Evolution of molecular switches for regulation of transgene expression by clinically licensed gluconate

Abstract Synthetic biology holds great promise to improve the safety and efficacy of future gene and engineered cell therapies by providing new means of endogenous or exogenous control of the embedded therapeutic programs. Here, we focused on gluconate as a clinically licensed small-molecule inducer and engineered gluconate-sensitive molecular switches to regulate transgene expression in human cell cultures and in mice. Several switch designs were assembled based on the gluconate-responsive transcriptional repressor GntR from Escherichia coli. Initially we assembled OFF- and ON-type switches by rewiring the native gluconate-dependent binding of GntR to target DNA sequences in mammalian cells. Then, we utilized the ability of GntR to dimerize in the presence of gluconate to activate gene expression from a split transcriptional activator. By means of random mutagenesis of GntR combined with phenotypic screening, we identified variants that significantly enhanced the functionality of the genetic devices, enabling the construction of robust two-input logic gates. We also demonstrated the potential utility of the synthetic switch in two in vivo settings, one employing implantation of alginate-encapsulated engineered cells and the other involving modification of host cells by DNA delivery. Then, as proof-of-concept, the gluconate-actuated genetic switch was connected to insulin secretion, and the components encoding gluconate-induced insulin production were introduced into type-1 diabetic mice as naked DNA via hydrodynamic tail vein injection. Normoglycemia was restored, thereby showcasing the suitability of oral gluconate to regulate in situ production of a therapeutic protein.


INTRODUCTION
Synthetic biology can help unlock the potential of gene and cell-based therapies to provide more efficient and safer treatment options for a wide range of chronic diseases, ena bling customiza ble tr eatment r egimens to fit each patient's needs. Despite the e v er-e xpanding pipeline of viral and non-viral gene therapies in clinical trials, this therapeutic modality has not yet achie v ed widespread clinical success. So far, most r esear ch efforts have been dir ected to vector design in order to improve cell and tissue targeting ( 1 ), and less attention has been gi v en to the molecular components that regulate the expression of the therapeutic payloads ( 2 ), which are often controlled by constituti v e or tissue-specific promoters ( 3 , 4 ). Incorporating synthetic gene-regulatory networks controlled by external signals to achie v e precision control over the therapeutic outputs can be a game-changer in the clinical success of gene therapies. Their use can also boost the outcome of other therapies, for example by providing ways to manage ov er-acti vation or uncontr olled pr oliferation of chimeric antigen receptor (CAR)-modified T cells after infusion into patients for inducible cancer imm unothera py ( 5 , 6 ) or by limiting the time window of CRISPR / Cas9 expression to reduce off-target effects in CRISPR / Cas9-mediated in vivo gene editing ( 7 ).
Se v eral mammalian synthetic gene switches have been designed to enable exogenous control over the magnitude and timing of therapeutic gene expression via specific input signals, ranging from various chemicals, such as antibiotics ( 8 ), cancer drugs ( 9 ), antivirals ( 10 , 11 ) and food components / additi v es (12)(13)(14), to traceless physical stimuli, such as light ( 15 , 16 ), temperature change ( 17 , 18 ), or electrical signals ( 19 , 20 ). Howe v er, the number of mammalian gene switches sensiti v e to small molecules that are suitable for in vivo applications still falls short, as most of the available trigger compounds suffer from pharmacokinetic challenges or undesirable side effects ( 21 ). Furthermore, despite the benefits of remote actuation for in vivo applications, using physical triggers has some disadvantages, such as the limited depth to which they can penetrate tissue to stimulate modified host cells or implanted engineered cells, or the need for complex equipment to a ppl y the stimulus ( 21 ).
Here, we focused on gluconate as a small-molecule inducer that is already clinically licensed, and set out to assemb le gluconate-responsi v e gene switches suitab le for both in vitro and in vivo applications. Gluconate is naturally found in small amounts in foods such as fruits, honey, wine or rice, and is also used in the food industry as an additi v e in many processed foods, featuring in the list of ingredients generally recognized as safe (GRAS) ( 22 ). Furthermore, gluconate is often used as a neutral carrier of ions, such as calcium, zinc or iron, in medications or nutritional supplements used to treat conditions such as anemia and hypocalcemia, or in combination with other drugs such as the anti-parasitic medication sodium stibogluconate used to treat leishmaniasis ( 23 ). Although little is known about the metabolism of gluconate in human cells, it is likely converted into the pentose phosphate pathway intermediate 6phosphogluconate by gluconokinase, as expression of this enzyme has been detected in se v eral human tissues ( 24 ) and cell cultures ( 25 , 26 ). This feature would provide better control over the timing / reversibility of transgene expression by pre v enting prolonged exposure to the inducer.
Some bacteria are capable of using gluconate as a source of carbon and energy in a glucose-restricti v e environment. The glucona te ca tabolic pa thway of E. coli is controlled by the transcriptional r epr essor GntR, which contains a DNAbinding helix-turn-helix domain at the N-terminus and an effector-binding domain (EBD) at the C-terminus ( 27 ). In E. coli , the regulatory regions of gluconate-inducible metabolic genes contain one or two copies of a highly conserved DNA sequence to which GntR binds ( 28 ). The affinity of GntR for these sites is impaired in the presence of gluconate due to a conformational change in the gluconatebound form of GntR, resulting in transcriptional activation.
In this work, we leveraged the ability of gluconate to induce changes in the GntR protein to assemble highperformance gene switches responsi v e to gluconate in mammalian cells. We applied a combination of random mutagenesis and high-throughput phenotypic screening to identify GntR variants with enhanced gluconate-inducible dimerization of GntR-fused split DNA-binding and transactivation domains, thereby obtaining a grea ter fold-activa tion of gene expression. Importantly, we demonstrated that the optimized molecular components could be used to assemble robust Boolean logic gates in cell culture and also to enable oral glucona te-media ted control of transgene expression in wild-type mice. As proof-of-concept, the gene switch was applied in the context of diabetes, in which the orthogonality of the calorie-free inducer could be explored to regulate insulin expression and help managing blood glucose le v els within the physiological range. We showed that orally administered gluconate, by itself, has no effect on the fasting or post-prandial glucose le v els of type-1 diabetic (T1D) mice, and confirmed that introduction of the components encoding gluconate-switched insulin production r estor ed normoglycemia in T1D mice fed with gluconate.

Plasmid construction
Design and cloning details for all genetic components utilized in this study are provided in Supplementary Tables S1 and S2 (Supporting Information). Plasmids were designed using Benchling ( www.benchling.com ) and constructed by standard molecular cloning techniques. For restriction enzyme-based cloning, plasmids were digested with the desired endonucleases (New England Biolabs); digested backbones were dephosphorylated with Quick CIP (M0525L, New England Biolabs) before ligation with T4 DNA ligase (EL0011, Thermo Fisher). All PCR reactions were performed using Q5 High-Fidelity DN A pol ymerase (M0491L, New England Biolabs). After ligation, the plasmids were transformed and amplified in E. coli XL10-Gold strain (Agilent) and DNA was extracted using a plasmid miniprep kit (Zymo Research) or Midiprep Kit (D4200, Zymo Research). Constructs were verified by Sanger sequencing at Microsynth AG. Synthetic gene fragments used in the study were codon-optimized for expression in human cells and synthesized by Twist Bioscience.

Cell culture
Human embryonic kidney cells (HEK-293T, ATCC: CRL-11268) were cultivated in Dulbecco's modified Eagle's medium (DMEM; 10566016, Thermo Fischer) supplemented with 10% v / v fetal bovine serum (FBS; F7524, Sigma Aldrich) under a humidified atmosphere containing 5% CO 2 at 37 • C. Routine sub-culture was performed e v ery 2-3 days when cell plates were around 80% confluent. Cells were detached using 0.05% trypsin-EDTA (25300054, Life Technologies) and re-seeded in fresh medium at the desired cell density. Cell number and viability were quantified using a CellDrop automated cell counter (DeNovix).

T r ansient tr ansfection
For transfection experiments in 96 / 48 / 24-well plates (Corning), cells were seeded at 0.1 ×10 6 cell / ml, the day before tr ansfection. The tr ansfection mixture for one well of 96well plates consisted of 50 l of FBS-free DMEM containing a 1:4 DN A:PEI (pol yethylenimine, MW 40000; 24765, Polysciences) mixture, with a total DNA amount of 120-150 ng. For larger plate formats, the quantities were scaled up accordingly. After 20 min incuba tion a t room temperature, the transfection mixture was added dropwise to the cells, which were then incubated overnight. The next morning, the culture medium was replaced with fresh FBScontaining DMEM with or without sodium gluconate (527-07-1, Merck) and cell supernatant samples were collected 24 h later for assay of human placental secreted alkaline phosphatase (SEAP) or nanoluciferase (NLuc) reporter, unless otherwise indicated.

Generation of monoclonal stable cell lines
HEK-293T cells were transfected with a hyperacti v e Sleeping Beauty (SB) transposase (pTS395) expression vector ( 29 ) in a 1:5 (w eight / w eight) ratio with pAna366 and pAna368 vectors, containing SB recognition sites and encoding a puromycin resistance marker and a blue fluorescent protein. The medium was exchanged 12 h after transfection and cells were incubated for 48 h before the addition of selection medium containing 4 g / mL puromycin. After three passages, the polyclonal population was sorted by fluorescence-activated cell sorting into single cells in 96-well plates and expanded to obtain monoclonal cell lines.

Random mutagenesis of GntR
The library of GntR mutants was synthesized using a Gen-eMorph II random mutagenesis kit (200552, Agilent Technologies) according to the manufacturer's instructions. A PCR reaction was performed to amplify the GntR sequence in the GntR-VPR fusion protein (pAna274), and the product was then cloned in the template plasmid, replacing the wild-type GntR. The library was transformed into competent XL10-Gold E. coli , plated into ampicillin-containing agar plates and incubated overnight at 37 • C. Colonies were picked and expanded during 24 h in deep 96-well plates containing 750 l of ampicillin-supplemented LB medium, at 37 o C with orbital shaking (220 rpm). The plasmid library was purified using the Zyppy-96 plasmid mini prep kit (D4041, Zymo Research) according to the manufacturer's instructions.

NLuc quantification
Secreted NLuc in cell culture supernatants was measured with the Nano-Glo Luciferase Assay System (N1110, Promega). In brief, a 7.5 l aliquot of each sample was mixed with 7.5 l buf fer / substra te mix (50:1) in the wells of 384-well plates (781076, Greiner Bio One), which were then incuba ted a t room tempera ture for 10 min. Luminescence was measured with a Tecan M1000 plate reader.

Insulin quantification
Recombinant mINS le v els in serum were quantified using a mouse insulin ELISA kit according to the manufacturer's instructions (10-1247-01, Mercordia). Optical density was measured at 450 nm on a Tecan M1000 plate reader and the corresponding concentrations were calculated in Prism 9 (GraphP ad Softwar e Inc) using a cubic-spline r egr ession based on the measured absorbances of manufacturerprovided standard solutions.

Animal experiments
C57BL / 6JRI mice (Janvier Labs Saint-Berthevin, France), weighing ∼25 g, were used. Animals were housed in a contr olled r oom a t 22 • C , 50% humidity, 12-h light-dark cycle with ad-libitum access to standard diet and drinking water. Animals were randomly assigned to experimental groups. To induce type-1-diabetes, C57BL / 6JRI mice (male, 8-9 w eeks) w er e tr eated during 5 days with streptozocin (S0130, Merck) at a daily dose of 75 mg / kg to deplete insulin-producing beta cells. Li v er function was evaluated by assaying the serum le v els of alanine transaminase (ALT, ab105134, Abcam) and aspartate aminotr ansfer ase (AST, MAK055, Sigma Aldrich) according to the manufacturers' protocols. The body weight of the animals was also regularly monitored. Cell implants. Engineered HEK-293T cells were encapsulated into coherent alginate-pol y-( L -l ysine)alginate beads (400 m; 200 cells / capsule) using an Inotech Encapsulator Research Unit IE-50R (Buechi Labortechnik AG) with the following parameters: 20 ml syringe operated at a flow rate of 400 units, 200 m nozzle with a vibration frequency of 1200 Hz, and bead dispersion voltage of 1.5 kV. Wild-type mice were intra peritoneall y injected with 1 ml of DMEM containing 5 ×10 6 cells. Sodium gluconate solutions were supplemented by oral gavage (250 l). Thai rice was pulverized, mixed with wa ter (ra tio 1:1) and administered to mice by o.g. at a dosage of 5 g / kg (equivalent to 375 g for a human weighing 75 kg). Hydrodynamic injection. Mice were injected with 2 ml of saline solution containing the plasmids pAna366 and pAna368 (4 mg / kg of DNA per mouse) 24 h before starting inducer trea tment. Glucona te administration was perf ormed twice / da y (e.g. 5 g / kg). Fasting blood glucose was measured after 4 h of food restriction using a clinically licensed glucometer (Accu-Check Instant, Roche). Glucose tolerance test (GTT) was performed by i.p. injection of glucose (2 g / kg), and glycemia was profiled at intervals for the following 2 h. Serum was isolated for analysis using BD Microtainer ® SST tubes according to the manufacturer's instructions (30 min incubation in the dark, centrifugation for 2 min at 10 000 × g; 365967, Becton Dickinson). All experiments involving animals were performed in accordance with the Swiss animal welfare legislation, approved by the veterinary office of the Canton Basel-Stadt (approval no. 2879 / 31996) and carried out by Shuai Xue (no. LTK4899) and according to the directi v es of the European Community Council (2010 / 63 / EU), approved by the French Republic (project no. DR2018-40v5 and APAFIS #16753), and carried out by Shuai Xue and Ghislaine Charpin-El Hamri (no. 69266309) at the Uni v ersity of Lyon, Institut Uni v ersitaire de Technologie (IUT), F69622 Villeurbanne Cedex, France.

Statistical analysis
Sta tistical evalua tion was conducted by using an unpaired Student's two-tailed t -test for comparing two sets of data as implemented in Prism GraphPad 9 (GraphPad Software Inc., San Diego, California, USA).

Gluconate gene switches based on GntR affinity for specific DNA sequences
We sought to build glucona te-regula ted gene expression systems with OFF-and ON-type behavior in mammalian cells by capitalizing on the glucona te-media ted GntR interactions with known DNA sequences in E. coli ( Figure  1 A). The OFF-type switch relies on GntR fused to the strong tripartite mammalian transactivator VP64-p65-Rta (VPR), while the ON-type switch relies on GntR fused to the transr epr essor Kruppel-associated box (KRAB) domain from human ZNF10. In the absence of gluconate, these fusion proteins bind to GntR-recognition DNA sequences (O GntR ), thereby activating or r epr essing, r especti v ely, the transcription of downstream genes. HEK-293T cells tr ansiently tr ansfected with the switch-OFF system controlling the expression of secreted alkaline phosphatase (SEAP) as a reporter protein showed significant gluconateresponsi v eness, producing 10-fold less SEAP in the presence of gluconate (Figure 1 B). We tested two GntR DNA recognition sites: (i) the sequence found in the promoter region of the gluconate transport operon (ATGTTACC-GATAACAG) and (ii) the consensus sequence (ATGT-T ACCCGT AACAT) ( 28 ). The gr eatest r esponsi v eness to gluconate was obtained with the first sequence (Supplementary Figure S1A). To further assess the specificity of GntR-VPR for the target sequence, we also deleted the whole O GntR operator site from the promoter region or inserted two base pairs in the center of this sequence. Both mutations abolished SEAP expression from the minimal promoter in the presence of gluconate ( Supplementary Figure S2). The switch-ON system showed repression of SEAP expression from the constitutive human cytomegalovirus promoter (P hCMV ) bearing downstream GntR-binding sequences when the cells were cultured in gluconate-free medium and GntR-KRAB was constituti v ely co-e xpressed (Figure 1 C). In the presence of gluconate, SEAP production was significantly increased. Two or three tandem repeats of the binding site in the promoter region did not improve the performance of the switches (Supplementary Figure S1A, B), and neither did replacing the constituti v e promoter P hCMV with the phospho gl ycerate kinase promoter (P PGK ) in the ON-type switch (Supplementary Figure   S1C). We also confirmed that each domain (GntR, VPR, KRAB) in the OFF-and ON-type switches is essential to obtain glucona te-media ted gene expression (Supplementary Figure S3A, B). The responsi v eness is improv ed when the O GntR sequence is placed upstream of the T AT A box in the OFF-type switch (rather than downstream), and downstream of the constituti v e promoter (P hCMV ) in the ONtype switch (rather than upstream) (Supplementary Figure  S4A, B). Next, we tested whether co-expression of the highaf finity glucona te transporter gntT from E. coli ( 28 ) would increase the sensitivity of the gene switches. Engineered cells cultured in the presence of different gluconate concentrations showed significant fold-change differences of SEAP expression a t concentra tions as low as 10 M for both the ON-and OFF-type systems when co-expressing GntT (Figure 1 D, E).

Gluconate-inducible gene expression relying on GntR dimerization
We have recently shown that some bacterial transcription factors (TFs) comprising helix-turn-helix (HTH) DNAbinding domains linked to an effector-binding domain (EBD) also show responsi v eness to their effectors when fused to another TF, relying on the interaction of the latter TF with its cognate DNA-binding sequence for the regulation of gene expression ( 30 ). To test whether this approach w ould also w ork for GntR, we transfected HEK-293T cells with constituti v ely e xpressed GntR fused to either TetR or to the VPR transactivator and analyzed gluconateinducib le SEAP e xpression from a tetracy cline-responsi v e promoter. The underlying hypothesis is that GntR dimerization brings VPR near the target promoter, thereby acti vating SEAP e xpr ession in a gluconate-r esponsi v e manner (Figure 2 A). We tried fusions of TetR and VPR at both termini of GntR and observed gluconate-inducible SEAP expression in cells transfected with GntR-TetR incorpora ting VPR a t the N-or C-terminus of GntR (Figure 2 B), affording a 3-or 4-fold increase in SEAP expression, respecti v ely. In a cytosolic dimerization assay based on split nanoluciferase (NLuc), in which the N-and Cparts of NLuc are fused to either terminus of GntR, we observed an increased NLuc signal in the presence of gluconate for three out of four combinations. The highest foldinduction was observed when both NLuc split parts were fused to the C-terminus of GntR (Supplementary Figure  S5), and ther efor e we adopted the GntR-VPR / TetR fusion configurations in all subsequent experiments. To regulate the expression of two proteins sim ultaneousl y, we combined this ON-type gene switch with the OFF-type gene switch based on a synthetic promoter bearing gluconateresponsi v e elements (Figure 1 B). Increasing concentrations of gluconate resulted in increased secretion of NLuc and decr eased secr etion of SEAP (Figur e 2 C). In this scenario, the synthetic transcription factor GntR-VPR functions sim ultaneousl y to control two opposite expression states: (i) in low gluconate-containing medium, SEAP expression is ON and NLuc expression is OFF, while in high gluconatecontaining medium, SEAP expression is OFF and NLuc expression is ON.  Data are shown as mean ± SD of n = 3 biolo gicall y independent samples, r epr esentati v e of 3 independent experiments. ns, not significant, * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001.

Optimizing gene switches via random mutagenesis of GntR and screening
We sought to improve the genetic circuit output by generating err or-pr one PCR-based random mutations of the GntR sequence in the GntR-VPR fusion protein. We evaluated the performance of the GntR-VPR mutant library by means of a high-throughput functional screening assay, and selected the best candidates in terms of sensitivity and fold-induction of NLuc expression from a tetracyclineresponsi v e promoter (Supplementary Figure S6A). HEK-293T cells co-transfected with the wild-type GntR-TetR and the library of GntR-VPR variants were assayed for NLuc secretion after incubation for 24 h in the presence or absence of gluconate (Supplementary Figure S6B). Six hits were validated in a secondary screening, and confirmed to provide significantly higher fold-inductions than the wildtype GntR-VPR fusion protein (Figure 3 A), due mostly to significant lower basal NLuc expression in the absence of gluconate. Inter estingly, the thr ee best performing mutants (m ut4 GntR D55V / E235K / G279V , m ut5 GntR A306S , and m ut6 GntR R285C / V289I ), have a single mutation or double mutations on the C-terminal EBD, spanning between residues 161 and 331 ( 31 ), and all of them involved replacement of at least one polar amino acid residue by a nonpolar residue, or vice versa (Supplementary Table S3). These changes might favor dimerization with the wild-type GntR fused to TetR, resulting in increased activation of NLuc expression.  constituti v e e xpression of GntR fused to T etR (P hCMV -GntR-T etR-pA or P hCMV -T etR-GntR-pA) and either constituti v e e xpression of GntR fused to VPR (P hCMV -GntR-VPR-pA or P hCMV -VPR-GntR-pA) or an empty vector, and then cultured for 24 h in the presence or absence of gluconate (10 mM). ( C ) Controlling the expression of two reporter proteins (SEAP and NLuc) in response to gluconate. Genetic constructs (left) and reporter analysis (right). HEK-293T cells were tr ansiently tr ansfected with the switch-ON system based on the TetR-GntR interaction with a TetO 7 DNA sequence dri ving NLuc e xpression and the switch-OFF system based on the GntR-VPR interaction with an O GntR DNA sequence driving SEAP expression. Supernatant levels of NLuc and SEAP were analyzed 24 h after incubation in the presence of dif ferent glucona te concentra tions. Da ta are shown as mean ± SD of n = 3 biolo gicall y independent samples, r epr esentati v e of three independent e xperiments. ns, not significant, *** P < 0.001, **** P < 0.0001.
To compare the glucona te-media ted inducibility of GntR heterodimers with GntR homodimers, we cloned the two GntR mutants that afforded the highest fold-inductions (GntR mut4 and GntR mut6 ) in the TetR construct. Interestingly, the inducibility was completely abolished when both dimerizing partners were based on the same GntR variant (Figure 3 B). In the cytosolic dimerization assay, while GntR mut6 and GntR wt fused to each NLuc moiety increased the NLuc luminescence in the presence of gluconate, use of the GntR mut6 variant in both split Nluc moieties did not result in increased luminescence in the presence of gluconate (Supplementary Figure S7), suggesting tha t glucona te-inducible GntR homodimeriza tion is less favorable for this mutant.
We characterized the dose-r esponse r elationships of the two best mutants, GntR mut4 and GntR mut6 , when they wer e co-expr essed with the gluconate transporter GntT (Figure 3 C). Both mutants afforded a high dynamic range, with up to 235-fold induction in NLuc le v els for 2 mM gluconate and 20-fold in the low micromolar range. We also tested the specificity of the GntR mut6 -based gene switch for gluconate b y ev aluating whether a set of different sugars and structurally similar molecules could activate reporter gene expression. None of the screened compounds triggered significant NLuc expression (Figure 3 D), suggesting high selectivity of the developed gene switch for the sugar acid gluconate. Next, we assessed how the genetic device responds to gluconate w hen stabl y integrated in the cell genome using the Sleeping Beauty transposase system. During this process we replaced VPR with the shorter transactivation domain VP16, which provided comparable functionality of the gene switch (Supplementary Figure S8). SEAP production by transgenic cells treated with gluconate increased dose-dependently and accumulated gradually in the Alto gether, random m utagenesis combined with functional phenotype screening enabled the identification of GntR variants that support high transgene e xpression le vels in the presence of gluconate while maintaining minimal expression in the absence of gluconate, for both transiently provided or stably integrated gene switches.

Construction of 2-input logic gates
The reliance of the gene switches on TetR binding to its co gnate DN A sequence and gluconate-inducible dimerization of GntR enabled us to build compact 2-input (gluconate and doxy cy cline (Dox)) logic gates to regulate the expression of one output. Therefore, we next assessed the performance of logic gates based on GntR wt and GntR mut6 in the presence or absence of gluconate and Dox. The Nimply gate (A AND NOT B, in which A is gluconate and B is Dox) using GntR mut6 performed better than the gate using GntR wt , with average ON / OFF ratios of 64 and 11, respecti v ely. As e xpected, the addition of Dox significantly decr eased r eporter gene expr ession in the pr esence or absence of gluconate for both GntR variant-based gates (Figure 4 A). We also assessed how this gate performs when stably integrated in the cell genome, and confirmed that the addition of Dox overrides the gluconate inducibility (Supplementary Figur e S9), ther eby constituting a 'safety-switch' in a gene / cell therapy setting to stop expression if required. Then, we replaced TetR with re v erse T etR (rT etR) to genera te logic ga tes tha t perform AND opera tion, in which the presence of both inducers is r equir ed to turn on gene expression. The circuit based on GntR mut6 performed robustly, showing an average of 27-fold induction between the ON and OFF states (Figure 4 B). To assemble gene switches that function as 2-input OR gates, i.e. the output is ON when either or both inputs ar e pr esent, the thr ee fusion proteins GntR mut6 -VPR, GntR wt -TetR and rTetR-VP16 were constituti v ely e xpr essed to r egulate NLuc expr ession. The gates functioned as expected, turning NLuc expression ON in the presence of at least one of the inputs (Figure 4 C). Although the output signals were higher in the presence of both inducers, the average ON / OFF ratios were high, especially for the gate based on the GntR mut6 variant (44-fold vs 12-fold for the GntR wt variant).

In vivo regulation of transgene expression
To demonstrate the functionality of our optimized glucona te-actua ted genetic switch in vivo , we encapsulated engineered cells in alginate and implanted them in wild-type mice, which were then orally gi v en different glucona te concentra tions (Figure 5 A). Analysis of blood samples re v ealed a gluconate-dependent increase in SEAP le v els (Figure 5 B). Conversely, cell implants harboring a non-functional gene switch (lacking the GntR-VP16 expression cassette) did not show gluconate-responsi v eness (Supplementary Figur e S10). Furthermor e, to test whether transgene expression could be inad vertentl y activated by diet, mice transplanted with functional switches were fed a large amount of rice, which is one of the richest gluconate-containing foods ( 32 ). SEAP le v els were not incr eased r elati v e to those of mice that did not recei v e any gluconate source (Supplementary Figure S10). We next sought to assess whether the gluconate-switchable device could be used to control the expression of thera peuticall y relevant cargos. We selected type-1 diabetes (T1D) as a disease model to showcase the potential of the device for regulating ad vanced thera pies in response to exo genous gluconate. The T1D mice recei v ed the switch components encoding gluconate-induced insulin production as naked DNA via hydrodynamic tail vein injection ( Figure 5 C), which is a well-established technique for efficient DNA uptake by hepatocytes (a pproximatel y 40% of the li v er cells have been reported to express the exogenous genes ( 33 )), with much lower vector delivery to other tissues, such as the heart, lungs or kidneys ( 33 , 34 ). To confirm successful DNA deli v ery into host cells and responsi v eness to gluconate, we first established that wild-type mice injected with the components encoding gluconate-switched NLuc expression exhibited higher luminescence in the abdominal region when treated with gluconate (Supplementary Figure S11). We also confirmed gluconate-dependent insulin secretion in transiently transfected cell cultures (Supplementary Figure S12). Gluconate intake by the engineered T1D mice ra pidl y r estor ed normo gl ycemia during a glucose tolerance test (GTT), while engineered T1D mice that did not take  Figure S13). To confirm tha t glucona te by itself does not affect the diabetic state, we measured fasting blood glucose levels and ran GTTs in T1D mice pr e-tr ea ted with glucona te or vehicle only. The response profiles were very similar in these two mouse groups (Supplementary Figure S14A, B). Collecti v ely, these results indica te tha t both the DNA-based and cell-based platforms are functional in vivo and can be regulated by exo genousl y supplemented gluconate.

DISCUSSION
Synthetic molecular switches enabling tight control over gene expression in mammalian cells in response to envi-ronmental signals are useful tools to study basic cellular functions of genes of interest, as well as to de v elop safer and more efficient gene and engineered cell therapies. In this work, we focused on the non-toxic and bioavailable inducer gluconate and we assembled a number of gluconatedependent transcriptional control devices featuring robust operation both in vitro and in vivo . Akin to the classical Tet-Off system, GntR fused to a transactivating domain binds to its cognate DNA-binding sequence O GntR and fully activates transcription of the downstream genes in the absence of the inducer. Fusion of GntR to a trans-silencer domain was r equir ed to r epr ess expr ession in mammalian cells cultured in gluconate-free medium. While ON-type switches ar e pr efer able for in vi vo applications, the reliance on inhibition of a full promoter by the ZNF10 KRAB domain is often associated with a high le v el of leaky expression and slow activation kinetics ( 35 , 36 ). This prompted us to assemble a second switch configura tion fea turing ONtype behavior, in which gluconate-inducible dimerization of two separate GntR polypeptides fused to the two domains of a split transcription factor (the DNA-targeting domain TetR and a transactivation domain (VPR or VP16)) come together to activate transgene expression in the presence of gluconate. Through random mutagenesis and functional phenotypic screening, we were able to identify pairs of GntR variants that showed minimal signal output in the absence of gluconate, and high activation in the presence of glucona te, grea tly surpassing the gene expression fold-changes achie v ed with the wild-type GntR protein. The best mutant switch can sense and respond to gluconate in the low micromolar range and achie v e e xpression changes of se v eral hundred-fold in the low millimolar range. Thus, randoml y m utating GntR and scr eening a r elati v ely small mutant library for the retention of function in the presence of the effector proved to be an efficient strategy to find variants that enab le improv ed glucona te-media ted transcriptional control. When we challenged the specificity of the interaction with gluconate by using a set of sugar molecules, none of the sugars tested induced transgene expression.
The optimized molecular components could be combined to assemble efficient higher-order gene circuits, including dual-input logic gates, which exhibited selective AND and OR operations conditioned by exposure to the two small molecules gluconate and Dox. These circuits are expected to be useful for applications requiring combinatorial computation in complex cellular environments.
In vivo experiments showed that orally available glucona te can stimula te engineered cell implants microencapsulated in alginate beads, as well as engineered host cells, to express the output signal. Administration of gluconate twice a day in T1D mice triggered the release of enough insulin to r estor e normo gl ycemia, as well as to ra pidl y attenuate postprandial glycemic excursions in glucose tolerance tests. The advantages of gluconate as inducer for controlling gene-and engineered cell-based therapies include its excellent safety profile, gi v en its e xtensi v e use in many drug formulations (e.g. 37 ) and as a common additi v e in many processed foods ( 38 ), such as those labeled with E576 (sodium glucona te), E577 (potassium glucona te), E578 (calcium gluconate), or E579 (ferrous gluconate). While we used hydrodynamic transfection for a proof-of-concept gene therapy application, other more clinically suitab le deli v ery v ehicles could alternati v ely be used. For example, the size of the gluconate-inducible insulin secretion device is compatible with the 8.5 kb packaging capacity of third-generation adenoviral vectors ( 39 ). Furthermore, for clinical translation as a cell therap y, mor e clinically r elevant cells would need to be considered, such as mesenchymal stem cells or memory B cells (40)(41)(42), engineered with the necessary components to produce the therapeutic output in response to gluconate ingestion. While we selected T1D as the disease model and insulin as the output for the proof-of-concept study, the glucona te-actua ted genetic switch should be readily adaptable to control in situ production and dosing of other therapeutic proteins for the treatment of a wide range of chronic diseases.

DA T A A V AILABILITY
The authors declare that all the data supporting the findings of this study are available within the paper and its supporting materials. All original plasmids listed in Supporting Table S2 are available upon request.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.