Targeting non-canonical antigens unlocks functional T-cell responses in renal cell carcinoma Authors: Jiahao Wang, Yihao Zhu, Xinyu He, Shenyi Yin, Yang He, Pengju Yao, Juan Li, Xuwen Li, Peiting Shi, Renzhi Qian, Zhiqing Xiao, Xiongjun Ye, Jianzhong Jeff Xi, Buqing Ye Published: 2026 ABSTRACT AbstractAbstractsBackgroundRenal cell carcinoma (RCC) frequently exhibits favorable responses to immunotherapy, despite a low tumor mutational burden, suggesting a critical role for non-mutational antigens presented by human leukocyte antigen class I in activating CD8+ T cells.MethodsTo systematically identify non-canonical tumor-specific antigens, we developed an integrated proteogenomic approach to RCC samples by combining immunopeptidomics with exome and transcriptome sequencing. We subsequently primed and expanded antigen-reactive T cells in vitro, isolated a panel of antigen-specific T-cell receptors, and characterized their functionality. Finally, we investigated the tumor-killing and regression of non-canonical antigen-reactive T cells using patient-derived tumor-like cell clusters and mouse models.ResultsWe discovered a diverse repertoire of non-canonical tumor-specific antigens derived from human endogenous retroviruses (hERVs) and long non-coding RNAs (lncRNAs), many of which were shared across patients. Single-cell RNA sequencing further revealed that T cells reactive to these antigens were enriched within exhausted subsets in the tumor microenvironment, indicative of persistent antigen-specific stimulation. Functionally, T cells engineered to recognize these non-canonical antigens mediated potent tumor cell killing both in vitro and in vivo. Our findings establish hERV-derived and lncRNA-derived antigens as key drivers of RCC immunogenicity and highlight their strong potential as targets for T-cell-based immunotherapy.ConclusionsOur work is the first to demonstrate that non-canonical antigens drive immunogenicity in low-mutation RCC, thereby resolving a key question in cancer immunology—how tumors with low mutational burden can provoke robust T-cell responses.Keywords: Human leukocyte antigen - HLA, Immunotherapy, Kidney Cancer WHAT IS ALREADY KNOWN ON THIS TOPIC WHAT THIS STUDY ADDS This study definitively identifies a repertoire of immunogenic, non-canonical tumor antigens derived from human endogenous retroviruses and long non-coding RNAs in RCC, and functionally proves that T cells engineered to target these antigens can mediate potent tumor killing. This study definitively identifies a repertoire of immunogenic, non-canonical tumor antigens derived from human endogenous retroviruses and long non-coding RNAs in RCC, and functionally proves that T cells engineered to target these antigens can mediate potent tumor killing. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Our findings resolve the key question of immunogenicity in low-TMB RCC and establish non-canonical antigens as promising, actionable targets for developing “off-the-shelf” T cell-based immunotherapies, potentially extending this approach to other low-mutation cancers. Our findings resolve the key question of immunogenicity in low-TMB RCC and establish non-canonical antigens as promising, actionable targets for developing “off-the-shelf” T cell-based immunotherapies, potentially extending this approach to other low-mutation cancers. Introduction Renal cell carcinoma (RCC), a common malignancy of the urinary system, exhibits a globally increasing incidence. A high tumor mutational burden (TMB) is an established biomarker predicting response to immune checkpoint blockade (ICB) in cancers such as melanoma and non-small cell lung cancer.1 Paradoxically, RCC demonstrates sensitivity to ICB despite having a relatively low TMB compared with other immunogenic tumors.2 3 Furthermore, although RCC is characterized by high levels of tumor-infiltrating effector T cells, this robust infiltration is often associated with poor prognosis,4 5 and the mechanisms underlying T-cell recognition in RCC remain incompletely understood. Even in a phase III trial where pembrolizumab significantly improved disease-free survival and overall survival, tumor recurrence remained frequent.6 These clinical features collectively position RCC as an ideal model for investigating alternative antigen sources, such as non-canonical tumor-specific antigens (TSAs), for immunotherapy. Tumor antigens presented on the cell surface can trigger immune surveillance by cytotoxic T lymphocytes. CD8+ T cells specifically recognize mutated or foreign epitopes presented by human leukocyte antigen class I (HLA-I) molecules to discriminate cancer cells from normal cells.7 8 Most known TSAs, particularly neoantigens, originate from canonical protein-coding regions and are strongly correlated with the efficacy of ICB.9 10 However, neoantigens are typically patient-specific, and only a small fraction of mutated epitopes are immunogenic.11,13 This necessitates the challenging task of identifying immunogenic neoantigens from the vast number of mutations in each patient,1112 14,16 posing significant obstacles for developing neoantigen-based vaccines, especially in low-TMB tumors like RCC. Recent studies have increasingly focused on the discovery of non-canonical antigens, which arise from aberrant translation of putatively non-coding transcripts and from aberrant or dysregulated transcription of non-coding genomic regions, untranslated regions (UTRs), or human endogenous retroviruses (hERVs).17,20 These irregular transcriptional and translational events generate peptide sequences that are absent from conventional protein sequence repositories.21 22 When such peptides are presented as tumor-specific and immunogenic HLA ligands, they could significantly expand the repertoire of targetable epitopes for immunotherapy, particularly in tumors with low TMB.20 23 24 hERVs, which are remnants of ancient retroviral infections constitute approximately 8% of the human genome, are of particular interest.25 While generally silenced in normal tissues, specific hERV families are aberrantly activated in many cancers, and their expression has been linked to T-cell responses.26 27 Notably, hERV expression correlates with improved survival and enhanced response to ICB, suggesting a potential role as shared TSAs.27 28 Although T-cell recognition of an hERV-derived antigen has been reported in RCC, technical limitations have hindered a comprehensive analysis of this antigen class. Moreover, it remains unknown whether TSAs derived from hERVs or long non-coding RNAs (lncRNAs) can serve as effective therapeutic targets to induce tumor regression in RCC. To address this gap, we applied an integrated proteogenomic approach, combining immunopeptidomics with exome and transcriptome sequencing, to systematically profile the HLA-I immunopeptidome of RCC tissues. This strategy enabled the identification of multiple immunogenic TSAs originating from hERVs and lncRNAs. We subsequently primed and expanded antigen-reactive T cells in vitro, isolated a panel of antigen-specific T-cell receptors (TCRs), and characterized their functionality. Finally, using patient-derived tumor-like cell clusters (PTCs) and mouse models, we demonstrated that TCR-engineered T cells targeting these non-canonical TSAs effectively mediate tumor cell killing and regression. Materials and methods Tissue specimens Tumor tissues and blood samples were obtained from patients who underwent tumor resection. All samples must have a clear pathological diagnosis, confirming RCC, and patient blood samples must be available for isolation of peripheral blood mononuclear cells (PBMCs). Detailed patient information is summarized in online supplemental table 1. Cell preparation and culture PBMCs were cultured in complete AIM-V (Gibco, Cat#12055091) medium supplemented with 1% penicillin-streptomycin (HyClone, Cat#SV30010), 1% GlutaMAX (Gibco Cat#35050061), 10 mM HEPES, 1 mM sodium pyruvate, and 10% fetal bovine serum (FBS) (HyClone, Cat#SH30406.05). HEK293T cells were cultured in complete Dulbecco’s Modified Eagle’s Medium (Gibco, Cat#C11995500BT) supplemented with 10% FBS and 1% penicillin-streptomycin. 786-O, Jurkat, T2, and K562 cell lines were cultured in complete RPMI (Gibco, Cat#C22400500BT) medium supplemented with 10% FBS and 1% penicillin-streptomycin. The 786-O, HEK293T, Jurkat, T2, and K562 cell lines were purchased from the Cell Resource Center, Peking Union Medical College. RCC primary cell line generation Thawed cryopreserved tumor cells were washed and plated in tissue culture plates containing OptiMEM medium (Gibco, Cat#31985070) supplemented with 10% FBS, 1 mM sodium pyruvate, 1% penicillin-streptomycin, and 5 ng/mL epidermal growth factor (Sigma-Aldrich, Cat#E9644). After 1 day, non-adherent cells (including immune cells) were removed by replacing the culture medium with fresh medium. Adherent cells were subsequently dissociated and passaged using Versene (Gibco, Cat#15040066). The expanding RCC cell lines tested negative for Mycoplasma and were verified as RCC cells by flow cytometry using anti-human CAIX-FITC (Santa Cruz, Cat# sc-365900). Liquid chromatography-tandem mass spectrometry analysis of immunopeptides This method was adapted from a published method.29 Briefly, W6/32 crosslinked protein A Sepharose (Thermo Fisher, Cat#101042) was prepared in-house. The lysis buffer consisted of 0.25% sodium deoxycholate, 0.2 mM iodoacetamide, 1 mM ethylenediaminetetraacetic acid, a 1:200 protease inhibitor cocktail, 1 mM phenylmethylsulfonyl fluoride, 1% octyl-b-D glucopyranoside, and phosphate-buffered saline (PBS). The tissues were cut into small pieces. Next, the lysate was slowly passed through a W6/32 column over 2 hours. The HLA-I peptide complex was eluted with 5–10 mL of 0.1 M acetate solution. HLA-I peptides were separated from larger HLA-I components (α-chain and β2-microglobulin) via a UHPLC 3000 (Thermo Fisher, USA). Finally, the fractions containing the immunopeptides were dried via vacuum centrifugation and stored at −20°C until analysis. The dried immunopeptides were resuspended in 20 µL of 0.1% formic acid and centrifuged at 20,000×g and 4°C for 20 min. Next, the supernatant was transferred to separate sample vials. The immunopeptides were separated via nanoflow high-performance liquid chromatography (LC) coupled to an Eclipse Orbitrap Eclipse Tribrid (T5 only) mass spectrometer (Thermo Fisher, USA) with a nanoelectrospray ion source. A 75 µm×15 cm C18 Acclaim PepMap RSLC column (Thermo Fisher, USA) was used for sample analysis. 5 µL of each sample were injected, and a technical replicate was measured simultaneously. Peptides were eluted with a linear gradient of 2–90% buffer B (80% acetonitrile and 0.1% trifluoroacetic acid in water) at a flow rate of 300 nL/min over 140 min. The top speed mode with a 2 s cycle was used for precursor selection, which isolated and fragmented the precursors via high-energy collisional dissociation. The resolution of the MS1 spectrum was set to 120,000, and the mass scan range was set to 300–1,800. The most intense ions were sequentially isolated and accumulated to an AGC target value of 5e5 with a maximum injection time of 120 ms. For the immunopeptide measurement, precursors with charge states between 1+ and 4+ were fragmented. The mass spectrometry (MS)/MS resolution was set to 60,000, the Automatic Gain Control (AGC) target value was set to 1.5×105, and the collision energy was set to 28%. All data analysis was performed via Peaks X Pro software (Bioinformatics Solutions). RNA sequencing and exome sequencing RNA was extracted using the Total RNA Isolation RNeasy Mini Kit with DNAse I (QIAGEN, Cat#74014) via the on-column digestion step. RNA quality was assessed using a Fragment Analyzer (Agilent Technologies). RNA sequencing (RNA-seq) libraries were prepared from 500 ng of total RNA using Illumina TruSeq Stranded messenger RNA reagents via a unique dual indexing strategy following the official protocol automated on a sciclone liquid handling robot (PerkinElmer). Libraries were quantified using a fluorometric method (QubIT, Life Technologies), and their quality was assessed using a Fragment Analyzer. Cluster generation was performed with the resulting libraries using Illumina HiSeq 3000/4000 PE Cluster Kit reagents. Libraries were sequenced on the Illumina HiSeq 4000 platform using HiSeq 3000/4000 SBS Kit reagents for 2×150 cycles. The sequencing data were demultiplexed using the bcl2fastq Conversion Software (V.2.20, Illumina). The collected samples were sent to the core facility of Novogene Technology (Beijing, China), for whole-exome sequencing (WES) analysis. Valid sequencing data were mapped to the reference human genome (UCSC hg19) using the Burrows-Wheeler Aligner software to obtain the original mapping results in BAM format. SAMtools30 mpileup, and BCFtools were used to perform variant calling and identify single nucleotide variants and InDels. Somatic single nucleotide variants were detected using muTect,31 whereas somatic InDels were identified using Strelka.32 HLA typing HLA typing high-resolution four-digit HLA-I typing was performed on extracted genomic DNA using the HLA amplification method with TruSight HLA V.2 A sequencing panel kit (Novogene, Beijing, China) was used. The data were analyzed using Assign TruSight HLA V.2.1 software (Novogene, Beijing, China). RNA sequencing analysis RNA-seq data were subjected to quality control, filtered, and aligned to the human genome. Salmon Quant was used for quantification to generate a transcriptome matrix based on annotations of hg38 reference genes. To quantify lncRNA expression, we performed quantification based on the hg38 lncRNA annotation to generate the lncRNA expression matrix. For ERV detection, the transcriptome data were aligned to the human hERV reference genome. Quantification was performed according to hERV annotation to obtain the hERV expression matrix. Based on lncRNA and hERV expression in the samples, we obtained gene sequences and translated lncRNA transcripts via the three-frame reading method and hERV transcripts starting from the initial codon to generate patient-specific lncRNA and hERV protein libraries. Finally, we exported the protein translation library in a fixed MS database format. RNA-seq analysis and lncRNA/hERV sequence database generation RNA-seq reads were aligned to the GRCH37/hg19 reference genome using RNA-Star (V.2.7.3a; https://github.com/alexdobin/STAR). Raw counts were transformed into Transcripts Per Million (TPM) values. Comprehensive gene annotation V.32 was downloaded from the GENCODE website, and chromosome position, transcript structure, and transcript information were obtained. Protein sequences were selected to define protein-coding and non-coding genes. For all plots, including the RNA-seq data, we used log2 transformation with a pseudocount of 1. In addition, we mapped the RNA-seq reads to non-coding or hERV regions, as previously described. Normalization for sequencing depth was performed for both the non-coding and hERV regions via the trimmed mean of M values method using the limma V.3.36.5 package of Bioconductor, and the number of genes was used as the library size. Expressed (TPM >0.0) lncRNA sequences were translated into three forward reading frames, as identified using a stop-to-stop strategy. The hERV sequences were translated from the initiation codon “ATG” via a stop strategy. Reference sequences, personalized protein-coding sequences, expressed lncRNAs, and hERV sequence entries were merged into a single, sample-specific, personalized proteome. Peptide synthesis and validation Peptides with 95% purity (produced by Fmoc solid-phase chemical synthesis) were purchased from Genscript Biotechnology and resuspended in dimethyl sulfoxide (DMSO) at 10 mg/mL. Analysis of candidate peptide immunogenicity The immunogenicity of the candidate peptides was assessed in vitro in autologous PBMCs using precoated 96-well Enzyme-Linked ImmunoSpot Assay (ELISpot) plates (Dakewei, Cat#110005), according to the manufacturer’s protocol and a published method. 4–1BB expression by CD8+ T cells was detected by flow cytometry after 10 days of stimulation using the procedure described in the interferon (IFN)-γ ELISpot assay section. Human TruStain FcX (BioLegend, Cat#422302), 7-AAD (BioLegend, Cat#420403), anti-human CD3-PE (BioLegend, Cat#300907), anti-human CD8-APC (BioLegend, Cat#300912), and anti-human 4-1 BB-PE (BioLegend, Cat#309804) were used to stain the stimulated PBMCs at 4°C for 30 min in the dark. Next, the cells were washed and resuspended in fluorescence-activated cell sorting (FACS) wash buffer. FACS data were acquired using a Fortessa flow cytometer (BD Biosciences) and analyzed using FlowJo V.10 software. The proportion of 4–1BB+ cells among CD3+ CD8+ T cells indicated immunogenicity of the candidate peptides. Peptide binding assay Multiple K562-HLA+-TAP1−/− cell lines were constructed in our laboratory and were used to determine the affinity of exogenous peptides for the HLA of interest.29 After incubation, the cells were washed and the surface levels of HLA-I were assessed by staining with Allophycocyanin (APC)-conjugated anti-human HLA-A, HLA-B, and HLA-C (BioLegend, Cat#311410) at 4°C for 15 min in the dark. HLA-bound fluorescence was measured using flow cytometry. The Fluorescence Index (FI) was calculated as follows: FI=(mean APC fluorescence with a particular peptide)/(mean APC fluorescence without the peptide). Identification of neoantigen-reactive T-cell receptors A modified Single-Cell Tagged Reverse Transcription sequencing (STRT-seq)-based method was used to identify neoantigen-reactive TCR sequences.33 34 First, T cells that upregulated an activation marker 4–1BB after co-culture of autologous PBMCs with peptides were FACS-sorted into wells of a 96-well plate containing lysis buffer. The lysis buffer contained 1 µL of 10 mM dNTP mixture (Vazyme, Cat#P031-01), 1 µL of 10 mM oligo dT primer (Tsingke), 1.9 µL of 1% Triton X-100 (Beyotime, Cat#P0096) and 0.1 µL of 40 U/mL RNase inhibitor (Vazyme, Cat#R301-01). Single-cell transcriptome amplification was performed following the Smart-seq2 protocol with a modified TSO primer. TCR-alpha and TCR-beta chains were further amplified with TCR-specific primers. The amplified TCR complementary DNA products were purified with clean DNA beads (Vazyme, Cat#N411-01) two times and quantified with a Qubit V.3.0 fluorometer (Thermo Fisher Scientific). Libraries were then constructed with the Nextera XT DNA Sample ll Preparation Kit (Illumina) according to the manufacturer’s instructions and sequenced on an iSeq sequencer (Illumina) with 150 bp paired-end reads. The full-length TCR sequences were assembled via the MiXCR and TraCeR methods from RNA-seq data. Construction of the full-length TCR Full-length neoantigen-reactive TCRs were constructed by fusing TRA V-J-encoding sequences and TRB V-D-J-encoding sequences with the modified mouse TRA and TRB constant chains, as previously described.34 Briefly, the partial TCR sequences obtained by sequencing methods were analyzed using the IMGT/V-Quest tool, which identified the TRAV and TRBV families with the highest likelihood of containing the identified segments identified by sequencing methods. The murine constant chains were modified to allow preferential pairing of the introduced TCR chains, enhance TCR surface expression and functionality, and facilitate identification of transduced T cells by flow cytometry using anti-mouse-TCR-β-chain antibodies (BioLegend, Cat#109205).35 36 The sequences for the TCR alpha and beta chains were linked together with a furin SGSG P2A linker and then codon-optimized, synthesized and cloned and inserted into the lentiviral vector pWLSV08. For TCRs with two putative TCR-alpha chains paired with one/two beta chains, all TCRs were constructed and evaluated for reactivity. Generation of T cells with transgenic TCRs To facilitate detection of transduction efficiency, we replaced the human TCR constant region with the mouse TCR constant region. HEK293T cells were transfected with the TCR cloning vector plasmid, as well as the pMD2.G and psPAX2 envelope plasmids using Lipofectamine 3000 (Thermo Fisher, Cat#L3000015). Lentiviral supernatants were harvested at 48 hours and 72 hours after transfection and then precipitated via a lentivirus precipitation solution (TransGen, Cat#FV101–01). T cells were sorted from donor PBMCs via CD3/28 DynaBeads (Thermo Fisher, Cat#11131D) and cultured in AIM-V medium supplemented with 100 U/mL interleukin (IL)-2. After 48 hours, activated T cells were co-cultured with precipitated lentivirus at 2,000 rpm for 2 hours and cultured with lentiviral supernatant for 3 days. At 5 days after transduction, the CD3/28 DynaBeads were removed, and the T cells were cultured in AIM-V medium supplemented with 20 U/mL IL-2. TCR expression efficiency was determined by flow cytometry using an anti-mTCRB-FITC antibody (BioLegend, Cat#1009206). Jurkat cells were grown on a 6-well plate 1 day before transduction to obtain Jurkat cells expressing a particular TCR. TCR-CD8-Jurkat cells were generated by treating Jurkat cells with high-titer TCRs and 8×NFAT-tdTomato-hCD8 lentiviruses. TCR-CD8-Jurkat cells were coincubated in a 96-well plate with antigen-primed K562-A02/11/24 cells at a 1:1 ratio. The percentage of tdTomato-expressing CD8+ mTCR+ Jurkat cells after 48 hours of coincubation was analyzed via flow cytometry. Functional assays for T-cell recognition of target cells The ability of TCR-T cells to recognize and eliminate primary tumor cells was assessed by flow cytometry.37 TCR-T cells were co-cultured with primary RCC tumor cells derived from patients in 96-well plates using complete AIM-V medium. Different effector to target ratios were used as indicated. As controls, we included mock-transduced T cells or primary tumor cells obtained from patients who did not harbor the HLA allele of interest. After 24 hours, viable cells were counted, and the percentages of target cells were determined by flow cytometry. The elimination index was calculated as follows: 1 − (number of residual target cells in the presence of TCR-T cells/number of residual target cells when seeded alone). Single-cell data analysis Single-cell expression matrices and single-cell TCR matrices were obtained via the CellRanger command from the raw data. We filtered out low-quality data by analyzing the distribution of expressed genes across all single cells and removed transcripts detected in fewer than three cells, cells with excessively low gene expression counts and excessively high mitochondrial expression. After standardizing, scaling and correcting the expression data, we used SCTransform normalization to obtain expression signature genes and performed principal component analysis (PCA) on the expression matrix. On the basis of the PCA results, we used Harmony to correct batch effects, clustered and annotated cells according to biological features, and reduced dimensionality via Uniform Manifold Approximation and Projection (UMAP). We integrated transcriptomic and TCR data via scRepertoire to calculate clonal abundance and clonal richness. To characterize antigen-associated T-cell populations, TCR sequences were further compared with different reference libraries according to sequence identity and annotation information, and each TCR was assigned to a corresponding category, such as lncRNA-associated and virus-associated. Based on these TCR annotations, T cells were grouped into distinct antigen-specific subsets for downstream comparative analysis. Differential expression analysis among different antigen-specific T-cell subsets was performed using the FindAllMarkers function in Seurat with the Wilcoxon test. Top ranked differentially expressed genes (DEGs) (by log-fold changes) with a log2 fold changes >0.8, expression in at least 25% of cells in the compared group, and adjusted p value<0.05. For each antigen-specific T-cell subset, marker genes were identified by comparing cells from one subset against virus-reactive T cells in the dataset. Gene Ontology (GO) enrichment analysis was performed on the DEGs of each antigen-specific T-cell subset. Enriched GO terms with adjusted p values<0.05 were considered significant and were used to interpret the functional characteristics and biological processes. Establishment of the PTC models PTC models were established as previously described.38 Surgically resected tumor samples were washed five times with PBS, and necrotic areas and adipose tissues were removed as much as possible. The tissues were then minced and digested in 5 mL of PBS buffer containing 1 mM EDTA and 1 mg/mL collagenase I, II, or IV (Yuanye Biology, Shanghai, China, Cat#S10053, Cat#S10054, Cat#S10056) at 37°C for 1 hour. The digestion mixture was pipetted every 15 min to obtain single cells, and 40 mm filters were used to collect dissociated cells. The cell suspensions were centrifuged (300×g, 10 min) and the cell pellets were resuspended in PTC growth medium. Malignant pleural effusions and ascites were centrifuged at 300×g for 5 min, and the cell pellets were resuspended in PBS and added to lymphocyte separation medium (MPbio, Cat#1692249) in a centrifuge tube. After centrifugation (400×g, 20 min), the middle cell layer was collected and resuspended in PTC growth medium. All cell suspensions from surgical resections, ascites, and pleural effusions in PTC growth medium were seeded in low-attachment-surface dishes at a concentration of 1×105 cells/cm2 and cultured in an incubator at 37°C and 5% CO2. PTC growth medium was changed every 2–3 days. PTCs were passaged via incubation in TrypLE Express (Thermo Fisher Scientific, Cat#12604013) for 5–10 min at 37°C to dissociate PTCs into single cells and replace them with fresh PTC growth medium. In vitro antitumor activity of TSA peptide pool in PTC models The lymphocyte-activated culture medium consisted of AIM-V (Thermo Fisher, A3021002), 10% FBS, and 20 IU/mL IL-2 (Cat#200-02-10; PeproTech) for 48 hours. To assess the inhibitory effects of TSA peptide pool on PTC models, we adapted the protocol established by Cattaneo et al.39 First, 100 PTCs were co-cultured with 1×105 activated HLA-paired lymphocytes seeded into U-bottom plates (Corning, Cat#7007) in 100 µL of medium. The medium was then treated with the negative control (no peptide) group (NC), or TSA peptide pool, and cultured at 37°C and 5% CO2. The co-cultured assay was maintained for approximately 7 days to allow sufficient interaction and observation of the effects of TSA peptide pool. Detection of PTC viability PTC viability assay was conducted in 96-well cell culture plates following a previously described procedure. Briefly, before co-cultured with lymphocytes, PTCs were seeded onto Teflon-modified chips (GeneX Health, Cat#GX-01). Images of each well were obtained using a Nikon Ti-U microscope system. After 7 days of co-culture, the plates were screened again using a Nikon Ti-U microscope system. Cell viability was determined using an image-based approach and calculated using the following formula: p(x)= St1/St0, p= 1n∑i=1np(x). St1 and St0 are the sum of PTC areas in a well after and before drug treatment, respectively, p(x) is the PTC cell viability of the single well, n is the number of replicates, and p is the average PTC cell viability of all replicates. The PTC areas were analyzed and quantified using ImageJ software. In vivo antitumor activity in a xenograft tumor model 786-O-A24 cell lines in which three endogenous HLA-A, HLA-B and HLA-C locus genes were knocked out and HLA-A*24:02 CDS regions were overexpressed. The following guides were selected: CRISPR-HLA-A: CGGCTACTACAACCAGAGCG CRISPR-HLA-B: AGATCACACTGACCTGGCAG CRISPR-HLA-C: AGGTCAGTGTGATCTCCGCA In the TAP−/− 786-O-A24 cell lines, where the TAP1 gene was knocked out, the following two guide RNAs were selected: CRISPR-TAP1-1: AGTTCGAAGCTTTGCCAACG CRISPR-TAP1-2: CATCATGTCTCGGGTAACAG The gRNAs were subsequently cloned and inserted into the LentiCRISPR V2 vector via the BsmBI sites. A total of 1×106 786-O cells were transfected with 3 µg of each plasmid guide construct via Lipofectamine 3000. After 14 days, HLA-I negative cells were sorted via the Pan HLA-I antibody (BioLegend, Cat#311410). Clones were grown from single cells, and the absence of HLA-I was verified by flow cytometry. 786-O-HLA−/− cells (1×106) were transfected with 10 µg of the HLA-A*24:02 CDS plasmid via Lipofectamine 3000. After 7 days, HLA-I positive cells were sorted via the Pan HLA-I antibody. 786-O-A24 cell clones were grown, and the expression of HLA was verified by flow cytometry. In order to obtain TAP−/− 786-O-A24 cell lines, we cloned TAP1 gRNAs into the LentiCRISPR V2 vector via the BsmBI sites. A total of 1×106 786-O-A24 cells were transfected with 3 µg of each plasmid guide construct via Lipofectamine 3000. After 14 days, TAP−/− 786-O-A24 cells were sorted by flow cytometry using a Pan HLA-I antibody. For the tumor model, highly immunodeficient female NCG mice (Viewsolid Biotech, Beijing, China) were used, and all the animals were housed under specific pathogen-free conditions. The mice were inoculated with 786-O-A24 cells into one of the flanks of each mouse subcutaneously. 2 weeks after engraftment, TCR-T cells (1×107 cells/mouse) were intravenously injected, and three doses of IL-2 (2×105 U/mouse) were subcutaneously injected into the 786-O-A24 tumor model. Mock-T cells transduced without TCR were injected as a negative control. Four doses of anti-programmed cell death protein 1 (PD-1) antibodies (αPD-1, 5 mg/kg per dose) were injected into additional TCR-T and mock-T cell groups intraperitoneally two times per week in the tumor model. The growth rates were measured by measuring the length and width of the tumors with calipers every two times a week. Tumor growth was monitored two times per week, and the volume of the tumors was calculated via the following formula: ½ length×width2. The experimental endpoint was set at the earliest stage when significant differences were observed among the treatment groups. The maximum limits of tumor burden include a tumor volume >4,000 mm3 in mice, a tumor weight >10% of the body weight, or ulceration, infection or necrosis of the tumor. The mice were sacrificed in a CO2 chamber. The tumors were separated from each mouse and weighed. The mice were ranked according to their initial tumor burden and then randomly assigned to the PBS, α-PD-1 alone, mock-T cell, or TCR-T cell treatment groups. The experimental endpoint was set at 2–3 weeks after the injection of TCR-T cells. Animal care was carried out in accordance with the guidelines of the Animal Care and Use Committee of the Institute of Microbiology, Peking University. Statistics and reproducibility Statistical analysis of multiple comparisons was performed using one-way analysis of variance, and comparisons between two groups were performed using Student’s unpaired t-test. The significance of overall survival was determined via Kaplan-Meier analysis with log-rank (Mantel-Cox) analysis. Data were analyzed using Microsoft Excel (V.16.16.27), and all statistical analyses were performed using GraphPad Prism V.9.0 (GraphPad). The error bars represent the SDs, and statistical significance was set at p<0.05. *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001 unless otherwise indicated. ns, not significant. All data presented are representative of two or more independent experiments. Results Identification of HLA class I-presented peptides in patients with RCC To identify tumor-derived peptides from hERVs and lncRNAs as potential antitumor targets, we first characterized the antigenic landscape of RCC (online supplemental figure 1). We performed immunopeptidomic analysis on 22 RCC samples to isolate and identify major histocompatibility complex class I (MHC-I)-associated peptides (MAPs). Patient clinical characteristics and HLA typing information are summarized in online supplemental table 1. To generate personalized reference databases, we performed WES on all 22 samples and RNA-seq on 14 of the available samples (P9–P22). MAPs were identified using LC-MS,29 40 and for each sample, we constructed personalized reference databases through in silico translation of sequencing data. These databases included the canonical proteome, either alone or combined with one of the following: (1) cancer-testis antigen (CTA) proteome, (2) mutation-derived tumor-specific antigen (mTSA) proteome, (3) lncRNA-derived proteome, or (4) hERV-derived proteome (figure 1A) . Landscape of the RCC immunopeptidomes. (A) Workflow for proteogenomic analysis of tumor tissues. HLA-I complexes were immunoprecipitated using the W6/32 antibody, and eluted peptides were analyzed by MS. Peptide identification was performed using personalized custom databases. (B) Length distribution of MAPs identified per patient. (C) Proportion of HLA-I binders in each sample, as classified by NetMHCpan4.1. (D) WES data for all 22 RCC samples. Top bar plot shows TMB; the heatmap depicts the top 20 frequently mutated neoantigen-associated genes; right bar plot indicates mutation types and counts for each gene. (E) Proteogenomic pipeline for identifying MAPs derived from lncRNAs and hERVs. (F) Number of predicted MAPs originating from lncRNA-derived and hERV-derived proteomes. (G) Summary of all MAPs identified across RCC samples. ATG, initiation codon;CDS, Coding Sequence; CTA, cancer-testis antigen; hERV, human endogenous retrovirus; HLA, human leukocyte antigen; lncRNA, long non-coding RNA; MAPs, major histocompatibility complex class I-associated peptides; MS, mass spectrometry; mRNA, messenger RNA; mTSA, mutation-derived tumor-specific antigen; RCC, renal cell carcinoma; SB, strong binder; TMB, tumor mutational burden; WB, weak binder; WES, whole-exome sequencing. This integrated approach identified over 120,000 unique MAPs mapping to the human UniProt proteome across the 22 tumors (mean: 5,628 peptides per sample; range: 1,400–12,814) (online supplemental figure 2A). The length distribution of these MAPs was consistent with typical immunopeptidome characteristics, showing a predominant fraction of 9–11-amino acid long peptides (figure 1B and online supplemental figure 2B). HLA-I typing revealed presentation by 15 different HLA-A, 23 HLA-B, and 15 HLA-C alleles, with over 50% of samples expressing one of three common alleles: HLA-A*02:01, HLA-A*11:01, or HLA-A*24:02 (online supplemental figure 2C, online supplemental table 1). To validate data quality, we performed deconvolution analysis using NetMHCpan4.1.41 As expected, the majority (80–90%) of MAPs from each sample were predicted to bind to their corresponding HLA alleles with various affinities (figure 1C). We next investigated the presence of mTSAs in the RCC immunopeptidome. WES was performed on the 22 tumor tissues and matched PBMCs, achieving average sequencing depths of 300× and 100×, respectively. Analysis showed that missense single nucleotide variants were the predominant mutation type, with an average TMB of 77.5 mutations per sample. Molecular profiling of this cohort confirmed typical RCC features, including mutations in VHL (12/22, 54.5%), PBRM1 (8/22, 36%), and BAP1 (4/22, 18%), with both clonal and subclonal alterations detected (figure 1D).42 Despite constructing personalized mutation-derived databases and applying a false discovery rate threshold of 0.05, no mTSAs were identified from the immunopeptidomes. This absence likely reflects the lower detection sensitivity of MS compared with next-generation sequencing rather than a true lack of mTSAs in RCC. Collectively, these results demonstrate the robustness of our immunopeptidomic pipeline for profiling HLA-I presented peptides. Proteogenomic identification of MAPs peptides from lncRNAs and hERVs in RCC Given prior evidence of elevated hERV expression in RCC and its correlation with response to ICB,28 alongside the known presence of non-canonical peptides in the HLA-I ligandome,43 44 we systematically characterized MAPs derived from hERVs and lncRNAs in RCC (figure 1E). We employed a proteogenomic approach involving three-frame translation for lncRNA transcripts and ATG to stop codon translation for hERV transcripts from the reported 3,173 hERV genes to predict potential peptides.45 46 To provide translational evidence for these candidates, we cross-referenced them with ribosome sequencing (Ribo-seq) data from RCC specific and pan-tissue datasets.47,49 This integrated proteogenomic approach identified 31 hERV-derived and 42 lncRNA-derived MAPs across 13 of the 14 RNA-seq profiled samples (figure 1F and G and online supplemental table 2). None of these novel peptides were present in the Immune Epitope Database (IEDB) database,50 confirming their non-canonical origin. Analysis of the 73 MAPs revealed that 9 were shared across multiple RCC samples (online supplemental table 3). Notably, two lncRNA-derived MAPs (P11-lnc1 and P13-lnc2) had been previously reported as non-canonical TSAs in melanoma, providing external validation. These results collectively demonstrate that transcripts from lncRNAs and hERVs can be translated and presented on HLA-I molecules in RCC. Next, we evaluated the immunogenicity of these MAPs by establishing an in vitro T-cell priming assay. In brief, autologous PBMCs were stimulated with candidate peptides, and T-cell reactivity was assessed via IFN-γ ELISpot and flow cytometry for the activation marker 4–1BB (figure 2A). And we obtained five hERV and two lncRNA MAPs in P21 (figure 1F and and online supplemental table 2). All peptides identified in P21 were subsequently selected for immunogenicity validation (figure 2A). We identified three immunogenic epitopes (P21-erv1, P21-erv5, and P21-lnc2), which elicited a significant enhancement in both IFN-γ production (figure 2B and D) and 4–1BB upregulation compared with the negative controls (*p<0.05) (figure 2C and E). Extending this analysis to all identified MAPs from the 13 samples revealed that a subset was immunogenic: 7 hERV-derived (figure 2F) and 9 lncRNA-derived (figure 2G) MAPs elicited specific T-cell responses in autologous settings (online supplemental figure 3, online supplemental table 5). These findings confirm that a proportion of peptides originating from hERVs and lncRNAs can function as immunogenic TSAs, capable of driving functional T-cell responses in RCC. Immunogenicity of lncRNA-derived and hERV-derived MAPs in RCC. (A) Workflow for in vitro T-cell stimulation. Autologous PBMCs were stimulated with peptides (10 µM) for 10 days, followed by flow cytometry and IFN-γ ELISpot analysis. (B and C) Representative T-cell responses to lncRNA and hERV MAPs after 10 days of peptide stimulation, assessed by IFN-γ ELISpot (B) and flow cytometry (C). (D and E) Summary of T-cell responses to lncRNA and hERV MAPs measured by IFN-γ ELISpot (D) and flow cytometry (E). Data are presented as mean±SD from three independent experiments. (F and G) Immunogenicity profiling of hERV (F) and lncRNA (G) MAPs in co-cultured PBMC assays. Responses were considered positive if ELISpot counts and 4–1BB+ CD8+ T-cell frequencies were significantly elevated relative to the negative control. Gray dots: MAPs with no significant response in either assay. Red dots: MAPs with significant responses in only one assay. Blue dots: MAPs with significant responses in both assays (ELISpot counts and 4–1BB+ CD8+ T-cell frequency assay). (H) Schematic of single-cell TCR sequencing performed on pooled antigen-reactive CD8+ T cells. (I) Manhattan plot showing VDJ recombination profiles of TCRα/β chains from reactive clones. CDR3 sequences of TCRβ (x-axis) and TCRα (y-axis) are displayed, with clone frequency represented on the z-axis. Data were analyzed by one-way ANOVA; *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. ANOVA, analysis of variance; ELISpot: Enzyme-Linked ImmunoSpot Assay; FACS, fluorescence-activated cell sorting; hERV, human endogenous retrovirus; IFN, interferon; IL, interleukin; lncRNA, long non-coding RNA; MAPs, major histocompatibility complex class I-associated peptides; Neg, negative control peptide; PBMC, peripheral blood mononuclear cell; PHA, phytohemagglutinin (positive control); RCC, renal cell carcinoma; TCR, T-cell receptor. Identification of HLA class I-presented peptides derived from CTAs Concurrently, we investigated the presentation of CTAs within the RCC immunopeptidome. By screening our data against the CTpedia database (containing 276 known CTAs),51 we identified 39 MAPs derived from reported CTAs across the 22 tumor samples (online supplemental figure 4A, online supplemental table 4). RNA-seq analysis confirmed significant upregulation of the corresponding CTA genes in tumor tissues compared with matched normal samples, although with detectable low-level expression in some normal tissues (online supplemental figure 4B). While most identified CTA peptides were patient-specific, a subset derived from genes such as PMEL, TEX15, SPAG9, and ODF2 were shared across multiple patients (online supplemental table 5). We subsequently evaluated the immunogenicity of these CTA-derived peptides using the same in vitro T-cell priming assay (figure 2B). In this assay, cytomegalovirus, Epstein-Barr virus, and influenza (CEF) peptide pools (comprising 31 HLA-I antigens from cytomegalovirus, Epstein-Barr virus, and influenza virus) served as a positive control for T-cell activation, while DMSO was used as a negative control. For instance, in patient P6, one of four tested peptides (P6-2) elicited a measurable T-cell response, as indicated by IFN-γ production (online supplemental figure 4C,E) and 4–1BB upregulation compared with the negative controls (*p<0.05) (online supplemental figure 4D,F). Extending this analysis across the cohort revealed that 8 of the 39 CTA peptides induced specific T-cell responses in autologous PBMCs (online supplemental figure 4G,5). A search of the IEDB confirmed that two of these immunogenic peptides (P1-2 and P4-1) represent novel epitopes. In summary, our immunopeptidomic profiling of low-TMB RCC tumors has identified a diverse repertoire of potential target antigens originating from lncRNAs, hERVs, and CTAs. The collective findings underscore the significant potential of targeting non-mutational antigen sources to develop immunotherapies for this malignancy. Transcriptional programs of non-canonical antigens reactive T cells in RCC To determine whether non-canonical antigen-reactive TCR clonotypes are enriched in activated T cells, we stimulated autologous PBMCs with antigen peptides (figure 2A). CD8+ T cells that upregulated the activation marker 4–1BB were then isolated by FACS and sorted into individual wells of a 96-well plate for subsequent single-cell TCR sequencing using an STRT-seq-based method (figure 2H).33 34 Sequencing of TCRα and β chains from cells stimulated with immunogenic antigens—such as the lncRNA-derived peptide P16-lnc2—confirmed the expansion of dominant antigen-specific TCRαβ clonotypes (figure 2I). In total, we successfully assembled eight full-length TCRαβ clonotypes reactive against three hERV-derived and five lncRNA-derived TSAs (online supplemental table 6). These results indicate that peripheral antigen-reactive CD8+ T cells can be effectively isolated based on a specific activation marker profile, enriching for antitumor TCR clonotypes. To characterize the phenotypic states of CD8+ T cells in the tumor microenvironment (TME), we performed single-cell RNA sequencing on tumor-infiltrating lymphocytes (TILs) from RCC samples (online supplemental figure 6A). Our dataset comprised transcriptomes from 40,775 single CD8+ T cells, predominantly from the seven RCC tumors for which we had identified antigen-reactive TCRs (online supplemental table 6). Unsupervised clustering revealed eight distinct CD8+ T-cell subsets (figure 3A–C, online supplemental figure 6B, onlinesupplemental tables 78), which were annotated using canonical T-cell gene signatures and cross-referenced with external human TIL datasets (figure 3D). The major subsets included exhausted T cells (Tex), progenitor exhausted T cells (Tex-pro), effector T cells, memory T cells (Tmem), and resident memory T cells (Trm). Minor populations included proliferating T cells, γδ T cells, and natural killer-like cells. Evaluation of exhaustion and NeoTCR scores indicated that the Tex and Tex-pro clusters exhibited the highest scores, suggesting an enrichment for antitumor antigen-reactive TILs (figure 3D and online supplemental figure 6C).52 53 Transcriptomic landscape of TSA-reactive CD8+ T cells from the autologous TME. (A) UMAP visualization of 40,775 CD8+ TILs from 7 RCC samples, colored by cell cluster identity based on correlation with published transcriptional signatures. (B) UMAP showing the distribution of single-cell TCR clonotypes across CD8+ TILs. (C) Bar plots showing the proportion of each T-cell cluster in each sample. Each cluster shown on the UMAP is denoted by a color code. No clusters were driven by a particular sample. (D) Relative expression of the most differentially expressed genes. (E) Bar graph illustrating the clonal expansion (TCR clonotype fraction) within each cluster. (F to H) UMAP projections highlighting the spatial distribution of T cells specific for hERV-derived TSAs (F), lncRNA-derived TSAs (G), and CEF viral antigens (H). (I and J) GO enrichment analysis of genes upregulated in hERV-reactive (I) and lncRNA-reactive (J) T cells compared with virus-reactive T cells. CEF: cytomegalovirus, epstein-barr virus, and influenza virus;GO, Gene Ontology; hERV, human endogenous retrovirus; lncRNA, long non-coding RNA; mRNA, messenger RNA; RCC, renal cell carcinoma; TCR, T-cell receptor; Teff, effector T cells; Tex, exhausted T cells; Tex-pro, progenitor exhausted T cells; TIL, tumor-infiltrating lymphocyte; TME, tumor microenvironment; Tmem, memory T cells; TNK-like, natural killer-like cells; Tprolif, proliferating T cells; Trm, resident memory T cells; TSA, tumor-specific antigen; Tγδ, γδ T cells; UMAP: uniform manifold approximation and projection. We next mapped the clonotypes of identified TCR—specific for two hERV antigens (P10-erv1, P21-erv5; representing 117 T cells) and four lncRNA antigens (P11-lnc3, P16-lnc2, P16-lnc3, P19-lnc2; representing 211 T cells)—onto the TIL clusters to define their phenotypic states (figure 3E and F). For comparison, we also mapped TCRs specific for four CEF viral antigens (112 T cells). The antigen-reactive T cells were predominantly localized within the Tex and Tex-pro clusters, whereas the viral antigen-reactive T cells were primarily found in Trm and Tmem clusters (figure 3F–H). CTA-reactive TCRs showed a similar enrichment in exhausted T-cell clusters (online supplemental figure 6E,F), consistent with previous studies where neoantigen-reactive TCRs mapped to exhausted subsets.53 54 Differential gene expression and GO analysis further revealed that hERV-reactive and lncRNA-reactive T cells significantly upregulated genes related to T-cell activation, cytokine production, and differentiation compared with non-tumor-reactive T cells (figure 3I and J). In summary, TCRs specific for both CTAs and non-canonical antigens were predominantly found within exhausted and progenitor-exhausted CD8+ T-cell populations in the TME. This contrasts with the less-differentiated memory phenotypes of viral antigen-reactive T cells and underscores the active engagement of these T cells with tumor antigens, despite their exhausted functional state. Functional validation of hERV-reactive and lncRNA-reactive T cells in vitro To functionally validate hERV-derived and lncRNA-derived antigens, we first determined the HLA-I allelic landscape of our RCC cohort. The three most frequent alleles—HLA-A*02:01, HLA-A*11:01, and HLA-A*24:02—were present in 15 samples, representing 68.2% of the cohort (figure 4A). TAP1 is a transporter protein that presents endogenous tumor antigen in the MHC complex. We then generated TAP1-deficient K562 cell lines stably expressing each of these HLA-A alleles to serve as antigen-presenting platforms for epitope affinity assays (figure 4B and online supplemental figure 8A,B). Using NetMHCpan4.1, we predicted the binding affinity of immunogenic hERV and lncRNA peptides, identifying 16 candidates with predicted binding to one of the three target HLA-A alleles (online supplemental table 9). To determine the optimal concentration for epitope affinity assay by TAP1-deficient K562 cell lines, we performed the titration of peptide concentration as described previously.55 We selected three epitopes with predicted HLA restrictions: P21-erv5 for HLA-A*02:01, P21-lnc1 for HLA-A*11:01, and P10-erv1 for HLA-A*24:02. For predicted epitopes, the binding affinity of K562 gradually increased at peptide concentrations as low as 0.1 µg/mL, with the maximum affinity reached at a concentration of 10 µg/mL (figure 4C–E and online supplemental figure 7). Therefore, 10 µg/mL was selected as the optimal peptide concentration for the following assay. The epitope affinity assay confirmed that 14 of these peptides exhibited significant HLA-I binding compared with DMSO controls (*p<0.05). Specifically, six antigens were restricted to HLA-A*02:01, four to HLA-A*11:01, and four to HLA-A*24:02 (figure 4F–K), thereby definitively establishing their HLA restriction. hERV and lncRNA-reactive T cells recognize tumor cells in vitro. (A) Frequency distribution of common HLA-A, HLA-B, and HLA-C alleles across patient samples. (B) Schematic of HLA-positive, TAP-deficient K562 cell line generation and peptide–HLA binding assay workflow. (C to E) Binding efficiency curves were generated by peptide titration using TAP-deficient K562 cells expressing HLA-A*02:01 (C), HLA-A*11:01 (D), or HLA-A*24:02 (E) pulsed with titrated concentrations of peptides. Binding capacity was presented as MFI of HLA-I staining. (F to H) Representative flow cytometry histograms from epitope binding assays using TAP-deficient K562 cells expressing HLA-A*02:01 (F), HLA-A*11:01 (G), or HLA-A*24:02 (H), pulsed with synthetic peptides. (I to K) Summary of epitope binding affinity, expressed as the ratio of peptide-pulsed MFI to DMSO-pulsed MFI, for HLA-A*02:01 (I), HLA-A*11:01 (J), and HLA-A*24:02 (K). Green dots indicate high-affinity epitopes. Data represent three independent experiments. (L) Schematic of the Jurkat-NFAT-tdTomato reporter assay using CD8+ Jurkat cells expressing a murine constant region TCR (mTCR). (M) Flow cytometry validation of hCD8+ and mTCRβ+ expression in engineered Jurkat-NFAT-tdTomato cells. (N to Q) NFAT-tdTomato activation in TCR-transduced Jurkat cells co-cultured with antigen-pulsed K562 cells. Negative control: K562 cells pulsed with an HLA-unmatched peptide. Data are mean±SD from three independent experiments; significance assessed by one-way ANOVA (**p<0.01, ***p<0.001). ANOVA, analysis of variance; DMSO, dimethyl sulfoxide; hERV, human endogenous retrovirus; HLA, human leukocyte antigen; lncRNA, long non-coding RNA; MFI, mean fluorescence intensity; TCR, T-cell receptor. Among the eight antigens for which we had previously identified reactive TCR clonotypes, five (P10-erv1, P21-erv5, P11-lnc3, P16-lnc2, and P19-lnc2) could be effectively presented by the engineered K562 lines (online supplemental table 6). To determine whether the identified TCRs recognize hERV and lncRNA antigens, we performed an in vitro stimulation assay using T cells engineered to express these TCRs. We first cloned the TCR genes into a CD8-overexpressing Jurkat cell line equipped with an NFAT-driven tdTomato reporter, which allows quantitative measurement of TCR activation via fluorescence (figure 4L–M).56 Next, HLA-positive, TAP-deficient K562 cells were pulsed with the hERV and lncRNA antigens and co-cultured with the corresponding TCR-engineered Jurkat cells (online supplemental figure 8C). After 16 hours, we observed a significant increase in NFAT-tdTomato expression in four distinct TCR-engineered Jurkat cells co-cultured with their cognate peptide-pulsed targets, compared with negative controls (*p<0.05) (figure 4N–Q and online supplemental figure 8D,E). This result confirms that T cells expressing these TCRs specifically recognize HLA-presented hERV and lncRNA antigens and mount a functional response. The successful activation of TCR-engineered T cells by these peptides underscores their potential as robust targets for T-cell-based immunotherapies in RCC, such as TCR-T cell therapy. hERV and lncRNA antigens demonstrate potent antitumor activity in PTCs To evaluate the therapeutic potential of the identified antigens, we used a clinically relevant PTC model for RCC (figure 5A). This model recapitulates the morphological and genomic features of the original tumors, enabling a personalized assessment of antigen-specific responses.57 We first established eight RCC PTCs from samples expressing the relevant HLA alleles (online supplemental table 1). Since TILs within PTCs often exhibit an exhausted phenotype and are functionally impaired,58 59 we introduced autologous peripheral lymphocytes into the co-culture PTC system as a source of responsive T cells (see Methods).57 TSA peptide pool (TSAs) targeting lncRNA and hERV antigens mediate tumor cell killing in HLA-matched RCC patient-derived models in vitro. (A) Workflow for co-culture of RCC PTCs with autologous PBMCs following TSA peptide pool treatment. PTCs with diameters >40 µm were selected for evaluation. (B to F) Functional assessment of TSAs in RCC PTC model 1. (B) Tumor cell viability after 7-day treatment. (C) IFN-γ levels measured by ELISA. (D) Flow cytometry analysis of TIL activation. (E) Immunofluorescence staining for pancytokeratin (PanCK, tumor cells), CD3 (T cells), and granzyme B (T-cell activation) after 7-day treatment. Scale bar: 50 µm. (F) Quantification of T cell, activated T cell, and tumor cell numbers following pool treatment. (G) Representative immunofluorescence images of responder and non-responder PTCs after pool treatment. Scale bar: 50 µm. (H to M) Summary of pool effects in responder (n=4) and non-responder (n=4) groups: tumor viability (H and I), IFN-γ secretion (J and K), and TIL activation in PTCs in the response (L) and non-response groups (M). Data represent mean±SD from three independent experiments; significance assessed by one-way ANOVA (*p<0.05). ANOVA, analysis of variance; CTA, cancer-testis antigen; hERV, human endogenous retrovirus; HLA, human leukocyte antigen; IFN, interferon; lncRNA, long non-coding RNA; NC: Negative control; PBMC, peripheral blood mononuclear cell; PTC, patient-derived tumor-like cell cluster; RCC, renal cell carcinoma; TIL, tumor-infiltrating lymphocyte; TSA, tumor-specific antigen (lncRNA/hERV-derived). We selected 14 immunogenic hERV and lncRNA peptides with confirmed high binding affinity for HLA-A*02:01 (figure 4F), HLA-A*11:01 (figure 4G), or HLA-A*24:02 (figure 4H and online supplemental table 9) to formulate a combined TSA peptide pool (TSAs). For comparison, a control pool was prepared using 14 selected CTAs (online supplemental table 10). In the RCC1 PTC model, treatment with the TSAs resulted in significant tumor growth inhibition (figure 5B), accompanied by increased IFN-γ production (figure 5C) and expression of the T-cell activation marker 4–1BB (figure 5E), compared with CTA and DMSO control treatments. Immunofluorescence analysis further confirmed that TSA treatment induced robust T-cell clonal expansion, secretion of cytotoxic granzyme B, and subsequent elimination of tumor cells within the PTCs (figure 5E and F). We extended this evaluation to seven additional RCC PTC models (RCC 2–8) (online supplemental table 1). The TSA pool mediated significant tumor growth inhibition in four of the eight models (RCC 1–4), while the remaining models (RCC 5–8) were non-responsive (figure 5G–I), underscoring the patient-specific nature of these therapeutic responses. Importantly, in the responsive models (RCC 1–4), treatment with the TSA pool resulted in increased IFN-γ production (figure 5J–K) and expression of the T-cell activation marker 4–1BB (figure 5L–M), compared with non-responsive models. In addition, we also introduced pre-expanded autologous peripheral lymphocytes reactive to TSA peptides (TSA-T cells) into the co-cultured PTC system (online supplemental figure 9A). As controls, pre-expanded reactive T cells treated without any peptide stimulation (NC) and with a control CTA peptide pool (CTA-T cells) were used. In the RCC9 PTC model, immunofluorescence analysis further confirmed that TSA-T cells treatment also induced robust T-cell clonal expansion, secretion of cytotoxic granzyme B, and subsequent elimination of tumor cells within the PTCs (online supplemental figure 9A,B). compared with CTA-T cells and control treatments, TSA-T cells resulted in significant tumor growth inhibition, accompanied by increased IFN-γ production (online supplemental figure 9C,D). Extended analysis of PTC viability in additional patient-derived models showed TSA-T cell-mediated significant tumor growth inhibition in three of five models (RCC 9–11), while the remaining two models (RCC12–RCC13) were non-responsive (online supplemental figure 9E–H). These findings are consistent with earlier T-cell response data (figure 5). In summary, testing in HLA-matched RCC PTC models revealed an overall response rate of 50% for the hERV and lncRNA TSA pool. This efficacy surpassed that of the CTA pool and control treatments, demonstrating the potent and broadly applicable antitumor activity of non-canonical antigens derived from hERVs and lncRNAs. Non-canonical Antigen-Specific T cells exhibit potent antitumor effects To evaluate the in vivo antitumor efficacy of T cells targeting non-canonical antigens, we established a xenograft model using the 786-O RCC cell line (figure 6A). We first genetically engineered these cells by knocking out all endogenous classical HLA class I genes (HLA-A, HLA-B, and HLA-C) and then introducing expression of either HLA-A*02:01, HLA-A*11:01, or HLA-A*24:02 (figure 6B). MS analysis confirmed the high presentation of the hERV-derived peptide P10-erv1 (hERV3895: LYDTVTHTF) on the 786-O-A24 cell line, identifying it as a suitable target for in vivo assessment (figure 6C). We subcutaneously implanted the engineered 786-O-A24 cells (1×106 cells per mouse) into NSG mice (figure 6A). Once tumors reached approximately 50 mm³, mice were randomized into either treatment groups or control groups (n=6 per group). We generated hERV3895 TCR-T cells by transducing healthy donor PBMCs with a lentiviral vector encoding the cognate TCRαβ genes, achieving a transduction purity of 38.4% (figure 6D). Mock-transduced T cells (1×107 cells) served as a negative control. To explore potential synergy with ICB, we included groups receiving TCR-T cells either alone or in combination with an anti-PD-1 antibody (αPD-1). Antitumor efficacy of hERV3895-specific TCR-T cells in a xenograft mouse model. (A) Experimental timeline. NSG mice were subcutaneously inoculated with 786-O-A24 cells (1×10⁶) expressing the hERV3895 antigen on day 0, followed by intravenous injection of TCR-T cells (1×10⁷) on day 15. Tumors were harvested and weighed at endpoint. (B) Validation of HLA-I expression and genetic engineering strategy. Endogenous HLA class I genes in 786-O cells were knocked out via CRISPR/Cas9, followed by lentiviral transduction of HLA-A*24:02. (C) MS/MS fragmentation spectra comparing endogenous and synthetic P10-erv1 peptides, with retention times indicated. (D) Representative flow cytometry results of mTCRαβ+ T cells after transduction with lentivirus encoding dominant TCRαβ sequences of hERV3895 antigen. (E) Images of the explanted tumor sizes in the six groups of mice treated with PBS, α-PD-1, mock-T, mock-T+α-PD-1, TCR-T, or TCR-T+αPD-1. (F to H) Tumor volume (F), tumor weight (G), and body weight (H) measurements across treatment groups. αPD-1: anti-PD-1 antibody (pembrolizumab); mock-T: transduced irrelevant TCR-T cells. Data shown as mean±SD (n=6 per group); significance assessed by two-tailed Student’s t-test (*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). hERV, human endogenous retrovirus; HLA, human leukocyte antigen; i.p., intraperitoneal; i.v., intravenous; MS, mass spectrometry; PBS, phosphate-buffered saline; PD-1, programmed cell death protein 1; s.c., subcutaneous; TCR, T-cell receptor. The results demonstrated that administration of P10-erv1-specific TCR-T cells significantly inhibited tumor growth compared with both PBS and mock-T cell controls (figure 6E). The combination of TCR-T cells with αPD-1 led to even greater tumor suppression, with substantially lower tumor volumes than in the group receiving mock-T cells plus αPD-1 (figure 6F and online supplemental figure 10A–F). Mice were euthanized at day 28, and excised tumors from the TCR-T cell treatment groups were markedly lighter than those from control groups (figure 6G). In addition, all groups showed no significant effects on total body weight (figure 6H) or organ weights (heart, liver, spleen, lung, kidney; online supplemental figure 10G–K), indicating a favorable safety profile. To demonstrate the potential clinical relevance of the hERV TCRs, it is essential to assess their activity against endogenously processed and presented antigens on primary tumor cells. Given that the antitumor function of the hERV3895 antigen (P10-erv1) was previously confirmed in tumor cell lines both in vitro and in vivo, and because abundant tumor tissue was available from patient P10, we established a patient-derived renal cell line (RCL-10) from frozen tumor suspensions.53 RCL-10 expressed the HLA-A*24:02 and HLA-A*02:01 alleles (online supplemental figure 11A, online supplemental table 1). For comparison, we also isolated RCL-17, which did not express the HLA-A*24:02 allele (online supplemental table 1). We then tested the ability of hERV3895 TCR-T cells to recognize primary tumor cells. In cytotoxicity assays, hERV3895 TCR-T cells effectively recognized and killed HLA-matched RCL-10 cells, whereas mock TCR-T cells did not (online supplemental figure 11B,C). In addition, neither hERV3895 TCR-T cells nor mock TCR-T cells showed recognition or cytotoxicity against HLA-mismatched RCL-17 cells (online supplemental figure 11D,E). To further assess antigen specificity and measure TCR avidity, hERV3895 TCR-T cells were co-cultured with TAP1-deficient 786-O-A24 cells pulsed with titrated concentrations of either hERV3895 peptide or control peptide. Across the peptide concentration range of 0.1–10 µg/mL, the binding affinity of 786-O-A24 increased in a dose-dependent manner, reaching its maximum at 10 µg/mL (online supplemental figure 11F). hERV3895 TCR-T demonstrated specific reactivity to cells presenting the cognate hERV3895 epitope, as well as to TAP1-expressing 786-O-A24 cells, but showed no reactivity toward cells pulsed with the irrelevant peptide or toward wild-type 786-O cells (online supplemental figure 11G). Furthermore, using graded peptide concentrations, we observed a positive correlation between the level of antigen presentation on tumor cells and the killing efficiency of the TCR-T cells (online supplemental figure 11G). These findings demonstrate that TCR-T cells targeting a non-canonical hERV-derived antigen can mediate significant antitumor activity in vivo and in vitro. The enhanced efficacy observed with anti-PD-1 combination therapy suggests potential synergy between TCR-based therapies and ICB for solid tumor treatment. Discussion Our study addresses a critical gap in low-TMB RCC research by employing a proteogenomic strategy to profile the HLA-I immunopeptidome. This approach identified a broad repertoire of MAPs derived from hERVs and lncRNAs. Additionally, we identified 39 peptides derived from conventional CTAs, some of which (eg, ODF2, PBK, PRAME1, PMEL) have known immunogenic potential in other cancers,60,62 underscoring the robustness of our discovery platform. Critically, unlike patient-specific neoantigens, hERV and lncRNA-derived antigens can be shared across patients. Among the 73 identified MAPs, 9 were presented across multiple RCC samples. The peptides P11-lnc1 and P13-lnc2, identified here, were also previously reported in melanoma.17 This shared nature positions non-canonical TSAs as highly attractive targets for developing “off-the-shelf” immunotherapies, offering a viable and scalable alternative to personalized neoantigen vaccines for low-TMB tumors like RCC. While our proteogenomic pipeline primarily targeted hERVs and lncRNAs, other cryptic translational events—such as alternative reading frames (altORFs) within annotated coding genes or upstream open reading frames (uORFs) in 5’ UTRs—may also contribute to HLA-I presented peptides in RCC. Emerging evidence indicates that these non-canonical proteomic sources can produce immunogenic epitopes across various cancers.18 Comprehensive identification of such peptides generally relies on ribosome profiling (Ribo-seq) coupled with personalized immunopeptidomics. Future studies that integrate Ribo-seq with expanded search databases will be essential to fully map the non-canonical antigen repertoire in RCC and to evaluate the relative roles of hERVs, lncRNAs, altORFs, and uORFs in tumor immunogenicity. We further employed a single-cell approach to identify TCRαβ clonotypes specific for these antigens. Our analysis revealed that hERV-reactive and lncRNA-reactive CD8+ T cells in the RCC TME exhibit distinct transcriptional profiles compared with viral antigen-reactive bystander T cells. These tumor-reactive T cells were predominantly enriched in exhausted subsets across patients, characterized by low expression of the memory marker IL7R and elevated expression of exhaustion markers (eg, PD-1) alongside cytotoxic molecules such as granzyme A (GZMA). GZMA is a serine protease capable of inducing caspase-independent cell death, suggesting that these exhausted T cells retain cytotoxic potential.63 This functional capacity underscores their potential value for adoptive cell therapies. To functionally assess antigen immunogenicity, we established an in vitro co-culture system by incubating patient-derived PBMCs with PTCs, simulating T-cell recruitment within the TME. Using this model, we demonstrated that hERV and lncRNA antigens elicit more effective tumor cell killing compared with other treatments. We propose that because these non-canonical antigens are not part of the thymically presented self-proteome, the immune system lacks central tolerance against them. This permits the persistence of a high-affinity T-cell repertoire, enabling potent antitumor responses through robust immune recognition of these “foreign-like” antigens. However, only approximately half of the PTC models responded to the TSA pool. These results underscore the complex and patient-specific mechanisms of tumor immune evasion, pointing to several underlying biological barriers. First, effective T-cell mediated cytotoxicity depends on robust antigen presentation. The frequent downregulation or loss of HLA class I expression—a well-documented immune escape mechanism in RCC—may critically limit the presentation of pool-targeted hERV and lncRNA peptides. Second, interpatient tumor heterogeneity strongly influences target antigen abundance. The specific non-canonical transcripts targeted by the TSA pool may be epigenetically silenced or expressed below the level required to sufficiently activate T cells in non-responsive models. Finally, the immunosuppressive TME likely plays a decisive role. High baseline expression of inhibitory immune checkpoints such as programmed death-ligand 1, the presence of suppressive cytokines like Transforming Growth Factor-β (TGF-β), and the accumulation of regulatory immune cells can profoundly impair effector T-cell activation, further exacerbating the exhausted phenotype commonly observed in RCC tumor-infiltrating lymphocytes. Therefore, addressing these concurrent deficiencies in antigen presentation, target transcript availability, and TME-mediated immunosuppression will be essential to enhance the broad clinical efficacy of TSA-based immunotherapies. Although our in vivo models demonstrated favorable safety profiles, the translation of hERV-targeted and lncRNA-targeted therapies will require rigorous evaluation of potential off-tumor, on-target toxicities. While these non-canonical genomic elements are generally epigenetically silenced in healthy cells, low-level or cryptic expression may still occur in certain normal tissues. Consequently, unintended recognition by highly potent TCR-T cells could provoke autoimmune reactions or collateral tissue damage. To mitigate such risks from a translational standpoint, future clinical development should incorporate comprehensive transcriptomic and proteomic profiling of candidate antigens across a wide range of normal tissues to confirm strict tumor-specific expression. Our study has several limitations. First, our focus on hERV-derived and lncRNA-derived MAPs may have overlooked antigens from other cryptic genomic sources, such as unannotated open reading frames within UTRs or alternative reading frames.18 Comprehensive detection of such peptides would require specialized approaches like ribosome profiling combined with proteogenomic analysis,48 which were not employed here. Second, the number of mapped antigen-specific T-cell clones was limited (eg, 117 cells for hERVs, 211 for lncRNAs). Future studies with larger cohorts and expanded antigen-discovery methods will be valuable to address these limitations. Supplementary material Acknowledgements We thank our patients for their consent, which allowed us access to their tissues for use in this study. We thank the National Center for Protein Sciences at Peking University in Beijing, China, and the High-Performance Computing Platform at Peking University for confocal microscopy, flow cytometry, sequencing, and computing resources, and Dr Guilan Li for assistance with NGS experiments. We are grateful to the staff at the Department of Urology, Cancer Hospital, Chinese Academy of Medical Sciences, and Peking Union Medical College for their assistance with this project. Footnotes Funding: This work was supported in part by the National Natural Science Foundation of China 2024YFF0507401 (JJX) and 82372885 (BY), and the Basic Science Center Program of the National Natural Science Foundation of China No. T2288102 (JJX). We also acknowledge the funding support from “Laboratory for Synthetic Chemistry and Chemical Biology” under the Health@InnoHK Program launched by Innovation and Technology Commission, The Government of Hong Kong Special Administrative Region of the People's Republic of China. Provenance and peer review: Not commissioned; externally peer-reviewed. Patient consent for publication: Not applicable. Ethics approval: This study was approved by the Medical Ethics Committee of the Cancer Hospital, Chinese Academy of Medical Sciences (approval number: NCC020381). Participants gave informed consent to participate in the study before taking part. Data availability statement No data are available. References Associated Data Supplementary Materials Data Availability Statement No data are available.