Email updates

Keep up to date with the latest news and content from BMC Bioinformatics and BioMed Central.

Open Access Highly Accessed Research article

Predicting HIV-1 broadly neutralizing antibody epitope networks using neutralization titers and a novel computational method

Mark C Evans, Pham Phung, Agnes C Paquet, Anvi Parikh, Christos J Petropoulos, Terri Wrin and Mojgan Haddad*

Author Affiliations

Monogram Biosciences Inc., 345 Oyster Point Blvd., South San Francisco, CA 94080, USA

For all author emails, please log on.

BMC Bioinformatics 2014, 15:77  doi:10.1186/1471-2105-15-77

Published: 19 March 2014

Abstract

Background

Recent efforts in HIV-1 vaccine design have focused on immunogens that evoke potent neutralizing antibody responses to a broad spectrum of viruses circulating worldwide. However, the development of effective vaccines will depend on the identification and characterization of the neutralizing antibodies and their epitopes. We developed bioinformatics methods to predict epitope networks and antigenic determinants using structural information, as well as corresponding genotypes and phenotypes generated by a highly sensitive and reproducible neutralization assay.

282 clonal envelope sequences from a multiclade panel of HIV-1 viruses were tested in viral neutralization assays with an array of broadly neutralizing monoclonal antibodies (mAbs: b12, PG9,16, PGT121 - 128, PGT130 - 131, PGT135 - 137, PGT141 - 145, and PGV04). We correlated IC50 titers with the envelope sequences, and used this information to predict antibody epitope networks. Structural patches were defined as amino acid groups based on solvent-accessibility, radius, atomic depth, and interaction networks within 3D envelope models. We applied a boosted algorithm consisting of multiple machine-learning and statistical models to evaluate these patches as possible antibody epitope regions, evidenced by strong correlations with the neutralization response for each antibody.

Results

We identified patch clusters with significant correlation to IC50 titers as sites that impact neutralization sensitivity and therefore are potentially part of the antibody binding sites. Predicted epitope networks were mostly located within the variable loops of the envelope glycoprotein (gp120), particularly in V1/V2. Site-directed mutagenesis experiments involving residues identified as epitope networks across multiple mAbs confirmed association of these residues with loss or gain of neutralization sensitivity.

Conclusions

Computational methods were implemented to rapidly survey protein structures and predict epitope networks associated with response to individual monoclonal antibodies, which resulted in the identification and deeper understanding of immunological hotspots targeted by broadly neutralizing HIV-1 antibodies.

Keywords:
HIV-1 antibody; Thick patch analysis; Bioinformatics algorithms; Boosting algorithm; Machine learning; Neutralization; in-silico epitope mapping; Epitope networks; Structural mapping; Sequence and structure analysis