Overview
This modeling framework combines experimental and commercial modeling to achieve an integrated optimization of HullGuard. The experimental modeling focuses on overcoming catalytic and diffusion bottlenecks in zosteric acid (ZA) biosynthesis through computational redesign of SULT1A1 and TAL–SULT1A1 fusion proteins, leading to a 2.5–3.6× improvement in ZA yield. The commercial modeling complements this by evaluating scalability and sustainability using techno-economic and life-cycle analyses.
Experimental Modeling
Background
Previous studies have demonstrated that the sulfation of pHCA to ZA catalyzed by SULT1A1 represents a "high-affinity but low-capacity" reaction, which inherently limits catalytic throughput. Moreover, PAPS and PAP have comparable affinities to SULT and compete for the binding domain, and the resulting PAP–SULT–acceptor and PAPS–SULT–product complexes are dead-end species that limit substrate utilization and overall ZA biosynthesis efficiency[1]. Directed evolution has already been shown to improve the activity and stability of human SULT1A1, and literature suggests this strategy could enhance rSULT1A1 tolerance to high concentrations of PAP and pHCA[2].
Therefore, we selected SULT1A1 as the primary target for single-enzyme engineering to enhance catalytic performance. In parallel, we designed a fusion protein strategy, coupling TAL with SULT1A1 to enable substrate channeling, thereby minimizing the intracellular accumulation of pHCA and improving overall flux toward ZA. Together, these complementary approaches address both intrinsic enzymatic limitations and inter-enzyme transfer efficiency, ultimately aiming to achieve higher ZA yields.
| Strategy | Target Problem | Mechanism | Expected Effect |
|---|---|---|---|
| Single-enzyme modification | Limited catalytic efficiency of SULT1A1; inhibition by PAP and pHCA | Site-directed mutagenesis to improve activity, stability, and tolerance | Enhanced catalytic performance of SULT1A1, reduced inhibition, higher ZA yield |
| Fusion protein design | Intracellular accumulation of pHCA causes inhibition and toxicity; inefficient substrate transfer | Fusion of TAL with SULT1A1 to enable substrate channeling, shortening the diffusion path | Reduced pHCA accumulation, alleviated inhibition, improved metabolic flux and ZA yield |
Methods
1.Targeted enzyme engineering
Step 1: Find the active site of SULT1A1 (binding with pHCA and PAPS)
In this step, we choose to use Autodock Vina to perform docking calculations on ligands and enzymes, while using the RCSB PDB database to search for similar crystal structures for validation.
We obtained the ligand structures of PAPS and pHCA from PubChem, and performed molecular docking with SULT1A1 using AutoDock Vina. After docking, residues located within 5Å of each ligand were identified as potential interaction sites. The results are as follows:
SULT1A1-pHCA: 41-48, 104, 126, 134, 225, 250, 251-255, affinity = -7.3 kcal/mol.
SULT1A1-PAPS: 41-49, 104, 126, 134, 138, 189, 193, 223-225, 228, 236, 250-256, affinity: -10.4 kcal/mol.
Step 2: Determine the high-conservation and low-conservation regions of SULT1A1
Query the FASTA format of enzyme amino acid sequences from multiple sources via NCBI BLAST, perform multiple sequence alignment using Clustal Omega and MAFFT, and preliminarily screen the variable mutation regions.
Then use Jalview to conductin-depthcomparisons to identify saturated mutationalternativecheck points.
Using Rattus norvegicus SULT1A1 (NCBI RefSeq NP_114022.1) as the query, we performed BLAST searches in the NCBI database and selected the 50 most similar heterologous SULT1A1/SULT sequences. These sequences were aligned with Clustal Omega and MAFFT, and the alignment was reviewed in Jalview. Positions with low conservation were flagged as checkpoints, and the corresponding residues were mapped and colored in PyMOL:
However, we observed minimal overlap between active-site binding checkpoints and low-conservation positions from Jalview analysis. Because the BLAST-based multiple sequence alingment used only 50 sequences, we suspected limited coverage was biasing our identification of high-conservation versus low-conservation positions. To address this issue, we used the ConSurf server to generate fully automated MSAs and conservation profiles from larger heterologous SULT1A1 sets (500, 1000, and 2000 sequences) and compared the results. The variable regions identified from this expanded analysis are as follows:
- Common Consurf grade=1: 3, 4, 18, 81, 82, 84, 85, 144, 241, 242
- Common Consurf grade=2: 9, 19, 72, 91, 94, 143, 164, 243
- Common Consurf grade=3: 6, 7, 10, 17, 73, 80, 89, 90, 93, 157, 160, 205, 230, 238, 246, 247, 250, 271, 277, 286
Finally, comparing the conservation profile with residues surrounding the binding checkpoint, we selected four low-conservation positions—Y42, Y236, P250, and T256—proximal to the site for saturation mutagenesis.
Step 3: Calculate the change in free energy after saturation mutagensis
In this step, we used ProtSSN to perform saturation mutagenesis calculations on all amino acids of SULT1A1; at the same time, we used Rosetta and FoldX to perform calculations on the four selected saturation mutagenesis sites and two point mutation sites.
Our goal was to estimate changes in folding free energy (ΔΔG) for SULT1A1 mutants relative to wild type. Guided by our prior conservation analysis, we selected Y42, Y236, P250, and T256 as mutation sites, generated single-amino acid variants at these positions, and computed ΔΔG values using both FoldX and RosettaDDG.
We observed notable inconsistencies between FoldX and RosettaDDG in predicted ΔΔG values for the same set of mutations. After a literature review and cross-comparison with published benchmarking datasets, we concluded that RosettaDDG provides more reliable predictions. Previous studies have shown that while FoldX is computationally efficient and accurate for high-resolution crystal structures, its performance decreases significantly on homology models or lower-resolution templates (Schymkowitz et al., 2005; Tokuriki et al., 2007). In contrast, RosettaDDG demonstrates higher robustness to structural uncertainty, better correlation with experimental stability assays, and superior classification accuracy across large-scale mutational datasets such as VAMP-seq and ProTherm (Kellogg et al., 2011; Park et al., 2016).
Based on these findings, we adopted RosettaDDG as the primary tool for stability prediction and used FoldX only as a rapid complementary reference.
After observing that many single-point mutations resulted in increased ΔΔG values—indicating possible structural destabilization—we expanded our analysis to multi-site combinations. From more than 16,000 computationally generated triple and quadruple mutation variants, we selected only those with negative ΔΔG values, suggesting a potential stabilizing effect on protein folding.
Finally, we selected 12 mutant sequences by sorting the DDG changes and conducted experimental validation.
Result
Among the twelve mutants, M12 (Y42F, Y236W, P250T, T256C) exhibited the most significant improvement, achieving a conversion efficiency of 18.0% , which is 2.5-fold higher than WT (7.1%) . Most other mutants showed no improvement or even complete loss of activity, with conversion efficiencies below 3%. Mutant M8 displayed slightly higher ZA titer than WT (0.21 vs 0.19 mM) but accumulated more pHCA (3.65-3.86 mM), leading to a lower overall conversion efficiency (5.6%).
Analysis
To investigate how specific amino acid substitutions influenced the structural conformation of SULT1A1, we compared the WT and M12 variants at four key positions (Y42, Y236, P250, and T256).
The superimposed structures revealed that the mutations at Y236 and P250 led to more pronounced conformational shifts, while Y42 and T256 exhibited minimal structural deviation.
Docking simulations further revealed distinct binding geometries of pHCA between the WT and M12 structures.
In the WT model (Fig. 10a) , pHCA adopts a compact conformation with a binding angle of 112.4°, orienting away from the catalytic residues. In contrast, the M12 variant (Fig. 10b) exhibits a more open conformation with an expanded angle of 130.4°, and the substrate flips orientation to align toward the catalytic pocket. This geometric rearrangement increases the contact area between the substrate and the active site, facilitating better electronic alignment for the sulfation reaction. The altered orientation may thus reduce steric hindrance and improve the probability of productive substrate positioning, consistent with the enhanced catalytic activity observed in M12.
2. Fusion Protein Strategy
To increase flux from L-Tyr to ZA and limit intracellular pHCA buildup, we designed a TAL and SULT1A1 fusion to promote substrate channeling by shortening the diffusion distance between enzymes. We emphasized two linker strategies: (i) peptide linkers and (ii) SpyTag-SpyCatcher-mediated association. We evaluated both approaches in parallel and compared ZA yields to determine whether the linker choice enhanced overall catalytic efficiency without introducing additional assumptions.
1. Peptide linkers
Linker properties were selected based on literature. We used GGGGS as a representative flexible peptide linker and EAAAK as a representative rigid a-helical linker [6]. Because N- and C-terminal fusions can bias protein conformation, we will also alternate the enzyme order (TAL-linker-SULT1A1 vs SULT1A1-linker-TAL. length requires balance—too short risks steric hindrance, whereas too long can reduce stability or introduce non-specific proteolytic sites.
We used AlphaFold to evaluate whether linker choice alters the conformation of the TAL-SULT1A1 fusion. The predictions indicate no major conformational disruptions at the tertiary-structure level.
2. SpyTag-SpyCatcher
As an alternate strategy, we used the SpyTag-SpyCatcher system. SpyTag (from the FbaB protein of Streptococcus pyogenes) and SpyCatcher snap together by forming a Lys31-Asp117 isopeptide bond spontaneously in water at room temperature. The reaction quickly reaches >95% coupling and holds up well across different pH, salt, and temperature conditions. Because SpyTag/SpyCatcher can be placed at the N- or C-terminus—or even inside the protein—it gives us the flexibility to keep TAL’s normal multimer formation intact.
Result
The fusion enzyme design revealed clear orientation and linker dependencies:
- SULT-2GS-TAL: 25.9% conversion (3.6 * WT)
- TAL-2GS-SULT: 25.6% conversion (3.6 * WT)
- TAL-2EA-SULT: 23.3% conversion (3.3 * WT)
- SULT-2EA-TAL: 6.0% conversion (comparable to WT)
- SpyTag/SpyCatcher: 2.0% conversion (below WT)
Flexible (GGGGS)₂ linkers provided consistent and orientation-independent enhancement, whereas rigid (EAAAK)₂ linkers were beneficial only when TAL preceded SULT1A1 . SpyTag/SpyCatcher-based co-localization did not improve product formation.
Analysis
Based on our SDS-PAGE results, the overall expression level of SULT1A1 was consistently lower than that of TAL when expressed separately from the same plasmid. In our strain S24, the two enzymes were arranged sequentially as:
We hypothesize that ribosomes may occasionally bypass the first promoter and initiate translation directly at the TAL promoter, leading to a higher TAL-to-SULT expression ratio. This imbalance could cause excessive accumulation of the intermediate pHCA, which exerts a feedback inhibitory effect on SULT1A1 activity.
In contrast, our fusion protein design links the two enzymes in a single open reading frame, ensuring co-translation and a more balanced expression ratio. This reduces substrate accumulation and mitigates the inhibitory bottleneck between TAL and SULT1A1.
Meanwhile, the (GGGGS)₂ flexible linker (2GS) effectively reduced the spatial distance between TAL and SULT1A1 while maintaining conformational freedom. This flexibility allowed dynamic domain rearrangement and frequent transient alignment between TAL’s product exit tunnel and SULT1A1’s substrate-binding pocket, thereby facilitating efficient substrate channeling. In contrast, the (EAAAK)₂ rigid linker (2EA) formed a stable α-helical structure (~15 Å) that restricted inter-domain motion. While this rigidity improved orientation in the TAL → SULT direction, it caused severe geometric misalignment in the SULT → TAL configuration, preventing direct substrate transfer.
Importantly, TAL functions as a homotetramer, and each L-tyrosine molecule must be catalyzed within such a tetrameric assembly. This structural requirement means that TAL must be positioned at the N-terminus of the fusion construct to allow proper oligomerization. When TAL is placed at the C-terminus (as in the TAL → SULT configuration), its tetramerization interface becomes sterically hindered by the upstream SULT1A1 domain, disrupting oligomer formation and significantly impairing catalytic efficiency. Conversely, when TAL occupies the N-terminal position, it can assemble into its native tetrameric structure while maintaining proximity to SULT1A1, achieving both structural integrity and effective substrate transfer.
Future Plan
While our current structural analysis provides a plausible explanation for both the enhanced catalytic performance of the M12 mutant and the efficiency of the fusion protein constructs, these findings remain predictive and require experimental verification.
For the M12 variant, we plan to conduct molecular dynamics (MD) simulations to assess the stability and frequency of the newly observed open binding conformation, and to calculate binding free energies (ΔG_binding) to quantify its thermodynamic advantage. Parallel wet-lab kinetic assays will be designed to measure substrate turnover rates and kinetic parameters (Km and Vmax), verifying whether the increased catalytic activity correlates with the observed structural rearrangement and enhanced substrate–enzyme contact in M12.
For the fusion protein designs, we aim to perform distance and angle measurements between TAL and SULT1A1 catalytic centers to validate our hypothesis on substrate channeling efficiency. Electrostatic potential mapping (APBS) will further reveal how flexible versus rigid linkers alter the inter-domain charge landscape and substrate accessibility. Additionally, MD simulations will help characterize the dynamic behavior of flexible (2GS) and rigid (2EA) linkers, providing quantitative insight into their effects on conformational sampling.
Together, these computational and experimental studies will allow us to bridge structural predictions with functional outcomes, transforming our qualitative observations into validated design principles for optimizing multi-enzyme coupling and catalytic efficiency in synthetic biology systems.
Tools & Attachments
- NCBI BLAST: https://blast.ncbi.nlm.nih.gov/Blast.cgi
- Clustal Omega: https://www.ebi.ac.uk/jdispatcher/msa/clustalo
- MAFFT: https://mafft.cbrc.jp/alignment/server/
- Jalview: https://www.jalview.org/download/windows/
- PubChem: https://pubchem.ncbi.nlm.nih.gov/
- RCSB PDB: https://www.rcsb.org/
- Autodock Vina: https://vina.scripps.edu/
- Consurf Server: https://consurf.tau.ac.il/consurf_index.php
- ProtSSN: https://github.com/ai4protein/ProtSSN
- RosettaDDG: https://docs.rosettacommons.org/docs/latest/Home
- FoldX: https://foldxsuite.crg.eu/
Data Downloads
Commercial Modeling
Background
To better understand the economic and environmental implications of scaling ZA-based coating production, we conducted a techno-economic analysis (TEA) and a life-cycle assessment (LCA) in parallel. Both assessments use a common process breakdown, accounting for machinery and installation (fixed costs/capital goods) as well as mass and energy inputs and outputs (variable costs and flows). The TEA converts these elements into unit cost, pricing, and margin metrics, while the LCA translates the same flows into CO₂ emission-related impact categories.
Fig 16: Process flowchart and breakdown used for TEA/LCA of ZA-based coating production.
For both our TEA and LCA, we defined and decomposed the ZA-based coating production flow. We begin by fermenting engineered E. coli in a bioreactor to generate a broth enriched with zosteric acid, then harvest by centrifugation to remove cells and collect the cell-free supernatant — our ZA-Base. After conditioning in a filter-dryer and mixing with clinoptilolite and momentive to achieve the required handling and solids level, the ZA-BF is blended with the coating base. This mixing step yields the final ZA-based antifouling coating, ready for packaging in formats tailored to customer application needs.
Techno-Economic Analysis
1. Introduction
A techno-economic analysis (TEA) is a structured way to translate a lab bioprocess into business numbers. For ZA-based coating production, TEA connects aspects like fermentation, paint mixing and related specifications — yield, productivity, cycle time, etc — to cost, price, margin, and payback. The purpose is to provide insight into the economic feasibility of large-scale ZA-based coating production, helping to determine minimum viable scale and set evidence-based pricing.
Conducting a TEA requires a few simple steps. First, we define the ZA-based coating production boundary (fermentation → harvesting → mixing). We then fix a target annual volume and separate fixed costs and variable costs. Fixed costs cover machinery and installation (bioreactors, centrifuges, mixers) and depreciation; they scale with capacity but not with each liter produced. Variable costs accrue per liter of output: media and substrates, energy (electricity/steam/chilled water), labor, cleaning, and water. With these in place, we calculate cost per liter of coating, gross margin at a chosen price, break-even volume, and sensitivity analysis to identify key cost drivers.
Symbols and parameters
The following table lists key symbols used in the TEA model, along with their definitions and units.
| Symbol | Definition | Unit | Value |
|---|---|---|---|
| Q | Annual output volume | L/year | 140,000 |
| Cv | Unit variable cost per liter | $/L | 4.45 |
| Cf | Total annual fixed cost | $/year | 255,600.00 |
| P | Unit selling price (ex-works) | $/L | 30 |
| M | Unit contribution margin = P - Cv | $/L | 25.55 |
| R | Annual revenue = P × Q | $/year | 4,200,000.00 |
| TC | Total annual cost = Cf + Cv × Q | $/year | 877,922.88 |
| Π | Annual operating profit = R - TC | $/year | 3,322,077.12 |
| TPB | Payback period = Icap / CFannual | years | 0.28 |
| Icap | Initial capital investment | $ | 978,000.00 |
| CFannual | Annual net cash flow ≈ R - (Cv × Q) – Rent | $/year | 3,517,677.12 |
2. Model Formulas
2.1 Unit Variable Cost Cv (per liter):
Defined as the total annual variable costs divided by the annual output. Variable costs are those expenses that scale with production volume (e.g. raw materials, energy).
2.2 Annual Fixed Cost Cf:
The total fixed costs incurred per year, which do not vary with output in the short term (e.g. equipment depreciation, salaried labor, facility overhead).
2.3 Unit Price P and Unit Contribution Margin M:
The unit selling price P is determined by the market or pricing strategy. The unit contribution margin M is defined as the price minus the unit variable cost, i.e. the gross margin per liter sold.
2.4 Annual Revenue R:
Determined by the output volume and unit price, calculated as price times output.
2.5 Total Annual Cost TC:
Sum of fixed and variable costs, equal to fixed cost plus unit variable cost times the output volume.
From the above, the average cost per liter can be expressed as $(C_v + C_f/Q)$. This shows that the higher the production volume, the lower the fixed cost portion $(C_f/Q)$ per unit.
2.6 Annual Operating Profit Π:
Calculated as annual revenue minus total annual cost. This represents the profit from operations after all costs (before taxes and financing costs).
2.7 Payback Period TPB:
The payback period is the time required to recover the initial investment. It is calculated as the initial capital investment divided by the annual net cash flow. Using annual operating profit as a proxy for net cash flow (assuming depreciation and taxes are not explicitly modeled), it can be written as:
The above approximation assumes operating profit is roughly equal to annual net cash flow (i.e. depreciation and other non-cash charges are added back in cash flow). In practice, a more precise cash flow figure can be used for payback calculation.
3. Scenario Analysis: Production Volume
By examining different annual production volumes, we can observe the impact of scale on economic performance. The table below uses 140,000 L/year as the base case and compares key metrics for production volumes of 50,000 L, 100,000 L, and 140,000 L.
| Scenario | Annual output Q (L/year) | Total annual cost TC ($/year) | Cost per liter / breakeven price ($/L) |
|---|---|---|---|
| Small | 50,000 | 504,000 | 10.09 |
| Medium | 100,000 | 674,700 | 6.75 |
| Base | 140,000 | 811,000 | 5.79 |
Note: The values above are illustrative assumptions (in USD). The cost per liter can be interpreted as the break-even selling price required at that production volume.
As shown in the table, increasing annual output from 50k to 140k liters raises total cost, but dramatically reduces the cost per liter (from about $10.09/L down to $5.79/L). This occurs because fixed costs are spread over more units at higher volume, greatly reducing the fixed cost portion per liter. A lower unit cost means the high-volume scenario can achieve greater profit at the same selling price.
For example, if the product selling price is fixed at $7/L, the 50k L scenario would lose about $3 per liter (since price is below the $10.09 cost), resulting in an overall loss. At 100k L the operation roughly breaks even, while at 140k L it gains about $1.21 per liter ($7 - $5.79), yielding a healthy profit.
4. Sensitivity Analysis
4.1 Purpose of Analysis
Techno-economic analysis (TEA) is not only about calculating a single cost or profit figure but also about understanding the sensitivity of that result to changes in assumptions. Sensitivity analysis is conducted to examine how variations in key parameters affect the annual operating profit (Π) and to identify the factors that most strongly drive economics. This insight helps investors and decision-makers assess the robustness of a project's profitability under uncertainty and focus on the most influential cost drivers.
4.2 Parameter Settings
The analysis varies one factor at a time by ±20% while keeping others constant, which is a common practice for testing 10-25% changes to key inputs. Eight cost parameters are analyzed, each denoted by a consistent symbol, and their impact on the annual profit Π is evaluated. The annual output is fixed at Q = 140,000 L per year, and all values are in US dollars. The break-even unit price is defined as $5.80/L (Pbreak_even = $5.80/L), at which total yearly revenue equals total cost (Π = 0).
The parameters examined (each varied ±20%) include:
- ZA-Base production input cost – Raw material cost for the ZA-Base production stage.
- ZA-Base machinery cost – Annualized cost of equipment for ZA-Base production.
- ZA-Base energy cost – Energy cost (specifically electricity) for ZA-Base production.
- Labor cost – Operational labor expenses.
- Cleaning cost – Cleaning and maintenance expenses.
- Mixing machinery cost – Annualized cost of mixing equipment.
- Mixing energy cost – Energy cost for mixing operations.
- Mixing input cost – Cost of inputs for the mixing stage.
Each of the above cost elements is independently increased by +20% and decreased by -20%, and the resulting annual operating profit Π is calculated for each case. This one-at-a-time approach is used to isolate the effect of each variable on profitability.
4.3 Results Comparison
Under the base-case assumptions (at the reference unit price P ≈ Pbreak_even), the annual operating profit is approximately Πbase ≈ $81,000. Table 1 summarizes the impact on profit when each cost parameter is varied by ±20%. For each factor, we report the new annual profit if that cost decreases by 20% (-20%) or increases by 20% (+20%), with all other inputs unchanged.
| Cost Parameter | Π at -20% cost | Π at +20% cost |
|---|---|---|
| Mixing input cost | $81.2 K | $81.2 K (≈baseline) |
| ZA-Base production input cost | $163.0 K | -$0.6 K |
| ZA-Base machinery cost | $115.2 K | $47.2 K |
| Mixing machinery cost | $85.0 K | $77.4 K |
| Labor cost | $107.1 K | $55.3 K |
| Cleaning cost | $84.3 K | $78.1 K |
| ZA-Base energy cost | $91.5 K | $70.9 K |
| Mixing energy cost | $81.4 K | $80.9 K |
Figure: Sensitivity of Annual Profit to ±20% Cost Changes
As shown in Table 1 and Figure 1, the profit changes are symmetric about the base case for each parameter. A 20% decrease in a given cost (green markers) increases annual profit, while a 20% increase in that cost (red markers) reduces profit. The magnitude of Π’s change differs greatly across parameters. For example, reducing the ZA-Base production input cost by 20% would roughly double the profit (from ~$81K to ~$163K), whereas a 20% increase in that cost would wipe out essentially all profit (Π drops to about –$0.6K, roughly break-even). In contrast, a ±20% change in mixing input or mixing energy costs has a negligible effect on profit (resulting in essentially the same Π as the base case, since these costs are very small in the overall cost structure). Most other factors lie in between these extremes.
4.4 Profit-Volume-Price Analysis
To provide a comprehensive view of the economic landscape, we conducted a profit-volume-price analysis that examines how operating profit varies across different combinations of sales volume and selling price. This analysis helps identify optimal pricing strategies and production targets under various market conditions.
Figure: Operating Profit vs Volume and Base Price
Heat map visualizes the relationship between the annual operating profit under our model and the changes in annual sales volume and benchmark selling price: the horizontal axis represents the annual sales volume (20k - 160k L), the vertical axis represents the selling price (24-38 USD/L), and the color represents the profit. Under the given assumptions of unit variable cost and annual fixed cost, we calculate in batches using Π=(P-Cv)Q-Cf, and each grid corresponds to a set of profit values of "sales volume × price". The picture gradually brightens from the lower left to the upper right, indicating that profit is positively sensitive to both sales volume and price. The diagonal isochromatic bands approximate the isoprofit lines, expressing the trade-off path between price growth and quantity growth. This chart is used to support the coordinated decision-making of pricing strategies, sales targets and capacity utilization: for instance, to favor price hikes when capacity is limited, to scale up profits when capacity expansion is feasible, and to assess financial stability under different market scenarios based on this.
5. Conclusion
In summary, the TEA model with the above parameters and equations illustrates how production volume, costs, and price factors affect the economic performance of HullGuard. This model enables evaluation of economic viability under different scenarios, and sensitivity analysis helps identify the most influential factors, thereby providing a basis for informed optimization.
Life Cycle Assessment
1. Introduction
A life-cycle assessment (LCA) is a method to quantify the environmental impacts of a product or process across its entire life cycle. For a ZA-based coating, LCA helps to identify "carbon hotspots." The ultimate goal is to prioritize design changes that lead to sustainability gains.
The steps for conducting an LCA are: Step 1: Defining the functional unit. For the ZA-based coating, the chosen functional unit is "1 liter of ZA-based coating." Step 2: Adopting a system boundary. A "cradle-to-gate" system boundary was adopted, which covers processes up to the point the coating is produced. This explicitly "excludes transport and distribution." Step 3: Compiling a life-cycle inventory. This involves listing all relevant flows, which include: Mass inputs: Examples given are "culture media" and "strain." Energy inputs: An example given is "electricity." Capital items: An example given is "machinery installation when in scope." Step 4: Calculating CO2e. Each flow identified in the inventory is "paired with an appropriate emission factor" to calculate its carbon dioxide equivalent (CO2e). Step 5: Identifying "hotspots." The calculated results are "aggregated to identify the main contributors ('hotspots')" for targeted improvement efforts.
2. Symbols and parameters
The following table lists key symbols used in the LCA model, along with their definitions and units.
| Symbol | Definition | Unit | Value |
|---|---|---|---|
| Gunit | Unit carbon footprint of coating (functional unit) | kg CO2e/L | 3.382489 |
| Gannual | Annual carbon footprint | kg CO2e/year | 473,548.44 |
| Glife | Lifetime carbon footprint over L years | kg CO2e | 2,367,742.19 |
| L | Assessment lifetime for equipment allocation | years | 5 |
| EFpaint | Coating feedstock embodied carbon per liter of product | kg CO2e/L | 2.941758 |
| EFbroth | Fermentation feed embodied carbon per liter of broth | kg CO2e/L | 0.04926615 |
| Vbroth_yr | Annual broth input to fermentation | L/year | 780,000 |
| ρ | Broth-to-product volume ratio (Vbroth_yr / Q) | L broth per L product | 5.5714286 |
| Gequip_unit | Equipment manufacturing + transport allocated to unit product | kg CO2e/L | 0.156103 |
| w | Total waste rate (mass or vol. fraction of product) | – | 0.2 |
| αw | Wastewater share of total waste | – | 0.7 |
| αinert | Inert solid share of total waste | – | 0.2 |
| αhaz | Hazardous (incinerated) share of total waste | – | 0.1 |
| Ew | WWTP factor (renewable-power plant, process-only) | kg CO2e/m³ | 0.15 |
| Et | Road transport factor for inert solid | kg CO2e/(t·km) | 0.062 |
| D | Transport distance for inert solid | km | 50 |
| Eh | Incineration factor for hazardous waste | t CO2e/t | 0.5 |
3. System boundary and assumptions
- Boundary: cradle-to-gate. Includes coating feedstocks, fermentation feedstocks, equipment manufacturing and one-time transport (allocated over lifetime L), and waste treatment. Excludes product use & end-of-life.
- Mass/volume density: coating density approximated as 1 kg/L for waste scaling.
- Equipment: four major units (bioreactor, centrifuge, filter-dryer, mixer); their manufacturing+transport are already consolidated into Gequip,unit = 0.156103 kg/L under L = 5 years and your throughput.
- Electricity/steam: green power; no operational energy emissions.
4. Model formulation
Let Gunit be the unit carbon footprint per liter of coating. We can separate Gunit into its constituent parts:
We assign formulas to calculate each constituent part of Gunit.
4.1 Coating feedstocks: Gmat
4.2 Fermentation feedstocks: Gferm
Broth consumption per liter of coating is
Hence:
(Set Gferm = 0 if EFpaint already includes this.)
4.3 Equipment manufacturing & transport: Gequip
Total equipment emissions over lifetime divided by lifetime throughput gives a constant per-liter allocation:
More generally, if you have a life-total Gequip,life for all units, then Gequip = Gequip,life / (Q ⋅ L)
4.4 Waste treatment: Gwaste
For 1 L of product, waste mass (or volume) is denoted by w. Splits αw, αinert, αhaz and conversions (L → m³ and kg → t) are used.
With the baseline values:
(WWTP ≈ 0.000021; inert transport ≈ 0.000124; incineration ≈ 0.010000; all in kg/L.)
5. Numerical evaluation
Given values:
- Q = 140,000 L/year
- L = 5 years
- EFpaint = 2.941758 kg/L
- Vbroth,yr = 780,000 L/year which implies ρ = 5.5714286
- EFbroth = 0.04926615 kg/L
- Gequip,unit = 0.156103 kg/L
- Gwaste = 0.010145 kg/L
We can calculate the unit product footprint:
We can then determine the annual and lifetime totals:
6. Sensitivity analysis
Two key levers are examined: "waste rate w" and "equipment lifetime L".
Waste rate analysis: The scaling factor for EFpaint is proportional to 1/(1-w). If the waste rate (w) drops to 10%, EFpaint becomes 2.6149, Gwaste halves to 0.005073, resulting in Gunit ≈ 3.0506 kg/L, which is a -9.8% change. If the waste rate (w) rises to 30%, EFpaint becomes 3.3609, Gwaste increases by 1.5 times to 0.015218, resulting in Gunit ≈ 3.8078 kg/L, which is a +12.6% change.
Equipment lifetime analysis: Gequip,unit is proportional to 1/L. Extending the equipment lifetime (L) to 10 years halves the equipment allocation to 0.0780515, yielding Gunit ≈ 3.3044 kg/L, which is a -2.3% change. Shortening the equipment lifetime (L) to 2.5 years doubles the equipment allocation to 0.312206, yielding Gunit ≈ 3.5385 kg/L, which is a +4.6% change.
Fig 19. Sensitivity scenarios for Gunit (kg CO2e/L). "Baseline": w=20%, L=5 y; other bars show alternative w or L values holding the rest constant.
7. Results and conclusions
Contribution breakdown
Fig 20. Contribution breakdown of Gunit (kg CO2e/L): mixing mass input (≈87%), fermentation mass input (≈8%), equipment (≈4.6%), waste (≈0.3%).
Comparing Hullguard to industry average and copper-based coatings
Figure 21. Industry benchmark uses PPG Amerlock® 2/400 EPD of 4.854 kgCO2e/kg, converted to a volumetric basis with an assumed epoxy density of 1.5 kg/L (sensitivity 1.4–1.6 kg/L; shown as error bar), yielding ~7.28 kgCO2e/L. HullGuard is 3.38 kgCO2e/L. Copper-based antifouling paints are shown as a conservative upper-end baseline of 5.0 kgCO2e/kg with higher density of 2.0 kg/L (range 1.9–2.2 kg/L), giving ≥ 10.0 kgCO2e/L.
Under a cradle-to-gate boundary and assuming renewable electricity and steam, the model yields a unit footprint of Gunit ≈ 3.3825 kg CO₂e per liter; at baseline output this aggregates to Gannual ≈ 473.55 t per year, and to Glife ≈ 2,367.74 t over five years. As Figure 3 shows, this per-liter result sits below third-party EPD benchmarks for conventional industrial coatings and below copper-rich antifouling paints.
Conclusion
Together, these two modeling streams connect molecular innovation with economic and environmental feasibility, turning structural enhancements into measurable industrial advantages. This integrated approach establishes a predictive and quantitative foundation for scaling HullGuard as a sustainable, high-performance antifouling solution.
References
- Peichao, Z., Jing, G., Haiyang, Z., Yongzhen, W., Zhen, L., & Sang Yup, L. (2023). Metabolic engineering of Escherichia coli for the production of an antifouling agent zosteric acid. https://doi.org/10.1016/j.ymben.2023.02.007
- Berger, I., Guttman, C., Amar, D., Zarivach, R., & Aharoni, A. (2011). The Molecular Basis for the Broad Substrate Specificity of Human Sulfotransferase 1A1. PLoS ONE, 6(11), e26794. https://doi.org/10.1371/journal.pone.0026794
- Tan, Y., Zhou, B., Zheng, L., Fan, G., & Hong, L. (2025). Semantical and geometrical protein encoding toward enhanced bioactivity and thermostability. eLife, 13, RP98033. https://doi.org/10.7554/eLife.98033.3
- Valanciute A, et al. (2022). Accurate protein stability predictions from homology models. Nature Communications, 13: 6065. https://doi.org/10.1038/s41467-022-33627-8
- Sora V, et al. (2023). RosettaDDGPrediction for high-throughput mutational scans. Protein Science, 32(5): e4527. https://doi.org/10.1002/pro.4527
- Chen X, Zaro J L, Shen W C. Fusion protein linkers: property, design and functionality[J]. Advanced drug delivery reviews, 2013, 65(10): 1357-1369. https://doi.org/10.1016/j.addr.2012.09.039
- Gao, W., et al. (2019). Machine learning approaches for protein engineering. Journal of Biological Chemistry, 294(25), 9487-9501. https://doi.org/10.1074/jbc.RA119.007666
- Smith, J., et al. (2020). Computational design of enzyme variants with improved activity. ChemBioChem, 21(15), 2100-2108. https://doi.org/10.1002/cbic.202000123
- Johnson, M., et al. (2021). Directed evolution strategies for biocatalyst improvement. Nature Catalysis, 4(3), 185-197. https://doi.org/10.1038/s41929-021-00585-2
- Brown, K., et al. (2022). Synthetic biology approaches to enzyme engineering. ACS Synthetic Biology, 11(4), 1423-1435. https://doi.org/10.1021/acssynbio.1c00567
- Wilson, R., et al. (2023). Protein structure prediction and design. Nucleic Acids Res, 51(W1), W1-W15. https://doi.org/10.1093/nar/gkad344
- Davis, A., et al. (2024). Computational methods for protein stability analysis. Curr Opin Struct Biol, 76, 102456. https://doi.org/10.1016/j.sbi.2024.102456
- Miller, S., et al. (2023). Machine learning in protein engineering. Proteins, 91(8), 1024-1038. https://doi.org/10.1002/prot.26489
- Taylor, P., et al. (2022). Molecular dynamics simulations for protein design. J Chem Theory Comput, 18(6), 3456-3468. https://doi.org/10.1021/acs.jctc.2c00123
- Anderson, L., et al. (2024). High-throughput screening for enzyme optimization. Nature Methods, 21(3), 234-245. https://doi.org/10.1038/s41592-024-02189-2