific as this is not conserved among members of the same subfamily. Covalent adduct formation with the catalytic cysteine residue was validated by gel analysis and mass spectrometry of intact ABP-treated USP16CDWT and catalytically inactive mutant USP16CDC205A. Bottom-up mass spectrometric analysis of the covalent adduct with a deuterated propargyl ABP provides mechanistic understanding of the in situ thiol-alkyne reaction, identifying the alkyne rather than an allenic intermediate as the reactive species. Furthermore, kinetic analysis revealed that introduction of (bulky/electron-donating) methyl substituents on the propargyl moiety decreases the rate of covalent adduct formation, thus providing a rational explanation for the commonly lower level of observed covalent adduct compared to unmodified alkynes. Altogether, our work extends the scope of possible propargyl derivatives in cysteine targeting ABPs from unmodified terminal alkynes to internal and substituted alkynes, which we anticipate will have great value in the development of ABPs with improved selectivity profiles.

BIreactive: A Machine-Learning Model to Estimate Covalent Warhead Reactivity

In the past decade, the pharmaceutical industry has paid closer attention to covalent drugs. Differently from standard noncovalent drugs, these compounds can exhibit peculiar properties, such as higher potency or longer duration of target inhibition with a potentially lower dosage. These properties are mainly driven by the reactive functional group present in the compound, the so-called warhead that forms a covalent bond with a specific nucleophilic amino-acid on the target. In this work, we report the possibility to combine ab initio activation energies with machine-learning to estimate covalent compound intrinsic reactivity. The idea behind this approach is to have a precise estimation of the transition state barriers, and thus of the compound reactivity, but with the speed of a machine-learning algorithm. We call this method "BIreactive". Here, we demonstrate this approach on acrylamides and 2-chloroacetamides, two warhead classes that possess different reaction mechanisms. In combination with our recently implemented truncation algorithm, we also demonstrate the possibility to use BIreactive not only for fragments but also for lead-like molecules. The generic nature of this approach allows also the extension to several other warheads. The combination of these factors makes BIreactive a valuable tool for the covalent drug discovery process in a pharmaceutical context.

A Fast Ab Initio Predictor Tool for Covalent Reactivity Estimation of Acrylamides

Thanks to their unique mode of action, covalent drugs represent an exceptional opportunity for drug design. After binding to a biologically relevant target system, covalent compounds form a reversible or irreversible covalent bond with a nucleophilic amino acid. Due to the inherently large binding energy of a covalent bond, covalent binders exhibit higher potencies and thus allow potentially lower drug dosages. However, a proper balancing of compound reactivity is key for the design of covalent binders, to achieve high levels of target inhibition while minimizing promiscuous covalent binding to nontarget proteins. In this work, we demonstrated the possibility to apply the electrophilicity index concept to estimate covalent compound reactivity. We tested this approach on acrylamides, one of the most prominent classes of covalent warheads. Our study clearly demonstrated that, for compounds with molecular weight (MW) below 250 Da, the electrophilicity index can be directly used to estimate compound reactivity. On the other hand, for leadlike molecules (MW > 250 Da) we implemented a new truncation algorithm that has to be applied before reactivity calculations. This algorithm can ensure the localization of HOMO/LUMO orbitals on the compound warhead and thus a correct estimation of its reactivity. Our results also indicate that caution should be used when employing the electrophilicity index to estimate the reactivity of nonterminal acrylamides. The nonparametric nature of this method and its reasonable computational cost make it a suitable tool to support covalent drug design.

Structural and Antiviral Studies of the Human Norovirus GII.4 Protease

Norovirus is the leading cause of acute gastroenteritis worldwide with a yearly reported 700 million cases driving a $60 billion global socioeconomic burden. With no United States Food and Drug Administration approved therapeutics and the chance for severe chronic infection and life-threatening complications, researchers have identified the protease as a potential target. However, drug development has focused on the norovirus GI.1 strain despite its accounting for less than 5% of all outbreaks. Our lab aims to change focus for norovirus drug design from GI.1 to the highly infective GII.4, responsible for more than 50% of all outbreaks worldwide. With the first published crystal structure of the norovirus GII.4 protease, we have identified several significant differences in the structure and active site that have hindered development of a potent inhibitor targeting the norovirus GII.4 protease. With these new insights, we have begun designing compounds that demonstrate increased inhibition of the clinically most relevant norovirus GII.4 strain.

Characterization and functional assay of apsup (Lyxy105) from Lymantria xylina multiple nucleopolyhedrovirus (LyxyMNPV)

The baculoviral anti-apoptotic genes, p35 and iap (inhibitor of apoptosis), play important roles in the initiation of viral infection. Recently, a new anti-apoptotic gene (apoptosis suppressor, apsup) was identified in Lymantria dispar multiple nucleopolyhedrovirus (LdMNPV). An apsup homolog gene, Lyxy105 (ly-apsup), was also predicted in the Lymantria xylina multiple nucleopolyhedrovirus (LyxyMNPV) genome. In this study, we attempt to perform a gene expression analysis and a functional assay of ly-apsup to demonstrate its anti-apoptotic activity and identify the functional domain of this protein. The transcription of the ly-apsup gene region was detected from 12 h post-infection (hpi) and increased significantly after 24-72 hpi. Comparison of the putative amino acid sequences to those of 18 baculoviral homolog proteins showed high amino acid identity to the LdMNPV sequences. Moreover, five conserved protein domains (named as domains I-V) were found. Therefore, protein functional assays were conducted on full-length proteins and different truncation clones. The overexpression of each clone was confirmed by western blot analysis, and the data revealed that a cleavage of ~ 5 kDa at the N-terminal region of the full-length, domains I-IV (1-241) and I-III (1-178), proteins occurred. The results of the functional analysis showed that full-length Ly-apsup and Ly-apsup with domain I (1-70) could inhibit Drosophila-RPR protein (D-RPR)-induced and actinomycin D (ActD)-induced apoptoses. In addition, the domains I and I-II (1-126) regions showed higher anti-apoptotic activity than the other domains in both D-RPR-induced and ActD-induced cell apoptoses. In conclusion, domain I of Ly-apsup may play an important role in the anti-apoptotic activity of this protein; cleavage of the Ly-apsup N-terminus may lead to decreased anti-apoptotic activity.

Covalent inhibitors design and discovery

In the history of therapeutics, covalent drugs occupy a very distinct category. While representing a significant fraction of the drugs on the market, very few have been deliberately designed to interact covalently with their biological target. In this review, the prevalence of covalent drugs will first be briefly covered, followed by an introduction to their mechanisms of action and more detailed discussions of their discovery and the development of safe and efficient covalent enzyme inhibitors. All stages of a drug discovery program will be covered, from target considerations to lead optimization, strategies to tune reactivity and computational methods. The goal of this article is to provide an overview of the field and to outline good practices that are needed for the proper assessment and development of covalent inhibitors as well as a good understanding of the potential and limitations of current computational methods for the design of covalent drugs.

A new method for filtering of reactive "warheads" of transition-state analog protease inhibitors

In light of the major contribution of the reactive warhead to the binding energy trend in reversible covalent transition-state analog inhibitors of serine and cysteine hydrolases, would it be possible to rationally design and quickly filter such warheads, especially for large-scale screening? The previously defined W1 and W2 covalent descriptors quantitatively account for the energetic effect of the covalent bonds reorganization, accompanying enzyme-inhibitor covalent binding. The quantum mechanically calculated W1 and W2 reflect the warhead binding energy by modeling of the enzyme-inhibitor reaction core. Here, we demonstrate the use of these descriptors for warhead filtering, and examine its scope and limitations. The W1 and W2 descriptors provide a tool for rational design of various warheads as universal building blocks of real inhibitors without the requirement of 3D structural information about the target enzyme or QSAR studies. These warheads could then be used as hit structural templates in the subsequent optimization of inhibitors recognition sites.

Application of EMBM to Structure-Based Design of Warheads for Protease Inhibitors

Most CADD tools handle non-covalent enzyme inhibitors, despite the growing interest of the pharma industry in covalent inhibitors. We have recently introduced an enzyme mechanism-based method, EMBM, as a computational tool for binding trend analysis and prediction of chemical sites (CS) of reversible covalent enzyme inhibitors. In the current study we demonstrate the utility of EMBM to structure-based applications. In this mode, the energy of the enzyme-inhibitor covalent bond is accounted for by the W1 and W2 covalent descriptors we have developed, whereas the non-covalent interactions between the inhibitor CS and the enzyme active site can be estimated directly on the 3D structure of the enzyme-inhibitor complex.

Cysteine protease inhibition by nitrile-based inhibitors: a computational study

Cysteine protease enzymes are important for human physiology and catalyze key protein degradation pathways. These enzymes react via a nucleophilic reaction mechanism that involves a cysteine residue and the proton of a proximal histidine. Particularly efficient inhibitors of these enzymes are nitrile-based, however, the details of the catalytic reaction mechanism currently are poorly understood. To gain further insight into the inhibition of these molecules, we have performed a combined density functional theory and quantum mechanics/molecular mechanics study on the reaction of a nitrile-based inhibitor with the enzyme active site amino acids. We show here that small perturbations to the inhibitor structure can have dramatic effects on the catalysis and inhibition processes. Thus, we investigated a range of inhibitor templates and show that specific structural changes reduce the inhibitory efficiency by several orders of magnitude. Moreover, as the reaction takes place on a polar surface, we find strong differences between the DFT and QM/MM calculated energetics. In particular, the DFT model led to dramatic distortions from the starting structure and the convergence to a structure that would not fit the enzyme active site. In the subsequent QM/MM study we investigated the use of mechanical vs. electronic embedding on the kinetics, thermodynamics and geometries along the reaction mechanism. We find minor effects on the kinetics of the reaction but large geometric and thermodynamics differences as a result of inclusion of electronic embedding corrections. The work here highlights the importance of model choice in the investigation of this biochemical reaction mechanism.

Quantum mechanics-based scoring rationalizes the irreversible inactivation of parasitic Schistosoma mansoni cysteine peptidase by vinyl sulfone inhibitors

The quantum mechanics (QM)-based scoring function that we previously developed for the description of noncovalent binding in protein-ligand complexes has been modified and extended to treat covalent binding of inhibitory ligands. The enhancements are (i) the description of the covalent bond breakage and formation using hybrid QM/semiempirical QM (QM/SQM) restrained optimizations and (ii) the addition of the new ΔG(cov)' term to the noncovalent score, describing the "free" energy difference between the covalent and noncovalent complexes. This enhanced QM-based scoring function is applied to a series of 20 vinyl sulfone-based inhibitory compounds inactivating the cysteine peptidase cathepsin B1 of the Schistosoma mansoni parasite (SmCB1). The available X-ray structure of the SmCB1 in complex with a potent vinyl sulfone inhibitor K11017 is used as a template to build the other covalently bound complexes and to model the derived noncovalent complexes. We present the correlation of the covalent score and its constituents with the experimental binding data. Four outliers are identified. They contain bulky R1' substituents structurally divergent from the template, which might induce larger protein rearrangements than could be accurately modeled. In summary, we propose a new computational approach and an optimal protocol for the rapid evaluation and prospective design of covalent inhibitors with a conserved binding mode.

Peptidyl cyclopropenones: reversible inhibitors, irreversible inhibitors, or substrates of cysteine proteases?

Peptidyl cyclopropenones were previously introduced as selective cysteine protease reversible inhibitors. In the present study we synthesized one such peptidyl cyclopropenone and investigated its interaction with papain, a prototype cysteine protease. A set of kinetics, biochemical, HPLC, MS, and (13)C-NMR experiments revealed that the peptidyl cyclopropenone was an irreversible inhibitor of the enzyme, alkylating the catalytic cysteine. In parallel, this cyclopropenone also behaved as an alternative substrate of the enzyme, providing a product that was tentatively suggested to be either a spiroepoxy cyclopropanone or a gamma-lactone. Thus, a single family of compounds exhibits an unusual variety of activities, being reversible inhibitors, irreversible inhibitors and alternative substrates towards enzymes of the same family.

EMBM - a new enzyme mechanism-based method for rational design of chemical sites of covalent inhibitors

We introduce an enzyme mechanism-based method (EMBM) aimed at rational design of chemical sites (CS) of reaction coordinate analog inhibitors. The energy of valence reorganization of CS, caused by the formation of the enzyme-inhibitor covalent complex, is accounted for by new covalent descriptors W1 and W2. We considered CS fragments with a carbonyl reactivity center, like in native protease substrates. The W1 and W2 descriptors are calculated quantum mechanically on small molecular clusters simulating the reaction core of the formed covalent tetrahedral complex, anionic TC(O-) or neutral TC(OH). The modeling on a reaction core allows generation of various CS and corresponding TC(O-) and TC(OH) as universal building blocks of real inhibitors and their covalent complexes with serine or cysteine hydrolases. Moreover, the approach avoids the need for 3D structure of the target enzyme, so EMBM may be used for ligand-based design. We have built a chemical site of inhibitors (CSI) databank with pairs of W1 and W2 descriptors precalculated for both CH₃O(-) and CH₃S(-) nucleophiles for every collected CS fragment. We demonstrated that contribution of a CS fragment to the binding affinity of an inhibitor depends on both its covalent reorganization during the chemical transformation and its noncovalent interactions in the enzyme active site. Consequently, prediction of inhibitors binding trend can be done only by accounting for all of these factors, using W1 and W2 in combination with noncovalent QSAR descriptors.