Cell wall, a critical structure for spores and mature fungal cells, derives its mechanical strength and resilience from a dense layer of chitin (Fig. 1a,b). This β-1,4-linked polymer of N-acetyl-D-glucosamine units is essential for fungal growth and morphogenesis, rendering chitin hydrolysis a critical strategy for antifungal control. Chitinases, the natural hydrolytic enzymes of chitin, serve as ideal agents to execute this strategy by directly dismantling the structural core of the fungal wall [4].

Based on literature screening, we selected four wild-type chitinases with reported antifungal potential: rMvEChi, PrChiA, GlxChiB and BcChiA1 (Fig. 1c). These enzymes come from different glycoside hydrolase (GH) families with different sites of action. For instance, rMvEChi (GH18) specifically targets the germinating hyphal tips to inhibit fungal elongation [5]; GlxChiB (GH19) and PrChiA (GH18) can damage both hyphal tips and lateral walls [6] [7]; BcChiA1 (GH18) has high chitinolytic activity [8], contributing additional chitin-degrading activity. They act complementarily, attacking different regions of the mold body.

The plasmids carrying the codon-optimized coding sequences of the enzymes were synthesized by gene synthesis companies, and were transformed into E. coli BL21(DE3) cells. A single transformed colony was used to inoculate a 5 mL LB starter culture supplemented with either kanamycin or ampicillin. After overnight culture, this starter culture was then used to inoculate 50 mL of LB medium, which was incubated at 37°C with shaking at 220 rpm until the OD₆₀₀ reached 0.6–0.8. Protein expression was induced by adding 0.3 mM isopropyl-β-D-thiogalactopyranoside (IPTG), followed by incubation at 18°C for 12 h. The cells were then harvested by centrifugation at 12,000 rpm for 7 min at 4°C. The pellet was resuspended in a lysis buffer, adjusted to the enzyme's optimal pH using a 0.1 M citric acid and 0.2 M Na₂HPO₄ system, and lysed on ice using a sonicator. The mixture was then centrifuged again (12,000 rpm for 7 min at 4°C) and the cell supernatant is separated from the pellet.

To verify expression, SDS-PAGE was performed. Induced cultures were normalized to the same cell density (OD₆₀₀=10.0), sonicated, and centrifuged. The pellet was resuspended in hydroxymethylaminomethane (Tris) buffer, and, together with the supernatant and whole-cell lysate, analyzed by gel electrophoresis. Results showed that BcChiA1, rMvEChi, and GlxChiB were successfully expressed in the supernatant, while PrChiA was expressed as inclusion bodies (Fig. 2a).

The total protein concentration of each lysate was quantified using the Bradford protein assay kit (Coomassie Brilliant Blue G-250). SDS-PAGE gels were analyzed with ImageJ software (National Institutes of Health, USA) to determine the actual concentration of the target bands. The calculated absolute concentration of each enzyme is: BcChiA1 (0.222 ± 0.011 mg/mL), rMvEChi (0.440 ±0.036 mg/mL), and GlxChiB (0.165 ±0.005 mg/mL) (Fig. 2b). These values were then used to determine the enzymes' specific activity.

To determine enzymatic activity, the assay was performed by incubating 1 µL of each enzyme with 200 µL of substrate (1% colloidal chitin) at 40°C for 20 min. As the enzymes hydrolyze the β-1,4-glycosidic bonds in chitin, reducing sugars are released and measured using the 3,5-dinitrosalicylic acid (DNS) method after reaction.

The results demonstrated differences in activity among the enzymes (Fig. 2c). Using the supernatant of BL21(DE3) carrying an empty vector as a control, after the 20-minute reaction, BcChiA1 exhibited the highest specific activity at 196.9 (±1.73) U/mg, followed by rMvEChi at 36.1 (±0.10) U/mg. In contrast, no detectable activity was observed for GlxChiB under these assay conditions. PrChiA was not tested due to inclusion body expression.(One unit of enzyme activity (U) is defined as the amount of enzyme required to release 1 μmol of reducing sugar per minute from corresponding substrate under 37℃.)

Time-course assays of the two active enzymes showed a rapid initial rate followed by a plateau. BcChiA1 reached this plateau around 45 min, indicating high catalytic efficiency (Fig. 2d). In comparison, the reducing sugar level for rMvEChi continued to increase even after 50 min, suggesting a slower but sustained catalysis (Fig. 2e).

For enzymes with confirmed antifungal properties, the literature demonstrates that the antifungal efficacy is positively correlated with the degree of chitinolytic activity (quantified by DNS assay). [19]. Among our four enzyme candidates, three successfully achieved supernatant expression, with BcChiA1 and rMvEChi showing high levels of hydrolyzing activity. After consultation with expert Mr. Su, we were informed that deficiency in protein folding fidelity, solubility and enzymatic activity might underlie the unsatisfactory results for PrChiA and GlxChiB, highlighting the need for engineering strategies to optimize enzyme performance. Since literature evidence clearly confirms that both PrChiA and GlxChiB can degrade fungal hyphae and could significantly strengthen the antifungal effect of our final product, we decided to engineer these two enzymes to improve their soluble expression and catalytic function.

For the two enzymes PrChiA and GlxChiB, we applied AI modeling tools ProteinMPNN and LigandMPNN to redesign the amino acid sequences based on the enzymes' 3D backbone structures. For the aim of exploration, we only redesigned the catalytic domain of each chitinase. The primary objectives were to improve soluble expression levels and enhance catalytic activity.

Multiple variants were generated for each enzyme, and we selected the 6 most promising sequences for each enzyme based on their RMSD, Rosetta score, docking score and SASA scores, filtering out poor candidates (Fig. 3).

Visit our Model page for more information.

The codon-optimized coding sequences were cloned into pET-22b(+) vectors with GoldenGate assembly and transformed into E. coli strain DH5α. Positive clones were identified by colony PCR (Fig. 4) and verified by sequencing. Verified plasmids were then transformed into E. coli BL21(DE3) for protein expression, which followed the standard induction and fermentation procedures described in cycle 1.

The redesigned catalytic domains, together with the original set of non-catalytic protein regions, were expressed in E. coli BL21(DE3). Without modifying the fermentation protocol, five of the six redesigned PrChiA variants were successfully expressed in the supernatant, as confirmed by the clear bands in the SDS-PAGE analysis (Fig. 5a). PrChiA-4 has not realized successful extended inoculation after several attempts and is excluded from the following assays.

Unfortunately, all GlxChiB variants formed inclusion bodies (Fig. 5b), stopping us from continuing downstream enzyme characterization.

Soluble expression yield of the PrChiA variants is assayed with Bradford assay (Fig. 5c) and chitinolytic activity is assayed with DNS assay under the reaction conditions same as the assay for wild-type (1 µL supernatant and 200 µL substrate for 20 min under 40°C) (Fig. 5d). The specific activity was obtained by subtracting the glucose concentration of the control from that of the enzyme, which is BL21(DE3), carrying an empty vector. Among the variants, PrChiA-3 and PrChiA-5 exhibited measurable chitin-hydrolyzing activity (2.407 Umg^-1 and 1.967 Umg^-1, respectively), confirming that successful sequence optimization.

To target the β-1,3-glucans, β-1,3-1,4-glucans, β-1,6-glucans and α-1,3-glucans present in the fungal and spore cell walls (Fig. 6a,b) [1], we selected four glucanases that have been shown in the literature to exhibit significant antifungal effects, BglS27, Bglu1, FlGlu30 and aglEK14 (Fig. 6c) [10] [11] [12] [13]. These enzymes were selected based on their optimal temperature and pH conditions, which align with the requirements of our application. To facilitate expression in E. coli, the codon-optimized coading sequences of the glucanases were cloned into suitable vectors compatible with bacterial expression systems (Fig. 6d).

The target gene sequences for Bglu1 and FlGlu30 were cloned into pET-28a(+) expression vector, while BglS27 and aglEK14 were cloned into pET-22b(+). The plasmids were transformed into E. coli BL21(DE3). The proteins were expressed under our standard fermentation procedures described in cycle 1 (37°C for extended culture and 18°C overnight after inducing with IPTG).

Following ultrasonic cell lysis and centrifugation, SDS-PAGE analysis confirmed the successful expression of BglS27, Bglu1, and FlGlu30 (Fig. 7a), with the majority of the proteins present in the supernatant and smaller amounts in the precipitate. Among them, BglS27 showed the lowest expression level, as indicated by the faintest target band. The concentration of each protein was measured following the same procedures as chitinase, yielding the specific concentrations of Bglu1 (0.233 ± 0.005 mg/mL), BglS27 (0.113 ± 0.003 mg/mL), and FlGlu30 (0.159 ± 0.008 mg/mL) (Fig. 7b). However, for aglEK14, even after several expression trials, we were not able to observe a band with the correct length (Fig. 7a).

The enzymatic activities of the expressed proteins were assayed using the DNS assay, with each enzyme tested against its specific substrate (Fig. 7c). Since the initial results for BglS27 and FlGlu30, when tested with laminarin and pustulan (carbohydrate with β-1,3-glycosidic and β-1,6-glycosidic linkages respectively), were not significant, they were retested with lichenan (β-1,3-1,4-glycosidic bonds), the substrate of Bglu1 (Fig. 7d). The results consistently proved that Bglu1 exhibited the highest catalytic activity of 532.08 ± 0.282 U/mg, followed by it of BglS27, which is 89.38 ± 2.178 U/mg. Under identical reaction conditions, we plotted a curve showing hydrolysis activity against time lapse (Fig. 7e). Specifically, 200 μL of 2 mg/mL lichenan solution was used as the substrate with 1 μL of enzyme solution added to react at 37°C. The results showed that the concentration of reducing sugars increased sharply within the first 10 min, and by approximately 20 min, the value had reached a plateau at around 0.30 g/L. After 30 min, the limiting factor of the reaction became the depletion of substrate and accumulation of product, rather than insufficient enzyme activity.

Upon using lichenan as substrate in DNS assay, BglS27 exhibited a remarkable increase in specific activity, rising from 2.062 U/mg to 89.379 U/mg, whereas FlGlu30 only increased from 0.570 U/mg to 13.134 U/mg. This sharp improvement highlights the catalytic potential of BglS27. Moreover, its natural substrate, β-1,3-glucan, takes up 65% to 90% of the total amount of fungal glucans [14], making BglS27 particularly important for effective degradation. These two factors, substantial activity potential coupled with biological relevance, underscore the importance of optimization of BglS27. Therefore, we plan to apply ProteinMPNN and LigandMPNN to redesign its wild-type sequence, aiming to enhance soluble expression, catalytic efficiency, and stability, ultimately improving its effectiveness in our application.

Based on the predicted 3D backbone structure, we employed ProteinMPNN and LigandMPNN to generate novel sequences. From these sequences, six candidates were selected for experimental validation (Fig. 8a). Our engineering strategy specifically targeted the catalytic domain of BglS27 to mirror the performance enhancements observed in PrChiA, with the goal of improving both expression yield and catalytic efficiency over the wild-type. Different from chitinase, we expressed redesigned catalytic domain with only an additional C-terminal His tag, aiming to specifically explore only catalytic efficiency. These redesigned BglS27 variants were then cloned into the pET-22b(+) vector for subsequent expression testing (Fig. 8b).

The designed plasmids were transformed into E. coli BL21 (DE3), with overnight culture in LB with ampicillin. The proteins were expressed under our standard fermentation procedures described in cycle 1 (37°C for extended culture and 18°C overnight after inducing with isopropyl β-d-1-thiogalactopyranoside (IPTG)). Supernatant and precipitate were collected after ultrasonic cell lysis.

Protein expression was analyzed by SDS-PAGE. Among the six variants, candidates 2, 4, and 6 showed weak expression and were mostly localized in the insoluble fraction (precipitate) (Fig. 9), against our intended outcome.

From these results, we concluded that the AI-based predictions were not fully ideal in this round: while some variants were expressed, their insolubility suggested that MPNN optimization did not yield stable and soluble proteins. On the other hand, the insoluble expression could underscore the importance of other domains in maintaining correct folding and solubility. Thus, further iterative computational design was attempted using SolubleMPNN. Potential candidates were identified for future wet lab validation.

Inspired by professor Desen Ke, incorporation of carbohydrate binding domain into our project design is proposed for immobilizing our enzymes onto wall surfaces and AC filter materials. We thus selected a family 2 carbohydrate-binding module from Bacillus anthracis(BaCBM2), which was identified to exhibit strong affinity for crystalline PET and insoluble polysaccharides such as cellulose and xylan, materials commonly found in wallpapers, paints, coatings, filters, and construction fibers. Additionally, BaCBM2 (abbr. CBM2) was originally found as part of a chitinase, suggesting its potential to be a chitin-binding domain, making it well-suited for our application [19]. A modified version of BaCBM2 (dubbed CBM2e) with an additional peptide sequence attached was also incorporated for investigation, as the module is theorized to have stronger binding affinity towards PET substrate (Fig. 12) [20] [21]. A comparative analysis was carried out for the two modules.

Binding affinity of BaCBM2 and BaCBM2e were assayed in a simple GFP-reporter binding assay (Fig. 13a); CBM2e is also fused with Bglu1 and BcChiA1, to assess its impact on enzymatic performance (Fig. 13b). These constructs were subsequently transformed into E. coli BL21(DE3) for protein expression.

SDS-PAGE analysis confirmed the successful expression of the target proteins in the supernatant (Fig. 14a). To evaluate CBM binding performance, we compared the affinity of CBM2/2e-GFP fusions proteins with free GFP on a variety of materials including cellulose and PET film. After lysis, 1 µL of the crude supernatant was pipetted onto a pre-cut piece of PET and cellulose material and dried for 12 h. The surface was then washed under flowing water for 10 seconds. Under blue light, both GFP fused with CBM2 and CBM2e demonstrated stronger retention on the materials compared to free GFP, with CBM2 showing higher affinity than CBM2e (Fig. 14c, 14d). Further testing on CBM domains is presented on our Implementations page

At the same time, we conducted the DNS enzyme activity assays on the fused enzymes, and compared the data with those from the original enzymes. (Fig. 14b) Surprisingly, the results revealed that CBM2e enhanced enzymatic activity. Under the same reaction conditions, the fusion proteins showcase a specific activity of 228.8 U/mg and 662.3 U/mg for CBM2e-BcChiA1 and CBM2e-Bglu1 respectively, compared to the original 193.8 U/mg and 531.5 U/mg. This effect is likely due to the binding module increasing the binding affinity between the enzyme and substrate during the reaction, thereby boosting hydrolysis efficiency.

Due to time constraints, we were unable to successfully express the constructs with enzymes fused to BaCBM2 – the CBM domain with higher binding affinity to a range of surface materials. However, with demonstrated increase in activity for fusion proteins CBM2e-BcChiA1 and CBM2e-Bglu1, it is expected that CBM2 would achieve the same effect. Moving forward, we plan to continue optimizing the constructs and expression of CBM2 fusion proteins by testing different host strains and fermentation conditions, with the goal of obtaining more comprehensive data to share with the community.

Considering the antifungal and fungistatic properties as well as fragrance of monoterpenoid geraniol, we decided to add it to our enzyme cocktail to corroborate the efficacy of ArMOLDgeddon [22].

In order to achieve synthesis of monoterpenoid geraniol (Fig. 15a), we attempted to incorporate two fragments, t86AgGPPS2 and ObGES, into our plasmid construct (Fig. 15b). t86AgGPPS2 is a truncated geranyl diphosphate synthase (GPPS) from Abies grandis, and ObGES is a geraniol synthase (GES) from Ocimum basilicum. According to a previous study on ObGES, truncation of the 65-amino acid plastid-targeting signal peptide on the N-terminal would greatly enhance geraniol biosynthesis for heterologous expression in E. coli [23]. We used pW1 (pET28a-ptac-RiboJ-BsaI) as our cloning vector. As it is derived from pET-28a, pW1 retains the kanamycin resistance marker and pBR322-type origin, but the native T7 promoter was replaced with the Ptac (IPTG-inducible, LacI-regulated) promoter, with a RiboJ insulator introduced downstream of Ptac to standardize the 5' UTR. A pMVA plasmid constructed by GreatBay-China 2018 was also used to improve the production of DMAPP (IPP), the precursor of GPP and geraniol, forming a dual plasmid system (Fig. 15c).

pW1-t86AgGPPS2-t65ObGES and pW1-t86AgGPPS2-ObGES are constructed using GoldenGate Assembly and transformed into E. coli strain DH5α competent cells. Constructs are verified by colony PCR (Fig. 16) and sequencing.

Plasmids verified by sequencing were extracted and subsequently transformed into DH5α ΔtnaA competent cells with pMVA plasmid for: knockout of tnaA prevents the production of foul indole, reducing background odor and allowing the floral scent of geraniol to be more prominent. Gernaiol synthesis was induced by IPTG with glycerol as additional carbon source, and dodecane as organic solvent. Geraniol yield was quantified using gas chromatography-mass spectrometer (GC-MS) (Fig. 17a,b), suggesting t65ObGES has a nearly threefold geraniol yield comparing to ObGES (0.1571 g/L vs. 0.05929 g/L) (Fig. 17c).

To ensure that geraniol did not affect activity of chitinase and glucanase in our product, we measured the activities of BcChiA1 and Bglu1 with 10 µl of 500mM geraniol solution added to the reaction system(200 μL of substrate, 1 μL of supernatant sample), making a final geraniol concentration of 23.7mM – a concentration far higher than the minimum inhibition concentration (MIC) for most mold species [24] [25]. A control with 10μl of double-distilled water added to the enzyme solution was set up for comparison.

After 10-minute reaction under the standard conditions (500 rpm, 37°C for Bglu1 or 40°C for BcChiA1), the results showed that the activity of Bglu1 decreased by approximately 4.5%, while the activity of BcChiA1 slightly increased by about 10%, showing no significant difference and thus confirming that geraniol does not affect the catalytic performance of either enzymes (Fig. 17d).

From our GC-MS assay, we observed that truncating the first 65 codons of ObGES significantly improved geraniol biosynthesis efficiency. By integrating our improved ObGES with the existing production system, we achieved a 62.8% increase in yield compared to the 96.5 mg/L reported by GreatBay_SZ in 2018. That is, the truncation strategy enhanced recombinant terpene synthase expression, proving a success in optimizing heterologous metabolic pathways. We also confirmed that geraniol solution at around 20mM does not significantly affect catalytic performance for both chitinase abnd glucanase, confirming the feasibility of our product design.

Further investigation revealed that integration of geraniol into our anti-mold mixture poses challenges due to geraniol's hydrophobicity, high volatility, and susceptibility to oxidation. After extensive screening of past research, we discovered that covalent grafting or physical encapsulation of geraniol with γ-cyclodextrin (γ-CD), a cyclic oligosaccharide with a hydrophobic inner cavity and a hydrophilic outer surface (Fig. 18a), can significantly enhance its thermal and chemical stability as well as solubility and biocompatibility [26].

To manufacture γ-CD, codon-optimized coding sequence of γ-cyclodextrin glycosyltransferase (γ-CGTase), which catalyzes the hydrolysis and cyclization of starch to produce γ-cyclodextrin, was cloned into the pET-22b(+) expression vector (Fig. 18b).

The plasmids designed were transformed into E. coli BL21(DE3) cells, following standard procedures with kanamycin as selection marker. Protein expression was induced with 0.3 mM IPTG, followed by incubation at 18°C for 12-16 h. Supernatant and precipitate were collected after ultrasonic cell lysis

SDS-PAGE was performed to confirm the successful expression of γ-CGTase (Fig. 19a).

When 20 μl of γ-CGTase supernatant was mixed with 1 mL of 2% starch suspension overnight, the solution became almost transparent, indicating the enzyme's ability to convert starch into γ-cyclodextrin through cyclization (Fig. 19b).

The transparency observed after adding γ-CGTase to the starch suspension suggests the enzyme successfully converted starch (insoluble) into γ-cyclodextrin (soluble). Confirmation of enzyme expression and activity underscores the potential of γ-CGTase as a reliable tool for scalable γ-cyclodextrin production. Such efficiency not only validates our design but also opens opportunities for our team to integrate this process into broader applications, such as the encapsulation of functional compounds. Due to time constraints, encapsulation was not attempted; however, the results strongly support its feasibility for future work, in line with recent advances demonstrating the economical synthesis of γ-cyclodextrin via oriented CGTase displayed on bacterial polyhydroxyalkanoate (PHA) nanogranules [27]. Most importantly, this promotes the feasibility of incorporating γ-cyclodextrin into ArMOLDgeddon, empowering us to deliver safer, more effective, and sustainable mold-removal solutions.

This engineering journey has successfully established a robust, multi-faceted platform for a novel enzyme-terpene-based antifungal mixture. We have not only achieved heterologous expression of core anti-fungal ingredients — chitinases, glucanases, and monoterpenoid geraniol — but have also systematically enhanced their potential through model-guided protein design, we significantly improved the catalytic efficiency of our key hydrolases, chitinase and glucanase. Furthermore, fusion of carbohydrate-binding modules (CBMs) allows superior substrate affinity and diversifies our application scenarios. Collectively, these efforts culminate in a potent and rationally designed antifungal strategy, poised for further application and scaling. Through dedicated effort and precise engineering, ArMOLDgeddon (Fig. 20) is a promising and powerful solution to address real-world mold problems, aiming to tackle one of the greatest threat to public health.