Chimeric Antigen Receptor-T Cells for Targeting Solid Tumors: Current Challenges and Existing Strategies

Following infusion into the systemic circulation, CAR-Ts are faced with the immediate obstacle of localizing to and infiltrating into the tumor parenchyma. Homing and tissue infiltrating is a multistep process governed by the expression and pairing of adhesion molecules present on both the T cells and the inflamed vasculature that act sequentially to mediate attachment, rolling, and extravasation of circulating lymphocytes towards a chemokine gradient produced by tumor cells. However, aberrant expression of adhesion molecules on the tumor endothelium as well as T cell chemokine receptor/tumor-associated chemokine incompatibility and hydrostatic pressure result in inefficient intratumoral T cell infiltration potentially causing treatment-related toxicities due to the accumulation of transferred cells in inflamed normal tissues, such as in the case of injury or autoimmune disease [20].

Several preclinical models demonstrated that the forced expression of a chemokine receptor complementary to tumor-associated chemokines enhanced the ability of CAR-Ts to traffic to and expand at the tumor site, consequently improving their antitumor efficacy [21, 22]. However, applicability of this approach is restricted by the fact that the chemokine landscape can be extremely heterogeneous both across disease entities and patients, underscoring the need to identify specific receptor candidates to enhance T cell infiltration into different cancer types [23]. Furthermore, chemokines are not restricted to the tumors, suggesting there could be diversion of the cells to other anatomical locations where the specific chemokine is present.

Although not always technically achievable, loco-regional delivery of CAR-Ts reduces trafficking restrictions without additional engineering while circumventing the transient pulmonary sequestration of intravenously administered T cells [24, 25]. In mouse models, intraperitoneal or intrapleural administration of CAR-Ts outperformed systemic infusion and, surprisingly, also impacted disseminated tumor sites attributed to a benefit of T cell activation shortly after delivery [26, 27]. Accordingly, several clinical trials are examining the safety of administration of loco-regional CAR-T therapies (discussed in Sect. 3), even though infiltration within solid tumor masses is not always improved by loco-regional delivery. Finally, nanoparticles expressing CARs, which bind to and re-program peripherally circulating T cells in vivo, were also recently developed to increase selectivity and distribution to distant organs [28].

Once they have successfully invaded the tumor parenchyma, CAR-Ts then have to contend with a highly hostile milieu for T cell antitumor effector function, replete with suppressive mediators (transforming growth factor [TGF]-β, interleukin [IL]-10, IL-4) and inhibitory molecules (programmed death-ligand 1 [PD-L1], cytotoxic T lymphocyte antigen 4 [CTLA-4], Fas-ligand [FASL]). Apart from TME remodeling, which should be induced by the combination with chemotherapy agents, a more specific combination strategy with programmed death 1 (PD-1)/PD-L1 or CTLA-4-blocking antibodies (the so-called checkpoint inhibitors commonly used in clinical studies with excellent outcomes [29]) and CAR-Ts can therefore potentially augment antitumor effects against solid tumors [17, 30, 31]. CAR-Ts can also be shielded to intrinsically resist immunosuppressive signaling by disrupting endogenous expression of inhibitory receptors through gene editing or transgenic expression of a dominant-negative form of those receptors or inhibitory antibodies [32‐36]. However, the abrogation of immunosuppressive signaling may be insufficient, prompting additional investigations into alternative approaches that can turn TME limitations into advantages for the transferred CAR-Ts. Co-expression of a chimeric receptor that converts an immunosuppressive signal into an immunostimulatory one could also extend CAR-T engineering beyond neutralization of inhibitory ligands to the active reversal of their effects. Exchanging the endodomain of inhibitory receptors such as IL-4 receptor (IL-4R) or PD-1 with signaling domains derived from stimulatory receptors (IL-7 receptor [IL-7R], CD28, or 4-1BB) improved in vivo antitumor efficacy of tumor-directed T cells [37‐39]. Importantly, CAR-T activation could be confined to the tumor site since triggering would require exposure to both the specific antigen and the tumor-derived factor. In addition to promoting function and survival of the modified T cells, the use of inhibitory-to-stimulatory switch receptors might present the advantage of depriving the TME of an immunosuppressive factor, potentially providing collateral benefits to endogenous exhausted tumor-infiltrating lymphocytes (TILs) [39, 40]. Although those additional engineering strategies proved effective in murine models, selective neutralization of a single immunosuppressive pathway might render a functional, albeit transient, antitumor state and fall short of preventing long-term relapse due to the upregulation of multiple inhibitory receptors by activated T cells, thus limiting the window of time that the CAR-Ts exert their function. On the other hand, as those receptors are important regulators of T cell homeostasis, the impact of such modifications on T cell effector function in humans remains to be determined, as well as any potential impact of the leverage of immune brake that could lead to uncontrolled lymphoproliferation or other immune-related adverse events.

Another methodology addresses the unfavorable TME by using CAR-Ts as production vehicles that secrete proinflammatory cytokines, such as IL-12 or IL-18, into the targeted tumor tissue, tuning the T cell response into a more acute one [41]. Beyond auto-stimulation of the transferred cells [42], release of effector cytokines by those so-called ‘TRUCKs’ (T cells Redirected for Universal Cytokine Killing) was shown to reshape the TME through multiple paracrine mechanisms including recruitment of additional tumor-reactive cells from the innate and adaptive immune systems [43‐45]. As tumor cell lysis by TRUCKs can generate new antigen-specific lymphocytes via epitope spreading, the concomitant local release of effector cytokines will support the effector function of these host immune cells and also recruit and activate innate immune cells [46, 47]. Despite all the expected benefits, the systematic delivery of proinflammatory cytokines may lead to significant toxicities [48], underscoring the critical need to restrict cytokine production to the lesion site by using a promoter that becomes active only upon CAR engagement. In addition, inducible expression systems are more likely to constrain cytokine levels within a therapeutic range as overactivation of T cells by supra-therapeutic cytokine levels will foster counterproductive exhaustion. However, in early-phase clinical trials, adoptive transfer of TILs genetically engineered to secrete IL-12 at the tumor site resulted in severe toxicities [49]. Therefore, the use of less stimulatory cytokines such as IL-18 might present a safer option as this cytokine was given intravenously at high biologically active doses to cancer patients with no occurrence of dose-limiting toxicities [50]. In addition, integration of suicide genes or safety switches is another option to mitigate toxicity potentially induced by such strategies (see Sect. 2.5).

Since emerging nanoscale-targeted drug carriers are able to remodel the TME without giving rise to the systemic toxicity, CAR-engineered T cells were also employed as active chaperones to successfully deliver adenosine receptor antagonist-loaded cross-linked multilamellar liposomal vesicles to TILs deep in the immunosuppressive TME, in order to prevent or rescue the emergence of hypofunctional CAR-Ts within the TME [51].

While the in vivo cell expansion and effectiveness of CD19 CAR-Ts seem to correlate in certain studies using CAR-T in hematological malignancies [52, 53], it is generally considered that the intrinsic qualities of infused lymphocytes are some of the determinants of success in CAR-T therapies. The ex vivo manipulation of T cells provides a unique opportunity to select for cellular subsets with enhanced potential for mounting durable antitumor responses [54]. Selection of CD8⁺ cytotoxic cellular subsets, ratios of CD4:CD8, or use of natural killer cells may increase broad effector activity [46, 55]. Although the ‘seed’ population optimally suited for the production of long-lived CAR-Ts is still a matter of debate, an emerging consensus postulates that less-differentiated phenotypes such as cells presenting naïve and central memory phenotypes have superior proliferative capacity and sustained survival and, as such, are more effective at regressing established tumors than late-differentiated effector memory and effector T cells [56]. Building on this concept, there is growing interest in developing protocols to conduct large-scale T cell amplification, while simultaneously preserving the functional features of early-memory T cells [57]. It was shown that reducing the duration of ex vivo culture to 3–5 days yielded less-differentiated cells with enhanced therapeutic potential compared with cells expanded using standard 9- to 12-day protocols [58]. An alternative strategy to limit cell differentiation during CAR-T manufacturing is the pharmaceutical blockade of the phosphoinositide 3-kinase (PI3 K)/AKT axis playing an integral role in T cell activation downstream of the TCR and co-stimulatory molecules [59, 60]. Another option would be to substitute IL-7 and IL-15 for IL-2 as the growth factor support during ex vivo generation of CAR-T products as this cytokine combination was shown to enrich for T memory stem cells [61]. In preclinical models, CAR-Ts expanded in IL-7 and IL-15 showed superior persistence and antitumor activity compared with counterparts grown in IL-2 [62].

Holding back the acquisition of full effector capacity ex vivo by the reduction of culture duration or modulation of T cell differentiation represents relatively easily translatable and widely applicable ways for the generation of early-memory CAR-Ts. The question is whether these cells have the therapeutic potential to be effective at lower infusion doses, potentially mitigating acute toxicity and commensurately trimming production costs [60].

The evolution of CAR design, to date, has focused predominantly on increasing signaling outputs through combinatorial modules of co-stimulatory domains fused in series to ITAM-bearing CD3ζ activation domain [63]. However, there is now a growing appreciation that functional tuning of CAR signaling has an upper limit. Above this limit, gains in the magnitude of effector outputs are negated by augmentation of T cell differentiation, exhaustion, and activation-induced cell death (AICD) [20, 21]. Accordingly, the next challenge for future CAR generations will be to calibrate CAR activation in order to achieve an optimal balance between effector and memory programs in T cells. Optimized configurations of CARs are being investigated to better recapitulate the dynamic process of natural T cell activation and co-stimulation, sharply differing from the 1:1 stoichiometry constraint within CAR designs currently under clinical investigation. For example, the expression of a CD28-based CAR along with 4-1BB ligand resulted in higher therapeutic efficacy, reconciling tumoricidal function afforded by CD28 co-stimulation with increased T cell persistence afforded by 4-1BB engagement [64]. Recently, a CD28-based CAR containing a single functional ITAM was shown to favor in vivo persistence of highly functional CAR-Ts, balancing the replicative capacity of long-lived memory cells with the acquisition of strong antitumor effector functions [65]. However, the optimal construct will likely depend on several factors, including affinity (avidity) for target, tumor access, and the type of TME.

Therefore, while several options to improve both persistence and expansion capacities of CAR-Ts are currently being investigated, no universal solution has yet been identified. To this end, the empirical testing of CARs remains the only option to evaluate the different potential schema of CAR/T cell phenotype/additional functionality such as TRUCKs.

Although not specific to solid tumors, due to the paucity of truly tumor-restricted antigens in solid tumor tissues, CAR-Ts will need to become capable of recognizing patterns of gene expression that are different between normal and malignant cells, rather than relying on single—though highly specific—antigenic markers. One approach that was investigated is to engineer CAR-Ts with dual specificity, whereby two receptors targeting distinct antigens act as ‘AND/NOT’ Boolean logic gates [66, 67] in order to prevent toxicity while maintaining efficacy, rather than irreversibly deleting CAR-Ts that are toxic against both tumor and host. The ‘AND’ gates require the successful recognition of a set of pairwise upregulated tumor antigens by two different CARs to initiate full immune cell functions [68, 69], whereas ‘NOT’ gates employ receptors that prevent T cell activation when engaging antigens found on healthy tissues [70]. While Boolean logical sensing may enhance the specificity of CAR-Ts towards tumors, this approach is still limited by the fixed antigen specificity of conventional CAR design, and by the fact that the therapeutic window will require an optimal expression pattern of multiple targets while a single target antigen loss could severely disable the system.

An alternative to this classical antibody-based CAR limitation would be to harness the multiple ligand-binding ability of physiological immune receptors such as NKG2D (natural killer group 2 member D). NKG2D recognizes several stress-induced ligands expressed within the TME of cancers from diverse origins, not only on the tumor cells themselves but also on tumor neovasculature and tumor-associated immune cells. Thus, a CAR bearing NKG2D as the targeting moiety holds the potential to eliminate a broad array of cancers, simultaneously altering the tumor and its supportive framework [71‐73]. A second ligand-based CAR approach targets the ErbB receptor family, for which at least one member is expressed in 88% of solid tumors [74‐77].

Another possibility is to target the CAR-Ts towards antigens expressed on tumor stroma and vasculature, which are expressed by multiple tumor types and would increase the homing into the TME [78, 79].

A first option to mitigate the potential on-target, off-tissue toxicity of CAR-Ts is the use of CAR-Ts with reduced persistence capacities such as transiently expressed CARs using non-viral approaches including messenger RNA (mRNA) electroporation [80], sleeping beauty transposition [81], and/or a multiple-dose schedule of short persisting CAR-Ts to control engraftment [80, 82]. Furthermore, the hypofunctionality of CAR-Ts within the TME may also be overcome by a multiple-dosing approach [16, 17, 83].

Equipping CAR-Ts with properties aimed at enhancing their potency or their infiltration into tissues should ideally be coupled with stringent safety attributes that allow for temporal regulation of activity or persistence of infused cells in the patients. Co-expression of suicide genes encoding surface molecules or enzymes conferring susceptibility to antibody- or drug-mediated cell death allows for selective and irreversible depletion of the transduced T cells after infusion into the patients [84‐86].

To avoid the irrevocable elimination of potentially therapeutic cells, several platforms have been developed to repeatedly turn on and off CAR-T activity at will after re-infusion into the patients (called ‘safety switch’ or ‘advanced cell programming technology’) to prevent and/or limit the likelihood of toxicity. These ‘switchable’ CAR-Ts are not directed to a cell surface target antigen and are per se inert but become operative strictly in the presence of a bispecific adaptor molecule that mediates formation of the immunological synapse between the target cancer cell and the lymphocyte [87‐92]. After rapid elimination of the adaptor molecule from the peripheral blood, CAR-Ts automatically turn off, thus providing a self-limiting safety switch. Moreover, the modularity of the switchable CAR-T approach provides options for altering specificity post-adoptive transfer by delivery of adaptor molecules targeting different antigens together with one single cellular product, which may be an effective strategy for addressing antigen loss relapse and heterogeneity of tumor populations. Furthermore, the ability to titrate CAR-T activity in vivo through adaptor molecule dosing paradigms offers the opportunity to achieve a gradual clearance of cancer cells, minimizing acute toxicity in high tumor burden patients. Finally, low-dose treatment with an adaptor molecule maintained a larger central memory compartment within CAR-Ts than did high-dose regimens, with the potential to boost in vivo cell endurance, as discussed earlier. However, the potential drawback of this approach lies in the need for multiple costly reagents and the challenge of ensuring that the engager and CAR-T meet in the correct location at a concentration of each entity sufficient to drive a therapeutic response. Within the parenchyma of a solid tumor, this would likely be a major dosing challenge.

The future of CAR-T cellular therapies for solid tumors resides in the alliance of wisely selected complementary approaches that will generate a cellular product with enhanced tissue penetration and homing, well-balanced effector and memory outputs, enhanced specificity/safety, and the ability to resist TME immunosuppression while concurrently reviving the endogenous host immune system (see Table 1). Using healthy donor cells instead of each patient’s cells, i.e., development of allogeneic approaches with a decreased risk of graft-versus-host disease (GvHD) and management of host-versus-graft disease (HvGD), may provide answers to some of these issues. The use of a single donor should provide a greater degree of product consistency, while the likely youthful healthy donor would potentially provide a T cell product that has not been skewed by the long-term exposure to tumor cells as would be the case for an autologous product. From a practical perspective, allogeneic CAR-T therapy may also provide economic benefits through reduced per patient costs and the fact that patients would not need to wait for the length of the manufacturing period before receiving the product. Earlier treatment of patients with acute disease could be of critical importance with respect to therapeutic readouts. One approach being pursued to generate an allogeneic CAR-T product is the complete elimination of TCR and human leukocyte antigen (HLA) molecules usually performed by gene-editing techniques [93, 94].

Yet, a major task is the transition from proof-of-concept studies employing human tumor cell line xenografts into immunocompromised mice to the development of clinically implementable technologies. Indeed, the clinical predictive power of such experimental systems is challenged by the fact that they imperfectly reflect the structural complexity and heterogeneity of established solid human tumors, poorly inform about potential cross-reactivity against healthy human tissues, and provide limited insights about how CAR-Ts interface with the host immune components. Patient-derived xenografts may represent more clinically relevant models but suffer from a variable engraftment rate and poor availability [95]. In addition, stromal cells from the original human tumor cannot proliferate continuously and are replaced by cells derived from the recipient mouse [96], thereby preventing investigations into the impact of therapy on TME. Ultimate validation of which combinatorial approaches or defined T cell subsets composition will achieve sustainable effective responses in the human context will only come from future clinical trials carried out to evaluate the resulting conclusion.