TGFβ Programs Central Memory Differentiation in Ex Vivo–Stimulated Human T Cells

The integration of stimulatory signals during T-cell activation programs the differentiation of effector and memory T cells. According to the progressive differentiation model, T cells differentiate depending on the nature and strength of activation signals following a one-way linear path from naïve to early memory (stem cell memory: Tscm; central memory: Tcm), effector memory (Tem), and finally terminally differentiated effector T cells (Teff; ref. 1). Hence, the gradual acquisition of effector features resulting from “strong” activation signals is associated with a decreased potential for long-term memory T-cell generation and persistence. Although challenged by evidence supporting the possibility to revert from effector to long-lived memory T cells (2), a consensus in T-cell–adoptive immunotherapy is to use ex vivo–expanded “early memory” Tcm and Tscm capable of proliferating and persisting in vivo after transfer (3–6). Thus, the use of various cytokine combinations, the alteration of metabolic pathways, and the modulation of signaling cascades involved in T-cell memory or effector fate determination are widely pursued to confer Tcm or Tscm characteristics to ex vivo–manipulated T cells for therapy (3–8).

The cytokine TGFβ has pleiotropic effects in the hematopoietic system (9, 10). Although primarily known for its immunoregulatory and antiproliferative properties, TGFβ orchestrates both regulatory T-cell (Treg) and inflammatory subset differentiation depending on the presence of additional signals (11–14). This pleiotropy is further exemplified by the contrasting prosurvival effects of TGFβ on naïve and memory T cells, and the proapoptotic and functional inhibitory effects on differentiated Teff (15–17). The role of TGFβ in memory T-cell differentiation remains incompletely understood, but given the potential of this cytokine to mitigate T-cell activation signals (18, 19), one may expect that TGFβ exposure during T-cell activation may favor early memory differentiation.

We show herein that exogenous TGFβ exposure during human T-cell stimulation ex vivo favored Tcm differentiation. In the presence of TGFβ, the transcriptional regulator of Teff differentiation, BLIMP-1, was suppressed, and ID3, a master regulator of T-cell memory differentiation, was induced, correlating with an early T-cell memory transcriptional signature (20, 21). The T cells generated in TGFβ-supplemented conditions expanded normally and displayed increased polyfunctional cytokine secretion relative to unexposed T cells in keeping with early memory differentiation (1). The adoptive transfer of activated T cells in immunodeficient mice revealed that TGFβ exposure in culture conferred an enhanced capacity to expand, persist, and mediate xenogeneic graft-versus-host disease (GVHD), as previously reported for cells with Tscm features (3). Similarly, chimeric antigen receptor (CAR)–modified T cells transduced and expanded in TGFβ-supplemented culture were more effective at controlling tumor growth in vivo. Hence, the TGFβ pathway can be used to program early memory differentiation in human T cells and has, therefore, immediate relevance for the field of adoptive immunotherapy.

Peripheral blood mononuclear cells (PBMC) were obtained by venipuncture after informed consent from a total of 39 healthy volunteers in accordance with institutional policies. A total of 1 × 10⁵ T cells (enriched using the Human T cell Enrichment Kit, StemCell Technologies) were stimulated in 96-well U-bottom culture plates with plate-bound anti-CD3e (5 μg/mL; UCHT1; BD Biosciences,) and soluble anti-CD28 (1 μg/mL; CD28.2; BD Biosciences) in T-cell media (Advanced RPMI1640, Gibco; 10% human serum and 1× l-glutamine, Sigma-Aldrich) with Penicillin (100 U/mL)–Streptomycin (100 μg/mL) Solution (Sigma-Aldrich) at 37°C and 5% CO₂. Recombinant human TGFβ (active form) was used at 5 ng/mL (Feldan). IL7 (10 ng/mL), IL15 (5 ng/mL; Miltenyi Biotec), IL2 (100 U/mL; Stem Cell Technologies), and the type I and II TGFβ receptor kinase inhibitor GW788388 (2.5 μmol/L; Selleckchem; ref. 22) were used as indicated. At days 3, 7, and 11, half of the media was replaced with fresh media alone or fresh media containing TGFβ or GW788388.

For pathogen-specific T-cell line generation, 1 × 10⁶ PBMCs from healthy volunteers were cocultured in 24-well flat-bottom culture plates with autologous irradiated (40 Gy) mature dendritic cells (DC) obtained as described in ref. 23 by monocyte isolation (plastic adherence method), culture in IL4- and GM-CSF–supplemented media for 7 days, and maturation for 48 hours following the addition of IL6, IL1β, TNFα, PGE2, and IFNγ. Mature DCs were loaded with a peptide library (1 μg/mL) consisting of 15-mers, overlapping by 11 amino acids and covering the entire Epstein–Barr nuclear antigen-1 (EBNA1) protein sequence (JPT peptides). Antigen-loaded DCs were cocultured with PBMC at a 1:10 ratio (stimulator:responder) in T-cell media supplemented with IL7 (10 ng/mL), IL15 (5 ng/mL), and TGFβ (5 ng/mL) when indicated. Restimulation of T cells with antigen-loaded DCs was performed weekly, and half-media changes were performed twice a week.

To test the specific reactivity of EBNA1-specific T-cell lines, ELISpot assays were performed at the end of the culture (28 days) using the Human IFNγ ELISpot PLUS Kit (HRP) ELISpot Assays (Mabtech, Inc.), and corresponding spot-forming cells and activity per 10⁵ cells were counted using a vSpot Reader Spectrum (AID) according to the manufacturer's instructions. Viability and apoptosis were evaluated by double staining of 10⁵ cells with Annexin V (AV; BD Biosciences) and propidium iodide (PI; 2.5 μg/mL; Invitrogen) after 7 and 14 days of culture with anti-CD3/CD28 stimulation (performed as described above). Staining for AV was performed in AV binding buffer (10 mmol/L HEPES pH 7.4, 140 mmol/L NaCl, 2.5 mmol/L CaCl₂) for 15 minutes at room temperature. PI was added to the cell suspension right before flow cytometry assessment of the percentage of viable (AV⁻PI⁻), apoptotic (AV⁺/PI⁻), and dead (AV⁺PI⁺, AV⁻/PI⁺) cells. Proliferation and viability data were acquired using an LSR II Flow Cytometer (BD Biosciences) and analyzed with DIVA Version 8.7 Software (BD Biosciences). For proliferation assays, 10⁵ T cells were labeled with 1 μmol/L CellTrace Violet (CTV; Invitrogen) prior to the cultures with anti-CD3/CD28 as described previously (24) and according to the manufacturer's instructions. Dye dilution was assessed by flow cytometry as above.

Antibodies targeting the following antigens (name of clone in parenthesis) were purchased from BD Biosciences: CD3 (SK7), CD4 (RPA-T4), CD8 (RPA-T8), CD45RA (5H9), CD45RO (UCHL1), CD95 (DX2), CCR7 (150503), CD62L (DREG-56), CD27 (M-T271), CD28 (CD28.2), CD127 (HIL-7F-M21), IL2 (MQ1-17H12), IFNγ (4S.B3), TNFα (MAb11), and FOXP3 (PCH101). The anti-CD271 (ME204) antibody was purchased from BioLegend. Cell surface staining was performed by incubating up to 10⁶ cells with the antibodies in PBS supplemented with 2% FBS in the dark at 4°C for 30 minutes. When required, intracellular staining was performed in the dark at 4°C for 30 minutes after staining for cell surface antigens, followed by permeabilization using the Cytofix/Cytoperm Buffer (BD Biosciences) as per the manufacturer's instructions. For functional analysis, 10⁵ T cells were restimulated with phorbol 12-myristate 13-acetate (PMA; 50 ng/mL), ionomycin (500 ng/mL; Sigma-Aldrich), and anti-CD28 and anti-CD49d (1 μg/mL; BD Biosciences) in the presence of brefeldin A at 7.5 μg/mL (Sigma-Aldrich) for 4 hours prior to cell surface and intracellular staining. All data were acquired on a LSR II flow cytometer, and sorting was performed using a FACS Aria III Sorter (BD Biosciences). Data were analyzed with DIVA Version 8.7 Software (BD Biosciences), Flow Logic Software (Inivai Technologies), or FlowJo V10 Software (Tree Star).

An equal number of cells (5 × 10⁵) per condition were used. Protein extraction was done in Laemmli Lysis buffer and resolved on 10% SDS-PAGE gels and transferred to polyvinylidene difluoride membranes. Anti-phospho SMAD3 (EP823Y; 1:2,000; Abcam), anti-SMAD3 (EP568Y; 1:2,500; Abcam), anti-ID3 (D16D10; 1:700; Cell Signaling Technology), anti–BLIMP-1 (C14A4; 1:700; Cell Signaling Technology), anti–β-actin (Ab-5; C4/actin; BD Biosciences), and Goat Anti-Rabbit IgG (H + L)-HRP–conjugated secondary antibody (1:3,000; Bio-Rad, catalog no. 1706515) were used. The Amersham ECL^TM Western Blotting Detection Reagent (GE Healthcare) was used for revelation.

T-cell subset qPCR (RT-PCR) analyses were performed as described previously (7). The RNA from sorted cells (FACS ARIAIII, BD Biosciences) was extracted using the RNeasy Micro Kit (Qiagen) according to the manufacturer's instructions. After DNase treatment (Ambion, Life Technologies), 0.5 μg of RNA was reverse transcribed with random primers using the High-Capacity cDNA Reverse Transcription Kit (Life Technologies). ID3 and PRDM1 gene expression was determined by qRT-PCR, in duplicate, using TaqMan Gene Expression Assays (Thermo Fisher Scientific, catalog no. 4331182; ID3 hs00171409, PRDM1 hs00153357), and TaqMan Fast Advanced Master Mix (Life Technologies) performed according to the manufacturer's instructions. The Viia7 qPCR Instrument (Life Technologies) was used to detect amplification. Relative expression (RQ = 2^−ΔΔC_t) was calculated using the Expression Suite Software (Life Technologies), and normalization was done using GAPDH (Thermo Fisher Scientific, catalog no. 4331182; GAPDH hs03929097).

For next-generation sequencing, all reagents and devices were from Thermo Fisher Scientific. Samples were prepared using 35 ng tRNA that was reverse transcribed using the SuperScriptVILO cDNA Synthesis Kit and amplified (Ion Ampliseq Transcriptome Human Gene Expression Core Panel targeting 20812 RefSeq genes). Libraries were generated and barcoded using Ion Xpress Barcode Adapters 1-16 kit, quantified with the Ion Library Quantitation Kit, and loaded on the Ion Chef instrument (Ion PI Hi-Q Chef Kit) for template preparation and loaded on Ion P1 v3 chips. Each sample was sequenced and aligned to the human reference genome (hg19) using the Torrent Suite software v5.0.5. Raw read counts were generated by the ampliSeqRNA plugins. The data were analyzed using Bioconductor packages (http://www.bioconductor.org/) and R statistical language (www.r-project.org). Differentially expressed gene analysis was performed using the DESeq2 package version 1.6.3 with raw read counts from AmpliSeq. The data can be found in the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo, accession number GSE87508). The T-cell–specific demethylation region (TSDR) analysis for FOXP3 promoter methylation status was performed on male T cells as described previously (25). Briefly, after DNA bisulfite conversion (EZ DNA Methylation-Direct Kit, Zymo Research) and PCR (Human FOXP3 Kit, Epigen DX), pyrosequencing was performed on a Biotage PyroMark Q96 MD Pyrosequencer (Qiagen) and analyzed with the Pyro Q-CpG software (Biotage).

Cytokine quantification in 25 μL of undiluted culture supernatants harvested at day 7 of anti-CD3/CD28 stimulation (as described above) was performed as described previously (7) using the Human Th1/Th2/Th17 Magnetic 8-Plex Panel (Life Technologies). Data were acquired on a MagPix Instrument (Bio-Rad).

T-cell–suppression assays used anti-CD3/CD28–stimulated or sorted peripheral blood Treg cells (CD4⁺CD25^highCD127⁻; suppressors) added to a mixed lymphoid reaction where CTV-labeled responder T cells (autologous to cultured cells) were mixed with allogeneic irradiated PBMCs (stimulator) at a 1:1:8 ratio (stimulator:responder:suppressor). After 6 days, the proliferation of responder T cells was analyzed by flow cytometry.

For GVHD induction, 8- to 12-week-old NOD/SCID/IL2Rγ^null (NSG) mice (Jackson Laboratory) received total body irradiation (250 cGy) 1 day prior to intravenous injection of 0.5 × 10⁶ T cells that had been stimulated for 7 days with anti-CD3/CD28 in the presence or absence of TGFβ (5 ng/mL). Human recombinant rhIL15 (1 μg, 2,000 U; Miltenyi Biotec) was administered intraperitoneally every 2 to 3 days for 3 weeks after transplant. Monitoring of circulating human cells was performed weekly by submandibular bleeding, cell surface staining, and flow cytometry as described above. When indicated, cellular suspensions from spleens were obtained by mechanical dissociation of the spleen performed in RPMI media (Gibco).

For evaluation of antitumor responses, second-generation B-cell maturation antigen (BCMA)–specific CAR (described in ref. 26) were transduced by lentivirus on day 2 of CD3/CD28 stimulation (TransAct, Miltenyi Biotec) in the presence of IL7 (10 ng/mL) and IL15 (5 ng/mL) ± TGFβ (5 ng/mL) and expanded for 7 days prior to adoptive transfer or cytotoxicity assays. To evaluate the specific cytotoxicity of BCMA-CART cells treated or not with TGFβ, human BCMA-expressing KMS11 cells were used. The KMS-11 cells were provided by Jonathan Bramson (McMaster University, Hamilton, Ontario, Canada) and were originally obtained from the ATCC and Kelvin Lee (Roswell Park Comprehensive Cancer Institute, Buffalo, NY). Jurkat cells (ATCC) were likewise provided by Jonathan Bramson (McMaster University, Hamilton, Ontario, Canada) and used as non–BCMA-expressing negative controls in our in vitro assays. The cell lines were not further authenticated. The cell lines were engineered with a transgene for enhanced firefly luciferase expression and allowed for puromycin selection (27). Both cell lines were tested for Mycoplasma (PlasmoTest Mycoplasma Detection Kit from InvivoGen) and were used after 7 to 13 passages. Cell lines were cultured in RPMI media (Gibco) supplemented with 10% FBS, 2 mmol/L l-glutamine (Sigma-Aldrich), penicillin (100 U/mL)–streptomycin (100 μg/mL; Sigma-Aldrich), and puromycin (8 μg/mL; InvivoGen) at 37°C and 5% CO₂.

For cell cytotoxicity assays, KMS-11 and Jurkat cells were labeled with either CTV or CellTrace Yellow (CTY; Invitrogen Life Technologies) and plated at equal numbers (2.5 × 10⁴) before coculture with BCMA-CART cells in a U-bottom 96-well plate (Sarstedt) at indicated effector:target ratio for 16 hours at 37 °C. After coculture, tumor cell viability was determined by flow cytometry using Flow Count Beads (Beckman Coulter). Tumor cell viability was calculated as: (100 − (Target cell alive/Target cell alone) × 100).

To assess the in vivo antitumor response of TGFβ-treated CART cells, a total of 2 × 10⁵ CART cells (approximately 2 × 10⁶ total T cells) were injected into NSG mice bearing enhanced firefly luciferase–expressing KMS-11 cells. A total of 10⁶ KMS-11 cells were injected intravenously 7 days prior to adoptive transfer. To delay the occurrence of GVHD, recipient mice did not receive irradiation. Bioluminescent imaging was performed weekly 10 minutes after the intraperitoneal injection of 150 mg/kg of fresh sterile d-Luciferin Solution (PerkinElmer). Dorsal and ventral views were obtained using an IVIS Spectrum (IVIS100 IVIS Lumina System, Caliper LifeSciences). Images were analyzed using Living Image Software version 4.5 for Windows (PerkinElmer). Peripheral blood sampling and flow cytometry assessments were performed weekly (as described above) to monitor CART cell persistence, and recipients were sacrificed at day 35 following tumor inoculation. Mice were maintained in a specific pathogen-free environment. The study was approved in accordance with the Canadian Council on Animal Care guidelines at Hôpital Maisonneuve-Rosemont (Montreal, Quebec, Canada; protocol #2016-FE-002).

Unless otherwise specified, all statistical analyses were paired to best assess the impact of test conditions for every donor and performed using the Wilcoxon signed-rank test or Student t test depending on data distribution (assessed by the Shapiro–Wilk test) using the IBM SPSS statistics 21 software or R statistical language (www.r-project.org). P < 0.05 was considered significant.

Total T cells from healthy donor volunteers were stimulated with plate-bound anti-CD3 and soluble anti-CD28 in the presence or absence of exogenous TGFβ. The expression of CD45RO and CD62L on both CD4⁺ and CD8⁺ T cells was used to assess for central memory (Tcm; CD45RO⁺CD62L⁺) and effector memory (Tem; CD45RO⁺CD62L⁻) differentiation. T cells were stimulated and incubated in parallel with the TGFβ receptor I and II kinase inhibitor GW788388 (22) given previous reports showing that autocrine TGFβ can have important effects on T-cell activation and differentiation (Fig. 1A–C; refs. 12, 28). The addition of TGFβ rapidly induced, whereas GW788388 suppressed, the phosphorylation of the canonical TGFβ mediator SMAD3 (Supplementary Fig. S1). After 7 days in the culture, nearly all cells could be identified as Tcm or Tem (Fig. 1A and B). The percentage of Tcm was significantly increased, and reciprocally, the percentage of Tem was decreased upon exposure to exogenous TGFβ for both CD4⁺ T and CD8⁺ T cells at day 7. The same pattern persisted after 14 days in culture, reaching statistical significance in CD4⁺ T cells. Likewise, CCR7, the other widely used Tcm marker, was expressed by a significantly greater percentage of CD4⁺ and CD8⁺ T cells after TGFβ exposure at both day 7 and day 14 (Fig. 1A and C). Conversely, the inhibition of autocrine TGFβ signaling with GW788388 favored Tem over Tcm accumulation. The impact of TGFβ supplementation or signaling blockade on the expression of other T-cell memory markers (CD27, CD28, and CD127) was also performed (Supplementary Fig. S2). A statistically significant difference was noted for the proportion of CD27-expressing T cells (lower in the TGFβ condition), and a slightly lower proportion of GW388788-treated T cells expressed CD127 and CD28 (CD8⁺ T cells only).

We next assessed whether TGFβ exposure could favor Tcm marker expression on different T-cell populations. To this end, we sorted naïve (CD45RA⁺CD45RO⁻CD62L⁺CCR7⁺CD95⁻), bulk memory (CD45RO⁺CD45RA⁻), as well as Tem (CD45RO⁺CD45RA⁻CD62L⁻CCR7⁻) and Teff (CD45RO⁻CD45RA⁺CD62L⁻CCR7⁻) CD4⁺ and CD8⁺ T cells. For both naïve CD4⁺ and CD8⁺ T cells, TGFβ exposure during stimulation yielded a higher percentage of Tcm relative to the control condition (Fig. 1D and E). The same observation was made following the stimulation of sorted CD4⁺ and CD8⁺ bulk memory T-cell populations (Fig. 1D and E). Autocrine TGFβ signaling blockade had no effect. Sorted Tem mostly kept their phenotype after stimulation irrespective of TGFβ exposure, whereas sorted and stimulated Teff mainly died or reverted to a Tem phenotype (CD8⁺ T cells from 2/4 donors) in both TGFβ-supplemented and control conditions (Supplementary Fig. S3). Globally, these data implied that TGFβ exposure favored Tcm differentiation in human T-cell cultures through effects on naïve and preexisting early memory T cells.

We then assessed whether exogenous TGFβ exposure would likewise confer Tcm features to T cells expanded using naturally occurring antigens. To this end, we expanded EBNA1-specific T cells ex vivo (29). Four weekly T-cell stimulations with autologous DCs loaded with peptide libraries were performed in the presence of cytokines previously shown to favor early memory T-cell differentiation (IL7 and IL15; ref. 3). Supplementation with TGFβ during the first 2 weeks increased the proportion of CD4⁺ and CD8⁺ T cells expressing CCR7 at day 14 by roughly 20% (Supplementary Fig. S4). Despite repeated antigen stimulations in the absence of exogenous TGFβ beyond day 14, CCR7⁺ T cells and CD62L-expressing CD4⁺ T cells were found in higher proportions in T-cell lines previously exposed to TGFβ at day 28. As opposed to T cells stimulated with anti-CD3/CD28, antigen-stimulated T cells in the presence of TGFβ expressed CD27 in a higher proportion relative to unexposed T cells, but no effects were found for CD127 and CD28 expression. Finally, the addition of TGFβ did not limit the generation of antigen-reactive T cells. We conclude that early TGFβ exposure during T-cell stimulation with either anti-CD3/CD28 or antigenic peptides globally favors Tcm marker expression.

It was previously found that TGFβ suppresses BLIMP-1 (encoded by PRDM1), a central regulator of Teff differentiation (30, 31). The transcriptional repressor BLIMP-1 inhibits memory differentiation, notably through TBX21 induction (32) and the repression of ID3, a key memory-associated transcription factor (21). Given the limited effect of autocrine TGFβ signaling inhibition in our system, we focused on the impact of TGFβ exposure on the expression of BLIMP-1 and ID3 in anti-CD3/CD28–stimulated T cells. Early after stimulation (72 hours), qPCR revealed that the TGFβ-exposed cells expressed lower PRDM1 (coding for BLIMP-1) and tended to have higher ID3 transcript levels relative to cells unexposed to TGFβ (Fig. 2A and B). This translated to significant differences in protein levels for both BLIMP-1 and ID3 (Fig. 2C and D). RNA sequencing performed on sorted CD4⁺ and CD8⁺ T cells after 7 days of culture linked this early pattern of BLIMP-1 and ID3 expression with the differential expression of several effector and memory genes between the two experimental conditions (Fig. 2E; refs. 1, 5, 6, 8, 20, 33, 34). Several transcripts associated with effector differentiation were downregulated in TGFβ-exposed cells relative to the reference condition. These included granzymes (GZM), FAS ligand (FASLG), STAT4, SOCS3, ID2, IFNG, and the master transcription factor of Teff differentiation transcripts PRDM1 and TBX21. Conversely, and consistent with Tcm differentiation, TGFβ exposure led to higher expression of the memory-associated transcription factors transcripts ID3, FOXP1, SOX4, and FOXO1. Globally, these results showed that during T-cell activation, TGFβ mitigated BLIMP-1 and associated effector gene expression, and increased ID3 expression, which was associated with the expression of T-cell memory–associated transcripts.

After establishing that TGFβ favored Tcm differentiation at the phenotypic and gene expression levels, we sought to determine whether TGFβ impacted T-cell growth and function. Because TGFβ is known to restrict cellular proliferation and promote apoptosis of Teff (14), we next assessed whether the effects of TGFβ on differentiation marker expression were biased by a restriction in cellular expansion (35). Cell counts and flow cytometry analysis performed at days 3, 7, and 14 revealed that TGFβ did not significantly affect cell accumulation or the proportion of CD4⁺ and CD8⁺ T cells (Fig. 3A and B). Dye dilution proliferation assays and AV/PI staining showed no significant differences in proliferation or apoptosis rates between experimental groups (Fig. 3C–E). Likewise, the addition of TGFβ did not compromise T-cell expansion relative to other commonly used cytokines for human T-cell stimulation/expansion (IL2, IL7, and IL15) at days 7 and 14 of culture (Supplementary Fig. S5), and the effects of TGFβ on Tcm proportions were preserved when combined with these cytokines. Thus, the impact of TGFβ on Tcm differentiation was largely independent of T-cell expansion, survival, or the presence of other cytokines.

We next aimed to determine whether TGFβ exposure during T-cell stimulation impacted functionality, as TGFβ can inhibit the secretion of IFNγ, TNFα, and IL2 (33, 36). However, Tcm are known to secrete all three cytokines, with a skewing toward monofunctional IFNγ secretion upon further effector differentiation (1). We assessed the impact of TGFβ exposure during T-cell stimulation on cytokine production by intracellular flow cytometry after briefly exposing days 7 and 14 T cells to PMA/ionomycin (Fig. 4A). To correct for interdonor variability, the percentage of cytokine-producing cells was arbitrarily set at 1 in the reference condition (T cells stimulated in the absence of TGFβ; Fig. 4A and B). The percentage of IFNγ-, TNFα-, and IL2-producing T cells was not significantly different between TGFβ-exposed and unexposed T cells at day 7. However, at day 14, a higher fraction of TGFβ-exposed CD8⁺ T cells were producing TNFα and IL2, but a lower percentage produced IFNγ relative to control. When we assessed for polyfunctionality, a higher percentage of TGFβ-exposed CD4⁺ and CD8⁺ T cells expressed all three cytokines at day 14 (Fig. 4C and D). Taken together, our results showed that TGFβ led to Tcm-associated cytokine secretion.

TGFβ has pleiotropic effects on T-cell differentiation, which may interfere with the potency of T-cell immunotherapies, notably through the induction of Tregs (37). Despite our data showing Th1/Tc1 cytokine secretion following activation in the presence of TGFβ, as well as previous work showing that TGFβ alone does not induce the Treg fate in human T cells (38, 39), we sought to rule out Treg skewing following T-cell activation in the presence of TGFβ. Our gene expression data confirmed that FOXP3, along with the other Treg gene transcripts, IKZF2 (coding for Helios) and IKZF4 (coding for Eos), were upregulated in T cells exposed to TGFβ relative to the control condition (Fig. 5A). Although, FOXP3 is expressed in both activated conventional T cells and Tregs in humans (40), it was imperative to show that TGFβ did not induce suppressive cells. A majority of anti-CD3/CD28–stimulated T cells expressed FOXP3 at the protein level after 7 days of culture, followed by a decline at day 14, in both control and TGFβ-supplemented cultures (Fig. 5B). TGFβ exposure alone (or with IL2) also did not lead to demethylation of the FOXP3 promoter, which is a hallmark of stable Treg differentiation (Fig. 5C). Suppression assays showed that the addition of activated T cells from both control and TGFβ-supplemented cultures at day 7 did not suppress but rather enhanced the proliferative response of autologous T cells (labeled) mixed with irradiated allogeneic targets. This contrasted with conditions where either unstimulated T cells or sorted Tregs (CD4⁺CD25⁺CD127⁻) from the same donors were added to the culture (Fig. 5D and E). IL10 was undetectable in culture supernatants, further arguing against Treg skewing (Fig. 5F). Finally, the cytokine arrays also failed to demonstrate significant or differential IL17 or IL9 production in TGFβ-supplemented cultures, respectively, indicating that the TGFβ-dependent Th17 and Th9 T-cell subsets are not expanded in our cultures (14). Our results confirmed that the addition of TGFβ during the stimulation of T cells does not induce alternative differentiation schemes in our system.

Early memory T cells have a greater capacity to expand, persist, and retain functionality following injection in immunodeficient mice (3, 4). To assess whether the observed Tcm bias conferred by TGFβ in vitro would translate into functional early memory T-cell features in vivo, T cells (5 × 10⁵) were harvested at day 7 of anti-CD3/CD28 stimulation in the absence or presence of TGFβ and injected intravenously into NSG mice. Polyclonal early memory T cells are predicted to cause severe xenogeneic GVHD as opposed to their more differentiated counterparts (3). Weekly assessments of peripheral blood and spleen assessments 4 weeks after adoptive transfer showed that TGFβ-conditioned T cells persisted at higher frequencies (Fig. 6A–C). This translated to weight loss (Fig. 6D) and signs of GVHD (ruffled fur, prostration) in the group that received TGFβ-exposed cells, but not in the control group, confirming that TGFβ-exposed cells had retained high functional xeno-reactive capacity. Hence, TGFβ exposure during the early stages of T-cell activation programmed T-cell persistence, differentiation, and function up to at least 1 month after infusion, as expected for early memory cells (3, 41). These data corroborated the in vitro phenotypic, functional, and gene expression characteristics of ex vivo–stimulated human T cells, linking TGFβ to early T-cell memory differentiation.

On the basis of our GVHD data, where a brief exposure to TGFβ during T-cell stimulation in vitro led to different outcomes several weeks after adoptive transfer, we evaluated whether the Tcm-promoting properties of TGFβ could be leveraged therapeutically. To this end, we used a myeloma xenograft model, where BCMA-expressing KMS-11 cells genetically engineered to express luciferase were injected intravenously into NSG mice 1 week prior to the adoptive transfer of human T cells expressing a BCMA-targeting CAR (BCMA-CART). We used a combination of anti-CD3/CD28 stimulation, IL7, IL15, and TGFβ (or not, in the reference condition) for transduction based on previous data highlighting the potential of IL7/IL15 combination to generate early memory T cells in this setting (42). Exogenous TGFβ enhanced the proportions of Tcm generated in the context of combined IL7 and IL15 supplementation, and did not affect the BCMA-CAR transduction rates (Supplementary Fig. S6). After transduction and a 7-day expansion, cytotoxicity was assessed in vitro using BCMA-expressing KMS-11 cells and Jurkat cells as BCMA-negative controls (Fig. 7A). Differentially labeled targets and control cells were coincubated for 16 hours with BCMA-CART effectors at different ratios. The BCMA-specific killing of target cells was robust and equivalent whether the effectors had been exposed to TGFβ or not during transduction and expansion (Fig. 7B), thereby confirming that a short exposure to TGFβ did not impede the acquisition of effector functions. Adoptive transfer of BCMA-CARTs in KMS-11 tumor-bearing NSGs revealed that BCMA-CARTs exposed to TGFβ in vitro prior to transfer outcompeted their unexposed counterparts in terms of tumor control or eradication (Fig. 7C and D; Supplementary Fig. S7 for dorsal views). Weekly peripheral blood assessments starting at day 14 after injection showed that TGFβ-exposed BCMA-CARTs were always found in greater numbers relative to unexposed BCMA-CARTs (statistically significant at day 14 after transfer; Fig. 7E). These data further support that ex vivo TGFβ exposure programmed T-cell fates associated with high functionality in the context of adoptive immunotherapy.

In the hematopoietic system, TGFβ is a quiescence factor that promotes stem cell as well as memory T-cell maintenance and naïve T-cell survival (10, 15, 37). TGFβ has been shown to dampen signaling downstream of the T-cell receptor (18, 43) and CD28 (44). We found that TGFβ increased the expression of early-memory markers in activated human T cells in multiple culture settings and improved polyfunctional cytokine production without altering T-cell expansion or leading to Treg generation. Our data showed that a brief TGFβ exposure could program a durable effect on T cells in vivo, as predicted for early memory T cells. In the two models used, TGFβ-exposed T cells ex vivo had a significant impact on so-called “hard” endpoints such as death from GVHD or antitumor responses. Globally, these results support that the gain in Tcm brought by TGFβ exposure is sufficient to mediate clinically meaningful effects.

Our results are in line with other reports establishing a role for TGFβ in early memory T-cell differentiation and maintenance in mice (15, 45). In the field of human adoptive immunotherapy, a previous study done with tumor-infiltrating lymphocytes (TIL) provides insights that overlap with some of our conclusions (46). In this context, exogenous TGFβ added to a rapid expansion procedure (REP) using feeder cells, anti-CD3 stimulation (OKT3), and IL2; enhanced functional antigen-specific CD8⁺ T cells; and prevented terminal CD8⁺ T-cell effector differentiation without leading to Treg expansion. Given that TILs are antigen-experienced T cells presenting evidence of exhaustion prior to expansion (47), which contrasts with the steady-state peripheral T cells that we used, this previous work supports that the memory-promoting effects of TGFβ may be extended to several adoptive immunotherapy approaches. However, our results contrast with another study primarily done in mice that reported that endogenous TGFβ signaling blockade in CD8⁺ T cells favored the expansion of Tcm cells relative to other subtypes (28). This is the reason why we tested paracrine TGFβ signaling blockade initially, which in contrast to the previous study, had opposite and marginal Tem-promoting effects. Such discrepancy may pertain to differences in the inhibitor used and/or experimental conditions. We used enriched T cells (vs. whole PBMCs) and combined anti-CD3/CD28 stimulation (vs. anti-CD3 alone), notably to reproduce the type of stimulation used in the CAR field. Our results following repeated antigenic peptide–loaded DC stimulations or culture using cytokine combinations corroborate the Tcm-promoting effects of TGFβ in several experimental conditions. Nevertheless, our results may not apply to adoptive immunotherapy models in mice, and optimizations may be required (concentration, timing of exposure, combination with other cytokines) before TGFβ supplementation is used in clinical scale ex vivo T-cell manufacturing.

Our gene expression studies, along with the suppression of BLIMP-1 and overexpression of ID3 following TGFβ exposure, are consistent with previous data contrasting effector versus memory signatures. These results linking TGFβ to the molecular underpinnings of early memory differentiation in human T cells extend our understanding of TGFβ and T-cell biology but raise several questions. Future studies will be required to determine whether TGFβ mostly acts indirectly by mitigating T-cell activation signals or whether a more direct relationship exists between canonical or noncanonical TGFβ signaling and BLIMP-1 suppression (18, 19, 30). We identified that two memory markers can be downregulated by TGFβ (modestly for CD27 and more robustly for KLF2). Although this had no functional consequences in our assays, it suggests that TGFβ may be explored to confer specific properties to T cells. Notably, the previously described direct effect of TGFβ on KLF2 expression during tissue-resident memory T-cell differentiation may be harnessed for immunotherapy (14, 48).

There are significant implications of our findings for the field of T-cell immunotherapy. The programming of early memory T cells is desirable to ensure robust in vivo expansion, generation of Teff, and self-renewal as early memory T cells leading to persistence of the infused cells (41). Likewise, the avoidance of Treg generation is paramount to the success of T-cell therapies. Our results showed that the supplementation of TGFβ along with anti-CD3/CD28 stimulation could achieve this balance. Our results also support that TGFβ operates according to the progressive T-cell differentiation model, whereby ex vivo–expanded early memory T cells originate from a pool of T cells that have not reached a Tem or Teff differentiation state. In addition to the targeted T-cell subsets preferentially leading to Tcm accumulation following TGFβ exposure, the activation of TGFβ signaling needs to be carefully timed. Although TGFβ can be used to program early memory T cells during T-cell activation, it could also impede Teff function and survival within infected or neoplastic microenvironments (16, 17, 33, 49, 50). Our data support the notion that the TGFβ-dependent acquisition of Tcm features during T-cell stimulation does not prevent the generation of potent T-cell responses in vitro or after transfer in NSG mice. A brief activation of the TGFβ pathway in T cells in vitro prior to adoptive transfer could harness the promemory properties of this master cytokine without compromising the therapeutic potential of the infused T cells.

J.-S. Delisle reports receiving a commercial research grant from and has ownership interest (including stocks and patents) in SpecificiT Pharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Dahmani, J.-S. Delisle

Development of methodology: A. Dahmani, K. Bezverbnaya

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Dahmani, C. Carli, C. Lamarche, M. Khalili, M. Goupil, J.L. Bramson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Dahmani, V. Janelle, C. Carli, J.-S. Delisle

Writing, review, and/or revision of the manuscript: A. Dahmani, V. Janelle, C. Carli, C. Lamarche, K. Bezverbnaya, J.L. Bramson, J.-S. Delisle

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Dahmani, V. Janelle, C. Carli, M. Richaud

Study supervision: A. Dahmani, J.-S. Delisle

The authors are grateful to the blood donors, the animal care facility personnel, Denis-Claude Roy and Vibuthi Dave for access to NSG mice, Martine Dupuis for expert flow cytometry support, Manuel Buscarlet and Lambert Busque for next-generation sequencing, Sébastien Lemieux for advice regarding the bioinformatics analysis, Jana Gillies and Megan Levings for FOXP3 promoter methylation studies, as well as Nathalie Labrecque, Claude Perreault, Sylvie Lesage, and Heather Melichar for helpful discussions and revision of the manuscript. J.-S. Delisle is a Fonds de recherche du Québec – Santé (FRQS) scholar, a Cole Foundation Early Career Transition Award laureate and a ThéCell network member, as well as a Canadian Donation and Transplant Research Program member. This work was supported by the National Science and Engineering Research Council (NSERC) of Canada through a Discovery Grant (418607-2012-RGPIN) and by a Leukemia/Lymphoma Society of Canada (LLSC) operating grant (#202379), both held by J.-S. Delisle.

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