Abstract
Wilms tumor 1 (WT1) is a tumor-associated antigen expressed in solid tumors and hematological malignancies. T-cell immunotherapies targeting WT1 are currently under development. To analyze endogenous T-cell responses against WT1, we trained computational models capable of detecting WT1-specific T-cell responses from T-cell receptor (TCR) sequencing data. We peptide-pulsed healthy donor and acute myeloid leukemia (AML) patient samples with VLDFAPPGA (VLD, WT1(37-45)) and RMFPNAPYL (RMF, WT1(126-134)) peptides, then sequenced the WT1 dextramer-positive CD8 + T-cells with single-cell RNA + TCRαβ sequencing. The TCRGP machine-learning TCR-classification method was trained with epitope-specific and control TCR repertoires, and we obtained AUROC values of 0.74 (VLD) and 0.75 (RMF), allowing reliable identification of WT1-specific T-cells. In bulk TCRβ sequenced patient samples (AML n = 21, chronic myeloid leukemia (CML) n = 26, and myelodysplastic syndrome n = 25), the median WT1-specific T-cell abundance was similar to healthy controls, but their VLD and RMF-specific TCR repertoires exhibited higher clonality with two patients presenting up to 13% of WT1-specific T-cells. ScRNA+TCRαβ sequencing of AML bone marrow T-cells revealed that WT1-specific T-cells predominantly exhibit an effector or terminal effector memory phenotype. In conclusion, our novel computational models enable large-scale WT1-specific T-cell identification from TCR sequencing datasets and leukemia-antigen-specific immune response monitoring.