Programmable synthetic receptors: the next-generation of cell and gene therapies

Cell-surface receptors and intracellular receptors

According to the location to bind ligands, synthetic receptors can be divided into cell-surface receptors (also known as transmembrane receptors) and intracellular receptors.³⁶

Cell-surface receptors generally comprise three types of transmembrane receptors (i.e., enzyme-linked receptors, G-protein-coupled receptors and ion channel-liked receptors), which span the plasma membrane and convert extracellular signals into intracellular signals.^37,38 For this category, each synthetic receptor contains an extracellular ligand-binding domain, at least one transmembrane domain and an intracellular effector domain. Besides the best-known CAR,^24,39 synNotch,^31,40,41 SNIPR (synthetic intramembrane proteolysis receptor),⁴² RASSL (receptor activated solely by a synthetic ligand),^43,44 Tango,^45,46 MESA (modular extracellular sensor architecture),^47,48 ChaCha,⁴⁹ TCS (two-component system),⁵⁰ chimeric cytokine receptor^51,52 and GEMS (generalized extracellular molecule sensor)⁵³ also fall into the category of cell-surface receptors (Table 1). This type of synthetic receptors can only sense external signals, making them suitable to detect cellular (e.g. CAR, synNotch and SNIPR) and systemic disease biomarkers (e.g. MESA and GMES).

Intracellular receptors can either locate in the cytoplasm or nucleus, or be anchored to the intracellular membrane of the cell. Cal-Light (calcium- and light-gated switch),⁵⁴ CHOMP (circuits of hacked orthogonal modular protease),⁵⁵ intrabody sensor,⁵⁶ RASER (rewiring of aberrant signaling to effector release),²⁷ LOCKR (latching orthogonal cage-key proteins),^57,58 COMET (composable mammalian elements of transcription)⁵⁹ and POST (phosphoregulated orthogonal signal transduction)⁶⁰ belong to the category of intracellular receptors (Table 1). Notably, the synthetic receptors discussed here could also be referred to as synthetic protein switches.⁶¹ Since intracellular receptors can only be activated by intracellular input, many of them (e.g., LOCKR and COMET) are designed as a switch induced by chemical molecules that can cross the plasma membrane.

Natural signaling-based receptors and orthogonal signaling-based receptors

Activated receptors trigger signal transduction via multiple downstream pathways. According to the downstream pathway actuated, either natural or engineered, we divide synthetic receptors into natural signaling-based receptors and orthogonal signaling-based receptors.

Natural signaling-based receptors rewire endogenous signaling pathways to either original or customized outputs. This category of synthetic receptors includes RASSL,^43,44 CAR,^24,39 chimeric cytokine receptor,^51,52 GEMS⁵³ and GEAR⁶² (Table 1). They inevitably activate native pathways to actuate outputs. Among them, RASSL, CAR and chimeric cytokine receptor redirect endogenous pathways to user-defined ligands to execute desirable functions, and are no longer responsive to their original ligands.^18–20 While GEMS and GEAR hijack endogenous pathways to generate additional customized outputs, they can simultaneously lead to activation of endogenous transcription networks.^18–20 On the other hand, the capacity to modulate transcriptional networks endows natural signaling-based receptors with unique advantages over orthogonal signaling-based receptors, allowing for the execution of natural and highly complex programs. For example, CARs are invented to mimic the function of T-cell receptor (TCR) to achieve antitumor activity⁶³ via modulating endogenous signaling pathways, but at the same time, additional transgenes can be induced by CAR activation (e.g., nuclear factor of activated T cells (NFAT)-driven cytokine expression), which could not be achieved by orthogonal systems.

By contrast, orthogonal signaling-based receptors are completely independent of endogenous signaling pathways. This attributes to the employment of orthogonal synthetic transcription factors (syn-TFs) or orthogonal signal transduction systems, as they cannot recognize endogenous regulatory elements and activate endogenous signaling cascades, respectively.^18–20 Nevertheless, some syn-TFs like dCas9-based TFs and TALE-based TFs can program endogenous gene expression without the requirement of extra synthetic promoters.^18,20 The currently available Tango,^45,46 MESA,^47,48 synNotch,^31,40,41 ChaCha,⁴⁹ Cal-Light,⁵⁴ intrabody sensor,⁵⁶ LOCKR^57,58 and COMET⁵⁹ employ syn-TFs for orthogonal signal pathway construction, while TCS⁵⁰ and POST⁶⁰ utilize orthogonal signaling systems derived from prokaryotes (Table 1).

Of note, the most widely used orthogonal TFs are of non-human origin (e.g., yeast and bacteria), therefore carrying high risks of immunogenicity in clinical applications. The use of humanized syn-TFs comprising both the DNA-binding domain and activation domain derived from human TFs will largely reduce the immunogenic risk. However, these fused TFs are still able to activate endogenous signaling, which reduces their orthogonality. Notably, Khalil and colleagues de novo designed a panel of fully humanized synthetic zinc finger transcription regulators (synZiFTRs) engineered by an array of zinc finger domains, which can specifically recognize their cognate short DNA-binding motifs, achieving genome-orthogonal specificities.⁶⁴ This kind of pioneering work will accelerate the progress of humanization of orthogonal signaling-based receptors, which is pivotal to bridge the translational gap to the clinic.

Soluble ligand-binding receptors and surface ligand-binding receptors

Synthetic receptors can bind to a range of soluble and surface-bound ligands with a high specificity and sensitivity. In line with the features of ligands, the corresponding synthetic receptors can be categorized into soluble ligand-binding receptors and surface ligand-binding receptors.^18,21

A large variety of ligands are in the soluble form, including most chemical molecules, hormones, cytokines, growth factors, intracellular soluble proteins, and some peptides cleaved from membrane proteins (e.g., carcinoembryonic antigen (CEA)⁶⁵ and amyloid-beta (Aβ)⁶⁶). The above listed soluble molecules can induce the activation of soluble ligand binding-receptors, such as Tango,^45,46 ChaCha,⁴⁹ MESA,^47,48 chimeric cytokine receptor,^51,52 CHOMP,⁵⁵ GEMS⁵³ LOCKR^57,58 and POST⁶⁰ (Table 1). By using small chemical molecules that can cross the plasma membrane, several intracellular synthetic receptors (e.g., POST, CHOMP and LOCKR) have been developed.²⁰ Meanwhile, they can also act as extracellular ligands when the ligand-binding domain of a receptor locates outside the plasma membrane, scilicet cell-surface receptors (e.g., Tango, ChaCha, MESA and GEMS).²⁰ For peptide/protein ligands, in most cases, they cannot enter the cell, and therefore act on cell-surface receptors to subsequently trigger downstream intracellular signaling cascades.

Surface-bound ligands fixed on the plasma membrane of sender cells can trans-activate cell-surface receptors on adjacent receiver cells.^18,20 CAR and synNotch are typical surface ligand-sensing receptors which theoretically cannot be activated by soluble ligands.^18,20 The different properties of these two types of synthetic receptors possess distinct mechanisms of action. Soluble ligand-binding receptors are usually activated by ligand-induced dimerization to trigger downstream signaling, whereas surface ligand-binding receptors require a mechanical force generated by ligand-receptor interaction to activate downstream signaling. Therefore, activation of soluble ligand-binding receptors (e.g., chimeric cytokine receptor and GEMS) can be induced by ligands after a long-distance transport, whereas surface ligand-binding receptors (CAR and synNotch) can only function in a juxtacrine manner.³⁴

Partially modular receptors and fully modular receptors

Over the past decade, an increasing number of synthetic receptor systems have been engineered and refined. Based on whether they are made up entirely of reconfigurable components, we divide synthetic receptors into partially modular receptors and fully modular receptors.

Partially modular receptors can be constructed by engineering the artificial sensor domain while retaining the original actuator domain, or vice versa. The former synthetic receptors can rewire endogenous signaling pathways to new ligands (e.g., CAR, chimeric cytokine receptor and GEMS), while the latter ones can activate an alternative pathway by natural ligands (e.g., Tango, ChaCha and dCas-synR)^18–20 (Table 1). Therefore, the former synthetic receptors can be categorized as natural signaling-based receptors and the latter ones as orthogonal signaling-based receptors (as discussed above).

Fully modular synthetic receptors, with both the sensor domain and actuator domain engineered, including synNotch,^31,40,41 SNIPRs,⁴² MESA^47,48 and LOCKR,^57,58 can execute novel functions without disrupting the endogenous pathways in an orthogonal way (Table 1). It is worth noting that the ‘building brick’ for modular assembly of fully modular synthetic receptors can be either derived from either pre-exiting natural components (e.g., synNotch, SNIPRs and MESA) or de novo designed ones (e.g., LOCKR).²⁰

Chimeric antigen receptors (CARs)

CARs are a best-known class of synthetic receptor systems, which have already been approved for CAR T cell therapies by the U.S. Food and Drug Administration (FDA)^6,68 (Supplementary Note 1). They represent a success of advances in synthetic biology to pioneer a new generation of therapeutics, motivating the continuous optimization of CAR-design and the development of new synthetic receptor systems throughout the field.

Currently, CARs have been widely armed in immune cells like T cells, NK cells and macrophages (reviewed in refs. ^5,69,70). Besides, innate T cells including invariant natural killer T (iNKT) cells, mucosal associated invariant T (MAIT) cells and gamma delta T (γδT) cells have been developed as promising immune effector cells, because they display intrinsic antitumor microenvironment (TME) capacities and minimal risks of graft versus host disease (GvHD) (reviewed in refs. ^71–73). CAR mainly consists of the extracellular domain (ECD), the transmembrane domain (TMD) and the intracellular domain (ICD) (Fig. 3a). The modular characteristics of its structure confer it advantages to be highly amenable to modification and redesign.

The ECD can be segmented into the signal peptide (SP) and the ligand-binding domain (LBD). The SP prompts the transmembrane receptor protein to be translocated to the plasma membrane, which usually is cleaved from mature CAR protein co-translationally^74,75 (Supplementary Note 2). The LBD, also called antigen-recognition domain, is typically a single-chain fragment variant (scFv).^76,77 The scFv is a compact artificial construct that comprises the immunoglobulin light and heavy chain variable regions connected by a flexible linker⁷⁸ (Supplementary Note 2). The scFv is widely adopted in CAR construction due to its compact size, high affinity and specificity for antigen recognition.⁷⁷ By changing the scFv, CARs can specifically recognize different antigens on various cancer cells (e.g., cluster of differentiation 19 (CD19), mesothelin, CEA, B-cell maturation antigen (BCMA), CD38) by one-to-one authentication,³⁹ and subsequently activate cancer-killing of CAR T cells. Besides simply replacing the scFv, CARs can be equipped with bispecific antibodies consisting of two different linked antigen-recognition domains (also referred to as ‘tandem CARs’)⁷⁹ (Fig. 3c and Table 1). As a result, tandem CAR T cells can recognize different antigens expressed in a single cancer cell (e.g., human epidermal growth factor receptor 2 (Her2) and interleukin 13 receptor alpha 2 (IL13Rα2),⁸⁰ CD19 and CD20⁸¹), and therefore reduce the possibility of tumor escape.^82,83

Recently, Wong and colleagues invented an intriguing generalized CAR platform, the split, universal, and programmable (SUPRA) CAR system for T cell therapy⁸⁴ (Fig. 3c and Table 1). In this system, the conventional CAR architecture is split into two elements: (1) a soluble zipFv by fusing a scFv to a leucine zipper and (2) a universal zipCAR containing the remnant transmembrane and intracellular domains attached to an extracellular cognate leucine zipper.⁸⁴ By adding different zipFv proteins, a unique zipCAR-expressing T cell can retarget different tumor antigens.⁸⁴ Moreover, SUPRA CARs can fine-tune T cell activation and program the Boolean logic operation, which enhances the safety and efficacy of T cell therapy.^84,85 Alternatively, various switchable CAR systems with bispecific adapters are under way (refs. ^86–95 and also reviewed in refs. ^96,97) (Fig. 3c and Table 1). More information about UniCAR is shown in Supplementary Note 3.

As mentioned previously, conventional CARs can only target surface-bound ligands while being unable to sense soluble ligands. To redirect the response of CAR T cells to soluble cues, Chang et al. engineered a CAR that enabled T cells to respond robustly to diverse soluble ligands via dimerization-induced mechanotransduction mechanisms⁹⁸ (Table 1).

The TMD is usually a hydrophobic α-helix that spans the plasma membrane and functions to anchor CAR proteins to the membrane.^68,99 Besides controlling membrane integration, the TMD also regulates key interactions between CARs, such as assembly, activation and high-order clustering.^68,99,100 Compared to ECD and ICD, TMD has received relatively less attention and research. Almost all the TMDs used in CARs are derived from natural T cell proteins, such as CD8, CD28, CD4 and CD3ζ.^99,101 Nonetheless, studies have indicated that the TMDs certainly impact the stability and function of CAR.^102,103 More recently, Elazar et al., demonstrated that the specific oligomeric state programmed by de novo designed TMDs can tune CAR function and CAR T cell activity, as well as decrease cytokine releasing relative to the commonly used CD28 TMD.¹⁰⁰

By linking the extracellular antigen-binding and transmembrane domains, the hinge region functions as a linker, providing the flexibility to access sterically hindered epitopes^99,101 (Supplementary Note 2). Studies have revealed much more crucial roles of the hinge region per se or it being coupled with the TMD than initially expected. These results demonstrate that the length and composition of the hinge can affect CAR functionalities, including fine-tuning CAR signaling activity, improving antitumor efficacy, and lowering cytokine release or neurotoxicity.^102–110 Therefore, in CAR T cell engineering, it is essential for systematic evaluation and optimization of the hinge and the TMD to ensure optimal performance and reliability.^{68,77,99,111–113}

The ICD transmits activation signals upon the antigen’s binding to the ECD. The ICDs of most well-studied CARs contain a CD3ζ-derived signaling moiety, which has three immunoreceptor tyrosine-based activation motifs (ITAMs)^77,114 (Supplementary Note 4). From the first-generation (1 G) of CAR derived in the 1990s¹¹⁵ to now, the structure of CAR per se has been constantly evolving up to the fifth-generation (5 G) (Fig. 3b and Table 1), aiming to increase specificity and minimize off-target toxicity.^114,116

The first-generation (1 G) CARs only contain ITAMs to provide activating signaling with none co-stimulatory domain¹¹⁵ (Fig. 3b and Table 1). Though 1 G CARs were proved to be able to activate T cells and control tumor in mice,^117–119 they failed to achieve antitumor responses in subsequent clinical studies.¹²⁰ The reason could be the CD3ζ alone is insufficient to activate resting T lymphocytes or trigger the production of optimal amounts of cytokines.^121,122 To solve these problems, second-generation (2 G) CARs were created by incorporating a costimulatory domain (Supplementary Note 4) on the basis of 1 G CARs (Fig. 3b and Table 1), which enables the activation of a second signal when stimulated by a tumor antigen.^123–128 This improvement has achieved the enhancement of cytokine production, CAR T cell persistence and antitumor efficacy.¹²⁵

And the third-generation (3 G) CARs are designed to combine multiple costimulatory domains to further enhance CAR T cell potency^129–132 (Fig. 3b). Although potential benefits including prolonged persistence and increased antitumor efficacy have been demonstrated both in vitro and in vivo,^130–137 some clinical results did not prove a significant superiority of 3 G CAR T cells to the 2 G CAR T cells.^138–140 As the available data was obtained from relatively small and heterogenic samples, it is still too early to draw a conclusion, and further large-scale studies are required to fully evaluate their feasibility.

The fourth-generation (4 G) CARs have been engineered from the 2 G CARs to constitutively or inducibly secrete cytokines (e.g., IL-12, IL-7, IL-15, IL-18, and IL-23), enhancing the immune modulating capacities^141–150 (Fig. 3b and Table 1). Upon CAR-mediated T cell activation, cytokines can be ideally released within targeted tumor locally, alleviating systemic side effects and solving the problem of insufficient production. Since the 4 G CARs can not only improve CAR T cell activation but also hijack host immune cells to enhance tumor killing,^151,152 the 4 G CAR T cells are also referred to as T cells redirected for universal cytokine killing (TRUCKs).^{114,116,151,152} Compared with conventional CAR T cells, TRUCKs show enhanced expansion and antitumor activity in preclinical studies, especially in animal models of solid tumors. However, in practice, systemic side effects of released cytokines may occur upon entry into circulation. For example, in a clinical study, Rosenberg and colleagues modified tumor-infiltrating lymphocytes (TILs) to express IL-12 under a NFAT-inducible promoter to treat metastatic melanoma. They observed severe toxicity induced in most patients, which is likely attributed to the secreted IL-12.¹⁵³ Here, more data derived from clinical trials are necessary to assess their safety and efficacy.

Distinct from TRUCKs, the immunomodulatory factor expression module of the 5 G CARs is replaced with a novel costimulatory domain which can activate a specific signaling pathway inside equipped CAR T cells^114,116 (Fig. 3b and Table 1). Based on this excellent principle, several approaches are emerging, of which the addition of IL-2 receptor β-chain (IL-2Rβ) into CARs is a notable example.¹⁵⁴ Upon activation by the antigen, the extra IL-2Rβ domain allows the activation of the JAK kinase and the STAT3/5 signaling pathways, which can empower CAR T cells to achieve superior antitumor effects with minimal toxicity in mouse models due to a better persistence and expansion in vivo. However, it could potentially increase the risk of CRS, thus requiring to be cautiously addressed in translational studies.¹⁵⁴

CAR T cell therapy has evolved and gradually matured during the past decades, showing great therapeutic potential in blood and bone marrow cancers. However, challenges remain in CAR-based solid tumor immunotherapy due to tumor heterogeneity, inefficient trafficking and tumor infiltration, and an immunosuppressive TME.^83,155–159 To facilitate CAR T cell therapy for solid tumors, several strategies have been developed. For instance, dual CAR, tandem CAR and UniCAR systems can recognize more than one antigen, helping mitigate tumor antigen escape. CAR T cells armed with matched chemokine receptor expression can permit trafficking and infiltration, achieving enhanced killing of solid tumors.^160–163 Also, engineering CAR T cells to express heparanase degrading extracellular matrix (ECM) can promote tumor T cell infiltration and antitumor activity.¹⁶⁴ In addition, the aforementioned TRUCKs were developed to overcome the drawbacks of the TME on CAR T cell therapy to treat solid tumors by immunomodulatory factors.^142,144

However, it remains challenging to minimize CAR-immune cells’ off-target and off-tumor toxicity. In this context, the development of next-generation CARs has already been underway^96,99,101 (Fig. 3c and Table 1). For example, the combinatorial logic control by dual CAR^165,166 or synNotch CAR^40,167,168 can enhance tumor targeting specificity via the presence of two or more antigens. On the other hand, a safety control by ON- and OFF-switch CAR can finetune CAR activity^169–172 and kill-switch CAR can manage the lifespan of CAR T cells.^173,174 These are promising strategies to improve the safety of CAR T cell immunotherapy in the future.

Furthermore, most CAR T cells under investigation currently are engineered by inserting the CAR construct into autologous T cells without disrupting the endogenous T-cell receptor protein (TCR) gene. Under this condition, the risk of GvHD, which is triggered by human leukocyte antigen (HLA)-TCR mismatching,¹⁷⁵ can be avoided. To facilitate allogeneic “off-the-shelf” CAR T cell transplantation, T cell receptor α chain (TRAC) deletion using endonucleases, thereby disrupting cell surface expression of the αβ T cell receptor (TCRαβ), can successfully prevent graft-versus-host responses.^175–180 Recently, CRISPR/Cas-mediated knockin technology enables the precise integration of CAR-encoding gene into TRAC locus in human peripheral blood T cells,¹⁸¹ which not only facilitates the production of allogeneic CAR T cells,^181–183 but also enhances T cell potency as the edited T cells outperformed conventionally engineered CAR T cells.^181,184 However, more recent studies revealed that the endogenous TCR plays a critical role in promoting long-term in vivo persistence of CAR T cells in not only animal models but also patients.^185,186 These results collectively indicate that it is crucial to balance the intricate effects of removing endogenous TCR from CAR T cells in tumor immunotherapy.

Synthetic Notch (synNotch) receptors

In-depth studies of Notch receptors provide critical insights into molecular mechanisms of Notch receptors.¹⁸⁷ And the intrinsic features including modularity and mechanical forces-triggered signaling independent of native ligands make Notch receptors ideally suitable for modular chimeric receptor engineering.¹⁸⁸ Taking advantage of it, Lim and colleagues created the innovative synNotch receptor system by utilizing the transmembrane core domain of native Notch receptors (Fig. 4a and Table 1), alongside the extracellular sensor domain and intracellular actuator domain.^31,40,41

Three intriguing works reported the modular synNotch receptors functioning with customized sensing and responsive behaviors in mammalian cells, including T cells.^31,40,41 Morsut et al. demonstrated that synNotch can function orthogonally to control cell differentiation, spatial patterning and Boolean decisions.³¹ Meanwhile, Roybal et al. reported that synNotch can sense tumor antigen and then drive the expression of CARs (synNotch CARs as mentioned above) (Fig. 3c and Table 1), which allows the engineering of AND-gate T cells activated only by dual antigen recognition.⁴⁰ Roybal et al. also reported that synNotch enables CAR T cells to yield customized therapeutic responses like secreting cytokines and antibodies in a very precise and localized manner.⁴¹ Compared to conventional CARs whose activation drives T cells to produce a native cytokine profile, synNotch can control the expression of extra user-defined cytokines.⁴¹ A more powerful combination of synNotch receptors with CARs for precise and effective cancer-killing is on the rise, which endows synNotch the great potential of becoming another synthetic receptor system applied in cancer immunotherapy.

Due to its superior performance, the synNotch platform has been widely adopted for designer cell-engineering,^25,33,34 particularly for the engineering of CAR T cells^{40,41,189,190} and the programming of self-organizing multicellular structures.^31,32,35 Although there are still challenges and limitations when it comes to practical biomedical applications, researchers are trying hard to figure out possible solutions. First, native Notch receptors have an inherent feature that both trans- and cis-interaction modes co-exist,³¹ so activation cannot be achieved due to cis-inhibition when the cognate ligand presents on the same surface as the receptor does (cis).^31,191 To avoid cis-inhibition, the ‘flippase-out’ strategy is employed to achieve the mutually exclusive expression of the synNotch and its ligand in flippase recombinase transgenic Drosophila.¹⁹¹ Second, synNotch activation requires mechanical forces triggered by surface-bound ligands, making synNotch receptors unable to sense soluble ligands.³¹ To address this, Toda et al. engineered anchor cells which can tether the soluble ligands (e.g., diffusible GFP), thus enabling synNotch receptors to respond to diffusible synthetic morphogens³⁵ (Fig. 4b and Table 1). Third, synNotch receptors display a high level of ligand-independent activation,¹⁹² which is also quite common for other synthetic receptor systems.²⁰ Excitingly, a recent work reported an improved version of synNotch, named enhanced synthetic Notch (esNotch) receptor. By adding a native Notch-derived intracellular hydrophobic sequence, an incredible reduction (14.6-fold) in the background activity level was achieved¹⁹² (Fig. 4b and Table 1). Impressively, Roybal and colleagues have achieved systematic and modular improvements of the synNotch architecture, including modifications of the extracellular sensor domain, TMD, intracellular juxtamembrane domain (JMD) and actuator domain.⁴² The evolved synNotch system is referred as synthetic intramembrane proteolysis receptors (SNIPRs) (Fig. 4b and Table 1), which realized background reductions and enhanced ligand-induced signals.⁴² Meanwhile, SNIPRs can be fully humanized with humanized modules, minimizing the risk of immune rejection.⁴² The use of humanized syn-TFs, including synZiFTRs, not only retains the orthogonality of SNIPRs, but also largely reduces the immunogenic potential for cell-based therapies.⁴²

In addition, Fussenegger and colleagues derived an orthogonal chemically activated cell-surface receptor (OCAR) system on the basis of synNotch by substituting conventional protein-specific LBDs with chemically induced dimerization (CID) domains (Fig. 4b). Induced by small molecules, the engineered OCARs on one cell can form heterodimers and trigger signal activation by releasing synTFs.¹⁹³ When the OCAR system is co-expressed with the conventional synNotch on the same cell, one part of OCAR will be sequestered by the synNotch receptor through coiled-coil interactions, making OCAR unable to be activated by the inducer. When synNotch receptor is activated by the presence of sender cells, the OCAR can restore its responsiveness to the inducer and thus further enhance synNotch signaling by adding inducers. Due to their mechanism of action, OCAR systems exhibit an intrinsic off-switch, which might be used as a safety switch in the case of toxicity or malfunction¹⁹³ (Fig. 4b and Table 1).

Modular extracellular sensor architecture (MESA)

Like synNotch receptors, the modular extracellular sensor architecture (MESA) receptors also take advantage of the proteolytic release of synTFs, which translocate to the nucleus and initiate transcriptional activation to execute orthogonal signals without interrupting endogenous signaling pathways^47,48 (Fig. 5a and Table 1). But different from synNotch receptors that require mechanical forces induced by surface-tethered ligands,³¹ MESA receptors signal via soluble ligand-triggered receptor heterodimerization to perform the subsequent proteolytic cleavage.^47,48

Each MESA receptor comprises two different single-pass transmembrane proteins, as their intracellular domains are distinct (Fig. 5a). MESA receptors also comprise three modular domains to transduce extracellular ligand-binding inputs into intracellular transcriptional outputs through synTF-releasing.^47,48

By engineering the extracellular sensor domain, MESA receptors can be redirected to new ligands.^47,48 However, the prototype of the MESA architecture also suffers from high level of ligand-independent activation possibly due to transient receptor dimerization during trafficking or on the cell surface.⁴⁸ To address this, Leonard and colleagues have been taking efforts on the improvement of the MESA system, especially to reduce undesired background signals. A systematical strategy was employed to elucidate and refine the MESA architecture, demonstrating that the TMD optimization can lower the background signaling and elevate the target signaling¹⁹⁴ (Fig. 5c and Table 1). In another study, authors exploited a computational design strategy to use the split-TEVp system for MESA optimization (Fig. 5c and Table 1), achieving both a low background and a high fold induction.¹⁹⁵ In parallel, a similar dCas9-based synthetic receptor system termed ‘dCas9-synR’ was reported (Fig. 5c and Table 1), which is capable of coupling native input signals with the direct activation of user-defined output response programs.¹⁹⁶

As discussed above, MESA and dCas9-synR are conceived as cell-surface receptors based on previous studies,^20,28,197 but a recent study challenged this conclusion.¹⁹⁸ In this study, the authors showed that both receptor systems did not work with purified protein ligands, but were only activated when using co-transfected ligands or the cell permeable molecule rapamycin as inducers. Therefore, they suspected that these two receptor systems function within the endoplasmic reticulum (ER), and should not be classified into cell-surface receptors.¹⁹⁸ By carefully checking previous publications, we find in three works relevant to MESA, the authors have shown that MESA receptors could be activated by purified vascular endothelial growth factor (VEGF) protein,^48,62 and MESA proteins expressed on cell surface were detected by flow cytometry.^48,194 And yet the inconsistent conclusion drawn by the recent study might attribute to the different components of MESA constructs (e.g., extracellular linker and TMD) used in research,¹⁹⁸ which could change the expression level of MESA on cell surface as indicated by previous works.^48,194 Though the early study done by Schwarz et al. showing that MESA could be activated by purified VEGF supported MESA as a cell-surface receptor,⁴⁸ a more recent work from Krawczyk et al. failed to replicate the original data.⁶² Using purified VEGF, few changes (in the range of about 1.1 ~ 1.2 fold) were observed, which strongly indicated that MESA did not achieve cell-surface receptor performance.⁶² Thus, as the bias in the literature exists,^{48,62,196,198} to give a precise characterization of both MESA and dCas9-synR systems, more replication attempts are definitely required.

Generalized extracellular molecule sensor (GEMS)

A clever insight into the EpoR architecture and signaling mechanisms^199–202 spurs the development of chimeric cytokine receptors based on EpoR scaffold^203–212 (Fig. 5d and Table 1). These chimeric cytokine receptors (e.g., SD1D2g-1A) contain extracellular scFv, extracellular EpoR D1D2 as well as TMD domains, and the intracellular domains of cytokine receptors (e.g., glycoprotein 130 (gp130) and EpoR)²⁰⁵ (Fig. 5d and Table 1). These chimeric cytokine receptors can employ robust ligand-dependent ON/OFF regulation and mimic the function of the native cytokine receptor system.^205,206

Based on this prototype, the GEMS system is rationally designed and engineered. GEMS receptors comprise a standard transmembrane scaffold derived from erythropoietin receptor (EpoR) with F93A mutation (Fig. 5b and Table 1), abolishing native ligand (erythropoietin) sensitivity.⁵³ Extracellular LBDs can be modularly replaced to sense and respond to a wide range of extracellular soluble ligands. Significantly, the ICDs employ various signaling domains to reroute inputs into distinct endogenous pathways (e.g., JAK/STAT, PI3K/Akt, PLCG and MAPK/ERK).⁵³ In recent reports, the GEMS platform has evolved into the superior generalized engineered activation regulators (GEARs)⁶² and advanced modular bispecific extracellular receptors (AMBERs)²¹³ (Table 1). The GEAR system combines endogenous signaling-induced synTF nuclear translocation and dCas9-based gene expression via the DNA-binding module (Fig. 5e). Various pathways (e.g., NFAT, nuclear factor kappa B subunit (NFκB), mitogen activated protein kinase (MAPK) or SMAD) activated by diverse receptors including GEMS receptors can be incorporated into engineered GEARs.⁶² The AMBER system is, in short, a type of programmable DARPin- and GEMS-based receptors (Fig. 5e), which inherits the characteristics and advantages of both systems. Thus, AMBERs can be engineered and refined to bind to new soluble ligands in a modular and predictable manner.²¹³ For proof of concept, the GEMS and its derivates have been applied to engineering designer cells in vitro and in vivo for disease treatment (Fig. 2d) (refs. ^{53,62,213,214} and reviewed in refs. ^18,19). Though showing promise in preclinical studies, challenges remain to be addressed before the translation of the GEMS system into clinics. A major challenge is the safety concern relevant to the immunogenicity of designer cells. Recently, Weber and colleagues engineered a material-genetic interface as the safety switch for mammalian therapeutic cells, in which they achieved the control of cell survival by allowing the designer cells to sense whether they were embedded in the hydrogels using a GEMS-based receptor.²¹⁵

As discussed above, synthetic receptors hijacking natural signaling pathways usually inevitably activate endogenous transcription networks to some extent. Using synthetic promotors, GEMS receptors can induce user-defined transgenic target expression.⁵³ The advantage is that GEMS receptors can generate high signal-to-noise ratios.¹⁸ However, the accompanying disadvantage is the perturbation of the gene regulatory network might affect the proliferation and survival of the cells being engineered.