Intracellular Artificial Neural Networks
Traditional genetic circuits implement Boolean logic — a given input combination produces a binary output: gene expressed or not. This is powerful for well-defined triggering conditions but brittle in real biological environments, where inputs are continuous, noisy, and rarely cleanly digital. An AND gate that requires both AHL and SRP to exceed a threshold will fail in any scenario where signals are graded or partially correlated.
Intracellular Artificial Neural Networks (IANNs) overcome this by implementing analog, weighted computation directly in biological hardware. Several specific advantages over Boolean circuits:
- Continuous input integration. IANNs compute a weighted sum of input signals before applying a nonlinear activation function. A cell implementing an IANN can respond proportionally to the magnitude of an environmental signal — not just to whether it crossed a threshold. In a lake system with gradients of H₂S, salinity, and phosphorus, this is directly relevant: a consortium member should modulate its output in proportion to the severity of each stressor, not just toggle on or off.
- Multi-input discrimination. Boolean circuits with n inputs require 2ⁿ combinatorial logic gates to implement arbitrary functions. An IANN with n input neurons implements any mapping through learned weights, enabling non-linear classification of complex environmental states that cannot be decomposed into simple Boolean combinations.
- Noise tolerance. Biological signal transduction is inherently stochastic. IANNs with sigmoid-like activation functions are naturally robust to input noise — they saturate smoothly at extremes rather than oscillating at a Boolean threshold where small perturbations produce large output swings.
- Multilayer computation. Hidden layers allow the network to extract abstract features from inputs before producing output. A single-layer perceptron can only implement linearly separable functions; a two-layer network can approximate any continuous function. This is the difference between detecting whether H₂S exceeds a threshold and detecting whether the biogeochemical state of a lake has entered a specific degradation regime that requires a coordinated response across all three strains simultaneously.
- In-principle learnability. If the weights of an IANN can be set by evolutionary selection or directed mutagenesis rather than by top-down design, the network can be trained to recognize environmental patterns that are too complex to specify analytically.
The AND-gate circuit in Strain C (AHL × SRP) is a Boolean approximation of what an IANN would implement more robustly: a continuous integration of quorum signal magnitude, phosphorus concentration, and potentially H₂S and dissolved oxygen levels, producing a graded PhoA expression output proportional to the severity of the combined environmental stress. The transition from Boolean to IANN is part of the long-term design vision for the Strain C circuit in Aim 2.
Application: Eutrophication state classifier for real-time bioremediation activation in Lake Budi.
Lake Budi does not exist in a single, stable degraded state — it oscillates seasonally between mesotrophic and eutrophic conditions depending on tidal intrusion, stratification depth, and temperature. A Boolean circuit triggered by a fixed SRP threshold will either activate prematurely (false positive, wasting metabolic resources) or fail to activate during early-stage eutrophication (false negative, missing the window for intervention).
An IANN implemented in Strain C could receive three continuous inputs:
- X₁ — AHL quorum signal (proportional to total consortium density; proxy for biofilm maturity and remediation readiness)
- X₂ — Soluble reactive phosphorus (SRP) (direct eutrophication signal, measured via phosphate-sensing two-component system)
- X₃ — Dissolved oxygen / H₂S proxy (redox state of the water column; high H₂S correlates with anoxia and peak internal P loading)
A two-layer IANN with sigmoid activation in the hidden layer would compute a weighted combination of these three inputs, producing a graded PhoA expression output — higher expression when all three signals indicate active eutrophication, lower expression when conditions are recovering or ambiguous. The network could be pre-trained on historical limnological data from the four DGA seasonal campaigns (LME-UChile, 2010) to classify lake states before deployment.
Limitations:
- Weight implementation. Analog weights in biological systems require precise control of transcription factor binding affinities, RBS strengths, and protein degradation rates — all of which have intrinsic variability that makes fine-grained weight setting difficult without directed evolution or cell-free prototyping.
- Intracellular signal range. The dynamic range of intracellular molecular signals is narrower than digital abstractions suggest. Saturating effects in gene expression limit the effective linear range of each node, compressing the analog computation the network can perform.
- Crosstalk. In a multi-strain consortium, signals produced by one strain (e.g., AHL from the consortium) diffuse freely and cannot be kept strictly intracellular — the IANN's input nodes will receive signal mixtures from the environment that the network was not trained to classify.
- Metabolic burden. Expressing a multilayer network of transcription factors, ribonucleases, and reporters imposes significant metabolic cost, potentially reducing the fitness of the IANN-carrying strain relative to simpler consortium members — destabilizing the auxotrophic ring.
The diagram below shows an intracellular multilayer perceptron designed for Füzi Poiesis — three environmental inputs (H₂S/heavy metals, salinity/dissolved oxygen, nutrients P/N) processed through two layers of transcription (Tx) and translation (Tl) nodes, producing three outputs: degrading enzymes (layer 2 left), antimicrobial action (layer 2 center), and GFP reporter (layer 2 right). Intercellular information vectors Z₁ (validation signal) and Z₂ (execution order) coordinate between cells.
Input layer: Three environmental sensors (X₁, X₂, X₃) each drive a Tx node that transcribes a sensor-responsive mRNA. The Tx nodes for X₁ and X₂ converge on a first hidden layer that integrates both signals before passing to the intercellular communication layer. X₃ feeds directly to the second cell (Z₂ pathway).
Layer 1 (hidden): Two Tx/Tl node pairs inside the first cell implement weighted summation of X₁ and X₂. The output of this layer is the intercellular validation signal Z₁ — an endoribonuclease (Csy4 analog) whose concentration encodes the weighted sum of inputs. Z₁ diffuses or is exported to layer 2.
Layer 2 (output): The second cell receives Z₁ and Z₂. Three Tl nodes integrate these intercellular signals through their mRNA regulatory elements — Z₁ cleaves at Csy4 sites to activate or repress each output independently. The three outputs (Y₁ degrading enzymes, Y₂ antimicrobial action, Y₃ GFP reporter) are produced in proportion to the combined input state. The endoribonuclease from layer 1 regulates GFP mRNA stability in layer 2, implementing the Csy4-based regulation specified in the assignment.
This multilayer perceptron architecture is the long-term analog of the Boolean AND-gate circuit currently implemented in Strain C of Füzi Poiesis. In the current design, PhoA expression is triggered by a two-input Boolean AND (AHL AND SRP > 0.5 mg/L). The IANN generalizes this to a weighted, multilayer computation that could simultaneously regulate alkaline phosphatase output (Strain C), SQR/PDO expression (Strain B), and GFP reporter intensity in proportion to the actual eutrophication state — rather than as a binary on/off switch. The GFP output (Y₃) maps directly to the fluorescent reporter that would allow field monitoring of consortium activity without sacrificial sampling.
Fungal Materials — Territorio Lafkenche
Fungal materials exploit the mycelium — the vegetative network of filamentous hyphae — as a structural scaffold. The most developed applications are:
This question has a specific answer from the territory where this project is grounded — the Lafkenche coast of Araucanía, at the edge of the Lake Budi basin. The generic answer about fungal synthetic biology is less useful here than the concrete answer about which organisms already grow in this watershed and what they are already doing.
Ganoderma australe, the southern bracket fungus known in Mapuche ethnobotany as a culturally significant species, grows natively in the valdivian temperate rainforest of the Budi basin. It produces extracellular laccases and peroxidases — the same class of enzymes that oxidize phenolic compounds and aromatic ring structures in lignin. These enzymes also attack polyphenolic complexes in anoxic sediments, potentially mobilizing iron-bound phosphorus into forms accessible for microbial transformation. Ganoderma australe is the priority organism for three reasons that align with the ethical architecture of Füzi Poiesis: it is native to the basin (no exotic species introduction), it is culturally present in Lafkenche territorial knowledge, and its enzymatic toolkit complements Strain B's SQR-mediated sulfide oxidation — attacking the problem from the solid phase (sediment) while the bacterial consortium acts in the water column.
What to engineer and why: The most valuable genetic engineering target in Ganoderma australe for Lake Budi applications would be enhanced laccase secretion under saline and anoxic conditions. Native laccases require oxygen as a cofactor — at the halocline and below, where anoxia is permanent, laccase activity drops to near zero. Engineering a version with reduced oxygen dependence (or co-expressing an alternative electron acceptor pathway) would allow the fungus to act at the sediment-water interface where iron-bound phosphorus release is most active. A second target would be upregulated phosphate-chelating organic acid secretion (oxalic acid, citric acid) to compete with the iron binding of phosphorus in sediment, releasing it into the water column where Strain C's PhoA can complete the remediation cycle.
Structural persistence. Mycelium forms a three-dimensional hyphal network that physically colonizes solid substrates — sediment, biochar, decomposing plant matter — creating stable, persistent structures that planktonic bacteria cannot. In an open aquatic system subject to tidal flushing and storm resuspension, mycelial networks anchored to the Bokashi del Budi matrix would maintain spatial organization that bacterial biofilms alone cannot sustain.
Eukaryotic secretory machinery. Fungi produce and secrete large, complex extracellular enzymes (laccases, peroxidases, cellulases) with post-translational modifications that bacterial expression systems struggle to replicate. Heterologous expression of laccases in E. coli typically yields misfolded, inactive protein; the same gene in its native Ganoderma host produces correctly glycosylated, fully active enzyme.
Cultural legitimacy. Introducing a genetically modified bacterial consortium into a Lafkenche sacred lake raises ethical questions that require years of community deliberation. Introducing a genetic modification into a fungus that already grows in the forest of the basin — one with existing cultural significance — is a meaningfully different ethical proposition. The organism is not foreign; the modification is a targeted enhancement of something already present in the territory.
First DNA Twist Order — Füzi Poiesis
This section documents the first DNA synthesis order for Füzi Poiesis — the codon-optimized phoA insert for Strain C's AND-gate remediation circuit — and the backbone vector into which it will be synthesized.
The insert sequence submitted for synthesis is the codon-optimized phoA coding sequence (1,410 bp), designed for expression in E. coli K-12 MG1655 as part of the AND-gate PhoA circuit in Strain C. Codon optimization was performed using the IDT Codon Optimization Tool, removing codons with usage frequency below 10% and matching GC content to the K-12 genome average of 51%.
| Parameter | Value |
|---|---|
| Gene | phoA (alkaline phosphatase) — E. coli K-12 MG1655 |
| Optimization | IDT Codon Optimization Tool — codons <10% frequency removed |
| Insert length | 1,410 bp |
| GC content | 51% (matched to K-12 genome average) |
| 5′ overhang | 20 bp Gibson Assembly overlap with P_lux (BBa_R0062) |
| 3′ overhang | 20 bp Gibson Assembly overlap with BBa_B0015 terminator |
| RBS validation | RBS Calculator v2.1 — target 10,000 a.u. translation initiation |
| RBS structural check | RNAfold MFE −22.60 kcal/mol — ribosome accessible |
| R-M shielding | EcoRI, SpeI, XbaI — 0 internal sites confirmed (NEBcutter v3) |
The phoA_opt insert is synthesized into the pANDgate backbone vector, which provides the regulatory architecture for the AND-gate circuit. The backbone includes the two promoter inputs (P_lux BBa_R0062 for AHL quorum sensing and P_phoB placeholder for phosphate sensing), a temperature-sensitive origin of replication (ts-ori, architectural placeholder — cold-inactivating variant specified as Aim 2 design requirement), and a KanR selection marker.
| Element | Part / Source | Function |
|---|---|---|
| Promoter input X₁ | BBa_R0062 (P_lux) | AHL/LuxR quorum sensing — activates at consortium density threshold |
| Promoter input X₂ | BBa_K116401 placeholder | Phosphate-excess sensor — Aim 2 implementation target |
| RBS | Designed — RBS Calculator v2.1 | ~10,000 a.u. translation initiation for sub-toxic PhoA expression |
| Insert | phoA_opt (1,410 bp) | Alkaline phosphatase — phosphorus remediation output |
| Terminator | BBa_B0015 (double terminator) | Transcriptional insulation |
| Origin of replication | ts-ori (placeholder) | Conditional replication — cold-inactivating variant for Aim 2 |
| Selection marker | KanR | Kanamycin resistance — compatible with auxotrophic ring constraint |
| Total plasmid size | pFP-C: 4,238 bp | Complete annotated sequence in Benchling |
As a global student at the SynBio USFQ Node (Universidad de La Frontera, Temuco, Chile), physical synthesis via Twist Biosciences is not executed within the scope of this course. The Twist order documented here represents the complete design specification — sequence, backbone, and part annotations — that constitutes the synthesis-ready deliverable for Aim 1. Physical synthesis of pFP-C is planned as the first step of Aim 2, in collaboration with UFRO-BIOREN, where the construct will be transformed into E. coli K-12 MG1655 for validation of AND-gate Boolean logic before progression to Halomonas elongata chassis integration.
pFP-A (4,626 bp) — Strain A · mcjABCD anti-coliform cassette
pFP-B (3,677 bp) — Strain B · sqr-pdo anti-H₂S cassette
pFP-C (4,238 bp) — Strain C · AND-gate PhoA anti-phosphorus plasmid (complete)