What is a repertoire?
The immune repertoire is the collection of unique immunoglobulin and T-cell receptor sequences present in an individual at a particular time. At Digital Proteomics we focus primarily on the immunoglobulin, or B-cell receptor repertoires of humans and other mammals. The total number of B cells and plasma cells that encode full-length immunoglobulins in an adult human is estimated to be 1010-1011, which puts an upper bound on the total size of the B-cell receptor repertoire. The antibody repertoire is dynamic, with sequence diversity and composition of the repertoire changing dramatically over time. In this post, we discuss the approach to immune repertoire construction that we employ in our Reptor and Alicanto services. Before repertoire construction and analysis can begin, the B-cell receptor transcripts must be sequenced. For a bit about that process, read our previous post on immunosequencing.
Due to high sequence diversity, immune repertoire sequencing and assembly of an individual repertoire is complex and requires specialized tools. It’s important to recognize the unique challenges of antibody repertoire sequence and analysis, which make them ill-suited for standard RNA-seq or target-enrichment workflows.
Repertoire construction overview
Read quality filtering
As in any bioinformatic pipeline, for immune repertoire analysis ‘garbage in = garbage out’. Read filtering is an important part of the process, and the best practices for any RNA/DNA sequencing analysis apply. This includes removing reads with low quality, removing reads that are generated from a spiked-in library (such as PhiX), and trimming primer and sequencing adapter sequences. Looking at overall statistics for your run using a free tool such as FastQC is also a great way to spot problems in your sequencing run.
A single amino acid change can dramatically alter the stability and affinity of an antibody. For that reason, constructing an error-corrected repertoire is a crucial first step to many downstream applications from antibody discovery to clone tracking. Both PCR amplification and sequencing can introduce errors, which must be corrected. A simple approach is to apply an abundance threshold and remove any sequences with fewer than a fixed number of reads. The idea behind this approach is that errors should be rare, and true antibody sequences should appear multiple times. Low abundance correct sequences will be dropped, removing real and important information about the repertoire.
Unique molecular identifiers (UMIs) have grown in popularity for correcting errors as well as accurately quantifying RNA abundance of each antibody locus (Khan et al. 2016, Turchaninova et al. 2016). At Digital Proteomics, our approach is based on the Hamming graph clustering approach to performing error correction developed at UCSD (Safonova et al. 2015). We create a node in the graph for each distinct antibody sequence, and create an edge between nodes if the sequences are within a predefined Hamming distance. In principle, erroneous sequences will appear in subgraphs also containing the correct sequence. A consensus sequence is created by collapsing dense subgraphs.
V(D)J labeling and complementarity-determining region identification
After error correction, the repertoire consists of a collection of corrected sequences. In order to further characterize the repertoire we must determine the germline sequences that gave rise to each antibody. This process is called V(D)J labeling, and open source tools such as IMGT/HighV-Quest, IgBLAST, MiXCR have been developed for this purpose. We apply the colored antibody graph technique (Bonissone and Pevzner, 2016) to determine the germline V, D, and J genes for each antibody sequence. In addition to identifying the original germ line sequence, aligning each antibody to the germline allows us to determine hypermutation sites.
Germline gene labeling also aides complementarity-determining region (CDR) identification. Each antibody chain has 3 CDRs, two of which reside fully in the V gene segment and a third which occurs at the junction of the V, D, and J (or V and J in the case of the light chains). Identifying these regions is important for understanding the relationships between antibodies in the repertoire, and their evolution in response to immunological challenge. In our process, we cluster nearly identical CDR3 sequences into a single group, or clone. The figure below demonstrates the process of V(D)J labeling, CDR3 identification, and CDR3 clustering.
What was once tens of gigabytes of unintelligible reads is now an annotated antibody repertoire. The world is your oyster for antibody repertoire analysis, and we will survey that world in a future post. Stay tuned!
Helpful resources for further reading
The Adaptive Immune Receptor Repertoire (AIRR) Community holds regular meetings and publishes standards for repertoire sequencing and analysis.