跳到正文
原文
Nature 健康护理主题·· 3 天前精选AI 评分66

工程化信号转导驱动闭环基因与细胞治疗:从设定点医学到临床转化

Engineering signal transduction for closed-loop gene and cell therapies

AI 导读

本文发表于Nature Reviews期刊,系统综述了闭环基因与细胞治疗的设计理念与应用进展。文章以传感器-处理器-效应器三层架构为基础,概述了代谢性疾病、炎症、感染、心血管疾病及肿瘤等领域中的代表性闭环系统设计,并讨论了AI辅助基因回路编程、多组学数字孪生与监督控制策略如何推动该类疗法向临床转化。

推荐理由

原文系统梳理了闭环基因与细胞治疗的核心架构及在多类疾病中的代表案例,可为同类研究的路线设计提供参考框架。

正文

Introduction

The conceptual roots of self-regulating therapy reach back to the earliest examples of gene control. The lac operon established the principle that DNA can encode conditional responses to molecular inputs, and the first synthetic gene circuits showed that regulation, oscillation and logic can be designed rather than merely observed.1,2,3 As cell and gene therapies began to mature, this logic naturally migrated toward medicine: if DNA can be programmed, then living cells can, in principle, be converted into therapeutic devices that sense, compute and act within the body.4

Synthetic gene circuits can be viewed as genetic software that programs how living systems sense inputs, process information and execute outputs, whereas mammalian therapeutic cells serve as the biological hardware that carries and runs this program.5,6,7 Because this hardware can take several forms, including mammalian designer cells, immune cells, stem cells and tissue-engineered cell systems, the same engineering logic can be adapted to different therapeutic settings.8,9,10 Disease intervention therefore becomes a problem of programming living systems according to pathophysiological context: metabolites, cytokines, hormones, hypoxia or physical stimuli can be linked to synthetic receptors, signal-processing modules and therapeutic effectors to create sophisticated living medicines that integrate sensing, decision-making and treatment within a single platform.11,12,13,14,15

Yet most therapeutic interventions still operate in an open-loop manner. Whether the modality is a pill, biologic agent, vector or engineered cell, the dosing schedule is usually determined empirically and then applied according to a fixed plan. This approach is effective in many settings, but is poorly matched to diseases whose critical biomarkers fluctuate over minutes, hours, circadian cycles or longer phases of progression. In these contexts, fixed dosing risks under-treatment during exacerbations, over-treatment during remission and cumulative toxicity when delivery cannot be titrated to physiological need.16,17

Synthetic biology offers a different design philosophy. Rather than treating signal transduction as background cellular noise, it treats endogenous signalling as a programmable substrate. Disease-associated inputs can be detected by natural or engineered receptors, routed through native or synthetic pathways, and translated into therapeutic outputs using modular genetic circuits.6,7,18,19,20,21,22 In its most advanced form, the therapy itself becomes a controller: it does not merely deliver a payload but continually measures deviations from a desired physiological state and adjusts its behaviour in response. This feedback-driven architecture embodies the principle of closed-loop control, whereby biological or engineered systems continuously sense physiological changes and autonomously modulate therapeutic outputs in response to deviations from a desired physiological state.8,11,12,13,23,24,25 Building on this foundation, setpoint medicine aims to maintain specific physiological variables within predefined healthy ranges (setpoints). If software-hardware integration provides the engineering framework, setpoint control defines the therapeutic objective: not merely to administer a therapeutic agent, but to continuously restore and maintain physiology within clinically meaningful target ranges. A defining trajectory in the field has been the progression from user-actuated optogenetic and electrogenetic control devices to biomarker-responsive designer-cell implants.26,27 As this progression implies, therapeutic autonomy did not emerge all at once. This capability emerged from sustained advances in clinically compatible input modalities, including metabolite-, drug-, optical- and radio-wave-responsive control systems,6,15,28,29,30,31,32 tunable promoters and scalable transcriptional platforms,5,33,34,35 engineered receptors and logic-gated immune-cell circuits,7,36,37,38,39,40 hybrid bioelectronic interfaces27,41,42,43 and implantable therapeutic cell chassis.13,44,45,46 It also reflects a broader maturation of the software-hardware paradigm in synthetic biology: as the control logic encoded in gene circuits became more sophisticated, the range of compatible mammalian therapeutic hardware expanded from cultured cells to implantable designer cells, engineered immune cells and tissue-associated constructs. Seen in this light, programmable signal transduction is not simply a collection of enabling tools, but the foundation of a new treatment logic that is moving from proof-of-concept toward targeted therapy.

In this review, we place that trajectory in a broader field-wide context, covering major advances that have conceptually expanded closed-loop control or that have moved closest to translational practice. We first discuss the shift from episodic treatment to setpoint medicine, then examine the sensor-processor-effector architecture that underlies therapeutic gene circuits, and finally survey representative applications in metabolism, inflammation, infection, cardiovascular homeostasis and cancer. This Review focuses primarily on mammalian closed-loop and context-responsive therapeutic systems, which currently provide the clearest translational archetypes for setpoint medicine. We define closed-loop therapeutic systems as those that incorporate measurable feedback signals to dynamically adjust therapeutic output and thereby maintain or restore a physiological variable around a defined setpoint. By contrast, context-responsive therapeutic systems, including feedforward, inducible and logic-gated architectures, operate primarily through predefined decision rules or context-dependent activation without continuous feedback-mediated adjustment of therapeutic output. Throughout, we emphasize a new therapeutic grammar in which continuous diagnosis, adaptive treatment and prevention become parts of the same system and, in some chronic indications, may ultimately support functional cure through durable restoration of homeostasis (Fig. 1).

Fig. 1

From episodic treatment to setpoint medicine. Conventional care separates diagnosis, treatment and prevention in time: biomarkers are measured intermittently, therapy is delivered by externally scheduled dosing and prevention relies on retrospective rules. Closed-loop gene and cell therapies fuse these functions within a single control system organized around an embedded sensor-processor-effector loop and a physiological setpoint. Embedded sensors provide continuous molecular diagnosis; processor modules convert that information into regulatory decisions; effector modules deliver adaptive therapy; and feedback suppresses excursions before they escalate into overt disease. If control is sufficiently durable to restore and maintain the setpoint, some chronic indications may begin to move from chronic management toward durable disease correction and functional cure. This figure was created using Adobe Illustrator

From episodic treatment to setpoint medicine

Closed-loop therapeutics reshape medicine by integrating diagnosis, treatment and prevention into a continuous adaptive process.21,23,47 In conventional care, diagnosis, treatment and prevention are separated in time: biomarkers are measured intermittently, doses are prescribed externally, and prevention depends largely on rules derived from past observations. In a closed-loop therapeutic, these functions converge. Sensing provides continuous molecular diagnosis, computation converts that information into adaptive treatment, and early correction suppresses excursions before they become severe, making control inherently preventive. If such therapeutic programmes remain sufficiently durable to maintain homeostasis without repeated rescue interventions, some chronic disorders may be susceptible to functional cure rather than indefinite symptomatic management.28,48,49,50

The first generation of mammalian therapeutic circuits addressed a more fundamental challenge: how to make transgene expression controllable in time and space rather than constitutive. This produced a rich toolbox of systems actuated by antibiotics, food additives, plant metabolites, cosmetics, inhaled molecules and physical cues.51,52,53,54,55,56 A decisive early example was the demonstration that optogenetic control could be coupled to metabolic therapy, enabling mammalian designer cells to be remotely instructed with high spatiotemporal precision to improve glucose homeostasis in vivo.15

Subsequent studies expanded this control repertoire through smartphone-guided far-red optogenetics, electrogenetic insulin release, smartwatch-programmed light delivery and progressively more sensitive optical systems.27,42,57,58 Related work extended the same logic to local cancer immunotherapy, sonogenetic tumour control and other physically actuated settings.59 Together, these platforms overcame problems of timing, reversibility and localization, and made it possible to test therapeutic constructs with unprecedented precision.

Open-loop systems, however, still leave the crucial decision outside the therapy. The patient, clinician or wearable device remains responsible for determining when control should be engaged.43,60,61 Recent user-friendly examples, such as nitroglycerin-responsive transdermal control and near-infrared-light-inducible intratumoural systems, show how far this paradigm can be refined.60,61 Even so, they also illustrate why autonomous control is fundamentally different rather than merely more convenient: in a truly closed-loop system, the therapy senses the disease state itself and computes the dose internally.

Autonomous control therefore emerged as a maturation of earlier gene-switch engineering rather than a rupture from it. The same advances that improved receptor design, promoter tuning, bioelectronic interfacing and implantable cell formats are now enabling biomarker-responsive circuits. The field is best understood not as a sudden leap to autonomy, but as a progression from remote instruction to local setpoint control.

Architectures that turn signal transduction into setpoint control

Most therapeutic gene circuits can be understood through a sensor-processor-effector framework.18,19,40,44,62 The sensor detects a molecular or physical input, the processor converts that input into a regulatory decision, and the effector executes the therapeutic response. Within the software-hardware paradigm, these layers define the therapeutic program, whereas the biological chassis determines where and how that program can operate in vivo.63 In closed-loop systems, the output also feeds back on the input or its downstream consequences, thereby stabilizing the system around a desired setpoint. This decomposition is not only conceptually tidy, but also practically useful, because each layer can be tuned or exchanged without redesigning the entire system.64,65

The sensor layer

The sensor technologies have expanded into an increasingly diverse repertoire of input modalities. Natural receptors such as G protein-coupled receptors (GPCRs), cytokine receptors and ion channels are attractive because evolution has already matched them to physiological concentration ranges.19,28 Synthetic receptors extend this repertoire by enabling the detection of antigens, soluble proteins or custom ligands not natively wired to therapeutic outputs.66,67 The generalized extracellular molecule sensor (GEMS) platform and later DARPin-based receptors showed that extracellular recognition can be modularized and coupled to downstream transcriptional programmes, a capability that is particularly relevant for cancer and immunotherapy.68,69 In parallel, intracellular sensors and dimerization systems enable circuits to respond to metabolites, hormones or protein-protein interactions within the cell.70

The processor layer

The processor layer integrates and processes sensory inputs to generate context-dependent therapeutic decisions. Many of the most successful mammalian circuits repurpose endogenous pathways, including cAMP/PKA/CRE, Ca2+/NFAT, JAK/STAT, SREBP/SRE and NF-κB. The strength of this strategy is that a disease-relevant input can often be coupled to a robust, well-characterized transcriptional response without having to build an entirely synthetic signalling cascade. A broad body of work has shown how endogenous pathways can be rewired through custom promoters, hybrid transcription factors and operator architectures to create robust sense-response devices.8,19,28,48,71,72 Although crosstalk is an obvious concern, earlier work suggested that overexpressed heterologous receptors can effectively monopolize their targeted signalling route, thereby improving functional insulation in defined contexts.73

The effector layer

The effector layer translates circuit decisions into therapeutic actions, including protein replacement, cytokine antagonism, metabolite clearance and targeted cytotoxicity. In feedback-controlled systems, therapeutic protein expression is dynamically adjusted in response to physiological signals. For example, insulin or GLP-1 production can be induced during hyperglycaemia and subsequently reduced as glucose concentrations return to the target range, thereby restoring glycaemic homeostasis while limiting unnecessary therapeutic output. Similarly, inflammation-responsive circuits can secrete antagonists such as soluble TNF receptor 1 or IL-1 receptor antagonist when inflammatory signals rise and attenuate their production as inflammation resolves, enabling adaptive restoration of immune homeostasis. Other closed-loop strategies act through the direct elimination of pathogenic metabolites. In cholesterol-, phenylalanine- and angiotensin-responsive shunt circuits, metabolic flux increases as substrate concentrations rise and decreases once they return to the physiological range, thereby enhancing clearance while minimizing unnecessary metabolic burden.71,72,74 In cancer therapy, the effector output may instead be targeted cytotoxicity. Here, activation can be restricted by microenvironmental cues or logic-gated recognition of disease-associated signals, thereby confining cell killing to pathological tissues and limiting damage to healthy cells.61,66,75 Such systems provide context-dependent therapeutic control rather than continuous feedback-mediated regulation. The modularity of this layer has driven a shift from one-transgene-one-disease thinking toward programmable therapeutic logic.

Closed-loop circuits

Closed-loop circuits can also be interpreted through control theory. Signal-dependent transcriptional activation provides a proportional-like component, protein accumulation and clearance can generate integral-like memory, and fast negative interactions or combinatorial logic can add derivative-like damping.16,17,76,77,78 This analogy is useful not because therapeutic circuits must literally recapitulate engineered PID controllers, but because it provides measurable performance goals: how quickly the therapy restores homeostasis, how much overshoot it generates, how stable it remains across repeated perturbations, and how robust it is to inter-individual variability. A useful clinical example is the fully closed-loop artificial pancreas developed by the Cambridge group, in which blood glucose is the regulated variable, the desired glycaemic range defines the target range, the deviation between measured or predicted glucose and this range constitutes the error signal, and insulin infusion provides the corrective output. Controller gain can be operationally defined as the change in insulin delivery elicited by a given glucose deviation, whereas restoration time describes how rapidly glucose returns to the target range following meals or other perturbations. Overshoot, stability and robustness can be quantified using post-correction hypoglycaemia, glucose variability, time in range and performance across repeated perturbations or individuals with differing insulin sensitivity.79,80 An analogous framework can be applied to living therapeutic circuits. In a urate-responsive probiotic circuit, urate is the regulated variable, the physiological urate range defines the target range, urate-dependent sensor activation provides the error-sensitive input, and urate oxidase expression constitutes the corrective output.81,82 Controller gain, restoration time, overshoot and robustness could therefore be quantified from the relationship among urate concentrations, therapeutic output and the kinetics of urate normalization following a hyperuricaemic challenge. Reporting such performance metrics would enable more rigorous and standardized comparisons of closed-loop living therapeutics across preclinical studies. These questions are often more clinically meaningful than any single endpoint measurement.

Safety architectures

Finally, safety is becoming a critical design feature rather than an afterthought. Multi-signal logic, reversible attenuation, kill switches, containment strategies and layered feedback architectures all aim to constrain therapeutic output to physiologically meaningful ranges.76,83,84,85 Suicide switches, such as inducible caspase-9 systems, enable the elimination of engineered T cells following severe toxicity, but their action is irreversible and may be compromised by transgene loss, silencing or mutation.86,87,88 NOT-gated circuits provide complementary protection by suppressing CAR-T-cell activity upon recognition of antigens expressed by healthy tissues.89,90 A canonical example is the inhibitory CAR (iCAR) strategy, in which T cells co-express an activating CAR directed against a tumour-associated antigen and an iCAR recognizing a protective antigen on healthy cells.89 More sophisticated multi-input gene circuits further enhance cellular discrimination by integrating tissue-specific and disease-associated signals. For example, a mammalian gene circuit combining liver-enriched transcription factors (HNF1A and HNF1B), tumour-associated SOX9 and SOX10 activity, and low let-7c microRNA abundance as a NOT input selectively identifies hepatocellular carcinoma cells.91,92 Activation of this logic gate drives expression of herpes simplex virus thymidine kinase, thereby enabling selective tumour-cell killing. Despite these safeguards, such systems remain susceptible to basal leakiness, antigen escape, circuit instability and immune-mediated selection pressures.93 This becomes even more important as circuits become more autonomous. A therapy that computes locally must also remain pausable, monitorable and overridable.

Classification of therapeutic control architectures

Therapeutic gene circuits can be classified according to how sensed inputs are coupled to therapeutic outputs, whether the output acts on the regulated variable, and whether control is autonomous or externally supervised.94,95,96 Homeostatic feedback systems directly oppose deviations from a physiological setpoint and enable graded, self-adjusting therapeutic responses. Metabolic-shunting and substrate-depletion systems likewise establish direct feedback by removing or converting the pathological input, although their performance may be constrained by enzyme saturation and metabolic burden.97 Inflammation-responsive circuits provide indirect negative feedback through the production of anti-inflammatory or tissue-protective factors, but their behaviour is often nonlinear and strongly context dependent.98 By contrast, context-gated and feedforward systems primarily determine where and when therapy is activated rather than continuously restoring a physiological variable.22,99 Externally supervised and hybrid systems incorporate devices, algorithms or clinical intervention, offering greater recalibratability and manual control at the cost of reduced autonomy and increased operational complexity. Logic and safety architectures improve specificity, containment and emergency shutdown, but do not necessarily regulate the regulated variable directly.100 The most appropriate architecture therefore depends on the disease biomarker, response dynamics, effector reversibility and required degree of clinical oversight (Table 1). In practice, effective therapeutic platforms may combine feedback control with logic gating, safety modules and external supervision.

Table 1 Comparison of regulatory paradigms for programmable therapeutic systems

Full size table

Metabolic and endocrine diseases as paradigms for closed-loop therapeutics

Metabolic disease has emerged as a leading proving ground for autonomous gene-circuit therapy because circulating biomarkers are accessible, therapeutic setpoints are clinically interpretable, and many therapeutic outputs are reversible or self-limiting. Landmark studies in obesity and diabetes established a design rule that now underpins much of the field: sense a disease-associated metabolite, convert that input into a programmable transcriptional response, and deploy an effector that drives the biomarker back toward the physiological range.8,11,48 Representative architectures across disease classes are summarized in Fig. 2, and autonomous closed-loop and context-responsive therapeutic systems reported to date are compiled in Table 2.

Fig. 2

Representative therapeutic architectures across disease classes. a Anti-obesity control. Elevated circulating fatty acids are sensed through a rewired lipid-responsive nuclear receptor system in which ligand binding to the engineered receptor complex relieves co-repressor recruitment and promotes co-activator assembly, thereby activating transcription of the effector gene pramlintide. Secreted pramlintide suppresses appetite and food intake, providing a feedback mechanism that reduces metabolic burden and helps counteract obesity-associated lipid excess. b A phenylalanine-responsive transcriptional circuit detects elevated phenylalanine (Phe) through a synthetic sensor module that promotes assembly of the transcriptional activator complex under high-Phe conditions. This activates expression of the catabolic enzymes phenylalanine ammonia lyase (PAL) and interleukin-4-induced gene 1 (IL4I1), which enhance phenylalanine degradation and thereby reduce pathological Phe accumulation in a substrate-depletion control mode. c Glycaemic control. Glucose- or hyperglycaemia-responsive designer cells convert elevated extracellular glucose into intracellular signalling events, including changes in membrane depolarization, Ca²⁺ influx and phosphorylation-dependent transcription factor activation, which drive expression of insulin and/or GLP-1. Secretion of these therapeutic effectors lowers blood glucose and restores normoglycaemia, establishing a homeostatic feedback loop around a physiological glycaemic setpoint. d A cholesterol-sensitive shunt circuit uses a modified SREBP–KRAB sensor platform to detect elevated cholesterol and regulate a synthetic expression cassette containing SRE binding sites. Under high-cholesterol conditions, ER retention of SREBP–KRAB relieves repression of the SRE-controlled anti-PCSK9 cassette, enabling anti-PCSK9 expression and promoting cholesterol clearance. As lipid levels normalize, sensor activity is attenuated, enabling closed-loop restoration and maintenance of cholesterol homeostasis. This figure was created using Adobe Illustrator

Table 2 Autonomous closed-loop and context-gated therapeutic systems reported to date

Full size table

Canonical closed-loop metabolic gene circuits

This principle first appeared in a fatty-acid-responsive circuit that drove pramlintide expression to reduce appetite and treat diet-induced obesity.11 It was then extended by β-cell-mimetic designer cells that coupled glucose sensing to insulin secretion and reproduced a canonical homeostatic function in vivo.8 A self-adjusting hyperinsulinaemia-responsive circuit further showed that closed-loop control can correct not only hormone deficiency but also pathological endocrine states such as insulin resistance, coordinating adiponectin release in response to excess insulin.48 Together, these studies established metabolic homeostasis as a tractable and clinically meaningful setting for therapeutic feedback control.

Hybrid and transitional control systems

Hybrid and transitional forms of control have also played an important role in this progression. Chemical interfaces such as green-tea-triggered regulation in diabetic mice and monkeys, together with cyber-physical systems based on smartphones, smartwatches, and electrodes, have helped bridge the gap between fixed dosing and fully autonomous therapy.27,42,57,58,101,102 Although these platforms are not autonomous in the strict sense, they helped translate programmable gene control into more adaptive therapeutic formats by embedding it within broader digital or wearable systems.

Metabolic shunting and flux reprogramming

More recent metabolic circuits suggest that the field is moving beyond endocrine replacement toward metabolic shunting. A cholesterol-responsive shunt used SRE-linked control to drive a PCSK9-directed effector and restore lipid homeostasis in experimental dyslipidaemia.71 A phenylalanine-responsive circuit extended the same logic to amino-acid homeostasis by activating PAL and IL4I1 only when phenylalanine rose into the pathological range.72 Recently, a hyperglycaemia-sensing circuit repurposed endogenous O-GlcNAc-YAP signalling to drive insulin or GLP-1 expression in diabetic mice, further expanding the repertoire of endogenous control axes that can be harnessed for setpoint regulation.103 Urate homeostasis was restored using a self-sufficient synthetic circuit in 2010, establishing an early example of a biomarker-responsive mammalian cell implant for autonomous therapeutic control. More recently, engineered probiotics have extended the same sense-and-respond principle to living microbial therapeutics. The probiotic-based urate-sensing and regulation platform PULSE was designed to detect elevated urate concentrations and dynamically increase urate-degrading activity, thereby restoring urate homeostasis.81 Similarly, the glucose-responsive probiotic platform GIFT used a HexR-based sensor to trigger glucose-dependent secretion of glucagon-like peptide-1 (GLP-1), improving glycaemic control in diabetic mice and non-human primates.104

Circadian and multi-timescale control

Circadian physiology adds a further dimension to metabolic control. Chronogenetic and circadian-biomarker-responsive systems show that oscillatory endocrine signals can be used either to phase-lock therapeutic output or to restrict actuation to biologically appropriate time windows.105,106,107 A dual-responsive inflammatory- and circadian-sensing OR-gate circuit further showed that therapeutic control can be coordinated across both pathological flares and daily rhythmic windows, extending chronobiologic design from timing alone toward multi-timescale regulation.108 These platforms are not disease-biomarker loops in the narrow sense, but they broaden the control paradigm by showing that physiology itself can determine when therapy should act.

Artificial pancreas

The artificial pancreas represents the most mature clinical implementation of closed-loop therapeutic control. Automated insulin-delivery devices have already shown how closed-loop control can reduce treatment burden and improve physiological stability when biomarker, controller, and effector are tightly coupled.109,110 Living closed-loop gene circuits extend the same logic inward: instead of relying on an external sensor-pump algorithm, the controller is embedded within engineered cells. Together with emerging hybrid biological-electronic systems, this supports a central premise of setpoint medicine: the next step is not simply to automate dosing, but to internalize physiological control within living therapeutics.

Inflammation, infection, tissue protection and cardiovascular homeostasis

Inflammation

i-ROBOT couples thiosulfate sensing to inflammatory-state recording and disease-responsive release of AvCystatin.111 In this system, the ThsS/ThsR two-component regulatory system detects the inflammation-associated biomarker thiosulfate and initiates the generation of heritable molecular records. In addition to enabling longitudinal monitoring of intestinal inflammation through fluorescence, sequencing and colorimetric readouts, i-ROBOT incorporates a second-generation base-editor-based recording module linked to translational activation of therapeutic protein expression. The engineered bacterium therefore extends beyond diagnosis to localized therapeutic intervention by producing effector molecules at sites of inflammation.111,112

NO-responsive microbial sensing and therapeutic circuits

Chua et al. engineered a probiotic therapeutic system in which the endogenous transcriptional regulator NorR senses the inflammatory biomarker nitric oxide (NO). Elevated NO concentrations activate NorR-dependent transcription from the pNorV promoter, thereby inducing expression of interferon-λ1 (IFN-λ1).113 Fusion to the YebF secretion signal enables extracellular delivery of IFNL1, which modulates local immune responses and attenuates intestinal inflammation. More recently, Ye et al. engineered Bacteroides thetaiotaomicron to respond to both NO and deoxycholic acid (DCA), enabling sensitive detection of intestinal inflammation.114 Integration of the NO- and DCA-responsive modules into an AND-gated gene circuit further restricted therapeutic activation to disease-associated intestinal conditions. Because intestinal bleeding is another hallmark of inflammatory bowel disease, blood-responsive gene circuits provide an additional strategy for triggering localized therapeutic production in the gut.115,116

A heme-responsive therapeutic circuit

Ge et al. engineered Escherichia coli with the heme transporter ChuA, the heme-responsive transcriptional repressor HrtR and the HrtR-regulated promoter (PL(HrtO)) to couple gastrointestinal bleeding to localized therapeutic protein production. In response to heme, the bacteria secrete the barnacle-derived adhesive protein CP43K together with trefoil factor 3 (TFF3), thereby promoting mucosal adhesion and intestinal barrier repair.117

A calprotectin-responsive therapeutic circuit

Engineered bacteria have been developed to sense faecal calprotectin, a widely used non-invasive biomarker of intestinal inflammation, thereby enabling sensitive detection of inflammatory activity in the gut.118,119 Coupling calprotectin-responsive sensing to production of the anti-inflammatory cytokine interleukin-10 (IL-10) enabled the bacteria to both detect disease-associated inflammation and promote resolution of chemically induced colitis.120 Although microbial systems differ from mammalian cell therapies in chassis and deployment, they demonstrate that the same sense–compute–act logic can be implemented across living therapeutic platforms.81,111,121

Tissue protection

In engineered cartilage, an NF-κB-responsive synthetic promoter was used to drive expression of interleukin-1 receptor antagonist (IL-1Ra), thereby establishing a self-regulating anti-inflammatory circuit.122 Inflammatory activation induces NF-κB-dependent IL-1Ra production, which suppresses IL-1 signalling and consequently reduces NF-κB activity, thereby closing the negative-feedback loop. Mechanogenetic tissues further showed that physical loading can serve as a physiologically meaningful trigger for local anti-inflammatory drug production.63

Infection

Infection-responsive systems extend therapeutic control to pathogen-specific sensing, real-time monitoring and autonomous antimicrobial intervention, including quorum-sensing-based biosensors and sense-and-destroy circuits.123,124,125 An engineered Lactococcus lactis strain expressing the synthetic transmembrane receptor HR4M specifically detects the Vibrio cholerae quorum-sensing signal cholera autoinducer-1 (CAI-1) in the intestine and activates β-lactamase reporter expression in response.126 Holowko et al. subsequently engineered an Escherichia coli biosensor by coupling the V. cholerae quorum-sensing system to CRISPR interference, enabling sensitive detection of V. cholerae-conditioned supernatant with tightly controlled green fluorescent protein expression.127 Building on this platform, Jayaraman et al. replaced the reporter with the gRNA, which binds to the constitutively expressed dCas9, and the resulting complex represses expression of the YebF–Art-085 effector construct.128 These systems illustrate context-responsive infection control rather than classical homeostatic feedback, because pathogen-associated signals trigger a predefined diagnostic or antimicrobial programme without continuous regulation of a physiological setpoint.

Pseudomonas aeruginosa is a major cause of respiratory and gastrointestinal infections and frequently displays broad resistance to antibiotics and other antimicrobial agents.129 To achieve pathogen-specific intervention, Escherichia coli was engineered to detect the P. aeruginosa quorum-sensing signal N-(3-oxododecanoyl)-L-homoserine lactone (3OC12-HSL). In response, the engineered sentinels produce and release the antibacterial toxin CoPy, enabling localized elimination of P. aeruginosa within the infection microenvironment. Together, these systems couple real-time pathogen sensing to on-demand production and delivery of antimicrobial effectors.130 They therefore exemplify context-responsive sense-and-destroy circuits rather than homeostatic closed-loop control, because pathogen-derived signals trigger a predefined antimicrobial programme without continuous regulation of a physiological setpoint.

Cardiovascular homeostasis

The cardiovascular arena is now emerging as another important setting for closed-loop therapeutics. Earlier hybrid feedback concepts explored behaviour-linked blood-pressure control,131 but a newly reported angiotensin-II/ACE2 hormone shunt is more directly biomarker-driven.74 In this system, the native AT1R receptor is rewired to a synthetic promoter that drives secretion-engineered soluble ACE2, thereby degrading angiotensin II and restoring normotension in experimental hypertension. The design is notable because it uses a disease-native sensor, routes the signal through endogenous GPCR-linked pathways and counters pathology by direct agonist elimination rather than simple receptor blockade.

The same antihypertensive shunt exemplifies the expanding therapeutic scope of closed-loop circuit design. The circuit operates across a clinically relevant angiotensin range, avoids hypotensive overshoot in mice, behaves predictably alongside conventional antihypertensive drugs and responds across angiotensin infusion, obesity-associated renin-angiotensin dysregulation and selected human sera.74 These are the kinds of features that move a system beyond elegant proof-of-concept toward a candidate for patient stratification and eventual clinical development. Cardiac injury is now entering the same design space: a cTnI-sensing designer-cell system used a synthetic troponin receptor to trigger tenecteplase secretion in a strict cTnI-inducible, doxycycline-repressible manner, providing a closed-loop proof-of-concept for early detection and treatment of acute myocardial infarction.132

What links urate control, liver protection, thyroid regulation, inflammatory implants, infection defence, hypertension and cardiac injury is not a shared effector but a shared control logic. Endogenous signalling becomes the decision layer. Signal transduction is not merely tolerated; it is repurposed as the physiological computation that determines when and how much therapy should be delivered. This logic is especially attractive in chronic disease because it turns diagnosis and treatment into a continuous preventive process (Fig. 3, Table 2). Beyond the peer-reviewed systems listed in Table 2, a recent study describing an Activin A inhibitor-secreting closed-loop cell therapy for fibrodysplasia ossificans progressiva suggests that autonomous therapeutic control may also extend to aberrant tissue remodelling disorders.133

Fig. 3

Representative closed-loop and context-responsive therapeutic circuits for inflammation, infection, tissue protection and cardiovascular homeostasis. Inflammation-sensing biosensors include a thiosulfate-responsive system (a), a nitric oxide (NO)-responsive biosensor for detecting inflammatory signals (b), a heme-responsive biosensor (c), a calprotectin-responsive biosensor (d), and an IL-1 signaling–responsive biosensor associated with tissue protection (e). Infection-responsive systems based on quorum sensing (QS) enable detection of Vibrio cholerae (f) and Pseudomonas aeruginosa PAO1 (g). For cardiovascular homeostasis, the angiotensin II receptor AT1R was rewired to a synthetic promoter to enable sensing of pathological angiotensin II signaling (h), supporting homeostatic regulation. This figure was created using Adobe Illustrator

Cancer and immunotherapy: from context-gated targeting to adaptive control

Cancer poses a different control problem from metabolic homeostasis. Its relevant inputs are spatially heterogeneous, often multimodal, and shaped by immunosuppression, stromal barriers and variable antigen density.134,135 For this reason, many of the most important therapeutic circuits in oncology are not yet fully closed-loop in the strict homeostatic sense.89 Instead, they are context-aware, logic-gated or externally actuated systems designed to reduce toxicity and sharpen specificity (Fig. 4).39,90,136,137,138 Even so, they belong to the same conceptual family because they transform signal transduction into programmable therapeutic decision-making.

Fig. 4

Synthetic genetic circuits for context-dependent cancer recognition and therapeutic control. Synthetic receptors and intracellular gene circuits enable engineered cells to integrate tumour-associated antigens, soluble factors, microenvironmental signals and oncogenic pathway activity, and to convert these inputs into selective therapeutic outputs. a Antigen-recognition platforms, including chimeric antigen receptors (CARs), synthetic Notch receptors (synNotch) and synthetic intramembrane proteolysis receptors (SNIPRs), couple cell-surface antigen binding to intracellular signalling or programmable transcription. b Multi-input logic circuits implement AND, OR and NIMPLY (A AND NOT B) operations to improve tumour discrimination and reduce activation in normal tissues. c Modular and switch-like receptors, including MESA-based systems and rewired cytokine or growth-factor receptors, convert soluble extracellular signals into user-defined transcriptional responses. d Tumour-microenvironment-responsive systems use protease activity or other local cues to unmask or activate therapeutic receptors selectively within the tumour microenvironment. e Cell-autonomous control includes activation-induced ADAM17-mediated shedding of AIR-CARs and CRISPR knock-in of therapeutic payloads under tumour-restricted endogenous promoters. f Oncogenic pathway sensors detect intracellular cancer-associated signalling or mutation-dependent RNA-processing events and trigger therapeutic gene expression or selective cell killing. Together, these platforms illustrate how iterative design and optimization of synthetic genetic circuits can improve the specificity, adaptability and therapeutic potency of engineered anticancer cells. This figure was created using Adobe Illustrator

Synthetic receptors for cancer therapy

Synthetic receptors have been engineered to recognize cell-surface antigens,9,66,139,140 soluble cytokines and inflammatory mediators,40,68,141,142,143 as well as extracellular signals processed through G protein-coupled receptor (GPCR)-based circuits.144,145

Synthetic Notch (synNotch) receptors

A canonical synNotch receptor comprises an engineered extracellular sensing domain, the Notch regulatory core and an intracellular transcription factor. Ligand binding generates mechanical force that initiates proteolytic cleavage of the Notch regulatory core, releasing the transcription factor to activate a user-defined genetic programme.7,9,38 Morsut et al. established the modularity of the synNotch platform by replacing both the native extracellular sensing domain and the intracellular transcriptional effector with user-defined components, demonstrating that receptor sensing and transcriptional outputs can be programmed independently.7 Since then, the synNotch platform has enabled combinatorial antigen recognition, conditional CAR expression,9,66 localized cytokine production,146 antigen-density discrimination147 and spatially restricted immunosuppression.148,149 Together, these advances have established synNotch as a versatile platform for coupling disease-associated signals to programmable therapeutic responses.

Synthetic intramembrane proteolysis receptors

A representative example is the synthetic intramembrane proteolysis receptor (SNIPR), a Notch-based platform engineered to detect soluble proteins.40 Unlike conventional synNotch receptors, which primarily respond to membrane-bound ligands, SNIPRs can be activated by soluble ligand binding, thereby initiating receptor processing and release of an intracellular transcriptional effector. Coupling SNIPR activation to CAR expression enables conditional activation of CAR-T cells in tumours enriched for soluble immunosuppressive mediators, improving tumour selectivity while limiting off-target activity. More broadly, SNIPRs extend the sensing repertoire of synthetic receptors beyond cell-surface antigens and provide a modular framework for engineering microenvironment-responsive cellular therapies.

Chimeric antigen receptor

A landmark example is the synNotch-to-CAR architecture, in which recognition of a tumour-priming antigen, such as ALPPL2, induces expression of a CAR directed against a tumour-associated target.66 This design addresses two major limitations of CAR-T-cell therapy for solid tumours: tonic signalling and on-target, off-tumour toxicity. By making CAR expression contingent on prior antigen recognition, the circuit preserves favourable T-cell states while enhancing tumour specificity. Unlike metabolic homeostatic feedback systems, this architecture exemplifies context-dependent therapeutic control. It functions as an autonomous on-demand eradication programme that is activated within the tumour microenvironment and returns to a quiescent state once the priming signal is no longer present.

More recently, Mackall and colleagues developed activation-induced release (AIR)-regulated CARs as a cell-autonomous negative-feedback strategy for dynamically tuning CAR signalling.150 Incorporation of a minimal CD62L-derived AIR motif into the CAR exploits endogenous ADAM17-mediated proteolysis to promote receptor shedding following T-cell activation. This activation-coupled negative-feedback mechanism continuously adjusts surface CAR abundance according to signalling intensity, thereby limiting excessive tonic and antigen-driven signalling, reducing T-cell exhaustion and activation-induced cell death, and ultimately improving T-cell fitness and antitumour efficacy.

G protein-coupled receptors (GPCRs)

Programmable antigen-gated G protein-coupled engineered receptors (PAGERs) employ a GPCR-based scaffold to couple soluble antigen recognition to inducible gene expression.145 Antigen binding relieves receptor autoinhibition, triggering receptor activation and subsequent induction of user-defined transcriptional programmes. Their modular architecture integrates soluble and cell-surface antigen sensing with pharmacological control, providing a versatile platform for programmable cellular therapies.

Beyond antigen recognition, synthetic GPCR-based receptors have been engineered to detect soluble factors within complex tissue microenvironments, enabling context-dependent therapeutic regulation while minimizing systemic toxicity. By combining customized ligand-binding domains with engineered signalling scaffolds, these receptors can respond to disease-associated mediators such as transforming growth factor-β (TGF-β) and vascular endothelial growth factor (VEGF),151,152 two key mediators of tumour immunosuppression and microenvironmental remodelling.

Tumour microenvironment-responsive gene circuits for cancer therapy

Complementary strategies have demonstrated that both tumour microenvironmental cues and engineered microbial chassis can further expand programmable control in oncology. Hypoxia-sensing CAR-T cells employ a dual oxygen-sensing switch to confine CAR expression to hypoxic tumour regions, thereby addressing one of the major limitations of solid-tumour immunotherapy: on-target, off-tumour toxicity.14 More recently, endogenous tumour-restricted promoters, including NR4A2 and RGS16, have been repurposed to localize cytokine expression within tumours, extending context gating from antigen-dependent activation to state-dependent transcriptional control.153

Living bacteria provide an orthogonal chassis for tumour-targeted therapy because they preferentially accumulate within tumours, where local physicochemical conditions can be exploited as endogenous regulatory inputs.154,155,156,157 Rather than implementing homeostatic feedback, these systems exemplify context-dependent therapeutic control by coupling tumour-specific environmental cues to localized therapeutic gene expression. Representative inputs include hypoxia, extracellular acidity and elevated lactate concentrations.158,159 For example, the hypoxia-responsive PfdhF promoter has been used to restrict anti-CD47 antibody expression to hypoxic tumour regions, thereby enhancing macrophage-mediated tumour clearance while minimizing off-target activity.160 Leventhal et al. engineered the attenuated probiotic strain SYNB1891 to express the Listeria monocytogenes diadenylate cyclase DacA, thereby producing cyclic di-AMP and activating the STING pathway within tumours.161 Hypoxia-responsive control has also been applied to essential bacterial genes such as asd, restricting bacterial proliferation to hypoxic tumour tissue and improving biosafety.162

Additional tumour-associated metabolic cues have likewise been harnessed for programmable control. Acid-responsive promoters, including PadiA and the CadC/cadBA regulatory system, drive therapeutic gene expression selectively under acidic conditions characteristic of the tumour microenvironment.163,164 Likewise, lactate-inducible circuits exploit the elevated lactate concentrations found in solid tumours to regulate therapeutic payloads independently of glucose and oxygen availability.165 Building on this strategy, Zou et al. combined lactate-dependent regulation of the essential growth gene asd and coagulase with quorum-sensing-controlled α-haemolysin expression in an engineered Escherichia coli Nissle strain, thereby improving both therapeutic efficacy and biosafety.166

Intracellular tumour signal-responsive gene circuits for cancer therapy

Beyond sensing extracellular cues, synthetic gene circuits have increasingly been engineered to monitor tumour-cell-intrinsic molecular states, enabling therapeutic responses that are directly coupled to oncogenic transcriptional, signalling and RNA-processing abnormalities. An AAV-delivered c-MYC-responsive gene circuit selectively reprogrammed tumour cells to express immunostimulatory payloads in response to aberrant c-MYC activity, illustrating how oncogenic transcriptional programmes themselves can serve as therapeutic control inputs.167 Extending this concept from transcriptional dysregulation to aberrant signalling pathways, the Rewiring of Aberrant Signalling to Effector Release (RASER) system converts hyperactive ErbB signalling into programmable therapeutic outputs.168 Using engineered protein pairs that recognize phosphorylated ErbB receptors and recruit an NS3 protease, RASER enables signalling-dependent release of therapeutic cargos, inducing apoptosis or activating CRISPR/Cas9-mediated cytokine expression selectively in ErbB-hyperactive cancer cells. Beyond transcriptional and signalling abnormalities, mutation-specific RNA processing has likewise been harnessed for tumour-selective control. Additionally, a RAS-responsive circuit couples RAS-GTP binding by CRAF RBDCRD–NarX fusion proteins to orthogonal NarL-dependent output expression, enabling preferential activation in RAS-driven cancer cells.169 A mutation-responsive splicing circuit exploits SF3B1 mutation-dependent RNA processing to drive selective HSV-TK expression, enabling ganciclovir-mediated elimination of SF3B1-mutant cancer cells while sparing genetically matched wild-type cells.170

Dietary small molecules, light, ultrasound and near-infrared-responsive bacteria have likewise been used to spatially restrict or temporally gate antitumour functions.59,61,75,158 These studies matter because they shift cancer therapy away from static potency and toward programmable context recognition, an essential requirement for any future autonomous oncology platform.

A fully closed-loop oncology platform will likely need to integrate antigen density, inflammatory tone, hypoxia, metabolic stress, stromal exclusion and treatment history. Sensor fusion of this kind is already becoming conceptually feasible through combinatorial protein circuits, synthetic receptor systems and CRISPR-based regulatory logic.61,67,70,171 The challenge is therefore not only to detect cancer, but to compute the appropriate therapeutic action under uncertainty while remaining within an acceptable safety envelope.

For that reason, cancer may be the disease area in which open-loop, context-gated and fully autonomous strategies will coexist for the longest time. Clinician override will remain important, but logic-gated cell therapies already show that targeted therapy can be defined not only by where a product goes or what it binds, but also by how it decides (Fig. 4, Table 2).

Why the field is now ready for the clinic: delivery, durability, oversight and trust

The central question for the field is no longer whether therapeutic gene circuits can work in principle. It is whether they can be delivered, stabilized, monitored and governed well enough to be trusted. Many of the most convincing autonomous demonstrations still rely on ex vivo engineered cells that are encapsulated and implanted. This format offers design freedom, retrievability and the possibility of device-like oversight, but it also imposes manufacturing burdens, limits implant longevity and introduces foreign-body responses.172

Vector-based delivery broadens the clinical design space through the use of viral vectors, viral-like particles, mRNA systems and plasmids, but autonomy places unusually strict demands on calibration. A constitutive transgene that drifts moderately may retain utility; a closed-loop controller that drifts may alter its threshold, leakiness or gain in clinically important ways.173,174 Seen through the software-hardware paradigm, translation depends not only on programming better genetic software but also on matching that software to hardware platforms whose persistence, accessibility, retrievability and immunological compatibility fit the clinical task. In setpoint medicine, delivery and control fidelity cannot be separated.

Durability is therefore more than a persistence problem. Long-term deployment exposes circuits to selective silencing, mutational erosion, epigenetic drift, altered cell-state composition and shifting host microenvironments. Engineering for low burden, redundancy, recalibratability and modular replacement will be as important as optimizing the initial control function. Safety switches remain necessary, but they are not sufficient: a controller must remain well tuned, not merely terminable.83,84

Clinical translation will also require a controller-based evaluation language. Beyond mean endpoint improvement, studies should report restoration time after perturbation, overshoot, oscillation, stability over months, robustness to co-medications and physiological variability, and the relationship between biomarker excursions and therapeutic output. The artificial-pancreas field has benefited from exactly this framing.109,110 Autonomous gene and cell therapies now need an analogous set of control-performance metrics, as outlined in Fig. 5 and Table 3.

Fig. 5

Translational roadmap from proof-of-concept to first-in-human setpoint therapy. The path to the clinic begins with selecting indications that have measurable biomarkers, interpretable setpoints and reversible or self-limiting effectors. Candidate circuits then require delivery formats that support stability and, where possible, retrievability or external override; preclinical studies should quantify restoration time, overshoot, oscillation, durability and robustness; manufacturing must preserve calibration across batches; and clinical deployment should incorporate layered safety, human oversight and access planning. An outer supervisory loop, including wearables, bioelectronics or digital modelling, may provide monitoring and override while the inner biological controller executes local setpoint control. This figure was created using Adobe Illustrator

Table 3 Translational requirements for early clinical closed-loop therapeutics

Full size table

Taken together, these are the reasons why the field now looks increasingly ready for clinic-oriented programmes. Logic-gated circuit therapies have already crossed the clinical threshold: the E-SYNC synNotch-CAR T-cell platform is being evaluated in a phase I study in patients with EGFRvIII-positive glioblastoma, showing that dynamic circuit control is no longer confined to proof-of-concept in rodents.66,175 Autonomous biomarker-responsive implants remain preclinical, but disease coverage is now broad and mechanistically diverse. Indications with interpretable biomarkers, bounded physiology, reversible or self-limiting effectors and retrievable or pausable delivery platforms should be considered realistic first-wave candidates (Fig. 5, Table 3).

Hybrid supervision may make early translation safer and more governable. Bioelectronics, wearables and patient-specific digital models can provide an outer supervisory loop even when a biological circuit executes the inner local control function.176,177,178,179,180,181,182 In this bounded-autonomy model, diagnosis becomes continuous, treatment becomes adaptive, prevention becomes automatic and durable disease correction becomes experimentally testable rather than purely aspirational (Fig. 5, Table 3).

AI-assisted design, calibration and supervisory control

AI-driven design of biological components

Foundation AI models, including AlphaFold, ESM and AlphaFold3, have transformed protein structure prediction, sequence representation and biomolecular interaction modelling.183,184,185,186,187 In parallel, generative frameworks such as RFdiffusion and ProteinMPNN enable de novo protein backbone generation, sequence design, binder engineering and functional-site scaffolding.188,189,190,191 Together, these advances provide a computational foundation for engineering sophisticated biological components, including protein switches, inducible dimerization modules, protease-responsive domains and aptamer-based sensors. For example, computationally designed amantadine-binding proteins have progressed from de novo homotrimers to small-molecule-responsive oligomeric systems, whereas chemically induced dimerization (CID) platforms provide broadly applicable mechanisms for ligand-controlled protein assembly.192,193,194,195,196 Likewise, light-switchable protein binders and machine learning-guided aptamer discovery illustrate how computational design can accelerate the development of inducible sensing modules.197,198 Importantly, these approaches do not replace experimental screening but instead improve the quality, diversity and functional relevance of candidate libraries, thereby accelerating downstream circuit engineering.

AI-driven genetic circuit programming

Computational frameworks increasingly enable integrated modelling of how promoters, regulatory elements, linkers, sensing modules and effector components collectively determine circuit behaviour. Engineering therapeutic gene circuits requires balancing multiple competing design objectives, including response threshold, dynamic range, transcriptional leakage, basal activity, output strength, robustness and safety. Automated circuit design platforms have demonstrated that high-level logical specifications can be translated into functional DNA-encoded circuit architectures.199 In parallel, machine learning-guided recommendation systems, Bayesian optimization and in silico design-build-test-learn (DBTL) workflows facilitate iterative refinement of circuit designs by integrating prior experimental knowledge with predictive modelling.200,201,202,203 In immune cell engineering, combining combinatorial chimeric antigen receptor (CAR) signalling-domain libraries with machine learning has enabled systematic mapping of the relationships between CAR architecture and cellular phenotypes,204 while computational tools such as CAR-Toner and CARMSeD facilitate the identification of tonic signalling and dysfunctional CAR designs.205,206 Collectively, these approaches illustrate that the principal contribution of AI within DBTL workflows is not to replace experimental validation but to maximize information gain during candidate selection, thereby accelerating the engineering of robust therapeutic gene circuits.

System-level modelling and supervisory control

Achieving reliable therapeutic performance requires extending AI-assisted circuit design to encompass system-level physiological modelling.207 Following in vivo deployment, circuit behaviour is dynamically influenced by the host biological context, delivery modality, spatiotemporal tissue distribution, immune microenvironment, systemic pharmacokinetics and pharmacodynamics (PK/PD), and interpatient variability. Mechanistic models, multiscale computational frameworks and physics-informed neural networks (PINNs) integrate reaction kinetics, molecular diffusion, mass transport and pharmacological processes to predict therapeutic circuit behaviour, demonstrating robust accuracy even under data-limited conditions.208,209,210,211 Digital twin frameworks further extend these capabilities by integrating longitudinal multi-omics, high-resolution imaging, single-cell profiling, PK/PD measurements and real-time device readouts to simulate therapeutic circuit trajectories across diverse physiological and pathological states.178,180,212,213 Building on these multiscale representations, model predictive control (MPC), adaptive feedback control and reinforcement learning have emerged as promising supervisory strategies for closed-loop therapeutics and targeted drug delivery systems.79,182,214,215 Together, these approaches establish a quantitative framework for detecting systemic instability, signal overshoot and functional drift, thereby enabling proactive circuit recalibration or timely clinical intervention before therapeutic performance deteriorates.

AI-enabled supervisory intelligence and setpoint medicine

This system-level perspective naturally converges with the concept of setpoint medicine, in which therapeutic objectives extend beyond single-dose interventions to the continuous maintenance of physiological variables within predefined, clinically meaningful ranges.79,182,214,215 Accordingly, AI is expected to operate across both the design and runtime phases of therapeutic systems: optimizing circuit architectures and delivery strategies during development, while enabling bounded autonomy, patient-specific monitoring and supervisory control following deployment. Within this framework, AI serves as a supervisory intelligence layer that links molecular design, system dynamics and clinical decision-making.

Successful clinical translation, however, requires equal emphasis on interpretability, transparency and governance. AI-assisted therapeutic control must be supported by auditable performance metrics, uncertainty quantification, explicit safety constraints and robust human oversight, rather than relying on opaque decision-making.216,217,218,219 Only under these conditions can AI enhance the safety, reliability and clinical implementation of programmable therapeutics across heterogeneous patient populations and complex disease settings.

Data limitations and future perspectives

Despite rapid methodological advances, the principal bottleneck in AI-enabled closed-loop therapeutics lies not in algorithmic complexity but in the quality, scale and translational relevance of empirical datasets.181,220,221,222,223 Critical gaps include sparse sequence-to-function maps, incomplete characterization of biological component dynamics, limited in vitro-to-in vivo predictability, inadequate pharmacokinetic and biodistribution data, and a scarcity of longitudinal safety datasets and clinical failure registries.224,225,226,227 To advance clinically reliable closed-loop therapeutics, predictive models must move beyond binary functional classifications to capture nonlinear response kinetics, temporal persistence, functional drift, safety thresholds and interpatient variability.212,228,229 Addressing these challenges will require the integration of uncertainty-aware modelling, standardized multimodal data infrastructures and interpretable performance metrics across experimental and clinical settings.230,231,232

Looking ahead, the central role of AI in this field will not be to support the unconstrained expansion of model complexity, but to enable verifiable, iterative and governable computational-experimental workflows. As high-throughput functional screening, automated biofoundries, single-cell and spatial multi-omics, wearable biosensors and longitudinal clinical registries mature, AI could progressively integrate molecular design, circuit kinetics, in vivo pharmacology and real-time patient-state monitoring within unified predictive frameworks.176,233,234,235,236,237,238,239 Closed-loop therapeutic systems may thereby evolve from predominantly static, hardwired genetic circuits towards adaptive architectures capable of model-guided recalibration, context-dependent adjustment and prospective risk mitigation (Fig. 6). Supported by robust data infrastructures, standardized validation platforms and evolving regulatory frameworks, AI could ultimately provide the integrative framework required to translate programmable therapies into adaptive, continuously supervised therapeutic systems.181,218,220,221,222,223,240,241

Fig. 6

Artificial intelligence across the closed-loop therapeutics pipeline. AI-enabled approaches support the development of closed-loop therapeutics from molecular component design to patient-level control. a Component design. Foundation models, generative design and structure-prediction tools facilitate the engineering of protein and RNA parts, protein switches, inducible dimerization modules, protease-responsive domains and aptamer-based sensors. b Circuit programming. Active learning, Bayesian optimization and machine-learning recommendation systems guide the design–build–test–learn cycle and help optimize promoter architecture, regulatory modules, sensing thresholds, dynamic range, basal leakage and safety windows. c System design and translation. Multiscale and physics-informed models, together with digital twins, integrate chassis selection, delivery, tissue distribution, pharmacokinetic/pharmacodynamic behaviour and patient heterogeneity to predict in vivo circuit performance. d Deployment and supervisory control. Model-predictive control, feedback-control algorithms and reinforcement learning can support longitudinal monitoring, detection of overshoot or functional drift, therapeutic recalibration and human override. e Clinical goal. These integrated capabilities aim to enable setpoint medicine by maintaining physiological variables within predefined and therapeutically desirable ranges. This figure was created using Adobe Illustrator

Concluding remarks

Over the past fifteen years, programmable mammalian therapeutics have progressed from elegant remote-controlled switches toward systems that increasingly resemble physiological controllers. One especially productive research trajectory has made this progression particularly visible, moving from optogenetic and physically actuated designer cells to biomarker-responsive circuits for metabolism, inflammation, infection, cardiovascular control and cancer.7,8,9,12,13,15,42 At the same time, the field is beginning to recognize a software-hardware paradigm in synthetic biology: genetic circuits provide the programmable control logic, whereas mammalian therapeutic cells and tissue-associated chassis provide the biological hardware on which this logic runs.4,242,243

Clinical translation and future implementation

Despite remarkable advances in synthetic biology and therapeutic gene circuit engineering, the clinical translation of programmable therapeutic systems remains constrained by several fundamental challenges. Among these, immunogenicity and host-circuit interactions represent major barriers to the long-term function of programmable therapeutic systems. Synthetic components, engineered receptors and non-self transgene products may be recognized by the host immune system, resulting in immune clearance, loss of circuit function or reduced therapeutic durability.244,245,246,247 To mitigate these risks, increasing efforts have focused on humanized and orthogonal biological components, including humanized synthetic receptors and minimally immunogenic regulatory elements, to reduce immune recognition while preserving circuit function.7,9,248

Importantly, immunogenicity is determined not only by circuit design but also by the choice of delivery modality. For example, adeno-associated virus (AAV)-based gene transfer can be compromised by pre-existing neutralizing antibodies and cellular immune responses, whereas repeated administration may further amplify immune recognition. Accordingly, optimized delivery vectors, transient expression strategies and tissue-targeted delivery approaches remain active areas of development.245,247,249,250 In parallel, biomaterial-based encapsulation and immune-isolation technologies have emerged as promising strategies to shield engineered cells or implantable devices from immune surveillance while maintaining bidirectional molecular exchange with the host environment. Such approaches may prolong therapeutic function while reducing the need for systemic immunosuppression.106,251,252

Additional translational opportunities may arise from deployment in immune-privileged or anatomically restricted sites, including the eye and skin, where immune exposure is comparatively limited.247,253 For cell-based therapies, autologous cells, hypoimmunogenic engineered cells and universal donor cell platforms may further reduce immune rejection and improve long-term persistence.248,254 Beyond immunogenicity, host-circuit compatibility also critically influences therapeutic performance. Differences in gene expression stability, signalling capacity, metabolic burden and tissue microenvironment across host cell types can alter the reliability and robustness of engineered circuits. Likewise, disease-associated changes in inflammation, nutrient availability and tissue architecture may reshape circuit behaviour in ways that remain difficult to predict from preclinical models.255,256,257,258 Improving chassis-circuit compatibility and engineering systems that maintain stable function across diverse physiological contexts will therefore be essential for achieving predictable and durable clinical performance.

Equally important are considerations of manufacturing, scalability and equitable access. Therapeutic platforms that depend on bespoke laboratory processes are unlikely to achieve widespread clinical implementation. Translational considerations should therefore be incorporated from the earliest stages of circuit design, including manufacturability, process robustness, batch-to-batch consistency, cost transparency and reimbursement strategy, rather than being treated as downstream challenges. Standardized cell banks, modular vector platforms and streamlined quality-control assays will be critical for enabling scalable and reproducible manufacturing. Addressing these factors early will help ensure that advances in therapeutic control translate into clinically meaningful and broadly accessible treatments.

Ultimately, the next generation of closed-loop therapeutics should be developed with clinical translation as a primary design objective rather than a downstream consideration. Achieving this vision will require the coordinated integration of synthetic biology, systems engineering, artificial intelligence, manufacturing science and regulatory innovation to realize programmable medicines that are not only intelligent, adaptive and safe, but also practical, scalable and broadly accessible in routine clinical care.

Building on these translational foundations, the next phase of closed-loop therapeutics should be driven by indication-led development programmes that define clinically meaningful setpoints, auditable performance metrics, layered safety architectures and scalable implementation strategies.259,260,261,262 Such programmes will also require hardware-aware design choices, because identical control logic may behave differently when deployed in implantable designer cells, engineered immune cells or other mammalian therapeutic chassis. AI-assisted design, virtual cell and tissue models, and supervisory bioelectronics may therefore become enabling infrastructure by helping to match software to hardware, anticipate functional drift, and support recalibration and clinical oversight.183,250,263

If these challenges can be overcome, the implications extend far beyond a new class of gene switches. Medicine could progressively shift from intermittent diagnosis followed by manual dosing towards continuous setpoint management, in which diagnosis, treatment and prevention are integrated within the therapeutic itself. In selected chronic diseases, durable restoration of physiological homeostasis may even begin to approximate functional cure. The next challenge is therefore clear: to transform closed-loop gene and cell therapies from isolated proof-of-concept studies into the first clinically credible platforms for setpoint medicine.

References

  1. Jacob, F. & Monod, J. Genetic regulatory mechanisms in the synthesis of proteins. J. Mol. Biol. 3, 318–356 (1961).

    Article  PubMed  CAS  Google Scholar 

  2. Elowitz, M. B. & Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature 403, 335–338 (2000).

    Article  PubMed  CAS  Google Scholar 

  3. Purnick, P. E. & Weiss, R. The second wave of synthetic biology: from modules to systems. Nat. Rev. Mol. Cell Biol. 10, 410–422 (2009).

    Article  PubMed  CAS  Google Scholar 

  4. Fischbach, M. A., Bluestone, J. A. & Lim, W. A. Cell-based therapeutics: the next pillar of medicine. Sci. Transl. Med. 5, 179ps7 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  5. Weinberg, B. H. et al. Large-scale design of robust genetic circuits with multiple inputs and outputs for mammalian cells. Nat. Biotechnol. 35, 453–462 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  6. Li, H. S. et al. Multidimensional control of therapeutic human cell function with synthetic gene circuits. Science 378, 1227–1234 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  7. Morsut, L. et al. Engineering Customized Cell Sensing and Response Behaviors Using Synthetic Notch Receptors. Cell 164, 780–791 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  8. Xie, M. et al. beta-cell-mimetic designer cells provide closed-loop glycemic control. Science 354, 1296–1301 (2016).

    Article  PubMed  CAS  Google Scholar 

  9. Roybal, K. T. et al. Precision Tumor Recognition by T Cells With Combinatorial Antigen-Sensing Circuits. Cell 164, 770–779 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  10. Toda, S. et al. Programming self-organizing multicellular structures with synthetic cell-cell signaling. Science 361, 156–162 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  11. Rossger, K., Charpin-El-Hamri, G. & Fussenegger, M. A closed-loop synthetic gene circuit for the treatment of diet-induced obesity in mice. Nat. Commun. 4, 2825 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  12. Saxena, P. et al. Synthetic gene network restoring endogenous pituitary-thyroid feedback control in experimental Graves’ disease. Proc. Natl. Acad. Sci. USA 113, 1244–1249 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  13. Schukur, L., Geering, B., Charpin-El Hamri, G. & Fussenegger, M. Implantable synthetic cytokine converter cells with AND-gate logic treat experimental psoriasis. Sci. Transl. Med. 7, 318ra201 (2015).

    Article  PubMed  Google Scholar 

  14. Kosti, P. et al. Hypoxia-sensing CAR T cells provide safety and efficacy in treating solid tumors. Cell Rep. Med. 2, 100227 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  15. Ye, H., Daoud-El Baba, M., Peng, R. W. & Fussenegger, M. A synthetic optogenetic transcription device enhances blood-glucose homeostasis in mice. Science 332, 1565–1568 (2011).

    Article  PubMed  CAS  Google Scholar 

  16. Del Vecchio, D., Dy, A. J. & Qian, Y. Control theory meets synthetic biology. J R Soc Interface. 13, 20160380 (2016).

  17. Filo, M., Chang, C. H. & Khammash, M. Biomolecular feedback controllers: from theory to applications. Curr. Opin. Biotechnol. 79, 102882 (2023).

    Article  PubMed  CAS  Google Scholar 

  18. Teixeira, A. P. & Fussenegger, M. Synthetic Gene Circuits for Regulation of Next-Generation Cell-Based Therapeutics. Adv. Sci. (Weinh.) 11, e2309088 (2024).

    PubMed  Google Scholar 

  19. Mansouri, M. & Fussenegger, M. Therapeutic cell engineering: designing programmable synthetic genetic circuits in mammalian cells. Protein Cell 13, 476–489 (2022).

    Article  PubMed  Google Scholar 

  20. Teixeira, A. P. & Fussenegger, M. Synthetic macromolecular switches for precision control of therapeutic cell functions. Nat. Rev. Bioeng. 2, 1005–1022 (2024).

    Article  CAS  Google Scholar 

  21. Kitada, T., DiAndreth, B., Teague, B. & Weiss, R. Programming gene and engineered-cell therapies with synthetic biology. Science. 359, eaad1067 (2018).

  22. Teng, F. et al. Programmable synthetic receptors: the next-generation of cell and gene therapies. Signal Transduct. Target Ther. 9, 7 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  23. Unal, G. & Fussenegger, M. Self-driving medicine: closed-loop therapeutics for autonomous disease control. Trends Biotechnol. (2026) online ahead of print.

  24. Aoki, S. K. et al. A universal biomolecular integral feedback controller for robust perfect adaptation. Nature 570, 533–537 (2019).

    Article  PubMed  CAS  Google Scholar 

  25. Frei, T. et al. A genetic mammalian proportional-integral feedback control circuit for robust and precise gene regulation. Proc. Natl. Acad. Sci. USA 119, e2122132119 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  26. Zhou, Y. et al. Optogenetics in medicine: innovations and therapeutic applications. Curr. Opin. Biotechnol. 92, 103262 (2025).

    Article  PubMed  CAS  Google Scholar 

  27. Krawczyk, K. et al. Electrogenetic cellular insulin release for real-time glycemic control in type 1 diabetic mice. Science 368, 993–1001 (2020).

    Article  PubMed  CAS  Google Scholar 

  28. Kemmer, C. et al. Self-sufficient control of urate homeostasis in mice by a synthetic circuit. Nat. Biotechnol. 28, 355–360 (2010).

    Article  PubMed  CAS  Google Scholar 

  29. Kim, T., Folcher, M., Charpin-El Hamri, G. & Fussenegger, M. A synthetic cGMP-sensitive gene switch providing Viagra((R))-controlled gene expression in mammalian cells and mice. Metab. Eng. 29, 169–179 (2015).

    Article  PubMed  CAS  Google Scholar 

  30. Wang, Y. et al. A versatile genetic control system in mammalian cells and mice responsive to clinically licensed sodium ferulate. Sci. Adv. 6, eabb9484 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  31. Li, H. S. et al. High-performance multiplex drug-gated CAR circuits. Cancer Cell 40, 1294–1305 e4 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  32. Stanley, S. A. et al. Radio-wave heating of iron oxide nanoparticles can regulate plasma glucose in mice. Science 336, 604–608 (2012).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  33. Ede, C., Chen, X., Lin, M. Y. & Chen, Y. Y. Quantitative Analyses of Core Promoters Enable Precise Engineering of Regulated Gene Expression in Mammalian Cells. ACS Synth. Biol. 5, 395–404 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  34. Lillacci, G., Benenson, Y. & Khammash, M. Synthetic control systems for high performance gene expression in mammalian cells. Nucleic Acids Res. 46, 9855–9863 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  35. Chen, W. C. W. et al. A synthetic transcription platform for programmable gene expression in mammalian cells. Nat. Commun. 13, 6167 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  36. Daringer, N. M., Dudek, R. M., Schwarz, K. A. & Leonard, J. N. Modular extracellular sensor architecture for engineering mammalian cell-based devices. ACS Synth. Biol. 3, 892–902 (2014).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  37. Schwarz, K. A., Daringer, N. M., Dolberg, T. B. & Leonard, J. N. Rewiring human cellular input-output using modular extracellular sensors. Nat. Chem. Biol. 13, 202–209 (2017).

    Article  PubMed  CAS  Google Scholar 

  38. Roybal, K. T. et al. Engineering T Cells with Customized Therapeutic Response Programs Using Synthetic Notch Receptors. Cell 167, 419–432 e16 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  39. Cho, J. H., Collins, J. J. & Wong, W. W. Universal Chimeric Antigen Receptors for Multiplexed and Logical Control of T Cell Responses. Cell 173, 1426–1438 e11 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  40. Piraner, D. I. et al. Engineered receptors for soluble cellular communication and disease sensing. Nature 638, 805–813 (2025).

    Article  PubMed  CAS  Google Scholar 

  41. Folcher, M. et al. Mind-controlled transgene expression by a wireless-powered optogenetic designer cell implant. Nat. Commun. 5, 5392 (2014).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  42. Shao, J. et al. Smartphone-controlled optogenetically engineered cells enable semiautomatic glucose homeostasis in diabetic mice. Sci Transl Med. 9, eaal2298 (2017).

  43. Huang, J. et al. An electrogenetic interface to program mammalian gene expression by direct current. Nat. Metab. 5, 1395–1407 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  44. Smole, A. et al. A Synthetic Mammalian Therapeutic Gene Circuit for Sensing and Suppressing Inflammation. Mol. Ther. 25, 102–119 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  45. Sedlmayer, F. et al. Designer cells programming quorum-sensing interference with microbes. Nat. Commun. 9, 1822 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  46. Teixeira, A. P. & Fussenegger, M. Synthetic biology-inspired therapies for metabolic diseases. Curr. Opin. Biotechnol. 47, 59–66 (2017).

    Article  PubMed  CAS  Google Scholar 

  47. Brooks, S. M. & Alper, H. S. Applications, challenges, and needs for employing synthetic biology beyond the lab. Nat. Commun. 12, 1390 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  48. Ye, H. et al. Self-adjusting synthetic gene circuit for correcting insulin resistance. Nat. Biomed. Eng. 1, 0005 (2017).

    Article  PubMed  CAS  Google Scholar 

  49. Khalil, A. S. & Collins, J. J. Synthetic biology: applications come of age. Nat. Rev. Genet. 11, 367–379 (2010).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  50. Weber, W. & Fussenegger, M. Emerging biomedical applications of synthetic biology. Nat. Rev. Genet. 13, 21–35 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  51. Huang, J. & Fussenegger, M. Programming mammalian cell behaviors by physical cues. Trends Biotechnol. 43, 16–42 (2025).

    Article  PubMed  CAS  Google Scholar 

  52. Gossen, M. & Bujard, H. Tight control of gene expression in mammalian cells by tetracycline-responsive promoters. Proc. Natl. Acad. Sci. USA 89, 5547–5551 (1992).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  53. Gitzinger, M. et al. The food additive vanillic acid controls transgene expression in mammalian cells and mice. Nucleic Acids Res. 40, e37 (2012).

    Article  PubMed  CAS  Google Scholar 

  54. Gitzinger, M. et al. Controlling transgene expression in subcutaneous implants using a skin lotion containing the apple metabolite phloretin. Proc. Natl. Acad. Sci. USA 106, 10638–10643 (2009).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  55. Weber, W. et al. Gas-inducible transgene expression in mammalian cells and mice. Nat. Biotechnol. 22, 1440–1444 (2004).

    Article  PubMed  CAS  Google Scholar 

  56. Bojar, D. et al. Caffeine-inducible gene switches controlling experimental diabetes. Nat. Commun. 9, 2318 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  57. Mansouri, M. et al. Smart-watch-programmed green-light-operated percutaneous control of therapeutic transgenes. Nat. Commun. 12, 3388 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  58. Zhou, Y. et al. A small and highly sensitive red/far-red optogenetic switch for applications in mammals. Nat. Biotechnol. 40, 262–272 (2022).

    Article  PubMed  CAS  Google Scholar 

  59. Gao, T. et al. Sonogenetics-controlled synthetic designer cells for cancer therapy in tumor mouse models. Cell Rep. Med. 5, 101513 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  60. Mahameed, M. et al. Nitroglycerin-responsive gene switch for the on-demand production of therapeutic proteins. Nat. Biomed. Eng. 9, 1129–1143 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  61. Qiao, L. et al. Engineered bacteria for near-infrared light-inducible expression of cancer therapeutics. Nat. Cancer 6, 612–628 (2025).

    Article  PubMed  CAS  Google Scholar 

  62. Nissim, L. et al. Synthetic RNA-Based Immunomodulatory Gene Circuits for Cancer Immunotherapy. Cell 171, 1138–1150 e15 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  63. Nims, R. J. et al. A synthetic mechanogenetic gene circuit for autonomous drug delivery in engineered tissues. Sci. Adv. 7, eabd9858 (2021).

  64. Jones, R. D. et al. Robust and tunable signal processing in mammalian cells via engineered covalent modification cycles. Nat. Commun. 13, 1720 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  65. Ma, Y. et al. Synthetic mammalian signaling circuits for robust cell population control. Cell 185, 967–979 e12.(2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  66. Hyrenius-Wittsten, A. et al. SynNotch CAR circuits enhance solid tumor recognition and promote persistent antitumor activity in mouse models. Sci Transl. Med. 13, eabd8836 (2021).

  67. Manhas, J., Edelstein, H. I., Leonard, J. N. & Morsut, L. The evolution of synthetic receptor systems. Nat. Chem. Biol. 18, 244–255 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  68. Scheller, L. et al. Generalized extracellular molecule sensor platform for programming cellular behavior. Nat. Chem. Biol. 14, 723–729 (2018).

    Article  PubMed  CAS  Google Scholar 

  69. Strittmatter, T. et al. Programmable DARPin-based receptors for the detection of thrombotic markers. Nat. Chem. Biol. 18, 1125–1134 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  70. Chen, Z. & Elowitz, M. B. Programmable protein circuit design. Cell 184, 2284–2301 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  71. Unal, G., Xie, Y. Q. & Fussenegger, M. A closed-loop cholesterol shunt controlling experimental dyslipidemia. Cell Metab. 37, 2066–2075 e3 (2025).

    Article  PubMed  CAS  Google Scholar 

  72. Galvan, S., Xie, Y. Q., Teixeira, A. P. & Fussenegger, M. Synthetic closed-loop gene circuit for phenylalanine regulation. Nucleic Acids Res. 53, gkaf1151 (2025).

  73. Ausländer, D. et al. A synthetic multifunctional mammalian pH sensor and CO2 transgene-control device. Mol. Cell 55, 397–408 (2014).

    Article  PubMed  Google Scholar 

  74. Unal, G. et al. A synthetic angiotensin II/ACE2-based hormone shunt controlling experimental hypertension. Nat. Commun. 17, 5116 (2026).

  75. Yang, L. et al. Engineering genetic devices for in vivo control of therapeutic T cell activity triggered by the dietary molecule resveratrol. Proc Natl. Acad. Sci. USA. 118, e2106612118 (2021).

  76. Hu, C. Y. & Murray, R. M. Layered feedback control overcomes performance trade-off in synthetic biomolecular networks. Nat. Commun. 13, 5393 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  77. Briat, C., Gupta, A. & Khammash, M. Antithetic Integral Feedback Ensures Robust Perfect Adaptation in Noisy Biomolecular Networks. Cell Syst. 2, 15–26 (2016).

    Article  PubMed  CAS  Google Scholar 

  78. Filo, M., Kumar, S. & Khammash, M. A hierarchy of biomolecular proportional-integral-derivative feedback controllers for robust perfect adaptation and dynamic performance. Nat. Commun. 13, 2119 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  79. Daly, A. B. et al. Fully automated closed-loop insulin delivery in adults with type 2 diabetes: an open-label, single-center, randomized crossover trial. Nat. Med. 29, 203–208 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  80. Boughton, C. K. & Hovorka, R. New closed-loop insulin systems. Diabetologia 64, 1007–1015 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  81. Gao, X. et al. Designer probiotic-based living drugs for uric acid homeostasis control in hyperuricemic mice and rats. Cell Rep. Med. 6, 102379 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  82. Wilkinson, S. P. & Grove, A. HucR, a novel uric acid-responsive member of the MarR family of transcriptional regulators from Deinococcus radiodurans. J. Biol. Chem. 279, 51442–51450 (2004).

    Article  PubMed  CAS  Google Scholar 

  83. Rossignoli, F., Hoffman, D., Atif, E. & Shah, K. Developing and characterizing a two-layered safety switch for cell therapies. Cancer Biol. Ther. 24, 2232146 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  84. Mecacci, S., Torregrosa-Barragán, L., Asin-Garcia, E. & Smith, R. W. Multilayered safety framework for living diagnostics in the colon. Front Syst. Biol. 3, 1240040 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  85. Park, H. B. et al. Improved safety of chimeric antigen receptor T cells indirectly targeting antigens via switchable adapters. Nat. Commun. 15, 9917 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  86. Bouquet, L. et al. Correction: RapaCaspase-9-based suicide gene applied to the safety of IL-1RAP CAR-T cells. Gene Ther. 32, 299 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  87. Bouquet, L. et al. RapaCaspase-9-based suicide gene applied to the safety of IL-1RAP CAR-T cells. Gene Ther. 30, 706–713 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  88. Xiao, W. et al. FGFR4-specific CAR-T cells with inducible caspase-9 suicide gene as an approach to treat rhabdomyosarcoma. Cancer Gene Ther. 31, 1571–1584 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  89. Fedorov, V. D., Themeli, M. & Sadelain, M. PD-1- and CTLA-4-based inhibitory chimeric antigen receptors (iCARs) divert off-target immunotherapy responses. Sci. Transl. Med. 5, 215ra172 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

  90. Srivastava, S. et al. Logic-Gated ROR1 Chimeric Antigen Receptor Expression Rescues T Cell-Mediated Toxicity to Normal Tissues and Enables Selective Tumor Targeting. Cancer Cell 35, 489–503 e8 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  91. Angelici, B. et al. An AAV gene therapy computes over multiple cellular inputs to enable precise targeting of multifocal hepatocellular carcinoma in mice. Sci. Transl. Med. 13, eabh4456 (2021).

    Article  PubMed  CAS  Google Scholar 

  92. Xie, Z. et al. Multi-input RNAi-based logic circuit for identification of specific cancer cells. Science 333, 1307–1311 (2011).

    Article  PubMed  CAS  Google Scholar 

  93. Son, H. I., Weiss, A. & You, L. Design patterns for engineering genetic stability. Curr Opin Biomed. Eng. 19, 100297 (2021).

  94. Xie, M. & Fussenegger, M. Designing cell function: assembly of synthetic gene circuits for cell biology applications. Nat. Rev. Mol. Cell Biol. 19, 507–525 (2018).

    Article  PubMed  CAS  Google Scholar 

  95. Cubillos-Ruiz, A. et al. Engineering living therapeutics with synthetic biology. Nat. Rev. Drug Discov. 20, 941–960 (2021).

    Article  PubMed  CAS  Google Scholar 

  96. Black, J. B., Perez-Pinera, P. & Gersbach, C. A. Mammalian Synthetic Biology: Engineering Biological Systems. Annu Rev. Biomed. Eng. 19, 249–277 (2017).

    Article  PubMed  CAS  Google Scholar 

  97. Nielsen, J. & Keasling, J. D. Engineering Cellular Metabolism. Cell 164, 1185–1197 (2016).

    Article  PubMed  CAS  Google Scholar 

  98. Zhao, N. et al. Synthetic biology-inspired cell engineering in diagnosis, treatment, and drug development. Signal Transduct. Target Ther. 8, 112 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  99. Zhu, B. et al. Synthetic biology approaches for improving the specificity and efficacy of cancer immunotherapy. Cell Mol. Immunol. 21, 436–447 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  100. Tan, X., Letendre, J. H., Collins, J. J. & Wong, W. W. Synthetic biology in the clinic: engineering vaccines, diagnostics, and therapeutics. Cell 184, 881–898 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  101. Yin, J. et al. A green tea-triggered genetic control system for treating diabetes in mice and monkeys. Sci. Transl. Med. 11, eaav8826 (2019).

  102. Xue, S. et al. A Synthetic-Biology-Inspired Therapeutic Strategy for Targeting and Treating Hepatogenous Diabetes. Mol. Ther. 25, 443–455 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  103. Zhang, Y. et al. A synthetic hyperglycemia-sensing gene circuit enhances blood glucose homeostasis in diabetic mice. Mol. Ther. 34, 2236–2255 (2026).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  104. Guan, N. et al. Glucose-responsive probiotics for glycaemic modulation in mice and monkeys. Nature 657, 477–486 (2026).

    Article  PubMed  CAS  Google Scholar 

  105. Pferdehirt, L. et al. A synthetic chronogenetic gene circuit for programmed circadian drug delivery. Nat. Commun. 16, 1457 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  106. Franko, N. et al. Regulation of therapeutic protein release in response to circadian biomarkers. Nat. Commun. 16, 9812 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  107. Cimino, A. et al. Programmable chronogenetic gene circuits for self-regulated circadian delivery of biologic drugs. J. Control Release 385, 113959 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  108. Cimino, A. et al. Dual-responsive synthetic gene circuit for dynamic biologic drug delivery via inflammatory and circadian signaling pathways. J. Biol. Eng. 19, 47 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  109. Boughton, C. K. et al. Fully automated closed-loop glucose control compared with standard insulin therapy in adults with type 2 diabetes requiring dialysis: an open-label, randomized crossover trial. Nat. Med. 27, 1471–1476 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  110. Lal, R. A., Ekhlaspour, L., Hood, K. & Buckingham, B. Realizing a Closed-Loop (Artificial Pancreas) System for the Treatment of Type 1 Diabetes. Endocr. Rev. 40, 1521–1546 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  111. Zou, Z. P. et al. Biomarker-responsive engineered probiotic diagnoses, records, and ameliorates inflammatory bowel disease in mice. Cell Host Microbe 31, 199–212 e5 (2023).

    Article  PubMed  CAS  Google Scholar 

  112. Buss, M. T. et al. Probiotic acoustic biosensors for noninvasive imaging of gut inflammation. Nat. Commun. 16, 7931 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  113. Chua, K. J. et al. An Engineered Probiotic Produces a Type III Interferon IFNL1 and Reduces Inflammations in in vitro Inflammatory Bowel Disease Models. ACS Biomater. Sci. Eng. 9, 5123–5135 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  114. Ye, M. et al. Engineered Bacteroides thetaiotaomicron sense gut inflammation and deliver therapeutic molecules to alleviate colitis in mice. Nat. Microbiol. 11, 2294–2309 (2026).

    Article  PubMed  CAS  Google Scholar 

  115. Gecse, K. B. & Vermeire, S. Differential diagnosis of inflammatory bowel disease: imitations and complications. Lancet Gastroenterol. Hepatol. 3, 644–653 (2018).

    Article  PubMed  Google Scholar 

  116. Mimee, M. et al. An ingestible bacterial-electronic system to monitor gastrointestinal health. Sci. (NY) 360, 915–918 (2018).

    Article  CAS  Google Scholar 

  117. Ge, C. et al. Engineered living glues secrete therapeutic proteins for treatment of inflammatory bowel disease. Nat. Biotechnol. (2026).

  118. Konikoff, M. R. & Denson, L. A. Role of fecal calprotectin as a biomarker of intestinal inflammation in inflammatory bowel disease. Inflamm. Bowel Dis. 12, 524–534 (2006).

    Article  PubMed  Google Scholar 

  119. Walsham, N. E. & Sherwood, R. A. Fecal calprotectin in inflammatory bowel disease. Clin. Exp. Gastroenterol. 9, 21–29 (2016).

    PubMed  PubMed Central  CAS  Google Scholar 

  120. Zhu, D. et al. Microbial Biosensor for Sensing and Treatment of Intestinal Inflammation. Adv. Sci. (Weinh., Baden.-Wurtt., Ger.) 12, e2504364 (2025).

    Google Scholar 

  121. Weibel, N. et al. Engineering a Novel Probiotic Toolkit in Escherichia coli Nissle 1917 for Sensing and Mitigating Gut Inflammatory Diseases. ACS Synth. Biol. 13, 2376–2390 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  122. Pferdehirt, L., Ross, A. K., Brunger, J. M. & Guilak, F. A Synthetic Gene Circuit for Self-Regulating Delivery of Biologic Drugs in Engineered Tissues. Tissue Eng. Part A 25, 809–820 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  123. Sedlmayer, F., Jaeger, T., Jenal, U. & Fussenegger, M. Quorum-Quenching Human Designer Cells for Closed-Loop Control of Pseudomonas aeruginosa Biofilms. Nano Lett. 17, 5043–5050 (2017).

    Article  PubMed  CAS  Google Scholar 

  124. Liu, Y. et al. Immunomimetic Designer Cells Protect Mice from MRSA Infection. Cell 174, 259–270.e11 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  125. Wang, Y. et al. Engineering antiviral immune-like systems for autonomous virus detection and inhibition in mice. Nat. Commun. 13, 7629 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  126. Mao, N., Cubillos-Ruiz, A., Cameron, D. E. & Collins, J. J. Probiotic strains detect and suppress cholera in mice. Science Translational Medicine. 10, eaao2586 (2018).

  127. Holowko, M. B., Wang, H., Jayaraman, P. & Poh, C. L. Biosensing Vibrio cholerae with Genetically Engineered Escherichia coli. ACS Synth. Biol. 5, 1275–1283 (2016).

    Article  PubMed  CAS  Google Scholar 

  128. Jayaraman, P. et al. Repurposing a Two-Component System-Based Biosensor for the Killing of Vibrio cholerae. ACS Synth. Biol. 6, 1403–1415 (2017).

    Article  PubMed  CAS  Google Scholar 

  129. Jurado-Martín, I., Sainz-Mejías, M. & McClean, S. Pseudomonas aeruginosa: An Audacious Pathogen with an Adaptable Arsenal of Virulence Factors. Int. J Mol. Sci. 22, 3128 (2021).

  130. Gupta, S., Bram, E. E. & Weiss, R. Genetically programmable pathogen sense and destroy. ACS Synth. Biol. 2, 715–723 (2013).

    Article  PubMed  CAS  Google Scholar 

  131. Rossger, K., Charpin-El Hamri, G. & Fussenegger, M. Reward-based hypertension control by a synthetic brain-dopamine interface. Proc. Natl. Acad. Sci. USA 110, 18150–18155 (2013).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  132. Si, Y. et al. Engineering mammalian cells for detection and treatment of cardiac injury. Mol. Syst. Biol. 22, 88–101 (2026).

    Article  PubMed  CAS  Google Scholar 

  133. Koirala, P. et al. A closed-loop cell therapy engineered to autonomously secrete Activin A inhibitor protects from fi brodysplasia ossifi cans progressiva. iScience 29, 116602 (2026).

  134. Junttila, M. R. & de Sauvage, F. J. Influence of tumour micro-environment heterogeneity on therapeutic response. Nature 501, 346–354 (2013).

    Article  PubMed  CAS  Google Scholar 

  135. Hanahan, D. Hallmarks of Cancer: New Dimensions. Cancer Discov. 12, 31–46 (2022).

    Article  PubMed  CAS  Google Scholar 

  136. Labanieh, L. et al. Enhanced safety and efficacy of protease-regulated CAR-T cell receptors. Cell 185, 1745–1763.e22 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  137. Choe, J. H. et al. SynNotch-CAR T cells overcome challenges of specificity, heterogeneity, and persistence in treating glioblastoma. Sci. Transl. Med. 13, (2021).

  138. Wu, C. Y. et al. Remote control of therapeutic T cells through a small molecule-gated chimeric receptor. Science 350, aab4077 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

  139. June, C. H. et al. CAR T cell immunotherapy for human cancer. Science 359, 1361–1365 (2018).

    Article  PubMed  CAS  Google Scholar 

  140. Baeuerle, P. A. et al. Synthetic TRuC receptors engaging the complete T cell receptor for potent anti-tumor response. Nat. Commun. 10, 2087 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  141. Edelstein, H. I. et al. Conversion of natural cytokine receptors into orthogonal synthetic biosensors. Nat. Chem. Biol. 21, 1719–1730 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  142. Engelowski, E. et al. Synthetic cytokine receptors transmit biological signals using artificial ligands. Nat. Commun. 9, 2034 (2018).

    Article  PubMed  PubMed Central  Google Scholar 

  143. Lu, G. et al. In vivo detection of immune responses via cytokine activity labeling. Cell 189, 939–955.e26 (2026).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  144. Kipniss, N. H. et al. Engineering cell sensing and responses using a GPCR-coupled CRISPR-Cas system. Nat. Commun. 8, 2212 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  145. Kalogriopoulos, N. A. et al. Synthetic GPCRs for programmable sensing and control of cell behaviour. Nature 637, 230–239 (2025).

    Article  PubMed  CAS  Google Scholar 

  146. Allen, G. M. et al. Synthetic cytokine circuits that drive T cells into immune-excluded tumors. Science 378, eaba1624 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  147. Hernandez-Lopez, R. A. et al. T cell circuits that sense antigen density with an ultrasensitive threshold. Science 371, 1166–1171 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  148. Reddy, N. R. et al. Engineering synthetic suppressor T cells that execute locally targeted immunoprotective programs. Science 386, eadl4793 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  149. Simic, M. S. et al. Programming tissue-sensing T cells that deliver therapies to the brain. Science 386, eadl4237 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  150. Bjelajac, J. R. et al. Cell-autonomous control of CAR signaling and receptor shedding via ADAM17-mediated proteolysis. Cell 189, 3883–3902.e23 (2026).

  151. Lim, W. A. & June, C. H. The Principles of Engineering Immune Cells to Treat Cancer. Cell 168, 724–740 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  152. Sadelain, M., Riviere, I. & Riddell, S. Therapeutic T cell engineering. Nature 545, 423–431 (2017).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  153. Chen, A. X. Y. et al. Rewiring endogenous genes in CAR T cells for tumour-restricted payload delivery. Nature 644, 241–251 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  154. Huang, X. et al. Bacteria-Based Cancer Immunotherapy. Adv. Sci. (Weinh.) 8, 2003572 (2021).

    PubMed  PubMed Central  CAS  Google Scholar 

  155. Yi, X. et al. Bacteria-triggered tumor-specific thrombosis to enable potent photothermal immunotherapy of cancer. Sci. Adv. 6, eaba3546 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  156. Zhou, S., Gravekamp, C., Bermudes, D. & Liu, K. Tumour-targeting bacteria engineered to fight cancer. Nat. Rev. Cancer 18, 727–743 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  157. Chen, Y. et al. Genetically engineered oncolytic bacteria as drug delivery systems for targeted cancer theranostics. Acta Biomater. 124, 72–87 (2021).

    Article  PubMed  CAS  Google Scholar 

  158. Niu, L. et al. Engineering oncolytic bacteria as precision cancer therapeutics: design principles, therapeutic strategies, and translational perspectives. Protein Cell 17, 279–303 (2026).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  159. Deng, Z. et al. Programmable living therapeutics for cancer immunotherapy. Cell Chem. Biol. 33, 597–622 (2026).

    Article  PubMed  CAS  Google Scholar 

  160. Xie, X. T. et al. Programmable engineered bacteria as sustained-releasing antibody factory in situ for enhancing tumor immune checkpoint therapy. Sci. Adv. 11, eadt7298 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  161. Leventhal, D. S. et al. Immunotherapy with engineered bacteria by targeting the STING pathway for anti-tumor immunity. Nat. Commun. 11, 2739 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  162. Chang, Z. et al. Bacterial immunotherapy leveraging IL-10R hysteresis for both phagocytosis evasion and tumor immunity revitalization. Cell 188, 1842–1857.e20 (2025).

    Article  PubMed  CAS  Google Scholar 

  163. Qin, W. et al. Bacteria-Elicited Specific Thrombosis Utilizing Acid-Induced Cytolysin A Expression to Enable Potent Tumor Therapy. Adv. Sci. (Weinh.) 9, e2105086 (2022).

    PubMed  PubMed Central  Google Scholar 

  164. Chien, T. et al. Enhancing the tropism of bacteria via genetically programmed biosensors. Nat. Biomed. Eng. 6, 94–104 (2022).

    Article  PubMed  CAS  Google Scholar 

  165. Zúñiga, A. et al. Engineered l-Lactate Responding Promoter System Operating in Glucose-Rich and Anoxic Environments. ACS Synth. Biol. 10, 3527–3536 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

  166. Zou, Z. P. et al. Self-Adjusting Engineered Probiotic for Targeted Tumor Colonization and Local Therapeutics Delivery. Adv. Sci. (Weinh.) 12, e06486 (2025).

    PubMed  PubMed Central  CAS  Google Scholar 

  167. Zhan, H. et al. Reprogramming the tumor microenvironment with c-MYC-based gene circuit platform to enhance specific cancer immunotherapy. Nat. Commun. 16, 7983 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  168. Chung, H. K. et al. A compact synthetic pathway rewires cancer signaling to therapeutic effector release. Science. 364, eaat6982 (2019).

  169. Senn, G. V., Nissen, L. & Benenson, Y. Synthetic gene circuits that selectively target RAS-driven cancers. Elife. 14, RP104320 (2026).

  170. North, K. et al. Synthetic introns enable splicing factor mutation-dependent targeting of cancer cells. Nat. Biotechnol. 40, 1103–1113 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  171. Mallozzi, A., Fusco, V., Ragazzini, F. & di Bernardo, D. A Biomolecular Circuit for Automatic Gene Regulation in Mammalian Cells with CRISPR Technology. ACS Synth. Biol. 13, 3917–3925 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  172. Bose, S. et al. A retrievable implant for the long-term encapsulation and survival of therapeutic xenogeneic cells. Nat. Biomed. Eng. 4, 814–826 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  173. Taghdiri, M. & Mussolino, C. Viral and Non-Viral Systems to Deliver Gene Therapeutics to Clinical Targets. Int. J Mol. Sci. 25, (2024).

  174. Kohn, D. B., Chen, Y. Y. & Spencer, M. J. Successes and challenges in clinical gene therapy. Gene Ther. 30, 738–746 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  175. ClinicalTrials.gov. Anti-EGFRvIII synNotch Receptor Induced Anti-EphA2/IL-13Ralpha2 CAR (E-SYNC) T Cells, https://clinicaltrials.gov/ct2/show/NCT06186401 (2026).

  176. Huang, G., Chen, X. & Liao, C. AI-Driven Wearable Bioelectronics in Digital Healthcare. Biosensors (Basel). 15, 410 (2025).

  177. Khoshfekr Rudsari, H. et al. Digital twins in healthcare: a comprehensive review and future directions. Front Digit Health 7, 1633539 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  178. Bunne, C. et al. How to build the virtual cell with artificial intelligence: Priorities and opportunities. Cell 187, 7045–7063 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  179. Qian, L., Dong, Z. & Guo, T. Grow AI virtual cells: three data pillars and closed-loop learning. Cell Res. 35, 319–321 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  180. Harkos, C. et al. Using mathematical modelling and AI to improve delivery and efficacy of therapies in cancer. Nat. Rev. Cancer 25, 324–340 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  181. Yin, J., Ma, X., Hu, L. & Ye, H. Translating synthetic gene circuits into the clinic: Challenges, opportunities, and future directions. Cell Syst. 16, 101425 (2025).

    Article  PubMed  CAS  Google Scholar 

  182. Jonathan Rivnay, R. R. et al. Integrating bioelectronics with cell-based synthetic biology. Nat. Rev. Bioeng. 3, 317–332 (2025).

    Article  Google Scholar 

  183. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  184. Varadi, M. et al. AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res. 50, p-D4–p-D44 (2022).

    Article  Google Scholar 

  185. Rives, A. et al. Biological structure and function emerge from scaling unsupervised learning to 250 million protein sequences. Proc Natl. Acad. Sci. USA. 118, (2021).

  186. Lin, Z. et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 379, 1123–1130 (2023).

    Article  PubMed  CAS  Google Scholar 

  187. Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  188. Watson, J. L. et al. De novo design of protein structure and function with RFdiffusion. Nature 620, 1089–1100 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  189. Dauparas, J. et al. Robust deep learning-based protein sequence design using ProteinMPNN. Science 378, 49–56 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  190. Wang, J. et al. Scaffolding protein functional sites using deep learning. Science 377, 387–394 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  191. Bennett, N. R. et al. Improving de novo protein binder design with deep learning. Nat. Commun. 14, 2625 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  192. Park, J. et al. De novo design of a homo-trimeric amantadine-binding protein. Elife. 14, 2625 (2019).

  193. Jin, Q. et al. De novo design of small molecule-regulated protein oligomers. Science 8, e47839 (2026).

    Article  PubMed  CAS  Google Scholar 

  194. Kang, S. et al. COMBINES-CID: An Efficient Method for De Novo Engineering of Highly Specific Chemically Induced Protein Dimerization Systems. J. Am. Chem. Soc. 141, 10948–10952 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  195. Spencer, D. M., Wandless, T. J., Schreiber, S. L. & Crabtree, G. R. Controlling signal transduction with synthetic ligands. Science 262, 1019–1024 (1993).

    Article  PubMed  CAS  Google Scholar 

  196. Stanton, B. Z., Chory, E. J. & Crabtree, G. R. Chemically induced proximity in biology and medicine. Science. 359, eaao5902 (2018).

  197. Gil, A. A. et al. Optogenetic control of protein binding using light-switchable nanobodies. Nat. Commun. 11, 4044 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  198. Bashir, A. et al. Machine learning guided aptamer refinement and discovery. Nat. Commun. 12, 2366 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  199. Nielsen, A. A. et al. Genetic circuit design automation. Science 352, aac7341 (2016).

    Article  PubMed  Google Scholar 

  200. Radivojevic, T., Costello, Z., Workman, K. & Garcia Martin, H. A machine learning Automated Recommendation Tool for synthetic biology. Nat. Commun. 11, 4879 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  201. Merzbacher, C., Mac Aodha, O. & Oyarzun, D. A. Bayesian Optimization for Design of Multiscale Biological Circuits. ACS Synth. Biol. 12, 2073–2082 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  202. Lawson, C. E. et al. Machine learning for metabolic engineering: A review. Metab. Eng. 63, 34–60 (2021).

    Article  PubMed  CAS  Google Scholar 

  203. Carbonell, P. et al. Bioinformatics for the synthetic biology of natural products: integrating across the Design-Build-Test cycle. Nat. Prod. Rep. 33, 925–932 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  204. Daniels, K. G. et al. Decoding CAR T cell phenotype using combinatorial signaling motif libraries and machine learning. Science 378, 1194–1200 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  205. Qiu, S. et al. CAR-Toner: an AI-driven approach for CAR tonic signaling prediction and optimization. Cell Res. 34, 386–388 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

  206. Ansari, M. S. et al. AI-guided CAR designs and targeted pathway modulation to enhance multi-antigen CAR T cell durability and overcome antigen escape. Nat. Commun. 17, 1549 (2026).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  207. Yang, J. et al. Advancing CAR T cell therapy through the use of multidimensional omics data. Nat. Rev. Clin. Oncol. 20, 211–228 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  208. Raissi, M., Perdikaris, P. & Karniadakis, G. E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J. Comput Phys. 378, 686–707 (2019).

    Article  Google Scholar 

  209. Karniadakis, G. E. et al. Physics-informed machine learning. Nat. Rev. Phys. 3, 422–440 (2021).

    Article  Google Scholar 

  210. Danhof, M. et al. Mechanism-based pharmacokinetic-pharmacodynamic modeling: biophase distribution, receptor theory, and dynamical systems analysis. Annu Rev. Pharm. Toxicol. 47, 357–400 (2007).

    Article  CAS  Google Scholar 

  211. Mager, D. E., Wyska, E. & Jusko, W. J. Diversity of mechanism-based pharmacodynamic models. Drug Metab. Dispos. 31, 510–518 (2003).

    Article  PubMed  CAS  Google Scholar 

  212. Katsoulakis, E. et al. Digital twins for health: a scoping review. npj Digit Med. 7, 77 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

  213. Yang, S. et al. Mechanistic data-informed multiscale quantitative systems pharmacology modeling framework enables the clinical translation and efficacy assessment of CAR-T therapy in solid tumors. J. Immunother. Cancer 13, e012331 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  214. Phillip, M. et al. Consensus recommendations for the use of automated insulin delivery technologies in clinical practice. Endocr. Rev. 44, 254–280 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  215. Mosquera-Lopez, C. et al. Enabling fully automated insulin delivery through meal detection and size estimation using artificial intelligence. npj Digit Med. 6, 39 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  216. Viceconti, M. et al. In silico trials: verification, validation and uncertainty quantification of predictive models used in the regulatory evaluation of biomedical products. Methods 185, 120–127 (2021).

    Article  PubMed  CAS  Google Scholar 

  217. Liu, X. et al. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Nat. Med. 26, 1364–1374 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  218. Vasey, B. et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat. Med. 28, 924–933 (2022).

    Article  PubMed  CAS  Google Scholar 

  219. International Medical Device Regulators Forum Artificial Intelligence/Machine Learning-enabled Working, G. Good machine learning practice for medical device development: guiding principles. (2025).

  220. Rajkomar, A., Dean, J. & Kohane, I. Machine learning in medicine. N. Engl. J. Med. 380, 1347–1358 (2019).

    Article  PubMed  Google Scholar 

  221. Kelly, C. J. et al. Key challenges for delivering clinical impact with artificial intelligence. BMC Med. 17, 195 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  222. Walsh, I. et al. DOME: recommendations for supervised machine learning validation in biology. Nat. Methods 18, 1122–1127 (2021).

    Article  PubMed  CAS  Google Scholar 

  223. Kapoor, S. & Narayanan, A. Leakage and the reproducibility crisis in machine-learning-based science. Patterns 4, 100804 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  224. Vamathevan, J. et al. Applications of machine learning in drug discovery and development. Nat. Rev. Drug Discov. 18, 463–477 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  225. Scannell, J. W. et al. Predictive validity in drug discovery: what it is, why it matters and how to improve it. Nat. Rev. Drug Discov. 21, 915–931 (2022).

    Article  PubMed  CAS  Google Scholar 

  226. Whalen, S., Schreiber, J., Noble, W. S. & Pollard, K. S. Navigating the pitfalls of applying machine learning in genomics. Nat. Rev. Genet. 23, 169–181 (2022).

    Article  PubMed  CAS  Google Scholar 

  227. Cheng, C. et al. A General Primer for Data Harmonization. Sci. Data 11, 152 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

  228. Alber, M. et al. Integrating machine learning and multiscale modeling-perspectives, challenges, and opportunities in the biological, biomedical, and behavioral sciences. NPJ Digit Med. 2, 115 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

  229. Acosta, J. N., Falcone, G. J., Rajpurkar, P. & Topol, E. J. Multimodal biomedical AI. Nat. Med. 28, 1773–1784 (2022).

    Article  PubMed  CAS  Google Scholar 

  230. Pahud de Mortanges, A. et al. Orchestrating explainable artificial intelligence for multimodal and longitudinal data in medical imaging. NPJ Digit Med. 7, 195 (2024).

    Article  PubMed  PubMed Central  Google Scholar 

  231. Wilkinson, M. D. et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci. Data 3, 160018 (2016).

    Article  PubMed  PubMed Central  Google Scholar 

  232. Matschinske, J. et al. The AIMe registry for artificial intelligence in biomedical research. Nat. Methods 18, 1128–1131 (2021).

    Article  PubMed  CAS  Google Scholar 

  233. Heo, Y. B., Park, J. S. & Woo, H. M. Architectures of emerging biofoundry platforms for synthetic biology. Curr. Opin. Biotechnol. 96, 103379 (2025).

    Article  PubMed  CAS  Google Scholar 

  234. Zhang, Q. et al. Integrating protein language models and automatic biofoundry for enhanced protein evolution. Nat. Commun. 16, 1553 (2025).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  235. Singh, N. et al. A generalized platform for artificial intelligence-powered autonomous enzyme engineering. Nat. Commun. 16, 5648 (2025).

    Article  PubMed  PubMed Central  Google Scholar 

  236. Longo, S. K., Guo, M. G., Ji, A. L. & Khavari, P. A. Integrating single-cell and spatial transcriptomics to elucidate intercellular tissue dynamics. Nat. Rev. Genet. 22, 627–644 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  237. Dunn, J. et al. Wearable sensors enable personalized predictions of clinical laboratory measurements. Nat. Med. 27, 1105–1112 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  238. Moor, M. et al. Foundation models for generalist medical artificial intelligence. Nature 616, 259–265 (2023).

    Article  PubMed  CAS  Google Scholar 

  239. Zhang, K. et al. Artificial intelligence in drug development. Nat. Med. 31, 45–59 (2025).

    Article  PubMed  CAS  Google Scholar 

  240. Food, U. S. & Drug, A. Marketing submission recommendations for a predetermined change control plan for artificial intelligence-enabled device software functions: guidance for industry and Food and Drug Administration staff (2025).

  241. Food, U. S. & Drug, A. Cybersecurity in medical devices: quality management system considerations and content of premarket submissions: guidance for industry and Food and Drug Administration staff (2026).

  242. Lienert, F., Lohmueller, J. J., Garg, A. & Silver, P. A. Synthetic biology in mammalian cells: next generation research tools and therapeutics. Nat. Rev. Mol. Cell Biol. 15, 95–107 (2014).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  243. Yan, X., Liu, X., Zhao, C. & Chen, G. Q. Applications of synthetic biology in medical and pharmaceutical fields. Signal Transduct. Target Ther. 8, 199 (2023).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  244. Rafiq, S., Hackett, C. S. & Brentjens, R. J. Engineering strategies to overcome the current roadblocks in CAR T cell therapy. Nat. Rev. Clin. Oncol. 17, 147–167 (2020).

    Article  PubMed  Google Scholar 

  245. Bulcha, J. T. et al. Viral vector platforms within the gene therapy landscape. Signal Transduct. Target Ther. 6, 53 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  246. Costa Verdera, H., Kuranda, K. & Mingozzi, F. AAV Vector Immunogenicity in Humans: A Long Journey to Successful Gene Transfer. Mol. Ther. 28, 723–746 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  247. Li, C. & Samulski, R. J. Engineering adeno-associated virus vectors for gene therapy. Nat. Rev. Genet. 21, 255–272 (2020).

    Article  PubMed  CAS  Google Scholar 

  248. Deuse, T. et al. Hypoimmunogenic derivatives of induced pluripotent stem cells evade immune rejection in fully immunocompetent allogeneic recipients. Nat. Biotechnol. 37, 252–258 (2019).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  249. Raguram, A., Banskota, S. & Liu, D. R. Therapeutic in vivo delivery of gene editing agents. Cell 185, 2806–2827 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  250. Bryant, D. H. et al. Deep diversification of an AAV capsid protein by machine learning. Nat. Biotechnol. 39, 691–696 (2021).

    Article  PubMed  CAS  Google Scholar 

  251. Vegas, A. J. et al. Long-term glycemic control using polymer-encapsulated human stem cell-derived beta cells in immune-competent mice. Nat. Med. 22, 306–311 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  252. Bochenek, M. A. et al. Alginate encapsulation as long-term immune protection of allogeneic pancreatic islet cells transplanted into the omental bursa of macaques. Nat. Biomed. Eng. 2, 810–821 (2018).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  253. Campochiaro, P. A. et al. Gene therapy for neovascular age-related macular degeneration by subretinal delivery of RGX-314: a phase 1/2a dose-escalation study. Lancet 403, 1563–1573 (2024).

    Article  PubMed  CAS  Google Scholar 

  254. Depil, S. et al. ‘Off-the-shelf’ allogeneic CAR T cells: development and challenges. Nat. Rev. Drug Discov. 19, 185–199 (2020).

    Article  PubMed  CAS  Google Scholar 

  255. Frei, T. et al. Characterization and mitigation of gene expression burden in mammalian cells. Nat. Commun. 11, 4641 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  256. Di Blasi, R. et al. Resource-aware construct design in mammalian cells. Nat. Commun. 14, 3576 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  257. Shakiba, N., Jones, R. D., Weiss, R. & Del Vecchio, D. Context-aware synthetic biology by controller design: Engineering the mammalian cell. Cell Syst. 12, 561–592 (2021).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  258. Labanieh, L. & Mackall, C. L. CAR immune cells: design principles, resistance and the next generation. Nature 614, 635–648 (2023).

    Article  PubMed  CAS  Google Scholar 

  259. Bashor, C. J. et al. Engineering the next generation of cell-based therapeutics. Nat. Rev. Drug Discov. 21, 655–675 (2022).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  260. Di Stasi, A. et al. Inducible apoptosis as a safety switch for adoptive cell therapy. N. Engl. J. Med. 365, 1673–1683 (2011).

    Article  PubMed  PubMed Central  Google Scholar 

  261. Elsallab, M. & Maus, M. V. Expanding access to CAR T cell therapies through local manufacturing. Nat. Biotechnol. 41, 1698–1708 (2023).

    Article  PubMed  CAS  Google Scholar 

  262. Kliegman, M. et al. A roadmap for affordable genetic medicines. Nature 634, 307–314 (2024).

    Article  PubMed  CAS  Google Scholar 

  263. Karr, J. R. et al. A whole-cell computational model predicts phenotype from genotype. Cell 150, 389–401 (2012).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  264. Bai, P. et al. A synthetic biology-based device prevents liver injury in mice. J. Hepatol. 65, 84–94 (2016).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  265. Brunger, J. M. et al. Genome Engineering of Stem Cells for Autonomously Regulated, Closed-Loop Delivery of Biologic Drugs. Stem. Cell Rep. 8, 1202–1213 (2017).

Download references

Acknowledgments

We thank Jianli Yin for generous assistance in figure design. We also acknowledge the use of ChatGPT (OpenAI) for language editing and proofreading.

Funding

Open access funding provided by Swiss Federal Institute of Technology Zurich.

Author information

Authors and Affiliations

  1. School of Life Sciences, East China Normal University, Shanghai, 200241, China

    Haifeng Ye

  2. Department of Biosystems Science and Engineering, ETH Zurich, Basel, 4056, Switzerland

    Martin Fussenegger

  3. Faculty of Science, University of Basel, Basel, 4056, Switzerland

    Martin Fussenegger

Authors

  1. Haifeng Ye
  2. Martin Fussenegger

Contributions

H.Y. conducted the literature review, curated the relevant literature, prepared the figures and wrote the original draft. M.F. conceptualized and supervised the work and critically reviewed and edited the manuscript. All authors have read and approved the article.

Corresponding authors

Correspondence to Haifeng Ye or Martin Fussenegger.

Ethics declarations

Competing interests

Martin Fussenegger is one of the Associate Editors of Signal Transduction and Targeted Therapy, and Haifeng Ye is one of the Editorial Board Members of Signal Transduction and Targeted Therapy, but they have not been involved in the process of the manuscript handling.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Ye, H., Fussenegger, M. Engineering signal transduction for closed-loop gene and cell therapies. Sig Transduct Target Ther 11, 515 (2026). https://doi.org/10.1038/s41392-026-03122-4

Download citation

  • Received: 12 April 2026

  • Revised: 03 October 2026

  • Accepted: 14 October 2026

  • Published: 17 November 2026

  • Version of record: 17 November 2026

  • DOI: https://doi.org/10.1038/s41392-026-03122-4

来源:Nature 健康护理主题 · nature.com