Generation of Pure Highly Functional Human Anti-Tumor Specific Cytotoxic T Lymphocytes With Stem Cell-Like Memory Features for Melanoma Immunotherapy

Adoptive immunotherapy based on the transfer of anti-tumor cytotoxic T lymphocytes (CTLs) is a promising strategy to cure cancers. However, rapid expansion of numerous highly functional CTLs with long-lived features remains a challenge. Here, we constructed NIH/3T3 mouse fibroblast-based artificial antigen presenting cells (AAPCs) and precisely evaluated their ability to circumvent this difficulty. These AAPCs stably express the essential molecules involved in CTL activation in the HLA-A*0201 context and an immunogenic HLA-A*0201 restricted analogue peptide derived from MART-1, an auto-antigen overexpressed in melanoma. Using these AAPCs and pentamer-based magnetic bead-sorting, we defined, in a preclinical setting, the optimal conditions to expand pure MART-1-specific CTLs. Numerous highly purified MART-1-specific CTLs were rapidly obtained from healthy donors and melanoma patients. Both TCR repertoire and CDR3 sequence analyses revealed that MART-1-specific CTL responses were similar to those reported in the literature and obtained with autologous or allogeneic presenting cells. These MART-1-specific CTLs were highly cytotoxic against HLA-A*0201+ MART-1+ tumor cells. Moreover, they harbored a suitable phenotype for immunotherapy, with effector memory, central memory and, most importantly, stem cell-like memory T cell features. Notably, the cells harboring stem cell-like memory phenotype features were capable of self-renewal and of differentiation into potent effector anti-tumor T cells. These “off-the-shelf” AAPCs represent a unique tool to rapidly and easily expand large numbers of long-lived highly functional pure specific CTLs with stem cell-like memory T cell properties, for the development of efficient adoptive immunotherapy strategies against cancers.


INTRODUCTION
Anti-CD19 CAR (Chimeric Antigen Receptor-modified)-T cells have recently encountered impressive therapeutic successes in malignant CD19 + hemopathies. They underline the potential clinical interest of being able of amplifying in vitro "natural" antitumor T lymphocytes (TLs) for adoptive cell therapy (ACT), especially in solid tumors such as melanoma (1). Specific in vitro T lymphocyte (TL) activation for ACT requires efficacious antigen presentation. Antigen presenting cells (APCs) deliver three essential signals to TLs, the first one being provided by specific MHC-peptide complexes interacting with T cell receptors (TCRs), the second one by accessory molecules, and the third one by cytokines and chemokines secreted in the T cell environment. To provide these three signals in vitro, Peripheral Blood Mononuclear Cell (PBMC) feeder cells, virus-transformed or CD40-activated B cells and monocyte-derived dendritic cells can be used as APCs (2)(3)(4)(5)(6)(7). TL activation with anti-CD3 and anti-CD28 antibodies and high doses of IL-2 can also be envisaged (8,9). Many of these approaches have already been used in different clinical trials. Nevertheless, these techniques require long and non-standardized procedures. Obtaining TLs in a sufficient number with suitable phenotype and function for successful immunotherapy remains difficult.
To circumvent these difficulties, different groups have thus developed standardized, "off-the-shelf" artificial antigen presenting systems to deliver the appropriate signals to rapidly generate in vitro anti-tumor TLs (10). With this aim, beads, allogeneic or xenogeneic cells have been used (11)(12)(13)(14)(15). Different works have already given hints that artificial antigen presenting systems based on beads complexed with HLA molecules or anti-CD3/anti-CD28 antibodies have lower capacities to activate TLs than more physiological systems such as cell-based systems (10,16). Among the different cell-based systems, human K562-based artificial antigen presenting cells (AAPCs) have been successfully used in phase I clinical trials in melanoma and are still optimized (14,17). Noteworthy, K562 cells expressing a truncated form of CD19 for expanding in vitro anti-CD19 CAR-T cells did not show superiority to conventional protocols in a phase I clinical trial (18) and most of the cell-based artificial antigen presenting systems are still in preclinical development (19).
Many artificial TL activation systems mentioned above lead to the expansion of some TLs which are not cells of therapeutic interest. For example, CAR-T cells expanded in vitro with anti-CD3/anti-CD28 beads are injected without further purification, along with autologous activated T cells of unknown specificity. The presence of non-specific TLs could decrease ACT efficiency and eventually generate side effects. This problem has been circumvented with either multiple stimulations to obtain a more specific response or the injection of tumor specificderived TL clones (20)(21)(22). However, numerous in vitro TL stimulations and/or in vitro selected clones without an appropriate phenotype could lead to impaired reactivity in vivo (23). The common source of TLs which are used in the different studies are tumor infiltrating lymphocytes (TILs), TLs from patients' draining lymph nodes (LNs) or PBMCs (2-4, 15, 20). Many difficulties are associated with TIL and LN manipulation, because they mainly contain terminally differentiated TL effectors, which have been shown to display deficient function, impaired proliferation and survival in vitro (24,25). Consequently, PBMCs were suggested as an interesting source of anti-tumor TLs, even as good, at least, as the tumor itself (4,15,17). A new idea has thus emerged that "younger", less differentiated TLs with long-lived phenotype would be better tools for ACT (26,27). ACT protocols had already proven that central memory TLs (T CM ) displayed more interesting antitumor capacities than effector memory TLs (T EM ) (27,28). More recently, in mice, non-human primates and humans, a subpopulation of memory TLs called stem cell-like memory TLs (T SCM ), which is less differentiated than T CM , was characterized. They express in humans different common markers found on naïve TLs (including CD45RA and CD62L) and a marker of already activated TLs, CD95, thus making them more differentiated than naïve TLs. Moreover, they display high proliferation, self-renewal, long-life and multipotency features, giving rise to all the other memory subsets (29)(30)(31)(32)(33).
However, obtaining anti-tumor TLs displaying T SCM features in an easy and reproducible manner in an in vitro protocol has not been reported yet. Recent works have proposed using anti-CD3/anti-CD28-based protocol to activate naïve TLs from peripheral blood in CMV infection and graft versus host disease (GVHD) contexts (31,32). Nevertheless, a protocol allowing the generation of tumor-specific TLs displaying T SCM features has not been proposed yet (34).
In this study, we used AAPCs generated in our laboratory, derived from murine NIH/3T3 fibroblasts and expressing the most frequent HLA class I molecule, HLA-A*0201, the three main accessory molecules (CD54, CD58, CD80) and a MART-1derived peptide, MART-1 being an overexpressed antigen in melanoma (15). We thus designed an AAPC-based preclinical procedure starting from peripheral blood TLs (PBTLs) as a source of specific TLs to obtain melanoma-specific CTLs usable for ACT. We obtained a pure population of MART-1-specific CTLs (M1-CTLs) in sufficient numbers for ACT. Phenotypic study revealed that these highly functional M1-CTLs harbored long-lived, renewable and non-exhausted T cell features. Therefore, in this preclinical study, by using our AAPC system, we propose a protocol which favors the generation of pure tumor-specific TLs displaying T EM , T CM and T SCM features which could prove very promising for an efficacious anti-tumor ACT protocol.

Recruitment of Healthy Donors and Melanoma Patients
Ten healthy donors and six grade II/III melanoma patients were recruited based on Human Leukocyte Antigen (HLA)-A2 expression assessed by flow cytometry. Healthy donors and patients were enlightened and signed informed consent under agreement from Rouen University Hospital Institutional Review Board "Comitéde Protection des Personnes Nord-Ouest I" which approved this study.
Both the expression of transgenes and the absence of mycoplasma contamination were ascertained as described (35) and by regular PCR and Hoechst staining, respectively, every three months.

Purification of T Lymphocytes and Co-Culture With AAPCs
PBMCs were collected by density centrifugation on a lymphocyte separation medium (Eurobio, Courtaboeuf, France). Irradiated AAPCs (10 5 cells per well, 25 Gy) were plated in 24-well plates in AIM-V medium (Invitrogen, Saint Aubin, France) supplemented with 5% DCS. The next day, total TLs were negatively selected from PBMCs using Dynabeads untouched human T cell Kit (Invitrogen) according to manufacturer's protocol.
Purified TLs (10 6 cells per well) resuspended in AIM-V medium were co-cultured with irradiated AAPCs at a ratio of 10 to 1 for 21 days as previously described (15,35). On day 7 of the co-culture and then, every third day, 20 IU/ml of IL-2 (R&D systems, Abington, UK) were added to the wells.
Immunoscope M1-CTLs obtained at day 35 were harvested and CD8 + TLs were purified using human CD8 microbeads (Miltenyi Biotec) according to manufacturer's protocol. CD8 + TL and M1-CTL RNA were isolated using GenElute Mammalian Total RNA miniprep kit (Sigma-Aldrich). Complementary DNA (cDNA) was obtained by RNA reverse transcription using M-MLV reverse transcriptase (Promega, Charbonnière, France) following manufacturer's protocol. Precise procedures and primer sequences to study complementary determining region 3 (CDR3) were described previously (36). PCR products were examined in an Applied Biosystem/Hitachi 3100 genetic analyzer (Hitachi Electronics Engineering, Tokyo, Japan). Data were analyzed with Sequencing Analysis software (Invitrogen). IOTest Beta Mark TCR Vb Repertoire Kit (Beckman Coulter, Villepinte, France) was used to study 24 Vb subfamilies following manufacturer's instructions.
All cells were resuspended in sorting buffer. Flow cytometry measurements were performed on FACSCanto and BD LSRFortessa systems (BD Biosciences) and analyses were made using FlowJo software (TreeStar Inc., OR).
For cytotoxicity assays, the final peptide concentration of 10 µM was used. Five thousand target cells were incubated with TLs at different ratios during 4h at 37°C.
For functional avidity assays, peptide concentrations ranging from 10 to 10 -6 µM were used. Five thousand T2 cells were incubated with TLs at a unique ratio of 10 to 1 during 4h at 37°C.
Expression of relevant antigens (HLA-A2, MART-1) by all cell lines and the absence of mycoplasma contamination were ascertained as described (35) and by regular PCR and Hoechst staining, respectively, every three months.

Statistical Analysis
We used Prism software (GraphPad software Inc, La Jolla, CA) in order to perform two way-unpaired t tests or Friedman tests with Dunn's post-tests. All error bars represent standard errors of the means (SEMs). ns (not significant) was used for p>0.05, * for p<0.05, ** for p<0.01.

RESULTS
Rapid Generation of Pure MART-1-Specific Cytotoxic T Lymphocytes Using a NIH/3T3-Derived Artificial Antigen Presenting Cell-Based Protocol NIH/3T3-derived AAPCs have been described as an efficient tool to expand antigen-specific CTLs (15). AAPCs reconstitute a human HLA-A*0201-restricted immunological synapse, including accessory molecules CD54, CD58, CD80 and present for this study MART-1 A27L analogue peptide on HLA-A*0201 molecule (AAPC M1m ). Total TLs were magnetically isolated with more than 95% of purity from PBMCs (data not shown) and cocultured with AAPC M1m (day 0 (D0), Figure 1A). At each step of our protocol, M1-CTL percentage was assessed by flow cytometry (Figures 1B-D). After one round of expansion, M1-CTLs were purified based on a PE-labeled Pentamer staining followed by incubation with anti-PE magnetic bead sorting (day 21 (D21), Figure 1A). Purified M1-CTLs were finally co-cultured with AAPC M1m to expand them until day 35 (D35, Figures 1A, D).
At D0, 10 6 TLs per ml were deposited on AAPC M1m ( Figure 1A). Total TLs from patients and healthy donors contained in average around 1.5% and 0.5% of M1-CTLs respectively, representing around 15x10 3 and 5x10 3 M1-CTLs per million TLs at D0 ( Figure 1B). Both M1-CTL percentages and absolute numbers at D0 were significantly higher in patients than in healthy donors (p=0.047 and p=0.045 respectively, Figure 1B).
At D21, we obtained in average around 3x10 6 and 5.6x10 6 TLs in patients and healthy donors respectively, indicating a significantly lower number of TLs in patients (p=0.0048, Figure 1C). Obtained TLs contained around 19% of M1-CTLs both in patients and in healthy donors ( Figure 1C). Although no significant differences in M1-CTL percentages and absolute numbers were found between patients and healthy donors, a significantly lower enrichment in M1-CTL population was observed in patients compared to healthy donors (p=0.026, Figure 1C) at D21.
At D35, at the end of our protocol, we obtained around 4.3x10 6 and 2.8x10 6 TLs in patients and healthy donors respectively ( Figure 1D). Obtained TLs contained around 96% of M1-CTLs both in patients and in healthy donors ( Figure 1D). No significant differences in M1-CTL percentages and absolute numbers were found between patients and healthy donors ( Figure 1D).
With our protocol, between the initial (D0) and the final step (D35), M1-CTL number fold enrichment was in average around 600 and 2500 in patients and healthy donors respectively, although no significant differences could be found anymore between both groups because of large inter-individual fold number variability ( Table 1 and Figure 1D) at D35.

Oligoclonal Expansion Restricted to Several Vb Subfamilies of Obtained Pure M1-CTLs
To characterize TCR usage in obtained M1-CTLs, we analyzed TL repertoire by flow cytometry and immunoscope.
First, we studied obtained TLs with antibodies specific of 24 different T cell receptor (TCR) beta chain variable subfamilies (Vb). We observed a large distribution of TLs in the different Vb subfamilies (Figure 2A, left panel). However, Vb2, 3, 7.1, 13.1, 13.2, 14, 17 and 22 subfamilies preferentially expanded with our protocol (Figure 2A, right panel).
Then, analysis of CDR3 length using immunoscope showed that all Vb subfamilies displayed an almost normal distribution in purified CD8 + TLs at D0 ( Figure 2B, upper panels) with variations in and between both patients and healthy donors. At D35, all Vb families displayed skewed patterns in M1-CTLs ( Figure 2B, lower panels), with the selection of numerous different clones.

Highly Cytotoxic Function of Obtained Pure M1-CTLs
In order to evaluate the cytotoxic potential of obtained M1-CTLs we first assessed that M1-CTLs secreted IFNg, TNFa and IL-2. With no significant difference between patients and healthy donors, we found that around 80% of obtained TLs were high producers of IFNg and TNFa ( Figure 3A, left and middle panels), while IL-2 production in our conditions was limited to around 10% of obtained TLs with donor-dependent variations ( Figure 3A, right panel).
Finally, using a functional avidity assay on T2 cells pulsed with decreasing concentrations of M1m peptide, from 10 to 10 -6 µM, we observed that M1-CTLs had similar functional avidity in patients and healthy donors with still measurable lysis at peptide concentrations of 10 -5 µM and even 10 -6 µM for patients ( Figure 3C).
Therefore, M1-CTLs obtained with our protocol displayed similar high cytotoxic molecule secretion and high cytotoxic function in both patients and healthy donors.

Suitable Phenotype for Immunotherapy of Obtained Pure M1-CTLs
In order to study the phenotype of obtained M1-CTLs, we performed a flow cytometry phenotypic study on M1-CTLs obtained at the end of our protocol (D35). We found that all obtained M1-CTLs expressed CD95 and that, among them, more than 85% TLs were positive for CD45RA ( Figures 4A, B). CD45RO was expressed by around 20% of all M1-CTLs, 80% of these CD45RO + cells expressing both CD45RO and CD45RA. Approximately 30% of all M1-CTLs expressed CD62L and 10% co-expressed CD62L and CCR7 (Figures 4A, B, and data not shown).
All obtained M1-CTLs both in patients and healthy donors expressed CD44, CD122 and Bcl-2. A majority of them, around 75%, expressed CD27. CD28 and CD127 were expressed by around 25% of obtained M1-CTLs. In addition, almost no obtained TLs expressed final differentiation and inhibition markers CD57, CTLA-4, PD-1, FasL and BTLA. Moreover, part of obtained M1-CTLs (15 to 20%) expressed the skin homing factors CLA, CCR4 and CCR10 (Figures 4A, B). For all of these phenotypic markers, no significant difference was observed between healthy donors and patients. Depending on the expression of CD45RA and CD62L, we defined TLs with stem cell memory (T SCM ), central memory (T CM ) and effector memory (T EM ) features in M1-CTLs ( Figures 4C, D) at the end of our protocol. In CD45RA + fraction, around 35% of TLs expressed CD62L, corresponding to TLs with T SCM features (around 30% of total TLs). In CD45RAfraction, around 50% expressed CD62L, corresponding to TLs with T CM features (around 10% of total TLs). Finally, in CD45RA + and CD45RAfractions, CD62L -M1-CTLs corresponded to TLs with T EM features (around 60% of total TLs). There were no significant differences between patients and healthy donors ( Figures 4C, D).
Since long-lived CTLs expressing CD62L have been shown to express high levels of CD27, CD28 and CD127 molecules (31), we compared the expression of these markers in CD62L + and CD62Lpopulations obtained with our AAPCs ( Figure 4E). Importantly, we found that CD62L + CTLs significantly expressed higher levels of CD28 and CD127 compared to their CD62Lcounterpart. The level of expression of CD27 was equally high in both populations ( Figure 4E). As the vast majority of M1-CTLs in our experiments expressed both CD45RA and CD95 markers, we could conclude that M1-CTLs with a T SCM phenotype according to CD95, CD45RA and CD62L expression indeed expressed high levels of CD27, CD28 and CD127.
In conclusion, by using our AAPC system, our protocol favored the rapid and efficient generation of antigen-specific TLs with long-lived and "young" phenotype.
High Proliferative Capacity, Self-Renewal, and Multipotency In Vitro of Fluorescence-Activated Cell Sorting-Purified M1-CTLs With T SCM Phenotype Features T SCM have recently been described as memory TLs with high capacities to proliferate, and both to renew and to differentiate into other memory subpopulations (31,32). Therefore, to further characterize the M1-CTLs displaying T SCM phenotype features (CD95 + CD45RA + CD62L + ) in our co-cultures, all cells being CD95+, obtained TLs were sorted according to PentM1m, CD45RA and CD62L staining to separate M1-CTLs into "T SCM " (CD45RA + CD62L + ) and "T EM " (CD45RA +/-CD62L -, Figure 5A), the main obtained populations. The purity of each population was always greater than 98% (data not shown). After 14 days of co-culture, expansion, phenotype and function of obtained "total"-, "T EM "and "T SCM "-derived CTLs were studied. Interestingly, we observed around a 2-fold higher number of M1-CTLs derived from purified "T SCM " [T(T SCM )] compared to purified "T EM "-derived ones (T(T EM ), Figure 5B). Phenotypic analysis of T(T EM ) CTLs revealed stable lower expression of CD62L ( Figure 5C, left panel) than in both T (T SCM ) and T(T Total ) populations, where a fraction of M1-CTLs maintained high expression of CD62L ( Figure 5C, middle and right panels respectively). Interestingly, cells in both T(T SCM ) and T(T Total ) populations could differentiate into T EM ( Figure 5C, middle and right panels respectively). Yet, Chromium release assays showed that T(T EM ) and T(T Total ) CTLs displayed slightly higher cytotoxic capacities than T (T SCM ) CTLs, less rich in functional T EM cells ( Figure 5D).

DISCUSSION
In this study, we describe, for the first time to our knowledge, an efficient preclinical protocol, based on the use of artificial antigen presenting cells (AAPCs), allowing the expansion of anti-tumor specific CTLs with stem cell-like memory features. Our "off-the-shelf" NIH/3T3 fibroblast-derived AAPCs are easily generated, rapidly expanded and could express any HLA molecule and antigen of interest. Noteworthy, NIH/3T3 fibroblast-based protocols are already used in clinics with no reported adverse effect due to xenoreactivity (37). Moreover, to be fully cautious, in our protocol, AAPCs are irradiated (they are totally cleared off anyway after TL activation) and specific CTLs are very highly purified (CTLs of unknown specificity could decrease ACT efficiency and even have deleterious effects). GMP grade NIH/3T3-based AAPCs have already been constructed and their use is now envisaged in different severe infections and cancers.
In this study, we chose to use peripheral blood TLs (PBTLs). Many studies have already shown that peripheral blood is a good source of antigen-specific TLs (4,14,15,17). PBTLs are easier to manipulate and less differentiated than tumor-infiltrating TLs.
The vast majority of M1-CTLs in the peripheral blood, not only of healthy donors but also of melanoma patients, even harbor the phenotype of naïve TLs (38). After stimulation, naïve TLs can indeed differentiate into different subsets of efficient long-lived memory TLs (31,32).
Using our AAPC M1m , we efficiently stimulated functional TLs in melanoma patients and in healthy donors. Significantly higher M1-CTL percentages and absolute numbers were detected at D0 in the peripheral blood of melanoma patients, certainly due to the ongoing anti-tumor natural immune response. But these patients' immune response might be negatively regulated because of tumor escape mechanisms, health problems or immunosuppressive treatments, which could affect M1-CTL initial proliferative capacities and explain M1-CTL lower FMP was used as a control for non-specific lysis. (B, C) Graphs represent the results obtained with ten healthy donors (left graphs) and six patients (right graphs), each donor having been studied in three independent experiments. Data are represented with standard errors of means. enrichment in some patients. Nevertheless, our results indicate that patients' peripheral blood M1-CTLs derived from patients' PBTLs globally preserve high in vitro proliferative capacities and functions after two AAPC stimulations, which might not be the case for M1-CTLs derived from TILs because of TIL full differentiation and intra-tumoral strong immune escape mechanisms. After a short time of co-culture, corresponding to only two rounds of stimulation, allowing limited CTL manipulations, we always obtained pure M1-CTL numbers compatible with ACT (10 8 to 10 9 pure M1-CTLs obtained from 100ml of peripheral blood) even when a low frequency of M1-CTLs (lower than 0.05%) was detected in the peripheral blood ( Table 1 and data not shown).
Purified M1-CTLs displayed a rather large TCR diversity revealed by a repertoire study which could be even more extensive before clinical application. Interestingly, all amplified Vb families have already been described after MART-1-specific CTL activation using autologous or allogeneic systems, indicating that our AAPC-based protocol did not induce any major bias in TL repertoire (4). Moreover, large TCR diversity has already been shown to be associated with better clinical outcome in melanoma patients (39,40). Efficient ACT requires sufficient number of TLs with strong functional capacities. We obtained with our protocol numerous pure M1-CTLs which were producing high amounts of IFNg, TNFa and IL-2, and were able to specifically lyse both MART-1pulsed T2 cells and melanoma-derived cell lines very efficiently, with a high functional avidity.
Efficient ACT also requires long-lived non-terminally differentiated CTLs for long term effects in vivo. We obtained with our protocol numerous pure M1-CTLs which expressed different long-lived TL markers (CD45RA, CD62L, CCR7, CD27, CD28 and Bcl-2) and no co-inhibitory or finally differentiated markers (CD57, CTLA4, PD-1 and FasL), PD-1 having recently been described as expressed by CD8+ progenitors giving rise to dysfunctional, exhausted-like TLs (41). Moreover, part of obtained M1-CTLs expressed skin homing factors (CLA, CCR4 and CCR10). All these phenotypic features could be of great interest for melanoma immunotherapy.
The subpopulation of memory TLs expressing CD95, CD45RA and CD62L in our study harbors stem cell-like memory TL (T SCM ) features, as shown after cell sorting in Figure 5. These TLs display high proliferation, self-renewal and multipotency potentials. This has already been reported in vivo in nonhuman primates, in immunocompetent mice, and in immunodeficient mice with gene-modified human anti-tumor TLs (31)(32)(33). In the latter mouse models, all injected human T cells were gene-modified, in order to get in vitro sufficient numbers of tumor-specific cells with T SCM features. Moreover, these cells differentiated in vivo into effector T cells, without retaining self-renewal capacities. This could be due, at least in part, to the absence of expression, in these mice, of the natural human ligands of CCR7 and CD62L expressed by human T cells with T SCM features, leading to the absence of preferential migration of these cells into secondary lymphoid tissues (31,32). To our knowledge, there is no available humanized mouse model yet (e.g. immunodeficient knock-in mice expressing human CCR7 and CD62L ligands in lymphoid tissues) allowing the in vivo thorough study of human T cells with T SCM features. Nevertheless, recent studies have proven that these T cells were indeed associated with long term protection in humans (42,43).
Obtaining in a reproducible manner stem cell-like memory T cells in vitro for ACT, without any TL gene modification, remains a challenge (34). All published relevant methods, to our knowledge, aiming at amplifying in vitro tumor-specific CTLs, including MART-1specific T cells, are based on the use of autologous antigen presenting cells, allogeneic cells, K562-derived antigen presenting cells or anti-CD3/anti-CD28 beads (14,17,(44)(45)(46)(47)(48)(49). They all require multiple stimulations for obtaining large numbers of specific CTLs and lead to the expansion of central or effector memory T cells (8,50). Indeed, to our knowledge, a clinical protocol to expand numerous antigen-specific CTLs with T SCM features and without TL gene modification does not exist yet.
Interestingly, TL stimulation with our NIH/3T3-derived AAPC-based protocol allowed the generation of CTLs with not only T CM and T EM but also T SCM features in an antigen-specific context from PBTLs of all tested patients and healthy donors. Together, the different CTL populations we obtained in large numbers with our protocol could greatly improve the anti-tumor response, allowing short and long-term effects after CTL injection.
Surprisingly, IL-7 or IL-15, known to promote stem cell-like memory T cells (51), alone or in all different combinations with or without IL-2 in our co-cultures, did not lead to greater numbers of T cells with T SCM features. IL-7 maintained the percentage of T cells with T SCM features without leading to cell expansion. IL-15 maintained the percentage of T cells with T SCM features and led to their expansion, but in a way comparable to IL-2 (Supplementary Material).
Altogether, our results are probably due, at least in part, to: (i) the particular immunological synapse created between specific CTLs and our NIH/3T3 fibroblast-derived AAPCs (expressing only potent accessory molecules at a high level), (ii) the few TL manipulations that our AAPC-based protocol requires (with only two stimulations in 35 days) and (iii) the low doses of IL-2 we added to our co-cultures, which did not trigger too strong TL activation (28).
MART-1 is a particular antigen. In the peripheral blood of melanoma patients and of healthy donors, M1-CTLs can be found at a higher frequency than other tumor-associated antigen (TAA)specific CTLs and these CTLs can more easily be expanded in vitro (52). Still, the study of M1-CTLs is highly relevant since MART-1 is over-expressed in many melanoma and M1-CTLs have been used in many clinical trials. Our AAPCs have already been described as capable of expanding functional specific CTLs against other virusand tumor-associated antigens such as pp65, gp100 and hTERT (15,53). Many TAA-specific CTLs, other than M1-CTLs, harboring a naïve phenotype, have been found, although at a lower frequency than M1-CTLs in the peripheral blood of healthy donors and cancer patients (54)(55)(56). Therefore, it should be possible to obtain T cells with T SCM features for these antigens, and it will be important in the near future to test this hypothesis in order to broaden the interest of our protocol.
In conclusion, we describe a novel and unique AAPC-based protocol allowing the activation and the expansion, in a specific manner, of sufficient numbers of highly functional anti-melanoma CTLs with a suitable phenotype for immunotherapy, i.e. not only with T EM and T CM but also with T SCM features, both in patients and in healthy donors, which could allow short and long term antitumor responses in vivo. We now intend to use GMP grade AAPCs and optimized CTL purification procedure in our protocol and to confirm these in vitro results in a relevant animal model, in order to be able to propose a clinical trial based on the use of these AAPCs for melanoma patients.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by Rouen University Hospital Institutional Review Board "Comitéde Protection des Personnes Nord-Ouest I". The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
MH and J-FC performed the experiments, analyzed the data, and wrote the paper. ED, FB, EF, PM, and AD helped to perform experiments. A-BD-M provided clinical data and samples. SA and OB helped to design the study and to write the paper. J-BL designed and supervised the study, analyzed the data, and corrected the paper. All authors contributed to the article and approved the submitted version.

FUNDING
MH and J-FC's PhD fellowships were provided by "La Ligue Contre le Cancer de Seine-Maritime", France. This study was supported by "La Ligue Contre le Cancer de Haute-Normandie", France.