Structurally Informed Fitness Landscapes for Surveillance of Emerging PRRSV Variants
Antibodies play a central role in neutralizing pathogens through direct interference with viral entry and recruitment of effector immune cells. However, viruses employ sophisticated escape mechanisms to evade these defenses, primarily through mutations in surface glycoproteins that reduce antibody binding affinity or alter critical functional domains. Antibody escape remains a formidable challenge in drug design efforts for Porcine Reproductive and Respiratory Syndrome Virus (PRRSV), a pathogen notorious for its rapid evolution and structural plasticity. Here we present EscaPRRS (Escape scoring for PRRS virus), a Bayesian variational autoencoder trained on ESM-2 protein language model embeddings to predict the fitness landscape and escape propensities of 52,622 mutants (recorded over 10 years) of the immunodominant GP5 glycoprotein encoded by PRRSV ORF5 gene. Unlike conventional models that estimate escape propensity only from sequence information, EscaPRRS circumvents the need for extensive alignments, integrating contributions from surface accessibility and biochemical dissimilarity at the binding interface. Our escape propensity scores demonstrate reliable structural and biological fidelity, with EscaPRRS scores correlating with binding affinities on seven different porcine receptor proteins (Pearson r = 0.74). Notably, EscaPRRS captures seasonal trends in immune evasion, highlighting its applicability in forecasting and surveillance of emerging/re-emerging PRRSV variants. IMPORTANCEPorcine Reproductive and Respiratory Syndrome Virus (PRRSV) remains the most economically detrimental illness for swine products, causing more than $1.2 billion in annual production loss in the United States, with indirect implications to food security and human health. PRRSV infection is known to severely affect porcine alveolar macrophages (PAMs), causing respiratory difficulties, following blockage of inflammatory signals that aids easy viral reproduction in the host cell. Identifying critical amino acid mutations at the highly variant GP5 glycoprotein attached to the viral cell membrane provides information on structural features linked to antigenic diversity and antibody neutralization, that potentially lead to clinical outbreaks in sow farms. Our effort employs machine learning approaches to learn patterns from a large number of sequences to map these critical domains to three-dimensional structures to score them for antibody escape tendencies. This greatly enhances our understanding of receptor and antibody binding mechanisms in PRRSV GP5.