Metabolic engineering of Escherichia coli for the production of 1,3-diaminopropane, a three carbon diamine

Bio-based production of chemicals from renewable resources is becoming increasingly important for sustainable chemical industry. In this study, Escherichia coli was metabolically engineered to produce 1,3-diaminopropane (1,3-DAP), a monomer for engineering plastics. Comparing heterologous C4 and C5 pathways for 1,3-DAP production by genome-scale in silico flux analysis revealed that the C4 pathway employing Acinetobacter baumannii dat and ddc genes, encoding 2-ketoglutarate 4-aminotransferase and L-2,4-diaminobutanoate decarboxylase, respectively, was the more efficient pathway. In a strain that has feedback resistant aspartokinases, the ppc and aspC genes were overexpressed to increase flux towards 1,3-DAP synthesis. Also, studies on 128 synthetic small RNAs applied in gene knock-down revealed that knocking out pfkA increases 1,3-DAP production. Overexpression of ppc and aspC genes in the pfkA deleted strain resulted in production titers of 1.39 and 1.35 g l−1 of 1,3-DAP, respectively. Fed-batch fermentation of the final engineered E. coli strain allowed production of 13 g l−1 of 1,3-DAP in a glucose minimal medium.

In Pseudomonas aeruginosa, 1,3-DAP is synthesized by the C 5 pathway using glutamate as a precursor, and is directly synthesized from spermidine by spermidine dehydrogenase encoded by the spdH gene 15 (Fig. 1a). In Acinetobacter baumannii, another 1,3-DAP biosynthetic pathway called a C 4 pathway is utilized, where 2-ketoglutarate 4-aminotransferase encoded by the dat gene 16 converts aspartate semialdehyde to L-2,4-diaminobutanoate, which is then converted to 1,3-DAP by L-2,4-diaminobutanoate decarboxylase encoded by the ddc gene 17 (Fig. 1a). Although P. aeruginosa and A. baumannii naturally produce 1,3-DAP to small amounts, there has been no previous effort to produce 1,3-DAP in any species. Since P. aeruginosa and A. baumannii are pathogenic bacteria and are not suitable for industrial application, E. coli was chosen as a host strain for the production of 1,3-DAP in this study.
In this paper, we report the development of an E. coli strain capable of producing 1,3-DAP ( Fig. 2) through systems metabolic engineering strategies that combine rational design of metabolic pathways and large-scale pathway engineering based on in silico flux analysis 18,19 and synthetic small RNA (sRNA)-based target screening 20,21 . Fed-batch culture of the final E. coli strain was also performed in glucose minimal medium for the production of 1,3-DAP.

Results
Tolerance of E. coli to 1,3-DAP. Putrescine and cadaverine, four and five carbon diamines, respectively, were previously reported to be toxic to E. coli 8,9 . Since 1,3-DAP , a three carbon diamine, might also be toxic to E. coli host, the tolerance of E. coli to 1,3-DAP was first assessed prior to strain development. Tolerance was tested by monitoring the growth profile of wild-type E. coli W3110 strain in R/2 medium supplemented with 0, 5,10,15,20,30,40, 50 g l −1 of 1,3-DAP, respectively (Fig. 3). The initial growth rate decreased by 7% compared to that of control (without 1,3-DAP) when 10 g l −1 of 1,3-DAP was present. Cells were able to grow until 20 g l −1 of 1,3-DAP was present, but the initial growth rate was decreased by 77%. No growth was observed when greater than 30 g l −1 of 1,3-DAP was added to the medium.
synthase encoded by the speE gene, might be the primary reason for the inefficiency of C 5 pathway. To test this hypothesis, we performed another simulation with the dAdoMet requirement artificially removed from the reaction (Fig. 1b). The specific production rate using C 5 pathway increased, indicating Figure 2. Metabolic engineering strategies for 1,3-DAP production using C 4 pathway. Red arrow indicates enforced flux achieved by overexpression of target gene using either plasmid-based expression or changing native promoter to trc promoter. Dotted arrow indicates feedback inhibition. Red X indicates genes knocked out, promoter change, or removal of feedback inhibition as indicated. The codon optimized dat and ddc genes are represented as dat opt and ddc opt , respectively. Abbreviations of genes and metabolites are as shown in Fig. 1. Other genes and metabolites shown here are: pfkA, 6-phosphofructokinase I; pfkB, 6-phosphofructokinase II; gltA, citrate synthase; aspC, aspartate aminotransferase; thrA, aspartokinase I; metL, aspartokinase II; lysC, aspartokinase III; asd, aspartate-semialdehyde dehydrogenase; ASPP, L-4aspartyl-phosphate.
Scientific RepoRts | 5:13040 | DOi: 10.1038/srep13040 that our hypothesis was correct. Based on these simulation results, the more efficient C 4 pathway was chosen for constructing the 1,3-DAP biosynthetic pathway in E. coli.
Removal of feedback inhibition and codon optimization. Two E. coli host strains, WL3110 22 and TH02 23 , were employed as the base strains for developing 1,3-DAP producing strains (Table 1). E. coli WL3110 is a lacI mutant of E. coli W3110 for the constitutive expression of genes, which are otherwise IPTG inducible. E. coli TH02 is derived from E. coli WL3110 by removing feedback inhibition of two major aspartokinases (encoded by the thrA and lysC genes) by changing the 1034 th base C with T in the thrA gene and the 1055 th base C with T in the lysC gene 23 . The removal of feedback inhibition was considered important since intracellular accumulation of inhibitory amino acids (i.e. L-threonine, L-isoleucine and L-lysine) could directly block the flux toward 1,3-DAP. To establish the C 4 biosynthetic pathway, original and codon optimized A. baumannii dat and ddc genes were cloned into pTac15K to construct plasmids p15DD and p15DD opt , respectively. The E. coli codon optimized dat and ddc genes were tested as inefficient translation caused by codon usage difference could limit 1,3-DAP production.
Reinforcing oxaloacetate flux toward 1,3-DAP biosynthesis. In order to further increase 1,3-DAP production, rational metabolic engineering was performed. It is well known that higher anaplerotic flux has a positive effect on enhanced production of oxaloacetate (OAA)-derived chemicals 23,24 by increasing OAA pool. Therefore, the ppc gene encoding the phosphoenolpyruvate carboxylase (PPC) was chosen as an overexpression target with an intention to increase oxaloacetate pool, which is a key metabolite toward 1,3-DAP biosynthesis. The native promoter of the ppc gene in the chromosome was replaced with the trc promoter. Unexpectedly, the resulting strain, DP01 (p15DD opt ), produced 0.182 ± 0.002 g l −1 of 1,3-DAP which is only marginally higher than that obtained with the control strain E. coli TH02 (p15DD opt ) (Fig. 5). As an alternative, the co-overexpression of the ppc and aspC gene was tried by replacing native promoters in the chromosome with trc promoter. The resulting E. coli DP02 (p15DD opt ) strain produced 0.282 ± 0.001 g l −1 of 1,3-DAP, which is 54.8% higher than that obtained with the control strain E. coli DP01 (p15DD opt ). We hypothesized that single overexpression of the ppc gene was not effective since the increased OAA pool did not contribute to increasing the flux toward 1,3-DAP Concentrations of 1,3-DAP are: black circle, control (0 g l −1 ); white circle, 5 g l −1 ; blue circle, 10 g l −1 ; green circle, 15 g l −1 ; yellow circle, 20 g l −1 ; orange circle, 30 g l −1 ; red circle, 40 g l −1 ; purple circle, 50 g l −1 . formation through aspartate aminotransferase (encoded by aspC) but increased the flux toward TCA cycle (Supplementary Discussion S1).
Having observed the positive effect of co-overexpression of the aspC and ppc genes on 1,3-DAP production, further overexpression of the ppc and aspC genes was examined in E. coli DP02 strain by the transformation of plasmid p15DD opt ppc and p15DD opt paspC, respectively. Surprisingly however, 1,3-DAP production was decreased by 10.53 and 23.97% compared to the control strain, respectively (Fig. 6). Co-overexpression of both ppc and aspC genes was attempted next. However, after repeated attempts, we could not construct the correct plasmid as spontaneous mutations persisted in the start codon region of ppc gene and stop codon region of aspC gene among other mutations. We consequently decided to use E. coli DP02 (p15DD opt ) for further improvement by metabolic engineering.
Synthetic sRNA-based screening of gene knockout targets. Besides overexpression of the ppc or aspC genes, several genes related with C 4 pathways such as the thrA, lysC, and asd genes were overexpressed to examine if 1,3-DAP production could further be enhanced. Also, the tynA gene that is related to 1,3-DAP degradation pathway was deleted to enhance 1,3-DAP production. However, all efforts were futile and did not result in a strain superior to E. coli DP02 (p15DD opt ) ( Supplementary Fig. S1). Thus, the synthetic regulatory sRNA knock-down system was employed to identify the target genes to be knocked down or knocked out to further increase 1,3-DAP production. For this endeavor, 128 synthetic sRNAs related to 1,3-DAP pathway or metabolic regulation were transformed into E. coli DP02 (p15DD opt ) strain and examined for 1,3-DAP production (Supplementary Table S1). Total of 11 genes showed positive effect ( Fig. 7a; Table 2). Genes related to phosphoenolpyruvate pool (the ptsI and pykF genes), pentose phosphate pathway flux (the pfkA gene), byproducts in downstream (the asnA and thrC gene), and degradation of 1,3-DAP (the tynA gene) showed positive effects. The five most effective targets were the pykF, lpxC, ptsI, pfkA and thrC genes, which upon sRNA knock-down led to 35.05, 13.87, 12.88, 12.87 and 12.65% increase in 1,3-DAP production titer (Table 2); based on these results, the genes were individually knocked out from the chromosome to completely eliminate expression of target genes. Since the lpxC and thrC genes are essential genes, the remaining three genes were knocked out. Unexpectedly, deletion of the pykF and ptsI genes did not show significant increase in 1,3-DAP titer. However, deletion of the pfkA gene resulted in the production of 0.680 ± 0.012 g l −1 of 1,3-DAP, which is a 169.5% increase compared to that obtained with E. coli DP02 (p15DD opt ) strain (Fig. 7b). This new strain is designated as E. coli DP09 (p15DD opt ).
Generally, deletion of the pfkA gene would result in increased flux through pentose phosphate pathway (PPP). The significant increase in 1,3-DAP titer is likely due to the increased NADPH pool through the enhanced PPP flux. To validate this, the PPP flux was measured using [1-13 C]glucose-based flux analysis. Experiments were designed and flux through PPP calculated as previously reported 24 . Briefly, flux through pentose phosphate pathway was determined by the fraction of labeled alanine. When PPP is used, the fraction of labeled alanine (m + 1 labeling pattern) is affected since C-1 is lost in the Gnd reaction in the form of CO 2 . As expected, the E. coli DP09 (p15DD opt ) strain had less m + 1 fraction ( Supplementary Fig. S2), suggesting that it has a stronger PPP flux. Further calculation revealed that 37.97% of total flux is flowing to PPP in the control strain E. coli DP02 (p15DD opt ), whereas 69.76% of total flux is going to PPP in E. coli DP09 (p15DD opt ) strain, the pfkA deletion mutant (Fig. 7c).  Combination of pfkA deletion and ppc/aspC overexpression. Based on the E. coli DP09 (p15DD opt ) strain, further engineering was tried to enhance tolerance of cell to 1,3-DAP by engineering nonspecific 1,3-DAP transporters. It was thought that PotE and/or CadB responsible for longer chain diamine transport might nonspecifically transport 1,3-DAP. Overexpression of PotE and CadB indeed enhanced 1,3-DAP production in flasks ( Supplementary Fig. S3). However, enhanced 1,3-DAP production was not achieved in fed-batch fermentations. Therefore, we applied another strategy for enhanced 1,3-DAP production (See Discussion).
Since aspartate aminotransferase encoded by the aspC gene indirectly requires NADPH as a cofactor (Fig. 2), insufficient NADPH pool was the principal reason why previous effort to enhance 1,3-DAP production by further plasmid-based overexpression of the ppc or aspC gene in E. coli DP02 (p15DD opt ) was not effective for enhancing 1,3-DAP production (Fig. 6); due to insufficient NADPH pool, more flux could not go through aspartate aminotransferase even though the ppc or aspC gene was further overexpressed. Therefore, plasmid-based overexpression of the ppc or aspC gene was re-examined in E. coli DP09 (p15DD opt ppc) and E. coli DP09 (p15DD opt paspC), respectively, in which NADPH pool is increased by the deletion of the pfkA gene. The synergistic effect of the overexpression of the ppc and aspC gene along with the deletion of the pfkA gene on 1,3-DAP production was thus observed; the engineered E. coli DP09 (p15DD opt ppc) and E. coli DP09 (p15DD opt paspC) strains were able to produce 1,3-DAP up to 1.391 ± 0.030 and 1.349 ± 0.039 g l −1 , respectively, which are 104.4 and 98.3% higher than that obtained with the control strain DP09 (p15DD opt ) (Fig. 6).

Fed-batch fermentation.
Since both E. coli DP09 (p15DD opt ppc) and DP09 (p15DD opt paspC) strains showed similar performance in flask cultivation, fed-batch fermentations of both strains were performed using pH-stat feeding strategy 25,26 . Briefly, the pH-stat feeding strategy is based on the rise of pH upon depletion of glucose due to the excretion of ammonium ions by the cells. When the pH rises beyond a set point of 7.02, glucose-containing feeding solution is automatically added (see Methods). At the end of batch phase, E. coli DP09 (p15DD opt ppc) and E. coli DP09 (p15DD opt paspC) strains produced 0.74 and 1.00 g l −1 of 1,3-DAP using 10 g l −1 of glucose in 16.33 and 23.83 h resulting in the yields of 0.074 and 0.100 g 1,3-DAP per gram glucose, respectively (Fig. 8); in the batch mode, 10 g l −1 of glucose was used to maintain 1,3-DAP concentration lower than the level negatively affecting cell growth as found in 1,3-DAP toxicity test. At the end of fed-batch culture, the DP09 (p15DD opt ppc) and DP09 (p15DD opt paspC) strains grew to the maximal OD 600 values of 110.5 and 143.5, corresponding to 61.88 and 80.36 g dry cell weight per liter, and produced 1,3-DAP to 12.80 and 13.06 g l −1 in 97.66 and 69 h, respectively (Fig. 8). The 1,3-DAP titers obtained with both strain were similar, but the 1,3-DAP productivity (0.19 g l −1 h −1 ) obtained with the DP09 (p15DD opt paspC) was higher than that (0.13 g l −1 h −1 ) obtained with the DP09 (p15DD opt ppc).

Discussion
Bio-based nylons and engineering plastics of various types, which have great industrial potential, can be manufactured by co-polymerization of diamines and dicarboxylic acids of different carbon lengths. We    and others previously reported production of 1,4-diaminobutane (putrescine) and 1,5-diaminopentane (cadaverine), four carbon-and five carbon-diamines, respectively, by metabolically engineered bacteria. However, three carbon-diamine, 1,3-DAP, has never been produced. This paper is the first report on The results of the toxicity test indicated that 1,3-DAP is toxic to E. coli. Although production of 1,3-DAP by fermentation is feasible, the final production titer of 1,3-DAP is limited to some extent. This problem can be addressed in the future by increasing the tolerance of cells to 1,3-DAP and/or by in situ removal of 1,3-DAP during the fermentation. Since 1,3-DAP is a non-natural metabolite in E. coli, an unknown nonspecific transporter is responsible for exporting 1,3-DAP out of the cell. We believe the gene encoding this transporter to be a key overexpression target for two reasons. First, the 1,3-DAP exporting activity would be insufficient since it is not specific to 1,3-DAP. Second, overexpression of transporter will decrease intracellular concentration of 1,3-DAP eliminating the toxic effect of 1,3-DAP to host cell. We hypothesized that the transporters of putrescine and cadaverine (PotE and CadB, respectively) might function as nonspecific transporters of 1,3-DAP due to their structural similarities. Indeed, overexpression of potE and cadB was effective in increasing 1,3-DAP production in flasks as mentioned in Results section. However, fed-batch fermentation of a cadB overexpressed strain (which showed higher 1,3-DAP titer in flask) showed significant growth retardation, resulting in lower production titer than the control strain (data not shown). Further experiments are required to overcome growth retardation caused by potE or cadB overexpression for improved 1,3-DAP production in fed-batch culture.
One of the key strategies employed for constructing 1,3-DAP overproducer was the sRNA knock-down technology. Although knock-down targets were identified, we decided not to introduce additional gene (encoding corresponding sRNA) to reduce cellular burden during industrial-scale fermentation in the future; thus, the corresponding target genes identified were knocked out from the chromosome instead. Excluding two essential genes, three genes (pykF, ptsI and pfkA) were identified as promising knock-out targets. At first, it was thought that intracellular phosphoenolpyruvate pool was key for enhanced 1,3-DAP production since both pykF and ptsI are responsible for the phosphoenolpyruvate pool. However, deletion of both pykF and ptsI was not effective for 1,3-DAP production (Fig. 7b). This result is probably due to the difference between knock-down using sRNA systems and complete knock-out by the deletion of gene from the chromosome; in the case of pykF and ptsI, knock-down of gene expression using sRNA system was effective over chromosomal deletion which can be explained as appropriate expression level of pykF or ptsI is required rather than no expression. Although not pursued in this study, optimal expression levels of pykF and ptsI will be beneficial for further increasing 1,3-DAP production.
After increasing the intracellular NADPH pool by pfkA deletion (identified by sRNA screening), further plasmid-based overexpression of ppc or aspC resulted in a dramatic increase of 1,3-DAP titer (Fig. 6). Unfortunately, plasmid-based co-overexpression of ppc and aspC could not be performed as we could not construct the plasmid due to spontaneous mutations in the genes (see Results). The reason for this phenomenon is currently unknown.
In silico metabolic simulation revealed that the maximum 1,3-DAP production rates using C 4 and C 5 pathways were 13.38 and 4.80 mmol per gram dry cell weight per hour, respectively, both at zero cell growth (Fig. 1b). Since the glucose consumption rate was set as 10 mmol per gram dry cell weight per hour during the simulation, the calculated theoretical maximum 1,3-DAP yields using C 4 and C 5 pathways are 1.338 and 0.480 mol per mole glucose, respectively; on gram basis, these values are equivalent to 0.550 and 0.197 g per gram glucose, respectively. Although these theoretical yields cannot be achieved as they are obtained at zero cell growth, the 1,3-DAP yield of 0.100 g per gram glucose obtained with our engineered strain DP09 (p15DD opt paspC) is much lower than the theoretical maximum yield; thus, further metabolic engineering is required to enhance 1,3-DAP production and yield. In this study, we reported development of engineered E. coli strain capable of producing 1,3-DAP. Identification of the optimal heterologous pathway (the C 4 pathway involving the dat and ddc genes) by in silico flux analysis, optimization of codons of the dat and ddc genes, removal of feedback inhibition by amino acids, reinforced flux from OAA toward 1,3-DAP biosynthesis, sRNA-based knock-down screening of candidate genes, and combination of all of these resulted in the final strain capable of producing 13.1 g l −1 of 1,3-DAP in glucose minimal medium by aerobic fed-batch fermentation. This strain can serve as a base strain for developing more efficient 1,3-DAP producer, and the strategy described here will be generally useful for developing microbial strains capable of producing other L-aspartate derived chemicals.

Methods
Bacterial strains, plasmids, and genetic manipulation. All bacterial strains and plasmids used in this study are listed in Table 1. E. coli NEB 10-beta (New England Biolabs, Ipswich, MA) strain was used for cloning studies. E. coli WL3110, which is a lacI mutant of E. coli W3110, and E. coli TH02, which is a mutant of E. coli WL3110 containing feedback resistant aspartate kinase Ι and III, were used as starting host strains for the construction of 1,3-DAP overproducing E. coli strains. Routine cultures for the construction of plasmids and strains were performed in Luria-Bertani (LB) broth or on LB plates (1.5%, w/v, agar) containing appropriate antibiotics. When needed, antibiotics were added at the following concentrations: kanamycine (Km), 25 μ g ml −1 ; ampicillin (Ap), 50 μ g ml −1 ; chloramphenicol (Cm), 17.5 μ g ml −1 . All DNA manipulations were performed by standard procedures.
All primers used in this study were synthesized at Genotech (Daejeon, Korea), and are listed in Supplmentary Table 2. The A. baumannii dat and ddc genes were amplified by PCR using the primers P1 + P2 and P3 + P4, respectively, and genomic DNA of A. bumannii as a template. Using the amplified fragments, plasmid p15DD was constructed by sequential insertion of the dat and ddc genes into pTac15k at EcoRI/KpnI and KpnI/PstI sites, respectively. Similarly, the E. coli-codon optimized A. baumannii dat and ddc fragments (Supplementary Table S3), which were designed by SynBioCodonOpt (http://pombe.kaist.ac.kr/codon/), were similarly cloned into pTak15k to make p15DD opt .
Plasmids p15DD opt ppc and p15DD opt aspC were constructed by cloning the E. coli W3110 ppc and aspC genes, respectively, at 16 bp downstream of the codon optimized ddc gene by Gibson assembly 27 . Plasmid p15DD opt and the ppc or aspC gene were mixed at molar ratio of 1:1 and incubated at 50 °C for 1 h, followed by transformation. The primers P5 and P6 were used to amplify the backbone plasmid p15DD opt and the primers P7 + P8 and P9 + P10 were used to amplify the ppc and aspC genes using genomic DNA of E. coli W3110 as a template. Plasmids p15DD opt ppc and p15DD opt aspC were obtained by Gibson assembly of the ppc and aspC genes, respectively, with p15DD opt . Finally, plasmid p15DD opt-paspC was constructed by adding the tac promoter in front of the aspC gene in p15DD opt aspC by Gibson assembly. For this, the primers P6 and P11 were used to amplify p15DD opt aspC, while the primers P12 and P13 were used to amplify tac promoter using pTac15k as a template; these were assembled to make p15DD opt paspC. Colonies were selected on LB agar plates containing Km. Recombinant strains harboring the correct plasmid was selected by colony PCR and further confirmed by sequencing.
All promoter change and deletion of chromosomal genes were carried out by one-step inactivation method 28 . A linear DNA fragment for changing the ppc promoter was amplified by PCR using the primers P14 and P15 and pMtrc9 (Table 1) as a template. Similarly, a linear DNA fragment for changing the aspC promoter was amplified by PCR using the primers P16 and P17 and pMtrc9 as a template. A linear DNA fragment for pfkA deletion was made by PCR using the primers P18 and P19 and pECmulox (Table 1) as a template. Detailed procedures for promoter change or deletion of the genes have been described previously 8 . Briefly, amplified fragments for target gene promoter change or deletion were eletroporated into E. coli strains harboring pKD46 (Table 1), which expresses λ -Red recombinase. Colonies were selected on LB agar plates containing Cm. Successful gene promoter change and deletion mutant having the lox71-Cm R -lox66 cassette were confirmed by direct colony PCR. The Cm resistant marker was subsequently eliminated by a helper plasmid pJW168 (Table 1), which contains a temperature sensitive replication origin and the IPTG-inducible Cre recombinase. The loss of Cm R was further verified by PCR.
Batch and fed-batch cultivation. Flask cultures were performed in R/2 medium (pH 7.0) supplemented with 10 g l −1 glucose and 3 g l −1 (NH 4 ) 2 SO 4 . R/2 medium (pH 7.0) contains per liter: 2 g (NH 4 ) 2 HPO 4 , 6.75 g KH 2 PO 4 , 0.85 g citric acid, 0.8 g MgSO 4 ·7H 2 O and 5 mL trace metal solution 8 . Glucose, MgSO 4 ·7H 2 O and (NH 4 ) 2 SO 4 were sterilized separately and added to R/2 medium later. Cells were grown in a test tube containing 10 mL of LB medium. After 8 h of cultivation, 1.5 ml culture was used to inoculate a 300 ml baffled flask containing 50 ml of culture medium, and cultivated at 37 °C and 200 r.p.m. in a shaking incubator. After 36 h, cells were harvested for further analysis. All flask cultures are done in triplicates.
Batch fermentation was conducted in a 6.6-l jar fermentor (Bioflo 3000; New Brunswick Scientific Co., Edison, NJ) containing 2 l of culture medium. Culture condition was the same as that for flask cultures. Seed culture (200 ml) was prepared by transferring 3 ml of culture broth grown for 8 h in a 250 ml Erlenmeyer flask containing 100 ml LB medium. Seed culture was transferred to a fermentor to have an Scientific RepoRts | 5:13040 | DOi: 10.1038/srep13040 initial OD 600 of ca. 0.25. Culture pH was adjusted to 7.0 by adding 10 M of KOH. Air was constantly flowed at 2 l min −1 and the dissolved oxygen (DO) level was kept at 20% of air saturation by automatically controlling the agitation speed from 200 to 1000 r.p.m..
Fed-batch fermentation started with batch fermentation described above. As in batch fermentation, pH of culture was kept 7.0 by adding 10 M KOH. Appropriate amount of feeding solution was added by employing the pH-stat feeding strategy as follows. When the pH of culture rose to 7.02, feeding solution was added until pH was lower than 7.02. The feeding solution contained 522 g l −1 of glucose, 8 g l −1 of MgSO 4 ·7H 2 O and 170 g l −1 of (NH 4 ) 2 SO 4 . The DO level was kept at 20% of air saturation by automatically adjusting agitation speed, and by supplying pure oxygen when the maximum agitation speed of 1000 r.p.m. was reached.
In silico flux response analysis. The genome-scale metabolic model EcoMBEL979 29 consisted of 979 metabolic reactions and 814 metabolites, which is a modified model of iJR904 30 , was used for in silico flux response analysis 23 . Since 1,3-DAP biosynthetic pathways (C 4 and C 5 pathways) do not exist in E. coli, heterologous metabolic reactions need to be additionally introduced to the model of EcoMBEL979. For the C 4 biosynthetic pathway, the following reactions were added to the EcoMBEL979 model: 'L-aspartate-semialdehyde + L-glutamate ↔ L-2,4-diaminobutanoate + alpha-ketoglutarate' , and 'L-2,4 diaminobutanoate + H + → 1,3-DAP + CO 2 ' . For the C 5 biosynthetic pathway, the following reaction was added: 'spermidine + an oxidized electron acceptor + H 2 O → 1,3-DAP + 4-aminobutanal + an reduced electron acceptor' . To investigate the theoretical effects of different metabolic pathways (C 4 and C 5 pathways) on 1,3-DAP production, constrained-based flux analysis was performed under pseudo steady-state assumption. Also, flux analysis was performed again to analyze the reason for the inefficiency of C 5 pathway by artificially removing dAdoMet and S-methyl-5'-thioadenosine from spermidine synthase reaction in the model, which modifies spermidine synthase reaction to 'putrescine → spermidine' . Cell growth rate was maximized as an objective function while gradually increasing the 1,3-DAP biosynthesis flux from minimum to maximum values. During the simulation, the glucose uptake rate was set at 10 mmol gDCW −1 h −1 .
Isotopic analysis. Sample preparation and gas chromatography-mass spectrometry (GC-MS) analysis were performed as described in a previous report 31 . The 13 C-labled biomass samples were harvested during mid-exponential growth phase (OD 600 of ~1.2) from flask cultivation containing 3 g l −1 of glucose and 1 g l −1 of [1-13 C] glucose. Cell pellets were hydrolyzed in 500 μ l of 6 M HCl at 105 °C for 24 h. The cell hydrolysate was dried at 95 °C for 5 h and then derivatized at 85 °C with 40 μ l dimethylformamide and 40 μ l N-tert-butyldimethylsilyl-N-methyltrifluoroacetamide for 30 min. The derivatized compounds were analyzed by the Agilent 6890N combined with a 5875XL MS system equipped with Agilent 7683 automatic injector and a column (HP-5MS, 30 m, ID 0.25 mm, film thickness 0.25 μ m, Agilent Technologies, PaloAlto, CA, USA). The oven temperature started at 140 °C, increasing to 310 °C at 12 °C min −1 , and held for 2 min. Naturally occurring isotopes of amino acids were corrected using a correction matrix 32 .
Analytical methods. Cell growth were monitored by measuring the optical density at 600 nm (OD 600 ) with Ultrospec 3000 spectrophotometer (Amersham Biosciences, Uppsala, Sweden). Glucose concentration was determined by glucose analyzer (model 2700 STAT; Yellow Springs Instrument, Yellow Springs, OH). Concentration of 1,3-DAP was measured by high performance liquid chromatography (1100 Series HPLC, Agilent Technologies, Palo Alto, CA). Automatic precolumn derivatization of 1,3-diaminopropane was performed by o-phthaldialdehyde (OPA; Sigma, St. Louis, MO) in HPLC. The OPA derivatization reagent was prepared as described previously 33,34 . Briefly, 0.20 g of OPA was dissolved in 9.0 ml of methanol, followed by the addition of 1.0 ml of 0.40 M (pH 9.0) borate buffer and 160 μ l of 2-mercaptoethanol (reducing regent). All samples, OPA reagent and borate buffer were filtered through a 0.2 μ m PVDF syringe filter (Whatman, Maidstone, UK). For derivatization, 1 μ l of the sample was mixed with 5 μ l of 0.40 M (pH 9.0) borate buffer. Following the addition of 1 μ l of the OPA reagent, the mixture was injected into HPLC. A Discovery C18 column (cat# 504955; 5 μ m, 4.6 mm x 15 cm) was used and operated at 25 °C. The mobile phase comprised solvent A (55% methanol in 0.1 M sodium acetate, pH 7.2) and solvent B (methanol), and was flown at 0.8 ml min −1 . The following gradient was applied: 1-6 min, 100% A; 6-10 min, a linear gradient of B from 0% to 30%; 10-15 min, a linear gradient of B from 30% to 50%; 15-19 min, a linear gradient of B from 50% to 100%; 19-23 min, 100% B; 23-25 min, a linear gradient of B from 100% to 30%; 25-28 min, a linear gradient of B from 30% to 0% (all in vol%). The standard 1,3-DAP used was purchased from Sigma (D23807), which was also treated for analysis as described above. The derivatized 1,3-DAP was detected by a variable wavelength detector (G1314A,