The traditional view of the immune system distinguishes innate immunity from adaptive or acquired immunity. Innate immunity is derived from cells expressing the receptors specific for molecules from microbial pathogens called pathogen-associated molecular patterns or self-molecules from the healthy or unhealthy individual called damage-associated molecular patterns. One of the key characteristics of innate immunity effector mechanisms is their capacity for very rapid response to pro-inflammatory cytokines such as IL-12, IL-18, and IL-33.

The effectors of adaptive immunity possess receptors characterized by highly diverse and specific antigens. A major feature of adaptive immunity consists in its serving as an essential support for the immunologic memory, which means that it can remember and quite effectively respond to an antigen long after having encountered it for the first time.

However, numerous works over the past 20 years have shown the distinction between innate immunity and adaptive immunity to be less clear-cut and more tenuous than it first appeared. This revised perception is based, in particular, on description of non-conventional T cells responding to stimuli that had previously been considered as being recognized solely by innate cells. These populations of T lymphocytes include not only certain T-cell receptor (TCR)-γδ cells but also TCR-αβ cells such as natural killer T (iNKT) cells and innate mucosal-associated invariant T (MAIT) cells [for a list of the different cells, see Ref. (1, 2)].

A new contingent of innate T cells was described in the early 2000s in the mouse, partially in the thymus, where they were termed «innate memory» (IM) CD8(+) T cells, and partially in the spleen, where they were termed «virtual memory» (VM) CD8(+) T cells (3, 4). Aside from possessing a phenotype of activated memory cells, one characteristic of these cells consists in their differentiating into memory cells independently of a foreign antigen. In parallel, CD8(+) T cells in humans were described as cells possessing innate characteristics including NK markers. They were found in human cord blood, a finding consistent with the hypothesis that their development does not depend on foreign antigens. These cells hence were termed NK-like CD8(+) T cells.

At the outset of this review, we shall compare the human NK-like CD8(+) T cells with IM/VM CD8(+) T cells in mice. On the basis of this comparison and with regard to humans, we shall focus first on expression of the transcription factor Eomesodermin (Eomes) as a lineage marker of that population of cells, and then on their innate functions (cytotoxicity and TCR-independent IFN-γ expression), along with their memory phenotype, and on the roles of IL-4- and promyelocytic leukemia zinc-finger factor (PLZF)-expressing T cells in differentiation of these cells, hereafter referred to as innate CD8(+) T cells. We shall discuss the use of membrane markers, particularly the α4-integrin CD49d, in order to obtain a more well-defined phenotype correlating with their functions and/or explaining their possible physiological role. Finally, we shall discuss the implication of innate CD8(+) T cells in anticancer immunity in humans.

Studies conducted shortly after 2000 by Forman et al. (5, 6) demonstrated the existence of CD8(+) T cells producing IFN-γ in response to innate signals occurring independently from the TCR. These CD8(+) T cells possessed a CD44(+) CD62L(−) CD122(+) memory phenotype and, in vivo, provided protection against Listeria monocytogenes (LM) infection (5–7). Their mobilization depended on the production of IL-12 and IL-18. Interestingly, the Forman team subsequently showed that this cell population was present in the thymus of C57BL/6 wild-type mice and that it was enriched in C57BL/6 H-2 K^b−/−D^b−/− mice not having undergone stimulation by foreign antigens (7).

A second series of studies having to do with Itk^−/− (inducing T cell kinase), Rlk^−/− (resting lymphocyte kinase), or Itk^−/−Rlk^−/− mice led to identification of a population of thymic CD8(+) T cells expressing an activated memory [CD44(+) CD62L(−) CD122(+)] phenotype and called IM CD8(+) T cells (8–12). Interestingly, these cells developed in the thymus and are exported to the spleen and the lymph nodes (LNs) where they could fulfill an anti-infective function against LM, particularly through production of IFN-γ following TCR-independent stimulation by IL-12 and IL-18. An important point in the studies dedicated to this population has been the demonstration that its differentiation depended on expression of the Eomes transcription factor and IL-15 (13–15). Eomes expression initiates the differentiation program of these cells and induces the expression of CD122 (the β chain of the IL-2 and IL-15 receptor). IL-15 has a critical role in the expansion of IM CD8(+) T cells (8, 16). It should also be noted that this population is present in Itk^−/−K^b−/−D^b−/− mice, a finding suggesting that at least some IM CD8(+) T cells are selected by non-classical major histocompatibility complex (MHC) class I molecules (9, 17–19).

A final series of studies has described in mouse spleens and peripheral lymphoid organs a population of activated [CD122(+)] and memory [CD44(+) CD62L(−)] CD8(+) T lymphocytes of which the differentiation into memory cells occurs independently of any recognition of a foreign antigen. This population consists in the so-called VM CD8(+) T cells. As is the case with IM T CD8(+) cells, Eomes and IL-15 are the two key factors in their differentiation. This population is capable of producing IFN-γ in response to innate stimulation by IL-12 + IL-18 (11, 20–22).

The expression by mouse thymic IM CD8(+) T cells of some integrins, such as CD49d [an α4-integrin, or VLA-α4, which is most often matched with a β7-integrin (which is bound to Madcam and VCAM-1 or CD106), or a β1-integrin], has been described (14, 21, 23, 24). Hence, CD49d is used to discriminate between IM and VM T cells (Figure 1A) arising from the thymus or the spleen, respectively. In this model, a possible filiation link between IM CD8(+) and VM CD8(+) T-cell populations remains to be investigated. A gating strategy to identify IM/VM CD8(+) T cells in mice, taking Eomes, CD44, and CD122 together with CD49d as delineating markers is depicted in Figure 1B and Figure S1 in Supplementary Material. Based on this gating strategy, in the C57BL/6 mouse strain, the vast majority of memory [CD44(+) Eomes(+) CD122(+)] CD8(+) T cells are IM CD8(+) T cells in the thymus vs. VM CD8(+) T cells in the spleen, as attested by their differential CD49d expression. Interestingly, in the BALB/c strain, thymic IM CD8(+) T cells are minority cells among memory CD44(+) CD122(+) Eomes(+) cells, raising the question of a possible link between IM and VM CD8(+) T cells, as well as the association of CD49d with the innate functions of IM/VM CD8(+) T cells (see CD49d, a Functional Marker of Innate CD8(+) T Cells in Humans).

IL-15 plays a key role in the homeostatic expansion of CD8(+) T cells, and it has been reported in several studies that this cytokine is implicated in differentiation into VM CD8(+) T cells after homeostatic proliferation (4, 25–28). However, August and his team have shown that the size of the IM/VM CD8(+) T population is not modified by T-cell depletion prior to bone marrow transplantation, a finding suggesting that lymphoid precursors are differentiated into IM/VM CD8(+) T cells following “tuning” by cells expressing MHC class I molecules. Moreover, in the same study, IM/VM CD8(+) T cells are distinguished from homeostatic proliferation CD8(+) T cells by a different transcriptional profile (29). Another study suggests that contrary to naive CD8(+) T cells and homeostatic proliferation CD8(+) T cells, acquisition of the phenotype of IM CD8(+) T cells necessitates engagement of their TCR (30).

In parallel with the previously described work on mice, several studies conducted in the early 2000s demonstrated the existence in humans of CD8(+) T cell-expressing markers and receptors of NK cells including CD56, KIR, NKG2A and NKG2C (CD159a and c), and CD94 (31–36). Several studies precisely characterized the phenotype of the KIR(+) CD8(+) T cells and showed that they possess an EMRA memory phenotype [CD45RA(+) CCR7(−) CD57(+)] (33, 34). Finally, Björkström et al. (33), showed that EMRA(+) KIR(+) CD8(+) T cells have a skewed repertoire using fewer different Vβ than their EMRA(+) KIR(−) CD8(+) T cell counterpart, a finding suggesting the role of antigenic pressure in the acquisition of this phenotype. An equivalent to this population of KIR/NKG2(+) CD8(+) T cells is present in human cord blood, where they possess an EMRA memory phenotype and rapidly express IFN-γ following TCR stimulation (36). In fact, this is a population of T cells that has been educated and has differentiated in the absence of foreign antigenic stimulation, into terminal effector memory T cells.

These KIR(+) CD8(+) T cells have a weaker response to TCR stimulation than their KIR(−) CD8(+) counterparts with regard to the expression of IFN-γ and TNF-α or to the degranulation evaluated by CD107a staining. The KIR(+) CD8(+) T cells expressing two different KIRs have a weaker response to TCR stimulation than cells expressing a single KIR or without KIR (33, 34). Remarkably, this CD56(+) [or KIR(+)] CD8(+) T cell subset responds quite effectively to innate stimuli, one example being the association of IL-12 with IL-18 (37). The same team reported that loss of this function in IL-12R^−/− patients is associated with a risk of severe infections from intracellular germs, particularly mycobacteria and Salmonella (37).

The study by Jacomet et al. (38) showed that KIR(+) and/or NKG2A(+) CD8(+) T cells preferentially express Eomes. These KIR/NKG2A(+) Eomes(+) CD8(+) T cells have a memory phenotype and share functional and phenotypic features (4, 38) with IM CD8(+) T cells in mice (8–11). Moreover, with respect to KIR/NKG2A(−) CD8(+) T cells, they possess an EMRA phenotype [CD45RA(+) CCR7(−)] and preferentially express the surface molecule CD57, which is a terminal differentiation marker (Figure 2A). Figure 2B shows a gating strategy designed to analyze these cells in human blood. Remarkably, the majority (around 60–70%) of KIR/NKG2A(+) Eomes(+) CD8(+) T cells express the T-box transcription factor T-bet, a phenotype comparable to the phenotype described for mouse VM CD8(+) T cells (22). These cells are characterized by increased frequency of CD16(+) cells in comparison to conventional memory [KIR/NKG2A(−) Eomes(+)] CD8(+) T cells. Moreover, CD16 expression is substantially increased in KIR/NKG2A(+) Eomes(+) CD8(+) T cells following 48 h of IL-15 in vitro treatment (Figure 3A).

Together with their marked NK-like phenotype, a hallmark of KIR/NKG2A(+) Eomes(+) CD8(+) T cells is their capacity to rapidly produce large amounts of IFN-γ in response to innate-like stimulation by IL-12 + IL-18 (38). So it is that among the CD8(+) T cells, 60–70% of those possessing a capacity for innate response to IL-12 + IL-18 are KIR/NKG2A(+) Eomes(+) CD8(+) T cells. Moreover, the frequency of the cells producing IFN-γ in response to IL-12 + IL-18 stimulation is four times greater than that of the same cells stimulated with anti-CD3 and anti-CD28 agonistic monoclonal antibodies (mAbs) (38). In the same way, these cells possess a cytotoxic arsenal: perforin and granzyme B. They possess an innate cytotoxic potential induced by an anti-CD16 antibody and revealed by the CD107a test, which assesses degranulation (38). In the final analysis, these data demonstrate that KIR/NKG2A(+) Eomes(+) CD8(+) T cells display innate activity by responding to innate stimuli with response efficacy greater than that of their response to adaptive stimuli (i.e., via their TCR), as has also been shown for innate T-cell populations such as dendritic epidermal T cells (39).

A final element appearing to favor the innate character of KIR/NKG2A(+) Eomes(+) CD8(+) T cells is their differentiation without any exogenous antigenic stimulation, as it is possible to show the existence of these cells in the fetal thymus (40), and as our team has shown them to be present in cord blood. Interestingly, in cord blood, they express an EMRA phenotype with a lower CD57 cell frequency than in the adult, suggesting that there exist supplementary steps in the terminal maturation of these peripheral cells (38). Functionally, as is the case in adult cells, these KIR/NKG2A(+) Eomes(+) CD8(+) T cord blood cells express IFN-γ after innate-like stimulation by IL-12 + IL-18.

All in all, we described for the first time a CD8(+) T cells population with innate features by associating Eomes and KIR/NKG2A markers with the capacity to respond to IL-12 + IL-18 stimulation. These features validate the KIR/NKG2A(+) Eomes(+) CD8(+) T-cell compartment as a new member of the innate T cell family in humans, and we have termed them innate CD8(+) T cells (41).

There exist a number of hypotheses on the mechanisms involved in the reprogramming of conventional T lymphocytes into innate CD8(+) T lymphocytes. One of these assumptions is based on a cross talk between IL-12R and TCR signalosome, in which IL-12 recruits Tyk2 and Fyn tyrosine kinases to activate CD3ζ–TCR signal transduction pathways (42).

Some studies conducted in mice suggest that IM and/or VM CD8(+) T cells could be selected by non-classical MHC class I molecules (9, 17–19). In humans, our results (Figure S2 in Supplementary Material) show that a relatively small fraction of KIR/NKG2A(+) Eomes(+) CD8(+) cells are MAIT cells. This finding suggests that at least some human innate CD8(+) T cells could be selected by non-classical class I MHC molecules, and it raises the possibility of the presence in this population of cells being restricted by non-classical MHC class I HLA-E molecules (18).

We have sought out other markers of innate CD8(+) T cells in humans. Among them, we tested CD56, a marker in humans associated with NK cells, but our results showed that this marker is no more effective in distinguishing the innate CD8(+) T-cell population than the KIR/NKG2A markers. More precisely, this marker delineate only 20–30% of the CD8(+) T cells expressing IFN-γ after innate stimulation by IL-12 + IL-18 (as opposed to the approximately 70% exhibited by the KIR/NKG2A markers) [data not shown; (38)]. We also tested CD161, of which the expression is a common feature of human innate T cell subsets including iNKT cells, TCR-γδ T cells, and MAIT cells (43). However, as for CD56, CD161 is expressed by only approximately 20% of the Eomes(+) KIR/NKG2A(+) CD8(+) T cells (Figure 3B).

Reasoning by analogy with the mouse model, we tested CD49d expression by innate CD8(+) T cells in humans (Figures 4A,B). Interestingly, the CD8(+) T cells with the KIR/NKG2A(+) Eomes(+) phenotype strongly expressed CD49d, as compared to the CD8(+) T-cell population taken as a whole (Figure 4A). Moreover, the majority (almost 70%) of the cells with a more pronounced expression of CD49d consisted in those expressing IFN-γ after innate-like stimulation by IL-12 + IL-18 (Figure 4B). These results show CD49d to be closely associated with the innate effector functions of the innate CD8(+) T cells. On the other hand, CD49d cannot substitute for Eomes and KIR/NKG2A, given the fact that only 20–40% of CD49d(+) cells are actually innate CD8(+) T cells (Figure S3 in Supplementary Material). However, as innate CD8(+) T cells arise from the thymus, the functional link between higher CD49d expression and IFN-γ secretion in response to IL-12 + IL-18 stimulation should be tested in cord blood, which is an accessible source of cells providing an approximate reflection of thymic cells in humans.

These results have led us to inquire about the biological meaning of CD49d expression by thymic vs. splenic (or IM vs. VM) CD8(+) T lymphocytes in mice. The results of several teams (14, 21) suggest that contrary to thymic IM CD8(+) cells, splenic (so-called VM) innate CD8(+) T cells do not express CD49d (14, 21) (Figure 1). We have observed, as in humans, that a large majority of splenic CD8(+) T cells expressing IFN-γ after innate-like stimulation by IL-12 + IL-18 were those harboring the IM [Eomes(+) CD44(+) CD122(+)] phenotype (Figure 4C). Moreover, these IFN-γ-producing cells were mostly those that more pronouncedly expressed CD49d as compared to the total CD8(+) T-cell population (Figure 4D).

Taken as a whole, these different results suggest that CD49d is associated with the innate-like functions of NK-like CD8(+) T cells as much in humans as in mice. According to the β chain with which CD49d is matched, the cells are variably liable to migrate toward different territories. While the α4β7-integrin is associated with the migration of T lymphocytes toward the digestive mucosa, the α4β1-integrin is associated with migration toward the oral mucosa, the salivary glands, the vaginal mucosa, and the central nervous system (44, 45). Indeed, therapeutic targeting of CD49d is used in treatment of multiple sclerosis (MS). Several studies have documented the implication of CD49d in the penetration of pathogenic T lymphocytes during MS or experimental allergic encephalomyelitis (EAE) (46, 47). Taken together, these different observations raise the possibility of the implication of the innate CD8(+) T cells in the pathogenesis of MS/EAE.

In mice, differentiation of IM CD8(+) T lymphocytes depends on soluble factors such as type I IFN and the IL-4 and IL-15 cytokines.

Several studies have shown that IL-4 favors the arising of VM CD8(+) T cells (48–51). More recent studies confirm the involvement of IL-4 in IM/VM CD8(+) T cell generation with a more critical role being assumed in the BALB/c (22, 30) than in the C57BL/6 mouse strain (14). IL-4-producing PLZF(+) T cells, including at least partially iNKT lymphocytes (52, 53), elicit the generation of IM CD8(+) T cells. Other results from the Hogquist group confirm the link between iNKT PLZF(+) cells, IL-4, and IM CD8(+) T cell generation by comparing different mouse strains (22). Figure 1B shows the higher frequency of T CD8(+) IM in the thymus of BALB/c in comparison with C57BL/6 mice. However, the August group shows results suggesting that IL-4 produced by differentiating CD4(+) CD8(+) double-positive thymocytes controls the generation of IM CD8(+) thymocytes in the Itk^−/− C57BL/6 background (16, 29). As concerns VM CD8(+) T lymphocytes, IL-4 is likewise at least partially implicated. IL-4 is likely to act by eliciting Eomes expression, of which the action would entail a heightened level of expression of CD122, and thereby sensitize the cells to IL-15 (14, 16).

Type I IFNs such as IFN-β and/or IFN-α (24) could favor the differentiation of lymphocytes to VM and/or IM CD8(+) T lymphocytes. The underlying mechanism described by the authors is the induction of Eomes expression by naive CD8(+) T cells.

Finally, IL-15 is a determining factor in the maintenance and/or homeostatic expansion of IM and VM CD8(+) T cells (8, 14, 16).

Few studies have addressed the role of these factors in the development of innate CD8(+) T cells in humans (38, 40). Interestingly, our own results (38) have shown a relation of proportionality between PLZF expression in T and iNKT lymphocytes and Eomes expression by innate CD8(+) T cells in cord blood, suggesting that iNKT cells control the differentiation of human innate CD8(+) T cells. Our results in vitro (41) suggest that IL-4 favors Eomes expression and the generation and/or expansion of human innate CD8(+) T cells.

There exist only sparse data in either humans or mice describing the functions of innate CD8(+) T lymphocytes. An initial set of results consisted in a demonstration in mice of a protective role of IM/VM CD8(+) T lymphocytes against viral (30, 54) and bacterial (11) infections. Our studies have been focused on the numeric/functional status of innate CD8(+) T lymphocytes during the multistep development of human tumors.

Taken together, having availed ourselves of different types of empirical data, we propose that in humans, innate CD8(+) T lymphocytes constitute a new cellular component that could have a role in antitumor immunity. Our results during CML and ovarian cancer are in favor of the existence of an axis composed of innate T cells with an antitumoral potential, which consist of iNKT cells and innate CD8(+) T lymphocytes. Differentiation of these unconventional CD8(+) T cells in humans is associated with Eomes expression and could depend on IL-4 and PLZF(+) iNKT cells. Our results suggest that during CML, these innate CD8(+) T lymphocytes could be controlled by immune checkpoints. While the results of our studies on solid cancers corroborate our hypothesis concerning the role of innate CD8(+) T lymphocytes in antitumor immunity, as of now we are unable to determine whether this role is protective, permissive, and/or detrimental in these other types of cancer.

All patients gave informed consent in a written form in accordance with the recommendations of “the Declaration of Helsinki” for participation in the study, which was approved by the scientific committees of the Clinic Investigator Center Inserm CIC-1402 (Poitiers, France), the Biological Resource Center of CHU of Poitiers (NF S96-900 certification since February 2014), the Biological Resource Center of CHU of Rennes (NF S96-900, certification since May 2009), and the Institute Curie Hospital (Paris, France).

AB, EC, and FJ designed the experiments, performed the experiments, analyzed and interpreted the data, and wrote the manuscript. LL, MA, NN, NP, BM, and SB contributed to sample preparation from patients and healthy controls, designed the experiments, performed the experiments, and analyzed and interpreted the data. EP, VC, and VL provided clinical samples and contributed to the interpretation of data. AH and J-MG together were responsible for the overall study design, supervised the project, and took primary responsibility for writing the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.