Overview

In this study, we characterized the extent to which xylitol metabolism is present in the human gut microbiota and multilaterally evaluated the ecological significance of the xylitol metabolic pathway we planned to introduce.

First, through metagenomic analysis, we estimated the distribution and relative abundance of genes involved in xylitol metabolism within the gut microbiome. Then, using the FriendlyNets model, we theoretically predicted the engraftment rate of E. coli strains engineered to possess xylitol metabolism.

Through this two-step analysis, we investigated both in silico (dry experiments) and in vitro (wet experiments) aspects of two key questions: “To what extent do gut bacteria perform xylitol metabolism?” and “How do artificially introduced strains behave within such an ecosystem?”

Background

Xylitol Metabolic Pathway and Analytical Objectives

The xylitol metabolic pathway consists of the following three steps:

1. Xylitol transporter (K17205, K17206, K17207): Transports xylitol into the cell
2. Xylitol dehydrogenase (K05351 or K00008): Oxidizes xylitol into xylulose
3. Xylulokinase (K00854): Phosphorylates xylulose and introduces it into the pentose phosphate pathway (PPP)

This pathway connects to the pentose phosphate pathway (PPP), linking xylitol metabolism to energy production and cofactor regeneration. Initially, we hypothesized that a dedicated xylitol transporter was essential for xylitol metabolism, and we searched for gut bacteria possessing all three key KOs. The result revealed that such bacteria were extremely rare, suggesting that xylitol metabolism scarcely occurs in the intestinal environment.

However, wet-lab experiments yielded results that contradicted our initial hypothesis: xylitol metabolism proceeded even in the absence of a specific xylitol transporter. This finding implies that a dedicated transporter may not be strictly required for xylitol uptake. Therefore, I considered and tested the following possible mechanisms that could explain this phenomenon.

• The E. coli strain used is Gram-negative, and its double-membrane structure may have influenced the efficiency of xylitol transport.[1]

Taking these factors into account, we reanalyzed the data under conditions excluding the KOs associated with xylitol transporters (K17205, K17206, K17207) in our dry analysis, in order to evaluate the potential for transporter-independent metabolic activity.

Method / Workflow

1. Definition of the Metabolic Pathway (Design Objective)

   • Target: Introduction of the conversion pathway from xylitol to D-xylulose-5-phosphate.
   • Components:
     • Transporter: K17205, K17206, K17207 (all three required; AND condition)
     • Dehydrogenase: K05351 OR K00008
     • Kinase: K00854 (required; AND condition)
   • Minimal introduction pathway (logical expression):
   $$\textbf{Xylitol Transporter: } K17205 \land K17206 \land K17207,\\\quad \textbf{Xylitol Dehydrogenase: } K05351 \lor K00008,\\\quad \textbf{Xylitol Kinase: } K00854$$ $$\textbf{Carrier} = (\text{Transporter}) \land (\text{Dehydrogenase}) \land (\text{Kinase})$$

2 KO Identification (KEGG)

• For each reaction in the pathway, the corresponding KO (KEGG Orthology) was identified using KEGG, and fixed as the logical expression described above.

3 Data Acquisition (CuratedMetagenomicData)

• Retrieved relative abundance (%) data from adult fecal samples (restricted to IBD samples when necessary).
• The data matrix has rows = taxonomic classification (s__ species level) and columns = samples.

4 Mapping KO to Species

• Obtained organism codes using keggLink("genes", "ko:Kxxxxx"), and normalized them to scientific names (genus_species).

• Extracted the Bacterial Unique Genome (BUG) matrix, which organizes the gut microbiome composition obtained from metagenomic analysis into a species × sample matrix.

• By comparing the list of s__level species actually observed in the samples with the species annotated as KO gene holders, the candidate species potentially involved in each KO were identified.

• By matching the s__ species in the BUG matrix, the candidate species for each KO were identified.

5 Application of the Logical Expression

• Transporter ∩ Dehydrogenase ∩ Kinase = species responsible for the metabolism.

6 Calculation of the Proportion of Metabolically Active Species

• For each sample, the sum of relative abundances of species responsible for the metabolism divided by the total relative abundance of all s__ species, multiplied by 100, was output as carriers (%).

Interpretation of Results

• Low carriers % across samples → complete pathway carriers are rare → considered suitable as a niche strategy.
• Interpretation: The estimation is based on the potential presence of KOs, not on the direct quantification of gene copy number or expression level (i.e., a conservative estimate).

Result

The relative abundance of bacteria in the gut possessing all three functions — xylitol transporter, xylitol dehydrogenase, and xylulokinase — was extremely low, with little inter-individual variation.

Results

• Total number of target KOs: 6
• Number of species analyzed: n = 261 (IBD patients only)
• Detection rate (relative abundance %):
  • Median: 0.0000%
  • Mean: 0.0077%
  • Maximum: 0.6491%

Table 1

In contrast, when the xylitol transporter was excluded and only the two functions — xylitol dehydrogenase and xylulokinase — were considered, the number of carrier species increased substantially, with a median of approximately 16% and a mean of about 18%. The large inter-individual variation suggests that a certain proportion of gut bacteria possess the latent potential for xylitol metabolism.

Results

• Total number of target KOs: 3
• Number of species analyzed: n = 261 (IBD patients only)
• Detection rate (relative abundance %):
  • Median: 16.44%
  • Mean: 18.88%
  • Maximum: 76.93%
  • Minimum: 0.62%

Table 2

In addition, when we examined bacterial species previously reported to exhibit xylitol metabolic activity the total relative abundance of these bacteria was found to be approximately 4%.[2]

From these results, we conclude that while around 18% of bacteria possess the genetic potential for xylitol metabolism, only about 4% are known from previous studies to actually exhibit xylitol metabolic activity.

Furthermore, when we investigated the Gram staining characteristics of each bacterial species along with the presence of xylitol metabolism–related KOs, we found a mixture of Gram-positive and Gram-negative species. This indicates that differences in cell envelope structure do not show a clear correlation with metabolic capability.

Table 3. Gram staining characteristics and presence of metabolism-related KOs in bacterial species known to possess xylitol metabolic activity.

conclusion

From this, the decrease in metabolic efficiency observed in wet-lab experiments is likely due to factors other than membrane permeability — for example, energetic burden from transporter overexpression or competition with existing broad-specificity transporters.

Overall, these analyses suggest that although bacteria capable of xylitol metabolism are a minor population within the gut microbiota, when genes related to xylitol uptake are not considered, a non-negligible proportion of bacteria possess some level of metabolic capability for xylitol utilization.

References

Overview

The Restriction–Modification (RM) system[1] is known to function as a mechanism that causes post-segregational killing (PSK) in cells that have lost a plasmid.

In this project, we designed a Kill Switch based on the RM system involving the restriction enzyme (REase) BsaI and two methyltransferases (MTases), BsaIM1 and BsaIM2 .

The effectiveness of the Kill Switch depends critically on the balance of enzyme expression levels, but experimentally determining and optimizing this balance is extremely challenging. Therefore, to estimate the range of enzyme expression levels required for proper operation of the Kill Switch, we performed simulations of the RM system.

We mathematically modeled the RM system and investigated how variations in the expression ratio between the restriction enzyme and the restriction–modification enzymes affect both the time required for the Kill Switch to fully activate and the overall behavior of the system.

Purpose

• Simulate how post-segregational killing (PSK) behavior in E. coli changes after the activation of a tetrathionate-responsive promoter is halted, using an RM system composed of one REase and two MTases.

• Evaluate the time-dependent changes in cleavage probability per recognition site and assess the time and efficiency required for the Kill Switch to fully activate, with and without degradation tags.

• Simulate the behavior of the Kill Switch in cases where the promoter exhibits leaky expression.

1. Enzyme degradation

In this project, the RM system is triggered when the concentration of tetrathionate decreases as inflammation subsides, leading to the cessation of BsaIM and BsaI expression regulated by TtrR . As tetrathionate is no longer supplied, the intracellular concentrations of BsaI ($[RE]$) and BsaIM ($[MT]$) begin to decay.

In this model, we first assume that cell division is the main factor responsible for the decay of both enzyme concentrations in the RM system. Because the dilution caused by cell division is a continuous process over time, it can be described using first-order reaction kinetics , yielding the following first-order differential equations .

By solving each of the differential equations,

Here,trepresents the elapsed time, with t = 0 corresponding to the point when expression ceases.

$[RE]_0$ and $[MT]_0$ denote the initial concentrations of each enzyme at the time expression stops.The degradation constant of BsaI is denoted by $\rho$, and that of the MTases (BsaIM1 and BsaIM2) is denoted by $\mu$.

By substituting the dilution condition caused by cell division, $t = \tau$ (doubling time) and $[RE] = \frac{1}{2}[RE]_0$, into equation (2) and solving for $\rho$, equation (3) is obtained.

2. Use of two types of restriction–modification enzymes

To prevent DNA cleavage, methylation of the recognition sites by MTase is required. The recognition sequence of BsaI is asymmetric: BsaIM1 modifies the adenine on the bottom strand, while M.BsaIB modifies the C5 position of cytosine on the top strand.

In this model, we assumed that DNA is protected if either of these two methyltransferases modifies the recognition site. The overall MTase activity was therefore defined as the sum of the initial effective activity constants of the two enzymes, expressed as $mtotal=mA+mB.m_{total} = m_A + m_B.$

3. Mathematical expression of the probability of PSK occurrence

The mathematical model of the RM system is derived by combining the probability that hemimethylated sites remain unmethylated during DNA replication cycles—due to the time-dependent decay of MTase activity—with the probability that these unmethylated sites are cleaved by the REase.

3.1 Probability of single-site cleavage（$P_1$）

To simplify the analytical treatment of the model, we assume that the decay rates of REase and MTase are identical ($\mu = \rho$), and derive the probability of single-site cleavage, $P_1(t, \theta)$.

The time-dependent change in the probability that a recognition site remains neither methylated nor cleaved is given by the product of the reaction rate for methylation or cleavage at a given time and the probability that the recognition site has not reacted up to that time.

$k_{MT}$ and $k_{RE}$ are the rate constants for methylation and cleavage reactions, respectively. By solving for $P_{NMT}$ and $P_{NRE}$, we obtain:

Here, $m = k_{MT}[MT]_*0$ and $r = k_*{RE}[RE]_0$. Therefore, the probability that neither DNA cleavage nor protection has occurred up to a given time is given by the product of these two probabilities:

In this context ,m represents the initial methylation rate, and r represents the initial cleavage rate.

At $t \neq 0$, the reaction rates decrease due to the reduction in enzyme concentrations, and thus the probabilities are expressed as follows:

The probability density of DNA cleavage at time $t+\theta$ is given by the product of the probability that the site has not reacted at time $t+\theta$ and the cleavage rate at that time.

Since $\rho = \mu$,

Therefore, by solving the differential equation, the following cleavage probability is obtained.

Therefore, the probability of cleavage is given by the product of the probability that the restriction–modification enzyme has not methylated the site up to time t and the probability that the site is cleaved before being methylated within the time interval $t+\theta$.

In this model, we assume that the decay is primarily due to dilution, and thus $\mu = \rho = \ln(2)/\tau$.

3.2 Overall probability of PSK occurrence ($P_{PSK}$)

To determine the probability of PSK occurrence across the entire genome from the single-site cleavage probability $P_1$, the total number of REase recognition sites on the genome, $N$ , is used. The number of recognition sites was obtained by searching for and counting sequences that match the BsaI recognition motif within the nucleotide sequence retrieved from GenBank.

PSK is defined as the probability that at least one site in the genome is cleaved by the REase.

3.3 Effects of leaky expression and the effective decay constant model

In this project, the Kill Switch functions through the inactivation of the promoter as the concentration of tetrathionate decreases. However, if this mechanism is imperfect, protein synthesis downstream of the promoter may still occur even under conditions where the promoter is expected to be inactive — a phenomenon known as leaky expression . When leakage occurs, enzyme expression does not completely stop, which may prevent the Kill Switch from functioning properly.

3.4 Approximation of leaky expression in the mathematical model: the effective decay constant model

The effect of leakage is incorporated into the model by assuming that the lack of enzyme concentration decay due to leaky expression is equivalent to a decrease in the apparent (effective) decay rate of the enzyme.The leak level is represented by L and is used in the following equation.

4. Various parameters of the model

Result

1. Differences in Kill Switch behavior depending on the r/m ratio

First, under conditions with no leakage, we plotted the time-dependent change in PSK probability for cases where the r/m ratio was 1.0 and 0.3.

As shown in Figure 1, there is a significant difference in the time required for the Kill Switch to fully activate depending on the r/m ratio.

• r/m ratio = 1.0: The PSK probability reached 100% approximately three hours after expression ceased.
• r/m ratio = 0.3: It took approximately five hours for the PSK probability to reach 100%.

2.Effect of promoter leaky expression on Kill Switch behavior

Next, we analyzed the behavior of the Kill Switch while accounting for the effects of leaky expression. The leak level was varied between 30% and 70%.

As shown in Figure 2, increasing the degree of leakage caused the curves to shift toward longer times. When the r/m ratio was 1.0, the Kill Switch functioned completely at a leak level of 30%, whereas at a leak level of 70%, the PSK probability barely increased and the Kill Switch failed to activate. When the r/m ratio was 0.3, the rise in PSK probability was slower compared to that of the r/m = 1.0 case at the same leak level.

For an r/m ratio of 1.0, the Kill Switch operated when the leak level was around 40%. However, for an r/m ratio of 0.3, even at a leak level of 30%, the Kill Switch did not fully function.

discussion

• A model was constructed and simulated for the case in which enzyme degradation occurs more slowly than usual due to residual enzyme expression caused by leakage.When degradation tags are present, the Kill Switch can still function even when the apparent decay rate of enzyme concentration is 0.6 times that observed in the absence of tetrathionate.Here, “apparent decay” does not refer to a decrease in the intrinsic degradation or dilution rate, but rather to the effect of leaky synthesis raising the lower bound of enzyme concentration, thereby delaying the time required to reach the DNA cleavage probability threshold.This suggests that the Kill Switch could activate even when inflammatory bowel disease (IBD) is only partially resolved and tetrathionate levels have not fully returned to baseline.However, this also raises the concern that premature activation of the Kill Switch might prevent complete recovery from IBD.Therefore, the strength of the degradation tag and the expression levels of the restriction–modification enzymes must be optimized to ensure proper functioning of the Kill Switch.

• For r/m ratios of 1.0 and 0.3, the Kill switch was activated at leak levels of 40% and 20%, respectively.When the leak level required for Kill Switch activation is high, the switch may still activate while inflammatory bowel disease (IBD) is only partially healed.Therefore, based on these results, unexpected activation of the Kill Switch is considered unlikely.Furthermore, adjusting the r/m ratio is important for identifying the leak level threshold that provides the most effective therapeutic outcome.

• In the model that accounts for leakage, the degree of leakage is represented as an increase in the lower bound of enzyme concentration ****— caused by residual enzyme expression due to leakage—and as a delay in reaching the threshold of DNA cleavage probability ****However, experimentally measuring the decay rate of enzyme concentration is difficult.Therefore, it is necessary to establish a connection between the model and measurable physical quantities—for example, by expressing a fluorescent protein in media containing sufficient tetrathionate and in media with low tetrathionate levels, and then correlating the fluorescence intensity ratio with the tetrathionate concentration.

References

1. Kozlova S, Morozova N, Ispolatov Y, Severinov K. Dependence of post-segregational killing mediated by Type II restriction–modification systems on the lifetime of restriction endonuclease effective activity. mBio. 15(8):e01408-24.

Overview

Crohn’s disease (CD) is a chronic inflammatory bowel disease where conventional therapies, such as 5-aminosalicylic acid (5-ASA), are widely used to manage inflammation [1]. However, these systemic drugs present a significant clinical challenge due to their narrow therapeutic window [2]. Achieving effective concentrations at the site of intestinal inflammation often requires high oral doses, which increases systemic drug exposure and the risk of side effects, forcing a difficult trade-off between efficacy and safety [3].

To address this challenge, our project developed an engineered E. coli designed to colonize the gut and locally produce Epidermal Growth Factor (EGF). EGF is a potent protein that promotes epithelial healing, and by delivering it directly to the site of inflammation [4]. Our approach aims to maximize therapeutic benefit while minimizing systemic exposure.

To quantitatively evaluate our strategy against the current standard of care, we constructed two distinct pharmacokinetic/pharmacodynamic (PK/PD) models. The first model was developed for conventional oral 5-ASA, establishing a quantitative baseline for its relationship between systemic exposure and clinical efficacy. The second model was built for our Xylego system, incorporating its biological features, including bacterial colonization dynamics, local EGF production, and a self-regulating Kill Switch.

5-ASA PK/PD Modeling

Methodology

We developed a parsimonious PK/PD model for oral mesalazine (5‑ASA) that captures (i) first‑order absorption from the gut lumen, (ii) systemic disposition of the parent drug and its primary metabolite (N‑Ac‑5‑ASA), and (iii) a delayed clinical response modeled via an indirect‑response equation for the Crohn’s Disease Activity Index (CDAI). The model is implemented as a system of ordinary differential equations (ODEs) and calibrated against published clinical data.

Model variables are the amounts in each compartment: $A_{gut}(t)$, $A_p(t)$ (parent), $A_m(t)$ (metabolite), and $CDAI(t)$. Concentrations used for outputs and PD are derived as $C_{plasma}(t) = A_p(t)/V_d$ and $C_{local}(t) = A_{gut}(t)/V_{lumen},$ where $V_d$ denotes the apparent central volume under oral sustained‑release (SR) conditions. Note: The true intravenous $t_{1/2}$ of mesalazine is short, while oral SR formulations can exhibit an apparent longer $t_{1/2}$ (flip-flop kinetics) as absorption becomes the rate-limiting step; our $k_{e,p}$ reflects this SR context.

The pharmacokinetics are described by

Here, $k_a$ is the first‑order absorption rate constant; $F_p$ and $F_m$ partition the absorbed amount into parent versus first‑pass metabolite; $k_{e,p}$ and $k_{e,m}$ are elimination rate constants. Oral doses of size $D$ are given at times $t_j$ (three times daily at 08:00/14:00/20:00 in the simulations), represented by Dirac impulses $\delta(\cdot).$ Unabsorbed fraction $F_{unabs} = 1 - (F_p + F_m)$ remains in the lumen and is implicitly accounted for in $A_{gut}$, thereby continuing to drive the topical PD effect.

The pharmacodynamics link the local (luminal) 5‑ASA exposure to improvements in disease activity via an indirect‑response model:

This structure captures the clinically observed delay between dosing and symptom improvement: the drug inhibits the “production” of disease activity rather than directly lowering CDAI instantaneously.

Parameterization was guided by authoritative sources (DrugBank DB00244; Japanese package inserts/monographs) and trial data used for PD anchoring[5][6][7] [8] [9]. For the sustained‑release (Pentasa®‑like) behavior we used $V_d = 18\,\mathrm{L}$, $t_{1/2,\,parent} \approx 6.4\,\mathrm{h}$, $k_{e,p} = \ln 2/6.4\,\mathrm{h}^{-1}$, $t_{1/2,\,met} \approx 7.5\,\mathrm{h}$, $k_a = 1.1\,\mathrm{h}^{-1}$, $F_p = 0.03$, $F_m = 0.18$, and $V_{lumen} = 1\,\mathrm{L}$. The PD baseline was set to $CDAI_0 = 275$, so $k_{in} = k_{out}\,\times 275$. We fit $k_{out}$, $E_{max}$, $EC_{50}$ to published mean $\Delta CDAI$ at weeks 2/4/8/12 for 4.8 g/day (−49.3, −61.8, −78.3, −101.1). The fitted values used in all simulations were

• $k_{out} = 1.6413\times10^{-3}\,\mathrm{h}^{-1}$ (time constant ~ 25 days),
• $E_{max} = 0.629$,
• $EC_{50} = 10.0\,\mathrm{mg}\,\mathrm{L}^{-1}$,
• $k_{in} = 0.451\,\mathrm{CDAI\,h}^{-1}$ to maintain the baseline $CDAI_0$.

Key calibrated parameters (units in SI where practical):

We simulated 12 weeks for three daily regimens: 1.5 g/day (3×500 mg), 3.0 g/day (3×1000 mg), 4.8 g/day (3×1600 mg). Numerical integration used SciPy’s solve_ivp; a stiff‑safe option is method='BDF' (results match RK).

Results

The plasma PK profiles, CDAI trajectories, and a therapeutic window visualization are shown below. All figures are generated directly from the implementation described above.

Figure 1: 5-ASA plasma concentration vs. time. Y-axis is in ng/mL; peaks scale approximately linearly with dose.

Plasma PK shows dose‑proportional systemic exposure and the expected multi‑peak pattern from thrice‑daily dosing. The simulated peak concentrations at steady cycling are approximately 1.35 µg/mL for 1.5 g/day, 2.69 µg/mL for 3.0 g/day, and 4.31 µg/mL for 4.8 g/day. These values scale roughly linearly with dose and are consistent with sustained‑release behavior reported in monographs when extrapolated from 1 g single‑dose metrics.

Figure 2: CDAI over time for different 5-ASA doses. The simulation shows that higher daily doses lead to a faster and larger improvement in disease symptoms over 12 weeks.

CDAI trajectories reproduce the characteristic delayed improvement. Calibration ensures the 4.8 g/day regimen approximates the published mean time course: at 2, 4, 8, and 12 weeks, the model predicts ($\Delta CDAI \approx$) −41/−65/−87/−94 (anchor: −49/−62/−78/−101). Lower daily doses achieve smaller and slower improvements, reflecting reduced local exposure.

Figure 3: Therapeutic window: efficacy vs. systemic exposure. Downward movement indicates better efficacy (ΔCDAI), while leftward movement indicates better safety (lower Cmax).

The therapeutic window plot places, for each regimen, the 12‑week efficacy ($\Delta CDAI$, more negative is better) against the peak plasma concentration (ng/mL, lower is safer). As dose increases from 1.5 to 4.8 g/day, points move downwards (greater efficacy) but rightwards (higher systemic exposure), quantifying the efficacy–safety trade‑off of conventional oral 5‑ASA.

Discussion

This PK/PD formulation provides a mechanistic and quantitative baseline for comparing conventional mesalazine to our engineered local EGF strategy. The indirect‑response equation encodes the clinically observed lag between exposure and symptom improvement, while the luminal‑driven PD term reflects 5‑ASA’s topical mode of action in the gut. Within this model, increasing dose broadens efficacy at the price of higher systemic peaks associated in the literature with renal, hepatic, and hematologic risks—illustrating the narrow therapeutic window challenge.

Assumptions were intentionally conservative to keep the model transparent and reproducible: systemic PK was summarized by a single central compartment for the parent and metabolite; coating and pH‑release nuances were abstracted into an effective ($k_a$); and inter‑individual variability was not modeled. Despite these simplifications, the model matches the shape and magnitude of published CDAI responses and yields dose‑proportional PK consistent with authoritative references. This makes it a suitable comparator for the next section, where we will show how local EGF delivery intends to shift the therapeutic window upward (greater efficacy) and leftward (lower systemic exposure).

EGF Local-Delivery Model

Methodology

We formulated a mechanistic PK/PD model for engineered E. coli (aka “Xylego”) that constitutively secrete EGF in the gut lumen. The model integrates: (i) bacterial population dynamics with an inflammation‑responsive Kill Switch; (ii) constitutive EGF production; (iii) lumen PK with proteolysis, minimal systemic absorption, and receptor‑mediated internalization tied to barrier access; and (iv) a damage–repair PD linking local EGF to CDAI.

State variables are bacterial count ($N$; cells), lumen EGF ($C_{lum}$; ng/mL), plasma EGF ($C_{pl}$; ng/mL), and damage level ($D$; dimensionless, $D=1$ maps to baseline CDAI).

Bacteria with Kill Switch (tetrathionate‑responsive):

The Kill Switch is triggered when tetrathionate, an inflammatory marker, is depleted; as $T$ falls during remission, k$_{kill}(T)$ rises and drives self‑termination.

Lumen and plasma EGF with receptor consumption and minimal absorption:

The mechanism by which EGF functions involves binding to its receptor, triggering a signal which is then internalized, leading to EGF’s degradation. In other words, its lifespan is quite short; this is captured by the $k_{int}$ and $k_{degrade}$ terms.

Damage–repair PD with EGF stimulation and high‑dose penalty (toxicity):

Mapping to CDAI uses $CDAI(t) = CDAI_{healthy} + s,D(t)$ with $s$ chosen so that $D=1$ yields ($CDAI_0=275$).

Parameterization reflects literature and wet‑lab guidance [10][11]. Rapid lumen proteolysis (half‑life ~20 min), 1% absorption to plasma, EGFR internalization on the order of tens of minutes, and an optimal EGF range near 50 ng/mL for wound healing. The $EC_{50}$ of 5 ng/mL reflects the concentration for half-maximal repair, while 50 ng/mL represents the upper end of the optimal therapeutic window before diminishing returns begin. Production per cell follows the OD0.01→16 ng/mL steady‑state heuristic (recalibrate once a direct secretion rate is measured).

We simulate 12 weeks for initial inocula $N_0$ in 1e8, 1e9, and 1e10.

Results

A quick steady-state calculation confirms the simulation’s output: with a bacterial load of ~1.2e10 CFU and a total EGF clearance rate of ~3.5 $\mathrm{h^{-1}}$ (from proteolysis and internalization), the expected lumen concentration is ~10.3 ng/mL, which closely matches the observed results, demonstrating the model is not a black box. The model is also robust to parameter changes affecting absorption; even a hypothetical 10-fold increase in the absorption rate ($k_{abs}$) results in peak plasma concentrations far below 1 ng/mL, confirming the therapy’s wide safety margin.

Figure 4: Lumen EGF concentration vs. time. Higher initial bacterial doses produce a faster and higher peak of local EGF.

The lumen EGF concentration rises in a dose-dependent manner before stabilizing as the bacterial population reaches its carrying capacity. Crucially, published experimental data indicate that EGF concentrations above 10 ng/mL are required for a therapeutic effect. Our model predicts that all tested doses achieve and sustain concentrations well above this known threshold, supporting the potential for a strong clinical response.

Figure 5: Disease activity (CDAI) vs. time. Higher initial bacterial doses lead to a faster and deeper initial improvement in symptoms.

The decline in CDAI is directly related to local EGF exposure, with larger inocula producing a more rapid initial response. This effect saturates, as the model includes a high-dose penalty to prevent unrealistic benefits at extreme concentrations. Notably, all three doses converge toward a similar improved state by 12 weeks, a dynamic that aligns with the behavior of probiotics, where sustained colonization provides a gradual but steady benefit.

Figure 6: Therapeutic window (Efficacy vs. Safety). The model predicts a wide safety margin, achieving efficacy with minimal systemic exposure.

This plot quantifies the trade-off between efficacy and safety. The y-axis shows the change in CDAI at 12 weeks (more negative is better), while the x-axis shows the peak plasma EGF concentration (lower is safer). All simulated doses achieve a strong therapeutic effect while remaining far to the left on the safety axis, indicating negligible systemic exposure compared to conventional therapies.

Discussion

This model captures the intended biological logic: Xylego gains a growth advantage from xylitol, continuously secretes EGF at the site of need, and the self‑limiting Kill Switch activates as inflammation resolves. EGF’s short life is represented via proteolysis and receptor‑mediated internalization, ensuring minimal systemic spillover. The PD reveals a practical optimum: sufficient local EGF to accelerate repair without entering a high‑dose regime where benefit wanes.

While robust, the model simplifies a complex biological system. The gut is treated as a single, well-mixed compartment, ignoring spatial gradients in the mucus layer or intestinal crypts. EGF internalization is modeled as a linear process, whereas EGFR binding is saturable and could be described with more detailed receptor kinetics. The model also simplifies the gut environment by omitting direct immune clearance, mucus turnover, competition with native microbiota, and the dynamic effects of dietary xylitol pulses. Furthermore, the “damage” variable is a phenomenological representation of Crohn’s disease, and some PD parameters are weakly identifiable without more targeted data. Finally, the model represents an “average” patient, not accounting for inter-patient variability, and focuses on PK/PD safety without modeling other potential risks like anti-EGF antibody formation or horizontal gene transfer.

A prospective simulation could explore optimizing dosing frequency. By modeling xylitol co-administration as a prebiotic that confers a competitive advantage (e.g., by increasing the effective growth rate $\mu$ or carrying capacity $K_{bac}$, we could investigate whether this leads to more sustained colonization. This would allow the therapeutic bacterial load to be maintained for longer, potentially reducing the required dosing frequency for long-term remission.

References