Engineering Cancer Antigen-Specific T Cells to Overcome the Immunosuppressive Effects of TGF-β

This article is featured in Top Reads, p.

Adoptive T cell therapy (ACT) uses T cells that recognize cancer Ags after being expanded ex vivo. Transfer of such cells into patients can elicit robust antitumor immunity, long-term stabilization of disease, and in some cases, complete remission (1). A promising class of target Ags for cancer immunotherapy are cancer-testis Ags because their expression is absent or low in normal adult somatic tissue and high in many tumors. Examples include New York esophageal squamous cell carcinoma 1 (NY-ESO-1), its homolog l-Ag family member (LAGE)-1A, and genes from the melanoma Ag gene (MAGE) family. NY-ESO-1 and LAGE-1A expression have been reported in a wide range of tumor types, including neuroblastoma, myeloma, metastatic melanoma, synovial sarcoma (SARC), bladder cancer, esophageal cancer, hepatocellular cancer, head and neck cancer, non–small cell lung cancer (NSCLC), ovarian cancer, prostate cancer, and breast cancer (2). The expression frequency of NY-ESO-1 and LAGE-1A differs greatly between tumor types, with the most commonly expressing tumors being myxoid and round cell liposarcoma (89–100%), SARC (80%), melanoma (46%), and ovarian cancer (43%). Other tumor types, including NSCLC, show protein expression of NY-ESO-1 in the range of 20–40% (2). We used engineered specific peptide enhanced affinity receptor TCRs that recognize the shared NY-ESO-1/LAGE-1A peptide SLLMWITQC (TCR: GSK3377794, formerly NY-ESO-1^c259) or the MAGE-A10–derived peptide GLYDGMEHL (TCR: ADP-A2M10, formerly MAGE-A10^c796) when presented by HLA-A*02:01. In a study of multiple myeloma, 19 of 25 patients responded favorably to ACT, with T cells expressing the NY-ESO-1 TCR when combined with autologous stem cell transplantation (3). Treated patients had a median progression-free survival of 13.5 mo and long-term persistence of transferred T cells (3). T cells carrying the same TCR also produced a 50% effective response rate in metastatic SARC (4). However, as in other similar trials, there is an unmet need to increase the depth and durability of responses, which varied greatly between patients (4–8).

Among the many factors that might negatively affect outcomes of ACT, one of the best documented is TGF-β [reviewed by Dahmani and Delisle (9)]. TGF-β has diverse and context-dependent functions; in healthy stromal tissue, TGF-β is mostly associated with tissue remodeling and regeneration [recently reviewed by Xu et al. (10)]. Its effects on T cells are generally suppressive and include the inhibition of TCR-driven proliferation, Th1 differentiation, T cell cytotoxicity, and production of IL-2 (9). In combination with other cytokines, TGF-β can drive the differentiation of CD4⁺ T cells toward immunosuppressive regulatory T cells (Tregs) (9).

TGF-β can function as a suppressor of tumor growth in the premalignant stage but can enhance transformation and tumor growth once malignant cells have developed resistance (11). TGF-β in tumors can be produced by activated Th17 cells and Tregs, tumor-associated myeloid-derived suppressor cells, and the malignant cells themselves (9). Tissue damage caused by therapeutic interventions such as radiation can also induce secretion of TGF-β by both cancer and stromal cells (12). In line with its pleiotropic nature, TGF-β can promote tumor growth and survival in multiple ways. Specifically, it can suppress CD8⁺ T cell effector responses directly, as well as indirectly, by promoting suppressive stromal cells and/or induction of a fibrotic barrier to restrict T cell infiltration (9). Increased TGF-β levels are associated with disease progression in a variety of cancers (13), including lung (14), prostate, and gastrointestinal cancers (15).

The TGF-β cytokine family is composed of three highly homologous isoforms (TGF-β1, TGF-β2, and TGF-β3) secreted as inactive complexes bound to latency-associated peptide. TGF-β is activated by release from these complexes after cleavage by extracellular proteases and/or low pH [reviewed by Dahmani and Delisle (9)]. The signaling-competent TGF-β receptor (TGFβR) is a tetramer, comprising two subunits each of TGFβRI and TGFβRII. The three isoforms of TGF-β vary in their use of coreceptors, but all isoforms require TGFβRII for signaling (16). On binding of TGF-β, TGFβRII phosphorylates the cytoplasmic tails of TGFβRI. This recruits the signal-transducer proteins similar to mothers against decapentaplegic (SMAD) 2 and SMAD3, which initiates a complex, downstream gene-expression profile (17, 18). TGF-β can also trigger signaling pathways that require alternative SMAD proteins or are SMAD independent. Together, this results in differential effects, depending on the TGF-β isoform, receptor, and target cell (11).

Deletion of the cytoplasmic tail of TGFβRII generates a dominant-negative receptor (dnTGFβRII) that can combine with itself or wild-type TGFβRII to form an inactive dimer, which can bind TGF-β, but when combined with homodimeric TGFβRI does not allow the tetrameric receptor to signal (19–21). T cells expressing a truncated TGFβR were resistant to the effects of TGF-β and more effective at eradicating tumors in animal models of ACT (22, 23), studies reviewed by Dahmani and Delisle (9), and clinically using EBV-specific T cells (24). GlaxoSmithKline is currently enrolling a phase Ib/IIa, multiarm, open-label pilot study (ClinicalTrials.gov: NCT03709706) of GSK3377794 as a monotherapy or in combination with pembrolizumab in HLA-A*02–positive patients whose tumors express NY-ESO-1/LAGE-1A. GlaxoSmithKline is also planning a phase 1b/IIA study for dnTGFβRII_NY-ESO TCR (GSK3845097) (ClinicalTrials.gov: NCT04526509), and it is important to understand the expression of TGF-β in different tumor indications that also express NY-ESO-1 and LAGE-1A to optimize the selection of patients whom this therapy may benefit.

Of particular interest is NSCLC, a common cancer with NY-ESO-1/LAGE-1A expression in 12–21% of patients (25–27), because this represents a significant number of eligible patients with unmet clinical need. A previous study using engineered T cells targeting NY-ESO-1 showed a partial response in a patient with advanced lung adenocarcinoma (LUAD) (28). Further in-depth analysis was required to understand TGF-β expression in NSCLC as a potential indication of interest. Our results support the use of dnTGFβRII together with engineered TCRs for improving ACT in patients with cancer with TGF-β–mediated immune suppression, initially in patients with SARC in whom there is high frequency of both NY-ESO-1 and TGF-β. Other indications, such as NSCLC, may also be considered, with the added challenge that expression of both is more heterogeneous.

The human biological samples were sourced ethically, and their research use was in accordance with the terms of the informed consents under an institutional review board/ethics committee–approved protocol.

mRNA sequencing data were retrieved from The Cancer Genome Atlas (TCGA) through OncoLand (Omicsoft, Cary, NC). The dataset consisted of 7421 paired samples of primary tumors and adjacent normal tissue. Reads were aligned and counted using Omicsoft Sequence Aligner and RNA sequencing by expectation maximization algorithm. After generating RNA sequencing by expectation maximization–normalized counts, the data were converted to fragments per kilobase million (FPKMs) values and normalized using linear scaling. A cutoff of 10 FKPMs was used to define samples as positive for TGF-β1 or for the combined expression of NY-ESO-1 and LAGE-1A. A one-tailed t test was used to detect upregulated tumor versus normal samples for log(FPKM + 0.1) expression values. Due to low numbers of normal samples, statistical analysis was not performed on ovarian serous cystadenocarcinoma (zero samples), SARC (two samples), and skin cutaneous melanoma (one sample). Sixty genes whose expression showed closest correlation with TGF-β1 in ∼6000 tumors were identified by using nearest neighbor analysis conducted using ArrayStudio Function “Find Neighbors” (Omicsoft) (Supplemental Table I). Statistical analysis was performed by using the t test function and standard software (Excel, R).

Human NSCLC samples (n = 35) from Avaden Biosciences (Seattle, WA) were used in accordance with the supplier’s patient consent documentation. A sequential semiautomated dual mRNA in situ hybridization–immunohistochemistry (ISH-IHC) assay was optimized on the Discovery Ultra (Ventana Medical Systems; Roche Tissue Diagnostics, Oro Valley, AZ). TGF-β1 mRNA was detected via ISH using an RNAscope dual-probe detection system (Advanced Cell Diagnostics, Newark, CA). Peptidylprolyl isomerase B (PPIB) probes were used as controls to ensure RNA quality and consistency of expression. All ancillary detecting reagents, including anti-CD3ε Ab and buffers, were from Ventana Medical Systems. Slides were scanned using the Axio Scan Z1 slide scanner (Zeiss, Cambridge, U.K.) and selected for analysis based on PPIB staining. Expression of TGF-β1 mRNA and cell surface CD3ε protein was quantified from 20× slide scans using Tissue Studio software (Definiens; MedImmune, Gaithersburg, MD).

RNA ISH on formalin-fixed paraffin-embedded human NSCLC tissue samples (Avaden Biosciences) was performed on an automated platform using the RNAscope 2.5 LS Duplex Reagent Kit from Advanced Cell Diagnostics according to the manufacturer’s instructions. In brief, 5-μm formalin-fixed paraffin-embedded tissue sections were pretreated with heat and protease before hybridization with oligoprobes for TGF-β1, CD4, CD8α, CD68, pan-cytokeratin, FOXP3, granzyme B (GZMB), collagen 11 A1, CD274, TNF receptor (TNFR) superfamily member 4, or ICOS. Preamplifier, amplifier, and HRP-/alkaline phosphatase–labeled oligos were hybridized sequentially, followed by chromogenic development. Samples were quality controlled for RNA integrity with a PPIB-RNA–specific probe and for background with a probe specific to bacterial dapB RNA. Samples were counterstained with Gill’s hematoxylin. Brightfield images were acquired using an Aperio AT2 digital slide scanner (Leica Microsystems, Milton Keynes, U.K.) equipped with a 40× objective. Quantitative analysis was performed using the HALO software (Indica Labs, Albuquerque, NM) and proprietary Advanced Cell Diagnostics software. For calculation of H score, cells were grouped into five bins based on the number of dots per cell. Clusters were divided by the typical probe signal area to calculate a dot number for the cluster. An H score on a scale of 0 to 400 was calculated according to the following equation: 0*(% of cells in bin 0) + 1*(% of cells in bin 1) + 2*(% of cells in bin 2) + 3*(% of cells in bin 3) + 4*(% of cells in bin 4).

Melanoma Mel624 cells from Thymed (Wendelsheim, Germany) and melanoma A375, LUAD NCI-H1755, bladder J82, colorectal adenocarcinoma Colo205, colorectal carcinoma HCT-116, and EBV-transformed lymphoblast IM-9 cells from American Type Culture Collection (LGC, Teddington, U.K.) cells were cultured at 37°C/5% CO₂. The medium, hereafter referred to as R10, was RPMI medium with 10% FBS, 100 U/ml penicillin/streptomycin, and 2 mM l-glutamine, all from Thermo Fisher Scientific (Loughborough, U.K.). J82 cells were transduced with a minigene to generate the line J82-ESO, which stably expressed the NY-ESO-1 peptide (29). When indicated, recombinant TGF-β1 (PeproTech, London, U.K.) was added. A375-GFP cells, used as targets in three-dimensional (3D) killing assays, were generated by lentiviral transduction of A375 cells to express GFP within the cytoplasm (Santa Cruz Biotechnology, Dallas, TX). A panel of 20 cancer cell lines were cultured in R10 for 72 h. The supernatants were then tested for production of TGF-β1, TGF-β2, and TGF-β3 using the Bio-Plex Pro TGF-β 3-Plex assay kit (Bio-Rad, Hertfordshire, U.K.) according to the manufacturer’s instructions.

NY-ESO-1 (30) and MAGE-A10 TCRs (31) have been described previously. Self-inactivating lentiviral vectors (32) were prepared, which expressed a human codon-optimized (GeneArt codon optimizer; Thermo Fisher Scientific) single open reading frame consisting of a dnTGFβRII–Furin-F2A module followed by affinity-enhanced cancer Ag-specific TCR-α- and β-chains, separated by a Furin-P2A cleavage site. Initial work was performed using a dnTGFβRII_NY-ESO-1 TCR construct containing a woodchuck hepatitis virus posttranscriptional regulatory element and a hemagglutinin-Ag tag on the N terminus of dnTGFβRII as described by Bollard et al. (33). Subsequently, and for mock clinical-scale cultures, constructs were used that had the human influenza hemagglutinin tag and the woodchuck hepatitis virus posttranscriptional regulatory element removed. A similar approach was used to generate the dnTGFβRII_MAGE-A10 TCR construct. Expression of the transgenes was driven from an elongation factor-1α promoter. Viral vectors were produced by transient transfection with the transgene plasmid, together with three plasmids separately expressing the lentiviral proteins Rev, Vesicular Stomatitis Virus G, and Gag/Pol in HEK293T cells. Supernatant was collected 48 h after transfection and concentrated by centrifugation.

Human primary T cells were obtained from PBMCs of healthy donors, stimulated overnight with anti-CD3/anti-CD28 Cell Therapy Systems (CTS) beads (Thermo Fisher Scientific) and 100 U/ml recombinant human IL-2 (rhIL-2; Proleukin; Prometheus Laboratories, San Diego, CA), and transduced on the following day with lentiviral particles (31, 34). Nontransduced (NTD) T cells were activated and expanded from the same donors for use as negative controls and for equalizing frequencies of transduced cells where required. T cells were produced at two scales using slightly different methods, as described later.

PBMCs from healthy donors were depleted of CD14⁺ cells using anti-CD14 MicroBeads (Miltenyi Biotec, Surrey, U.K.). T cells were cultured overnight with anti-CD3/anti-CD28 CTS beads in IMDM with GlutaMAX supplemented with 10% FBS and 100 U/ml penicillin-streptomycin (all from Thermo Fisher Scientific), in the presence of 100 U/ml rhIL-2. Titrated concentrations of lentiviral particles were added to achieve equivalent transduction as assessed by TCR β-chain V region (Vβ) flow cytometry. CTS beads were removed after 5 d, and the T cells were expanded for 12–14 d, replacing the IL-2–containing medium regularly.

T cells were expanded using a process similar to that used for clinical manufacturing. T cells were isolated from PBMCs by anti-CD3/anti-CD28 CTS Dynabead purification, incubated overnight with anti-CD3/anti-CD28 beads in TexMACS medium (Miltenyi Biotec) in the presence of 100 U/ml rhIL-2, and transduced with lentiviral vector the following day. T cells were expanded for a total of 12–13 d, with transfer from tissue culture flasks to a Xuri W25 bioreactor (GE Healthcare, Amersham, U.K.) on day 5 or 6. CTS beads were removed on the day of harvest.

T cells were rested in RPMI with 0.5% FBS for 2 h, then stimulated with TGF-β1 (PeproTech) for 30 min at 37°C at the concentrations indicated. Cells were incubated with LIVE/DEAD Fixable Aqua Dead Cell Stain (Invitrogen, Paisley, U.K.), followed by a fluorescein-conjugated Ab against CD3, CD8 (BD Biosciences, Swindon, U.K.), or CD4 (eBioscience, Hatfield, U.K.), and a PE-conjugated Ab against TCR Vβ13.1 (NY-ESO-1 TCR), Vβ13.2 (MAGE-A10 TCR) (Beckman Coulter, High Wycombe, U.K.), or allophycocyanin-conjugated TGFβRII (Thermo Fisher Scientific). Cells were fixed and permeabilized using Phosflow Lyse/Fix Buffer and Phosflow Perm Buffer III and stained with a SMAD2/3 Ab (all from BD Biosciences). Flow cytometry of live gated cells was performed on BD-LSRFortessa X-20 using FACSDiva software (both from BD Biosciences) for acquisition. FlowJo software (FlowJo, Ashland, OR) was used for data analysis.

To demonstrate expression of dnTGFβRII, we purified total RNA from 5 × 10⁶ transduced T cells (Qiagen RNeasy kit; Qiagen USA) and reverse transcribed (Invitrogen SuperScript IV; Thermo Fisher Scientific, USA). PCR amplification was then performed with Platinum PCR SuperMix (Thermo Fisher Scientific, USA) with primers specific for detection of the dnTGFβRII sequence and not the endogenous TGFβRII gene sequence because the reverse primer was designed to bind downstream of the stop codon.

Frozen mock clinical-scale expanded T cells were thawed and rested for 26 h in RPMI without tryptophan (Pan Biotech, Wimborne, U.K.) supplemented with 1 mg/ml BSA, 10 μg/ml human insulin, 5 μg/ml human holo-transferrin (Sigma-Aldrich, U.K.), and 1% glutamine. T cells were labeled with 1 μl/ml violet proliferation dye 450 (BD Biosciences). A total of 0.5 or 1 × 10⁶ labeled T cells in a final volume of 1 ml medium were incubated alone or mixed at a 5:1 ratio with irradiated HLA-A*02:01 NY-ESO-1 and MAGE-A10–positive A375 cells and cultured for 3 d in the presence of variable concentrations of TGF-β1. All experiments included a negative control, consisting of the cell line HCT-116, and a positive control, consisting of target cells pulsed with either a heteroclitic variant of the NY-ESO-1 peptide (SLLMWITQV) or the MAGE-A10 (GLYDGMEHL) peptide (Peptide Protein Research, Hampshire, U.K.). Cells were harvested and stained with 7-aminoactinomycin D and conjugated Abs against CD3, CD4, CD8, TCR Vβ13.1, or TCR Vβ13.2 (BD Biosciences). Cells were manually gated using FlowJo software. The division index (DI) was calculated as follows: $DI=Σ0i(afi* i)Σ0iafi$⁠, where $afi=fi2i$ and i represent the round of division and f its frequency, respectively.

Transduced and NTD T cell batches had similar percentages of CD8⁺ cells, but transduction was less efficient for vectors containing the dnTGFβRII_TCR module compared with TCR alone. To normalize the transduced T cell frequencies, we added NTD T cells from the same donor where required and confirmed matching percentages of transduced T cells by flow cytometry.

T cells and target cells were cocultured in R10 with different concentrations of TGF-β in 96-well flat-bottom plates at 37°C/5% CO₂ for 24–72 h. Supernatants were assessed for cytokine production using either 17-Plex Human Cytokine MAGPIX kit (Bio-Rad) or DuoSet ELISA kits (Bio-Techne, Abingdon, U.K.) for analysis of IFN-γ, IL-2, and GZMB. The plates were developed using either tetramethylbenzidine substrate or Glo luminescence HRP substrate (Bio-Techne), and each plate was incubated for 5 min before being read on the FLUOstar Omega plate reader (BMG LABTECH, Cary, NC). Data analysis was conducted in the Omega data analysis software version 3.10R6 (BMG LABTECH).

To assess 2D cytotoxicity, we cultured target cells overnight in 96-well flat-bottom plates in R10 at 37°C/5% CO₂. The following day, T cells were added with 5 μM CellPlayer Caspase 3/7 apoptosis reagent (Essen BioScience, Royston, U.K.), then cultured for 100 h in IncuCyte FLR or IncuCyte ZOOM (Essen BioScience) imaging every 2–4 h. Target cell death was assessed using a size exclusion gate of >100 μm to differentiate between apoptotic T cells and target cells. Normalized area under the curve (AUC) at 40 h after T cell addition was used to compare between donors.

For 3D killing assays, A375-GFP cells were seeded in ultra-low-adhesion plates (Brand, Wertheim, Germany) at a density of 200–250 cells/100 μl/well to generate microtissues 500–550 µm in diameter, 5–6 d later. Fifty thousand T cells per well were added and normalized for equal frequencies of transduced cells. Recombinant human TGF-β1 was added to final concentrations of 5–10 ng/ml at the same time. Alternatively, target cells were preincubated with TGF-β1 during spheroid formation to ensure even TGF-β distribution throughout the microtissue. Total volumes of medium were adjusted to 200 μl/well. 3D spheroid microtissue killing assays were performed by time-lapse imaging as described previously. Raw fluorescent images were exported using Essen BioScience proprietary software. The microtissue area was analyzed using a custom AxioVision software macro (Zeiss), and cytotoxic killing was quantified as the change in total GFP fluorescent area.

To obtain an estimate of how widely applicable a dnTGFβRII approach in ACT might be, we analyzed data in TCGA for the expression of TGF-β1 and TGF-β–induced (TGF-BI) mRNAs in 17 cancer indications (each 103–1107 primary tumor samples). Compared with its abundance in matched normal tissues, TGF-β1 mRNA was significantly upregulated in cancers of breast, esophagus (esophageal carcinoma), glioblastoma (glioblastoma multiforme), head and neck (head and neck squamous cell carcinoma), and kidney (kidney renal clear cell carcinoma). TGF-β1 mRNA levels in other cancers showed smaller differences between tumor and adjacent normal tissue, which were not statistically significant (Fig. 1A, upper panel). Basal levels of TGF-β were higher in normal lung and pancreatic tissue than in other tissues, which reflects the importance of TGF-β in maintaining normal tissue homeostasis, especially at mucosal sites (35). Expression of TGF-BI, a gene induced by TGF-β pathway activation (36), generally mirrored that of TGF-β (Fig. 1A, lower panel), supporting the premise that TGF-β mRNA in tumors corresponds to biologically active TGF-β.

Tumor samples were then analyzed for expression of NY-ESO-1 and LAGE-1A. A sample was considered positive for a gene when transcript levels of NY-ESO-1/LAGE-1A were >10 FPKMs. By this definition, the frequency of HLA-A*02:01 was ∼40% in NY-ESO-1/LAGE-1A–positive and NY-ESO-1/LAGE-1A–negative tumors. Eighty percent of NY-ESO-1/LAGE-1A–positive tumors were positive for TGF-β1 mRNA, either alone or together with TGF-β2 and TGF-β3 (Fig. 1B).

To obtain additional confirmation for the biological activity of TGF-β, we analyzed 6000 TCGA tumor samples for expression of a subset of 60 genes known to be regulated by, or associated with, TGF-β expression and TGFβR signaling (Fig. 1C, Supplemental Table I). This analysis was focused on tumor types that expressed >10% combined NY-ESO-1 and LAGE-1A. These eight indications were selected because they have been shown to commonly express NY-ESO-1 and LAGE-1A (2) and may be considered suitable for clinical trials using the NY-ESO-1 TCR with GSK3377794 and GSK3845097. It is therefore important to understand coexpression of TGF-β in these indications.

An example of the analysis is shown (Fig. 1C), which indicates that the TGF-β–associated genes were consistently upregulated in SARC and skin cutaneous melanoma, while they showed expression that was more heterogeneous in other indications, including esophageal carcinoma, NSCLC, and ovarian serous cystadenocarcinoma.

We assessed the expression of TGF-β by a panel of 20 tumor cell lines from different tissue origins (Supplemental Fig. 1A). TGF-β1 was produced by a number of cancer cell lines from a range of tissue origins (skin, prostate, colorectal, pancreatic, liver, brain, and esophageal) with the melanoma line A375 producing around 5300 pg/ml, the highest levels of TGF-β1 compared with background levels found in medium containing bovine serum, of the 20 cancer types tested. TGF-β2 showed more restricted production, with five of the cell lines producing TGF-β2 above background levels. TGF-β3 was produced at much lower levels and by fewer cell lines than either TGF-β1 or TGF-β2. Overall, the majority of cancer cell lines tested produced at least one of the three isoforms of TGF-β. However, because TGF-β can be secreted in complex with latency-associated peptide that maintains TGF-β in an inactive state (37), it is unclear whether this reflects the levels of truly functional TGF-β produced by the cells.

Because of the large number of patients with NSCLC known to express NY-ESO-1 or LAGE-1A, further in-depth analysis was performed to understand TGF-β expression in NSCLC as a potential indication of interest. To investigate the expression of TGF-β and regulated genes in NSCLC tumors at a cellular level, we performed duplex IHC/RNA ISH (RNAscope) analysis (Fig. 2A). The percentages of T cells (CD3⁺) and TGF-β ⁺ cells varied across patient samples, without any obvious correlation between them. However, CD3⁺ cells were frequently surrounded by TGF-β⁺ cells, in two different regions of the biopsy (Fig. 2B), although there was heterogeneity between samples (Fig 2A).

Further analysis of the cellular composition of NSCLC tumor nests and stroma as assessed by a pathologist revealed cells positive for CD4, CD8, CD68 (macrophages), programmed death-ligand 1 (PD-L1), FOXP3 (Tregs), and COL1A (collagen 1A–producing stromal cells), as well as an abundance of TGF-β⁺ cells (Fig. 2C, upper panel). Slightly higher levels of TGF-β were seen in the samples from LUAD compared with lung squamous cell carcinoma (data not shown). Analysis of cells that were dual positive for TGF-β and other selected markers indicated that the main sources of TGF-β were CD68⁺ macrophages and pan-cytokeratin⁺ tumors and stroma (Fig. 2C, lower panel). Not shown are the frequencies of GZMB, CD274, TNFR superfamily member 4, and ICOS single-positive cells, which were rare (median frequency, <3% of all cells) and were not a significant source of TGF-β in stroma or tumor (median frequency of double-positive cells, <2%).

Analysis of TGF-β1 and IFN-γ mRNA specifically in LUAD and lung squamous cell carcinoma (both n = 15) by duplex RNAscope showed abundant TGF-β1 expression, but only rare IFN-γ–producing cells, both in tumor and stromal cells (Fig. 2D, 2E).

Together, these data suggest that expression of active TGF-β is common in stroma and tumors, where it could contribute to the suppression of infiltrating cytotoxic T cells. Strategies to reduce the sensitivity of cytotoxic CD8⁺ T cells to TGF-β should therefore improve their efficacy in certain tumor microenvironments.

We generated lentiviral vector particles coexpressing dnTGFβRII with affinity-enhanced TCRs. In line with previous observations (38), higher concentrations of lentiviral particles were required to achieve equivalent transduction with the larger dnTGFβRII_TCR vector constructs. Where necessary, in functional assays, samples transduced with TCR only were normalized using NTD T cells to give equivalent frequencies of transduced cells.

To show that the dnTGFβRII was expressed only by transduced T cells and to allow us to differentiate between the transduced dnTGFβRII and the endogenous TGFβRII, we performed PCR analysis using primers capable of discriminating between the endogenous TGFβRII and the dnTGFβRII (Supplemental Fig. 1B). This confirmed that the dnTGFβRII was expressed only by cells transduced with the dnTGFβRII–containing vector and not by cells transduced with the NY-ESO TCR alone or the NY-ESO TCR alongside a control sequence.

Transduced T cells were produced under conditions suitable for a clinical-scale product in a Xuri W25 Bioreactor, stimulated with anti-CD3/anti-CD28 beads, and IL-2 and T cells expressing the dnTGFβRII alongside the NY-ESO-1 TCR expanded comparably with those expressing the NY-ESO-1 TCR alone.

We then examined those T cells by flow cytometry for surface expression of TGFβRII and the transduced TCR. The Ab did not discriminate between the endogenous TGFβRII and the transduced dnTGFβRII. Natural levels of TGFβRII were high on resting NTD T cells and T cells transduced with NY-ESO-1 TCR alone. On transduction with vectors encoding dnTGFβRII_NY-ESO-1 TCR, there was a consistent increase in the median fluorescence intensity (MFI) of cell surface TGFβRII expression compared with T cells transduced with the NY-ESO-1 TCR vector (Fig. 3A), indicating that dnTGFβRII was expressed. This was particularly noticeable on CD4⁺ T cells, where expression of the endogenous TGFβRII was lower. Importantly, increased expression was observed only on the TCR Vβ13.1⁺ (transduced) T cells. When the MFI was assessed across three donors (Fig. 3A, lower panel), a statistically significant increase in TGFβRII expression was seen for the CD4⁺ TCR Vβ13.1⁺ T cells. Increased expression of TGFβRII was not observed with TCR Vβ13.1⁺ cells when transduced with NY-ESO-1 TCR alone, indicating that the increased detection was likely a result of coexpression of dnTGFβRII. The CD8⁺ TCR Vβ13.1⁺ T cells showed a trend toward increased expression; however, expression of the endogenous TGFβRII was naturally much higher on CD8⁺ T cells than on CD4⁺ cells.

With the dnTGFβRII_MAGE-A10 TCR system, we observed a significant increase in the level of TGFβRII expression on T cells from multiple donors, on TCR Vβ13.2⁺, CD4⁺, and CD8⁺ T cells compared with that seen on either NTD or cells expressing TCR alone (Supplemental Fig. 2A). These two sets of data were performed using different lentiviral constructs, using T cells from different donors, at different times with different batches of Ab. It is therefore not possible to directly compare levels of TGFβRII between the NY-ESO and MAGE-A10 TCR systems.

When TGF-β binds to TGFβRII and forms a complex with TGFβRI, one of the key signaling events leading to immune suppression is phosphorylation of SMAD2 and SMAD3 proteins (11). We tested whether T cells expressing dnTGFβRII together with affinity-enhanced TCRs to suppress TGF-β signaling had altered SMAD phosphorylation by costaining with Abs against the transduced Vβ-chain and phosphorylated SMAD2/3. Maximal SMAD2/3 phosphorylation in TGFβRII–negative cells was achieved by adding 10 ng/ml TGF-β. SMAD2/3 phosphorylation in response to 10 ng/ml TGF-β1 was reduced in both CD4⁺ and CD8⁺ TCR Vβ13.1⁺ cells coexpressing dnTGFβRII with NY-ESO-1 TCR (Fig. 3B). In contrast, both CD4⁺ and CD8⁺ TCR Vβ13.1⁺ and Vβ13.1^– T cells transduced with NY-ESO-1 TCR alone showed increased SMAD2/3 phosphorylation in response to TGF-β1, indicating that the TGF-β signaling pathway was intact in those cells. Importantly, these data also showed that although the level of TGFβRII expression was not significantly elevated on CD8⁺ T cells transduced with dnTGFβRII (Fig. 3A), phosphorylation of SMAD2/3 in response to TGF-β1 was reduced (Fig. 3B), indicating that dnTGFβRII was expressed in CD8⁺ T cells.

These data were confirmed using the MAGE-A10 TCR system (Supplemental Fig. 2B). On addition of either 10 or 50 ng/ml TGF-β to the MAGE-A10 TCR T cells, the MFI of SMAD2/3 phosphorylation uniformly increased compared with the untreated control cells. In contrast, T cells expressing dnTGFβRII_MAGE-A10 TCR showed little or no upregulation of SMAD2/3 phosphorylation on addition of exogenous TGF-β, demonstrating that the TGF-β signaling pathway was not functional in those cells.

We next assessed the effect of dnTGFβRII coexpression on the proliferation of T cells expressing affinity-enhanced TCRs when cocultured with tumor cells expressing cognate Ag. In the absence of TGF-β1, T cells expressing NY-ESO-1 TCR with or without dnTGFβRII proliferated similarly (Fig. 4A). However, TGF-β1 at a concentration of 5 ng/ml significantly reduced the proliferation of T cells expressing affinity-enhanced TCR alone (Fig. 4A, 4B). Conversely, although addition of 5 or 10 ng/ml TGF-β1 to T cells transduced with dnTGFβRII_NY-ESO-1 was associated with a trend for a reduction in proliferation compared with proliferation in the absence of TGF-β, significance was not reached (Fig. 4A, 4B).

Similar results were obtained with the dnTGFβRII_MAGE-A10 TCR (Supplemental Fig. 3A), suggesting that coexpression of dnTGFβRII with high-affinity TCRs permits T cells to proliferate in response to Ag-expressing tumor cells in the presence of otherwise inhibitory concentrations of TGF-β1. MAGE-A10 TCR-transduced cells were inhibited with both 10 and 50 ng/ml TGF-β1, while the proliferation of cells expressing MAGE-A10 TCR together with dnTGFβRII remained largely unaffected in the presence of TGF-β1.

We next sought to determine how TGF-β1 influenced cytokine release by T cells expressing high-affinity TCRs when cocultured with A375 melanoma cells or J82 bladder carcinoma cells overexpressing NY-ESO-1 (J82-ESO). As a preliminary experiment, T cells from a single donor were stimulated with A375 or J82-ESO tumor cell lines. Addition of 50 ng/ml TGF-β1 reduced the secretion of IFN-γ, IL-2, and MIP-1β by T cells expressing NY-ESO-1 TCR alone (Fig. 5A–C). However, secretion of these cytokines was not impaired when T cells coexpressed dnTGFβRII. Interestingly, TGF-β1 did not inhibit IL-1β secretion, irrespective of whether dnTGFβRII was expressed (Fig. 5D), whereas G-CSF secretion was inhibited regardless of the expression of dnTGFβRII (data not shown), suggesting that different cytokines may be impacted differently by the presence of TGF-β.

We confirmed these data by testing NY-ESO-1 TCR T cells from multiple donors and showed significant differences in IFN-γ and IL-2 production between T cells with and without coexpressed dnTGFβRII (Fig. 5E, 5F). Further experiments using T cells manufactured at mock clinical scale confirmed that IFN-γ production by T cells expressing dnTGFβRII_NY-ESO-1 were completely resistant to inhibition by up to 50 ng/ml TGF-β1 when cocultured with three different tumor cell lines (Supplemental Fig. 3B).

T cells expressing dnTGFβRII_MAGE-A10 showed a similar pattern of cytokine production as seen with the dnTGFβRII_NY-ESO TCR, although protection from inhibition by TGF-β was not complete (Supplemental Fig. 3C). In summary, T cells coexpressing dnTGFβRII and high-affinity TCRs retained Ag-induced cytokine production, at least partially, in the presence of otherwise inhibitory concentrations of TGF-β1.

We then looked at the impact of TGF-β1 on the cytotoxic activity of T cells. T cells expressing either NY-ESO-1 TCR or dnTGFβRII_NY-ESO-1 TCR rapidly induced tumor cell death, as assessed by caspase-3/7 activation in A375 cells, with maximal apoptosis achieved after ∼40 h (Fig. 6A). Killing activity of T cells expressing NY-ESO-1 TCR was compromised by addition of 10 ng/ml exogenous TGF-β1, whereas T cells coexpressing dnTGFβRII with NY-ESO-1 TCR retained their activity. Similar results were obtained with transduced T cells from five different donors (Fig. 6B), as shown by the increased AUC of target cell apoptosis in the presence of TGF-β1 for T cells expressing dnTGFβRII_NY-ESO-1 compared with NY-ESO-1 TCR only. TGF-β1 reduced the T cell–mediated killing of A375 cells by ∼30–40% when T cells expressed NY-ESO-1 TCR alone. However, T cells coexpressing dnTGFβRII and NY-ESO-1 TCR were not inhibited under the same conditions.

In agreement with these data, T cells cocultured with various tumor Ag-expressing cell lines and either 5 or 50 ng/ml TGF-β1 showed reduced GZMB secretion when expressing NY-ESO-1 TCR, while those expressing dnTGFβRII_NY-ESO-1 TCR remained unaffected (Supplemental Fig. 4A).

To further assess the cytotoxic function of these T cells in a more challenging, potentially more physiologically relevant system and over a longer time frame, we used A375 target cells grown as 3D spheroid microtissues (Fig. 7). T cells transduced with the NY-ESO-1 TCR or coexpressing dnTGFβRII were capable of killing A375-GFP spheroids in the absence of exogenous TGF-β1 (Fig. 7A, top panel). The presence of 5 ng/ml exogenous TGF-β had little effect on the ability of T cells expressing TCR alone or dnTGFβRII_TCR to control the growth of the spheroids (Fig. 7A, middle panel). However, with 10 ng/ml TGF-β, expression of dnTGFβRII notably improved the cytolytic function of the T cells (Fig. 7A, bottom panel) with only the T cells coexpressing dnTGFβRII with NY-ESO-1 TCR capable of eliminating all the spheroids by the end of the assay.

This effect was more pronounced when the target cells were preincubated with TGF-β1 during spheroid formation (Fig. 7B), rather than adding TGF-β1 and T cells simultaneously. In the presence of 5 ng/ml TGF-β1 (Fig. 7B, middle panel), the dnTGFβRII_NY-ESO TCR T cells showed slightly increased efficacy against the spheroids, compared with the NY-ESO TCR T cells, although this became more pronounced at the later time point. Again, the differences in cytotoxicity against spheroids were increased in the presence of 10 ng/ml TGF-β (Fig. 7B, bottom panel), particularly at 303 h. At the later time point in the presence of 10 ng/ml TGF-β, the NY-ESO TCR T cells were no longer capable of controlling growth of the spheroids, while the spheroids cocultured with T cells expressing dnTGFβRII_NY-ESO TCR were almost completely destroyed. Similar results were obtained in an independent experiment with T cells from a second donor (data not shown). Because of the 3D nature of the spheroids, reporting the data as the area of the spheroid (as we have done), rather than the volume of the spheroid, is likely to be an underestimate of the true impact of the T cells.

These data were also supported by 3D killing data using the dnTGFβRII_MAGE-A10 TCR (Supplemental Fig. 4B). Most of the A375 spheroids were killed by T cells expressing dnTGFβRII_MAGE-A10 TCR, while T cells expressing the MAGE-A10 TCR were less effective at controlling growth of the spheroids in the presence of 5 ng/ml TGF-β1.

Together, these data, obtained with two different affinity-engineered TCRs and T cells from different donors, show that dnTGFβRII can improve key effector functions of T cells in vitro in the presence of otherwise inhibitory concentrations of TGF-β1.

In this study, we demonstrated that TGF-β expression is elevated in a number of different cancer indications that might be relevant for clinical trials using NY-ESO-1 TCR-transduced cells. We then showed that addition of TGF-β was capable of inhibiting the function of T cells expressing engineered TCR in vitro, while cells engineered to coexpress dnTGFβRII alongside the TCR were, at least partially, resistant to inhibition.

Clinical trials using ACT with T cells transduced with engineered TCRs or chimeric Ag receptors have shown promising clinical responses in hematological cancers (1). However, a need remains to improve the rate, depth, and durability of responses in solid tumors with immune-compromised phenotypes (39). TGF-β, especially the most prevalent isoform, TGF-β1, is a key factor in several immune resistance mechanisms. Moreover, TGF-β overexpression was associated with a lack of response to checkpoint inhibitor therapy across cancer types (40). In this study, we found that mRNA for TGF-β and a panel of TGF-β–induced genes was elevated in a variety of common tumors, indicating the presence of biologically active TGF-β. At the cellular level, T cells were frequently surrounded by TGF-β⁺ cells in NSCLC samples and rarely expressed IFN-γ. Close association of T cells and CD68⁺ TGF-β⁺ cells was recently reported in gastric cancer and appeared to play a role in immune suppression (41).

A variety of inhibitors of TGF-β signaling have been developed and tested, targeting different steps of the canonical pathway and using various modalities (42). The elevated expression of TGF-β in normal tissues, particularly lung (Fig. 1A), where it is known to mediate normal homeostasis and immunosuppressive function, could indicate a potential risk for systemic TGF-β–targeting therapies, while the context-dependent and pleiotropic nature of TGF-β makes it a challenging target for systemic treatments. Specifically, animal studies have raised concerns about long-term adverse effects in the form of de novo appearance of unrelated neoplasms or enhanced outgrowth and metastasis of the primary tumor [reviewed by Akhurst (43)].

More targeted approaches aim to make cancer-specific T cells resistant to suppression by TGF-β (44). One of the most extensively studied is the reduction of endogenous TGFβR signaling through expression of a dnTGFβRII transgene (19–21). Some groups reported that transgenic mice expressing high copy numbers of dnTGFβRII under the control of CD2 or CD4 promoters developed a lymphoma-like proliferation of CD8⁺CD44⁺ T cells with age [reviewed by Oh et al. (45)]. Although the mechanism is not entirely clear, chromosome 15 amplification and overexpression of c-Myc suggested an underlying developmental defect (46), possibly related to the requirement of TGF-β during normal thymic development of CD8⁺ T cells. No cases of lymphoma have been reported from clinical trials using adoptive transfer of differentiated T cells expressing a dnTGFβRII transgene. In contrast, expansion of transduced T cells was temporary, and their persistence in peripheral blood ranged from 2 to 51 mo (24). The authors also reported that the T cells expressing dnTGFβRII were more efficacious than the unmodified T cells (24). Importantly, these results were achieved without preconditioning chemotherapeutic lymphodepletion, a common requirement for ACT at present. Several other trials with dnTGFβRII are either ongoing or have been completed (47–49), but no other clinical data have been published to date.

In this study, we observed that TGF-β1 inhibited in vitro proliferation and effector functions of T cells expressing two different affinity-enhanced TCRs, whereas coexpression of dnTGFβRII overcame some, or all, of these inhibitory effects. To facilitate the use of this strategy in ACTs, we designed a lentiviral vector that allows different genes to be coexpressed alongside TCRs or chimeric Ag receptors of different Ag specificities. Such genes, through improving T cell function, persistence, or resistance to inhibition, may benefit patient populations with different immunological tumor types (13). A current limitation is lentiviral transduction efficiency, which decreases with increasing vector size (38). We used higher concentrations of lentiviral particles to achieve equivalent transduction and saw no difference in the phenotype of T cells generated with the two vector constructs.

We have shown in this article that expression of dnTGFβRII is compatible with engineered affinity-enhanced TCRs of different specificities, using T cells manufactured using a clinically relevant process. Using RT-PCR, we confirmed that the dnTGFβRII was expressed only in the cells that were transduced with the dnTGFβRII_NY-ESO-1 constructs. By flow cytometry, it was not possible to distinguish between the native TGFβRII and the truncated dnTGFβRII version. However, we also demonstrated that transduction with the truncated receptor led to increased cell surface expression of TGFβRII on CD8⁺ and CD4⁺ T cells using both the NY-ESO-1 and MAGE-A10 TCR systems. The reduced SMAD2/3 phosphorylation in response to TGF-β would suggest that either most of the endogenous TGFβRII was replaced at the cell surface by the dnTGFβRII version or that overexpression of the dnTGFβRII was sufficient to outcompete for binding to the TGFβRI. In either case, it is clear that by binding to and inhibiting inactivating signaling through TGFβRI/II within the cell, this acts as a true dominant-negative receptor.

We found that TGF-β1–3 was secreted by a range of tumor cell lines from different tissues of origin. These assays required acidification of the supernatants, which activates TGF-β by release from association with latency-associated peptides, a process that usually occurs through the activity of tumor-derived proteases and reduced pH (37). Acidification allows for detection of total TGF-β but makes determining the true levels of functional TGF-β difficult (37). The lack of inhibition of T cells and corresponding lack of apparent effects of the dnTGFβRII in the absence of exogenous TGF-β suggest that the TGF-β that was secreted by the tumor cell lines was probably not active. We are unaware of any studies describing the concentration of active TGF‐β directly within tumors. Circulating levels of TGF-β in human plasma, as measured by ELISA or receptor-binding assays, vary between 0.5 and 25 ng/ml, with higher concentrations often seen in samples from patients with cancer compared with healthy donors (14, 15). However, determining the concentration of active TGF-β is difficult because it suffers from the same caveat of requiring activation to allow for detection. It is highly likely that TGF‐β concentrations are higher in tumors than in plasma. In this study, we used a range of concentrations between 5 and 50 ng/ml, which are similar to those used by others studying the impact of TGF‐β on T cells (24, 49). We also observed that maximal SMAD2/3 phosphorylation in T cells occurred when T cells were treated with ≥10 ng/ml TGF-β.

TGF-β can also increase the sensitivity of CD8⁺ T cells to other inhibitory factors. For example, TGF-β–induced upregulation of CD73 has been reported to interfere with therapeutic costimulation with agonists for 4-1BB and possibly other TNFR family members (50). Moreover, TGF-β was reported to decrease the efficacy of programmed death-1/PD-L1 blockade by upregulation of programmed death-1 expression on T cells (9). Studies in diverse in vivo tumor models have shown that targeting both PD-L1 and TGF-β simultaneously can be more effective than targeting either pathway alone (51–53). These data emphasize the importance of TGF-β in tumor resistance and the potential for targeting TGF-β therapeutically in combination with other immunotherapeutic agents. High levels of TGF-β within a tumor can also lead to failure of ACT by both T cell suppression and immune cell exclusion (54). The latter will likely require additional treatment strategies to improve homing or access of the T cells to the tumor.

In addition to the results reported in this article, we also performed an extensive package of work to investigate whether addition of the dnTGFβRII moiety might have an adverse impact on the safety profile of T cells transduced with either the NY-ESO-1 or MAGE-A10 TCR, both of which have been well tolerated in clinical trials. In brief, no evidence was found of cross-reactivity, alloreactivity, or increased sensitivity, as measured in a battery of in vitro tests performed as described by Border et al. (31) and Sanderson et al. (55). Importantly, T cells transduced with dnTGFβRII maintained T cell function in the presence of TGF-β without displaying signs of T cell overactivation. Because this approach is designed to “inhibit an inhibitor,” rather than agonistic stimulation, we would expect it to be relatively safe. Therefore, it has the potential to enhance the depth and durability of ACT in a wide range of cancers where TGF-β overexpression is common.

We acknowledge the contribution of many people who contributed to this work either through useful discussions or provision of reagents, including Gwen Binder, Victoria Anderson, Helen Tunbridge, and Thomas Weissensteiner from Adaptimmune, and David Caballero Pradas, David Krull, Paul Fisher, Neil Sheppard, Ian Catchpole, Marie Davis, Aiman Shalabi, Matthew Roberts, and Russ Poe from GlaxoSmithKline. Editorial support for this manuscript was provided by Thomas Weissensteiner (Adaptimmune) and Elevate Scientific Solutions (Envision Pharma Group), which was contracted and compensated by Adaptimmune for these services.

Adaptimmune and GlaxoSmithKline have a strategic collaboration and license agreement. In September 2017, GlaxoSmithKline exercised its option to exclusively license the right to research, develop, and commercialize Adaptimmune’s NY-ESO SPEAR T cell therapy program, and data from the NY-ESO program are included in this publication. GlaxoSmithKline’s funding and involvement in the underlying research and the associated GlaxoSmithKline authors are disclosed in the submitted work. All authors are current (or former) employees of Adaptimmune Ltd. or GlaxoSmithKline and receive (or received) salary and stock options for their work.