flexIble proteiN desigN and accUratE biNDing estimatiOn – Innuendo

Computational protein design (CPD) consists of designing proteins accomplish- ing certain tasks. The case of non-covalent interactions requires designing molecules with a specified binding affinity range. However, this design problem is especially challenging due to the exponential size of the design space in terms of amino acid choices, the role of flexibility which determines interaction affinity and specificity, and the difficulty to reliably estimate the binding affinity, a subtle thermodynamic quantity. Methods-wise, we ambition to advance the state-of-the-art of flexible computational protein design and binding affinity estimation, which raise difficult high dimensional geometric problems. Our methodological contributions will encompass three tasks.

The first focuses on the design of so-called movesets (proposals in the realm of Monte Carlo methods) to generate conformations of whole proteins. Our methods will combine recent insights on the geometry of protein backbones in torsion angle space, and constraints inherent to the geometry of side chains. The second concentrates on novel sampling algorithms to estimate reliably the partition functions involved in the definition of the dissociation constant. The punchline will consist of embedding our novel movesets into state-of-the-art Monte Carlo Markov Chain methods computing volumes and densities of states in high-dimensional spaces. The third one deals with the design of a novel CPD pipeline for flexible protein design, based on the previous two tasks. Overall, the novel algorithms will make it possible to explore a larger design space, while at the same time reducing the experimental burden, via superior binding affinity estimates. All methods will be made available in the Structural Bioinformatics Library (http://sbl.inria.fr), which provides both low level algorithms and applications for biologists and biophysicists.

These methods will be used to develop high affinity neutralizing (and protective) nanobinders against SARS-CoV-2 and others sarbecoviruses. With up to 6.8 million deaths worldwide in three years, the COVID-19 crisis has indeed demonstrated the necessity to better combat such respiratory viruses. Since early 2020, numerous studies have developed neutralizing nanobinders (monoclonal antibodies, nanobodies, artificial and miniproteins) targeting the spike of SARS-CoV-2 to block infection. However, the main limitation of these molecules is their relatively narrow recognition spectrum, and the lack of recognition of new viral variants. Taking advantage of the methods developed in the project, our objective is to select artificial proteins able to recognize with high affinity the receptor binding domain (RBD) of a large spectrum of sarbecoviruses. These reactives should be able to neutralize these viruses efficiently and to be tested as antivirals in an animal infection model, the hamster. We will thus use the previous selection of artificial proteins (named alphaReps) that we selected against SARS-CoV-2, displaying affinity in the nanomolar range to the spike and able to neutralize a D614G SARS-CoV-2 strain and protect hamsters from the infection. We will determine the X-ray and cryo-EM structures of several neutralizing alphaReps - RBD or spike of SARS-CoV-2 to define the residues defining the binding surfaces. In parallel, we will measure the affinity between our neutralizing alphaReps and RBDs representing the sarbecoviruses diversity. These data will be used as validation tools for the algorithms generated in the project for binding affinity estimates.

By combining into a virtuous loop our novel methods, as well as experiments for structure (cryo- EM, X-ray crystallography) and thermodynamics (affinity measurements), we will design broad spectrum nanobinders against circulating sarbecoviruses, also harboring limited sensitivity to immune escape mutations.