The GBA refers to the bidirectional pathway between the enteric nervous system (ENS) and the central nervous system (CNS), relating neural functions in the brain to intestinal functions in the gut and vice versa (Carabotti et al., 2015). It has been postulated that this connection may be critical to the investigation of mood disorder pathogenesis and treatment, a traditionally neuro-centered field of research, as imbalances in gut microbiota (dysbiosis) have been found to contribute to the development of various mood disorders such as generalized anxiety disorder, chronic stress, and MDD (Carabotti et al., 2015; Caspani et al., 2019).

Current monitoring and diagnostic efforts of the GI tract are primarily focused on fecal metabolomics. This method is noninvasive, inexpensive, and repeatable, however, multiple studies have shown discrepancies in the accuracy of using fecal samples as a proxy for mucosal microbiota representation (Rangel et al., 2015; Ringel et al., 2015; Tap et al., 2017). On the other hand, sampling directly from the GI tract as is achieved with endoscopic biopsies, is a more reliable and accurate method of characterizing this unique environment. However, this procedure poses the issue of extreme invasiveness which results in low patient compliance (Nagata et al., 2019). As such, present methods involve significant drawbacks and ultimately do not provide an accurate enough reflection of the composition of the gut microbiota (Tang et al., 2020).

Ingestible biosensors mitigate many of these concerns as they are relatively non-invasive, patient-friendly, and easily repeatable, all while still sampling directly from the GI tract. By monitoring internal molecular components, ingestible biosensors have the potential to equip researchers with a more accurate way of investigating the GBA, providing a means for early detection of diseases that have not yet externally manifested, or providing confirmation of disorders that are currently measured by qualitative methods such as mood disorders (Haleem et al., 2021).

According to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), in order to be diagnosed with MDD, a patient needs to experience five or more depressive symptoms for more than 2 weeks; these symptoms must impair an individual’s ability to function on a daily basis, conduct their typical work and navigate relational and educational pursuits (Table 1). As such, diagnosis is primarily qualitative, heavily relying on patient reporting and physician comprehension. This results in subjective symptomatology, especially as the professional understanding and categorization of MDD has changed over time. In 2013, the DSM-IV transitioned to become the DSM-V, which introduced many changes such as discrimination between single and recurrent depressive episodes (drawing more attention to persistent episodes) and acknowledged that bereavement can potentially lead to symptoms characteristic of MDD (as first seen in the DSM-IV-TR when the DSM-IV excluded bereavement from depressive disorders) (Table 1). These alterations in diagnostic thresholds can lead to over- and or misdiagnosis in MDD, and lack consistency which is needed for accurate diagnosis.

The number of individuals diagnosed with MDD has seen a huge increase since 2005; especially in the past 3 years, a greater disease burden has been observed among adolescents and young adults (Proudman et al., 2021). Many factors have been linked to the increased prevalence of the condition including: greater awareness and education of mental health, growing economic concerns in different regions, and the widely experienced reality of social isolation and limited access to healthcare during the COVID-19 pandemic.

Lockdown and other public health recommendations in place from 2020 to 2022, have had a tremendous impact on peoples of all ages, with feelings of loneliness and isolation being a common finding. Results from a cross-sectional online survey distributed in 101 countries indicated that 21% of respondents experienced severe and debilitating loneliness during the pandemic, correlating to a 28% increase in the number of individuals affected by MDD in a single calendar year (O’Sullivan et al., 2021; World Health Organization, 2022a; World Health Organization, 2022b).

However, the burden of depression extends beyond an individual’s experience of the condition, as it can shape the lives of those in their community and social circles. In Canada, for example, it is estimated that the sole burden of mental illnesses and addictions in Ontario is more than 1.5 times that of all cancers, including the combined local burden of breast, colorectal, lung, and prostate cancers (Ratnasingham et al., 2013).

Depressive mood disorders have also been reported as one of the leading causes of disability worldwide, costing billions of dollars per year through losses in productivity during working hours and associated healthcare expenses. Canadians living with anxiety and depression can face unemployment rates of up to 30%, which is further hindered by challenges to receive an accurate diagnosis and have access to necessary health services (‘Unemployment, Mental Health and Substance Use’, 2014). According to the World Health Organization, depression is also a major contributor to suicide across the globe (World Health Organization, 2017), however, there remains serious gaps in appropriate support, treatment, and funding of such programs.

Thus, there is a need to complement current qualitative methods of MDD diagnosis, with a more objective and quantitative approach. This has the potential to improve diagnosis accuracy and gain a better understanding of the disorder physiologically, feeding into more informed and personalized treatment plans.

There exists in literature, a clear link between MDD and the GBA (Figure 2). Taxonomic association studies in humans have identified a high prevalence of microbiome dysbiosis in MDD patients compared to healthy individuals (Ressler and Mayberg, 2007; Naseribafrouei et al., 2014; Jiang et al., 2015; Zheng et al., 2016; Caspani et al., 2019), and numerous studies using germ-free mice have also revealed the detrimental effects of uncolonized and dysbiotic gut microbiomes (Clarke et al., 2013; Luczynski et al., 2016; Matcovitch-Natan et al., 2016). One study by J. Zhang et al. (2020) demonstrated that depression-like behaviors can be induced in germ-free mice who have received transplantation of microbiota derived from MDD patients.

While the pathophysiology of MDD in relation to the GBA is not fully elucidated, the vagus nerve has been highlighted as a critical player in facilitating the effect of neuroactive gut metabolites on ENS development and CNS signaling (Braniste et al., 2014; Erny et al., 2015; Ogbonnaya et al., 2015; Morais et al., 2021). Additionally, the GBA relies on proper functioning of the blood-brain barrier (BBB) and the intestinal epithelial barrier (IEB), which are both held together by tight junction proteins, and is highly sensitive to inflammatory cytokines and endotoxins (Geng et al., 2020). Consequently, endotoxins that permeate through the IEB via the vagal afferent nerve, can disrupt neuronal signaling pathways within the brain.

Reporter systems are coding regions that can be linked to a promoter and subsequently produce outputs that are clearly distinguishable from the background activity of a cell (Li et al., 2018). They are meant to provide a means of measuring the promoter activity, and by extension, molecular species or cell pathways which regulate that promoter. With the emergence of DNA writing and sequencing technologies, researchers have developed long-term molecular recording platforms with genetic based reporters in living cells which hold great potential to serve as a reporter system for our ingestible sensor.

SCRIBE (Synthetic Cellular Recorders Integrating Biological Events), a strategy developed by Farzadfard and Lu (2014), generates intracellularly expressed single-stranded DNA (ssDNA) in response to various regulatory signals. The ssDNAs are recombined into genomic DNA at specific loci by co-expressed recombinases, resulting in precise point mutations which accumulate as a function of signal magnitude and duration. However, due to its limited recombination rate, SCRIBE requires actively dividing cells to achieve population-memory; thus, the platform is more suitable for recording across collective genomes of cellular populations, rather than editing at the single-cell level.

To overcome this dependence on large cell populations, Tang and Liu (2018) developed two CRISPR-mediated analog multi-event recording apparatus (CAMERA) systems, enabling analog recording within each cell. The systems use the CRISPR-Cas9 nuclease and Cas9-derived base editors, which are expressed on writing plasmids in response to stimuli of interest, to generate permanent DNA sequence modifications at guide RNA-specific loci on recording plasmids. Both systems achieved stable recording of the amplitude or duration of various signals; this can be extended to record multiple metabolite concentrations through assignment of associated mutations to unique locations on the plasmid. The ratio of edited to unedited DNA on these recording plasmids were shown to accurately record the duration and intensity of both chemical and light-based stimuli.

The DNA-based Ordered Memory and Iteration Network Operator (DOMINO) is another genetic based recording platform that uses CRISPR base-editing technology to manipulate DNA in response to signals of interest (Farzadfard et al., 2019). By linking the guide RNA and the Cas9 complex to separate inducible promoters, the system can be controlled by an operational signal and another independent input, thereby forming a DOMINO operator that can only generate mutations when both inducers are present.

We aim to achieve targeted sensing of the large intestine for two main reasons: 1) the metabolites we aim to study have been predominantly characterized in fecal samples, which are generally considered a proxy for large intestine conditions due to their close proximity (Gevers et al., 2014), and 2) bacterial populations in the large intestine have greater numbers, increased diversity, and more stability than the small intestine (Kastl et al., 2019).

The main goal of this model is to mathematically simulate the proposed base editing system to analyze its behavior in different environments, verifying that the system achieves AND-gate functionality, and thus, spatiotemporal specificity.

The model predicts how riboflavin and metabolite concentrations induce environment-specific plasmid editing in the large intestine. From the sensing of riboflavin and individual metabolites up to plasmid editing, the following major reactions were characterized in the model: riboflavin-regulated BE2 and sgRNA1 production, metabolite-regulated sgRNA2 production, sgRNA and BE2 dimerization, BE2 and recording plasmid complexation, and base editing (Supplementary Figure S1).

To simplify the simulation of the base editing system: 1) only one bacterium was included in the model, and 2) the sensing and recording of only one metabolite, indole, was simulated as a placeholder for other metabolites of interest.

Based on our kinetic simulation, the relationship between metabolite concentration and base-editing is logarithmic, not linear (Figure 9A). Furthermore, variations in riboflavin concentration would have a direct impact on base editing at the metabolite-associated recording site (Figure 9B). Thus, a normalization model was constructed to achieve two aims: 1) controlling the confounding effect of riboflavin concentration when using base-editing ratios to more accurately infer metabolite concentrations, and 2) creating a linear relationship between actual and predicted indole concentrations in order to more accurately assess fold-changes in metabolite concentrations. Data from the kinetic simulation at 10 h was used to determine the relationships and constants required for this model (Supplementary Table S2). This normalized model is intended to be deployed after extracting and reading the biosensor output; it uses final base editing ratios for each site as inputs, and returns a relative metabolite level with arbitrary units that has been normalized for riboflavin concentrations.

Based on exploratory analysis (Supplementary Figure S2), the relationship between riboflavin concentrations and base-editing at each of the two locations can be defined by: Ratio 1 = β 1   ⋅ Riboflavin + α 1 (8) Ratio 2 = β 2   ⋅   log Riboflavin + α 2 (9)

Hence, to nullify the relationship between riboflavin and ratio 2, both sides are divided by the right-hand side of Eq. 9: Ratio 2 β 2   ⋅ log Riboflavin = 1 (10)

Substituting from Eq. 8: Normalized   ratio = Ratio 2 β 2   ⋅ log Ratio 1 −   α   1   β 1   2 + α 2 (11)

Given the logistic relationship between the normalized ratio and indole concentrations (Supplementary Figure S3), the prediction was linearized using: P r e d i c t e d   = exp   N o r m a l i z e d   r a t i o − β 3 α 3 (12)

Linear regression showed that this model achieves a strong, linear relationship between the actual and predicted indole concentration when varying indole levels (Figure 9C: R² = 0.992, p = 1.19 × 10⁻⁹). Furthermore, this model successfully removes the confounding effect of riboflavin: when indole was held constant and riboflavin was increased, a logarithmic increase in base-editing at the indole site (Figure 9B) was normalized such that a linear regression model demonstrated no significant relationship between riboflavin and predicted indole levels (Figure 9D: p = 0.149).

The short chain fatty acid (SCFA), butyrate, is a gut-derived metabolite that has shown promising potential in alleviating depressive symptoms in patients with MDD through its anti-inflammatory properties. Additionally, studies have shown that butyrate-treated microglia respond to LPS stimulation by decreasing the production of cytokines such as TNFa and overall dampening neural inflammation (Park et al., 2019; Silva et al., 2020; Caetano-Silva et al., 2023).

Studies also point to the role of butyrate in the stimulation of afferent fibers of the vagus nerve, which has been shown to suppress depressive symptoms in mild and severe MDD patients (Peña et al., 2014; Stilling et al., 2016; Bonaz et al., 2017; 2018; Cawthon and De La Serre, 2018). Though the mechanism of action is not fully elucidated, some hypothesize that this stimulation alters the expression of receptors such as GPR41, which is found on enteric nerve fibers and interacts with SCFAs to allow for direct communication from the gut to the CNS (Schroeder and Bäckhed, 2016; Bruning et al., 2020).

Studies have shown that butyrate-producing Faecalibacterium and Coprococcus were consistently found at lower levels in MDD patients compared to controls (Jiang et al., 2015; Zheng et al., 2016; Valles-Colomer et al., 2019). Specifically, F. prausnitzii, a butyrate producing bacteria that comprises up to 14% of the total fecal gut microbiota (Rivière et al., 2016), was found at relatively decreased levels across multiple studies (Jiang et al., 2015; Zheng et al., 2016; 2020; Chen et al., 2018; 2020; Rong et al., 2019; Stevens et al., 2021). This was further supported by a study with 165 young adults, which showed a negative association of fecal butyrate concentrations with depressive symptoms (r = −0.195) (Müller et al., 2021).

Tryptophan, an essential amino acid necessary for protein synthesis, plays a key role in the GBA; specifically, its metabolic pathway produces three critical metabolites: kynurenine, indole, and serotonin (Roth et al., 2021). The majority of tryptophan (95%) undergoes conversion into kynurenine which is subsequently degraded into neuroprotective kynurenic acid and neurotoxic quinolinic acid (Badawy, 2017; Savitz, 2019), whereas serotonin and indole metabolites are produced to a lesser extent. The intricate relationship between the tryptophan pathway and MDD is demonstrated by a breadth of research.

Tetrahydrofolate (THF), a folate (vitamin B9) derivative and precursor to L-methylfolate (5-MTHF), is a metabolite produced by gut microbiota for various essential biosynthetic processes such as DNA and amino acid synthesis (Khan and Jialal, 2022). Neural cells also utilize 5-MTHF, taken up through folate receptors and transporters, for crucial processes such as neural cell division, and tissue repair (Balashova et al., 2018). Additionally, 5-MTHF binding triggers specific signaling mechanisms such as neurulation (Ami et al., 2016), and modulates synaptic activity. THF deficiency can therefore lead to impairment of cell division and neurological disorders (Khan and Jialal, 2022).

Approximately one-third of depressed individuals are folate deficient (Young and Ghadirian, 1989; Bottiglieri et al., 1992; Bottiglieri et al., 2000). Presently, studies support the link and suggest that folate supplementation, particularly with 5-MTHF, is beneficial for depressed patients (Nelson, 2012; Bender et al., 2017; Altaf et al., 2021). Furthermore, a mutation in the MTHFR gene, an enzyme required to convert 5-MTHF from THF, was found in MDD patients leading to 5-MTHF unavailability (Gilbody et al., 2007a; Gilbody et al., 2007b). Additionally, 5-MTHF adjunct therapy with selective serotonin reuptake inhibitors (SSRIs) or serotonin-norepinephrine reuptake inhibitors (SNRIs) was investigated and proven effective in MDD patients (Altaf et al., 2021; Lam et al., 2022), increasing remission rates by 39% compared to patients who only received SSRIs or SNRIs.

It is believed that 5-MTHF supplementation benefits MDD patients through tetrahydrobiopterin (BH4) regeneration (Altaf et al., 2021) as 5-MTHF is further processed into BH4 in the brain (Akoglu et al., 2008; Miller, 2008). BH4, a rate-limiting cofactor for hydrolase enzymes such as tyrosine and tryptophan hydrolase, is required for the synthesis of various monoamine neurotransmitters including serotonin, dopamine, norepinephrine, and epinephrine (Miller, 2008). Lower BH4 levels are also found in depressed individuals (Coppen et al., 1989; Anderson et al., 1992; Nasser et al., 2014), further implying a relationship between BH4 and MDD.

Finally, we propose tetrathionate as a novel indicator for MDD. Tetrathionate is a commonly used marker of gut inflammation for diseases such as inflammatory bowel disease (IBD). There exists a fair library of research linking MDD to IBD and its subtypes, Crohn’s disease (CD) and ulcerative colitis (UC). Shared factors between the two disorders include the dysregulation of the GBA (Fadgyas-Stanculete et al., 2014), microbiota and IEB function (Jeffery et al., 2012; Kelly et al., 2015; Yuan et al., 2021), stress (Mawdsley and Rampton, 2005; Goodhand et al., 2012), immune dysregulation (Persoons et al., 2005), and increased circulating proinflammatory cytokines (Alexakis et al., 2017; Facanali et al., 2023). Additionally, IBD and MDD also share disease implicated metabolites such as SCFAs and those involved in the tryptophan pathway (Banfi et al., 2021). Therefore, the relationship between tetrathionate and IBD can rationally be extended to MDD.

Studies show higher rates of depression in patients diagnosed with IBD compared to healthy controls (Goodhand et al., 2012; Byrne et al., 2017). Specifically, one study by Addolorato et al. (1997) showed a significantly higher percentage of individuals with depression in CD and UC patients than in controls (41.9% versus 11.1%, and 50.0% versus 11.1%, respectively). Some studies also indicate that disease activity is influenced by depression and that disease duration can be inversely correlated with MDD presence (Goodhand et al., 2012; Byrne et al., 2017; Facanali et al., 2023). Further, a prospective study by Persoons et al. (2005) found that after treatment with infliximab, an anti-TNF therapy, patients with MDD at baseline achieved remission significantly less (29%) in comparison to non-depressed patients (70%) and required re-treatment.

The pathology of IBD has eluded researchers for years, however, recently the possibility of MDD as a causal or resultant factor of inflammatory gut disorders has been introduced. For example, UC may potentially be caused by a disturbed inflammatory response toward a set of specific gut microbiota, or, gut microbiota may interact with the CNS through the immune system and circulating metabolites such as neurotransmitters can subsequently affect an individual’s mood (Yuan et al., 2021). Thus, there exists a space to explore the bidirectionality of the GBA and the many gut-brain or brain-gut mechanisms that may be in play (Collins, 2020; Günther et al., 2021).

One of the primary concerns regarding the use of ingestible biosensors lies in the logistics of device encapsulation and device retention within the GI tract. A majority of ingestible biosensors are composed of rigid and non-degradable materials, which can potentially result in damage to the GI tract should the device be retained for more than six to eight hours depending on individual GI movement (Watson, 2020; Lu et al., 2023). Approximately 1.4% of patients experience capsule retention, therefore, precautions such as utilizing soluble and malleable materials can be exercised to reduce rates of retention and blockage (Lu et al., 2023). This is important to consider as trauma to the gut can cause gastrointestinal perforations, leading to internal bleeding or peritonitis (Revell et al., 2018). However, as described earlier, our biosensor pill will be administered in a highly controlled clinical setting with stringent subject inclusion criteria. This will include screening out individuals with digestive or GI motility issues which can be monitored using barium X-rays as described previously (Palmer et al., 2011). If the pill is retained in the GI tract, it can be removed by either medications or endoscopic retrieval to prevent any potential complications associated with long-term capsule retention (Rondonotti et al., 2010).

The implementation of a killswitch has yet to be added to our system; this is important to consider as the system is biological instead of chemical or electrical. While the issue of patient privacy is avoided without the use of electronics, the concern of potential contamination either to the environment or the human body is of concern and must be addressed. Thus, future studies should consider implementing a system that can neutralize the bacteria via expression of toxins or lysis protein within the pill in the presence of an external chemical stimulus, as proposed by Rottinghaus et al. (2022). This ensures that in the given scenarios where a patient desires to withdraw from the study, the engineered bacteria can be safely disengaged.

It is important to note that this system as a whole remains an unvalidated hypothesis, therefore initial tests should focus on validating its functionality in a tightly-controlled in vitro environment. Furthermore, our design is also conducive for use in animal models to further test for functionality and safety (Turner et al., 2011). Specifically, large animal models are more appropriate than small rodent models, as the pill would not need to be dramatically resized and risk impacting its capacity and functionality. Additionally, porcine or canine models have demonstrated a more analogous GI tract to humans than mice (Ziegler et al., 2016; Ribitsch et al., 2020).

There are several key factors that must be considered before any practical application is possible.

First, the ability to distinguish biologically relevant differences from non-ideal behavior of the bacteria (i.e., signal-to-noise ratio) is a large concern. Thus far, genetic reporter systems have been validated in vitro using high inducer concentrations to elicit strong promoter responses. When seeking to translate this technology for in vivo applications, a number of environmental effects will become relevant and influence the behavior and efficiency of the system. Differences in metabolites and salts in the bacterial environment, for example, may affect the ability of the Cas9 complex to associate with DNA (Zhang, S., et al., 2020). Cell-specific machinery may also exhibit inhibitory effects that can further modulate base-editing (Wang et al., 2017). At the same time, inducer concentrations will vary less drastically between groups, especially when investigating mood-disorder related biomarkers which often have small effect sizes that render many results non-significant (Kennis et al., 2020). Hence, the effect of these variations in comparison to the effect size of metabolites being studied must be considered to determine the degree of precision that is required to produce the desired response. It is also important to note that any quantitative data from our sensor that is collected is not useful in and of itself. The collected data must be correlated to MDD symptomatology and severity by monitoring individuals over time to provide useful insights applicable to studying the disease.

Second, the growth rate of the engineered sensing bacteria within each compartment must be considered as base editing is strongly dependent on the bacterial generation of plasmids containing each sensing system. There are a number of factors that can influence bacterial growth, of which the growth medium is the most pertinent. Based on past ingestible biosensors which have utilized a semi-permeable membrane to separate the bacterial compartment from the external environment (Mimee et al., 2018), nutrient availability will greatly depend on local GI environments; this can influence bacterial growth rate and therefore the generation of sensing plasmids. Additionally, the duration of time between ingestion to excretion, and within each local GI environment, can vary from individual to individual depending on a number of factors. As such, further studies using ex vivo and in silico models to simulate the gut microbiome and device trajectory through specific GI segments can be used to correlate ingestion duration with optimal bacteria growth phases. This can also be used to provide crucial information pertaining to the initial seeding density of bacteria within each compartment and determine if an optimal concentration is required prior to entering the target environment for optimal metabolite recording. In this way, the base editing output can also be normalized post hoc to the bacterial growth rate.

The total manufacturing and operational cost of our pill design can be calculated by summing the total material, manufacturing and operational cost for each use of the pill.

The three-layer outer coating of the pill requires a 0.2694 mL solution consisting of Eudragit L100 dissolved in isopropanol and acetone, as described by Khan and Jialal, 2022. For the internal scaffolding, approximately 1,278 mm³ Type I bacterial cellulose is required for the construction of each pill. The sensing system is primarily composed of E. coli which after initial construction is indispensable and would not need to be factored into the cost. Using these metrics, the material cost per pill is approximately $0.427 USD.

According to well established next-generation sequencing platforms like Illumina, the operational costs of high throughput sequencing or targeted gene expression sequencing is approximately $23.00 USD per 150 bp, for a sequencing depth of 50, 000 copies. Thus, for our 400 bp sample, the approximate sequencing cost per pill is $61.33 USD.

It is difficult to present an exact measure of the labor and overhead costs associated with manufacturing the material, however, an educated estimate based on previous literature concerning similar products such as ingestible pills, biosensors, and other sequencing material would expect our metabolite biosensor cost to be around $90 USD - $100 USD per pill (Koklu et al., 2022). Therefore, the overall cost to produce our biosensor would be approximately $160 USD per pill.