Improving the Production of L-Phenylalanine by Identifying Key Enzymes Through Multi-Enzyme Reaction System in Vitro

L-Phenylalanine (L-Phe) is an important amino acid used in both food and medicinal applications. We developed an in vitro system that allowed a direct, quantitative investigation of phenylalanine biosynthesis in E. coli. Here, the absolute concentrations of six enzymes (AroK, AroL, AroA, AroC, PheA and TyrB) involved in the shikimate (SHIK) pathway were determined by a quantitative proteomics approach and in vitro enzyme titration experiments. The reconstitution of an in vitro reaction system for these six enzymes was established and their effects on the phenylalanine production were tested. The results showed that the yield of phenylalanine increased 3.0 and 2.1 times when the concentrations of shikimate kinase (AroL) and 5-enolpyruvoyl shikimate 3-phosphate (EPSP) synthase (AroA) were increased 2.5 times. Consistent results were obtained from in vivo via the overexpression of AroA in a phenylalanine-producing strain, and the titer of phenylalanine reached 62.47 g/l after 48 h cultivation in a 5-liter jar fermentor. Our quantitative findings provide a practical method to detect the potential bottleneck in a specific metabolic pathway to determine which gene products should be targeted to improve the yield of the desired product.

L-Phe is an essential amino acid for humans and most livestock. It is used in feed, food additives, taste and aroma enhancers, pharmaceuticals or as building blocks for drugs, dietary supplements, nutraceuticals, and ingredients in cosmetics 1 . Notably, L-Phe is used in the production the sweetener aspartame, which has a steadily increasing world-wide demand 2 . Currently, L-Phe is obtained by chemical, enzymatic or microbial processes. In recent years, there has been increased interest in producing L-Phe through microbial fermentation, especially by metabolically engineered strains of E. coli, which have a high growth rate and well-defined physiological characteristics 3 .
L-Phe biosynthesis and its regulation have been extensively investigated in E. coli. The shikimate pathway serves as the primary source of production of L-Phe. Two shikimate kinases, AroK and AroL, catalyze the formation of shikimate 3-phosphate (S3P) from shikimate and ATP. The expression of aroL is regulated by TyrR with tyrosine or tryptophan as a co-repressor 4,5 . In contrast to AroL, the activity of AroK in the cell is independent of both the amount of extracellular aromatic amino acids and the level of tyrR gene product 6,7 . 5-Enolpyruvyl shikimate 3-phosphate synthase (AroA) catalyzes the transfer of the enolpyruvoyl moiety from phosphoenolpyruvate (PEP) to the hydroxyl group of carbon 5 of S3P with the elimination of phosphate to produce EPSP [8][9][10][11] . This is an addition-elimination reaction, which introduces the three-carbon fragment destined to become the side chain of phenylalanine. The branch point compound that serves as the starting substrate for the three terminal pathways of aromatic amino acid biosynthesis, chorismic acid (CHA), is obtained through the conversion of EPSP to chorismate catalyzed by chorismate synthetase (AroC) 12,13 . This enzyme catalyzes the phosphate elimination by a 1,4-elimination mechanism that proceeds with anti-stereochemistry 14 . Additionally, this enzyme is oxygen sensitive and may remain inactive under aerobic conditions 15 , but it can be activated by a reduced flavin adenine dinucleotide (FAD)-regenerating system in an atmosphere of H 2 . This was accomplished by the reduction of FAD with reduced nicotinamide adenine dinucleotide (NADH) and mammalian diaphorase or bacterial NADH dehydrogenase 16 . Bifunctional chorismate mutase / prephenate dehydratase (PheA) carries out the second step in phenylalanine biosynthesis as well as in the parallel biosynthetic pathways for the production of the aromatic amino acids tyrosine and phenylalanine. The native enzyme is a dimer of identical subunits each containing a dehydratase active site, a mutase active site and a phenylalanine binding site 17,18 . L-Phe was shown to feedback-inhibit both the chorismate mutase and prephenate dehydratase activities of the enzyme by an allosteric mechanism 6 . Finally, a tyrosine-repressible aromatic amino acid aminotransferase (TyrB) catalyzes the transamination of L-Glutamic acid and phenylpyruvate to yield L-Phe [19][20][21] .
To some extent, it is possible to enhance the specific productivity of L-Phe in E. coli by manipulating known catalytic enzymes. Due to the complexity and interconnectivity of metabolic networks, the creation of a highly efficient cell factory for the production of a target protein often requires multiple alterations, including overexpressing key enzymes, deleting genes involved in off-pathway processes or that lead to byproducts, increasing the expression level of efflux proteins for the removal of feedback inhibition, or introducing external enzymes with high activity [22][23][24][25][26] . For instance, Backman et al. engineered E. coli for L-Phe production, based on overexpressing the aroF WT and pheA fbr genes and developed an efficient fermentation process within 36 h, with a final L-Phe titer of 50 g/l 27 , and Marjan et al. applied direct reduction of the carbon flow to acetate by knocking out ackA-pta and poxB 28 . Although there are a number of studies on the relationships between the flux of a metabolic system in vivo and the absolute concentrations of particular enzymes, the measurement of precise concentrations and activities of the enzymes in the cells are quite challenging. Nevertheless, in vitro enzyme reaction system has been well applied in recent years to achieve the guide of engineering the strains 29 . Moreover, label-free proteomics tools will provide an opportunity for the measurement of absolute protein concentrations inside the cells, which will be a new application in the field of metabolic engineering.
Collectively, we took all of these advantages and developed a much easier method to analyze L-phe biosynthesis pathway quantitatively. In this study, the absolute quantities of proteins in the phenylalanine production strain HD-1 were determined using label-free proteomics. In addition, to establish a multi-enzyme reaction system, we expressed and purified six enzymes from the shikimate pathway (AroK, AroL, AroA, AroC, PheA, and TyrB) and verified their activities. We then reconstituted the shikimate pathway reaction system in vitro base on the absolute concentrations of shikimate enzymes obtained from the proteomics data. We found that AroL and AroA were bottlenecks for increasing phenylalanine synthesis flux. We tested overexpression of the aroL and aroA genes in the production strain and found that overexpression of aroA improved the yield of L-Phe by 45.18%. The results showed that the in vitro multi-enzyme system was a rational strategy for reconstitution of the shikimate acid pathway components in vivo that regulate the production of L-Phe.

Materials and Methods
Plasmid, bacterial strains and medium. Plasmid pET28a was reserved in our lab. The host bacterial DH5α and BL21(DE3) were purchased from invitrogen company. The E. coli HD-1 strain for phenylalanine production was derived from E.coli W3110 strain that initially undergone multiple random mutagenesis for the enhanced production of L-phenylalanine and then stepwise rational metabolic engineering. The genome sequences of the HD-1 strain were determined and compared (Table 1 and Supplementary Tables 2-3) with our knowledge. All the bacteria are kept in 30% of the glycerin and stored in − 80 °C before used.
All kits and markers used for the construction of clones were from Omega Bio-tek, Inc (USA).
Plasmid Construction. The genomic DNA of the phenylalanine producing E. coli strain HD-1 was used as the template for amplification of the phenylalanine synthesis genes. All enzymes genes were amplified by PCR using primers as described in the Supplementary Table 1. After amplification, the DNA fragments were purified from agarose gel, and subsequently inserted into the pET28a vector using restriction enzymes and T4 DNA ligase. The plasmids and strains used in this study are shown in Table 1.
Enzyme expression and purification. To  . Bound proteins were then eluted with elution buffer (26 mM NaH 2 PO 4 , 0.5 M NaCl, 500 mM imidazole, pH 7.4). The recombinant protein-containing fractions were dialyzed against storage buffer (100 mM Tris-HCl, 5 mM β -mercaptoethanol) in order to remove the imidazole. After purification, the purified enzymes concentrations were determined using the BCA kit. The purified proteins were stored at − 80 °C until use.

Fermentation of phenylalanine production strains in a 5L-bioreactor. A crude enzyme extraction
was prepared from the high phenylalanine-producing strain HD-1. After growing 7 hours in a shaking flask of LB, the cells were then inoculated into a 3.5 L fermentation medium in a 5L bioreactor. During the entire fermentation process, the pH was kept at 7.0 with the addition of 25% ammonia water and the temperature was set at 33 °C until the OD 600 reached 35 (mid log phase) and then was up-shifted to 38 °C to induce the expression of enzymes responsible for L-Phe production 1 . The dissolved oxygen (DO) level was upheld near 40% by adjusting the agitation speed (300-900 rpm) and the aeration rate (2-3 vvm). The glucose concentration was maintained between 0-5 g/l by feeding 800 g/l glucose. The fermentation broth was sampled every two hours to detect phenylalanine concentrations, cell density analysis, and glucose consumption. A sample collected at 32 h was used for the crude enzyme extraction experiment. The cells were harvested by centrifugation for 15 min at 10,000 g at 4 °C. For the crude enzyme extract preparation, the cell pellet was washed three times with enzyme assay buffer. Next, the cells were resuspended in lysis buffer (binding buffer containing 1% cocktail). The crude extracts were obtained by sonication of the cell suspension. Cellular debris was removed by centrifugation (30 min, 10,000 g, 4 °C) and the supernatant was dialyzed at 4 °C. The final crude enzyme extracts were stored at − 80 °C. L-phenylalanine overproducing strain. It was derived from E. coli W3110 strain, which initially undergone multiple random mutagenesis and then overexpressing aroF WT and pheA fbr . The genome sequences of the HD-1 strain were determined and compared (Supplementary Tables 2-3 (triethylammonium bicarbonate)] to a final concentration of 1 mg/mL. Equal aliquots were then digested with trypsin (Promega, Madison, USA) overnight at 37 °C.
A NanoLC system (NanoLC 2D Ultra, Eksigent, Massachusettes, USA) equipped with a Triple TOF 5600 mass spectrometer (AB SCIEX, Massachusettes, USA) was used for analysis. Peptides were trapped on a NanoLC pre-column (Chromxp C18-LC-3 μ m, size 0.35 × 0.5 mm, Eksigent) and then eluted onto an analytical column (C18-CL-120, size 0.075 × 150 mm, Eksigent) and separated by a 120 min gradient from 5 to 35% Buffer B (Buffer A: 2% ACN, 98% H 2 O, Buffer B: 98% ACN, 2% H2O, 0.1% FA) at a flow rate of 300 nL/min. Full-scan MS was performed in positive ion mode with nano-ion spray voltage of 2.5 kv from 350 to 1500 (m/z), with up to 30 precursors selected for MS/MS (m/z 100-1500) if exceeding a threshold of 125 counts per second (counts/s). Peptides with + 2 to + 5 charge states were selected for MS/MS. The collision energy (CE) for collision-induced dissociation (CID) was automatically controlled using an Information-Dependent Acquisition (IDA) CE parameter script to achieve optimum fragmentation efficiency. Finally, the relative proportions of the enzymes in the cell could be obtained by comparing the amount of the peptides with the protein molecular masses after the proteomic analysis 30  Part two determined the enzymatic activity of TyrB catalyzing the conversion of phenylalanine to phenylpyruvate acid 19 . The reaction system was as follows: 40 mM sodium cacodylate (pH 7.6), 80 μ M pyridoxal phosphate and 2.5 mM phenylalanine, 5.0 mM 2-oxoglutarate, and TyrB at an appropriate concentration (2-5 μ M). After incubation of mixtures at 37 °C for 30 min, the reaction was terminated by the addition of 0.2 ml of 1 M formic acid, and phenylpyruvic acid was determined as described above.
The activity of the crude enzyme extract was assayed by detection of phenylalanine. The reaction systems were as follows: Tris-HCl 100 mM, pH 7.5, SHIK (8 mM), ATP (8 mM), KCl (50 mM), MgCl 2 (5 mM), PEP (4 mM), FAD (1 mM), NADH (5 mM), β -mercaptoethanol (25 mM), L-glutamic acid (4 mM), pyridoxal phosphate (1 mM) and appropriate crude enzyme extraction. The reaction was conducted at 37 °C for 30 min. The enzymes were inactivated by heating at 100 °C for ten minutes when the reaction was complete. The enzymes were then precipitated by centrifugation at 10,000 g for 5 min and the supernatant was used for detection of phenylalanine by HPLC, based on amino acid analysis with OPA derivatization and detection at 338 nm using Zorbax Eclipse-AAA columns on an Agilent (Santa Clara, USA) 1100 HPLC. The mobile phase consisted of two elutions: A (40 mM Na 2 HPO 4 , pH 7.

Validation of the key enzyme in shikimate pathway in vivo.
For the validation of the key enzyme identified in the multiple enzyme system in vivo, aroL and aroA genes were cloned into p15A1 to construct plasmids p15A1-aroL1/2/3 and p15A1-aroA1/2/3, respectively, using primers aroL(A)-LF11/12/13 and aroL(A)-LR11/12/13 (shown in Supplementary Table 1). AroC gene was cloned into p15A1 to construct plasmids p15A1-aroC1/2/3, used as controls. All the integration vectors (Table 1) used in this study were constructed by a sequence-independent "simple cloning" method without the need for restriction and ligation enzymes 32 . The host strains HD-1 containing the different overexpression plasmids were fermented, separately, and the strain HD-1 containing plasmid p15A1-aroA2 was fermented in a 5L-bioreactor.

Results and Discussion
L-phenylalanine-overproducing strain. In this study, E. coli HD-1 strain was chosen as a host strain.
It was derived from E. coli w3110 strain, which initially undergone multiple random mutagenesis and then overexpressing aroF WT and pheA fbr based on system-wide analysis of metabolism results in gradual increase in L-phenylalanine production throughout the strain engineering steps 27 . The genome sequence of the HD-1 strain was determined and subjected to comparative genomic analyses (Supplementary Tables 2-3). With the genomic information of the w3110 strain as a control, pairwise genome alignment for the w3110 and HD-1 strains revealed that a total 47 genes in the w3110 strains appeared to be either deleted or have modified sequences in the HD-1strain, the most effects of which on the L-phenylalanine production phenotype are not clear. Four genes   involved in the L-phenylalanine biosynthesis of the HD-1 strain, including pykA, aroG, pta and pheA genes, were found to have nonsynonymous SNPs, in comparison with the E. coli w3110 strain, which have likely affected L-phenylalanine production in the HD-1 strain. Relatively genomic differences between E. coli w3110 and HD-1, and mutations in many of these L-phenylalanine biosynthesis-related genes indicate that the HD-1 strain seems to be better optimized for the L-phenylalanine overproduction.
Enzymatic assays for enzymes in the shikimate pathway. The five-step metabolic pathway for phenylalanine synthesis from SHIK in E. coli is illustrated in Fig. 1. To investigate the contribution of different enzymes to the production of L-Phe, the crude enzyme extract obtained from the cells at 32 h (Fig. 2) was used. The enzymes involved in the phenylalanine synthesis pathway were expressed in E. coli (DE3) and purified. The SDS-PAGE results from the E. coli express system are shown in Fig. 3. Specifically, the activities of two  Table 2. The determination of purified enzyme activity. Incubation mixture was described in the "Assay of the enzyme activities". Phenylpyruvate was determined as described in the "Determination of phenylpuruvate acid". isoenzymes, namely AroK and AroL, were determined. The specific activity of AroK was 18 μ mol/(min.mg), which was much lower than that of AroL (61 μ mol/(min.mg), as shown in Fig. 4. This result indicated that AroL has a much better catalytic ability than AroK and may be the dominant enzyme of the pathway in vitro. The activities of the first four enzymes (AroL, AroA, AroC, and PheA) and the last one (TyrB) were quantitated by assaying phenylpyruvic acid production. The results revealed that all the purified enzymes were active, as shown in Table 2. Additionally, we measured the activities of the crude enzyme extract and detected a peak that appeared at relatively the same time as the phenylalanine standard (Fig. 5), indicating that all enzymes of the shikimate pathway present in the crude enzyme extract were functional and active.
Determine the absolute enzyme concentrations. Mass spectrometry data were processed using the Pioplite software with protein identification and quantification. The number of fragment-iron spectra acquired for all the peptides of a protein correlates with the expression level of each protein. However, as the differences in composition and size of each protein can be profound, these results are inadequate to accurately represent the expression level of different proteins. To overcome this problem, we used APEX 33 to process the original label-free proteomics data to acquire reliable data on the ratios between different proteins. The relative amounts of enzymes (normalized to AroK) are shown in Fig. 6a. To determine the absolute intracellular concentrations of the various enzymes, at least one enzyme must be known in advance. We used AroK as such a standard, because the amount of AroK in the crude extract was easily measured by fluorescence since NADH is involved in its catalyzed reaction. Different amounts of purified AroK were added into the diluted (1:3) crude extract and the reaction rate was measured. The results revealed that the concentration of AroK in the crude enzyme extract was 9.6 μ M, as can be  To ensure that the reaction rate was not affected by the low substrate concentration during the reaction, all the substrates were added at concentrations much higher than that of their Km values (Fig. 7). In order to evaluate the influence of different enzymes, each purified enzyme was separately titrated into the reaction system and the concentration of the corresponding enzymes was then increased to 2.5 times based on their initial level. After the addition of the purified enzymes, yields of phenylalanine were increased to different levels, indicating the different impact The yields of phenylalanine were increased to different levels after the addition of single purified enzymes to 2.5 times the level from the absolute enzyme concentrations measured in the crude cell extract, indicating the different impact of enzymes on the production of phenylalanine. The control is the production of phenylalanine in the system constituted by crude enzyme extract. The data is from three experimental replicates. of each enzyme on the production of phenylalanine. The reaction system without adding any purified enzymes was used as control, as shown in Fig. 8. The production of phenylalanine in the samples in which AroL and AroA were added, was higher than in the other three groups, indicating that the two steps of the shikimate acid pathway (catalyzed by AroL and AroA, respectively) were potentially the limiting steps in the synthesis of phenylalanine. Thus, we reasoned that the genes aroL and aroA might be the bottlenecks causing the inefficiency in the production of phenylalanine.
Validation of the key enzyme in vivo. The metabolic relevance of our observation in vitro was tested through overexpression of targeted genes in vivo. Nine strains (HD-L (A, C)1/2/3) carrying plasmids containing aroL, aroA or aroC under the control of different promoters were constructed. When we transformed the HD-1 strain with p15A1-aroA3, which expresses the aroA gene under the strong promoter BBa_J23118, cell growth was severely reduced, likely due to overflow metabolism. As expected, as shown in Fig. 9a, the strains HD-C1/2/3 overexpressing aroC did not improve the production of phenylalanine compared with HD-1. Unexpectedly, strains overexpressing aroL, HD-L1/2/3, did not lead to increased phenylalanine production. The HD-L2/3 strains in particular produced 5.7 ± 0.13 g/L and 5.262 ± 0.16 g/L phenylalanine, which was slightly lower than the control strain HD-1. We hypothesized that this is mainly due to the inhibition of AroL by shikimate acid and tryptophan, which are not produced in vitro 4,34,35 . For example, the activity of AroL decreased by approximately sevenfold when the shikimate concentration was increased from 1 to 10 mM, suggesting an inhibition by high levels of the substrate 4 . Interestingly, the production of the other two resulting strains, HD-A1/2, reached 7.542 ± 0.16 g/l and, approximately 27% and 32% higher than the control strain HD-1, respectively. Actually, in order to ensure that AroL had been successfully overexpressed in the strains HD-L1/2/3, the enzyme AroL was purified and the activity of it was determined by coupling the release of ADP from the shikimate kinase-catalyzed reaction to the oxidation of NADH using pyruvate kinase and lactate dehydrogenase as coupling enzymes. As shown in Fig. 9b, the catalytic rate of the enzyme in strains HD-L1/2/3 was significantly higher than that of the original strain HD-1. The result showed that the amount of AroL in the strains HD-L1/2/3 was increased after the overexpression of the enzyme. At the same time, it was proved that AroL was successfully overexpressed in the strains HD-L1/2/3. The production of L-Phe by HD-A2 and HD-1 (the control) was also investigated in a 5 L-bioreactor, and the titer of phenylalanine reached 62.47 g/l after 48 h cultivation (Fig. 10). The engineered strain HD-A2 produced approximately 38.82% more phenylalanine compared with the original strain HD-1, and the phenylalanine yield on glucose was increased from 0.186 g/g (18.6%) to 0.236 g/g (23.62%) ( Table 3). EPSP synthase (AroA) catalyzes the transfer of the enolpyruvoyl moiety from phosphoenolpyruvate (PEP) to the hydroxyl group of carbon 5 of shikimate 3-phosphate. Therefore, we hypothesize that when the amount of AroA increases, it can improve the binding rate of PEP, thereby improving the reaction efficiency. This demonstrated that increasing expression of AroA, the key enzyme identified through in vitro metabolic control analysis, increased phenylalanine production in vivo.

Conclusion
In this study, we have demonstrated that a metabolic pathway can be reconstituted in vitro from purified enzymes with cofactors and coenzymes, which is an ideal module to investigate its regulation since many parameters can be varied. The precise measurement of the absolute intracellular concentration of the enzymes in vivo and Figure 10. Fermentation process of phenylalanine production stain HD-A2. L-Phe production by HD-A2 was investigated in a 5L-bioreactor, and the titer of phenylalanine was measured after 48 h cultivation. The data is from three experimental replicates.   Table 3. Comparison of the fermentation results of the original HD-1 strain and the aroA overexpressed strain HD-A2 by fed-batch fermentation. The OD, Phe concentration and yield were compared between HD-1 and HD-A2. The data is from three experimental replicates.
the possibility of fine-tuning of enzyme amounts in vitro make it possible to quantitatively evaluate the effect of different enzymes on the production of a specific pathway based on metabolic control analysis. In addition, key enzymes in the pathway can be determined and the overexpression of these enzymes is more likely to be an effective modification strategy to improve the carbon flux into the synthetic pathway of the targeted metabolites in vivo. The enzyme AroA exhibited the largest impact on phenylalanine synthesis in the phenylalanine producing strain HD-A2. The production in the modified strain HD-A2 resulted in an increase of 38.82% in the L-Phe titer compared with the production in the original strain. This method, based on the reconstitution of purified stable enzymes in vitro, has unique advantages over blindly modifying living microorganisms in vivo. These include easy access and control, high product yield, fast reaction rate, tolerance of toxic compounds and products, great engineering flexibility for in vitro assembly and shifting unfavorable reaction equilibrium. This approach may shed light on a rational strategy for engineering strains and can serve as a new biomanufacturing platform evolving from fundamental research tools. Furthermore, it has a wide potential application for the efficient construction of industrial strains.