Sequencing and data processing

The library was sequenced as 2 × 150 bp paired-end readout with NovaSeq 6000 System^® (illumina). The fastq files were randomly subsampled to a size of 1 Gb using Seqtk (version 1.3-r106). CDR3 sequences were called by MiXCR (version 3.0.13)²⁶ using the following options: -s Homosapiens –starting-material dna –adapters adapters-present –impute-germline-on-export –5-end v-primers –3-end j-primers –receptor-type trb –region-of-interest CDR3 –only-productive –align "-OvParameters.geneFeatureToAlign = VRegion" –assemble "-OaddReadsCountOnClustering = true" –verbose.

Clones with non-human epitopes were filtered out by a screen against 7,276,705 CDR3 amino acid sequences downloaded from TCRdb²⁷, McPAS-TCR²⁸, PIRD²⁹, TCR3d³⁰, and VDJdb^31,32. Some clones were excluded if junction sequence of the clone does not contain proper conserved residues (e.g. V-Cystein at 104th or J-Phenylalanine at 118th to 129th). The average number of unique CDR3 amino acid sequences for each replicate in patients was 12,917.4, ranging from 2984.3 to 82,681.7.