Document resource
Background Identifying and engineering T cell receptors (TCRs) specific to neoantigens in acute myeloid leukemia (AML) is a significant challenge in advancing adoptive immunotherapy. We present a structure-guided pipeline for designing neoantigen-specific TCRs and evaluating their presence and potential specificity within patient-derived repertoires.Methods We focused on the TP53 Y220C-derived neoantigen (VVPCEPPEV), restricted by HLA-A*02:01. Starting from a melanoma TCR-pHLA complex (PDB: 2BNQ), we substituted the peptide with the AML neoantigen and used ProteinMPNN, a deep learning-based protein sequence design, to redesign residues in the CDR1, CDR2, and CDR3 loops, generating 50,000 candidate TCR sequences optimized for binding. Unique patient-derived TCRs were identified using a developed script from single-cell TCR sequencing (scTCRseq) data and defined as putative when multiple alpha or beta chain sequences were observed in the same barcode ( figure 1A-B). These patient TCRs were clustered with the designed sequences using GLIPH2, which identified one cluster containing over 90% of patient samples and two designed TCRs sharing a common CDR3 motif (figure 1C). To improve confidence and account for features beyond CDR3 motif, additional clustering analyses using TCRdist3 and SwarmTCR are underway.Results Patient TCRs were prioritized based on reliable cell annotations, non-putative status, and co-occurrence with TP53 mutations ( figure 1D-E). Six designed TCRs—selected based on structural confidence and clustering—and 62 patient TCRs from the dominant cluster were structurally modeled using TCRmodel2 (figure 1F). All modeled TCRs are currently being evaluated with STAG (Structural TCR-Antigen Generator) to assess their predicted binding specificity and interface similarity with the TP53-HLA-A*02:01 complex. Initial clustering suggests convergence between designed and patient TCRs in recognizing the TP53 neoantigen. Structural modeling further supports their compatibility with the target pHLA complex. Ongoing STAG analysis will provide binding likelihood scores and inform candidate selection for functional validation.Conclusions Our integrative strategy—combining structure-based TCR design, patient repertoire mining, and predictive binding evaluation—offers a reproducible and scalable framework to guide the discovery of personalized TCR candidates for AML immunotherapy.Abstract 1083 Figure 1Analysis of designed TCR and repertoire in AML. A. Cell annotation. B. T-cell clonality. C. Proportion of TCRs across clusters. D. Highlighted cells with TCRs from clusters 38 and 160. E. Clonality of cells in D. F. TCRmodel2 structures; top: patient TCRs, bottom: designed TCRs, colored by pLDDT