Rapid Construction of Antitumor T-cell Receptor Vectors from Frozen Tumors for Engineered T-cell Therapy

Tumor antigen–specific T cells recognize cancer targets via heterodimeric T-cell receptors (TCR) that recognize tumor antigen–derived peptides loaded on MHC molecules on cancer cells. Diverse sequences in both TCR α and β chains, especially in their complement-determining region 3 (CDR3), determine MHC restriction and peptide specificity. Adoptive transfer of autologous tumor antigen–specific T cells into cancer patients is a promising therapeutic strategy for treatment of cancer patients (1–7). Because it is difficult to expand sufficient numbers of autologous tumor antigen–specific T cells from patients, methods have been developed to engineer peripheral bulk T cells to express tumor antigen–specific TCR genes (8–10). It has been widely demonstrated that TCR gene–engineered T cells have antitumor effects comparable with the parental T-cell clones against cancer targets. Clinical trials testing TCR gene–engineered T cells have demonstrated feasibility, safety, and therapeutic effects in multiple tumor types (11–14). However, only a limited number of therapeutic antitumor TCR genes have been developed, which limits the broad application of this therapeutic strategy to cancer patients (15, 16).

Traditionally, tumor antigen–specific TCR α and β chain genes are obtained from well-characterized tumor antigen–specific T-cell clones expanded in vitro. However, establishing tumor antigen–specific T-cell clones targeting a broad array of tumor antigens and MHC restriction elements is laborious and technically challenging in a high-throughput manner. Single-cell approaches such as single-cell PCR (17–21) and emulsion PCR (22, 23) can identify tumor antigen–specific TCR pairs. However, obtaining high-quality antitumor T cells from cancer specimens requires collection and processing of large amounts of freshly resected surgical specimens, which may not be feasible in all patients. Alternatively, next-generation sequencing (NGS) has been utilized to identify paired TCR α and β chain sequences from frozen tumor specimens (24–27). In this method, sets of TCR α and β sequences for tumor-infiltrating T cells are obtained, and pairing of TCR α and β chain genes is predicted based on matched frequencies in each specimen. Estimating absolute frequencies for TCR genes is still challenging with this approach because many T cells express two TCR α chain genes (28). In both single-cell– and NGS-based approaches, end-point results are often nucleotide sequences for many candidate TCR pairs. Laborious procedures such as synthesizing the TCR-expressing cassettes, cloning in expression vectors, and testing reactivity against target antigens are then required to identify candidate therapeutic TCR genes. More rapid identification of tumor-reactive TCR genes would facilitate personalized adoptive T-cell therapy.

To speed identification of tumor-reactive TCR genes, we developed a method for creating a library of randomly paired TCR genes in retroviral vectors. The TCR library was expressed in peripheral T cells for screening of antigen-specific TCR genes. We obtained TCR gene–engineered antitumor T cells within 2 weeks from frozen, stored specimens. We applied the method to identify tumor antigen–specific TCR α and β chain pairs from frozen tumor tissues and demonstrated the therapeutic potential of the library-derived TCR genes in a tumor xenograft model. We therefore demonstrate that antitumor TCR vectors can be obtained from small frozen tumor specimens without knowledge of the TCR sequences.

All procedures in this study were approved by the Roswell Park Cancer Institute (RPCI)'s Institutional Review Board and the Institutional Animal Care and Use Committee. Informed consent was obtained from all patients. Ovarian tumor specimens were obtained at the surgery at RPCI, frozen in liquid nitrogen, and stored at −80°C. Generation of NY-ESO-1–specific T cells was performed as described previously (29). Total RNA from NY-ESO-1–specific T cells was obtained using TRI Reagent followed by Phenol/Chloroform extraction or the Direct-zol RNA MiniPrep Kit (Zymo Research). Total RNA from tumor specimens (70–100 mg weight) was obtained by using tissue homogenizer in TRI Reagent followed by column purification with the Direct-zol RNA MiniPrep Kit. Yield of total RNA was between 137 and 268 μg. Reverse transcription was performed using the RevertAid First Strand cDNA Synthesis Kit using an oligo-dT primer (Thermo-Fisher) from 5 μg total RNA in a 20 μL reaction scale.

Primer sequences were listed in Supplementary Table S1. To enable unbiased PCR amplification of TCR-coding genes, typically, 2 μL cDNA from tumor tissues was mixed with multiplexed forward primers that are flanked by the common anchor sequences before the start codon (HTTCR#F for TCR α chain or HTTCR#C for TCR β chain) in separate tubes in 1 × Phusion polymerase reaction master mix (Thermo-Fisher). A single cycle of 98°C for 40 seconds, rapid cooling to 72°C then slow (–0.1°C/second) cooling to 66°C, 66°C for 30 seconds, and 72°C for 5 minutes was performed to synthesize the second-strand DNA. Unused multiplexed primers and all single-strand cDNA were destroyed by incubation with Exonuclease I at 37°C for 15 minutes followed by enzyme inactivation at 85°C for 15 minutes. To each reaction, the forward primer which binds to the common anchor sequences in HTTCR#F and #C (HTTCR#D for TCR α chain or HTTCR#A for TCR β chain) and the reverse primer for the TCR constant region (HTTCR#E for TCR α chain or HTTCR#B for TCR β chain) were added in 1 × Phusion polymerase reaction master mix. PCR was performed by 1 cycle of 98°C for 30 seconds; 2 cycles of 98°C for 10 seconds, 62°C for 30 seconds, and 72°C for 30 seconds; 30 cycles of 98°C for 10 seconds and 72°C for 60 seconds; and 1 cycle of 72°C for 2 minutes. The reaction was loaded on 1% agarose gel containing SYBR Safe DNA Gel Stain (Thermo Fisher Scientific) and electrophoresed at 90 V for 30 minutes. The main band of TCR variable fragments at around 450 bp was excised under the transilluminator (Invitrogen), and DNA fragments were extracted using the Zymoclean Gel DNA Recovery Kit (Zymo Research). DNA concentration was measured by absorbance at 260 nm. To amplify integrated TCR transgenes from T cells transduced with TCR genes, genomic DNA from TCR-transduced T cells was mixed with vector-specific primer pairs amplifying the entire TCR-expressing cassette (Forward: CGAATTCCCAAACTTAAGCTTGGTACCG and Reverse: GCAGCGTATCCACATAGCGTAAAAGG) in 1 × Phusion polymerase reaction mix. The PCR was performed by 1 cycle of 98°C for 30 seconds; 35 cycles of 98°C for 10 seconds, 71°C for 30 seconds, and 72°C for 40 seconds; and 1 cycle of 72°C for 2 minutes. Then, 1 μL of the reaction was mixed with Vβ-anchor–specific forward primer (HTTCR#A) and Cα-specific reverse primer (HTTCR#E) in 1 × Phusion polymerase reaction mix and cycled for 1 cycle of 98°C for 30 seconds; 35 cycles of 98°C for 10 seconds and 72°C for 70 seconds; and 1 cycle of 72°C for 2 minutes. Amplified DNA fragments were isolated and quantified as described above.

The DNA fragment encoding cysteine-modified TCR Cβ-P2A fusion protein was amplified by PCR from a plasmid containing this fragment and purified by gel electrophoresis. Description of the destination plasmid vector is provided in Supplementary Fig. S1. Essentially, the destination plasmid was based on the MSCV retroviral vector. The splice acceptor site from the human elongation factor 1α promoter was introduced before the TCR cloning site. A residue in the TCR α chain constant region was mutated to a cysteine to enhance pairing with the cysteine-modified TCR β chain. The linearized destination plasmid (50 ng), which was treated with NotI and PspOMI and gel purified, was mixed with equimolar amounts of Vβ, Cβ-P2A (obtained by PCR using HTTCR#F and HTTCR#G primers from the destination plasmid), and Vα fragments in 10 μL 1 × NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs) and incubated at 50°C for 60 minutes. To clone the TCR-expressing cassette that was amplified from genomic DNA of transduced T cells, linearized plasmid and TCR-expressing cassette insert were mixed at 1:2 molar ratio in 10 μL 1 × NEBuilder HiFi DNA Assembly Master Mix and incubated at 50°C for 60 minutes. The assembled product was used to transform chemically competent Escherichia coli, NEBStable, after purification by a DNA Clean & Concentrator kit (Zymo Research). Transformed E. coli were spread over three 10 cm agar plates and incubated for 14 to 16 hours at 37°C. Confluent E. coli colonies in all three plates were pooled, and plasmids were purified by the ZymoPURE Plasmid Midiprep Kit (Zymo Research). Quality of this bulk plasmid preparation was examined by restriction enzyme treatment with NotI and PacI, which excise the TCR-expressing cassette from the plasmid backbone, followed by electrophoresis in an agarose gel. In some experiments, plasmids obtained from pooled E. coli colonies were used to retransform competent E. coli to obtain single colonies. Some colonies were tested by DNA fingerprinting for TCR transgene by direct colony PCR using OneTaq (New England Biolabs) using a primer pair HTTCR#A and HTTCR#E; the reaction was then treated with AluI or MspI restriction enzyme (Thermo Scientific).

Retroviral particles were produced by cotransfection of TCR-encoding transfer plasmids and pVSV-G envelope plasmids into the GP2-293 packaging cell line (Clontech) by Lipofectamine 2000 (Invitrogen-Thermo Scientific). Packaging cells were coincubated with plasmids for 7 hours, and culture medium was replaced. After 36 hours, supernatant was harvested, centrifuged for 5 minutes at 400 × g, and immediately used for transduction of T cells. Peripheral blood mononuclear cells (PBMC) were obtained from healthy donors' buffy coat using the density gradient method using lymphocyte separation medium and stored in a liquid nitrogen tank in 90% FBS plus 10% DMSO. PBMCs were preactivated by 10 μg/mL phytohemagglutinin (Remel) for 40 hours in RPMI1640 medium supplemented with 10% FBS, penicillin, streptomycin, and l-glutamine in the presence of rhIL2 (10 U/mL; Sigma), rhIL7 (10 ng/mL; R&D Systems), and rhIL12p70 (20 ng/mL; Peprotech). Typically, preactivated PBMCs (1 × 10⁵) were harvested, counted, and plated on 96-well flat-bottom plate precoated overnight with Retronectin (10 μg/mL) and mAbs to human CD3 (5 μg/mL, OKT3; eBioscience) in the presence of rhIL2, rhIL7, and rhIL12. Typically, 125 μL retroviral supernatant was added to transduce T cells, which were then cultured for 24 hours. Cells were expanded in the presence of rhIL2 and rhIL7 without rhIL12 and used for evaluation within 7 days after transduction. Transduction of Jurkat (E6-1; ATCC) or J.RT3-T3.5 (ATCC) was performed similarly but without activating reagents and cytokines and using 6 to 12 μL retroviral supernatant.

NY-ESO-1–specific T cells were detected by specific MHC/peptide tetramer reagent (Ludwig Center for Cancer Research, University of Lausanne). TCR gene–transduced T cells were washed in PBS containing 1% FBS and incubated at 37°C for 15 minutes in the presence of phycoerythrin (PE)-conjugated tetramer (6 μg/mL) in 1% FBS-PBS. Cells were then stained by allophycocyanin (APC)-conjugated anti-CD4 and PerCP/Cy5.5-conjugated anti-CD8 (Biolegend) at 4°C for 15 minutes. Fluorescent signals were acquired by FACSCalibur instrument and analyzed by FlowJo software. In some experiments, tetramer⁺ T cells were sorted using FACSAria instrument. Genomic DNA of sorted cells was obtained using a Quick-gDNA MicroPrep kit (Zymo Research). Cytokine production from TCR gene–transduced T cells was tested by intracellular cytokine staining. Target cells were NY-ESO-1–expressing melanoma cell lines (HLA-A*02⁺DRB1*01⁺DPB1*04⁺ SK-MEL-37; HLA-B*35⁺ SK-MEL-52; HLA-DRB1*04⁺DP*04⁺ COLO 316 retrovirally transduced with NY-ESO-1 and CIITA genes) or NY-ESO-1–negative HLA-Cw*03⁺ MZ-MEL-12. To test Cw*03-restricted NY-ESO-1–specific reactivity, a Cw*03-negative and NY-ESO-1–negative A2780 ovarian cancer cell line was engineered by the Sleeping Beauty transposon system. Cw*03 and NY-ESO-1–coexpressing transposon plasmid was constructed by inserting human elongation factor 1α promoter followed by Cw*03-P2A-EGFP(A206K)-T2A-NY-ESO-1–expressing cassette and the rabbit globin polyadenylation signal into pT2/BH (a gift from Perry Hackett: Addgene plasmid # 26556). A2780 was electroporated using the 4D-Nucleofector system (Lonza) with pCMV(CAT)T7-SB100 (a gift from Zsuzsanna Izsvak: Addgene plasmid # 34879) and pT2-EF-Cw3-GFP-ESO. GFP⁺ clones were obtained by limiting dilution. To induce immunoproteasome expression, Cw*03⁺NY-ESO-1⁺ A2780 cells were pretreated with 1,000 U/mL rhIFNγ (Peprotech) for 2 days. Before coculture with T cells, some aliquots of target cells were pulsed overnight with 10 μg/mL synthetic NY-ESO-1 peptide (Genscript) or 10 μg/mL recombinant NY-ESO-1 protein (Ludwig Institute for Cancer Research) and washed in RPMI1640 medium. T cells were cocultured with target cells for 6 hours in the presence of GolgiStop (BD Biosciences). Cells were stained by fluorescein isothiocyanate–conjugated anti-CD4 and PerCP/Cy5.5-conjugated anti-CD8, permeabilized using the BD Cytofix/Cytoperm Plus Fixation/Permeabilization Kit (BD Biosciences), and stained with PE-conjugated anti-TNFα and APC-conjugated anti-IFNγ (BioLegend). Cells were analyzed by FACSCalibur instrument and FlowJo software. SK-MEL-37 and SK-MEL-52 were obtained from the Ludwig Institute for Cancer Research. MZ-MEL-12 was obtained from Dr. Jäger (Krankenhaus Nordwest, Frankfurt Germany). Ovarian cancer cell lines (Colo 316 and A2780) were from our cell bank. Cancer cell lines were tested for mycoplasma contamination using the Mycoalert Kit (Lonza Walkersville). A2780 was certified to have no mycoplasma contamination by IDEXX BioResearch.

NOD/SCID/common γ chain–deficient (NSG) mice (The Jackson Laboratory) were bred at the Laboratory Animal Resource at RPCI. Mice were inoculated with 1 × 10⁶ IFNγ-treated Cw3⁺NY-ESO-1⁺ A2780. Therapeutic T cells were generated by retroviral transduction of the Cw*03-restricted NY-ESO-1(92-100)–specific secondary TCR library (tumor #3). On day 3, mice received 4 × 10⁶ TCR-transduced or -untransduced T cells, or untreated. On days 3 to 5, all T-cell–infused animals were i.p. injected with 5 × 10⁴ units IL2 (Peprotech). Tumor growth was measured every other day. Tumor volume was calculated by a formula 0.5 × (longer diameter) × (shorter diameter)². Animals were sacrificed when tumor volume reached 2,000 mm³.

The two-tailed paired or unpaired t test was used to evaluate statistically significant differences between the values in two groups. P values are annotated as follows: *, P < 0.05 and **, P < 0.01.

A schematic TCR-expressing cassette designed for construction of retroviral vectors is shown in Fig. 1A. To enable stoichiometric expression of TCR α and β chains from a single transcript, TCR α and β chains were genetically connected via the P2A translational-skipping site (30, 31). The start codon for the TCR β chain was preceded by the Kozak consensus sequence (GCCACC) for efficient translation (32). The constant regions of both chains were modified by a cysteine residue to create an artificial disulfide bond that enhances paring of transgenic TCR α and β chains and inhibits pairing with endogenous TCRs (33). In the TCR-expressing cassette, the TCR α and β chain variable (Vα/Vβ), joining (Jα/Jβ) and the TCR β chain diverse (Dβ) regions that are critical for antigen recognition, are required to be obtained from antigen-specific T cells. In contrast, the constant regions can be prepared as stocked fragments that contain artificial modifications such as cysteine modification and fusion to the P2A site. In addition, the TCR α chain constant region can be included in the destination plasmid vector to reduce the number of fragments to be assembled. A destination retroviral expression plasmid was constructed as shown in Supplementary Fig. S1A. The destination plasmid contains a TCR-expressing cassette in which the TCR Cα region was modified to contain a PspOMI-recognizing site (GGGCCC). The TCR-expressing cassette in the destination plasmid was excised by treatment with NotI and PspOMI restriction enzymes. The artificial PspOMI-recognizing site was corrected to the natural Cα sequence during the assembling reaction (Supplementary Fig. S1B).

The 5′ parts of TCR α and β chains, Vα and Vβ, are highly variable. Therefore, without knowledge of TCR sequences, a large number of multiplexed forward primers are required for PCR amplification. Two sets of multiplexed forward primers were designed for all known Vα and Vβ reported in the IMGT database (34). Sequences of primers used in this study are listed in Supplementary Table S1. The TCR Vβ–specific primer set consisted of 45 primers that are flanked with the common 5′ sequence [the 3′ region of the destination plasmid vector from the restriction enzyme (NotI) site; designated as Vec' in Fig. 1B] before the start codon. Similarly, 49 Vα-specific primers had another common sequence (the 3′ region of the P2A site). These common flanked sequences serve as anchors for PCR amplification by the common primer sets. The procedures for amplification and cloning of TCR genes are depicted in Fig. 1B. In all experiments for this study, we used cDNA prepared from total RNA using oligo dT primers and a reverse transcriptase (Step 1). Because PCR amplification by multiplexed primers can cause an amplification bias due to different efficiencies for each primer, amplification of Vα and Vβ regions was performed using the anchor-specific forward primers which extend the 5′ anchor by flanked sequence (Vec'' in Fig. 1B) and reverse primers specific for the TCR constant region (Cα and Cβ). To end this, the second-strand TCR cDNA was synthesized by a single-cycle polymerase reaction primed by multiplexed primers, thereby adding an anchor sequence to the 5′ end of the second-strand cDNA (Step 2), followed by elimination of excess primers by Exonuclease I treatment (Step 3). Then, Vα and Vβ fragments were amplified by PCR using an anchor-specific forward primer and a reverse primer specific for the TCR common region (Cβ or Cα; Step 4). The 5′ region of the resulting Vβ fragment has a 30-nucleotide overlap with the 3′ region from a NotI cloning site in the destination plasmid vector, whereas the Vα fragment has an overlap with the 3′ region of P2A sequence. The cysteine-modified constant region for TCR β chain linked to the P2A sequence was PCR-amplified from the destination plasmid containing this fragment. Three fragments, VDJβ, Cβ-P2A, and VJα, were assembled with a linearized destination retroviral plasmid vector containing the cysteine-modified Cα fragment by a modified Gibson assembly using NEBuilder HiFi DNA Assembly Master Mix (Step 5). Using frozen stocked Cβ-P2A and linearized destination vector fragments, we prepared assembled vectors for transformation of competent E. coli cells within 4 hours.

To validate the rapid TCR-cloning method, we constructed TCR-expressing retroviral vectors from established T-cell clones that have unique combinations of TCR α and β chain genes. Four CD8⁺ T-cell clones (“AL”: HLA-A*02–restricted NY-ESO-1(157-165)–specific; “JD”: HLA-A*02–restricted NY-ESO-1(157-165)–specific; “KQ”: HLA-B*35–restricted NY-ESO-1(94-102)–specific; and “PP”: HLA-B*35–restricted NY-ESO-1(94-104)–specific) and five CD4⁺ T-cell clones (“SB”: HLA-DR*01–restricted NY-ESO-1(95-106)–specific; “JM”: HLA-DP*04–restricted NY-ESO-1(157-170)–specific; “3B5”: HLA-DP*04–restricted NY-ESO-1(157-170)–specific; “5B8”: HLA-DP*04–restricted NY-ESO-1(157-170)–specific; and “PB-T”: HLA-DR*04–restricted NY-ESO-1(111-143)–specific) were selected for experiments. All T-cell clones were established from peripheral blood or tumor specimens obtained from ovarian cancer patients at RPCI as described (30, 35, 36). As a control T-cell clone, the Jurkat T-lymphoma cell line (ATCC; TIB-152) was included. Vβ and Vα fragments were amplified as a single band for all T-cell clones tested (data not shown). The assembled TCR-expressing vectors were used to transform chemically competent E. coli. Transformed cells were spread and incubated overnight on agar plates to produce confluent colonies. To confirm that the TCR-expressing cassette was correctly assembled, bulk plasmids that were obtained from pooled colonies were digested by NotI and PacI restriction enzymes, which excise full-length TCR-expressing cassettes (Fig. 1B). A single band with the expected size for the TCR-expressing cassette was excised from the plasmid at around 1.8 kb (data not shown), indicating that our cloning procedures correctly assembled fragments as the expressing cassette.

To test the functionality of the cloned TCR, healthy donor T cells were polyclonally activated and infected with retroviruses generated from bulk plasmids containing T-cell clone–derived TCR-expressing cassettes. After a single infection, 25% to 35% of T cells expressed transduced TCR, as determined by the increase in TCR Vβ subtype expression where appropriate antibodies were available (Vβ16 for JD-TCR and Vβ5.3 for KQ-TCR; Fig. 2A). Functional expression of antigen-specific TCR α and β chain pairs was determined by MHC/peptide tetramer staining. Staining of untransduced and irrelevant TCR gene–transduced T cells was negligible. All four HLA class I–restricted TCR expressed on CD8⁺ T cells were stained by corresponding tetramers (HLA-A*02/NY-ESO-1(157-165) for AL and JD, HLA-B*35/NY-ESO-1(94-102) for KQ, or HLA-B*35/NY-ESO-1 (94-104) for PP; Fig. 2B; Supplementary Fig. S2). TCR-transduced T cells produced both IFNγ and TNFα upon coculturing with the antigen and HLA-expressing cancer cell lines (Fig. 2C). To assess functional expression of HLA class II–restricted NY-ESO-1–specific TCRs for which MHC/peptide-tetramer reagents were not available, TCR gene–transduced T cells were cocultured with antigen-pulsed target cells or MHC class II and NY-ESO-1–coexpressing cancer cells followed by intracellular cytokine staining. As shown in Supplementary Fig. S3, HLA class II–restricted TCR gene–transduced T cells produced IFNγ and TNFα upon antigen stimulation. These results demonstrate that our cloning protocol constructs functional TCR-expressing vectors for all T-cell clones tested without the need of TCR sequence information.

Next, we applied our TCR-cloning protocol to construct randomly paired TCR-expression libraries from polyclonal T-cell populations. We first tested the feasibility of constructing TCR-expression libraries from polyclonal T cells from PBMC of three healthy donors. Both TCR Vα and Vβ were amplified from PBMC cDNA (data not shown). Assembled plasmids were amplified in competent cells and purified from pooled colonies. The size of the library from one 50 μL vial of the competent cells was determined to be 2.8 ± 0.5 × 10⁵ (mean ± SD for 3 donors) by serial dilution of the transformed cells.

To investigate whether TCR gene amplification using the common primer sets against anchor-specific forward and common TCR constant region-specific reverse primers enables unbiased amplification of different TCR species, we compared usage of TCR Vβ subtypes in CD3⁺ T cells in PBMC and plasmid library by flow cytometry using Vβ subtype–specific antibodies. TCR Vβ usage in the TCR-expressing retroviral library was investigated by infecting retroviral particles into a Jurkat T-lymphoma subline with a mutated TCR β chain [J.RT3-T3.5 (J.RT3); ATCC: TIB-153]. J.RT3 was transduced with retroviral library at suboptimal viral titers that transduce less than 30% of cells to minimize multicopy transduction. Expression of cell-surface TCR Vβ was tested by flow cytometry using 24 different antibodies against Vβ subtypes. Figure 3A shows the relationship between the frequency of each Vβ subtype in CD3⁺ T cells in PBMC and library-transduced J.RT3 for 3 independent libraries from different donors. Overall, Vβ usage in CD3⁺ T cells was retained in the library. Figure 3B compares mean percentages of Vβ usage with results of statistical analyses by paired t tests. Although there were a few significant differences in the Vβ usage between PBMC and TCR library, such as overrepresentation of Vβ5.1 in the library, and several minor differences, the Vβ usage in peripheral T cells was well reproduced in the library. These results suggest that the majority of TCR gene species in the polyclonal T-cell population were PCR-amplified and assembled without bias.

Next, we tested whether our protocol could identify a tumor antigen–specific TCR gene from a polyclonal T-cell population without isolating tumor antigen–specific T cells. Immunogenic tumors are enriched in tumor antigen–specific T cells (29). Therefore, we hypothesized that randomly assembled TCR libraries from tumor specimens could be used to identify tumor antigen–specific TCR genes (Fig. 4A).

Even though tumors are enriched with tumor antigen–specific T cells, the probability that correct tumor antigen–specific TCR pairs would form through random pairing is low. Therefore, we tested the feasibility of recovering correctly paired tumor antigen–specific TCR from randomly paired TCR library in two ways. First, we mixed the HLA-A*02–restricted NY-ESO-1(157-165)–specific TCR-expressing plasmid (AL-TCR) with an irrelevant TCR-expressing plasmid at varying ratios (10%, 1%, 0.1%, and 0.01% for AL-TCR–expressing plasmid). Retroviral vectors were produced using mixed plasmids and were used to infect activated T cells from a healthy donor. As expected, specific TCR-expressing T cells as detected by a HLA-A*02/NY-ESO-1(157-165) tetramer decreased as the ratio of the AL-TCR–expressing plasmid decreased and became nearly undetectable when that ratio was at and below 0.1% (Supplementary Fig. S4). To isolate the NY-ESO-1–specific TCR-expressing transgene, the tetramer-stained T cells were sorted by flow cytometry, and genomic DNA with the integrated retroviral TCR transgene was extracted. The TCR transgene was amplified by nested PCR of genomic DNA and was reassembled into a TCR-expressing plasmid. As expected, we obtained sufficient number (358 cells) of tetramer-stained cells from 1% AL-TCR mixture from 6 × 10⁵ transduced T cells, whereas only 27 and 4 cells were obtained from 0.1% and 0.01% mixtures, respectively. Furthermore, although nested PCR yielded a product of the expected size from 27 tetramer-sorted cells from the 0.1% mixture, it failed to recover the transgene from 4 cells from the 0.01% mixture. As shown in Supplementary Fig. S4, all vectors generated by reassembling of nested PCR products in the TCR-expressing plasmids similarly transduced A*02-restricted NY-ESO-1(157-165)–specific TCR. These results demonstrated that even though frequencies of correctly paired TCRs in the primary library are as low as 0.1%, tumor antigen–specific TCR genes can be identified by isolation of TCR-transduced cells with desired functions, such as tetramer binding, and PCR amplification of integrated TCR transgenes by nested PCR. Assuming unbiased PCR amplification and assembly of TCR α and β chain fragments, 0.1% of a correctly paired TCR-expressing plasmid should be present in the library when a specific T-cell clone is present at 3.3% among total TCRαβ-expressing T cells.

Next, we tested whether we could recover correctly paired TCR from cDNA from the same AL clone and an irrelevant T-cell clone mixed at varying ratios (30%, 9%, 2.7%, and 0.8% for AL cDNA). Randomly paired TCR-expressing libraries were prepared from the cDNA mixtures, and T cells were transduced by retroviral libraries. As described for plasmid mixtures, we stained transduced T cells with a specific tetramer (Fig. 4B), sorted tetramer-stained T cells, and amplified TCR-expressing transgenes to clone into the retroviral plasmid. As shown in Fig. 4C, correctly paired antigen-specific TCR was recovered from the 2.7% AL cDNA mixture, whereas nested PCR of tetramer-sorted cells (8 cells from 0.6 × 10⁶ transduced T cells) from 0.8% AL cDNA mixture failed to amplify the TCR transgene. Therefore, in accordance with the experiments using plasmid mixtures, we expected that correctly paired antigen-specific TCR could be recovered when antigen-specific T-cell clones are present at or above 3% among total TCRαβ-expressing T cells in the specimen.

Using our TCR amplification and assembly protocol described above, we constructed a TCR gene library from 3 frozen tumor specimens that were known to be infiltrated by high-frequency HLA-Cw*03–restricted NY-ESO-1–specific CD8⁺ T cells (29). Percentages of Cw*03/NY-ESO-1(92-100) tetramer–stained cells among CD8⁺ T cells were 20.9%, 26.2%, and 16.4% for tumors #1, #2, and #3, respectively. However, percentages of tetramer-stained cells among total CD3⁺ T cells were not available. TCR variable fragments were amplified from cDNA of frozen ovarian tumor specimens and assembled as retroviral plasmid vectors. Then, retroviral particles were produced using the bulk plasmid library.

Polyclonally activated T cells were infected once with the tumor-derived TCR-expression library at suboptimal viral titers to minimize multicopy transduction. As shown in Fig. 4D, transduction with libraries derived from tumors #2 and #3 only slightly increased tetramer-stained cells over background staining, and transduction with the library derived from tumor #1 induced no detectable tetramer-stained cells. The secondary TCR library was prepared using nested PCR products from tetramer-sorted T cells and was used to prepare retroviral vectors to transduce primary T cells. A large fraction of the tumor #2 and #3–derived secondary TCR library–transduced T cells were reactive to the cognate antigen, as demonstrated by specific tetramer staining (Fig. 5A) and cytokine release against peptide-pulsed target cells (Fig. 5B). In contrast, the tumor #1–derived library did not contain tetramer-reactive TCRs (data not shown). TCR clones in the secondary library were characterized by TCR transgene DNA fingerprinting by digestion with restriction enzymes. As shown in Supplementary Fig. S5, the secondary TCR library for tumor #2 was significantly enriched by a single clone (7/14 clones), whereas tumor #3 was enriched with 2 clones (clonotype 3A: 7/14 and clonotype 3B: 4/14). Sequences for selected clones for each clonotype were shown in Supplementary Table S2. Clonotypes 3A and 3B share the same TCR β chain but have unrelated α chains. Transduction with the enriched TCR clone from tumor #2 and the clonotype 3A but not 3B from tumor #3 induced tetramer⁺ T cells (Supplementary Fig. S6). It is unclear why clonotype 3B was enriched in the secondary library but did not bind tetramer.

In contrast to TCRs obtained from established T-cell clones or single-cell–based approaches, our strategy based on random pairing could generate artificial TCR pairs that could cross-react to self-antigens and cause toxicity in patients. In order to determine specificity of library-derived TCRs, we tested recognition of glycine-substituted peptides for the Cw*03-binding NY-ESO-1(92-100: LAMPFATPM) epitope. As shown in Supplementary Fig. S7, substitution of any amino acid residue in the central part (MPFATPM) of the epitope almost completely abrogated recognition by both tumor #2– and #3–derived TCRs, indicating NY-ESO-1-specificity. In silico search for the TCR-recognizing motif (MPFATPM) through the ScanProsite tool (http://prosite.expasy.org/scanprosite/) found that only NY-ESO-1 has this motif among mammalian proteins.

We next investigated if a newly identified TCR could be functional and capable of rejecting tumors. We utilized an in vivo tumor xenograft model to study the therapeutic effect of the tumor #3–derived secondary TCR library–transduced T cells. Consistent with a report that the NY-ESO-1(92-100) epitope is generated by immunoproteasome (37), in vitro recognition of the A2780 ovarian cancer cell line (which was engineered to express Cw*03 and NY-ESO-1) by engineered T cells was significantly enhanced by IFNγ treatment of cancer cells (Fig. 5C). NSG mice were inoculated subcutaneously with HLA-Cw*03⁺NY-ESO-1⁺ A2780. On day 3, when palpable tumor was established, mice were intravenously infused with 4 × 10⁶ T cells either untransduced or transduced with the tumor #3 TCR library. Whereas untransduced T cells showed no antitumor effects, TCR gene–transduced T-cell products eliminated established tumors in all animals (Fig. 5D).

Adoptive transfer of autologous tumor antigen–specific T cells is an effective therapeutic treatment for cancer patients (38, 39). Gene engineering of patients' peripheral T cells with tumor antigen–specific TCR or chimeric antigen receptor (CAR) gene can overcome challenges in obtaining sufficient numbers of tumor antigen–specific T cells from patients' specimens (16). In solid tumors, because most known tumor-specific antigens are intracellular proteins, such as cancer-testis antigens, TCR rather than CAR genes may be more suitable for manufacture of therapeutic T-cell products. However, because of the variety of HLA types and tumor antigen expression patterns, a large panel of TCR genes specific for different tumor antigens and HLA restriction elements would be required for the treatment of a population of patients with different tumor types. To accelerate the discovery of tumor antigen–specific TCR α and β chain pairs, we have developed a method to construct a library of randomly paired TCRs from tumor tissues that are infiltrated by tumor antigen–specific T cells.

In contrast to other methods that require single-cell suspension, our method only requires snap-frozen tumor specimens, which can be prepared at most clinical sites. Furthermore, as this protocol requires only a few micrograms of total RNA from tumor specimens, we expect that the protocol could be applied to small specimens such as needle biopsies from which isolation of infiltrating T cells may be challenging. Following standard procedures of total RNA extraction and reverse transcription, variable regions of TCR α and β chains were amplified by the common primers, but not TCR Vα-c or Vβ–specific multiplexed primers, in order to minimize PCR bias. These variable fragments were assembled as expressing cassettes and cloned. Through our method, a tumor-infiltrating T-cell–derived TCR library can be prepared as retroviral expression plasmids within 24 hours. We then retrovirally transduced the TCR library into peripheral T cells, in order to screen for tumor antigen–specific TCRs. We successfully identified tumor antigen–specific TCR pairs from 2 of 3 frozen ovarian tumor specimens. The TCRs obtained by this approach were functional and able to recognize tumor in vitro. We demonstrated the therapeutic potential of one of the library-derived TCRs (tumor #3) by adoptively transferring the T cells in a tumor xenograft model.

Our method identified tumor-reactive TCRs despite low frequencies of T cells infected by tumor-derived TCR-expression library at suboptimal viral titers. One of the tumor-derived libraries (from tumor #1) did not contain many tumor-reactive TCRs, although the original tumor specimen contained a high frequency of tetramer-reactive CD8⁺ T cells (data not shown). It is possible that the specimen that was used for single-cell suspension for T-cell staining and that for RNA extraction contained different fractions of tetramer-reactive T cells. Alternatively, tetramer-reactive T cells could be composed of oligoclonal populations, for which the probability to form functional TCR pairs in the library exponentially decreases. Although the multiplexed primers were designed to amplify all known TCR variable regions, it may be possible that the tetramer-reactive TCR gene had unknown polymorphic mutations in their primer-binding regions. Here, we used a small library (3 × 10⁵) to screen on a limited scale. Although this library size is considered to be sufficient to obtain correct TCR pairs from a T-cell clone of more than 1% frequency among total T cells, the library size could be expanded by the use of electrocompetent cells to identify specific TCR pairs from less frequent T-cell clones, for example, those in peripheral blood.

Randomly paired libraries of immunoglobulin heavy and light chains have been used to identify antibodies against therapeutic targets including cancer (40). In general, immunoglobulin heavy and light chains are PCR-amplified and randomly fused via a linker peptide to generate single-chain variable fragments. Then, pairs with the desired specificity are isolated by screening the library for binding to target antigens. However, this method has not been tested to identify antigen-specific TCR heterodimer genes. Our results demonstrated that a tumor-derived randomly paired TCR library is a useful resource to identify tumor antigen–specific TCR pairs. In addition, we were able to quickly generate viral vector constructs containing the new TCRs, which not only sped up the screening process, but also provided a tool to genetically engineer T cells for adoptive transfer studies.

In contrast to other methods used to obtain paired TCR α and β chain sequences from a single cell by single-cell PCR, emulsion PCR, or the pairSEQ platform, our method has the possibility of generating artificial TCR pairs that recognized cancer targets with higher affinity than the natural tumor antigen–specific TCR pair (41). Artificial TCR pairs can cross-react with unexpected antigens, including some expressed in normal tissues. Off-target toxicity of the TCR gene–engineered T-cell product is a problem with nonnatural TCRs such as affinity-enhanced TCRs or murine TCRs (42–44). The TCRs we identified here from tumor specimens are NY-ESO-1–specific, as determined by the TCR-recognizing motif using glycine-substituted peptides. Although testing off-target reactivity is important for any therapeutic TCR gene, candidates for therapeutic TCR genes identified by our method can be extensively tested for cross-reactivity against a panel of normal tissues and genes that contain homologous sequences of the TCR epitope.

In summary, we established a method for discovery and identification of relevant TCRs that can be utilized in a viral vector construct form for downstream translational validation toward therapeutic adoptive cell therapy for cancer.

T. Tsuji, R.C. Koya, and K. Odunsi have ownership interest in a patent application by Roswell Park Cancer Institute. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Tsuji, R.C. Koya, K. Odunsi

Development of methodology: T. Tsuji, A. Yoneda

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Tsuji, J. Matsuzaki, A. Miliotto, C. Ryan, K. Odunsi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Tsuji, K. Odunsi

Writing, review, and/or revision of the manuscript: T. Tsuji, J. Matsuzaki, A. Miliotto, C. Ryan, R.C. Koya, K. Odunsi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Odunsi

Study supervision: K. Odunsi

This work was supported by RPCI, NCI grants P30CA016056 and P50CA159981, Roswell Park Alliance Foundation, Ovarian Cancer Research Fund Alliance (327679), and the New York State Stem Cell Science (NYSTEM: C030158).

We appreciate technical supports from E. Cheryl and A.J. François. We thank members of Center for Immunotherapy at RPCI for valuable comments.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.