Rational Identification of Neuropilin-1 Inhibitors: From Deep-Learning to Drug Repurposing

Abstract

Neuropilins (NRP)—comprising neuropilin-1 (NRP1) and neuropilin-2 (NRP2)—are type I transmembrane glycoproteins with multifunctional extracellular regions. NRP1 functions as a multifunctional, non-enzymatic co-receptor that potentiates signalling from vascular endothelial growth factor (VEGF), Semaphorins, transforming growth factor beta (TGF-β) and other axes central to angiogenesis, tumour progression, viral entry and immune regulation. Despite this compelling biology, no potent NRP1 inhibitors have been approved, owing to broad physiological expression, multi-ligand/co-receptor complexity, and a shallow, highly polar b1-domain cleft that favours C-terminal arginine (C-end Rule; “CendR”) motifs. Inhibitors were sought that disrupt interactions between NRP1 and VEGF-A165, as well as other CendR-bearing ligands, by targeting hotspot residues within the b1 pocket. In this thesis, that gap is addressed by the rational identification of peptide and small-molecule inhibitors that can disrupt engagement at the NRP1-b1 pocket, and by the establishment of methodological principles for their discovery and validation. A multimodal framework was assembled in which deep-learning–guided generative design, AIdriven protein modelling, molecular dynamics simulations and structure-based virtual screening were integrated with biochemical and biophysical validations, followed by cell-based functional assays. In this way, the constraints of a shallow, highly polar and charged ligand-binding site could be engaged. The included side project also points out opportunities for selectivity and targeted delivery were informed by heterogeneity in NRP1 and NRP2 expression. Feasibility of CendR-independent peptide inhibition was demonstrated through designing de novo peptides that were generated by state-of-the-art tool, RFdiffusion. The peptides were designed to fit the NRP1-b1 pocket and engage hotspot residues identified by the canonical interaction between NRP1 and VEGF-A165. Their binding poses and apparent affinities were assessed, together with inhibitory potency against angiogenic and tumour cell phenotypes. Additionally, a disease-informed strategy was investigated, in which perturbed interfaces between Charcot–Marie–Tooth–associated aminoacyl tRNA synthetase mutants and the NRP1-b1 pocket were modelled to extract short binding segments. In parallel, small molecule inhibition was pursued through structure-based virtual screening of an FDA-approved library using focussed and blind docking with multiple algorithms. Although no definitive hits were confirmed experimentally, interaction hypotheses and prioritisation rules were refined to complement peptide development. Also another small molecule can be developed based on the identified compound. Finally, heterogeneous NRP1 and NRP2 expression across normal and tumour tissues was examined at single-cell resolution to identify cell types with pronounced expression. Co-expressed markers were also noted from differential gene-expression analyses within different cancer types, which can benefit development of a targeted therapy. Collectively, a discovery framework is delivered that couples deep-learning-guided peptide identification and repurposing-oriented screening with validation and disease-context mapping. With this ‘proof-of-concept’ study, the work provides a basis for subsequent optimisation and for translation of NRP1-targeted inhibitors toward therapeutic development.