γδ T, NKT, and MAIT Cells During Evolution: Redundancy or Specialized Functions?

Mammalian immune systems can fight a wide range of threats, from foreign attacks to metabolic or epithelial homeostasis imbalances. To achieve this, lymphocytes differentiate into a variety of effectors that play overlapping but nonredundant roles. On one end of the spectrum, innate lymphoid cells (ILCs) mostly reside in tissues, are activated by conserved determinants through germline encoded receptors, and are poised to produce rapid effector responses. On the other end, mainstream T cells patrol through the body and specifically recognize very diverse Ags thanks to their somatically recombined Ag receptors assembled during development, the TCRs. After priming in secondary lymphoid organs, their activation leads to the generation of delayed but highly specific effector responses adapted to both the quality and the intensity of the threats. In-between mainstream αβ T cells and ILCs are subsets of immune effectors collectively named “innate-like T cells.” Although generated by somatic rearrangements similar to mainstream T cells, their TCRs are less diverse than those of mainstream T cells. These innate-like T cells acquire functional features corresponding to specific differentiation programs and tissue locations during their development, without requiring priming in secondary lymphoid organs.

In jawed vertebrates, two kinds of T cells are distinguished by the nature of their TCRs, αβ and γδ. Most αβ T cells are mainstream adaptive T cells, whereas most γδ T cells are innate-like T cells. Moreover, evolutionarily conserved subsets of innate-like αβ T cells named “preset αβ T cells” have been identified in mammals (1) and amphibians (2). The term “preset” stems from the acquisition in the thymus of specific effector functions and homing to specific tissues by selection on hematopoietic cells and not on epithelial cells (1). Unlike mainstream αβ T cells that recognize peptides presented by polymorphic MHC (MH) molecules, mammalian preset αβ T cells recognize nonpeptidic compounds presented by nonpolymorphic MH1-like molecules, CD1d and MR1. iNKT cells recognize glycolipids presented on CD1d, and mucosal-associated invariant T (MAIT) cells recognize 5-amino-ribityl-uracil derivatives generated during vitamin B₂ synthesis in bacteria and yeasts on MR1. The mechanism of antigenic recognition by γδ T cells is distinct from the αβ TCR one, because it generally does not rely on MHC or MHC-like presenting molecules. It may instead be more similar to the antigenic recognition by Abs. Like most ILCs, preset αβ T cells and γδ T cells often reside within peripheral tissues and are poised to produce rapid effector responses.

This review provides an evolutionary perspective on innate-like T cells as compared with mainstream T cells and ILCs. We briefly summarize the appearance and characteristics of these subsets in the earliest vertebrates before describing innate-like T cells in mammals and amphibians. We describe the different Ag recognition systems used and discuss the way they evolved. We show here that, although ILC-, αβ-, and γδ-like T cells were present in the earliest vertebrates, iNKT and MAIT cells are probably recent additions in mammals. We demonstrate that all these cells use common effector modules linked to the expression of specific transcription factors (TFs), leading to overlapping functions. Nonetheless, the evolutionarily conserved relationships between effector functions, Ag recognition systems, and tissue distributions suggest nonredundant functions of innate-like T cells.

The defining feature of the adaptive immune system of all vertebrates is the use of very diverse Ag receptors made by somatically randomly rearranged genes, enabling the recognition of any foreign molecule (Table I). Each cell expresses only one receptor, and each receptor is expressed by a small number of cells that can tremendously expand after encountering their specific Ag. Variable, diversity, and joining gene segments of the Ig superfamily and RAG-dependent recombination are used by jawed vertebrates, whereas RAG-independent copying of variable lymphocyte receptors (VLRs) by gene conversion are found in jawless vertebrates (3). Interestingly, in both cases, two kinds of receptors are made in the thymus or its equivalent: αβ and γδ TCR in jawed vertebrates and VLRA and VLRC in jawless vertebrates such as lampreys. Interestingly, γδ T cells such as VLRC⁺ lymphocytes are enriched in barrier tissues (4).

In jawed vertebrates, αβ TCRs recognize pathogen-derived peptides bound to MHC molecules. Although syntenic regions of the MHC locus referred to as proto-MHC are present in the invertebrate deuterostome amphioxus (Branchiostoma lanceolatum) (5, 6), no real MHC homolog has been found to date. In jawless vertebrates, no direct VLRA binding to native Ags has been observed, except in one instance (7), suggesting that VLRAs also recognize Ags bound to some kind of presenting molecules. In addition, no MHC-like encoding gene has been discovered in lamprey, despite many efforts. Nonetheless, both VLRA and VLRC repertoires show evidence of selection (at least on length) (4, 8). This suggests the existence of selecting molecule(s), yet to be identified. ILCs are also evolutionarily ancient and were probably present in the earliest deuterostomes. In the absence of TCR, ILCs are more challenging to identify because they do not express lineage-specific genes. Still, at least three ILC subsets were recently identified in zebrafish (9), and cytotoxic lymphocytes resembling NK cells have been identified in many jawed vertebrates (10). In lamprey, a heterogeneous lymphocyte subset that lacks VLRs includes granular NK-like cells and nongranular ILC2-like cells (M. Hirano, J. Li, M.D. Cooper, unpublished observations, and 4, 10). Evidence of NK receptors and NK-like cells has also been reported in invertebrates, including Ciona (11) and Botryllus (12).

Thus, ILCs and T cells are evolutionarily ancient. As discussed below, it is probable that the effector modules expressed by ILCs and T cells were also present in early deuterostomes. The occurrence and preservation of two types (αβ/VLRA and γδ/VLRC) of Ag receptors suggests nonredundant functions. αβ/VLRA recognition would require presentation of the Ags by specialized molecules such as MHC molecules, whereas γδ/VLRC may recognize unprocessed or native Ags, in an Ab-like fashion.

Preset αβ T cells acquire specific effector functions in the thymus. Unlike mainstream αβ T cells that recognize peptides presented by polymorphic MHC molecules, mammalian preset αβ T cells recognize nonpeptidic compounds presented by nonpolymorphic MH1-like molecules.

An MHC genomic locus containing genes encoding MHC class I and class II polymorphic molecules has been identified and characterized across all jawed vertebrates from sharks and bony fishes to amphibians, reptiles, and birds (reviewed in [13]). In addition, in all jawed vertebrates, multiple MHC class I families following various and complex evolutionary patterns have been found and are still discovered through the availability of new genomic data (14). This is underscored by the recent reports of large families of nonclassical MHC genes across 13 families of cartilaginous fishes (15). Such a complexity is most likely related to diversified immune and nonimmune roles of MHC molecules, as well as to the constant pressure exerted by pathogens to escape Ag presentation.

To date, the amphibian Xenopus (X. tropicalis and X. laevis) is the only species outside mammals in which preset T cells and nonpolymorphic MH have been characterized genetically and functionally to some degree. In both Xenopus species X. tropicalis and X. laevis, besides a unique polymorphic MHC classical class Ia gene located in the bona fide MHC locus, there is an expanded family of 23 nonpolymorphic MH1b genes outside the MHC locus in the telomeric region of the same chromosome 8 (16–19). In X. laevis, polymorphic MH1a molecules are not significantly expressed on the surface of most cells, including the thymus, until the onset of metamorphosis, when they become ubiquitously expressed (20–22). Moreover, the tadpole thymus lacks significant expression of the LMP7 gene, a critical component of the immunoproteasome subunit, until the onset of metamorphosis (22). However, the X. laevis tadpole thymus exhibits transcripts of numerous phylogenetically distinct nonpolymorphic or oligomorphic MH1b genes. Most of these MH1b genes exhibit a limited tissue-specific expression. Furthermore, several of these MH1b genes are expressed by thymocytes rather than the thymic epithelium, which is a characteristic of MH1 molecules selecting preset αβ T cells (17, 23).

Notably, TCR repertoire analysis by deep sequencing has revealed that tadpoles largely produce T cells with invariant or very limited TCRα diversity in germinal configuration (e.g., no N diversity), whereas adult frogs shift to increased production of T cells expressing the diversified TCR repertoire canonically associated with adaptive immunity (24). In tadpoles at 2 wk of age, six unique invariant TCRα rearrangements represent >80% of the TCRα repertoire used by the CD8^−/dim T cell pool (24). Thus, T cell immunity in tadpoles appears predominantly governed by innate-like T cells.

In support of this hypothesis, two distinct innate-like T cell subsets, iVα6 and iVα45 T cells, requiring distinct MH1b molecules (XNC10 and XNC4, respectively) for their development and functions have been characterized by reverse genetics approaches (24–26) and tetramer technology (2, 24). XNC10 deficiency induced by transgenesis using RNAi as well as CRISPR/Cas9 not only abrogates the development of iVα6 T cells but also markedly increases the susceptibility to infection by the ranavirus FV3 (a major amphibian pathogen) in tadpoles and adults (24, 26, 27). Furthermore, injection of XNC10 tetramers into tadpoles transiently depletes iVα6 T cells and impairs tadpole antiviral immune responses (28). XNC10 also plays a complex role in tumor immunity with transient iVα6 T cell depletion by XNC10 tetramers weakening tadpole host antitumor immune responses (29).

The function of the second MH1b, XNC4, appears to be more complex. It is required for the development of a T cell subset that expresses an invariant rearranged TCR Vα45-Jα1.14 (iVa45-T) chain (2, 25, 30). However, although targeting the Vα45-Jα1.14 TCR rearrangement or disrupting the Jα1.14 segment clearly shows that this T cell subset is critical for resistance to infection by mycobacteria (Mycobacteria marinum), the deficiency of XNC4 increases tadpole susceptibility to both mycobacterial and viral infections.

MR1 appeared in the earliest mammals because it is not present in platypus, reptiles, or birds (31). MR1 is also absent in the genomes of the two Xenopus species (X. laevis and X. tropicalis). Interestingly, turtles harbor an MH1 gene made of an α1 domain displaying several typical MR1 characteristics, including lysine 43 in the MR1 groove that binds and stabilizes 5-OP-RU (32). However, the α2 domain of this gene does not have the features necessary to bind 5-OP-RU and is more similar to classical MH1 molecules (32). It is unclear whether this molecule binds small metabolites. In mammals, unlike classical MH1 molecules that constantly drift during evolution, MR1 is subjected to purifying selection. The high conservation of MR1 leads to interspecies cross-reactivity: Human and mouse MAIT cells can be activated by mouse and human MR1 loaded with 5-OP-RU (33). Similarly, bat MAIT cells can kill both bat and human target cells loaded with 5-OP-RU (34).

Interestingly, MR1 has been lost independently three times: in carnivores (cat, dog, and panda), in armadillo, and in lagomorphs. MR1 coevolved with TRAV1 (the TRAV gene used by MAIT cells) because in all the species that have lost TRAV1, MR1 is either lost or pseudogenized. Conversely, TRAV1 is present in all the species that have a functional MR1 (31), which strongly suggests that the only function of MR1 under evolutionary pressure is the selection and restriction of the TRAV1⁺ MAIT cells (35).

Although MR1 is generally considered as quasi-nonpolymorphic, some genetic variations have recently been identified in human populations (36, 37). In particular, an H17R polymorphism is present in ∼20% of the alleles (36). This polymorphism is a source of an MH2 restricted minor alloantigen detected during graft-versus-host responses (38). However, the H17 residue is not conserved among species, and molecular modeling indicates that H17 does not interact with the TCR. Four other, much less frequent coding MR1 alleles have also been found. One of them, R9H (homozygous frequency, 0.0001 to 0.0002), prevents the binding of 5-OP-RU but not of Ac-6-FP (37). The absence of MAIT cells in a homozygous MR1^R9H/R9H patient confirms the requirement of MR1–5-OP-RU complexes for MAIT cell development (37, 39, 40). The functional consequences of the absence of MR1 and MAIT cells are difficult to ascertain because the index patient also harbored a homozygous deficiency in IFIH1, leading to a loss of function of the MDA5 protein involved in viral RNA sensing.

The mammalian CD1d molecule belongs to the family of CD1 and PROCR (endothelial protein C receptor) that harbors a narrow and hydrophobic cleft, allowing these molecules to bind various lipids. CD1/PROCR genes have not been found in amphibians or fishes, and the prototypic CD1/PROCR seems to be derived from an ancestral classical MH gene more than 375 million years ago before the separation between lungfishes and tetrapods (32). The CD1/PROCR family is already present in birds and reptiles, where it structurally looks similar to classical MH1 molecules (reviewed in [32]). In several mammalian species, CD1d underwent a very important pseudo-allelic diversification (31, 41), indicating specialized function in binding diverse sets of lipids (42).

Notably, despite the cross-reactivity of mouse and human iNKT cells toward mouse and human CD1d loaded with glycolipid Ags (43), the amino acid sequence similarity of mouse and human CD1d is much lower than that of mouse and human MR1. This suggests that the structural constraints in the CD1d cleft required to bind the glycolipid(s) are lower than those necessary for MR1 to bind 5-OP-RU and/or that MR1 is subjected to additional constraints on the outer part of the molecule.

Interestingly, another MH1-like molecule, called “MHX,” sharing evolutionary features with MR1 (e.g., purifying selection, nonpolymorphism in the species where it is present), appeared in the first eutherian species (31). Like MR1 and TRAV1 that coevolved in mammalian species (31), MHX seems to have coevolved with another invariant TCRα chain (TRAV41-TRAJ38) found in the rabbit, a species that has neither NKT nor MAIT cells (31). MHX is absent from marsupials but present in most eutherians, except for mice and humans. There are no data available on the occurrence of the invariant rearrangement TRAV41-TRAJ38 in species other than the rabbit, and no data are available on the ligand that could be presented by MHX. Still, the coevolution of an MH1-like molecule with a TRAV segment suggests the presence of another innate-like αβ T cell subset.

In summary, the two MH1-like molecules restricting the two best-known preset subsets, iNKT and MAIT cells, appeared with the emergence of mammalian species. Another pair of TRAV/MH1-like molecules (TRAV41/MHX) presents similar evolutionary features. It is tempting to hypothesize that preset T cells emerged in mammals to fulfill important new functions related to placentation, such as tolerance to conceptus. Constant higher temperature (37°C compared with ectotherm) may be another contributing factor.

γδ T cells were considered early on as a family of prototypic innate-like T cells because of the limited diversity of their TCRs, their enrichment at barrier sites, their effector/memory phenotype, and their polyclonal reactivity in multiple pathological contexts, conferring upon them a general function as a first line of defense of the organism (44, 45). In mice and humans, most if not all γδ T cells can be considered as innate-like, either because they acquire effector functions during development similar to preset αβ T cells or because their TCRs recognize conserved Ags. Interestingly, in some species such as ruminants, γδ T cells are the predominant subset of T cells even in blood circulation and may fulfill the function of mainstream T cells (46).

γδ T cells are found in all classes of jawed vertebrates, although they have been mostly studied in mice and humans. The characteristics of γδ T cells in terms of location, effector function, and activation modality are closely linked to the V genes composing their TCR (44, 45). This correlation is partly due to constrained expression of specific Vγ and Vδ TCR genes, together with specific functional features, during defined windows of development. Agonist selection by conserved endogenous TCR ligands, either in the thymus or in the periphery, also contributes to the acquisition of effector functions by γδ T cells (47).

Unlike αβ TCRs, γδ TCRs exhibit a limited repertoire diversity, largely due to the small number of TCR genes available for recombination, a programmed TCR gene expression during ontogeny, and the likely existence of conserved endogenous selecting ligands (44, 48). Furthermore, early fetal-derived γδ T cells develop with low expression of TdT in mice and humans, and thus their TCR is generated with limited junctional diversity (49, 50). γδ TCR genes are highly divergent among species, but they present several conserved features. In all species (except those few that have two TRD loci) the TRD locus is inserted into the TRA locus, which leads to the loss of the TRD locus during recombination of TRA. Moreover, the distal (3′) Vα/δ-genes can rearrange with either TRDJ or TRAJ genes and consequently can be part of either TCR-δ or TCR-α chains. This is analogous to the lamprey, where the VLRC locus is located within the VLRA locus (51). TRG loci show high variability in size and composition, and some ruminants have two separate TRG loci (52). In some species, such as cattle (53), goat (54), and swine (55), the number of TRDV genes is greatly expended (more than 50 genes in cattle), probably enabling a greater diversity of γδ TCRs. Although γδ TCRs are functionally similar to αβ, they also share striking features with BCRs. Both TCR-δ and IgVH can have very long CDR3s compared with αβ CDR3s (56). This is in part due to the use of several TRDD genes to compose the CDR3δ. In many species, a large percentage of TCRD genes bear the IgV region of the H chain (of the BCR) as the recognition element (for review, see [13]). This is because the IgH and TCRD loci were presumably derived from cis-duplication, and IgVH has further been transferred into the TCRD locus in several vertebrates. In addition, trans-rearrangement of IgM, IgW, and IgH V segments to TCRD and J segments of TCRD genes, yielding hybrid receptors, are found at high frequencies in the shark (57, 58). These findings corroborate the view that γδ TCR binding is structurally more akin to that of Ig receptors than MHC-restricted αβTCR.

Because only few TCR ligands have been described for γδ T cells in mouse and human to date, general rules of the γδ TCR antigenic recognition remain to be fully established (44, 59). Nevertheless, the Ag recognition mode by γδ TCRs is distinct from the αβ TCR one because it generally does not rely on MHC or MHC-like presenting molecules. Recent work has identified several members of the extended family of butyrophilin (BTN) molecules as binding partners of some γδ TCRs, both in mouse and in human (including BTN, BTN-like, and Skint molecules). Although this work is still ongoing, the emerging mechanism is that distinct heterodimers of BTN molecules interact with specific germline encoded V regions of TCRγ chains (and perhaps additional regions), similar to αβ T cell super-Ags (44). BTN molecules are part of the extended B7 family of immunomodulatory molecules. Similar to γδ TCR genes, BTN molecules have homologs in all jawed vertebrates. BTN genes are encoded in clusters that have evolved by gene duplication, deletion, diversification, and pseudogene formation in a way that is consistent with rapid evolution by selection (60). Although members of the BTN family are not highly conserved throughout evolution (60), the proposed mode of interaction between distinct γδ TCRs and their cognate BTN dimer ligands appears conserved from mouse to human (61–64). The strong relationship between γδ TCR and BTN molecules is exemplified by the coevolution between the Vγ9Vδ2 TCR chains and their putative interacting BTN across mammals (65). Thus, it is possible that γδ TCR genes have individually coevolved with their cognate BTN ligands, leading to the diversification of both families of molecules and their divergence among species. It is presently unknown whether all γδ TCRs can recognize BTNs, but genetic arguments support that the interactions between molecules of the BTN family and γδ TCRs have been conserved throughout evolution.

γδ TCRs can also recognize diversified ligands, generally corresponding to stress-regulated endogenous molecules (59). Their recognition is mediated by variable regions of the TCR generated by somatic recombination (mostly the Vδ CDR3 region) and involves both germline and nongermline encoded residues (66–69). As such, recognition of each of these ligands is restricted to either clonal or polyclonal populations of γδ T cells, indicative of a more adaptive function of γδ T cells leading to clonal or polyclonal immune responses (e.g., to CMV [70]). Strikingly, in some ruminant species where γδ T cells are more numerous than αβ T cells in the blood, CDR3δ are particularly diverse (46). In these species, γδ T cells may have evolved to fulfill the function of mainstream T cells by recognizing very diverse ligands.

A particular case of diversified γδ T cell reactivity is the recognition of MR1 by some human γδ T cells (67, 68). The mode of binding is unusual or atypical because the γδ TCRs recognize the underside (68) or the side of the MR1 molecule (67). The γδ TCRs do not recognize 5-OP-RU or ac-6-FP per se, which is only required to enable translocation of MR1 to the cell surface (67–69, 71). Interestingly, the mode of recognition of MR1 by human γδ TCRs shares some similarities with the way other MH1b molecules are recognized by either mouse or human γδ TCRs (67) (e.g., EPCR [72], T22 [66], CD1d [69]). The growing number of MH1b molecules recognized by γδ TCRs and the recurrence of the reactive γδ TCRs among individuals (66–68) may suggest the existence of a conserved selection process that would favor the emergence of MH1b reactive clones. Alternatively, γδ TCRs can bind non-MHC molecules (e.g., annexin A2 [73], EPHA2 [74]), including molecules that are unlikely to have any physiological relevance (e.g., PE [75]), in accordance with an adaptive-like mode of recognition by γδ TCRs.

At least several γδ TCR clones recognize both BTN molecules and variable Ags (62, 74). However, the involvement of each interaction in physiological and pathological contexts is still unclear. Both interactions are unlikely to occur at the same time on a single TCR molecule (62), and thus these two interactions may serve a distinct purpose. The coevolution between specific γδ TCR genes and genes encoding their cognate BTN ligands, as well as the conservation of the general mode of interaction between TCRs and BTN dimers, implies an important function of the BTN in the biology of γδ T cells. Indeed, TCR–BTN interactions play key roles in thymic selection (e.g., mouse DETC [76]), extrathymic maturation (e.g., mouse and human gut intraepithelial lymphocytes [77]), and effector functions of γδ T cells (e.g., phospho-Ag detection by human Vγ9Vδ2 T cells [78]). Importantly, the expression of given BTN molecules is often restricted to specific cell types (76, 77, 79). Therefore, distinct subsets of γδ T cells likely encounter their cognate BTN dimers at specific developmental stages or in specific environments, which may lead to distinct functional outcomes. This general mechanism thus provides a rationale for the innate-like features of γδ T cells, the association of γδ T cells bearing given Vγ TCR chains (and to some extent Vδ) with functional features (e.g., location, effector functions), and the nonclonal immune responses of some γδ T cell subsets (e.g., human Vγ9Vδ2 T cells). One likely possibility is that innate interactions with BTN play primary functions in shaping organ-specific γδ T cell repertoires and innate-like phenotypes, whereas interactions with diverse Ags further contribute to adaptive-like responses. It is presently unknown whether the innate ability to recognize BTN molecules is shared by all γδ T cells or whether some γδ T cells are bona fide adaptive effectors.

It is noteworthy that in the lamprey, the VLRC repertoire is more restricted than the VLRA one (4, 8). Furthermore, VLRC⁺ lymphocytes express orthologs of the two components of very late Ag-4 (4), the expression of which correlates with adherence of human γδ T cells to epithelial cells (80) and, as one would expect, are enriched in barrier tissues (4). The convergence between VLRC⁺ lymphocytes and γδ T cells may indicate that despite the absence of BTN molecules in the lamprey, other molecules may similarly shape the repertoire and innate-like functions of VLRC⁺ lymphocytes.

The effector functions of innate-like, preset, and mainstream T lymphocytes correspond to distinct differentiation programs or modules that are shared among some of these subsets and with ILCs: For instance, cytotoxicity through perforin/granzyme release is found in NK and effector CD8 T cells, whereas type 1 immunity associated with IFN-γ secretion is found in NK, ILC1, CD8, most γδ T cells, and a subset of CD4 as well as by preset T cell subsets such as NKT1 and MAIT1 cells. These differentiation programs are related to the expression of specific TFs such as T-bet (encoded by Tbx21) or Eomes for type 1 immunity or ROR-γt (encoded by Rorc) for type 17 immunity.

The effector modules used by mainstream and innate-like T cells are evolutionarily ancient because they are also used by ILCs, and many defining genes can be found in the earliest vertebrates. In mammals, three main differentiation programs correspond to the three main ILC subsets:

• Th1 (antimicrobial) program, often associated with targeted cytotoxicity, controlled by T-bet (Tbx21). The cytotoxic program is controlled by the TFs Eomes and T-bet and depends on perforin and granzymes.

• Th2 (antiparasites such as helminth) program, controlled by GATA-3. A homolog is also expressed in lamprey VLRA/C⁺ lymphocytes (4).

• Th17 (mucosal-associated) program, controlled by ROR-γt and associated with tissue repair properties.

IL-17 is of very ancient origin and one of the most conserved cytokines throughout evolution. It is present in many mollusks and is associated with inflammation, particularly in the gut, because IL-17 can be produced by mucosal epithelial cells (81, 82). The Th1 (IFN-γ, IL-12, TNF-α), Th2 (IL-4, IL-5, IL-13), and Th17 (IL-17, IL-22) cytokine genes were already present during the emergence of cartilaginous fishes (83), indicating that T cells could generate sets of responses as diverse as in mammals.

In mice, the canonical preset αβ T cells, iNKT and MAIT cells, differentiate into two main subsets expressing either T-bet or ROR-γt (84, 85). In contrast, human MAIT and NKT cells simultaneously express T-bet, ROR-γt, Helios, and Eomes (86, 87), leading to versatile and context-dependent effector functions (88). In bats, MAIT cells appear similar to human MAIT cells with simultaneous expression of T-bet and ROR-γt (34). This is also true for MAIT cells from cattle, sheep, and opossum (H. Bugaut, O. Lantz, and F. Legoux, unpublished observations). Thus, the mouse MAIT cell program appears to be more the exception than the rule in mammals.

Similar to mouse preset αβ T cells, mouse γδ T cells differentiate into two main subsets that produce Th1 and Th17 cytokines, respectively. Although ROR-γt is specifically required for Th17 production, T-bet deficiency affects the function of both Th1 and Th17 producers (89). Th2 cytokines such as IL-4 can be produced together with Th1 cytokines, although the contribution of GATA-3 in this program is unknown (90). In humans, γδ T cells are mostly characterized by their natural cytotoxicity and secretion of Th1 cytokines. They can also be polarized in vitro to produce Th2 and Th17 cytokines (91, 92).

Foxp3 is not expressed by ex vivo innate-like T cells or ILCs. The significance of FOXP3 expression by human γδ T cells after in vitro stimulation (93) is unclear, because FOXP3 can also be transiently expressed by mainstream T cells after in vitro stimulation (94). Thus, on first approximation, innate-like T cells and ILCs can express all the effector modules of mainstream T cells except Foxp3.

Although lymphocytes may have already been capable of diverse effector responses in ancestors of cartilaginous fishes, it is unknown whether they did and whether this diversification first appeared in innate leukocytes or adaptive lymphocytes.

TCF-1 (and to a lesser extent LEF-1, a redundant factor of the same family) is a key T cell factor in mice and humans. It is rapidly induced by Notch signaling in thymic settling progenitors (95–97), is necessary and sufficient to initiate T cell specification (95), and imprints the epigenetic landscape used by all mature T cells (98). TCF-1 (and LEF-1) is proposed to broadly regulate T cell function, including the acquisition of effector functions in mainstream and preset αβ T cells (99), as well as in γδ T cells (100). Interestingly, a TCF-1 ortholog is expressed in lamprey VLRA⁺ lymphocytes (3), suggesting that its functions are evolutionarily conserved.

More recently, TCF-1 was identified as an early controller of ILC development in the mouse (101). TCF-1 is the first “shared TF” expressed in both T cells and ILCs during development (95, 101). TCF-1 has some shared functions in T cells and ILCs (102), and its expression in lymphoid precursors is sufficient to trigger a transcriptional program shared by T cells and ILCs (95). TCF-1 is thus likely responsible for the developmental acquisition of similarities between the two lineages, which include their diverse effector programs. Although TCF-1 is required for late differentiation of ILCs and acquisition of effector programs, it is dispensable for ILC specification (102). ILCs may thus have originally evolved without TCF-1 and may have acquired it together with the diverse effector programs of adaptive T cells. Consistent with this scenario, in the lamprey, TCF-1 is expressed in VLRA⁺ lymphocytes but not in VLRC⁺ or VLR^neg lymphocytes that include ILC-like cells (4). Furthermore, TCF-1 expression is induced by distinct mechanisms in T cells and ILCs (97), supporting that this factor was acquired independently by the two lineages. Finally, the acquisition of a TCF-1–dependent T cell transcriptional program by ILCs would further explain the intriguing observation that ILCs express many genes related to TCR function, including TCR genes and TCR signaling components (103). Thus, TCF-1 is likely responsible for the acquisition of diverse effector programs shared by all innate and adaptive T lymphocytes.

The hallmark of mammalian preset αβ T cells is the expression of the TF PLZF (encoded by Zbtb16) (1) that induces a tissue residency differentiation program (104). PLZF has many nonimmune functions, but its expression in preset αβ T cells and ILCs is related to a conserved enhancer that binds Runx1 (105). The induction of PLZF is linked to selection by CD4⁺CD8⁺ (DP) thymocytes (104, 106). This process requires SLAM-dependent interactions and TCR triggering by ligands presented on DP thymocytes (107, 108). Moreover, forced expression of classical MHC on DP leads to the development of PLZF-expressing T cells restricted by these MHC molecules (109). The adaptor SAP that mediates SLAM signaling is necessary for the development of both iNKT and MAIT cells in mice, but only for NKT cells in humans because SAP-deficient patients harbor normal numbers of MAIT cells (110). Although not expressed in most γδ T cells (90), PLZF is expressed and required by some mouse γδ T cell subsets that are characterized by Vγ1Vδ6.3 (111) or Vγ6 TCRs (112) (Heilig and Tonegawa nomenclature) and that acquire their effector functions in the thymus. SAP is required for Vγ1 but not Vγ6 T cells (111, 113). SAP is also required for Vγ4 and Vγ5 T cells, which do not require PLZF but also acquire their effector functions in the thymus (113). It is tempting to hypothesize that selection on DP thymocytes selects and preprograms some γδ T cell subsets such as Vγ1Vδ6.3 T cells, whereas alternative mechanisms likely control effector programming of other γδ T cell subsets that also complete their maturation in the thymus.

PLZF function in ILCs is unresolved. It is transiently expressed during the development of all helper ILCs (114). Lineage-tracing experiments using a Zbtb16-Cre strain show that LTi and NK cells exhibit a limited history of PLZF expression (114). The lineage-tracing approach may lack sensitivity if Zbtb16 is expressed at low levels or for a short time in some ILC lineages. Nonetheless, ILCs do not need PLZF for their early development (103, 114), and PLZF requirement at steady state is restricted to specific ILC lineages and specific tissues (e.g., ILC2, liver ILC1 [114]). Given the conservation of PLZF expression between lamprey ILC-like cells and vertebrate ILCs (10), it is tempting to speculate that PLZF regulates the acquisition of important functional ILC features that remain to be characterized. The precise function of PLZF remains unclear. However, PLZF ectopic expression in mouse DN3 or DP thymocytes impairs mainstream T cell development and promotes the acquisition of preset features by mainstream αβ T cells, including an effector phenotype and tissue-resident features (104, 115). Therefore, although PLZF is a key upstream controller of the acquisition of innate functional features in developing T cells and some ILCs, alternative mechanism(s) must be used by PLZF-independent γδ T cells and ILC.

Innate-like T cells generally acquire their effector functions after TCR signaling. However, some γδ T cells undergo functional maturation independently of their TCRs, similar to ILCs. Fetal-derived mouse Vγ4+ γδ T cells acquire their ability to produce IL-17 and their tropism toward epithelia in the thymus in the absence of functionally recombined TCR. This acquisition is mediated by the TF SOX13 (116). Although SOX13 is generally induced downstream of γδ TCR signaling and is necessary and sufficient for the effector differentiation of γδ T cells into IL-17 producers (117), the mechanism responsible for the TCR-independent induction of SOX13 in some γδ T cells is unknown. Strikingly, SOX13 is also expressed in lamprey VLRC⁺ lymphocytes, which have been proposed to be functionally equivalent to γδ T cells (4). Thus, SOX13 may play key roles in the acquisition of evolutionarily conserved innate-like features (e.g., skin homing). Interestingly, SOX13 ectopic expression in T cell precursors disrupts αβ T cell development. The proposed mechanism is that SOX13 interacts with and sequesters TCF-1 (117). Likewise, in the lamprey, SOX13 and TCF-1 expression appears to be mutually exclusive: SOX-13 is specifically expressed by VLRC⁺ lymphocytes, whereas TCF-1 is expressed only by VLRA⁺ lymphocytes (4). SOX13 may thus control the acquisition of an ancestral innate functional program by adaptive lymphocytes, antagonistic to the one controlled by TCF-1.

Regarding TCR specificities and restricting molecules of innate-like T cells, two main evolutionary scenarios can be distinguished. For γδ T cells, TCR V genes probably coevolved with their BTN ligands, leading to concerted divergence during evolution. By contrast, the semi-invariant TCR and MH1b (CD1d and MR1) molecules of preset αβT cells have most likely been subjected to purifying selection, probably because they must interact together without pressure for polymorphism as their bacterial ligands correspond to structural chemicals displaying limited diversity.

The functions of preset T cells that are under evolutionary pressure could be related to their differentiation programs, Ag specificity, tissue location, and specialized roles in species with external developmental stages relying on small numbers of T cells or any combination thereof. However, the various effector modules of preset T cells (the prototypic ones being types 1 and 17 immunity related to T-bet and ROR-γt expression, respectively) are shared with ILCs, some γδ T cell subsets and tissue-resident memory mainstream T cells. PLZF expression, which indicates a SLAM/SAP-dependent selection by CD1d/MR1 expressed on DP thymocytes during development in mice (104, 106), is expressed by iNKT cells, MAIT cells, and some γδ T cell subsets in both mice and humans (1, 118). Notably, in humans, PLZF is also expressed at a significant level by circulating mainstream CD8 T cells and at still higher expression levels in liver CD4 and CD8 mainstream T cells, without, however, reaching the levels of MAIT cells (85). Thus, preset T cells do not harbor specific effector modules and may in fact display versatile functions according to the context (i.e., cytokine environment) of TCR stimulation (88). Like for γδ T cells and mainstream tissue-resident memory T cells, it is not even clear whether TCR triggering is necessary for many functions of preset T cells (119, 120). Thus, Ag specificity may be implicated in the functions of iNKT and MAIT cells only under certain circumstances.

The fact that both iNKT and MAIT cells were retained during mammalian evolution implies a critical nonredundant role for these two subsets. Given that the effector programs of iNKT cells and MAIT cells are similar, we can speculate that the feature that provides a selective advantage is the antigenic specificity of iNKT and MAIT cells. Because of the very different nature of ligands presented by CD1d and MR1 (glycolipids versus riboflavin metabolites), iNKT and MAIT cells will likely receive TCR signals and elicit immune responses in distinct contexts. Notably, iNKT cells can be activated by bacteria, both by upregulation of endogenous ligands and by provision of bacterial glycolipids (121). In contrast, MAIT cells display strong reactivity exclusively toward exogenous microbial compounds because 5-OP-RU is not produced by mammalian cells. This differential reactivity between iNKT and MAIT cells toward endogenous or exogenous ligands may explain the conservation of both subsets during mammalian evolution.

The authors have no financial conflicts of interest.