The promise of γδ T cells and the γδ T cell receptor for cancer immunotherapy

γδ T cells form an important part of adaptive immune responses against infections and malignant transformation. The molecular targets of human γδ T cell receptors (TCRs) remain largely unknown, but recent studies have confirmed the recognition of phosphorylated prenyl metabolites, lipids in complex with CD1 molecules and markers of cellular stress. All of these molecules are upregulated on various cancer types, highlighting the potential importance of the γδ T cell compartment in cancer immunosurveillance and paving the way for the use of γδ TCRs in cancer therapy. Ligand recognition by the γδ TCR often requires accessory/co-stimulatory stress molecules on both T cells and target cells; this cellular stress context therefore provides a failsafe against harmful self-reactivity. Unlike αβ T cells, γδ T cells recognise their targets irrespective of HLA haplotype and therefore offer exciting possibilities for off-the-shelf, pan-population cancer immunotherapies. Here, we present a review of known ligands of human γδ T cells and discuss the promise of harnessing these cells for cancer treatment.


INTRODUCTION
The adaptive immune compartment comprises of three distinct cell subsets, namely B cells, T cells expressing an ab T cell receptor (TCR) and T cells expressing a cd TCR. These cells generate their defining B cell receptors (BCRs), or TCRs using somatic V(D)J recombination which enables them to recognise a vast spectrum of antigens. Despite the fact that gene segments used in rearrangement leading to ab and cd TCRs were discovered almost at the same time, 1 the elucidation of ab T cell biology progressed rapidly, while the complexities of cd T cells have been slower to emerge. Nevertheless, the tripartite organisation of the lymphocytic immune system appears to be a fundamental requirement for efficient function, as this organisation is present both in jawed and jawless vertebrates, even though the manner of generating the defining receptors completely differs between these lineages. 2 The unique functions of cd T cells, particularly in terms of antigen recognition and kinetics of response, provide further evidence that cd T cells represent an important and non-redundant lymphocyte subset.
V(D)J recombination of the c and d chain genes is shown in Figure 1 (IMGT nomenclature 3 ). There are 4-6 (depending on haplotype) functional variable (V)c, and 8 Vd gene segments in humans. 4 Some Vd segments can be used to generate both the a and d chains of the TCR, as they are located within the tcra locus. 5,6 The number of V segments that can be used for cd T cells is much smaller than that for ab T cells (46 Va and 48 Vb segments). However, the potential diversity of cd TCR surpasses that of ab TCR, owing to extensive N-region nucleotide additions and presence of distinct D segments (present only in tcrd but not tcrg locus) which can be used simultaneously and read in all three frames. This junctional variability results in the generation of hyperdiversity focused on the complementarity determining region (CDR)3 loops which are crucial for antigen recognition. 7 Furthermore, the length of the CDR3s of both a and b chains is constrained, due to the requirement to make a well-defined contact with peptide-MHC complexes, while CDR3 in the d chain is usually more variable and longer than its c counterpart. 8 With regard to CDR3 length, the cd TCR resembles the BCR more than ab TCR. This greater variability of cd TCRs may translate into recognition of both proteins and smaller molecules. The CDRs form loops in the cd TCR structure to provide a highly variable antigen-binding domain at the membrane-distal end of the molecule (Figure 2).
After arising from a common progenitor in the thymus, the maturation pathways of cd and ab T cells diverge. Notably, the development of cd TCR 1 thymocytes does not require the expression of Aire, 9 a transcriptional regulator crucial for the negative selection of autoreactive ab T cells. The mechanism by which T cells become committed to the ab or cd lineage is not yet fully understood as thymocytes rearrange b, c and d genes at the same time which can lead to simultaneous expression of the cd TCR and pre-TCR (invariant Ta paired with TCR-b). 10 However, recent evidence suggests that thymocytes adopt the cd T cell lineage after receiving a strong signal via cd TCR, which can be additively enforced by additional signalling via pre-TCR -thus enabling weak ligands to drive cd T cell lineage commitment as well. 11 If cells fail to receive this survival signal they silence the cd TCR and undergo TCR-a rearrangement. 12 This signal strength model implies that cd T cells need to encounter a cognate ligand in the thymus. However, to date only one molecule, namely Skint-1, has been described as a thymically expressed ligand necessary for development of a subset of mouse cd T cells. 13 The identity of other ligands required for positive selection of cd remains to be elucidated. Strong cd TCR-mediated interactions in the thymus have been shown to result in upregulation of CD73, the earliest identified marker of cd lineage commitment. 14 CD73 is expressed by the vast majority of cd T cells in the periphery, supporting the notion that recognition of the ligand in the thymus is a common occurrence in cd T cell development. Another striking difference in development between ab and cd T cells is the acquisition of effector functions. Conventional ab T cells acquire their effector phenotype, in terms of produced cytokines, upon interactions with their targets in the periphery, while cd T cell functions, like their anatomical location, appear to be pre-determined in the thymus by the chain usage of their TCR. 15 In humans, cd T cells constitute 0.5-10% of T cells in peripheral blood but are substantially enriched in epithelial tissues (e.g. in skin, lungs, intestine). The majority of peripheral blood V-J recombination at tcrg locus V-D-J recombination at tcra/tcrd locus  Figure 1 V(D)J recombination at the tcrg (upper panel) and tcra/tcrd (lower panel) locus. Only the functional gene segments are shown. The TCR-c chain is produced using only a single V-J recombination, with P/N additions occurring at the V-J junction. The TCR-d chain is produced using V-D-J recombinations that can involve either 2 or 3 D segments, leading to the creation of up to 4 N diversity regions. For the clarity of the figure, only the gene segments that can be used in TCR-d chain production are presented (lower panel). The organisation of loci tcrg and tcra/tcrd was adapted from IMGT database. 3 cd T cells express the Vd2 chain, while the tissue-resident cd T cells are mainly Vd1 pos or Vd3 pos . The precise reasons for this tissue specificity and the mechanisms that underlie it are yet to be elucidated. The role of cd T cells in infection is well established and has been amply reviewed elsewhere. 16 More recent discoveries suggest that cd T cells play a role in anticancer immunity, and it has been known for over a decade that mice lacking the cd TCR are more susceptible to some cancers through unknown mechanisms. 17 Here, we review recent advances in discovering cancer-associated antigens recognised by human cd T cells and the evidence that there is a broad spectrum of cd T cell ligands waiting to be discovered. We also provide a brief overview of therapeutic applications of cd T cells in cancer immunotherapy.
cd TCR RECOGNITION OF CANCER-ASSOCIATED ANTIGENS Known ligands of human cd TCR are scarce when compared with the vast spectrum of antigens recognised by ab T cells. However, significant progress in this area has been made in the recent years, linking the observed recognition of tumour cells to specific ligands confirmed by biochemical and biophysical data. The main targets encompass phosphorylated prenyl antigens, endo-and exogenous lipids presented by CD1-family proteins, and cell stress molecules that can indicate DNA damage, viral infection or malignant transformation.

Recognition of phosphoantigens
Phosphorylated isoprenoid metabolites, commonly referred to as phosphoantigens (PAgs), can be produced by both bacterial and eukaryotic cells, using non-mevalonate and mevalonate biosynthetic pathways, respectively. The accumulation of PAgs, such as isopentenyl pyrophosphate (IPP), is a result of metabolic dysregulation that commonly occurs in tumour cells 18 and therefore the enzymes of the mevalonate pathway are a valid target for anticancer drugs (reviewed in Clendening and Penn 19 ). Importantly, PAgs have long been known to be recognised by cd T cells expressing Vc9Vd2 TCR. The presence of this peripheral blood subset of cd T cells is restricted to higher primates, with a few exceptions. 20 Importantly, mice and other rodents do not possess any corresponding T cell subsets that respond to PAgs. The mode of PAg recognition has only recently been resolved. Early studies indicated that recognition of PAgs by human Vc9Vd2 T cells required cell surface presentation by a species-specific molecule. 21 One of the proposed molecules presenting the antigens was F1-ATPase which is expressed on the surface of a wide range of tumour cells. 22 Further research suggested that PAgs could bind F1-ATPase in the form of a nucleotide derivative, increasing the efficiency of T cell activation. However, as antibody blocking of F1-ATPase did not abrogate T cell recognition, other molecules were thought to be involved in PAg presentation. 23 The cell surface-expressed Butyrophilin molecules are encoded within the MHC class I locus and offer attractive potential candidates as PAg-presenting molecules. Butyrophilin-3A (BTN3A/CD277) is present in humans in three isoforms (BTN3A1, BTN3A2 and BTN3A3). 24 Recent studies have shown that antibodies specific for BTN3A1 could either mimic PAg-mediated activation of the TCR (antibody 20.1) or abrogate this stimulatory effect (antibody

103.2)
. 25 Biophysical analysis of the underlying interactions suggested that 20.1 antibody induced/stabilised a TCR-activating conformation of BTN3A1. 26 Additionally, PAgs and 20.1 antibody activated the same intracellular signalling pathways in the responding cd T cells, suggesting a common recognition process. 27 Interestingly, transduction studies revealed that while the expression of BTN3A1 alone is sufficient for the activation mediated by 20.1, additional genes located on chromosome 6 are required for PAg-mediated recognition. 28 These studies suggest the possible existence of molecules responsible for efficient loading of PAgs into butyrophilin-3A1 in a conceptually similar manner to loading peptides into MHC, or that there is a requirement for a co-stimulatory ligand. Two main models explaining the role of BTN3A1 have been proposed, involving intracellular or extracellular binding of pyrophosphate antigens ( Figure 3a). Sandstrom 30 In direct contrast, De Libero and colleagues resolved the crystal structure of two PAgs (IPP and HMBPP) in complex with the extracellular domain of BTN3A1 where the antigen was bound in a shallow surface pocket (Figure 3b). 31 The binding affinity of Vc9Vd2 TCR to BTN3A1 was relatively weak (K D , 0.1 mM) but increased in the presence of IPP. These two mechanisms need not be mutually exclusive and it remains possible that PAgs bind to both the intracellular and extracellular domains of BTN3A1 to provide a failsafe against aberrant TCR triggering and autoimmunity. The binding affinities of IPP to BTN3A1 and the Vc9Vd2 TCR interaction with BTN3A1-IPP are summarised in Table 1. Further investigation will be required to determine the exact mechanism of antigen presentation by BTN3A1.
cd T cells recognise specific ligands in the context of antigenpresenting molecules from the CD1 family Four members of the human CD1 family of proteins (CD1a-CD1d) present both endogenous and exogenous lipids that can be recognised by both ab and cd T cells. 32 Early reports demonstrated that mucosal cd T cell clones showed broad reactivity towards these four CD1 isoforms loaded with exogenous phospholipids. 33 Further studies showed that T cells expressing the Vd3 chain were capable of killing tumour cells expressing CD1d (but not other CD1 isoforms) without addition of exogenous lipids. 34 Additionally, the cd T cell compartment of peripheral blood was shown to recognise CD1d in complex with self-lipids termed sulphatides (sulphated galactosylceramides). 35 CD1d-reactive cd T cells expressed similar Vd1-Jd1 chains paired with different Vc chains. Notably, some of the generated clones could bind CD1d without any loaded lipid. Recent studies provided the structural basis for CD1d-sulphatide recognition, demonstrating that the binding occurred solely via the d chain of the TCR -the germline-encoded residues made contact with the CD1d molecule whereas the CDR3d loop interacted with sulphatide ( Figure 4a). 36 As the CDR3 region is the most variable part of the cd TCR, it is possible that a subset of cd T cells can recognise diverse lipid cargoes presented by CD1 molecules, in a similar way to ab TCR recognition of peptides bound to MHC. A similar mode of interaction was reported for CD1d-a-galactosylceramide (aGalCer) and a Vd1 TCR (9C2) where CD1d was bound by germline-encoded residues in the d chain while the lipid was recognised by the hypervariable CDR3c loop (Figure 4b). 37 Interestingly, Pellicci et al. have recently discovered a novel T cell subgroup recognising CD1d-aGalCer complex, expressing a Vd1 domain linked to permissive Ja-Ca segments, paired with diverse TCR-b chains (termed d/ab T cells). 38 Again, the binding of the TCR to CD1d was mainly driven by germline-encoded residues in the Vd1 domain while the specific recognition of the bound lipid was exclusively mediated by the b chain of the TCR (Figure 4c). The affinity of the interaction between TCRd/ab and CD1d-aGalCer was, however, much higher than described for 9C2 cd TCR (the binding affinities of TCRs to CD1d complexed with lipid ligands are summarised in Table 2). Importantly, functional d/ab T cells comprise a high proportion of human CD1d-reactive T cells. aGalCer has been so far investigated only in context of boosting NK T cell anticancer activity, 39 but it could potentially be used to activate aGalCer-reactive cd T cells, similarly to using PAgs. Notably, a recent study showed that small quantities of aGalCer and other a-glycosylceramides can be endogenously produced in mammalian species to modulate the immune response. 40 Additionally, CD1d may be an important target for cancer immunotherapy as its high expression was reported on chronic lymphocytic leukaemia cells, correlating with disease progression. 41 Similarly, CD1c on acute leukaemia cells was shown to present a novel class of immunogenic self-lipids, called methyl-lysophosphatidic acids (mLPAs). Recognition of CD1c-mLPA complex by cd T cells has yet to be demonstrated, 42 but recent studies showing that CD1 family members can present self-lipids make these molecules attractive targets for future exploration.
cd TCR recognition of general stress ligands cd T cells are known to expand in response to some viral infections, with cytomegalovirus (CMV) providing the best-studied example to date. CMV infection results in preferential expansion of cd T cells expressing Vd1, Vd3 and Vd5 (commonly termed 'non-Vd2 T cells'). 43,44 These cells can also recognise and lyse intestinal tumour cells, in line with the fact that they are enriched in epithelial tissues where they perform stress surveillance. 44 Interestingly, infection with CMV, regarded as a treatment-related complication in kidney transplant recipients, serves a protective role against cancer in those patients. 45 Taken together with the fact that Vd2 neg compartment is selectively expanded in CMV carriers and does not diminish with age, it is possible that CMV-mediated activation of cd T cells confers an adaptive-like, lifelong reduction of cancer risk. 46 The ligands of dual reactive cd T cell clones recognising CMV-infected and transformed cells remain unknown, with one notable exception. Recently, Déchanet-Merville and colleagues identified the ligand of one of such clone, named LES. The LES cd T cell clone expressed Vc4 and Vd5 TCR chains and was substantially enriched (25% of peripheral blood T cells) in a CMV-infected transplant recipient. LES was shown to recognise a range of solid tumour lines and CMV-infected cells via its TCR rather than natural killer (NK)-type receptors. The LES TCR was demonstrated to bind endothelial protein C receptor (EPCR). 47 EPCR exhibits sequence and structural homology with the MHC-like protein family CD1, and can present phospholipids bound in the antigen-presenting groove. 48 EPCR plays a dual role in cancer, as it can both promote and inhibit metastases, presumably depending on whether it is expressed on tumour cells or endothelium (reviewed in Mohan Rao et al. 49 ). Additionally, EPCR signalling can activate antiapoptotic or Table 1 The interaction affinities of IPP binding to the intracellular or extracellular domains of BTN3A1, and of TCR binding to BNT3A1 in presence of IPP.  proapoptotic pathways through unresolved mechanisms. Importantly, direct binding of the LES TCR to EPCR was confirmed by surface plasmon resonance, showing relatively low binding affinity (K D < 90 mM) (Table 3). EPCR, however, does not appear to be a common, innate-like target for dual reactive cd T cell clones, as the recognition of the tumour lines by such clones, isolated from other patients, was not abrogated by antibody blocking of EPCR. Moreover, the LES TCR recognition was mediated by the hypervariable CDR3 loop of the Vc4 chain, supporting the hypothesis that EPCR binding was part of an adaptive response. Interestingly, the expression of EPCR was not sufficient to induce T cell activation, suggesting that functional recognition required a cell stress context and additional co-stimulatory molecules. Indeed, CMV infection conferred the recognition of fibroblasts and endothelial cells without altering EPCR expression suggesting that recognition of this receptor occurs in a stress-related co-stimulatory context. The identity of these co-stimulatory ligands remains unknown. EPCR is not the only cell stress marker that requires an additional co-stimulatory context for efficient recognition by cd T cells. Other examples described so far encompass the human homologue of the bacterial MutS (hMSH2) protein, 50 being a part of the DNA mismatch repair system (MMR), and MHC class I polypeptide-related sequence A (MIC A), 51 a broadly recognised stress marker. Both molecules are recognised by both highly variable cd TCR and the invariant NKG2D receptor (discussed in the next section). hMSH2 is a key protein involved in repairing mutations resulting from DNA recombination or replication. Unsurprisingly, defects in hMSH2 have been reported in various tumour histologies including lung, colon, breast and prostate cancers. 52 Although hMSH2 is normally restricted to the nucleus, it can be translocated during cellular stress. 53 Recent studies have demon-strated that hMSH2 can be a ligand for some Vd2 cd T cells when it is ectopically expressed, inducing cytotoxicity and IFN-c production. 54 Surface expression of hMSH2 can be induced by Epstein-Barr virus (EBV) infection, 54 oxidative stress and proinflammatory cytokines such as IL-18 corroborating its role in generalised stress.
Interestingly, oxidative stress promoted the expression of MIC A in a similar manner to hMSH2, involving p38/JNK signalling pathways. 55 Another group reported MIC A/B upregulation resulting from ataxia telangiectasia mutated (ATM) protein kinase signalling in response to DNA damage, contributing to cd T cell-mediated lysis of ovarian cancer cell lines. 56 Additionally, the level of MIC A expression on breast cancer cell lines correlated with their susceptibility to cd T cell-mediated cytotoxicity. 57 Induction of surface expression of MIC A/B was also reported in primary glioblastoma after chemotherapy. 58 EBV infection induces MIC A expression as well, showing further similarities between expression of MIC and hMSH2.

cd TCRS RECOGNISE TUMOURS IN A CO-STIMULATORY STRESS CONTEXT
As mentioned above, the recognition of cell stress markers by cd T cells appears not to be driven solely by the interaction between the TCR and its cognate ligand but rather requires a wider stress context provided by co-stimulatory receptors on the T cell and additional self-antigens. In the case of the LES clone recognising EPCR, part of the co-stimulatory effect was provided by LFA-1 on the T cells binding to ICAM-1 (overexpressed as a result of CMV infection). 47 CD2 interaction with CD58 (LFA-3) was also implicated. Blocking of TCR binding to EPCR completely abrogated the recognition while disrupting the ICAM-1-LFA-1 and CD2-CD58 axes merely decreased T cell activation, demonstrating the dominance of TCR signalling and the possible co-existence of complementary co-stimulatory pathways. Furthermore, another group showed that the CD8aa homodimer could act as a coreceptor for recognition of CMV-infected cells by Vd1 TCR chains (Figure 5a). 59 This observation is in line with the findings that T cells expressing CD8aa play an important role in immunosurveillance of the epithelial tissues against viral 60 and bacterial 61 infections.
The requirement for co-stimulation is well established for ab T cells and it has been known for some time that CD8ab heterodimer acts to co-receive peptide-MHC class I, stabilising the TCR interaction 62 and ensuring full phosphorylation of the CD3f chain 63 to increase sensitivity to antigen by up to a million-fold 64 (reviewed in Cole et al. 65 ). Thus the ab T cell coreceptors, CD4 and CD8, ensure that this T cell subset is MHCrestricted 66,67 and determine the fate of the developing T cell. 68 In an analogous fashion, co-stimulation via stress-induced ligands might maintain the correct focus and function of the cd T cell compartment. In keeping with this concept, both hMSH2 and MIC A/B are dually recognised by cognate TCRs and the NKG2D receptor (Figure 5b). NKG2D is commonly expressed on the surface of NK cells, ab T cells and cd T cells, and binds to cell stress molecules from the UL16-binding protein (ULBP1-6) family (reviewed in Ullrich et al. 69 ). The  interactions of germline-encoded NKG2D with such diverse ligands are possible because of conserved fragments within the a1a2 superdomain in the ligands. The involvement of the cd TCR and NKG2D in recognition of the MIC A molecule has recently been studied. Xu et al. showed that both TCR and NKG2D bound overlapping fragments of MIC A, with different affinity and kinetics -for NKG2D, the affinity was much higher than for the TCR (Table 3). 70 Nevertheless, the complex between TCR and MIC A was particularly stable, suggesting a sequential model of binding where the initial contact by NKG2D is followed by the more stable TCR:MIC A complex. The role of TCR chains in complex formation remains to be fully resolved as the studies implicated partial contributions of the germline-encoded CDR loops from Vd1 TCR 70 while other studies showed that MIC A-associated recognition of tumour cells could be mediated by both Vd1 71 and Vd2 56,57,72 TCRs. Notably, TCR engagement has been shown to be indispensable for cd T cell-mediated cytotoxicity whereas NKG2D played only a co-stimulatory role; thus indicating the importance of TCR:MIC A complex formation. 73 ULBP molecules may be recognised in a similar manner as it has been shown that ULBP-4 engages both NKG2D and Vc9Vd2 TCRs. 74 The sequential recognition of different targets by cd T cells could therefore play an important role in immunosurveillance as it allows the T cell to rapidly scan the target cells for the signs of stress, indicating a possible infection or transformation. The requirement for a multicomponent stress context for full T cell activation could then provide fail-safe protection against autoimmunity. The apparent co-existence of diverse co-stimulatory axes decreases the chances of immune evasion.

HARNESSING cd T CELLS FOR THERAPY
Stimulating cd T cells in vivo with phosphoantigens As mentioned above, the Vc9Vd2 T cell subset that predominates in human peripheral blood is capable of responding to prenyl pyrophosphates. This feature has been employed to redirect Vc9Vd2 T cells to tumours by manipulating isoprenoid metabolism in the cancer cells. Such manipulation can be achieved using aminobisphosphonates (e.g. zoledronate, pamidronate, risedronate) which are structural analogues of prenyl pyrophosphates that contain an amino moiety. As a result, aminobisphosphonates act indirectly by inhibiting farnesyl pyrophosphate synthase (FPPS) in the mevalonate pathway leading to the accumulation of prenyl pyrophosphate substrates. 75,76 Recently, Idrees et al. screened over 50 tumour cell lines, showing a direct correlation between zoledronate-induced FPPS inhibition and tumour recognition by Vc9Vd2 T cells. 77 The FPPS-inhibiting concentration of zoledronate was on average two orders of magnitude lower than the concentration required for inhibition of tumour proliferation, thus indicating that T cell activation was not the result of cell death. Another group showed that risedronate caused PAg accumulation in the tumour cells which were then recognised by Vc9Vd2 T cells. 78 Upon recognition, T cells produced IFNc which caused upregulation of ICAM-1 on tumour cells, contributing to a positive feedback loop and subsequent cytotoxicity. Additionally, aminobisphosphonates, 79  phenotype of professional antigen-presenting cells upon activation. 80 Therefore, they may be used as a potent anticancer vaccine in addition to their broad antineoplastic cytotoxicity.

Redirecting T cells to tumours
In the absence of known ligands, cd T cells can be redirected to tumours using antibodies. The efficacy of bispecific antibodies, where one part recognised a tumour surface marker (for example, EpCAM on liver tumours or HER2/neu on pancreatic cancer) while the other binding site targets CD3 or the Vc9 chain of the TCR, has been demonstrated in pre-clinical models. 81 87 CAR-mediated signalling resulted in a similar cytokine secretion profile as TCR activation, and induced unbiased expansion of cd T cell subsets. Notably, CAR technology has been combined with the generation of induced pluripotent stem cells (iPSC) from human peripheral blood T cells. 88 Such cells exhibited a phenotype closest to cd T cells of all lymphocyte lineages and exerted similar in vivo antitumour activity to CAR-transduced cd T cells.
Finally, T cells can be redirected to tumours by introducing an exogenous TCR of known anticancer specificity into patient-derived peripheral lymphocytes prior to adoptive transfer of these cells. Most TCR gene studies have involved transduction of ab TCRs into ab T cells. 89 However, this strategy comes with the inherent risk of ab TCR mispairing between the endo-and exogenous TCR chains, resulting in receptors of unknown and potentially autoreactive specificities (Figure 6b). Such autoreactivity has been observed both in ex vivo human samples 90 and in mouse models. 91 cd T cells offer an attractive way to circumvent this problem as tumour-reactive ab TCRs can be introduced into these cells without the risk of mispairing ( Figure 6c). 92,93 Additionally, cd T cells transduced with ab TCRs retain the functionality of their original TCR and respond to stimuli transferred via either TCR with rapid, cdlike kinetics. 94 The main obstacle associated with ab TCR transfer, however, is that cd T cells usually do not express CD4 or CD8 co-receptors required for the efficient recognition of peptide-MHC. Thus efficient function might require co-transduction with a coreceptor 95 or use of enhanced affinity TCRs. 96 It is also possible to transduce peripheral lymphocytes (both of cd and ab origin) with a specific cd TCR. Zhao et al. demonstrated that T cells transduced with a Vc9Vd2 TCR, modified with a CDR3d loop specific for unidentified antigens on ovarian carcinoma, exerted an anticancer activity in vivo. 97 Similarly, the transduction of a PAg-reactive Vc9Vd2 TCR into ab T cells successfully redirected them towards cancer cells, as well as led to downregulation of their endogenous ab TCRs, thus abrogating alloreactive responses. 98 It is therefore likely that the detailed characterisation of cancer-specific cd TCRs will bring about new studies examining the efficacy of such TCRs in the TCR gene transduction setting.

Generating clinically relevant numbers of tumour-reactive cd T cells
The main hurdle that needs to be overcome in order to use cd T cells as a therapeutic is obtaining these cell in clinically meaningful numbers. For ab T cell-based therapies, this is a relatively straightforward task as the requirements for effective ab T cell expansion have been widely studied, and the cognate ligands can be readily used to stimulate the expanding cells. 99 In the case of cd T cells, the only known common ligands are PAgs which are recognised by the Vc9Vd2 subset. As aminobisphosphonates such as zoledronate cause accumulation of PAgs in the cells, they can be used to preferentially expand that T cell subset both ex vivo, prior to re-infusion, and directly in patients. However, PAgs and aminobisphosphonates expand only the Vc9Vd2 subset of cd T cells, and this bias may have a negative impact on clinical efficacy. In an alternative approach, immobilised anti-cd TCR antibody has been demonstrated to promote the expansion of all cd T cell subsets, without modulating their antitumour function. 100,101 This method, however, may lead to mitogen-induced T cell death and has not yet been used in a published clinical trial. Deniger et al. have recently proved that T cells with polyclonal cd TCR repertoires can be expanded to large numbers (.10 9 ) using IL-2, IL-21 and clinical grade artificial antigen-presenting cells, engineered to express co-stimulatory molecules CD19, CD64, CD86, CD137L and membrane-bound IL-15. 102 Notably, the antigen-presenting cells were derived from a tumour line K562, broadly recognised by cd T cells, and cirradiated prior to T cell exposure. The irradiation step could potentially enhance the recognition even further, as it leads to expression of cell stress molecules. Polyclonal cd T cell lines showed superior toxicity compared to those expressing only Vd2 chains, thus highlighting the possibility of enhancing cd T cell-focused immunotherapy. Similarly, Anderson and colleagues used the same antigen-presenting cell line to generate unbiased cd T cell products from peripheral blood of cancer patients. 84 cd T cells for cancer immunotherapy M Legut et al.
Finally, manipulations of the CD3 co-receptor can be applied to increase therapeutic efficacy of cd T cells. Dopfer et al. recently showed that two anti-CD3 antibodies affect the functional activity of cd T cells differentially. 103 The reported results indicated that an anti-CD3 antibody, OKT3, induced strong proliferation and cytokine secretion with decreased cytotoxic potential, while another anti-CD3 antibody, UCHT1, enhanced tumour killing via Ras/Erk and PI3K/Akt pathways, without an increase in cytokine secretion or proliferation. These results might pave the way for new therapeutic approaches focused on manipulation for CD3-mediated signal transduction to favour expansion of large numbers of cd T cells, with desired effector functions such as cytotoxicity.
Lessons learned from clinical trials Numerous attempts to use cd T cells in cancer immunotherapy have been made over the past decade, with variable efficacy and a good overall safety profile.
Two recent trials investigated the anticancer effect of zoledronate combined with low dose IL-2. In renal cell carcinoma, the combined treatment resulted only in a minor expansion of Vd2 pos cells which diminished after repeated treatment cycles without achieving the objective response in any of the enrolled patients. 104 A similar study by Kunzmann and colleagues showed no objective response in patients with solid tumours (melanoma and renal cell carcinoma), and a partial response in acute myeloid leukaemia patients. 105 Surprisingly, the treatment resulted in increased vascular endothelial growth factor levels, an indication of augmented angiogenesis, having a negative impact on the therapeutic outcome. No severe side effects (apart from grade 4 fever) were observed. A recent phase IV clinical trial of zoledronate in cancer-free patients demonstrated that the severity of transient inflammation-related side effects, caused by the treatment, could easily be predicted by ex vivo measurements of IFN-c production by zoledronate-activated peripheral blood cells. 106 Notably, repeated zoledronate treatment led to the depletion of the central memory subset of blood Vc9Vd2 T cells, as well as a long-lasting decrease of their absolute count. 106,107 This phenomenon may be caused by peripheral blood neutrophils which have been shown to ingest zoledronate and subsequently inhibit the proliferation of the T cells by production of hydrogen peroxide. 108 Repeated stimulation with PAgs, either accumulated in response to aminobisphosphonates or as a result of an accelerated mevalonate pathway, can also lead to a differentiation shift of Vc9Vd2 T cells and functional exhaustion. 109 Zoledronate therefore appears to be insufficient to generate an effective antitumour response per se -nevertheless, two multicentre studies demonstrated that treatment with zoledronate combined with chemotherapy improved the survival of breast cancer and multiple myeloma patients. 110,111 The adoptive cell transfer (ACT) of ex vivo expanded cd T cells appears to be a more effective treatment option than using zoledronate as a single agent. In metastatic renal cell carcinoma, two groups reported re-infusing autologous PAgexpanded Vd2 pos T cells. 112,113 When the T cells were infused with IL-2 only, 6 out of 10 patients experienced stabilisation of the disease whereas the infusion of the cells with both IL-2 and zoledronate resulted in 5 out of 10 disease stabilisations and 1 complete response. However, as both studies included only a small number of patients, it is possible that the observed superior efficacy of using both IL-2 and zoledronate was purely coincidental. When cells were expanded using IL-2 and zoledronate but infused without further exogenous stimulation, the partial response was observed in 3 out of 10 patients with various solid tumours. 114 As the patients were treated with chemotherapy together with ACT, the response could be contributed to the additive/synergistic effect of the combination therapy. A similar effect was observed by Nicol and colleagues as the complete response (1/17) and partial response (2/17) patients received both cell infusion and chemo/hormone therapy. 115 However, the patient cohorts in these studies were small and highly diverse, in terms of cancer type, additional therapy and composition of the T cell infusion. These variables make it difficult to assign the exact clinical contribution of cd T cells. ACT of cd T cells and other cytotoxic cell types can also increase the efficacy of radiofrequency ablation. 116 The aforementioned trials utilised patients' autologous peripheral blood T cells for ACT -however, in some instances the effector functions of autologous T cells are severely impaired. 109 A recent study by Wilhelm et al. demonstrated that it is possible to infuse patients with a cd T cell product obtained from haploidentical donors, with no evidence of graft versus host disease, showing substantial clinical efficacy (3 out of 4 patients with advanced haematological malignancies experienced complete regression). 117 The donor cd T cells persisted for 28 days and expanded in vivo, in response to exogenous IL-2 and zoledronate. Another group also showed that prolonged persistence of infused cd T cells can occur without IL-2 supplementation, probably relying on endogenous IL-15. 118 Moreover, exogenous IL-18 can be a potent co-stimulator of cd T cell proliferation. 119,120 Overall, while cd T cell therapy has a good safety profile, clinical performance to date has been underwhelming. 121 Anticipated discoveries of the various ligands recognised on tumour cells by the Vd2 neg T cell subset will lead to a better understanding of how these cells operate and an increased capacity to beneficially harness these mechanisms for immunotherapy.

CONCLUSIONS AND FURTHER DIRECTIONS
The evolutionary conservation of cd T cells in all jawed vertebrates and of an analogous third lymphocyte lineage in their jawless ancestors 2 attests to the critical importance of these cells in the maintenance of immune integrity. cd T cells are known to play an important role in infection 16 ; further recent discoveries have shown that this T cell lineage can respond to stress signals and thereby play an important role in cancer immunosurveillance. Indeed, mice lacking the cd TCR have increased cancer risk. 17 Despite the importance of cd T cells, very few ligands for the human cd TCR have been confirmed by biochemical and biophysical data and this field is still in its infancy. Interestingly, certain similarities exist between the few known human cd TCR ligands that might point to some generalised mechanisms by which these cells patrol the periphery. BTN3A1, MIC A/B and CD1-family proteins belong to the immunoglobulin superfamily (while EPCR is highly homologous to CD1d), and both infections and cellular stress seem to play a role in activating cd T cells via these immunoglobulin structures. Microbial infections lead to a stimulatory conformational change and/or direct metabolite presentation by BTN3A1 while herpesvirus (CMV, EBV) infections cause upregulation of EPCR, MIC A and hMSH2, respectively. Similarly, cell stress, including neoplasms, can involve deregulation of the mevalonate pathway 122 and subsequent PAg activation of BNT3A1 in addition to ectopic expression of cell stress markers such as EPCR, MIC and hMSH2, while CD1-family proteins can present endogenous lipids acting as a marker of malignant transformation (CD1c) 42 or viral infection (CD1d). 123 Interestingly, both CD1d and EPCR are capable of lipid presentation, and both can be recognised by cd T cell clones regardless of the lipid cargo -thus indicating that both molecules can act as an immunogenic marker per se. 35,47 Requirement for a wider cell stress context for cd T cell recognition potentially restrains these immune cells and provides a failsafe against autoimmune reactions. Cancer-associated cd TCR targets are essentially self-antigens. Accordingly, cd TCR interactions with the known targets are of low affinity (K D 5 10 -4 -10 -6 M; Tables 1-3), similar to ab TCR affinities for self-derived peptides. 124,125 To date, only two structures of human cd TCR in complex with a ligand have been resolved ( Figure 4). In both cases (CD1d-sulphatide and CD1d-aGalCer), the germline-encoded CDR loops engaged the antigen-presenting CD1d molecule while hypervariable CDR3 loops made contacts with the lipid ligand, suggesting that discrimination between subtly different lipid cargoes is possible. In contrast, PAg recognition may be mediated mostly by germline-encoded CDR loops from Vd2 and/or Vc9 chains as PAg-BTN3A1 can be recognised by a wide range of different Vd2 Vc9 T cell clones. However, the exact mode of PAg recognition awaits a BTN3A1: cd TCR complex structure.
In summary, cd T cell therapy for cancer has demonstrated a good overall safety profile although results to date have not lived up to their early promise. Nevertheless, we anticipate that many more ligands will be found for these important immune cells in the next 5 years thereby greatly expanding the therapeutic horizon. Therapeutic approaches to cancer treatment based around the ab TCR in gene transfer 99 or as soluble bispecifics 126,127 have burgeoned in the last two years. Importantly, cd TCRs do not cd T cells for cancer immunotherapy M Legut et al.
appear to suffer the limitation of being HLA-restricted so might offer exciting possibilities for new off-the-shelf, pan-population cancer immunotherapies in the very near future.

Mateusz Legut is a Cancer Research UK funded PhD student, David
Cole is a Wellcome Trust Career Development Fellow and Andrew Sewell is a Wellcome Trust Senior Investigator.