Comprehensive Approach for Identifying the T Cell Subset Origin of CD3 and CD28 Antibody–Activated Chimeric Antigen Receptor–Modified T Cells

Adoptive T cell immunotherapy with chimeric Ag receptor (CAR)-modified T cells targeting tumor Ags has been incorporated into cancer treatment with promising efficacy in distinct settings (1–4). CARs are genetically engineered immunoreceptors comprising a single-chain variable fragment linked to cytosolic endodomains from costimulatory receptors and/or the TCRζ chain (5–7). The structure of the CAR, including the affinity of the single-chain variable fragment, the type of spacer and costimulatory endodomains, the design of the clinical protocol, and the disease targeted profoundly affect the fate and function of CAR-modified T cells, as does the manufacturing protocol that determines the character of the T cell product infused (2–5, 8–21). Data regarding the best T cell subset from which to derive CAR-modified T cells for infusion are inconclusive and controversial, and most patients receive CD4⁺ and CD8⁺ T cells whose subset derivation is unknown (2–5, 11–19). The ultimate objective of T cell therapy is to transfer a long-lived T cell population with the capacity to differentiate into potent tumor-specific effectors and to self renew (8, 22). Short-lived effector T cells (T_EFFs) possess potent effector function in vitro; however, they appear to be less attractive for adoptive immunotherapy because of their limited proliferation and engraftment in vivo (23–25). Memory T cells subsets have been shown to expand substantially in vivo and are long-lived, with their self-renewal capacity being inversely proportional to their differentiation state (26). Recently, it has been reported that Ag-experienced memory T cell subsets directly promote the phenotypic and functional differentiation of naive T cells (T_naives), which, as a consequence, lost antitumor potential when transferred in vivo (27).

Expression of the lymph node–homing molecules CCR7 and the leukocyte common Ag (CD45) isoforms RA and RO distinguishes memory T cells from T_naives and allows the dissection of the memory/effector T cell compartment into at least four main subsets (28, 29): memory stem T cells (T_SCMs), central memory T cells (T_CMs), effector memory T cells (T_EMs), and terminally differentiated T_EFFs (T_EMRAs) (22, 28, 29). T_CMs coexpress CCR7 and CD45RO, having lost CD45RA during naive → memory transition (30). Upon antigenic restimulation, T_CMs lose CCR7 expression and differentiate into T_EMs (30, 31) and finally into T_EMRAs, which are considered to be terminally differentiated. T_EMRAs lack CCR7 and CD45RO and re-express CD45RA (32). A fourth memory subset, T_SCMs, resides phenotypically within the naive-like T cell compartment (CD45RO⁻CD45RA⁺CCR7⁺) and can be distinguished from T_naives by their expression of CD95 (Fas) (22, 29). Each T cell subset has distinct engraftment capacities and function following adoptive transfer in preclinical trials (29–31, 33). In particular, T_CMs are thought to have superior engraftment and persistence compared with more differentiated memory T cell subsets (22, 26, 28–31, 33–37). The recently described T_SCM subset may represent the earliest stage of memory T cell differentiation, and it may have the ability to transfer stem cell–like T cells for improved long-term efficacy (38, 39).

To identify the characteristics and subset derivation of CAR-modified T cells polyclonally expanded on CD3 and CD28 Ab-coated plates, as used in our clinical studies (2–5, 11–19), we sorted each T cell subset and followed its fate and function after activation, CAR transduction, and culture alone and after reconstitution into the corresponding subset-depleted polyclonally activated bulk PBMCs. In a proof-of-concept study, we demonstrate that each T cell subset is sensitive to CAR transduction, and each displays a specific functional profile. T_naive-derived populations showed the most rapid expansion and dominated the cultures by the end of the culture period, but they had reduced effector functions and killing compared with memory subsets. Furthermore, T_naive-derived cells cultured in the presence of memory T cells differentiate more than when cultured alone and show coincidentally reduced effector cytokine production and ability to proliferate in response to CAR stimulation. T_SCMs show the most rapid expansion of all subsets; however, because of their low frequency at the start of culture, they represented only a minor fraction at the end. Finally, we found that, compared with IL-2, a combination of IL-7 and IL-15 increased the yield of T_SCM- and T_CM-derived T cells within the bulk cultures. Irrespective of the protocol used for expansion, our comprehensive approach reveals the characteristics of CAR-modified T cells polyclonally expanded from PBMCs, demonstrates the functional properties of each expanded T cell subset, and underscores the importance of culture conditions to influence the desired T cell populations. To this end, our strategy allows a meticulous assessment of the effects of manufacturing changes on the subset contribution to in vitro–expanded T cells.

All PBMC and T cell cultures and all assays were performed using complete media (45% RPMI 1640 [HyClone, Logan, UT], 45% Click’s Medium [Irvine Scientific, CA]) supplemented with 5 mM l-glutamine (Invitrogen, Carlsbad, CA), penicillin (100 IU/ml), and streptomycin (HyClone), containing 10% FBS (HyClone), in humidified incubators at 37°C and 5% CO₂.

For all experiments, blood samples were collected with informed consent from healthy volunteers using protocols approved by the Baylor College of Medicine Institutional Review Board. PBMCs were isolated from blood by Lymphoprep density gradient centrifugation (STEMCELL Technologies, Vancouver, BC, Canada).

PBMCs were enriched for CD3⁺ T cells with the Pan T Cell Isolation Kit II (Miltenyi Biotec), following the manufacturer’s instructions. Cells were labeled with fluorescent Abs to CD3, CD45RO, CD95, CCR7 (all purchased from BioLegend), and CD45RA (Beckman Coulter). FACS was performed for T_CMs (CCR7⁺CD45RA⁻), T_EMs (CCR7⁻CD45RA⁻), T_EMRAs (CCR7⁻CD45RA⁺), T_SCMs (CD45RA⁺CCR7⁺ → CD95⁺), and T_naives (CD45RA⁺CCR7⁺ → CD95⁻) (22, 28, 40) on a BD FACSAria II SORP (BD Biosciences). Postsorting analysis of purified subsets revealed >98% purity.

NIH 293T, JEKO, and HDLM-2 cell lines (American Type Culture Collection, Manassas, VA) were cultured in IMDM (BioWhittaker, Walkersville, MD) supplemented with 10% FCS (HyClone) and 5 mM l-glutamine (Thermo Fisher, Waltham, MA). Second-generation GD2.CAR (14g2a.CD28ζ CAR)–transduced T cells were directed to the disialoganglioside (GD2) to assess specific in vitro re-expansion and effector function. Expanded second-generation CD19.CAR⁺ (FMC63.CD28ζ CAR) and CD19.CAR-NGFR⁺ (FMC63.CD28ζ-I-NGFR CAR) T cells were analyzed for their ability to produce cytokines in response to CAR Ag-expressing target cells, which constituted the JEKO (CD19⁺) and HDLM-2 (CD19⁻) cell lines.

We used CAR constructs directed to the disialoganglioside, GD2, and the B cell marker CD19 that have been targeted in many clinical studies. Second-generation GD2.CAR (14g2a.CD28-ζ) and the CD19.CAR (FMC63.CD28ζ) cloned into SFG vectors were described previously (41–43). The CD19.CAR construct was used alone and/or cloned upstream of an internal ribosome entry site (IRES)–nerve growth factor receptor (NGFR) or IRES mOrange to facilitate selection and detection.

Transient retroviral supernatants were produced by cotransfection of NIH 293T cells with the MoMLV gag/pol expression plasmid PeqPam3(-env), the RD114(-env) expression plasmid RDF, and SFG vectors at a ratio of 3:2:3, with a total of 10 μg of DNA. The transfection was facilitated with GeneJuice reagent (Calbiochem). The supernatant was harvested 2 and 3 d after transfection, filtered (using a 0.45-μm filter), snap-frozen, and stored at −80°C in 5-ml aliquots.

PBMCs or FACS-sorted T cells were activated on CD3 (from the OKT3 hybridoma; ATCC CRL-8001) and CD28 (BD, Franklin Lakes, NJ) Ab-coated plates at 500 μl of 1 μg/ml each in the presence of recombinant human (rh)IL-2 (50 IU/ml) or rhIL-7 and rhIL-15 (each at 10 ng/ml). In some experiments, CD3/28 Ab–activated subsets were transduced on day 2 after activation and then reconstituted into CD3/28-activated T cell subset–depleted PBMCs on day 3 of culture at their initial frequency or cultured separately. The medium and cytokines were changed every 3 d during culture or when passaging the T cells for splitting during expansion.

T cells or T cell subsets were transduced 2 d after activation: nontissue-treated 24-well plates were coated with 500 μl of 1 mg/ml retronectin (Takara Biochemicals, Shiga, Japan) overnight. After removing the retronectin, 1.5–2 ml of retroviral supernatant was added per well and centrifuged at 2000 × g for 1 h at 25°C to allow vector adherence. The supernatant was removed prior to the addition of CD3/28-activated peripheral blood–derived T cells or FACS-sorted T cell subsets. Cells were incubated for 2 d on the virus-coated plate in the presence of 50 IU/ml rhIL-2 (Proleukin; Chiron, Emeryville, CA) or 10 ng/ml rhIL-7 and 10 ng/ml rhIL-15 (PeproTech) and then transferred to tissue culture–treated plates. Where indicated, 1 d after transduction, each T cell subset was reconstituted at its initial frequency into the corresponding subset-depleted population.

Following Ag specific stimulation, cytokine production (IFN-γ, TNF-α, IL-2) was determined by intracellular fluorescence staining. All Abs were purchased from BioLegend, unless indicated otherwise. GD2⁺ JF cell lines (serving as stimulator cells for second-generation GD2.CAR T cells) and CD19⁺ JEKO cell lines (serving as stimulator cells for second-generation CD19.CAR T cells) were added to T cells at a 10:1 stimulator/responder ratio. HDLM-2 cells lacking GD2 and CD19 were used as controls, and respective background responses have been subtracted from CAR-specific cytokine production. For effector cytokine detection, cultured T cells were restimulated for 6 h in the presence of 1 μg/ml brefeldin A (Sigma-Aldrich). After harvesting, phenotype staining was performed using mAbs for CD3 (OKT3), CD4 (SK3), and CD8 (RPA-T8). To define memory subsets, T cells were extracellularly stained for CCR7 (G043H7), CD45RA (HI100), CD45RO (UCHL1), and CD95 (DX2). In particular experiments, an anti-NGFR (C40-1457; BD) Ab was used to stain for NGFR expression on CD19.CAR-NGFR⁺ T cells. Transduction efficacies were assessed by staining for CAR expression on the T cell surface using Abs targeting GD2 (IA7) (44) or CD19.CAR [goat anti-human IgG (H+L); Jackson ImmunoResearch, West Grove, PA]. To exclude dead cells, live/dead-discrimination staining dye (Invitrogen) was added. Subsequently, cells were permeabilized with Permeabilizing Solution 2 (BD Biosciences) and stained for IFN-γ (4S.B3), TNF-α (MAb11), and IL-2 (MQ1-17H12). Cells were analyzed on a LSR II flow cytometer using FlowJo version 10 software (TreeStar). Lymphocytes were gated based on the forward scatter (FSC) versus the side scatter profile and subsequently gated on FSC (height) versus FSC to exclude doublets.

A modified VITAL assay was used for cytotoxicity testing, as described previously (22, 45, 46). Briefly, the JEKO cell line served as a CD19⁺ target cell for second-generation CD19.CAR T cells, whereas HDLM-2 cells that lack CD19 were used as control targets. JEKO cells were labeled with 10 μM CFSE (Molecular Probes). As controls, CD19⁻ HDLM-2 cells were labeled with 5 μM dimethyldodecylamine oxide-succinimidyl ester (Invitrogen). Cells were cocultured in duplicate for 16 h at T cell/target cell ratios of 1:1 and 10:1. Labeled cells were analyzed as duplicates using an LSR II flow cytometer. Samples without T cells, containing only target cells (CD19⁺ JEKO cell lines or CD19⁻ HDLM-2 cells), served as internal controls. Cells were gated using live/dead discrimination staining–dye negative cells (Invitrogen). The mean percentage of survival of CD19⁺ JEKO target cells was calculated relative to CD19⁻ HDLM-2 controls. Percentage target cell lysis was calculated as follows: mean percentage survival of targets in cultures containing defined numbers of T_EFFs in comparison with control cells without T cells.

Statistical analysis was performed with GraphPad Prism version 6. Data were analyzed using repeated-measures one-way ANOVA, two-way ANOVA, or a paired t test, as indicated, after verifying Gaussian distribution with the Kolmogorov–Smirnov test. The p values ≤ 0.05 were considered significant: *p < 0.05, **p < 0.01, ***p < 0.001. T cell subset fold expansion expresses T cell subset numbers (assessed by FACS) in relation to total cell numbers on the indicated days after initial stimulation and day 1.

To analyze the characteristics of each T cell subset transduced with a GD2.CAR, we compared their proliferation and phenotype following CD3/28 Ab activation (in the presence of IL-7 and IL-15) and cytokine production following stimulation via the CAR using the CAR Ag (disialoganglioside: GD2). First, CD3⁺ T cells were enriched from PBMCs by magnetic bead sorting and T_CMs (CCR7⁺CD45RA⁻), T_EMs (CCR7⁻CD45RA⁻), T_EMRAs (CCR7⁻CD45RA⁺), T_SCMs (CD45RA⁺CCR7⁺ → CD95⁺), and T_naives (CD45RA⁺CCR7⁺ → CD95⁻) were isolated by polychromatic FACS (Fig. 1A). The average frequency of each T cell subset from presorted CD3⁺ T cells was 21.5 ± 6.2%, 17.8 ± 8.8%, 9.8 ± 5.5%, 0.9 ± 0.2%, and 50.5 ± 9.2%, respectively (Fig. 1B). To determine whether all T cell subsets isolated from PBMCs were sensitive to retroviral transduction, we transduced each subset with a second-generation CAR (14g2a.CD28-ζ) targeting GD2 (41) on day 2 after stimulation.

As presented in Fig. 1C, GD2.CAR could be expressed in all T cell subsets, with median transduction efficacies of 60.2 ± 10.2%, 60.1 ± 7.2%, 66.6 ± 7.9%, 61.2 ± 7.1%, and 64.6 ± 6.1% in CD4⁺ PBMC-derived T cells (T_PBMCs), T_CMs, T_EMs, T_EMRAs, and T_naives, respectively, and 65.6 ± 20.7%, 70.8 ± 15.4%, 75.3 ± 12.3%, 61.7 ± 16.8%, 66.6 ± 19.2%, and 82.7 ± 7.9% in CD8⁺ T_PBMCs, T_CMs, T_EMs, T_EMRAs, T_SCMs, and T_naives, respectively (summarized in Fig. 2A). We observed a trend toward higher transduction efficacies in CD8⁺ T cells compared with CD4⁺ T cells (70.4 and 62.5%, respectively) (Fig. 2A). As shown in Fig. 2B, we detected a preferential expansion of CD8⁺ T cells compared with CD4⁺ T cells within T_PBMC, T_CM, T_EM, T_EMRA, and T_SCM, but not T_naive, populations. As reported previously for T_SCM-derived cells (22), we observed almost exclusive proliferation of CD8⁺ T_SCMs, whereas CD4⁺ T_SCM expansion could not be detected. However, all other expanded T cell subsets showed a balanced CD4⁺/CD8⁺ ratio (Fig. 2A, 2B). When cultured individually, all T cell subsets proliferated in response to CD3/28 stimulation, confirming their viability after separation, and all transduced T cell subsets showed rapid expansion, with the highest expansion at day 14 from T_SCM-derived cells (311 ± 63-fold), followed by T_naive-derived cells (210 ± 24-fold) and T_CM-derived cells (150 ± 65-fold) and lower expansion of T_EM-derived cells (88 ± 31-fold) and T_EMRA-derived cells (61 ± 22-fold) (Fig. 2C). All transduced subsets differentiated during culture, so that by day 14, CCR7⁻CD45RO⁺ T_EMs predominated in T_EMRA- and T_EM-derived subsets, whereas a substantial proportion of CCR7⁺CD45RO⁺ T_CMs was present in T_CM-, T_naive- and T_SCM-derived cultures (Fig. 2D), demonstrating that T_naives develop a memory-like phenotype in response to CD3/28 stimulation in vitro.

To examine CAR-specific effector functions, we stimulated each transduced subset on day 14 with the GD2⁺ JF neuroblastoma cell line and a GD2⁻ Hodgkin lymphoma cell line (HDLM2) and measured IFN-γ cytokine expression in intracellular cytokine assays (Fig. 2E). GD2-specific IFN-γ production from CD4⁺ and CD8⁺ T_CM-, T_PBMC-, and T_SCM-derived cells was greater than from T_naive-, T_EMRA-, and T_EM-derived cells. No IFN-γ was produced by nontransduced T cells in response to JF cells (data not shown). To compare T cell subset proliferation in response to CAR stimulation, each transduced subset was stimulated with plate-bound GD2 (used to avoid tumor-derived activating or inhibitory signals) in the absence of exogenous cytokines on day 14 of culture. Intriguingly, T_EMRA- and T_naive-derived cells failed to proliferate, whereas T_SCM-, T_CM-, and T_EM-derived cells showed robust re-expansion in response to CAR Ag over 7 d of culture (Fig. 2F). However, T_naive-derived cells did proliferate (18-fold expansion) when IL-7 and IL-15 were added to the re-expansion culture (data not shown), indicating greater dependence on exogenous cytokines than memory-derived T cells.

In summary, all tested T cell subsets were susceptible to CAR transduction and expanded when activated via the TCR, with greatest fold expansion by T_SCM-derived > T_naive-derived > T_CM-derived > T_EM-derived > T_EMRA-derived cells. Although lower in T_EM- and T_EMRA-derived cells, all subsets produced IFN-γ following CAR stimulation and acquired a memory T cell phenotypic profile. T_SCM-, T_CM-, and T_EM-derived cells showed a greater capacity to expand in response to CAR stimulation than did T_EMRA- and T_naive-derived cells.

To evaluate the characteristics of each T cell subset when reconstituted into the bulk subset-depleted PBMC population, each T cell subset was purified from PBMCs by FACS (Fig. 3A, 3B), activated with CD3 and CD28 Abs, and transduced with a retroviral vector encoding a second-generation CD19.CD28ζ CAR and a truncated NGFR (CD19.CAR-NGFR), separated by an IRES to allow flow cytometric detection following expansion. In parallel, each subset-depleted PBMC population was activated with CD3/28 Abs and transduced with the same second-generation CD19.CAR, but without NGFR (Fig. 3C, 3D). One day after transduction, each T cell subset was reconstituted at its initial frequency into the corresponding subset-depleted population. This design allowed us to distinguish the depleted bulk-transduced PBMCs from the reconstituted transduced T cell subset by flow cytometric detection of NGFR (Fig. 3D). All T cell subsets were sorted to >98% purity (Fig. 3B, 3C) and transduced with frequencies > 80% (Fig. 3E). CD19.CAR transduction did not affect the survival or expansion of any T cell subset compared with nontransduced subsets cultured alone in parallel (data not shown).

Differences in the rate of proliferation of each transduced subset, cultured alone or within the transduced bulk population, could not be detected (Fig. 4A). However, the rate of expansion of T cell subsets within the bulk culture was much greater for T_naive-derived cells (CD3⁺ 234 ± 36-fold expansion) and T_SCM-derived cells (CD3⁺ 621 ± 189-fold expansion) than for T_EMRA- and T_EM-derived cells (CD3⁺ 44 ± 6-fold and 54 ± 9-fold expansion), whereas T_CM-derived cells showed an intermediate rate of expansion (CD3⁺ 204 ± 48-fold expansion) (Fig. 4A). By day 14, T_naive-derived cells dominated the cultures (76 ± 19% of CD3⁺ T cells), with T_CM-derived cells accounting for 30 ± 12% of CD3⁺ T cells (Fig. 4B, 4C). CD4⁺ and CD8⁺ T_EMRA- and T_EM-derived cells could barely be detected because of the other more rapidly expanding subsets (Fig. 4B, 4C). Hence, bulk CD3/28-activated T cells were derived almost entirely from naive and T_CM CD4⁺ and CD8⁺ T cells, with little or no contribution from T_EM and T_EMRA subsets (Fig. 4B, 4C). Of note, despite their high rate of expansion when cultured in bulk PBMCs, T_SCM-derived cells represented <2% of the bulk cultures on day 14 because of their low starting frequency.

By day 14 of culture, T cells with a T_EM phenotype (CCR7⁻CD45RO⁺) dominated all subset-derived cultures (Supplemental Fig. 1). However, a substantial proportion of T cells with a T_CM phenotype could be detected within expanded T_CM-derived cells (CD4⁺ 20.4 ± 2.9%; CD8⁺ 13.5 ± 2.6%), T_naive-derived cells (CD4⁺ 45.2 ± 6.9%; CD8⁺ 18.5 ± 2.4%), and CD8⁺ T_SCM-derived cells (19.4 ± 2.7%), as well as in bulk PBMC–derived cells (CD4⁺ 12.4 ± 2.8%; CD8⁺ 15.7 ± 3.9%).

To evaluate cytokine production in response to CAR stimulation by T cell subset–derived cells expanded within bulk PBMCs, we performed intracellular flow and costained with Abs to NGFR (the marker for the subset) and to the spacer/hinge region of the CD19.CAR (Fig. 5A, 5B) to distinguish subset- and bulk-derived T cells. Fig. 5C and 5D show IFN-γ, TNF-α, and IL-2 production from each T cell subset (CD4⁺: red bars, CD8⁺: blue bars) and from the T cell subset–depleted bulk population (CD4⁺: black bars, CD8⁺: white bars) in response to a CD19⁺ mantle cell lymphoma (JEKO) and a CD19⁻ Hodgkin lymphoma (HDLM2). T_naive-derived cells produced relatively low levels of all cytokines compared with bulk-derived T cells and with other subsets, which showed similar capacities to produce cytokines in response to CAR stimulation, even when their frequencies within the cocultures were low (Fig. 5C–E).

We next tested the killing capacity of each subset cultured alone, as well as its respective subset-depleted and subset-reconstituted PBMC cell preparation (Fig. 6A). Fig. 6B shows the killing capacity of each T cell subset cultured alone, T cell subset–depleted bulk population, and T cell subset–reconstituted bulk population in response to a CD19⁺ mantle cell lymphoma (JEKO) and a CD19⁻ Hodgkin lymphoma (HDLM2). Intriguingly, killing of CD19⁺ JEKO cells was significantly reduced in cultures in which T_CMs and T_EMs were depleted from the bulk PBMC population (Fig. 6B, top panels). In contrast, T cell subset–reconstituted populations showed little variance in their killing (Fig. 6B, middle panels). Comparison of the T cell subsets cultured alone revealed a significantly lower killing capacity of T_naive-derived cells (Fig. 6B, bottom panels).

In summary, T_naive-derived cells expand preferentially within polyclonally activated PBMCs and express a memory T cell surface marker profile, but they show reduced effector cytokine production and killing compared with memory-derived subsets.

To evaluate the phenotypic and functional characteristics of each T cell subset cultured in the presence of the whole PBMC preparation compared with T cells cultured alone, we followed the experimental setup presented earlier, which included the transduction of each T cell subset using a retroviral vector encoding a second-generation CD19.CD28ζ CAR and a truncated NGFR (CD19.CAR-NGFR) and in-parallel transduction of each subset-depleted PBMC population with the same second-generation CD19.CAR but without NGFR. Initially, we followed the fate of individually sorted CAR-transduced T cell subsets after activation, transduction, and culture alone or after reconstitution into the relevant subset-depleted population (Fig. 7A). Each T cell subset differentiated during culture; by day 7 of culture, T cells with T_CM and T_EM phenotypes (CCR7⁺CD45RO⁺ and CCR7⁻CD45RO⁺, respectively) dominated all subset-derived cultures (Fig. 7). We defined the T_CM proportion of each T cell subset as a measure of differentiation. When comparing the phenotype of each transduced subset cultured alone or within the transduced bulk PBMC population, we found significantly higher proportions of phenotypically less-differentiated T_CMs in separately cultured T_CM- and T_naive-derived cells (day 14: T_CM-derived cells cultured in bulk: CD4⁺ 13.1 ± 2.6%; CD8⁺ 12.9 ± 1.4% versus T_CM-derived cells cultured alone: CD4⁺ 21.7.1 ± 9.9%; CD8⁺ 21.2 ± 5.4% and T_naive-derived cells cultured in bulk: CD4⁺ 14.8 ± 4.7%; CD8⁺ 11.2 ± 1.1% versus T_naive-derived cells cultured alone: CD4⁺ 39.8 ± 7.8%; CD8⁺ 29.1 ± 6.6%) (Fig. 7). A substantial proportion of T cells with a T_CM phenotype could also be detected within expanded CD8⁺ T_SCM-derived cells (day 14: T_SCM-derived cells cultured in bulk: 15.3 ± 1.7% versus T_SCM-derived cells cultured alone: 22.6 ± 0.9%) (Fig. 7). Of note, T cell cultures derived from less-differentiated T_SCM-, T_CM-, and T_naive-derived cultures had more stable CAR expression in long-term cultures up to day 21, whereas T_EM- and T_EMRA-derived cultures showed partial loss of their CAR (Supplemental Fig. 2).

When comparing intracellular cytokine production in response to CAR stimulation (using the CD19⁺ mantle cell lymphoma [JEKO]) of each transduced subset cultured alone or within the transduced bulk PBMC population, T_naive-derived cells produced less IFN-γ when cultured within the transduced bulk PBMC population than when cultured alone, whereas all other subsets showed similar IFN-γ production, regardless of whether they were cultured in the presence of the whole PBMC preparation or alone (Supplemental Fig. 3). T_naive-derived cells produced relatively low levels of IFN-γ compared with other T cell subsets that produced similar cytokine levels in response to CAR stimulation (Supplemental Fig. 3).

Finally, we compared T cell subset proliferation in response to CD19.CAR stimulation. Each transduced subset cultured alone or T cell subset–derived cells expanded within bulk PBMCs were stimulated on day 14 of culture using CD19⁺ JEKO cells in the presence or absence of exogenous cytokines (IL-7 and IL-15) (Fig. 8A), and T cell subset expansion was measured 7 d later. To assess the expansion of each subset cultured in the presence of the whole PBMC preparation, we used Abs to NGFR (T cell subset) and to the spacer/hinge region of CD19.CAR before and after CAR stimulation (Fig. 8A). In the absence of exogenous cytokines, T_EMRA- and T_naive-derived cells failed to proliferate, whereas T_SCM-, T_CM-, and T_EM-derived cells expanded to a similar degree, regardless of whether they were cultured alone or within the transduced bulk population (Fig. 8B). When cultured with cytokines, only T_EMRAs did not expand. As expected, the expansion of the remaining subsets was much greater than without cytokines (Fig. 8C) and, notably, T_naive-derived cells showed significantly greater expansion when cultured alone than when cultured within the transduced bulk population (65-fold for CD4⁺ and 36-fold for CD8⁺ alone compared with 43-fold for CD4⁺ and 15-fold for CD8⁺ cultured in bulk) (Fig. 8C). In contrast, all other subsets showed slightly greater expansion when cultured in bulk (Fig. 8C).

In summary, T_naive-derived cells differentiate more when cultured in the presence of memory T cells than when cultured alone, and they show coincidentally reduced effector cytokine production and re-expansion capacity in response to CAR stimulation. Overall, T_naive-derived cells show reduced effector cytokine production compared with memory-derived subsets, and they lack the ability to respond to stimulation in the absence of cytokines.

Because the T_naive-derived CAR-modified cells that dominated our cultures showed inferior CAR-mediated effector function and may have limited longevity, we aimed to increase the frequencies of T_CM- and T_SCM-derived cells during expansion of the cultures. Therefore, we compared the effects of the two cytokine regimens on relative T cell subset expansion. To determine how IL-2 could influence the outgrowth of T_CMs and T_SCMs within bulk cultures, we reconstituted FACS-purified T_CMs and T_SCMs, as described earlier (Fig. 9A), followed by polyclonal activation with CD3 and CD28 Abs in the presence of IL-2 (50 IU/ml) or IL-7 and IL-15 (each at 10 ng/ml) and transduction with a retroviral vector encoding mOrange. We reconstituted each subset into its respective subset-depleted population on day 3 and then assessed the fate of T_CMs and T_SCMs by tracing their path during the course of expansion (Fig. 9B). The medium and cytokines were changed every 3 d during culture and expansion. The frequency and absolute numbers of each subset were assessed weekly for up to day 35 after the initial stimulation (Fig. 9C). Although the trend toward higher frequencies of CD4⁺ or CD8⁺ T_CM- and T_SCM-derived cells induced by culture in IL-7 and IL-15 (Fig. 9C) did not reach significance, we plan to enhance this trend using cytokine modifications and to use our separation approach to validate the changes. Owing to the higher overall fold expansion rate of T cells expanded in IL-7 and IL-15, the absolute numbers of T_CM- and T_SCM-derived cells were significantly higher than in IL-2–grown cultures (Fig. 9D). Furthermore, T cells cultured in IL-2 showed an earlier contraction, as defined by a reduction in cell numbers from day 20, whereas T cells cultured in IL-7 and IL-15 expanded long-term, with no signs of culture contraction up to day 35 (Fig. 9D). These results confirm that IL-7 and IL-15 better supported the expansion of T_CMs and T_SCMs within bulk cultures (22, 39).

The central purpose of our study was to understand the contribution of naive and effector/memory T cell subsets to the bulk in vitro–stimulated CAR-stimulated T cell products that are eventually infused into patients in our clinical trials. To this end, we developed an approach to assess the fate of each T cell subset by tracing its path during the course of expansion after transduction with a GD2.CAR or a CD19.CAR and cultured alone or after reconstitution into the relevant subset-depleted PBMCs that had been activated and transduced in parallel (e.g., T_SCM into T_SCM-depleted PBMCs). In our proof-of-concept study, we show that all subsets were similarly susceptible to transduction, and the rate of T cell subset expansion was unchanged by transfer into the subset-depleted population. The greatest rate of expansion was observed in T_SCM-derived T cells, but because T_SCMs represented <1% (0.9 ± 0.2%) of the starting population, their descendants made up <2% of the final product on day 14. Naive-derived T cells showed the second greatest rate of expansion, and this population dominated the cultures by day 14, accounting for up to 89% of CD19.CAR-modified T cells. T_CMs showed the third greatest rate of proliferation and accounted for up to 44% of the final population. T_EM- and T_EMRA-derived populations proliferated relatively weakly and were poorly represented in the final population (<5 and <3%, respectively). Despite the greatest proliferative response to TCR stimulation, the dominant naive-derived T cell majority showed relatively poor cytokine production, killing, and poor proliferative responses to CAR stimulation. Further, T_naive-derived cells differentiated to a greater extent when cultured in the presence of memory T cells compared with when cultured alone and showed reduced re-expansion capacity in response to CAR stimulation. IL-7 and IL-15 induced greater total fold expansion of CD3/28-activated T cells and produced higher frequencies of T_CM- and T_SCM-derived cells than did IL-2.

Successful CAR-modified T cell therapy depends on multiple cell- and patient-dependent factors. Among these, the T cell subset origin of the final T cell product infused may be critical to the final outcome by contributing to postinfusion T cell expansion, effector function, and long-term persistence (26, 34, 37). However, although the effector memory phenotype of the final CAR-modified T cell product infused is often described, the T cell subset of origin is rarely investigated, so that functional dissection of the infusion product (T_EFF products) remains a prerequisite to better understand and correlate therapy outcome with T cell product input. Although individual subsets have been expanded in isolation, neither their expansion and function after reconstitution into the bulk population nor the subset contribution to the final T cell product have been reported. Of note, we analyzed the T cell subset contribution only of PBMCs activated on CD3 and CD28–coated plates in the presence of IL-7 and IL-15 or IL-2, which are the methods that we have used to expand CAR-modified T cells in our clinical trials. Many other methods are used by other groups, including CD3/28-coated beads, soluble CD3 and CD28 or PHA, and these may produce different results. Regardless, our strategy presents a means to evaluate the effects on T cell subsets resulting from any changes in culture conditions.

There is intense discussion about the optimal phenotypic T cell profile and optimal T cell subset for immunotherapeutic use with regard to longevity, engraftment, and antitumor effector function (22, 26, 28, 31, 34–37). Berger et al. (30) showed that CMV-specific T_EFF clones derived from T_CMs exhibited superior engraftment capacities compared with clones derived from T_EMs, and they were long-lived and established persistent memory T cells in a primate model in which TCR stimulation was provided by endogenous CMV. Other groups have suggested that the T_naive population is important, because this subset may contain precursors for tumor-specific Ags that could contribute to tumor elimination (47). The question of whether naive-derived T cells displaying memory T cell phenotypic characteristics have the capacity for long-term survival in vivo, particularly if they do not encounter their cognate Ag, remains an important subject for investigation. Naive to memory T cell conversion is a very complex process that requires multiple encounters with Ag, and it occurs in secondary lymphoid organs that provide an environment for such conversion (48–51). Several studies reported effective in vitro T cell priming for multiple viral and tumor Ags; however, their clinical efficacy remains to be demonstrated (51–60). In our studies, naive-derived T cells developed effector functions in vitro, but these were less potent than memory-derived T cells. Whether such naive-derived T cells with a memory phenotype have long-term memory potential in vivo and whether in vivo CAR stimulation by tumor cells can induce true memory conversion remain to be demonstrated.

Recently, Sommermeyer et al. (61) showed that when modified with a CD19.CAR, a combination of CD4⁺ naive- and CD8⁺ T_CM-derived T cells showed the most potent antitumor function in vitro and in preclinical models and in combination had synergistic antitumor effects in an EBV⁺ lymphoma xenograft model (Raji). This combination is currently under clinical evaluation. A possible advantage of the T_CM population is that it contains T cells specific for endogenous viruses that may provide antigenic stimulation after infusion.

Gattinoni and colleagues (29) described the existence of stem cell–like T cells (T_SCM) within the phenotypically naive (CD45RA⁺CCR7⁺) T cell population. The expression of CD95 distinguished T_SCMs (CD45RA⁺CCR7⁺CD95⁺) from T_naives (CD45RA⁺CCR7⁺CD95⁻) and, in a murine xenograft model, only T_SCMs were able to serially transfer graft-versus-host disease (39). The distinct functional and unique homeostatic properties of T_SCMs have also been demonstrated in a nonhuman primate model during the course of SIV infection. In that report, T_SCMs were described as the least-differentiated memory subset, functionally distinct from conventional memory cells, and they served as precursors of T_CMs (38). In a human study, Xu et al. (36) suggested that the frequency of T_SCM-like cells within the infused T cell product correlated with in vivo expansion following adoptive transfer, whereas Biasco et al. (62) showed that T_SCMs produced the greatest contribution to long-term reconstitution in gene therapy studies using T cells. Hence, T_SCMs may have even greater potential for expansion and long-term persistence than T_CMs (39), but the relative benefits of T_CM- and T_SCM-derived CAR-modified T cells in clinical trials have not been reported. Our data suggest that standard strategies for CAR-modified T cell manufacturing may preferentially expand T_naives, with some contribution from T_CMs but little contribution from T_SCMs, despite the fact that they expand exponentially in vitro when cultured alone or within PBMCs. Further studies are required to evaluate the long-term fate of T_CM- and T_SCM-derived cells, both in culture and in patients.

The ultimate T cell subset composition will likely also be affected by the CAR structure and endodomains, but our system provides a way to test these effects on the final infusion product. We show that culture conditions profoundly affect the subset composition of expanded T cells. IL-2 expansion produced lower total fold expansion than did IL-7 and IL-15 and, thus, decreased the numbers of T_SCMs and T_CMs in the final product. Although not significant, IL-7 and IL-15 also produced higher frequencies of T_CM- and T_SCM-derived cells than did IL-2. Further efforts to improve the relative frequency of T_CM/T_SCM-derived cells over naive-derived T cells could be validated using our strategy. Other culture manipulations, including culture vessels, media formulation, and type of initial stimulus, will likely also influence the degree to which each subset contributes to the final population. Indeed, in our prior study, we found that T_SCM-derived CMV-specific T cells did not expand within the bulk population if they were stimulated with CMV peptides rather than with CD3 and CD28 Abs (22). A more direct way to obtain the desired T cell subset is to separate it from the starting population prior to culture, but this increases the expense and complexity of manufacture (61). It would be instructive to know whether the manufacturing strategies used in various clinical trials promote the outgrowth of T_CMs and T_SCMs, and it would useful to develop cell culture strategies that enhance their expansion within bulk cultures, without the requirement for separation strategies. We provide a strategy to evaluate the subsets of origin of T cells grown under any conditions using PBMCs from patients or healthy donors. Our study supports the importance of IL-7 and IL-15 for the expansion of T_CM- and T_SCM-derived T_EFFs, at least using our manufacturing strategy (36).

Although the different T cell subsets were not differentially transduced by retroviral vectors, their dissimilar rates of proliferation led to changes in their relative frequencies in the final population, which was dominated by naive-derived T cells (76 ± 19% of CD3⁺ T cells). T_CM-derived populations maintained their frequencies (30 ± 12% of CD3⁺ T cells), and T_SCM-derived populations were maintained or slightly increased (mean of 1.5 ± 0.3% of CD3⁺ T cells), whereas T_EM and T_EMRA frequencies were reduced (mean of 4.7 ± 1.6% and 2.8 ± 1.3% of CD3⁺ T cells, respectively). Although the majority of each subpopulation differentiated into cells with a T_EM-like phenotype, their functionalities differed with respect to cytokine production and proliferation in response to CAR stimulation. T_naive- and T_EMRA-derived cells produced lower levels of cytokines and proliferated poorly in response to CAR stimulation compared with T_CM- and T_SCM-derived cells. This may be of little importance for T_EMRAs, because they make up only a minor fraction of the final product; however, a high frequency of poorly functional T_naive-derived cells may be less effective after infusion. Our results confirm reports suggesting that naive-derived T cells possess poor effector functions (37). However, we cannot exclude the possibility that naive-derived T cells will acquire improved function after multiple stimulations in vivo via CAR, although they responded relatively poorly to a single CAR stimulation in vitro. Our data confirm the recent report by Klebanoff et al. (27) showing enhanced phenotypic and functional differentiation of naive-derived T cells that have been cultured in the presence of Ag-experienced memory T cells. Naive-derived T cells cultured in bulk PBMC preparations show reduced re-expansion capacity in response to CAR stimulation and, hence, may lack antitumor potential when transferred in vivo. Nevertheless, when we cultured naive-derived T cells separately, re-expansion to CAR stimulation was reduced compared with T_SCM-, T_CM-, and T_EM-derived cultures. Hence, if T_naives are to be used for adoptive immunotherapy, we would propose culture in isolation to avoid the negative effects of culture with Ag-experienced memory T cells. Further in vitro and in vivo studies over longer time periods are required to determine the long-term consequences of infusing a preponderance of T_naive-derived cells; however, without some form of differential gene marking, this may be difficult.

Our strategy could also be used in cases in which short-lived effectors are desirable for safety reasons (e.g., to test a novel T cell manipulation). Regardless, our strategy can be a powerful tool to evaluate the effects of manufacturing changes on the subset contribution to in vitro–expanded T cells. Certainly, a one-size-fits-all strategy for the generation of CAR-modified T_EFFs may not be expected because of tumor cell biology and patient characteristics, but a better understanding of the characteristics of the generated CAR-modified T cell products will help to identify the subsets with most clinical efficacy. To this end, our strategy provides a powerful tool to elucidate the characteristics of CAR-modified T cells polyclonally expanded from PBMCs on CD3 and CD28 Ab-coated plates and to reveal the functional properties of each expanded T cell subset.

We thank Dr. Jason Millward (Institute of Medical Immunology, Charité University Medicine Berlin and Experimental and Clinical Research Center, a joint cooperation between the Charité Medical Faculty and the Max-Delbrück Center for Molecular Medicine) for statistical advice. This project was supported in part by the Cytometry and Cell Sorting Core at Baylor College of Medicine and the expert assistance of Joel M. Sederstrom and team. We thank Dr. Juan F. Vera (Center for Gene Therapy, Baylor College of Medicine) for the generous provision of the SFG.IRES.mOrange construct, as well as Tatiana Goltsova and Dr. Amos Gaikwad (Texas Children’s Hospital Flow Cytometry Core Laboratory) for technical flow cytometry assistance.

C.M.R. is on the scientific advisory board of Cell Medica. The other authors have no financial conflicts of interest.