IL-7 and IL-15 instruct the generation of human memory stem T cells from naive precursors

Adaptive immunity is a potent and flexible system able to combat microbes and cancer cells.^1,2  In the presence of infections or cancer, antigen-specific lymphocytes expand and differentiate into effectors devoted to rapidly clearing the pathogen and memory cells able to persist long-term to patrol the entire organism for recurrence and minimal residual disease.^3,4  However, the mechanism and hierarchical differentiation path underlying the generation of memory precursors and terminal effector cells remain to be fully elucidated.⁵  This process has been proposed to involve a self-renewing, stem cell–like memory T-cell subset capable of differentiating into effectors on antigen reencounter.^6,7  This T-cell subset, referred to as memory stem T cells (T_SCM), and initially described in mice,^8,9  begins to be unveiled in humans.¹⁰  T_SCM potential biodistribution and long-term persistence represent appealing features to overcome the current limitations of cancer adoptive immune-gene therapy.^11-13  At present, clinical-grade protocols able to obtain or preserve T_SCM functional and phenotypic characteristics remain to be defined. We previously showed that costimulation of unselected T cells and culture with γ-chain cytokines allow the preferential generation of gene-modified T cells with a functional central memory (T_CM) phenotype, superior to effector/effector memory (T_EM) counterparts for expansion potential and antitumor activity.^14,15  Compared with T_CM and T_EM lymphocytes, naive T cells (T_N) are endowed with the highest developmental plasticity and are unique in the ability to generate daughter cells with potential to enter the entire spectrum of immunologic memory, including T_SCM. We thus hypothesized that, starting from naive precursors, we could differentiate and genetically engineer human T_SCM. We report that IL-7 and IL-15 support the generation of postmitotic costimulated CD8⁺ T cells with molecular and functional features of T_SCM cells. These cells—defined by the expression of CD45RA, CD45R0, CD62L, CCR7, IL-7Rα, and CD95—can be identified among healthy subjects, are selectively enriched in hematopoietic stem cell transplant (HSCT) recipients, and reveal a phenotypic and functional profile distinct from that of T_CM and T_EM cells for extensive expansion capacity and ability to differentiate into potent effectors, self-renew, and expand in serial transplant recipients.

Peripheral blood mononuclear cells (PBMCs) were harvested from healthy donors after informed consent in accordance with the Declaration of Helsinki. PBMCs were isolated by Ficoll-Hypaque gradient separation (Lymphoprep; Fresenius), and enriched for CD3⁺ or CD8⁺ cells with the Pan T Cell Isolation Kit II or the CD8 Isolation Kit, respectively (Miltenyi Biotec) following the manufacturer's instructions. Cells were labeled with anti-CD3, anti-CD45RA (BD Biosciences), and anti-CD62L (Exbio) fluorescent antibodies and FACS-purified into T_N, T_CM, and T_EM on a MoFlo MLS cell sorter (DAKO). Postsorting analysis of purified subsets (comprising CD45RA, CD62L, and CD45R0 flow cytometric evaluation) revealed greater than 95% purity.

Sorted T lymphocytes were activated by 30 ng/mL plate-bound anti-CD3 antibody (OKT3; OrthoBiotech) or anti-CD3/CD28–conjugated magnetic beads (bCD3/CD28, ClinExVivo; CD3/CD28, Invitrogen) in a 3:1 bead/T-cell ratio and then cultured with: (1) recombinant human (rh) IL-2 at 300 IU/mL; (2) rhIL-7 at 5 ng/mL (PeproTech), or (3) rhIL-7 and rhIL-15 at 5 ng/mL each (PeproTech). Cytokines and medium were replaced every 3-4 days. Transduction was performed 48 hours after initial stimulation using either a HSV-Tk/ΔLNGFR-encoding retroviral vector (SFCMM-3 Mut2, MolMed)¹⁴  or a lentiviral vector (LV) encoding a transgenic WT1-specific TCR.¹⁶  Transduction efficiency was assessed at day 7 after stimulation by flow cytometry. Cells were counted every 3-4 days by Trypan blue dye exclusion.

Cells were labeled with fluorescent antibodies against CD3, CD4, CD8, CD45RA (Biolegend); CD45R0, CD27, CD28, CD95, CD271, CCR5, CCR7, CXCR4, HLA-DR, IL-2Rβ, 7AAD (BD Biosciences); CD62L (Exbio); CXCR3 (R&D Systems); IL-7Rα, Vβ21 (Beckman Coulter). CMV pp65-specific T cells were detected with HLA-A*0201-NLVPMVATV PE-conjugated pentamers (Proimmune). For intracellular staining, cells were stimulated with 50 ng/mL PMA (Sigma-Aldrich) and 1 μg/mL ionomycin (Sigma-Aldrich), brefeldin 10 μg/mL (Sigma-Aldrich) was added for 2 hours, cells were stained with surface antibodies, washed and fixed with 1% paraformaldehyde. Intracellular staining was performed with anti–IL-2, γ-IFN, granzyme A, and perforin-specific antibodies (BD Biosciences) in PBS containing 0.05% saponin (Sigma-Aldrich). Data were acquired using a FACSCanto II (BD Biosciences) and analyzed with FlowJo Version 8.7 software (TreeStar).

Two weeks after bCD3/CD28 stimulation and transduction with an LV encoding for a WT1-specific TCR,¹⁷  lymphocyte subpopulations (0.5 × 10⁶) were cocultured with WT1_126-134-pulsed (100μM) irradiated (100 Gy) T2 cells (0.2 × 10⁶) in the presence of irradiated (30 Gy) autologous PBMCs (2 × 10⁶). Cells were cultured with IL-7 and IL-15 (5 ng/mL) for 16 days. Cytokines were replaced every 3-4 days. Lymphocytes were then tested simultaneously for cytokine production and memory subset composition, on short in vitro stimulation with cognate peptide. Briefly, 0.2 × 10⁶ T cells were preincubated with 50μM TAPI-2 (Enzo Life Sciences) for 1 hour as previously described.¹⁸  Cells were then coincubated with WT1_126-134–pulsed (100μM) irradiated T2 cells in 1:1 ratio for 4 hours. After adding brefeldin 10 μg/mL for an additional 2 hours, cells were stained with surface antibodies, fixed, and stained for intracellular IL-2 and γ-IFN.

Real-time quantitative PCR (qPCR) for single-joint TCR excision circles (sjTRECs) was performed as described previously,¹⁹  using GAPDH as a control to standardize DNA content. Briefly, amplification reactions were performed in a final volume of 25 μL containing 50 ng of genomic DNA isolated from the different T-cell subsets, TaqMan universal PCR master mix (PerkinElmer/Applied Biosystems), and the appropriate primers and probes. The number of sjTRECs per 100 ng of DNA was determined on the basis of a standard curve developed in-house by cloning the sequence of α1 circles from human cord blood genomic DNA into a plasmid vector and diluting the plasmid into human DNA from a cell line devoid of TRECs (K562).

The protocol was approved by the Institutional Animal Care and Use Committee (IACUC). At day −1, 6- to 8-week-old female NOD/Scid mice (Charles River Italia) were given 1 mg of blocking anti-mouse IL-2Rβ monoclonal antibody intraperitoneally (IP) to neutralize residual natural killer activity. At day 0, mice received total body irradiation (250 cGy) and were infused IP with 5 × 10⁶ T cells. Mice were clinically monitored and weighed 3 times per week. Human chimerism on peripheral blood was assessed weekly by flow cytometry. At time of severe GVHD, defined as the concomitant presence of human chimerism greater than 10% and weight loss greater than 5%, mice were killed, and chimerism was assessed on bone marrow (BM) and spleen. In serial transplantation experiments, BM- and spleen-derived cells were reinfused in secondary recipients after the same procedure.

Transcriptional profiling was determined in 12 samples of T cells, using Affymetrix HG-U133 Plus 2.0 GeneChip arrays. RNA was isolated using the RNeasy Plus Mini kit (QIAGEN) and reverse transcribed by the GeneChip 3′IVT Express kit. Affymetrix HG-U133 Plus 2.0 GeneChip array hybridization, staining, and scanning were performed using Affymetrix standard protocols. Fluorescence signals were recorded by Affymetrix scanner 3000, and image analysis was performed with the GeneChip Operating Software (GCOS) software. All data analyses were performed in R using Bioconductor libraries and statistical packages. Probe-level signals have been converted to expression values using the robust multiarray average procedure (RMA).²⁰  Specifically, intensity levels have been background adjusted, and normalized using quantile normalization, and log2 expression values were calculated using median polish summarization and the custom-definition files for Human Gene U133 Plus 2.0 arrays based on Entrez (Version 13; http://brainarray.mbni.med.umich.edu/Brainarray/default.asp).²¹  The final dataset was composed of 12 samples [3 replicates each for T_N, T_CM, T_(TN), and T_(TCM)] and 18 185 Entrez gene IDs. Differentially expressed genes have been identified using significance analysis of microarray (SAM) algorithm coded in the samr R package. In SAM, we estimated the percentage of false-positive predictions with 1000 permutations and selected those transcripts whose q value (ie, false discovery rate [FDR]) was equal to 0 (or ≤ 0.05) and absolute fold change was ≥ 2. Gene set enrichment analysis was performed with GSEA software (http://www.broadinstitute.org/gsea/index.jsp)²²  on log2 expression data of T_N, T_CM, T_(TN), and T_(TCM) samples. Gene sets comprise BioCarta pathways derived from the Molecular Signatures Database (http://www.broadinstitute.org/gsea/msigdb/index.jsp) and the genes up-regulated in T_CM cells of the CD8⁺ T-cell memory signature (supplemental Table 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). Signatures have been considered significantly enriched at FDR ≤ 25% when using default parameters and 5000 permutations of gene sets. Gene Ontology classification has been assessed using DAVID functional annotation (http://david.abcc.ncifcrf.gov/), grouped under broad categories, and plot as pie chart. The list of human cell differentiation molecules has been obtained from UniProt (http://www.uniprot.org/docs/cdlist). The independent dataset from Gattinoni et al¹⁰  has been downloaded as.CEL files from GEO GSE23321 and the expression values quantified using RMA and the custom definition files for Human Gene 1.0 ST arrays based on Entrez (Version 13). Entrez-based probe sets for Human Gene U133 Plus 2.0 and 1.0 ST arrays have been matched before merging the 2 datasets obtaining a meta-dataset comprising 21 samples (12 samples from this study and 9 samples for T_N, T_SCM, T_CM from GSE23321) and 17 186 Entrez gene IDs. The meta-dataset has been normalized using ComBat, an empirical Bayes framework for adjusting data for batch effects (http://jlab.byu.edu/ComBat).²³  Microarray data generated in this study are available at Gene Expression Omnibus GSE41909.

Sorted T cells were loaded with CFSE 5μM (Invitrogen) and then cultured for 5 days with titrated (1-0.001 μg/mL) plate-bound OKT3, with or without anti-CD28 antibody (2 μg/mL; BD Biosciences) in the presence of IL-2 (300 IU/mL), IL-7, and IL-15 (5 ng/mL). Cells were then harvested, stained with 7AAD to exclude nonviable cells, and analyzed by flow cytometry.

Seven patients with high-risk acute leukemias referred to our Hematology Unit and treated with myeloablative allo-HSCT were studied. Patients received a treosulfan- and total body irradiation–based myeloablative conditioning followed by an unmanipulated peripheral blood graft from a matched unrelated donor (3 of 7) or an HLA-haploidentical related donor (4 of 7). The conditioning regimen included in vivo T- and B-cell depletion by ATG-Fresenius (10 mg/kg for 3 times) and Rituximab (a single 500-mg dose); GVHD prophylaxis consisted of Rapamycin (target level: 8-15 ng/mL, until day +60) and Mycophenolate Mofetil (15 mg/kg 3 times a day, until day +30). Peripheral blood was harvested 1 month after transplantation (median day after transplantation 33, range 25-50), on written informed consent.

Statistical analysis was performed with Prism 5 (GraphPad Software). Data are shown as the mean ± SEM. One-way ANOVA with Bonferroni correction and the 2-tailed unpaired t test were used for comparison of 3 or more groups, or 2 groups, respectively. Correlation was estimated by 2-tailed Pearson coefficients and significance. GVHD-free survival of mice infused with human lymphocytes was analyzed with log-rank and Gehan-Breslow-Wilcoxon tests.

T_N, T_CM, and T_EM were FACS-sorted to greater than 95% purity and activated with anti-CD3/anti-CD28–conjugated cell-size beads and low doses of IL-7 and IL-15 (Figure 1A). We then tested the expansion potential and sensitivity to genetic manipulation of individual T-cell subsets. All subsets revealed sensitive to both lentiviral and retroviral transduction (Figure 1B), with naive-derived transduced CD3⁺ T cells (T_(TN)) expanding significantly more than T_CM- and T_EM-derived transduced cells (T_(TCM) and T_(TEM), respectively; P < .05; Figure 1C and supplemental Figure 1A for CD8⁺ cells). Of note, among T_(TN) cells, a consistent fraction of CD8⁺ lymphocytes retained surface expression of CD45RA and CD62L (Figure 1D), a phenotype traditionally associated with naive T cells, for up to 30 days in culture (P < .05; Figure 1E). Most of these genetically modified CD62L⁺CD45RA⁺ cells also acquired the expression of CD45R0, the CD45 isoform associated with antigen-experienced T lymphocytes (Figure 1F-G), while preserving CCR7 expression (supplemental Figure 1B). This was in contrast to T_(TCM) and T_(TEM) cells, which mostly appeared CD45RA⁻ and CCR7⁻ (P < .001; supplemental Figure 1B). Expression of CD27 and CD28 costimulatory molecules was not significantly different among CD8⁺ T_(TN) and T_(TCM), whereas was lower in T_(TEM) cells (P < .01; supplemental Figure 1C). While displaying a naive-like surface phenotype, functionally T_(TN) revealed hallmarks of memory cells, such as undetectable sjTREC content and the ability to proliferate on TCR triggering in the absence of costimulation (supplemental Figure 2 panels A and B, respectively). To reveal the biologic determinants associated with this phenotype and dissect their relative contribution, we applied to FACS-purified CD45RA⁺CD62L⁺ cells different stimulation (costimulation vs TCR triggering alone) and culture conditions (IL-7 and IL-15 together, IL-7, IL-15, or IL-2 alone). Costimulation was critical for T_N cell expansion because cells failed to robustly proliferate in response to OKT3 (P < .001; Figure 1H). Furthermore, while IL-7 was critical for CD8⁺ T cells to preserve a CD45RA⁺CD62L⁺ phenotype in response to bCD3/CD28 (P < .01; Figure 1I), IL-15 best supported their expansion (Figure 1H). Thus, T_N appear unique for the ability to differentiate into T cells characterized by the stable expression of CD45RA, CD45R0, CD62L, CCR7 surface phenotype and high proliferative capacity. Mechanistically, costimulation, IL-7, and IL-15 appear critical and nonredundant for the instruction and expansion of this population.

To further define the phenotype and function of T_(TN) cells, we analyzed the expression of IL-7Rα, a marker for T-cell fitness and persistence,^24,25  and of chemokine receptors implied in homeostatic (CXCR4) or inflammation-driven (CCR5) migration on T_(TN), T_(TCM), and T_(TEM) CD8⁺ subsets. For T_(TN), we analyzed the entire cellular population as well as the specific CD45RA⁺CD62L⁺ cellular fraction. IL-7Rα was expressed by a higher proportion of CD8⁺ T_(TN) than T_(TCM) and T_(TEM) (P < .01; Figure 2A) for up to 30 days in culture (P < .001; Figure 2B). Likewise, CD8⁺ T_(TN) expressed higher levels of CXCR4 than CD8⁺ T_(TCM) and T_(TEM), while CCR5 expression was lower in T_(TN) than in CD8⁺ T_(TCM) and T_(TEM) cells (Figure 2C). CD8⁺ T_(TN) displayed also a significantly lower expression of HLA-DR compared with other memory counterparts (P < .01; Figure 2C). In contrast, CXCR3 and IL-2Rβ were similarly expressed in all T-cell subsets (supplemental Figure 1 panels D and E, respectively).

In functional analyses, we found that a substantial proportion of CD8⁺ T_(TN) secreted IL-2 in the absence of IFN-γ, and stained mostly negative for granzyme-A and perforin (Figure 2D-E). By sharp contrast, a larger proportion of CD8⁺ T_(TCM) and T_(TEM) cells produced IFN-γ, granzyme-A, and perforin (Figure 2D-E). Thus, T_(TN) revealed a less differentiated phenotype compared with T_(TCM) and T_(TEM) cells and lacked immediate effector functions. Accordingly, IFN-γ production negatively correlated with CD45RA/CD62L or IL7-Rα surface expression (r² = 0.8077 and P = .0002; r² = 0.7954 and P = .0002, respectively; Figure 2F). Thus, genetically modified T_(TN) cells display phenotypic and functional characteristics of early differentiated cells.

The phenotypic and functional profile suggests that T_(TN) cells represent a population of antigen-experienced cells at an early stage of differentiation. Therefore, we evaluated their expansion potential, differentiation, and self-renewal abilities in an antigen-specific model. Purified T-cell subsets were activated with bCD3/CD28 and engineered to express a transgenic TCR specific for a HLA-A2–restricted peptide from the WT1 tumor antigen.¹⁷  TCR-transduced lymphocytes were then stimulated with T2 cells pulsed with cognate peptide (Figure 3A). Although all gene-modified cells proliferated at high and similar extent on antigen encounter (Figure 3B), T_(TN) cells expanded significantly more than T_(TCM) and T_(TEM) (P < .05; Figure 3C). At day 16 after antigen-specific stimulation, TCR-transduced T cells were challenged with the cognate peptide and simultaneously assessed for cytokine production and memory subset composition. Of note, while the majority of T_(TN) differentiated into T_CM- and T_EM-like cells (Figure 3D), a substantial fraction of T_(TN) cells preserved the CD45RA⁺CD62L⁺ phenotype, which was absent in T_(TCM) and T_(TEM) counterparts. According to surface phenotype, within T_(TN) cells, differentiated CD45RA^+/−CD62L⁻ T cells produced high amounts of IFN-γ (Figure 3E), while the CD45RA⁺CD62L⁺ T cells were characterized by enhanced IL-2 secretion and low IFN-γ production (P < .05; Figure 3F). Altogether, these data suggest that T_(TN) are able to proliferate, differentiate, and self-renew on antigen encounter.

To further investigate the self-renewal potential of T_(TN) cells, we moved to an in vivo setting and evaluated T_(TN) ability to mediate xenogeneic GVHD²⁶  in serial transplantation experiments. We adoptively transferred equal numbers (5 × 10⁶ CD3⁺/mouse) of T_(TN) and T_(TCM) into irradiated (250 cGy) and NK-ablated immunodeficient mice. As controls, cohorts of mice were infused with unmanipulated FACS-sorted T_N, T_CM or with unselected lymphocytes in vitro activated with bCD3/CD28 plus IL-7 and IL-15 [T_(PBLs)]. The CD4:CD8 ratios of infused cells were: T_(TN), 0.71 ± 0.42; T_(TCM), 1.65 ± 0.65; T_N, 0.63 ± 0.24; T_CM, 3.57 ± 1.73; T_(PBLs), 2.12 ± 0.91.

One week after infusion, comparable frequencies of circulating human T cells were observed in all mice (supplemental Figure 3). After an additional week, T_(TN), but not T_(TCM) or T_(PBLs) cells persisted in vivo to levels comparable with those of unmanipulated human T cells (Figure 4A). Human chimerism correlated with GVHD. Indeed, T_(TN), but not T_(TCM), induced severe GVHD in the majority of mice, with kinetics and intensity comparable with those observed in mice infused with unmanipulated cells (Figure 4B). At time of GVHD, the majority of circulating CD8⁺ T_(TN) cells revealed a central memory (CD45RA⁻CD62L⁺) or effector memory (CD45RA⁻CD62L⁻) phenotype, indicating their ability to differentiate in vivo (Figure 4C). Of note, in T_(TN)-infused mice, a fraction of lymphocytes retained CD45RA and CD62L coexpression up to the day of sacrifice, whereas CD45RA⁺CD62L⁺ cells could not be detected in mice that received other T-cell subsets. Circulating T_(TN) cells showed higher frequencies of IL-7Rα, compared with T_(TCM) (P < .001; Figure 4D). This correlated with a higher human CD3⁺ cell chimerism in the bone marrow (Figure 4E) and in the spleens of T_(TN)-infused mice, which was approximately 5-fold higher than that found in mice infused with T_(TCM) cells (Figure 4F). Thus, T_(TN) cells engraft and expand in vivo and mediate a potent xeno-GVHD activity.

Given the rapid onset of lethal GVHD, we could not evaluate long-term persistence of T_(TN) lymphocytes. We thus performed serial transplantations. BM- and spleen-retrieved human CD3⁺ cells (5 × 10⁶ CD3⁺/mouse) from engrafted primary recipients were injected into irradiated and NK-ablated secondary mice. Circulating T lymphocytes were detected 1 week after infusion in secondary recipients of T_(TN) cells. T_(PBLs) also engrafted but to a lesser extent, while T_(TCM) failed to be detected (P < .05; Figure 4G). Accordingly, only T_(TN)-infused mice experienced lethal GVHD (P < .01; Figure 4H), characterized by a faster kinetics compared with first recipients. Most of circulating CD8⁺ lymphocytes in T_(TN)-infused mice acquired a central memory phenotype (mean percentage of CD45RA⁻CD62L⁺ CD8⁺ T cells: 77.2%, SEM 0.96), with a small fraction of the cells maintaining IL-7Rα expression (mean percentage of IL-7Rα⁺ CD8⁺ T cells: 4.1%, SEM 0.53; Figure 4I). T_(TN) proved able to home to the spleen and BM also in secondary recipients (Figure 4J). Thus, T_(TN) cells engraft, expand, and exhibit xenoreactivity in serial transplantations, suggesting self-renewal abilities.

To further characterize T_(TN) cells, we performed a genome-wide microarray analysis of CD8⁺ T_(TN) and T_(TCM) subsets and compared them to the recently reported T_SCM¹⁰ and to unmanipulated CD8⁺ T_N and T_CM as controls. We first compared the expression pattern of natural T_N and T_CM to overcome the noise secondary to transduction and in vitro culture of T_(TN) and T_(TCM). SAM algorithm identified a human memory signature defined by a set of 65 genes up- and down-regulated during the differentiation of naive CD8⁺ cells into memory lymphocytes (supplemental Table 1). The memory signature grouped T_N, T_CM, T_(TN), and T_(TCM) samples as depicted in Figure 5A. In particular, hierarchical clustering showed that T_(TN) lymphocytes were partitioned in the memory branch of the dendrogram, indicating that T_(TN) are part of the memory compartment. However, within the memory arm, while T_CM and T_(TCM) clustered together, T_(TN) constituted a separate branch which still conserved some homology with unmanipulated T_N cells. To precisely place T_(TN) in the T-cell hierarchy, we analyzed their relationships with the other subsets. First, to further reinforce that T_(TN) cells represent a memory subpopulation different from naive precursors, we used gene set enrichment analysis (GSEA). Specifically, we interrogated the genes of the CD8⁺ T-cell memory signature up-regulated in the T_CM subset (T_CM gene set; supplemental Table 1) and gene sets extracted from the Molecular Signature Database for statistical association between gene signatures and our T-cell subsets. If T_(TN) cells represent a memory subset with a naive surface phenotype, then the T_CM gene set of the CD8⁺ T-cell memory signature should be more enriched in T_(TN) than in T_N cells. Indeed, GSEA evidenced that the T_CM gene set is significantly overrepresented in T_(TN) (FDR q value = 0), confirming the clustering indication that T_(TN) cells are relatively more differentiated than their natural counterpart (Figure 5B). Next, we sought to validate also at the molecular level that T_(TN) are hierarchically superior to T_CM and T_(TCM). Indeed, a closer comparison of gene expression profiles between T_(TN) and the more differentiated T_CM and T_(TCM) revealed that genes less represented in T_(TN) have known roles in effector T-cell functions, (eg, EOMES, T-bet, NK receptors and granzyme family members; supplemental Table 2). All genes differentially expressed (2-fold cutoff and q value ≤ 0.05) between T_(TN) and memory cells [T_CM and T_(TCM)] were functionally classified under broad categories based on information of the Gene Ontology and of the Molecular Signature Database (enrichment analysis using GSEA). These analyses showed that T-cell differentiation and inflammation-related pathways are inversely correlated to the T_(TN) phenotype (Figure 5C). Thus, by gene profiling analysis, T_N-derived CD62L⁺CD45RA⁺CD45R0⁺ T cells prove to represent a novel memory subset distinct and hierarchically superior to T_CM and T_(TCM) cells and similar to naturally occurring putative T_SCM (supplemental Figure 4). Finally, we validated selected genes from a list of cluster of differentiation (CD) molecules differentially expressed in T_N compared with unmanipulated T_CM (supplemental Table 3) by flow cytometry. We confirmed the differential expression of CCR7 and CCR5 in T_(TN) and T_(TCM). Among the markers screened, we identified CD95, a molecule associated with memory T cells,²⁴  as one of the top-ranked antigens expressed by T_CM and not by T_N (supplemental Table 3). Flow cytometric analysis confirmed optimal expression of CD95 on T_(TN), T_(TCM), and T_CM memory cells (Figure 5D-E). In contrast, only a minority of natural CD8⁺CD45RA⁺CD62L⁺ T_N cells expressed CD95 (Figure 5D-E).

Among circulating CD45RA⁺CD62L⁺CD8⁺cells harvested from healthy donors, CD95⁺ cells expressed higher levels of CD45R0 than CD95⁻ cells (measured as relative fluorescence intensity [RFI], P < .001; Figure 6A-B), thus resembling in vitro–generated T_(TN) cells. Furthermore, FACS-sorted CD45RA⁺CD62L⁺CD95⁺ CD8⁺ lymphocytes, but not their CD95⁻ counterpart, proliferated at very low concentrations of plate-bound anti-CD3 antibody in the absence of costimulation, recapitulating a memory hallmark (Figure 6C-D). Thus, a distinct subset of CD45RA⁺CD62L⁺CD45R0^dim CD95⁺ cells displaying a naive surface phenotype associated with memory functional features can be identified in vivo.

We also investigated the representation of CD45RA⁺CD62L⁺CD45R0^dimCD95⁺ cells in patients with hematologic malignancies treated with allogeneic HSCT. This in vivo clinical setting is characterized by systemic inflammation and severe host lymphopenia.²⁷  Alloreactive T_N lymphocytes present in the graft are exposed to costimulation-competent antigen-presenting cells and elevated levels of IL-7²⁸  and IL-15,^29,30  an environment that closely mimic the culture conditions adopted in our study (Figure 6E). Results indicate that 1 month after HSCT, the majority of circulating CD45RA⁺CD62L⁺CD8⁺ T lymphocytes express CD95 (P < .01; Figure 6F-G), and represent a high fraction of CD8⁺ T cells (P < .05; Figure 5H), if compared with healthy controls. Finally, we investigated whether CD45RA⁺CD62L⁺CD45R0^dimCD95⁺ could also be defined within antigen-specific T cells. To this, we studied the phenotype of cytomegalovirus (CMV) pp65_(NLVPMVATV)–specific CD8⁺ T cells in seropositive HLA-A02⁺ donors (Figure 6I), and found that within CD45RA⁺CD62L⁺CD8⁺ cells, CMV-specific T cells were significantly enriched for CD95⁺ lymphocytes compared with polyclonal cells (P < .01; Figure 6J). Furthermore, CD95⁺CD45RA⁺CD62L⁺CD8⁺ CMV-specific lymphocytes expressed significantly higher levels of CD45R0 compared with CD95⁻ polyclonal counterparts. Thus, these data support the existence of CD45RA⁺CD62L⁺CD45R0⁺ CD95⁺ lymphocytes in vivo, and their accumulation in an IL-7/IL-15–enriched milieu. Finally, having identified CD95 as the discriminative marker for naive and memory CD45RA⁺CD62L⁺ lymphocytes, we formally proved that T_(TN)/T_SCM could be differentiated from natural CD95⁻ T_N cells. CD95⁺CD62L⁺CD45RA⁺CD45R0⁺ T cells were originated by bCD3/CD28 stimulation in the presence of IL-7 and IL-15 starting from purified naive CD95⁻ precursors. Expansion, sensitivity to genetic manipulation, and phenotypic and functional features of CD62L⁺CD45RA⁺CD45R0⁺ T cells differentiated from CD95⁻ naive precursors were similar to those of T_(TN) lymphocytes generated from unfractionated CD62L⁺CD45RA⁺ cells (supplemental Figure 5, Figures 1-2). Likewise, the critical role of costimulation, IL-7, and IL-15 for the instruction of postmitotic CD62L⁺CD45RA⁺CD45R0⁺CD95⁺ T cells was confirmed starting from purified naive CD95⁻ precursors (supplemental Figure 6). Costimulation was necessary for CD95⁻ T_N cell expansion regardless of the cytokine used (supplemental Figure 6A-C). IL-7 was instrumental for the generation of T_(TN)/T_SCM cells (supplemental Figure 6D), but their expansion required either IL-15 or IL-2 (supplemental Figure 6A). Importantly, IL-15 proved superior to IL-2 in supporting expansion coupled to preservation of the T_SCM phenotype (P < .05; supplemental Figure 6B). Notably, interference with the glycogen synthase kinase-3β/Wnt pathway was significantly less effective than the combination of IL-7/IL-15 in expanding and sustaining T_SCM cells generated from naive precursors (supplemental Figure 7). Altogether, our data support the existence of CD45RA⁺CD62L⁺CD45R0⁺CD95⁺ lymphocytes with stem-cell memory-like functions that differentiate from naive precursors and accumulate preferentially in the presence of IL-7 and IL-15. T_SCM cells can be easily differentiated, expanded, and genetically modified thus representing optimal candidates for adoptive T-cell therapy.

We identified a postmitotic CD8⁺ lymphocyte population, generated in vitro from T_N precursors but yet traceable in nature, characterized by lack of immediate effector functions, high expansion potential, and self-renewal abilities. T_(TN) are best defined as CD8⁺ lymphocytes expressing the CD45RA⁺CD45R0⁺CD62L⁺CCR7⁺CD95⁺ phenotype, and can be either induced starting from naive precursors in response to CD3 and CD28 engagement in the presence of IL-7 and IL-15 or isolated from the peripheral blood of healthy subjects. In clinical conditions characterized by high availability of IL-7 and IL-15, such as in HSCT recipients, T_SCM cells are enriched.

The existence and origin of memory T cells with stem cell–like qualities have been debated for some time.^6,7  In mice, memory stem T cells with the capacity of self-renewal and differentiation into T_CM and T_EM subsets have been identified among antigen-experienced CD44^low CD62L^high CD8⁺ T cells.^8,9  In humans, memory T cells with stem cell–like qualities have been recently reported.¹⁰  Recently described CCR7⁺CD62L⁺CD45RA⁺CD45R0⁻CD27⁺CD28⁺IL-7Rα⁺ putative T_SCM cells mediated superior antitumor responses than other memory T cells in adoptive T-cell therapy attempts, thus representing valuable candidates for adoptive T-cell therapies. The naive-derived CD45RA⁺CD45R0⁺CD62L⁺CCR7⁺IL-7Rα⁺CD95⁺ lymphocytes described here (T_(TN)) share many of the features reported for T_SCM lymphocytes, as they preserved the CD45RA⁺CD62L⁺ phenotype and refrained to acquire effector functions. They solely differ from putative T_SCM for CD45R0 expression. This discrepancy could be plausibly ascribed to the differences in the activation status at the time of CD45R0 assessment. While T_(TN) are analyzed at the peak of their activation (ie, 16 days after stimulation), putative T_SCM are circulating lymphocytes and therefore reasonably resting T cells. Indeed, we observed that putative T_SCM, once activated, up-regulate CD45R0 (supplemental Figure 7C), supporting the hypothesis that CD45R0 expression in T_SCM fluctuates as a function of T-cell activation status.

Here, we show that it is possible to generate, expand, and gene-modify T_SCM lymphocytes, while preserving their critical stem cell–like features. Most importantly, we demonstrate for the first time, in serial transplantation studies, the multipotency, the differentiation ability, and possibly the self-renewal capacity of T_(TN)/T_SCM lymphocytes, declined in a clinically relevant setting, such as xeno-GVHD. Notably, while gene expression profiling clusters our expanded T_(TN) cells with recently reported natural T_SCM cells¹⁰  (supplemental Figure 4), in a direct comparison with the alternative proposed method for ex vivo T_SCM generation,¹⁰  we show that costimulation, coupled to IL-7 and IL-15, is superior to glycogen synthase kinase-3β/Wnt inhibitors in promoting the expansion and transduction of T_(TN)/T_SCM cells (supplemental Figure 7). In addition, gene-modified T_(TN) cells described here express markers predictive of clinical outcome in adoptive immunotherapy trials^31,32  such as IL-7Rα, CD27, and CD28.

We identified a natural counterpart of in vitro–expanded cells, as the analysis of healthy subjects and of CMV⁺ donors revealed that CD45RA⁺CD45R0^dimCD62L⁺CCR7⁺CD95⁺ lymphocytes represent a natural T-cell subset, distinct from naive and memory cells for phenotypic appearance, and clearly behaving as memory lymphocytes, for activation requirements.³³  In contrast to T_N, T_SCM cells can proliferate in the absence of costimulation. We compared different activation stimuli and cytokines for the ability to foster T_SCM generation and expansion in vitro. While CD28 costimulation and IL-15 were critical for optimal expansion of these cells, in line with their role in homeostasis and proliferation of CD8⁺ T cells,^34-37  IL-7 appeared uniquely capable of instructing lymphocytes toward the stem-like phenotype. Accordingly, the finding that IL-7Rα was preserved by self-renewing T_(TN) in serial transplantation experiments in vivo further supports IL-7 as a critical determinant for T_SCM generation and maintenance. Among other possibly relevant signaling events, the glycogen synthase kinase-3β/Wnt pathway has been proposed as critical for the generation and maintenance of memory and stem cell–like properties. However, we failed to record up-regulation of Wnt-dependent genes in the T_(TN) gene-profiling analysis. This might indicate the existence of posttranscriptional mechanisms of regulation or of alternative signaling pathways, possibly emanating from the IL-7R. Future challenges will be needed indeed to define the molecular pathways controlling T_SCM generation. IL-7 has a pivotal role in controlling the size of the lymphoid population, and T cells compete for a limited amount of constitutively produced IL-7.^38,39  Hence, it is plausible that, during a physiologic primary immune response, in which proinflammatory factors prevail, only the few T_N cells that do not down-regulate IL-7Rα and gain access to IL-7, might give rise to T_SCM cells. By contrast, our clinical-grade culture system, in which access to IL-7 is not limited, may enable a higher fraction of cells to differentiate into functional T_SCM. In the HSCT clinical setting, IL-7 bioavailability is increased because of lymphodepletion²⁸  in a context of systemic inflammation and immune activation²⁷  that closely recapitulates our in vitro model. In this scenario, we documented a selective expansion of the T_SCM compartment, which represented the totality of CD45RA⁺CD62L⁺ putative naive CD8⁺ lymphocytes. This clinical observation endorses the central role of IL-7 in T_SCM generation.

Our study raises important questions on the mechanism of memory generation and maintenance. Our results suggest a precursor-product relationship between the CD45RA⁺CD62L⁺CD95⁺ postmitotic T_(TN) cells and the other memory subsets, consistent with a linear differentiation model in which T_N cells may differentiate first to T_SCM, then to T_CM and T_EM cells. However, we cannot exclude that in vivo priming of T_N cells could result in the generation of a mixed population of effector and memory cells, perhaps by asymmetric division, as suggested by other groups.^40,41 

An important aim of our study was to identify the optimal source of cells to be used for the genetic manipulation of lymphocytes suitable for adoptive T-cell therapy (ACT).^12,42,43  When we directly compared FACS-sorted T_N, T_CM, and T_EM, only T_N cells gave rise to significant numbers of T_SCM-like cells. Accordingly, T_(TN) revealed higher in vivo engraftment, persistence, and xenoreactivity when directly compared with T_(TCM) cells in primary and secondary transplant recipients. Thus, targeting naive T cells might have a direct impact on ACT.

In recent years, genetic manipulation of T cells produced important advances in the treatment of cancer patients. Expression of suicide genes, tumor-specific TCR, or chimeric antigen receptors, have significantly increased the safety and efficacy profile of T-cell products, and already produced objective clinical responses.^44-46  Recent technological advances, including the use of zinc finger nucleases, allow today a sophisticated genetic surgery of T lymphocytes, aimed at substituting, instead of only adding, specific biologic functions, such as antigen specificity.¹⁶  In this context, unique in the ability to couple antitumor efficacy with enhanced safety, preservation of a T_SCM functional phenotype will be critical for the generation of cellular products not only able to efficiently target cancer cells, but also capable of persisting and mediating a dynamic cancer immunosurveillance in treated patients. A rapid implementation of our protocol will be ensured by the availability of clinical-grade reagents required for T_(TN) generation and genetic manipulation. Thus, T_SCM cells could be readily edited with tumor antigen specificities and expanded according to our clinical-grade protocol, while preserving their expansion, multipotency, and differentiation abilities.

The authors thank Prof T. Tanaka (Osaka University, Japan) for kindly providing the TMβ-1 hybridoma.

This work was supported by the Italian Ministry of Health (GR07-5 BO and RO10/07-B-1), the Italian Ministry of Research and University (FIRB-IDEAS, linked to ERC starting grants), Fondazione Cariplo, the Italian Association for Cancer Research (AIRC, AIRC 5x1000), and the EU-FP7 program (ATTACK, PERSIST).

N.C. conducted this study as partial fulfillment of the PhD in Cellular and Molecular Biology, San Raffaele University (Milan, Italy). G.O. conducted this study as partial fulfillment of the PhD in Molecular Medicine, San Raffaele University (Milan, Italy).

Contribution: N.C. and C. Bonini designed the study and wrote the manuscript; N.C. conducted laboratory experiments; B.C., E.P., and A.B. helped to perform in vivo experiments; F. Cocchiarella, F.M., and A.R. performed microarrays; M.F. and S.B. performed bioinformatics analysis; G.O. performed and analyzed sjTREC quantification; C. Bordignon and A.M. contributed to manuscript drafting; and J.P., F. Ciceri, and M.T.L.-S. provided clinical data and samples.

Conflict-of-interest disclosure: C. Bordignon is an employee of MolMed SpA. C. Bonini has a research contract with MolMed SpA. The remaining authors declare no competing financial interests.

Correspondence: Dr Chiara Bonini, Experimental Hematology Unit, San Raffaele Scientific Institute, via Olgettina 60, Milano, Italy; e-mail: bonini.chiara@hsr.it.