Metabolic profiling analysis of the vitamin B12 producer Propionibacterium freudenreichii

Abstract Vitamin B12 (VB12) is an indispensable cofactor of metabolic enzymes and has been widely used in the food and pharmaceutical industries. In this study, the effects of medium composition on VB12 production by Propionibacterium freudenreichii were evaluated and optimized based on statistical experiments. The results showed that glucose, yeast extract, KH2PO4, and glycine have significant effects on VB12 production. The final titer of VB12 reached 8.32 ± 0.02 mg/L, representing a 120% increase over the non‐optimized culture medium. We employed a metabolomics approach to analyze the differences of metabolite concentrations in P. freudenreichii cells cultivated in the original medium and optimized fermentation medium. Using multivariate data analysis, we identified a range of correlated metabolites, illustrating how metabolomics can be used to explain VB12 production changes by corresponding differences in the overall cellular metabolism. The concentrations of many metabolic intermediates of glycolysis, the Wood–Werkman cycle, the TCA cycle, and amino acid metabolism were increased, which contributed to the synthesis of propionic acid and VB12 due to an improved supply of energy and precursors.


| INTRODUC TI ON
, cobalamin), the only vitamin containing a metal element, is an essential cofactor of key enzymes that catalyze crucial biological activities such as the synthesis of DNA, amino and fatty acids in living organisms (And & Ragsdale, 2003;Eschenmoser, 1988;Fang et al., 2017). However, humans and other animals cannot synthesize VB 12 de novo (Nielsen et al., 2012). Accordingly, nutritional deficiencies of VB 12 can lead to a variety of complications in humans, including neuropsychiatric symptoms and various forms of cancer (Berg et al., 2013;Lechner et al., 2005). Therefore, VB 12 has been widely used in the pharmaceutical, food, and feed industries, due to its special nutritional and economic value.
The industrial production of VB 12 is mainly dependent on the fermentation of microorganisms that can produce VB 12 via a de novo synthesis pathway (Kojima et al., 1993;Rodionov et al., 2003;Swithers et al., 2012). Pseudomonas denitrificans and Propionibacterium freudenreichii are the most widely used VB 12 producers in industrial applications (Martens et al., 2002). Previous reports revealed two very complex biosynthetic pathways of VB 12 , named the aerobic and anaerobic pathway, both of which comprise at least 25 steps with uroporphyrinogen III as the same initial precursor . The adenosylcobalamin molecule consists of three parts: the upper ligand, the central corrin nucleus, and the lower ligand ( Figure A1). Fang et al. (2017) reviewed both pathways of VB 12 biosynthesis. The first committed precursor of the central corrin nucleus is 5-aminolaevulinic acid, which can be synthesized either from glycine and succinyl-CoA in the C4 pathway or from glutamate in the C5 pathway. Uroporphyrinogen III is synthesized from eight molecules of 5-aminolaevulinic acid. After cobalt chelation and eight peripheral methylation reactions with adenosylmethionine as the methyl donor, cob(II)yrinic acid a,c-diamide is formed. Then, the upper ligand is attached to the cobalt atom of cob(II)yrinic acid a,c-diamide to form adenosyl cobyrinic acid a,c-diamide. After four stepwise amidation reactions at different carboxyl groups, adenosyl cobyric acid is produced from adenosyl cobyrinic acid a,c-diamide.
The lower ligand of VB 12 is derived from aminopropanol, nicotinic acid mononucleotide, and DMBI. Although P. freudenreichii can de novo synthesize DMBI from riboflavin, oxygen is required for this process (Fang et al., 2017). Thus, Propionibacterium needs both anaerobic and aerobic conditions for effective vitamin B 12 production.
Furthermore, many studies added DMBI to the medium as an important precursor (Guo & Chen, 2018).
Adenosylcobalamin is coproduced by P. freudenreichii together with the main product propionic acid (PA). Many studies attempted to increase the production of PA and VB 12 in Propionibacterium (Belgrano et al., 2018;Suwannakham & Yang, 2005). Several studies had attempted to increase VB 12 production by adding some precursors such as cobalt ions (Quesada-Chanto et al., 1994;Seidametova et al., 2004;Yongsmith et al., 1982) and DMBI (Marwaha & Sethi, 1984). In addition to these two factors, other media components also a play significant role in VB 12 production, including the carbon source, yeast extract, casein hydrolysate, calcium pantothenate, NaH 2 PO 4 , and so on. Accordingly, many research efforts on improving the synthesis of VB 12 are based on optimizing fermentation processes to reduce cell growth inhibition by PA, such as cell immobilization and in situ product removal. . The anaerobic synthesis pathway of VB 12 in Propionibacterium has been studied intensively. Several reports showed that VB 12 production could be increased by the homologous or heterologous overexpression of genes located in the hem, cob, and cbi operons, or gene clusters involved in VB 12 synthesis, but the concentration was limited to between 0.96 and 1.68 mg/L (Piao et al., 2004). Little progress has been made on improving the VB 12 yield of Propionibacterium. Further studies are needed to discover the potential metabolic regulation mechanisms, which would help remove the bottlenecks of VB 12 production in Propionibacterium.
As a powerful analytical tool, metabolomics has been used to investigate the global metabolism of Propionibacterium for improving PA production (Cardoso et al., 2007;Choi & Mathews, 1994). Guan et al. found the key metabolic nodes affecting PA production by comparing metabolic profiles of wild-type and genome-shuffled mutant strains of Acidipropionibacterium acidipropionici, and then, the addition of key exogenous metabolites (precursors and amino acids) could increase the PA titer from 23.1 ± 1.2 to 35.8 ± 1.0 g/L (Guan et al., 2015). However, metabolomic studies of P. freudenreichii for enhancing VB 12 production have not been reported.
In this study, P. freudenreichii CICC 10019 was used for VB 12 production with the fermentation medium reported by  as the original medium. As reported before, glycerol and amino acids, such as glycine, may be potential bottlenecks in VB 12 biosynthesis.
Thus, glycerol was added as the carbon source, yeast extract as the amino acid source, and glycine as the precursor of methyl group donors to optimize the fermentation medium. Fermentation medium optimization was first performed using a statistical experimental design to analyze the effects of key medium components on VB 12 production by P. freudenreichii. Then, comparative metabolic profiling was applied to investigate the metabolic changes of P. freudenreichii grown on the original and the optimized fermentation medium. We then analyzed the intracellular metabolite concentrations at three time points of the fermentation process (logarithmic growth phase, stationary phase, and VB 12 production phase) in the two media. This study deepens our understanding of the metabolic regulation of VB 12 synthesis in P. freudenreichii.

| Microorganism, medium, and culture conditions
Propionibacterium freudenreichii CICC10019 was purchased from the Chinese Industrial-Microorganism Conservation Center and stored in a pre-culture medium supplemented with 20% glycerol at −80°C.
Fermentations were performed in an anaerobic box (MGC C-31; Mitsubishi Gas Chemical Co., Inc., Tokyo, Japan). An aliquot of the cryopreserved cells of P. freudenreichii CICC10019 was streaked on an agar-based solid medium and incubated at 30°C for 5 days to obtain single clones of the activated strain. Then, the single clones were used to inoculate 100-mL sealed anaerobic bottles containing 90 mL pre-culture medium and grown for 24 h at 30°C. Finally, fermentation in 100 mL flasks containing 90 mL fermentation medium was inoculated with 10% (v/v) of the pre-cultures and grown at 30°C for 122 h. The pH value should be adjusted to approximately 7.0 with NaOH solution every 12 h until the end of the fermentation. After 84 h, 5,6-dimethylbenzimidazole was added to a final concentration of 0.9 mg/L, and the fermentations were finished at 122 h.

| Medium optimization
Medium optimization was performed in two steps using Design Expert 10.0 software (Stat-Ease, Inc., MN, USA). Firstly, a Plackett-Burman design was first used to determine the effects of eight variables on VB 12 production. Eleven variables, including 8 medium components (A, glucose; B, corn steep liquor; C, glycerol; E, yeast extract; F, 1% CoCl 2 ·6H 2 O; G, KH 2 PO 4 ; I, (NH 4 ) 2 SO 4 ; J, glycine) and three dummy variables (D, H, K), were screened in twelve trials. The experiments were carried out according to the matrix obtained using Design Expert 10.0 as shown in Table A1. Next, four independent variables that showed significant correlations (glucose, yeast extract, KH 2 PO 4 , glycine) were optimized using a Box-Behnken design. The experimental design, model calculation, graph drawing, and other analyses were performed using Design Expert 10.0. The design matrix and responses are shown in Table A3. A total of 29 experiments were carried out to accurately estimate the errors in the response surface methodology model, and the center point in the design was repeated five times. Each obtained response was used to develop the model of the response. A quadratic polynomial model was applied to evaluate the response of the dependent variables using the following equation: where Y i is the response value, X i is the coded value of the factor, β 0 is a constant coefficient, β i is the linear coefficient, β ii is the quadratic coefficient, and β ij is the interaction coefficient (Box et al., 1978;Rodrigues et al., 2012). X 1 , X 2 , X 3 , and X 4 represent the concentrations of glucose, yeast extract, KH 2 PO 4 , and glycine, respectively.
The pH value of all media was adjusted to a value between 6.8 and 7.0 by adding 12% NaOH solution before autoclaving.

| Metabolomic samples, quenching, and metabolite extraction
The samples for metabolic profiling analysis with the same amount of biomass were collected at the logarithmic phase (31.5 h), stationary phase (56 h), and VB 12 production phase (106 h). The cellular metabolism was quenched by immediately adding the samples into 1 mL of 40% methanol (−20°C) and mixing gently. Then, the cells were pelleted by centrifugation (4,000 g, 4°C, 1 min), resuspended in 0.8 mL methanol (−80°C), and rapidly frozen in liquid nitrogen. For analysis, the cells were harvested by centrifugation of the quenched mixture for 10 min at 0°C, and the supernatant was collected. Then, the cells were resuspended in 50% methanol (−40°C). The mixtures were frozen in liquid nitrogen and thawed three times. After centrifugation (10,000 g, 4°C, 10 min), the supernatant was collected. The pellet was resuspended in 50% acetonitrile (−40°C), sonicated for 10 min in an ice bath, and then centrifuged (10,000 g, 4°C, 10 min).
Finally, all the supernatants were combined in a single tube, filtered through a Teflon filter with 0.22 μm pore-size, lyophilized overnight, and stored at −80°C until LC-MS/MS analysis.
An electrospray ionization source was used for detection in the negative mode under the following conditions: ion voltage, 4500 V; declustering potential, 80 V; source temperature, 550°C; curtain gas pressure, 35 psi; nebulizer gas pressure, 55 psi; heater gas pressure, (1) and MSMS data were exported to identify putative metabolites by searching their accurate masses against the E. coli Metabolome Database. Then, the accuracy of mass measurement, isotopic fit, and LC peak quality were evaluated. Metabolites were identified using the following criteria: precursor mass accuracy <15 ppm; sufficient isotopic fit and good LC peak shape; fragments in the MSMS spectra match those in the in-house library or online databases.

| Pathway enrichment analysis
MetaboAnalyst (https://www.Metab oAnal yst.ca/) was used for further metabolic pathway enrichment analysis of the differential metabolites to obtain insights into the metabolic regulation in response to medium optimization. In the MetaboAnalyst analysis, Over Representation Analysis was set to "Hypergeometric Test" and Pathway Topology Analysis was set to "Relative-betweenness Centrality." The closest available pathway library was Escherichia coli K-12 MG1655. The significant pathways were also identified based on a p-value <0.05 and FDR correction <0.05 and listed in descending order according to the value of the "Impact" factor, while those with an Impact value of zero were excluded.

| Other analytical methods
The optical density (OD) was determined using a V-1600 spectrophotometer at 600 nm. The total concentration of the whole VB 12 both inside and outside of cells was determined by HPLC as previously described by Fang et al (Fang et al., 2018). The fermented broth (20 mL) was mixed with 2 mL of 8% (w/v) NaNO 2 and 2 mL of glacial acetic acid, after which the mixture was boiled for 30 min and filtered through a Teflon filter with 0.22 μm pore-size. The final sample was resolved on a reverse-phase C-18 column (4.6 × 250 mm, 5 µm, Agilent) using an Agilent 1260 HPLC operating at 30°C and monitored at 361 nm. The mobile phase consisted of water and methanol (70:30 [v/v]) at a flow rate of 0.8 mL/min. The VB 12 standard was from Sigma-Aldrich (USA).
Organic acids and glucose present in the fermentation broth samples were analyzed by HPLC equipped with an organic acid analysis column (HPX-87H, Bio-Rad) operated at 55°C with 5 mM H 2 SO 4 as the mobile phase at 0.5 mL/min. Organic acids and glucose standards were used to create a calibration curve. Averages of three biological replicates were reported.

| Statistical analysis
In this study, three independent experiments were carried out for each condition in medium optimization, while four biological replicates were used to perform metabolic analysis for each sample condition. The experimental data were reported as the mean value with the error indicated by the standard deviation. Metabolomics data were normalized and analyzed for statistical significance using MATLAB and Excel software (Microsoft Corp., USA). The normalized data were imported into SIMCA-P software (Ver 14.1; Umetrics, Umea, Sweden) for multivariate statistical analysis. PCA was conducted after mean-centering and unit variance scaling. Hierarchical clustering analysis was performed using MeV software.

| RE SULTS
3.1 | Screening of significant variables for VB 12 production using a Plackett-Burman design To identify the factors that significantly affect VB 12 biosynthesis, eight components of the fermentation medium were selected for the Plackett-Burman design, and the low and high levels of each factor were listed in Table A1. As shown in Table A2, factors with confidence levels above 95% (p < 0.05) were considered to have significant effects. The results showed that glucose (p = 0.0021), yeast extract (p = 0.0007), KH 2 PO 4 (p = 0.0102), and glycine (p = 0.0256) all have significant effects on VB 12 production. Table A2 shows the regression coefficient for each factor. A positive regression coefficient indicates that a higher level of the factor has a beneficial effect on the production of VB 12 , while a negative regression coefficient indicates the opposite relationship. High levels of glucose (40 g/L), yeast extract (10 g/L), and glycine (2 g/L) had an apparent positive effect on VB 12 production, while a low level of KH 2 PO 4 (3 g/L) was also beneficial for VB 12 production. Then, these four components were selected for further experiments. The concentrations of the other factors in the medium for further optimization were set as follows: corn steep liquor 30 g/L, CoCl 2 ·6H 2 O 0.005 g/L, and (NH 4 ) 2 SO 4 2 g/L.

| Model fitting and statistical analyses
To obtain a better medium formulation for VB 12 production, we used response surface methodology to further optimize the concentra- and R 2 summary statistics (adjusted R 2 = 0.9728, predicted R 2 = 0.9275) indicated a good fit. The significance of the response surface methodology model was evaluated using analysis of variance (ANOVA), as shown in Table A4. It was found that all linear and quadratic effects of glucose, yeast extract, KH 2 PO 4 , and glycine should be significant in the model (p < 0.05). The results also indicated that there are significant interactions between glycine and yeast extract as well as between glycine and KH 2 PO 4 . Then, the three-dimensional response surface graphs were plotted to illustrate the optimum levels of the tested components for each pair of factors by keeping the other two factors constant at their middle level (Figure 1). When the concentration of yeast extract was increased from 5 to 25 g/L, VB 12 production first increased and then decreased ( Figure 1a). KH 2 PO 4 , like yeast extract, showed a strong effect on VB 12 production. By contrast, the concentration of glycine had little effect on VB 12 production (Figure 1c), which was consistent with the ANOVA analysis ( The model for VB 12 production was represented using Equation (2): The maximum point of Equation (2) should be in a downward open parabola due to negative signs of all the quadratic coefficients.
Thus, the best concentrations of all the factors in Equation (2) were calculated as follows: glucose (A) 54.3 g/L, yeast extract (B) 17.6 g/L, KH 2 PO 4 (C) 2.7 g/L, and glycine (D) 3.5 g/L. The model predicted a maximum of 8.283 mg/L VB 12 .

| Comparative analysis of VB 12 fermentation using the original and optimized media
To verify the validity of Equation (2), fermentation of P. freudenreichii was performed using the predicted fermentation conditions (repeated 3 times). The final titer of VB 12 reached 8.32 mg/L (0.47 mg/L/OD) ( Figure 2) and was inside the 95% confidence inter- Response surface for the production of VB 12 by P. freudenreichii grown at 30°C for 122 h, with pH adjustment to 7.0 every 12 h, according to the Box-Behnken design. Response surface (a) and contour plot (b) of the combined effects of yeast extract and glycine on VB 12 production. Response surface (c) and contour plot (d) of the combined effects of KH 2 PO 4 and glycine on VB 12 production final titer of VB 12 in the optimized group reached 8.32 ± 0.02 mg/L, which was 2.2-fold higher than in the original group (3.81 mg/L). The titer of VB 12 per unit OD also increased from 0.28 mg/L/OD (original group) to 0.47 mg/L/OD (optimized group). Furthermore, the concentrations of propionic, succinic, and acetic acid in the optimized group were 17.6, 7.9, and 6.8 g/L, which was slightly higher than the corresponding values of the original group, which were 15.3, 7.0, and 5.7 g/L, respectively.

| Comparative metabolic profiling analysis of VB 12 production in the original and optimized groups
Metabolic profiles of P. freudenreichii were analyzed at 31.5, 65, and 106 h of fermentation in original and optimized media to explain how metabolic regulation of the cell affects VB 12 synthesis. As shown in Table A6, a total of 69 intracellular metabolites including sugars, organic acids, amino acids, and other compounds were detected and then identified via LC-MS/MS. Then, principal component analysis (PCA) was first applied to explore the metabolic data as shown in Thus, the metabolic data were suitable for further analysis of intracellular metabolism changes during VB 12 fermentation in the optimized and original medium.

| Different metabolites and pathway enrichment analysis
The log base twofold change values (log 2 (FC)) between the optimized and original groups were calculated for each time point. As shown in Table A7, differential metabolites were identified using the following criteria: log 2 (FC) > 1 or <−1, and -log(P) > 1.301 (pvalue < 0.05). During the logarithmic growth phase, a total of 46 metabolites exhibited differential levels, 43 and 3 of which were, respectively, up-and downregulated in the optimized group. Similarly, 11 and 15 different metabolites were, respectively, up-and downregulated in the optimized group during the stationary phase, as well as 9 and 14 differential metabolites during the VB 12 production phase. As shown in Figure 4, many more marker metabolites were upregulated, especially in the logarithmic growth and stationary phases, while relatively few were downregulated, mainly in the stationary and VB 12 production phases. Next, MetaboAnalyst was used for further metabolic pathway enrichment analysis of the differential metabolites to obtain insights into the metabolic regulation in response to medium optimization.
During the logarithmic phase, almost all differential metabolites were upregulated in the optimized medium, while no significant pathway was enriched among the few downregulated metabolites.
Thus, downregulation of metabolites failed to identify any metabolic pathways by enrichment analysis. The pathway analysis of upregulated metabolites is summarized in Table A8. The most significant metabolic pathways revealed by MetaboAnalyst were propanoate F I G U R E 2 VB 12 fermentation in anaerobic flasks using the original and optimized medium at 30°C for 122 h, with pH adjustment to 7.0 every 12 h. Green: biomass, yellow: glucose concentration, black: propionic acid concentration, blue: succinic acid concentration, orange: acetic acid concentration, and red: VB 12 concentration. Solid circles represent the original medium (org), and triangles represent the optimized medium (opt). Data were presented as means * standard deviations (SD) calculated from three independent experiments F I G U R E 3 PCA-score scatter 3D plot of metabolic profile of P. freudenreichii grown in an original and optimized medium at 30°C for 122 h, with pH adjustment to 7.0 every 12 h. "A" and "B," respectively, indicated the fermentation samples in the original and optimized medium. "l," "s," and "p," respectively, mean the fermentation samples obtained at the logarithmic phase (31.5 h), stationary phase (56 h), and VB 12 production phase (106 h) metabolism, glycine, serine and threonine metabolism, alanine, aspartate and glutamate metabolism, butanoate metabolism, citrate cycle, and glycolysis or gluconeogenesis. C1 metabolism (glycine and serine) and metabolic pathways that provide precursors for VB 12 (glycine, serine, and threonine metabolism, alanine, aspartate, and glutamate metabolism) were more important revealed according to MetaboAnalyst. Therefore, the supplemental nitrogen source and glycine in the optimized medium promoted the PA metabolism and cell growth of P. freudenreichii and were also beneficial for the supply of precursors and methylation donors for the VB 12 synthesis pathway. Secondly, the pathway analysis with MetaboAnalyst showed that medium optimization enabled the bacteria to maintain more active PA synthesis, which helped maintain higher cell activity and energy supply during the stationary phase (Tables A9-A10).
MetaboAnalyst indicated that the pentose phosphate pathway was downregulated. Finally, MetaboAnalyst obtained similar results for the up-and downregulated differential metabolites in the production phase and stationary phase (Table A11). The pentose phosphate pathway was identified as significant for downregulated differential metabolites. All the detected metabolites were subjected to hierarchical cluster analysis as shown in Figure 5. All metabolites were clustered into four categories (I, II, III, IV) according to their change trends in the different fermentation periods. Cluster I mainly included cofactors such as ATP and NADH; cluster II included various amino acids; cluster III mainly included the metabolites involved in glycolysis, pentose phosphate pathway, and the TCA cycle; and cluster IV was mainly related to PA synthesis. Only 12 metabolites exhibited positive correlations of trends between the optimized and original groups based on correlation coefficients, (red stars), which indicates that the trends of most differential metabolites were altered by medium optimization. As shown in the metabolic network of VB 12 synthesis in P. freudenreichii (Figure 6), the synthesis of VB 12 relies on various metabolic pathways, mainly providing cofactors (ATP, GTP, NADH, NAD, and FADH 2 ) and amino acids (glutamic acid, glutamine, S-adenosyl methionine, and threonine).

| Analysis of key metabolic pathways associated with VB 12 and PA production
Next, we analyzed the changes of glycolysis, pentose phosphate pathway, the TCA cycle, pyruvate metabolism, and different amino acid metabolism pathways according to the above results of pathway enrichment analysis, and analyzed their possible effects on VB 12 and PA synthesis. The central metabolic pathways in P. freudenreichii were presented in Figure 6.
As shown in Figures 5 and 6, the metabolites involving glycolysis (glucose 6-phosphate, fructose 6-phosphate, fructose 1,6-bisphosphate, glyceraldehyde 3-phosphate, glycerate 1,3-biphosphate, 2-phosphod-glycerate, phosphoenolpyruvate, and pyruvate) could be clustered into different groups according to their change trends during the fermentation time. Firstly, the F I G U R E 4 Venn diagrams of the upregulated (log 2 (FC) > 1) and downregulated (log 2 (FC) < −1) differential metabolites at the logarithmic phase (l), stationary phase (s) and VB 12 production phase (p). log 2 (FC) means the log base twofold change values. Differential metabolites were identified using the following criteria: log 2 (FC) > 1 or <−1, and −log(P) > 1.301 (p-value < 0.05) In P. freudenreichii, the TCA cycle and the Wood-Werkman cycle have many common intermediate metabolites, including acetyl-CoA, succinic acid, fumaric acid, and malic acid. In addition, propionyl-CoA, R-methylmalonyl-CoA, and PA exclusively take part in the Wood-Werkman cycle, while citric acid, cis-aconitic acid, and oxoglutaric acid only take part in the TCA cycle. As can be seen in Figure 5, the intermediate metabolites of the Wood-Werkman cycle, especially the coenzyme intermediates, were significantly more abundant in the optimized medium than in the original medium, especially during the initial phase of fermentation. The production of PA by P. freudenreichii is characterized by growth coupling, and the energy generated during the formation of PA is beneficial to the growth of bacteria and the synthesis of VB 12 . The role of succinic acid in the PA synthesis pathway is consistent with the metabolic role of PA itself and is different from other intermediate metabolites, which may be attributed to the fact that succinic acid consumption and PA production are directly coordinated by the enzyme E2 ( Figure 6). Therefore, the formation of succinic acid is the most important factor for PA synthesis.
The metabolism of amino acids was significantly different in the two different media ( Figure 5). Except for l-asparagine, l-cysteine, and lhistidine, all other detected amino acids (l-aspartic acid, l-glutamine, l-glutamic acid, l-tryptophan, l-serine, l-threonine, l-methionine, l-valine, l-isoleucine, l-lysine, l-arginine, l-proline, l-tyrosine, and lglycine) were significantly more abundant in the optimized group, which was due to the addition of yeast extract as a rapidly available nitrogen source for cell growth, which not only accelerated the growth of bacteria but also enhanced the conversion of the carbon source into precursors of VB 12 synthesis. It is worth noting that in the early phase of fermentation, the bacterial cells in the original group contained more l-asparagine, l-cysteine, and l-histidine than those in the optimized group. One of the most important physiological functions of l-asparagine is to store and transport nitrogen, which may affect nitrogen metabolism and stress resistance. Cysteine can be produced from methionine, which is reported to be capable of storing reducing power and has a certain anti-reverse effect. The metabolism of cysteine may also affect the formation of S-adenosyl methionine (SAM) from methionine, which affects the synthesis of VB 12 . Histidine is synthesized from d-ribulose 5-phosphate in the pentose phosphate pathway, which is converted into phosphoribosyl pyrophosphate after several steps, in a pathway that is competitive with purine and adenosine. Higher histidine abundance in the original medium is likely due to higher flux through the pentose phosphate pathway. In particular, the VB 12 synthesis process consumes large amounts of methyl groups, the direct source of which is SAM, while the synthesis of SAM depends on the methionine and folic acid cycle. The one-carbon moiety that enters the folic acid cycle can be derived from glycine, histidine, serine, or cysteine. This is one of the main reasons why the addition of glycine and yeast extract in the optimized medium increased the yield of VB 12 . Finally, cofactors such as ATP and NAD(P)H also play important roles in the synthesis of VB 12 . In the optimized medium, especially during the period of VB 12 production after the growth phase, the total intracellular cofactor content of the cells in the optimized medium was significantly higher than in the original medium. The increase in the overall abundance of cofactors is beneficial for the synthesis of VB 12 and may also be related to the addition of phosphorus.

| DISCUSS ION
Propionibacterium freudenreichii has been used to produce therapeutically active VB 12 , and increasing the production of VB 12 by Propionibacterium has received a lot of attention. Many studies optimized the fermentation medium by adding precursors of cobalamin biosynthesis to enhance VB 12 production by P. freudenreichii, and these studies revealed DMBI and cobalt ions play crucial roles in the biosynthesis of VB 12 (Murooka et al., 2005;Paulina et al., 2017;Roman et al., 2001;Wang et al., 2015;Thirupathaiah et al., 2012).
Other media components were also optimized in many studies to improve the productivity of VB 12 using statistical experimental designs. A 93% increase of VB 12 concentration (4 mg/L) was obtained following medium optimization with glycerol as the carbon source (Kosmider et al., 2012). Five of 13 tested medium components (calcium pantothenate, NaH 2 PO 4 , casein hydrolysate, glycerol, and FeSO 4 ) were found to have significant effects on VB 12 production.
Another 43% increase of VB 12 production by medium optimization with glucose as the carbon source was also reported (Chiliveri et al., 2010). Optimization of the eight most significant influencing factors F I G U R E 5 Hierarchical cluster analysis of all detected metabolites in the metabolic profile of P. freudenreichii grown in the original and optimized medium at 30°C for 122 h, with pH adjustment to 7.0 every 12 h. "A" and "B," respectively, indicated the fermentation samples in the original and optimized medium. "l," "s," and "p," respectively, mean the fermentation samples obtained at the logarithmic phase (31.5 h), stationary phase (56 h), and VB 12 production phase (106 h). Blue circles indicate differential metabolites, and red stars indicate metabolites with a significant positive correlation between the two media. Four groups (I, II, III, and IV) are clustered (FeSO 4 , inoculum size, (NH 4 ) 2 HPO 4 , glucose, DMBI, yeast extract, sodium lactate, and CoCl 2 ) was performed using the Taguchi method.
Waste frying sunflower oil was used as a medium for VB 12 production by P. freudenreichii, whereby DMBI, CoCl 2 , FeSO 4 , and calcium chloride were found to have the most significant effects on VB 12 production during medium optimization (Hajfarajollah et al., 2014). Figure 6, other precursors such as glycine, threonine, 5-aminolevulinic, or choline were also important for the production of VB 12 (Piwowarek et al., 2018). Many compounds such as DMBI and cobalt ions are related to VB 12 synthesis as direct precursors.

As shown in
Many studies have provided insights into the effects of these compounds on the synthesis of vitamin B 12 with clear conclusions. Our study focused more on the effect of whole-cell metabolic changes induced by precursor additives on VB 12 synthesis. Thus, these compounds were not considered in our medium optimization. The original medium used for VB 12 production was reported by Peng, Wang, Liu, et al. (2012), based on which glycerol, yeast extract, and glycine were added to the optimized medium in this study (Peng, Wang, Liu, et al., 2012).
Experimental results of the response surface design indicated that the yield of VB 12 could be significantly increased by adequate supplementation of glucose, yeast extract, KH 2 PO 4, and glycine.
Yeast extract contains a variety of amino acids, which are more easily available and fast-acting nitrogen sources for P. freudenreichii than the corn steep liquor in the original medium. The additional KH 2 PO 4 may promote the synthesis of metabolites containing phosphate moieties, such as metabolic cofactors and nucleic acids. Glycine acts as a primary methyl group donor for VB 12 synthesis. According to the identified clusters and pathway analysis, the metabolism of P.
freudenreichii underwent complex changes in response to the modified medium, mainly affecting the sugar metabolism pathways (glycolysis, pentose phosphate pathway, TCA cycle, and PA metabolism), amino acid metabolism, and cofactor metabolism. The changes in metabolic pathways not only affected the synthesis of VB12 but also led to changes in cell growth and PA production.
Usually, inhibiting PA biosynthesis or removing PA from the culture medium was necessary for improving VB 12 production, but Wang et al. (2015) found that cobalamin production still requires a certain concentration of PA (10-20 g/L during the initial stage of fermentation and 20-30 g/L at a later stage) (Wang et al., 2015).
It therefore seems that PA does not always have a negative effect on VB 12 synthesis. Guan et al. (2015) investigated PA production in P. freudenreichii using a metabolomic approach, which revealed key metabolites and pathways that are positively correlated with PA production. Notably, these were surprisingly similar to the ones that promoted VB 12 production in this study. The upregulation of glycolysis, pentose phosphate pathway, and the common metabolites between the TCA cycle and Wood-Werkman cycle benefited not only PA production as overflow metabolism but also VB 12 synthesis due to increased supply of energy and precursors. It is worth noting that amino acid metabolism plays an important role in both PA and VB 12 synthesis. Guan et al. (2015) found that PA production increased by 39.9% when 20 mM arginine and aspartate were added, while leucine, threonine, γ-aminobutyric acid, citrulline, methionine, and serine had no significant effects on PA production (Guan et al., 2015).
The strength of aspartate metabolism in P. freudenreichii was found F I G U R E 6 Metabolic pathways of PA biosynthesis in P. freudenreichii to be highly strain-dependent (Thierry et al., 2011). Aspartate can be deaminated to fumarate and further reduced to succinate, with concomitant production of NAD + and ATP (Crow, 1986(Crow, , 1987. The metabolism of other amino acids, such as glutamate, glycine, and threonine, can provide precursors for VB 12 production. Overall, PA and VB 12 are not produced by strictly competing pathways, but their synthesis is correlated in P. freudenreichii. For example, a high energy yield is provided by PA fermentation via the Wood-Werkman cycle for cell growth and VB 12 synthesis (Thierry et al., 2011). As the cofactor for methylmalonyl CoA mutase, VB 12 also plays an important role in the conversion of succinyl-CoA into methylmalonyl-CoA, which is a penultimate step of the Wood-Werkman cycle (Piwowarek et al., 2018).

| CON CLUS IONS
A two-step medium optimization method based on a statistical experimental design was applied to increase the VB 12 production by more than 118%. High levels of glucose (40 g/L), yeast extract (10 g/L), and glycine (2 g/L) had an apparent positive effect on VB 12 production, while a low level of KH 2 PO 4 (3 g/L) was also beneficial for VB 12 production. This study provides in-depth insight into the central carbon metabolism of P. freudenreichii. The upregulation of glycolysis, the pentose phosphate pathway, amino acid metabolism, and the common metabolites between the TCA cycle and Wood-Werkman cycle benefited not only PA production as overflow metabolism but also VB 12 synthesis due to increased supply of energy and precursors. PA metabolism indirectly contributes to VB 12 synthesis through growth coupling, which depends on succinic acid.

E TH I C S S TATEM ENT
None required.

CO N FLI C T O F I NTE R E S T
None declared. writing-review and editing (equal).

DATA AVA I L A B I L I T Y S TAT E M E N T
All data generated or analyzed during this study are included in this published article.