Antitumor Activity Associated with Prolonged Persistence of Adoptively Transferred NY-ESO-1 ^c259T Cells in Synovial Sarcoma

Synovial sarcoma represents approximately 5% of all soft-tissue sarcomas and is characterized by a translocation between SYT on the X chromosome and SSX1, SSX2, SSX4 on chromosome 18 (1). Seventy percent of synovial sarcoma diagnoses occur in patients under 40 years old. Standard therapy for localized disease includes surgical resection with or without radiotherapy, and there is no clearly established role for neoadjuvant or adjuvant chemotherapy (1). In patients with metastatic disease, cytotoxic chemotherapy with agents such as anthracyclines, ifosfamide, trabectedin, or pazopanib can stabilize disease or induce regression in some patients, but the expected 5-year overall survival (OS) is dismal, highlighting the need for more effective therapies (1).

T cells can mediate potent and long-lived antitumor effects, as illustrated by the prolonged duration of benefit following ipilimumab in melanoma (2) and PD-1/PD-L1 blockade in an increasing number of tumor histologies (3–7). However, response to checkpoint blockade varies widely, and checkpoint-responsive histologies typically demonstrate high levels of immunogenicity as measured by nonsynonymous somatic mutation (nSSM) burden (8–11), T-cell infiltration, or PD-L1 expression (7). Like other translocation-driven cancers, synovial sarcoma demonstrates a relative paucity of nSSMs (12), which likely explains its unresponsiveness to checkpoint blockade (13). In contrast, 70% to 80% of synovial sarcomas express NY-ESO-1 (14–16), an immunogenic cancer testis antigen that has been targeted using vaccines (17–21) and adoptive transfer of T cells engineered to express NY-ESO-1^c259, an affinity-enhanced T-cell receptor (TCR) targeting NY-ESO-1/LAGE1a (22–25).

Previous experience with adoptively transferred NY-ESO-1^c259T cells plus high-dose IL2 reported clinical responses in 11 of 18 patients (61%) with refractory synovial sarcoma (22, 23). In those studies, a correlation between response and expansion of the adoptively transferred cells as measured at 30 days was not observed, and persistence of NY-ESO-1^c259T cells beyond 30 days was not reported. Here, we report clinical outcomes along with the characteristics of persistent transduced cells from the initial cohort in a pilot trial assessing NY-ESO-1^c259T cells (SPEAR T cells) in patients with advanced synovial sarcoma.

The study protocol specified that this cohort contain 10 evaluable patients. We screened 120 patients with synovial sarcoma for HLA-A*02 and NY-ESO-1 expression. Fifteen patients (median age, 29 years; range, 18–51 years) with recurrent and/or metastatic synovial sarcoma were enrolled and underwent leukapheresis (Supplementary Fig. S1). Of the 15 enrolled patients, 14 were HLA-A*02-01 and 1 was HLA-A*02-06 positive. The majority of patients had received at least 2 lines of prior chemotherapy. Three patients had rapid disease progression following enrollment and did not receive treatment. The remaining 12 patients received a lymphodepleting preparative regimen, comprised of fludarabine 30 mg/m² on days −4 to −1 and cyclophosphamide 1,800 mg/m² on days −2 and −1, followed by cell infusion on day 0. Table 1 summarizes the demographics, clinical response, and cell infusion doses for these 12 patients. The median-infused NY-ESO-1²⁵⁹T-cell dose was 3.6 × 10⁹ (range, 0.45–14.4 × 10⁹) transduced cells. Manufactured products meeting the minimum protocol-specific dose of 1 × 10⁹ NY-ESO-1²⁵⁹T cells were successfully generated from 10 of 12 patients; the remaining 2 patients were infused with cell numbers below the protocol-specified dose.

Patients infused with NY-ESO-1^c259T cells on this trial demonstrated a median time to initial response of 6.2 weeks (range, 4–9 weeks) and a median duration of response of 30.9 weeks (range, 13–72 weeks; Table 1). The overall response rate (ORR) in this study cohort was 50%, with 1 confirmed complete response (CR) and 5 confirmed partial responses (PR; 95% Clopper–Pearson confidence interval, 0.21–0.79). The median progression-free survival (PFS) was 15 weeks (range, 8–38 weeks). One responder (patient 205) underwent surgical resection prior to progression and therefore is censored for this endpoint from the time of resection. The current estimate of the median OS is approximately 120 weeks (range, 37 weeks–undetermined value, as the upper bound has not been reached; Fig. 1A). Five patients are alive (at March 30, 2017, data cutoff), including 2 patients (patients 202, 204) who progressed following response but underwent surgical resection at the time of progression (Fig. 1B). An additional patient (patient 201) underwent surgical resection at the time of progression, but subsequently died. Therefore, the proportion censored at the time of data cutoff is 5/12 (42%), and a longer follow-up time is needed to better estimate the median OS with precision. Two patients who received a subtarget (<1 × 10⁹) transduced cell dose progressed within 12 weeks.

The pace of antitumor response was not compatible with antitumor effects mediated solely by chemotherapy. Maximal effects of chemotherapy are typically observed within 4 weeks of treatment; however, 7 patients experienced continued decreases in tumor burden following the 4-week evaluation point (Fig. 1C). Maximal antitumor responses occurred in 4 patients more than 3 months following cell infusion. Several features of the clinical course of patients treated on this study demonstrate that antitumor effects were immune-mediated. Patient 201 demonstrated transient radiographic worsening of miliary lung metastases within 48 hours following cell infusion associated with increased circulating C-reactive protein levels, followed by a CR lasting until 9 months postinfusion (Fig. 2A and B). We also observed direct evidence of tumor trafficking by NY-ESO-1^c259T cells in patient 206, who received a dose of NY-ESO-1^c259T cells below the protocol-specified minimum dose (Table 1), then developed signs and symptoms consistent with a delayed cytokine release syndrome (CRS) on day 56 following cell infusion. On day 60, surgical resection of an enlarging tumor mass (Fig. 2C) and drainage of pleural fluid was performed, which demonstrated NY-ESO-1^c259T cells within both the tumor mass and the pleural fluid (Fig. 2D).

There were no fatal serious adverse events in this treatment cohort. Adverse events ≥ grade 3 were reported in all 12 treated patients, and these were deemed treatment related in 11 patients. The most common adverse events ≥ grade 3 among the 12 treated subjects were lymphopenia (100%), leukopenia (92%), neutropenia (83%), anemia (83%), hypophosphatemia (75%), and thrombocytopenia (67%). The incidence of ≥ grade 3 febrile neutropenia was 17%. Five patients experienced CRS of grades 1 (n = 2), 2 (n = 1), and 3 (n = 2). CRS occurred within a median of 4 days (range, 0–11 days; 1 instance occurred on the day of treatment). The median duration of CRS was 10 days (range, 8–28 days). No events of seizure, cerebral edema, or encephalopathy were reported. Patients were monitored by PCR and found negative for replication-competent lentivirus. In patients in which more than 1% of the peripheral blood mononuclear cells (PBMC) were vector-positive at 6 months postinfusion (n = 4), clonality assessment was carried out to exclude insertional oncogenesis as a mechanism for persistence. All analyzed samples, collected from 6 to 21 months postinfusion, showed a high level of polyclonality of the transduced population and the absence of dominant clones.

We used PCR to quantitate circulating NY-ESO-1^c259 vector copies following T-cell infusion. NY-ESO-1^c259T cells were detectable in all patients following infusion, with peak levels measured within the first 10 days after infusion. Peak NY-ESO-1^c259 vector copy levels (P_max) were statistically significantly higher (P = 0.0411) in responders (median, 106,174 vector copies/μg; range, 76,185–192,445) compared with nonresponders (median, 30,601 vector copies/μg; range, 11,265–119,883; Fig. 3A). Using 75,000 as a post hoc cutoff, 6 of 6 responders demonstrated P_max >75,000 vector copies/μg of DNA, and 1 of 6 nonresponders demonstrated P_max >75,000 vector copies/μg of DNA. Among 7 patients for whom monitoring continued beyond 200 days, circulating NY-ESO-1^c259TCR-expressing T cells were detected (Fig. 3B and C; Supplementary Fig. S2). The persisting cells were polyfunctional, as determined by intracellular cytokine staining (Supplementary Fig. S3). CD4⁺ and CD8⁺ NY-ESO-1^c259T cells persisting in the blood remained virtually negative for exhaustion markers such as PD-1 and LAG3 for the duration of the analysis period (Supplementary Fig. S2). Thus, NY-ESO-1^c259T cells expand significantly in vivo in responding patients and demonstrate long-term persistence and functionality (Fig. 3D) associated with clinically meaningful antitumor effects.

We assessed the memory cell phenotype of NY-ESO-1^c259TCR-expressing T cells in the manufactured product and following infusion in a subset of patients. As expected, due to the ex vivo expansion method used, NY-ESO-1^c259T cells present in the manufactured product predominantly comprised effector memory (TEM; CCR7⁻CD45RA⁻) or effector memory RA⁺ (TEMRA; CCR7⁻CD45RA⁺) cells, with lower proportions of central memory (TCM; CCR7⁺CD45RA⁻) or stem cell memory (TSCM; CCR7⁺CD45RA⁺) cells (Fig. 4A–C; Supplementary Fig. S2). Although the power to detect correlates of response is diminished due to small numbers of patients, we found no differences between manufactured products administered to responders versus nonresponders with regard to proportions of CD4 versus CD8, proportions of TEM, TEMRA, TSCM, or TCM, other cell-surface markers, or NY-ESO-1–induced cytokine production. Following adoptive transfer, we consistently observed an increase in the frequency of TSCM and TCM subsets within the circulating CD4⁺ and CD8⁺ NY-ESO-1^c259T-cell pool, which occurred within the first month and persisted for the duration of our evaluation. Whereas both the CD4⁺ and CD8⁺ NY-ESO-1^c259T-cell pool were enriched for TSCMs, the CD4⁺ NY-ESO-1^c259T-cell pool was significantly enriched for the TCM subset compared with the CD8⁺ NY-ESO-1^c259T cells (P = 0.044). Compiled data for the entire cohort of patients are shown in Fig. 4C.

A high proportion of NY-ESO-1^c259T cells persisting as long as 28 months postinfusion in patient 202 enabled functional analysis in this patient (Fig. 3C). Cells maintained robust expression of CCR7⁺ and CD45RA⁺ (Supplementary Fig. S4). Exposure to antigen ex vivo induced robust killing in the absence of exogenous cytokines (Fig. 3D), and although more differentiated effector cells were generated after killing, a significant pool of CCR7⁺ and CD45RA⁺ cells was maintained, suggesting self-renewal and differentiation ability. Despite persistence of circulating functional NY-ESO-1^c259T cells, this patient's disease relapsed, and at resection the recurrent tumor continued to stain positive for NY-ESO-1 expression (Supplementary Fig. S5). However, we observed no evidence for infiltration of the tumor by NY-ESO-1^c259T cells. Additional postrelapse biopsies were assessed for antigen expression, and NY-ESO-1 continued to be expressed in 5 of 6 samples tested from patient 202 and 1 of 1 sample tested from patient 209 (Supplementary Fig. S6). Although differing levels of expression were observed, this observation may be attributed to tumor heterogeneity.

To probe the origin of the long-lived NY-ESO-1^c259T cells, TCRBV sequencing was performed on sorted, nontransduced CD8⁺ subsets within the apheresis product and NY-ESO-1^c259CD8⁺ subsets for patient 201, for whom ample cell numbers were available (Fig. 5A). Because the dextramer reagent used to identify NY-ESO-1^c259T cells binds more reliably to CD8⁺ T cells compared with CD4⁺ T cells, this analysis was limited to the CD8⁺ subset. Where possible, naïve (N), TSCM, TCM, TEM, and TEMRA subsets were analyzed. NY-ESO-1^c259T cells bearing a TSCM phenotype (CCR7⁺CD45RA⁺) were nearly absent from the manufactured product but were found circulating postinfusion, whereas TEMRA cells, which constituted roughly 50% of the manufactured product, were present at low levels in the circulating pool postinfusion (Fig. 4C). Using the inverse Shannon entropy score as a measure of clonality, we observed that the TEM subset within the apheresis product and TEM cells in the manufactured product manifested diminished clonal diversity compared with other subsets, consistent with previous reports (26). Remarkably, however, postinfusion NY-ESO-1^c259 N, TCM, TSCM, and TEMRA subsets demonstrated similar diversity to their nonengineered counterparts within the apheresis product and showed no evidence for clonal contraction over time. Similar results were observed for TCM and TEM populations in other patients (Supplementary Fig. S7). These findings are consistent with a model wherein the fitness of the NY-ESO-1^c259T-cell pool is similar to that of nontransduced T-cell subsets and wherein a robust regenerative pool maintains diversity of endogenous TCRs despite significant sustained expansion in vivo.

We next sought to track clonal progeny over time using the endogenous TCRBV sequence in NY-ESO-1^c259T cells as a barcode to identify cells with shared clonotypes and to compare subset distribution of long-lived versus random clonotypes. For these analyses, we used a cutoff of 10 copies of the unique TCRBV sequence present at the 6-month time point to define a “long-lived clonotype,” and we used a bootstrap simulation to calculate the proportional subset distribution of 19 long-lived versus 19 randomly selected clonotypes (Fig. 5B and C). Results shown did not depend on the cutoff used. This approach controls for potential bias related to the increased proportion of specific subsets at particular time points (e.g., high levels of TEMRA in the manufactured product). The results revealed clear distinctions between the subset distribution of persistent versus random clonotypes at all time points (χ² test, P < 0.01). TCM subsets were enriched for persistent clonotypes at every time point examined (Fig. 5). TSCM subsets were enriched for persistent clonotypes in the apheresis product, manufactured product, and at 1-month postinfusion time points, but thereafter we observed a reduction. In addition, we also evaluated the distribution of persistent clones across cell subsets at each time point. We observed that the distribution of cell proportions was roughly uniform (and did not significantly diverge from a uniform distribution), whereas at 5 of 6 time points analyzed, the distribution of the random clones was significantly different from a uniform distribution.

Figure 5D illustrates the subset distribution and total number of 11 of the 19 persistent clones, selected because these clonotypes were identified at every time point analyzed. The data illustrate that persistent clonotypes are rarely found in only one subset but rather are present in multiple T-cell subsets at each time point. Persistent clonotypes were predominantly found in both a regenerative subset (e.g., TSCM and/or TCM) as well as an effector subset at the same time point, including in the apheresis and manufactured product. This pattern would be expected if long-lasting clones were derived from a preexisting pool of antigen-experienced T cells with a presence in a self-maintaining regenerative pool that continually and efficiently gives rise to differentiated T-cell populations in response to antigen.

Adoptive cell therapy has emerged as one of the most promising approaches for harnessing the power of the immune response against cancer. Advances in vector design, receptor engineering, and cell production have improved the feasibility of generating activated T-cell products and diminished the time required to deliver products to patients. This progress is exemplified by the impressive results following CD19 chimeric antigen receptor (CAR) cell therapies for B-cell acute lymphoblastic leukemia and diffuse large B-cell lymphoma (27–32), the former of which uses a similar method of T-cell engineering and expansion as this study and has now been approved by the FDA. However, CAR-based therapeutics have not yet mediated benefit in significant numbers of patients with nonhematopoietic solid tumors. Adoptive transfer of tumor-infiltrating lymphocytes (TIL) can reproducibly mediate antitumor effects in melanoma, but TILs are difficult to generate from nonmelanoma histologies. Beyond small studies demonstrating activity of virus-specific TILs in cervical cancer (33) or a TIL product enriched for mutations in hepatobiliary carcinoma (34), the efficacy of adoptive therapy using TILs has not been convincingly demonstrated beyond melanoma. Engineered TCRs represent a promising approach to overcome the challenges associated with generating TILs and the theoretical limitations of low-affinity TCRs for tumor-associated self-antigens present in the natural T-cell repertoire (35).

In this study, safety, feasibility, and clinical and biological effects of autologous T cells engineered to express an affinity-enhanced TCR recognizing the NY-ESO-1–derived peptide SLLMWITQC (NY-ESO-1^c259T cells) were tested in synovial sarcoma. We demonstrate expansion of NY-ESO-1^c259T cells postinfusion in all patients and persistence of NY-ESO-1^c259T cells for 6 months in all responders and in 1 of 6 nonresponders. Previous studies of NY-ESO-1^c259T cells in synovial sarcoma reported similar response rates, but long-term persistence of NY-ESO-1^c259T cells was not reported. In addition, previously reported studies included administration of high-dose IL2, which is associated with significant toxicity (22, 23). In patients with relapsed or high-risk multiple myeloma, NY-ESO-1^c259T cells administered following a myeloablative preparative regimen in the context of an autologous stem cell transplant in an aging population demonstrated significant expansion and persistence associated with clinical benefit. Although clinical responses appeared greater than would be expected with transplant alone in the 20 patients studied, antitumor activity attributable to the infused cell product was difficult to definitively distinguish from the chemotherapy effect (36). A follow-on study is open to investigate the effect of the NY-ESO-1 SPEAR T cells without transplantation (NCT03168438). In the study presented here, NY-ESO-1^c259T cells administered without exogenous IL2 demonstrate impressive clinical activity associated with expansion and persistence following a reduced preparative regimen. Demonstration of transient increase in the size of metastatic lesions consistent with lymphocyte-induced inflammation followed by regression, trafficking of NY-ESO-1^c259T cells into the tumor bed, and a prolonged progressive clinical response in several patients confirms that the antitumor effects observed were immune-mediated. Responders demonstrated enhanced NY-ESO-1^c259T-cell expansion compared with nonresponders.

The impressive persistence of the NY-ESO-1^c259T cells in these patients, associated with continued antitumor effects in vivo and polyfunctionality ex vivo, provided an opportunity to more deeply analyze biological features associated with persistently functional T-cell populations following adoptive transfer. We observed dramatic differences between the subset composition of the NY-ESO-1^c259T-cell pool in the manufactured product, which nearly exclusively comprised TEM and TEMRA cells, versus the composition of the persisting NY-ESO-1^c259T-cell pool in vivo, which was comprised predominantly of TSCM and TCM subsets. Current models of CD8⁺ T-cell differentiation hold that TEMRA and TEM cells derive from self-regenerating pools of naïve, TSCM, and/or TCM cells (37–40). According to these models, CD8⁺ T-cell subset development is linear and hierarchical, such that naïve cells can generate naïve and/or TSCM cells, but TSCM cells do not generate naïve cells. Similarly, TSCM cells can generate TSCM and/or TCM cells, but TCM cells do not generate TSCM or naïve cells. Neither TEM cells nor TEMRA cells are capable of self-regeneration to any significant extent, thus explaining their inferiority for use in the setting of adoptive cell therapy (41, 42). We used fate mapping via next-generation sequencing of endogenous TCRs to test the veracity of this model as it relates to persisting NY-ESO-1^c259T cells following adoptive transfer.

Analysis of the diversity of endogenous TCRs in the NY-ESO-1^c259T cells expanding and persisting in vivo demonstrated similar repertoire diversity compared with nonengineered populations and no evidence of clonal contraction over time, demonstrating that a robust repertoire of T cells contributes to the NY-ESO-1^c259T-cell pool in vivo. This observation suggests that long-term persistence is independent of host factors that might drive survival through the endogenous TCR receptor, such as latent viruses like Epstein–Barr virus or cytomegalovirus. Such approaches have been exploited to assist in long-term engraftment of CAR T cells but are not likely to be at play here, because such a mechanism would be expected to increase the clonality of the persisting population, which was not observed (43, 44). The results also demonstrate that persistent T cells were not derived from one cell subset, but rather their clonotypes were present across all T-cell subsets examined, including naïve, TEM, TSCM, TCM, and TEMRA. These results, working within the traditional model of CD8⁺ T-cell differentiation, lead to the conclusion that persisting clonotypes derive from a naïve subset, capable of both robust regeneration of naïve cells as well as efficient generation of progressively more differentiated subsets in response to persistent antigen exposure in vivo. However, unexpectedly, persistent clonotypes were less enriched in the naïve subset of the preengineered apheresis product than random clonotypes. We were unable to perform this study in additional patients to establish this pattern more robustly, due to sample limitations. However, the data suggest it may be that current standard phenotypic profiles used to define T-cell subsets do not accurately identify pools with capabilities limited to the hierarchical, linear differentiation model described above.

A common feature of clinical studies that demonstrate prolonged persistence of engineered T cells is use of a manufacturing method that simultaneously provides activation and costimulation to the manufactured T cells (45–47). Costimulation through the CD28 molecule increases metabolic capacity of T cells, their telomerase activity, and the antiapoptotic protein BCL-X(L), which contribute to activity and longevity of the infused cells (48–50). Fate mapping, such as that performed here, cannot precisely determine the subset from which individual effectors are derived. Nonetheless, the observation that persistent clonotypes were consistently enriched within naïve, TSCM, and TCM pools compared with random clonotypes provides evidence that phenotypes identified previously to define these pools identify subsets with enhanced regenerative capacity. This raises the prospect that administering selected populations of regenerative populations in future trials may enable dose reduction and potentially enhanced efficacy. Furthermore, the observation that persistent clonotypes distribute more uniformly across regenerative and effector subsets than random clonotypes, due to the efficient generation of effector pools, illustrates that NY-ESO^c259T exhibits a balance of efficient effector T-cell generation and robust self-renewal.

The results reported herein provide evidence that adoptive T-cell therapy for cancer using an affinity-enhanced TCR, without IL2, can provide a robust and long-lasting pool of antitumor effectors that can mediate clinically meaningful antitumor effects. The NY-ESO-1^c259T cells maintained killing potential and demonstrated no evidence for T-cell exhaustion in the peripheral blood compartment, despite prolonged persistence and exposure to antigen in vivo. Further analyses revealed evolution from an infused pool comprised predominantly of effector cells to a persisting pool dominated by TCM and TSCM subsets. Next-generation TRBV sequencing of sorted cell subsets confirmed enrichment of persistent NY-ESO-1^c259T-cell clonotypes in TCM subsets at all time points analyzed and enrichment within TSCM subsets. The data presented are consistent with a model wherein the engineered NY-ESO-1^c259T-cell products administered in this trial contained a stem/progenitor pool capable of both self-renewal and prolonged efficient generation of potent antitumor effectors.

The data and results presented are from Cohort 1 of an ongoing pilot, phase I/II, nonrandomized, open-label study (NCT01343043). This study was conducted in compliance with the Declaration of Helsinki (with amendments), and in accordance with local legal and regulatory requirements. Patients were recruited from 3 academic centers in the United States. The protocol was approved by each center's Institutional Review Board, and all patients signed informed consent forms. Enrollment for the cohort reported has been completed. Of the other 3 cohorts in the study, 2 are ongoing and 1 has been terminated. The results from these 3 cohorts will be reported following study completion.

The primary endpoint was ORR, which was defined as the proportion of patients with a confirmed CR or PR. Secondary endpoints assessing efficacy include the following: time to response (time from T-cell infusion to earliest documentation of confirmed PR or CR), duration of response (time from first documented evidence of confirmed CR or PR to the earliest documentation of disease progression or death from any cause), PFS (time from T-cell infusion to the earliest documentation of disease progression or death from any cause), and the OS rate (time from T-cell infusion to death from any cause). Additional endpoints were safety and tolerability of NY-ESO-1^c259T as well as correlative studies to evaluate expansion and persistence of gene-marked cells and the T-cell subsets.

Key eligibility criteria included patients 4 years or older who have histologically confirmed synovial sarcoma that is unresectable, metastatic, progressive, persistent, or recurrent (advanced disease), are HLA*02 positive, and have tumors that express NY-ESO-1 tumor antigen (≥50% of tumor cells express 2+ or 3+ staining by IHC). Although various HLA*02 haplotypes bind the target peptide and are then recognized by the NY-ESO-1^c259TCR, the affinity of this binding varies. Patients in this cohort were HLA*02:01 or HLA-A2*02:06, which have similar binding affinities. Patients must have previously received an ifosfamide and/or doxorubicin-containing regimen and have measurable disease according to RECIST version 1.1. Patients had Eastern Cooperative Oncology Group performance status 0 to 1 or for children ≤10 years of age Lansky ≥60, had a life expectancy of >3 months, and had left ventricular ejection fraction of ≥40%. Laboratory assessments for eligibility were as follows: absolute neutrophil count ≥1,000/mm³, platelet count ≥75,000/mm³, serum bilirubin <2 mg/dL, ALT and AST ≤2.5× upper limit of normal, and creatinine clearance of ≥60 mL/minute. HLA by high-resolution testing was performed at a local or central laboratory. NY-ESO-1 testing was performed at a Clinical Laboratory Improvement Amendments (CLIA)–certified pathology laboratory at the NCI, Bethesda, MD. Disease was classified according to RECIST v 1.1, and radiological disease assessments were performed at weeks 4, 8, 12, and every 3 months thereafter. Patients who progressed were followed up for long-term toxicity until death or for 15 years postinfusion.

Cells and vector were manufactured as described previously (36).

Engineered T cells were manufactured at the Cell and Vaccine Production Facility at the University of Pennsylvania (Philadelphia, PA) for patients 201 and 202. The remaining patients’ cells were manufactured at Progenitor Cell Therapy, following qualification of the transferred process. Engineered T cells were generated from CD25-depleted CD4⁺ and CD8⁺ T cells that were activated and expanded using αCD3/αCD28 antibody-conjugated beads (Life Technologies). T cells were transduced at a multiplicity of infection of 1 transducing unit per cell. Manufacturing time ranged from 28 to 35 days, including release testing.

The lentiviral vector is a self-inactivating vector derived from HIV-1 containing a woodchuck hepatitis virus posttranscriptional regulatory element (WPRE). An EF1α promoter drives transgene expression. Vector was produced at the City of Hope (Duarte, CA) using transient transfection with four plasmids expressing the transfer vector, rev, VSV-G, and gag/pol, in 293T cells. Supernatant was collected at multiple time points, clarified, treated with Benzonase, and concentrated by tangential flow filtration and centrifugation. Transduction potency was measured on primary T cells.

The NY-ESO-1–expressing A375 tumor cell line was obtained from the ATCC (February 2016). Cells have been routinely screened for Mycoplasma contamination (Mycoplasma Experience Ltd.) and authenticated by short tandem repeat (STR) analysis (LGC Standards). The cells were last authenticated by STR analysis in January 2017. Peripheral blood lymphocytes were obtained from human healthy volunteers following Ficoll–Hypaque density gradient separation (Lymphoprep, Nycomed Pharma AS). T cells from healthy donors were activated, expanded, and transduced identically to the clinical manufacturing process.

Research sample collection and initial processing were performed at clinical sites and then transferred to Cambridge Biomedical, a commercial laboratory operating in compliance with Good Laboratory Practices (GLP), as described previously (36), for further processing and qPCR analyses. For multiparametric flow cytometric analyses, samples were subsequently sent to Caprion, a commercial GLP-compliant laboratory. Assay performance and data reporting conforms with MIATA guidelines (51).

Samples were collected as described previously (36). Peripheral blood samples were collected in sodium heparin CPT tubes (BD Vacutainer) for qPCR analysis or flow cytometric analysis. Samples were delivered to the commercial laboratory from the clinical sites at room temperature and in insulated containers. For flow cytometric analyses, PBMCs were purified, processed, and stored in liquid nitrogen prior to transfer to the commercial laboratory.

Genomic DNA was isolated from PBMCs via QIAamp DNA Blood Mini Kit (Qiagen), quantified by spectrophotometer, and stored until testing at −80°C. qPCR analyses were performed in bulk and in triplicate for each time point using ABI TaqMan technology (6FAM-labeled WPRE probe with MGB quencher) and a validated assay to detect the WPRE sequence present in the lentivirus backbone, using 100 ng genomic DNA per replicate. A standard curve was established using serial dilutions of a plasmid containing the WPRE sequence. The limit of detection for the assay is 100 copies/μg genomic DNA. The white blood cell count was used as an amplifier to calculate the number of gene modified cells in the blood. An assumption of 1 vector copy per transduced cell was based upon the multiplicity of infection of 1 used for cell transduction; however, the gene modified cell population is likely to contain a subset of cells with multiple integration events.

Following receipt at a commercial laboratory, cells were stored in liquid nitrogen vapor phase and thawed on the day of assay. Multiparametric immunophenotyping was performed using approximately 1 × 10⁶ cells/condition. Cells were acquired using a BD LSR II cytometer (BD Biosciences). Compensation values were established using single antibody stains and BD compensation beads (BD Biosciences). Data were analyzed using FlowJo v9.6.2 (Treestar), R v3.3.1 (R Foundation for Statistical Computing; ref. 52), Pestle (NIH), and SPICE (NIH) software packages.

The following antibodies were used for T-cell phenotyping: CD8-BV650 (BD), CD4-BV605 (BD Biosciences), CD3-Alexa Fluor700 (BD Biosciences), CD95-BV711 (BD Biosciences), CD45RO-PerCPCy5.5 (BD Biosciences), CD25-APC-Cy7 (BD Biosciences), CD127-BV421 (BioLegend), CCR7-PE-Cy7 (BioLegend), CD45RA-ECD (Beckman Coulter), PD-1-BV785 (BioLegend), LAG3-FITC (R&D Systems), and TIM3-APC (R&D Systems). The dead cell exclusion stain (Live/Dead Aqua) was purchased from Invitrogen. To detect transduced NY-ESO-1^c259TCR-expressing cells, PE-conjugated pentamer reagents specific for the HLA-A*02:01 SLLMWITQC complex (ProImmune) were used at the manufacturer's recommended concentrations.

Frozen patient PBMCs and healthy donor NY-ESO-1 TCR transduced T cells were thawed in RPMI with 10% FBS. These cells were then stained with a pentamer specific to NY-ESO-1^c259TCR-transduced T cells (A*02:01 SLLMWITQC) for 20 minutes at 4°C, followed by subsequent staining with antibodies for an additional 15 minutes at 4°C. 7-AAD was used as the live/dead marker and was added at least 10 minutes prior to sorting. Fluorescence minus one controls were prepared and used for correct gate compensation settings. Cells were sorted using an Aria Fusion cell sorter (BD Biosciences). Data were analyzed using FACSDiva (BD Biosciences) software postsort. The following antibodies were used for identification of transduced T cells for sorting and subsequent postsort phenotypic analysis: CD8-QDOT655 (Life Technologies), CD4-AF780 (eBioscience), CD3-AF700 (eBioscience), CCR7-PE-CF594 (BD Biosciences) and CD45RA-FITC (eBioscience), CD127-BV421 (BioLegend), CD95-BV711 (BD Biosciences). The NY-ESO-1 Pro5 MHC-pentamer conjugated to PE (A*02:01 SLLMWITQC) was purchased from ProImmune. For lineage-tracing studies, NY-ESO-1 TCR-transduced T cells were stained with NY-ESO-1 dextramer conjugated to PE (A*02:01 SLLMWITQC) and purchased from Immudex. 7-AAD was purchased from BD Biosciences.

Killing assays were carried out with the IncuCyte FLR-Platform (Essen BioScience), as described previously (53, 54). Adherent A375 cells were plated at 2 × 10⁴ cells per well and incubated overnight at 37°C/5% CO₂ in RPMI-1640 medium supplemented with 10% heat-inactivated FBS and GlutaMAX (Life Technologies) in a 96-well flat-bottom plate. The following day, cells were washed twice and incubated with 3 × 10⁴ sorted NY-ESO-1^c259TCR-transduced T cells from either a patient or healthy donor per well and 3.3 μmol/L IncuCyte Caspase-3/7 reagent (Essen BioScience). Cells were imaged over 130 hours at 3-hour intervals to detect apoptosis. Data were analyzed using IncuCyte ZOOM 2015A software (Essen BioScience) to distinguish apoptotic A375 cells from apoptotic T cells.

Serial 5-μm sections of formalin-fixed, paraffin-embedded tissues were analyzed for the following single markers by IHC: NY-ESO-1, Pan-CK (Clone AE1/AE3/PCK26, Ventana), CD45 (Clone 2B11 + PD7/26, Ventana), CD3 (Clone 2GV6, Ventana), CD8 (Clone C8/144B, Dako), CD4 (Clone 1F6, Novocastra), PD-L1 (Clone SP142, Ventana), and PD-1 (Clone SP269, Spring Bioscience). Staining was performed either at a CLIA-certified clinical laboratory (QualTek Labs) or a CLIA-certified and Belgian Accreditation Organization and College of American Pathologists–accredited laboratory (HistoGeneX). For some samples, NY-ESO-1 IHC testing was performed at the NCI (Bethesda, MD).

Double immunofluorescence (IF) labeling assays were performed at a commercial laboratory on the Ventana Discovery XT platform. In the first assay, 5 μm–thick sections were incubated for 2 hours at 37°C with a mix of mouse monoclonal anti-human CD8 (Clone C8/144B, 1/100, DakoCytomation) and rabbit monoclonal anti-human Ki67 (Clone 30.9, 1/4, Ventana). Visualization was performed using Alexa488-labeled goat anti-mouse (1/100, Molecular Probes) and Cy3-labeled goat anti-rabbit (1/1,000, Jackson Immunoresearch Laboratories) antibodies. The second immunofluorescent assay combined rabbit monoclonal anti-human CD3 (clone 2GV6, 1/4, Ventana) with mouse monoclonal FoxP3 (clone 236A/E7, 1/60, eBioscience). Visualization was established using Cy3-labeled goat anti-mouse (1/100, Jackson Immunoresearch Laboratories) and Alexa488-labeled goat anti-rabbit (1/100, Invitrogen Molecular Probes) antibodies. Counterstain for both assays was performed using Hoechst 33342 (1/2,000, Invitrogen Molecular Probes). As a negative control, host immunoglobulins (negative control mouse, clone M1G-45, 1/60–1/200, Abcam and negative control rabbit, 1/5,000–1/25,000, Cell Signaling Technologies) were used to replace the primary antibody.

Virtual images were generated using ZEN 2 blue edition software (Carl Zeiss Microscopy GmbH 2011) on an Axio Scan.Z1 Imaging microscope (Carl Zeiss) using the same exposure times for all samples. Each image consisted of 3 image layers, one for each fluorescent tag (layer 1 for FOXP3/Ki67, layer 2 for CD3/CD8, layer 3 for Hoechst). IF images were quantified within a pathologist-defined tumor region via object-oriented image analysis using Definiens Architect XD v2.1.1 Tissue Studio IF with action library v3.6.1 (Definiens). A specific rule set was written in Definiens Architect XD v2.1.1 Tissue Studio IF to determine the total number of nuclei with expression of signal in layer 1 (FOXP3 or Ki67) and/or layer 2 (CD3 or CD8) of the image. Manual delineation of the tumor region, guided by a histopathology report, was done in Definiens Architect XD v2.1.1 Tissue Studio IF. Artifacts due to tissue processing and whole-slide imaging, such as autofluorescence, tissue tearing, tissue folding, and out-of-focus regions, were excluded. Nuclear objects were generated by Definiens Architect XD v2.1.1 Tissue Studio IF based on the Hoechst layer (layer 3) of the image and the expression of signal present on the other two image layers (layer 1 and layer 2) was evaluated within these nuclear objects. The analysis method considered signals to be positive when signals above the background level (determined via an IgG isotype control) were present.

All stained slides were scanned with whole-slide scanner 250 Flash II (3DHISTECH) to obtain a whole-slide image. A 20 × Plan Apo objective (0.80 numerical aperture) and a CIS (VCC-FC60FR19CL) charge-coupled device progressive scan color camera with a resulting image resolution of 0.24 μm/pixel was used. JPEG image encoding with quality factor 80 and an interpolated focus distance of 15 with stitching in the scan options were chosen. For every slide, a specific scan profile was configured. Scanned images were examined in Pannoramic Viewer (3DHISTECH) to check for image quality and whole-tissue coverage.

Object-based image analysis for Pan-CK, CD45, CD3, CD8, CD4, CD163, CD20, PD-L1, and PD-1 was done with the commercial software VIS version 6.2 (Visiopharm) at a commercial laboratory. Manual delineation of the tumor region, guided by a histopathology report, was done in the Image Analysis module of VIS. Artifacts due to tissue processing and whole-slide imaging, such as ruptured tissue, dirt, or out-of-focus regions, were excluded. For every marker, an APP was written that calculated the area of the manually delineated tumor region and the area of the marker-positive regions.

Flow-sorted populations of cells were snap-frozen on dry ice and submitted for either Survey or Deep resolution sequencing of the CDR3 regions of TCRβ and/or TCRα (immunoSEQ, Adaptive Biotechnologies). Briefly, genomic DNA was extracted from these dry pellets and subjected to multiplexed amplification of TCR sequences using proprietary combinations of V and J gene primers, which are validated to reduce amplification bias. The resulting amplification products were used for next-generation sequencing through the CDR3 region, such that read counts for a given clonally rearranged TCR sequence allow quantification of clone size.

Clonality is calculated as the inverse normalized Shannon entropy, defined as ⁠, where N is the total number of unique clones and p_i is the proportion of clone i. A clone is defined as long-lived if there are more than 10 copies of that clone at the month 6 time point. Results described are not sensitive to the particular threshold defining a long-lived clone. For the sample considered herein, there were 19 long-lived clones.

At each time point, a bootstrapped collection of 19 long-lived clones or control clones (uniformly selected from all clones present) was simulated from the entire set of clones present at that time across all cell types. The proportion of this sample with at least one copy found in each cell subtype was calculated. Clones found in multiple cell subsets were counted toward each. One thousand bootstrapped samples were performed at each time point for both long-lived and control samples, and the results of this calculation were not sensitive to this parameter. Significance was determined with a two-tailed Student t test, and P values were corrected for multiple hypothesis testing via the Bonferroni method.

The primary endpoint is ORR per RECIST v1.1, defined as the proportion of subjects with a confirmed CR or PR relative to the total number of subjects. The ORR was based on confirmed responses from the investigator assessment of overall response. Subjects with unknown or missing response were treated as nonresponders; that is, they were included in the denominator when calculating the proportion. The ORR was evaluated using 95% Clopper–Pearson confidence intervals. The primary analysis population was the modified intention-to-treat population, defined as subjects who received the NY-ESO-1^c259T-cell therapy.

The secondary endpoints of peak expansion, time to response, duration of overall response, PFS, and OS were evaluated using descriptive statistics and/or graphical displays such as line plots, box plots, or Kaplan–Meier curves, as appropriate. Peak expansion is the maximum persistence value after NY-ESO-1^c259T-cell infusion for each subject. Time to response is defined as the interval between NY-ESO-1^c259T-cell infusion and earliest date of first documented confirmed response. Duration of overall response is defined as the time from confirmed response until the first date of progressive disease per RECIST v1.1. OS is defined as the interval between the date of NY-ESO-1^c259T-cell infusion and death due to any cause. For this data cutoff, if the subject was known to be alive on March 30, 2017, OS was censored at this date. Median survival was calculated using the log–log estimate and the corresponding 95% confidence interval. Peak expansion was compared between subjects who responded and did not respond to NY-ESO-1^c259T-cell therapy using the exact based Wilcoxon test. Flow cytometry was assessed using the Friedman test.

Sandra P. D'Angelo reports receiving commercial research support from Adaptimmune. J. Glod is a consultant/advisory board member for GSK. S.A. Grupp reports receiving commercial research support from Adaptimmune and is a consultant/advisory board member for the same. W.D. Tap is a consultant/advisory board member for Adaptimmune and Immune Design. S. Basu has ownership interest (including patents) in Adaptimmune LLC. G. Kari has ownership interest (including patents) in Adaptimmune. T. Holdich has ownership interest in Adaptimmune (share options). C.L. Mackall is a member of the Adaptimmune Scientific Advisory Board. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.P. D'Angelo, L. Melchiori, M.S. Merchant, S. Grupp, K. Chagin, G.K. Binder, H. Zhang, T. Trivedi, L. Pandite, C.L. Mackall

Development of methodology: L. Melchiori, M.S. Merchant, R. Kaplan, S. Basu, N. Bath, G. Betts, H. Zhang, T. Trivedi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.P. D'Angelo, L. Melchiori, M.S. Merchant, J. Glod, R. Kaplan, S. Grupp, W.D. Tap, N. Bath, A. Tipping, G. Betts, I. Ramachandran, H. Zhang, E. Van Winkle, T. Holdich, R. Amado, C.L. Mackall

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.P. D'Angelo, L. Melchiori, M.S. Merchant, W.D. Tap, K. Chagin, G.K. Binder, S. Basu, D.E. Lowther, R. Wang, A. Tipping, J.-M. Navenot, H. Zhang, D.K. Wells, T. Trivedi, T. Holdich, L. Pandite, R. Amado, C.L. Mackall

Writing, review, and/or revision of the manuscript: S.P. D'Angelo, L. Melchiori, M.S. Merchant, J. Glod, R. Kaplan, S. Grupp, W.D. Tap, K. Chagin, G.K. Binder, S. Basu, D.E. Lowther, I. Ramachandran, J.-M. Navenot, H. Zhang, D.K. Wells, E. Van Winkle, G. Kari, T. Trivedi, T. Holdich, L. Pandite, R. Amado, C.L. Mackall

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Melchiori, D. Bernstein, K. Chagin, S. Basu, H. Zhang, E. Van Winkle

Study supervision: M.S. Merchant, S. Grupp, W.D. Tap, K. Chagin, G.K. Binder, L. Pandite, C.L. Mackall

Other (clinical operations for the study): G.K. Binder

We are grateful to the patients who participated in this study. This trial is registered on ClinicalTrials.gov, identifier NCT01343043. This work was supported, in part, by the Intramural Research Program of the NCI. This clinical trial was sponsored by Adaptimmune.