The peripheral blood samples of patients diagnosed with GC admitted to Shanxi Cancer Hospital from 2020 to 2021 and healthy people were collected. Inclusion criteria include (1) patients who were diagnosed with GC for the first time on admission, and (2) patients without other tumors and without any treatments. The inclusion criteria of the control population include (1) the physical examination population of our hospital who were similar in age to GC patients, and (2) those who had no important potential diseases. As experiments involve human materials, all sample requests, collection, and processing are informed and agreed in writing by patients and in accordance with the ethical guidelines and norms of academic institutions. This study was approved by the Ethics Committee of Shanxi Province Cancer Hospital.

The peripheral blood of 68 cancer patients was collected for TCR beta-chain repertoire sequencing, and the peripheral blood from 49 healthy subjects was obtained as controls (Supplementary Table 1). The blood sample of each sample was divided into 2 ml of peripheral anticoagulant blood and 2 ml of non-anticoagulant blood. One milliliter of non-anticoagulant blood was used for DNA extraction, and the rest was used for cell surface molecular detection. Serum was separated by centrifugation from the anticoagulant blood and stored at −80°C for later use. All the above operations were completed within 24 h.

Genome DNA was extracted from 1 ml of non-anticoagulant peripheral blood using the DNeasy Blood & Tissue Kit (QIAGEN,166039685) according to the manufacturer’s protocol. The concentration and purity of extracted DNA were determined by Nanodrop 2000 and stored at −80°C for later use. For each sample, 200–500 ng of DNA was used for library preparation and sequencing. The first multiplex PCR goes 10 cycles. Multiplex PCR primers include 30 TRB-V primers and 13 TRB-J primers in QIAGEN Multiplex PCR Master Mix, target the conserved regions of TRBV and TRBJ genes, and were designed to specifically amplify the complementarity-determining regions—CDR3. The second simple PCR utilizes a universal primer and goes 25 cycles, which is attached with the 3’ adapter “AAGTCGGAGGCCAAGCGGTCTTAGGAAGACAA” and the 5’ adapter “AAGTCGGATCGTAGCCATGTCGTTCTGTGAGCCAAGGAGTTG” to generate sequencing libraries, and then cooled to 4°C. The TCR beta-chain sequencing libraries were constructed using the Phusion High-Fidelity PCR Master Mix with HF Buffer (NEB) according to the instructions, and then were sequenced on the BGISEQ-500 platform with 100-bp paired-end reads (18).

Exome library construction was performed with 400 ng of DNA of each sample; the library was enriched by array hybridization (MGIEasy Exome Capture V4 Probe Set) and then sequenced on the MGISEQ-2000 platform with 100-bp paired-end reads, according to the manufacturer’s standard procedure.

Flow cytometry was used to determine the percentage and absolute number of lymphocyte subtypes in whole blood samples by using a six-color fluorescent monoclonal antibody kit combined with an absolute count microsphere kit (QuantoBio, Beijing, China). Briefly, peripheral blood cells were stained with the six-color fluorescent monoclonal antibody including PerCP-Cy5.5-anti-CD45, FITC-anti-CD3, PC7-anti-CD4, APC-Cy7-anti-CD8, APC-anti-CD19, and PE-anti-CD16+56+ according to the manufacturer’s protocol. After that, the samples were treated with hemolysin and lysed erythrocytes. Before flow cytometry detection, absolute counting microspheres of the same volume as the test sample were added and thoroughly mixed. Then, the cells were analyzed by the BD LSRFortessa cell analyzer (Becton-Dickinson FACSCanto II), and the percentage and absolute number of lymphocyte subtypes were analyzed by CellQuest Pro software (Becton-Dickinson FACSCanto 3.1).

Commercial EBV EA-IgG ELISA kit (Autobio, Zhengzhou, China) was used to detect EBV EA-IgG in serum, strictly according to the instructions.

The raw sequencing data were filtered with SOAPnuke (19) (v1.5.3) to remove reads with adapter contamination, reads with more than 10% N bases, and low-quality reads containing more than 50% low-quality bases (base quality value < 15, Supplementary Table 2). After the data filtration, the clean sequencing data were aligned to the human TRBV, TRBD, and TRBJ reference database using MiXCR (20)(v3.0.13, Supplementary Table 3) and then mapped reads were assembled into clonotypes, which refer to the unique TRBV–TRBJ CDR3 nucleotide sequence. The MiXCR applied a heuristic multilayer clustering method to correct the PCR and sequencing error. The TCR clonotypes with clone count and fraction were exported for further analysis. The R package “immunarch” (21) (v0.6.6) was employed to perform repertoire analysis, such as the relative abundance of CDR3, TRBV/TRBJ gene usage, estimation of repertoire diversity, and antigen-specific annotation of CDR3. The statistical analysis was performed using R software (v3.6.0), and the Wilcoxon test was adopted to compare the difference between the two groups. The whole-exome sequencing (WES) data were filtered using SOAPnuke with default settings and then HLA typing was performed with OptiType (22) (v1.3.1). We used the Fisher test to evaluate the enrichment of each HLA allele in the tumor group and control group.

On average, 9.27 × 10⁶ TCR-β sequences were obtained from each individual (7.71 × 10⁶ in the healthy control group and 1.04 × 10⁷ in the tumor group). The mean number of unique CDR3 of the healthy and the tumor group is 11,655.4 and 12,875.8, respectively (Supplementary Figure 1C). Also, we found that there is no significant difference in the absolute number of CD3⁺ cells between the tumor and control group (Wilcoxon test, p = 0.629). The length of CDR3 in both healthy controls and cancer patients showed a normal distribution (Supplementary Figure 1A). We defined the CDR3 with a frequency of >1% to be hyper-expanded CDR3, the CDR3 with a frequency of >0.1% to be large CDR3, the CDR3 with a frequency of >0.01% to be medium CDR3, and the CDR3 with a frequency of ≤0.01% to be small CDR3 (23). The proportion of both large clones and highly expanded clones was comparable between the healthy controls and the cancer group (Supplementary Figure 1D).

We further divided TCR CDR3 into three groups (“T&C”, “T/C”, and “Individual”) based on the shared CDR3 between the healthy control group and the tumor group. The CDR3 in the “T&C” group is shared by both the healthy control group and the tumor group (shared by at least one control sample and at least one tumor sample), the “T/C” group involved the CDR3 shared by the tumor group or the control group (if the CDR is identified in at least two control samples, but does not exist in the tumor samples, or identified in at least two tumor samples, but does not exist in the control samples), and the “Individual” group contains the unique CDR3 of each individual; 13.6% of CDR3 are shared by the tumor group and the control group, 32.0% are shared by the control group, 49.2% are shared by the tumor group, and an average of 46.1% of CDR3 are unique for each individual (Figure 1A; Supplementary Table 4). Most hyper-expanded CDR3 are individual-specific in both the tumor group and the healthy control group. The control group has a significantly higher percentage of the small, medium, large, and hyper-expanded CDR3 shared by both the tumor group and the control group (p = 0.046, p = 0.023, p = 0.026, p = 0.015), and these CDR3 display different distributions via different stages (p = 0.014, Kruskal–Wallis test, Figure 1D). The large individual-specific CDR3 of the tumor group accounts for a significantly higher proportion than the healthy control group (p = 0.0007, Figure 1B). CDR3 specifically shared by the tumor group occupied a higher proportion than CDR3 shared by the control group, and shows different distribution via different stages (p = 5.8e-5, Kruskal–Wallis test, Figure 1C). The distribution of the expanded (medium, large, and hyper) individual-specific CDR3 has a significant difference (p = 0.02, Kruskal–Wallis test, Figure 1E).

The gene usage of TRBV and TRBJ was determined by the clonotypes assembled by MiXCR. TRBV20-1 (9.24%) and TRBV6-1 (8.42%) are the most frequently used TRBV genes, and TRBJ2-1 (19.2%) and TRBJ2-7 (13.9%) are the most frequently used TRBJ genes. The tumor group has a significantly lower gene usage of TRBV30, TRBV12-4, TRBV4-1, TRBV4-2, TRBV18, TRBJ1-1, TRBJ1-5, and TRBJ2-2. The TRBV3-1, TRBV6-4, TRBV12-3, TRBV7-2, and TRBJ2-5 are more frequently used in tumor groups (Figures 2A, B). Moreover, the tumor group has a higher TRBV3-1 gene usage frequency than the control group regardless of tumor stage (p = 0.0013, Kruskal–Wallis test, Figure 2C). As only A*02:01 is significantly enriched in the tumor group (Supplementary Table 5), we compared the gene usage of TRBV and TRBJ between samples with or without A*02:01 to identify the TRBV or TRBJ gene usage bias, only TRBJ2-3 has a significant relationship with A*02:01, but it has no difference between the tumor group and the control group (Supplementary Figure 2). We also compared the TRBV–TRBJ combination usage, and the tumor group and control group differ greatly. The highest V–J pairing usage in the tumor group is TRBV6-1–TRBJ1-6, while TRBV30–TRBJ2-2 is the most frequently used V–J pair in the control group (Supplementary Figure 3). The combined usage of TRBV12-3–TRBJ2-2, TRBV12-3–TRBJ2-5, TRBV4-1–TRBJ1-1, and TRBV15–TRBJ1-2 is significantly associated with tumor stage (Supplementary Figure 4, Kruskal–Wallis test, p = 2.8e-05, p = 0.00042, p = 4.8e-06, p = 0.00038).

A total of 1,175,884 unique CDR3 amino acids were identified from the 117 individuals. Among them, 461,554 CDR3 were found in healthy controls, 38,922 CDR3 were shared between healthy controls and cancer patients, 675,408 CDR3 were found in cancer patients, and 63,712 CDR3 were generated by at least two cancer patients; among these CDR3, we identified 374 hyper-expanded tumor-specific CDR3. After removing CDR3 that were recorded by the ImmuneCODE™ database (24), 292 CDR3 were retained to calculate the abundance of tumor-specific hyper-expanded CDR3, which is significantly higher in the elder group (Figure 3A), but comparable in different tumor stage (Figure 3B) and gender (Figure 3C). The tumor-specific CDR3 “CASSFRSGVYNEQFF” is hyper-expanded across Gr044, Gr053, Gr054, Gr063, and Gr066 (Figure 3D); the CDR3 cluster shared antigen specificity with the CDR3 “CASSFRSGVYNEQFF” is widespread among the tumor group (52/68), this CDR3 cluster contains 103 CDR3, the sequence logo of this cluster showed that positions 6–9 are diverse, and the first four amino acids and the last five amino acids are conserved (Supplementary Figure 5). “CASSQGGEVVEHEQFF” is enriched in Gr044, Gr054, and Gr063. The frequency of CDR3 “CASSLIEGHRYF” is high across Gr005, Gr011, Gr012, Gr013, Gr015, and Gr023. The CDR3 “CASSPSLAGGLCYNEQFF” was identified in 21 tumor patients, and hyper-expanded across Gr012, Gr030, Gr041, and Gr043. These hyper-expanded tumor-specific TCR CDR3 may be important for cancer immunotherapy.

The TCR CDR3 were annotated with antigen-specific immune receptor databases VDJdb (25), McPAS-TCR (26), and PIRD TBAdb (27), and 688 EBV (Epstein–Barr Virus)-specific TCR CDR3 were identified, including the hyper-expanded CDR3 “CSARVGVGNTIYF” in Gr061 (1.9%), the hyper-expanded “CASSPLAGVTDTQYF” in Gr069 (1.0%), and the large CDR3 “CASSLTSATGELFF” in Gr083 (0.7%) (Figure 4A). As the limitation of incomplete antigen-specific immune receptor databases, we clustered the CDR3 amino acid sequences with iSMART (28), which aggregated T-cell receptor CDR3 sequences into antigen-specific clusters based on the pairwise local CDR3 alignment scores. The CDR3 clusters containing EBV-specific CDR3 were identified as EBV-specific CDR3 clusters. We defined the EBV-specific CDR3 abundance as the frequency of all CDR3 in the EBV-specific CDR3 clusters. As Camago et al. have reported that male patients under the age of 60 have a higher incidence of EBV-positive GC (29), we compared the EBV-specific CDR3 abundance in age, gender, tumor stage, and other risk factors. The male group has a significantly higher EBV-specific CDR3 abundance than the female group (t-test, p = 0.031, Figure 4D), and the elders (age ≥; 60) tend to have lower EBV-specific CDR3 abundance (Figure 4C). The stage IV GC patients have a significantly higher EBV-specific CDR3 abundance than the healthy control group (t-test, p = 0.007, Figure 4B), which indicated that the high level EBV-specific CDR3 is a potential biomarker in cancer monitoring or diagnosis, although the EBV-positive GC was not significantly associated with tumor stage in the meta-analysis of 5,081 GC (30).