Multiplex Design of the Metabolic Network for Production of l-Homoserine in Escherichia coli

In this study, the bottlenecks that sequentially limit l-homoserine biosynthesis were identified and resolved, based on rational and efficient metabolic-engineering strategies, coupled with CRISPR interference (CRISPRi)-based systematic analysis. The metabolomics data largely expanded our understanding of metabolic effects and revealed relevant targets for further modification to achieve better performance. The systematic analysis strategy, as well as metabolomics analysis, can be used to rationally design cell factories for the production of highly valuable chemicals.

aspartate metabolic pathway (Fig. 1). The accumulation of L-homoserine was derived from an L-threonine-or L-methionine-producing strain. It was found that most efforts were focused on regulating gene expression of the L-aspartate metabolic pathway (8,9). Generally, three aspartokinase isoenzymes (AKI, AKII, and AKIII) that catalyze the first step in L-homoserine biosynthesis were overexpressed to enhance the carbon flux to L-homoserine production. In addition, the transporter system was enhanced to improve survivability under the stress of L-homoserine (10). Due to the lack of optimization upstream of glycolysis, as well as branched pathways (11), L-homoserine was produced with a low yield by mutant and metabolically engineered strains, including Corynebacterium glutamicum (12) and Escherichia coli (13,14). Until now, the optimization of each gene participating in different pathways and the identification of bottlenecks limiting L-homoserine biosynthesis in E. coli have rarely been reported.
Currently, pathway design and optimization are essential to obtain microbial cell factories with excellent performance for industrial production of commercial chemicals (15,16). A series of attempts have been made to construct novel producers for biobased chemicals as the technologies in support of synthesizing new materials and manipulating genes have matured (17)(18)(19). Indeed, some strategies have been applied to optimize the biosynthetic pathways, which consist of large numbers of genes, or to reprogram gene expression to manipulate complex phenotypes. Prominent examples of such methods include the multidimensional heuristic process (MHP) (20), multiplex navigation of global regulatory networks (MINR) (21), and multibranched and multilevel regulated biosynthetic pathways (MBMRPs) (22), which rapidly engineer organisms with desired capabilities without the trial and error of iterative experimentation. On the other hand, "omics" profiling technologies, such as metabolomics, have facilitated an overview of cell metabolism, allowing a more in-depth insight into intracellular mechanisms in modified-organism analysis (23).
Here, the initial L-homoserine-producing strain was obtained by blocking the degradative and competitive pathways and overexpressing thrA (encoding homoserine dehydrogenase) based on an O-succinyl homoserine-producing strain (24). Then, ratio- nal design, including reinforcement of the transport system, redirection of the carbon flux, the CRISPR interference (CRISPRi) system, and module integration strategy, were used to construct an L-homoserine-producing strain. The iteratively modified strain HS33 could produce 7.25 g/liter L-homoserine from glucose in a shake flask. Metabolomics analysis further revealed the intracellular behaviors (metabolic differences) in response to genetic modifications to direct carbon to L-homoserine formation. Based on the metabolomics analysis, the anaplerotic route afforded by pyruvate carboxylase was introduced to direct carbon toward L-homoserine, which resulted in accumulating 8.54 g/liter L-homoserine (0.33 g/g glucose, 62.4% of the maximum theoretical yield) in shake flask cultivation. Finally, 37.57 g/liter L-homoserine was produced under fedbatch fermentation, with a yield of 0.31 g/g glucose.

RESULTS
Construction of an initial L-homoserine-producing strain. The "pull-push-block" strategy is an efficient method to engineer microorganisms involved in biosynthesizing target products by modifying metabolic networks (25,26). In our previous study, the E. coli strain ΔJIB* TrcmetL with overexpression of metL and deletion of the metJ, metI, and metB genes was constructed to produce O-succinyl homoserine from L-homoserine and succinyl-coenzyme A (CoA) (24). To construct the basic L-homoserine-producing strain HS1, L-homoserine-converting pathway-related genes (thrB, encoding homoserine kinase, and metA, encoding homoserine O-succinyltransferase) were successively deleted to "block" L-homoserine degradation. Then, the L-homoserine-converting pathway was further strengthened by overexpression of thrA to "push" the carbon flux to L-homoserine production. Subsequently, the lysine-auxotrophic strain HS2 was generated by deleting lysA to investigate the effect of eliminating a precursor competing metabolic pathway on L-homoserine production. As shown in Fig. 2, HS1 could accumulate 1.85 g/liter of L-homoserine in MS (minimal salt) medium. The disruption of L-lysine biosynthesis increased the production of HS2 to 2.01 g/liter with the same specific production of 0.33 g/g cell dry weight (CDW) when L-lysine (0.025 g/liter, optimized amount) was added (see Fig. S1 in the supplemental material).
Modification of the transport system for improvement of the L-homoserine titer. Based on observations that L-homoserine inhibits the activity of aspartokinase (encoded by metL) and L-glutamate dehydrogenase (encoded by gdhA) (27,28), toxicity stress arising from gradual accumulation of product results in repressing cell growth and product yield (29,30). Therefore, strengthening the capability of the L-homoserine transport system and the transformation of other toxic intermediate metabolites are top priorities. rhtA, encoding the inner membrane transporter that is involved in the export of L-homoserine (31), was overexpressed chromosomally by replacing the native promoter with the trc promoter to obtain strain HS3 (Trc-rhtA). Strain HS3 increased the production of L-homoserine by 30.9% to 2.63 g/liter (Fig. 3). It was reported that overexpression of eamA confers resistance to toxic chemicals (32). In order to further increase the L-homoserine export capacity and relieve the growth burden of homoserine-producing strains to enable survival, the native promoter of the eamA gene was replaced by the trc promoter in strain HS4 (Trc-eamA). Batch cultivation of HS4 resulted in the production of 2.17 g/liter L-homoserine. Moreover, two rhtA gene copies (the native rhtA gene and replacement of the lacI gene) and eamA were overexpressed under the trc promoter in the chromosome to construct strain HS5 (ΔlacI::Trc-rhtA Trc-rhtA Trc-eamA). Under batch culture, strain HS5, with modification of the transport system and construction of a constitutive expression system, could produce 3.14 g/liter L-homoserine, which was 54.2% higher than strain HS2 production. In addition, the specific production of strain HS5 was also increased (Fig. 3), which confirmed that enhancing the transport system was beneficial for the improvement of L-homoserine productivity.
Effect of a glyoxylate shunt on L-homoserine production. The deletion of the iclR gene can activate the glyoxylate shunt, which increases the availability of malate as the precursor of oxaloacetate (2,33). To further direct the carbon flux to L-homoserine, the transcriptional regulator IclR was removed to construct strain HS6. As shown in Fig. 4, the final L-homoserine concentration obtained in strain HS6 with iclR deleted was equal to that in the control strain. Additionally, the AspA-catalyzed reaction is an efficient alternative to the AspC-catalyzed reaction in the conversion of fumarate and oxaloacetate to L-aspartate (34,35). Therefore, the deletion of fumarase genes (fumA, fumC, and fumB) and overexpression of aspA were implemented. Under batch culture, both strains HS7 (with fumAC deleted) and HS8 (with aspA overexpressed) showed lower L-homoserine production than HS6 (Fig. 4).
Repression of candidate genes by the CRISPRi system to further enhance L-homoserine production. To identify the target genes to be further manipulated, the CRISPRi system for sequence-specific control of gene expression was used on the L-homoserine production strain HS6. Specifically, the homoserine-forming pathway was partitioned into three modules, which were separated at GAP (D-glyceraldehyde 3-phosphate) and OAA (oxaloacetate). These metabolic nodes are involved in the glycolytic pathway, by-product production, and amino acid biosynthesis. The results for strains with target gene downregulation were measured and compared with the control harboring the pTarget-null plasmid without an N20 sequence [HS6 (pdCas9, pTarget-null)]. As shown in Fig. 5a, 39 genes were chosen to study the effects of target gene repression on L-homoserine production. Compared with the control group, HS6(pdCas9, pTarget-null), the strains with single guide RNAs (sgRNAs) targeting genes exhibited differences in the accumulation of L-homoserine. The L-homoserine production of strains with sgRNA-directed downregulation of ptsH, ptsI, crr, ptsG, tktA, rpe, talB, argA, argG, proB, and gadA in three modules were increased by more than 100%. On the other hand, the sgRNAs targeting zwf, pta, and poxB also increased the accumulation of L-homoserine by 50% to 100% (Fig. 5a).
On the basis of these results, candidate genes with promising positive targets for L-homoserine yields were selected for further individual deletion. As such, ptsG, ptsH, ptsI, crr, tktA, rpe, talB, zwf, ackA, pta, ilvA, poxB, argG, gadA, and proB were removed individually based on strain HS6 to construct a series of novel L-homoserine producers (strains HS9 to HS23). As shown in Fig. 5b, the cell growth of these engineered strains was variable; in particular, growth of the strain with deletion of ptsI was seriously inhibited. Additionally, the results showed that the strains with individual deletions of ptsG, ptsH, crr, tktA, rpe, pta, argG, and proB accumulated higher L-homoserine content than the control strain. In particular, strains HS9 (ΔptsG) and HS12 (Δcrr) with perturbation of the phosphotransferase system (PTS) showed L-homoserine production at 5.27 g/liter and 5.52 g/liter, respectively, which was higher than that of other mutations. These positive genes, which were screened out by the CRISPRi system, should be further investigated to identify the rate-limiting steps blocking L-homoserine biosynthesis. Restoration of glucose uptake by non-PTS sugar transporter modification. On the basis of the above-mentioned results, the effects of the iterative genetic modifications of the selected genes not participating in L-homoserine biosynthesis were investigated. In this study, combinational genetic perturbation targeting the PTS and the pentose phosphate pathway was performed. To provide ample upstream pathway strength, a new set of five strains (HS24 to HS28) were engineered on the basis of perturbation of the PTS equilibrium, on which iterative genetic modifications were carried out between the PTS and the pentose phosphate pathway. Unexpectedly, the results showed that the L-homoserine yield was reduced in these further engineered strains (Fig. 6a). Given the differences in cell growth and major by-product accumulation between HS9 and HS12, the ΔptsG strain (HS9) was chosen as the candidate for further investigation (Fig. 6b). Due to the decrease in glucose uptake caused by the ptsG gene deletion, the promoter region of the glk gene (encoding glucokinase) was replaced with the trc promoter, and inactivation of galR led to constitutive expression of a non-PTS sugar transporter to restore glucose transport capacity for efficient L-homoserine production (see Fig. S2 in the supplemental material). The L-homoserine titer observed with the HS29 strain was 6.27 g/liter, which was a 19.0% increase compared to the HS9 strain. The major by-products, including acetic acid and ␣-ketoglutarate, were produced at 1.86 g/liter and 2.39 g/liter, respectively, in flask fermentations.
Flux reinforcement by enhancing circulation of the amino donor. It should be clear that glutamate functions as the amino donor for the transformation of target genes, including glycolysis, by-products, and branch pathways, were investigated in this study. The results were drawn from combined data for the single gene target among the genes. Relative changes in L-homoserine yield compared with the negative control (HS5 harboring a pTarget-null plasmid without N20 sequence) are represented as colored circles. Blue circles represent no significant change (Ϫ20% to 5%) in L-homoserine production compared to the negative control. Targets that increased L-homoserine production are shown in red. (b) Cell growth and homoserine production in deletion of positive targets. Control (con) refers to strain HS6. The cells were cultured in MS medium at 28°C for 48 h. The error bars represent SD. oxaloacetate into L-aspartate. On the other hand, ␣-ketoglutarate would be accumulated from L-glutamate under conditions of nitrogen limitation. Not only is ␣-ketoglutarate responsible for catalyzing the first reaction in ammonia assimilation, it coordinates carbon and nitrogen utilization by rapid modulation of the glycolytic flux to alter import of glucose and consumption of phosphoenolpyruvate (PEP) (11). Therefore, in order to convert ␣-ketoglutarate into L-glutamate, strain HS30 was constructed by replacing the wild-type promoter of the gltBD operon (encoding L-glutamate synthase) with a strong trc promoter in strain HS6, and the L-homoserine yield was increased by 23.1% to 3.95 g/liter. This showed that the amino donor (L-glutamate) supply was considered another bottleneck in the biosynthesis of L-homoserine. Additionally, another set of two strains (HS31 and HS32) were constructed by deleting argG and proB based on the gltBD gene upregulation strain HS30 to increase the availability of intracellular L-aspartate and L-glutamate, respectively. However, not only were cell growth and glucose consumption severely repressed, but less accumulation of L-homoserine was observed, especially for strain HS32 (Fig. 7a).
These observations prompted the construction of strain HS33, which was iteratively modified by upregulating the expression of L-glutamate synthase based on strain HS29 to remove the deeper rate-limiting step. The successful overexpression of selected genes was subsequently verified by reverse transcription-quantitative PCR (RT-qPCR) (see Fig. S3 in the supplemental material). Notably, the observed L-homoserine production by strain HS33 was increased by 15.6% to 7.25 g/liter without affecting cell growth and the consumption of glucose (Fig. 7b). Also, the concentration of the by-product ␣-ketoglutarate was further decreased to 1.63 g/liter (Fig. 7b), indicating that upregulation of the gltBD operon could effectively facilitate the conversion of ␣-ketoglutarate into L-glutamate. It was concluded that the multidirectional control of  metabolic pathways associated with metabolic strategies and CRISPR-based transcriptional regulators enabled efficient carbon utilization and recovery of the amino donor for production of L-homoserine.
Metabolic variation response to iterative genetic modifications. An intracellularmetabolite profiling study was performed in E. coli W3110 and HS33 to further investigate the effectiveness of the above-mentioned iterative genetic modifications and to identify important metabolic pathways that are closely associated with overproduction of L-homoserine (Fig. 8). In strain HS33, the intracellular levels of glycolytic intermediates, such as glucose-6-phosphate (G6P), pyruvate, and acetyl-CoA, increased 2.3-, 2.2-, and 18.2-fold, respectively. Additionally, a 3.3-fold increase in AMP was also detected in the strain. These results indicated that in strain HS33, disruption of the PTS (deletion of ptsG) effectively reduced consumption of PEP, which was beneficial in increasing the formation of precursor for L-homoserine biosynthesis. Moreover, the activation of a non-PTS by glk gene promoter replacement and galR gene deletion increased the consumption of ATP, resulting in the elevation of AMP. Also, 2.3-and 2.2-fold increases of fumarate and malate, as well as a 35.0% decrease in isocitrate, were observed in strain HS33, which indicated the activation of a glyoxylate shunt in the TCA cycle. The concentrations of ␣-ketoglutarate and L-glutamate increased 5.9-and 3.2fold, respectively, in strain HS33. For two metabolites, aspartate and ␣-ketoglutarate, biosynthesis occurs through transamination between oxaloacetate and L-glutamate. Also, L-glutamate, as the donor of an ␣-amino for the biosynthesis of amino acids, can be synthesized from ␣-ketoglutarate by a transamination reaction. The increased L-glutamate and ␣-ketoglutarate pools indicated that the overexpression of the gltBD cluster could effectively provide an amino donor for the L-homoserine biosynthetic pathway and adjust the circulation of their pools within the proper range. With the increase of the L-homoserine pool, a decrease of aspartate was also observed. These results demonstrated that enhancement of the L-aspartate metabolism pathway associated with disruption of competitive pathways could effectively activate the biosynthesis of L-homoserine from L-aspartate. The concentration of the by-product 2,6diaminopimelate (DAP) increased 2.2-fold due to knocking out lysA. Based on metabolome analyses, a better understanding is needed of the new limitations, such as carbon redirection and L-homoserine efflux, which would facilitate flux reinforcement by introducing an anaplerotic pathway or other regulatory mechanisms for moreefficient production of L-homoserine.
Replenishment of oxaloacetate from pyruvate toward further improvement based on metabolomics analysis. While obtaining a global overview of metabolic variation, we found that intracellular levels of acetyl-CoA for strain HS33 were higher (more than 18-fold) than those for the wild-type strain, suggesting the carbon flux was not efficiently diverted to oxaloacetate or citrate for production. To address this new bottleneck, two strategies were investigated: (i) introducing the pyruvate carboxylase gene into the engineered strain to direct pyruvate to oxaloacetate rather than acetyl-CoA and (ii) introducing the citrate synthase gene into the engineered strain to direct acetyl-CoA to the TCA cycle. Since E. coli lacks the anaplerotic enzyme pyruvate carboxylase, pyruvate accumulation in the strain could not be directly altered by directing pyruvate to the aspartate branch (Fig. 9a). Although the thrA gene was previously overexpressed by replacing the trc promoter, the transcription level was increased by about 2-fold (see Fig. S3). To further enhance the flux of the L-aspartate branch, we overexpressed an aspartate-insensitive pyruvate carboxylase (encoded by pyc P458S ) from C. glutamicum and threonine-insensitive bifunctional aspartokinase/ homoserine dehydrogenase (encoded by thrA G433R ) and aspartokinase (encoded by lysC) from E. coli in plasmid pACYC 184. Additionally, E. coli citrate synthase functions as a trimer of dimeric subunits, and its activity is allosterically inhibited by NADH and 2-oxoglutarate (36). In contrast, citrate synthase from Gram-positive bacteria and all eukaryotes is a simple dimer that is not allosterically regulated (37). Therefore, we overexpressed various citrate synthase genes, including citA bs (Bacillus subtilis), gltA cg (C. glutamicum), and gltA ec (E. coli), in plasmid pTrc99A. Four plasmids were transformed into strain HS33, resulting in strain HS33/pACYC-pyc P458S -thrA G433R -lysC, HS33/pTrc-citA bs , HS33/pTrc-gltA cg , and HS33/pTrc-gltA ec (Table 1).
Five strains, including HS33, were cultivated in MS medium for 48 h. Among the tested strains, HS33/pACYC-pyc P458S -thrA G433R -lysC could produce 8.54 g/liter L-homoserine (0.33 g/g glucose), which was increased by 17.8% compared to strain HS33. Pyruvate carboxylase plays an anaplerotic role for carbon redistribution by directing pyruvate to the aspartate branch, which was shown to be efficient in reducing the accumulation of ␣-ketoglutarate (0.26 g/liter) and acetate (1.35 g/liter). However, the other three strains that overexpressed citrate synthase produced smaller amounts of L-homoserine than the corresponding strain, HS33 (Fig. 9b). The overexpression of citrate synthase by B. subtilis resulted in reducing the accumulation of ␣-ketoglutarate, but the L-homoserine yield was higher than that of the control (0.30 versus 0.23 g/g glucose). In contrast, the overexpression of citrate synthase by C. glutamicum and E. coli resulted in improved growth and reduced by-product (acetate) accumulation. However, ␣-ketoglutarate accumulated significantly, which was probably due to the fact that the flux was altered by the enzyme by directing oxaloacetate to the citrate branch. Fed-batch production of L-homoserine. In order to scale up for L-homoserine production by HS33/pACYC-pyc P458S -thrA G433R -lysC from glucose, fed-batch culture was implemented in a 5-liter bioreactor containing 2 liters of medium. Cell growth, L-homoserine production, and residual glucose during the fed-batch culture are shown in Fig. 10. A total of 450 ml feeding medium containing 500 g/liter glucose was added to the bioreactor during the cultivation, and the residual glucose concentration was maintained below 10 g/liter. The cell growth entered stationary phase within 40 h, and the optical density at 600 nm (OD 600 ) fluctuated around 30 till the end of cultivation. Within 108 h of the fermentation, L-homoserine production continuously increased and was independent of cell growth, and the maximum titer achieved was 37.57 g/liter with a yield of 0.31 g/g glucose, representing a 4.4-fold increase compared to the titer achieved in flask shake cultivation. By-products of amino acids were almost undetectable. However, further effort is needed to increase the titer and shorten the fermentation time, such as optimizing the expression of the recognized genes and the culture conditions.

DISCUSSION
This work showed the construction of a basic homoserine producer through a combination of the traditional pull-push-block effects of metabolic engineering and rational genetic modifications. In accordance with the results reported previously, the disruption of competitive and degradative pathways (L-threonine, L-lysine, and L-methionine pathways) contributed to increasing the production of L-homoserine (13). When it comes to the accumulation of target products that are toxic to cells or that inhibit the activity of enzymes in the metabolic pathway, strengthening the exportation of these metabolites is of great importance to relieve the cell burden and enhance production (32,(38)(39)(40). Here, strengthening the transport system and constructing a constitutive expression system not only allowed constitutive expression of modified genes without the addition of inducer, it also avoided suppressing cell growth caused by the inducer and L-homoserine (32). Additionally, observations after deleting the transcriptional regulator IclR indicated that the TCA cycle could operate smoothly when activating the glyoxylate shunt flux. However, the diversion of flux away from forming oxaloacetate from fumaric acid by deletion of fumarase genes or overexpression of aspA dramatically reduced the production of L-homoserine, which may be attributed to the toxicity of fumaric acids (41).
In addition to the above-mentioned rather obvious genetic modifications, we also tested the impacts of another 39 different genes on L-homoserine production by using CRISPRi-based gene repression methodology. The results of gene repression indicated that use of the CRISPRi system was a feasible method to regulate gene expression with less labor and time consumption. Specifically, downregulation of the PTS is widely approved for reinforcing the biosynthesis of PEP-derived products (42,43). The repression of zwf, tktA, rpe, and talB disrupted the pentose phosphate pathway, which can direct more carbon to the synthesis of precursors in the glycolytic pathway. argA-, argG-, proB-, and gadA-repressing regulation may contribute to accumulating L-glutamate and L-aspartate and therefore provide more precursors for L-homoserine biosynthesis (44). Also, the repression of ackA, pta, and poxB contributed to reducing secretion of acetate, which blocked the by-product formation pathway and released acid stress. Moreover, validation of screened candidate genes by individual gene deletion further indicated that sequence-specific control of gene expression on a genome-wide scale is a practical tool for engineering gene-regulatory systems in L-homoserine biosynthesis (45). However, we observed that deletions of some genes led to a decrease in L-homoserine production, which was not consistent with the results showing that these genes were repressed by CRISPR-dCas9. The possibility of completely inhibiting gene expression with CRISPRi is low, and in some cases, even a low expression level of certain genes located in metabolic pathways could potentially contribute enough enzyme activity to maintain its flux and thus cover the effect of targeting such genes (46).
Based on CRISPRi-based systematic analysis, integration of glucose uptake and recovery of L-glutamate were adopted to develop a stable strain with better performance. In the metabolism of the carbon skeletons of L-homoserine, PTS disruption is beneficial to improve the size of the intracellular pool of PEP, which is a direct precursor of L-aspartate family amino acids (42). However, the observed phenotype of PTS disruption mutations other than ptsG mutation was in accord with the reports that low growth rates were observed as the strains would consume ATP for phosphorylation of glucose through a non-PTS system (47). Therefore, restoration of glucose transport capacity could efficiently improve L-homoserine production. In terms of engineering amino acid supply pathways, enhancing circulation of amino donors by converting ␣-ketoglutarate to L-glutamate could also efficiently enhance L-homoserine production. However, increasing the availability of the intracellular amino acids L-aspartate and L-glutamate by blocking the metabolic pathways failed to increase the accumulation of L-homoserine. The results indicated that the regulation of L-aspartate and L-glutamate metabolism is complicated, as they are precursors involved in more than 10 pathways, which made it difficult to identify the major bottleneck for L-homoserine accumulation.
Metabolomics analysis revealed that the multiplex design of the metabolic network caused less flexibility of flux distribution on the metabolome level. The significant variations of strain HS33 directed more carbon toward L-homoserine biosynthesis after the multiple genetic modifications. Additionally, they facilitated understanding of the underlying metabolic effects and provided relevant targets for strain modification to achieve a preferable performance. The results demonstrated that introducing the anaplerotic pathway to direct pyruvate to oxaloacetate was superior to the strategy of directing acetyl-CoA to the TCA cycle. The metabolic bottleneck for the production of L-homoserine could be efficiently relieved by modification of metabolic flux at both the oxaloacetate and pyruvate nodes. Directing pyruvate to the L-aspartate branch could directly realize carbon redistribution, while regulation of the TCA cycle was quite complex, as it is a crucial component of metabolism in organisms (48).
In conclusion, 5 levels of bottlenecks, i.e., the transport system, transcriptional regulation, carbon utilization, amino donor circulation, and the anaplerotic path, that sequentially limited L-homoserine biosynthesis were identified and resolved after systematic searching and multiplex design of the L-homoserine biosynthesis pathway. The L-homoserine production of the optimal strain HS33/pACYC-pyc P458S -thrA G433R -lysC reached 8.54 g/liter (0.33 g/g glucose, 62.4% of the maximum theoretical yield) and 37.57 g/liter (0.31 g/g glucose, 58.6% of the maximum theoretical yield) in shake flask cultivation and fed-batch fermentation without optimizing the expression of the selected genes or the culture conditions (Table 2). Further, it is expected that L-homoserine production by the optimal strain constructed in this study can be further enhanced by optimizing the expression of the positive genes and the fermentation process, which would unlock the potential of the strain for the production of L-homoserine and its derivatives on a large scale.

MATERIALS AND METHODS
Strains, plasmids, and culture conditions. All the strains and plasmids used in this study are listed in Table 1. E. coli BL21 was used for plasmid construction, and the OSH-producing strain (24) was used as the original strain. The low-copy-number plasmid pACYC184 was used as the backbone for overexpressing the mutated thrA gene (G433R) and lysC gene from E. coli W3110 and the mutated pyc gene (P458S) from C. glutamicum ATCC 13032. Plasmid pTrc99A was used for overexpressing citrate synthase genes from E. coli W3110, B. subtilis 168, and C. glutamicum ATCC 13032. Luria-Bertani medium (10 g/liter tryptone, 5 g/liter yeast extract, and 10 g/liter NaCl) was used for cell cultivation. When needed, kanamycin and chloramphenicol were supplemented at final concentrations of 50 g/ml.
For expression of citrate synthase, the corresponding genes, gltA ec (NCBI gene ID 12930947), citA bs (NCBI gene ID 936270), and gltA cg (NCBI gene ID 3344398), were amplified from the genomic DNA of E. coli, B. subtilis, and C. glutamicum using the primers GltA ec _F/GltA ec _R, CitA bs _F/CitA bs _R, and GltA cg _F/ GltA cg _R. Then, the three fragments were assembled with the pTrc99A backbone fragment obtained using the primers pTrc99A-F/pTrc99A-R to yield pTrc-gltA ec , pTrc-citA bs , and pTrc-gltA cg . The primers used for plasmid construction are listed in Table 3.
Construction of an sgRNA-expressing plasmid for the CRISPRi system. The pdCas9 plasmid with a catalytically dead Cas9 mutant was stored in our collection (45). The original plasmid, pCas, and pTargetF, containing the Cas9 gene (51) and sgRNA, were donated by Sheng Yang from the Institute of Plant Physiology and Ecology (Chinese Academy of Sciences, Shanghai, China). The sgRNA-expressing plasmid library targeting selected genes was constructed according to previously reported methods (22). Briefly, sgRNA cassettes were generated through PCR-based site-directed mutagenesis using the original pTarget as a template. Primer sequences used for generation of sgRNA cassettes and the corresponding sgRNA expression vectors are listed in Table 4. Every sgRNA-expressing plasmid for downregulation of target genes was transformed into electrocompetent cells containing plasmid pdCas9 to obtain a series of strains: HS6(pdCas9, pTarget-X), where X refers to the genes in the central metabolism and selected amino acid biosynthetic pathways.
Genetic manipulation of E. coli derivatives using CRISPR-Cas9. Gene insertion and deletion and chromosomal promoter replacement were achieved by using the CRISPR-Cas9 system, as previously described (24). The detailed protocol of gene editing is described in the supplemental material. The primers used in this study are listed in Table 5. Electrocompetent cells were prepared as previously described (51).  Culture conditions. MS medium was used for L-homoserine production by E. coli W3110-derived strains, consisting of (per liter) 40 g glucose, 17 g (NH 4 ) 2 SO 4 , 4 g yeast extract, 1 g KH 2 PO 4 , 1 g MgSO 4 , 25 g CaCO 3 , 5 mg FeSO 4 ·7H 2 O, 5 mg MnSO 4 ·7H 2 O, and 5 mg ZnSO 4 . Shake flask fermentation was performed in 500-ml flasks containing 20 ml MS medium at 28°C for 48 h at 150 rpm. When necessary, the medium was supplemented with 50 g/ml chloramphenicol or kanamycin. L-Methionine and L-threonine were supplemented to final concentrations of 0.05 and 0.2 g/liter. Additionally, L-lysine was added to MS medium at final concentrations ranging from 0 to 0.5 g/liter (0, 0.0125, 0.025, 0.05, 0.125, 0.25, and 0.5 g/liter), depending on the experiment, as described in the text. For expression of the CRISPRi system, strains were induced with 0.1 mM IPTG (isopropyl-␤-D-thiogalactopyranoside) when the OD 600 reached 0.4 to 0.6. The cells were cultivated aerobically at 28°C in a shaking incubator at 150 rpm and supplemented with 50 g/ml antibiotics when necessary.
In fed-batch fermentation, a 5-liter bioreactor (Baoxing, Shanghai, China) was used for L-homoserine, with an effective working volume of 2.0 liters. The medium for fed-batch fermentations consisted of (per liter) 15 g glucose, 14 g (NH 4 ) 2 SO 4 , 2 g yeast extract, 2 g KH 2 PO 4 , 0.5 g MgSO 4 , 0.5 g L-threonine, 0.2 g L-methionine, 0.1 g L-lysine, 5 mg FeSO 4 ·7H 2 O, 5 mg MnSO 4 ·7H 2 O, and 5 mg ZnSO 4 . The feeding medium contained 500 g/liter glucose, 12.5 g/liter KH 2 PO 4 , 10.0 g/liter NaHCO 3 , 4 g/liter L-threonine, 1 g/liter L-methionine, and 0.5 g/liter L-lysine, and the feed was started when the residual glucose concentration was below 10 g/liter. The fermentation temperature was maintained at 28°C, and the pH was adjusted to 6.8 with 40% ammonia water throughout the process. The dissolved oxygen level was maintained above 20% by adjusting aeration and agitation rates. aspA ATACTAGTACACCATAGTAGGCATCAGCGTTTTAGAGCTAGAAATAGC pTarget-gadA gadA ATACTAGTGTGAATCGAGTAGTTCTGAGGTTTTAGAGCTAGAAATAGC pTarget-null b ATACTAGTGTTTTAGAGCTAGAAATAGC pTarget R-common ACTAGTATTATACCTAGGACTGAGC a Each pTarget primer has the same reverse primer, pTarget R-common (listed last). b pTarget-null represents the primer used to construct pTarget-null without the N20 sequence, used as a negative control. c The italic letters represent the restriction enzyme cutting site (SpeI), and the boldface letters represent the 20-bp region complementary to the targeting region of the selected gene.
concentrations of amino acids. By-product measurement was performed with an Aminex HPX-87H column (300 by 7.8 mm), using 5 mM H 2 SO 4 as the mobile phase, with a flow rate of 0.6 ml/min and detection via refractive index or UV absorption at 210 nm (6).
RT-qPCR for mRNA quantification. E. coli strains W3110 and HS33 were cultured for 18 h in 20 ml MS medium at 150 rpm and 28°C. Samples were collected during the exponential growth phase, prepared by centrifugation, and flash frozen in liquid nitrogen and were sent to Tsingke Biological Technology (Wuhan, China) for RT-qPCR analysis with 16S rRNA as a control for normalization between samples. The fold changes of target genes were calculated as 2 ϪΔΔCT according to the method of Schmittgen and Livak (52).
Metabolome analysis. Metabolomics was applied to characterize the biological variation of the intercellular metabolites between E. coli W3110 and HS33. The samples were collected from cultures grown in 6 technical replicates per strain during the exponential growth phase. The cells were washed twice with ice-cold phosphate-buffered saline (PBS) and flash frozen in liquid nitrogen for further metabolite extraction. Statistical analysis of metabolite profiles was performed by BGI Tech Solutions Co., Ltd. (Shenzhen, China). The intracellular metabolite concentrations of E. coli W3110 were used as a control for normalization between samples. The detailed protocol of the liquid chromatography-mass spectrometry (LC-MS) system is described in the supplemental material. The screened metabolomics data were subjected to partial least squares discrimination analysis (PLS-DA). Significantly changed metabolites (SCM) (defined as follows: base VIP [variable important for the projection], Ͼ1.0; fold change, Ͼ1.20 or Ͻ0.83; q-value, Ͻ0.05) were selected for subsequent chemical structure identification.
Statistical analysis. All experiments were conducted in triplicate. The data were averaged and are presented as means Ϯ standard deviations (SD). P values of Ͻ0.01 were considered statistically significant. All the figures were prepared using Origin Software version 8.0 (OriginLab Corp., Northampton, MA, USA).
Data availability. The raw data supporting the conclusions of this article will be made available, without undue reservation, to any qualified researcher.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.2 MB.