## Introduction

The field of synthetic biology, now on the verge of its third decade, is emerging from its formative tool-building phase into one that is application focused1,2,3. This trend toward practical implementation is setting the stage for biology to address unmet needs in health and food security4,5,6,7,8,9, green manufacturing and energy10, amongst many others11,12. Work to date in the field has largely relied on encoding functions into cells, which can take months or years to reach a final product13. This has led to efforts to move synthetic biology out of the cell using cell-free protein expression systems which vastly accelerates the design-test cycle14,15,16 and enables these biotechnologies to be housed in a biosafe format that can be freeze-dried for distribution and use without refrigeration4,9,17,18,19. Using the cell-free enzymatic machinery of transcription and translation, these systems allow for gene circuits to operate as they do inside living cells17,20.

Efforts to develop cell-free technologies have been particularly successful for point-of-care (PoC) diagnostics, environmental sensing4,17,21,22,23,24, on-demand manufacturing of protein-based therapeutics9,18,19,25 and education26,27. One of the most exciting aspects of these efforts is the potential to improve access to health care. This is especially relevant for PoC diagnostics where the principles of rational design from synthetic biology are enabling the rapid and low-cost development of sensors for pathogens and small molecules. These advances are bringing real-time surveillance of outbreaks and low-cost tools for accessible global health closer to reality. As demonstrated by the SARS-CoV-2 outbreak, access to diagnostics is critical and shortages can be a significant bottleneck to containment, leading to restrictions in movement and normal life28. While gene circuit-based sensors provide a promising solution, they need to be paired with companion technologies to be used outside of the laboratory.

Recognizing the importance of this challenge, we identified the blood glucose meter as a potential solution. As one of the most widely used PoC monitoring devices, it has significantly improved the lives of millions of people with diabetes by enabling the portable quantification, and therefore personal management, of blood sugar levels. These devices are based on a glucose-dependent electrochemical signal generated by an enzyme, such as glucose oxidase, in their test strips29. The widespread adoption of glucose monitors has resulted in a global network of device manufacturing and consumable distribution, as well as broad acceptance by patients and clinicians. Inspired by this model, we aimed to leverage this established diagnostic infrastructure as a universal interface for emerging diagnostic toolsets from the field of synthetic biology. Previous work has shown that glucose meters can be used for the detection of analytes other than glucose, particularly through the release of a pre-existing, DNA-invertase conjugate from magnetics beads. However, to date30,31,32,33,34,35 the potential of a universal glucose-based output platform for gene circuit-based sensors and their use in real-world applications has yet to be demonstrated. Importantly, here we report the de novo design and broad application of gene circuit-based sensors capable of the in situ expression of glucogenic reporter enzymes in response to target analytes.

Below we describe the development of a gene circuit-based system that translates the presence of a target analyte into a signal that can be easily measured by a glucose meter. This is achieved by replacing conventional reporter proteins for gene circuit-based sensors (e.g. green fluorescent protein) with enzymes that convert inert glucose-containing precursors into glucose monomers that can be detected by an off-the-shelf glucose meter. The development of this platform begins with identifying glucogenic enzymes that are compatible with cell-free systems and that can serve as novel reporter proteins. Using a typical commercial glucose meter, we screened candidate enzymes and found that the expression of select enzymes can generate glucose in as little as 60 min in coupled protein synthesis-glucose generation reactions. We also demonstrate an enzyme-mediated method with the potential for pre-clearing endogenous glucose from samples (e.g. blood). Next, we use the substrate-specific nature of reporter enzymes, along with other strategies, to allow simple sensor multiplexing in cell-free reactions. With the glucose-mediated interface established, we then demonstrate the successful operation of gene circuit-based sensors for small molecules and synthetic RNAs. To show the potential of this interface for global health applications, we utilize the glucose meter to detect RNA sequences for typhoid, paratyphoid A and B, and related drug resistance genes. Using the related RNAs, we describe robust signal within 60 minutes, sensor multiplexing and detection of Salmonella typhi itself at clinically relevant concentrations. Finally, to demonstrate the potential of this platform for response to public health emergencies, we present the development and validation of sensors for the SARS-CoV-2 virus, the causative agent of the current global COVID-19 pandemic, and demonstrate detection from clinical samples.

## Results

### Platform development

We first assessed the feasibility of glucose detection in the biochemical context of a cell-free system (CFS) (Fig. 1a). We chose a recombinant CFS (PURExpress, NEB) that is comprised of purified proteins and does not contain enzymes from glucose metabolism that would interfere with glucose generation in our system. By simply doping glucose into the CFS we tested whether an off-the-shelf glucose meter with test strips could detect the presence of the sugar. Our initial efforts were confounded by a high degree of read variability, which, after buffer optimization, was resolved by the simple addition of 0.0125 % Tween-20, a non-ionic detergent. The detergent may serve to ensure that the CFS efficiently wicks into the capillary channel of the test strips.

With the concept of measuring glucose from CFS established in principle, our next step was to develop reporter enzyme systems with the capacity to generate glucose in the presence of substrate. With these in-hand, gene circuit-based sensors can be designed to generate glucose in response to activation (Fig. 1b). We began by screening the de novo expression and catalytic activity of 40 enzymes under the buffer conditions of the CFS. Leveraging the capacity of the CFS for rapid prototyping, we tested the expression and corresponding substrate catalysis by the enzymes in an overnight incubation. Using the glucose meter to detect glucose output, we found three enzymes that yielded significant glucose level increases in CFS (Fig. 1c): trehalase (C. japonicus, tre37a), lactase (E. coli, lacZ) and phosphatase (E. coli, ybiV), using their respective substrates trehalose, lactose, and glucose-6-phosphate. These results were corroborated using a separate colorimetric assay for detection, based on the conversion of nicotinamide adenine dinucleotide (NAD) to NADH by glucose dehydrogenase (GDH) in the presence of glucose (Fig. 1d).

To tackle the potential issue of endogenous blood glucose background in patient samples, we used the enzyme glucose dehydrogenase (B. subtilis, variant E170K/Q252L36). This enzyme from the oxidoreductase family converts D-glucose into D-glucono-1,5-lactone, which is inert to the glucose meter. For every molecule of glucose catabolized, this enzyme requires a molecule of NAD. We hypothesized that this 1:1 stoichiometry between substrate and cofactor might allow us to selectively clear fixed amounts of glucose from incoming samples by adding equimolar amounts of the NAD co-factor. To test the concept, we added glucose to the CFS and evaluated the effect of expression of glucose dehydrogenase in the presence of various concentrations of NAD. After an overnight incubation at 37 °C, glucose meter measurements showed that glucose concentration reduction was dependent on glucose dehydrogenase and proportional to supplemented NAD concentration (Fig. 1e). These results highlight that NAD availability can be used effectively to tune glucose dehydrogenase-mediated glucose clearance.

Toehold switches are a class of programmable riboregulators that can be used to control reporter protein expression in a sequence-specific manner, allowing them to serve as RNA sensors37. We previously used toehold switches for the recognition of pathogen sequences using colorimetric protein reporters for optical detection4. Here, to produce a glucose output from gene circuit-based sensors, we placed the sequence for the trehalase reporter enzyme downstream of a rationally designed toehold switch-based RNA sensor37. DNA encoding this toehold switch-based sensor was added to CFS containing trehalose (20 mM) and incubation was limited to one hour to simulate practical applications. At the shorter one hour incubation, CFS reactions contribute a non-specific background signal that is not present in overnight reactions (Fig. 2a). The use of control reactions containing the sensor alone resolves this potential challenge (see below) by making the detection of target sequences unambiguous.

Glucose meters provide a quantitative measurement of glucose concentration from 10 to 600 mg/dL which introduces the possibility of creating high and low glucose reporter levels to enable multiplexed output. With this in mind, we evaluated the potential for tuning the level of glucose output from the sensors, using synthetic sensor A and its respective target (Supplementary data sets 2-3)37. We started by testing the effect of varying the concentration of the substrate trehalose and found that substrate concentration had a significant impact on the glucose signal generated, with higher concentrations providing greater signal-to-noise ratios and stronger signals (Fig. 2a). We also found that the glucose output level was responsive to the concentration of template DNA for the toehold switches (Fig. 2b). Looking to further control glucose production, we investigated the potential of different reporter enzymes to generate variable glucose output levels. When DNA coding for two reporter enzymes (lactase or trehalase) with two different upstream toehold switches was added to the same CFS reaction, different glucose levels were produced depending on which trigger RNA was present in the reaction (Fig. S1a).

Using the rapid (1 h) and high dynamic range of the trehalase reporter, we then took advantage of the gene circuits themselves to modulate glucose production. As we have recently reported38, toehold switch sequences can influence the expression level of reporter enzymes to provide an additional layer of control in the design of cell-free sensors. By combining a toehold switch with a low expression output (Tre A) and another with a high expression output (Tre B), we show that the sequence-specific detection of the respective RNAs can be distinguished using a common enzyme/substrate pair. Activation of the switches with an equal concentration of RNA triggers A and B (5 nM) results in distinct low and high glucose outputs, respectively (Supplementary Fig. S1b). Importantly, when both target RNA sequences are present, the combined glucose production itself can also be clearly distinguished from the presence of either RNA inputs alone (Supplementary Fig. S1b).

### Glucose meter-mediated small molecule detection

With the development of the underlying glucose meter-based interface complete, we next focused on demonstrating the potential of this tool for the gene circuit-mediated detection of a small molecule other than glucose. To create a sensor for the antibiotic tetracycline (TC) we adapted a bacterial regulatory gene circuit for antibiotic resistance. In wild-type systems, the transcription of the tetracycline efflux pump (TetA) is regulated through the TetA operator (TetO) sequence, which is repressed by the binding of a repressor protein (TetR) in the absence of tetracycline. When tetracycline is present, TetR dissociates from TetO enabling the expression of TetA39. We recapitulated this circuit in vitro by placing the TetO regulator between a T7 promoter and the trehalase reporter enzyme to create a glucose-generating sensor that is responsive to tetracycline (Fig. 2c). In the absence of TetR, reactions generate a strong glucose signal in comparison to the negative control, but when the TetR protein is added to the control sensor circuit, glucose production is significantly reduced (Fig. 2d). Finally, we show that in the presence of both TetR and tetracycline (TC), high glucose measurements are re-established, demonstrating the ability of the glucose meter to effectively sense the presence of this small molecule.

### Proof-of-concept diagnostic sensors

#### Antibiotic resistance gene sensors

To test the potential of the glucose meter for PoC molecular diagnostics, we used toehold switches designed for the detection the genes responsible for ampicillin (AmpR) and spectinomycin (SptR) resistance17 to regulate trehalase reporter expression. DNA encoding the respective switches was then tested in the presence or absence of the RNA target in CFS containing trehalose. Both sensors yielded a strong glucose response in the presence of synthetic target RNA sequences (Supplementary Fig. S2).

With the potential for molecular diagnostics established, we set out to develop a panel of tests for typhoid, paratyphoid A and B and the associated fluoroquinolone antibiotic resistance gene QnrS40. These infectious diseases are common in global regions where access to clean water is limited. It is estimated that these diseases, collectively referred to as enteric fever, annually infect over 14 million people, causing >135,000 deaths41. Despite the widespread impact on health, it remains challenging to effectively diagnose and track these infections due to the lack of low-cost, point-of-care diagnostics42.

Twelve toehold switches for each of the four target genes (corresponding to typhoid43, paratyphoid A and B44, and fluoroquinolone resistance40) were computationally designed and placed upstream of the trehalase reporter gene. Adapting the colorimetric GDH-based glucose assay described above (Fig. 1d), we rapidly screened for switch performance using linear DNA (50 nM). Toehold switches with a significant glucose signal output were identified for each target gene (Supplementary Fig. S3) and the top-performing switches were then validated using the glucose meter detection system (Fig. 3a).

We then set out to evaluate the detection threshold of the glucose meter interface when paired with an upstream isothermal amplification step. Previous work has shown that the addition of an amplification step can improve detection sensitivity by orders of magnitude, allowing engineered molecular sensors to detect target nucleic acids well within clinically relevant concentrations4. For target amplification we chose Nucleic Acid Sequence-Based Amplification (NASBA), which is a primer-directed method that yields significant amplification and a second sequence-specific check point for the diagnostic45,46. Using NASBA (1 h) followed by a reaction containing the typhoid-specific toehold switch (targeting the STY1607 gene), the glucose meter interface achieved detection of target RNAs down to the low attomolar range (Fig. 3b). This positive detection of STY RNA from 56 aM samples (1 µL input) represents an initial input of approximately 34 total RNA copies. Similar results were found for paratyphoid A and B, and QnrS targets, with detection thresholds in the low attomolar range (Supplementary Fig. S4). Orthogonality of these sensors was also tested using the glucose meter interface and, as expected, showed specific activation (Supplementary Fig. S5).

We next sought to demonstrate the potential of the glucose meter interface for simple multiplexed diagnostics using the typhoid (STY) and QnrS (Q) sensors. As above, we used NASBA isothermal amplification prior to toehold switch reactions to determine if we could detect each RNA target alone or in combination from low starting concentrations (500 copies in 1 µL). We took advantage of differential glucose outputs from the two sensors to distinguish between detection events. NASBA and toehold switch reactions were performed and using the glucose meter to monitor outputs, we found that STY and Q RNAs yield significantly distinct glucose signals (Fig. 3c). Similarly, when STY and Q RNA inputs were combined (500 copies each), the sensor platform generated a third and unique glucose concentration that allowed for distinction from the individual RNAs. Thus, we showed that the glucose meter interface has the potential for multiplexed detection from input RNAs in the attomolar range. This demonstrates a possible real-world application—a single test capable of independently detecting the presence of both typhoid and a relevant antibiotic resistance gene.

To bring this multiplexed detection capability one step further, we wanted to provide health care professionals with an unambiguous interpretation of the results to provide users with the appropriate course of action. To meet this challenge, we developed a proof-of-concept online tool for interpreting the glucose values (https://www.pardeelab.org/gm.html). Here, using a smartphone or computer, users enter the measured glucose concentration values (mg/dL) for control and test reactions (Fig. 3d, left). These values are compared to look-up tables associated with the tests, and definitive results are displayed (Fig. 3d, right).

Moving the potential of this sensor platform into the hands of users in de-centralized clinics poses challenges, including how to reliably lyse the cell wall of these gram-negative bacteria using simple protocols and equipment. We therefore developed accessible protocols for the detection of endogenous RNA from whole-cell S. typhi. For this we developed a workflow that combines heat and detergent treatments of S. typhi, which is compatible with NASBA and subsequent CFS detection of target RNA (Fig. 4a, b)47. Using the glucose meter to monitor reaction outputs, we found that the sensor platform could clearly detect target STY RNA from serum-containing mock samples with 103 CFU/mL S. typhi. This corresponds to levels found in patients who tested positive for typhoid infections using the most common detection protocol, lab-based blood culture, which has been quantified to include samples in the range of 1000 – 43,500 DNA copies/mL48. With lab-based method such as blood culture and PCR difficult to implement in low-resource settings, we envision the glucose meter-based interface providing a more rapid and point-of-need method of typhoid detection42,48.

The glucose meter interface also holds great potential in providing de-centralized molecular diagnostic capacity to front-line responders in public health emergencies. With the current coronavirus (SARS-CoV-2) outbreak49,50, there is an urgent need for low-cost, PoC molecular diagnostics that can be distributed beyond centralized facilities. With this in mind, and to demonstrate the adaptable capability of the glucose meter platform for the detection of novel and emerging pathogens, we designed and assembled 24 toehold switches specific to six target regions in the SARS-CoV-2 genome (ORF1b, RdRP, E, and three locations within the N gene: N1, N2, N3). Using a lactase reporter combined with the colorimetric substrate chlorophenol red-beta-D-galactopyranoside (CPRG), we screened all 24 toehold switches for response to the corresponding synthetic trigger RNAs and found 6 high-performing candidates (Supplementary Fig. S6, supplementary dataset 2). Top-performing switches were then tested for trigger-mediated glucose output production using a glucose meter for quantification (Supplementary Fig. S7).

Employing the toehold switch (N3 B) specific for the viral gene N (target region N3), we then demonstrated the highly sensitive and specific detection of SARS-CoV-2 (Fig. 5). Using genomic RNA isolated from the virus, coupled with an upstream NASBA amplication step, we found that the N3 sensor could detect as few as 100 RNA copies (1 µL of RNA at 166 aM, Fig. 5a). When this sensor was challenged with off-target viral RNA (1.66 pM) isolated from other respiratory viruses (influenza A subtypes H1N1 and H7N9, MERS), sensor activation and glucose production was highly selective for SARS-CoV-2 (Fig. 5b). Similarly sensitive and selective detection was also observed for the other sensor evaluated (gene E, Supplementary Fig. S8). As a final demonstration, we tested the N3 sensor using clinical RNA samples from six patients, in parallel with RT-qPCR (Fig. 5c). The glucose meter-based diagnostic results matched RT-qPCR performance, with clear discrimination of three SARS-CoV-2 positive patient samples (samples 4-6) from negative control patient samples (samples 1–3). It is worth highlighting that the assembly and prototyping of these sensors was done in a matter of days, illustrating the potential for synthetic biology-based solutions to respond to urgent public health needs.

To match the portability of the glucose meter, we developed a companion portable incubator to provide temperature control at the point of use. This compact instrument is comprised of a temperature-resistant polycarbonate, 3-D printed outer shell that contains a heated aluminum lid and block controlled by a simple four-way switch (Fig. 5d). To operate, users simply set the switch to the corresponding incubation condition for lysis (100 °C), NASBA (65 °C then 41 °C) or cell-free reaction (37 °C). Using a thermoelectric Peltier element  and a PTC heating element, temperature transitions between these settings is rapid, taking only 1–3 min. LEDs on the front of the device indicate heating (red), cooling (blue) and correct target temperature (green). Here we demonstrate the portable incubator (PI) in a side by side comparison with a lab-based thermocycler (TC) in the detection of SARS-CoV-2 viral RNA (Fig. 5e).

## Discussion

We have developed a robust and versatile system to interface gene circuit-based sensors with off-the-shelf glucose meters to create a new potential class of point-of-care diagnostic tools. The utility of this platform is confirmed through a series of proof-of-concept experiments that demonstrate accurate and sensitive detection of global health threats such as S. typhi and the SARS-CoV-2 virus in as little as two hours (Figs. 35). To tackle the challenge of endogenous glucose in patient samples, we presented three possible use case modes where such possible interference could be avoided with the use of programmable glucose reduction (Fig. 1e), sample dilution (Fig. 4) and nucleic acid purification (Fig. 5).

Taken together, we show that the glucose meter interface has the potential to distribute gene circuit-based sensing capacity outside of the laboratory. This capability could have near-term benefits for global health, especially in the context of the SARS-CoV-2 outbreak and the associated diagnosis bottlenecks across the globe. Much of the current gene circuit-based sensing capacity developed thus far is limited to lab-based assays. By developing this glucose meter interface, we hope to provide an accessible and low-cost means of distributing the capacity of cell-free gene circuit-based sensors to applications beyond the laboratory environment.

### Statistical analysis

Unless otherwise indicated, experimental data sets were compared using two-tailed Welch’s t-test. All statistical analysis and graphing was done using Graphpad Prism 7 software.

## Materials

### General materials

Unless otherwise noted, all chemicals were purchased from Sigma-Aldrich (St. Louis, MO, USA). DNA oligonucleotides were purchased from Integrated DNA Technologies (Coralville, IA, USA) and used without further purification. Cell-free expression buffers and commercial restriction enzymes were purchased from New England Biolabs (NEB, Ipswich, MA, USA). NASBA kits for isothermal amplification were purchased from Life Sciences Advanced Technologies Inc. (St Petersburg, FL, USA). Enzyme substrates were purchased from BioShop Canada Inc. (Burlington, ON, Canada): lactose (LAC234), trehalose (TRE222), glucose 6-phosphate (GPS095) NAD (NAD001).

### Enzyme genes

The gene for lactase was sourced from Pardee et al. (2014)17. The gene for the glucose phosphatase (E. coli, ybiV) was provided by the Yakunin lab at University of Toronto. Trehalase gene (C. japonicus, tre37a) was acquired from DNASU (Arizona State University), contributed by Kelley Moremen (University of Georgia, NIH grant RR005351). Glucose dehydrogenase mutant gene (B. subtilis, variant E170K/Q252L)36 was synthesized by IDT.

### Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.