Project Summary
Post-acute sequelae of COVID-19 (PASC), colloquially known as long covid, has been reported
in 31%-69% of COVID-19 patients. The prevalence of PASC among patients, even those with
mild infection, and those with prior vaccination, underscores the importance of principled
approaches for uncovering and addressing the cause of these issues. We propose to conduct a
detailed analysis of the immune repertoire dynamics of PASC patients and compare them to
both COVID-19 patients without PASC, and healthy individuals. These will be secondary
analyses, conducted on existing data from a previous study with the Institute of Systems Biology
and Swedish Health Services, which includes longitudinal deep immunophenotyping with single
cell and plasma multiomics, repertoire sequencing, electronic health records, viremia
measurements, and antibody titers for 209 COVID-19 patients. In Aim 1, we will conduct
inference, analysis, and comparison of T-cell receptor repertoire dynamics in PASC. Using
interpretable statistical, biophysical, and machine learning approaches, we will conduct a
detailed analysis of T-cell repertoires aimed at finding PASC specific clonotypes and their
corresponding receptor features. This involves building cohort specific models of thymic
selection, examining how these models differ between cohorts and over time, inferring the
dynamics of repertoire size, sharing and diversity, and uncovering the receptor features which
drive these differences. In Aim 2, we will develop new methods for the integration of T-cell
repertoire and single cell dynamics. The existing breadth of multiomics data allows us to explore
new methodologies for integrating different modalities of longitudinal data. For T-cell repertoires
in particular, we plan to extend existing models of thymic selection to include gene expression of
relevant T-cell specific genes and to study how the expansion and contraction of clonotypes
affects the dynamics of T-cells in gene expression space. By using interpretable biophysical and
machine learning methods, we can construct generative models of TCRs including gene
expression values and study how these distributions change in time. Results have strong
potential to accelerate our understanding of the etiology of immune-based PASC responses,
which is essential for prioritizing potential therapeutic targets for prevention and treatment.
Further, results will advance methods for future research across a wide range of infectious
diseases and immune-mediated medical conditions.
