EXPLORING THE DRIVERS OF PUBLIC T-CELL RECEPTORS USING DEEP LEARNING AND TRANSCRIPTOMICS

Abstract

The adaptive immune system's remarkable capacity for targeted defense relies on a vast and diverse repertoire of T-cell receptors (TCRs) generated through a stochastic process. This theoretically implies a near-limitless pool of potential TCR sequences estimated at about 10^61 unique TCRs. Despite this potential for immense diversity, identical TCR sequences, termed public TCRs, are recurrent across individuals in the population. The mechanisms driving this phenomenon and the intrinsic functional identity of T cells expressing public TCRs remain poorly understood. This dissertation addresses this gap by integrating deep learning on large-scale repertoire data with single-cell multi-omics analysis of healthy human immune cells.

First, we developed a convolutional neural network (CNN) that accurately classifies TCRs as public or private based on their nucleotide sequence, thereby minimizing inherent generative properties from confounding antigen exposure history. Subsequent sequence-level analyses revealed that TCR publicness is governed by intrinsic generative biases, including shorter CDR3 lengths, preferential V(D)J gene usage, and specific, positionally-constrained nucleotide motifs. Structural modeling further suggested that these sequence features result in unique CDR3β loop conformations, particularly when interacting with MHC Class I molecules.

Second, we leveraged this classifier to investigate the functional phenotypes of public and private T cells using single-cell RNA sequencing. While public and private CD4+ T cells were transcriptionally indistinguishable, public CD8+ T cells exhibited a distinct natural-killer-like and relatively more potent cytotoxic effector phenotype. They were characterized by the significant upregulation of cytotoxic molecules (GZMB, GZMH, PRF1), Killer-cell Immunoglobulin-like Receptors (KIRs), and effector markers (CX3CR1), coupled with the downregulation of naive/memory-associated genes (IL7R, LEF1). Furthermore, intercellular communication analysis revealed that public CD8+ T cells function as major signaling hubs, showing significantly enhanced signal reception, primarily through MHC I and KIR-ligand pathways.

Together, these findings suggest that public TCRs are not a random artifact of repertoire generation. Instead, public CD8+ T cells may represent an evolutionarily conserved, functionally distinct population endowed with a pre-programmed, innate-like cytotoxic capacity. Their heightened state of immune readiness suggests they form a critical bridge between innate and adaptive immunity, providing a new framework for understanding immune surveillance with significant implications for vaccine design and T-cell-based immunotherapies.