Population variability in the generation and selection of T-cell repertoires

Abstract

The diversity of T-cell receptor (TCR) repertoires is achieved by a combination of two intrinsically stochastic steps: random receptor generation by VDJ recombination, and selection based on the recognition of random self-peptides presented on the major histocompatibility complex. These processes lead to a large receptor variability within and between individuals. However, the characterization of the variability is hampered by the limited size of the sampled repertoires. We introduce a new software tool SONIA to facilitate inference of individual-specific computational models for the generation and selection of the TCR beta chain (TRB) from sequenced repertoires of 651 individuals, separating and quantifying the variability of the two processes of generation and selection in the population. We find not only that most of the variability is driven by the VDJ generation process, but there is a large degree of consistency between individuals with the inter-individual variance of repertoires being about ∼2% of the intra-individual variance. Known viral-specific TCRs follow the same generation and selection statistics as all TCRs.

Probing T-cell response by sequence-based probabilistic modeling

“…We also consider a recent computational tool, SONIA [15] that infers selection factors acting on the TRB chain from RepSeq datasets. SONIA quantifies selection pressures on amino acids in the TCR sequences in terms of position-specific “selection” or q -factors, by comparing the probability of finding a given sequence in a post-stimulation repertoire of interest compared to a baseline repertoire.…”

“…Specifically, SONIA fits the statistics of count-weighted CDR3 datasets by an independent-site model, fixing the same “background” distribution for both datasets to the probability of generation of these sequences P gen (Eq.7 in Methods). We used SONIA models with Left+Right feature encoding [15] (Fig. 2B).…”

“…Response scores can also be defined in terms of the TCR probability at day 21 relative to baseline distribution P gen ( σ ) [21]. These scores capture information about the overall selection pressures acting on the sequence, including thymic and antigen specific selection [15] (see Methods). The advantage of this choice is that P gen can be estimated in silico [21], without the need for longitudinal datasets that include sequence data at time points before stimulation, which are not always available.…”

“…A significant correlation is recovered by progressively filtering out the low-abundance sequences from the testing set, shown on the example of stimulation by p =NA of Pt3 PBMC. For both RBM and SONIA (shown on the right), the correlation is robust to the choice of different background distributions: P 0 (the probability learned on the same patient’s dataset at day 0), P gen (the probability of generating a given CDR3, described by OLGA [21]) and P post (the post-thymic selection CDR3 distribution, sampled here from a default human TRB model available in SONIA [15]). The small differences between the 3 curves are mainly due to sampling, see also Methods.…”

“…We propose a set of probabilistic-modeling approaches, based on Restricted Boltzmann Machines [12–14] and previously defined selection factors [15], to characterize features of responding TCR from sequence data. We apply the method to T-cell subrepertoires from [16] that are in vitro stimulated by patient-specific tumor neoantigens in the peripheral blood of seven long-term survivors of pancreatic ductal adenocarcinoma (PDAC).…”