Alzheimer's disease (AD), a devastating neurodegenerative disease, constitutes a major global health challenge. For more than a century, AD research has largely centered on one of its core pathological hallmarks: the extracellular deposition of β-amyloid (Aβ), which aggregates into amyloid plaques. However, therapeutic strategies targeting Aβ have yielded only modest clinical efficacy, compelling the scientific community to reconsider the complete and complex etiology of AD and to explore more fundamental pathogenic drivers.
Among the numerous risk factors for late-onset Alzheimer's disease (LOAD), which accounts for over 99% of all cases, the ε4 allele of the apolipoprotein E (APOE) gene is recognized as the strongest genetic risk factor (Blumenfeld et al., 2024). Recent studies, particularly in-depth, cell-type-specific analyses, have elucidated the key mechanisms of action for APOEε4. APOEε4 is expressed in multiple cell types within the central nervous system—including neurons, astrocytes, and microglia—where it induces a range of detrimental functions. A key downstream effect is the initiation and exacerbation of chronic, destructive neuroinflammation. Compared to the functional ε3 allele, the ε4 variant is significantly associated with more severe gliosis, elevated release of pro-inflammatory cytokines, and the transition of microglia to a disease-associated microglia (DAM) phenotype. This persistent, APOEε4 -driven neuroinflammation leads directly to excessive synaptic pruning, neuronal dysfunction, and ultimately cell death, culminating in progressive cognitive decline (Zhou et al., 2023).
Consequently, a more forward-thinking therapeutic paradigm for AD is emerging. This approach marks a shift from focusing merely on Aβ as a pathological manifestation to intervening in the core pathological process driven by the genetic wellspring of APOEε4 : neuroinflammation (Hering et al., 2025). Directly targeting and mitigating neuroinflammation provides a more robust rationale and novel therapeutic opportunities for protecting neurons at their source and slowing disease progression. Motivated by this understanding, we are dedicated to developing an innovative therapy capable of precisely responding to AD-associated pathological signals and selectively inhibiting neuroinflammation.
Figure 1. Cell type-specific APOE4 cascade model of AD and related therapeutic strategies. (Reproduced from Blumenfeld, J., Yip, O., Kim, M.J. et al. Nat Rev Neurosci 25, 91–110 (2024).)
Butyrate, a short-chain fatty acid (SCFA) produced by the gut microbiota through the fermentation of dietary fiber, possesses biological functions that extend far beyond traditional energy metabolism. It serves not only as a potent anti-inflammatory mediator but also as a key epigenetic modulator, capable of acting as an inhibitor of histone deacetylases (HDACs) to regulate the expression of genes associated with neuronal plasticity and memory formation.
The therapeutic potential of butyrate in Alzheimer's disease (AD) has been demonstrated in relevant animal models. A pivotal study revealed that intraperitoneal injection—a form of systemic administration—into the APPswe/PS1dE9 mouse, a classic AD model, completely reversed existing contextual memory deficits (Kilgore et al., 2010). This crucial in vivo evidence strongly indicates that butyrate, or its downstream effectors, can effectively act on the central nervous system and substantially ameliorate AD-related cognitive impairments.
To further validate the molecular and cellular basis of this therapeutic strategy, we conducted two lines of supporting investigation. First, to directly verify the prerequisite for butyrate's action in the brain, we developed an in silico model to predict its blood-brain barrier (BBB) permeability. The simulation results confirmed that butyrate molecules can indeed traverse this critical physiological barrier. Second, focusing on its anti-inflammatory function, our preliminary in vitro cellular assays demonstrated that in inflammatory models of microglia and hippocampal neurons induced by lipopolysaccharide (LPS), butyrate effectively suppressed the release of inflammatory cytokines and protected neurons from damage.
By integrating the in vivo cognitive benefits reported in the literature with our own supporting evidence from both in silico modeling of BBB permeability and in vitro anti-inflammatory assays, we have established butyrate as an ideal therapeutic molecule capable of targeting the core AD pathology of neuroinflammation. This provides a direct and solid foundation, supported by both computational and experimental data, for the overall design of our project.
Concurrently, for the final output module—the butyrate synthesis pathway—we adopted a strategy of adaptation and integration. Rather than creating a pathway from scratch, we opted to implement the thoroughly validated and highly efficient butyrate synthesis scheme published by Wu et al. in the journal Gut Microbes. This scheme establishes a complete, five-step pathway from acetyl-CoA to butyrate by heterologously expressing four key enzymes (Hbd, Crt, Ter, BCoAT) and leveraging the endogenous yeast protein Erg10 (Wu et al., 2024).
Figure 2. Butyrate production of yeasts with recombinant plasmids. The chart shows the butyrate production for engineered yeast strains (JWY1-JWY5) expressing various combinations of genes. Strain JWY2, notably, exhibited a butyrate titer of 266 mg/L. Reproduced from Wu et al. (2024)
The specific steps of this synthesis pathway are as follows:
Step 1 (Thiolysis):
Two molecules of acetyl-CoA are condensed to form one molecule of acetoacetyl-CoA. This reaction is catalyzed by the endogenous yeast protein Erg10 and thus requires no exogenous gene.
Step 2 (Reduction):
Acetoacetyl-CoA is reduced to 3-hydroxybutyryl-CoA. This step is catalyzed by 3-hydroxybutyryl-CoA dehydrogenase (Hbd).
Step 3 (Dehydration):
3-hydroxybutyryl-CoA is dehydrated to form crotonyl-CoA, a step catalyzed by crotonase (Crt).
Step 4 (Reduction):
Crotonyl-CoA is subsequently reduced to butyryl-CoA, catalyzed by trans-2-enoyl-CoA reductase (Ter).
Step 5 (CoA Transfer):
Butyryl-CoA is converted to the final product, butyrate. This reaction is catalyzed by butyryl-CoA:acetate CoA-transferase (BCoAT).
Figure 3. The Butyrate Synthesis Pathway.
Adopting this established pathway effectively reduces the technical risks associated with the output module, allowing us to concentrate our efforts on the innovative design of the core sensing and regulatory systems.
The gut-brain axis is a critical bidirectional communication network linking the intestinal microbiota to the functions of the central nervous system (CNS). Recent research has clearly revealed that this axis is not merely a regulator of physiological functions but also a highly promising "actionable therapeutic target," offering a novel avenue for intervening in neurodegenerative diseases. Given that patients with Alzheimer's disease (AD) commonly exhibit gut dysbiosis and impaired intestinal barrier function, direct and precise intervention at the intestinal level to influence cerebral pathology has emerged as a highly compelling, cutting-edge strategy (Loh et al., 2024).
However, the effective delivery of beneficial metabolites such as butyrate to the brain poses significant pharmacokinetic challenges. Traditional oral supplementation with butyrate (e.g., in capsule form) is inefficient because upon reaching the distal gut, it is extensively consumed as the primary energy source for colonocytes. This pronounced first-pass effect results in only a minimal concentration of butyrate entering systemic circulation, making it exceedingly difficult to achieve a therapeutically effective concentration in the cerebrospinal fluid.
To overcome this bottleneck, we propose an innovative Live Biotherapeutic Products (LBPs) strategy. Instead of relying on conventional external administration, we will utilize an engineered probiotic—our intended chassis is Saccharomyces boulardii—designed to colonize the patient's gut. There, it will function as an "intelligent drug factory" for the in situ, continuous, and responsive production of butyrate. This synthetic biology-driven approach of in situ synthesis aims to alter the pharmacokinetic profile, bypass the first-pass effect, and thereby stably elevate butyrate concentrations in both systemic circulation and, ultimately, the cerebrospinal fluid, achieving a more efficient and precise therapeutic delivery.
Figure 4. Microbiota-gut-brain axis in Alzheimer's disease.(Reproduced from Loh, J.S., Mak, W.Q., Tan, L.K.S. et al. Signal Transduct Target Ther 9, 37 (2024))
To ensure both the engineering efficiency and the practical viability of this project's final application, we have devised a two-step yeast chassis selection strategy: "prototype first, then transplant." The rationale for this approach has been successfully validated in prior cutting-edge research on live biotherapeutics (Chen et al., 2020).
The ultimate therapeutic goal requires our Engineered Living Therapeutic (ELT) to function safely and effectively over the long term within the complex environment of the human gut. In this context, Saccharomyces boulardii stands out as the indisputably ideal application chassis. As the only yeast approved by the U.S. Food and Drug Administration (FDA) for use as a probiotic, S. boulardii is not only supported by extensive clinical data confirming its safety. Furthermore, it thrives at 37°C and exhibits enhanced tolerance to harsh intestinal conditions such as gastric acid, making it better suited for colonization and persistence in the human body.
However, the de novo development and iterative optimization of complex gene circuits directly in S. boulardii present significant technical challenges. The genetic toolbox for S. boulardii is far less extensive than that for model organisms, and its transformation efficiency is extremely low—sometimes hundreds of times lower than that of standard laboratory yeast. Furthermore, due to practical laboratory constraints, we do not have immediate access to an auxotrophic strain of S. boulardii, which prevents direct engineering as we cannot effectively screen for positive transformants.
Therefore, adopting the workflow from this successful precedent, we rationally selected the laboratory model yeast—Saccharomyces cerevisiae INVSc1—as our initial "prototyping chassis." S. cerevisiae INVSc1 provides an ideal platform for accelerating the Design-Build-Test-Learn (DBTL) cycle for complex gene circuits, owing to its well-defined genetic background, vast library of standardized biological parts, and mature, high-efficiency transformation methods. Moreover, this strain is auxotrophic for four markers (leucine, tryptophan, uracil, and histidine), facilitating immediate and effective selection of transformants.
In summary, our overall strategy is to first complete the validation of all core functional modules on the efficient "prototyping platform" of INVSc1 before transferring the perfected system to the "clinical application platform" of S. boulardii. The core principle of this approach is the decoupling of two complex scientific challenges: innovative gene circuit design and final application chassis adaptation. This allows us to accelerate the Design-Build-Test-Learn (DBTL) cycle for the core functionalities, thereby advancing the project from blueprint to practical application with minimal risk and maximum efficiency.
The design of our sensor began with a series of rigorous evaluations into the scientific reliability and engineering feasibility of the primary input signal, trimethylamine (TMA). We selected TAAR5 as the TMA receptor based on its well-defined biological function: as a trace amine-associated receptor (TAAR), it recognizes and responds to volatile amines in the olfactory system. A key GWAS study provided further compelling genetic evidence, demonstrating a direct correlation between variants in the human TAAR5 gene and the ability to perceive TMA, which offered a solid theoretical basis for our design (Gisladottir et al., 2020).
First, regarding reliability, our literature review revealed a key discrepancy: although gut dysbiosis is prevalent in AD patients, its net effect on total TMA production was unclear. We therefore analyzed public metagenomic datasets to compare the gut microbiota of AD patients (at the MCI stage) with that of healthy controls, specifically targeting the key TMA-producing functional genes, cntA and cutC. The analysis uncovered a specific alteration in TMA-generating pathways: while the abundance of functional genes from different pathways varied, the key gene from the choline metabolic pathway, cutC, was elevated in the gut of MCI-stage AD patients.
Figure 5. Relative abundance of TMA-producing functional genes in the gut microbiota of patients with Mild Cognitive Impairment (MCI) versus healthy controls. The figure shows the relative abundance of several key functional genes involved in the trimethylamine (TMA) metabolic pathway in healthy controls and MCI patients. The results clearly indicate that, compared to the healthy group, the abundance of the key gene from the choline pathway, cutC, is elevated in the gut microbiota of MCI patients.
Second, to evaluate the engineering feasibility of this signal from both a structural and a systemic perspective, we conducted in-depth in silico simulations. On one hand, we obtained a high-quality predicted structural model of the TMA receptor, TAAR5, from the AlphaFold Protein Structure Database and performed molecular docking simulations. On the other hand, we developed a parallel physiologically based pharmacokinetic (PBPK) model. However, these two lines of analysis converged on a potential issue: the molecular docking predicted a high dissociation constant (K_d), implying that a high concentration of TMA would be required for activation, while the PBPK model estimated that physiological TMA concentrations in the gut of both healthy individuals and AD patients were unlikely to reach this theoretical activation threshold.
This simulation result stood in stark contrast to published large-scale human genetic evidence, which shows a very strong association between functional polymorphisms in the TAAR5 gene and the human olfactory perception of TMA (Gisladottir et al., 2020). This strongly suggests that TAAR5, a trace amine-associated receptor, should possess sufficient sensitivity to TMA under physiological conditions. Faced with this conflict between our in silico model and published evidence, we concluded that direct wet-lab experimentation was necessary to definitively determine the true response capacity of our engineered TMA sensor. Based on this reasoning, we designed a specific gene circuit to experimentally address this critical question.
Figure 6. Plasmids for the single-input TMA sensor system. (A) The sensor plasmid pESC-URA-PTEF1-TAAR5. (B) The reporter plasmid pGBKT7-PFBP1-EGFP. (C) The reporter plasmid pGBKT7-PMSN2-EGFP. (Generated with SnapGene®.)
The core of this system is a dual-plasmid sensing strategy. The sensor plasmid, pESC-URA-PTEF1-TAAR5, was constructed on a high-copy pESC-URA backbone with a URA3 selection marker. On this plasmid, the strong constitutive promoter PTEF1 drives the continuous, high-level expression of the human TMA receptor TAAR5, which is fused to an N-terminal yeast membrane-targeting peptide. TAAR5 is a classic G protein-coupled receptor (GPCR) that, with this N-terminal modification, is designed to be anchored to the yeast cell membrane as a "signal antenna."
Our decision to add an N-terminal leader sequence to TAAR5 is based on an established strategy for optimizing heterologous GPCR expression in yeast. Studies have shown that expressing human GPCRs in yeast can lead to issues such as incorrect protein processing, retention in the endoplasmic reticulum, or improper localization to the cell membrane. To overcome this bottleneck, fusing a leader sequence to the N-terminus of a GPCR has been proven to facilitate its correct insertion and functional expression in the plasma membrane (Lengger & Jensen, 2020). We therefore incorporated this design to enhance the expression abundance and functional integrity of TAAR5 on the yeast cell surface.
The accompanying reporter plasmids, pGBKT7-PFBP1-EGFP and pGBKT7-PMSN2-EGFP, were built on the pGBKT7 backbone with a TRP1 selection marker. They control the expression of the downstream Enhanced Green Fluorescent Protein (EGFP) using the PFBP1 and PMSN2 promoters, respectively, which are presumed to be responsive to changes in cAMP signaling.
At this point, a detailed and theoretically highly viable design for a single-input sensor was complete. However, the robustness of a theoretically sound design under real-world physiological fluctuations remained a critical unknown that would determine its success or failure and had to be examined through more rigorous stress testing.
Our experimental strategy was to co-transform the sensor plasmid pESC-URA-PTEF1-TAAR5 with either the pGBKT7-PFBP1-EGFP or the pGBKT7-PMSN2-EGFP reporter plasmid, thereby creating two engineered strains for parallel testing. In these strains, we hypothesized that extracellular TMA would bind to the TAAR5 receptor, activating its coupled G-protein, which in turn would modulate intracellular concentrations of the second messenger cAMP and ultimately activate the corresponding PFBP1 or PMSN2 promoter to drive EGFP expression.
After the model was constructed, our further research led to a key conclusion. Gut bacteria produce TMA from various substrates, including TMAO, choline, phosphatidylcholine, L-carnitine, and betaine, making the presence of these precursors a prerequisite for TMA production. Differences in the intake of TMA precursors and long-term dietary patterns—which in turn establish differentiated gut microbiota—can cause significant fluctuations in intestinal TMA levels among individuals. This posed new requirements for the design of our signal-responsive element (Malinowska et al., 2017).
The value of this cycle lies in the fact that it subjected our top-level design to a successful "stress test" before we committed resources to wet-lab validation, revealing a critical flaw that required correction.
We clearly demonstrated that although the baseline signal strength of TMA was sufficient (as established in the "Design" phase), the physiological "noise" (concentration fluctuations) caused by daily diet was equally substantial, leading to a significant overlap between the signal and noise ranges. This finding meant that our initially designed sensor system (plasmids pESC-URA-PTEF1-TAAR5 and pGBKT7-PFBP1-EGFP/pGBKT7-PMSN2-EGFP), while theoretically elegant, would be unable to effectively distinguish between a true pathological signal and normal dietary fluctuations in a practical application. Its signal-to-noise ratio was too low, disqualifying it as a reliable therapeutic.
Therefore, the final "Learn" outcome of this cycle was the falsification of the rationale for a single-input sensing, top-level design. To address the core challenge of the low signal-to-noise ratio, it became imperative to introduce more complex regulatory logic. This directly guided the design of our next cycle: to incorporate a second input signal more strongly correlated with the AD pathological state, construct an AND-gate logic, and thereby substantially improve the precision and reliability of the therapy through dual-signal verification.
Based on the AND-gate strategy conceptualized in the first cycle, the core task of this round was the parallel experimental validation of the sensing modules for the two input signals: reactive oxygen species (ROS) and trimethylamine (TMA).
1. Design of the ROS sensing module: When selecting the second input signal for the AND-gate system, our literature review confirmed that in the early stages of AD, drastic changes in the gut microbiota are often accompanied by impaired intestinal barrier function and elevated oxidative stress in epithelial cells. This leads to a significant dysregulation of reactive oxygen species (ROS) concentrations in the gut environment, making ROS an ideal marker for the pathological state (Das & Ganesh, 2023). Accordingly, we selected the ROS-responsive yeast promoter PTRR1 and designed a single-plasmid reporter system to validate its function.
Figure 7. Structure of plasmid pGBKT7-PTRR1-EGFP. This system utilizes the PTRR1 promoter to drive the expression of the downstream Enhanced Green Fluorescent Protein (EGFP). (Generated with SnapGene®.)
We conducted an in-depth literature review of the molecular mechanism by which PTRR1 responds to ROS. Research has shown that this is an intricate signal transduction cascade accomplished through the collaboration of multiple proteins. Early studies first discovered that the transcription factor Yap1, in concert with another transcriptional regulator Skn7, acts as the core effector for activating TRR1 gene expression (Lee et al., 1999). However, subsequent research further revealed that Yap1 itself cannot be directly oxidized by hydrogen peroxide (H₂O₂). The true signal "sensor" is the endogenous yeast peroxidase Gpx3. When intracellular H₂O₂ levels rise, the active site cysteine (Cys36) of Gpx3 is oxidized to form a sulfenic acid (Cys36-SOH). This sulfenic acid intermediate then reacts with Cys598 of Yap1 to form a transient inter-protein disulfide bond (Gpx3-S-S-Yap1). Finally, this inter-protein bond is resolved by a nucleophilic attack from Yap1's own Cys303, ultimately forming an activating intra-protein disulfide bond (Cys303-S-S-Cys598) within the Yap1 protein. The activated Yap1 forms a complex with Skn7, and together they bind to the promoter of the TRR1 gene (i.e., PTRR1), thereby efficiently initiating the transcription of downstream genes (Delaunay et al., 2002). Our design aims to leverage this endogenous, efficient, and well-characterized signaling pathway, which consists of "Gpx3 sensing and Yap1/Skn7 execution."
2. Design of the TMA sensing module: To validate the TMA signal, we used the dual-plasmid sensing scheme designed in the first cycle. The core of this system is a dual-plasmid sensing scheme. The sensor plasmid was constructed on a high-copy pESC-URA backbone with a URA3 selection marker, on which the strong constitutive promoter PTEF1 drives the continuous, high-level expression of the human TMA receptor, TAAR5.
We commissioned a commercial company to synthesize all designed plasmids and subsequently employed a chemical transformation method to introduce these plasmids into Saccharomyces cerevisiae INVSc1 cells.
1. Successful validation of the ROS module: During the functional testing phase, the successfully transformed ROS-sensing strain was cultured and then treated with various concentrations of H₂O₂ as the ROS-inducing agent. Flow cytometry analysis clearly showed that, compared to the control group, the H₂O₂-treated group exhibited a significant and dose-dependent increase in EGFP fluorescence signal, a result that was highly consistent across two detection platforms.
Figure 8. Validation of the ROS-responsive module by fluorescence microscopy. Yeast cells containing the reporter plasmid pGBKT7-PTRR1-EGFP were imaged after a 6-hour incubation with or without 5 mM H₂O₂. A significant increase in EGFP fluorescence was observed in H₂O₂-treated cells compared to the untreated control. (Scale bar: 20 µm)
Figure 9. Dose-dependent activation of the ROS-responsive module measured by flow cytometry. The percentage of EGFP-positive cells carrying the pGBKT7-PTRR1-EGFP plasmid was quantified after 1 hour (left) and 2 hours (right) of incubation with increasing concentrations of H₂O₂. The data show a clear dose-dependent increase in the EGFP-positive population upon H₂O₂ stimulation.
2. Failure to construct the TMA module: In stark contrast to the smooth progress with the ROS module, we encountered an unexpected obstacle during the construction of the TMA module. We attempted multiple strategies, including single-plasmid transformations (with pGBKT7-PFBP1-EGFP, pGBKT7-PMSN2-EGFP, or pESC-URA-PTEF1-TAAR5) and dual-plasmid co-transformation or sequential transformation (pGBKT7-PFBP1-EGFP + pESC-URA-PTEF1-TAAR5, and pGBKT7-PMSN2-EGFP + pESC-URA-PTEF1-TAAR5). The results showed that while yeast strains transformed with the reporter plasmids pGBKT7-PFBP1-EGFP or pGBKT7-PMSN2-EGFP alone could be successfully selected and grew normally, all attempts to transform cells with the sensor plasmid pESC-URA-PTEF1-TAAR5 failed. Whether transformed alone or into a strain already containing a reporter plasmid, we were unable to select any positive transformants, or the transformation efficiency was effectively zero, even after multiple optimizations of the experimental protocol.
This round of parallel testing yielded insights on two levels: the functionality of the ROS sensing module was successfully validated, while the failure to construct the TMA sensing module led us to unexpectedly discover and pinpoint a core technical bottleneck.
The success of the ROS module demonstrated that the PTRR1 promoter is an effective and reliable ROS-responsive element, providing the first experimentally validated functional component for our AND-gate system.
However, the failure to construct the TMA module compelled us to pause for in-depth analysis. We hypothesized that this toxicity stemmed from the overexpression of the human G protein-coupled receptor (GPCR), TAAR5, in the yeast cells. In our initial design, the combination of the strong constitutive promoter PTEF1 and the high-copy-number plasmid backbone pushed the expression dosage of TAAR5 to a level that was intolerable to the yeast cells, leading to transformation failure.
This unexpected failure proved to be more valuable than an anticipated success. It led us to the critical realization that the straightforward strategy of expressing a heterologous GPCR at high dosages is not viable. This key insight defined the precise direction for our next round of engineering: we must resolve the expression toxicity of TAAR5 by replacing the promoter and rewiring the signaling pathway, in order to build a TMA sensing module that is both functionally reliable and host-compatible.
In DBTL Cycle 2, while we successfully validated the ROS-responsive module, we more importantly discovered that the TMA sensing module exhibited fatal expression toxicity. Therefore, the core objective of this cycle was highly focused: to fundamentally resolve the expression toxicity of the TAAR5 receptor and to construct a TMA sensing module that is both functionally reliable and host-compatible.
Based on an in-depth analysis of the cause of failure in Cycle 2 (a strong promoter combined with a high-copy plasmid), we designed a novel, single-plasmid system that integrates TMA sensing, signal transduction, and fluorescent reporting into multiple expression cassettes.
Figure 10. Structure of plasmid pRS416-TAAR5-GPA1-EGFP. (Generated with SnapGene®.)
This design incorporated four key strategic modifications:
Through this series of interlinked modifications, we expected the new system to not only be free of cytotoxicity but also to channel the TMA signal through the rewired and highly efficient "TAAR5-chimeric Gpa1-MAPK pheromone pathway," ultimately generating a precisely detectable fluorescent signal from the PFUS1 promoter.
Figure 11. Schematic of the native pheromone-induced GPCR signaling pathway in yeast. The binding of a pheromone to its G protein-coupled receptor (GPCR) on the cell membrane activates a downstream MAP kinase (MAPK) signaling cascade. This signal ultimately activates the transcription factor Ste12 to induce the expression of pheromone-responsive genes, triggering the yeast mating response.(Reproduced from Lengger, B. & Jensen, M.K. FEMS Yeast Res 20, foz087 (2020).)
We commissioned a commercial company to synthesize the newly designed plasmid, pRS416-TAAR5-GPA1-EGFP. Although the initial transformation efficiency was low, likely due to the large size of the plasmid, we ultimately succeeded in obtaining positive transformants by optimizing the experimental conditions.
In the subsequent functional tests, we induced the engineered strain by adding various concentrations of TMA to the culture medium and monitored EGFP expression using both confocal fluorescence microscopy and flow cytometry. The results clearly demonstrated that, compared to the untreated control group, the TMA-induced engineered strain produced a significant fluorescent signal, which successfully validated the functionality of our rewired MAPK signaling pathway.
Figure 12. TMA-induced EGFP expression in the rewired yeast sensor. Cells containing the sensor plasmid pRS416-TAAR5-GPA1-EGFP were imaged by fluorescence microscopy after a 6-hour incubation with or without 20 mM TMA. A significant induction of EGFP fluorescence was observed in the TMA-treated group compared to the untreated control. BF, Bright Field. Scale bar: 20 µm.
Figure 13. Dose-response analysis of the TMA sensor by flow cytometry. Yeast cells carrying the redesigned sensor plasmid (pRS416-TAAR5-GPA1-EGFP) were incubated for 6 hours with the indicated concentrations of TMA (0–50 mM). The histograms illustrate the green fluorescence intensity distribution for each population. The percentage of EGFP-positive cells is indicated for each condition, and the data demonstrate a significant sensor response even at low TMA concentrations. WT denotes the wild-type INVSc1 strain without the plasmid, which served as a negative control.
The success of this experimental round signifies that, through a series of engineering modifications, we have completely resolved the expression toxicity issue of the TAAR5 receptor and successfully constructed a functionally reliable TMA sensing module.
Notably, the positive results from this experiment (Figures 12 and 13) also fully resolved the core conflict we identified in the first cycle: although our initial in silico model predicted that physiological TMA concentrations would be insufficient to activate the sensor, the current experimental results, consistent with literature reports, clearly demonstrate that our engineered yeast can respond effectively to physiologically relevant concentrations of TMA.
With the successful resolution of the technical bottleneck in the TMA sensing module, combined with the previously validated ROS-responsive module, we have now assembled all the core sensing components required to build the final intelligent-response system; all core technical obstacles have been cleared. This milestone achievement allowed us to shift our focus from "functional implementation" to "long-term safety." Based on a forward-looking consideration of the long-term safety of live biotherapeutics, we proposed that a negative feedback loop should be introduced into the system as a safety redundancy. To achieve this goal, a separate DBTL cycle (Cycle 4) will be initiated, dedicated to the design and validation of this safety module.
Following three progressive DBTL cycles, we have successfully designed and validated all the necessary core sensing components. Building on this foundation, we integrated these validated modules with the adapted butyrate output pathway to establish the final design blueprint for a complete "AND-gate-butyrate" intelligent-response system. This blueprint employs a cascade regulation mechanism, intricately linking signal perception with product synthesis through a multi-plasmid system:
1.Sensing and Transduction Module:
Working Principle: This module functions as the first-level control switch for the AND-gate logic, relying entirely on the sophisticated utilization of the endogenous yeast ROS-response pathway. When ROS (e.g., H₂O₂), a pathological marker, is present, the endogenous yeast peroxidase Gpx3 acts as a specific sensor. Its active-site cysteine residue (Cys36) is oxidized by H₂O₂ to form a highly reactive sulfenic acid intermediate (Cys36-SOH). This intermediate then reacts with the C-terminal cysteine residue (Cys598) of the transcription factor Yap1, forming a transient, covalent inter-protein disulfide bond between Gpx3 and Yap1. Subsequently, another cysteine residue at the N-terminus of the Yap1 protein (Cys303) performs a nucleophilic attack on this inter-protein disulfide bond. This resolves it, via a thiol-disulfide exchange reaction, into a stable intra-protein disulfide bond (Cys303-S-S-Cys598) within Yap1. The formation of this intra-protein disulfide bond is the molecular signature of full Yap1 activation. Activated Yap1 then forms a complex with the transcriptional regulator Skn7, and together they bind to our selected PTRR1 promoter to specifically initiate the expression of the downstream TMA receptor, TAAR5. Therefore, the cell will only express the TMA sensor in the presence of an ROS signal. As the chimeric G-protein Gpa1 is designed for constitutive expression driven by the PTEF1 promoter, the cell possesses the complete machinery necessary to activate the downstream MAPK pathway once the TAAR5 receptor is expressed.
Figure 14. Design of plasmid pRS416-PTRR1-TAAR5-GPA1. (Generated with SnapGene®.)
2. Output Module Plasmid Design (Butyrate Synthesis Pathway):
The presence of homologous DNA sequences on a plasmid, such as repeated promoters, is a key factor that can induce endogenous homologous recombination in the host cell. This recombination process can lead to the deletion of the genetic segment between the two homologous sequences, thereby compromising the structural integrity of the plasmid. Therefore, selecting different constitutive promoters for each expression cassette is a critical design strategy to eliminate these recombination hotspots, ensuring the long-term stability and functional integrity of the multi-gene expression system.
Working Principle and Synthesis Pathway: This heterologous pathway synthesizes butyrate from the endogenous yeast precursor, acetyl-CoA, through a five-step enzymatic reaction. The key enzymes are expressed from the designed plasmids and are under the strict control of the signaling pathway.
Step 1 (Thiolysis): Two molecules of acetyl-CoA are condensed into acetoacetyl-CoA. This step is catalyzed by the endogenous yeast protein Erg10 and requires no exogenous gene.
Step 2 (Reduction): Acetoacetyl-CoA is reduced to 3-hydroxybutyryl-CoA. This step is catalyzed by 3-hydroxybutyryl-CoA dehydrogenase (Hbd). The hbd gene is located on plasmid pGBKT7-ter-PFUS1-hbd, and its expression is strictly controlled by the PFUS1 promoter, serving as the final output checkpoint for the entire AND-gate logic.
Step 3 (Dehydration): 3-hydroxybutyryl-CoA is dehydrated to form crotonyl-CoA. This step is catalyzed by crotonase (Crt). The crt gene is located on plasmid pESC-LEU-bcoat-crt and is driven by the constitutive promoter PADH1.
Step 4 (Reduction): Crotonyl-CoA is reduced to butyryl-CoA. This step is catalyzed by trans-2-enoyl-CoA reductase (Ter). The ter gene is located on plasmid pGBKT7-ter-PFUS1-hbd and is driven by the constitutive promoter PCYC1.
Step 5 (CoA Transfer): Butyryl-CoA is converted to the final product, butyrate. This step is catalyzed by butyryl-CoA:acetate CoA-transferase (BCoAT). The bcoat gene is located on plasmid pESC-LEU-bcoat-crt and is driven by the strong constitutive promoter PTEF1.
Figure 15. Plasmids designed for the butyrate synthesis pathway. (A) Plasmid pESC-LEU-bcoat-crt, containing the expression cassettes [PTEF1 - bcoat - Terminator] and [PADH1 - crt - Terminator]. (B) Plasmid pGBKT7-ter-PFUS1-hbd, containing the expression cassettes [PCYC1 - ter - Terminator] and [PFUS1 - hbd - Terminator]. (Generated with SnapGene®.)
3. Signal Transduction Cascade:
Fig 16. Schematic of the dual-input AND-gate for butyrate production in engineered yeast. The system integrates two signals: Reactive Oxygen Species (ROS) and Trimethylamine (TMA). The presence of ROS primes the system by inducing the expression of the TMA receptor (TAAR5). Subsequent binding of TMA activates the MAPK pathway, leading to the expression of hbd, a key enzyme in the engineered butyrate synthesis pathway. The circuit is only switched on to produce butyrate when both signals are present.
This integrated design blueprint is logically sound and presents manageable risks, marking the substantial completion of our initial development and iterative optimization phase. However, a powerful system must also be a safe one. Based on a forward-looking consideration of the long-term safety of live biotherapeutics, we immediately initiated a fourth DBTL cycle, specifically dedicated to designing and evaluating a necessary safety redundancy module for this system.
After establishing the complete "AND-gate-butyrate" system blueprint in Cycle 3, we initiated this cycle out of a high regard for the long-term safety of live biotherapeutics. The core objective was to ensure the long-term safety of the engineered yeast by designing a negative feedback inhibition module capable of responding to the concentration of the product (butyrate) and automatically shutting down the production pathway.
To this end, we designed a dual-plasmid regulatory circuit based on the TetR repressor system, analogous to an automated factory containing both a "monitoring department" and a "production department."
1.Monitoring and Management Plasmid:
Figure 17. Structure of plasmid pESC-HIS-lrp-PPcha-tetR, containing the expression cassettes [PTEF1 - lrp - Terminator] and [PPcha - tetR - Terminator]. (Generated with SnapGene®.)
Working Principle: This plasmid utilizes the strong promoter PTEF1 for the constitutive expression of the leucine-responsive regulatory protein, Lrp. In bacteria, the Lrp protein has been confirmed to be a molecular sensor that can directly bind butyrate (Nakanishi et al., 2009). To verify its suitability for our design, we first obtained a high-quality predicted structural model of Lrp from the AlphaFold Protein Structure Database. We then conducted molecular docking simulations, and the results predicted a favorable binding interaction between butyrate and the ligand-binding domain of Lrp, providing structural support for its use as the sensor component in our feedback loop. In our design, as the intracellular concentration of butyrate accumulates due to production, butyrate molecules act as ligands, binding to the ligand-binding domain of the Lrp protein. This induces a conformational change in Lrp, activating our selected PPcha promoter (Nakanishi et al., 2009) and thereby initiating the expression of the downstream tetracycline repressor protein, TetR.
2.Production and Execution Plasmid (A modification of the Cycle 3 design):
Figure 18. Structure of plasmid pGBKT7-pTETR-bcoat-crt, containing the expression cassettes [pTETR - bcoat - Terminator] and [PADH1 - crt - Terminator]. (Generated with SnapGene®.)
Working Principle: The TetR protein is a transcriptional silencer derived from prokaryotes that specifically binds to its corresponding DNA operator sequence (tetO). In our design, we replaced the promoter for bcoat—a key enzyme catalyzing the final, rate-limiting step of the butyrate synthesis pathway—with the TetR-regulated pTETR promoter, which contains tetO sequences. When the TetR protein produced by the "monitoring department" is expressed, it binds to the tetO sites on the pTETR promoter. This binding sterically hinders the recruitment of the transcriptional machinery, thereby potently inhibiting the transcription of the bcoat gene (Welman et al., 2007). Once bcoat expression is blocked, the entire butyrate synthesis pathway is interrupted.
Figure 19. Schematic of the two-plasmid negative feedback circuit for autoregulation of butyrate synthesis. (A) The monitoring circuit. Intracellular butyrate acts as a ligand, binding to the Lrp protein and inducing a conformational change that activates the PPcha promoter to drive the expression of the repressor protein, TetR. (B) The production and execution circuit. The expressed TetR protein binds to the tetO sites on the pTETR promoter, potently inhibiting the transcription of bcoat, a key enzyme in the synthesis pathway. Together, these circuits form a closed-loop system that automatically shuts down production in response to product accumulation. Our core hypothesis was that this negative feedback loop could effectively sense the intracellular accumulation of butyrate and halt its further increase by shutting down the final step of butyrate synthesis, thus forming a sophisticated, self-regulating closed-loop system.
The 'Build' phase for this cycle was designed to involve the synthesis of the two key plasmids for the negative feedback circuit—the monitoring plasmid (pESC-HIS-lrp-PPcha-tetR) and the modified production plasmid (pGBKT7-pTETR-bcoat-crt)—followed by their transformation into the yeast chassis. However, this experimental construction phase was not executed.
Consequently, as the construction phase was not executed, the 'Test' phase—which was intended to evaluate the functionality of the negative feedback loop—was also cancelled. This decision was based on a parallel theoretical analysis, the conclusions of which are detailed in the 'Learn' section.
In parallel with the design of this negative feedback module, our team completed pharmacokinetic (PK) modeling for butyrate. The results of this key theoretical analysis directly determined the final outcome of this cycle. The modeling showed that even if the engineered yeast produced butyrate at its maximum theoretical rate, its concentration in the gut environment would remain far below any known hazardous threshold. This meant that the fundamental premise for our complex negative feedback mechanism—designed to address the potential risk of "butyrate overdose"—was invalid. Therefore, this design was rationally cancelled.
**Figure 20. Pharmacokinetic Partitioning Model of Butyrate Flux.**This figure simulates the partitioning of butyrate produced by engineered yeast between two primary fates. The x-axis represents the butyrate production rate of the engineered yeast (mmol/h), with the origin (0) corresponding to the baseline physiological state without the engineered yeast. The y-axis represents the butyrate flux (mmol/h). The red curve (Φ_efflux) denotes the flux absorbed into the portal vein, while the green curve (Φ_metab) represents the flux metabolized by colon epithelial cells. The marked "Design point" indicates the predicted performance of the engineered yeast at a specific, predefined production rate. At this "Design point," the model shows that the majority of the butyrate flux is preferentially metabolized (Φ_metab = 13.5 mmol/h), while the flux entering the portal vein is considerably lower (Φ_efflux = 5.8 mmol/h), indicating that the metabolic activity of the intestinal epithelium significantly limits butyrate's entry into systemic circulation.
More importantly, this cycle, albeit unexecuted, triggered a deeper level of thinking about our safety strategy. We realized that controlling the product concentration might not be the optimal safety strategy. This is because the engineered yeast itself, as a living organism, could pose more uncontrollable risks—such as its colonization, proliferation, or potential for genetic mutations within the gut—than an overproduction of its product.
This conceptual upgrade pointed to a new and more fundamental direction for the fifth cycle's safety design: to abandon product regulation in favor of cell clearance. The core task of the fifth cycle will therefore be to design a user-inducible "kill switch." This system must meet stringent criteria, including high reliability, low leakage, and user controllability (e.g., being triggered by the oral administration of a safe, common small molecule), to provide the user with an ultimate and proactive safeguard.
The theoretical analysis from Cycle 4 upgraded our safety strategy from "product control" to the more fundamental "cell population control." Therefore, the core objective of this cycle is to design a user-inducible, highly efficient, and low-leakage "kill switch" to provide the ultimate safeguard for the entire live biotherapeutic.
We designed a system for programmed cell death (apoptosis) that is inducible by tetracycline (or its derivatives, such as doxycycline). The system consists of a constitutively expressed regulatory module and a tightly regulated effector module. The essence of its design is to achieve the goal of being "completely silent under basal conditions, yet highly lethal upon induction."
Regulatory Module: For reasons of genetic safety and stability, we placed the kill switch system on a recombinant plasmid, pRS403, to maximally ensure it would not be lost or leak. This module uses the strong constitutive promoter PADH1 to continuously express the core regulatory protein rtTA3 (reverse tetracycline-controlled transactivator). We chose rtTA3 because it is an evolved version of rtTA that has been demonstrated to have higher sensitivity to the inducer doxycycline, allowing activation at lower drug concentrations (Markusic et al., 2005). This engineered fusion protein includes the key SV40 nuclear localization signal (NLS). The SV40 NLS is the most thoroughly studied prototype of a classic NLS (cNLS) and is efficiently recognized by the cell's importin alpha/beta proteins, ensuring that rtTA3 is actively transported into the nucleus to perform its transcriptional regulation function after synthesis (Cross et al., 2024). Furthermore, the C-terminus of rtTA3 is fused to the potent VP16 transcriptional activation domain from the herpes simplex virus, which can efficiently recruit the host's transcription complex to ensure strong activation of the downstream gene under induced conditions (Vojnic et al., 2011).
Effector Module: The core of this module is the potent pro-apoptotic human protein, BAX. The expression of the BAX gene is controlled by a tetracycline-responsive promoter (pTETR) that contains multiple tetO operator sequences. The binding of TetR and its derivatives (like rtTA3) to the tetO sequence is one of the most extensively studied and widely used systems in the field of gene expression regulation (Welman et al., 2007). To ensure this switch is tightly controlled, we designed additional insulator elements flanking both sides of the pTETR promoter to actively recruit transcriptional repressors, thereby maximally suppressing any background leaky expression in the non-induced state (Welman et al., 2007).
Figure 21. Gene circuit design for the kill switch plasmid, pRS403-rtTA3-pTETR-BAX. The plasmid contains two expression cassettes: [PADH1 - rtTA3 - tCYC1] and [insulator - pTETR - BAX(human) - tADH1 - insulator]. (Generated with SnapGene®.)
Figure 22. Schematic of the tetracycline-inducible apoptotic system in yeast.(A) In the absence of tetracycline (non-induced state), the rtTA3 protein is inactive and cannot bind the TETR promoter. Flanking insulator elements ensure that BAX gene expression is completely repressed, allowing the cell to remain viable. (B) Upon the addition of tetracycline, the drug activates rtTA3, causing it to bind to the TETR promoter and strongly drive BAX gene expression. The BAX protein targets the mitochondria, forms pores in the outer membrane, causing the release of cytochrome c and ultimately triggering apoptosis.
Working Principle:
Non-induced State (in the absence of tetracycline): In the absence of an inducer, although the rtTA3 protein is located within the nucleus, its conformation prevents it from binding to the tetO sequences on the TETR promoter. Simultaneously, the insulator elements block potential activation of the system by transcriptional activation domains or other cis-acting elements (such as enhancers) from the host genome. As a result, BAX protein expression is completely repressed, allowing the yeast cells to grow and function normally. Conversely, the insulator elements can also serve a protective role for the system, preventing it from being silenced by potential silencing effects in the yeast genome and thereby ensuring that the BAX gene can be properly expressed to exert its killing effect when needed.
Induced State (upon oral administration of tetracycline by the user): When the user takes tetracycline (or doxycycline), the drug molecules enter the engineered yeast and bind to the rtTA3 protein as an allosteric effector. This binding event induces a conformational change in rtTA3, enabling it to bind with high affinity to the tetO sequences on the TETR promoter. Subsequently, its fused VP16 potent transcriptional activation domain efficiently recruits the yeast's transcriptional machinery, powerfully initiating the expression of the BAX protein. The expression of BAX in yeast has been proven to be lethal (Manon et al., 1997). Its mechanism of action is as follows: the expressed BAX protein targets and inserts into the outer mitochondrial membrane, where it oligomerizes to form pores. These pores disrupt the integrity of the mitochondrial membrane, leading to the release of key pro-apoptotic factors, particularly cytochrome c, from the mitochondria into the cytoplasm (Zhang et al., 2017). The release of cytochrome c is the critical initiating signal that triggers the apoptotic cascade, ultimately leading to cell death and thus efficiently clearing the engineered yeast from the body.
We plan to commission a commercial company to synthesize the aforementioned gene expression cassette on a yeast recombinant plasmid. Subsequently, we aim to stably integrate it into the genome of the INVSc1 yeast via homologous recombination to ensure the stable inheritance of this safety function.
A killing efficiency test will be performed. This will involve adding various concentrations of tetracycline (or doxycycline) to the culture medium, followed by calculating the number of surviving colonies (CFU) via plate spreading to precisely quantify the kill switch's killing efficiency and response sensitivity.
The "Learn" outcome of this cycle is the completion of the full theoretical design for a logically rigorous and functionally powerful final safety module. This design not only directly addresses the higher-level safety requirement raised in Cycle 4—to control the living cells themselves—but its dual consideration for both "low leakage" and "high efficiency" also makes it a reliable engineered solution.
At this point, all the necessary core functional modules for this project—the AND-gate sensing module (TMA+ROS), the butyrate output module, and the user-controllable kill switch module—have now been either fully designed or preliminarily validated. This marks the substantial completion of the project's initial research and iterative optimization phase, laying a solid foundation for the next step: the final integration of all modules and their migration into the therapeutic probiotic chassis, Saccharomyces boulardii.
With the design and validation of the core functional modules nearing completion, the objective of this cycle shifts to the project's final application and implementation. Specifically, the goal is to plan the strategy and pathway for transferring the gene circuits, developed and validated in the laboratory model yeast (S. cerevisiae INVSc1), into the final therapeutic chassis with probiotic functions (Saccharomyces boulardii). To this end, we have established a rigorous two-step "Prototype-then-Transfer" research and development strategy.
The first phase is rapid prototyping in INVSc1. We selected S. cerevisiae INVSc1 as the initial development platform based on its significant advantages as a model organism, aiming to maximize the speed of DBTL cycle iterations and minimize early-stage research and development risks. These advantages include its excellent genetic tractability, well-defined genetic background, rapid growth cycle, and a vast toolbox of standardized biological parts. This strategy successfully decouples the complex scientific problems of "gene circuit design and validation" and "final chassis adaptation," enabling us to efficiently complete the initial innovative work.
The second phase is the functional transfer into Saccharomyces boulardii. As the only yeast approved by the FDA for use as a probiotic, S. boulardii is the ultimate vehicle for achieving this project's therapeutic goals. Its key features include clinically proven safety, high tolerance to the harsh gut environment, and inherent probiotic functions such as gut microbiota modulation and anti-inflammatory effects, which can create a therapeutic synergy with the butyrate produced by our engineered yeast. Our core hypothesis is that the gene circuits successfully validated in INVSc1 can, after appropriate modification and optimization, replicate their core functions in S. boulardii.
We will need to perform an analysis of genetic tool compatibility, which may require changing plasmid backbones; codon-optimize all heterologous genes to match the expression preferences of S. boulardii; and optimize the transformation protocol for S. boulardii (e.g., using electrotransformation) to overcome its low transformation efficiency. To ensure the long-term genetic stability required for a live therapeutic, our long-term goal is to stably integrate all functional modules into the S. boulardii genome. Considering the difficulty of multi-plasmid co-transformation or transforming a single large plasmid, we plan to adopt a stepwise, sequential integration strategy. This involves integrating and validating each functional module (e.g., sensing, output, and safety modules) one by one to ensure the success of each step, ultimately leading to the construction of the final, complete engineered strain.
After a successful transfer, a comprehensive functional re-validation is mandatory. This includes repeating the dual-response tests for ROS and TMA signals in S. boulardii and quantifying butyrate production in a simulated gut environment. Furthermore, we must systematically evaluate the "chassis effect"—potential performance differences (such as sensitivity, response speed, maximum yield, etc.) resulting from the change in the host organism—and perform targeted optimizations accordingly.
The "Learn" outcome of this planning cycle is twofold. First, we have established a clear R&D pathway from "laboratory prototype" to "preclinical product," a critical step toward the project's practical application. Second, we have foreseen potential technical bottlenecks during the chassis migration process, such as changes in gene expression levels, metabolic flux redistribution, and tool compatibility issues, which has allowed us to develop a clear risk mitigation plan for future development.
This planning cycle directly guides our subsequent work. We recommend that the development of the S. boulardii genetic manipulation platform be carried out in parallel with the completion of the "kill switch" module from Cycle 5. Concurrently, to mitigate risks, a stepwise transfer strategy should be adopted: first, transfer a simple reporter system (such as plasmid pGBKT7-PTRR1-EGFP) into S. boulardii for initial validation. Only after its success should the more complex AND-gate and butyrate synthesis modules be transferred.
After completing the design, construction, and experimental validation of the core functional modules through the DBTL cycles, our focus shifted to a more ambitious goal: could we, at the in silico level, integrate the intracellular gene circuit's response, the cell's overall metabolic production capacity, and the complex in vivo delivery process of the therapeutic molecule to prospectively evaluate the ultimate efficacy of our entire live biotherapeutic design?
To answer this question, we undertook a series of system-level, multi-scale modeling efforts. This work served not only as a theoretical validation and extension of our previous wet-lab experiments but also as a critical step in elevating our project from a "functionally validated biological device" to an "engineered therapy with predictable efficacy."
1. Kinetic Modeling of the Gene Circuit and Molecular Docking
To deeply understand the dynamic response behavior of our designed AND-gate gene circuit, we constructed an ordinary differential equation (ODE) model describing the entire signaling pathway. As described in the first cycle, this model, in conjunction with parallel molecular docking simulations, confirmed the ideal theoretical feasibility of our sensor system: the TAAR5 receptor can effectively bind TMA, and the entire AND-gate logic possesses highly specific switching characteristics, enabling it to precisely recognize the combination of pathological signals. This provides a solid theoretical basis for the system's efficient in vivo operation.
2. Metabolic Network Modeling: Predicting the Theoretical Production Potential of Butyrate
A precision "switch" must be connected to an efficient "factory." To evaluate the production potential of the engineered yeast as a "cell factory," we employed a genome-scale metabolic model (GEM) for predictive simulations. We incorporated the heterologous butyrate synthesis pathway into the yeast's metabolic network and reconfigured the network based on culture conditions that simulate the gut microenvironment, as defined by published literature data.
By solving and simulating the metabolic network, we systematically calculated the theoretical flux range for butyrate production by the engineered yeast under various oxygen concentrations and growth rate limitations. This series of simulations not only revealed the metabolic bottlenecks and potential optimization avenues for the butyrate synthesis pathway but also provided crucial input parameters—specifically, the predicted theoretical yield at given growth rates—for the subsequent integrated pharmacokinetic model used to evaluate final therapeutic efficacy
3. Cross-Barrier Delivery Prediction: A Machine Learning Study of Butyrate's Blood-Brain Barrier Permeability
Before conducting the final pharmacokinetic modeling, we had to resolve a critical unknown parameter: the blood-brain barrier (BBB) permeability coefficient (logBB) of butyrate. To tackle this challenge, we first employed a machine learning approach. Using a large dataset of water-soluble small molecules, we specifically built and validated a machine learning model capable of predicting small molecule logBB values through feature engineering and model training. This model ultimately provided us with a reliable predictive range for butyrate's BBB permeability, completing the final piece of data required for our integrated modeling.
4. Integrated Pharmacokinetic Modeling: Prospective Evaluation of the Therapeutic Endpoint
After obtaining all critical input parameters—including the switching characteristics of the AND-gate circuit, the steady-state butyrate yield predicted by the GEM, and the logBB value predicted by the machine learning model—we finally established an integrated, steady-state pharmacokinetic model.
This model acts as a bridge, systematically connecting the entire process from the "cellular factory in the gut" to the "therapeutic target in the brain." Based on literature and physical laws, it describes the complete journey of butyrate: from its release by the yeast, to its absorption by the intestinal epithelium, its entry into blood circulation, and finally, its penetration across the blood-brain barrier. Through this multi-scale, multi-model integrated analysis, we ultimately generated a theoretically-grounded, semi-quantitative prediction for the steady-state therapeutic concentration of butyrate achievable in the cerebrospinal fluid. This prospective in silico validation connects all of our micro-level design efforts to an assessable, clinically relevant, macro-level therapeutic endpoint, providing strong theoretical evidence to support the ultimate feasibility of our therapeutic design.
Building upon the forefront of Alzheimer's disease (AD) therapeutics—targeting the neuroinflammation driven by the APOEε4 allele—this project has not only established a functionally comprehensive and logically rigorous design blueprint for an engineered yeast but, more importantly, has contributed a systemic solution and a new design paradigm for the field of Engineered Living Therapeutics (ELTs).
Our work moves beyond the limitations of traditional single-target, passive-administration approaches. Through a series of rigorous DBTL cycles, we have developed a novel design of a highly specific, dual-input AND-gate sensing module, offering a novel engineering strategy to solve the "signal-to-noise ratio" challenge for biosensors in complex in vivo environments. Furthermore, we have proactively integrated multiple safety mechanisms, including a user-controllable "kill switch," into the top-level design and have planned a meticulous migration path from the "prototyping chassis" to the "application chassis." This directly confronts and addresses the core challenges of safety and practicality on the path to clinical translation for live biotherapeutics. In conclusion, this project is not merely an innovative therapeutic attempt for AD but also a successful application of systems engineering thinking in the field of complex disease treatment, which could support the future development of smarter and safer living medicines.
Building on the current solid theoretical and design foundation, we have charted a clear, pragmatic, and promising translational pathway to advance this engineered yeast from a design blueprint to a preclinical candidate.
First, we will complete the experimental validation and systems integration of all core functional modules. This includes the construction and testing of the user-controllable "kill switch" and a gut colonization module, ensuring that the final engineered strain not only possesses an ultimate safety assurance but can also function stably as a long-term "drug factory" within the complex gut environment.
Subsequently, we will initiate the critical migration to the clinical application chassis. We will employ a rigorous, stepwise integration and validation strategy to transfer all optimized modules into the probiotic yeast S. boulardii. During this process, we will systematically address the technical challenges potentially posed by the "chassis effect" to ensure that the function of the entire complex genetic network can be fully replicated in the new host.
Finally, upon completing the construction and in vitro functional validation of the engineered S. boulardii, we will advance the project to the preclinical research stage. We firmly believe that this gut-brain axis-based live biotherapeutic, capable of intelligently sensing and responding to complex pathological signals, not only holds the promise of a revolutionary breakthrough for AD treatment but that its "sense-compute-actuate" design philosophy and "multiple safety redundancies" engineering concept are poised to become a scalable platform technology. In the future, this platform could be applied to treat a range of major diseases associated with gut-brain axis dysregulation, such as Parkinson's disease and depression, heralding a new era of precise and controllable live biotherapeutic medicine.
Details
Blumenfeld, J., Yip, O., Kim, M. J., & Huang, Y. (2024). Cell type-specific roles of APOE4 in Alzheimer disease. Nature Reviews Neuroscience, 25, 91–110.
Chen, K., Zhu, Y., Zhang, Y., Hamza, T., Yu, H., Saint Fleur, A., Galen, J., Yang, Z., & Feng, H. (2020). A probiotic yeast-based immunotherapy against Clostridioides difficile infection. Science Translational Medicine, 12(543), Article eaax4905.
Cross, E. M., Akbari, N., Ghassabian, H., Hoad, M., Pavan, S., Ariawan, D., Donnelly, C. M., Lavezzo, E., Petersen, G. F., Forwood, J. K., & Alvisi, G. (2024). A functional and structural comparative analysis of large tumor antigens reveals evolution of different importin α-dependent nuclear localization signals. Protein Science, 33(2), Article e4876.
Das, T. K., & Ganesh, B. P. (2023). Interlink between the gut microbiota and inflammation in the context of oxidative stress in Alzheimer’s disease progression. Gut Microbes, 15(1), Article 2206504.
Delaunay, A., Pflieger, D., Barrault, M.-B., Vinh, J., & Toledano, M. B. (2002). A thiol peroxidase is an H₂O₂ receptor and redox-transducer in gene activation. Cell, 111(4), 471–481.
Gisladottir, R. S., Ivarsdottir, E. V., Helgason, A., Jonsson, L., Hannesdottir, N. K., Rutsdottir, G., Arnadottir, G. A., Skuladottir, A., Jonsson, B. A., Norddahl, G. L., Ulfarsson, M. O., Helgason, H., Halldorsson, B. V., Nawaz, M. S., Tragante, V., Sveinbjornsson, G., Thorgeirsson, T., Oddsson, A., Kristjansson, R. P., ... Stefansson, K. (2020). Sequence variants in TAAR5 and other loci affect human odor perception and naming. Current Biology, 30(23), 4643–4653.e3.
Hering, H., Bussiere, T., Liu, C. C., Glajch, K. E., Weihofen, A., Grogan, J., & Walsh, D. M. (2025). A manifesto for Alzheimer’s disease drug discovery in the era of disease-modifying therapies. Molecular Neurodegeneration, 20, Article 88.
Kilgore, M., Miller, C. A., Fass, D. M., Hennig, K. M., Haggarty, S. J., Sweatt, J. D., & Rumbaugh, G. (2010). Inhibitors of class 1 histone deacetylases reverse contextual memory deficits in a mouse model of Alzheimer's disease. Neuropsychopharmacology, 35, 870–880.
Lee, J., Godon, C., Lagniel, G., Spector, D., Garin, J., Labarre, J., & Toledano, M. B. (1999). Yap1 and Skn7 control two specialized oxidative stress response regulons in yeast. Journal of Biological Chemistry, 274(23), 16040–16046.
Lengger, B., & Jensen, M. K. (2020). Engineering G protein-coupled receptor signalling in yeast for biotechnological and medical purposes. FEMS Yeast Research, 20(1), Article foz087.
Loh, J. S., Mak, W. Q., Tan, L. K. S., Ng, C. X., Chan, H. H., Yeow, S. H., Foo, J. B., Ong, Y. S., How, C. W., & Khaw, K. Y. (2024). Microbiota–gut–brain axis and its therapeutic applications in neurodegenerative diseases. Signal Transduction and Targeted Therapy, 9, Article 37.
Malinowska, A. M., Szwengiel, A., & Chmurzynska, A. (2017). Dietary, anthropometric, and biochemical factors influencing plasma choline, carnitine, trimethylamine, and trimethylamine-N-oxide concentrations. International Journal of Food Sciences and Nutrition, 68(4), 488–495.
Manon, S., Chaudhuri, B., & Guérin, M. (1997). Release of cytochrome c and decrease of cytochrome c oxidase in Bax-expressing yeast cells, and prevention of these effects by coexpression of Bcl-xL. FEBS Letters, 415(3), 273–276.
Markusic, D., Oude-Elferink, R., Das, A. T., Berkhout, B., & Seppen, J. (2005). Comparison of single regulated lentiviral vectors with rtTA expression driven by an autoregulatory loop or a constitutive promoter. Nucleic Acids Research, 33(6), Article e63.
Nakanishi, N., Tashiro, K., Kuhara, S., Hayashi, T., Sugimoto, N., & Tobe, T. (2009). Regulation of virulence by butyrate sensing in enterohaemorrhagic Escherichia coli. Microbiology, 155(2), 521–530.
Vojnic, E., Mourão, A., Seizl, M., Simon, B., Wenzeck, L., Larivière, L., Baumli, S., Baumgart, K., Meisterernst, M., Sattler, M., & Cramer, P. (2011). Structure and VP16 binding of the Mediator Med25 activator interaction domain. Nature Structural & Molecular Biology, 18, 404–409.
Welman, A., Barraclough, J., & Dive, C. (2007). Tetracycline regulated systems in functional oncogenomics. Translational Oncogenomics, 2, 17–33.
Wu, J., Huang, H., Wang, L., Gao, M., Meng, S., Zou, S., Feng, Y., Feng, Z., Zhu, Z., Cao, X., Li, B., & Kang, G. (2024). A tailored series of engineered yeasts for the cell-dependent treatment of inflammatory bowel disease by rational butyrate supplementation. Gut Microbes, 16(1), Article 2316575.
Zhang, M., Zheng, J., Nussinov, R., & Ma, B. (2017). Release of cytochrome c from Bax pores at the mitochondrial membrane. Scientific Reports, 7, Article 2635.
Zhou, J., Wang, Y., Huang, G., Yang, M., Zhu, Y., Jin, C., Jing, D., Ji, K., & Shi, Y. (2023). LilrB3 is a putative cell surface receptor of APOE4. Cell Research, 33, 116–130.
