Bicyclic peptides : targeting toxins and pathogens to accelerate anti-infective discovery

Abstract

The pore forming toxins pneumolysin and а-hemolysin play a key role in the pathogenicity of Streptocoocus pneumonia and staphylococcus aureus respectively. We used the phage display platform at BicycleTx Ltd to screen these toxins for potential bicyclic peptide inhibitors. Despite multiple efforts with multiple different versions of target protein no bicyclic peptide binders to pneumolysin could be identified. Selections against а-hemolysin were more successful with bicyclic peptides which bound to and inhibited the pore forming activity of а-hemolysin being identified. An extensive optimisation procedure was undertaken including the production of affinity maturation libraries as well as testing non-canonical amino acids. A 1.98Å co-crystal structure gave useful insights into the epitope and mechanism of binding as well as interesting similarities to a previous α-hemolysin antibody. The final optimised hit of BCY00034709 was produced and its mechanism of inhibition characterised. We were able to show that this novel bacterial-anti-toxin is capable of protecting red blood cells as well as human lung cells from the toxic effects of а-hemolysin in vitro. We also describe the discovery of novel bicyclic peptide inhibitors of the type 1 signal peptidase of staphylococcus aureus produced by bicyclic peptide phage display selections. We show that these bicyclic peptides are capable of inhibiting SpsB in a FRET assay which mimics the proteins endogenous function. Through the process of affinity maturation this produced our most potent hit which had an IC50 of 22nM. Computational modelling and assay data provided insights into the mechanism of action of BCY00041582 as well as the potential role of crucial residues. Despite this potent inhibition in vitro no activity against a clinical S. aureus strain was seen that could be attributed to SpsB inhibition and we explore potential reasons for this asymmetry as well as a possible strategy to bridge this gap. The limitations of phage display selections against SpsB and detergent solubilised membrane proteins as a whole are also discussed. Focusing on the production of the false positives we observed that show S. aureus growth inhibition whilst not displaying any inhibition of SpsB, along with the potential mechanism and strategies for overcoming this. Keywords: Graph limit theory, Complex networks, Machine learning, Deep learning, Persistent homology, Topological data analysis, Statistical mechanics, Condensed matter physics, Soft matter, Molecular dynamics simulations.