High-level production of vitamin K2 in Escherichia coli via modular molecular engineering

Abstract

Menaquinone-7 (MK-7), a key form of vitamin K2 with wide-ranging nutritional and pharmaceutical applications, has attracted increasing interest for microbial production. Here, we developed an integrated modular metabolic engineering strategy in Escherichia coli to enhance MK-7 biosynthesis. Cellular membrane capacity and acetate metabolism were rewired to improve precursor supply for the mevalonate (MVA) pathway, while arabinose induction was applied to overexpress three critical enzymes, including BsHepPPS (Bacillus subtilis), EcMenA (E. coli), and BsUbiE (B. subtilis). Among them, EcMenA was identified as a major bottleneck. Rational protein engineering based on folding free energy analysis and consensus design yielded the EcMenA mutant G110W, which produced 102.55 mg/L MK-7 in shake-flask fermentation, a 57.2 % increase compared with the wild-type (WT) enzyme. Further active-site hotspot random mutagenesis generated a G110W-Q57T double mutant, raising MK-7 production to 176.38 mg/L, a 72 % increase compared to the single mutant. Optimization of EcMenA expression cassette by ribosome binding site redesign using a generative network further improved MK-7 titer to 227.53 mg/L in shake flasks. Finally, scale-up fermentation in a 50-L bioreactor, combined with optimized fermentation strategies, achieved a maximum MK-7 titer of 2.18 g/L. This study establishes a systematic framework integrating metabolic rewiring, enzyme engineering, and expression optimization, providing a robust platform for industrial-scale MK-7 production in microbial hosts.

Graphical abstract

1. Introduction

Vitamin K is an essential fat-soluble nutrient for the human body, existing mainly as phylloquinone (Vitamin K1) and a series of menaquinones (Vitamin K2). Vitamin K2 comprises several homologs (MK-n, where n represents the number of isoprenoid units in the side chain), among which menaquinone-7 (MK-7) is considered one of the most biologically active and beneficial form [1]. MK-7 plays a central role in precisely regulating calcium metabolism and maintaining coagulation functions [2]. It strengthens bones and protects blood vessels by activating osteocalcin (OC) and matrix Gla protein (MGP), directing calcium deposition to the bones to increase bone density and prevent osteoporosis [3]. Simultaneously, it inhibits calcium deposition in blood vessel walls, reducing the risk of arteriosclerosis [4]. Given these significant health benefits, MK-7 has garnered substantial commercial interest and represents a high-value product (Fig. 1). The global market for Vitamin K2, driven predominantly by MK-7, was valued at approximately USD 203.4 million in 2024 and is projected to exceed USD 708.7 million by 2034, underscoring its rapid growth and economic importance (Global Market Insights, Report ID: GM13710). Traditional methods of Vitamin K production include plant extraction, chemical synthesis, and microbial fermentation [5,6]. Natto is a rich source of Vitamin K2, containing up to 1200 μg per 100 g. [7,8]. However, the extraction of Vitamin K2 from natto remains costly. Chemical synthesis uses menadione, isoprene, and trans-phytonadione as substrates to successfully produce menaquinone-7 (MK-7) with a purity of 99.9 % [9]. However, the synthesis process can lead to the formation of harmful cis-isomers and byproducts, posing risks to human health and the environment [10]. Additionally, in industrial-scale chemical synthesis of MK-7, only about 30 % of the product is in the highly active all-trans configuration (the natural form is 100 % all-trans) [8]. Given these challenges, biosynthesis holds significant potential for the production of natural metabolic products like Vitamin K2. However, current biosynthesis yields are relatively low, which highlights the need for engineering host strains to enable high-level synthesis of Vitamin K2.

Fig. 1.

Construction and optimization of the MK-7 metabolic pathway in modified E. coli.

Schematic presentation of the MK-7 metabolic pathway in engineered E. coli. The metabolic pathway responsible for MK-7 biosynthesis is primarily divided into three modules (the MVA pathway, DHNA pathway, and MK-7 pathway). Black arrows represent the natural metabolic routes existing in E. coli. Green arrows denote heterologously expressed metabolic pathways in the host. Brown arrows indicate endogenously expressed metabolic routes of host genes. Red text indicates overexpression of the relevant genes on the genome. Blue text signifies overexpression of the relevant genes on plasmids. Purple scissors represent the knockout of the relevant genes. The brown box highlights the primary application areas of MK-7, including promoting bone development in children, preventing osteoporosis, inhibiting vascular calcification, and exerting antioxidant effects, among others. PoxB, pyruvate oxidase; MvaE, HMG-CoA synthase; MvaS, HMG-CoA reductase; MVK, mevalonate kinase; PMK, phosphomevalonate kinase; and MVD, mevalonate pyrophosphate decarboxylase; Idi, IPP isomerase; SAM, S-adenosyl-l-methionine; SAH, S-adenosyl-l-homocysteine; DMK-7, demethylmenaquinone-7; UbiE: ubiquinone/menaquinone biosynthesis methyltransferase; HMG-CoA, 3-hydroxy-3-methyl-glutaryl-coenzyme A; M-5P, phosphomevalonate; M-5PP, diphosphomevalonate; IPP, isopentenyl pyrophosphate; DMAPP, dimethylallyl pyrophosphate; FPP, farnesyl pyrophosphate; MenA, DHNA heptaprenyltransferase; DHNA, 1,4-dihydroxy-2-naphthoate; Heps, heptaprenyl pyrophosphate synthetase I; ispA, Isopentenyl pyrophosphate transferase; MetK, Methionine adenosyltransferase.

The biosynthesis of MK-7 requires two main metabolic pathways: the methylerythritol phosphate (MEP) pathway to produce isopentenyl pyrophosphate (IPP), and the 1,4-dihydroxy-2-naphthoate (DHNA) pathway to produce DHNA [11] (Fig. 1). MK-7 biosynthesis has been extensively studied in Bacillus subtilis [12], while fermentation in this organism typically requires more than 100 h [13,14]. By contrast, Escherichia coli is a well-studied microbial chassis that offers easier genetic manipulation and a shorter fermentation time of less than 50 h [15]. Under anaerobic conditions, E. coli naturally synthesizes three quinone types: ubiquinone-8 (Q-8), menaquinone-8 (MK-8), and demethylmenaquinone-8 (DMK-8), with MK-8 being the predominant product due to the lack of heptaprenyl pyrophosphate synthase (HepPPS). [16]. Gao et al. first expressed a B. subtilis-derived HepPPS (BsHepPPS) in E. coli to enable MK-7 biosynthesis, and further disrupted ubiC, ubiA, and ispB which are key genes in competing pathways, achieving 8.8 mg/L MK-7 production in a 5-L bioreactor [11]. In subsequent work, Gao et al. optimized the expression of key genes, introduced a novel heterologous ubiE, and leveraged E. coli's native DHNA and MEP pathways to produce the MK-7 precursors DHNA and farnesyl pyrophosphate (FPP). Notably, the genetic engineering strategies involving knockout of PoxB and NlpI enhanced MK-7 production by approximately 20 % and 26 %, respectively [17]. In this pathway, HepPPS catalyzes the conversion of FPP to heptaprenyl pyrophosphate (HepPP), MenA converts HepPP to DMK-7, and UbiE methylates DMK-7 to yield MK-7. This strategy achieved 1.35 g/L MK-7 in a 1-L bioreactor. Beyond monoculture approaches, Yang et al. developed a co-culture system combining E. coli and Elizabethkingia meningoseptica, the latter overexpressing octaprenyl pyrophosphate synthase to overcome limitations of single-strain synthesis [18]. This method achieved 32.6 mg/L MK-7 in shake-flask fermentation. Together, these studies have established a strong foundation for using E. coli as a chassis for high-level MK-7 production.

The rapid development of bioinformatics tools has greatly accelerated enzyme engineering, which remains one of the most effective strategies for optimizing microbial biosynthetic capacity [19,20]. Huang et al. identified MenD and MenA as key enzymes in MK-7 biosynthesis in B. subtilis, and introduced mutations to improve enzyme performance [21]. Modifying MenD enhanced both its expression and stability, while introducing the T290M mutation in MenA further improved catalytic activity, resulting in 338.4 mg/L MK-7 production in a 3-L bioreactor. Gao et al. also highlighted MenA and UbiE as promising enzyme targets for E. coli MK-7 optimization, findings that were later experimentally confirmed [17]. Enzyme engineering typically focuses on three main aspects: thermostability, catalytic activity, and solubility [22]. These properties directly influence enzyme persistence in the cell and the total amount of substrate converted. Thermostability engineering often employs strategies such as folding energy optimization, surface charge modification, and consensus design [23]. Catalytic activity engineering is generally more challenging, as it requires detailed molecular dynamics (MD) simulations of enzyme–substrate interactions to identify hot-spot residues for mutagenesis [22,24]. Such enzyme engineering strategies have been widely applied in the metabolic optimization of vitamins, amino acids, and other high-value products [25].

In this study, we sought to further optimize MK-7 biosynthesis in E. coli by building upon previous research and focusing on engineering key enzymes. We first reconstructed the biosynthetic pathway based on earlier studies and knocked out competing metabolic genes [11,17]. We targeted the previously reported key enzyme EcMenA and the downstream enzyme BsUbiE for modification (Fig. 1). We designed to enhance the stability of EcMenA and BsUbiE through folding energy engineering and consensus design, finding that significantly improved MK-7 production by EcMenA modification. Using the most beneficial EcMenA variant, we then applied AI-assisted pocket key binding residue finding to identify hot spots for catalytic activity enhancement. Guided by these predictions, we created 2000 random mutations and identified four variants that markedly increased MK-7 titers. Recognizing the central role of EcMenA, we subsequently optimized its expression cassette by AI-assisted de novo promoter and ribosome binding site (RBS) design. Finally, we scaled up production using the best variant in a 50-L bioreactor, achieving substantially improved MK-7 yields. We conducted MD simulations to further explain the potential catalytic mechanism of EcMenA.

2. Materials and methods

2.1. Chemicals and reagents

Methanol and menaquinone-7 (MK-7) standards, used for high-performance liquid chromatography (HPLC) analysis, were obtained from Sigma-Aldrich (Steinheim, Germany). PCR amplification was carried out with the Vazyme High-Fidelity PCR Kit. The EasyPure® PCR Purification Kit, employed for purifying PCR products, was acquired from TransGen Biotech Co., Ltd. (Beijing, China).

2.2. Strains and plasmid

All bacterial strains and plasmids utilized in this work are provided in Supplementary information.

2.3. Gene manipulation

The key genes involved in menaquinone biosynthesis, including MenA and UbiE and their variants, were cloned into the pSB1a/mena/ubie plasmid for expression in E. coli BL21(DE3). This plasmid was constructed previously by inserting a synthetically synthesized mena/ubie gene cassette into the pBAD-MVA vector at NcoI and XhoI restriction sites. Site-directed mutagenesis was performed to generate specific mutants of MenA, UbiE, and their variants, utilizing PCR amplification with the pSB1a/mena/ubie plasmid as the template. The primers for mutagenesis are listed in Supplementary information. The amplified mutant fragments were purified using the EasyPure® PCR Purification Kit (TransGen Bio, Beijing, China). To modify the host strain and optimize precursor supply, the NlpI and PoxB genes were knocked out via CRISPR-Cas9 gene editing as follows: sgRNAs targeting each gene were designed using the CHOPCHOP online tool and cloned into the pTarget plasmid. The donor DNA fragments, containing homologous arms flanking the deletion regions, were synthesized and co-electroporated with the pCas9 plasmid into E. coli. After recovery, cells were plated on LB agar containing appropriate antibiotics, and successful knockout mutants were screened by colony PCR followed by DNA sequencing to confirm the deletions.

The acs gene, responsible for converting acetate to acetyl-CoA, was overexpressed by replacing its native promoter with the strong, constitutive J23119 promoter. This was achieved through a homologous recombination-based promoter replacement, following standard techniques. The resulting recombinant strain contained the NlpI and PoxB deletions, along with acs overexpression. Additionally, this study tested multiple promoters for driving the expression of the ECidi gene. The ECidi gene was cloned downstream of these promoters to construct expression units, which were then co-transformed into host strains along with plasmids carrying EcMenA, BsUbiE mutants and other MK-7 pathway genes. The stationary-phase promoters used in this study included P_katE, P_fic, and P_osmY, with reference to the research by Wang et al. [26]. Furthermore, two additional promoters, P_Bad [27] and J23119 [28], were also employed for comparative analysis.

All gene cloning, mutagenesis, and strain modifications were confirmed by PCR and DNA sequencing. The engineered strains were used for subsequent fermentations to evaluate MK-7 production.

2.4. Cultivation

E. coli strains were cultured in LB (Luria–Bertani) medium at 37 °C with shaking at 220 rpm. Antibiotics were added as necessary, including ampicillin at a concentration of 50 mg/L. For protein expression, starter cultures were transferred into TB medium at a 1 % inoculation rate and grown overnight at 30 °C with shaking at 220 rpm. The medium was supplemented with 0.2 % arabinose and the appropriate antibiotic (100 mg/L ampicillin). For the bioconversion of MK-7, cells were harvested following an 18 h cultivation period via centrifugation at 6000×g for 5 min. The pellet was then resuspended in M9 medium (pH 7.0) supplemented with 20 g/L glucose, 12.8 g/L Na₂HPO₄·7H₂O, 3 g/L KH₂PO₄, 0.5 g/L NaCl, 1 g/L NH₄Cl, 0.1 mM CaCl₂·2H₂O, and 5 mM MgSO₄·7H₂O, adjusting the cell density to an OD₆₀₀ of either 10 or 20. The bioconversion process was carried out at 37 °C under continuous shaking at 220 rpm for a duration of 10 h.

2.5. Fermentation of MK-7 production in 50-L bioreactor

Fermentation for MK-7 production was conducted in a 50-L bioreactor (Womei biology, Shanghai, China). A monoclonal colony was first inoculated into 75 mL of LB medium for seed culture and grown overnight. This pre-culture was then transferred into 1500 mL of LB medium. Once the OD₆₀₀ reached 3, the cells were inoculated into a 50 L minimal medium containing 20 g/L glucose, 14 g/L KH₂PO₄, 4 g/L (NH₄)₂HPO₄, and 1.8 g/L citric acid monohydrate. Fermentation was initiated at 37 °C. When the OD₆₀₀ attained 70, a 100 × arabinose solution (200 g/L) was added in a single dose to achieve a final concentration of 2 g/L in the broth, thereby inducing protein expression at 30 °C. After 12 h of induction, the temperature was shifted to 37 °C to promote MK-7 synthesis. Throughout the process, the pH was maintained at 7.0, and dissolved oxygen (Li et al.) was kept between 20 % and 60 % by modulating agitation speed and the feeding rate. This range achieves an optimal balance between avoiding oxygen limitation and oxidative damage, thereby supporting both cell growth and product synthesis. The initial glucose concentration in the fermenter was 20 g/L. Glucose feeding commenced at 9 h and was carefully controlled to maintain low residual levels throughout the fermentation, ensuring optimal activity of the pBAD promoter.

2.6. Analytical methods

Cell growth was assessed by recording the optical density at 600 nm (OD₆₀₀). To extract total MK-7, E. coli cells were collected via centrifugation at 12,000 rpm and room temperature, followed by a single wash with distilled water. MK-7 was isolated using a mixture of n-hexane and 2-propanol (2:1, v/v). Quantitative analysis was carried out using an HPLC system equipped with a UV detector set at 254 nm and a reversed-phase column (Thermo Fisher Scientific, Hypersil™ ODS C18; dimensions: 4.6 × 250 mm, particle size: 5 μm). Separation was conducted at 30 °C with 100 % methanol as the mobile phase, pumped at a flow rate of 1.0 mL/min. The injection volume was 10 μL.

2.7. Computational analysis

The folding free energy calculations were performed using Rosetta Cartesian_ddg and custom Python scripts[29], which help assess the flexibility and rigidity of the protein structure. By analyzing homologous sequences through consensus design provided by a web server, non-conserved residues were identified, and site-directed mutagenesis was conducted based on the scoring results. The binding pocket of the enzyme was characterized using PrankWeb [30].

2.8. MD simulation

The 3-D structure of EcMenA was modeled using AlphaFold-2 (AF-2) [31]. MD simulations were conducted in Gromacs-2020 [32] with the FF14SB force field [33] applied to describe the protein. The system was solvated within a cubic box of SPC/E water molecules, maintaining a minimum distance of 12 Å from the protein to the box edges. Following neutralization with Na⁺ or Cl⁻ ions, energy minimization was performed via the steepest descent algorithm. Subsequently, equilibration was carried out under NVT and NPT ensembles at 300 K for 100 ps and 200 ps, respectively. A production MD run of 100 ns was executed at the same temperature, and the resulting trajectories were subjected to analysis.

3. Results

3.1. Establishing MK-7 biosynthesis in E. coli

This study builds upon previous reports to optimize the biosynthesis of MK-7 in E. coli. Based on studies by Gao et al. and the membrane-accumulating nature of MK-7, knockout of the NlpI gene in Escherichia coli helps maintain cell membrane structure to increase MK-7 storage capacity, thereby indirectly enhancing its production. Additionally, knockout of the PoxB gene reduces acetate accumulation, a byproduct of pyruvate oxidase activity that disrupts normal cellular metabolism, which is also beneficial for MK-7 synthesis. Therefore, we knocked out both genes to achieve effects consistent with those reported by Gao et al. [11,17]. We also overexpressed acetyl-CoA synthetase (acs), an enzyme that converts acetate into acetyl-CoA. This strategy reduces acetate accumulation while increasing the availability of acetyl-CoA for the mevalonate (MVA) pathway (Fig. 1). The PoxB knockout and acs overexpression were achieved by replacing their native promoter with the strong, constitutive J23119 promoter.

For precursor supply, we focused on the MVA pathway to provide isopentenyl pyrophosphate (IPP) and the DHNA pathway to provide DHNA (Fig. 1). The MVA pathway, incorporating EfmvaE, EfmvaS, MmmvK, ScpmK, and ScmvD, was assembled into a single expression cassette and integrated into the genome at the msbB site. The enzyme Idi, which converts IPP to farnesyl pyrophosphate (FPP), was crucial for further conversion by HepPPS to form heptaprenyl pyrophosphate (HepPP) (Fig. 1). While previous studies have used the pBad promoter to overexpress ECidi, in the present study we tested and compared three different promoters for ECidi expression: the strong constitutive J23119 promoter, the stationary-phase promoter P_katE, and the pBad promoter.

Finally, the key genes BsHepPPS, EcMenA, and BsUbiE were cloned in sequence into the pBad plasmid, following the same gene order as that used by Gao et al. [17], to ensure a consistent basis for further investigation. Among them, BsHepPPS encodes heptaprenyl pyrophosphate synthase, which catalyzes the conversion of farnesyl pyrophosphate to heptaprenyl pyrophosphate; EcMenA encodes a prenyltransferase that catalyzes the conjugation of heptaprenyl pyrophosphate with DHNA to form demethylmenaquinone; and BsUbiE encodes a methyltransferase that catalyzes the methylation of demethylmenaquinone to produce MK-7. Since the catalysis occurs after cell density accumulation, ensuring that all precursors are available, we tested alternative stationary-phase promoters including P_katE, P_osmY and P_fic to enhance enzyme expression [24]. The plasmid was transformed into E. coli ΔNlpIΔPoxB::acs, along with pBad-ECidi, J23119-ECidi, and P_katE-ECidi constructs. Our results indicated that overexpression of ECidi driven by J23119 and pBad promoters yielded similar results, both outperforming the stationary-phase promoter P_katE (Fig. 2). Moreover, induction of key catalytic enzymes (BsHepPPS, EcMenA, and BsUbiE) using the pBad promoter resulted in nearly three times higher expression levels than with the stationary-phase promoter P_katE (Fig. 2A). The highest-producing strain achieved 65.23 mg/L MK-7 production (Fig. 2B).

Fig. 2.

Schematic diagram of MK-7 production in strains with different promoter combinations.

(A) MK-7 yield resulting from combinations of different genomic promoters and plasmid-based promoters. (B) Schematic representation of the promoter engineering combinations. The blue bars below delineate the expression loci: Ecidi enzyme is genomically integrated with three distinct promoters tested, while the key catalytic enzyme is plasmid-borne with four different promoters evaluated. In the promoter combinations, the term before the plus sign denotes the type of promoter integrated into the genome, while the term after the plus sign indicates the promoter located on the plasmid.

3.2. Identification of MenA as a key enzyme in MK-7 synthesis

The pathway we developed demonstrated the capability to biosynthesize MK-7, with the three catalytic enzymes BsHepPPS, EcMenA, and BsUbiE receiving FPP, IPP, and DHNA to synthesize MK-7. While MenA is known to be central to this pathway, its role as a rate-limiting enzyme has not been confirmed [17]. To address this, we focused on EcMenA and BsUbiE, both direct contributors to MK-7 synthesis, to evaluate their catalytic dominance. Increasing an enzyme's stability by making its structure more rigid extends its half-life, improves stress tolerance, and boosts soluble expression, thereby expanding the functional enzyme pool and overall catalytic capacity. A resulting increase in product titer upon implementing a stabilized variant defines this enzyme as the pathway's flux bottleneck. We engineered these enzymes using rational design strategies to improve their stability. To enhance enzyme stability, we used Rosetta ddG to engineer structural rigidity and employed a consensus design approach for evolutionary design targeting stability. We performed virtual whole-residue saturation mutagenesis using Rosetta ddG, selecting the top 20 candidates based on energy ranking. Based on the inherent properties of the enzymes and to enhance the comprehensiveness of our screening, we selected variants with maximum ddG values greater than −4 and −3 for EcMenA and BsUbiE, respectively (Fig. 3A). Consensus design was conducted, selecting conservation rates of 65 % and 80 % for EcMenA and BsUbiE (Fig. 3B), respectively. Single mutations were performed on both enzymes, revealing that none of the BsUbiE variants showed more than a 10 % increase in MK-7 production. In contrast, three variants of EcMenA contributed more than 30 % to MK-7 yield, with G110W achieving 102.55 mg/L of MK-7 in shake flask fermentation, a 57.21 % increase over the WT strain (Fig. 3C).

Fig. 3.

Analysis of key enzymes in the MK-7 biosynthetic pathway.

(A) Bar chart of the top twenty ddG values selected for MenA and UbiE mutants, respectively, after virtual comprehensive residue saturation mutagenesis using Rosetta ddG. (B) Schematic diagram of sequence conservation design for MenA. (C) MK-7 production levels in strains harboring single-point mutations in MenA or UbiE. The left panel shows EcMenA mutant strains, with orange bars representing MK-7 production and light green curves indicating the corresponding OD₆₀₀ values. The right panel displays BsUbiE mutant strains, with green bars representing MK-7 production and blue curves indicating the corresponding OD₆₀₀ values. Error bars represent standard deviation. ddG denotes the change in free energy difference.

Having identified EcMenA as the key rate-limiting enzyme and obtained the significantly stabilized G110W variant, our subsequent efforts focused on further enhancing its catalytic activity through semi-rational design.

3.3. Engineering the activity of MenA

Building on the beneficial variants, we sought to further enhance the activity of EcMenA. Using the structure- and AI-based binding site prediction tool PrankWeb [30], we identified key binding pockets in the enzyme. Visualization of the predicted pocket revealed overlap with catalytic residues previously reported [34] (Fig. 4A). For semi-rational design, we selected eight residues including V287, Q57, S60, A138, L283, V158, L197, and G126 within the predicted binding region as mutational hotspots (Fig. 4A). Random mutagenesis was performed on these sites in the G110W, generating a library of 1000 variants. Initial screening based on experimental and liquid chromatography data identified nine mutants with MK-7 yield improvements exceeding 10 % (Fig. 4B). Upon re-screening, seven of these consistently showed improvements above 10 % (Fig. 4B). Notably, four mutants exhibited activity enhancements greater than 50 %, with the best-performing variant Q57T achieving a 72 % increase in MK-7 production, reaching 176.38 mg/L in shake-flask fermentation (Fig. 4B). In the second round, saturation mutagenesis was conducted on the G110W-Q57T double mutant, and another 1000 variants were screened. However, no significant improvement in MK-7 production was observed beyond the levels achieved in the first round (Fig. 4C).

Fig. 4.

Engineering of the active site in EcMenA.

(A) The key binding pocket (yellow) in EcMenA was visualized using PyMol based on structure- and AI-based binding site prediction via PrankWeb. Eight residues within the active pocket were selected as mutation hotspots (blue). (B) Violin plot displaying the production distribution of 1000 mutants from the first round of high-throughput screening. Nine beneficial sites were identified. Bar chart shows the production levels of these nine beneficial sites upon re-screening. With the exception of G110W, all subsequent variants were double mutants. (C) Violin plot showing the production distribution of another 1000 mutants from the second round of high-throughput screening.

With the high-performance EcMenA variant G110W-Q57T successfully engineered through multiple rounds of stability and activity improvements, we next sought to evaluate the performance of these engineered strains under conditions more relevant to industrial production by turning to the optimization of the fermentation process.

3.4. Rational design of the expression cassette to accelerate recombinant expression of MenA

Given the essential role of EcMenA, we further optimized its expression cassette to improve recombinant protein production. The pBad promoter was retained to allow induction of EcMenA expression, ensuring effective enzyme activity during the stationary phase. Instead, we focused on enhancing translation efficiency by redesigning the RBS. The RBS region was computationally optimized and ligated into the original plasmid (Fig. 5A). Employing an AI-based algorithm [35] that accounts for both upstream and downstream contextual sequences, we generated 100 novel RBS variants and validated the top 10 highest-ranked sequences, predicting their capacity for expression maximization. The feature sequences of the top ten RBSs are listed in Fig. 5A alongside their corresponding MK-7 production levels, which reflect their relative strengths. Among these, six were found to enhance MK-7 accumulation. The top-performing RBS variant increased MK-7 production by 29 %, achieving a titer of 227.53 mg/L in shake flask cultures (Fig. 5A). Additionally, we evaluated a separate pBad promoter for expressing BsHepPPS and BsUbiE following the EcMenA cassette. This configuration slightly improved MK-7 production compared to a polycistronic cassette containing all three genes.

Fig. 5.

Enhanced recombinant expression efficiency, increased MK-7 production through optimization of the expression cassette and schematic diagram of the fed-batch fermentation process of the engineered strain in a 50-L fermenter.

(A) The RBS sequence between the pBad promoter and the gene EcMenA was predicted using an AI-based tool. Due to the prediction method accounting for both upstream and downstream regions of the RBS, the full sequence is relatively long; only characteristic segments are shown in the figure. The MK-7 production levels in the bar graph correspond to the top ten RBS sequences on the left. (B) The time-course profiles illustrate the glucose feeding rate (orange line, DO level (blue line), OD₆₀₀ (brown line), and MK-7 production (red line) over time. When the OD₆₀₀ reached 70, arabinose was added to a final concentration of 2 g/L for induction, and the temperature was adjusted to 30 °C. After 12 h of induction, the temperature was shifted to 37 °C to further promote MK-7 synthesis. Throughout the process, the pH was maintained at 7.0. Glucose feeding was initiated after 9 h and was carefully controlled to maintain low residual levels.

3.5. MK-7 production in 50-L bioreactor

Based on the optimized strategies developed in previous sections, we constructed the final engineered strain by integrating the superior EcMenA variant and the optimized expression cassette with the enhanced RBS. To evaluate the performance of this strain and achieve high-level MK-7 production, we conducted fed-batch fermentation in a 50-L bioreactor (Fig. 5B). The fermentation process was initiated at 37 °C to promote rapid cell growth. Feeding began at 9 h post-inoculation when the OD₆₀₀ reached approximately 40. The feeding rate was dynamically controlled based on real-time residual glucose levels to maintain a low concentration, ensuring balanced nutrient supply and metabolic stability. To boost biomass and avoid resource competition during enzyme production, induction was initiated at an OD₆₀₀ of 90 by shifting the temperature to 30 °C and adding l-arabinose (2 g/L). This induction strategy ensured that the key enzymes (particularly EcMenA) were expressed during the stationary phase, minimizing metabolic burden during the growth phase.

After 12 h of induction, the temperature was restored to 37 °C to initiate in situ whole-cell catalysis, thereby facilitating the efficient conversion of accumulated precursors into MK-7. The fermentation process was continuously monitored over time. After 64 h, the MK-7 titer peaked at 2.06 g/L, while the OD₆₀₀ began to decline after reaching a maximum value of 101. Subsequently, within an additional 8 h, the DO level rebounded to 95 %, indicating markedly reduced metabolic activity and the onset of stationary phase. At this point, the fermentation was terminated. The final MK-7 production reached 2.18 g/L, a 9-fold increase from the shake-flask level of 227.53 mg/L. This significant enhancement clearly demonstrates the scalability of both our engineered strain and optimized process strategies. The high titer achieved can be attributed to the synergistic effects of EcMenA engineering, metabolic pathway balancing, and precisely controlled fermentation conditions.

3.6. MD analysis of combination mutants

To further elucidate the mechanism behind the enhanced enzyme stability conferred by the mutation, MD simulations were performed for both the WT and the G110W-Q57T mutant. As shown in Fig. 6A, during extended simulation times, the mutant exhibited longer periods of low RMSD values. Both the average and final RMSD values of the WT (average: 0.3450; final: 0.3931) were higher than those of the EcMenA mutant (average: 0.2842; final: 0.2675). These results indicate that the mutant possesses greater structural rigidity, thereby improving enzyme stability. However, RMSF analysis revealed no significant changes in the residual flexibility at the mutation site between the mutant and the WT (Fig. 6B). Notably, a decrease in RMSF values was observed in other regions, particularly in the adjacent peptide chains near the mutation site within the three-dimensional structure. Following the stability-driven conservative design, and taking the spatially adjacent residues A49 and S166 with reduced RMSF as an example, the spatial distance between them decreased from 9.9 Å and 14.8 Å to 6.4 Å and 8.5 Å, respectively (Fig. 6C). Hydrogen bond frequency analysis showed that the average number of hydrogen bonds formed near these two residues adjacent to the mutation site in the mutant was higher 5.07, compared to only 4.68 in the WT. The mutation may have optimized the local conformational structure, allowing the backbones or side chains of adjacent residues to adopt more favorable positions, thereby facilitating the formation of more and stronger hydrogen bonds. This "pre-organized" structure exhibits reduced dynamics. The involvement of adjacent residues in a more stable hydrogen-bonding network restricts their motion, ultimately enhancing the overall stability of the enzyme.

Fig. 6.

Molecular dynamics simulation analysis of EcMenA and the double mutant.

(A) RMSD plot of the EcMenA (WT, black line) and G110W-Q57T (red line). (B) RMSF plot of the EcMenA (black line) and G110W-Q57T (red line). (C) Distance between the mutation site G110 and the adjacent peptide chain. Left: schematic of the EcMenA structure; Right: schematic of the double mutant. Residues A49, S166, G110, and G110W are shown in blue, and spatial distances are indicated with yellow dashed lines. (D) Illustration of the distance within the aspartate-rich structural region near the active pocket of EcMenA. Left: structure of the EcMenA; Right: structure of the double mutant. Residues D68 and D208 are colored blue, and spatial distances are marked with yellow dashed lines.

Subsequently, to further investigate the structural changes near the active pocket resulting from the high-throughput screening-derived mutation, cluster analysis was conducted for both the mutant and the WT with an RMSD cutoff of 0.18 nm (i.e., 1.8 Å). This means that two structures were considered similar and grouped into the same cluster if their RMSD was below this threshold. This process yielded 54 structural clusters for the mutant and 69 for the WT. The most predominant conformational states of the enzymes in solution were selected for further analysis. These clusters are representative and suggest that the enzyme predominantly interacts with the substrate in this form. Previous MD simulations of BsMenA revealed that enzymes in this family possess an essential aspartate-rich motif near the active pocket[34]. A similar motif was observed in EcMenA (D68xxxxxD74, D204xxxD208), with D68 and D208 being spatially corresponding. By comparing the spatial distances between these corresponding aspartate residues near the gate of the active pocket in the WT and the mutant, we found that the distance decreased from 8.2 Å to 6.0 Å, resulting in a narrower gate structure (Fig. 6D). This change may enhance the enzyme's ability to grasp the substrate, thereby contributing to the improved activity of the mutant. The insights gained from this study offer a theoretical basis for the rational design of this enzyme in the future.

4. Discussion

In this study, we engineered the central metabolism of E. coli by knocking out nlpI and poxB and overexpressing acs, while integrating a heterologous MVA pathway and optimizing DHNA supply to ensure sufficient precursor availability. A modular engineering approach was employed to systematically construct and optimize the MK-7 biosynthesis pathway. Subsequently, based on previous studies, conservative residue mutations were introduced into EcMenA and BsUbiE to enhance their stability. However, following the stability-enhancing mutations, only the EcMenA mutants led to an increase in MK-7 production, whereas mutations in BsUbiE did not significantly improve the yield, confirming that EcMenA is the key rate-limiting enzyme. Therefore, we further employed PrankWeb for structure- and AI-based prediction of the active pocket in EcMenA, which identified the key residues. Through semi-rational design and saturation mutagenesis, the double mutant G110W-Q57T further enhanced the production. However, subsequent high-throughput screening did not yield any mutants with further improved activity. We switched to enhancing EcMenA expression via rational RBS design. An AI-assisted algorithm generated 10 novel RBS sequences, with the best performer increasing MK-7 production to 227.53 mg/L in flasks. Finally, the production performance of the top-performing strain was evaluated and validated in a 50-L bioreactor, achieving a yield of 2.18 g/L of MK-7, which demonstrates its promising potential for industrial application.

A key finding of this work is the identification and modification of EcMenA as a critical bottleneck in MK-7 biosynthesis. Previous studies have demonstrated that MK-7 can be produced in microbial systems through the integration of multiple metabolic pathways [36,37]. Previous research centered on metabolic and enzyme engineering in B. subtilis [38,39], alongside membrane engineering [40], yielding notable progress. Gao et al. constructed a biosynthetic pathway for MK-7 in E. coli and proposed that UbiE and MenA [11] might represent key metabolic nodes requiring optimization. The following year, the same team significantly improved MK-7 production through metabolic engineering and the combined overexpression of endogenous MenA and heterologous UbiE [17], further underscoring the importance of these two enzymes. Subsequently, Huang et al. engineered a stability-enhanced mutant of MenA (T290M) through screening and combined it with a variant of MenD [21], leading to increased MK-7 production. In our study, for the first time, we directly compared the effects of directed evolution on both EcMenA and BsUbiE, clearly demonstrating the pivotal role of MenA in MK-7 synthesis. The G110W mutant of EcMenA alone increased MK-7 production by 57 %, and a combination of mutations further enhanced the yield by 72 %.

Enzyme engineering is a core strategy in synthetic biology for enhancing the performance of microbial cell factories and the titers of target products. Through rational design, directed evolution, and artificial intelligence-assisted prediction, key enzymes can be precisely optimized. Wang et al. successfully obtained mutants of l-sorbosone dehydrogenase (SNDH) with significantly enhanced activity through a computer-aided semi-rational design approach [41], which included expanding the entrance of the substrate pocket and engineering the residues within it. However, improving enzymatic activity remains more challenging than enhancing stability. Luo et al., for instance, identified non-conserved residues near the active site of 3α-HSDH for saturation mutagenesis [42], employed deep learning to predict mutant K_cat values, and experimentally validated a variant with a 10-fold increase in catalytic efficiency after multiple rounds of screening. In line with these approaches, our study also utilized bioinformatic tools (including AI-assisted binding site prediction via PrankWeb and AI-driven RBS design [43,44]), coupled with semi-rational design and expression optimization, to significantly improve MK-7 production in EcMenA-mutant strains.

To elucidate the molecular mechanism by which the beneficial mutations enhance enzymatic function, molecular dynamics simulations were performed on both the wild-type EcMenA and the G110W-Q57T mutant. The results revealed that the mutant exhibits a lower overall RMSD, indicating enhanced structural rigidity. This provides a more stable catalytic scaffold, enabling key catalytic residues to maintain their optimal spatial orientations more persistently, thereby potentially reducing the activation energy barrier of the reaction. Concurrently, a key spatial distance at the entrance of the active site pocket, such as that between D68 and D208, decreased from 8.2 Å to 6.0 Å, resulting in a narrower gate structure. This structural change may not only improve substrate specificity but also compel the substrate to adopt a conformation more closely resembling the transition state upon binding, thereby optimizing catalytic efficiency. These two characteristics mutually reinforce each other: the enhanced global rigidity stabilizes the constricted gate conformation, while the strengthened local hydrogen-bonding network provides the structural basis for this stabilization. Collectively, this "pre-organized" effect creates an active center with higher geometric precision and reduced dynamic fluctuations, fundamentally optimizing the catalytic microenvironment of EcMenA and leading to a significant increase in production.

In conclusion, this study demonstrates the effective enhancement of MK-7 production in E. coli using integrated metabolic engineering and enzyme evolution strategies. We systematically identified EcMenA as a pivotal bottleneck and improved its catalytic performance through structure-informed mutagenesis, achieving a significant increase in titer. Despite these improvements, limitations remain, such as potential redox cofactor imbalances and the need for finer dynamic regulation of precursor pathways. Additionally, the current strain may exhibit reduced genetic stability under prolonged fermentation. Future work will focus on deploying these engineered modules in more robust microbial hosts, applying real-time metabolic flux control, and leveraging machine learning for high-throughput enzyme design. Further efforts will also aim to streamline the entire synthesis pathway toward industrial-scale production, ultimately strengthening the biomanufacturing platform for vitamin K₂ and related high-value products.

CRediT authorship contribution statement

Zelin Lu: Writing – review & editing, Writing – original draft, Validation, Methodology, Investigation, Formal analysis. Zhongshi Huang: Writing – review & editing, Software, Formal analysis, Data curation, Conceptualization. Zhengyin Wu: Writing – review & editing, Software. Zhengwen Zhu: Writing – review & editing, Software, Methodology, Formal analysis. Yibo Zhu: Validation, Investigation, Formal analysis, Conceptualization. Xiaonuo Teng: Validation, Methodology, Investigation. Huyang Chen: Supervision, Resources, Formal analysis.Jingwen Zhou: Writing – review & editing, Resources, Methodology. Fuqiang Ma: Writing – review & editing, Resources, Conceptualization. Xinglong Wang: Writing – review & editing, Supervision, Resources, Project administration, Methodology, Funding acquisition.

Data availability statement

The authors declare that all data supporting the findings of this study are available in the article and its supplementary files.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Xiaonuo Teng and Huyang Chen are currently employed by Suzhou Womei biology Co., Ltd.

Appendix A. Supplementary data

The following is the supplementary data to this article: