Design and applications of synthetic information processing circuits in mammalian cells

Synthetic biology as a discipline merging life sciences with engineering has since its inception proved to be a valuable tool in research as well as in industrial and medical applications. While wholly synthetic genomes are no longer a fiction,^1,2  it is now becoming possible to not only replace an organism's genetic material with synthetic DNA, but to also design the function of that synthetic DNA and produce genetic circuits that either respond to combinations of external or internal signals in therapeutic applications^3,4  or provide us with valuable insights into the complex relationships between DNA, RNA and proteins in the functional cell.^5,6  The principle idea behind synthetic biology is in fact a simple one—to define functionally separate basic parts that behave in a predictable way and can be joined and recombined as modules into higher system, which will perform much more complex functions, but in turn also act predictably. Basic modules for gene editing, transcriptional control, translational control and functional control of proteins were first developed in prokaryotes, and drawing a parallel to electronic systems, transistors,⁷  Boolean logic gates,⁸  oscillators,⁹  switches,¹⁰  band-pass filters¹¹  and many other regulatory modules and devices were assembled. Many of these devices have then been transferred or independently reinvented to function in eukaryotic cells. In this review we will focus on synthetic biology in mammalian cells and aim to present an overview of different mechanism of gene editing and transcription regulation (Fig. 1A) through direct control of transcription factors or through rewired signaling pathways, we will provide a window into construction and function of modules operating on the level of RNA and translation regulation (Fig. 1B), and will describe attempts at construction of faster responsive elements through posttranslational control of proteins and their functions (Fig. 1C). Finally, we will describe examples of interesting and complex devices that represent an important advance towards potentially useful therapeutic applications of mammalian synthetic biology.

The first step in designing a controllable synthetic biological circuit is the selection of appropriate input signals. Often, proof of principle circuits are designed so that the input signal is provided by the presence or absence of certain constitutively active elements.^12–14  On the other hand, circuits can be designed to conditionally respond to a wide range of externally supplied or intracellular signals, specific to a disease state, physiological conditions or cell type. The latter include the presence of certain DNA elements,¹⁵  short RNAs,¹⁶  or endogenous proteins,^17–19  and circuits can respond to the signals present locally or systemically in an organism, such as metabolites,^20–23  neurotransmitters²⁴  or cytokines.^25,26  Development of physiologically relevant synthetic sensors is fundamental to development of therapeutic circuits and we will cover some examples of these input systems along with the examples of therapeutic use in a later section of this review. In these opening section however, we will focus on the introduction of synthetic input systems, designed to be orthogonal to the mammalian cell biology.

For therapeutic applications of biologic devices as well as for investigation purposes we often want to design systems that respond only to selected input signals without sensing or interfering with the state of the chassis cell or organism in unpredictable ways. This allows us to exert control over the target cells, either to perturb the investigated system or to safely initiate or terminate production of therapeutic effectors. The earliest developed and most well characterized orthogonal sensing systems are based on small chemical inducers. In genetic circuits, the inducers can exert their effect directly on transcription factors (TFs) by allosterically affecting their binding affinity for target DNA.²⁷  Many such transcription factors are derived from bacteria²⁸  and modified to function in mammalian cells by fusion with repression domains (such as the Krüpel associated box—KRAB²⁹ ) or activation domains (such as the virally derived VP16 domain³⁰  or the human p65 domain³¹ ). Additionally, appropriate binding sites for those transcription factors have to be inserted into the promoter region of the gene of interest. The most widely used example of a bacterial transcription factor adapted for mammalian use, and one that superbly illustrates the principles of modular design, is the tet system. The TetR protein, originaly found in bacteria, has first been fused to the VP16 domain to generate a ligand dependent activator that drives expression from a minimal promoter with the tetO binding sites.³²  TetR binds to tetO only in the absence of its ligand, doxycyclin, therefore the TetR-VP16 fusion (tTA) creates the so called tet-OFF system that activates transcription in the absence of doxycycline (Fig. 2A). If only transient expression is desired, the tet-OFF system places an inconvenient burden on both the researcher and the cell in having to maintain and tolerate a constantly high concentration of doxycyclin when no protein is being produced. To circumvent this and activate the target gene only in the presence of doxycyclin, two other design options are available—either the fusion of TetR to the KRAB in combination with a constitutively active tetO containing promoter, which could be described as derepression in the absence of the ligand and was termed the TET-dependent transsilencer tTS³³  (Fig. 2B), or the reverse tet transcription activator rtTA, which binds to the target DNA only in the presence of the ligand and thereby activates transcription with the help of the VP16 fusion domain³⁴  (Fig. 2C). These options differ in kinetic properties and leakage, which are therefore the key factors that need to be considered during the design of circuits containing the tet systems. Importantly, there is a whole range of bacterial transcription factors responding to different small molecule ligands described, however for most of them mutants that bind to target DNA either in the absence or the presence of ligands are not available, so fusion with either VP16 or KRAB is the preferred strategy in harvesting bacterial transcription factors for the design of OFF and ON systems in mammalian cells, respectively.²⁷  Many other small molecule responsive transcriptional regulators have been harvested from nature, mainly bacteria, such as acetaldehyde responsive elements, which enable cell stimulation through volatile components,³⁵  a urate responsive system, where in addition to the natural sensor addition of a transporter improved the responsiveness,²³  and the sensing of the flavonoid phloretin, an apple metabolite, by the regulator from Pseudomonas putida.³⁶ 

Another strategy of controlling synthetic biological circuits is by their coupling to endogenous signaling networks. The receptor type that lends itself well to this strategy is the G-protein coupled receptor (GPCR) also known as the serpentine receptor for its seven membrane passing helical domains.³⁷  The greatest merit of this receptor family is the number of its members and the range of stimuli they can respond to. GPCRs are the largest and most diverse group of membrane receptors in eukaryotes and can bind neurotransmitters, hormones, pheromones, odors and various other synthetic or biological molecules and some of them even respond to light, which requires binding of light-sensitive cofactors.³⁸  Upon ligand-induced activation, the GPCR acts as a guanine exchange factor for the coupled GTPase, which initiates downstream signaling, for example by activating adenylyl cyclase.³⁹  Ausländer et al. have designed synthetic mammalian sensors by coupling this native GPCR signaling pathway to transcriptional control of transgenes by placing them under a cAMP-responsive promoter³⁸  (Fig. 2D). Unfortunately, the downside of this type of signaling in design of synthetic circuits is that the GPCR signaling networks are not mutually orthogonal, resulting in crosstalk between signalization through various GPCR pathways if executed in a single cell. Müller et al. have been able to circumvent this by using designed cell consortia that convert signals from GPCRs into orthogonal soluble mediators that can be combined in the downstream processing cells.³⁵ 

The activated GPCR is eventually phosphorylated in what is usually a negative feedback loop or desensitization mechanism. The phosphorylated GPCR is recognized and bound by arrestins which prevent it from further binding to G-proteins and facilitate its internalization with eventual dephosphorylation and recycling to the membrane. It is this binding of the activated GPCR to arrestin that can be utilized for the design of synthetic signaling pathways. Fusion of a transcription factor to the GPCR through a protease specific cleavage site, and the fusion of the cognate protease to arrestin, result in a system that upon GPCR activation leads to the cleavage of the TF from the membrane, its translocation into the nucleus and activation of transcription of target genes (Fig. 2E). This transcriptional activity persists well after the signal from the GPCR is deactivated or its ligand is no longer present, which depending on the specific circuit topology can either be considered an advantage or a disadvantage.⁴⁰  Additionally, while arrestins are believed to bind fairly exclusively to phosphorylated GPCRs, synthetic signaling as described above has been shown to be relatively leaky. A split protease strategy with one protease fragment attached to arrestin and the other fragment to the GPCR together with the cleavage site and TF were designed to compensate for this problem. Additionally, cleavage site modifications and arrestin truncations were employed with varying efficiency, depending on cell type and the exact GPCR used.⁴¹ 

Following the principle of complete orthogonality, an effort was made to transplant bacterial two-component systems that are also able to sense a wide range of inputs, into mammalian cells. These function through a histidine-aspartate phosphorelay between a receptor histidine kinase and a response regulator transcription factor. While Hansen et al. have been able to show the effectiveness of the phosphorelay, inducible activation of two-component systems has not yet been achieved in mammalian cells and it remains unclear whether the constitutive activity of the transplanted signaling circuit was a result of some unpredicted process in the chassis cell or the presence of activators in mammalian cell media.⁴² 

Chemically inducible dimerization (CID) offers a powerful tool to control and manipulate target proteins in living cells. A typical CID system consists of two different protein domains which dimerize upon ligand binding. The most widely used CID today is the FKBP–FRB system, based on the human FKBP12 and mTOR proteins which heterodimerize upon binding of rapamycin or its analogs.⁴³  This system has been used to generate rapamycin inducible translocation⁴⁴  and activation of transcription factors,⁴⁵  inteins,^46,47  phosphatases⁴⁸  and kinases⁴⁹  and a number of reviews describing its development and applications are available.^50–52  In order to simultaneously control and combine multiple processes, however, orthogonal CIDs are required. Recently two new CID systems were developed using the plant hormones abscisic acid and gibberellin.^53,54  Liang et al. modified proteins of the plant abscisic acid signaling pathway (PYL1 and ABI1) in order to chemically induce proximity of intracellular proteins in mammalian cells⁵³  and the gibberellin induced CID was developed by optimizing GID1 and GAI heterodimerizing proteins. Additionally, to effectively control the dimerization of GID1 and GAI in mammalian cells, chemical modifications of gibberellin were required. At physiological pH, the carboxylic group of gibberellin is negatively charged, decreasing its efficiency of internalization across the cellular membrane. Esterification improved membrane permeability while retaining the full heterodimerization potential of gibberellin due to processing by intracellular esterases.⁵⁴  Interestingly, abscisic acid required no such modification in spite of the fact that it is also acidic. The rapamycin-, abscisic acid- and gibberellin-induced systems are completely orthogonal and thus well suited for control of separate signaling events.⁵⁵  Gibberellin and rapamycin were used to independently induce protein translocation to mitochondria and the plasma membrane respectively,⁵⁴  while abscisic acid and rapamycin were used to translocate proteins to the nucleus or the plasma membrane respectively.⁵³  Finally, abscisic acid and gibberellin were combined in design of inducible Cas9, behaving either as an OR or an AND gate in respect to the two inputs.⁵⁵ 

Perhaps the most spectacular example of inducible systems for mammalian cells was introduced by light sensitive receptors. Light regulated activation of neurons represents an extremely powerful tool to study and engineer the nervous system.⁴⁴  The advantage of light in synthetic circuits is the precision with which stimulation can be controlled both temporally and spatially. Although stimulation by light in living tissue remains a challenge due to the tissue's opacity for light, light exhibits fewer side effects than chemical stimulation.

The Arabidopsis light-sensitive cryptochrome 2 (CRY2) and its dimerization partner CIB1 or their truncated mutants have been established as the most useful heterodimerization domains due to the absence of a requirement for exogenous chromophores and fast kinetics of protein association upon light stimulation and dissociation in dark.⁵⁶  This fast and reversible system was thus used to generate light-inducible transcription factors in mammalian cells coupled to the yeast GAL4 DNA-binding domain⁵⁶  as well as the designable TALE⁵⁷  and the programmable CRISPR/Cas9 DNA-binding domains.⁵⁸  The possibilities for synthetic circuit design are further expanded by a light inducible Cre recombinase.⁵⁶ 

In order to generate orthogonal signaling circuits using only light as an input, one must pay attention to the wavelengths of light that induce any particular system. The CRY-CIB system is induced by irradiation with blue light and two other systems induced by blue light exist, based either on FKF1-GIGANTEA interaction or on the LOV domain's inducible conformational shift. The FKF1-GIGANTEA is a heterodimerization system that differs from the CRY-CIB system most importantly in the reaction kinetics and reversibility. While CRY-CIB quickly dissociates to inactive monomers in the dark, FKF1 and GIGANTEA remain associated for at least 1.5 hours, increasing their activity even after short stimulation, but making the system not suitable for precise temporal control.⁵⁹  The LOV domain on the other hand utilizes a completely different mechanism of activation. Upon blue light stimulation it undergoes a conformational change, exposing its C-terminal Jα helix.⁶⁰  This allowed the engineering of tunable light-inducible dimerization tags (TULIPs)—peptide tags buried inside the LOV protein while in the dark state, but exposed after the unfolding of the Jα domain, enabling interaction with a PDZ domain. Point mutations in the LOV domain were required to increase the dynamic range of the system by reducing the background Jα undocking while point mutations in the PDZ domain provided a wide range of association affinities.⁶¹  Interestingly, Müller et al. used the PDZ domain with a low affinity to their advantage in designing a circuit with three orthogonal light inputs.⁶² 

The phytochrome (Phy-PIF) system as one of the other two systems used in the circuit with three light inputs, responds to red light. Two other important features distinguish this system from the ones described above: (1) it requires the presence of a chromophore not available in mammalian cells, which therefore must be supplied exogenously or biosynthesized, and (2) once activated by red light, it does not passively return to an inactive state, rather it can be actively recovered through illumination with far red wavelengths. The latter has in many cases proven to be an advantage as it effectively creates a stable switch and allows even more precise control over the system's activation.⁶³ 

Another approach to engineering light sensitive systems originates from the plant stress response to ultraviolet light. Absorption of UV-B wavelengths allows the use of this system orthogonally to the blue and red light-inducible systems or fluorescent proteins without crosstalk. The system utilizes the unique properties of the UVR8 protein, which forms photolabile homodimers. Induction with UV-B light causes disruption of salt-bridges in the dimer structure and thus leads to formation of protein monomers.⁶⁴  The most interesting application of the UVR8 light sensitive system was the development of a light-triggered protein secretion system in mammalian cells in which the cargo was fused to several copies of UVR8. Association of these proteins into clusters prevented their transport in membrane vesicles, resulting in release of the protein only upon UV irradiation and dissociation of the clusters.⁶⁵ 

Although the ability to respond to light has evolved across kingdoms, all of the light responsive systems described here thus far were, not surprisingly, derived from plants. The choice of an evolutionarily distant source organism almost assures that the systems will be orthogonal to the mammalian chassis. However, a native mammalian receptor-based light-inducible system does also exist, involving chimeric GPCR proteins sensitive to light, named optoXR. Similarly to the GPCR-based systems described above, optoXR was designed to control receptor initiated signaling pathways. In the chimeric protein, intracellular loops of the light-sensitive GPCR rhodopsin were replaced with intracellular domains of other GPCRs, retaining the opsin activation and transduction of the light signal, but diversifying the output by the use of any one of the large family of GPCRs as an effector. Up to date, two optoXRs that selectively recruit distinct signaling pathways were characterized, one based on the human α1a-adrenergic receptor and the other based on the hamster β2-adrenergic receptor. Optical stimulation resulted in activation of adenylyl cyclase (production of cAMP) or phospholipase C (production of IP3 and DAG), respectively, leading to activation of downstream signaling.⁶⁶ 

While input signals form the initial stage for control of biologic circuits, the circuits themselves are defined by the interaction between the biological molecules that transduce and convert the signal. These molecules can range from DNA and RNA to proteins in either closed loop or open loop topologies. In other words, the interplay between the transcription, translation and protein interactions/modifications is what determines the behavior and function of a biological circuit. Input signals can be applied to control any of these levels. In the following sections we are going to review the most commonly used methods to control transcription, translation and protein modification in synthetic circuits, before highlighting some of the most interesting circuit topologies.

The highest number of synthetic biological circuits in prokaryotes as well as eukaryotes has been developed with transcriptional regulation as the principle underlying mechanism. Since proteins are usually introduced into foreign cells and organisms by means of genetic modification, control of their transcription provides an easily accessible way of circuit design. While the abundance of bacterial transcription factors provides a toolbox of regulatory elements, the development of designable targeted DNA binding domains, such as zinc fingers, transcription activation like effectors (TALEs) and the Cas9 protein of the bacterial clustered regularly interspaced short palindromic repeats (CRISPR) system provides a highly important advance. The design of targeted DNA binding domains provides proteins that have very similar properties yet can be designed to recognize almost any DNA sequence, which means that any endogenous sequence in the complex mammalian genome can be targeted. Zinc fingers were the first developed designable modular DNA binding domains. One zinc finger repeat recognizes and binds to a DNA base-pair triplet and longer sequences can be targeted by zinc finger fusions, called polydactyl zinc fingers. However, cooperativity and target-sequence overlap between tandem fusions of zinc finger domains present limitations to zinc finger modularity and complicate the design of arbitrary DNA sequence targeting domains, so several methods to simplify zinc finger synthesis were developed.⁶⁷  An important breakthrough occurred by the elucidation of the DNA recognition code by transcription activator like effectors (TALEs). These proteins were derived from plant pathogens⁶⁸  and have a more straightforward DNA-recognition code with each nucleotide recognition mapping to two residues in a 34-residue repeat.^69,70  Stringing of these repeats creates DNA binding modules recognizing typically 18 base pair long sequences. Fast TALE assembly platforms have been developed that enabled construction of large numbers of TALEs,^71–76  including TALEs targeting all human gene coding regions.⁷⁷ 

This highly efficient platform was however overshadowed by the CRISPR/Cas system where the recognition is based on the complementarity between the target site and a guide RNA (gRNA). CRISPR/Cas proteins therefore simplify the construction of large arrays of targeting designs by the fact that they do not need to be re-designed for each individual target. Providing several sgRNAs to a cell expressing Cas9 allows easy multiplexing of different targets.⁷⁸ 

The area of research most important to a mammalian circuit designer concerning these DNA binding domains are modifications that allow them to function as transcription factors in mammalian cells. First of all, the native Cas9 protein is a nuclease, so in order to function as an epigenetic effector rather than a genome editing tool, it had to be modified with two mutations knocking out its catalytic activity and resulting in the so called dCas9.⁷⁹  All of the designed DNA binding proteins can then be used as transcription repressors simply in virtue of sterically blocking transcription (a mechanism termed roadblock) if they are targeted to a region closely downstream of a promoter.^79–82  Alternatively, similarly to the bacterial transcription factors described above, they can be fused to activation and repression domains. The virally derived VP16 domain was improved by identification of the minimal activation region and then stringing of multiple copies of this region to result in more potent activation domains termed VP64⁸³  or VP192. This was further improved by the addition of the p65 domain derived from the human nuclear factor κB and another virally derived transcription activation domain Rta, resulting in a much more potent, but also significantly larger, tripartite activation domain VPR.⁸⁴  Tanenbaum et al. developed another interesting approach that avoids increasing the size of the protein. They fused only short epitope peptides to the Cas9 molecule and then supplied it with separately encoded VP64 fused to scFv antibodies binding to the epitope tag. In this way they were able to direct a large number of activation domains to the same DNA locus and increase transcription activation.⁸⁵  VP64 or other effector domains can also be recruited to the gRNA-Cas9 complex via RNA-binding proteins (RBPs).⁸⁶  The advantage of this method is that by selecting orthogonal RBPs with either activation or repression domains, and different copy numbers of RBP binding sites in each gRNA, one could selectively upregulate or downregulate a number of different genes in parallel with only one type of Cas9 molecule.⁸⁷  The KRAB domain, which is also abundantly present in endogenous human transcription factors and functions by directing chromatin remodeling, remains the most widely used repression domain in synthetic mammalian circuits. However, it has to be mentioned that the KRAB domain silences gene expression in a wide region spanning tens of kilobases around the target site.⁸⁸  Fusion of designed DNA binding domains with methylation and demethylation domains have also been used to influence gene expression by targeted epigenetic modification.^89–94 

Another important property that was introduced into the designed DNA binding domains is responsiveness to external stimuli. This was accomplished by the above described CIDs. Small molecule-inducible as well as light inducible zinc fingers,⁹⁵  TALEs^57,96,97  and Cas9^55,58,98,99  have been described on the basis of induced heterodimerization of the DBD with an activation or repression domain (Fig. 3A). Interestingly, both the TALE and the Cas9 DBDs themselves have been split and reconstructed, Cas9 inducibly with the FKBP-FRB system⁹⁸  and TALEs constitutively via inteins.¹³  Additionally, TALEs have been made inducible through intein- or FKBP-mediated circularization. In this case, the superhelical nature of the TALE interaction with DNA was exploited by locking the TALE into a circular conformation, thus preventing it from winding around the DNA (Fig. 3B). The ability to bind the DNA was restored and transcription of target genes upregulated only upon ligand- or protease-induced linearization of the TALE.¹⁰⁰  On the other hand, a mechanism of control that is unique to the Cas9 molecule is the dependence of the nuclease function on the length of gRNA. Kiani et al. have demonstrated that a catalytically active Cas9-VPR fusion can be used to either activate transcription from its target DNA when directed by a short gRNA or to cut the target and thus knock out target genes when directed by a longer gRNA.¹⁰¹ 

On the other hand, several members of the nuclear hormone receptor family function as native mammalian ligand dependent transcriptional regulators. Their mechanism of action involves activation through binding of the ligand (a hormone or its synthetic analog) and a subsequent intramolecular homodimerization, resulting in nuclear transport and the binding of target DNA to activate transcription¹⁰²  (Fig. 3C). Ligand binding domains (LBDs) of the nuclear receptors can be fused to selected DNA-binding domains, such as zinc fingers,⁹⁵  TALEs⁹⁷  and CRISPR/Cas9,¹⁰³  to create targeted ligand dependent transcription factors. The human nuclear receptors for estrogen (ER), progesterone (PR) and the retinoid X receptor (RXR), as well as the Drosophila ecdysone receptor (EcR) are most widely used to generate ligand dependent artificial transcriptional regulators.^95,97,103  Because their natural ligands may possess biological activity and stimulate unwanted gene networks, modified ER and PR LBDs were designed, that bind synthetic antagonists (4-OHT and RU486) which do not activate endogenous pathways.^104,105 

While many of the above mechanisms represent standalone sensors and simple devices, their true value for synthetic biology is realized only upon consideration of connectivity or composability. Assembly of synthetic circuits such as switches, oscillators or complex logic gates requires layering and feedback loops, imposing the requirement that the output of one information processing module serves as the input of another.

This is most readily realized with genetic circuits, where the modules are transcription units and the target of the first transcriptional regulation is itself a transcription factor controlling the next regulatory unit. Layering transcription repressors in this way leads to an interesting, albeit relatively simple system in which an odd number of repressors always leads to the final target repression while an even number of layered repressors leads to transcriptional activation of the final target (Fig. 4A–C). Due to the inherent level of noise in biological systems and imperfect on/off ratio such layering will eventually lead to a loss of signal, but a low number of layered repressors still offers some interesting circuit topologies. Most strikingly, one can arrange a target gene to be acted upon by two regulation cascades, one consisting of a single repression step and the other of a double inversion step, thus seemingly resulting in both activation and repression of the target gene upon stimulation. However, if these repressors are tuned such that the direct repressor has a lower affinity for binding, the two pathways will respond to different concentrations of inducers, resulting in a band-pass filter¹⁰⁶  (Fig. 4D).

Allowing the genetic outputs to be controlled by more than one input results in assembly of logic gates. Kramer et al. designed OR, NOR, AND, NAND, implication and nonimplication gates by combining ON (buffer gates) and OFF (NOT gates) systems with small molecule inputs.¹⁰⁷  Using only repressors, NOR gates can be constructed (Fig. 5A), and since these are functionally complete, by layering of just NOR gates all other Boolean operations can be constructed, for example NIMPLY gates (Fig. 5B) and OR gates (Fig. 5C). TALEs again represent a good transcription factor to use in such circuits, since (1) a large number of TALEs with high specificity can be designed, allowing construction of many layered or orthogonally parallel gates, (2) orthogonal TAL-KRAB effectors are likely to exert a similar degree of repression, resulting in high predictability of circuits, which is not necessarily true of different types of bacterial transcription factors and (3) TAL effectors are proteins which themselves can be controlled through transcriptional regulation, in contrast to the CRISPR-based repressors that require RNA for their targeting. Transcription of short RNA is performed by the RNA polymerase III, which requires the U6 promoter and is more challenging to regulate than mRNA transcription. For this reason, development of logic gates based on CRISPR lags behind other applications of CRISPR/Cas9. Methods to layer CRISPR-based genetic circuits include insertion of gRNA into introns of controlled target mRNA genes or design of Cas9-regulated polymerase III promoters, but both of these proved to be relatively noisy and no higher order gates were constructed.¹⁰⁸  An approach using ribozymes to express gRNA as a longer mRNA and then autocatalytically remove the unnecessary 5′ and 3′ sequences along with the protective caps, seems to offer more promising results, but has not yet been peer reviewed.¹⁰⁹  While the CRISPR system exceeds any other platforms with respect to the ease of construction of large libraries and multiplexing, TALEs demonstrated at least comparable activation and repression potency and are probably more efficient in construction of complex multilayered circuits.⁸¹ 

Arranging transcription repressors in such a way that they act upon each other in a closed loop results in even more complex behavior, such as amplification, switching or oscillation. Interestingly, it is again the number of nodes that determines this behavior—an odd number of nodes results in oscillation and an even number results in multiple steady states.^110,111  For the bistable switches it has been predicted¹¹²  and demonstrated¹¹³  that in addition to mutual repression, nonlinearity is required and is often provided in the form of cooperativity. Since most bacterial transcription factors bind to their target DNA as dimers in a cooperative manner, the fulfillment of this requirement is already built into the system and bistable mutual repression switches based on bacterial transcription factors in mammalian cells have been described¹¹⁴  (Fig. 6A). On the other hand, TALEs bind to DNA as monomers^115,116  and therefore provide no cooperativity, so in order to construct a bistable switch with designable transcription factors the addition of a feedback loop based on competitive binding of repressors and activators was required¹¹³  (Fig. 6B). Modularity of transcription factor design with VP16 and KRAB domains, but utilizing the same DBDs, greatly facilitated this effort. Another example of a genetic switch with a positive feedback was designed to be inducible in only one direction, thus providing cellular memory for encounters with external inputs or physiologically relevant signals, such as hypoxia or DNA damage¹¹⁷  (Fig. 6C).

Although any odd number of repressor arranged in a closed loop theoretically results in oscillation, a three node repressor based oscillator (repressilator, Fig. 6D) has only been demonstrated in bacteria⁹  and to date no higher order repressilators were constructed successfully. However, there are two other interesting oscillating circuit architectures described in mammalian cells. Swinburne et al. described a single-node architecture with TetR inhibiting expression of its own gene (Fig. 6E), where in order to increase the likelihood of oscillations additional RNA and protein destabilization were included, although it has not been shown that these features are indeed required for oscillation.¹¹⁸  Tigges et al. developed a two-node oscillator with TetR-VP16 (tTA) driving its own expression and the expression of another transcription activator (PIT) driving the expression of antisense mRNA to knock down tTA (Fig. 6F). They showed that oscillatory behavior depended crucially on gene copy number ratio, while the absolute quantity of the transfected genes influenced the amplitude and period of oscillations.¹¹⁹  They later explored a different approach using small interfering RNA (siRNA) instead of antisense mRNA, which resulted in a slower oscillator and relieved the strict requirement for fine balancing of gene copy number.¹²⁰ 

One level downstream of genetic circuits, RNA circuits retain the simplicity of sequence targeting design by complementary base pairing, but allow fine-tuning of circuits with additional types of control due to the ability of RNA to form catalytically active tertiary structures. The synthetic biology of RNA in general has been reviewed in the first volume of this book,¹²¹  so we will only focus on the mechanisms of RNA that are particularly useful to circuit design. These include aptamers, ribozymes, RNA binding proteins, and RNA interference.

Aptamers are the RNA ligand-response element, able to bind to external signals ranging from small molecules to large biochemical polymers. Aptamers can be designed by the evolutionary method SELEX¹²²  to bind to selected targets (for a recent review see Darmostuk et al.¹²³ ), but the most well characterized and widely used aptamer remains the theophylline binding aptamer¹²⁴  due to its high affinity, selectivity and robust performance in live mammalian cells. The other key property of aptamers is the fact that they can be incorporated into other RNA molecules, making these responsive to external stimuli. For example a secondary structure rearrangement upon ligand binding can cause a transcription start site in mRNA to become obstructed or interfere with ribosomal scanning and thus prevent translation.¹²⁵  Interestingly, translational activation is not usually achieved through ligand binding to 5′ UTR aptamers but instead relies on ribozymes.

Ribozymes in the context of RNA circuit design refer to short RNA sequences with the ability to autocatalytically excise themselves from a longer RNA molecule. They can be used to generate short RNAs from longer transcription cassettes, for example to facilitate mRNA degradation by removing the 5′ cap or the 3′ poly(A) sequence from the transcribed mRNA. Importantly, ribozymes can cut either to their 5′ or their 3′ terminus. These functions are performed for example by the hepatitis delta virus ribozyme and the hammerhead ribozymes, respectively, and the use of both types of ribozymes enables the removal of both end modifications.¹⁰⁹  Ribozymes achieve even more features when aptamers are included in their sequence, making them ligand-dependent. In this way, translation of target proteins can be inhibited by ligand-induced ribozyme (aptazyme) cleavage^126,127  (Fig. 7A) or initiated by ligand-induced aptazyme deactivation¹²⁸  (Fig. 7B).

Just like transcription factors bind DNA, RNA-binding proteins (RBPs) can bind RNA in a sequence specific manner. Most often, the bacteriophage MS2 coat protein^129,130  and the archaeal L7Ae protein^131,132  are used to bind target RNA, even though their target sequence cannot be selectively designed. These two proteins have been used to inhibit translation when bound in the 5′ UTR¹³³  as well as to control mRNA splicing¹⁹  (Fig. 7C). A strategy to convert the RBP OFF-switch into an ON-switch also exists and is based on the inclusion of a 5′ RBP binding site, a premature stop codon and an internal ribosome entry site (IRES) before the coding sequence for the protein of interest. In this way the mRNA is degraded through nonsense-mediated mRNA decay in the absence of an RBP, but in its presence, the stop codons are obstructed and the protein in translated from the IRES¹³⁴  (Fig. 7D).

TetR can also be used to bind to RNA, introducing ligand responsiveness and allowing interesting interplay between the DNA and RNA levels.¹³⁵  Parallel to the modular structure and designable targeting of TALEs to DNA, designable targeting of RNA can be achieved with the use of PUF proteins, which are also composed of modular repeats, each binding to a specific RNA base in a predictable manner,^136–138  and can be fused to repression and activation domains to either downregulate or upregulate protein expression.¹³⁹  Additionally, PUFs can be used as designable splicing factors¹⁴⁰  and even as designable RNA-endonucleases to parallel TAL endonucleases.¹⁴¹ 

Finally, the naturally prominent method of RNA regulation mechanism, RNA interference, can also be adapted to synthetic biologic circuits. Antisense RNA strategies have largely been replaced with short RNA interference, but with the caveat that the latter relies on the DICER processing and the RISC complex, only present in eukaryotes.¹⁴²  Endogenous microRNAs (miRNAs) can be used for circuit regulation, resulting in the recognition of the host cell type, for example for identification and selective targeting of cancer cells.¹⁶  Artificial miRNAs and short hairpin RNAs (shRNAs) can be used as synthetic circuit inputs, especially when combined with other strategies, such as aptamers and aptazymes, to make them ligand responsive.^{17,18,143,144} 

Significant improvements of genetic circuits can be obtained through the transcription and translation regulation interplay. Using translation regulation of reporters with RNA-binding proteins which were themselves under the transcriptional regulation of inducible transcription factors, Ausländer et al. designed NIMPLY logic gates, where one small-molecule input signal was used to repress transcription of a reporter with an MS2 or L7Ae binding site and another input signal was used to repress transcription of the RBP (Fig. 8A). In the presence of only one input, the reporter was transcribed and translated, but if both inputs were absent, the transcription of RBP prevented translation of the reporter, while the presence of both inputs repressed the reporter directly. Wiring two NIMPLY gates such that each input activated transcription of the reporter in one gate and of the RBP in the other gate, resulted in an XOR gate, which allowed for the assembly of a half-adder and a half-subtractor.¹⁴⁵ 

A seminal work by Deans et al. demonstrated that tight, tunable and reversible control of transgene expression can be obtained through coupling of two transcription repressors (TetR and LacI) with shRNA interference. A circuit was designed in which the gene of interest was repressed by LacI and at the same time protein expression from any leaky transcription was knocked down by shRNA. The LacI repressor also controlled the expression of a TetR gene, which, when transcribed, inhibited transcription of shRNA (Fig. 8B). In this way induction with IPTG relieved both the repression and the knockdown and allowed efficient expression of the protein of interest, while uninduced cells produced no detectable quantities of target protein, making this system safe for expression of even highly toxic proteins.¹⁴⁶  A similarly safe system was designed by Greber et al., who cloned a tagged gene of interest under a promoter controlled by a small molecule inducible transcription factor. An intronic siRNA against the tag was inserted into the gene coding for the transcription factor. In this way, the translation of the protein of interest was constitutively knocked down, which resulted in tighter regulation, although at the cost of reduced maximum expression. Importantly, a toggle switch with mutual repression supplemented with the siRNA control mechanism exhibited a much higher dynamic range than without siRNA.¹⁴⁷ 

Different cell types and particularly cancer cells have different expression profiles of endogenous miRNA, but to differentiate between them with high certainty, it is important to not only sense a single miRNA with high expression but also those that are absent in a particular cell type. As miRNAs act as repressors of translation by facilitating mRNA degradation, they can be considered inverters (NOT gates). Combining a single and double inversion circuit with miRNA inputs, Xie et al. designed a cell sorter. In this circuit the miRNAs expressed at low levels in target cells were designed to inhibit translation of the reporter gene directly and miRNAs expressed at high levels inhibited expression of a LacI repressor, which in turn also controlled the gene of interest (Fig. 9A). Interestingly, miRNAs in the single inversion step can be designed to act on the same transcript (NOR gate), while the double repression miRNAs must act each on its own LacI transcript to result in an AND gate (Fig. 9B), otherwise an OR gate would be obtained (Fig. 9C). The system was used to design a complex gate integrating information from a large number of endogenous mRNA (Fig. 9D) and further improved by placing LacI under the control of an activator, whose translation was repressed by the same highly expressed miRNAs. Additionally, an exogenous miRNA was encoded along with LacI to enhance repression. Replacing the reporter gene with a suicide gene, resulting in selective destruction of a selected cell type, has a high therapeutic potential.¹⁶ 

Strikingly, the existence of self-replicating RNA molecules and subgenomic promoters that allow transcription of RNAs from RNAs, can even make RNA circuits completely independent of the genetic encoding at the level of DNA. Synthetic logic circuits, sensors and switches can thus be operated entirely by RNA and proteins encoded in it. Wroblewska et al. re-designed the cell classifier by encoding it on an alphaviral RNA replicon and replacing the LacI repressor with an L7Ae protein binding in the 5′ untranslated region. They also designed a double and triple inversion circuit and a mutual repressor switch using the MS2-CNOT7 fusion protein, which also acts as an inhibitor of translation when bound to the 3′ untranslated region of mRNA.¹⁴⁸  These types of circuits are safer in comparison to DNA encoded circuits, because they pose no risk of genomic integration.

While most engineered genetic circuits are based on transcriptional and translational regulation, there are advantages to regulation on the protein level. Most importantly, regulation of proteins through transcription or translation creates a delay, but some natural circuits are able to respond faster by using signaling pathways based on protein interactions or their modifications. In some applications, fast response is absolutely crucial, especially for therapeutic purposes e.g. release of hormones, such as insulin, or neurotransmitters, where the response is required within minutes or seconds.

Fast signal processing in cells is mainly performed by the phosphorylation/dephosphorylation cascades catalyzed by protein kinases and phosphatases. Phosphorylation/dephosphorylation cascades are rapid and reversible, however design of selective small molecules for pharmacological perturbation of native protein kinases or phosphatases still remains a challenge. One option for design of synthetic phosphoregulation pathways is through small molecule-controlled kinases and phosphatases.¹⁴⁹  Camacho-Soto et al. prepared split tyrosine-kinases (TK)⁴⁹  and split tyrosine-phosphatases (TP)⁴⁸  as two functionally inactive fragments appended to CIDs (rapamycin, abscisic acid and gibberellin inducible dimerization domains). This allowed their control with external inputs, but did not enable assembly of a complete signaling pathway orthogonal to endogenous signaling. In fact, few attempts at design of fully synthetic phosphorylation cascades have been made as kinase specificity selective enough to maintain complex orthogonal signaling pathways has proven difficult to engineer. Ryu and Park took advantage of the three-tiered mitogen-activated protein kinase cascade (MAPK) to design modular protein-protein interaction signaling pathways in mammalian cells.¹⁵⁰  In MAPK signaling three consecutive kinases act upon one another and the flow of information is often directed by a scaffold protein, but functional domains of the scaffold often overlap.¹⁵¹  In the engineered MAPK system, the kinases of the yeast mating pathway were transferred to a mammalian chassis and the native scaffold protein Ste5 was replaced with well characterized and modular protein interaction domains, such as PDZ and MTD.¹⁵⁰  This represents a step toward rewiring of MAPK pathways, but the complexity of the MAPK interactions has not yet been understood enough to allow generation of completely orthogonal circuits with designed inputs and outputs.

Phosphoregulation has been used to design some interesting biologic devices in combination with transcriptional regulation. If multiple targets for a transcription factor are present, binding to any of the targets will sequester the transcription factor and reduce its availability to bind to others. This creates a load that interferes with the circuit operation. Even though synthetic biology aims to create modular designs in which each module acts predictably upon modules downstream of it, such retroactive loads disrupt this predictability and complicate circuit design. Mishra et al. addressed this retroactivity by combing transcriptional control with phosphoregulation cascades. A load driver was engineered in which a quasi-steady state is established faster than in the transcription regulation system, thus allowing the device to respond more predictably to input changes.¹⁵² 

In comparison to phosphorylation/dephosphorylation, proteolysis is also fast but irreversible. However, to our benefit, the principles of protease target recognition and cleavage are well understood and orthogonal protease-proteolytic target pairs are easier to engineer. The tobacco etch virus protease (TEV) is the best characterized and most widely used protease in synthetic biology. A split version of TEV protease with inducible reconstitution was engineered and offers an attractive method to regulate the protease activity.¹⁵³  In addition to the widely used TEV protease, other proteases with orthogonal cleavage sites^154–157  could be used to provide a tool for modulating fast multiple signaling processes in mammalian cells.

Proteases have been used for selected target protein degradation, for example by exposing cryptic degrons or, conversely, can induce target binding or catalytic activity by removal of inhibition domains.¹⁵⁶  Stein and Alexandrov developed an autoinhibited hepatitis C virus protease, which could be activated by the removal of the autoinhibition peptide by the TVMV protease,¹⁵⁷  while Shekhawat et al. developed autoinhibited coild-coil pairs, in principle useful for induction of dimerization through proteolytic removal of the inhibitory peptide, and demonstrated its use on the in vitro example of a split luciferase reporter.¹⁵⁸ 

On the other hand, conditional protein depletion can be achieved through the use of N-terminal degrons. These protein domains induce protein degradation based on the N-end rule through recognition of the N-terminal degron residues by ubiquitin ligase which mediates polyubiquitination, marking the protein for proteasomal degradation. The earliest described N-terminal degron was a peptide derived from the yeast temperature sensitive dihydrofolate reductase (DHFR).¹⁵⁹  In synthetic circuits, degrons have been designed to be regulated for example by proteolytic cleavage and a genetic circuit using potyviral proteases to control degrons was implemented in bacteria, but not in mammalian cells.¹⁵⁶  A small-molecule inducible system for rapid protein depletion was developed based on the plant auxin dependent degradation pathway. Auxins bind to the TIR1 protein and induce interaction between TIR1 and the auxin inducible degron (AID). TIR1 recruits the multi-protein E3 ubiquitin ligase complex SCF, therefore interaction between TIR1 and the AID degron leads to ubiquitination and protein depletion. Because the TIR1-SCF interaction is based on highly conserved protein domains but TIR1 is only present in plants, the auxin-inducible degron (AID) enables orthogonal, reversible and tunable protein degradation in a range of eukaryotic cells, from yeast to mammals.¹⁶⁰ 

While RNAs can be independent of genetic encoding and posttranslational regulation is fast and often reversible, recombinases enable a completely different approach to biologic circuit design as they result in permanent genetic modifications. By their mechanism of action, recombinases are divided into two groups; tyrosine mediated site-specific recombinases, which can mediate gene expression by excision of a transcription terminator,¹⁶¹  and serine mediated site-specific integrases, which can influence gene expression by inversion of the gene of interest.¹⁶²  Tyrosine recombinases Cre/loxP and Flp/FRT, and serine recombinases including PhiC31 and Bxb1 integrases are fully orthogonal. Additionally, inducible split version of both Flp and Cre recombinases are available.

Recombination-based cellular memory and counters have first been developed in bacteria, but here we would like to point out some unintuitive applications in mammalian cells. Lapique et al. used recombinases to create a time delay circuit. They realized that upon transfection of an inducible target gene along with its repressor for an inverter or double inverter gate, a pulse of high expression will occur before sufficient repressor accumulates to inhibit transcription of the target gene. Only after that pulse does the expression of target gene become predictably inducible by depression. To avoid this pulse, the target gene was placed between recombination sites in a reverse orientation relative to its promoter and cotransfected with both a repressor and a recombinase. In the time that the recombinase is expressed and inverts the target gene, enough repressor accumulates to prevent the intial pulse. They showed that such a circuit can be used to safely express a toxic gene in response to specific cell type inputs, such as miRNA, thus improving the cell-classification circuit described previously by the same group.¹⁶³ 

Site-specific recombinases can act as activators or repressors of transcription within the same cell, integrating signals on a single transcriptional layer. However, implementation of complex functions requires design of heterospecific recognition sites, which are orthogonal in respect to each other, but can be processed by the same recombinase. The most complex circuit utilizing orthogonal recombinases and heterospecific recombination sites to date is a logic circuit designed by Weinberg et al. (Fig. 10). While logic gates are usually designed ad hoc with multiple promoters or copies of output gene as required, they decided on a more general approach in which a number of genetic “addresses” are designed, corresponding to each possible combination of input signals (four addresses for a two-input system, eight for a three input system and so on). Each address can code for zero, one or more output genes and which address gets transcribed is determined by the action of recombinases, inducible by input signals. This ensures that the basic coding architecture for all logic gates will be the same, but does not yet represent a general system, as the outputs inside the addresses again need to be recoded ad hoc for each specific logic gate. They therefore took it one step further and encoded a transcriptionally inactive output into each address, but each flanked by a pair of recombination site orthogonal to all other recombination sites in the system. In this way it was demonstrated that all possible two-input Boolean logic gates can be encoded by the same genetic construct, from which one can select any one gate by the action of four recombinases acting on the addresses to repair the output gene at that address. The final output is then produced by the two inputs inducing another two orthogonal recombinases selecting which address will get transcribed. In its final implementation this system thus requires a large number of orthogonal recombinases, some of them with the requirement for heterospecific target sites, but the existence of such an orthogonal set has also been demonstrated.¹⁶⁴ 

Engineered synthetic circuits in mammalian cells provide powerful approaches to sense and control levels of crucial disease-relevant metabolites and they offer promising strategies for cell-based therapeutic applications as the additional level of control provides higher accuracy, efficiency as well as safety. In this final section, we will review some of the proof of concept studies of treatment of human diseases with biologic networks.

Gene therapy based on engineering cells from patients has been the topic of research for more than a decade, however the simple introduction of a selected gene into engineered cells has been surpassed by the design of autonomous therapeutic devices able to sense a combination of different signals, process this information and produce the desired therapeutic effectors (Fig. 11). In addition to engineered patient's cells, encapsulated (e.g. in alginate beads) universal therapeutic cells have been tested in several different animal studies to regulate hormones in diabetic patients, treat obesity,²²  metabolic syndrome,¹⁶⁵  and hypertension²⁴  as well as diseases such as psoriasis,²⁵  gout²³  and colitis.²⁶ 

Unless treated in time, chronically deregulated blood glucose levels linked to a complex and progressive metabolic disorder diabetes mellitus, can initiate downstream pathologic cascades. Since glucose levels in the body are in constant flux, a designer sensor-effector cellular device for glycemic control and insulin delivery represents an important synthetic biology state of the art strategy for treatment of diabetes. Extrapancreatic cells were reprogrammed to sense physiological blood glucose levels by mimicking the β-cell glucose sensing cascade through glycolysis mediated calcium entry with voltage-gated calcium channels (CaV1.3). The calcium channel CaV1.3 enables β-cells as well as engineered cells to profile physiological blood glucose levels in a precise and reversible manner. Ca⁺² influx as a result of opening of CaV1.3 channels due to nonphysiological levels of glucose, promotes intracellular calmodulin-calcineurin signaling and nuclear translocation of calcium responsive transcription factors to induce gene expression, resulting in insulin secretion. Coupling of CaV1.3-based glucose sensing to insulin production and secretion resulted in β-cell mimetic designer cells that provide closed-loop glycemic control.¹⁶⁶ 

Another interesting cellular device employing therapeutic gene circuits was developed for sensing and suppressing inflammation. A complex biological response involving the innate immune system is usually elicited by pathogens, injury or damaged tissue and serves to eliminate pathogens and restore homeostasis. However, aberrant inflammatory signaling is harmful for the organism and may lead to chronic diseases, such as rheumatoid arthritis and multiple sclerosis, due to amplified signals through positive feedback loops.¹⁶⁷  Fast and precise suppression of inflammation is therefore a key challenge for successful therapy of inflammatory diseases. Smole et al. developed an anti-inflammatory cell device based on engineered encapsulated mammalian cells able to detect inflammatory mediators at physiologically relevant levels and suppress inflammation by production of anti-inflammatory proteins. Furthermore, to enable a fine-tuned and precise response, the device additionally contains a signal amplifier with a positive feedback thresholder (Fig. 11A). This module is designed so that activation of the sensor results in transcription of a positively autoregulated transcription factor, but when the signal is low, a competitive DNA-binding domain prevents activation of the amplifier. To prevent long term systemic inhibition of beneficial inflammatory signaling an externally inducible switch off system is included and represents a major improvement of this therapeutic cellular device compared to already established gene therapy.²⁶ 

It is worth noting that synthetic biologic sense-and-respond circuits have not only been proposed for use in therapies, but also for veterinary practice to facilitate livestock breeding. Artificial insemination is standard practice in kettle breeding, however reproduction in mammals is narrowly time-regulated through hormone controlled female ovulation, and sperm delivery must match this fertility window.¹⁶⁸  Ovulation in mammals is triggered with the release of luteinizing hormone (LH) from the pituitary gland and its binding to the LH receptor activates GPCR signaling, resulting in release of an oocyte. To address the challenges of sperm delivery timing in cattle artificial insemination, a synthetic biology device which enables coordination of ovulation and sperm delivery was engineered and tailored to dairy cows. The device is composed of sensor cells and spermatozoa encapsulated in cellulose sulphate capsules and implemented into a cow's uterus. Sensor cells expressing LHR monitor levels of LH and upon detection of oestrus produce cellulase, which degrades the capsules and enables release of sperm. The artificial insemination device was validated in cows, however fine tuning of the device could facilitate its use in other species.¹⁶⁹ 

Probably the greatest success of therapeutic application of synthetic biology is represented by the CAR T-cell-based cancer immunotherapy. Recently, clinical implementation of two important advances in cancer immunotherapy were reported: engineered T-cell receptors (TCRs)¹⁷⁰  and T-cells with chimeric antigen receptors (CARs).¹⁷¹  In their latest generation, CARs consist of an engineered extracellular recognition domain, which is most often a single chain antibody against a selected cancer antigen, the cytoplasmic activation domain CD3ζ and costimulatory domains like CD28 and 4-1BB. The activation and costimulatory domains trigger T-cell activation and expansion that enables retargeting of T-cells towards cancer elimination.^172–174  Despite broad applicability, however, CAR T-cell therapy displays several drawbacks, including potential severe side effects and a limited number of absolutely tumor specific antigens. Several additional synthetic biology tools have thus been recently developed to precisely regulate CAR T-cell proliferation to prevent side effects such as cytokine storms.¹⁷¹  Small molecule control¹⁷⁵  and engineering of dual-antigen sensing to create AND logic gates¹⁷⁶  with improved specificity (Fig. 11B), as well as kill switches¹⁷⁷  to deplete the therapeutic cells when they are no longer needed, were implemented.

In currently on-going clinical trials of CAR T-cell cancer immunotherapy, targeting of cancer cells is based on the cancer specific antigen CD19, which is present in both healthy and malignant B cells.¹⁷⁸  To bypass the problem of cross-reactivity to healthy cells, a strategy with dual-antigen sensing using two independent CARs in the same cell was developed. A CAR specific for the B-cell marker CD19 is used to mediate CD3ζ activation, while a CAR specific for the prostate stem cell antigen (PSCA) is used for costimulation through the CD28 and 4-1BB signaling domains. The implementation of an AND gate should enhance ON-target activity by binding of both CARs at the same time to nearby target cells and result in specific activation of T-cells. However, the anti-CD19 CAR was able to activate T-cells without costimulatory signals from the second PSCA-specific CAR. By minimization of the T-cell activation through introduction of CARs with diminished activity and by switching from the combination of CD19 and PSCA antigens to PSCA and PSMA, more specific T-cell activation was achieved.¹⁷⁹ 

An even more robust dual-receptor AND gate was created using CARs in combination with synthetic Notch receptor signaling (synNotch). The extracellular domain of synNotch was engineered to target CD19, but unlike with CARs, binding does not trigger T-cell activation. Upon antigen binding, the intracellular part of the synNotch receptor is cleaved, which results in release of a transcription factor that drives inducible expression of a CAR specific for the second antigen. The synNotch-gated CAR expression displayed a highly specific response to multiple antigens and precise therapeutic discrimination in vivo. A high level of modularity of synNotch receptor components provides flexibility in engineering cells with customized sensing and response. Thus, the synNotch-CAR dual receptor facilitates immune recognition of a wider range of tumors.¹⁷⁶ 

Additionally, the safety of cell-based therapy, which includes stem cells and cancer immuno-therapy, is an important issue. Several kill switches have been developed to terminate the introduced cells upon application of a signal (e.g. doxycycline) in order to prevent excessive activation of engineered T-cells or development of cancer from stem cells.¹⁸⁰  However, many of the proposed solutions are quite sensitive to escape mutations, which can inactivate the kill switch, and such mutated cells would preferentially survive and multiply in the organism as a result of selective pressure. Safety switches that bypass this mutation sensitivity based on incorporation of unnatural aminoacids have been implemented in bacterial cells^181,182  and it is likely that a similar approach could function in mammalian cells and in in vivo applications in patients.

Medical applications represent some of the most exciting areas of the development of synthetic biology. A wide array of different tools to regulate cellular response and process combinations of inputs have been developed and their function tested in mammalian cells and often also in animal models. Most of the tools that work in bacteria could also be implemented in mammalian cells. The most important remaining challenge of tool development in mammalian cell biology is the construction of fast response circuits with response times in seconds to minutes that are required in many medically relevant conditions. It is also worth noting, that apart from cancer immunotherapy, no other therapeutic applications of synthetic biology have yet been used in the clinic. Based on the available tools, robust and precise regulation could soon be translated into the clinical trials and practice.

While cell therapy represents a therapeutic platform different from the existing pharmacological pillars composed of small molecules and biological drugs, major pharmacological companies are actively exploring and developing cell-based therapy. However, personalized cell therapy is bound to remain expensive due to the costs of individualized cell culturing. Introduction of encapsulated universal cells represents a tempting alternative as it could provide a more cost effective and well validated therapeutic approach, but novel delivery agents need to be developed. It is likely that we will see clinical applications of synthetic biology within the next decade.

This work was supported by funding from the Slovenian Research Agency grant number P4-0176, J3-6791 and J1-6740.