Abstract 1161 Natural anosmia-inducing compounds for control of mosquito biting behaviors: Machine learning-assisted determination of structural determinants for ORco ligands antagonizing odorant receptor function

Insect odor receptors (ORs) are heteromeric ligand-gated cation channels composed of an obligatory, highly conserved receptor subunit, ORco, and one of many variable subunits, ORx, in as yet undefined molar ratios. When expressed alone in various cell culture systems, ORco forms homotetrameric channels gated by ORco-specific ligands acting as channel agonists. In previous studies, we identified small ligands acting ex vivo as specific ORco channel antagonists, orthosteric or allosteric relative to a known agonist binding site, and cause severe inhibition of the olfactory function in mosquitoes in vivo. Therefore, we sought to design a virtual screening protocol to identify candidate ORco antagonists that could be used as vector behavior control agents. Initially, based on the previous results, we developed a ligand-based pharmacophore describing the 3D arrangement of orthosteric antagonist features necessary for blocking ORco's biological response. The specific pharmacophore model was required to match all orthosteric input molecules, while keeping the number of false positives at a minimum. Four features were found to meet these requirements best, two hydrophobic centroids, one hydrogen bond acceptor and one hydrophobic atom. Then, a new collection of 49 volatile organic compounds (VOCs) of natural origin was both computationally and functionally screened. The pharmacophore's sensitivity was 86%, while its specificity was 57%, leaving room for improvement. To improve the pharmacophore's performance, we calculated a set of 2D descriptors for all pharmacophore hits and generated a support vector machine (SVM) capable of discriminating between true (ex vivo confirmed) orthosteric antagonists from false positive hits. The best SVM model, trained on a collection of 104 functionally-characterized natural compounds, included a topological descriptor encoding the branching of molecules, and a descriptor reflecting the extend of hydrophobic or hydrophilic effects on the surface area of molecules. The statistically estimated cross-validation out-of-sample misclassification rate of this SVM was 3.13%. The validity of employing the two-step in silico protocol to identify ORco orthosteric antagonists is currently under investigation on a third collection of 247 previously untested VOCs of natural origin with a combination of ex vivo activity inhibition and in vivo repellence assays. The first results are encouraging. Moreover, the generation of a second protocol to predict ORco allosteric antagonists is in progress. Supported by the Hellenic Foundation for Research and Innovation (H.F.R.I.) grant "3D-ORrco" (HFRI_FM17_637); the Competitiveness, Entrepreneurship and Innovation Operational Programme (NSRF 2014-2020) grant "OPENSCREEN-GR" (MIS 5002691); and the European Commission grant "LIFE CONOPS" (LIFE12 ENV/GR/000466).