Machine learning identifies T cell receptor repertoire signatures associated with COVID-19 severity

Abstract

T cell receptor (TCR) repertoires are critical for antiviral immunity. Determining the TCR repertoire composition, diversity, and dynamics and how they change during viral infection can inform the molecular specificity of host responses to viruses such as SARS-CoV-2. To determine signatures associated with COVID-19 disease severity, here we perform a large-scale analysis of over 4.7 billion sequences across 2130 TCR repertoires from COVID-19 patients and healthy donors. TCR repertoire analyses from these data identify and characterize convergent COVID-19-associated CDR3 gene usages, specificity groups, and sequence patterns. Here we show that T cell clonal expansion is associated with the upregulation of T cell effector function, TCR signaling, NF-kB signaling, and interferon-gamma signaling pathways. We also demonstrate that machine learning approaches accurately predict COVID-19 infection based on TCR sequence features, with certain high-power models reaching near-perfect AUROC scores. These analyses provide a systems immunology view of T cell adaptive immune responses to COVID-19.

Comparative single‐cell RNA sequencing analysis of immune response to inactivated vaccine and natural SARS‐CoV‐2 infection

“…TCR repertoires analysis unveils a robust TCR expansion subsequent to the initial In‐Vac administration, indicating a significant T cell response to vaccination, which is consistent with previous findings 19 . However, the disparate expansion patterns of paired VJ genes, noted in samples post‐In‐Vac administration versus those from COVID‐19 patients, underscore the varying effects on T cell‐mediated immunity 14,72 . This divergence likely arises from different antigen epitopes in terms of abundance and/or variance between In‐Vac and COVID‐19, which lead to distinct preferences for VJ genes crucial in recognizing specific antigens 14 .…”

“…19 However, the disparate expansion patterns of paired VJ genes, noted in samples post-In-Vac administration versus those from COVID-19 patients, underscore the varying effects on T cell-mediated immunity. 14,72 This divergence likely arises from different antigen epitopes in terms of abundance and/or variance between In-Vac and COVID-19, which lead to distinct preferences for VJ genes crucial in recognizing specific antigens. 14 Prolonged antigen exposure and inflammatory signals often correlate with the decayed T cell functionality.…”

Architecture of the SARS-CoV-2-specific T cell repertoire

“…Limited diversity of T cell repertoire seems to be associated with severe disease ( 38 , 39 ), whereas higher diversity is more likely to result in successful elimination of the virus. Multiple studies have found that the overall diversity of non-antigen-specific TCRs in blood samples taken from patients with COVID-19 is lower than that of healthy donors ( 40 , 41 ), and is even lower in patients with severe disease. For example, a cohort of patients with pneumonia had a slightly less diverse overall TCR repertoire compared to those with mild disease ( 38 ) presumably due to expansion of SARS-CoV-2 specific T to defend against the infection cells in symptomatic and hospitalized individuals ( 42 ) On the other hand, low repertoire diversity may be a prognostic factor and explain the higher risk of serious illness and death in elderly patients ( 43 – 47 ), as it is well known that TCR repertoire diversity declines with aging, and this is also known to affect the antiviral response to other pathogens, such as the human influenza A virus ( 48 , 49 ).…”

“…Numerous studies have shown that the TCR repertoire in patients with mild COVID-19 infection remains relatively diverse within CDR3 central region, with high generation probability compare to severe patients ( 38 ). This leads to a broad range of SARS-CoV-2-specific sequences observed in mild disease, with many public CDR3 sequences ( 19 , 36 , 38 , 54 ) This potentially explains why pneumonia patients have TCRs with longer CDR3 regions arising from lower-probability V(D)J-rearrangement events relative to the SARS-CoV-2-associated TCR repertoires in patients with mild disease, which also tend to prominently feature public clonotypes ( 38 , 40 ). TCR repertoire profiles in asymptomatic infection is similar to mild disease ( 36 ).…”

Robust diagnosis of infectious disease, autoimmunity and cancer from the paratope networks of adaptive immune receptors

“…For use as a liquid biopsy, the diverse AIR signals in a given donor must be formulated as a single feature vector that can be compared to those of other donors. In recent years, statistics-and ML-based approaches have been actively explored in order to construct such features in order to classify donors according to their disease status [11][12][13][14][15][16][17][18][19][20] .…”