The human gut microbiota contains a large number of interacting species of bacteria, archaea, bacteriophages, eukaryotic viruses, and fungi which together create a complex ecosystem able to influence human physiology, pathologies, and behavior (Oliphant et al., 2019; Fan and Pedersen, 2020). The microbiome plays important roles in human homeostasis mostly related to metabolism (Smith et al., 2013). Multiple studies have linked abnormal-gut microbiota with altered metabolite profiles of patients with different diseases such as metabolic liver disease (Jiang et al., 2015; Schwenger et al., 2019), inflammatory bowel disease (Duboc et al., 2013; Nikolaus et al., 2017; Franzosa et al., 2019) as well as metabolic disorders like obesity and malnutrition (Ridaura et al., 2013; Smith et al., 2013; Sharon et al., 2014).

The analysis of feces metabolites has opened a new window on the complex interactions occurring within the gut (Wang et al., 2011; Patterson et al., 2016; Woting and Blaut, 2016; Jia et al., 2018). In the clinics, simple and quantitative tests enable measurements of fatty acid content or malabsorption of carbohydrates by analyzing the pH of feces (Caballero et al., 1983; Eherer and Fordtran, 1992). The detection of fecal proteins, particularly calprotectin, helps diagnose and monitor inflammatory bowel diseases (Manceau et al., 2017). In addition, bacterial infections of the gut can be detected using culture-based or molecular genotyping strategies (Karu et al., 2018). Finally, mass spectrometry coupled with liquid chromatography (LC-MS) has been successfully applied to measure the levels of metabolites in human feces allowing the prediction of key associations between diet and microbiome (Bjerrum et al., 2015; Bar et al., 2020). While LC-MS is increasingly used in clinical diagnosis and allows for general and precise metabolic profiling, it is still impractical and expensive for daily monitoring of metabolites (Seger and Salzmann, 2020). Additionally, heavy methods such as LC-MS are not deployable in the field or at home, restricting large-scale prospective routine monitoring of patients. The development of innovative point-of-care (POC) testing could enable real time clinical decision-making, eliminating requirements for specialists to perform and analyze the test. Therefore, new technologies are needed to support fast, field-deployable, and cost-effective detection of the metabolites produced by the microbiome in human samples such as feces.

Programmable bacteria present an attractive technology for engineering portable biosensor devices that could help address these challenges. A biosensor is composed of a biological sensing component, which recognizes a chemical or physical change, coupled to a transducing element that produces a measurable signal in response to the environmental change (Daunert et al., 2000). A wide number of bacterial biosensors have been engineered to detect different types of analytes by connecting natural transcriptional responses to different reporter genes (Chang et al., 2017; Hicks et al., 2020). Bacterial biosensors have significant potential in applications for medical diagnosis as they perform analyte detection with a robust response, high sensitivity, and can be optimized to detect molecules in complex media such as clinical human samples (Courbet et al., 2015; Watstein and Styczynski, 2017; Chang et al., 2021). Moreover, they are inexpensive and easy to manipulate and store. Synthetic biology has enabled bacterial biosensor improvement by providing a large number of standardized genetic parts, together with systematic strategies for organism engineering, resulting in biosensors with a higher specificity, able to detect molecules in a relevant range of concentration (Hicks et al., 2020).

In addition, synthetic biology has demonstrated the promising in vivo application of engineered bacteria for the treatment of diseases, including metabolic disorders (Nelson et al., 2021), infections (Daeffler et al., 2017; Riglar and Silver, 2018), and modulation of the tumor microenvironment (Guan et al., 2021). Characterizing and optimizing the sensing performance of these bacteria under physiological conditions could improve their in vivo applications and therapeutics abilities.

Here we developed a rapid workflow for functional assessment of bacterial biosensors in fecal samples. We characterized the sensing performance of five bacterial biosensors on fecal solutions of IBD patients. Three of the biosensors evaluated are based on cytosolic transcription factor systems: a biosensor responding to benzoate, based on BenR activator and the pBEN promoter (Libis et al., 2016; Zúñiga et al., 2020); a biosensor responding to anhydrotetracycline (aTc), based on TetR repressor and the pTET promoter (Lutz and Bujard, 1997; Courbet et al., 2015), and a biosensor responding to l-lactate, based on LldR regulator and the pALPAGA promoter (Zúñiga et al., 2021). The other two biosensors, TcpP/TcpH and VtrA/VrtC, correspond to transmembrane chimeric receptors for bile salts, activated via ligand-induced dimerization (Chang et al., 2021).

We have assessed the matrix effect of feces on their response, which changed depending on the type of biosensor and the target molecule. Using this method, we were able to detect exogenously added metabolites in presence of fecal solution. Even more, we detected endogenous bile salts in some samples, demonstrating that the performance of bacterial biosensors could be optimized for fecal metabolite detection and used as a monitoring strategy. In the future, this method could be used for rapid pre-prototyping of engineered bacteria designed to operate in the gut, with applications to in vivo diagnosis and therapeutics.

Fecal samples preparation for matrix effect assessment on bacterial biosensor performance.

We first aimed to evaluate the matrix effect of feces on bacterial biosensors functionality. The first step in this process was to determine a simple and robust protocol to prepare a feces-derived solution (fecal solution thereafter) to which biosensors can be exposed. Our aim was to obtain from feces a sample that contains feces-derived metabolites yet is liquid so can be easily and reproducibly handled for testing with bacterial biosensors. The solution samples from 4 patients were recovered (see materials and methods for details) and pooled, to reduce the sample variability in the matrix effect estimation. Four treatment methods were applied to feces solution: 1) no filtration, meaning no change in the original microbial composition of the feces, 2) no filtration + antibiotics, to inhibit the growth of the endogenous bacterial microbiome during the experiment, 3) filtration at 0.45 µm and 4) filtration at 0.2 µm, to eliminate host-derived microorganisms in the samples (Figure 1A). Analysis of colony forming units (CFU) and absorbance at OD600 in LB medium confirmed growth of microorganisms only in fecal solutions without filtration (Supplementary Figure S1A), keeping a portion of the microbial composition able to grow under the conditions in which the bacterial biosensor performs. We evaluated the global matrix effect of pooled fecal samples by using a BenR-pBEN benzoate biosensor, based on a soluble regulator, due to the low concentration of this metabolite in human feces (Jenner et al., 2005; Muñoz-González et al., 2013; Gutiérrez-Díaz et al., 2018). We assessed the activity of the BenR-pBEN biosensor, with a limit of detection of 80 µM (Libis et al., 2016; Zúñiga et al., 2020), to sense exogenously added benzoate in presence of fecal solutions diluted at final concentration of 75%, 50%, and 25% vol/vol (Figure 1A) (see methods). We then evaluated the GFP fluorescence signal of the biosensor in the presence or in the absence of 100 µM benzoate (Figure 1B, Supplementary Figure S1). After 16 h of induction at 37°C, considerable inhibitory effects were observed in the benzoate biosensor output signal in 75% and 50% of feces (Supplementary Figure S1B). On the other hand, 25% of fecal solution allowed for a better sensing performance (Figure 1B). However, the fluorescence measurement was significantly lower in non-filtered samples compared to samples filtered with 0.45 and 0.2 µm (Figure 1B). In addition, a fraction of cells with no detectable fluorescence was observable in non-filtered feces induced with benzoate, probably corresponding to live or dead endogenous host microorganisms present in the matrix, reaching the 30% of the total population in the condition non-filtered but with antibiotics and more than 46% in condition with non-filtered fecal sample (Figure 1B and Supplementary Figure S1). Since the filtering of fecal samples at 0.2 µm did not show a significant improvement from the filtered at 0.4 µm, we chose the filtering at 0.4 µm to reduce the matrix effect of feces as a final protocol step. These results demonstrate that feces samples have significant inhibitory matrix effects on bacterial biosensors’ performance but by filtering and diluting fecal solutions, this effect can be reduced allowing bacterial biosensors to detect exogenously added metabolites.

We then evaluated the sensing performance of different bacterial biosensors in presence of feces samples. We evaluated five different biosensors for: aTc; l-lactate; primary bile acids, taurocholic acid (TCA); secondary bile acids, glycodeoxycholic acid (GDCA); and benzoate. We evaluated the matrix effects on the sensing performance of these biosensors by adding feces samples at 2, 4, and 10-fold dilutions (50%, 20%, and 10% feces respectively, Figure 2A, see methods and Supplementary Figure S2). To observe the variability of the matrix effect we used solutions from six different patients (samples S1-S6). For TcpP/TcpH and VtrA/VtrC, in order to avoid interference from endogenous cognate ligands on the biosensor, the samples were chosen for their low content in total bile acids, as measured by an enzymatic assay, corresponding to samples S4-S6 (Figure 2A, Supplementary Table S2). All biosensors were able to detect their specific inducers in 10% feces samples. However, most of them failed to produce fluorescence in 50% of feces, except for lactate biosensors that showed a higher fluorescence when a higher concentration of feces was added. This was due to the endogenous presence of lactate in the samples (Supplementary Table S2), confirmed by enzymatic assay. Indeed, a feces sample with low measured lactate concentration (Figure 2A, patient S2) did not exhibit such increase in fluorescence. The different bacterial biosensors showed diverse degrees of sensitivity to fecal matrices (Figure 2B, Supplementary Figure S3, Table 1, and Supplementary Table S3). TetR was strongly inhibited (more than 50%), even in 10% feces. Interestingly, the VtrA/VtrC system exhibited a non-homogeneous response and was strongly inhibited too, although relying on a similar architecture to TcpP/TcpH, which was not significantly affected. This difference might be due to the fact that the behavior of TcpP was previously optimized by circuit tuning and directed evolution (Chang et al., 2021).

Importantly, a clear patient-to-patient variability was observed for every biosensor, suggesting that feces matrix effects responsible for biosensor inhibition could be due to specific factors having varying abundance in different patients. These results show that several bacterial biosensors can operate in diluted fecal solution, yet have diverse sensitivity to fecal matrices, requiring case-by-case optimization.

In this work, we provide an optimized method to prepare fecal solutions for prototyping bacterial biosensors in human feces. We showed that a simple filtration step is enough to remove host microbes and reproducibly obtain a physiological-derived media retaining essential characteristics of human feces, such as matrix effects and endogenous metabolites (e.g. bile acids and l-lactate). We found significant inhibitory matrix effects of feces on the bacterial biosensors tested, although the robustness of the different biosensors varied. In addition, matrix effects varied significantly from patient to patient. This patient-to-patient variability could be due to host or microbiome-derived molecules that interfere with biosensor physiology. It is also possible that some medications inhibit the biosensors. Detailed knowledge of patients’ full clinical picture and current treatments will be essential to interpret the biosensor’s response. In all, the biosensors tested here are highly sensitive to fecal samples, and the optimal working conditions in our studies were general at a 10% feces sample dilution in liquid culture. However, we show that encapsulating bacteria in alginate leads improves the robustness of the biosensor against inhibitory matrix effects. We observed a significant improvement in the sensing performance of the benzoate biosensor in alginate beads compared to non-encapsulated cells in presence of 50% of fecal solution, without the need for any other optimization. This improvement may be due to two complementary factors; first, hydrogel beads can locally concentrate bacteria at a higher density, while providing water and nutrients to the cells and allows the diffusion of the inducer molecules to sense. Second, the alginate hydrogel itself offers a physical protection under stress conditions while conserving bacterial biosensing ability, as already shown in previous work by our group and others (Choi et al., 2013; Courbet et al., 2015; Li et al., 2017; Tang et al., 2021).

From our study, we can highlight several important points for future bacterial biosensor prototyping in feces: 1) because of biosensor-specific sensitivity to fecal matrices, case-by-case optimization of every new bacterial biosensor for operation in feces is required, 2) testing the biosensor over various individual samples coming from different patients is critical to obtain a bacterial biosensor working over a wide range of real-world conditions, and 3) encapsulation of biosensors can improve their performance in fecal solution by reducing their susceptibility to matrix effects. Using our workflow, these assays can be performed rapidly.

Interestingly, the TcpP/TcpH bile salts biosensor, that detects primary bile salts, was capable of detecting high concentrations of endogenous bile salts in samples from five different patients, in accordance with enzymatic measurements. Patients with IBD have altered fecal bile salts profiles (Duboc et al., 2013; Torres et al., 2018; Lavelle and Sokol, 2020), with lower levels of secondary bile salts but higher levels of primary (Jansson et al., 2009; Jacobs et al., 2016; Franzosa et al., 2019). For other samples, data was not in accordance with the enzymatic measurements, possibly because of different, patient-specific bile salts profiles, some bile salts not being detected by our bacterial biosensor. A more definitive answer could be provided in the future by using LC/MS to determine the exact bile salts species distribution in each sample. Nevertheless, as a proof of concept, these data demonstrate, for the first time to our knowledge, the possibility of using bacterial biosensors to detect endogenous metabolites in human feces.

How could bacterial biosensors operating in fecal samples be optimized in the future? First, other reporters having a higher signal-to-noise ratio, such as luciferase, might be evaluated. Yet, unless using the luxCDABE operon, which has lower performance, optimized luciferase systems such as nanoluc, while providing a lower limit-of-detection, work better after cell lysis, which would complicate the assay protocol (Lopreside et al., 2019). Second, amplifying genetic devices such as recombinase switches or hrp transcription factors might help combat matrix effects and enable operation at higher concentrations, thereby supporting lower limits of detection (Courbet et al., 2015; Wan et al., 2019). Recombinase-mediated inversion or excision could also allow post facto analysis of biomarker presence through DNA sequencing or PCR (Courbet et al., 2015; Wan et al., 2019). Furthermore, an appropriate coating on hydrogel beads complemented with genetic devices biocontainment (e.g. auxotrophs) (Steidler et al., 2003; Chan et al., 2016; Moya-Ramírez et al., 2022) could not only allow for a better performance of the biosensor on feces but also prevent bacteria leakage in the environment allowing the deployment of these biosensors as a point-of-care metabolite monitoring device.

The method shown here could be performed on a lab-on-chip device enabling successive feces samples filtration, dilution, and sensing assay in an automated manner (Wu et al., 2017, 2018; Arshavsky-Graham and Segal, 2020). Such devices would open the door to field-deployable, point-of-care gut metabolite detection either for diagnostics or epidemiological purposes (Figure 5).

Finally, another potential and compelling application of our method is its use for rapid and simple prototyping of engineered “smart” gut probiotics. Engineered bacteria have recently been developed to detect and/or treat many pathological conditions such as inflammation, diabetes, phenylketonuria, hyperammonemia, and cancer (Isabella et al., 2018; Riglar and Silver, 2018). As of now, these strains have been evaluated in animal or in cellular co-culture models (Mimee et al., 2016; Daeffler et al., 2017; Taketani et al., 2020; Nelson et al., 2021). While providing valuable information, these models present limitations in terms of time, physiological relevance, and amenability to screening. The use of human fecal samples could complement these approaches by providing a fast and efficient method to assess the matrix effects of fecal matter on bacterial sensors and therapeutics and optimize their behavior.