Thymic selection of T-cell receptors as an extreme value problem

T lymphocytes (T cells) orchestrate adaptive immune responses upon activation. T cell activation requires sufficiently strong binding of T cell receptors (TCRs) on their surface to short peptides (p) derived from foreign proteins, which are bound to major histocompatibility (MHC) gene products (displayed on antigen presenting cells). A diverse and self-tolerant T cell repertoire is selected in the thymus. We map thymic selection processes to an extreme value problem and provide an analytic expression for the amino acid compositions of selected TCRs (which enable its recognition functions).

💡 Research Summary

The paper tackles the fundamental problem of how a diverse yet self‑tolerant T‑cell receptor (TCR) repertoire is generated in the thymus. The authors propose that thymic selection can be framed as an extreme‑value statistical problem. They begin by modeling the binding affinity between a TCR and a peptide‑MHC (pMHC) complex as an energy variable E, assuming that each TCR encounters a large number (N) of self‑derived peptides whose individual binding energies are independent random draws from a common distribution f(E). Positive selection requires that the minimum binding energy (E_min) among these N encounters be lower than an upper threshold E_pos, while negative selection demands that the maximum binding energy (E_max) be higher than a lower threshold E_neg. Because E_min and E_max are the minima and maxima of many independent samples, they follow Gumbel‑type extreme‑value distributions.

Using the known properties of the Gumbel distribution, the authors derive analytical expressions for the location and scale parameters in terms of the moments of f(E) and the sample size N. These expressions allow them to compute the probability that a given TCR survives both selection steps. Crucially, they link the survival probability to the amino‑acid composition of the TCR’s complementarity‑determining region (CDR). Each amino acid i contributes a characteristic mean binding energy μ_i and variance σ_i². The extreme‑value constraints translate into a Boltzmann‑like weighting for each residue:

p_i ∝ exp