Background We and others have shown that validated neoantigen-specific CD8+ TIL from several cancer types express high levels of CXCL13 and ENTPD1 (CD39), and low level of IL7R. Notably, among a small number of functionally validated tumor-reactive TIL, those from ICB-resistant tumors were transcriptionally distinct. A major limitation impeding assessment of this observation in a larger cohort is the expensive and cumbersome nature of functional T cell assays detecting human tumor-reactive TIL. We therefore developed a conserved, 3-gene score to distinguish tumor-specific from bystander TIL, providing insights into tumor-reactive TIL biology independent of functional assays.

Methods Neoantigen- and CEF-specific TIL from 3 melanoma specimens (Oliveira et al, Nature, 2021) were used to train a gene score weighted for CXCL13, ENTPD1 and IL7R, which was validated with neoantigen- and CEF-specific TIL from 2 lung cancers (Caushi et al, Nature, 2021). SAVER was used to recover missing data. Single patient models were constructed initially with linear regression and random forest algorithm on imputed and non-imputed data. Combined voting models were built. ROC curves and AUC were used for model evaluation. Due to strong inter-patient heterogeneity, patient-specific cutoffs for defining MANAscore^hi TIL were set based on MANAscore distributions.

Results Our three-gene MANAscore algorithm out-performed other RNAseq-based algorithms in identifying validated neoantigen-specific CD8+ TIL. Most MANAscore^hi TIL expressed a tissue resident memory (TRM) program, consistent with validated neoantigen-specific TIL. Putative tumor-reactive cells (pTRC) identified via MANAscore in anti-PD-1-treated lung cancers had higher expression of immune checkpoint and cytotoxicity-related genes relative to non-pTRC, and their frequency was significantly correlated with ICB response. pTRC in pathologically responding tumors showed higher IL7R expression, lower checkpoint expression, and shared trajectories originating from high expression of GZMK and moving toward a stem-like phenotype. Moreover, MANAscore accurately identified T cells specific for multiple classes of tumor antigens across different cancer types, including cancer-testis antigen-specific TIL from oral cancers, neoantigen-specific TIL in metastatic cancers (including breast cancer, colorectal cancer, melanoma), and viral oncogene-specific TIL in virus-driven Merkel cell carcinomas. Application of MANAscore on larger scRNAseq datasets integrating additional melanomas and lung, oral, sinonasal, and head and neck cancers revealed shared expression patterns of pTRC across tumor types, which provides a unique opportunity to discover novel T cell-specific therapeutic targets and biomarkers.

Conclusions MANAscore is a robust tool for enriching candidate tumor-specific TIL. It can be used to understand the functional programming and underlying biology of tumor-reactive TIL in different antigen-specific contexts.

Acknowledgements The Mark Foundation for Cancer Research, Bloomberg~Kimmel Institute for Cancer Immunotherapy, The Mark Foundation Center for Advanced Genomics and Imaging, Cancer Research Institute, Lung Cancer Foundation of America, LUNGevity, American Lung Association, Swim Across America, Commonwealth Foundation, Bristol-Myers Squibb, National Institutes of Health grants R37CA251447 (K.N.S.), R01HG010889 (H.J.), R01HG009518 (H.J.), P01 CA255517, T32 CA080416, and P30 CA006973, Kelsey Dickson Team Science Courage Research Award: Advancing New Therapies for Merkel Cell Carcinoma (MCC), MCC Patient Gift Fund, National Foundation for Cancer Research.