Monooxygenase-dehydrogenase cascade for sustained enzymatic remediation of TMA in salmon protein hydrolysates

Monooxygenase-dehydrogenase cascade for sustained enzymatic remediation of TMA in salmon protein hydrolysates Rasmus Reea, Øivind Larsena, Sushil Gaykawada, Sreerekha S. Ramananda, Antonio GarcíaMoyanoa, Irina Elena Chiriacb, Pål Puntervolla, Gro Elin Kjæreng Bjergaa a NORCE Climate & Environment, NORCE Research AS, Bergen, Norway b Leitat Technological Center, Terrassa, Barcelona, Spain. Abstract Fish protein hydrolysates hold great promise as nutraceuticals, yet their application as food ingredients or nutraceuticals is currently limited by their fish-like odor. This odor is mainly due to the presence of trimethylamine (TMA), a volatile biogenic amine resulting from the breakdown of naturally occurring trimethylamine-N-oxide (TMAO) in marine fish. The bacterial trimethylamine monooxygenase mFMO can oxidize TMA into TMAO using molecular oxygen and the cofactor nicotinamide adenine dinucleotide phosphate (NADPH). We have established an enzyme cascade which takes advantage of glucose dehydrogenase to recycle NADPH from NADP+, significantly decreasing the cost of the reaction and paving the way for using the enzyme system in fish protein hydrolysates targeted for human consumption. We demonstrate that the dual enzyme system works in an industrially relevant substrate. Salmon protein hydrolysate treated with an mFMO/glucose dehydrogenase cocktail showed a 75% reduction in TMA. A trained sensory panel perceived an improved odor across several parameters, including a reduction in the characteristic TMA smell. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Introduction The Norwegian aquaculture and fisheries industry produced more than 3.2 million tons of seafood in 2023 (1). Of this, over 1 million tons were byproducts, defined as any product which is not the main product produced from a raw material, such as heads, frames and backbones. Fish protein hydrolysis is a way of utilizing such byproducts through controlled proteolysis, often using subtilisins (2). The product, fish protein hydrolysates (FPH) is a mix of amino acids and small peptides which are suitable for human consumption. It is an excellent protein source, with protein content varying depending on production method; salmon byproducts typically contain 69-89% protein (3–5). FPH has been explored as a source of bioactive peptides with antioxidant, anti-hypertensive and anti-inflammatory effects (6, 7). However, according to statistics of Norwegian use, FPH is mainly used in pet food and feed formulations for farmed fish (1) rather than in human nutrition. The main barriers to consumer acceptance of FPH as food appears to be its fishy odors and flavors (1, 2, 8). Consequently, there is a demand for novel strategies which enable the use of these nutritious fish byproduct-derived proteins in the higher-value food market. Several compounds are known to contribute to fish smell, of which trimethylamine (TMA) is a key component. TMA is a volatile, biogenic amine which is formed postmortem from trimethylamine-N-oxide (TMAO). TMAO is a naturally occurring metabolite in fish from cold and deep-sea environments and is, importantly, odorless. TMAO is believed to function as an osmolyte (9) and as a piezolyte, counteracting pressure-mediated inhibition of protein function (10, 11). After slaughtering, TMAO in fish is converted to TMA by TMAO-reducing bacteria, contributing to its fish-like odor (12–14). The FPH industry has identified TMA as a key target for improving the sensory properties of FPHs. To remove it, some industrial actors currently use nanofiltration of FPH, but this untargeted method has the drawback of causing significant protein loss and altering the nutritional composition (15). An interesting alternative to filtration is to use an enzyme to specifically target TMA. Previously, we have shown that the bacterial trimethylamine monooxygenase mFMO can be used to remove most of the TMA from salmon FPH in a targeted approach (16, 17). The mFMO enzyme is a flavincontaining monooxygenase (FMO) (18) isolated from the marine gammaproteobacterium Methylophaga aminisulfidivorans (19). It belongs to a family of closely related bacterial FMOs which catalyze the oxidation of TMA (16, 20, 21). These bacterial FMOs oxidize TMA using molecular oxygen and NADPH as a cofactor, leaving TMAO and the oxidized cofactor NADP+ as products (Figure 1, top) (19, 21). To enhance compatibility with industrial .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint processing conditions, a thermostable mFMO variant, termed mFMO_20, was generated through structure-based engineering and shown to perform well at up to 65°C (17). The oxidation reaction catalyzed by mFMO uses one molecule of NADPH per molecule of TMA. As NADPH is very expensive, mFMO-assisted removal of TMA in fish protein hydrolysates will not be industrially viable unless cofactor consumption is managed in a cost-effective manner. Two strategies have been explored to reduce the cost of cofactors in reactions that depend on them: enzyme-cofactor engineering and cofactor recycling. Enzyme engineering can be used to improve the affinity for a cheaper cofactor with the same redox function, such as nicotinamide adenine dinucleotide (NADH) (22) or nicotinamide mononucleotide (23). However, mutation may compromise other aspects of enzyme functionality, such as reaction rate and stability, requiring substantial screening to achieve a useful enzyme reaction rate (24). To our knowledge, successful attempts to change the cofactor specificity of FMOs from NADPH to NADH have not been reported. Furthermore, although NADH is cheaper than NADPH it remains prohibitively expensive for an industrial process (25). Cofactor recycling is thus an attractive strategy for cofactor management and has been successfully implemented in several systems (25). It involves regenerating the redox cofactor, enabling its reuse in multiple reaction cycles. This can be achieved through direct chemical reductive regeneration (26), homogenous (27, 28) and heterogenous (29) regeneration using hydrogen and organometallic catalysts, photocatalytic regeneration (30), and electrochemical regeneration (31). Alternatively, enzymatic regeneration (32–34) uses a secondary enzyme reaction, which reduces the cofactor while oxidizing a sacrificial substrate, to maintain cofactor availability and drive the main reaction. It has several advantages: it can be highly specific and, depending on the choice of sacrificial substrate, toxic components and catalysts can be avoided. (25). For food production involving enzymatic TMA-removal, the sacrificial substrate and products must food safe, and the regeneration enzyme must be compatible with both the processing conditions and the buffer requirements of the TMA oxidizing enzyme. Moreover, glucose is an inexpensive substrate, and this recycling strategy allows the use of the cheaper cofactor NADP+ instead of NADPH, significantly reducing costs associated with cofactor supplementation. Various enzymes have been used to regenerate NADPH, including alcohol dehydrogenase, phosphite dehydrogenase, glucose dehydrogenase (GDH) and glucose-6-phosphate dehydrogenase (35). Phosphite dehydrogenase has been used in fusion constructs to enable .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint sustained catalysis mediated by mFMO (36), and a Baeyer–Villiger monooxygenase (37). Like FMOs, the latter enzyme belongs to the class B flavoprotein monooxygenases (38). While GDH has not yet been reported for use with FMOs, it is an attractive cofactorregenerating enzyme due to its widespread application, its simplicity, and the low cost of its substrate, glucose. Of note, glucose dehydrogenase B (GdhB) has been used to regenerate NAPDH during enzymatic Tyrian purple production by an FMO (39). For the application area in this study, it is important to note that both the glucose substrate and the product, glucono-1,5-lactone (also known as glucono-delta-lactone, or GDL), are recognized as safe and widely accepted food ingredients (Figure 1, bottom) (40). In this study, we couple the activity of the thermostable mFMO_20 to the activity of the glucose dehydrogenase GdhB from Priestia megaterium (previously known as Bacillus megaterium) (41, 42). We demonstrate that this enzyme cascade, in the presence of excess glucose and the oxidized cofactor NADP+, effectively depletes TMA in salmon protein hydrolysate while recycling NADPH and producing GDL. Our approach demonstrates the utility of cofactor recycling for cost reduction in TMA remediation in an industrially relevant context. Results and discussion Establishing a cofactor recycling system for a TMA monooxygenase When setting up a cofactor recycling system for enzymatic removal of TMA in FPH, we chose the GdhB enzyme and glucose as sacrificial substrate, and to use NADP+ as the added cofactor. To guide enzyme dosage and cofactor concentration in the dual enzyme system, we characterized the kinetics of GdhB for glucose. The Km of GdhB for glucose was 68.7 mM (Table S1), a factor of 8 x 105 higher than that of mFMO_20 for TMA (17). The Vmax was about half that of mFMO_20. To compensate for the lower efficiency of GdhB, we set up the recycling system using a glucose concentration of 50 mM and a 10:1 ratio of GdhB to mFMO_20. Since NADH is more stable and less costly than NADPH, activities of a mFMO_20 and GdhB were assessed using both NADPH and NADH as electron donors. GdhB can catalyze oxidation of glucose by using both cofactors and can as well catalyze the reverse reaction using NAD(P)H and GDL (Supplemental Figure S1A). A sufficiently high glucose concentration is this required to drive the reaction towards GDL formation and NADP+ reduction. However, mFMO_20 does not accept NADH as a cofactor (Supplemental Figure S1B), necessitating use of NADP+. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Figure 1: TMA oxidation and glucose dehydrogenation by FMO and GDH. Scheme of the TMAto-TMAO and glucose-to-GDL reaction cycle. NADPH and NADP+, reduced and oxidized nicotinamide adenine dinucleotide phosphate, respectively. To demonstrate that the mFMO_20/GdhB enzyme system catalyzed the expected reactions, we conducted an enzyme assay using the two enzymes and NADP+ as a cofactor. Two key features of the experimental design support this demonstration. First, we used NADP+ as a cofactor, rather than NADPH, ensuring that any TMAO production depended on GdhB activity. Second, TMA was added in a 5:1 molar excess relative to the cofactor, so that the formation of TMAO in quantities exceeding the initial cofactor concentration would provide direct evidence of cofactor recycling. The substrates and products were quantified using liquid chromatography/mass spectrometry (LC/MS) (Figure 2, Supplemental Table S3-S4). TMAO and GDL were formed when both enzymes, substrates and a cofactor were present. As expected, TMAO was not formed when any of the components were removed. When 100 μM NADPH was used as cofactor in the absence of GdhB, only 58 µM TMAO was formed, and as expected, no GDL was produced. Cofactor recycling was demonstrated by the formation of 191 µM TMAO a reaction with only 100 µM NADP+. This confirms that cofactor recycling enabled the oxidation of TMA in molar amounts exceeding the initial cofactor concentration. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Figure 2: TMA oxidation driven by cofactor recycling Enzyme assays (n=3) with the indicated components (+, present; -, absent): mFMO_20, GdhB, NADP+ (or NADPH), glucose and/or TMA at the given concentrations, were incubated for 1 hour at 25°C and analyzed by LC/MS for the presence of TMA, TMAO, glucose or GDL, which were quantified against a standard curve. The GDL concentrations were lower than expected, as we anticipated GDL and TMAO to increase in concert. This discrepancy may be due to spontaneous hydrolysis of GDL to gluconic acid, a process reported in the literature (40). Although this possibility was considered during LC/MS method development, no ions corresponding to gluconic acid were detected. A disadvantage of using GdhB for cofactor recycling in this system is its relatively low catalytic activity compared to mFMO_20, necessitating a relatively high concentration of glucose to drive the recycling reaction. We used 50 mM glucose and 100 μM NAD+, which is consistent with the 100 mM glucose and NAD+ concentrations of 10-500 mM used in previous studies with this enzyme (41, 43, 44). The final glucose concentration in the protein hydrolysate obtained in this study was relatively high at 1.75% w/w, assuming a 10% dry weight content .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint (see next section). However, this level is comparable to the 1% xylose concentration used with heat to achieve browning and caramelized flavor for odor masking of salmon protein hydrolysate (45). In comparison,). To achieve better TMA remediation with lower levels of glucose in the final product, it could be useful to improve the activity and substrate affinity of GdhB through enzyme engineering (46) or through immobilization (47, 48). The dual enzyme system reduces TMA levels in salmon protein hydrolysates To demonstrate that the mFMO_20/GdhB enzyme cascade can deplete TMA in an industrially relevant protein hydrolysate, a lab-scale protease-driven hydrolysis of salmon heads and frames (byproducts) was performed (Figure 3). Figure 3: Production and evaluation of enzyme-treated FPH at lab scale. The rectangles indicate process steps and analyses, while diamonds show branching points, and the parallelograms contain intermediate and final products, and indicates which figure contains the associated result. The flow of control FPH fractions are shown in red and enzyme-treated FPH fractions in blue. FPH: fish protein hydrolysate (salmon); LC/MS: liquid chromatography/mass spectrometry. Whole, fresh salmon were filleted, and the heads and frames were minced in a meat grinder to produce the byproduct feedstock. This mince was mixed with 50% water (w/w) and 0.5% .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Alcalase 2.4L (v/w biomass), a commercial subtilisin endoprotease with broad specificity. After hydrolysis and heat inactivation of the protease, the sample was centrifuged to isolate the watersoluble fraction containing hydrolyzed peptides, hereafter referred to as FPH. To protect the added mFMO and GdhB enzymes from proteolytic cleavage and lipid interference, they were introduced only after hydrolysis and centrifugation, along with glucose and NADP+ (Figure 3). We obtained a total of 4860 ml liquid hydrolysate which served as the substrate for enzymatic TMA removal treatment (Figure 3, top middle box). The dry matter content was estimated to be 9.3%.LC/MS analysis of FPH samples collected during the enzymatic TMA removal process showed that treatment with mFMO_20 and GdhB depleted TMA in a time-dependent manner, reducing it to less than 1% of the initial intensity (Figure 4A, Supplemental Table S5). The TMA intensity was unchanged in controls where enzymes, glucose, and NADP+ were absent. The TMAO intensity was stable throughout the time course, likely reflecting the high TMAO content in the freshly prepared FPH. The glucose concentration also remained stable, as it was added in great excess. GDL intensity increased over time in parallel with the reduction of TMA, confirming successful cofactor regeneration. The final dried control and enzyme-treated FPHs were prepared by evaporation until partially dried, followed by freeze drying until completely dry. Analysis of the evaporate (Figure 4B, Supplemental Table S6) revealed approximately 50 µM TMA remaining in the control FPH, while no TMA was detected in the enzyme-treated evaporate. This demonstrates that although drying may assist in TMA removal it is not sufficient to fully deplete it. Quantification of TMA in the dried FPHs showed that 58 ppm remained in the mFMO_20/GdhB-treated FPH, whereas 208 ppm was retained in the control (Figure 4C). Hence, the enzyme treatment oxidized approximately 75% of the TMA content. A previous study reported that the application of nanoand diafiltration reduced TMA from 700 ppm to 100 ppm in cod FPH and from 400 ppm to 100 ppm in salmon FPH (15). Reaching a level of 100 ppm was associated with improved TMA taste intensity. We did not observe an increase in TMAO concomitant with the decrease in TMA in the enzyme-treated FPH. The 150 ppm TMAO produced may not make enough of a difference to be detected between samples, given that the TMAO concentration was measured between 1344 and 1615 ppm, and the variation between replicate injections was between 4.9 and 12.2% (Supplemental Table S6). Neither did we detect GDL in the enzyme-treated FPH. The lower limit of quantitation of GDL was 4.2 pmol, while the reduction in TMA was measured at 7.62 pmol (150 ppm). Given that the TMA conversion exceeded the GDL quantitation limit, we would have expected to detect GDL. One possible explanation for this apparent discrepancy is hydrolysis of GDL into gluconic acid, which would prevent GDL accumulation and detection. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Figure 4: Cofactor recycling-driven TMA oxidation in salmon protein hydrolysate. A) Time course analysis of the enzymatic TMA removal process. TMA, TMAO, glucose and GDL were quantified at each time point by LC/MS. Blue: enzyme-treated FPH (presence of mFMO_20/GdhB); red, control FPH (absence of mFMO/GdhB). n=6 for each time point (0, 30, 60, 90, 120 min) for control FPH and n=4 for enzyme-treated FPH. Error bars show standard deviation between replicates. AUC: area under the curve. B) TMA in the enzyme-treated and control FPH evaporates measured by LC/MS (n=3). Error bars show standard deviation between evaporate batches. C) TMA, TMAO, glucose and GDL quantified by LC/MS in dried enzyme-treated FPH and control FPH; n=2 injections of the same sample. Error bars show standard deviation. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint The mobile phase is 10 mM ammonium acetate, pH 8.0. Buffer A has 3% acetonitrile and buffer B has 90% acetonitrile. The column is an ACQUITY Premier BEH Amide VanGuard FIT HILIC column (Waters), with particle size 1.7 µm and internal dimensions of 2.1 x 100 mm, fitted with a 2.1 x 5 mm guard column. 3 µl sample or standard were loaded. The method takes 14 minutes and uses a 0.2 ml/min flowrate, starts at 85% B, going to 80% over 1.6 minutes, then to 40% at 10 minutes, holding at 40% until 12.7 minutes, while final conditioning at 80% B happens between 12.8 and 14 minutes. An Orbitrap Q-Exactive (Thermo Fisher) is connected online to a Dionex UltiMate 3000 UHPLC system (Thermo Fisher). The mass spectrometer is operated in the positive mode for full ion scan, with scheduled switching to negative mode when GDL elutes at approx. 2.2 minutes. Quantification of the selected ions is performed in Excalibur Quant Browser (Thermo Fisher) by taking the area under the curves of the extracted ion chromatograms from samples and standards in defined RT windows (Table S2). The enzyme treatment of the hydrolysates was performed in batches, and these batches served as replicates for the statistical treatment of the quantified ions. Sensory assessment by trained panel Dried enzyme treated and control FPH were assessed by a trained sensory panel at Nofima AS, using a Quantitive Descriptive Analysis (ISO 13299:2016) and according to a Generic Descriptive Analysis (55). The sensory panel consists of 8 judges, trained and satisfying the requirements of ISO 8586-1:2012. The rooms where the test was conducted were built according to ISO 8589:2007, and have individual judging booths, standardized lighting, and its own ventilation system. The judges were presented with the dried hydrolysates (either control or enzyme-treated), blinded as to which sample was which. Samples were served in a blocked and randomized fashion to each judge. 1 gram of sample in white cups with metal lids were served. They were evaluated on the smell parameters of total smell intensity, trimethylamine smell, boullion smell, sweet smell, sour/fermented smell, seaweed smell, feed smell, oxidized smell, and rancid smell. The parameters are described in Table S3. Each judge scored the intensity of the samples 1-9 for each parameter, with 1 being least intense and 9 being most intense. Panel averages were compared by ANOVA, and correction for multiple testing was done by Tukey’s test. Difference between control and enzyme-treated FPH was deemed significant if the corrected p-value was <0.05. Consumer recruitment for sensory testing A total of 70 adult consumers were recruited from among the employees of Leitat Technological Center, located in the Province of Barcelona (Spain), to take part in a sensory evaluation focused on odor perception. Recruitment was carried out internally by sending an email invitation to staff member through the organization’s internal mailing system. The email included a brief description of the study’s objective, the nature of the sensory evaluation (limited to smelling two samples), and the criteria for participation. Eligible participants were adults (≥18 years) with no known olfactory impairments. Employees interested in participating were directed to an online registration form, which collected demographic information (age, gender, location), verified eligibility, and allowed individuals to indicate preferred time slots. Although no strict quotas were applied, efforts were made to ensure demographic diversity among the participants. Before the evaluation, all participants signed an informed consent form, in compliance with ethical research standards approved by the relevant institutional review board. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Consumer panel evaluation During the sensory session, each participant was presented with two blinded hydrolysate samples contained in odor-isolated vessels. Sample A (enzymatically treated) and sample B (no treatment). Both samples have been prepared under the same conditions, ensuring uniformity in quantity and presentation and have been given to the panelists in a room at a constant temperature and isolated from external odors to avoid interferences. Panelists evaluated specific odor-related parameters: fish smell, smell intensity, smell freshness, sulfur like smell and ammonia like smell, using intensity and acceptability scales of seven points. The parameters are described in Table S4, and the intensity descriptions are in Table S5. A preference test between samples was also performed (56). Consumers were asked to rank the samples in order of preference. To obtain the ranking of each sample, each rank position was multiplied by the number of consumers that had selected it, and the sum of the rankings of each sample was calculated. Low values in rank sum of samples indicated that the sample has mainly been ranked in the first order of preference. Acknowledgements All authors received funding from the European Union’s Horizon 2020 research and innovation program under Grant Agreement 101000607 (OXIPRO). LC/MS analyses were performed at the Department of Bioscience, University of Bergen, and the authors thank Ersilia Bifulco (University of Bergen) for support with the LC/MS data collection. Sensory assessment by the trained panel was performed at Nofima AS. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint References 1. Myhre M, Richardsen R, Nystøyl R, Strandheim G. 2024. Analyse marint restråstoff 2023 SINTEF Ocean AS. 2. Kristinsson HG, Rasco BA. 2000. Fish protein hydrolysates: Production, biochemical, and functional properties. Crit Rev Food Sci Nutr 40:43–81. 3. Opheim M, Slizyte R, Sterten H, Provan F, Larssen E, Kjos NP. 2015. Hydrolysis of Atlantic salmon (Salmo salar) rest raw materials - Effect of raw material and processing on composition, nutritional value, and potential bioactive peptides in the hydrolysates. Process Biochem 50:1247–1257. 4. Idowu AT, Benjakul S, Sinthusamran S, Sookchoo P, Kishimura H. 2019. Protein hydrolysate from salmon frames: Production, characteristics and antioxidative activity. J Food Biochem 43:e12734. 5. Ramakrishnan VV, Hossain A, Dave D, Shahidi F. 2024. Salmon processing discards: a potential source of bioactive peptides - a review. Food Prod Process Nutr 6:22. 6. Neves AC, Harnedy PA, O’Keeffe MB, FitzGerald RJ. 2017. Bioactive peptides from Atlantic salmon (Salmo salar) with angiotensin converting enzyme and dipeptidyl peptidase IV inhibitory, and antioxidant activities. Food Chem 218:396–405. 7. Guan M-Q, Yu L, Gu H, Fu Q, Liu M-M, Li K, Yang X-R, Framroze B, Guo J-H, Wei J-J, Li Y-L. 2025. Protein hydrolysate from Atlantic salmon (Salmo salar) improves aging-associated neuroinflammation and cognitive decline in rats by reshaping the gut microbiota and Th17/Treg balance. Int J Biol Macromol 306:141270. 8. Lindsay RC. 1990. Fish flavors. Food Rev Int 6:437–455. 9. Yancey PH. 2005. Organic osmolytes as compatible, metabolic and counteracting cytoprotectants in high osmolarity and other stresses. J Exp Biol 208:2819–2830. 10. Yancey PH, Fyfe-Johnson AL, Kelly RH, Walker VP, Auñón MT. 2001. Trimethylamine Oxide Counteracts Effects of Hydrostatic Pressure on Proteins of Deep-Sea Teleosts. J Exp Zool 289:172–176. 11. Julius K, Al-Ayoubi SR, Paulus M, Tolan M, Winter R. 2018. The effects of osmolytes and crowding on the pressureinduced dissociation and inactivation of dimeric LADH. Phys Chem Chem Phys 20:7093–7104. 12. Barrett EL, Kwan HS. 1985. Bacterial reduction of trimethylamine oxide. Annu Rev Microbiol 39:131–149. 13. Lidbury I, Murrell JC, Chen Y. 2014. Trimethylamine N -oxide metabolism by abundant marine heterotrophic bacteria. Proc Natl Acad Sci U S A 111:2710–2715. 14. Sun J, Mausz MA, Chen Y, Giovannoni SJ. 2019. Microbial trimethylamine metabolism in marine environments. Environ Microbiol 21:513–520. 15. Steinsholm S, Oterhals Å, Thoresen L, Underhaug J, Kousoulaki K, Aspevik T. 2021. Reduction in flavor-intense components in fish protein hydrolysates by membrane filtration. J Food Sci 86:3855–3867. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint 16. Goris M, Puntervoll P, Rojo D, Claussen J, Larsen Ø, Garcia-Moyano A, Almendral D, Barbas C, Ferrer M, Bjerga GEK. 2020. Use of Flavin-Containing Monooxygenases for Conversion of Trimethylamine in Salmon Protein Hydrolysates. Appl Environ Microbiol 86:e02105-20. 17. Goris M, Cea-Rama I, Puntervoll P, Ree R, Almendral D, Sanz-Aparicio J, Ferrer M, Bjerga GEK. 2023. Increased Thermostability of an Engineered Flavin-Containing Monooxygenase to Remediate Trimethylamine in Fish Protein Hydrolysates. Appl Environ Microbiol 89:e00390-23. 18. Paul CE, Eggerichs D, Westphal AH, Tischler D, van Berkel WJH. 2021. Flavoprotein monooxygenases: Versatile biocatalysts. Biotechnol Adv 51:107712. 19. Alfieri A, Malito E, Orru R, Fraaije MW, Mattevi A. 2008. Revealing the moonlighting role of NADP in the structure of a flavin-containing monooxygenase. Proc Natl Acad Sci U S A 105:6572–6577. 20. Choi HS, Kim JK, Cho EH, Kim YC, Kim J Il, Kim SW. 2003. A novel flavincontaining monooxygenase from Methylophaga sp. strain SK1 and its indigo synthesis in Escherichia coli. Biochem Biophys Res Commun 306:930–936. 21. Chen Y, Patel NA, Crombie A, Scrivens JH, Murrell JC. 2011. Bacterial flavincontaining monooxygenase is trimethylamine monooxygenase. Proc Natl Acad Sci U S A 108:17791–17796. 22. Cahn JKB, Werlang CA, Baumschlager A, Brinkmann-Chen S, Mayo SL, Arnold FH. 2016. A general tool for engineering the NAD/NADP cofactor preference of oxidoreductases. ACS Synth Biol 6:326–333. 23. Black WB, Zhang L, Mak WS, Maxel S, Cui Y, King E, Fong B, Sanchez Martinez A, Siegel JB, Li H. 2020. Engineering a nicotinamide mononucleotide redox cofactor system for biocatalysis. Nat Chem Biol 16:87–94. 24. Brinkmann-Chen S, Flock T, Cahn JKB, Snow CD, Brustad EM, McIntosh JA, Meinhold P, Zhang L, Arnold FA. 2013. General approach to reversing ketol-acid reductoisomerase cofactor dependence from NADPH to NADH. Proc Natl Acad Sci U S A 110:10946–10951. 25. Wang X, Saba T, Yiu HHP, Howe RF, Anderson JA, Shi J. 2017. Cofactor NAD(P)H Regeneration Inspired by Heterogeneous Pathways. Chem 2:621–654. 26. Jones JB, Sneddon DW, Higgins W, Lewis AJ. 1972. Preparative-scale Reductions of Cyclic Ketone and Aldehyde Substrates of Horse Liver Alcohol Dehydrogenase with in situ Sodium Dithionite Recycling of Catalytic Amounts of NAD. J Chem Soc Chem Commun 856–857. 27. Maenaka Y, Suenobu T, Fukuzumi S. 2012. Efficient Catalytic Interconversion between NADH and NAD + Accompanied by Generation and Consumption of Hydrogen with a Water-Soluble Iridium Complex at Ambient Pressure and Temperature. J Am Chem Soc 134:367–374. 28. Abril O, Whitesides GM. 1982. Hybrid Organometallic/Enzymatic Catalyst Systems: Regeneration of NADH Using Dihydrogen. J Am Chem Soc 104:1552–1554. 29. Wang X, Yiu HHP. 2016. Heterogeneous Catalysis Mediated Cofactor NADH Regeneration for Enzymatic Reduction. ACS Catal 6:1880–1886. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint 30. Brown KA, Wilker MB, Boehm M, Hamby H, Dukovic G, King PW. 2016. Photocatalytic Regeneration of Nicotinamide Cofactors by Quantum Dot−Enzyme Biohybrid Complexes. ACS Catal 6:2201–2204. 31. Hildebrand F, Kohlmann C, Franz A, Lütz S. 2008. Synthesis, Characterization and Application of New Rhodium Complexes for Indirect Electrochemical Cofactor Regeneration. Adv Synth Catal 350:909–918. 32. Rafter GW, Colowick SP. 1957. Enzymatic Preparation of DPNH and TPNH. Methods Enzymol 3:887–890. 33. Jia Q, Zheng Y-C, Li H-P, Qian X-L, Zhang Z-J, Xu J-H. 2022. Engineering Isopropanol Dehydrogenase for Efficient Regeneration of Nicotinamide Cofactors. Appl Environ Microbiol 88:10.1128/aem.00341-22. 34. Xu Z, Jing K, Liu Y, Cen P. 2007. High-level expression of recombinant glucose dehydrogenase and its application in NADPH regeneration. J Ind Microbiol Biotechnol 34:83–90. 35. Bachosz K, Zdarta J, Bilal M, Meyer AS, Jesionowski T. 2023. Enzymatic cofactor regeneration systems: A new perspective on efficiency assessment. Sci Total Environ 868:161630. 36. Rioz-Martínez A, Kopacz M, Gonzalo G De, Pazmiño DET, Gotor V, Fraaije MW. 2011. Exploring the biocatalytic scope of a bacterial flavin-containing monooxygenase. Org Biomol Chem 9:1337–1341. 37. Pazmiño DET, Snajdrova R, Baas B-J, Ghobrial M, Mihovilovic MD, Fraaije MW. 2008. Self-Sufficient Baeyer–Villiger Monooxygenases: Effective Coenzyme Regeneration for Biooxygenation by Fusion Engineering. Angew Chemie Int Ed 47:2275–2278. 38. van Berkel WJH, Kamerbeek NM, Fraaije MW. 2006. Flavoprotein monooxygenases, a diverse class of oxidative biocatalysts. J Biotechnol 124:670–689. 39. Lee J, Kim J, Song JE, Song WS, Kim EJ, Kim YG, Jeong HJ, Kim HR, Choi KY, Kim BG. 2021. Production of Tyrian purple indigoid dye from tryptophan in Escherichia coli. Nat Chem Biol 17:104–112. 40. Ramachandran S, Fontanille P, Pandey A, Larroche C. 2006. Gluconic Acid: Properties, Applications and Microbial Production. Food Technol Biotechnol 44:185– 195. 41. Heilmann HJ, Mägertmann and col HJ, Gassen HG. 1988. Identification and isolation of glucose dehydrogenase genes of Bacillus megaterium M1286 and their expression in Escherichia coli. Eur J Biochem 174:485–490. 42. Pauly HE, Pfleiderer G. 1975. D-Glucose Dehydrogenase from Bacillus megaterium M 1286 : Purification, Properties and Structure. Hoppe Seylers Z Physiol Chem 356:1613–1623. 43. Baik SH, Ide T, Yoshida H, Kagami O, Harayama S. 2003. Significantly enhanced stability of glucose dehydrogenase by directed evolution. Appl Microbiol Biotechnol 61:329–335. 44. Baik SH, Michel F, Aghajari N, Haser R, Harayama S. 2005. Cooperative effect of two .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint surface amino acid mutations (Q252L and E170K) in glucose dehydrogenase from Bacillus megaterium IWG3 on stabilization of its oligomeric state. Appl Environ Microbiol 71:3285–3293. 45. Kouakou C, Bergé J-P, Baron R, Lethuaut L, Prost C, Cardinal M. 2014. Odor Modification in Salmon Hydrolysates Using the Maillard Reaction. J Aquat Food Prod Technol 23:453–467. 46. Hua L, Qianqian B, Jianfeng Z, Yinbiao X, Shengyu Y, Weishi X, Yang S, Yupeng L. 2022. Directed evolution engineering to improve activity of glucose dehydrogenase by increasing pocket hydrophobicity. Front Microbiol 13:1044226. 47. Chansaenpak K, Kamkaew A, Lisnund S, Prachai P, Ratwirunkit P, Jingpho T, Blay V, Pinyou P. 2021. Development of a Sensitive Self-Powered Glucose Biosensor Based on an Enzymatic Biofuel Cell. Biosensors 11:16. 48. Zhou L, Ouyang Y, Kong W, Ma T, Zhao H, Jiang Y, Gao J. 2022. One pot purification and co-immobilization of His-tagged old yellow enzyme and glucose dehydrogenase for asymmetric hydrogenation. Enzyme Microb Technol 156:110001. 49. Cho CE, Taesuwan S, Malysheva O V., Bender E, Tulchinsky NF, Yan J, Sutter JL, Caudill MA. 2017. Trimethylamine-N-oxide (TMAO) response to animal source foods varies among healthy young men and is influenced by their gut microbiota composition: A randomized controlled trial. Mol Nutr Food Res 61:1600324. 50. Leonardos G, Kendall D, Barnard N, Leonardos G, Kendall D, Barnard N, Threshold O, Leonardos G, Barnard N. 1969. Odor Threshold Determinations of 53 Odorant Chemicals Odor Threshold Determinations Of 53 Odorant Chemicals 2470. 51. Dai W, He S, Huang L, Lin S, Zhang M, Chi C, Chen H. 2024. Strategies to reduce fishy odor in aquatic products: Focusing on formation mechanism and mitigation means. Food Chem 444:138625. 52. Liu L, Zhao Y, Zeng M, Xu X. 2024. Research progress of fishy odor in aquatic products: From substance identification, formation mechanism, to elimination pathway. Food Res Int 178:113914. 53. Zakeri B, Fierer JO, Celik E, Chittock EC, Schwarz-Linek U, Moy VT, Howarth M. 2012. Peptide tag forming a rapid covalent bond to a protein, through engineering a bacterial adhesin. Proc Natl Acad Sci U S A 109:E690–E697. 54. Macheroux P. 1999. UV-Visible Spectroscopy as a Tool to Study Flavoproteins, p. 1– 7. In Methods in molecular biology, vol. 131: Flavoprotein Protocols. 55. Lawless HT, Heymann H. 2010. Sensory Evaluation of Food. Springer New York, New York, NY. 56. Kramer A, Kahan G, Cooper D, Papavasiliou A. 1974. A non-parametric ranking method for the statistical evaluation of sensory data. Chem Senses Flavor 1:121–133. .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Supplementary figures and tables Figure S1: A) Activity of purified GdhB (500 nM) with glucose (50 mM) and the indicated oxidized cofactor (100 µM), and with the indicated reduced cofactor and GDL (50 mM), measured by absorbance change at 340 nm. B) Activity of purified mFMO_20 (50 nM) with 100 µM trimethylamine (TMA) and 100 µM reduced NADPH or nicotinamide adenine dinucleotide (NADH), measured by absorbance change at 340 nm. Table S1: Kinetic parameters of GdhB using glucose as a substrate. NADP+ concentration. Parameters of mFMO_20 are included for reference. Enzyme Vmax (s-1) Km (mM) Specificity constant (s-1 M-1) Reference GdhB 0.429 68.68 (glucose) 6.25 This study mFMO_20 0.93 0.83 x 10-3 (TMA) 1.11 x 106 Goris et al, 2023 Table S2: Metabolites and related analytical information quantified by LC/MS. Metabolite Molecular formula Mass (g/mol) Ion (m/z, charge) Retention time (minutes) TMA C3H9N 59.112 60.08160 (+H) 3.82 TMAO C3H9NO 75.11 76.07630 (+H) 4.58 Glucose C6H12O6 180.156 198.09677 (+NH4) 3.78 GDL C6H10O6 178.14 177.03991 (-H) 2.17 Table S3-S6: Quantification of selected ions by LC/MS. Excel sheet available online. Contains quantitation results and standard curves (Table S3) of TMA, TMAO, glucose and GDL used in Figure 2 (Table S4) and Figure 4 (Table S5-S6). .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint Table S7: Description of smell criteria for dry hydrolysates, employed by trained sensory panel. Parameter Description Total smell intensity Intensity of all smells in the sample Trimethylamine smell The smell of trimethylamine (TMA) Sweet smell Related to a sweet smell Sour/fermented smell A fermented sour smell, spoiled (the smell of a sour dishrag) Mineral smell Related to smells of plaster, lime, chalk and dryness Seaweed smell Related to fresh and dried seaweed and greens, green tea Feed smell Related to the smell of fish feed Oxidized smell Related to an oxidized smell which reminds you of dust, drawers and cardboard Rancid smell The intensity of all rancid smells (grass, hay, candle wax, paint, tallow, soap) Table S8: Hedonic scale description, used in the consumer panel. Score Acceptability description 1 Dislike very much 2 Dislike moderately 3 Dislike slightly 4 Neither like nor dislike 5 Like slightly 6 Like moderately 7 Like very much Table S9: Description of odor parameters used in the consumer panel. Odor parameter Description Fishy smell Degree to which a fishy odor is perceived Smell intensity Overall intensity of the sample’s odor Smell freshness Perception of how fresh the smell is (pleasant, non-stale) Sulfur-like smell Perception of any sulfur-related notes (rotten egg, pungent) Ammonia-like smell Detection of ammonia or any sharp chemical odors .CC-BY 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted June 19, 2025. ; https://doi.org/10.1101/2025.06.16.658034doi: bioRxiv preprint