Localized delivery of anti-tumor molecules using logic-gating

One advantage of applying logic gating to biological therapies is to trigger localized secretion of ant-tumor molecules, within the tumor microenvironment. Logic gated molecule release results in the highest concentration being delivered directly at the tumor site, rather than the molecule being sequestered in other tissues. Second, the specific delivery of molecules may allow for the application of a wider range of antitumor molecules, which may be unsuitable for systemic delivery. Additionally, well-established non-logic gated systems such as nuclear factor of activated T cells (NFAT)-dependent (T cells redirected for antigen-unrestricted cytokine-initiated killing (TRUCKs;fourth generation CAR T cells)17 (reviewed18 19) and chemically controlled systems20 21(reviewed22) can also be employed for site specific delivery of modulating factors.

There are several non-specific biological agents which have demonstrated efficacy in promoting anti-tumor responses. Interleukin-12 has potent anti-tumor efficacy and has shown to synergise with CAR T cell therapy in glioblastoma,23 yet is systemically toxic.24 Similarly, HER-2-specific CAR T cells secreting the dendritic cell growth factor Fms-like tyrosine kinase 3 ligand (Flt3L) induced epitope spreading in host T cells against non-HER-2 tumors in a transgenic OT-I murine system, particularly with the addition of immune adjuvants.25 CD19-CAR T cells engineered to express CD40L also facilitated the recruitment of endogenous immune cell populations in a syngeneic immunocompetent mouse model.26 These agents may be promising candidates for gated expression in the tumor microenvironment to enhance immune responses.

However, there must also be stringent pre-clinical evaluations of such therapies. The final doses delivered at the tumor site may remain unknown or be under limited control and therefore the half-life of these molecules must be determined to ascertain potential toxicity risks. Despite these considerations, there are numerous possibilities in the types of treatments which could be delivered via logic gating.

Limiting antigen escape and combating tumor heterogeneity

Antigen escape is the process of specific downregulation of a tumor expressed protein resulting from selective pressure exerted by targeted therapy or the outgrowth of pre-existing antigen negative clones. While antigen escape has been observed in hematological malignancies,2 27 ,28 this remains a greater challenge for solid tumors, due to the increased tumor antigen heterogeneity. Tumor heterogeneity has been observed both between patients (interpatient heterogeneity) and within patient tumors (intratumoral heterogeneity).29 Patient tumor sampling and biomarker testing, particularly before enrolment in relevant clinical trials, have demonstrated activating mutations present in proteins such as KRAS,30 BRAF31 and EGFR32 within tumors and between patients. Tumor heterogeneity is a contributing factor to resistance against mono-targeted therapies,33 and with advancing technology this type of tumor analysis has been further enhanced by single-cell technologies.34 Such technology enables monitoring of a patient’s tumor over the course of targeted therapy to allow earlier identification of target antigen downregulation and antigen escape and potential re-evaluation of the treatment plan. The phase I EGFRvIII CAR T cell trial (NCT02209376) allowed patients to participate in the trial with EGFRvIII positivity ranging from 6%–96% (calculated by RNA sequencing of the number of EGFRvIII reads/ total EGFR and EGFRvIII reads).3 With a patient inclusion criterion of tumors with EGFRvIII expression as low as 6%, it is tempting to speculate that patients with low target expression would not have the same potential to benefit from treatment as those with a higher EGFRvIII percentage and target expression levels are of important consideration in future clinical trial design.

One of the most difficult challenges associated with CAR T cell therapy is the selection of an appropriate tumor antigen to target. Target antigens can be broadly classified as TSA, tumor associated (TAA) or Embryonic Cancer Germline Antigens.35 Therapeutic targeting of TSAs provides a greater level of safety as the specific expression of these antigens on malignant tissue reduces the likelihood of on-target/off-tumor targeting. An example of a TSA is EGFRvIII, a mutation in the Epidermal Growth Factor Receptor expressed in a subset of Glioblastoma tumors.36 Comparatively, TAAs are expressed at a low level on healthy tissues but have a higher level of expression on malignant tissues. Targeting these antigens can result in on-target/off-tumor toxicity.37 CAR T cell therapy has been greatly successful in the treatment of some types of hematological malignancies by targeting CD19,38 a TAA expressed on both healthy and malignant B cells. Germline antigens include targets such as IL-13Rα239 and are frequently expressed on malignant cells but also demonstrate minimal expression on healthy adult tissue. Therapeutic targeting of an antigen which is also expressed on healthy tissue risks damage to other non-malignant cells in the body.

Published results of glioblastom-specific CAR T cell clinical trials have reported only transient antitumor responses in trials targeting several different antigens.3–6 Targeting of the high-affinity cytokine receptor IL-13 receptor α2 (IL-13Rα2), expressed on over 80% of glioblastoma tumors, caused reductions in IL-13Rα2-expressing tumor cells in two of three patients with few side effects.40 Similarly, CAR T cells targeting EGFRvIII in glioblastoma patients also resulted in reduced EGFRvIII-expressing tumor cells in five of seven patients. However, in both studies, an antigen-negative population remained, and this antigen escape allowed the majority of the tumor to persist, ultimately resulting in poor overall survival.3 40

While targeting TSAs in the context of solid tumors provides the greatest level of safety to avoid off tumor toxicity, the heterogeneous and potential low expression of TSAs has contributed to the lower success rate for CAR T cell therapy. However, using these antigens as induction signals of a logic gated system, rather than the target of therapy may provide greater anti-tumor function and more complete tumor cell elimination than what is induced by conventional CAR T treatment. Importantly, logic gating is still a form of targeted therapy and therefore, these systems are not immune from inducing target antigen downregulation or loss via antigen escape.

Modulation of immune responses

The application of various logic gating systems has the potential to modulate immune responses. Depending on the type of system designed and the form of logic gating which is most suitable, immune responses such as CRS and off-tumor targeting of healthy tissue can be reduced, by the direct and local influence on tumor microenvironment.

CRS is another important consideration for CAR T cell therapy, particularly for hematological malignancies. The large quantity of proinflammatory cytokines such as IFN-γ systemically produced by highly activated CAR T cells activates bystander myeloid cells to release additional proinflammatory cytokines; including the potent IL-6.41 Therefore logic-gating the local secretion of anti-inflammatory mediators may assist in controlling systemic CRS.

Cases of CRS vary in severity, but the most severe cases can lead to multi-organ failure and death.41 The use of a monoclonal antibody (tocilizumab, α-IL-6) has shown efficacy in treating CRS without compromising CAR T cell efficacy.41 The availability of such anti-inflammatory medications is currently quite topical as tocilizumab is efficacious in treating patients infected with COVID-19.42 43 This has led to widespread redeployment of the medication for treating COVID-19 infections and consequently manufacturing delays and shortages.44 While this is an issue currently, and over time will resolve, the sudden reallocation of such medication further strengthens the case for approaches such as logic gating, whereby immune responses can be modulated without the external delivery of the drug itself, rather it can be produced locally by the infused logic gated T cell product.

Some degree of inflammatory cytokine release can enhance antitumor responses by activating bystander cells, however, a fine balance must be achieved to avoid excessive cytokine release and damage to tissues. Using logic gating systems, CRS could be controlled with the gated release of anti-inflammatory cytokines such as IL-10, monoclonal antibodies tocilizumab or mediators within the tumor microenvironment. This would mean that the tumor cells could still be eliminated, but the adverse patient responses of CRS may be avoided.

Logic gating approaches using inhibitory signaling pathways can further aid discrimination between healthy and malignant tissue. Genetic circuits built to only be induced in the absence of the ligation of a healthy antigen targeting CAR may prevent the off-tumor toxicity which can be observed when the primary target is minimally expressed on healthy cells. These types of circuits could prevent off-tumor toxicity before they occur. The application of inhibitory logic gating circuits may also work to increase the possible pool of target antigens by providing a checkpoint of ‘no healthy antigen’ on the cells before the destruction of the tumor cell, increasing safety.

Novel approaches are still needed to provide a greater level of regulation and control over engineered and infused T cells. Logic gating provides an additional level of control and safety and depending on the gating mechanism, even multi-antigen, or combination approaches. There are three main logic gating approaches described in this review and we will provide examples from the literature where they have been applied to engineered T cells and discuss some reflections on future approaches.

AND-gating

One of the early forms of exerting control over engineered T cells was the development of ON switches. In this system, the administration of small molecules enables the completion of a circuit to allow T signalling.45 One such system was developed by Wu et al46 which they termed ON CARs.

In this system, the antigen-binding domain and the signaling domain (containing the CD3ζ domain and ITAMs) are two separate molecules that only associate in the presence of a small molecule. This small molecule is a modified version of rapamycin with reduced immunosuppressive activity,47 which the authors refer to as rapalog.46 In the presence of the target antigen and rapalog, the functional anti-CD19 CAR unit is formed; an example of AND-gating. An advantage of this system is that the level of functional CARs can be modulated by titration of rapalog, with an increased dose leading to a greater reduction in tumor burden in mouse models of leukemia treated with ON CAR T cells.46

One year after the publication of pharmacological control of ON switched CARs, the technology was enhanced by the ‘IF-THEN’ or AND logic gating system. The concept of AND-gating requires two distinct antigens to be detected to trigger the outcome or downstream event; one to induce the logic gated system, the other to kill the target cell. The benefit of genetic approaches such as AND-gating in biological systems results in the presence of potentially potent anti-tumor molecules being modulated within the tumor environment and thus limiting systemic expression which may be intolerable to patients.24

In 2016, the design of the ‘AND-gated’ Synthetic Notch receptor (SynNotch) was developed by Morsut et al.14 ,15 In this system, the ligand binding domain is linked to the Notch core (a natural ligand in the body with a role in development48 within the transmembrane domain and a transcription factor. While there are multiple transcription factors which can be used in this system, the authors investigated the specific function of the Gal4-VP64 transcription factor in human cells. This system is an enhancement on the 2015 publication of ON switches by providing control over CAR expression, without the need for systemic delivery of recombinant antigen binding domains or small molecules.

On ligation of the scFv with its target antigen, the transcription factor Gal4-VP64 is cleaved from the SynNotch receptor by the natural Notch cleavage enzyme γ-secretase and the transcription factor is translocated into the nucleus of the cell where it becomes bound to the Upstream Activating Sequence. After binding, Gal4-VP64 drives the transcription of the desired cellular program, designed by the user (figure 2). In Roybal’s seminal 2016 publication, programs of cytokine production, monoclonal antibody production, transcription factors and even the expression of a second CAR were all shown to have the potential to be driven via SynNotch circuitry,49 providing evidence of a powerful circuit to regulate multiple anti-tumor mediators. Though one important characteristic of the system to consider is the inherent kinetic delay in the time from receptor ligation to transcription of the genetically coded signal (be that CAR expression, cytokine release, etc). Roybal et al noted an interval of approximately 6 hours from SynNotch activation to CAR expression (as determined by GFP expression of the tagged CAR).15 So while induction of the second protein may be slow, the extent of the kinetic delay may change with alternative induction antigens and future receptor designs.

The SynNotch system has been applied to multiple cancer models, including leukaemia,49–51 lung adenoarcinoma,52 mesothelioma,51 ovarian cancer51 and glioblastoma,53 in which tumor heterogeneity can be a contributing factor to unsuccessful targeted therapy. The glioblastoma tumor-specific protein EGFRvIII-SynNotch mediated induction of a Tandem CAR—a single CAR structure with dual targeting scFvs, specific for EphA2 and IL-13Rα2,53 was recently shown to induce more complete clearance of the glioblastoma orthotopic xenograft model GBM6 and result in longer survival compared with conventional single specific EGFRvIII, EphA2 or IL-13Rα2 CARs alone.53 Additionally, myelin oligodendrocyte glycoprotein was also used to mediate the induction of the EphA2 and IL-13Rα2 tandem CAR; thus providing evidence that normal antigens at the tumor site can also be used as induction signals in the SynNotch system.

The SynNotch system has also been applied to mesothelioma and ovarian cancer using the same ‘pan’ tumor antigen; ALPPL2, as the activation signal.51 The use of a more broadly applicable ‘pan’ cancer antigen was also investigated by Cho et al, who designed both a CAR and a SynNotch receptor specific for the receptor tyrosine kinase Axl50—expressed in breast, lung, colon and pancreatic cancers. The authors demonstrated the feasibility of this Axl system, achieving specific Axl-mediated CAR activation and induction of IL-10 on SynNotch activation.50 These studies have identified and characterized additional induction antigens which are amenable to the SynNotch system, both of which have the potential for a wide application to multiple cancer indications. The system can then be mixed and matched for the most efficacious genetic program to be induced (such as cytokines or additional CARs), dependent on the malignancy type and the tumor microenvironment to which it is applied.

Logic gating has also been shown to have a positive effect on T cell phenotype. There is evidence to suggest that the fine balance of signals, received by the T cell induces changes the differentiated state and indeed, tonic signaling induced by a CAR has been shown to influence T cell fate.54 55 Some early evidence has been published suggesting that logic-gating T cells can influence T cell fitness. In both the Hyrenius-Wittsten et al51 and Choe et al53 publications, the authors demonstrate logic gated T cells display a more ‘naïve’ like phenotype (CD62L⁺ CD45RA⁺) and reduced expression of checkpoint molecules. Hyrenius-Wittsten et al, propose that this less differentiated state is due to the removal of tonic signaling by changing the receptor design (compared with a conventional CAR).51 53 Hence, it is possible that logic gated technology may be a powerful tool to improve therapy efficacy by preserving the T cell population in a less differentiated state, while allowing modulation of the tumor microenvironment.

To date, Roybal et al have been the dominant investigators in the SynNotch research field, thus, the research field is small but growing. Other research groups have also applied SynNotch logic gating in their studies. Stanley Riddell’s group have developed an ‘AND’ gated ROR-1 specific CAR T cell which improved the safety of targeting this TAA by requiring the coexpression of another protein; EpCAM, or, for more tumor specificity B7-H3, preventing direction of cytotoxicity towards ROR-1 expressing stromal cells.56 Interestingly, protection of ROR-1 stromal cells was only found to occur when there was a spatial segregation between malignant and healthy cells. This paper nicely demonstrates a limitation of the SynNotch system, in that if healthy cells expressing a low level of the target antigen (TAA such as HER-2) are not spatially separated from the tumor cells, the healthy cells are likely to be destroyed by the T cell.56

The benefit of systems such as AND-gating broadens the applicability of many anti-tumor or immune activating molecules which may have been prevented from clinical use because of their systemic adverse effects. Stringent safety testing will be required to fully characterize the induction and half-life of the induced protein production. AND-gating is yet to be tested in clinical trials and so the ‘real world’ success is to be determined.

OR-gating

The concept of OR-gating is an example of a multi-antigen approach requiring the recognition of either one or more of the targeted CAR T cell antigens (figure 2). This strategy expands the CAR T cell antigen repertoire and therefore increases the likelihood of tumor eradication for heterogeneous tumors. There are a variety of different approaches to the OR-gated CAR T cell concept, such as pooling a combination of single antigen targeting CAR T cells or engineering a single CAR receptor to target multiple antigens termed tan- or biCARs. CAR T cells can also be engineered to express CARs targeting two different antigens (dual CAR T), as well as three or more antigens such as triCARs or quad-CARs.

Multiantigen targeting CAR T cell capabilities have demonstrated enhanced anti-tumor elimination compared with single targeting strategies.57 The simplest of these strategies is the use of pooled CAR T cells, which involves combining two single-specific CAR T cell populations.58 This approach has been used in both hematological cancer models, by targeting CD19 and CD123 for the treatment of B-cell acute lymphoblastic leukemia (B-ALL)58 and in solid tumors such as glioblastoma by targeting both HER-2 and IL-13Rα2.59 In both studies, pooling two populations of single-specific CAR T cells significantly enhanced tumor elimination both in vitro and in in vivo mouse models, particularly when the cancer models heterogeneously expressed different antigens.

Dual targeting antigens can be particularly advantageous to mitigate antigen escape or antigen loss. Clinically, CD19 antigen reduction or loss following treatment with CD19-targeted immunotherapies is often observed in relapsed and/or refractory B-ALL.60 This was shown to be mitigated by sequentially targeting a second antigen with maintained expression, using CD22 targeted CAR T cells, which resulted in 73% complete remission,61 though CD22-negative relapses were later observed.62 In addition, simultaneous antigen targeting by CD19/CD22 dual-specific CAR T cells, resulted in 88% complete remission.62 Several clinical trials using dual CAR T cells are currently ongoing.61 63 Furthermore, tandem CARs or tanCARs, in which a bispecific CAR receptor targets two different antigens (figure 3), has shown to further enhance tumor elimination,64–66 with a CD19/CD20 tanCAR clinical trial demonstrating promising results.67 In the treatment of solid tumors using a U373 glioblastoma model, Hegde et al demonstrated that tanCAR T cells controlled tumor growth significantly longer than pooled and dual CAR T cells.59 64

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

CAR modality effects anti-tumor function. Chimeric Antigen Receptor (CAR) T cells expressing a CAR specific for a single antigen demonstrate the lowest anti-tumor efficacy. Pooled CAR T cells are more efficacious than single specific CAR T cells and are generated by pooling two populations of single specific CAR T cells. Dual CAR T cells express multiple CARs on the one T cell. The most effective CAR modality is TanCAR T cells, with each T cell expressing one CAR engineered with multiple antigen recognition domains (scFvs). scFv, short-chain variable fragment.

The benefit of the OR-gated designs of multiantigen targeted CAR T cells increases recognition capacity by CAR T cells and enables the genetic targeting of heterogeneous tumors, minimizing antigen escape. However, while these approaches offer hope to eliminate heterogeneous tumors, OR-gated designs have been associated with significant on-target/off-tumor adverse events. In studies of B-cell malignancies, the administration of dual CAR T cells resulted in significant CRS in multiple clinical trials, including CD19/CD22 dual CAR T cells,68 and CD19/CD20 tanCAR T cells.67 This demonstrates the importance of antigen target selection, and the use of additional safety measures, some of which have been engineered into the gated CAR T cell design, such as those using AND-gated or NOT-gated approaches.

NOT-gating

Many proteins which are highly expressed on tumor cells, may also be expressed at low levels on healthy tissue. Whist the inclusion of suicide switches in genetically engineered cells provides a level of safety, this activation is irreversible and binary; either on or off. Once the suicide gene has been activated, the transduced cells are destroyed, no longer exerting anti-tumor functions, or providing benefit to the patient.

Unlike AND-circuits or OR-circuits which result in activation, the inhibitory NOT- logic gate prevents T cell activation on target recognition. NOT-gating can be applied to tumor antigens which may not be exquisitely tumor specific and display a low level of expression on healthy cells.

Inhibitory CARs (iCARs) are designed with inhibitory signaling domains such as those found in PD-1 and CTLA-469 (well characterized T-cell inhibitory receptors). iCARs are used in combination with CARs specific for TAAs to reduce the likelihood of healthy tissue damage. Biological function (cytotoxic T cell responses) will only occur if the TAA-CAR AND-NOT- the healthy antigen CAR are ligated. In this way, the biological gating system is a combination of AND-NOT systems.

The function of iCARs has been evaluated in vitro and in vivo. Fedorov et al demonstrated in vitro that iCARs had greatly reduced cytokine production and cytotoxic capabilities when cocultured with induced pluripotent stem cell-derived differentiated into fibroblasts which displayed both healthy (PSMA) and tumor (CD19) antigens compared with levels measured with coculture with target antigen (CD19) expressing cells.69 To determine the functional capacity of the iCARs in vivo, the authors generated a NALM-6 model of leukemia, expressing both CD19 and the healthy antigen PSMA. Mice bearing tumors lacking PSMA expression showed a reduced tumor burden by bioluminescence imaging compared with mice bearing CD19+/PSMA +tumors when treated with the iCARs.69 Flow cytometric analysis of the bone marrow confirmed similar proportions of dual positive tumor cells remaining in iCAR treated mice compared with the non-T cell treated control. These results demonstrated the applicability of this system in vitro and in vivo. However, one critical consideration of this system is that both the iCAR and the target were required to be highly expressed for optimal suppression of CAR function,69 which is unlikely to be a possible combination for all tumor antigens.

Despite these promising findings, it is important to consider the potential implications of genetic circuits on T cell health and phenotype. The premise of the iCAR is that the inhibitory signal generated by ligation of a ‘healthy/normal protein’ by the iCAR will override any effector ‘kill’ signal induced by ligation of a ‘tumor targeting’ CAR. Therefore, the resulting T cell function is dependent on the balance of activation and inhibitory signals. Mackall et al have shown preliminary evidence to suggest that T cells expressing both the CD93 NOT-gated CAR T cells and CD19 iCAR T cells exhibited decreases in PD-1 and TIM-3 expression compared with mock transduced and CD93-28z CAR T cells at baseline, however, further phenotypic analysis on T cell fitness post iCAR activation was not investigated.70

Opposing most of the biological logic gating systems which promote T cell toxicity, there is another class of logic gating inhibiting T cell function. This family of receptors is referred to as OFF-switch. OFF-switches are generally pharmacologically mediated.45 71 72 One example of the OFF-switch system is the Chemically Regulated-SH2-delivered Inhibitory Tail (CRASH-IT). These small molecule mediated switches can be applied both to engineered CAR T cells but also within the T cell receptor (TCR) complex. The switch was composed of an SH-2 binding domain fused to the intracellular signaling domain of PD-1 to inhibit signaling. Degradation of the inhibitory complex was achieved via the creation of a fusion protein containing a protease and a degron (a protein component targeting it for degradation by the proteosome). The benefit of the CRASH-IT switch is that it can be repeatably activated without resulting in engineered cell destruction.73

The authors also showed titratable function of the switch using a small molecule; asunaprevir (an anti-viral drug approved for use in combination with Daclatasvir in the treatment of Hepatitis C.74 In this study, in the absence of asunaprevir, the link between the degron and the protein of interest was degraded. However, the presence of the small molecule held the entire complex together, targeting it for degradation. This provides the same benefit of a suicide switch in preventing activation of the system, but in a non-binary manner and allowing reactivation of the system. While an excellent proof of principle, whether or not this finding can be recapitulated in vivo is yet to be reported.

Inhibitory circuits, or NOT-gates, provide an opportunity for the field to begin to investigate the anti-tumor function of therapies targeting highly expressed tumor antigens, which may have some level of expression on healthy tissue. The inbuilt safety step of either triggering an inhibitory signaling event or an inhibitory signaling event and degradation of the CAR could provide some level of reassurance that any targeting of healthy tissue can be halted before tissue destruction occurs.