Results

Adressing PFAS degradation

Per- and polyfluoroalkyl substances (PFAS) are fluorinated compounds widely used in industry, including in non-stick cookware, firefighting foams, and food packaging. They are highly valued for their exceptional stability, resistance, and unique amphiphilic properties (both lipophilic and hydrophilic). However, these same properties make PFAS extremely persistent and chemically stable, earning them the nickname “forever chemicals.”

Because of their useful characteristics, PFAS have spread into countless industrial applications, making them ubiquitous in the environment. Some compounds, such as PFOA and PFOS, have been extensively studied. Recent findings from the International Agency for Research on Cancer (IARC) in Lyon classified PFOA as “carcinogenic to humans” (Group 1) and PFOS as “possibly carcinogenic to humans” (Group 2B) (Zahm et al., 2024). Such studies have driven stricter regulations. For example, the European Directive 2020/2184 requires monitoring of 20 PFAS compounds in drinking water and sets a maximum concentration of 0.10 µg/L for their combined levels. Current legislation is progressively restricting industrial use of PFAS, with discussions to expand bans to emerging compounds such as trifluoroacetic acid (TFA), which is not yet regulated as evaluations by WHO and ANSES are still ongoing.

Even as the use of regulated PFAS declines, new alternatives continue to appear until they, too, fall under regulation. This continuous cycle leads to the accumulation of PFAS in the environment, posing significant health risks. Studies have linked even low concentrations of PFAS to cancer, endocrine disruption, and reproductive toxicity (Richard et al, 2021).

Among PFAS, short-chain compounds such as TFA are particularly challenging. TFA can form as a product of abiotic degradation of longer-chain PFAS (Jeschke, 2017), or from pesticide degradation (EFSA, 2014). Due to its high polarity and persistence, removing TFA from water is nearly impossible except through reverse osmosis, which is prohibitively expensive. Moreover, while reverse osmosis can concentrate TFA, treating the concentrate itself remains problematic. The main disposal method is incineration, but studies have shown that PFAS incineration (e.g., polytetrafluoroethylene at 500 °C, Ellis et al., 2003) can actually produce TFA. Complete mineralization of TFA requires temperatures above 1400 °C, far beyond the capacity of most facilities. Consequently, incineration is not a viable solution for TFA destruction, prompting exploration of biological and enzymatic alternatives.

We began by reviewing literature on microbial PFAS degradation. Several bacterial strains show potential to interact with fluorinated compounds. For example, Pseudomonas strain PS27 can bioaccumulate 30–40% of PFHxS within 7 days, although without degradation (Presentato et al, 2020). Other strains demonstrate defluorination activity, such as Acidimicrobium sp. strain A6, which slowly reduces PFOA concentrations, with fluoride release suggesting defluorination (Huang et al., 2019, Fig. 1). Similarly, Labrys portucalensis F11 can decrease PFOS concentrations from 10 mg/mL to nearly zero over 194 days, linked to defluorination activity (Wijayahena et al., 2025).

Figure 1 : Results of PFOA 0.24 mM (100 mg/L) incubations with pure A6 and A6 enrichment cultures (Figure taken from Huang et al., 2019).

However, these processes often generate shorter-chain PFAS metabolites (PFHpS, PFHxS, PFHxA, PFPeA, PFBA, PFPrA), effectively shifting the problem to other persistent and potentially toxic compounds. This highlights the urgent need for novel synthetic biology approaches capable of achieving complete PFAS mineralization, rather than partial transformation.

Figure 2 : Plot showing the decrease in PFOS and the corresponding increase in fluoride ions (A), the formation of shorter-chain PFAS metabolites (PFHpS, PFHxS, PFHxA, PFPeA, PFBA and PFPrA) from PFOS biodegradation (B) (Figure taken from Wijayahena et al., 2025).

In the scientific literature on bioremediation, directed evolution has often been considered as a strategy to adapt bacteria so that a pollutant could serve as an energetic substrate. The energy potential of a substrate lies in its ability to donate electrons, and the difference in redox potential between the electron donor and the acceptor determines the amount of energy released.

However, fluorinated substrates present a particular challenge. Fluorine is the most electronegative element in organic chemistry. As a result, a defluorination reaction does not release electrons: instead, the electron remains tightly bound to fluorine, forming a fluoride ion (F⁻). For this reason, our initial consideration of a directed evolution approach appeared unlikely to succeed.

Bacterial degradation also has significant drawbacks. These include the long timescales required for most reactions and the lack of control over degradation products, which may be as toxic or even more toxic than the original compounds. This led us to prioritize an enzymatic approach, which offers better control over PFAS transformation, improved characterization of the reactions, and faster processing times.

The literature highlights several enzymes capable of breaking C–F bonds. As noted by (Wackett, 2024) or (Harris et al., 2024), such enzymes like laccases, peroxydase and dehalogenases have demonstrated activity in catalyzing defluorination of PFAS moieties, providing a promising starting point for engineering more efficient systems :

After further analysis of the available mechanisms, dehalogenases appeared to be the most promising and straightforward option for defluorination. These enzymes can defluorinate compounds containing one or two fluorine atoms, such as fluoroacetate and difluoroacetate, which are structurally very close to TFA. Among them, haloacid dehalogenases (HADs) seemed the most likely to act on TFA. These enzymes are extremely well characterized (Khusnutdinova et al., 2023, Farajollahi et al., 2024), making them attractive candidates for engineering.

We decided to focus on two haloacid dehalogenases :

RPA1163 (Uniprot: Q6NAM1), which efficiently defluorinates fluoroacetate with a defluorination activity of ~6 µmol/mg of protein after 2-hour reaction (Fig. 3). Its crystal structure has been resolved to high quality (1.05Å for 3R41), providing a strong basis for rational design.
DAR3835 (Uniprot: Q479B8), which demonstrates strong activity against difluoroacetate (Fig. 3, (Khusnutdinova, 2023)).

Figure 3. Enzymatic activity after 2 hours of absorbance monitoring at 540 nm in the presence of 5 mM fluoroacetate and 20 µg of purified enzyme. Figure adapted from Khusnutdinova (2023).

These enzymes share the same catalytic mechanism. The reaction involves a conserved Asp–His–Asp catalytic triad (Fig. 4, Chan & al, 2011). The aspartate residue performs an SN2 nucleophilic attack, displacing fluoride. The resulting covalent intermediate is hydrolyzed by a water molecule, activated by the histidine base, with assistance from the second aspartate residue.

Figure 4: Proposed two-step reaction mechanism of fluoroacetate dehalogenase by Chan & al, 2011. “Catalysis involves the conserved aspartate-histidine-aspartate catalytic triad. The reaction cycles from the free enzyme, Michaelis complex (I), covalent ester intermediate (II), enzyme-product complex (III), and then back to the free enzyme. First, fluoride is displaced by the aspartate nucleophile in an SN2 attack. The resulting covalent intermediate is then hydrolyzed by a water molecule activated by the histidine base. This step is assisted by the second aspartate residue.” from Chan & al, 2011.

Other enzyme classes also attracted our attention. For example, horseradish peroxidase (HRP) catalyzes the oxidation of phenolic substrates in the presence of H₂O₂, generating highly reactive radical intermediates capable of promoting secondary reactions with a broad range of organic molecules. Colosi et al. (2009) reported that HRP could degrade PFOA in the presence of 4-methoxyphenol (a phenolic co-substrate), producing smaller molecules. However, most degradation products remain fluorinated, and there is no evidence that HRP can directly cleave C–F bonds. To broaden the range of fluorinated substrates we are considering, we selected one of the fluorinated products formed during HRP-catalyzed PFOA degradation—1,1,1-trifluoro-2-butanone— as a candidate substrate for testing with our enzymes.

Reductive dehalogenases are enzymes that remain relatively underexplored, and their reaction mechanisms are not yet fully understood (Hu & Scott, 2024). To date, several reductive dehalogenases have been shown to reduce organochlorine, organobromine, and organoiodine compounds; however, there is currently no direct evidence of reductive defluorination. Nevertheless, a detailed thermodynamic study suggests that defluorination could be feasible (Parsons et al., 2008). In addition, computational simulations indicate that the enzyme T7RdhA from Acidimicrobium—a bacterium capable of slight PFOA and PFOS defluorination—may interact with PFOA. This makes it an excellent candidate for explaining the bacterium’s ability to attack PFAS (Huang & Jaffé, 2019). This is one of the enzymes we planned to test as an exploratory work to assess its defluorination capacity.

Finally, laccases were also considered. These copper-containing enzymes transfer a single electron from a mediator molecule, producing a radical capable of oxidizing other compounds while oxygen is reduced to water. Luo et al. (2018) showed that laccases, supplemented with the co-substrate 1-hydroxybenzotriazole, could degrade ~50% of PFOA over 157 days. The degradation products were mainly shorter-chain alcohols and aldehydes, still partially fluorinated. Because of the long timescales required and the limited efficiency, we decided not to pursue laccases further.

As mentioned before, the TFA concentration has been increasing over the last few years. Finding a way to remove the smallest PFAS from the environment becomes extremely urgent. Dehalogenases such as RPA1163 and DAR3835 can catalyse the defluorination of fluoroacetate (1 fluoride atom) and difluoroacetate (2 fluoride atoms) (Khusnutdinova et al., 2023; Farallajolahi & al, 2024). But when addressed to trifluoroacetate, no defluorination is observed (Farallajolahi & al, 2024, Fig. 5). However, docking studies performed by Farallajolahi & al (2024) indicate the same active site orientation for fluoroacetate, difluoroacetate, and trifluoroacetate, but the more the compounds are fluorinated less it is catalysed, suggesting that progressive fluorination has a negative impact on defluorination rate.

Figure 5 : “Binding of MFA (A), DFA (B), and TFA (C) in DeHa4. Structures are constructed by AutoDock Smina. The active site residues of the protein are shown in sticks, and the ligand (MFA) is shown in sticks and spheres, with C in cyan, O in red, N in blue, and F in pink.” from Farallajolahi & al, 2024.

To overcome this barrier we discussed with researchers specialized in fluorine chemistry to find a way to manage this issue. After hours of discussions, the preferred solution would be to add a halogenated cycle in the molecule to weaken the three carbon-fluorine bonds. Indeed, the halogenated cycle can delocalize some charges and desymmetrize the C–F bonds, which could help lower the energy barrier for an F⁻ to leave. However, this solution is hypothetical, and a new challenge arises: identifying an enzyme able to catalyse this reaction.

We found the lipase SpL as the perfect candidate to catalyze the formation of the halogenated cycle (Zeng & al, 2018, Fig. 6). This enzyme can catalyze the amide formation from carboxylic acid and amine.

Figure 6 : Amide formation from a carboxylic acid and an amine (from Zeng & al, 2018).

We would like to use this enzyme to catalyse the following reaction (Fig. 7). This reaction can be monitored by HPLC-UV, as both the reagent 4-chlorobenzylamine and the product N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide absorb in the UV range due to their aromatic rings.

Figure 7 : Amide formation from TFA and to produce N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide.

The product of the reaction, N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide, is significantly larger than the initial substrate, trifluoroacetic acid (TFA). This increase in molecular size poses a potential risk: reduced accessibility to the enzyme’s catalytic site. Our molecular docking experiments using YASARA (with AutoDock Vina) confirmed this concern. The product preferentially binds outside the catalytic site, whereas TFA, being smaller, is able to travel through the tunnel leading directly to the active site.

1. Structural alignment

After several discussions with researchers in protein design, we were advised to explore structural alignment as a strategy to identify enzymes with similar overall folds and conserved active sites, even if the surrounding amino acids are not identical. The idea was to find a structurally related enzyme that has never been tested, but that might still act on fluorinated substrates such as FA, DFA, TFA, or N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide.

We performed structural alignment using Foldseek, with DAR3835 as the template enzyme. Based on Foldseek scoring parameters (RMSD, TM-score, E-value, and sequence identity), we identified a single protein meeting our criteria: an alpha/beta-hydrolase from Chloroflexota bacterium (Uniprot entry A0A6N8Z7V7), that we will called “A0A6”.

2. Molecular docking using YASARA

In parallel, we performed molecular docking on RPA1163, the enzyme with the most resolved structures, including several co-crystallized with ligands such as glycolate (a reaction product) and fluoroacetate. The objective was to enhance RPA1163’s affinity for various fluorinated substrates identified in our literature review—namely DFA, TFA, PFOA, 1,1,1-trifluoro-2-butanone, and N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide—in order to increase binding energy at the catalytic site and promote SN2-mediated defluorination.

Key Criterion: All mutants must maintain a distance <3.5 Å to the catalytic aspartate oxygen to enable SN2 reaction.

Out of the 250 mutants we manually generated using YASARA, we selected 10 representative cases for further analysis—those showing the highest binding energy at the active site of mutated RPA1163, exceeding that of the wild-type enzyme. These include:

1 mutant to improve affinity with DFA
1 mutant to improve affinity with TFA
2 mutants to improve affinity with PFOA
2 mutants to improve affinity with 1,1,1-trifluoro-2-butanone
4 mutants to improve affinity with N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide

DFA

The mutation of Lysine 152 to Phenylalanine increased the binding energy from 4.80 to 4.86 kcal/mol while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. Although the improvement is small, this was one of the very few cases where affinity with DFA was enhanced, which justified its selection.

TFA

The mutation of Isoleucine 253 to Phenylalanine increased the binding energy from 5.40 to 5.48 kcal/mol, also maintaining a distance <3.5 Å to the catalytic aspartate oxygen. Despite the modest gain, it was selected as it was one of the rare cases showing improved affinity with TFA.

PFOA

Since PFOA is a hydrophobic molecule, like many PFAS compounds, its interactions predominantly involve hydrophobic amino acids such as isoleucine (Ile), valine (Val), and leucine (Leu). To facilitate the opening of the tunnel-like active site, alanine (Ala) mutations were also introduced.

The combined mutations Lysine 181 → Phenylalanine, Tryptophan 185 → Alanine, Isoleucine 253 → Valine, and Histidine 155 → Alanine resulted in a binding energy of 10.29 kcal/mol. Remarkably, PFOA was unable to enter the wild-type (WT) tunnel leading to the active site (see figure), but these mutations enabled access while keeping the <3.5 Å distance to the catalytic aspartate. This mutant was named “Flammenkuche.”
The mutations Tryptophan 185 → Alanine, Isoleucine 253 → Alanine, and Histidine 155 → Isoleucine yielded a binding energy of 8.99 kcal/mol under similar conditions (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Picon.”

1,1,1-Trifluoro-2-butanone

The mutations Histidine 155 → Isoleucine and Isoleucine 253 → Tryptophan increased the binding energy from 4.0 to 5.56 kcal/mol (see figure), while preserving the <3.5 Å catalytic distance. This mutant was named “RPA by-product.”
The mutations Histidine 155 → Isoleucine, Isoleucine 253 → Tryptophan, and Aspartate 114 → Asparagine increased the binding energy from 4.0 to 5.28 kcal/mol (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Choucroute.”

N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide

Upon addition of chlorinated aniline to TFA, the compound's polarity shifted, resulting in increased hydrophobicity (XLogP3-AA = 2.8, representing the logarithm of its partition coefficient between octanol and water (PubChem)). To enhance substrate binding affinity at the enzyme's active site, mutations introducing hydrophobic residues were performed. Furthermore, due to the presence of an aromatic ring in the molecule, mutations promoting pi–pi stacking interactions, such as substitutions with phenylalanine (Phe), were also introduced.

The mutations Serine 184 → Valine, Tryptophan 185 → Alanine, Isoleucine 253 → Tyrosine, Valine 177 → Phenylalanine, Tryptophan 156 → Phenylalanine, Lysine 181 → Alanine, and Tyrosine 154 → Alanine increased the binding energy from 5.06 to 7.26 kcal/mol (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Luffy.”
The mutations Serine 184 → Valine, Isoleucine 253 → Threonine, Methionine 155 → Phenylalanine, Tryptophan 185 → Alanine, Lysine 181 → Alanine, and Tyrosine 154 → Alanine increased the binding energy from 5.06 to 7.73 kcal/mol (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Sangoku.”
The mutations Isoleucine 253 → Alanine, Lysine 152 → Serine, Methionine 155 → Phenylalanine, Tryptophan 185 → Alanine, Lysine 181 → Alanine, and Tyrosine 154 → Alanine increased the binding energy from 5.06 to 7.68 kcal/mol (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Naruto.”
The mutations Tryptophan 185 → Glycine, Histidine 155 → Phenylalanine, Tyrosine 154 → Alanine, Arginine 111 → Isoleucine, Phenylalanine 40 → Alanine, Lysine 181 → Alanine, and Lysine 152 → Alanine increased the binding energy from 5.06 to 7.29 kcal/mol (see figure) while maintaining a distance <3.5 Å to the catalytic aspartate oxygen responsible for the SN2 mechanism. This mutant was named “Madara.”

3. Molecular docking using AI

We selected the nine top-ranked mutants generated by Xavier Robert based on the AI-driven scoring function (CNN_VS score). All of them ensured a distance <3.5 Å from the catalytic aspartate oxygen involved in the SN2 mechanism. These mutants were named XR1 to XR9, with the most interesting mutations being:

XR1: R111I, R114I, K152V, H155F, W185L, I253M
XR2: R111I, R114I, K152M, H155F, K181L, W185L, I253Q
XR3: R111I, R114V, K152I, H155Y, W185L, I253M
XR4: R111V, R114V, H155F, W185L, I253M
XR5: R111I, R114I, K152L, H155F, I253N
XR6: R111I, R114I, H155Y, K181L, I253Q
XR7: R111I, R114V, K152S, H155F, W185L, I253T
XR8: R111I, R114V, K152M, H155F, K181L, I253Q
XR9: R111V, R114I, K152V, Y154L, H155Y, W185L, I253S

Thanks to structural alignment, molecular docking, and literature review, we established a list of 25 enzymes to be experimentally characterized, ranked by testing priority. For RPA1163 mutants acting on a single substrate, enzymes are sorted based on their binding energy to the active site of the mutated RPA1163. The top priority is the degradation of cyclic TFA, therefore enabling TFA amidation with Lipase SpL and degradation of the reaction product (cyclic TFA).

Number	Enzyme Name	Correspondance	Notes
1	Lipase SpL	U1	Requires mass spectrometry (MS) analysis
2	RPA1163 WT	A1	Used as a positive control
3	XR1	O1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.76)
4	XR2	M1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.76)
5	XR3	N1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.76)
6	XR4	D1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.76)
7	SANGOKU	T1	RPA1163 mutant – Manually designed with YASARA for cyclic TFA
8	NARUTO	R1	RPA1163 mutant – Manually designed with YASARA for cyclic TFA
9	XR5	J1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.65)
10	XR6	L1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA (CNN_VS > 5.65)
11	T7RdhA	X1	Requires MS analysis – reductive dehalogenase
12	A0A6	V2	Protein never tested – identified via structural alignment
13	PICON	B1	RPA1163 mutant targeting PFOA
14	MADARA	S1	RPA1163 mutant – Manually designed with YASARA for cyclic TFA
15	LUFFY	Q1	RPA1163 mutant – Manually designed with YASARA for cyclic TFA
16	CHOUCROUTE 2	I1	RPA1163 mutant targeting PFOA degradation byproduct (1,1,1-Trifluoro-2-butanone)
17	TFA253F	G1	RPA1163 mutant targeting TFA
18	DFA152F	E1	RPA1163 mutant targeting DFA
19	FLAMMENKUCHE	C1	RPA1163 mutant targeting PFOA
20	CHOUCROUTE	K1	Targets PFOA degradation byproduct (1,1,1-Trifluoro-2-butanone)
21	XR7	F2	RPA1163 mutant – Enzyme generated by AI for cyclic TFA
22	XR8	H1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA
23	XR9	P1	RPA1163 mutant – Enzyme generated by AI for cyclic TFA
24	DAR3835 WT	W2	Second positive control
25	Peroxidase	—	Requires MS – purchased dehydrated, not a priority

For the laboratory experiments, our objectives are as follows :

Reproduce literature results with native enzymes RPA1163 and DAR3835 on FA, DFA, and TFA (we do not expect to observe defluorination on TFA).
Evaluate lipase activity to determine whether it can catalyze the addition of a halogenated ring to TFA, forming N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide, by HPLC-UV.
Test engineered mutants that showed the best docking scores with our target ligands (FA, DFA, TFA, N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide, PFOA, and 1,1,1-trifluoro-2-butanone) to assess whether any mutant is effective in defluorinating these molecules, by measuring fluorine concentration.
Test uncharacterized native enzymes, such as T7RdhA and A0A6, on fluorinated substrates.

Concretely, this involves expressing and producing the selected enzymes in various E. coli strains, followed by purification in sufficient quantities. Each enzyme will then be tested using the appropriate analytical method—such as HPLC-MS, HPLC-UV, colorimetric assays, or fluorine probes—to detect the target compound. These targets include the formation of cyclic TFA or the release of protons (H⁺) resulting from SN2 defluorination reactions. The ultimate goal is to construct a bioreactor containing immobilized enzymes to remove TFA and PFOA from drinking water.

To enable the construction of an enzymatic bioreactor, it was essential to design a strategy that not only allowed the large-scale production of degradation enzymes but also facilitated their rapid purification. For this purpose, we turned to the recently described targeting system VNP15 (Eastwood et al., 2023). This system relies on a 20–amino acid peptide tag that directs proteins into extracellular vesicles. These vesicles can be readily isolated from the culture medium and provide a microenvironment that supports both the stability and long-term storage of functional recombinant proteins.

The rationale was that by fusing our enzyme of interest to a VNP15 tag and an additional 6×His purification tag, we could harvest vesicle-enriched supernatants, lyse them by sonication, and subsequently purify the target enzyme via nickel affinity chromatography. This strategy, however, required adding two additional peptide sequences (20 residues from VNP15 and 6 residues from the His tag), potentially increasing the risk of steric hindrance and negatively affecting protein folding or catalytic activity. To evaluate this risk, we compared the predicted folded structures of tagged versus native proteins using AlphaFold and YASARA, observing a RMSD < 3 Å. This indicated no major disruption in folding.

Nonetheless, to minimize potential interference with catalytic activity, we introduced TEV protease cleavage sites between the VNP15 or His tag and the enzyme (Figure 1, Figure 2). This design allows for a streamlined purification workflow: secretion of vesicles → filtration and recovery → sonication → nickel column purification. The addition of TEV protease directly onto the nickel column (the protease itself also carrying a His tag and thus retained on the column) cleaves the enzyme from its tags, releasing a purified, tag-free enzyme directly in the eluate.

Figure 8 : Illustration of the purification steps enabled by our design. (A) Protein release into vesicles mediated by the VNP15 tag. (B) Recovery of vesicles from the supernatant, followed by sonication to disrupt them. (C) Schematic view of the protein domain organization at this stage of purification. (D) Nickel column purification and elution through TEV protease cleavage, with the protease itself retained in the column due to its own 6×His tag.

Transformation tests

Protocol optimization was performed using a fluorescent control construct (e.g., Vnp-mNeonGreen-6xHis) to compare two plasmids, pLR81 and pIBA37. This control allowed us to monitor protein production dynamics and to determine which plasmid backbone provided the most suitable expression profile for our system. The fluorescence signal also enabled visualization of vesicle formation and facilitated the tracking of each step of the downstream workflow, including filtration and purification.

Cloning and transformation were initially performed in E. coli DH5α, which served as the chassis for preliminary wet-lab validation.

For protein overexpression, several wild-type E. coli strains: BL21, HB42, and MG55 were tested to identify the most efficient host. Among these, MG55 exhibited the highest protein yield and was therefore selected for the remainder of the project

Transformation of E.Coli K16 MG55 Bacteria

The following experiments can be partially summarized in the table below.

	Cloning	Sequencing	Transformation	Overproduction	Purification	Activity testing
A1 : RPA1163 WT
B1 : Picon
C1 : Flammekueche
D1 : XR4
E1 : DFA152F
F2 : XR7
G1 : TFA253F
H1 : XR8
I1 : Choucroute 2
J1 : XR5
K1 : Choucroute
L1 : XR6
M1 : XR2
N1 : XR3
O1 : XR1
P1 : XR9
Q1 : Luffy
R1 : Naruto
S1 : Madara
T1 : Sangoku
U1 : Lipase SpL
V2 : AO46
W2 : DAR3835 WT
X1 : TH7Rdha WT

Table 1: Summary of results for cloning, sequencing, transformation, overproduction, purification, and activity testing — success (green) or failure (red), unknown (purple).

Plasmids whose sequences were validated by sequencing were transformed into the E.Coli K16 MG55 strain, a wild-type strain that we had previously made competent and which had proven to produce large quantities of proteins. Transformations were performed from mini-preps, following the standard protocol. Transformation success was verified by PCR. We were able to transform the following plasmids: RPA WT, RPA DFA, RPA XR7, RPA XR5, RPA XR2, RPA XR1, Naruto, Madara, and DAR WT. Due to time constraints, other plasmids were not transformed.

Overproduction

Bacterial Cultures

The first cultures were performed from colonies expressing Vnp.mNeonGreen and the Vnp.6xHis control. The objective was to optimize culture conditions to maximize protein and vesicle production via the Vnp sequence. For large-scale cultures, a starter culture was first prepared in a test tube containing 5 mL of LB, ampicillin, and a colony, incubated overnight at 37°C at 200 rpm. The next day, this culture was poured into 500 mL of TB with ampicillin in a 5 L flask (Fig. 9), incubated under the same conditions. Induction was performed at an OD of 0.5 with ATC, followed by overnight incubation. The wide flask allowed better aeration, which favored protein production. Strong green fluorescence was observed under a transilluminator in the Vnp.mNeonGreen culture (Fig. 10), while no fluorescence was detected in the Vnp.6xHis control.

Figure 9 : 500mL culture performed in a 5L flask.

Figure 10 :Observed fluorescence after an overnight incubation for the Vnp.mNeonGreen culture

Cultures were then distributed into 10 Falcon tubes of 50 mL, then centrifuged at 3000 g for 10 minutes at 4°C. Cell pellets (Fig. 11) were stored at -80°C and supernatants at 4°C. Supernatants from Vnp.mNeonGreen also showed green fluorescence, suggesting the presence of free proteins or proteins contained in released vesicles. No signal was detected in the Vnp.6xHis control supernatant.

Figure 11 :Observed fluorescence after an overnight incubation for the Vnp.mNeonGreen culture

This protocol was then used to produce proteins from the following transformants: RPA1163 WT, RPA1163 DFA, RPA1163 XR7, RPA1163 XR5, RPA1163 XR2, RPA1163 XR1, Naruto, Madara, and DAR3835 WT.

Overproduction test

Since the studied proteins are not fluorescent, overproduction tests were performed to verify their expression. A starter culture was first prepared for each transformant. In disposable tubes allowing optical density reading, 5 mL of TB, 5 µL of ampicillin, and 100 µL of starter culture were added, then incubated at 37°C at 200 rpm. At an OD of 0.5, cultures were induced with 2 µL of ATC for 2 h. An uninduced control was also prepared. After induction, a volume corresponding to an OD of 2 was collected and centrifuged at 3000 g for 10 minutes at 4°C. Pellets were resuspended in 200 µL of cracking buffer. Samples were then analyzed by SDS-PAGE. An intense band at the expected size was observed in induced samples compared to uninduced controls, confirming overproduction for the proteins RPA WT, RPA DFA, RPA XR7, RPA XR5, RPA XR2, RPA XR1, Naruto, Madara, and DAR WT.

Growth curves

To assess the impact of A1 and U1 protein overproduction on MG1655 growth, we monitored the growth of MG1655 strains overexpressing U1, A1, and the wild-type (WT) strain over a day. No visual differences were observed in the optical density (OD) curves over time (Fig. 12), indicating that enzyme overproduction does not affect MG1655 growth. This observation suggests two possible hypotheses: either the enzyme is not toxic to MG1655, or it is secreted into the extracellular medium via vesicles. The first hypothesis appears more likely—although we cannot exclude the possibility of partial secretion—since SDS-PAGE analysis of A1 and U1 cell pellets shows that the proteins are produced in relatively high quantities after induction with aTc, something that is likely absent in the WT strain (Fig. 13, 14 and 15). We also observed that the protein quantity appears stable from 7.25 hours onward, suggesting that overnight cultures can be used without affecting production levels.

Figure 12 : OD monitoring of MG1655 WT, A1 and U1.

Figure 13 : SDS-Page monitoring of our WT culture between 2:45 and 22:30 of growth. The sample numbers are described above, 0.4ODu per well.

Figure 14 : SDS-Page monitoring of our A1 culture between 2:45 and 22:30 of growth. The sample numbers are described above, 0.4ODu per well.

Figure 15 : SDS-Page monitoring of our U1 culture between 2:45 and 22:30 of growth. The sample numbers are described above, 0.4ODu per well.

Two purification protocols from vesicles were tested, named protocol A and protocol C.

Protocol A

In protocol A, supernatants were filtered at 0.45 µm, then vesicles were lysed by adding 0.1% Triton X-100. The solution was then filtered again at 0.45 µm to remove membrane debris. The obtained filtrate, containing proteins in soluble form, was incubated with Ni-NTA resin for affinity purification. After binding, the resin was transferred to a microspin column, washed with NPI-20 buffer (20mM imidazole), then proteins were eluted with NPI-500 buffer (500mM imidazole). Elution fractions were then dialyzed and concentrated by centrifugation on Vivaspin 500 columns, then stabilized with glycerol for storage at -80°C. Samples were taken at each step and analyzed by SDS-PAGE. However, fluorescence detection showed no signal for either Vnp.mNeonGreen or the Vnp.6xHis control. This protocol, tested several times, never allowed protein purification and was therefore abandoned.

Protocol C

Protocol C followed a similar logic with some modifications. The supernatant was first filtered at 0.45 µm. Vesicles were then concentrated by filtration on a 0.1 µm membrane, then manually recovered by washing with sterile PBS. They were then lysed with 0.1% Triton X-100. As before, 0.45 µm filtration removed debris, and the filtrate containing soluble proteins was incubated with Ni-NTA resin.

The resin was transferred to a microspin column, washed with NPI-20, then proteins eluted with NPI-500. Samples were dialyzed, concentrated, and stabilized with glycerol for storage at -80°C. Despite several attempts, this protocol also failed to purify proteins: no fluorescence detected, for either Vnp.mNeonGreen or Vnp.6xHis. Protocol C was therefore also abandoned.

An alternative protocol, named protocol B, was tested from cell pellets. Cells were lysed with Cell Lytic, then the soluble fraction was incubated with Ni-NTA resin for binding via the His-tag. After binding, the resin was transferred to a microspin column, washed with NPI-20 buffer, then proteins eluted with NPI-500 buffer. Samples were then dialyzed, concentrated, and stabilized with glycerol.

Samples were taken at each step (culture, pellet, lysate, flow-through, wash, elution, dialysis) to monitor purity by SDS-PAGE (Fig. 16 and 17). An intense band at approximately 27 kDa, corresponding to the expected size of Vnp.mNeonGreen, was clearly observed. The Vnp.6xHis control showed no intense band. This protocol proved effective for purifying proteins, although several non-specific bands were also present, particularly in elution fractions (B5E) and after dialysis (B6), suggesting protein contamination.

Figure 16 : Protocol B Vnp.mneongreen

Figure 17 : Protocol B Vnp.6xhis

Protocol B Optimization
Due to the time required and inefficiency of protocols A and C, protocol B was selected for further work. It is faster and allows obtaining purified proteins. To reduce contamination, several numbers of washes with NPI-20 buffer were tested: 2, 3, 4, and 5 washes. With 2 washes, significant contamination was still visible (Fig. 18) With 4 (Fig. 20) and 5 washes (Fig. 21), proteins of interest, including Vnp.mNeonGreen, were completely lost. With 3 washes (Fig. 19), a slight band corresponding to Vnp.mNeonGreen at 27 kDa was visible, with few other non-specific bands. This compromise was deemed optimal.

Figure 18 : Protocol B Vnp.mneongreen - 2 wash

Figure 19 : Protocol B Vnp.mneongreen - 3 wash

Figure 20 : Protocol B Vnp.mneongreen - 4 wash

Figure 21 : Protocol B Vnp.mneongreen - 5 wash

Next, another strategy was tested: performing two washes with NPI-20, followed by a wash with NPI-100 (or NPI-250), before elution with NPI-500. However, under these conditions, the 27 kDa band lost intensity and non-specific bands reappeared, indicating loss of proteins of interest and increased contamination (Fig. 22 and Fig. 23).

Figure 22 : SDS-Page purification of VNP-m-Neo-Green with NPI100

Figure 23 : SDS-Page purification of VNP-m-Neo-Green with NPI250

Final Choice
For subsequent work, it was decided to continue with protocol B using three washes with NPI-20, allowing retention of sufficient protein quantity while limiting contamination.

Purification of Our Enzymes
Protein purification can vary in complexity depending on the nature of the protein. In the case of RPA proteins, limited data is available, which necessitated several trials to develop an effective protocol. An initial purification of RPA WT was performed from bacterial pellets according to protocol B described above. The only difference concerns the elution step. After washes, the resin was incubated overnight at 4°C with 500 µL of NPI-20 buffer containing 1 µL of TEV protease. This protease cleaves the 6-His tag, thus facilitating protein elution. The next day, centrifugation was performed and the protocol continued normally. Results were analyzed by SDS-PAGE. Three intense bands were observed: one at approximately 37 kDa corresponding to RPA without the TEV site or His-tag, a band below possibly corresponding to RPA without the Vnp site, and a band at approximately 28 kDa probably corresponding to TEV protease. The same protocol performed without TEV addition shows only low-intensity bands at these molecular weights. TEV addition therefore appears to significantly improve protein elution.

Figure 24 : SDS-Page purification of A1 with TEV

Figure 25 : SDS-Page purification of A1 without TEV (ctr)

The same protocol was tested on RPA XR1 and RPA XR2. The purified proteins show several distinct forms. A second purification was performed with a Magne-His kit to be more precise, in case the observed bands corresponded to TEV that had not been removed during washes. After analysis, the bands do not correspond to TEV and magnetic purification does not provide a better solution.

Figure 26 : SDS-Page purification of A1 with TEV comparison

Comparative analyses show that these bands likely correspond to: Correctly cleaved proteins, partially cleaved Vnp15, and uncleaved proteins with their tag intact.

Figure 27 : SDS-Page purification of A1 in comparison with 0.2µL of TEV

Figure 28 : SDS-Page purification of A1 in comparison with 2µL of TEV

Figure 29 : SDS-Page purification of A1 in comparison with 1µL of TEV

Figure 30 : SDS-Page purification of A1 in comparison with 0.5µL of TEV

Tests with different amounts of TEV (0.2 µL to 2 µL) maintain this two-band profile, confirming that the second band is not TEV itself (Fig. 27 to 30).

Protein Assay
Total protein quantification was performed with a ThermoScientific kit compatible with glycerol. This colorimetric test estimates protein concentration by comparing sample absorbance to a BSA standard curve (Fig. 31) ranging from 0 to 2000 µg/mL. Samples are diluted in a buffer containing 50% glycerol to remain compatible with the kit. The reactive solution is prepared just before use by mixing two reagents provided in the kit in a 50:1 ratio. Each sample or standard is then mixed with this solution, incubated at 37°C for 30 minutes, then absorbance is measured at 520 nm. Concentration is determined from the calibration curve. This protocol was tested on Vnp.mneongreen protein purified from bacterial pellet. A concentration of approximately 452.6 µg/mL was obtained (average of 1/10 and 1/50 dilutions).

Figure 31 : BSA calibration curve for first experiment

The same protocol was applied to RPA WT, RPA XR1, and RPA XR2 proteins (Fig. 32 and Tab 2).

Figure 32 : BSA calibration curve during second experiment

Table 2 : Enzyme concentrations using BSA assay

The quantities of proteins obtained are, however, insufficient for kinetic tests. Many purifications will therefore be necessary. To reduce the cost associated with using Cell Lytic for cell lysis, sonication tests were performed.

After optimizing the sonication protocol, the following parameters were retained: five cycles of ten pulses each, at a power setting between 2 and 5 (to prevent foaming), with a duty cycle of 30%. The resuspension volume was adjusted based on the number of optical density (OD) units in the bacterial pellets. Volume and cycles did not seem to impact subsequently the protein quantity. Moreover, lysis using CellLytic buffer was compared (Condition 9 in Fig. 33). No major differences were observed between sonication combined with lysozyme and CellLytic condition, indicating that sonication is an adequate method. However, in the absence of lysozyme, protein yield appeared lower, likely due to incomplete bacterial lysis— some proteins may remain trapped within cells.

Figure 33: SDS-PAGE of sonication tests. Clear lysate is present in the wells. The tested conditions are described below.

Condition number, Culture (mL), resuspension PBS volume (mL), power, cycles
Condition 1, 50 mL, 5 mL PBS, Power 8, 10 × 10 pulses
Condition 2, 250 mL, 25 mL PBS, Power 8, 10 × 10 pulses
Condition 3, 50 mL, 5 mL PBS, Power 10, 10 × 10 s continuous
Condition 4, 50 mL, 5 mL PBS, Power 4, 10 × 10 pulses
Condition 5, 50 mL, 5 mL PBS, Power 8, 10 × 10 pulses
Condition 6, 100 mL, 10 mL PBS, Power 8, 10 × 10 pulses
Condition 7, 100 mL, 5 mL PBS, Power 8, 10 × 10 pulses
Condition 8, 100 mL, 10 mL PBS, Power 0, No sonication
Condition 9, 50 mL, 5 mL CellLytic, /, Chemical lysis

Subsequent pellet purifications were performed using this cell lysis protocol. After comparing protein presence in bacterial pellets and culture supernatants (Fig. 34), we confirmed once again that the proteins are strongly overproduced in the pellets. On the SDS-PAGE gel, no protein bands were visible in the culture supernatant. This absence may be due to the low concentration of proteins in the supernatant, as the culture medium volume is large compared to the bacterial biomass. Therefore, even if some proteins are secreted, they may be too diluted to be detected without prior concentration.

Figure 34: SDS-PAGE of bacterial pellets (PL) and culture supernatants (SP) of culture n°6 in notebook.

For protein coming from culture n°6, subsequent protein production was successful for A1 and U1, whereas not successful for O1 with everything in the insoluble pellet. However, O1 could be purified from the culture of the 07/09 week. It is unexplained why work once and not after (change in medium impact: first in LB, second in TB).

So, all enzymes were successfully purified subsequently even if proteins are lost due to lack to Ni-NTA resin. We can notice that the resin is not saturated only for WT purification (because no overproduced enzymes). Indeed, it is the only case where TEV protease is not visible and thus not co-eluted with the target protein (A1, U1 or O1). A larger volume of resin should have been used.

For proteins produced from culture n°6, subsequent expression was successful for A1 (Fig. 35) and U1 (Fig. 36), but not for O1 (Fig. 37), which was entirely found in the insoluble pellet. However, O1 was successfully purified from the culture dated 07/09 (Fig. 39). The reason for this discrepancy remains unclear, though a change in culture medium may have played a role, LB was used in culture n°6, while TB was used in the 07/09 culture.

Ultimately, all enzymes were successfully purified, although some protein loss occurred due to insufficient binding to the Ni-NTA resin. Notably, resin saturation was only absent during WT purification (Fig. 38) and O1.1 purification (Fig. 37), likely because no overexpressed enzyme was present. This is also the only condition where TEV protease was not detected, because binded to the resin, and therefore did not co-elute with the target protein (A1, U1, or O1). A larger volume of resin should have been used to improve protein and TEV binding.

Figure 35: SDS page of purification steps of MG1655 A1.3 (from culture n°6)

Figure 36: SDS page of purification steps of MG1655 U1.1 (from culture n°6)

Figure 37: SDS page of purification steps of MG1655 O1.1 (from culture n°6)

Figure 38: SDS page of purification steps of MG1655 WT (culture n°6)

Figure 39: SDS page of purification steps of MG1655 O1.3 (from culture of week 07/09)

Protein were then quantified using bovine serum albumine (BSA) protein quantification on SDS-Page gel (Fig. 40). The imageJ analysis allowed to build a standard range curve with BSA (Fig. 41) and link protein concentration (µg/mL) with band intensity. The protein quantity associated with the above purification (Fig. 35, 36 and 39) are as follows :
A1 (3) concentration = 1617.95 µg/mL
O1 (3) concentration = 1617.95 µg/mL
U1 (1) concentration = 1468.02 µg/mL

These concentrations are significantly higher than those obtained using the CellLytic lysis protocol. For A1 and O1, which are dehalogenases, the quantities are sufficient to perform a colorimetric assay to monitor defluorination. However, they are not adequate for use with the fluorine probe. Indeed, 2.5 mL must be collected for each measurement, meaning that a five-point analysis requires 25 mL (two times 12.5 mL), with a target protein concentration between 100 µg/mL and 200 µg/mL. Under these conditions, using purified proteins is challenging.

An alternative approach was therefore considered. Since proteins appeared to be present in relatively high amounts in the clear lysate (Fig. 33), this fraction might be sufficient to observe defluorination. Before testing this hypothesis with the fluorine probe, it was first evaluated using a colorimetric assay in a much smaller volume (200 µL) following the protocol from []. For U1, the lipase SpL, the situation is favorable: HPLC-UV analysis requires only a small volume (approximately 10 µL), making the available enzyme quantity sufficient for characterization.

Figure 40: SDS page of purified enzymes (A11, A12, A13, O13 and U11). The numbers 1, 2, 3, 4, and 5 correspond to the following BSA concentrations in µg/mL, respectively: 25, 250, 750, 1500, 2000

Figure 41: BSA standard range - Band intensity as a function of BSA concentration (µg/mL) - corresponds to the analysis of gel on Figure 39

Prior to performing the colorimetric assay, a dedicated culture was prepared to obtain only the clear lysate—without proceeding through the full purification protocol—for future colorimetric assay. However, enzyme overproduction was not observed (Fig. 42, showing clear lysates from cultures of A1, U1, O1, and WT MG1655), possibly due to the use of aged bacterial cultures.

Following ImageJ analysis, a calibration curve correlating band intensity with BSA concentration was generated (Fig. 43), allowing estimation of enzyme concentrations:
A1: 130.45 µg/mL
U1: 474.26 µg/mL
O1: 12.73 µg/mL
WT: 10.86 µg/mL

These concentrations are insufficient for defluorination assays. Therefore, bacterial strains should be re-streaked on fresh plates, and new cultures should be prepared to ensure optimal protein expression.

Figure 42: SDS page of clear lysates from 07/10/25 (A1, O1, U1 and WT). The numbers 1, 2, 3, 4, and 5 correspond to the following BSA concentrations in µg/mL, respectively: 25, 250, 750, 1500, 2000

Figure 43: BSA standard range - Band intensity as a function of BSA concentration (µg/mL) - corresponds to the analysis of gel on Figure 41

First of all, cyclic TFA is not soluble in water; a 90 mM solution was only soluble in a mixture of 62% methanol and 38% water. However, methanol is not ideal for enzymatic reactions and should be replaced with DMSO. Solubility must be tested both in the stock solution (original concentrated preparation) and in the working solution (the final diluted mixture used in the assay). When 5 µL of the 90 mM stock was added to 200 µL of fully aqueous medium, the compound was no longer soluble. Therefore, more extensive solubility and enzyme activity tests in partially organic media are required.

Given these limitations, the following experiments focused on fluoroacetate—the natural substrate of the dehalogenase RPA1163 (A1). Both A1 and O1 enzymes were tested to determine whether the wild-type enzyme could defluorinate its natural substrate and whether the engineered variant retained this activity. The colorimetric assay was performed using the following purified enzymes: A1 (from culture no. 6), O1 (from the week of 07/09), WT (from culture no. 6), and the clear lysate obtained on 07/10/25.

RPA1163 (A1) catalyzes an SN2 reaction that releases H⁺ ions into the medium as a reaction product. Phenol red was used as a pH-sensitive dye to follow pH change, and a standard curve was established using HCl concentrations ranging from 0 to 1 mM (see Fig. 44). Each well contained 5 µL of enzyme or clear lysate, combined with 195 µL of buffer supplemented with phenol red and fluoroacetate. Absorbance at 540 nm was monitored over a 2-hour period, and the resulting data were plotted as normalized absorbance curves over time.

Fig. 44. HCl standard range from 1mM HCl (yellow) to 0mM HCl (pink)

Normalized absorbance was calculated as the difference between the absorbance at 540 nm of the sample condition and that of the corresponding blank. It was computed as follows :

A_sample = A_{mean sample} − A_{mean blank}(t₀)

To reset the baseline to zero at the initial time point:

A_norm = A_sample − A_sample(t₀)

The resulting HCl calibration curve was successfully generated and exhibited a linear relationship (R² = 0.9898, Fig. 44).

Fig. 45 HCl standard range - Normalized absorbance at 540 nm as a function of HCl concentration (mM).

As expected, no overproduction was observed during culture (Fig. 41) for the clear lysates (Fig. 46), resulting in an insufficient enzyme quantity. This experiment should be repeated using a subsequently overproduced culture, first tested via colorimetric assay following a literature protocol, and then analyzed using the fluorine probe.

Fig. 46. Normalized absorbance variation over time for different concentrations of fluoroacetate (0, 5 and 10mM) in clear lysate corresponding A1, O1, and WT MG1655. Curves obtained after a colorimetric assay.

The purified O1 enzyme (Fig. 47), tested at 0, 5, and 10 mM fluoroacetate, exhibited behavior similar to the control (WT). Two hypotheses may explain this observation: (1) The enzyme was engineered to preferentially bind cyclic trifluoroacetate (TFA), which may have reduced its affinity for fluoroacetate (FA), preventing FA from effectively binding to the active site. (2) The enzyme may not be active. Since it was not produced under the same culture conditions as A1 and WT, direct comparison of enzymatic activity is not valid due to differences in experimental protocols.

Fig. 47. Normalized absorbance variation over time for different concentrations of fluoroacetate (0, 5 and 10mM) in purified enzymes corresponding A1.3, O1.3, and WT MG1655. Curves obtained after a colorimetric assay with absorbance taken at 540nm.

Acidification was observed with enzyme A1 (Fig. 47 and 48) at both 5 and 10 mM fluoroacetate, indicating that A1 is active and capable of catalyzing the defluorination of fluoroacetate.

Fig. 48. Normalized absorbance variation over time for different concentrations of fluoroacetate (0, 5 and 10mM) in purified enzymes corresponding A1.3 (RPA1163). Curves obtained after a colorimetric assay.

Fig. 49: H+ concentration as a function of time for purified A1 (RPA1163) with three fluoroacetate concentrations : 0, 5 and 10mM.

This result aligns with findings reported in the literature. (Khusnutdinova et al.,2023) reported an enzymatic activity of over 5.5 µmol/mg for RPA1163 (A1) at 5 mM fluoroacetate over a 2-hour incubation. By tracing the [H⁺] concentration over time (Fig. 49), we calculated enzymatic activity values of 4.32 µmol/mg and 2.98 µmol/mg for 5 mM and 10 mM fluoroacetate, respectively at 2 hours. These values are slightly lower but remain within the same order of magnitude as the published data. It was calculated as:

n(H⁺) ⁄ m(E), (µmol/mg) , where n(H⁺) is the amount of reaction product released in the well after 2 hours, and m(E) is the quantity of enzyme used.

Enzyme activity was also calculated per minute :

n(H⁺) ⁄ (t × m(E)) = V₀ × V_well ⁄ m(E), (µmol/min/mg) , where n(H⁺) is the amount of reaction product released in the well during the selected time interval t, and m(E) is the quantity of enzyme used.

Focusing on the linear portion of the [H⁺] over time curve, the enzymatic activity was estimated at approximately 0.055 and 0.045 µmol/min/mg for 5 mM and 10 mM fluoroacetate, respectively. When dividing the literature value by 120 minutes, the resulting activity falls between 0.045 and 0.05 µmol/min/mg—again, closely matching our observations. The slightly reduced activity may be attributed to suboptimal enzyme loading. At the time of the experiment, the exact enzyme concentration was unknown, and 5 µL of enzyme solution was used. For purified enzymes, this corresponded to approximately 7.5 µg, which is lower than the 20 µg used in the referenced study. Although enzymatic activity does not necessarily scale linearly with enzyme concentration, a follow-up experiment using a higher enzyme amount could provide a more accurate activity estimate.

In conclusion, enzyme A1 is active and capable of defluorinating fluoroacetate. The purification protocol using sonication lysis does not impair enzymatic function, as the enzyme retained its catalytic activity. Future steps include repeating the colorimetric assay using cyclic TFA and the O1 variant (a mutant of RPA1163), alongside the fluorine probe, to determine whether the enzyme remains active in a partially organic solvent—and, most importantly, whether O1 catalyzes the defluorination of cyclic TFA.

Given that A1 was confirmed to be active, it is reasonable to hypothesize that U1 (Lipase SpL) is also active, as both enzymes were cultured under identical conditions. The next objective is to assess whether U1 can catalyze the addition of a chloroaniline group to TFA, and to detect this cycle formation using our established HPLC-UV method. Looking ahead, these enzymes could potentially be integrated into enzymatic bioreactors to automate and facilitate the removal of TFA from potable water.

Thanks to NMR H analysis, we were able to confirm the synthesis of cyclized TFA, which we then analyzed and measured using HPLC-UV. A standard curve for cyclized TFA was created so that we could ultimately measure this molecule in our future reaction samples.

Synthesis of cyclized TFA

Target product: N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide (cyclized
TFA)
Reagents: 4-chlorobenzylamine (4 mmol) + ethyl trifluoroacetate (4 mmol)
Conditions: reaction at room temperature, stirring overnight in toluene
Yield: 75.9% (0.7209 g obtained from 0.004 mol theoretical)
Purification: column chromatography, isolated product isolated as a pure
white powder
Control: NMR confirming the structure of the product
CCM observation: presence of the product and elimination of the reagent
after purification

Figure 50 : NMR spectrum of synthesized N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide.

HPLC-UV analyses

Objective: to determine the retention times of the various target molecules and develop an analytical method for quantification.

Optimized gradient: 20 min in reverse phase
Retention time (rt):

4-chlorobenzylamine: ~7.6 min
Cyclic TFA: ~9.2 min

4-chlorobenzylamine and cyclic TFA were first tested several times separately for retention times to ensure the reproducibility of our results. Next, an HPLC vial was prepared with 500μL of 4-chlorobenzylamine + 500 μL cyclic TFA:

Figure 51 : HPLC Chromatogram Showing Retention Times of 4-Chlorobenzylamine (Reagent) and Cyclized TFA (Product).

Injection into HPLC of 10 μL

Separation: co-injection confirming the clear distinction between reagent and product

Calibration curve: performed with cycled TFA and toluene as internal standard

Linearity: R² = 0.978

y=0.0863*x + 0.339

Figure 52 : Calibration curve of cyclized TFA (N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide).

Adressing PFAS detection

Rational Design of E. coli-Based Biosensors for PFOA Detection via Transcriptomics-Guided Promoter Selection

PFAS detection is one of the major challenges limiting their study in water. Detection and quantification require complex protocols, costly experiments, and specific controls.

Our objective was to improve available detection methods by designing a biosensor targeting the two PFAS we studied: PFOA and TFA. Our strategy was to expose a PFAS-resistant bacterium to different concentrations of these compounds and analyze its transcriptional response. Promoters of genes whose expression increased with higher PFAS concentrations would then be identified and used to build a genetic construct. A reporter gene placed downstream of these promoters would produce a measurable signal proportional to the PFAS concentration in the culture medium.

A bacterium particularly known for its resistance to PFAS is Labrys portucalensis F11 (Science of the Total Environment, 2024). One major challenge we faced was establishing a transformation protocol for this strain, which, to our knowledge, has not yet been successfully transformed or even attempted in the literature. Another challenge was the extraction of RNA from this non-model, encapsulated bacterium.

Central step of our detection project:

Identification of promoters sensitive to PFAS concentrations during bacteria exposure to PFAS.

First, we needed to identify a minimal medium in which Labrys could grow properly without interference between the medium components and the PFAS. We tested different dilutions of TSB and found that the 1/5 dilution was the lowest dilution that did not alter the growth curve of Labrys portucalensis (Figure 1).

We initially aimed to perform growth curves in media containing PFOA and TFA. However, due to laboratory safety regulations, this was not possible: the growth tubes used for aerobic cultures allow air exchange, which posed a risk of contaminating the entire incubator with PFAS. Using sealed Falcon tubes could have addressed this issue, but their size was incompatible with our optical density measurement equipment. As a result, we relied solely on growth in TSB to estimate the time points at which to harvest the bacteria and apply RNAprotect for RNA stabilization.

Figure 1 : Growth curve of labrys portucalensis for different dilution of TSB with sterilized water.

We established an RNA-seq workflow in collaboration with our sponsor BioMérieux. Labrys cultures were exposed to PFOA and TFA at two concentrations: a “high” condition (100 µM) and a “low” condition corresponding to the EU regulatory limit for PFAS in drinking water (0.10 µg/L). Bacterial pellets (~7.5 × 10^8 cells) were harvested, RNA stabilized with RNAprotect®, and sent to BioMérieux for extraction and sequencing (Figure 2). Although preliminary optimization was initially planned, workflow changes required us to provide pellets directly. Each condition was prepared in quadruplicate, with one replicate set aside as a test sample for protocol adjustments.

Figure 2 : RNA-seq on Labrys portucalensis F11 conditions layout.

Despite several extraction attempts, RNA quality thresholds could not be reached, preventing us from pursuing the study of Labrys portucalensis transcriptional response to PFOA or TFA. This limitation likely stems from two factors: (i) the lack of precise calibration between OD and cell density, and (ii) the capsulated nature of L. portucalensis (Carvalho et al., 2008), which makes RNA purification particularly resistant to lysis. For future work, a better characterization of the OD-to-cell density relationship is essential, along with the implementation of stronger lysis strategies (e.g., combined mechanical and chemical double lysis) to ensure successful RNA extraction.

Although RNA-seq on Labrys portucalensis was unsuccessful, we still had time and resources to attempt a second trial using a model bacterium on which transcriptomic analysis would be easier: E. coli MG1655. We also identified the study by Wintenberg et al. (2025), which performed RNA-seq on E. coli MG1655 exposed to 100 µM of PFOA. Based on these results, we decided to design our PFOA biosensor using E. coli as a chassis. The only missing information was the transcriptional response of E. coli to TFA, which we set out to investigate.

We therefore performed growth tests in both minimal medium (M9) and TSB medium diluted with sterile water. We chose to perform RNA-seq in TSB 1:2 medium. An E. coli preculture was diluted 1:100 into TSB medium diluted 1:2 with sterile water, supplemented either with 100 µM TFA or with 0.87 nM TFA (Figure 3). The following table summarizes the experimental conditions:

Figure 3 : RNA-seq on E.coli MG1655 conditions layout.

The protocol for harvesting bacteria at the selected time points was established in collaboration with the IGFL laboratory, which performed the sequencing. A culture volume corresponding to approximately 10^9 bacteria was centrifuged for 10 minutes at 4000 g and immediately frozen in liquid nitrogen. The bacterial pellets were then stored at –80 °C. Samples were transported to IGFL in dry ice using a certified courier.

RNA purification was performed at the sequencing facility using the Illumina Stranded Total RNA Prep, Ligation with Ribo-Zero Plus kit. A single-end, reverse-stranded RNA-seq was performed with an average sequencing depth of 28,518,648 reads per sample.

Data processing from raw reads to the generation of the count table was carried out on Galaxy Europe. The first step consisted of assessing data quality using the tool Falco (Galaxy Version 1.2.4+galaxy0).

Figure 4 : Mean quality score of all samples across read positions.

Given the high quality of the raw data and the issues encountered when generating the count matrix after trimming (Figure 4), we decided, in consultation with a supervising instructor, to omit the trimming step. Reads were mapped using the “Map with BWA” tool (Galaxy Version 0.7.19), and read counting was performed with HTSeq-count (Galaxy Version 2.0.9+galaxy0). This analysis workflow was validated by our instructors based on the good quality of the raw data.

The resulting count matrix was subsequently analyzed using DESeq2 in R, which provided, among other outputs, the log2 fold change (L2FC) between two experimental conditions, as well as the adjusted p-value. These results allow the assessment of both the magnitude of transcriptional changes for each gene and their statistical significance.

For both the data generated by Wintenberg et al. (2025) and our own experiments, we applied the same analysis workflow. Starting from a DESeq2 output matrix (log2 fold change, adjusted p-value) and the raw count matrix, we selected genes with a log2 fold change (L2FC) > 1.5 and an adjusted p-value < 0.05. These genes were then ranked from highest to lowest L2FC.

Among these candidates, we focused on genes with very low read counts under control conditions (untreated / H20) but high read counts under PFAS treatment, in order to minimize background noise. Importantly, we avoided selecting genes with strong L2FC values when their expression in the untreated condition was already high, since these would reduce biosensor sensitivity.

For each PFOA treatment condition analyzed by Wintenberg et al. (2025), we selected two genes meeting these criteria and belonging to distinct signaling or metabolic pathways, to reduce the likelihood of co-regulation. Specifically:

6 h exposure to 100 µM PFOA: we selected appY (DLP12 prophage; DNA-binding transcriptional activator AppY) displaying a log₂ fold change = 1.5, and malE (maltose ABC transporter periplasmic binding protein) displaying a log₂ fold change = 2.28. These genes belong to different pathways, making co-regulation unlikely. This property is particularly important for biosensor design (see next section).
24 h exposure to 100 µM PFOA: we selected thrA (bifunctional aspartate kinase/homoserine dehydrogenase 1) displaying a log₂ fold change = 5.28, and mqsA (antitoxin of the MqsRA TA system / DNA-binding transcriptional repressor MqsA) displaying a log₂ fold change = 2,6 , which also belong to distinct pathways and are therefore unlikely to be co-regulated.

Later, when our own sequencing results became available, we applied the same approach to the conditions where E. coli MG1655 was exposed to 100 µM TFA for 4 h and 8 h.

Figure 5 : Volcano plot displaying genes that are significantly differentially expressed compared to the control condition. The gene symbol is labeled when the absolute log2 fold change is greater than 1.5. A) Differential gene expression analysis between the H2O condition and TFA 100 µM after 8 hours of exposure. B) Differential gene expression analysis between the H2O condition and TFA 100 µM after 24 hours of exposure.

At 4 h, the data did not reveal any particularly strong candidates, with the exception of ER3413_28, corresponding to carB (carbamoyl-phosphate synthetase large subunit). The average read count for this gene increased from 1,501 under control conditions (H₂O) to 4,629 under TFA exposure (log₂ fold change = 1,6). However, a baseline read count as high as 1,501 would generate excessive background noise, making this gene potentially unsuitable for use in a biosensor due to the important background.

At 8 h, several more upregulated genes were identified. Notably, ER3413_2576, corresponding to purM (phosphoribosylformylglycinamide cyclo-ligase), increased from an average of 429 reads (control) to 3,079 reads (TFA) (log₂ fold change = 2,8) , and ER3413_3765, corresponding to xanP (xanthine:H⁺ symporter), increased from 184 reads (control) to 3,116 reads (TFA) (log₂ fold change = 4,08). Although purM and xanP are not part of the same operon, both are repressed by purR and are therefore co-regulated. To avoid potential co-regulation, we replaced purM with ER3413_1665, corresponding to gtrS (CPS-53 (KpLE1) prophage; serotype-specific glucosyl transferase YfdI), which belongs to a distinct metabolic pathway and is unlikely to be co-regulated with xanP. it increases from an average of 433 reads to 1345 reads (log₂ fold change = 1,5).

All of these selected genes were significantly differentially expressed (padj < 0,05).

Although we were unable to integrate these results into our constructs due to time limitations, we are confident that the promoters selected for our design have the potential to reveal the presence of TFA in a given sample.

A well-designed biosensor must meet two essential criteria:

Specificity – it should respond exclusively to the signal of interest and not be activated by unrelated signals.
Sensitivity – it should be able to detect the molecule of interest within relevant concentration ranges.

In designing our biosensor, we prioritized specificity first, and subsequently optimized sensitivity while maintaining the same level of specificity.

To ensure specificity, we chose to rely on not just one but two promoters of genes upregulated in the presence of PFAS under each condition. The principle is that a detectable signal should appear only if both promoters are simultaneously active. This requirement explains our earlier strategy of selecting genes from distinct metabolic pathways. The goal was to maximize the likelihood that the set of molecules/signals capable of activating both promoters simultaneously would be restricted to the PFAS of interest (see schematic below).

For the reporter gene, we selected luciferase, since its luminescent output is directly proportional to its expression level. This is not the case for fluorescent proteins, which are less suitable for transmitting a quantitative signal proportional to promoter activity—one of our primary objectives.

The luciferase operon typically contains five genes: luxA, luxB, luxC, luxD, and luxE.

The LuxC–LuxD–LuxE trimer forms the fatty acid reductase complex, which synthesizes aldehydes whose chains are elongated before being processed by the LuxA–LuxB dimer.
The LuxA–LuxB dimer catalyzes the oxidation of reduced flavin mononucleotide (FMNH₂). The reaction products include oxidized flavin mononucleotide, a fatty acid chain, and visible blue-green light as the energy output.

Figure 6 : Luciferase mechanism

To ensure that the luciferase operon is expressed only if both promoters are active simultaneously, we designed a strategy in which the operon would be split into two functional modules. The genes responsible for synthesizing the luciferase substrate (luxC–luxD–luxE) would be placed under the control of the first promoter, while the genes encoding the luciferase enzyme itself (luxA–luxB) would be placed under the control of the second promoter (Figure 7, Figure 9).

With this configuration, activation of only one promoter—for example, in response to a natural metabolic signal—would lead to the expression of only part of the luciferase system. In such cases, no functional luminescent signal would be produced, since both modules are required. A measurable luciferase signal would therefore occur only when both promoters are active at the same time (see schematic with modeled graphs).

To complement this system, we also envisioned incorporating a fluorescent protein reporter into each synthetic operon, the whole construct is called “the biosensor construct”“. This would allow real-time monitoring of promoter activity and provide immediate diagnostic feedback in cases where no luciferase signal is observed, indicating which promoter is inactive (Figure 8).

Figure 7 : Illustration of the reflexive workflow explaining why we selected two different promoters from genes upregulated under PFAS exposure, each belonging to a distinct signaling pathway.

Figure 8 : Construct illustration : The biosensor construct for robust sensor response.

Figure 9 : Diagram illustrating the robustness of the response to a single signal.

We transformed E. coli DH5α either with the biosensor construct (see our collection part) integrated in a low copy plasmid (pSEVA261) or with an empty plasmid. We first tested our plasmid using the inducible promoters pLac and pTet to validate our design and exposed both strains to several inducer conditions to evaluate the system’s output under inducible promoter control.

Figure 10 : Construct illustration : The biosensor construct for robust sensor response with inducible promoter as positive control.

Response of single induction either IPTG 20µM or 10 ng/mL aTc:

A. IPTG induction

Figure 11 : Normalized GFP signal in DH5α transformed with either the biosensor construct or the empty plasmid, under single induction (20 µM IPTG) or without induction.

* : p < 0,05 between the condition “induction” and “non induction”, ** : p < 0,01 for the same conditions. # : p < 0,05 between the condition “induction” and “Empty plasmid”, ## : p < 0,01 for the same conditions. † : p < 0,05 between the condition “empty plasmid” and “non induction”, †† : p < 0,01 for the same conditions.

Figure 12 : Normalized luminescence signal in DH5α transformed with either the biosensor construct or the empty plasmid, under single induction (20 µM IPTG) or without induction.

Fluorescence and luminescence values are normalized to the optical density (OD).

Simple IPTG induction results in a GFP signal very similar to the uninduced condition. This illustrates a well-documented phenomenon in the literature, which we will describe in more detail later: the pLAC promoter exhibits considerable basal leakage. Despite this, a stronger GFP signal is observed under inducer conditions.

Regarding the luminescence signal, although it is initially higher than in the control condition, it tends to approach the control level as E. coli growth progresses. This preliminary observation suggests that the second promoter, pTet, exhibits little to no leakage, since the luminescent signal appears only when both promoters are activated.

As expected from our experimental design, activation of only one of the two promoters resulted in a weak luminescent signal (< 6000). This emission level is considerably lower than the values observed at the same time point under double induction (Figure X), confirming that the combined activation of both promoters is required to achieve a strong luminescent response.

B. Anhydrotetracycline induction

Figure 13 : Normalized mCherry signal in DH5α transformed with either the biosensor construct or the empty plasmid, under single induction (10 ng/mL aTc) or without induction.

Figure 14 : Normalized luminescence signal in DH5α transformed with either the biosensor construct or the empty plasmid, under single induction (10 ng/mL aTc) or without induction.

Simple anhydrotetracycline induction results in an mCherry signal significantly different from the uninduced condition (Figure X). Even though the uninduced condition shows a higher signal compared to the control, it remains significantly lower than the induced condition, so the Ptet promoter only leaks a little bit.

Regarding the luminescence signal, very high levels are observed even though only a single induction was performed. This can be explained by the conclusion drawn in the following paragraph: the pLac promoter exhibits strong leakage. As a result, in a condition where the pTet promoter is induced, both promoters are effectively active, since pLac is already leaking leading to a luminescence signal.

Comparison of luminescence output under no inducer, single inducer, and dual inducer conditions:

Figure 15 : Normalized luminescence signal in E. coli DH5α transformed with either the biosensor construct or the empty plasmid, under different induction conditions: single induction with 10 ng/mL aTc or 20 µM IPTG, double induction with 20 µM IPTG and 10 ng/mL aTc, or without induction.

* : p < 0,05 between the condition “Induction 20µM IPTG + 10 ng/mL aTc” and “non induction”, ** : p < 0,01 for the same conditions. # : p < 0,05 between the condition “Induction 20µM IPTG + 10 ng/mL aTc” and “Induction 10ng/mL aTc”, ## : p < 0,01 for the same conditions. † : p < 0,05 between the conditions “Induction 10ng/mL aTc” and “empty plasmid”, †† : p < 0,01 for the same conditions.

Under IPTG induction, the response was very low and indistinguishable from the baseline, consistent with the absence of pTet leakage and comparable to the uninduced condition. This matches the construct’s design, which prevents luminescence when only one promoter is active. In contrast, aTc induction triggered a strong signal because of pLac leakiness, effectively mimicking a dual activation scenario.

Dual induction nevertheless produced the highest luminescence, exceeding the levels observed under either single induction. While aTc alone generated a detectable signal, it remained significantly weaker than with both inducers combined. Overall, these results confirm that the construct favors specificity through simultaneous activation of both promoters, while also underlining the influence of pLac leakage and the need for further optimization.

This construct design emerges as a powerful template for enhancing signal specificity, with potential applications across a wide range of systems. While the constitutive leakiness of one promoter poses challenges and calls for further refinement, the overall architecture achieves a dramatic improvement in generating a uniquely specific response, demonstrating the promise of this strategy for precise synthetic control.

Since we conducted many experiments in parallel across different laboratories, we tested the reporter construct design and the PFOA-sensitive promoters separately. Unfortunately, we did not have time to integrate these promoters into the full construct, so they are assessed individually in the following section.

We only had time to work with two of the promoters identified in the RNA-seq study of E. coli exposed to PFOA (Wintenberg et al., 2025): the promoters of the thrA (b0002) and mqsA (b3021) genes. We constructed reporter plasmids (low - intermediate copy plasmid with p15A ORI) in which these promoters drove the expression of an RFP reporter. E. coli MG1655 ΔRM strains were transformed with these plasmids, resulting in one strain carrying the plasmid with the b0002 promoter driving RFP expression, and another strain carrying the plasmid with the b3021 promoter driving RFP. The control strain was unmodified E. coli MG1655 (WT).

Figure 16 : Illustration of transformed bacteria used to assess the powerfulness of the PFOA sensitive promoter.

We performed a time-course experiment (between t = 3h and t = 11h, then we measure it at t = 24h, t = 27h and a final time course between t = 27h and t = 35h) monitoring OD600 and RFP fluorescence (excitation 532 nm, emission 584 nm) in bacteria exposed to varying concentrations of PFOA in the growth medium.

The goal was to identify the time point where fluorescence response differed with PFOA concentration by analyzing the fluorescence/OD ratio in the presence versus absence of PFOA, to determine when the biosensor was most effective and whether fluorescence scaled proportionally with PFOA levels.

We identified a single time point (t = 12 h) that best met the response criteria for the b0002 promoter, and a single time point (t = 24 h) for the b3021 promoter.

Promoter b0002 :

Figure 17 : RFP fluorescence/OD response of E. coli MG1655 ΔRM transformed with the reporter construct containing the b0002 promoter, when exposed to different PFOA concentrations at t = 12h. (n = 4)

Promoter b3021 :

Figure 18 : RFP fluorescence/OD response of E. coli MG1655 ΔRM transformed with the reporter construct containing the b3021 promoter, when exposed to different PFOA concentrations at t = 24h. (n = 4)

E. coli MG1655 ΔRM transformed with the reporter construct containing promoter b3021 partially displays a PFOA-dependent response at t = 24 h. This response is observed at initial PFOA concentrations of 4 mM (200 µM final) and 20 mM (1 mM final). Unexpectedly, fluorescence under control conditions was higher than at 1.28 µM, 6.4 µM, and 32 µM, which we currently cannot fully explain. Possible causes include contamination of the control water or errors in premix preparation. Discussions with microbiologists suggest this is likely a measurement artifact or preparation mistake. We allow ourselves to interpret the results without this point, which most likely represents an error, while keeping in mind that the experiment should be repeated several times to confirm this. After removal, a clear PFOA-dependent fluorescence response is evident (Figure 19), way better than the previous promoter. Significance was assessed using a t-test after verifying homoscedasticity (n = 4 per condition).

Figure 19 : RFP fluorescence/OD response of E. coli MG1655 ΔRM transformed with the reporter construct containing the b3021 promoter, when exposed to different PFOA concentrations at t = 24h, when removing the water point. (n = 4)

The data suggest that there may be a proportional relationship between the PFOA concentration and the Fluo RFP / OD signal at 24 hours for b3021.

We want to predict the PFOA value using the Fluo/OD value, which implies swapping the axes and plotting PFOA (the variable to be predicted) on the Y-axis and Fluo/OD (the predictor variable) on the X-axis, because regression models optimize the sum of squared residuals along the vertical axis.

Figure 20 : Graphical representation of PFOA concentration as a function of the measured RFP fluorescence/OD ratio, along with the fitted polynomial regression curve.

With 7 reliable data points per condition, each being the average of four replicates, we chose a second-degree polynomial fitted curve because the coefficient of determination (R²) is the best we got. Specifically, an R² of 0.997 indicates that 99,7% of the variance between these two variables is explained by the second-degree polynomial model. However, using a second-degree polynomial requires some caution regarding overfitting. This risk is limited by the fact that the model is only second-degree and that each data point represents the average of four replicates.

In conclusion, and pending repetition of the experiment for confirmation, there is a second-degree polynomial relationship between the RFP fluorescence ratio (Excitation: 532 nm, Emission: 584 nm, gain = 115 on the Tecan Pro 200) normalized by OD₆₀₀ and the PFOA concentration ranging from 1.28 µM to 20 mM, at t = 24 h, in 190 µL of LB culture supplemented with 10 µL of PFOA-containing sample, incubated at 37 °C with agitation (Figure 21):

[PFOA] = 1.63 × 10⁻⁴ × ( RFP_Fluo
OD )² − 1.92 × ( RFP_Fluo
OD ) + 5443.9

By knowing the ratio of RFP Fluo / OD, it is possible to estimate the PFOA concentration with a 95% confidence interval using the predict function in RStudio, which computes the following formula to determine the confidence interval:

SE(Ŷ) = \sqrt(x₀ᵀ Cov(β̂) x₀) x₀ is the column vector containing the variables xxx raised to the appropriate powers (x⁰, x¹, x²) Cov(β̂) is the covariance matrix of the coefficients. lower bound = Ŷ - t 1 - α⁄2, n - p \cdot SE(Ŷ) upper bound = Ŷ + t 1 - α⁄2, n - p \cdot SE(Ŷ)

Example : With R, we want to determine the PFOA concentration corresponding to Fluo/OD = 9000. We created a polynomial model in R (see attached R Markdown file). After checking the homogeneity and normality of the residuals, we used the predict function on our polynomial model, which provides a lower and upper bound.

R returned an upper bound of 2029 µM and a lower bound of 572 µM. In our experimental data, we observed a Fluo RFP / OD ratio of 9000 corresponding to 800 µM of PFOA, which illustrates the accuracy of the model.

Figure 21 : Graphical representation of PFOA concentration as a function of the measured RFP fluorescence/OD ratio, with the fitted polynomial regression curve and the 95% confidence interval shown in red.

This method, however, should be validated using independent datasets that were not used to construct the polynomial model—for example, by repeating the experiment multiple times with the same bacterial strain and verifying whether the estimated concentrations consistently fall within the predicted confidence interval.

Although the confidence interval obtained with this method remains relatively broad, this approach represents, to our knowledge, the very first attempt at enabling biological detection and approximate quantification of PFOA concentrations. Such a completely novel strategy could be applied for preliminary water screening, allowing rapid, low cost and widespread monitoring across numerous sampling points. This, in turn, would make it possible to generate much denser contamination maps while still providing an approximate estimate of PFOA levels.

However, this model relies on a single promoter, which may also be activated by other compounds. Incorporating the same approach into a construct with two promoters could improve signal specificity.

Looking further: testing PFOA measurement in a complex matrix (environmental water) :

We also measured fluorescence in the same bacteria (with the promoter b3021 and b0002) exposed to different concentrations of PFOA diluted in environmental water to assess whether promoter activation was specific to PFOA or interfered by compounds present in environment water. Fluorescence was recorded at four time points: t = 0, t = 20 h, t = 24 h, and t = 40 h. We will focus on the result at t = 24h as we have selected this time point for our model.

Figure 22 : Average RFP fluorescence/OD response at different PFOA concentrations diluted in environmental water (complex matrix). (n = 4)

While the b3021 promoter exhibited a strong response even in the presence of water alone, the b0002 promoter showed a significant fluorescence increase (p < 0.05) when the PFOA concentration in the well was equal to or greater than 1 nM. However, the response was not proportional to PFOA concentration. In this complex environmental water matrix, it is likely that several compounds can activate the b3021 promoter, as illustrated by its strong response in the water-only condition. This highlights the importance of enhancing specificity. Our divided luciferase operon design could provide a solution by requiring the activation of two promoters simultaneously, thereby optimizing the response and restricting specificity to PFOA.

References

Chan, P. W. Y., Yakunin, A. F., Edwards, E. A., & Pai, E. F. (2011). Mapping the Reaction Coordinates of Enzymatic Defluorination. Journal Of The American Chemical Society, 133(19), 7461‑7468. https://doi.org/10.1021/ja200277d

Colosi, L. M., Pinto, R. A., Huang, Q., & Weber, W. J. J. (2008). Peroxidase‐mediated degradation of perfluorooctanoic acid. Environmental Toxicology And Chemistry, 28(2), 264‑271. https://doi.org/10.1897/08-282.1

Eastwood, T. A., Baker, K., Streather, B. R., Allen, N., Wang, L., Botchway, S. W., Brown, I. R., Hiscock, J. R., Lennon, C., & Mulvihill, D. P. (2023). High-yield vesicle-packaged recombinant protein production from E. coli. Cell Reports Methods, 3(2), 100396. https://doi.org/10.1016/j.crmeth.2023.100396

EFSA Guidance Document for evaluating laboratory and field dissipation studies to obtain DegT50 values of active substances of plant protection products and transformation products of these active substances in soil. (2014). EFSA Journal, 12(5). https://doi.org/10.2903/j.efsa.2014.3662

Ellis, D. A., Martin, J. W., Muir, D. C. G., & Mabury, S. A. (2003). The use of 19F NMR and mass spectrometry for the elucidation of novel fluorinated acids and atmospheric fluoroacid precursors evolved in the thermolysis of fluoropolymers. The Analyst, 128(6), 756. https://doi.org/10.1039/b212658c

Farajollahi, S., Lombardo, N. V., Crenshaw, M. D., Guo, H., Doherty, M. E., Davison, T. R., Steel, J. J., Almand, E. A., Varaljay, V. A., Suei-Hung, C., Mirau, P. A., Berry, R. J., Kelley-Loughnane, N., & Dennis, P. B. (2024). Defluorination of Organofluorine Compounds Using Dehalogenase Enzymes from Delftia acidovorans (D4B). ACS Omega, 9(26), 28546‑28555. https://doi.org/10.1021/acsomega.4c02517

Harris, B. A., Zhou, J., Clarke, B. O., & Leung, I. K. H. (2024). Enzymatic Degradation of PFAS : Current Status and Ongoing Challenges. ChemSusChem. https://doi.org/10.1002/cssc.202401122

Hu, M., & Scott, C. (2024). Toward the development of a molecular toolkit for the microbial remediation of per-and polyfluoroalkyl substances. Applied And Environmental Microbiology, 90(4). https://doi.org/10.1128/aem.00157-24

Huang, S., & Jaffé, P. R. (2019). Defluorination of Perfluorooctanoic Acid (PFOA) and Perfluorooctane Sulfonate (PFOS) by Acidimicrobium sp. Strain A6. Environmental Science & Technology, 53(19), 11410‑11419. https://doi.org/10.1021/acs.est.9b04047

IARC Monographs evaluate the carcinogenicity of perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS). (s. d.). https://www.iarc.who.int/news-events/iarc-monographs-evaluate-the-carcinogenicity-of-perfluorooctanoic-acid-pfoa-and-perfluorooctanesulfonic-acid-pfos/

Jeschke, P. (2017). Latest generation of halogen‐containing pesticides. Pest Management Science, 73(6), 1053‑1066. https://doi.org/10.1002/ps.4540

Khusnutdinova, A. N., Batyrova, K. A., Brown, G., Fedorchuk, T., Chai, Y. S., Skarina, T., Flick, R., Petit, A., Savchenko, A., Stogios, P., & Yakunin, A. F. (2023). Structural insights into hydrolytic defluorination of difluoroacetate by microbial fluoroacetate dehalogenases. FEBS Journal, 290(20), 4966‑4983. https://doi.org/10.1111/febs.16903

Luo, Q., Yan, X., Lu, J., & Huang, Q. (2018). Perfluorooctanesulfonate Degrades in a Laccase-Mediator System. Environmental Science & Technology, 52(18), 10617‑10626. https://doi.org/10.1021/acs.est.8b00839

Parsons, J. R., Sáez, M., Dolfing, J., & De Voogt, P. (2008). Biodegradation of Perfluorinated Compounds. Reviews Of Environmental Contamination And Toxicology, 53‑71. https://doi.org/10.1007/978-0-387-78444-1_2

Presentato, A., Lampis, S., Vantini, A., Manea, F., Daprà, F., Zuccoli, S., & Vallini, G. (2020). On the Ability of Perfluorohexane Sulfonate (PFHxS) Bioaccumulation by Two Pseudomonas sp. Strains Isolated from PFAS-Contaminated Environmental Matrices. Microorganisms, 8(1), 92. https://doi.org/10.3390/microorganisms8010092

PubChem. (s. d.). Acetamide, N-[(4-chlorophenyl)methyl]-2,2,2-trifluoro-. PubChem. https://pubchem.ncbi.nlm.nih.gov/compound/580806

Rickard, B. P., Rizvi, I., & Fenton, S. E. (2021). Per- and poly-fluoroalkyl substances (PFAS) and female reproductive outcomes : PFAS elimination, endocrine-mediated effects, and disease. Toxicology, 465, 153031. https://doi.org/10.1016/j.tox.2021.153031

Wackett, L. P. (2024). Confronting PFAS persistence : enzymes catalyzing C–F bond cleavage. Trends In Biochemical Sciences. https://doi.org/10.1016/j.tibs.2024.11.001

Wijayahena, M. K., Moreira, I. S., Castro, P. M. L., Dowd, S., Marciesky, M. I., Ng, C., & Aga, D. S. (2025). PFAS biodegradation by Labrys portucalensis F11 : Evidence of chain shortening and identification of metabolites of PFOS, 6 : 2 FTS, and 5 : 3 FTCA. The Science Of The Total Environment, 959, 178348. https://doi.org/10.1016/j.scitotenv.2024.178348

Wintenberg, M. E., Vasilyeva, O. B., & Schaffter, S. W. (2025). Comparative Transcriptomic Analysis of Perfluoroalkyl Substances-Induced Responses of Exponential and Stationary PhaseEscherichia coli. bioRxiv (Cold Spring Harbor Laboratory). https://doi.org/10.1101/2025.02.18.638913

Zahm, S., Bonde, J. P., Chiu, W. A., Hoppin, J., Kanno, J., Abdallah, M., Blystone, C. R., Calkins, M. M., Dong, G., Dorman, D. C., Fry, R., Guo, H., Haug, L. S., Hofmann, J. N., Iwasaki, M., Machala, M., Mancini, F. R., Maria-Engler, S. S., Møller, P.,. . . Schubauer-Berigan, M. K. (2023). Carcinogenic.

Zhou, X., Mehta, S., & Zhang, J. (2020). Genetically Encodable Fluorescent and Bioluminescent Biosensors Light Up Signaling Networks. Trends In Biochemical Sciences, 45(10), 889‑905. https://doi.org/10.1016/j.tibs.2020.06.001

Adressing PFAS degradation

Litterature overview

Focus on TFA degradation

Structural alignment and molecular docking

1. Structural alignment

2. Molecular docking using YASARA

DFA

TFA

PFOA

1,1,1-Trifluoro-2-butanone

N-[(4-chlorophenyl)methyl]-2,2,2-trifluoroacetamide

3. Molecular docking using AI

Final enzyme list and wet-lab objectives

Plasmid design

Transformation results

Transformation tests

Transformation of E.Coli K16 MG55 Bacteria

Overproduction and growth tests

Overproduction

Growth curves

Protein purification of vesicles using cell lytic buffer

Purification of pellets with cell lysis using cell lytic buffer

Purification using sonication-based lysis in combination with lysozyme

Defluoration monitoring

HPLC-UV and NMR methods

Synthesis of cyclized TFA

HPLC-UV analyses

Adressing PFAS detection

Central step of our detection project:

RNAseq Labrys portucalensis

RNAseq on E.coli

Identify promoters sensitive to PFOA and TFA

Biosensor design for PFOA

Assess the robustness of the biosensor construct

A. IPTG induction

B. Anhydrotetracycline induction

Assess the ability of the PFOA-sensitive promoter to respond to PFOA exposure

References

References