# Structural Constraints Acting on the SARS-CoV-2 Spike Protein Reveal Limited Space for Viral Adaptation

## TL;DR

This study shows that the SARS-CoV-2 spike protein has limited ability to adapt due to strict structural constraints, despite rapid mutations.

## Contribution

The paper introduces a machine learning model combining structural and evolutionary constraints to assess substitution viability in SARS-CoV-2.

## Key findings

- Structural constraints on the SARS-CoV-2 spike protein have changed little despite significant phenotypic evolution.
- Signature mutations in variants of concern remain under constant structural constraints.
- A machine learning model confirms that substitution viability is largely unaffected by variant structure differences.

## Abstract

The SARS-CoV-2 pandemic resulted in an unprecedented scientific response. The scale of global genome sequencing, protein structural determination, and targeted studies of variant dynamics has resulted in a unique dataset, providing a valuable resource for studying viral evolutionary dynamics. Previous analysis of SARS-CoV-2 evolution has revealed apparently saltatory dynamics, with viral variants arising following evolutionary jumps without genetic intermediates represented in the sequence database. We utilize rich SARS-CoV-2 datasets to interrogate the role of protein structural constraint in SARS-CoV-2 evolution and whether saltatory dynamics result from the spike protein accessing previously nonviable sequence space. We apply multiple computational predictors of structural constraint across different structural backgrounds and assess how constraint has changed during SARS-CoV-2 variant evolution. These predictions are validated using substitution data from the SARS-CoV-2 global sequence database. We find that the structural constraint experienced by specific sites has undergone limited change, despite significant phenotypic evolution of the SARS-CoV-2 S protein. The structural constraints acting on signature mutations of variants of concern remain constant regardless of which viral variant structure is used to make predictions. We also develop a machine learning model to assess substitution viability, combining predictors of evolutionary constraint with information about local structural context. This confirms our conclusions, with model performance largely unaffected by the use of different viral variant structures. These results suggest that despite its rapid rate of mutation, the SARS-CoV-2 S protein is subject to strict structural constraints and exhibited limited genomic plasticity following zoonotic transmission into the human population.