The Adaptive Immune Receptor Repertoire Knowledge Commons: An invitation to the community

Abstract

Sequencing data elucidating adaptive immune receptor (AIR) repertoires, genomic loci encoding AIR genes, and AIR antigens are hosted in siloed repositories, limiting integrative analyses. We are in the process of creating the AIR Repertoire Knowledge Commons (AKC) by merging data from community-backed repositories and applying knowledge-generation algorithms to integrate and enrich data and metadata.

Introduction

The antigen-recognition system of B and T cells relies on a highly diverse repertoire of receptors, called adaptive immune receptors (AIRs), that are the foundation of both the high degree of specificity exhibited by B and T cell responses and their ability to respond to a vast array of distinct antigens. Characterizing AIRs and AIR repertoires is key to the development of a wide range of clinical applications. These include an FDA-approved assay for monitoring minimal residual disease (https://www.clonoseq.com) and many other applications with clinical potential in the near to middle term, such as predicting transplant success and responses to cancer immunotherapy, designing vaccines and diagnostics, and developing therapeutic antibodies.

AIR repertoires are highly diverse because the genes encoding AIRs are somatically generated through a DNA recombination process that combines multiple, germline-encoded genes to produce an essentially unique gene sequence in each developing B and T cell. The total potential diversity is tremendous, for example, ~10¹⁸ unique αβ T cell receptors (TCRs).¹ There is also significant interindividual diversity as a result of extensive germline polymorphism and structural variation in AIR loci.²

Sequence data elucidating three key facets of AIRs and AIR repertoires are freely, publicly available and have contributed to basic science discoveries and the development of diverse clinical applications. The Adaptive Immune Receptor Repertoire (AIRR) Community formed in 2015 to develop the standards and other resources needed for making these data findable, accessible, interoperable, and reusable (FAIR).³ The three key types of sequencing data and corresponding AIRR Community-backed data repositories are as follows:

1. Rearranged AIR genes and AIR repertoires: AIR repertoire sequence (AIRR-seq) data elucidate the nature of rearranged AIR genes and the diversity and clonal composition of AIR repertoires. These data are critical for understanding the AIR repertoire properties associated with health and for learning the AIR repertoire signatures associated with specific immune exposures, such as specific diseases or vaccinations. The AIRR Community established the AIRR Data Commons, a distributed system of repositories for publicly sharing AIRR-seq data (https://docs.airr-community.org/en/stable/api/adc.html). The AIRR Data Commons can be queried via an application programming interface (API)⁴ or either of two graphical interfaces (GUIs): the VDJServer Community Data Portal⁵ or the iReceptor Gateway.⁶ The AIRR Data Commons currently contains >5.25 billion rearranged AIR gene sequences from >10,000 AIR repertoires.

2. AIR germline genes and loci: AIR germline gene and locus sequence data elucidate the complex polymorphic and structural variation of AIR loci. These data are critical for understanding disease or immune-response associations with AIR allelic, non-coding, and structural variants. Additionally, complete and accurate reference databases of AIR germline genes are critical for the precise analysis of AIRR-seq datasets. The AIRR Community promotes (1) the development of complete and accurate sets of reference germline AIR genes and (2) the accurate inference and reporting of germline genes identified from repertoires. It has produced two databases. The Open Germline Receptor Database (OGRDB) is a database of AIR germline alleles with supporting evidence and review history.⁷ It also hosts reference germline sets—collections of identified alleles for particular species or subspecies. OGRDB currently contains 421 human and 2,450 mouse reference sequences compiled into 48 reference sets. VDJbase contains individual- and population-level AIR genotypes and haplotypes inferred from AIR repertoires and from AIR locus sequencing data, along with statistics on allele occurrence.⁸ VDJbase currently contains details of 3,765 allelic variants, 2,441 genotypes, and 854 haplotypes collected from 2,441 human B and T cell receptor loci.

3. Antigen specificity: Data on AIR antigen specificity elucidate the binding and specificity principles underpinning AIR-antigen interactions. These data are critical for designing interventions, such as vaccines, therapeutic antibodies, and cellular therapies, such as CAR T cells. The Immune Epitope Database (IEDB) catalogs experimental data on AIR epitopes, the parts of antigens bound by AIRs.⁹ The IEDB predates the AIRR Community, but its developers have been active contributors to AIRR Community efforts since the community’s inception. The IEDB currently holds data on >1.5 million epitopes from almost 75,000 antigens and almost 200,000 AIRs. The Immune Receptor Antigen Database (IRAD; https://irad.research.cchmc.org) curates AIR sequences with their antigen specificity. For experimental reasons, AIR specificity is often determined to the resolution of only the antigen but not the epitope. Thus, the IRAD fills a receptor-specificity curation gap not met by the IEDB. The IRAD currently contains entries for 138,013 human B cell receptor sequences.

For all three data types, the amount of data generated and available is continually increasing.

Although these resources have made significant contributions to advancing scientific knowledge within their respective domains, they were designed for using each data type independently of the other two. Thus, it is challenging to utilize them for cross-cutting research questions that require the integration of data from two or more of these resources. For example, the AIRR Data Commons cannot be queried for AIR repertoires containing AIRs with known antigen specificities, nor can associations between antigen specificities and germline alleles be easily discovered. Addressing such cross-cutting questions requires querying multiple repositories, downloading the data, and developing and implementing data-integration algorithms. This requires extensive offline work not easily accomplished at the full scale of these resources, leading individual research groups to implement solutions for specific subsets of the data and causing significant duplication of effort and inconsistency from study to study in how the data are processed and integrated. Additionally, many investigators lack the computational expertise and/or resources to perform the processing and integration.

We believe there is tremendous opportunity to accelerate research and discovery in areas related to B and T cell responses and AIR-based clinical intervention through the creation of a comprehensive knowledge commons with a full-scale integration of these data. Therefore, with sponsorship from the National Institute of Allergy and Infectious Diseases, we are in the process of creating the AIRR Knowledge Commons (AKC; https://airr-knowledge.org). The foundation of the AKC will be data integrated from these existing, community-supported data repositories and harmonized around a common data model based on widely adopted ontologies and data standards. The AKC will apply existing and novel cross-cutting knowledge-generation algorithms to enrich the integrated data and metadata with new cross-cutting relationships and inferred knowledge. New data and knowledge will be continually added, as the AKC strives to assemble the critical mass of data required for developing highly accurate predictive algorithms for long-standing questions of critical importance (e.g., predicting AIR specificity) and to ask questions across a large and diverse set of subjects with a variety of health and disease phenotypes. The AKC will enable the query, exploration, analysis, visualization, and download of the integrated data and associated knowledge via APIs and GUIs. To ensure that the AKC addresses the scientific needs of a broad user community, we invite community participation via the submission of data and use cases and via feedback on usability and functionality.

Data harmonization and integration via widely adopted standards

To enable data integration across the three domains, we are establishing an encompassing set of data and metadata standards and defining a common data model (CDM). We are relying on existing community standards, ontologies, and other widely used terms with standardized identifiers (IDs) as much as possible. In particular, we are relying on the AIRR Community standards^4,10 and ontologies in the OBO Foundry (https://obofoundry.org). Using ontologies to establish a common semantic model has many benefits, particularly the creation of a data-type- and application-independent representation of the domain, providing interoperability with other resources using the same ontologies, as well as extensibility in the future. The CDM will provide an integrated view of the domain and be implemented in a single, formal representation in the Linked Open Data Modeling Language (LinkML) (https://github.com/linkml/linkml), which supports exports to other formats. All specifications and code will be made publicly available so that any existing or future repositories with relevant data can support the integration of their data and metadata into the AKC.

Cross-cutting knowledge generation

We will enhance the value of the integrated data and metadata by generating new cross-cutting relationships to drive increased connectivity and deeper data integration, as well as the inference of new knowledge. We have identified four high-priority areas for early implementation:

1. Inferring novel AIR germline gene alleles, genotypes, and haplotypes from all appropriate AIR repertoires. We will infer AIR germline gene alleles, genotypes, and haplotypes for all repertoires with sufficient sequence length and coverage by using, e.g., the method reported by Gadala-Maria et al.¹¹ This will rapidly increase the rate at which new AIR alleles, genotypes, and haplotypes are discovered and made available to the community, accelerating the discovery of AIR variants with disease and other associations (e.g., see Collins et al.¹²). The significant increase in the number of available alleles, genotypes, and haplotypes will greatly enhance our knowledge of population associations, including with genetic ancestry, geographic location, and health or disease state.

2. Relating AIRs to epitope and antigen specificity. We will relate orphan AIRs, i.e., AIRs lacking specificity information from the AIRR Data Commons, to AIRs with known specificity from the IEDB or IRAD. We will initially relate orphan AIRs to epitopes and antigens when orphan and non-orphan AIRs share V and J germline allele assignments and identical complementarity determining region 3 (CDR3) amino acid sequences. A full-length sequence can also be used and could be particularly important for B cell receptors (BCRs) and antibodies (Abs), which are subject to somatic hypermutation. We might also use various metrics of sequence similarity, as well as consider algorithms for predicting specificity. The key will be identifying the best metrics and threshold criteria and providing AKC users with metadata to guide interpretation. Relating AIRs from the AIRR Data Commons to specificity will improve our understanding of specificity co-occurrence in a single repertoire, how that might shape and be shaped by repertoire features (such as diversity and clonality), and the role of the subject’s genotype and/or haplotype in determining the space of possible specificities for an individual. These associations will further our understanding of how specificity patterns relate to genetic ancestry and health or disease states. Although the percentage of orphan AIRs that initially map to specificity might be small, it will grow quickly as the data holdings for all contributing repositories rapidly increase.

3. Calculating probabilities of AIR generation by using the CDR3 sequence and AIR genotype and haplotype information. Using tools such as OLGA, we will calculate the probability that any given AIR sequence is generated through rearrangement during the AIR gene recombination process.¹³ The utility of this probability has been demonstrated in various contexts, such as identifying disease-specific AIR.

4. Characterizing AIR overlap across all repertoires in the AKC and the cross-repertoire occurrence frequency for each AIR. We will characterize, at an unprecedented scale, public clone frequencies, as well as the frequency and nature of the cross-subject appearance of highly similar but non-identical AIRs.

We will closely monitor the field for new algorithms, and we invite community feedback on additional algorithms and types of knowledge for implementation and prioritization.

Anticipated functionality and an example use case

The AKC will contain a continuously updated knowledgebase of integrated data and computationally derived new knowledge, enabling (1) cross-data-type browsing; (2) querying over the integrated data, metadata, and knowledge via APIs and GUIs; and (3) the retrieval of integrated data, metadata, and knowledge in standardized, parsable forms ready for state-of-the-art computational tools. Example cross-cutting research questions that the AKC will support are as follows:

1. Specificity of interest: Users will be able to query the AKC for all AIRs with a specificity of interest, such as influenza hemagglutinin glycoprotein epitopes, and then analyze repertoires containing the relevant AIRs to identify the corresponding germline alleles and determine which of the specific AIRs have the highest probability of being generated.¹³ This could inform vaccine design by identifying epitopes against which effective immune responses can be most readily elicited in the largest number of people and in diverse populations.

2. Germline allele associated with a disease: Users will be able to query the AKC for a germline allele with a known disease association (e.g., Parks et al.¹⁴), determine whether the allele is present in AIRs with known (or predicted) specificities, and examine the allele’s expression and functional patterns. For example, is the allele overrepresented in certain genetically related populations? Is the allele expressed in healthy individuals or individuals with another disease phenotype? Are B or T cells bearing AIRs derived from the allele clonally expanded? In which cellular subsets are these AIRs observed? Which antibody isotypes or somatic hypermutation patterns are associated with BCR alleles? Are B or T cells bearing AIRs derived from the allele overrepresented in specific tissues or at specific timepoints after an immune exposure or therapy?

3. AIR repertoires from individuals with a particular phenotype: Users will be able to query the AKC for AIR repertoires from subjects with a particular phenotype, such as lupus or asymptomatic SARS-CoV-2 infection, and determine whether particular germline alleles, genotypes, or haplotypes or particular antigen specificities are overrepresented. Results of interest could be followed by questions from examples 1 and 2.

Research based on these questions could lead to early diagnostic approaches, the identification of predisposing risk factors, a better understanding of disease etiology, novel interventions, or approaches to personalized therapeutic decision-making.

Example use case

Rheumatic heart disease (RHD) can be triggered by a group A Streptococcus (GAS) infection.¹⁵ Genome-wide association studies (GWASs) revealed a significant association between RHD and nonsynonymous variants present in the IGHV4-61*02 allele of the IGHV4-61 gene.¹⁴ Hypotheses that could be pursued with data from the AKC include the following:

• The IGHV4-61*02 allele predisposes individuals to autoimmune disease.

• The IGHV4-61*02 allele predisposes individuals to autoimmune diseases with cardiac involvement, such as RHD or lupus.

• Populations with a higher incidence of lupus cardiac involvement have high frequencies of the IGHV4-61*02 allele.

• AIR repertoires from patients with RHD or lupus have high IGHV4-61*02 usage and highly expanded IGHV4-61*02-expressing clones.

• BCRs and Abs utilizing IGHV4-61*02 have cross-reactive specificity for GAS epitopes or antigens and human heart tissue epitopes or antigens.

In the absence of the AKC, pursuing these hypotheses requires the following:

• Querying multiple repositories multiple times to ensure that no data are missed as a result of terminology, API, or other differences between repositories.

• Downloading the data and comparing original source identifiers (e.g., PubMed) and other metadata to identify and resolve duplicate data across repositories.

• Making germline gene assignments for all repertoires to ensure comparability.

• Mapping AIRs to specificities for all AIRs not sourced from repositories of specificity.

• Resolving terminology differences for data sourced from different repositories so that relevant comparator groups can be formed for analysis via aggregation, for example, resolving terminology differences for metadata fields such as subject demographics (e.g., race and/or ethnicity codes) or phenotype (e.g., lupus and systemic lupus erythematosus [SLE]) or for specificity annotations.

Under the proposed AKC, results could be returned with a single query and little to no post-processing of the data.

To illustrate the synergistic complementarity resulting from the AKC data integration, we queried the component repositories by using the query term “IGHV4-61*02” (Figure 1). The AIRR Data Commons currently contains ~10.3 million BCR rearrangements with IGHV4-61*02 as the utilized V gene allele. These rearrangements are derived from 764 AIR repertoires from 306 subjects with 17 disease diagnoses, including chronic lymphocytic leukemia, COVID-19, hepatitis B, SLE, celiac disease, and multiple sclerosis as the top six. In VDJbase, the IGHV4-61*02 allele is found in the genotype or haplotype of 141 subjects with eight diagnoses, including myasthenia gravis, celiac disease, SLE, and multiple sclerosis. VDJbase contains data from an additional 24 subjects who have a homozygous deletion for the IGHV4-61 gene, as well as IGHV4-61*02 allele frequencies among subjects with various phenotypes, including presumed healthy status, any diagnosis of an autoimmune disease, or a specific diagnosis of SLE. Specificity data from the IEDB and IRAD are available for 452 IGHV4-61*02 BCR rearrangements with 36 known epitope and antigen specificities, including epitopes and antigens from multiple viruses, such as HIV, hepatitis C virus, and Ebola virus, as well as cardiolipin and molecules associated with multiple sclerosis and Kawasaki disease. After the AKC is established, these data will be returned in an integrated fashion in response to a single query. Additionally, many of the ~10.3 million IGHV4-61*02-containing IGH sequences originating from the AIRR Data Commons will be mapped to a specificity, whereas others will have predicted specificities based on sequence similarity or other distance metrics. Only 34 of the corresponding subjects are shared with VDJbase. Thus, genotype and haplotype data will be added for many of the 272 additional subjects from the AIRR Data Commons.

Figure 1. Results of querying the AIRR Data Commons, VDJbase, IEDB, and IRAD for IGHV4-61*02.

Rings with a solid fill correspond to query results, whereas rings with a patterned fill indicate the absence of query results. This contrast highlights the unique data and knowledge contributed by each type of repository. From the center outward: sequences for IGH rearrangements utilizing IGHV4-61*02; specificities for receptors containing an IGHV4-61*02-encoded variable region; repertoires that include IGHV4-61*02 rearrangements; subjects from whom IGHV4-61*02-containing repertoires were collected; the health status or diagnosis annotations associated with those subjects; and IGHV4-61*02 allele frequencies among VDJbase subjects with annotations of presumed healthy status, any autoimmune disease, or systemic lupus erythematosus (SLE).

Conclusions

The amount of data already hosted by the component repositories is large, and these data stand ready to immediately support a tremendous amount of knowledge generation. Through the extension and application of existing algorithms and the development of new algorithms, data and metadata from the repositories will be integrated and enriched with new knowledge capturing the relationships between the core data types. APIs will enable queries of the AKC component repositories and integrated knowledgebase. Web-based GUIs will provide efficient access for a broad user community. These interfaces will enable querying of each data type via both its standard metadata and the new computationally derived, cross-cutting knowledge. Queries will return integrated results ready for exploration and analysis with state-of-the-art computational tools within the AKC or offline after download. The AKC will be open and extensible, allowing additional data resources to be added over time. Extensibility will be enabled by containerized applications, community-developed ontologies and standards, APIs designed for interoperability with other NIH data commons efforts (such as the NCI Cancer Research Data Commons), and open-source code, all supported by comprehensive documentation. The AKC will be a cloud-based knowledgebase that makes integrated AIR germline, repertoire, and specificity data and new, computationally derived knowledge accessible to the broad research community.

Invitation to the community

The AKC mission is to make data and knowledge about AIR repertoires, receptor gene polymorphisms, and receptor specificities freely, easily, and publicly available in a unified fashion to facilitate cross-cutting, cutting-edge research on the antigen recognition system of the adaptive immune response and its role in health and disease. Accomplishing this mission and maximizing the resource’s impact will require community involvement. We invite members of the community to contribute directly to the AKC’s development by contributing data, knowledge-generation algorithms, and use cases and by responding to surveys and participating in planned virtual and live events designed to assess and solicit feedback on functionality and usability. Information about contacting the AKC team and ways to get involved will be posted at https://airr-knowledge.org/.