For this disease never took any man the second time so as to be mortal.

—Thucydides, The History of the Peloponnesian War (translation by Thomas Hobbes)

Immunological memory—the ability to remember and respond rapidly and more vigorously to a pathogen upon subsequent encounters—has long been recognized in human history. The first documentation of immunological memory came from the Greek historian Thucydides, who vividly described the plague that struck the city of Athens in 430 bc, recounting that “this disease never took any man the second time”1. It took us more than two millennia to understand that immunological memory is a fundamental property of the adaptive immunity conveyed by B and T lymphocytes2.

Despite the enormous progress in our understanding of basic aspects of T cell immunity, the ontogeny of memory T cells remains a matter of active debate3,4. It is clear, however, that immunological memory and protective immunity can last several decades and perhaps a lifetime, even in the absence of re-exposure to the pathogen5,6. This astonishing stability of T cell memory, despite the high cellular turnover that characterizes immune responses and the lack of replenishment of antigen-specific T cells from hematopoietic stem cells (HSCs)—owing to constraints imposed by stochastic recombination of the T cell receptor (TCR) and thymic involution—has sparked the idea that T cell immunity could be maintained via stem cell–like memory T cells7. Over the past decade, the realization that memory T cells share a core transcriptional signature with HSCs8 and display functional properties found in stem cells, such as the capacity to divide asymmetrically to generate cellular heterogeneity9, has further strengthened the view that T cells, akin to all somatic tissues, might be organized hierarchically and sustained by antigen-specific T memory stem cells10.

In this Review, we outline emerging findings demonstrating that a subset of minimally differentiated memory T cells behave as antigen-specific adult stem cells. We also discuss recent evidence placing these TSCM cells at center stage in many physiological and pathological human processes. Finally, we highlight ongoing efforts aimed at either harnessing the therapeutic potential of TSCM cells for adoptive immunotherapies or, conversely, at destabilizing the TSCM cell compartment to eliminate drug-resistant viral reservoirs or treat adult T cell leukemia and autoimmune diseases. The conceptual work and key discoveries that have shaped this field of investigation are summarized in Figure 1.

Figure 1: T cell stemness and TSCM cells: milestones and key discoveries.
figure 1

TSCM cells, T memory stem cells; GVHD, graft-versus-host disease; HIV-1, human immunodeficiency virus type 1; SIV, simian immunodeficiency virus; HTLV-1, human T cell lymphotropic virus type 1; CAR, chimeric antigen receptor; SNP, single-nucleotide polymorphism.

The discovery of TSCM cells

Advances in multiparameter flow cytometry over the past 20 years have enabled dissection of the heterogeneity of the T cell compartment with ever-increasing precision11. In a seminal study, van Lier and colleagues identified human naive, memory and effector T cell subsets on the basis of the combinatorial expression of CD27 and CD45RA, with naive cells expressing both molecules, and memory and effector cells expressing only CD27 or CD45RA, respectively12. Subsequent work by Sallusto et al.13 revealed the presence of two major functional subsets within the CD45RA− memory T cell pool: central memory T (TCM) cells, which express the lymph node homing molecules CCR7 and CD62L and have limited effector functions, and CCR7−CD62L− effector memory T (TEM) cells, which preferentially traffic to peripheral tissues and mediate rapid effector functions.

The idea that memory T cells might not be confined solely to the CD45RA− T cell compartment, but might also be present within what was considered to be a naive T cell population, began to take shape following the identification in mice of a memory T cell population that is characterized by a naive-like phenotype, but that expresses high amounts of stem cell antigen 1 (SCA1) and the memory markers interleukin-2 receptor β (IL-2Rβ) and chemokine C-X-C motif receptor 3 (CXCR3)14. These cells were termed TSCM cells because it was observed that they were capable of sustaining graft-versus-host disease (GVHD) upon serial transplantation into allogeneic hosts, and that they could reconstitute the full diversity of memory and effector T cell subsets while maintaining their own pool size through self-renewal14. Identifying the human counterpart of TSCM cells, however, has not been straightforward, mainly owing to the lack of a human ortholog of SCA1, the prototypical marker of mouse TSCM cells. Although it was known that a substantial fraction of long-lived antigen-specific CD8+ and CD4+ memory T cells displayed a naive-like phenotype (CD45RA+CCR7+CD27+) years after infection with EBV15 or vaccination with attenuated smallpox or yellow fever (YF) viruses16,17, a precise set of surface markers with which to pinpoint this elusive memory phenotype in humans was missing. The breakthrough came with the demonstration that mouse TSCM cells could be generated successfully in vitro from naive precursors by activating the Wnt-β-catenin signaling pathway using the Wnt ligand, Wnt3a or inhibitors of glycogen synthase kinase-3β (ref. 18). By using this strategy to characterize extensively the phenotype of candidate human TSCM cells generated in vitro, it was possible to identify key surface markers that can distinguish naturally occurring human TSCM cells from the naive T cell pool19. Similar to their mouse counterparts, human and nonhuman primate (NHP) TSCM cells are clonally expanded cells that express a largely naive-like phenotype in conjunction with a core of memory markers, such as CD95, CXCR3, IL-2Rβ, CD58 and CD11a19,20. These cells represent a small fraction of circulating T lymphocytes (≈2–3%). Notably, the frequency of circulating TSCM cells does not vary substantially with age21, but it seems to be heritable and associated with single-nucleotide polymorphisms (SNPs) at a genetic locus containing CD95 (ref. 22), which suggests a potential role of FAS signaling in the regulation of TSCM cell homeostasis. TSCM cells exhibit all the defining properties of memory cells, including a diluted content of TCR excision circles, the ability to proliferate rapidly and release inflammatory cytokines in response to antigen re-exposure, and a dependence on IL-15 and IL-7 for homeostatic turnover19,23. Despite being functionally distinct from naive T cells, TSCM cells share with them similar recirculation patterns and distribution in vivo, as evidenced by detailed compartmentalization studies in NHPs24. For instance, TSCM cells are found more abundantly in lymph nodes than in the spleen and bone marrow, and they are virtually absent from peripheral mucosae24. Thus, TSCM cells represent a subset of minimally differentiated T cells characterized by phenotypic and functional properties that bridge naive and conventional memory cells (Fig. 2).

Figure 2: Hierarchical model of human T cell differentiation.
figure 2

After antigen priming, naive T (TN) cells progressively differentiate into diverse memory T cell subpopulations, and ultimately, into terminally differentiated effector T (TTE) cells. T cell subsets are distinguished by the combinatorial expression of the indicated surface markers. As TN cells differentiate progressively into the TTE cell type, they lose or acquire specific functional and metabolic attributes. TSCM cell, T memory stem cell; TCM cell, central memory T cell; TEM cell, effector memory T cell; ΔΨm, mitochondrial membrane potential.

TSCM cells: evidence of stemness

The concept of stemness embraces the capacity both to self-renew and to generate the entire spectrum of more differentiated cells25. When Fearon and colleagues7 initially postulated the existence of a stem cell pool of memory T lymphocytes, the authors pointed to TCM cells as putative T memory stem cells. This assumption was based on the evidence that TCM cells are less differentiated than TEM and effector cells, as shown by their longer telomeres and lower expression of perforin, granzymes and other effector molecules13. Furthermore, it was intuitive to assume that the pool of T memory stem cells should be confined to lymph nodes and secondary lymphoid organs, and TCM cells were, at that time, the only antigen-experienced T cells known to express CCR7 and CD62L. The notion that TCM cells might function as T memory stem cells received further support from subsequent findings that demonstrated that TCM cells have superior immune-reconstitution capacity and a greater ability to persist in vivo than TEM cells26. Recent clonogenic experiments in mice based on single-cell serial transfer have formally demonstrated the ability of mouse TCM cells to self-renew and generate TEM and effector progeny in vivo27,28. By contrast, TEM cells were unable to serially reconstitute the host, even when transferred at 100-fold higher numbers, and so showed a limited capacity for self-renewal. Although these experiments did not evaluate TSCM cells, these results, when combined with those of sophisticated experiments tracking T cell fates in mice through genetic barcoding29 and on single naive T cell transfer30, provided strong support for the progressive model of T cell differentiation originally developed by Sallusto and Lanzavecchia31. Indeed, three separate models have been proposed to explain memory T cell differentiation3: according to the first two models, memory T cells originate from effectors either after26 or before32 the peak of T cell expansion. The progressive differentiation model, on the contrary, suggests that memory T cells are derived directly from naive lymphocytes upon priming, and further differentiate into shorter-lived effector subsets in a hierarchical differentiation tree, similarly to that of other organ systems31 (Fig. 2). By using hematopoietic stem cell transplantation (HSCT) from haploidentical donors as a model system to study T cell differentiation, two independent groups have shown recently at polyclonal, antigen-specific and clonal levels that human TSCM cells differentiate directly from naive precursors and emerge early upon in vivo priming33,34. By multiparametric flow cytometry and TCR sequencing, it was possible to trace and quantify the in vivo differentiation landscapes of transferred naive T cells, which showed that more than 30% of naive T cells undergoing priming differentiate into TSCM cells33,34. Indeed, discrete T cell subsets traced across HSCT behaved preferentially within a progressive framework of differentiation. Notably, only naive T cells and TSCM cells were able to reconstitute the entire heterogeneity of memory T cell subsets, including TSCM cells33. A fraction of cells originally designated as TEM cells reverted to a TCM cell phenotype33. By contrast, only a very limited number of TCM and TEM cells converted to the TSCM cell type33. Echoing these findings, the transfer of genetically modified virus-specific T cells reconstituted the full diversity of the T cell memory compartment— inclusive of TSCM, TCM and TEM cells—only when TSCM cells were present within the infused cell product35. All together, these results strengthen earlier in vitro observations in humans19 and NHPs24 showing that the potential to form diverse progeny is progressively restricted as the cell type proceeds from TSCM to TCM and TEM cells. Thus, granting some level of plasticity to the system, these data point to a progressive model of T cell differentiation, in which TSCM cells are at the apex of the hierarchical tree. In line with this concept, the gene expression profile of human T cell subsets partitions TSCM cells with antigen-experienced T cells and places them at a hierarchically superior level over the TCM cell type19,23,36,37.

The concept of stemness also involves self-renewal and implicates long-term persistence25. The long-term persisting ability of TSCM and other antigen-experienced T cells cannot be addressed easily in humans because naive T cells are generated continuously, and several antigenic contacts might occur after the initial encounter. Longitudinal monitoring of genetically engineered lymphocytes infused as antigen-experienced cells, and distinguishable from endogenous lymphocytes thanks to the retroviral integration and transgene expression, has recently enabled the tracking of single T cell clonotypes over time. In patients afflicted with the adenosine deaminase (ADA)-deficient form of severe combined immunodeficiency (SCID), genetically engineered TSCM cells persisted and preserved their precursor potential for decades38. Engineered lymphocytes were tracked for up to 14 years in patients with leukemia who were treated with haploidentical HSCT and donor lymphocytes transduced retrovirally to express a suicide gene35. This study revealed that the extent of expansion and the amount of persisting gene-marked T cells are tightly correlated with the number of TSCM cells infused, which indicates that this subset of memory cells is endowed with enhanced proliferative potential, immune-reconstitution capacity and longevity35. Notably, the same observation has been reported in a clinical trial based on the infusion of autologous T cells that have been genetically engineered to express a chimeric antigen receptor (CAR)39, which underscores that this phenomenon is not confined to the HSCT model. In patients treated with suicide-gene therapy, it was possible to detect circulating gene-modified T cells from 2 to 14 years after treatment. Viral integration and TCR-α and TCR-β clonal markers were used to trace longitudinally single, gene-modified T cell clones, sorted according to the T cell differentiation phenotype in the infused products and in patients, at long-term follow-up. It was thus possible to show that dominant T cell clones detected long term originate preferentially from infused TSCM cells, and to a lesser degree, from TCM clones35. Taken together, these results indicate that human TSCM cells have an exceptional capacity to persist long term. Similar conclusions were reached by monitoring T cell subset dynamics in NHP models of infection24 and patients with HIV-1 undergoing antiretroviral therapy (ART)40; two experimental settings in which antigen load and time of antigen exposure can be controlled precisely. By taking advantage of the peculiar biology of the Tat-specific epitope TL8, which uniformly undergoes escape mutation within 4–5 weeks after infection with simian immunodeficiency virus (SIV), Lugli et al.24 investigated the persistence of different memory Tat-specific T cell subsets in the virtual absence of any stimulation with antigen. In this setting, TSCM cells were able to persist at unchanged levels for up to 70 d after infection, whereas TCM and TEM cells contracted tenfold and 100-fold, respectively24. Similarly, pharmacological antigen withdrawal in ART-treated patients with HIV-1 was associated with a decline of HIV-1-specific TEM cells and terminally differentiated effector (TTE) cells, whereas the TSCM cell type gradually increased in number under these conditions40. Mirroring these findings, after YF vaccination virus-specific TTE cells underwent a more pronounced contraction than TEM cells, which in turn declined more steeply than TCM cells36. Remarkably, the frequency of YF-specific TSCM cells was maintained stably even 25 years after vaccination36. Taken together, this series of studies provides compelling evidence that human TSCM cells are generated directly from naive lymphocytes and are endowed with long-term self-renewal capacity and multipotency.

TSCM cells in host defense

Pathogen-specific TSCM cells have been increasingly identified in human acute and chronic infections caused by viruses, bacteria and parasites19,35,36,40,41,42. These results demonstrate that TSCM cells are commonly generated during natural immune responses against foreign pathogens, but the underlying mechanisms remain poorly understood. Human studies are limited in that the exact time of infection is usually unknown, which makes it difficult to study T cell priming and kinetics. By contrast, active vaccination offers the possibility of inducing an immune response in a supervised fashion. Smallpox and YF vaccines are particularly suitable models of human primary acute viral infection because they consist of live, attenuated, replication-competent viruses capable of inducing strong immune responses with consequent clinical symptoms43. By using YF vaccination as a model system, the kinetics of TSCM cell formation and long-term maintenance have recently been studied in great detail36. Consistent with findings from studies of SIV infection in NHPs24, YF-specific TSCM cells were detectable at early time points after vaccination when the immune response was dominated by effector T cells36. These TSCM cells persisted at stable levels and became the major YF-specific memory T cell population in the circulation decades after the initial immunization36. Considering that YF vaccination provides life-long protection43, it is reasonable to assume that TSCM cells have a central role in the maintenance of long-term T-cell memory.

The presence of a relevant pool of TSCM cells might also be essential for the control of persisting infections, in which effector T cells undergoing functional exhaustion and replicative senescence need to be replenished continuously by less differentiated T cell subsets44,45,46,47. Notably, recent studies in chronic viral40,48 and parasitic infections42 revealed a negative correlation between the severity of disease and the frequency of circulating TSCM cells. It is unclear whether these observations result from the inability of TSCM cells to be maintained under conditions of strong inflammation and high antigenic load, or vice versa, that the presence of insufficient numbers of TSCM cells impairs the ability of the immune system to control pathogen replication. However, emerging findings suggest that TSCM cells are crucial to the maintenance of immune homeostasis; high levels of infection of the TSCM cell compartment and its subsequent functional perturbation have been linked to the development of symptomatic immune deficiency following SIV and HIV-1 infections49,50. Indeed, high quantities of SIV DNA were found in CD4+ TSCM cells from rhesus macaques, who typically develop an AIDS-like clinical picture when left untreated, but they were not found in CD4+ TSCM cells from SIV-infected sooty mangabeys, a group of NHPs that are refractory to clinical or laboratory signs of immune deficiency even when high levels of virus circulate in the peripheral blood49,51,52. Resonating with this observation, viremic nonprogressors—a rare group of untreated patients with HIV-1 who develop high levels of HIV-1 replication in the absence of clinical immune deficiency—exhibit reduced levels of HIV-1 DNA in CD4+ TSCM cells in comparison to patients with HIV-1 who show ordinary rates of disease progression50. All together, these results underscore a crucial function of TSCM cells in the sustenance of long-lasting cellular immunity against acute and chronic microbial infections.

Given the pivotal role of TSCM cells in the maintenance of life-long immunological memory, it would be desirable to develop vaccines that are capable of inducing substantial numbers of TSCM cells. The majority of clinical vaccine formulations designed to stimulate CD8+ T cell–mediated immunity induce predominantly TEM, and few memory, cells53,54. These vaccines are rarely efficacious as compared to those that induce protective antibodies2,55. Indeed, current T cell vaccines seem inefficient at triggering mechanisms that are key for the development of memory T cells, including optimal signaling via the TCR and the induction of appropriate metabolic programs, transcription factors and chromatin reorganization56. Considering that the activation of CD8+ T cells under conditions of low-level inflammation enhances memory cell formation, one might surmise that vaccines should, ideally, stimulate T cells without triggering the excessive release of proinflammatory cytokines57. It is, however, debatable whether optimal generation of memory T cells requires the avoidance of effector cell differentiation. This is illustrated by the fact that natural infections generate sound memory T cell responses, including TSCM cells, despite the initial predominance of effector cells43. Indeed, the emergence of TSCM cells was recently observed after the administration of a subunit cancer vaccine capable of inducing a rapid and robust expansion of effector T cells58. Much work remains to be done in this area; however, the induction of TSCM cells by novel vaccines should not be at the expense of more differentiated TEM and tissue-resident memory cells, which assure immediate protection at the entry site of re-infection in peripheral tissues59,60,61. Ideally, new vaccines will be able to recreate the full spectrum of memory cell phenotypes that human pathogens and their pathophysiological properties induce in vivo62,63.

TSCM cells can exacerbate human disease

The complex biology of TSCM cells can make it difficult to discriminate between their protective and pathogenic effects because the very characteristics that enable TSCM cells to represent the backbone of life-long cellular immunity under physiologic conditions might empower these cells to drive disease pathogenesis64. This seems particularly relevant in the setting of a growing list of immune-mediated diseases associated with aberrant and autoreactive memory T cells. For instance, recent correlative studies have suggested an increased frequency and activation state of CD8+ TSCM cells in individuals with aplastic anemia, a disease mediated by autoreactive cytotoxic T cells targeting hematopoietic progenitors, as compared to healthy individuals65. Moreover, an elevated number of CD8+ TSCM cells after immunosuppressive treatment was associated with treatment failure and subsequent aplastic anemia relapse65. Elevated numbers of TSCM cells were also noted in patients with uveitis, but not in those with systemic lupus erythematosus, an immune-mediated disease characterized primarily by autoreactive humoral responses65. Further pointing toward a role of TSCM cells in the pathogenesis of autoimmune diseases and other illnesses of the lymphatic system, a recent genome-wide association study found a strong association between genetic polymorphisms affecting susceptibility to juvenile idiopathic arthritis or chronic lymphocytic leukemia, and the frequency of CD4+ TSCM cells22. How TSCM cells can influence autoimmune diseases will have to be studied in dedicated investigations, but on the basis of current knowledge, it is reasonable to hypothesize that long-lasting autoreactive or abnormally activated TSCM cells might induce self-renewing inflammatory cellular responses that are responsible for the durable, and in most cases life-long, persistence of such diseases66. The possible role of TSCM cells in other diseases with profound disturbance of cellular immune responses, such as autoimmune hepatitis, thyroiditis, type 1 diabetes and certain types of glomerulonephritis, are currently unknown but represent a high priority area of future research.

In addition to their role in autoimmunity, TSCM cells might have a distinct role in viral diseases in which T cells represent the predominant targets, such as infections caused by CD4+ T cell tropic retroviruses. Notably, work in the context of HIV-1 infection has shown that CD4+ TSCM cells can effectively support both productive viral replication and a transcriptionally silent form of infection67. Moreover, by infecting long-lived CD4+ TSCM cells, HIV-1 is able to exploit their stemness to establish an extremely durable, self-renewing viral reservoir that can persist for decades, despite ART, and continuously replenish virally infected cells, thus perpetuating a disease that they are meant to restrict68. Indeed, the half-life of HIV-1-infected TSCM cells in ART-treated individuals has been estimated to be around 277 months, a time period substantially longer than that observed for viral reservoirs established in more short-lived T cell populations69. In line with these observations, phylogenetic studies demonstrated close associations between viruses circulating early after HIV-1 infection and viral sequences isolated from CD4+ TSCM cells after almost a decade of suppressive ART68. Notably, the ability to use CD4+ TSCM as a long-term viral reservoir also seems to occur in individuals infected with HTLV-1, a retrovirus related to HIV-1 that is the primary cause of adult T cell leukemia (ATL). Emerging data indicate that transformed, HTLV-1 infected CD4+ TSCM cells can act as progenitors for dominant circulating ATL clones and can efficiently propagate ATL upon transplantation in animal models70. This suggests that they can serve as a cancer stem cell population responsible for the propagation and maintenance of HTLV-1-infected malignant cells.

Targeting TSCM cells for therapy

Harnessing TSCM cells for adoptive T cell therapy. The extreme longevity, the robust proliferative potential and the capacity to reconstitute a wide-ranging diversity of the T cell compartment make the TSCM cell type an ideal cell population to employ in adoptive immunotherapy (Fig. 3). Driven by the growing success of clinical trials that are based on the transfer of naturally occurring and genetically engineered tumor-reactive T lymphocytes, adoptive immunotherapies are rapidly becoming a real therapeutic option for patients with cancer71,72. Although these regimens can induce complete and durable tumor regressions in patients with advanced cancer, current response rates remain mostly inadequate, which underscores the need for further improvements71,72. There is now extensive evidence indicating that objective responses are strongly correlated with the level of T cell engraftment and peak of expansion early after transfer73,74,75,76,77,78,79. T cell persistence, although not strictly indispensable in certain conditions74,75,76,77,80, has also been associated with the likelihood of objective responses in numerous trials78,79,81,82,83,85 and might be required to sustain durable remissions86. These parameters are influenced considerably by the composition of the infused T cell product because T cell subsets differ widely in terms of proliferative capacity, immune reconstitution and long-term survival10,87. Indeed, the administration of cells with longer telomeres83,88 or cell products comprising higher fractions of CD62L+, CD28+ or CD27+ T cells has been shown to correlate with objective tumor responses in patients83,88,89,90, which suggests that less differentiated T cells are therapeutically superior to TTE cells. Notably, the engraftment and expansion of T cells engineered to express a CD19-specific CAR39 or a suicide gene35 was correlated with the frequency of infused CD8+CD45RA+CCR7+ TSCM cells. Adoptive transfer experiments in mice, using defined T cell subsets, have demonstrated formally that the infusion of less-differentiated CD62L+ T cell populations results in enhanced T cell engraftment, expansion and persistence, which leads ultimately to more profound and durable tumor regressions18,19,91,92,93,94,95. Consistent with the developmental hierarchy, minimally differentiated TSCM cells mediate more potent antitumor responses than TCM cells, which, in turn, are more effective than highly differentiated TEM cells18,19,96. Some level of plasticity, however, must be granted to the hierarchical model of memory T cell differentiation. In NHPs, genetically engineered CMV-specific effectors derived from purified TCM cells proved superior to effectors derived from TEM cells in terms of expansion and persistence in vivo, which shows that even after manipulation in vitro and, apparently, a similar degree of terminal differentiation, T cells maintain some characteristics of the subset of origin, and can possibly, at least in part, revert to that original phenotype and function97.

Figure 3: TSCM-cell-based therapeutic interventions for human diseases.
figure 3

TSCM cells can be either disrupted (left) to treat TSCM-driven diseases, such as autoimmunity, T cell leukemia and T cell tropic infections, or exploited (right) to potentiate T cell–based immunotherapies against cancer and infectious diseases. Left, Wnt antagonists or short hairpin RNA (shRNA) targeting key molecules involved in Wnt signaling, such as T cell factor 7 (TCF7) could be used to disrupt long-lasting, self-renewing TSCM cell reservoirs by driving them to differentiate into short-lived subsets, such as TEM cells. Nanoparticle or aptamer technology could be employed to target CD4+ T cells or virally infected T cells specifically. Right, patient- or donor-derived naive-like T cells can be used to generate and expand TSCM cells in vitro with or without gene engineering. Gene modifications include the insertion of tumor or virus-specific chimeric antigen receptor (CAR) or T cell receptor (TCR) genes, tumor or virus-specific TCR gene editing, suicide-gene transfer for the option to eliminate the transferred TSCM cells and their progeny in case of overwhelming toxicity, and CCR5 gene editing in the setting of HIV-1 infection. Virus-specific TSCM cells can also be expanded from the naturally occurring antigen-specific TCR repertoire through sensitization protocols in vitro favoring the generation of TSCM cells. APC, antigen-presenting cell.

Despite overwhelming preclinical data indicating a therapeutic advantage to transferring tumor-reactive CD62L+ T cell subsets18,19,91,92,93,94,95, clinical trials have largely employed unselected intratumoral or peripheral blood mononuclear cell (PBMC)-derived T cell populations. Tumor-infiltrating lymphocytes are typically in a state of terminal differentiation and functional exhaustion, which makes the isolation of early memory T cell subsets impractical98,99. However, the selection of less differentiated T cell subsets becomes realistic and desirable in the context of immunotherapies that are aimed at conferring tumor reactivity to circulating T cells via TCR or CAR gene engineering. The isolation of less differentiated T cell populations also has the advantage of reproducibly generating more defined T cell products. Indeed, PBMC composition can vary substantially between individuals as a consequence of age100, pathogen exposure101 and prior systemic treatments102. Moreover, unselected populations containing high proportions of TEM and effector cells might fail to generate viable clinical products owing to poor in vitro cell expansion103. Recently, a few clinical trials in which CD19-specific CAR T cells were generated from isolated TCM cells have been reported86,104,105. This strategy led to the generation of infusion products composed of substantially more TEM cells than those originating from unselected PBMCs, which indicates that, in the absence of culture conditions restraining T cell differentiation18,106,107,108,109,110, the benefit of depleting highly differentiated T cell subsets is outweighed by the concomitant removal of naive and TSCM cells104. Notwithstanding the reduction of less differentiated T cell subsets, the rates of objective remissions in patients with acute lymphoblastic leukemia (ALL) were comparable to results of trials that used unselected T cell populations74,75,78,104,111,112. Whether differences in manufacturing and T cell product composition will affect the rates and duration of clinical responses in other diseases and settings remains to be shown.

So far, the clinical exploitation of TSCM cells has been hindered by their relative paucity in the circulation>19,20 and the lack—until recently—of robust, clinical-grade manufacturing protocols that are capable of generating and maintaining this cell type in vitro. These strategies rely on programming and redirecting TSCM cells from naive-like T cells isolated from PBMCs23,113 (Fig. 3). Although the isolation of naive T cells adds complexity to the manufacturing process, it is a crucial step because the presence of more differentiated T cell subsets during naive T cell stimulation accelerates naive T cell differentiation into TEM and TTE cells114. It should also be considered that purifying large numbers of specific cell subsets over multiple parameters under good manufacturing practice conditions is becoming increasingly accessible thanks to recent developments in clinical cell-sorting technologies87,115. IL-7 and IL-15 have been used successfully to generate tumor-redirected or suicide-gene-modified TSCM cells from naive cell precursors23 (Fig. 3). IL-7 is essential for the development of these cells23,116, whereas IL-15 primarily sustains their expansion23. IL-7 and IL-15-programmed TSCM cells possess a core gene signature of naturally occurring TSCM cells, display an enhanced proliferative capacity as compared to other T cell subsets and are uniquely capable of expanding and mediating GVHD upon serial transplantation23. This cytokine combination could also be employed to generate large numbers of TCR-gene-edited TSCM cells by combining zinc-finger nuclease sets specific for the endogenous TCR gene loci with viral vectors encoding tumor-specific TCRs117 (Fig. 3). Moreover, the ability of IL-7 and IL-15 to support the formation and expansion of TSCM cells makes it an ideal strategy for generating TSCM cells without the need to redirect their specificity. This might be particularly suitable for the generation of virus-specific TSCM cells for the treatment and prevention of life-threatening infections after transplantation (Fig. 3), given that infection control can be obtained by transferring relatively small numbers of virus-specific memory cells118. A demonstration that IL-7 and IL-15 could be employed successfully to generate and expand virus-specific TSCM cells, starting from isolated naive-like cells, was provided recently by Volk and colleagues119. This protocol could also be adapted to generate CAR-modified virus-specific TSCM cells, which might lower the risk of GVHD, given the restricted TCR repertoire, and which may exhibit additional proliferative and survival advantages as a result of the triggering in vivo of the native virus-specific TCRs by antigens from persistent viruses82,120. Another clinical-grade strategy promoting the generation of tumor-reactive TSCM cells is based on the activation of naive-like lymphocytes in the presence of IL-7, IL-21 and the Wnt agonist TWS119 (ref. 113). Although both IL-15 (refs. 121,122) and IL-21 (refs. 123,124,125) have been implicated in the generation and maintenance of memory T cells, IL-21 is more effective in restraining T cell differentiation107, owing to its specific ability to activate signal transducer and activator of transcription 3 (STAT3) signaling126 and to sustain the expression of the Wnt-β-catenin transcription factors TCF7 and LEF1 (ref. 107). TWS119 has a synergistic effect with IL-21 to induce maximal expression of TCF7 and LEF1 by stabilizing β-catenin113. CAR-modified TSCM cells generated under these culture conditions are phenotypically, functionally and transcriptionally equivalent to their naturally occurring counterparts113. Moreover, they exhibit metabolic features, such as a high spare respiratory capacity127 and low glycolytic metabolism128, that are characteristic of long-lived memory T cells. Although these culture conditions profoundly inhibit T cell proliferation, TSCM cells can be redirected efficiently against a tumor antigen and expanded to clinically relevant numbers113. More importantly, CAR-modified CD8+ TSCM cells mediated superior and more durable anti-tumor responses than cells generated with protocols currently employed in clinical trials113. CAR-modified TSCM cells might also provide an attractive approach for immunotherapy in the setting of nonmalignant diseases, such as HIV-1 infection or other chronic viral illnesses129,130 (Fig. 3). All together, these studies provide both a strong scientific rationale and practical methodologies for the rapid advancement of TSCM cells in human clinical trials of adoptive immunotherapy131.

Disrupting TSCM cell reservoirs in retroviral infections and autoimmune diseases. The emerging role of CD4+ TSCM cells in the pathogenesis of chronic viral infections such as HIV-1 and HTLV-1 infection might also offer novel opportunities to prevent, treat or cure these diseases. In the context of HIV-1 infection, specific interventions that eliminate HIV-1-infected CD4+ TSCM cells might allow for the destabilization of HIV-1 reservoirs by reducing the number of HIV-1-infected source cells from which new HIV-1+ viral and cellular progeny can continuously originate, despite suppressive ART. As the molecular programs that govern the stem cell–like behavior of TSCM cells continue to be understood, new molecules regulating proliferation and self-renewal of TSCM cells might represent attractive targets for reducing viral persistence in CD4+ TSCM cells. For instance, Wnt-β-catenin signaling has been identified as a key driver for the homeostasis of TSCM cells18, and pharmaceutical inhibition of this pathway might therefore translate into a more limited ability of HIV-1 to use the TSCM compartment for maintaining the survival of virally infected cells (Fig. 3). This approach might be facilitated by the availability of existing pharmacological inhibitors of Wnt-β-catenin designed to target cancer stem cells132. Although such a strategy might be not entirely specific to the elimination of HIV-1-specific CD4+ TSCM cells, advances in nanotechnology might enable selective delivery of Wnt-β-catenin antagonists or short hairpin RNAs targeting key mediators of Wnt signaling to CD4+ T cells or virally infected cells via nanoparticles or aptamer-based targeting systems133,134 (Fig. 3). Similar strategies are also conceivable for targeting HTLV-1-infected TSCM cells in the setting of ATL or to disrupt long-lasting reservoirs of autoreactive TSCM cells in autoimmune diseases. Additionally, recent advances in gene editing ex vivo might enable the design of CD4+ TSCM cells that are intrinsically resistant to HIV-1, through, for example, targeted deletion of the chemokine receptor CCR5, which is necessary for viral entry135, thus mimicking the CCR5Δ32 mutation known to confer resistance to HIV-1 infection136 (Fig. 3). Such a population of long-lasting, HIV-1-resistant CD4+ T cells could be used in adoptive immunotherapy strategies to establish a durable cellular immune system that is no longer able to support HIV-1 infection and that might lead to drug-free remission of HIV-1 infection.

Concluding remarks

TSCM cells are rare, antigen-experienced T cells, probably generated directly from naive lymphocytes and endowed with long-term self-renewal capacity and multipotency. Compelling evidence in mice, NHPs and humans points toward a scenario in which TSCM cells represent the apex of the memory T cell differentiation tree. Their longevity and their capacity to reconstitute the entire heterogeneity of the T cell memory compartment entail a double-edged—protective or pathogenic—role for TSCM cells in human diseases. Their increasingly recognized protective role in acute and chronic infections makes them optimal candidates for therapeutic exploitation in vaccination and adoptive T cell therapy against infectious diseases and cancer. Conversely, their relevance in the pathogenesis of autoimmunity, adult T cell leukemia and HIV-1 makes them an attractive target to tame for these pathological conditions. Several issues regarding TSCM cell biology remain to be addressed: characterization of their metabolic requirements, epigenetic and transcriptional programs and anatomical niches (Box 1) will guide innovative TSCM cell-based therapeutic interventions for human diseases.