Collective fluctuations underlying nanobody inhibitory activity targeting B. anthracis S-layers revealed by multiscale simulations

Nanobodies (Nbs) that depolymerize bacterial surface-layers (S-layers) offer a route to antivirulence therapeutics, but their mechanisms have been difficult to infer from binding structures alone. In B. anthracis, several Nbs bind to Sap, the S-layer protein that assembles into a paracrystalline lattice surrounding the cell. Despite similar binding poses and sequences, only a subset of these (inhibitory) Nbs induce disassembly of the protein lattice, ultimately abrogating pathogenicity. In this study, we leveraged multiscale simulations to test whether representative Nb-induced fluctuations local to the binding site are sufficient to reproduce and explain lattice-scale depolymerization. Using a divide-and-conquer strategy, we first developed a bottom-up coarse-grained (CG) model of the multidomain Sap monomer. We then compared machine learning (ML) and information theoretic approaches to identify Nb-induced collective fluctuations that are predictive and potentially causative for depolymerization. We found that motions encoding Nb rigidification and partial clamping of the binding site instigate early-stage Sap depolymerization when propagated into lattice-scale computational depolymerization assays. Furthermore, the model informed by correlation-grouped ML analysis correctly reproduced both inhibitory and non-inhibitory phenotypes for 10 out of 12 (Nb)-Sap systems. Combined time-resolved defect and strain analyses revealed that inhibitory Nbs cooperatively apply a critical amount of tensile stress that destabilizes Sap-Sap interfaces parallel to the direction of strain and proximal to the Nb binding site, thereby supporting a mechanism in which local, Nb-imposed fluctuations propagate into lattice-scale mechanical instability. More broadly, this work demonstrates how multiscale simulations combined with ML analysis can test whether molecular-scale conformational signatures are sufficient to drive emergent phenotypes in large protein assemblies. In the future, this general approach can be adapted for mechanistic study and subsequent rational design of therapeutics that rely on dynamical interventions of protein virulence factors, such as through rigidification or assembly-disrupting modes of action.