In silico identification of praziquantel as a putative vasopressin V2 receptor antagonist for autosomal dominant polycystic kidney disease

Background: Tolvaptan, a vasopressin V2 receptor (V2R) antagonist, is the only approved disease-modifying therapy for autosomal dominant polycystic kidney disease (ADPKD). Its clinical use is limited by an FDA Black Box warning for idiosyncratic drug-induced liver injury (DILI), creating an unmet need for V2R antagonists with improved safety profiles. We additionally test whether the screening methodology generalizes by reapplying the entire pipeline to a second, unrelated protein target. Methods: We performed a virtual screen of 3,790 chemically diverse drug-like ChEMBL molecules (max_phase ≥ 2, MW 250–550) against the recently solved tolvaptan-bound V2R structure (PDB 9HAP) using AutoDock Vina (exhaustiveness 8) on a consumer laptop CPU. Candidates passing the tolvaptan score threshold (≤ −8.43 kcal/mol) were filtered through staged drug-likeness, PAINS, and reactive-group filters, then categorized into priority tiers using Tanimoto novelty against known vaptans and ChEMBL clinical-phase metadata. Top candidates were structurally validated by computing receptor-residue contact overlap with the tolvaptan reference pose, cross-validated by independent AI-based blind docking with DiffDock-L, and assessed for predicted hepatotoxicity using admet-AI. Cross-receptor selectivity was characterized by Vina counter-docking against AlphaFold-predicted V1a, V1b, and oxytocin (OTR) receptors using toggle-tryptophan-centered docking boxes. Atom-level binding-mode comparison between praziquantel and tolvaptan was performed by distance-based interaction analysis on the Mode-1 docked poses. To probe methodology generalization, the entire pipeline was reapplied to an unrelated protein target — secreted phospholipase A2 group IIA (sPLA2, PDB 1OXL), an enzyme implicated in snake-venom lethality — with varespladib (a clinically investigated pan-sPLA2 inhibitor) as a reference-anchor compound at every triage step and a five-move triage funnel applied to the 3,724-ligand library: druglikeness; cross-target Vina-bias diagnostic; ADMET; active-site catalytic-residue (His48 + Asp49) contact validation; decoy-distribution check; snake-vs-human selectivity counter-screen against human PLA2G2A (PDB 1KVO); and DiffDock-L rescore. Pipeline reproducibility was independently assessed by docking varespladib twice via independent paths (explicit reference dock and incidental presence in the ChEMBL library). Results: Three Tier-1 candidates were selected for full validation: praziquantel, olaparib, and glimepiride. Praziquantel — an FDA-approved antiparasitic on the WHO Essential Medicines List — emerged as the strongest cross-validated candidate: Vina score −9.75 kcal/mol (vs tolvaptan −8.43), DiffDock-L rank-1 confidence −1.01 (vs tolvaptan −1.35; both moderate confidence), and 75% binding-pocket residue overlap with tolvaptan including engagement of the W6.48 toggle switch. Critically, admet-AI predicted approximately half the DILI risk of tolvaptan (0.466 vs 0.918) along with reduced predicted hERG cardiac liability and lower lipophilicity. Olaparib showed cross-method binding agreement but matched tolvaptan's high predicted DILI; glimepiride was rejected by DiffDock disagreement. No prior literature was identified linking praziquantel to V2R or ADPKD. Counter-screen against V1a, V1b, and oxytocin receptors confirmed praziquantel as V2-preferring (V2R−V1A delta = −4.09 kcal/mol, more selective than tolvaptan's −2.82), with the caveat that the V2R−OTR delta (−2.05 kcal/mol) is the weakest of the three discriminations. Atom-level interaction analysis revealed that praziquantel engages the same critical V2R residues as tolvaptan (Q89, Q93, Q116, W6.48 toggle tryptophan, K113) with three hydrogen bonds versus tolvaptan's two, despite a structurally distinct pyrazinoisoquinoline scaffold (versus tolvaptan's benzazepine), providing a mechanistic rationale for the predicted hepatotoxicity differential. A targeted SAR scan around the praziquantel cyclohexyl substituent generated 37 designed analogs evaluated through the same full pipeline; three analogs beat praziquantel on at least two of four target criteria, most notably a spiro[2.5]octan-6-yl variant with a 64% reduction in predicted DILI (0.169 vs 0.466) and reduced OTR engagement at the cost of a modest decrease in predicted V2R binding strength. Methodology generalization to sPLA2 yielded 3,721 successful dockings (99.92%) over 6.07 hours of wall-clock screening, with varespladib (Vina −9.55 kcal/mol) ranking in the top ~15% of the library — broadly consistent with its established biological role. After the five-move triage funnel, a four-way validated eight-candidate shortlist emerged. The strongest snake-selective candidate is cipargamin, an antimalarial spiroindolone in Phase 2 trials (snake-sPLA2 Vina −10.77 kcal/mol, snake-vs-human selectivity Δ = −2.34 kcal/mol, His48-engaging active-site STRONG verdict, DiffDock-L confidence comparable to or better than varespladib); drospirenone, an approved oral progestin, emerged as an unexpected separate Tier-1 candidate with the best ADMET profile in the panel (predicted DILI 0.08, hERG 0.12). Two methodology disclosures arose from the sPLA2 work: (i) initial human-counter-receptor docking boxes centered on a 216-atom substrate-analogue ligand drifted 24 Å from the catalytic histidine, producing a uniform selectivity artifact caught only because the varespladib reference anchor produced a biologically-implausible result; the fix is to anchor docking boxes on a catalytic-residue side-chain atom (His48 NE2) when the bound inhibitor spans the substrate channel; (ii) DiffDock-L is not absolutely calibrated for small (~125-residue) secreted enzymes, so its sPLA2 confidences must be interpreted relative to a reference compound rather than against the published absolute threshold. Reproducibility-by-duplicate: varespladib's Vina Mode-1 scores agreed to 0.056 kcal/mol across two independent dockings (explicit reference dock and incidental ChEMBL-library presence) — well within Vina's normal run-to-run variance. Conclusions: We report computational evidence that praziquantel is a putative V2R-binding repurposing candidate with predicted hepatotoxicity substantially lower than tolvaptan and a V2-selectivity profile more favorable than the current standard of care; we additionally disclose three computationally-designed praziquantel analogs predicted to further improve on the parent compound's ADMET profile. The screening methodology generalizes to a second unrelated protein target (sPLA2), where it produced a four-way validated eight-candidate shortlist headed by cipargamin and drospirenone as secondary, separately-pursuable repurposing hypotheses for snake-envenomation sPLA2 inhibition; signal quality is target-dependent and two methodology disclosures (catalytic-residue-anchored docking boxes for small enzymes; reference-anchor sanity checks at every triage step) are reported in support of the generalization claim. These predictions warrant experimental validation: V2R radioligand binding, functional cAMP suppression assays (including parallel OTR counter-screening), ADPKD cell-line cyst models, and prospective synthesis-and-test of the designed analogs for the V2R arm; sPLA2 inhibition assays and in vivo snakebite-envenomation models for the sPLA2 arm. The full pipeline is reproducible on a consumer CPU/GPU laptop using public open-source tools. We make no clinical claims; in vitro and in vivo validation are required before any therapeutic implications can be assessed.