It is well established that protein structure is more conserved than sequence on an evolutionary timescale.  This fact allows functional inferences to be drawn from proteins that share the same fold, even when their sequence similarity is quite low. In the case of B cell receptors (BCRs), which consist of two protein chains (“heavy” and “light”), the relationships between sequence, structure and function are more complex. All BCRs look similar globally but differ in the details of their antigen-binding regions. These differences arise from combinatorial arrangement of genes and random mutations that occur upon antigen encounter. Traditionally, bioinformatics analysis of BCR sequences involves clustering sequences that arise from the same heavy and light chain genes into “lineages” in order to group those that target a common antigen.  The diversity of BCRs has been estimated to exceed 1013 in humans, which means that it is very unlikely that this approach can be extended to multiple donors, even after exposure to the same antigen. Our hypothesis is that structural clusters of BCRs arising in different donors are more likely to target the same antigen than BCRs in different clusters.  Indeed, x-ray crystallographic studies have demonstrated that structurally similar BCRs targeting common antigens can arise in different donors using different genes. In the last year high-throughput sequencing methodologies have emerged that can deliver on the order of 104 paired (heavy-light chain) BCR sequences in a single experiment. Clearly, x-ray crystallography, which typically required ~6 months per structure, will not be able to cope with so many emerging BCR sequences in a high-throughput manner. Thus, there is a strong motivation to leverage structural bioinformatics in order to infer structure and functional similarities. Using a novel alignment and 3D rendering method (Repertoire Builder), we could reduce the time required to build an atomic-resolution BCR model to under six seconds, corresponding to over 17,000 atomic resolution models per day on a single CPU. BCR and T cell receptor (TCR) benchmarks demonstrate that Repertoire Builder is significantly more accurate than any tested web sever. We further show that human BCRs in multiple donors acquired post flu vaccination indeed display strong structural convergence. Even more striking, subsets of human BCRs exhibit highly significant structural similarities to BCRs acquired from vaccinated mice. These findings suggest that BCR modeling, in combination with high-throughput sequencing may be useful to identify diverse sequences targeting common antigens across donors and across species.
 

Prof. Standley received his PhD in Chemistry from Columbia University in 1998. He then joined Schrodinger, Inc. where he worked as a scientific software developer for five years. In 2003 he moved to the Institute for Protein Research, Osaka University as a Senior Scientist. He joined the Immunology Frontier Research Institute (IFReC) as a Principal Investigator in 2008 and, after a two-year cross-appointment at Kyoto University’s Institute for Virus Research, became a Professor full time at the Research Institute for Microbial Diseases in 2016.