Two healthy volunteers provided informed consent. Peripheral blood mononuclear cells (PBMCs) were isolated from healthy volunteers by non-mobilized leukapheresis and Ficoll-Paque (GE Healthcare, Buckinghamshire, UK) gradient centrifugation and used to generate T cells, DCs and LCLs. To generate monocytes, non-adherent cells of PBMC were removed following a 2-hour incubation at 37°C, and adherent cells were scraped from culture dishes. Monocytes were cultured in 6-well plates at 3×10⁶ cells/well in X-VIVO 15 medium containing 1× GlutaMAX, 500 U/mL IL-4 and 1000 U/mL GM-CSF for 5 days to obtain immature DCs. All experiments used fresh immature DCs. LCLs were established by infecting PBMCs with B95-8 cell supernatant according to standard EBV transformation protocols.19 In brief, PBMCs were cultured with RPMI 1640 medium including 20% fetal bovine serum, streptomycin, penicillin and B95-8 cell supernatant. The cells were passaged every week until stable growth was exhibited. For LCL-based transfectants, EBV-LMP2A (from EBV B95-8 strain) and GFP cDNA were linked together using 2A sequence derived from foot-and-mouth disease virus and cloned into the pCDH-EF1-Luc2-T2A-TdTomato lentiviral vector. LCLs of donor 1 and donor 2 were transduced with this vector to generate LMP2A-overexpressing LCLs (LCL_LMP2A). HEK 293-FT (Life Technologies, USA), a packaging cell line used to produce lentivirus supernatants, were obtained from the American Type Culture Collection (Manassas, Virginia, USA).

DNA of donors’peripheral blood was extracted with DNeasy Blood & Tissue Kit (QIAGEN, Germany) according to the manufacturer’s protocol. Genotypes of HLA alleles were performed using high-resolution, high-throughput HLA genotyping with deep sequencing (BGI Diagnosis, Shenzhen, China). The HLA types of donors were outlined in online supplemental table S1.

Staining steps were performed at room temperature for 15 min with PBS washes between steps. Flow cytometry antibodies were used (all anti-human; clone IDs are given in parentheses): Fixable Viability Stain 780 (FVS780), CD3 (UCHT1), CD4 (RPAT4), CD8 (RPA-T8), CD45RA (HI100), CCR7 (150503), CD25 (M-A251), CD137 (4B4–1), CD107A (H4A3), CD107b (H4B3). All antibodies were from BD Biosciences, except for anti-mouse TCR-β constant region (clone H5-597, eBioscience, USA).

Fresh biopsy samples from Patient-Derived Xenograft (PDX) model were fixed in 10% formalin and embedded in paraffin tissue blocks. Then, tissues were sectioned at 5 μm and prepared according to standard methodology for routine histology. For immunohistochemistry, the investigated antibodies included mouse monoclonal anti-CD3 (clone LN10; ZSGB-BIO, Beijing, China) and mouse monoclonal anti-PD1 (clone UMAB199; ZSGB-BIO). Counterstaining was performed using H&E and bluing reagent from ZSGB-BIO; slides were mounted with coverslips and air-dried.

PBMCs was incubated for 7 days in T-cell media consisting of X-VIVO 15 (Lonza, USA), Glutamax (Life Technologies), IL-2 (50 U/mL, Perprotech, USA), IL-7 (10 ng/mL, Perprotech) and IL-15 (10 ng/mL, Perprotech). And then 5×10⁶ PBMCs (S0) were cocultured with 5×10⁵ autologous irradiated (3000 rad) DCs pulsed with commercially available LMP2A peptide mix (Miltenyi Biotec, Germany) at a concentration of 10 µg/mL in T-cell medium for 7 days, after which they were re-stimulated with autologous irradiated (3000 rad) DCs pulsed with commercially available LMP2A pepmixes one more time to generate LMP2A-reactive T cells (S2). Furthermore, CD3+ T cells of S2 were sorted into 96-well PCR plates by single-cell sorting using BD FACS Aria Il (Special Order System). And then, 96-well PCR plates were placed into liquid nitrogen and conserved at −80°C prior to performing single-cell TCR sequencing.

TCRs present in CD3+ T cells of S2 were identified using single-cell TCR sequencing as described in our previous study.20 Briefly, TCR sequences from the sorted single cells were obtained using a series of two nested PCR reactions. PCR products were purified and sequenced by Sanger sequencing method with Cα and Cβ region primers. The TCR sequences were analyzed using IMGT/V-Quest tool (http://www.imgt.org/).

S0 and S2 in each donor were harvested and used for high-throughput TCRβ sequencing. Detailed information of TCRβ sequencing has been described in our previous study.21 22 Briefly, the TCRβ was amplified and sequenced using Multiplex PCR and Illumina HiSeq2500 platform (MyGenostics, Beijing,China) from >500 ng of genomic DNA for each sample. The TCRβ information was identified based on the definition established by the International ImMunoGeneTics collaboration, and the V, D, J segments were discerned by a standard algorithm.23 Only productive reads that did not contain frameshifts or stop codons were used for statistical analysis. On average, 2,186,966 TCR templates were detected (range 1,678,257–2,695,675) from four samples of two donors, representing an average of 42,879 unique clonotypes (range 25,489–60,270) (online supplemental tables S2–S5). To quantify the diversity of TCR repertoire, we adopted clonality index that provides a diversity measurement as a function of distribution of TCR frequencies and is independent of sampling depth. The clonality is defined as 1—normalized Shannon’s entropy with values ranging from 0 (most diverse) to 1 (least diverse).24 25

Human IFN-γ ELISPOT Kit (Abcam, USA) was performed as the manufacturer’s protocols. Briefly, 2×10⁴ T cells, rested in cytokine-free media overnight, were incubated with 1×10⁴ PBS-washed autologous DCs pulsed with LMP2A peptide mix at a concentration of 10 µg/mL for approximately 20 hours in the absence of exogenous cytokines. The number of colored spots was calculated by ImmunoSpot plate reader and associated software (Cellular Technologies, USA).

For approximately 20 hours, 1×10⁶ T cells were incubated with 1×10⁵ autologous LCL_LMP2A or LMP2A peptide pulsed DCs in the absence of exogenous cytokines. IFN-γ level in coculture supernatant was measured using human IFN-γ ELISA kit (ExCell Bio, China) as the protocol’s procedure.

CFSE-based cytotoxicity assay was performed as shown in our previous study.20 LCL_LMP2A were labeled with 5 µM CFSE (BD Biosciences) for 15 min at 37°C and then cocultured with T cells at room temperature for 4 hours, at E:T ratio of 0.5:1, 2.5:1 and 10:1. After the coculture, 1 µg/mL propidium iodide (BD Biosciences) was added to evaluate the ratio of target cell death, and then the samples were analyzed by flow cytometry.

TCRα/β chains were synthesized (GenScript) and cloned into our lentivirus vector.20 TCRs were constructed in a β-α chain order and their constant regions were replaced by mouse counterparts modified with hydrophobic substitution and added disulfide bond as previously described, which not only was convenient for detection of TCR-T but also improved TCR pairing and TCR/CD3 stability.20 26 To obtain TCR-T, T cells were transduced with lentivirus as previously described.20 Briefly, PBMCs were stimulated in T-cell media supplemented with 50 ng/mL OKT3 and 1 µg/mL anti-CD28 for 2 days before transduction. TCR lentivirus were obtained by cotransfection of 293-FT cells with both lentivector and packaging plasmids using PEI MAX 40000 (Polysciences, USA).20 The lentiviral supernatants were harvested at 48 and 72 hours after transfection and concentrated using ultracentrifugation method with 20,000 g, 90 min at 4°C.20 And then activated T cells were transduced by concentrated lentivirus with 8 µg/mL polybrene (Sigma-Aldrich, USA). Two days later, the transduction efficiency was evaluated using mouse TCR-β chain constant region staining by flow cytometry.

Each NOD/SCID mouse subcutaneously received 2×10⁶ LCL_LMP2A tumor cells. Three days later, each mouse was intravenously infused twice with 3×10⁶ TCR-Ts at 2-day interval (on days 3 and 5). Intraperitoneal administration of IL-2 (1000 IU per mouse) was given on days 3, 5, 7, 9 and 11. Calipers were used to measure tumors, and the products of the perpendicular diameters were recorded.

Statistical analysis was performed using GraphPad Prism V.7.0 (GraphPad Software, California, USA) and Stata V.11.0 (Stata Corp). Statistical comparison was conducted with unpaired two-tailed Student’s t test and one-way analysis of variance with Bonferroni post-test. All tests were two-sided and p value <0.05 was considered statistically significant.

Raw sequencing data were submitted to the Sequence Read Archive (BioProject No. PRJNA642688).

To obtain LMP2A-specific T cells, we first enrolled two healthy donors and obtained their peripheral T cells and DCs, and then peripheral T cells of each donor were stimulated with autologous irradiated DCs pulsed with a pool of LMP2A peptides for twice (figure 1). Flow cytometry showed that phenotypic characteristics of prestimulated and poststimulated T cells (namely S0 and S2) hugely varied in each donor, and per cent of CD8+ T cells and effect memory T cells significantly increased (figure 2A,B). In addition, ELISA and ELISPOT assays demonstrated that poststimulated T cells could specifically identify LMP2A peptides (figure 2C,D).

Since frequencies of LMP2A-specific T cells should significantly increase after stimulation, high-throughput TCRβ sequencing of prestimulated and poststimulated T cells could identify significantly increasing TCRβ sequences, which could be β chains of LMP2A-specific TCRs, but corresponding TCRα sequences were unknown. Hence, poststimulated T cells were also sorted into 96-well plates and were amplified using single-cell PCR to obtain their TCRs, which could identify corresponding TCRα of significantly increasing TCRβ (figure 1). High-throughput TCRβ sequencing of prestimulated and poststimulated T cells demonstrated that more clonal V-J usage in S2 than S0 T cells (figure 3A and online supplemental figure S1A). In addition, we plotted the Lorenz curve for S0 and S2 to assess any potential skewing and diversity of the TCR repertoire composition, which demonstrated that diversity of TCR repertoires in S0 cells was higher than that in S2 cells (figure 3B and online supplemental figure S1B). Indeed, hyperexpanded T-cell clonotypes (HEC) of S0 cells occupied a median of 13.6% of the total TCR repertoire; reciprocally, the HEC of S2 cells contributed to a median of 30.5% of the total TCR repertoires (figure 3C). To systematically and quantitatively assess the repertoire clonality of S0 and S2 cells, we found that the clonality of S2 cells was higher than that of S0 cells (figure 3D). In general, these findings suggested that frequencies of LMP2A-reactive T cells could significantly increase after repeated stimulation with LMP2A peptide mix. Furthermore, to identify candidate LMP2A-reactive T cells, LMP2A-reactive TCRs could meet two criteria: frequencies of TCRs in poststimulated T cells were >0.1% and meanwhile frequencies of TCRs in poststimulated T cells was >10-fold higher than those in prestimulated T cells, based on high-throughput TCRβ sequencing results of prestimulated and poststimulated T cells (online supplemental tables 2–5). Based on these criteria, we found 36 and 57 eligible TCRβ sequences for donor 1 and donor 2, respectively (online supplemental tables 6 and 7). To identify corresponding TCRα sequences for these TCRβ sequences, we performed single-cell TCR sequencing of poststimulated T cells for donor 1 and donor 2, respectively (online supplemental table 8). Hence, we integrated high-throughput TCRβ sequencing and single-cell TCR sequencing results and found two and five candidate LMP2A-reactive TCRs for donor 1 and donor 2, respectively (online supplemental table 9). In final, top two candidate LMP2A-reactive TCRs were used for function validation for each donor (table 1). To optimize expression of introduced TCRs in T cells to obtain candidate LMP2A-reactive TCR-Ts, TCRs were constructed in a β/α chain order and constant regions were replaced by mouse counterparts modified with hydrophobic substitutions and interchain disulfide bonds previously described (figure 3E). These TCRs were lentivirally transduced into peripheral T cells to manufacture TCR-Ts with transduction efficiency of more than 50% (online supplemental figure S2).

To evaluate the ability of these TCR-Ts to specifically identify and mediate effector functions in response to LMP2A in vitro, ELISA and ELISPOT assays both showed high IFN-γ secretion in all four TCR-engineered T cells on coculture with LMP2A-pulsed DCs (figure 4A,B). Furthermore, gating on CD3+ T cells, CD25 and CD137 upregulation was also observed in all four TCR-engineered T cells cocultured with LMP2A-loaded DCs (figure 4C).

To verify that the LMP2A peptides are endogenously processed and presented on the cells with corresponding HLA molecules, we transfected LMP2A-encoding lentivector into LCLs to generate LMP2A-overexpressing LCLs (LCL_LMP2A) for each donor. As expected, IFN-γ production and CD137 upregulation were observed in all four TCR-engineered T cells cocultured with LCL_LMP2A (figure 5A). In addition, all four TCR-engineered T cells revealed cytotoxic activity against LCL_LMP2A (figure 5B). These findings indicated that the LMP2A epitopes were expectedly processed and presented on the cell surface and could be specifically recognized by LMP2A-reactive TCR-Ts.

Next, since all four TCR-engineered T cells of two donors demonstrated in vitro recognition and cytotoxicity against LCL_LMP2A, showing feasibility and effectiveness of screening individualized LMP2A-reactive TCR approach, to simplifiy in vivo function validation of LMP2A-reactive TCR-Ts, we only evaluated therapeutic benefit in vivo of TCR1 and TCR2-engineered T cells using a xenogeneic model subcutaneously engrafted with LCL_LMP2A. NOD/SCID mice that were subcutaneously engrafted with LCL_LMP2A underwent adoptive infusion with untransduced T cells, TCR1-Ts, TCR2-Ts on day 3 (figure 5C). Mice treated with TCR1-Ts or TCR2-Ts had a reduced tumor progression compared with those with untransduced T cells, drawing significantly lower tumor area by day 40 (figure 5D). In addition, we found that infused TCR-T cells were infiltrated into tumors but mock T cells did not, when mice were killed (online supplemental figure S3A). Although infused TCR-T cells were infiltrated into tumor, we did not find upregulation of exhaustion marker, for example, PD-1 (online supplemental figure S3B). However, this might be due to lower number of infiltrated TCR-T cells. Unexpectedly, TCR2-Ts demonstrated more stronger in vivo antitumor ability than TCR1-Ts, although TCR1-Ts and TCR2-Ts showed similar in vitro cytotoxicity against LCL_LMP2A (figure 5D).