TL;DR
This paper introduces a new computational method, SPACE, that uses antibody structural modeling to group antibodies that bind to the same epitope, even if their sequences are very different. This approach reveals a higher degree of convergent immune responses to coronaviruses than previously understood and offers a powerful way to functionally annotate large antibody datasets.
What Problem Does This Paper Address?
Identifying which antibodies bind to the same "epitope" (the specific part of a pathogen an antibody targets) is crucial for developing effective antibody therapies and understanding immune responses. Traditional methods, like "clonotyping," rely on sequence similarity and can miss antibodies with similar binding functions but different genetic origins. This is a significant limitation because antibodies from different lineages can still converge on the same epitope, especially for highly conserved targets like those on coronaviruses. The sheer volume of newly identified antibodies, particularly in response to the COVID-19 pandemic, makes experimental epitope mapping prohibitively time-consuming and expensive. Therefore, there's a need for computational methods that can accurately group antibodies by their functional binding sites, even when sequence similarity is low.
Methods
The core of this work is the development of the Structural Profiling of Antibodies to Cluster by Epitope (SPACE) algorithm. The process involves several key steps:
- Antibody Modeling: For antibodies with known sequences but no solved structures, their variable (Fv) regions were modeled using the ABodyBuilder tool. This tool leverages the FREAD software to predict CDR loop structures based on homologous templates. A subset of 1,500 high-quality models was selected for further analysis.
- Structural Clustering: The SPACE algorithm first clusters antibodies based on the lengths of their six CDR loops. Then, it performs a greedy clustering based on the pairwise structural root-mean-square deviation (RMSD) of the CDR regions between antibody models. This groups antibodies with similar structural Fv regions.
- Domain Consistency Classification: Structural clusters were classified as "domain-consistent" or "domain-inconsistent" based on the epitope metadata of the antibodies within them. A cluster was considered domain-consistent if all its members bound to the same antigen and/or the same domain of that antigen.
- Clonotyping (for comparison): Lenient VH-only and Fv clonotyping protocols were also performed to compare the performance of structural clustering against sequence-based methods.
Datasets
The primary dataset used in this study is the Coronavirus Antibody Database (CoV-AbDab). The version used was timestamped to March 11, 2021, and contained approximately 2,304 antibodies and 420 nanobodies that bind to coronaviruses.
- Solved Structures: The study also analyzed 60 experimentally solved X-ray crystal structures of antibodies and nanobodies bound to SARS-CoV-2 antigens, primarily the spike receptor-binding domain (RBD).
Key Insights and Ideas
- Structural Similarity Trumps Sequence Similarity for Epitope Binding: The study strongly suggests that antibodies with similar 3D structures are more likely to bind to the same epitope, even if their amino acid sequences are quite different. This challenges the traditional reliance on sequence-based methods like clonotyping.
- Convergent Evolution is More Pervasive Than Previously Thought: The high percentage of "domain-consistent" structural clusters indicates that immune responses to coronaviruses exhibit significant convergence. Different lineages of antibodies are evolving to recognize the same critical viral epitopes.
- SPACE Algorithm's Predictive Power: The SPACE algorithm demonstrates a high accuracy (92%) in grouping antibodies that bind to consistent domains, suggesting its utility for predicting epitopes of uncharacterized antibodies.
- Functional Annotation of Large Datasets: This structural approach offers a scalable way to functionally annotate vast antibody databases, providing insights that sequence-based methods miss.
- Identifying "Public" Antibodies: The method can identify groups of antibodies from different individuals and lineages that target the same immunodominant epitopes, which are of great interest for vaccine design and therapeutic development.
Key Results
- High Accuracy of Structural Clustering: 92% of multiple-occupancy structural clusters identified by SPACE were "domain-consistent," meaning antibodies within these clusters were predicted to bind to the same antigen domain.
- Identification of Sequence-Dissimilar, Functionally Similar Antibodies: The analysis of solved structures revealed instances where antibodies with low CDRH3 sequence identity (e.g., 43%) bound to the same epitope with similar modes.
- Broad Coverage of CoV-AbDab: The study successfully modeled and clustered 1,500 Fv sequences from CoV-AbDab, representing a significant portion of the available antibody data.
- Pan-Coronavirus Epitopes: Approximately 37.5% of the structural clusters contained antibodies shown to bind to multiple coronaviruses, highlighting conserved vulnerabilities.
- Bridging Species Barriers: The study identified a structural cluster (SC62) that grouped human and murine antibodies targeting the SARS-CoV-2 RBD, suggesting epitopes can be targeted by antibodies from different species.
Limitations and Future Work
- Model Accuracy: The accuracy of the structural clustering relies on the quality of the homology models. While efforts were made to select high-quality models, inherent limitations in modeling exist.
- Epitope Metadata Completeness: The accuracy of "domain-consistent" classification depends on the completeness and accuracy of the epitope metadata available in CoV-AbDab. Some "false positives" might be due to inaccuracies in experimental epitope labels.
- Epitope Dependence: The degree to which antibody structure needs to be conserved to engage the same epitope is likely epitope-dependent and can be influenced by factors like glycosylation.
- Future Work: The authors suggest that as more antibody structures become available, the accuracy of homology models and epitope binning will improve. They also anticipate advancements in single-domain protein modeling techniques will further enhance antibody structure prediction and epitope profiling. The SPACE algorithm is presented as a versatile tool applicable to other disease-focused antibody datasets.
Summary updated