Immunosequencing identifies signatures of cytomegalovirus exposure history and HLA-mediated effects on the T cell repertoire

Abstract

An individual's T cell repertoire dynamically encodes their pathogen exposure history. To determine whether pathogen exposure signatures can be identified by documenting public T cell receptors (TCRs), we profiled the T cell repertoire of 666 subjects with known cytomegalovirus (CMV) serostatus by immunosequencing. We developed a statistical classification framework that could diagnose CMV status from the resulting catalog of TCRβ sequences with high specificity and sensitivity in both the original cohort and a validation cohort of 120 different subjects. We also confirmed that three of the identified CMV-associated TCRβ molecules bind CMV in vitro, and, moreover, we used this approach to accurately predict the HLA-A and HLA-B alleles of most subjects in the first cohort. As all memory T cell responses are encoded in the common format of somatic TCR recombination, our approach could potentially be generalized to a wide variety of disease states, as well as other immunological phenotypes, as a highly parallelizable diagnostic strategy.

Predicting the spectrum of TCR repertoire sharing with a data‐driven model of recombination

“…Model training is done by finding model parameters that maximize the likelihood of the data, equal to the product of generation probabilities of the observed TCRs in the dataset. Here, we used IGoR to infer a generation model from the non‐functional reads in the datasets from which the productive reads used for the sharing analysis came, for human,30 and mice 23…”

“…Human dataset was used from the CMV study in 30. 658 individual samples were used, with mean number of reads ≈200 000, and mean number of unique CDR3s ≈180 000.…”

“…Such sharing of specific TCRs is expected from the relatively low diversity of antigen‐specific sequences revealed by in vitro multimer‐staining experiments 11, 12. A very similar idea has been exploited by several groups to identify TCRs specific to the Cytomegalovirus,30 Type‐1 diabetes,48, 49 arthritis50 and other immune diseases 51. In these studies, there is no theoretical expectation from the recombination model.…”

“…Specifically, we re‐analyzed published TCR β‐chain nucleotide sequences of 14 Black‐6 mice23 and 658 human donors30 (Section 7). Individual samples comprised 20 000‐50 000 unique sequences for mice, and up to 400 000 for humans.…”

“…We performed the sharing analysis on 30 patients with bladder cancer, on TCR repertoires sequenced from blood samples 54. We compared it with 30 healthy individuals, chosen at random among the individuals studied in 30 to have similar sample sizes. We then downsampled the reference repertoires of the healthy individuals to have the exact same sample sizes as the cancer patients to guarantee a fair comparison.…”