The potential of adoptively transferred immune cells to cure cancer was first demonstrated over 50 years ago, when leukaemia was found to be eradicated in irradiated mice receiving allogeneic but not syngeneic bone marrow transplants . Much later, it became clear that the same cells that caused graft-versus-host disease (GvHD) could also mediate a graft-versus-leukaemia effect , with a critical role for T cells being confirmed by the finding of increased leukaemic relapse in recipients of T-cell depleted bone marrow transplants [3, 4]. The major goal of cellular immunotherapy has subsequently been to harness the cytotoxic and other immunomodulatory capabilities of T cells to eliminate cancers or viral infections without eliciting undesired effects such as GvHD. An approach that was quickly identified involves the adoptive transfer of CD8+ T cells that are specific for antigens expressed only, or at much higher levels, in cancer or virally infected target cells. Both autologous and allogeneic T cells have been isolated, expanded in vitro in conditions that favour cells specific for a target antigen, and infused into patients successfully to treat Epstein–Barr virus (EBV) , cytomegalovirus (CMV) [6, 7] and melanoma . Essential to this strategy is the isolation of T cells with high functional avidity for the relevant antigens. The techniques of T cell receptor (TCR) gene therapy offer an elegant solution: the transfer into a T cell of genes for the α and β chains of a specific TCR causes redirection of the specificity of that cell to the target of the transferred TCR , allowing large populations of antigen-specific T cells to be generated by transduction with the genes for a TCR with high avidity for the antigen of interest. The therapeutic potential in humans of such gene-modified T cells was first shown by Morgan et al. , who documented the regression of metastatic melanoma in two out of 15 patients treated with adoptively transferred autologous T cells that had been transduced with a TCR specific for the tumour-associated antigen MART-1.
The principle of using gene therapy to redirect the specificity of T cells for therapeutic purposes has been extended by the use of artificial TCRs. These receptors comprise a domain that binds strongly to the antigen being targeted, linked to intracytoplasmic structures that are usually derived from the signalling domains of conventional immune receptors and co-receptors. This allows the effects of ligand binding to be transduced through existing pathways to activate the T cell. Because these artificial receptors combine structures from several different molecules, they are known as chimeric antigen receptors (CARs); these have been reviewed recently [11, 12].
Whether using TCRs or CARs, gene therapy to redirect T cells has the same requirements for success. The gene-modified T cells must have optimal avidity for their cognate antigen, determined by the affinity with which the introduced receptor binds its antigen, and the level of its expression on the T cell surface; they must persist in vivo sufficiently to exert the desired therapeutic effect; and they must be safe to use, lacking on-target or off-target toxicity.
In this review, we discuss the challenges of T cell gene therapy, including the optimization of affinity and surface expression of the introduced TCR, the choice of T cell subpopulations for gene transfer, and the promotion of persistence of gene-modified T cells in vivo. We review the safety concerns surrounding the use of gene-modified T cells in patients, discussing emerging solutions to these problems, and describe the increasingly positive results from the use of gene-modified T cells in clinical trials of adoptive cellular immunotherapy.
Maximizing surface expression of introduced TCR
T cells are activated and gain effector function when a threshold number of serial engagements of their TCRs with peptide-major histocompatibility complexes (MHCs) is reached [13, 14]. Although the level of this threshold is altered by co-stimulation , T cell activation remains critically dependent on the number of TCRs on the cell surface and the affinity with which each TCR binds its antigen: optimizing these two attributes is critical to the success of TCR gene therapy.
The artificial nature of CARs brings both advantages and disadvantages. The binding domain of a CAR is often derived from the antigen-binding fragment (Fab) of a monoclonal antibody, with the two chains of this variable region being artificially linked to form a single chain variable fragment (Figure 1). Although this allows binding of antigens on cell surface molecules with high affinity, CARs are unable to recognize antigens derived from intracellular molecules that have been processed and presented in the context of human leukocyte antigen (HLA): this excludes many antigens that may be specific and of functional importance to virally infected and cancer cell targets. This is not the case with native αβ TCRs, which recognize their cognate antigens only in the restrictive context of specific HLA molecules; however, this benefit of a wider range of potential antigens is partially offset by the additional difficulties that HLA restriction poses for clinical translation, and the potential for tumours to escape immune recognition by down-regulating HLA . The lack of structural similarity of CARs to endogenous TCR chains abolishes the possibility of mispairing, overcoming a major concern in TCR gene therapy; however, the nonphysiological means of intracellular signal transduction, particularly in first-generation CARs lacking co-stimulatory domains, may diminish their efficacy , and the potential immunogenicity of nonhuman engineered sequences may limit the persistence of CAR-transduced cells in vivo [17, 18].
The assembly and surface expression of an introduced TCR is a complex process, requiring pairing of the introduced α and β chains to form a heterodimer that then associates with the four invariant CD3 chains, γ, δ, ε and ζ (Figure 1). The concentration of CD3, particularly the ζ component, is limiting ; incomplete TCR–CD3 complexes are degraded in the endoplasmic reticulum . In transduced T cells, the introduced α or β chains have the potential to form heterodimers with endogenous complementary β or α chains, giving rise to a TCR of unpredictable specificity in a process known as mispairing. Mispaired TCRs may compete with the introduced TCR for CD3, inhibiting the redirection of the T cell to antigens recognized by the introduced TCR; more seriously, mispairing may produce TCRs specific for self-antigens, generating autoreactive T cells that have not been subjected to central tolerance and may cause a lethal GvHD-like syndrome .
The surface expression of any introduced TCR is also affected by properties that are intrinsic to the TCR: a ‘strong’ TCR will be expressed at higher levels than a ‘weak’ endogenous TCR, replacing it on the cell surface, whereas two ‘strong’ TCRs will be co-expressed [22, 23]. The structural differences that confer the range of strengths found within the endogenous TCR repertoire remain incompletely understood; however, several modifications to the structure of introduced TCRs have been shown successfully to increase surface expression, either alone or in combination [24, 25].
Structural modifications to enhance expression and reduce mispairing
Following the unexpected observation that human lymphocytes showed increased antigen-specific activity when transduced with a fully mouse-derived TCR compared with the same TCR modified to contain human constant α and β chain regions , Cohen et al.  showed that a hybrid TCR with human variable domains and mouse constant regions functioned better in human cells than its fully human counterpart. This reflected both enhanced stability of the hybrid TCR–CD3ζ complex and preferential pairing of the murine constant regions with themselves rather than the endogenous human TCR chains. Although initial studies used TCRs specific for antigen derived from melanoma, the approach has subsequently been adopted for other TCRs . However, the presence of nonhuman peptide sequences in an introduced TCR may elicit immune responses directed against these regions , potentially reducing the longevity of engineered T lymphocytes in vivo. More recently, the few amino acid residues responsible for the improved function of murine TCRs have been identified, allowing TCRs to be minimally modified to reduce immunogenicity [30, 31].
Minimal structural modifications
The α and β chains of wild-type TCRs are covalently linked by a disulphide bond in the constant region. Replicating a strategy originally used to stabilize soluble TCRs for crystallization , two groups used point mutations to replace the threonine-48 on the α chain and serine-57 on the β chain with cysteines, allowing an additional disulphide bond to be formed between the constant regions [33, 34]. The modified peptides pair more efficiently, give rise to increased TCR expression at the cell surface and mediate improved antigen-specific responses [33, 34]. Recently, the additional disulphide bond has been shown to reduce autoimmune pathology caused by mispairing with endogenous TCR chains .
Detailed analysis of the crystal structure of the TCR constant domains has allowed identification of two residues that interact at the interface of the α and β chain constant regions: a glycine on the α chain and an arginine on the β chain interact in a way sterically and electrostatically analogous to a ‘knob-in-hole’ configuration. Swapping these residues between α and β chains, giving a ‘hole-in-knob’ configuration, favours selective assembly of the introduced TCR at the same time as preserving its function .
Optimizing equimolar translation of introduced TCR α and β chains
Even minimal structural modifications have the potential to increase the immunogenicity of an introduced TCR. An alternative approach is to focus on improving the translation of the transgenic TCR mRNA. The use of codons optimized for expression in the host species has been demonstrated to improve protein expression in many systems , and the same is true when expressing TCRs in transgenic human T cells [24, 37, 38], with consequent improvements in antitumour responses .
Although it is possible to transduce a T cell with both α and β chains of a TCR in two separate vectors, or in a single vector with two separate promoters, both these approaches risk imbalanced expression of the introduced TCR chains, increasing the possibility that a chain expressed in excess will mispair with endogenous TCR chains. Although the internal ribosome entry site (IRES) sequence of encephalomyocarditis virus has been widely used to construct bicistronic viral vectors, IRES-mediated translation is relatively inefficient  and, in recent years, the use of ‘self-cleaving’ 2A peptide sequences derived from picornaviruses or porcine teschovirus to achieve equimolar expression of both introduced TCR chains has been almost universally adopted. Through a ribosomal skip mechanism, these sequences induce translation of two separate peptides from a single mRNA transcript, achieving near stoichiometric production of each peptide . Although a short 2A peptide sequence remains attached to the C-terminus of the first translated peptide, there have been no reports of consequent immunogenicity or impaired function.
Enhancing TCR function through co-transducing with CD3
We have recently shown that the provision of endogenous CD3 is rate limiting for the expression of introduced TCR in gene modified T cells. Using two different TCRs modified by codon optimization and incorporation of an additional disulphide bond, we demonstrated that co-transduction of murine T cells with the genes encoding the γ, δ, ε and ζ chains of the CD3 complex alongside those for the TCR resulted in 16–20-fold enhanced TCR expression, increased avidity for cognate antigen, improved tumour clearance and more effective memory responses . Although it is possible that the enhancement of TCR expression through reduced competition for CD3 may also increase the expression of mispaired TCRs, we have previously found that the use of strong TCRs completely suppresses the surface expression of endogenous TCR .
Down-regulation of endogenous TCRs
Rather than increasing supply of the CD3 complex, an alternative strategy to limit the effects of competition for CD3 on expression of an introduced TCR is to reduce demand for CD3 by down-regulating the expression of endogenous TCRs. This has the theoretical advantage of simultaneously decreasing the potential for mispairing with endogenous TCR chains. Okamoto et al.  designed a vector encoding the α and β chains for a TCR recognizing the tumour-specific antigen MAGE-A4, alongside small-interfering RNAs (siRNA) for conserved elements of wild-type TCR α and β chain constant regions; down-regulation of the introduced MAGE-A4 TCR was elegantly avoided through codon optimization of its encoding sequences, rendering it resistant to RNA interference . Primary human T lymphocytes transduced with the MAGE-A4-TCR-siRNA vector demonstrated increased expression of the introduced TCR and a reduced, but not eliminated, expression of endogenous TCR. More recently, the addition of siRNA to a Wilms tumour antigen 1 (WT1)-specific TCR vector was shown to confer increased antileukaemia activity in vivo; a reduced tendency to lose antigen specificity after expansion was also noted, possibly reflecting diminished production of mispaired nonspecific TCRs . These promising results from RNA interference lend theoretical support to a novel strategy that uses zinc finger nucleases to disrupt endogenous TCR α and β chain genes at the DNA level. Introduction of zinc finger nucleases into these genes has been shown completely to abrogate expression of endogenous TCRs ; subsequent transduction with a WT1-specific TCR generates cells that recognize this target but are devoid of residual endogenous TCR reactivity, including the alloreactivity that is responsible for unwanted effects such as graft-versus-host disease .
Identifying high avidity T cells and increasing TCR affinity
So far, the main focus in the development of TCR gene therapy has been to target cancer related disease . Some antigens expressed by cancer cells, including those that are the products of abnormal fusion genes such as bcr-abl in chronic myeloid leukaemia, are specific to the cancer (‘tumour-specific antigens’): these antigens evoke a strong response from T cells. However, many markers of cancer cells are also found, at lower density, in normal tissues (‘tumour-associated antigens’); autologous T cells recognizing these ‘self’ antigens are usually of low avidity, with the high-avidity T cell clones having been deleted through tolerogenic processes that serve as a natural safety mechanism to delete self-reactive T cells and hence prevent autoimmunity.
To overcome this obstacle to isolating T cell clones with high avidity for tumour-associated antigens (TAA), researchers have resorted to using novel systems where a particular MHC/TAA peptide combination is not expressed during the establishment of tolerance.
Isolation of high-avidity T cell clones from artificial nontolerized environments
This method, first described by Theobald et al. , exploits peptide sequence differences between mouse and human proteins involved in cancer. Although CD8+ T cell clones with high avidity for the TAA p53 have generally been deleted in humans, the vaccination with human p53 peptide of mice transgenic for the human major histocompatibility molecule HLA-A2 causes expansion of murine CD8+ T cell clones with high avidity for human p53; the TCR genes from these clones can be isolated and used to redirect human CD8 or CD4+ T cells against p53 expressing cells [47, 48]. Using the same method, T cell clones with high avidity for other TAA, including MDM2 , carcinoembryonic antigen (CEA)  and gp100 , have been isolated. The potential immunogenicity of high-avidity TCRs isolated in murine systems may be avoided using mice transgenic for both human TCR and MHC genes .
Isolation of high-avidity allo-MHC-restricted T cell clones from MHC-mismatched donors
An alternative, allo-MHC-restricted approach exploits the natural repertoire of lymphocytes from HLA-A2 negative donors to isolate T cell clones with high avidity for TAA presented by HLA-A2 , a strategy first used in mice . Using this approach, human T cell clones with high avidity for several TAA, such as cyclin-D1 , WT1  and MDM2 , have been isolated and their TCR genes cloned into vectors for gene therapy [55-57]. Gene transfer of allo-HLA-A2/TAA specific TCR genes into polyclonal human T cells has been shown to redirect specificity towards a broad range of tumour cell lines [26, 52] and cancer cells isolated from patients with leukaemia . Allo-HLA-A2/TAA specific TCR-transduced human T cells have also been shown to eradicate human tumours in xenograft mouse models .
The use of TAA peptide-loaded HLA-A2 positive stimulator cells to isolate TAA peptide-specific T cell clones from HLA-A2 negative donors is unsuccessful in many cases because allogeneic stimulator cells often provoke dominant CD8+ T cell responses against allogeneic epitopes unrelated to the A2-presented TAA epitope of interest. There have been new methods to simplify and refine the process of identifying allo-HLA-A2/TAA specific T cell clones. One different approach is to tag HLA-A2 recombinant molecules containing the relevant peptide to B cells from HLA-A2 negative hosts for T cell isolation . This approach allows the introduction of the allogeneic HLA-A2/TAA complex as the sole antigen, precludes the need to use peptide loaded cell lines and reduces the risk of contamination. Another strategy, developed by Wilde et al. , is to use engineered RNA encoding HLA-A2 and a complete tumour protein to transfect HLA-A2 negative donor-derived professional antigen-presenting cells so that HLA-A2/TAA complexes are expressed. This system has the advantage of negating the need to identify the immuno-dominant peptide within the tumour protein, and has recently been used to identify T cell clones against a novel TAA, survivin . Lastly, the advent of HLA-A2 multimer, an engineered construct comprising multiple HLA-A2 recombinant proteins containing the relevant peptide, has made it possible to identify and isolate antigen-specific T cells directly from HLA-A2 negative donors [61, 62] or from HLA-A2 positive leukaemic patients who received mismatched HSCT from HLA-A2 negative donors . Using HLA-A2 multimers, Amir et al.  isolated from one such patient T cell clones that were specific for the HLA-A2/melanoma TAA PRAME.
The issue of antigen specificity in TCRs derived from T cells that did not undergo maturation and thymic selection in the context of HLA-A2 is important: these TCRs may show poor discrimination of different peptides presented by HLA-A2. Recent studies have gone to great lengths to address this but have achieved different outcomes. From a patient with severe GVHD after HLA-A2 mismatched donor lymphocyte infusion, all but one of the 50 allo-HLA-A2 restricted T cell clones recognized a single HLA-A2/peptide complex, indicating fine antigen specificity . A detailed structural study of one allo-HLA-A2-restricted TCR further reinforces the notion that the TCRs from allo-HLA-A2/TAA-restricted T cells have antigen specificity focused around the TAA peptide . However, a recent study found that allo-HLA-A2-restricted T cell clones isolated from HLA-A2 negative patients previously vaccinated with WT-1 peptide also recognized other HLA-A2 peptides with high avidity, suggesting a significant degree of antigen promiscuity . The different conclusions were likely to be related to the different settings and methodology used in these studies, but nevertheless stringent testing of antigen specificity is required for allo-HLA-A2/TAA-restricted T cells before clinical use.
Because TAA may be expressed in normal tissues, allo-restricted genetically modified T cells may also have the unintended consequence of targeting normal tissues. The allo-HLA-A2/PRAME-specific T cell clones isolated by Amir et al.  were not only highly specific towards a broad range of tumour cell lines and leukaemia, but also had reactivity towards kidney epithelial cells. In a different study, Leisegang et al.  found that HLA-A2 positive T cells transduced with an allo-HLA-A2-restricted TCR targeting survivin underwent extensive apoptosis while in culture: survivin was expressed on the same T cells, survivin was expressed on the same T cells, resulting in HLA-A2 restricted fratricide. Although it is possible to reduce the risk from undesired targeting of vital tissues by inclusion of suicide genes, it is much more difficult to circumvent the problem of HLA-A2 restricted fratricide if the TAA is also expressed in lymphocytes and therefore this situation should be avoided.
In vitro affinity maturation of TCRs
Affinity maturation refers to a natural process by which B cell clones achieve higher affinity B cell receptors to cognate antigen, and are conferred with a survival advantage. The affinity of TCRs cloned from naturally circulating T cells is low (10–4 to 10–6m) compared with the affinity of antibodies for cognate antigen (10–9 to 10–12m), reflecting the fact that T cells do not undergo affinity maturation in vivo. Therefore the ability to use in vitro methods to ‘affinity mature’ TCRs provides an attractive avenue of improving TCR gene therapy.
Affinity maturation strategies using libraries of mutants to cover different combinations of amino acid substitutions have been adapted for the purposes of increasing affinity of a few well characterized TCRs . Not surprisingly, the most effective mutations were directed against antigen-binding regions of the TCR [68-71]. Techniques used include error-prone PCR , site-directed mutagenesis  and structure-based design [73, 74]. Affinity matured TCRs were displayed using yeast, phage or mammalian T cells, initially with single short chain gene constructs containing the variable (V) regions of the TCR α and β chains (Vα-linker-Vβ). Systems using yeast [68, 75] and phage [76, 77] allowed large libraries of short-chain TCRs containing mutations to be expressed with a yeast or phage surface protein. However, the expression, folding and association of single chain Vα and Vβ TCR chains in non-mammalian cells often require additional genetic changes in the framework regions of the TCR . A recent innovation was the ability to express full-length TCR chains in a multi-cistronic vector for phage display . The more physiological approach of screening TCR mutants by transduction into TCR-deficient murine T cells has been used, although the screening output is much reduced compared to yeast or phage display .
Following display on a cell surface, mutated TCR chains are selected on increased ability to bind multimer. The cDNA of the affinity matured TCR is subsequently characterized and can be used for gene transfer. The current affinity matured TCRs all have slower MHC dissociation rates [72, 76, 77, 79, 80], a property associated with increased functional avidity . Usually single or dual amino-acid substitutions within the antigen-binding region of the TCR are sufficient to increase affinity several hundred fold [76, 77, 80]. Although many of the high affinity TCR retain antigen specificity , there are examples in the literature of cross-reactivity to self-peptides , alloreactivity  and reactivity to antagonistic peptide . Functional data using affinity matured TCR expressed in T cells or cell lines have lagged behind biochemical studies. Surprisingly, screening in T cell hybridomas showed that some high affinity TCRs have no (or even a negative) effect on peptide sensitivity , whereas, in others, there was only modest enhancement .
A recent study addressed in greater detail the extent to which affinity matured TCRs improve the ability of primary CD8+ T cells to mount antigen-specific responses . Using a TCR specific for TAX peptide presented by HLA-A2, and three affinity matured variants with up to 700-fold increase in affinity, it was shown that CD8+ T cells with high affinity TCR have faster responses. However, CD8+ T cells expressing the high affinity TCR failed to respond to low concentration of antigen, indicating that further affinity maturation does not necessarily translate to improved sensitivity.
Some studies have aimed to improve T cell function by modifications outside the antigen-binding regions of the TCR. One group found that an amino acid substitution at position 107 in the TCR framework region stabilizes the CD3β loop and enhances antigen sensitivity in several TCRs . Even manipulating TCR constant regions by removing N-glycosylation sites can have profound effects on enhancing antigen sensitivity . Beyond the TCR, molecules involved in TCR signalling, such as linker of activated T cells, can be engineered to be more resistant to degradation, enhancing T cell function .
The modifications to optimize antigen binding, enhance expression and improve signal transduction of the TCR complex are summarized in Figure 2.
Generating antigen-specific helper, killer and suppressor T cells
Although early efforts have focused on generation of antigen-specific CD8+ T cells, more recent work has broadened the targets of gene therapy to include CD4 helper and regulatory T cells, with the aim of redirecting their antigen specificity and function in a way that can be controlled for therapeutic use. The most common non-CD8+ T cell engaged for TCR gene therapy is the CD4+ T cell. CD4+ T cells engage peptide presented by class II MHC, which is found on professional antigen-presenting cells, such as dendritic cells. While the primary function of the CD8+ T cell is as a cytotoxic effector, CD4+ T cells have a more diverse role in orchestrating the adaptive immune system, augmenting CD8+ T cell function and inducing long-term memory.
Even though TCRs isolated from T cells with high avidity for tumour antigens are mostly MHC class-I restricted and function optimally in the presence of CD8 co-receptor, many have been found to function in CD4+ T cells where the CD8 co-receptor is absent [55, 86, 87]. There has been a great interest in this finding because few natural CD4+ T cells recognizing tumour targets have been isolated. Gene transfer using MHC class I-restricted TCRs enables the generation of tumour-specific CD4+ T cells, augmenting the ability of tumour specific CD8+ T cells to kill tumour [55, 86, 88, 89]. The mechanism for this form of CD4+ help is uncertain, although it may involve production of the T cell growth cytokine interleukin (IL)-2, or activation of dendritic cells through CD40/CD40L interaction [55, 87]. The function of MHC class I-restricted TCRs in CD4+ T cells can be improved by introducing the CD8 co-receptor [86, 87, 89].
Depending on their cytokine milieu, naive CD4+ T cells differentiate into distinct populations, including Th1, Th2, Th17 and regulatory T cells (Tregs). These populations have different cytokine profiles, which can then influence other immune cells, including CD8+ T cells . In general, Th1 cells are important for immunity to viral infections, Th2 cells for immunity to bacterial and helminthic infections, and Tregs for suppression of autoimmunity. Therefore, it is crucial to utilize the appropriate CD4 cell type for the relevant therapeutic indication. Many of the TCR gene transfer protocols using retroviral vectors require the activation of T cells with concanavalin A or anti-CD3 antibodies, inducing a Th1 cytokine signature of interferon (IFN)-γ, tumour necrosis factor (TNF)-α and IL-12 [86, 88, 89, 91], which co-incidentally is appropriate for tumour immunotherapy. Although most studies on the tumour protection effect of transferring CD4+ tumour-specific T cells have used unsorted CD4+ T cells, some studies have removed existing Treg populations to prevent suppression of the desired effects [88, 91, 92]. CD4+ T cells gave a similar level of tumour protection to CD8+ T cells transduced with the same TCR in one murine study . There is currently no published clinical trial using gene modified CD4+ T cells, although adoptive transfer of CD4+ T cells originating from naturally occurring tumour-infiltrating-lymphocytes (TILs) achieved long-term tumour regression in one patient , suggesting that CD4+ T cells can have beneficial anti-tumour effects in the absence of CD8+ T cells.
Two studies have looked into the mechanisms of tumour protection afforded by CD4+ T cells taken from transgenic mice [95, 96]. Both showed that the cytotoxic effect of these TCR transgenic CD4+ T cells was sufficient to eradicate established tumours in mice; anti-tumour effects were augmented when Tregs were depleted. The CD4+ T cells had features characteristic of effector CD8+ T cells, expressing perforin, granzyme, CD107 and Th1 cytokines. An additional effect was the up-regulation of MHC class II on tumour cells, allowing CD4+ T cells to engage tumour directly . Endogenous CD8+ T, B, natural killer (NK), NKT and endogenous IFN-γ producing cells were not required for the anti-tumour effect, although IFN-γ production by CD4+ T cells was crucial .
Apart from introducing TCR genes into CD4+ T cells to target tumour cells, TCR-transduced CD4+ T cells can also be utilized to suppress autoimmunity. We demonstrated this in a murine model of arthritis, using both isolated natural CD4+ CD25+ Tregs and conventional CD4+ T cells that were ‘converted’ into Tregs by transducing with the FoxP3 transcription factor . Although natural Tregs suppressed the accumulation of pathogenic Th-17 cells more effectively than converted Tregs, both types of TCR gene-modified Tregs were able to reduce joint swelling and inflammation in an antigen-specific manner, resulting in complete remission of disease. Other studies have shown the efficacy of using CAR-transduced natural Tregs to ameliorate colitis . The feasibility of utilizing Tregs for TCR gene transfer has also been recently studied in the human setting, using both natural Tregs  or Foxp3-transduced ‘converted’ conventional CD4+ T cells . However, clinical studies using TCR gene modified Tregs have not yet been performed.
Promoting in vivo persistence of gene-modified T cells
One of the most important challenges of successful cancer immunotherapy is for the genetically modified T cells to persist beyond a few months after transfer. This has proved to be a greater challenge for T cells modified with CAR genes than T cells modified with conventional TCR genes. The data from two phase I human trials using CAR-transduced T cells showed that in vivo persistence may be quite limited: from as low as 1–7 days in patients with bulky disease  up to 6–12 weeks for most patients [101, 102]. TCR-transduced T cells, on the other hand, have been shown in murine  and human studies [10, 50] to have a greater propensity to persist after transfer. This difference is probably related to the inferior signal transduction of earlier CAR constructs: recent molecular engineering of CAR constructs to include the co-stimulatory domains CD28  or CD27  have resulted in improved persistence.
Strategies commonly used in clinical protocols to increase persistence of transferred T cells include administration of exogenous IL-2 [10, 50, 102, 106] and non-myeloablative lymphodeletion with radiotherapy  and/or chemotherapy [10, 108, 109]. It is generally accepted that lymphodepleting conditioning therapies increase the availability of T cell growth cytokines by decreasing competition from endogenous T cells. Animal studies have shown that, without lymphodepletion, transferred T cells do not persist and were not able to eradicate tumours; the alternative strategy of vaccine-induced activation of transferred genetically modified T cells was inferior to sublethal irradiation .
Systemic IL-2 cytokine treatment is widely recognized to cause significant toxicity, especially when high doses are used [102, 106, 110], and this limits the duration of treatment to a few days. Innovative attempts to circumvent this problem include transducing cytokine genes and receptors into transferred T cells, although the results obtained have been mixed: the IL-2 gene was transduced into TILs taken from melanoma patients but did not increase the ability to persist beyond 4 months when the cells were reintroduced . In a separate study, introduction of the IL-12 gene into pmel-1 transgenic CD8+ T cells conferred increased anti-tumour effects without the need for IL-2, but also did not increase survival . Although there are in vitro data showing that increasing the surface expression of cytokine receptors such as IL-7R , as well as the use of chimeric receptors with IL-2/IL-15 signalling domains , can improve the expansion of CD8+ T cells, no effect has so far been shown on persistence in vivo.
Most ex vivo cell preparation protocols for gene transfer into T cells have involved the use of exogenous IL-2 to expand or activate T cells. Other γ-chain cytokines, including IL-7, IL-15 and IL-21, whether used singly or in combination, were all superior to IL-2 in not driving CD8+ T cells into the terminal differentiated state during T cell activation . In particular, IL-21 was shown to be most potent influence in suppressing terminal differentiation [115, 116]. Although IL-21 primed naïve CD8+ T cells produce less IFN-γ in vitro than IL-2 or IL-15 primed cells, they were more effective at eradicating tumour cells in an in vivo animal model .
The finding that differentiated T cells were paradoxically less effective for tumour eradication led other studies to focus on the use of less differentiated T cells. One study showed, in a TCR transgenic model, that naïve CD8+ T cells mediated better anti-tumour immunity than central memory T cells (Tcm) . Another population of naturally occurring T cells, termed memory stem cells (Tscm) can be generated using a GSK kinase inhibitor (TWS119) to induce the WNT β-catenin pathway in naïve T cells [119, 120]. Genetically-modified Tscm expressing tumour-specific CAR were superior to memory T cells in persisting and controlling tumour in a xenograft model . We have transduced haemopoietic stem cells (HSC) with a TCR specific for WT1, and found that these cells can undergo positive thymic selection to become mature CD8+ T cells that display rapid antigen-specific responses . Other studies have confirmed that TCR gene-transduced HSC can give rise to CD8+ T cells with highly potent anti-tumour effects .
Different populations of memory T cells arise after initial encounter with antigen. To study which memory population was superior in persistence after adoptive transfer, Berger et al.  obtained CMV specific CD8+ T cells from nonhuman primates and sorted them into CD62L-hi central memory T cells (Tcm) or CD62L-lo effector memory T cells (Tem). In vitro activation and transduction of marker genes resulted in both T memory groups increasing the expression of granzyme and perforin and down-regulating CD62L . However, upon re-infusion, Tcm but not Tem antigen-specific CD8+ T cells reacquired CD62L and homed to lymph nodes, persisting for substantially longer periods.
Another promising way to influence the persistence of CD8+ T cells is by manipulating metabolic pathways. Two recent studies have shown that rapamycin, which prevents mTOR from forming the mTORc1 complex , and metformin, which stimulates mitochondrial fatty acid oxidation , can enhance the development of memory T cells. Subsequent to these studies, the combination of rapamycin treatment with lymphodepletion has been shown to increase the anti-tumour efficacy of naïve OT-1 transgenic CD8+ T cells .
Addressing safety concerns
The therapeutic potential of adoptively transferred redirected T cells is accompanied by the possibility of toxicity. On-target therapeutic effects such as tumour lysis syndrome , on-target side-effects as a result of low-level expression of a targeted antigen on normal tissues [50, 128, 129] and off-target effects as a result of unforeseen cross-reactivity of engineered T cells  have all been described. Stable expression of a transgene requires integration into the host genome, which is an event that risks insertional mutagenesis . Although broadly T-cell depleting therapies, including antibodies and immunosuppressive agents such as corticosteroids, will eliminate undesired autoimmunity, they will also impair beneficial immune responses . More specific strategies have therefore been developed to allow targeted eradication of engineered T cells.
Incorporation of herpes simplex virus thymidine kinase (HSV-TK)
Transduction of allogeneic donor lymphocytes with HSV-TK has been used to confer sensitivity of these cells to ganciclovir, with the aim of allowing their elimination in the event of GvHD following donor lymphocyte infusion after allogeneic stem cell transplantation . Limitations of this approach include immunogenicity of the viral proteins, failure of HSV-TK to impart ganciclovir sensitivity to slowly proliferating cells in patients with chronic GvHD , loss or silencing of the HSV-TK transgene , and removal of ganciclovir from the options to treat CMV and EBV infections.
Addition of antibody targets
An alternative approach has been to express on the T cells surface molecules that can function as targets for therapeutic antibodies. The first such molecule was CD20, a B cell marker that is the target for rituximab, a humanized antibody that is in widespread clinical use [135, 136]. Complement-dependent cytotoxicity and antibody-dependent cellular cytotoxicity are considered to be important mechanisms of rituximab killing, with the former dependent on high CD20 expression , giving rise to the possibility that T cells transduced to express a low level of introduced TCR and CD20 might still be able to exert a toxic effect at the same time as remaining resistant to rituximab. Other disadvantages include the depletion of normal B cells and hence humoral immunity after rituximab treatment, and the as yet unconfirmed possibility that expression of a B cell marker in T cells might subvert their function. A refinement of this approach is to transduce T cells with a construct encoding the minimal epitope of CD20 required for recognition by rituximab, reducing the risk of functional disturbance in the transduced cells (M. Pule, personal communication).
A similar strategy has used a ten-amino acid sequence of the human c-myc protein to provide a target for anti-myc antibodies. Incorporation of this ‘myc tag’ at the N-terminus of the TCR α chain variable region allowed efficient depletion of adoptively transferred T cells, rescuing mice from lethal autoimmunity in a model of diabetes .
Introduction of apoptotic proteins
Conditional alleles based on intermediates within the apoptotic pathway have been used to regulate cellular apoptosis. Target pro-apoptotic proteins, such as FADD (Fas-associating protein with death domain)  are fused to a domain, such as FK506-binding protein, that binds a small lipid-permeable ligand . Introduction of this ligand causes dimerization of the target protein, activating the apoptotic pathway . This approach allowed the successful elimination of autologous transferred T cells in a primate model, although resistance to apoptosis was seen in cells expressing only low levels of the suicide construct . Molecules such as Fas or FADD are upstream of several inhibitors of apoptosis (e.g. bcl-2, bcl-xL, c-FLIP) that may be up-regulated ; recent interest has focused on the use of inducible caspase 9, a late-stage apoptosis pathway molecule that is less subject to anti-apoptotic regulation. In human recipients of haploidentical stem cell transplants for relapsed leukaemia, ligand-induced dimerization of transduced caspase 9 has been successfully used to eliminate over 90% of T cell effectors of GvHD at the same time as preserving antiviral responses .
Clinical experience with TCR and CAR therapy
The observation that adoptive transfer of autologous TILs can mediate objective clinical responses in patients with metastatic melanoma  was an early spur to the field of T cell immunotherapy for cancer, with subsequent studies showing response rates in melanoma improving to 72% with conditioning regimens comprising more aggressive lymphodepletion . However, the difficulty of generating TILs for other cancer types, and the finding that tumour-reactive TILs could be isolated from only half of resected melanoma samples , has intensified efforts to bring to clinical use autologous T cells transduced to express TCRs or CARs against tumour antigens. Twenty-eight such clinical trials are currently registered with the US National Institutes of Health (http://clinicaltrials.gov/), with others already being concluded; the more significant results are summarized below.
The first study to report outcomes after adoptive transfer of TCR gene-modified T cells into patients used autologous T cells transduced with a TCR recognizing MART-1, a melanoma/melanocyte differentiation antigen. Two out of 15 patients (13%) showed full clinical regression of metastatic melanoma, with gene-transduced cells persisting in the circulation for 1 year after treatment . Subsequent to this pioneering study, the same group reported objective cancer regressions in 30% of patients using a TCR with higher avidity for the MART-1 antigen ; however, in contrast with the previous trial, this high-avidity TCR also caused destruction of normal melanocytes in skin, eye and ear, with some patients requiring local steroid administration to treat uveitis and hearing loss. To address this problem, the group has targeted NY-ESO-1, a cancer/testis antigen that is expressed in many cancers, including melanoma and synovial cell sarcoma. Unlike MART-1, the expression of NY-ESO-1 in normal adult tissues is restricted to cells in the testis that lack class I major histocompatibility antigens, and so are not recognized by T cells directed against the antigen. In a trial that used lymphodepletion and IL-2 therapy to promote persistence of transferred autologous anti-NY-ESO-1-directed T cells, objective clinical responses were seen in five of 11 patients with melanoma and four of six with synovial cell sarcoma, with no evidence of toxicity attributable to the transferred cells .
The potential danger of using high-avidity TCRs that target antigens found on normal tissues in addition to tumours has been confirmed in a recent study using T cells engineered to express a TCR with high avidity for CEA. CEA is a tumour-associated antigen overexpressed in many epithelial cancers, including colorectal adenocarcinoma, and is also expressed on normal colonic epithelial cells. Although there was regression of metastatic colorectal cancer in one of three patients treated with autologous gene-modified anti-CEA T lymphocytes, all three developed a severe and dose-limiting colitis .
Similar on-target toxicity has been documented in early trials of T cells redirected with CARs. The first three patients with metastatic renal cell carcinoma to receive T cells transduced with a CAR recognizing the tumour-associated antigen carbonic anhydrase IX (CAIX) developed liver enzyme disturbances as a result of CAIX expression on the bile duct epithelial cells . More seriously, Morgan et al.  recently reported the death of a patient treated with anti-ERBB2 directed T cells for metastatic adenocarcinoma of the colon: the patient developed rapid pulmonary infiltration and a cytokine storm, considered to be a result of first-pass localization of the administered cells to the lung, where low levels of ERBB2 on lung epithelial cells may have triggered cytokine release .
A further challenge to the implementation of therapy with gene-modified T cells is that of poor persistence after adoptive transfer. T cells modified with a CAR targeting the α-folate receptor were circulating in large numbers two days after transfer to patients with ovarian cancer, although they declined to barely detectable levels by 1 month, in tandem with development of an inhibitory factor in the serum, most likely an antibody . Other studies have documented anti-transgene immune responses against CAR-transduced cells in more detail, showing both humoral and cellular responses that neutralized function and limited persistence of cells transduced with CARs recognizing CAIX, CD19 or CD20 [17, 18]. Although these studies underline the need to minimize immunogenicity of both transgene and vector, one study in mice has shown that a therapeutic response can be achieved even when CARs are expressed only transiently in transferred cells through RNA transfer , suggesting that the persistence of redirected cells is not always critical.
The therapeutic potential of CAR-redirected T cells has been shown more clearly in a study that targeted T cells against the GD2 antigen, expressed on most neuroblastomas. T cells that had previously been enriched for specificity against EBV by ex vivo culture with EBV-transformed stimulator cells, and polyclonal T cells, were separately transduced with an anti-GD2 CAR. Eleven patients received equal numbers of transduced EBV-specific and polyclonal T cells; half of the eight patients with evaluable tumours showed regression or necrosis of the cancer . The CAR-transduced EBV-specific CD8+ T cells persisted longer than the polyclonal T cells, although the levels of both had become extremely low or undetectable by 6 weeks after transfer . Although this was initially thought to suggest an effect on the EBV-specific CD8+ T cells of continuing native receptor stimulation by latent EBV antigen on endogenous antigen-presenting cells, longer-term follow-up has shown persistence to correlate more with the proportion of CD4+ helper cells and CD45RO+CD62L+ central memory cells present in the transferred populations: EBV-specific and polyclonal T cells have been shown to persist for up to 96 or 192 weeks, respectively, after transfer, with the length of persistence correlating with an increased time to progression of disease .
Finally, several groups have used CARs to direct T cells against lymphoid cancers expressing the B cell markers CD19 or CD20 [17, 127, 151-154]. Kochenderfer et al.  transferred T cells transduced with an anti-CD19 CAR to eight patients with advanced, progressive B-cell malignancies, giving IL-2 to promote persistence. Six of the patients obtained remission, four showed reversible toxicity associated with IFN-γ and TNF-α release, and one developed long-term depletion of normal polyclonal CD19+ B cells. The findings are consistent with a therapeutic effect of the CAR-transduced cells, although some of the anti-tumour effects may have been a result of chemotherapy given as part of the trial protocol . A similar study showed superior in vivo persistence of cells transduced with a second-generation anti-CD19 CAR containing a co-stimulatory CD28 endodomain compared to cells expressing an otherwise identical CAR lacking this domain . The benefits of such co-stimulation are also evident in a recent report of treatment using T cells that expressed an anti-CD19 CAR incorporating a 4-1BB (CD137) signalling domain : not only did the transferred cells persist to at least 6 months after infusion, but also the patient developed marked regression of cancer, associated with the tumour lysis syndrome, at a delayed timepoint that suggests a clear separation from the effects of chemotherapy.
These promising clinical results suggest that adoptive immunotherapy with redirected T cells is finally coming of age. The potential of therapeutic T cells to traffic to sites of disease, to expand and to persist after a single treatment remain major advantages over the currently available immunotherapies that use monoclonal antibodies. As research to optimize the functional avidity of TCR and CAR-transduced cytotoxic T cells continues, the field has expanded: beyond the antigen receptor, encompassing modifications to crucial coreceptors and manipulations of metabolic pathways to enhance function and longevity of responses; beyond the cytotoxic T cell, to include redirection of memory T cells, stem cells and regulatory T cells; and beyond cancer, to the enhancement of antiviral responses and the treatment of autoimmune disease. The increasing sophistication of measures to ensure the safety of engineered T cells is accompanied by an increasing number of clinical trials: these will be essential to guide the effective translation of cellular immunotherapy from the laboratory to the bedside.
This review was supported by grants from Leukemia and Lymphoma Research, Experimental Cancer Medicine Centre, ATTACK EU Consortium, and the Medical Research Council. H.J.S. is a consultant for Cell Medica. The remaining authors declare that they have no competing financial interests.