Covariant Fitness Clusters Reveal Structural Evolution of SARS-CoV-2 Polymerase Across the Human Population

Abstract

Understanding the fitness landscape of viral mutations is crucial for uncovering the evolutionary mechanisms contributing to pandemic behavior. Here, we apply a Gaussian process regression (GPR) based machine learning approach that generates spatial covariance (SCV) relationships to construct stability fitness landscapes for the RNA-dependent RNA polymerase (RdRp) of SARS-CoV-2. GPR generated fitness scores capture on a residue-by-residue basis a covariant fitness cluster centered at the C487-H642-C645-C646 Zn2+ binding motif that iteratively evolves since the early phase pandemic. In the Alpha and Delta variant of concern (VOC), multi-residue SCV interactions in the NiRAN domain form a second fitness cluster contributing to spread. Strikingly, a novel third fitness cluster harboring a Delta VOC basal mutation G671S augments RdRp structural plasticity to potentially promote rapid spread through viral load. GPR principled SCV provides a generalizable tool to mechanistically understand evolution of viral genomes at atomic resolution contributing to fitness at the pathogen-host interface.

Profiling Genetic Diversity Reveals the Molecular Basis for Balancing Function with Misfolding in Alpha-1 Antitrypsin

“…The VSP analysis was performed as previously described 1, 41–43 using gstat package (V2.0) 66, 67 in R. VSP is built on Gaussian process regression (GPR) based machine learning. A special form of GPR machine learning that has been developed in geostatistics, Ordinary Kriging 68 , is used to model the spatial dependency as a variogram to interpolate the unmeasured value to construct the phenotype landscape for AAT.…”

“…To probe the molecular basis for the balance between misfolding and function of AAT, we measured the intracellular monomer and polymer levels, secreted monomer and polymer levels, and NE inhibitory activity for 75 AAT variants distributed across the protein sequence that include both pathogenic and benign variants found in the population. To address the complex folding dynamics likely differentially impacted by these variants, we applied a Gaussian process regression (GPR) based machine learning approach termed variation spatial profiling (VSP) 1, [41][42][43][44] .…”

Tracing genetic diversity captures the molecular basis of misfolding disease

“…To begin to understand the mechanism by which GRP94 modulation corrects the AAT fold on a residue-by-residue basis at atomic resolution, we applied our Gaussian process (GP)-spatial covariance (SCV) (GP-SCV) principled machine learning approach through variation spatial profiling (VSP) 1,[17][18][19][20][21] . GP-SCV principled relationships generated through VSP is based on a statistical paradigm used to find value in complex physical landscapes (see "Methods").…”

“…To address the central problem of information flow from the genome to the proteome in human biology in the context of the complex folding dynamics impacted by heat shock protein 90 (Hsp90) family members 16 that constitute a central ATP-dependent chaperone system in the cell, we developed a Gaussian process (GP) regressionbased machine learning 1,[17][18][19][20][21] approach to learn the basal state of the protein fold and its response to proteostasis management. GP can use a sparse collection of variants across the worldwide population and their associated phenotypes as a collective input to generate spatial covariance (SCV) maps of the response of every residue in the protein sequence and its residue-residue responses to environmental changes, including pharmacological intervention 1,[17][18][19][20][21] . GP-SCV principled modeling captures as a collective the variant changes that report on the evolutionary trajectory of the entire wild-type (WT) protein fold at atomic resolution that impacts health and disease of the host in response to the environment and natural selection 1,[17][18][19][20][21] .…”

“…GP can use a sparse collection of variants across the worldwide population and their associated phenotypes as a collective input to generate spatial covariance (SCV) maps of the response of every residue in the protein sequence and its residue-residue responses to environmental changes, including pharmacological intervention 1,[17][18][19][20][21] . GP-SCV principled modeling captures as a collective the variant changes that report on the evolutionary trajectory of the entire wild-type (WT) protein fold at atomic resolution that impacts health and disease of the host in response to the environment and natural selection 1,[17][18][19][20][21] .…”