Rational Metabolic Engineering of Escherichia Coli for High-yield L-serine Production

Chenyang Wang Shanghai Advanced Research Institute Junjun Wu Nanjing Agricultural University Lei Wang Jiangnan University Xiaojia Chen ShanghaiTech University Qinyu Li ShanghaiTech University Li Li Shanghai Tramy Green Food (Group) Co. Ltd Xinxin Ma Shanghai Tramy Green Food (Group) Co. ltd Jiping Shi Shanghai Advanced Research Institute Zhijun Zhao (  zhaozj@sari.ac.cn ) Shanghai Advanced Research Institute https://orcid.org/0000-0003-1196-553X


Background
L-serine(L-Ser) is a vital amino acid in metabolism in humans and other animals and is widely used in the food, pharmaceutical and cosmetic industries [1]. Additionally, L-Ser has been identi ed as one of the top 30 most interesting biochemical building block. Currently, L-Ser production relies on enzymatic or direct fermentation [2,3], and the global L-Ser production capacity (350 tons per year) is well below the expected market demand (3000 tons per year) [4]. Therefore, it is necessary to develop a more effective L-Ser production method. Furthermore, demand exists for developing a direct fermentation approach that can be implemented with low-cost substrates and will can simplify operational puri cation procedures and reduce pollution, as enzymatic conversion always utilizes the expensive precursors glycine and methanol [5]. L-Ser production by microbial fermentation has been extensively studied in Corynebacterium glutamicum. In 2007, Petra Peters-Wendisch et al. engineered a C. glutamicum strain with an industrial production capacity 36 g/L L-Ser [6][7][8]. The strain was constructed by overexpressing L-Ser pathway genes, deleting the L-Ser dehydratase sdaA, and reducing the expression of serine hydroxymethyltransferase (SHMT) encoded by glyA and was cultured with an external folate supply [6][7][8]. Replacement of folate with corn steep liquor led to the production of an L-Ser titer of 43 g/L in 96 h by another C. glutamicum strain with random mutation and minimization of by-product synthesis [9]. However, as a host, Escherichia coli has attracted attention due to its well-characterized genetic background, amenability to genetic manipulation, and faster growth rate, higher fermentation intensity and better utilization rate of glucose in direct fermentation [10]. More importantly, L-Ser can be produced from glucose by fermentation with a higher theoretical yield (approximately 1.34 g/g glucose) in E. coli than in C. glutamicum (1.16 g/g) [11].
Recently, Hemanshu Mundhada et al. developed a strain of E. coli with a production capacity of 11.7 g/L L-Ser via ultraviolet radiation on a 1 L scale [1]. Then, this group increased the production capacity to 37 g/L by adaptive laboratory evolution [12]. Subsequently, translation initiation optimization led to the production of an L-Ser titer of 50 g/L with a yield of 0.36 g/g from glucose in 60 h by strain ALE-5 via evolution engineering this L-Ser production capacity is the highest of any strains to date [13]. However, the application of many random mutagenesis evens makes the description of e cient L-Ser synthesis mechanisms di cult. Furthermore, previous studies have shown that repeated random mutation often leads to unknown mutations at some locations in the genome together with targeted mutagenesis [14]. The impact of these unknown mutations is di cult to determine. Moreover, whether the unknown mutations affect universally industrialized fermentation. All these factors limit the application of these strains in industrial production.
In this study, an L-Ser-producing strain was constructed from E. coli W3110 by introducing a series of de ned genetic manipulations (Fig. 1). A basic production strain was constructed by strengthening the L-Ser biosynthesis. Then L-Ser degradation and conversion pathways were optimized. Higher production was further achieved by engineering the L-serine transporter system. The rational design strategies described here signi cantly improved Lserine productivity and yield in fed-batch fermentation.

Results And Discussion
2.1 Construction of the L-serine production strain from E. coli W3110 As a prerequisite for L-Ser production, the activity of the branch pathway leading to L-Ser biosynthesis, which involves serA, serC and serB, was enhanced. PGDH, encoded by serA, catalyzes the initial reaction in L-Ser biosynthesis and the catalytic activity of PGDH can be regulated by feedback inhibition by L-Ser in E. coli [23]. The feedback inhibition was overcome by mutation of two residues (344 and 346) to alanine, as previously described, which would remove the hydrogen bonds between L-Ser and the regulatory binding domain. This led to the mutated gene named serA fr [24]. The feedback resistance of the enzyme PGDH, encoded by serA fr , was investigated by overexpressing these genes in BL21(DE3) via the pT7-7 vector. The activity of serA fr could was sustained at 95% with 80 mmol/L L-Ser, whereas the activity of the wild-type enzyme remained at only 10% ( Fig. 2A). Then, serA fr , serC, and serB were overexpressed in the low copy number pSC vector containing the PR or PL promoter with resulting in SP-01, SP-02 and SP-03 (Fig. 2B).
To produce L-Ser, the sdaA gene encoding the L-Ser-speci c dehydratase was rst deleted from E. coli W3110 to construct the SSW-01 strain. Subsequent deletion of glyA, encoding SHMT, which converts L-Ser to glycine, resulted in the double knockout SSW-02 strain. To evaluate the L-Ser production capacity, the resulting plasmids SP-01 (SP-serA fr ), SP-02 (SP-serA fr C) and SP-05(SP-serA fr BC) were transformed into SSW-02. As shown in Fig. 2C, strain SSW-02/SP-01 was grown in M9-yeast medium supplemented with 50 mmol glucose, and the nal concentration of L-Ser was 155 mg/L after 15 h in a shake ask. An L-Ser concentration of 220 mg/L, 42% higher than that obtained by culturing SSW-02/SP-01, was obtained by culturing SSW-02/SP-02.
SSW-02/SP-05 attained the highest L-Ser concentration, 270 mg/L, which exhibited 1.74-fold increase compared to only overexpressing serA fr . The L-Ser accumulation pro le shown in Fig. 2C, indicates that the production of L-Ser increased as more biosynthetic genes were overexpressed. Furthermore, previous studies have shown that only 15% of the carbon assimilated from glucose is directed into the L-Ser biosynthetic pathway in E.
coli [3]. Hence, SP-serA fr BCpgk (SP-08) was then constructed to increase the carbon ux from glucose to L-Ser and improve L-Ser productivity via ampli cation of phosphoglycerate kinase encoded, by pgk (Fig. 2B). Flask culture of the recombinant SSW-02/SP-08 strain produced a nal L-Ser concentration of 311 mg/L, 15% higher than that obtained by culturing SSW-02/SP-05 (Fig. 2C). Thus, overexpression of pgk effectively improved the L-Ser production capacity of the strain. To further examine L-Ser production of SSW-02/SP-08, fed-batch fermentation was performed in a 5-L fermenter. The highest L-Ser concentration, 17.7 g/L, was observed at 32 h with a yield of 24% from glucose (Fig. 2D).
2.2 In uence of mutations in glyA on L-serine production and cell growth A previous study showed that attenuation of glyA transcription resulted in increased L-Ser accumulation, a decrease in the purine pool, poor growth and cell elongation (Fig S1, Additional le 1) [25,26]. The same phenomenon was observed in this study; SSW-02 cells were elongated and exhibited unstable growth at the early stage of fermentation. We reprogrammed the predominant one-carbon source metabolism with suppressed SHMT activity to increase the stability of the strains. A series of error-prone PCRs were employed to construc a glyA mutation library [27]. Different reductions in SHMT activity were obtained and examined by transforming the recombinant plasmids harboring glyA mut into BL21(DE3). As shown in Table 4, SHMT encoded by glyA mut (K229G) showed an activity of 0.13 U, which decreased by 41% compared to wild type. The mutant K229G was modeled by SWISSMODEL based on the wild type SHMT (PDB ID, 1DFO). As shown in Fig. 3A, close view of the SHMT K229G mutant compared with the wild type SHMT complexed with cofactor LPL (pyridoxal 5′-phosphate) and THFA (PDB ID, 1DFO). The side chain of the K229, which involved the degradation of L-Ser, was removed to obtain the mutant K229G [28]. Sequentially, the glyA gene in SSW-01 was replaced with the appropriate glyA mut (K229G) via CRISPR/Cas9 to generate SSW-03 (△sdaA glyA mut ). Then, the L-Ser biosynthesis plasmid SP-08 was transformed into SSW-03, and cell growth and L-Ser production were evaluated. As shown in Fig. 3B, glyA mut introduction resulted in a 24% increase in biomass, and cultured cells maintained stable growth throughout repeated experiments. SSW-03/SP-08 produced 21.6 g/L of L-Ser, an increase of 22% compared to SSW-02/SP-08.

In uence of sdaB, ilvA, tdcB and tdcG deletion on L-serine production
The L-Ser production capacity of E. coli was signi cantly increased by overexpression of serA fr , serB, serC and pgk via knockout of the sdaA and mutation of glyA. The four genes other than sdaA and glyA, i.e., sdaB, ilvA, tdcB and tdcG, have been reported to transform L-Ser to pyruvate in E. coli [29,30]. However, previous studies focused mainly on decreasing the degradation of L-Ser by deleting all of these genes simultaneously, and few studies have systematically investigated the individual contribution of these degradation genes to L-Ser production. To prevent the degradation and improve the production of L-Ser, sdaB, ilvA, tdcB and tdcG were knocked out individually in the SSW-03 background to generate strains SSW-05, SSW-06, SSW-07 and SSW-08 (Fig. 4A). The plasmid SP-08 was transformed into these mutant strains to produce L-Ser. As shown in Fig. 4B, strain SSW-05/SP-08, which had sdaB deletion, showed the highest L-Ser production of 26.5 g/L, an increase of 23% compared to SSW-03/SP-08.This result was expected, because the SSW-05/SP-08 biomass was also increased by 16%, and SSW-05/SP-08 showed an L-Ser productivity of nearly 0.87 g/L/h at 28 h. While the ilvA gene was knocked out, the growth of the strains was severely inhibited, and production could not be induced during fermentation of SSW-07/SP-08 (Fig. 4C).
The growth restriction of SSW-06/SP-08 may be due to disruption of branched chain amino acid synthesis by deletion of ilvA [31]. Regarding the tdcB gene, the marginal difference in the L-Ser titer and biomass between the SSW-03/SP-08 and SSW-07/SP-08 strains indicated that deletion of tdcB is insu cient to improve L-Ser production (Fig. 4D). However, fermentation of deletion of tdcG exhibited unexpected results. This tdcG gene knockout strain, SSW-08/SP-08, showed a same biomass and 42% lower L-Ser production than the SSW-03/SP-08 (OD 600 ~ 36, 21.6 g/L) ( Fig. 4B and Fig. 4E). The complex phenomenon associated with SSW-09/SP-08 may be caused by regulation of the expression of the interrupted operon tdcABCDEFG by deletion of tdcG. These results suggested that only deletion of sdaB improved L-Ser production, increasing the L-Ser titer by 23%; thus, SSW-05 with only deletion of sdaB was selected for the following experiment, which would avoid severely affected in cell growth by knockout all L-Ser dehydratases.
2.4 Effect of engineering L-serine transport system on strain productivity Moreover, engineering amino acid transport system is also important to further improving strain productivity by blocking reuptake of amino acid and reducing futile cycles [32,33]. In E. coli, four genes, sstT [34], cycA [35], sdaC [36]and tdcC [37], have been reported to be involved in L-Ser uptake. Notably, sdaC is the only gene described as a highly speci c serine transporter, and deletion of sdaC was found to improve L-Ser production in our recent studies [38]. Thus, the highly speci c L-Ser uptake gene sdaC was deleted from SSW-05 to reduce the unwanted futile cycles caused by L-Ser reuptake; this deletion resulted in strain SSW-10. As shown in Fig. 5A, the SSW-10/SP-08 produced 30 g/L L-Ser with a yield of 0.37 g/g from glucose, 16% higher than that of SSW-05/SP-08. In addition, the nal L-Ser productivity of SSW-10/SP-08 was approximately 0.84 g/L/h, which was almost 1.15-fold that of SSW-05/SP-08. E ux pump is an important component of amino acid transport system and it is known to increase strain tolerance by accelerating the export of amino acid from cells. However, no research to date has reported well-characterized L-Ser exporters in E. coli. ThrE has been identi ed as an L-Ser/L-threonine exporter in C. glutamicum [39]. And thrE family identi ed as amino acid exporters in select bacteria, archaea and eukaryotes, but no homologues were found in E.coli [40]. Here, heterologous expression of thrE was performed to verify if it works in E.coli. Thus, thrE was cloned into the constructed expression vector SP-08 adjacent to the PR promoter, resulting in the plasmid SP-09 (Fig. 5B). This recombinant plasmid was then transformed into SSW-10. Figure 5B shows the fermentation process of SSW-10/SP-09. Overexpression of thrE resulted in a 9% decrease in the nal OD 600 value and an 16% increase in L-Ser production compared to those of SSW-10/SP-08. Although the L-Ser production by the nal strain SSW-10/SP-09 (35.1 g/L) was lower than L-Ser production by the strains constructed by Maja Rennig (50 g/L), the yield (42%) of L-Ser from glucose of strain SSW-10/SP-09 was higher than that of the strains constructed by Maja Rennig (36%) [13]. Moreover, strain SSW-10/SP-09 exhibited highest productivity and yield of L-Ser from glucose observed to date.

Transcriptomic analysis of of E. coli W3110 and SSW-10/SP-09
To investigate the effect of L-Ser fermentation on intracellular metabolism, transcriptomic analyses of E. coli W3110 and SSW-10/SP-09 were performed in the exponential phase. A total of 1679 transcripts were found to be signi cantly different under two criteria (p-value < 0.05 and fold change > 2.0). Transcription levels in central carbon metabolism, including glycolysis, tricarboxylic acid(TCA) cycle and amino acid pathways related L-Ser synthesis, were compared.
Expression of the genes related to most reactions in the glycolysis such as pgi, fabAB, tpiA, eno, and pyk was downregulated in SSW-10/SP-09, while that of pgk, encoding phosphoglycerate kinase, was upregulated due to its expression in plasmid SP-09 (Fig. 6A). In the TCA cycle, expression of most genes were also downregulated in SSW-10/SP-09 (Fig. 6B). As a main machinery for adenosine triphosphate(ATP) synthesis, TCA cycle could produce 12.5 ATP molecules per pyruvic acid (PYR) molecule with important intermediates such as oxaloacetate(OAA) and acetyl-CoA(AcCoA) [41]. Downregulation of TCA cycle might cause inferior growth with less energy supply. However, the mqo gene encoding malate dehydrogenase that convert malate with quinone to oxaloacetate and reduced quinone was upregulated. Reduced quinone could signi cantly decrease global DNA methylation level cells, and cause acute oxidative damage [42]. Reduced quinone rise in SSW-10/SP-09 may be another reason for biomass decrease. In this study, the OD 600 of SSW-10/SP-09 was 24, a decrease of 35% compared to that of E. coli W3110 (OD 600 = 37). Gene sdhC, encoding succinate dehydrogenase (ubiquinone) cytochrome b560 subunit, was related to in oxygen availability and upregulated in SSW-10/SP-09 [43].
Next, we analyzed changes in the expression of genes related to L-Ser production in SSW-10/SP-09 (Fig. 6C). The expression levels of serA, serC and serB increased in varying degrees. Expression of the gene glnA related to conversion from L-glutamic acid (L-Glu) to L-glutamine(L-Gln), which provided NH4 + for L-Ser biosynthesis, was upregulated. It caused damping reaction in L-Glu, L-Gln and 2-oxoglutarate(2-OXO) such as gltB and gltD. Expression of the dsdA encoding D-Ser ammonia-lyase was upregulated. However, expression of cysEKO, ltae and trpAB involved in L-cysteine(L-Cys) and L-tryptophan(L-Trp) biosynthesis were not change. Likewise, SSW-10/SP-09 showed downregulation of glycine cleavage(Gcv) system genes such as gcvT, gcvP and gcvH due to less intracellular glycine(Gly) (Fig. 6D). It could result in a decreased amount of one-carbon units and poor growth [44]. However, metF, encoding 5,10-CH2-THF reductase, involved in one-carbon metabolism drastically increased, which could compensate for one-carbon unit [45]. Expression of the betB encoding the enzymes that converts betaine aldehyde to betaine was upregulated. Betaine could regulate intracellular osmotic pressure and provide methyl [46]. With supplement betaine, production of L-threonine, cobalamin and L-lactate were increased [47]. Expression of genes related to metabolism of L-threonine(L-Thr), a downstream amino acid of L-Ser, was analyzed (Fig. 6E). The expression levels of ilvA, which was involved in both L-Thr and L-Ser dehydration, was decreased.

2.6
Intermediate metabolite analysis of of E. coli W3110 and SSW-10/SP-09 As shown in Fig. 7A, a set of 17 intracellular metabolites, including glycolytic intermediates, intermediate metabolite in TCA cycle and amino acid related L-Ser, were measured. A score plot of the PCA model using 17 intracellular metabolites showed the discrimination of metabolite pro les depending on different strains (Fig. 7B). In the PCA model, the intracellular metabolite pro les of E. coli W3110 and SSW-10/SP-09 were clearly discriminated. Along the axis of PC1 of the score plot, the metabolite pro les of E. coli W3110 were located on the positive side, while the metabolite pro les of SSW-10/SP-09 were located on the negative side. Intracellular glucose-6-phosphate (G6P) concentration of SSW-02/SP-08 increased. It may be caused by downregulated of most downstream genes such as pgi, fabAB and eno in glycolysis ( Fig. 8A and Fig. 9A). Intracellular PYR concentration decreased due to weak glycolysis and e cient carbon ux on L-Ser. In the TCA cycle, 2-OXO concentration and malic acid (MAL) concentration showed no signi cant changes between SSW-10/SP-09 and E. coli W3110. Intracellular L-Ser concentration was 472.5 µg/L/g (DCW) , which was 32-fold of control. Consumption of L-Gln, pitched into the second step of L-Ser biosynthesis, caused damage of its precursor L-Glu. High intracellular L-Thr concentration was in favor of maintaining L-Gly concentration [12,51]. It also was the reason for the lessened concentration of L-valine (L-Val), L-leucine (L-Leu) and L-isoleucine (L-Ile). Higher intracellular L-Thr concentration also caused downregulation of thrABC (encoding homoserine dehydrogenase I, homoserine kinase, and threonine synthetase), which was consistent with the result showed in Fig. 8C, due to its feedback inhibition [21]. Intracellular L-phenylalanine (L-Phe) concentration of SSW-10/SP-09 increased 182% when compared to it of E. coli W3110. While there was no distinct relationship reported between L-Phe and L-Ser production to date.

Conclusions
In this study, a systematic investigation was performed in E. coli to construct an L-serine-producing strain with de ned genetic modi cation. Pure rational metabolic engineering of L-serine-producing strain would provide clearly information for the further improvement, which are di cult to be applied on the random mutagenesis strains because of unknown mutations in their genomes. The key genes (serA fr , serC, serB and pgk) for L-serine biosynthesis were overexpressed. The transformation pathways were optimized by introducing a glyA mutation (K229G) and deleting sdaA and sdaB. Furthermore, L-Ser uptake gene sdaC was deleted and L-serine/L-threonine exporter ThrE was overexpressed. L-serine production of 35 g/L with the highest productivity of 0.98 g/L/h and yield of 0.42 g/g glucose was nally achieved. The analysis of transcriptome and intermediate metabolites was performed to further understand the regulatory mechanisms of L-serine production. The fermentation-based process described herein provides an important step towards the industrial production of L-serine directly from glucose. Moreover, further strain development can be achieved through genetic optimization of SSW-10/SP-09 with completely de ned genomic traits.

Strains, media and materials
Wild-type E. coli W3110 was used as the parent strain for serial engineering of L-Ser production. E. coli 5α was used for cloning and propagation of plasmids. E. coli BL21(DE3) was used for enzyme assays. Further strains constructed in this study are shown in Table 1. For strain construction, cultures were grown at 30 °C or 37 °C C in Luria-Bertani medium (LB; 10 g/L tryptone, 10 g/L NaCl, 5 g/L yeast extract) and supplemented with antibiotics as appropriate.

Construction of gene knockout mutants
Genes were knocked out singly or in combination using the method reported by Kirill A. Datsenko and Barry L. Wanner [15]. The primers used for ampli cation of the kanamycin cassette from the plasmid pKD4 are shown in Table S1, Additional le 2. The plasmids helper pKD13, pKD46 and pCP20 were used for construction of knockout mutants. All gene knockout strains were veri ed by sequencing with primers shown in Table S2, Additional le 3.

Construction of the glyA mutant library
Random mutagenesis was induced by error-prone PCR, and the mutation rate was controlled at 0.66%. Genomic DNA of E. coli W3110 was utilized as the template with the primers glyA-F/R (Table 3). PCR reagents were mixed in a volume of 50 µL according to the following composition: 10 × reaction buffer, 10 pmol each primer, 2 µmol MnCl 2 , 2 µmol MgCl 2 , 1 µmol Taq DNA polymerase and unbalanced dNTPs. PCR products were puri ed and digested with Nde I and Hind III and were then subcloned into the expression vector pT 7-7. BL21(DE3) cell transformed with these expression vectors grew in LB at 30 °C. Sequentially, the glyA sequence in pT7-7-glyA was replaced with these different glyA mutation constructs using site-directed mutagenesis with primers shown in Table S3, Additional le 4.

Chromosomal integration of gly mut constructs
Pcas and PtargetF [16] were synthesized by GenScript (Nanjing, China).The sgRNA primer, including N20 sequences followed by the protospacer adjacent motif (PAM) sequence and donor DNA primer, glyA-D, used in this study are shown in Table 3. Genes were replaced using the method reported by Yu Jiang et al [16]. All gene knockout strains were veri ed by sequencing.

Plasmid construction for overexpression of L-serine biosynthetic pathway components
All plasmids used for plasmid construction are described in Table 2. The low copy number vector SP is a laboratory stock plasmid and contains the temperature-sensitive lambda repressor cItS857 gene and the lambda PR and PL promoters. The vector SP was used as the backbone for all plasmids constructed in this study. The L-Ser biosynthetic genes serA, serB and serC were ampli ed from E. coli W3110 using the primers shown in Table 3. The serA fr mutant was generated by mutating two residues in serA, His344 and Asn346, to alanines by site-directed mutagenesis with the primers shown in Table 3. SerA-p1 and serA-p2 were used to clone serA fr into the Xba I/Nhe I site in SP under the control of the PL promoter, yielding the plasmid SP-01. This plasmid was later used to clone serC, generating SP-02. The gene serB was cloned into the SP-02 vector at the Bgl II and Sca I site, generating SP-05. Subsequently, the gene pgk, encoding phosphoglycerate kinase, was cloned into the SP-05 vector backbone, yielding the vector SP-08. The gene thrE, encoding the L-Ser/L-threonine exporter, was ampli ed from C.glutamicum ATCC 13032. And the resulting 1.7-kb fragment was cloned into the corresponding restriction sites in SP-08, generating in the vector SP-09. 4.6 PGDH and SHMT enzyme assays BL21(DE3)/pT7-7-serA fr cells were harvested at mid-exponential growth phase through centrifugation after induction by isopropyl-beta-Dthiogalactopyranoside (IPTG), and crude extracts were obtained using ultrasonication. PGDH in crude extracts was puri ed by ion exchange chromatography (AKTA) on a Sepharose Fast Flow column, and diethylaminoethyl dextran gel (DEAE) was used as the anion exchange agent [17]. PGDH activity was determined by determination of α-ketoglutarate (α-KG) reductase activity instead of glyceric acid-3-phosphate dehydrogenase activity. The 1-mL reaction system contained 40 mmol/L potassium phosphate buffer (pH = 7.5), 1.0 mmol/L DL-dithiothreitol (DTT), 0.25 mmol/L NADH, 5 mmol/L α-KG and 10-30 µg of the puri ed crude extract [18]. BL21(DE3)/pT7-7-glyA mut growth was induced by IPTG at an OD 600 of 0.5, and the culture was centrifuged to obtain bacterial cells at an OD 600 of 4.
SHMT activity was determined by a continuous spectrophotometric assay using DL-3-phenylserine hydrate and phosphopyridoxal as substrates [19]. Hydrolysis of DL-3-phenylserine hydrate by SHMT was monitored spectrophotometrically at 279 nm to assess the formation of benzaldehyde. The standard curve was generated with a benzaldehyde concentration gradient by spectrophotometry 279 nm in the dark. The assay buffer contained 1 mg of centrifuged bacterial cells and 1 mL of substrate (50 mmol/L DL-3-phenylserine hydrate, 30 µmol/L phosphopyridoxal, pH = 8.0) at 37 ℃. After culture at 30℃, 200 rpm for 1 h, the assay buffer was centrifuged at 5000 rpm for 10 minutes, and the supernatants were evaluated at 279 nm. The production of 1 mol of benzaldehyde per hour with 1 g wet weight of the cell in 1 L assay buffer buffer was de ned as one unit.

Shake ask and fed-batch fermentation
For shake ask studies, a single clone was rst grown in 5 mL of LB for 12-14 h, and 5 mL of the culture was transferred to 100 mL of M9 medium with supplemented 2 g/L yeast extract and 9 g/L glucose for culture in a 500-mL shake ask at 30 °C and 200 rpm. Each culture was induced after 3 h by heating to 38 °C. The shake ask studies were repeated at least three times.
Fed-batch fermentation was conducted in a 5-L bioreactor (Biostat A Plus, Sartorius Stedim, Germany). A single clone was precultured in 50 mL of LB and shaken at 33 °C and 200 rpm for 12 to 14 h. The culture was inoculated into 2.5 L of fermentation medium at a 1:20 (v/v) inoculum:medium ratio atan initial temperature of 33 °C. L-Ser production was induced at an OD 600 of 20 by heating to 38 °C. The agitation, air supplementation and feed rate were changed to maintain the dissolved oxygen (DO) concentration above 30% saturation. The pH was controlled at 6.8 using 30% (w/v) NH 3 ·H 2 O. The DO-stat feeding strategy was employed to supply exhausted nutrients to the fermenter. The feeding solution contained 40% (w/w) glucose.

Sample preparation and extraction for intermediate metabolite analysis
Bioreactor-grown cells were harvested at exponential growth phase after induction. 5 mL of culture was injected into the 20 mL quenching solutions (glycerol/saline, 60/40, v/v) and directly centrifuged at 12,00 rpm at − 20 °C for 3 min. After the removal of the supernatant, cell pellets were resuspended by 5 mL saline and cells were collected by centrifugation at 12,00 rpm at − 20 °C for 3 min. Subsequently, cell pellets were extracted three times by cold methanol as reported previously [20]. Cell debris was removed by centrifugation for 5 min at 12,000 rpm.

Analytical methods
Bacterial growth was monitored by measuring the OD 600 in a spectrophotometer (Beckman Germany), and the glucose concentration was analyzed using an SBA sensor machine (Institute of Microbiology, Shandong, China). The L-Ser from fermentation solution was determined by precolumn derivatization HPLC as follows. Two hundred microliters of cell-free supernatants, 100 µL 1 mol/L triethylamine (TEA, with acetonitrile as the solvent) and 100 µL of 0.2 mol/L phenylisothiocyanate (PITC, with acetonitrile as the solvent) were added to into a 1.5-mL microcentrifuge tube. Then, 400 µL of n-hexane was added to the tube and incubated at room temperature for 1 h. Then, the lower solution layer (200 µL) was mixed with 800 µL of deionized water and ltered through a 0.22-µm PTFE membrane lter (Hydrophilic PTFE Syringe Filter, ANPEL Laboratory Technologies Inc.) [21]. Finally, the ltrates were used for HPLC analysis with a Shimadzu Separations module connected to a Shimadzu SPD-M20A detector set to 256 nm and were separated on an Agilent Organic acids within the cell were determined using AB SCIEX QTRAP 5500 system with WATERS T3 column (4.6 × 150 mm, 3 µm) at 40 °C. Conditions of mass spectrometer were as follows: curtain gas, 35 psi; ion source gas 1, 55 psi; ion source gas 2, 55 psi; source temperature, 550℃; polarity, negative; ionspray voltage, 4500 V. Mobile phase A was 0.1% formic acid in acetonitrile whereas mobile phase B consisted of 0.1% formic acid in water.

Transcriptome data sets
For transcriptome analysis, we only used the transcriptome data sets that were obtained at the exponential growth phase after induction. The samples were frozen immediately in liquid nitrogen and sent to Sangon Biotech (Shanghai, China) for transcriptome sequencing. Differentially expressed genes (DEGs) were identi ed according to the following rules: a log2 fold change (FC) > 2 and a p value < 0.05 [22].   Availability of data and material All data generated and analyzed during this study are included in this published article and its supplementary information les.   The data are presented as the means ± SDs from three measurements. a pT-glyA was used as the positive control.