Adoptive cell therapy targeting common p53 neoantigens in human solid cancers


 Adoptive cell therapy (ACT) targeting neoantigens can achieve durable clinical responses in patients with cancer. Most neoantigens arise from rare mutations, requiring highly individualized treatments. To broaden the applicability of ACT targeting neoantigens, we focused on TP53 mutations commonly shared across different cancer types. Here, we describe a library of T cell receptors (TCRs) that can target TP53 mutations shared among 7.3% of patients with solid cancers. These TCRs recognized tumor cells in a TP53 mutation- and human leucocyte antigen (HLA)-specific manner both in vitro and in vivo. Patients with chemorefractory epithelial cancers treated with ex vivo-expanded autologous tumor infiltrating lymphocytes (TILs) naturally reactive with mutant p53 experienced limited clinical responses (2 PRs/12 patients), and we detected low frequencies, exhausted phenotypes, and poor persistence of the infused mutant p53-reactive TILs. Alternatively, we treated one patient with a chemorefractory breast cancer with ACT by transducing autologous peripheral blood lymphocytes with an HLA-A*02-restricted anti-p53R175H TCR. The infused cells exhibited an improved immunophenotype and prolonged persistence compared to the TIL ACT and the patient experienced an objective tumor regression (-55%) that lasted 6 months. Collectively, these data demonstrate the feasibility of off-the-shelf TCR-engineered cell therapies targeting shared p53 neoantigens to treat human cancers.

By iterating this approach for the additional 77 samples, we identi ed a total of 21 unique TIL reactivities (27%) including 8 that were previously reported 10-12 : Redundancies of TCRs recognizing the same mutations led us to identify a total of 39 TCRs, including 16 HLA class I-restricted TCRs and 23 HLA class IIrestricted TCRs (Table 1 and Supplementary Table 2). Two different patient TILs reacted against the same TP53 mutation (p53 R175H and p53 Y220C ) with the same HLA restriction. All the TCRs listed in the library were validated to have mutant speci city (no detectable recognition of wild-type peptides at 10 ng/mL or lower concentrations). Because some TCRs targeting "hotspot" mutations, including R175H, Y220C, and R273C, were paired with common HLAs, such as A*02:01 and DPB1*04:02, they could potentially be used to treat a broad range of patients. Overall, 7.3% of patients with solid cancers are estimated to share the corresponding TP53 mutations and HLAs as the TCRs listed in the library and can potentially bene t from off-the-shelf use of the TCRs (Table 1).
The antigen and HLA-speci c target recognition by mutant p53-reactive TCRs We further characterized the anti-mutant p53 TCRs in the library. We rst explored whether T cells transduced with a TCR targeting the HLA-A*02:01-restricted "hotspot" p53 Y220C neoepitope (referred to as Y220C-TCR) could recognize allogeneic or autologous tumor cells. When co-cultured with the cancer cells, healthy donor T cells expressing the anti-p53 Y220C TCR speci cally upregulated 4-1BB against cells that were doubly positive for p53 Y220C and A*02:01 (Fig.   2a). The remaining HLA class I-restricted TCRs with available autologous or allogeneic tumor cells also recognized the tumor cells in an antigen and HLAspeci c manner (TCRs from patients 4141, 4196, 4266, 4324, and 4350; Table 1). In line with that effort, 4 HLA-A*02:01-restricted TCRs targeting p53 R175H were functionally compared using a panel of cell lines. All 4 TCR-expressing T cells speci cally upregulated 4-1BB against SAOS2-R175H, TYK-nu and KLE cells, which were p53 R175H+ and HLA-A*02:01 + , but not against the control tumor cells that were singly positive for p53 R175H or HLA-A*02:01 (Fig. 2b). Whereas 4-1BB upregulation on CD8 + cells against the p53 R175H+ HLA-A*02:01 + tumor cells was similar between the 4 TCRs, the 4196 AV6/BV11 TCR secreted the highest level of IFN-γ when co-cultured with HLA-A*02:01 + T2 cells pulsed with serially diluted peptides (Fig. 2b) and was subjected to further preclinical testing (hereafter, referred to as R175H-TCR). TYK-nu cells successfully grew in NSG (NOD-scid IL2Rg null ) mice but engraftment of KLE cells was not observed. We evaluated the e cacy of ACT using human PBL transduced with the R175H-TCR against engrafted TYK-nu cells. NSG mice were subcutaneously injected with TYK-nu cells (2 to 3x10 6 ). When the tumor reached ~30 mm 2 in size, they received control PBLs transduced with the irrelevant Y220C-TCR or PBLs transduced with the R175H-TCR. First, we tested this TCR by transducing two different healthy donor PBLs. The healthy donor PBLs (2 or 5x10 6 cells) expressing the R175H-TCR led to tumor regression and improved overall survival relative to the control-TCR transduced PBLs ( Fig. 2d-g). PBLs from a patient transduced with the R175H-TCR also inhibited tumor growth and showed speci city for p53 R175H+ TYK-nu cells compared to control p53 Y220C+ 4259 PDX (Fig.  2h,i); however, even at higher dose levels, patient PBLs failed to achieve tumor regression to the same degree as the healthy donor PBLs.
Treatment of patients with epithelial cancers with ACT using autologous TILs targeting p53 neoantigens Next, we tested the feasibility, safety, and e cacy of ACT targeting p53 neoantigens in 12 patients with chemorefractory epithelial cancers by transferring ex vivo-expanded TILs that contained varying numbers of T cells naturally reactive with mutant p53 ( Table 2). The TP53 mutations that were targeted included both "hotspot" and "non-hotspot" mutations. Four and seven patient infusion products contained HLA class I-restricted and HLA class II-restricted anti-mutant p53 TILs, respectively, and 1 infusion product contained 2 different TIL reactivities with each restriction. Two patients (4127 and 4343) showed a partial response (PR) by RECIST 1.0 criteria with response durations of 4 and 6 months, respectively. After a preliminary analysis of TIL infusion products demonstrated elevated levels of PD-1 expression, a pre-transfer dose of pembrolizumab was incorporated into the clinical protocol to prevent inhibition of the transferred cells 4 ( Table 2). The median frequency of mutant p53-reactive TILs in the infusion products determined by deep sequencing of CDR3B was low [median 8.9% (range, 1 to 50.8%), Fig. 3a]. The two patients who had a PR had 2.8% (4127) and 1% (4343) of mutant p53-reactive cells in their infusion products. The majority of 4343 TILs were against another neoantigen (FAM105A Y202C , Table 2). The 12 patients received a median of 8.1x10 10 autologous bulk TILs or 4.0x10 9 mutant p53-reactive cells (range, 7.7x10 8 to 5.2x10 10 ) per patient (Fig. 3a). The median persistence of mutant p53-reactive cells at 6 weeks post-treatment was 0.01% (range, 0 to 1.45%) (Fig. 3a). The phenotype of the infused TILs showed a high degree of exhaustion phenotype: A median of 43%, 33%, and 93% of the bulk TILs were positive for PD1, TIM3, and CD39, respectively (Table 2). Conversely median CD62L expression, a marker for less differentiated naïve and central memory T cells, was 5.02% (range, 0.89 to 29.24%) ( Table 2). Recently, it has been demonstrated that CD39 -CD69 -T cells with stem-like properties were associated with complete melanoma regression but CD39 + CD69 + T cells showed a differentiated phenotype and were associated with poor TIL persistence 18 . We speci cally interrogated the differentiation phenotype of mutant p53-reactive TILs based on CD39 and CD69 expression using HLA class I tetramers, given their low and variable frequencies within the infusion products ( Fig. 3b): Among the patients who received HLA class I-restricted mutant p53-reactive TILs, the infusion products for patients 4266, 4324 and 4350 that had HLA-A*68:01, C*06:02, and A*11:01 restriction, respectively, were chosen based on the availability of tetramers and leukapheresis products. The tetramer + CD8 + TILs in the infusion products for patients 4266, 4324, and 4350 consisted mostly of differentiated CD39 + CD69 + T cells (97.6, 62.8 and 86.8%, respectively) and very few stem-like CD39 -CD69 -T cells (0.045, 2.17 and 2.06%, respectively) (Fig. 3c).
The improved frequency, phenotype and persistence of TCR-engineered PBL We hypothesized that engineering "young" PBLs comprised mostly of naïve and central memory cells to express an anti-mutant p53 TCR might increase the frequencies of mutant p53-reactive T cells and improve their differentiation/exhaustion phenotypes and in vivo persistence. We tested this hypothesis by engineering autologous PBL with an allogeneic anti-mutant p53 TCR for experimental treatment of patient 4349 with a chemorefractory breast cancer. A 48year-old HLA-A*02:01 + patient, who had progressed through 10 prior lines of hormonal and chemotherapies (Supplementary Table 3), was treated with autologous PBLs engineered with the R175H-TCR that showed anti-tumor e cacy in a murine model. WES of a lymph node metastasis identi ed 119 nonsynonymous somatic mutations, including p53 R175H , but no mutant p53 R175H -reactive TILs were detected. The frequency of R175H-TCR + cells in the infusion product following retroviral transduction and rapid expansion was 64%, which was higher than that of any of the selected TIL infusion products ( Fig. 3a and Table 2). The patient presented with advanced disease at the time of treatment: fungating masses that had eroded the tissue of bilateral breasts, a left pleural effusion, and cardiac tamponade requiring a surgical pericardial window 6 days before the ACT treatment. The patient received a standard lymphodepleting chemotherapy followed by ACT of 5.3x10 10 transduced cells with no post-infusion IL2 due to her severely compromising cardiopulmonary disease. Upon cell infusion the patient developed an acute cytokine release syndrome (CRS) (Extended Data Fig. 4a). The day after cell infusion, she received one dose of dexamethasone, which signi cantly improved her symptoms. At follow-ups 6 and 14-weeks post-cell therapy, target lesions were down 37% and 55%, respectively by RECIST 1.1 criteria. The metastases in the pericardium, chest wall, and the subcutaneous tumor deposits signi cantly decreased and all detectable skin lesions completely resolved by day 60 post-cell therapy (Fig. 4a,b). The genetically engineered infusion product for patient 4349 contained higher numbers of stem-like CD39 -CD69cells (32.7%) relative to the TILs from the naturally selected TIL patients, 4266 (0.045%), 4324 (2.17%) or 4350 (2.06%) ( Fig. 3c and Extended Data Fig. 3). Immunostaining of the skin biopsies at day 0 before the cell therapy but after the lymphodepleting chemotherapy revealed healthy tumor cells with high expressions of p53 in tumor cells (Fig. 4c). At day 0, we did not detect any tumor in ltrating CD8 + T cells, PD1 + T cells or PDL1 + tumor cells. However, at day 6 post-cell therapy, we saw necrotic tumor cells, a decrease in p53 + tumor cells, and a dramatic increase in CD8 + T cells as well as PD1 + T cells and PDL1 + tumor cells. Targeting the 3' UTR sequence of the MSGV1 retroviral backbone used to express the R175H-TCR by RNAscope, we determined that many tumor in ltrating T cells were indeed MSGV1 + cells expressing the R175H-TCR (Fig. 4d). In addition to the increase in PD1 + T cells in the skin biopsies, we detected increased PD1 + R175H-TCR + T cells in the peripheral blood and pleural effusion at day 14 (Extended Data Fig. 4b), and the patient received 1 dose of pembrolizumab at day 16. Twelve hours later, she experienced fever, respiratory distress, kidney dysfunction, and skin rashes, which improved within 2 weeks. Her symptoms coincided with a sharp increase in the serum cytokine levels of IFN-g, IL6 and IL10 (Extended Data Fig. 4a).
Persistence of R175H-TCR + T cells at 6 weeks post-ACT was higher than that seen in any of the patients receiving TIL treatments ( Fig. 3a and Extended Data  Table 4 and 6). In line with these data, we detected functional R175H-TCR + T cells secreting IFN-g upon co-culture with imDC pulsed with the p53 R175H ME from the PBL collected at 4 months post-ACT, reinforcing the presence of circulating R175H-TCR + memory T cells (Extended Data Fig. 5c). At 6 months post-ACT, the patient progressed with new cutaneous metastases on her bilateral breasts. A skin biopsy con rmed the presence of tumor cells with intact expression of p53 and pan-HLA class I, as well as a lack of R175H-TCR + T cells (Extended Data Fig. 6); however, WES of the biopsy revealed that a portion of chromosome 6 was lost, resulting in HLA loss of heterozygosity (LOH), including HLA-A*02:01 ( Fig. 4h and Extended Data Fig.7). We did not detect HLA LOH from the 2 metastases we had originally resected to generate TILs. Additionally, we orthogonally validated the loss of HLA-A*02:01 in the progressing lesion by RNAscope. In contrast to high expression of HLA-A*02:01 in the tumor cells in the pretreatment biopsy, a vast majority of the tumor cells in the progressing lesion did not express HLA-A*02:01 (Extended Data Fig. 8). The patient died at 8 months post-ACT due to complications secondary to disease progression.

Discussion
Despite recent advances in targeted cancer therapies, TP53 mutations, the most frequent mutations in cancer, remains "undruggable" 26 . ACT may provide a unique opportunity to target TP53 mutations but its effectiveness against TP53 mutations has not been systematically tested. In the present study, we describe a library of TCRs targeting shared TP53 mutations, including both "hotspot" and "non-hotspot" TP53 mutations. These well-de ned TCRs will be readily available to transduce patient's PBL to target p53-mutated cancers in the same manner as chimeric antigen receptor (CAR) T cells are produced; therefore, this approach is expected to reduce the time required for neoantigen screening and extensive cultures of T cells, which can take several months to a year. More advanced off-the-shelf ACT approaches involving allogeneic donor PBL could be available and are being tested extensively in the CAR setting 27,28 . The broad applicability of the anti-mutant p53 TCR library contrasts with targeting private neoantigens which are only applicable for a single autologous treatment 5,6 . We are making the full sequences of the TCRs available (Supplementary Table 2) and are continuing to add new TCRs to the library. Some of the anti-mutant p53 TCRs reacted with autologous or allogeneic cancer cells in a TP53 mutation and HLA-speci c manner. Notably, ACT of the R175H-TCRengineered T cells led to a signi cant decrease in the growth of TYK-nu cancer cells in NSG mice. TYK-nu cells have been shown to have low cell surface expression of the p53 R175H epitope:HLA-A*02:01 complex (1.5 copies/cell) 14 ; however, the R175H-TCR-engineered PBLs effectively targeted the tumor cells in vitro and in vivo. Various studies have shown that CD4 + T cells can be e cacious in treating cancers in both preclinical and clinical settings 1,29-32 ; however, because tumor cells do not normally express HLA class II, the CD4 + TILs or TCRs were not tested against tumor cells in this study. Because APCs are required for CD4 + T cells to indirectly recognize tumor antigens, humanized mouse xenograft models with intact human APCs may be useful to study the e cacy of ACT of CD4 + TCRs targeting p53 mutations.
Our analysis of the 12 infusion products used for autologous TIL treatment identi ed the limitations of naturally selecting mutant p53-reactive TILs, including low frequencies of mutant p53-reactive T cells, exhausted/differentiated immunophenotypes, and a lack of persistence of the infused T cells. Chronic antigen exposure and/or an immunosuppressive tumor microenvironment may contribute to the exhausted/differentiated phenotypes of TILs, which in turn may limit the proliferative potential and persistence of TILs post-infusion 33 . For instance, ~98% of p53 R248W -reactive cells in the infusion product for patient 4266 exhibited an exhausted/differentiated phenotype of CD39 + CD69 + . Despite the large number of p53 R248W -reactive cells (5.3x10 10 cells) given to patient 4266, we detected very few in circulation at 6 weeks post-infusion (Table 2). In contrast, genetically engineered PBLs with TCRs showed improved persistence. Additionally, our single cell whole transcriptome analysis of patient 4349's TCR-engineered PBLs at 6 weeks-post ACT demonstrated that a cluster of T cells had acquired a central memory phenotype with features of stem-like T cells, which might have a long-lasting impact on anti-tumor immunity. The sequential skin biopsies from patient 4349 demonstrated that "cold" tumor cells with no in ltrating T cells before cell infusion became heavily in ltrated with R175H-TCR + T cells, suggesting TCR-engineered PBLs were able to home to the tumor site. Given the patient's multiple prior chemotherapy regimens, it is unlikely that a single cycle of lymphodepleting chemotherapy contributed to the tumor regression.
This single case is insu cient to determine the causal relationship between the young and exhausted phenotypes of TCR-engineered PBL and TIL therapies, respectively, and their clinical responses; however, two patients (4141 and 4196, see Table 2) who received autologous TIL therapies targeting p53 R175H (A*02:01-restricted) did not have clinical responses while the ACT with PBL transduced with the R175H-TCR, one of the three anti-p53 R175H TCRs isolated from patient 4196's TIL, led to an objective response in patient 4349, indicating the importance of the phenotype of T cells for effective ACT. Furthermore, our preclinical testing of the R175H-TCR-transduced T cells showed that ACT using healthy donor PBLs led to complete regression of TYK-nu tumors in mice whereas ACT using patient PBLs even at a higher dose level did not cause tumor regression, again pointing to the role of T cells phenotypes in successful ACT.
Leukapheresis prior to systemic chemotherapies may help capture less differentiated/exhausted PBLs than collecting T cells that have undergone multiple cycles of depletion and reconstitution following cytotoxic chemotherapies 34 .
Allogeneic use of TCRs targeting neoantigens and their potential toxicity have not previously been explored. Although a single case is insu cient to draw a rm conclusion, we did not detect any off-target toxicity of the transferred autologous PBLs expressing the allogeneic R175H-TCR except for CRS associated with on-target effects similar to CAR T cell therapies 35 . Patient 4349's tumor recurred with LOH of a portion of chromosome 6 spanning the HLA-A*02:01 locus. No HLA LOH was detected in the initial metastases we had resected to generate TILs; however, our data could not determine whether a relatively small subclonal population of tumor cells with HLA LOH existed before the ACT or whether the tumor cells rapidly evolved to lose HLA-A*02:01 following the ACT. Targeting multiple neoantigens with more than one HLA restriction and prescreening patients for intact HLA before ACT will likely be bene cial. Collectively, our data demonstrate the potential of ACT using the library of anti-mutant p53 TCRs to treat p53-mutated advanced cancers.

Subjects and clinical protocols
Written, informed consent was granted from all study participants, and all studies were conducted in accordance with The Declaration of Helsinki, The Belmont Report, and the U.S. Common Rule. This study was approved by the Investigational Review Board at the National Cancer Institute in accordance with an assurance led with and approved by the U.S. Department of Health and Human Services and was registered at https://clinicaltrials.gov under NCT00068003, NCT01174121 and NCT03412877. Adults age 18-70 with upper or lower gastrointestinal, hepatobiliary, genitourinary, breast, or ovarian/endometrial cancer (both protocols), or glioblastoma (NCT01174121) or endocrine tumor, neuroendocrine tumor (NCT03412877) refractory to standard chemotherapy were recruited. Patients underwent surgery and leukapheresis and were treated with either ACT of autologous TILs (NCT01174121) or TCR-engineered autologous PBLs (NCT03412877).
Whole-exome sequencing (WES) and whole transcriptome library prep, next-generation sequencing and data analysis Identi cation of somatic tumor mutations, including TP53 mutations, by WES and whole transcriptome (RNA-Seq) analysis was previously described 5,10,12,36 . In brief, genomic DNA and total RNA was puri ed using the AllPrep DNA/RNA kit (80204, QIAGEN, USA) for fresh tumors and matched normal apheresis samples following manufacturer's suggestions. Whole-exome library construction and exon capture of approximately 20,000 coding genes was prepared using SureSelect XT HS Target  (https://sites.google.com/site/strelkasomaticvariantcaller/), and Mutect (https://www.broadinstitute.org/gatk/) were used to call variants. Next, VCF les were merged using GATK CombineVariants tools and annotated using Annovar (http://annovar.openbioinformatics.org). Somatic copy number alterations were determined using sequenza R package with a mutation frequency threshold adjusted to 0.08 from the default of 0.1 to account for intratumor heterogeneity and normal cell contamination.

Generation of clinical TIL infusion products
Generation of clinical TIL infusion products for ACT has been described previously 2,4 . Brie y, 24 TIL fragment subcultures were ex vivo-expanded as described above. Following neoantigen screening, subcultures with highest mutant p53 reactivity or reactivity against other neoantigens were subjected to a rapid expansion protocol in which target TILs were incubated with irradiated healthy donor peripheral blood mononuclear cells (PBMC), anti-CD3 antibody (clone sorted separately through CD3 + CD4 + CD8 − (for CD4) and CD3 + CD4 − CD8 + (for CD8) gates using SH800S or MA900 (Sony Biotechnology, Japan).

HLA typing and haplotype-speci c copy number analysis of HLA loci
Patient's WES data of tumor and germline (peripheral blood) samples were mapped to human reference genome (hg19) using Novoalign (Novocraft).
Cellularity and purity of these samples were estimated using Sequenza 37 . HLA typing of patients' germline data was computationally determined by taking consensus of predictions from two HLA-typing algorithms, HLA_PRG_LA 38 , and PHLAT 39 . With patient's HLA-typing, cellularity & ploidy estimates and tumor & germline BAM les, we checked for loss of heterozygosity (LOH) in the HLA class I alleles using an adjusted version of the original LOHHLA tool 40 . This custom version of the LOHHLA tool is deposited in Bitbucket (https://bitbucket.org/SENTISCI/lohhla/src/master/). Bulk next-generation TCRB sequencing TCRB survey or deep sequencing was performed from genomic DNA by Adaptive Biotechnologies (USA). Frozen, pelleted PBMCs or TILs (5e4 to 1e6 cells) were submitted for sequencing. Analysis of productive TCR rearrangements was performed using ImmunoSEQ Analyzer 3.0 (Adaptive Biotechnologies).

Single cell TCRA/TCRB next-generation sequencing
Single-cell TCR sequencing was performed using the Takara SMARTer® Human scTCR a/b Pro ling Kit -96 (Takara Bio, USA) according to manufacturer instructions. Brie y, single cells from populations of interest were sorted into wells of a 96 well plate and subjected to cDNA synthesis and ampli cation using SMART technology to incorporate cellular barcoding. cDNA corresponding to TCRA and TCRB transcripts was further ampli ed and prepared for sequencing. Sequencing was performed on an Illumina MiSeq® instrument with paired-end, 2x300bp reads using the MiSeq Reagent Kit v3 (600 cycle) (MS-102-3003, Illumina, USA). Read extraction and clonality counts were determined by the MiXCR software package (Milaboratory, Russia).

TCR cloning and retroviral T cell transduction
Reconstructed variable regions of TCRB and TCRA sequences were linked with a P2A spacer sequence, codon optimized and cloned into an MSGV1 vector with murinized constant region sequences (Genscript, USA) 41 . Transduction of healthy donor PBLs using fresh viral supernatant was described previously 36 .
The R175H-TCR used in the treatment of patient 4349 was previously described 12 . Clinical grade GMP retroviral supernatants were obtained from the Vector Production Facility at Cincinnati Children's hospital. Transduction of patient 4349 PBL was performed as previously described 42 . Transduction e ciency was determined by ow cytometry using an anti-mTCRβ antibody as described above.

Co-culture of T cells with APCs or tumor cells
Screening of TILs for reactivity against TP53 mutations was performed as described previously 36 . Brie y, autologous imDCs or B-cells were electroporated with 5-10 µg of in vitro-transcribed TMG RNA (1x10 5 cells/well) and rested overnight or pulsed with peptide (5x10 4 imDCs or 1e5 B-cells/well) for 2 to 4 hours. Differential primary cell counting was done based on cell size using an AOPI dye and Auto T4 or Cellaca MX cell counter (Nexcelom Bioscience, USA). Target cells (imDCs or B-cells) were washed twice and resuspended in 50/50 media and co-cultured with 2-3x10 4 T cells in ELISpot plates (EMD Millipore, MA).
Phorbol 12-myristate 13-acetate (PMA) and ionomycin (Cat # 00-4970-93, Thermo Fisher Scienti c, USA) were included as a positive control. Co-cultured cells were stained and analyzed by ow cytometry as described above, and IFN-g ELISpot plates were processed according to the manufacturer's instructions (Mabtech, Sweden). Tumor cells were co-cultured at a 1:1 to 2:1 (target:effector) ratio with T cells (2 to 10 x10 4 cells/well) overnight in at-bottom 96 well plates. Following co-culture supernatant was analyzed to assess IFN-secretion by ELISA (EHIFNG, Thermo Fisher Scienti c, USA) and the co-cultured cells were analyzed by ow cytometry as described above.
Immunohistochemistry and RNAscope analysis of patient 4349's biopsies Bioinformatic analysis for single cell sequencing data Single cell transcriptome sequencing data were rst processed using Cell Ranger pipelines (v3.1.0; 10X Genomics, USA). The demultiplexed sequencing data were mapped to the reference genome database (human reference GRCh38-2020-A, 10X Genomics). Gene expression matrices were generated from the unique molecular identi er-collapsed read counts on individual cell barcodes. Gene expression matrices with error-corrected hdf5 reads and cell barcode included cells and annotated genes. Low-quality cells or doublets were ltered out based on the feature distribution, and genes with fewer than 5 read counts across all cells were removed. The expression matrix was converted to TPM using R (3.6.3). Further analyses were performed using R package Seurat v4 43,44 . A common gene expression matrix was generated for patient 4349's infusion product and 6-week-post-treatment PBL cells. The expression matrix was standardized by regressing out highly variable genes and mitochondrial genes followed by normalization and scaling. We excluded all TRAV/TRBV genes to remove endogenous TCR expression as a source of clustering bias prior to processing by Seurat 18 . Principal components (PCs) were generated, and elbow and Jackstraw plots were used to de ne signi cant PCs for clustering. Uniform Manifold Approximation and Projection (UMAP) plots were generated by the clustered PCs. Cluster markers were obtained for individual clusters using default parameters. Heatmap of genes of interest were generated by DoHeatmap function using a downsample parameter of 100 (Extended Data Fig.5A). The featureplot function was used to display genes of interest on the UMAP (Fig. 4g and Extended Data Fig. 5b). Single-cell gene set enrichment analysis (scGSEA), a rank-based gene signature metric that computes the expression score of a gene list relative to all other genes using single cell RNA expression, was performed on patient RX and PBL single cell transcriptome data as described before 18,45 . In brief, normalized scRNA gene expression matrices with barcodes were used as input in conjunction with 72 single cell gene set lists (Supplementary Table 4) or gene sets from the bulk RNA sequencing (496 gene signatures from ImmunesigDB, Supplementary Table 5 and 6). Only CD8 + cell signatures were considered. scGSEA scores for all gene signatures were calculated using GSVA function and z-scaled for cross-signature comparisons across cells and samples. Clustered correlation matrices between various gene signatures from single T cells were generated using the R package Corrplot (https://github.com/taiyun/corrplot). Cell types with highest correlation scores with cluster markers generated using patient RX and PBL_6w in conjunction with differentially expressed genes were used to assign cell-types to the RX and PBL_6w cells.

Flow cytometry-based multiplex blood serum cytokine analysis
The serum samples were longitudinally collected for patient 4349 at intervals indicated in Extended Data Fig.4A. The levels of cytokines, including IFN-g, IL-6, and IL-10, were analyzed using LEGENDplex CD8/NK panel (740267, Biolegend, USA) per the manufacturer's instruction by ow cytometry.

ACT of human TYK-nu cancer cells in NSG mice
Animal experiments were approved by the Institutional Animal Care and Use Committees of the NCI and performed in accordance with the National Institutes of Health guidelines. Six-week-old female NSG mice were obtained from NCI Frederick, USA. Two to three million human ovarian cancer TKY-nu cells or 5 million #4259 colon PDX tumor cells in 100 ul of PBS were injected subcutaneously into the right ank area of NSG mice. Following tumor cell injection, all of the mice were randomized. Two to three weeks after tumor inoculation, 2 or 5 million healthy donor PBLS or 10 or 20 million patient 4349's PBLs genetically engineered with the R175H-TCR, the Y220C-TCR or untransduced PBLs in 500 ul of PBS were intravenously injected into tumor bearing NSG mice. Three daily doses of 180,000 IU of recombinant human IL2 in 500 ul of PBS were injected intraperitoneally following the T cell transfer. Tumor growth was measured twice a week, and tumor size was calculated as the product of two perpendicular measurements. All experiments were conducted in a double blinded manner and under animal study protocol approved by the Animal Care and User Committee of NCI.
Generation of Fig. 1a The schematic of neoantigen identi cation and TCR isolation in Fig. 1A was created using BioRender.com.

Statistical analysis
All mouse experiments were randomized after tumor implantation before ACT without investigator binding, and sample sizes of 4, 5 or 10 mice per experimental arm were used to ensure adequate power. Two-way ANOVA was used to determine statistical analysis of tumor growth between experimental arms using Prism 8. Log-rank test was performed for statistical analysis of mouse survival using Prism 8.

Supplementary Materials
Extended Data Fig. 1

Declarations
Acknowledgments: We thank the CCR Genomics Core, NCI for providing the sequencing service for the single cell whole transcriptome analysis. We thank the NCI Surgery Branch Cell Production Facility for generating clinical TIL and TCR-engineered PBL infusion products. We also thank the Surgery Branch Vector Production Facility for validating the GMP viral vector used in this study, and Arnold Mixon and Shawn Farid for their assistance with FACS. Data and materials availability: The single cell transcriptome data are currently being deposited in dbGaP. The patient tumor samples and phereses are unique materials that will not be available to be shared due to privacy concerns and limited availability. The PDX cell lines and the genetically engineered SAOS2 cell lines are available upon request.