Modelling

Abstract

Our modeling work supports the development of a biocontrol agent for cacao. We use curated databases and computational analysis to select relevant Phytophthora species and AMPs, ensuring our experimental work is robust and broadly applicable.

Phytophthora Database

As a first step, we compiled an overview of Phytophthora species that occur in nature. To support our modeling, a database was constructed to identify which species are most relevant for protecting cacao pods, and which could serve as alternatives in case cocoa-infecting isolates were unobtainable.
This database provides information on common hosts, geographic distribution, associated diseases, and lifestyle traits. For entries involving cacao or related hosts, these parameters were examined in greater detail to align with black pod disease and to consider conditions where cacao losses are most severe. In practice, this allowed us to shortlist species that infect cacao, and exhibit trait profiles representative of the broader threat landscape, ensuring that early tests are relevant beyond a single species.
All information was obtained from literature and public repositories such as the NCBI, the Phytophthora Database and UniProt [1], [2], [3]. By comparing species based on features critical for control strategies, the database directly informed both our agent-based modeling and subsequent wet lab experiments.

AMP Database

Our antimicrobial peptide (AMP) database is the backbone of our project, guiding both the design of our sensing and secretion system, as well as the selection of AMPs that we will test in the lab. Each candidate AMP carries a unique identifier from databases such as the Database of Antimicrobial Activity and Structure of Peptides, UniProt, and the Antimicrobial Peptide Database[3], [4], [5].
For every entry, we have identified the Phytophthora target species, its molecular target, the biological role of the target, and the mechanism by which the AMP could inhibit it. Alongside these annotations, we stored the amino acid sequence, its length, the citation to the primary source, and an internal status tag that tracks the peptide’s journey from discarded to shortlisted. Codon optimized sequence for B. subtilis, as well as notes for synthesis and assembly are also included.
In its current form, the database contains roughly forty peptides, with sequence lengths ranging from short motifs to larger proteins. Target species annotations allowed us to pick AMPs that are active against (related) species that could potentially lead to black pod rot. To visualize our shortlisted AMPs, we used the amino-acid sequences in the AMP database and predicted the structures with AlphaFold, then analyzed the models in UCSF Chimera and ChimeraX [8].

AMP visualizations

Seed-derived vicilin fragment shown to inhibit Phytophthora hyphal growth; likely acts by binding cell-wall carbohydrates.

Short, cationic plant peptide (Eucommia-type) that rapidly permeabilizes pathogen membranes.

α-helical AMP that inserts into oomycete membranes and forms pores.

Carbohydrate-binding (hevein-type) lectin from Euonymus europaeus that targets β-glucan or cellulose-rich cell walls.

Invertebrate-derived antifungal peptide with robust membrane activity and proteolytic resilience.

Our five shortlisted AMPs are visualized and colored by Kyte-Doolittle hydrophobicity [9]. Strongly charged hydrophilic residues, aspartate (-3.5), glutamate (-3.5), arginine (-4.5) and lysine (-3.9) are shown in blue. Off-white denotes the mildly hydrophilic residues serine (-0.8), threonine (-0.7), and glycine (-0.4). Hydrophobic residues with strong lipid affinity such as valine (4.2), isoleucine (4.5), leucine (3.8) and phenylalanine (2.8) are colored yellow. This scheme highlights amphipathic patterning across the peptides, helping to visualize potential membrane-interaction surfaces and pore-forming motifs.

AMP Interaction with the Phytophthora membrane

To model the interaction of the AMP with the Phytophthora membrane, we have used the AlphaFold PDB outputs of our predicted AMPs and combined this with the input from the PPM 3.0 server [10]. We determined the orientation of the selected AMP helix relative to a phospholipid bilayer and uploaded the output in UCSF Chimera to obtain ach membrane placement consistent with an amphipathic, surface-seeking peptide. For the visualisations, we also used iCn3D to render the PPM-oriented peptide and membrane boundaries [11]. Specifically, we applied the colour by normalized hydrophobicity scheme with dark green for hydrophobic residues (Trp, Phe, Leu, Ile, Tyr, Met, Val, Cys), light green for polar residues (Pro, Thr, Ser, Ala, Gln, Asn, Gly), and grey for charged, non-hydrophobic residues. As shown in the video, the hydrophobic face is aligned with the bilayer core, while grey charged residues cluster near the polar head group interface, this is where pore-forming is likely to occur.
For our AMP01, vicilin-like antimicrobial peptides 2-2, the predicted hydrophobic thickness was approximately 31.4 ± 0.6 Å with a tilt of 25 ± 5° relative to the bilayer and a transfer free energy of around −25.8 kcal mol⁻¹, indicating a strongly favourable association with the membrane. The membrane is represented by white dotted slabs which create the bilayer boundaries. The AMP helix sits obliquely across this 31 Å core, with its hydrophobic face buried and its lysine and arginine-rich face in the upper membrane. This geometry supports our predicted membrane-disruptive mode of action, typical of cationic AMPs, where interfacial binding and tilted insertion facilitate local thinning and the perturbation of the Phytophthora membrane.

Sensing System

For our sensing system, we have chosen an AND logic gate, that requires two inputs. Once these two environmental cues are present at the same time, it will activate transcription of our AMPs. In our design, these two inputs are elicitins and reactive oxygen species (ROS), signals that together indicate the presence of Phytophthora and trigger the production of antimicrobial peptides.
Elicitins are highly conserved, small proteins secreted by Phytophthora species during plant infections. These proteins are well-studied and are known to trigger strong necrotic activity, accompanied by early defense responses such as ROS bursts and the expression of pathogenesis-related proteins [12]. Importantly, elicitins act by binding to plant ligands and directly activate plant defense signaling. This makes them highly specific markers of Phytophthora activity, rather than general stress indicators.
ROS, in turn, plays a central position in plant immunity. They act both as local messengers and systemic signals, controlling defense activation and hypersensitive response (HR)-associated cell death. However, Phytophthora species secrete effector proteins such as RxLRs to manipulate this system. Li and his team showed that ectopic expression of RxLR effectors can intensify ROS accumulation, causing premature cell death. While this may appear counterintuitive, it reflects the hemi-biotrophic lifestyle of Phytophthora, in which the pathogen exploits necrotic, dead tissue as a nutrient-rich substrate to sustain growth and spread.
This interplay results in a paradox: what should have been the plant’s defense (ROS), becomes a double-edged sword that favors the pathogen. Crucially, this manipulation creates unusually high, sustained ROS levels during early infection. Such concentrations are practical to detect and serve as a strong signal that the plant is under Phytophthora attack [13], [14].
However, ROS alone could also be a false positive as it is an indicator of general microbial infection and elicitins alone may be below critical thresholds in the initial stages of infection but when taken together, elicitins and ROS represent two complementary, pathogen-specific indicators. Elicitins provide specificity by being unique Phytophthora-derived proteins, while ROS provide robustness as a measurable, amplified plant response. The co-detection of elicitin and ROS mimics the plant’s natural signaling logic which means our dual-input system replicates the plant’s own immune logic. This is a strong validation of our idea of a synthetic biosensing system [16].
Since we target both, it allows us to design a sensing system that is not only reliable but also resilient against false positives caused by general plant stress [17]. This dual strategy overcomes the limitations of each individual signal and offers a powerful advantage for early and accurate detection of Phytophthora infection. Now that the signals are selected and their secreted concentrations are validated to be enough to be detected by our model.

Figure 1: Split-T7 AND gate integrates elicitin sensing and ROS to drive promoter activation

We move on to the next step, which is how this information which was sensed will be processed. We have designed a simple yet elegant plan using the split T7 RNAP. A split T7 RNAP has been widely used to create synthetic gene circuits, especially due to its transcriptional activity and promoter malleability which makes it suited for use in synthetic gene circuits.
The T7 RNAP was obtained from bacteriophage T7 and is an RNA polymerase which recognizes a very specific T7 promotor. It's an orthogonal system, which means that native T7 RNAP will not transcribe the native genes driven by host promoters. Neither can the polymerases in the host cell recognize the promoter of the T7 RNAP, which means the two transcriptional systems would not transcribe each other [18].
In the split version of the T7 RNAP, there usually is a C-terminal fragment that contains the catalytic core and the DNA binding domain. On the other hand, the N-terminal fragment is needed for transcript elongation. Ikeda and his team found that if these two fragments are cut as N terminal and C terminal fragment, these fragments would be inactive of their own and cannot transcribe [19]. However, if they are expressed in the same cell, then they dimerize and drive transcription from the T7 promotors.
Another option to bring the split T7 polymerase together is through inteins. An intein is a protein that can excise itself from a host protein and splice the remaining parts of the protein together. Split inteins can be used to covalently join protein sub-units. The inteins are split and then fused to two different protein domains, once the two protein domains are expressed in the cell, an auto catalytic excision of the inteins occurs and the two split domains are joined together by a peptide bond into a full functional proteins [20].
We designed our sensing system inspired by a split intein based T7 RNAP. Since we already established two signals to be detected as ROS and elicitins, we would design our system by splitting the T7 RNAP into halves: N terminal fragment and the C terminal fragment. Each of these fragments would be then connected by a split intein fragment. The expression of the two fragments of T7 RNAP will be controlled by two different promotors, one for each signal. Once the RNAP is assembled, it would bind to the promoter to initiate transcription of the gene encoding our AMPs, which are located downstream of this sequence [21].
For ROS, there will be a N-terminal fragment of the T7 RNAP coupled to the members of the PerR regulon PkatA, PmrgA and PzosA which are strongly induced due to the presence of ROS [22]. Even after extensive literature analysis, there was no evidence of an inducible promoter which is specific for elicitins, however we found plant-based receptors which can detect the presence of elicitins [23]. This inspired us to design our own receptor-based sensing system for the second signal.
Our receptor-based system should be feasible, since similar plant receptors recognize a highly conserved domain of elicitins. One example is the elicitin response-like (ELR) protein, isolated from Solanum microdontum. It associates with immune co-receptors like BAK1/SERK3 and mediate broad-spectrum recognition of elicitins [24].
A next step in creating this sensing system would be to design a bacterial receptor based on ELR, coupled to a synthetic signaling cascade which induces the expression of the C-terminal fragment of the RNAP in the presence of elicitins. Once the two fragments are expressed, they would then be spliced together by the inteins, forming the complete T7 RNAP. This would then bind to the T7 promoter leading to a stronger expression of the AMPs.

Figure 2: Split-T7 reconstitution drives tagged AMP co-expression and secretion

Even though the split T7 RNAP is a robust system, we do acknowledge that this system could come with some challenges, such as high energy demand, reduced growth of the chassis and imbalanced transcription, translation and folding along with leakage of the gene of interest [25]. However, we do address this issue by using a fragmented T7 RNAP and a synthetic receptor-based system which finetunes and improves control over the system. Our dual layered kill switch also ensures robust biosafety and prevents horizontal gene transfer. The T7 promoter also leads to a very high level of transcription and can cost a lot of energy to the cell, however fine tuning the expression conditions such as low inducer or optimizing induction time, can help mitigate the issue of high energy consumption.
The dual-input sensing system minimizes false positives while adding an additional layer of specificity and robustness to our design. This approach not only enhances the reliability of our biosensor but also provides a framework that future iGEM teams can build upon to detect pathogenic infections. Our design represents a valuable step toward the next generation of biosensors, that not only sense a pathogen, but will also act upon encountering one.

Agent-based model: AND-Logic gate

To determine the expected behaviour of our engineered B. subtilis and AMP secretion, we modelled our design before building it. We used NetLogo to create our agent-based model [26]. In each simulation, concentration of the signal fields can be visualized in the plots with AMP levels drawn in red, ROS in yellow, Phytophthora remaining in purple, elicitins in green and β-glucans in pink.

The pathogen, is represented by the maroon patch in the center of the screen, while the bacteria roam around and are changing from red to yellow upon activation. They create a brightening halo when they are actively secreting AMPs. As elicitin and ROS levels are rising and reach threshold, β-glucan fragments released by the degradation of the pathogen's cell wall which further amplify ROS and activate the AND gate. As a result, the cells increase secretion above baseline levels. Diffusion and evaporation, set by dehydration rate, shape the halo’s spread and the pathogen dissapears continuously.

Once the pathogen is cleared, the bacteria return to the ON/OFF cycle with random activation every 30 time steps. Consequentlty, secretion drops to a baseline level only produced by active cells. Elicitin, glucan, and ROS concentrations remain near zero as there is no more Phytophthora present, the AND gate never gets triggered and there is no longer boosted AMP secretion.

In the previous cases, dehydration levels were low meaning the environment was high in moisture. Due to the changes set in the environment, dehydration, there is a sharp drop in diffusion rates while evaporation rates rise. The AMP secretion halo is much smaller and fades quicker than in moist conditions.

In this example, the kill-switch is triggered mid-simulation. A uniform toxin pulse instantly wipes out turtles on any patch at or above the lethal dose. Our bacteria vanish almost simultaneously, leaving only the diffusing AMP field to decay under evaporation. After the kill-switch, the blob continues to decline for a short window from residual AMP and the model’s natural decay term, then stabilizes. Indeed, with no living bacterium left, plots for secretion flatten, and the display shifts toward a dimmer background as AMP evaporates and the remaining pathogen persists.

Overview & Quick Start

Our agent-based model (ABM) simulates a population of engineered bacteria that protects against a Phytophthora infection by secreting AMPs. The secreted AMPs diffuse and evaporate over a 2D lattice. Each patch of the lattice carries state variables for AMP concentration, pathogen biomass, three pathogen related recognition signals elicitin, ROS and β-glucan fragments, with an optional toxin that can be used as a kill-switch. Each bacterium is an individual that senses local signals through an intracellular AND logic gate. When the protein output is above the threshold, cells turn ON and start secretion of our AMPs. The antimicrobial field spreads by diffusion and kills the pathogen with a sigmoidal effector function.
The environmental stress in this model was designed with a dehydration factor that simultaneously reduces diffusion, increases evaporation and attenuates gene expression. The model is designed to visualize a causal link from pathogen signals to single-cell decisions, from AMP accumulation to pathogen clearance, with optional kill-switch interventions and base line activity dynamics when no pathogen is present. Our model does not consider bacterial replication.
To run this model, first choose the initial bacterial population size and set the dehydration levels, then decide whether to start with or without the presence of Phytophthora. It is recommended move the slider to the right to improve the speed of the simulation. To test the kill switch, press the kill-switch button which instantly triggers the toxin and kills all bacteria present.
To run the model in NetLogo, download NetLogo (https://www.netlogo.org/) and run this code after downloading it on ModellingCommons (http://blog.modelingcommons.org/browse/one_model/7735#model_tabs_browse_info).

Pathogen & Environment

In our model, the pathogen is represented as a disk at the center, with its size easily adjustable. To simulate environmental stress, the roughness of its boundary increases under dry conditions, reflecting reduced suitability for growth. The pathogen’s decline over time is modeled through a combination of natural decay and a Hill-type antimicrobial effect, which limits the rate of cell loss per time step. This approach prevents unrealistic, instantaneous die-off while maintaining a response once antimicrobial concentrations exceed a critical threshold.
Environmental transport and stress in our model are governed by diffusion and evaporation rates that depend on moisture levels. As dehydration increases, diffusion slows while evaporation accelerates, reflecting how dry conditions limit both spread and persistence. Under high moisture, however, Phytophthora thrives - humidity promotes sporulation, zoospore movement, and splash-mediated dispersal, while prolonged wetness lowers infection barriers. In these conditions, our model assigns a high diffusion rate and low evaporation rate, allowing AMPs to spread efficiently and persist longer in the boundary layer. As the environment dries, diffusion weakens and evaporation increases, causing AMP concentrations to recede more quickly. This dynamic coupling allows environmental moisture to naturally modulate AMP deployment, aligning the synthetic response with real-world Phytophthora infection dynamics.

Signal Fields

Elicitin is produced in proportion to pathogen biomass and gradually decays over time. β-glucan fragments are released when antimicrobial peptides damage the pathogen’s cell wall; these fragments then stimulate the production of reactive oxygen species (ROS), which also decay at their own rate. Together, these signal-field parameters define how recognition unfolds and how long signals persist. Elicitin-prod and elicitin-decay determine how quickly the PAMP signal rises upon pathogen contact and how long it remains active, while glucan-release-rate and glucan-decay control the generation and lifetime of damage fragments. ROS production is modelled as a combination of three terms: ROS-prod-amp, ROS-prod-phyto, and ROS-prod-glucan. This was counteracted by ROS-decay. Increasing the glucan-driven term enhances the system’s sensitivity to pathogen damage.

Intracellular Logic Gate

Figure 3: From sensing to clearing: AND logic gate and β-glucan feedback drive AMP-mediated pathogen clearance. Created in BioRender.com

The intracellular logic gate integrates two inputs, elicitin and ROS, which are each processed through Hill functions. Elicitin is produced in proportion to pathogen biomass and decays over time, while β-glucan fragments are released when antimicrobial peptides (AMPs) damage the pathogen’s cell wall. These fragments stimulate additional ROS production, which also declines at its own decay rate.
The cell’s transcription rate combines a basal leak term with a gated response, scaled by a dehydration factor that reduces expression under dry conditions. Translated proteins accumulate from mRNA and gradually decay over time. Once protein levels exceed a defined threshold, the cell is considered activated, triggering an increase in AMP secretion. A probabilistic switch is included to bias activation toward the infection center-ensuring that strong secretion occurs where it is most needed, while a low baseline response remains active elsewhere through an ON/OFF system.
Model parameters determine when and how cells activate. The basal transcription rate (a₀) defines background expression in the absence of inputs, while aₘₐₓ controls the maximum transcriptional gain when both inputs are high. The translation rate (β) converts mRNA to protein, and the decay rates (dₘ and dₚ) govern how quickly mRNA and proteins degrade—faster decay shortens and weakens responses, while slower decay introduces persistence or memory. Input sensitivities (K₁ and K₂) represent the half-maximal constants of the Hill functions for elicitin and ROS, where larger values require stronger signals to activate. The steepness parameters (n₁ and n₂) shape how abrupt this activation is, with higher values producing a more switch-like response. Finally, the decision threshold (y-threshold) sets the protein level required for activation; increasing it filters out weak or transient signals. In our model, once this threshold is surpassed, the probability of switching ON decreases with distance from the infection center, ensuring that AMP production is concentrated where Phytophthora pressure is highest.
For an input s with parameters K for sensitivity and n for cooperativity, the Hill response is given by: s^n/(K^n+s^n)
By applying separate Hill functions to elicitin and ROS and then multiplying their outputs, we captured the logic that neither input alone can strongly activate transcription. Only when both signals are simultaneously high does their product approach one, triggering the aₘₐₓ term-the defining behaviour of our AND gate.

Rationale Behind the AND gate

We chose an AND gate because it best represents the biological rule we aim to capture. Our antimicrobial peptides (AMPs) should be secreted strongly only when both independent cues indicate a genuine Phytophthora infection. This design prevents unnecessary secretion and ensures accurate spatial control. In our model, ROS levels rise most where AMPs are actively interacting with Phytophthora, so the product of the two Hill terms peaks in these regions. This leads to a targeted secretion boost exactly where the pathogen pressure is highest, making the response both efficient and effective.
Alternative logic configurations fail to meet these criteria. A NAND gate would invert the response, suppressing secretion when both cues are high and triggering it in less relevant areas. An OR gate or any single-input system would act prematurely, responding to partial or background signals and causing false positives even without a true infection. A NOR gate, on the other hand, would deactivate secretion precisely when evidence of attack appears. The AND gate therefore offers the simplest and most reliable framework to achieve a pathogen-specific and biologically meaningful response.

Secretion Logic

Secretion in our model operates on two levels to balance safety and specificity. The parameter q defines the baseline secretion rate, active even during pathogen-free periods when some bacteria may randomly release small amounts of AMPs. The parameter q_boost represents the additional secretion produced by activated cells during infection and is scaled by Euclidean distance from the infection center, using an exponential decay so that secretion is concentrated near the pathogen. A master switch variable controls whether secretion logic is active at any time.
Throughout all phases, each bacterium moves forward in small, slightly randomized steps with a gentle inward nudge near the boundary to prevent clustering at the edges. This motion naturally redistributes AMP secretion in space. When no pathogen is present, the population periodically reshuffles which bacteria are active based on a Bernoulli draw, defined by reshuffle-period and next-reshuffle-tick. As a result, background secretion remains heterogeneous and dynamic, simulating the stochastic behavior expected in real microbial populations.

Kill-switch

Our model also includes a toxin field to support the following kill-switch scenario. A toxin dose is injected on a uniform level into all patches when activated and the lethal dose fixes the threshold above which any turtle on a patch dies immediately. More information on the kill switch can be found on the safety and security page.

Outcomes

The following parameters define the outcomes and visualization of our model. The effective concentration C_eff is the antimicrobial level at which a patch is considered effective. The target-protected-area sets how many patches must be above C_eff to declare operational protection, and time-to-protection records the first tick when that target is reached. The amp-max and amp-max-attack scalars normalize the colour mapping of AMP in pathogen-free and pathogen-present phases so that halos remain informative over different ranges. Finally, plotting variables such as amp-secreted-per-tick and the means of m and y provide a link between molecular-level dynamics and population-level outcomes.

Experimental validation

In this project, no laboratory testing of the proposed system was performed. The next crucial step is experimental validation, as bench-scale and in vivo testing are essential to confirm the model’s predictions and verify that the system is safe, specific, and effective under real-world conditions. Demonstrating functionality in a controlled laboratory environment will lay the groundwork for translating our design into a reliable and practical biocontrol solution.