Substrate-Guided AlphaFold2 Repositioning Identifies Goserelin as an Mpro Inhibitor

The rapid identification of antiviral candidates is a challenge in chemical biology and molecular drug discovery, especially when emerging infectious diseases move faster than conventional therapeutic development. Disease X represents a scenario in which a serious epidemic may be caused by a pathogen not yet known to cause human disease, with the earliest phase of response may depend on limited biological information instead of comprehensive structural, pharmacological, or clinical knowledge. Under such conditions, the challenge is to discover a new inhibitor and establish a strategy that can move fast from sequence-level information to experimentally testable drug candidates within a short timeframe. Coronaviruses provide a useful model for examining this challenge because their replication depends on proteolytic processing of viral polyproteins. The main protease, M^pro, plays a critical role in this process by recognizing and cleaving conserved amino acid sequences within the viral polyprotein, thereby releasing functional proteins required for viral replication and transcription. As a result, M^pro has become an important target for antiviral drug design. Structure-based virtual screening depends strongly on the quality of the protein conformation used for docking, and the ligand-binding pocket of a protease is not a fixed geometric cavity. It is shaped by substrate recognition, local side-chain rearrangement, hydrogen-bonding networks, and the dynamic accommodation of peptide-like chemical groups. AlphaFold2 offers a powerful way to predict protein structures directly from amino acid sequences, which is attractive in an outbreak setting where sequence data may become available much earlier than crystallographic structures. However, the conformation of a predicted binding site may not always be sufficiently accurate for ligand docking, because even small side-chain deviations can alter binding poses, docking scores, and candidate prioritization. This creates a methodological gap between rapid structure prediction and reliable drug repositioning. In a recent research paper published in Physical Chemistry Chemical Physics, Dr. Huixuan Zhao, Dr. Wentao Qi, Dr. Ke Liu, Dr. Jiayi Zhao, Associate Professor Xueping Hu and Professor Weiqiao Deng from Shandong University addressed this gap through an innovative drug discovery strategy that combines AlphaFold2-predicted structures, molecular dynamics refinement, an FDA-approved drug database, and molecular docking with conformer-dependent charge.

Briefly, the research team generated an M^pro –peptidyl substrate complex using AlphaFold2 from the amino acid sequences of the protease and the substrate sequence SAVLQSGFRKM.  Because M^pro naturally recognizes peptide substrates, placing the peptidyl substrate into the prediction-and-refinement process gave the binding pocket a biologically relevant reference for adopting a substrate-compatible conformation.

Molecular dynamics simulations then served as the decisive refinement step. Across 200 ns simulations, the M^pro –substrate systems reached structural stability, and binding free-energy analysis identified a representative complex with a strong substrate-binding profile. Several residues contributed to pocket stabilization, including Thr26, Glu166, His164, Cys145, Asn119, Gln19, His163, Gln189, Gly143, and Phe140. The refined complex displayed a dense hydrogen-bonding pattern, including several substrate contacts with high occupancy across the simulation. Using the peptidyl substrate to induce the pocket before screening converted a static AlphaFold2 prediction into a more chemically organized binding environment for ligand docking. The authors then used the optimized M^pro structure to screen 2005 FDA-approved drugs from DrugBank. Standard precision docking provided an initial filtration step, after which a smaller group of compounds was examined using the authors’ MDCC method. MDCC incorporates conformer-dependent RESP charges into docking, thereby treating different conformers of the same ligand as electronically distinct rather than forcing a single charge description across all conformational possibilities. From the top-ranked compounds, the authors prioritized peptide-based or amide-containing molecules because the M^pro active site is naturally adapted to peptide-bond recognition. After considering binding mode, availability, and commercial relevance, six candidates were purchased for experimental testing: Goserelin, Lypressin, Pentagastrin, Cefoperazone, Carbetocin, and Piperacillin.

The team conducted biological assays and found that at 10 μM, Goserelin inhibited SARS-CoV-2 M^pro activity by 75%, whereas the other five selected drugs showed inhibition below 50%. Dose-response analysis gave an IC50 of 3.79 μM at pH 7.5. The authors then tested Goserelin under pH 6.6, a condition linked in the paper to the nasal environment, and observed a lower IC50 of 2.05 μM with smaller deviation. The pH-dependent retention and enhancement of activity supported the authors’ suggestion that Goserelin could be further examined for nasal delivery. To understand why Goserelin emerged from the screen, the authors examined its M^pro -bound state through further molecular dynamics simulations. Goserelin adopted an extended conformation within the binding pocket and occupied the S1, S2, S4, and S1′ regions. Its tyrosine group fitted into S1 and formed a hydrogen bond with His163, while tert-butyl-serine and leucine moieties entered S2 and S4. Additional hydrogen bonds involved Gln189, Thr26, Glu166, Asn142, and Asn119, while hydrophobic contacts included residues such as Leu141, Leu167, Pro168, and Cys145. The comparison with reported noncovalent M^pro inhibitors and the peptidyl substrate strengthened the molecular interpretation: Goserelin reproduced several key binding-site occupation patterns associated with M^pro recognition.

To summarize, the most direct chemical engineering application of the Shandong University researchers is the construction of a rapid computational screening pipeline for emergency antiviral discovery. In a Disease X situation, waiting for experimentally solved protein structures can delay the first round of therapeutic exploration. The new reported strategy offers a more practical route: begin with pathogen-derived amino acid sequences, generate a protein–substrate complex with AlphaFold2, refine the binding pocket by molecular dynamics, and use the resulting structure for drug repositioning against an approved-drug library. From an engineering perspective, this creates a modular workflow that can be adapted quickly once a viral protease sequence and, when available, its substrate recognition motif is known. The value is both speed and operational organization: sequence input, pocket refinement, docking, candidate ranking, purchase of available compounds, and biochemical testing are arranged as a defined discovery process rather than as disconnected computational and experimental steps.

A second application of the authors’ findings can be in improving virtual screening infrastructure for targets whose binding pockets are sensitive to conformational change. The substrate-induced refinement used in this paper provides a way to engineer more realistic docking models for proteases, especially when the natural substrate can guide the active-site geometry. The use of MDCC further strengthens the computational side by accounting for conformer-dependent charge during docking, which is particularly relevant for larger, flexible, peptide-like approved drugs. Such a workflow could be useful in academic drug discovery centers, public-health preparedness programs, and pharmaceutical repurposing platforms where a first-pass candidate list must be produced quickly and rationally. The medical value of the approach is that it could help clinicians and translational researchers identify realistic intervention candidates early in an outbreak, before a dedicated antiviral development program has matured. Because such compounds already carry pharmacological and manufacturing information, the route from computational discovery to practical medical evaluation becomes more direct. For respiratory viral disease, the strategy also encourages attention to site-specific therapy: an inhibitor that retains activity under conditions relevant to the upper airway could be examined for local delivery, exposure at the mucosal surface, and early suppression of viral replication. Indeed, the new approach may be especially relevant for viruses that depend on protease-mediated polyprotein processing, where substrate-guided pocket modeling can connect viral biology directly to therapeutic screening.