While distinct input and output modules of logic gated ACT are defined by their biological thresholds, each intact circuit should eventually comply with clinical tolerability. This aspect is particularly important while weighing the actual expression of a given antigen by tumor vs. normal cells, as the outcome of antigen binding critically depends not only on its expression level, but also on the overall avidity of interaction, the fidelity of the signaling cascade and the response mechanism. With this in mind, attempts are made to identify combinations of surface antigens that are expressed by all cells in a given tumor and by most tumors of a given type and, desirably, by different tumor types, but not by any vital tissue. As the number of input signals grows, precision is expected to increase, but at the expense of the fraction of tumors complying with all inputs and of applicability, as additional modules would inevitably require increasingly improved gene transfer vectors.

Acute myeloid leukemia (AML) is a heterogeneous class of diseases characterized by particularly high genetic instability that gives rise to multiple immunophenotypes in a single patient, rendering AML notoriously refractory to conventional immunotherapies (23). In attempt to circumvent this heterogeneity, Perna et al. (24) performed meticulous proteomics and transcriptomics analysis of large surfaceome datasets from malignant and normal tissues, aiming at identifying antigen pairs which could serve as targets for combinatorial CAR therapy. While a number of antigens have been previously reported as potential AML CAR targets, none met the criteria defined in this study for an optimal single target. Following a stringent refining process, four antigen pairs: ADGRE2+CD33, CCR1+CLEC12A, CD70+CD33, and LILRB2+CLEC12A, yielded nearly optimal results, for all of which the dual targeting score significantly exceeded that of either antigen alone.

More recently, Dannenfelser et al. carried out a comprehensive in-silico screen of >2.5 million dual and 60 million triple antigen combinations across 33 tumor types and 34 normal tissues, assessing their ability to discriminate tumors from normal tissues via either AND or NOT gates (25). For prioritizing candidate antigen combinations the authors created a numerical ruler, integrating precision and recall (fraction of targetable tumor samples), with which they compared single antigens to double and triple gates. For every cancer type assessed they identified at least 25 high-score antigen pairs, which substantially outperform single ‘clinical’ antigens currently explored in CAR therapy of the same cancer. Furthermore, for several tumor types, triple antigen combinations show near ideal precision, yet, with an anticipated decline in recall. Although this study is based on RNA-seq databases rather than actual surface protein expression, it underscores the discriminatory power of combinatorial antigen recognition, paving the way for clinical evaluation of numerous CAR therapies targeting new antigen combinations in virtually all cancer types.

Another class of candidate antigens for combinatorial recognition are expressed by tumor-supporting cells at the TME, including cancer-associated fibroblasts, tumor-associated macrophages, myeloid-derived suppressor cells, regulatory T cells and additional immune cells (9–11), as well as tumor endothelial cells (TECs) (26–29). Antigens that are preferentially expressed by these cells (such as VEGFRs, PSMA, ALCAM, CD13, CLEC14A, RGS5, TEMs, fibronectin EIIIB splice variant, Endothelin B receptor, α_vβ₃ integrin and others) are particularly attractive, as, unlike tumor antigens, they are not subjected to genetic and epigenetic instability. Among these, TECs draw special attention, as they are highly accessible to the therapeutic cells and are less affected by immune-suppressive conditions dominating the TME. While none of these antigens is utterly specific, collectively they provide an additional layer of potential targets for gated therapy.

Loss of heterozygosity (LOH) characterizes the majority of human cancers. It is manifested in multiple losses of full chromosomes, entire chromosomal arms or sub-chromosomal regions and is often associated with the loss of a normal copy of a tumor suppressor gene (30–33). Many LOH events already occur prior to malignant transformation creating a loss signature that is shared by premalignant cells and all descendant tumor cells in a given patient (34). An inevitable outcome of LOH is the concomitant loss of all other genes residing on the deleted chromosomal material, and these naturally include heterozygous alleles of protein-coding genes. This early LOH-driven antigenic landscape is consequently irreversible and is not affected by tumor heterogeneity. LOH-based discrimination of tumor from normal cells can be achieved by NOT gates based on either CARs that are directed at an allelic variant encoding an extracellular epitope, which can be distinguished from the homologous one or TCRs, recognizing a linear peptide that is presented on normal but not tumor cells by one of the patient’s HLA-I products. A special class of LOH events involve the HLA gene locus on chromosome 6q21 and are apparently exploited by tumors as an escape mechanism from CD8 T cells (35–37). It now emerges that this mechanism also allows tumor cells to avoid recognition by adoptively transferred T cells, as reported, for example, in TIL therapy of a patient with metastatic colorectal cancer, targeting a peptide harboring the KRAS G12D mutation presented by HLA-C*08:02 (38). LOH is source for a universe of antigenic combinations that can be targeted by NOT modules and pioneering attempts to explore this new route to cancer immunotherapy have recently been published (39, 40).

Hypoxia is a prominent feature of the TME, resulting from rapidly dividing cancer cells and aberrant vascularization (41). Tumor cell adaptation to hypoxic conditions is governed by hypoxia-inducible factors (HIFs) via the hypoxia pathway (42). Under normoxic conditions, prolyl-hydroxylases (PHDs) hydroxylates conserved prolines of HIF-1α, leading to ubiquitination and proteasomal degradation. However, during hypoxic conditions, the PHDs are inactive, allowing HIF-1α accumulation, translocation to the nucleus, dimerization with HIF-1β and formation of a transcriptional complex. The HIF complex binds to the promoter region of hypoxia-responsive elements (HRE) and triggers transcription, which regulates several biological functions of tumor cells (43). These two hypoxia-induced mechanisms, acting independently at the transcription and post-translational levels, can be exploited for confining CAR activity to hypoxic tissues.

Pro-inflammatory cytokines secreted by a variety of cells at the TME induce cellular responses that are associated with tumor initiation, progression, tissue invasion, metastasis and evasion from an immune response. In parallel, potent immunosuppressive mediators that are overproduced by tumor-supportive immune cells act in many solid tumors to counteract the antitumor response. Consequently, the TME is often characterized by elevated levels of cytokines, including TNF-𝛼, IFN-γ, IL-6, IL-8, IL-17, IL-21, TGF-𝛽 and IL-10 (44–46), which can serve as TME biomarkers and guide AND gated therapies exploiting cytokine-responsive receptors or transcriptional control elements.

Continuous interactions between tumor and nontumor cells at the TME result in the upregulation of a battery of pericellular proteases, which play a critical role in tumor angiogenesis, invasion and metastasis (47–49). These include various matrix metalloproteinases, cathepsins, elastase, granzyme B, tissue and urokinase plasminogen activators and others, creating a protease-rich niche that is distinguishable from most other normal tissues. This unique characteristic of the TME paves the way for the design of antitumor antibodies or CARs whose recognition moieties are masked by a protease-cleavable peptide, which is liberated at the tumor site, creating a bona-fide logic AND gate.

Tumor-specific as well as tissue- or state-specific promoters can be harnessed to confine the expression of therapeutic genes of interest (GOIs) to designated target cells in-vivo and, as such, offer another class of cues for gated cancer therapy. However, as with tumor antigens, no single promoter appears to be solely restricted to a given type of tumor. Nonetheless, dividing the specificity task between two promoters, both simultaneously active only in the target cells, could confer the desired precision of action (50–52). In addition, synthetic promoters comprising distinct transcription factor (TF)-binding elements can assure that transcription of a therapeutic gene takes place only in cells expressing all corresponding TFs (53, 54).

NOT gates are designed to protect normal cells that express a selected tumor antigen targeted by an activating CAR (aCAR) or TCR), as well as a second antigen (the ‘protective’ antigen) that is NOT expressed by the tumor. An effective NOT device would counteract T cell activation triggered by the tumor antigen when simultaneously engaging the protective antigen on the same normal cell. Creating a safe NOT gate entails the incorporation of an inhibitory module specific to the second antigen, which can operate through either strong inhibitory signaling or potent competition with the activating module while guaranteeing continuous dominance over the latter. It is mandatory that the effect of the NOT module on the therapeutic T cell is transient and fully reversible, as repeated inhibitory signaling can anergize these cells and abolish their therapeutic efficacy.

The implementation of the inhibitory approach for producing a NOT gate was pioneered by Fedorov at al (77)., who recruited the signaling domains of the T cell inhibitory receptors CTLA-4 and PD-1 to create inhibitory CARs (iCARs) ( Figure 1 ). Since then a number of studies exploring NOT gates have examined iCARs incorporating inhibitory signaling elements that are derived from T or NK cell-associated inhibitory receptors, including PD-1 (40, 78), BTLA (79) and LIR1 (39).

Employing competitive inhibition, the SUPRA CAR technology offers a ‘Cell Selector’ application ( Figure 1 ) (59). Here, the presence of a protective antigen on normal cells recruits a designated zipFv harboring a zipper, which binds avidly to the activating zipFv zipper, blocking its binding to the zipCAR, thereby preventing activation. Another competition-based NOT module is offered by the Co-LOCKR system ( Figure 1 ), allowing the simultaneous recognition of antigen A AND antigen B but NOT antigen C, creating an AND-NOT gate (62). This scenario is addressed through a third soluble component specific to antigen C, which carries a ‘Decoy’ segment. Upon antigen binding on the cell to be protected, the Decoy binds tightly to the adjacent ‘Key’ and prevents binding of the latter to the ‘Cage’, so that release of the recruitment peptide and subsequent T cell activation cannot take place. Another antigen-driven AND-NOT circuit was recently reported ( Figure 1 ), utilizing a synNotch receptor, which governs the expression of the pro-apoptotic factor tBID (truncated BID, a member of the BH3-domain-only subgroup of BCL-2 family) upon antigen binding as a novel, gated suicide mechanism (66).