Neoantigen-specific CD8 T cells with high structural avidity preferentially reside in and eliminate tumors

The study investigates whether the strength of antigen recognition by CD8 T cells—quantified as TCR structural avidity (pMHC–TCR dissociation kinetics) and functional avidity (antigen sensitivity)—explains their preferential presence in tumors and their anti-tumor efficacy. Prior work suggests neoantigen-specific T cells may escape negative thymic selection and exhibit superior functionality relative to tumor-associated antigen (TAA)-specific cells. Clinical efficacy of TIL-ACT correlates with persistence of transferred clones, but how TCR avidity affects tumor homing and engraftment remains unclear. Here, the authors profile virus-, TAA-, and neoantigen-specific TCRs from tumors and blood across multiple cancers, test the relationship between avidity and tumor residency/infiltration, evaluate the role of CXCR3, and develop an in silico predictor of TCR structural avidity to identify high-avidity, tumor-engrafting T cells.

Background evidence includes: (i) neoantigen burden and TMB correlate with immunotherapy benefit and TIL-ACT efficacy; (ii) structural and functional avidities of CD8 T cells correlate and influence performance; (iii) TIL-ACT outcomes link to in vivo persistence and distinct gene expression; (iv) CXCR3 and tissue-residency integrins (CD103, CD49a) promote tumor homing/retention; (v) epitope immunogenicity prediction based solely on antigen presentation (binding/stability) often poorly matches T cell recognition, whereas models incorporating TCR recognition (e.g., PRIME) and peptide dissimilarity-to-self better predict immunogenicity; (vi) high-affinity/avidity TCRs can enhance efficacy but functional assays can be confounded by activation/exhaustion status. These works collectively motivate measuring structural avidity and exploring biophysicochemical features to identify potent tumor-reactive T cells.

- Cohorts and antigens: Generated 371 CD8 T cell clones specific to 19 neoantigens, multiple TAAs, and viral epitopes from 16 patients (melanoma, ovarian, lung, colorectal) and 6 healthy donors. Antigen-specific cells were sorted using reversible pMHC multimers (NTAmers) to avoid loss of high-avidity cells, then single-cell cloned and expanded. - Structural avidity: Determined pMHC–TCR dissociation half-life (T1/2) using NTAmers. Rapid decay of multimers to monomeric pMHC enables off-rate measurements on living cells. For validation and consistency, selected TCRαβ pairs were cloned into healthy donor T cells to compare avidity rankings. - Functional avidity: Measured antigen sensitivity by IFN-γ ELISpot dose–response to derive EC50 (peptide concentration for half-maximal response). - pMHC binding control: In vitro pMHC refolding assays validated peptide–HLA binding to exclude confounding from poor peptide–MHC affinity; compared experimental avidity to in silico presentation predictors (NetMHCpan, MHCflurry, MixMHCpred, NetMHCstabpan) and immunogenicity metrics (PRIME, DisToSelf). - Paired PBL vs TIL analyses: Seven pairs of clones recognizing the same epitope were derived from blood and tumor of the same patients to compare structural/functional avidities; TCR repertoire sequencing assessed clonal overlap and frequencies in each compartment. - Animal models and ACT: Engineered primary human CD8 T cells with NY-ESO-1(157–165) TCR variants of low (V491), intermediate (WT), and high (DMB) structural avidity; adoptive transfer into IL-2 NOG mice bearing Me275 melanoma. Tumor growth, survival, and intratumoral CD8 densities were assessed. Anti-CXCR3 blocking antibody was used to test chemokine receptor dependence; additional neoantigen TCR models corroborated findings. - Chemokine receptor/integrin profiling: Compared expression of CXCR3, CCR7, CD103, CD49a and others on matched low- vs high-avidity clones at baseline and after short stimulation. - Molecular modeling: Rosetta-based TCRmodel and consensus scoring with DOPE to model TCR–pMHC complexes; quantified favorable polar and apolar contacts to correlate with experimental T1/2. - Sequence-based clustering and prediction: Extracted CDR3β 4-mer biophysicochemical motifs (Atchley factors), hierarchical clustering (UPGMA) across 58 TCRs (12 specificities) with known avidities; identified motif hotspot enriched in high-avidity TCRs. Built a logistic regression classifier using solvent-exposed CDR3β residue presence (R, N, D, G, I, L, F; solvent accessibility >30%) to predict high vs low avidity; performance evaluated by AUC and cross-validations (leave-20%-out and leave-one-epitope-out). - Clinical sample validation: Applied predictor to bulk TCR repertoires from blood and tumors of four melanoma patients to assess enrichment of predicted high-avidity TCRs in tumors; experimentally validated two KIF1B(5918F)-specific TCRs predicted as high vs low avidity in biophysics and in ACT against autologous tumor in mice. - Statistics: Pearson/Spearman correlations, Mann–Whitney, Wilcoxon paired tests, log-rank tests; customized randomization test for enrichment (Fig. 4c).

- Broad heterogeneity of structural and functional avidities across virus-, TAA-, and neoantigen-specific CD8 T cell clones; structural avidity (T1/2) generally higher for neoantigen-specific than TAA-specific T cells (e.g., overall comparisons significant: P=0.003 and P=0.023 in figure panels). - Structural avidity correlated with functional avidity (T1/2 vs EC50), though functional assays showed higher variability. - TILs vs PBLs: For seven paired epitope-specific comparisons, TIL-derived clones displayed significantly higher structural avidity and antigen sensitivity than blood-derived counterparts across multiple cancers; holds when considering unique clonotypes. - Clonotype-level evidence: In Lung1 patient, among UTP20 neoepitope-specific clonotypes shared between blood and tumor, clonotype 5 dominated in TILs (58.8%) but was 1.4% in PBLs; its structural avidity was highest and correlated with tumor frequency. Similar results for PHLPP2 in CRC1. - Molecular modeling: Higher-avidity TCRs established more favorable TCR–pMHC interactions (greater polar/apolar contacts), correlating with experimental T1/2. - In vivo ACT: NY-ESO-1 TCR variants: V491 (T1/2=7 s), WT (27 s), DMB (252 s). Despite similar in vitro tumor reactivity of WT and DMB vs V491, only structural avidity predicted superior tumor control and intratumoral engraftment; DMB showed significantly better tumor control (log-rank P<0.0001) and higher CD8+ infiltration than WT and V491 (e.g., infiltration P=0.0002 and P=0.0008 vs V491 and WT). - CXCR3 linkage: High-avidity clones expressed more CXCR3 and upregulated it upon stimulation; also increased CD103 and CD49a. Blocking CXCR3 in vivo reduced tumor control (P=0.01) and CD8+ tumor infiltration, confirming CXCR3 dependence of high-avidity T cell homing. - Predictive modeling: CDR3β-based hierarchical clustering identified a hotspot enriched for higher-avidity TCRs (T1/2>10 s), not explained by random clustering (P<0.001). Logistic regression classifier achieved AUC=0.96 (sensitivity 0.91, specificity 0.97) on 58 TCRs; cross-validations across antigen classes and HLA achieved >70% success. - Clinical relevance: In four melanoma patients, predicted high-avidity TCRs were enriched in tumors vs blood (cumulative P=0.05). For two KIF1B(5918F)-specific TCRs from patient Mel8, the model correctly predicted high vs low avidity; only the high-avidity TCR mediated tumor control in ACT against autologous tumor (log-rank P<0.0001), despite similar in vitro tumor reactivity of both.

The findings demonstrate that high structural avidity of TCR–pMHC interactions is a key determinant of tumor residency and effective tumor control by CD8 T cells. Neoantigen-specific T cells are enriched for high structural avidity compared to TAA-specific cells, which explains their preferential detection within tumors and links neoantigen recognition to enhanced functionality and infiltration. Compared with functional readouts (EC50), structural avidity provides a robust, activation-independent biomarker of T cell potency that correlates with in vivo engraftment and efficacy. Mechanistically, high structural avidity associates with enhanced CXCR3 expression and tissue residency integrins, and tumor control depends in part on CXCR3-mediated recruitment. The biophysicochemical commonalities among high-avidity TCRs enabled a sequence- and structure-informed predictor that identifies high-avidity TCRs without prior antigen specificity, and these predicted high-avidity TCRs are enriched in patient tumors and confer superior antitumor efficacy in vivo. Together, these results support prioritizing high-structural-avidity neoantigen-specific T cells (preferentially from TILs) for ACT and TCR-engineering strategies, and highlight the value of integrating structural avidity measurements and sequence-based prediction into T cell selection pipelines.

This study establishes that CD8 T cells with high structural avidity preferentially reside in tumors and mediate superior tumor control, particularly in the context of neoantigen recognition. It validates structural avidity (pMHC–TCR off-rate) as a robust, reproducible biomarker of T cell potency, links high avidity to CXCR3-dependent tumor homing, and introduces a biophysicochemical, CDR3β-based predictive model (AUC 0.96) to identify high-avidity TCRs from sequence alone. Clinically, predicted high-avidity TCRs are enriched in tumors and outperform low-avidity counterparts in ACT. Future directions include expanding training datasets to improve predictor generalizability, incorporating additional CDR regions and structural features, integrating predictors with antigen presentation/immunogenicity models, and prospectively validating high-avidity TCR selection in clinical ACT trials.

- Functional assays (EC50) depend on T cell activation/exhaustion status and show variability; structural avidity is more robust but still measured in vitro. - The sequence-based predictor, while accurate in the current dataset, requires further validation on larger, external cohorts and diverse epitopes/HLA alleles; current cross-validation success (>70%) indicates room for improvement. - Modeling relies on homology-based TCR–pMHC structures with estimated contacts; structural inaccuracies or limited template availability may affect predictions. - In vivo validation largely used IL-2 NOG mouse models and selected tumor lines; translation to human clinical settings may be influenced by tumor heterogeneity, microenvironment, and antigen presentation variability. - CXCR3 is a key contributor but not the sole determinant of tumor homing; other chemokine axes and retention factors were not exhaustively dissected.