*Purpose: Antibody-mediated rejection (AMR) remains one of the major causes of allograft failure and our understanding of this disease process is poor. The generation and development of T cell clones specific to kidney alloantigens play a central role in transplant rejection. Frequently, assumptions regarding intra-allograft immune mechanisms are inferred from studies performed on the peripheral blood. We developed a computational tool to score each T cell clone and compared the clonal distribution in biopsies with paired peripheral blood samples.

*Methods: The 10X Genomics platform with 5-prime VDJ chemistry was used to generate T cell CDR3 sequence data from single T cells from human kidney transplant biopsies (n=2) or paired peripheral blood (n=2). CDR3 amino acid sequences from each sample were used to generate a score for each clone based on Kidera factors.

*Results: We developed a variational autoencoder, a type of deep learning model, that generates an embedding space for each T cell clone based on the CDR3 amino acid Kidera factors. This allowed for vectorization of CDR3 sequences and subsequent dimensional reduction in 2D space to compare clones from biopsy and blood. TCRB chains from biopsies and blood were distributed in a significantly different pattern in 2d space (p-value = 1.075e-12, 2-dimensional Kolmogorov-Smirnov)(figure 1).

*Conclusions: T cell clonal distribution in peripheral blood does not represent the intra-allograft T cell clonal repertoire in kidney transplantation. Peripheral blood T cell clonal repertoire analysis cannot accurately infer the intra-allograft T cell repertoire.

« Back to 2022 American Transplant Congress