Prediction of VRC01 neutralization sensitivity by HIV-1 gp160 sequence features

摘要

Author summary The two Antibody Mediated Prevention (AMP) clinical trials are testing whether intravenous infusion of VRC01 (an antibody that can neutralize most HIV-1 viruses) can prevent HIV-1 infection. Since the outer envelope (Env) protein of HIV-1 is highly genetically diverse, the AMP trials will evaluate in an amino acid sequence sieve analysis whether VRC01 prevents infection differentially depending on Env amino acid features of exposing viruses. To maximize power of sieve analysis, the number of amino acid features tested should be limited to those most likely associated with whether the virus is sensitive to neutralization by VRC01. We used machine learning to analyze a database of 611 HIV-1 Envelope pseudoviruses, with data on how well VRC01 neutralizes each pseudovirus, to identify models that best predict neutralization sensitivity from the amino acid features and to identify the most predictive features. We identified models that could predict HIV-1 sensitivity (as opposed to resistance) to VRC01 very well, and found that several amino acid residues in Env locations where both VRC01 and the CD4 receptor bind were important for making correct predictions. Our modeling approach will enable a focused AMP sieve analysis and may be useful for studying the use of VRC01 in the treatment of HIV-infected persons.