A Self-Regulated Network for Dynamically Balancing Multiple Precursors in Complex Biosynthetic Pathways

Abstract

Microbial synthesis has emerged as a promising and sustainable alternative to traditional chemical synthesis and plant extraction. However, the competition between synthetic pathways and central metabolic pathways for cellular resources may impair final production efficiency. Moreover, when the synthesis of target product requires multiple precursors from the same node, the conflicts of carbon flux have further negative impacts on yields. In this study, a self-regulated network was developed to relieve the competition of precursors in complex synthetic pathways. Using 4-hydroxycoumarin (4-HC) synthetic pathway as a proof of concept, we employed an intermediate as a trigger to dynamically rewire the metabolic flux of pyruvate and control the expression levels of genes in 4-HC synthetic pathway, achieving self-regulation of multiple precursors and enhanced titer. Transcriptomic analysis results additionally demonstrated that the gene transcriptional levels of both pyruvate kinase PykF and synthetic pathway enzyme SdgA dynamically changed according to the intermediate concentrations. Overall, our work established a self-regulated network to dynamically balance the metabolic flux of two precursors in 4-HC biosynthesis, providing insight into balancing biosynthetic pathways where multiple precursors compete and interfere with each other.

1. Introduction

Microorganisms take in carbon sources from the environment, breaking them down into smaller molecules and generating energy through catabolism to sustain physiological activities. Some of these small molecules, such as phosphonyl pyruvate (PEP), pyruvate, and acetyl-CoA, serve as central metabolites and are subsequently used by the cells for anabolism to synthesize various functional metabolites (Calabrese et al., 2021). The introduction of specific exogenous enzymes allows engineered microbes to convert these central metabolites into target products. Because of the rapid growth of microbes and their high compatibility of complex pathways, microbial synthesis approaches have been widely employed in the production of high-value compounds and offered promising alternatives to traditional chemical synthesis (Cravens et al., 2019; Shah et al., 2019; Peralta-Yahya et al., 2012). However, exogenous enzymes often exhibit sub-optimal performance, especially in complicated synthetic pathways, resulting in rate-limiting steps. In addition, the intrinsic regulatory network of the microbes may not be able to properly control the heterologous synthetic pathways, posing challenges for achieving optimal production (Teng et al., 2022). Furthermore, heterologous pathways bring additional metabolic stress on the cells, creating a competition between the central metabolism and synthetic pathways, leading to an imbalanced carbon distribution, which can negatively impact on both cell growth and productive performance (Li et al., 2022).

To address the above problems, multiple engineering strategies have been applied to enhance the production of microbial-based compounds. For example, the protein expression level can be adjusted at the transcriptional or translational level by adopting engineered promoters and ribosomal binding sites (RBS) with varying strength (Jones et al., 2015). The competing pathways and degradation pathways of end-product can also be eliminated through gene knockouts (Zhang et al., 2021; Bang et al., 2023). Moreover, directed evolution (Scheiblbrandner et al., 2017), enzyme compartmentalization (Wei et al., 2020), and protein scaffold (Zhang et al., 2017) are designed to increase the efficiency of synthetic pathways. However, the changing metabolic status and the living environment of microbes make it difficult for static regulation to accommodate the varying conditions. To achieve a more precise control, dynamic regulation was proposed and demonstrated capable of responding to the changing factors in real-time, such as the concentration of certain metabolites, temperature, pH, and cell density (Teng et al., 2022). Therefore, this strategy can be efficiently applied to dynamically optimizing enzyme expression, relieving the accumulation of toxic intermediates, and balancing the cell growth with enhanced production.

Dynamic regulation has been developed to address many problems in synthetic pathways due to its high sensitivity to metabolic state, especially the imbalance problem of metabolic flow (Brockman et al., 2015). Most synthetic pathways utilize central metabolites as precursors and divert carbon flux to the target products. When excessive carbon flux is diverted from a node to support additional compounds synthesis, the carbon flux at that node may not be sufficient to sustain downstream central metabolic pathways or other branches, which disturbs normal processes (Wu et al., 2016). If the intermediates or final products in a synthetic pathway have cellular toxicity, this stress will be even more severe. To mitigate the conflict of carbon flux distribution, Farmer et al. developed an acetyl phosphate biosensor to dynamically activate the limiting steps in lycopene pathway. Thus, the lycopene would not be synthesized until extra carbon flux enters waste metabolites pathway (Farmer et al., 2000). To minimize the carbon consumption on cell growth, Yang et al. developed a bifunctional dynamic control network basedon an end product-responsive biosensor. By dynamically activating synthetic pathways and reducing the unnecessary consumption of the central metabolite PEP, the initial growth of cells is ensured, and the yield of muconic acid is increased (Yang et al., 2018). In another study, Xu et al. engineered a pyruvate-responsive biosensor to monitor the carbon flux amount in glycolysis, with a bifunctional regulation circuit, the production of glucaric acid was improved (Xu et al., 2011). However, comparing with these cases, many natural pathways use multiple central metabolites as precursors simultaneously, and the carbon flow pass through different branches before converging on the final product, making the situation even more complicated (Xu et al., 2011; Zhang et al., 2011). In this situation, not only will the synthetic pathway compete with the central pathway for carbon flow, but different synthetic branches will compete as well, resulting in unbalanced precursors and blocked synthesis. To date, using dynamic regulation strategies to balance the accumulation of multiple precursors in one pathway is still under explored.

In this study, we aimed to construct a self-regulated network to dynamically balance multiple precursors in a single synthetic pathway. The 4-hydroxycoumarin (4-HC) pathway was utilized as proof of concept due to its requirement for two precursors, salicylate, and malonyl-CoA, both derived carbon flux from glycolysis. More importantly, 4-HC type anticoagulants play an important role in treating thromboembolic diseases clinically, and 4-HC serves as an immediate precursor of the anticoagulants. In glycolysis, PEP is a crucial intermediate that can be converted to pyruvate and enter the tricarboxylic acid (TCA) cycle to support cell growth. Meanwhile, PEP is a key point from which various natural or synthetic pathways can be derived, such as the shikimate pathway (Xiberras et al., 2019). The intermediate molecule salicylate in the shikimate pathway is derived from PEP, while malonyl-CoA originates from acetyl-CoA, which is a downstream product of pyruvate. The two precursors compete for carbon flow from glycolysis, and an inefficient allocation of this flow can lead to decreased production. To optimize the carbon utilization, we rewired the metabolic background to obtain a salicylate high producer. Then, we employed a salicylate-responsive biosensor system to dynamically regulate the supplies of two precursors malonyl-CoA and salicylate, In our design, the carbon flux was distributed depending on the concentration of salicylate. This ensured the accumulation of salicylate and the release of enough malonyl-CoA for the complete conversion of salicylate. At the same time, crucial genes involved in 4-HC biosynthetic pathway were dynamically controlled by the CRISPRi coupled biosensor system. As a result, we achieved an improved 4-HC production, and transcriptomic analysis further confirmed the function of this regulation at the transcriptional level. Most importantly, this study provides novel insights for the regulation of complicated pathways with multiple precursors derived from central pathways.

2. Results

2.1. Metabolic design for multiple precursors balanced bioproduction.

To build the bioproduction platform for balancing multiple precursors, the key point is to rewire carbon metabolism and regulate the release of each precursor with proper carbon flux level separately. The carbon flow derived from PEP supports various central metabolic pathways, including the TCA cycle and fatty acid synthesis. PEP also drives carbon flux to many metabolic branches such as the shikimate pathway, glycerol fermentative pathway, and oxaloacetate anaplerotic reaction catalyzed by PPC. We proposed that it is possible to balance multiple downstream pathways simultaneously by rewiring the PEP node. As a shikimate pathway downstream product, 4-HC can be produced by expressing a set of heterogenous enzymes in E. coli (Fig. 1). In the last two steps of 4-HC synthesis, as one precursor, salicylate is converted into 4-HC catalyzed by a β-ketoacyl-ACP synthase III (FabH)-type quinolone synthase (PqsD) and salicyl-CoA synthase (SdgA), one molecule of malonyl-CoA is condensed as another precursor (Lin et al., 2013). Basically, both precursors come from the PEP node, but through different, competing branches. It is noteworthy that the releasing of salicylate is accompanied by the formation of one molecule of pyruvate, which is in a position between PEP and malonyl-CoA on central pathway. Exactly, the carbon flux intersection in pyruvate gives us room for balancing the generation of two precursors. To provide adequate carbon flow and reducing force for the synthesis of aromatic compounds, we chose glycerol as the carbon source (Xiberras et al., 2019).

Fig. 1.

Schematic representation of the established 4-HC biosynthetic pathway. DHA, Docosahexaenoic acid. G3P, Glycerol-3-phosphate. DHAP, Dihydroxyacetone phosphate. GA3P, Glyceraldehyde 3-phosphate. E4P, Erythrose 4-phosphate. PEP, Phosphoenolpyruvic acid. DAHP, 3-Deoxy-D-arabino-heptulosonate 7-phosphate. OAA, Oxaloacetic acid. PYR, Pyruvate. glpK, encodes glycerol kinase. gldA, encodes glycerol dehydrogenase. glpD, encodes aerobic glycerol 3-phosphate dehydrogenase. dhaKLM, encodes dihydroxyacetone kinase subunit. tpiA, encodes triose-phosphate isomerase. tktAB, encodes transketolase. talab, encodes transaldolase. gapA, encodes glyceraldehyde-3-phosphate dehydrogenase. pgk, encodes phosphoglycerate kinase. gpm, encodes phosphoglycerate mutase. eno, encodes enolase. ppc, encodes phosphoenolpyruvate carboxylase. pck, encodes phosphoenolpyruvate carboxykinase. pykA and pykF, encode pyruvate kinase. ppsA, encodes phosphoenolpyruvate synthetase. aroG, encodes 3-deoxy-7-phosphoheptulonate synthase. aroA, encodes 3-phosphoshikimate 1-carboxyvinyltransferase. aroC, encodes chorismate synthase. aroL, encodes shikimate kinase 2. entC, encodes isochorismate synthase. pchB, encodes isochorismate pyruvate lyase. aceEF, encodes pyruvate dehydrogenase. lpd, encodes lipoamide dehydrogenase. accAD, encodes acetyl-CoA carboxyltransferase subunit. sdgA, encodes salicyl-CoA synthase. pqsD, encodes beta-ketoacyl-ACP synthase III (FabH)-type quinolone synthase. Genes colored red represent deletion or repression. Genes colored blue represent overexpression. Genes colored green represent dynamical regulation.

We started our design by focusing on shikimate branch for establishing a salicylate high producer. The isochorismate pyruvate lyase (PchB) mediates the conversion of isochorismate to salicylate, also generates one molecule of pyruvate. According to previous studies, it is able to save PEP flow and recycle pyruvate from aromatic synthetic pathway by disrupting the pyruvate generation genes on glycolysis, concluding that cutting off most pyruvate native sources would make salicylate pathway obligatory to pyruvate generation and cell growth (Wang et al., 2019; Noda et al., 2016; Noda et al., 2017; Ponce et al., 1998). By coupling bioproduction with cell growth, salicylate synthesis can be enhanced as a salvage pathway while saving carbon flux. In order to block most pyruvate native sources in E. coli, we cut off two main pyruvate generation pathways: The glycerol fermentative pathway was shut down by deleting glycerol dehydrogenase encoded gene, gldA. And the pyruvate generation reaction on glycolysis was shut down by deleting pyruvate kinases encoded genes, pykA and pykF. The disruption of pykA and pykF will also save PEP for salicylate releasing (Lin et al., 2013). With this engineered circuit, carbon source will be utilized efficiently to produce the 4-HC precursor, salicylate. Meanwhile, pyruvate would be generated within salicylate pathway as a byproduct to support cell growth and another 4-HC precursor, malonyl-CoA.

2.2. Screening of 4-HC producers with enhanced salicylate accumulation.

We started to establish the 4-HC high producer from strain BW25113 F’, a derivative of E. coli K-12 strains. To create the shikimate branch enhanced strain, the glycerol dehydrogenase encoded gene gldA was deleted in BW25113 F’ to produce strain ZZ1 Subsequently, the pyruvate kinase gene pykA and pykF were knocked out separately in ZZ1 to acquire ZZ2 and ZZ3. Both isozymes PykA and PykF have pyruvate kinase activity, knocking out one of them did not have severe impact on cell growth. According to previous studies, PykF contributes more than PykA to cell growth (Shen et al., 2017), which is consistent with our results that ZZ2 grew better than ZZ3 (Fig. 2a). To fully block this reaction, we further knocked out pykF in ZZ2, yielding strain ZZ4. By disrupting the main pyruvate generation genes including gldA, pykA, and pykF, cellular pyruvate supply was thought to be highly hindered in strain ZZ4. When culturing the cells in minimal M9 medium with glycerol (20 g/L) as the carbon source, ZZ4 showed serious growth defect (Fig. 2a). In order to further confirm that the growth defect of ZZ4 was caused by pyruvate deficiency, the medium was supplemented with 4 g/L pyruvate at the beginning of inoculation. As we expected, the cell growth of ZZ4 was almost recovered to normal level with the supplement of exogeneous pyruvate (Fig. 2b).

Fig. 2.

Growth performance and Production performance of different strains. a. Growth curve of different strains. ZZ1 (BW25113 F’△gldA), ZZ2 (BW25113 F’△gldA△pykA), ZZ3 (BW25113 F’△gldA△pykF), ZZ4 (BW25113 F’△gldA △pykA△pykF), ZZ5 (BW25113 F’△ppc) b. Growth curve of different strains after feeding 4 g/L pyruvate. c. Production performance of SA in different strains at 48-hour. d. Production performance of 4-HC in different strains at 48-hour. All error bars represent standard deviation. (n = 3). The experiments are biological replicates.

Given the limited cell growth attributed to pyruvate deficiency, we reasoned that plasmid-based overexpression of salicylate pathway in ZZ4 strain could alleviate pyruvate deficient to enhance the accumulation of salicylate as the pathway is obligated to pyruvate synthesis. At the same time, knocking out pykA and pykF decreased the consumption of PEP on biomass and fatty acid accumulation, guiding more PEP into salicylate synthesis pathway. With the optimized metabolic background, salicylate production will be further promoted. To verify our hypothesis, we adopted a previously constructed salicylate synthesis plasmid in our lab (Lin et al., 2013). The high copy number plasmid pZE12-luc contains two operons both controlled by pLlacO1 promoter (Lutz and Hermann., 1997), the first operon includes entC and pchB, encoding two enzymes responsible for converting chorismate into salicylate. Another operon contains aroL, ppsA, tktA, and aroG^fbr, which are for chorismate boosting. The BW25113 F’, ZZ2, ZZ3, and ZZ4 were cultivated in M9 medium with 5 g/L yeast extract for optimal production. As expected, the triple knockout strain ZZ4 produced the highest titer of 1.22 g/L salicylate in 48h fermentation, which is approximately two-fold improvement compared with BW25113 F’ strain (Fig. 2c).

Additionally, ZZ2 and ZZ3 achieved moderately increased salicylate yield of 0.78 g/L and 0.70 g/L, separately (Fig. 2c). This proved that, with the strategy of supplementing pyruvate via producing target products, the engineered strains were able to achieve improved salicylate production significantly, even without directly enhancing any synthetic pathways.

As the salicylate high accumulation strain ZZ4 was established, we subsequently investigated whether it has the potential to be an efficient 4-HC producer. We adopted the 4-HC synthesis pathway developed by Lin et al. In the first step, salicylate is catalyzed by SdgA to form salicoyl-CoA. Then, PqsD converses salicoyl-CoA into 4-HC, also requiring a malonyl-CoA in this step. Based on the pathway design, salicylate and malonyl-CoA were thought to be two precursors for 4-HC synthesis. As malonyl-CoA is derived from glycolysis products, inevitably, the generation of malonyl-CoA will compete the central carbon flux with shikimate pathway that contributed to salicylate accumulation. Meanwhile, malonyl-CoA also benefits from salicylate generation reaction catalyzed by PchB since that the extra released pyruvate in this step also compensated downstream metabolites including acetyl-CoA and malonyl-CoA. To preliminary assess the 4-HC conversion capacity of the candidate strains, we employed the previously constructed plasmid pCS-PS, which containing PqsD and SdgA under the control of pLlacO1 promoter (Lin et al., 2013). We transferred pZE-EP-APTA, responsible for synthesizing SA, and pCS-PS, responsible for downstream synthesis, into the platform strains. As the native control group, the BW25113 F’ strain could produce 0.40 g/L 4-HC in a 48-hour fermentation (Fig. 2d). To our surprise, although ZZ4 had advantage of high intermediate accumulation, it only produced 0.26 g/L 4-HC and remained 0.33 g/L salicylate that was not fully utilized (Fig. 2d). We assumed that the low production of 4-HC and insufficient conversion of salicylate was caused by the inadequate malonyl-CoA. Albeit the pathway we introduced was able to replenish some pyruvate, this may not be sufficient to compensate for the pyruvate loss caused by the gene knockout, and this may be more pronounced when carbon flow enters malonyl-CoA. In other words, the accumulation of salicylate and malonyl-CoA, as two direct precursors, was not well balanced. We noticed that ZZ2 strain has a good basal production of 0.52 g/L 4-HC and a certain amount of salicylic acid accumulation, which made this strain a 4-HC producing candidate (Fig. 2d). Therefore, ZZ2 possesses the potential to be further modified to increase the supply of malonyl-CoA on the basis of maintaining the level of salicylate.

2.3. Design and characterization of repression regulatory network

Since simultaneously blocking pykA and pykF highly restricted the pyruvate generation and the malonyl-CoA supply, we planned to implement the dynamic regulation strategy to compromisingly replenish the loss of malonyl-CoA, meanwhile, ensuring the high accumulation of salicylate. Based on the results of 4-HC producer screening, ZZ2 was selected as the starting strain, and the gene pykF was selected as the first regulation target. According to our design, the intermediate SA serves as the inducing molecule. During the early growth state, no regulation will be applied, and cells could consume substrate to support the cell growth. Meanwhile, the carbon flux was able to serve for the accumulation of salicylate and ensure the generation of malonyl-CoA. When SA accumulates to a threshold, the pyruvate flux would be dynamically turned off by inhibiting pykF, and the cellular metabolism would move to the production state. During the production state, as all the native pyruvate releasing reactions are blocked, carbon flux would be derived to 4-HC synthetic pathway, and the pyruvate generated on SA pathway would support the cell growth. (Fig. 3b). We assume that the downstream pathway from SA to 4-HC could be further improved because of the enhanced accumulation of malonyl-CoA and SA. Basing on the first strategy, we hoped to figure out another target to further mitigate the production burden of malonyl-CoA even during the production state. Since TCA cycle could be regarded as a competing route to the synthesis of 4-HC, we still wanted to limit the carbon consumption on TCA cycle. By considering the aforementioned preconditions, we tried to block the phosphoenolpyruvate carboxylase encoded gene, ppc. However, deletion of gene ppc from wild-type BW25113 F’ caused serious growth retardation and non-detected production of 4-HC (Fig. 2a). Thus, the ppc gene was selected as another dynamic regulation target to maintain the supply of malonyl-CoA and SA. Similarly, the cell would consume carbon flux to support the cell growth and other metabolisms without disturbance during the early growth state. When the concentration of SA reach to the threshold, the cellular metabolism would move to the production state due to the dynamical inhibition on ppc, resulting in the deficiency of TCA cycle. Meanwhile, the carbon flux would be directed into the synthetic pathway of malonyl-CoA and SA, and the regenerated pyruvate from SA biosynthesis would be also driven to the generation of malonyl-CoA (Fig. 3b). Consequently, such improved synthetic pathway of malonyl-CoA and SA could further contribute to the downstream synthesis of 4-HC. At this point, we are looking forward to seeking out a proper metabolic engineering tool to realize the above designs.

Fig. 3.

Design and characterization of dynamic repression networks. a. Dynamic performance of dynamic repression circuits. PC, positive control, the plasmid pHA-MCS-eGFP was co-transferred with the plasmid pMK-MCS. SA, salicylate. The red dot represents terminator. 0.5mM IPTG was added to activate the expression strength of pLlacO1. b. Schematic representation of dynamic regulation on 4-HC biosynthetic pathway. PEP, Phosphoenolpyruvic acid. OAA, Oxaloacetic acid. PYR, Pyruvate. c. Production performance of dynamic repression networks at 48-hour. NC-ppc, sgppc controlled by pLlacO1. NC-pykF, sgpykF controlled by pLlacO1. PC, positive control, the plasmid pHA-EP-APTA, pCS-PS-lpp0.5-marR, and pLC-pLlacO1-sgpykA were co-transferred into ZZ7. 0.5mM IPTG was added to activate the expression strength of pLlacO1. All error bars represent standard deviation (n = 3). The experiments are biological replicates.

Transcriptional factor-based biosensors (TFBs) have recently been demonstrated and widely used in dynamic regulation for enhanced compound productions (Zhang et al., 2011; Jiang et al., 2022). Due to the high accumulation of salicylate in ZZ2, we tried to adopt a salicylate-responsive biosensor system to regulate the synthetic pathway of 4-HC. According to our previous results, the multiple antibiotic resistance regulator (MarR) from E. coli can recognize the binding sequence located on the PmarO promoter and inhibit its expression (Zou et al., 2021). Such inhibition can be released after adding a certain amount of salicylate. After the characterization and exploration of this MarR-PmarO biosensor system, we obtained a hybrid promoter I12AII14T with improved dynamic performance (Zou et al., 2021). However, the characterization was previously conducted under a double-plasmids system, which might cause the production burden during the biosynthesis process. At this point, we re-constructed the hybrid promoter I12AII14T with a fluorescent reporter gene eGFP in a medium copy number plasmid pMK-MCS, resulting in pMK-I12AII14T-eGFP. Subsequently, the corresponding regulator MarR controlled by a constitute promoter lpp0.5 was inserted into pMK-I12AII14T-eGFP, engendering pMK-I12AII14T-eGFP-lpp0.5-marR. As the control group, the hybrid promoter I12AII14T was substituted by the wild-type promoter PmarO, yielding pMK-Pmaro-eGFP-lpp0.5-marR. In order to investigate the re-constructed MarR-PmarO biosensor system, the plasmid pMK-Pmaro-eGFP-lpp0.5-marR and the plasmid pMK-I12AII14T-eGFP-lpp0.5-marR were transferred into BW25113 F’, respectively. Gradient concentrations of salicylate were added to test their dynamic performances. In the absence of salicylate, the PmarO biosensor system and I12AII14T biosensor system exhibited the expression strength of 179.02 a.u. and 1204.19 a.u., respectively (Supplementary Fig. 1). By adding 2 g/L salicylate, the responsive strength of PmarO biosensor system and I12AII14T biosensor system reached to 805.56 a.u. and 3181.65 a.u., respectively, demonstrating that the re-constructed MarR-PmarO biosensor systems still possess reliable dynamic performance (Supplementary Fig. 1). Consequently, the re-constructed biosensor system with wild-type promoter PmarO has limited leakage and I12AII14T biosensor system possesses higher responsive strength.

To dynamically inhibit the gene pykF or ppc, a MarR-PmarO biosensor-based dynamic repression regulatory network was designed and characterized by cooperating with CRISPRi system. In our previous study, different recognizing positions may result in different dCas9 repression efficacies on target genes (Wang et al., 2021). Especially, it could achieve nearly 100% repression efficacy when the recognizing sequence contains NGG at the start codon area, which could be regarded as the deletion of endogenous genes (Doench et al., 2014). To figure out the best repression efficacy on the targeted gene, we employed Streptococcus pyogenes dCas9 (SpdCas9) with different sgRNA variants to investigate their repression performance.

Specifically, we chose the spacer targeted on the start codon area, and two more sgRNA spacers targeted on the intragenic area of a green fluorescent protein encoded gene (eGFP), resulting in sgeGFP1, sgeGFP2, and sgeGFP3 (sgRNA sequences shown in Supplementary Table. 2). Because of the high efficacy of CRISPRi system, the wild-type MarR-PmarO biosensor system possessing limited leakage was used to regulate the expression of sgRNA variants, obtaining plasmids pMK-PmarO-sgRNAs-lpp0.5-marR. Then, the PmarO-sgRNAs biosensor systems were co-transferred with pHA-MCS-eGFP into ZZ6 strain (BW25113 F’ integrated with dCas9) to detect the repression efficacy. And the plasmid pHA-MCS-eGFP was co-transferred with Pmk-MCS as the positive control. As shown in Fig 3.a, the expression level of eGFP was repressed when adding SA. Specifically, the variant sgeGFP2 (targeted on the middle of the egfp sequence) showed the highest repression efficacy at 86.64% after adding 1 g/L SA compared with the repression efficacy of sgeGFP1 at 83.03% and sgeGFP3 at 81.27%. These results demonstrated that expression level of targeted genes can be dynamically repressed by SA when controlled by the engineered MarR-PmarO biosensor system.

Based on the same design principles of egfp dynamic repression circuits, sgRNA variants targeted on gene pykF or gene ppc were obtained (sgRNA sequences shown in Supplementary Table. 2). To apply the dynamic CRISPRi system, we integrated the SpdCas9 into 4-HC producers ZZ2 to obtain ZZ8. As shown in Supplementary Fig. 2., integration of SpdCas9 did not influence the cell growth and production performance in wild-type BW25113 F’ strain. However, when harboring the integration of SpdCas9, ZZ8 showed remarkably decreased production of 4-HC and inhibited cell growth performance. We speculated that the integration of Spdcas9 would have an unknown negative impact on the growth of pykA knocking out strain. To address this problem, we designed a sgRNA variant to inhibit the expression of pykA as a substitution to pykA knocking out. The sgRNA variant sgpykA was controlled by pLlacO1 and expressed in a low copy number plasmid pLC-MCS, resulting in pLC-pLlacO1-sgpykA. The effect of the sgpykA-Spdcas9 system on 4-HC production performance was detected in ZZ6 and ZZ7 by introducing 4-HC biosynthetic plasmids pZE-EP-APTA and pCS-PS. As shown in Supplementary Fig. 3., there were no obvious differences of 4-HC production and cell growth performance between pykA knockout or inhibiting strains. Consequently, an optimized 4-HC biosynthesis producer ZZ8 with the introduction of pLC-pLlacO1-sgpykA was obtained and applied in the following research.

To dynamically regulate the expression of pykF or ppc, we designed a series of sgRNA variants (sgpykF1–3 or sgppc1-3) and integrated them with the biosensor system, respectively, leading to pCS-PS-Pmaro-sgRNAs-lpp0.5-marR. With the introduction of plasmid pZE-EP-APTA, production of 4-HC and remaining SA were detected after incubation for 48 hrs. The sgRNA spacers with nearly 100% repression efficacy on pykF or ppc controlled by pLlacO1 were chosen as the negative control (NC) group. As the results shown in Fig. 3c, both NC-ppc and NC-pykF produced less 4-HC comparing with positive control (PC) group, suggesting that the constant repression effect on pykF or ppc limits the pathway performance as well as the cell growth. Although sgppc1, sgppc2, and sgpykF1 did not show higher 4-HC production comparing with PC group, the remaining SA in each experimental group decreased, demonstrating the influence of the dynamic regulatory network. Similarly, due to the high repression efficacy on pykF or ppc, sgppc1, sgppc2, and sgpykF1 showed limited cell growth and did not possess superiority in 4-HC synthesis. Surprisingly, sgppc3, sgpykF2, and sgpykF3 can produce 0.67 g/L, 0.52 g/L, and 0.69 g/L, respectively, lower concentrations of remaining SA comparing with PC group were detected at the same time (Fig. 3c). These three superior 4-HC producers possessed improved 4-HC synthesis performance with robust cell growth. Accordingly, 4-HC production was enhanced by 1.5-fold with the dynamic repression networks harboring sgpykF3. The results proved that malonyl-CoA was enough to convert most remaining SA into 4-HC in these dynamic regulation groups, and two precursors were well balanced to enhance the final production. Nevertheless, small amounts of SA still remained in the 4-HC biosynthesis process, which indicated that the final production of 4-HC can be further increased.

2.4. Bifunctional network for improved 4-HC biosynthesis performance

The remaining SA observed in those superior 4-HC producers hinted the opportunity of further improvement in 4-HC production. Since the conversion of SA to 4-HC is catalyzed by SdgA and PqsD, we speculated that the premature expression of the two enzymes when SA does not accumulate will bring extra metabolic burden to the cells, leading to suboptimal 4-HC production. Basing on this speculation, we hypothesized that dynamically turning on the expression of SdgA and PqsD when SA accumulate to certain level would fine-tune the cellular metabolism to relieve the metabolic burden. Thus, we employed the SA-responsive promoters PmarO and I12AII14T respectively to regulate sdgA and pqsD, resulting in dynamic activation plasmids pCS-PmarO-PS-lpp0.5-marR and pCS-I12AII14T-PS-lpp0.5-marR. The dynamic activation plasmids were co-transferred with pZE-EP-APTA to test the production performance. As the results shown in Supplementary Fig. 4, the 4-HC production could achieve 0.60 g/L with the downstream genes dynamically activated by hybrid promoter I12AII14T, which is a 1.31-fold improvement by static regulation pathway. But similar improvement did not appear in the PmarO controlled system, which only shows 0.08 g/L of 4-HC production. Limited expression strength of native promoter PmarO might lead to this poor production performance. Above results demonstrated that dynamically activation of the pathway downstream genes sdgA and pqsD by hybrid promoter I12AII14T could effectively improve the production.

To take both the advantages of dynamic repression circuits and dynamic activation circuits, we planned to construct a self-regulated bifunctional network. Firstly, we tried to verify the feasibility of the bifunctional network by introducing both the PmarO controlled CRISPRi system targeted on egfp and I12AII14T controlled red fluorescent protein (RFP). As shown in Fig. 4a, the expression of eGFP could be inhibited and the expression of RFP could simultaneously be activated after adding elevated concentrations of SA. Additionally, different inhibition efficacies and activation rates were observed in different configurations of bifunctional circuits. Similar to the results of dynamic repression circuits, bifunctional circuit involving sgeGFP2 still showed the highest repression efficacy 80.4% comparing with 60.6% repression efficacy of sgeGFP1 and 72.2% repression efficacy of sgeGFP3 (Fig. 4a). Activation rate of various bifunctional circuits did not show obvious difference, which was owing to the stable dynamic performance of I12AII14T-MarR sensor system. Due to the leakage of hybrid promoter I12AII14T, RFP could still be expressed without inducing SA. Especially, the highest expression level of RFP was able to be detected in sgeGFP3-rfp, which showed 9172.14 ± 430. 75 a.u. by adding 1 g/L SA (Fig. 4a). To summary, the constructed bifunctional network could be dynamically regulated by SA to simultaneously achieve gene repression and gene activation.

Fig. 4.

Characterization of bifunctional networks. a. Dynamic performance of bifunctional networks. SA, salicylate. Red dot represents terminator. b. Production performance of bifunctional network at 48-hour. SR, static regulation circuit. DR, dynamic repression circuit. DB, dynamic bifunctional circuit. 4-HC, 4-Hydroxycoumarin. c. Dynamic fluctuations of SA concentrations upon time in 4-HC synthesis. NC, negative control, 4-HC synthesis regulated by static circuit. d. Time course of 4-HC production performance of bifunctional networks. NC, negative control, 4-HC synthesis regulated by static circuit. All error bars represent standard deviation (n = 3). The experiments are biological replicates.

Next, we applied this self-regulated bifunctional network in ZZ7 to test its production performance on 4-HC biosynthesis. Correspondingly, the I12AII14T controlled rfp was substituted with sdgA and pqsD as the dynamic activation module. Three screened dynamic repression circuit (sgppc3, sgpykF2, and sgpykF3) with better 4-HC production performance were selected to be integrated with dynamic activation module respectively, resulting in pCS-I12AII14T-PS-PmarO-sgRNAs-lpp0.5-marR. Three plasmids containing bifunctional circuits were co-transferred with pZE-EP-APTA separately to conduct a 48-hour fermentation in shake flasks. As the results shown in Fig. 4b, all strains harboring different self-regulated bifunctional networks could achieve enhanced 4-HC production when comparing with strains harboring dynamic repression circuits only. Specifically, the strain harboring bifunctional circuits with sgppc3 could produce 0.67 g/L 4-HC at 48 hours (which was 0.58 g/L in dynamic repression circuit only) and the strain harboring bifunctional circuits with sgpykF2 could produce 0.54 g/L 4-HC at 48 hours (which was 0.51 g/L in dynamic repression circuit only). It is noteworthy that the strain harboring bifunctional networks with sgpykF3 reached the highest 4-HC production of 0.78 g/L (which was 0.69 g/L in dynamic repression circuit only), which consistently demonstrated the superiority of the bifunctional networks in 4-HC biosynthesis. Since the remarkable production performance was observed in bifunctional networks containing sgppc or sgpykF, we assumed that the production performance of 4-HC could be further improved by combining these two regulation targets. To verify this hypothesis, we constructed sgpykF2 and sgpykF3 into the plasmid pCS-I12AII14T-PS-PmarO-sgppc3-lpp0.5-marR separately, obtaining pCS-I12AII14T-PS-PmarO-sgppc3-PmarO-sgpykF2-lpp0.5-marR and pCS-I12AII14T-PS-PmarO-sgppc3-PmarO-sgpykF3-lpp0.5-marR. The results shown in Supplementary Fig. 5 illustrated that the strains containing double-site dynamic repression targets did not reach higher titers of 4-HC. Specifically, 4-HC productions of the strain harboring sgppc3-sgpykF2 achieved 0.52 g/L and the strain harboring sgppc3-sgpykF3 reached 0.55 g/L of 4-HC, both of which were lower than the superior strain harboring sgpykF3 only. It is possible that the deficient cell growth of two strains that harbors double-site dynamic repression targets caused the poor production performance. To further investigate the improved production performance in bifunctional networks, we conducted the time course experiments to detect the dynamic fluctuation of the accumulation of intermediate SA and the final product 4-HC. We selected the plasmid pCS-PS-lpp0.5-marR as the static regulation circuit to be co-transferred with pZE-EP-APTA resulting in negative control group. As the results shown in Fig. 4c and d, the dynamic fluctuations of SA accumulation reflected the effect of bifunctional networks. As no disruption applied on WT strain, it showed constantly but low SA level during the whole fermentation cycle, and the production of 4-HC was obtained with a low rate. The other three strains with dynamic bifunctional circuits have similar trends. Specifically, the SA and 4-HC did not show obvious increase before 9-hour. From 9-hour to 12-hour, the delayed repression on pykF or ppc drove carbon flux from cell growth to SA synthesis, resulting in the delayed but drastic elevation of SA accumulation, and the maximum accumulation of SA were achieved at 12-hour. Meanwhile, as the dynamic activation of pqsD and sdgA, the accumulated SA started to be converted to 4-HC, resulting in the gradual decrease of SA concentrations and gradual increase of 4-HC concentrations from 12-hour till the end. The strain harboring sgpykF-2 is an exception since the increase of 4-HC stopped at around 18-hour. It is worth mentioning that owning to the leaking expression of I12AII14T, we could observe the 4-HC accumulation even at the initiation of fermentation process.

To further characterize this SA-responding bifunctional network in 4-HC biosynthesis from the transcriptional level, we conducted a transcriptomic analysis to examine the variations of the genes under dynamic regulations, sdgA and pykF. Here, we selected the 16S ribosomal RNA (encoded as rrsA) as the reference gene because the transcription of rrsA is constant in E. coli. We chose the superior producer with bifunctional self-regulated network contained sgpykF3 to investigate the transcriptional changes during the 48-hour fermentation of 4-HC biosynthesis. Accurate quantifications of transcripts changes were determined by RT-qPCR and normalized with reference gene. As the results shown in Fig. 5a and b, before 6-hour, the transcription of pykF and sdgA elevated following the increase of the cell density. From 12-hour, as the SA gradually accumulated to the maximum level, the transcription of pykF decreased due to the dynamic repression of sgpykF-SpdCas9 system induced by SA. Meanwhile, the transcription of sdgA increased due to the dynamic activation circuits induced by SA as well. The results of transcriptomic analysis coincide with the variations of metabolites, which further indicated and explained the process of engineered bifunctional networks leading to increased production performance in 4-HC biosynthesis.

Fig. 5.

Transcriptomic analysis of gene pykF and sdgA. a. Transcriptional changes of pykF in dynamically controlled 4-HC biosynthesis. b. Transcriptional changes of sdgA in dynamically controlled 4-HC biosynthesis. All error bars represent standard deviation (n = 3). The experiments are biological replicates.

Overall, we constructed and engineered a bifunctional self-regulated network for dynamical balancing multiple precursors on complicated synthetic pathway. By applying the dynamic networks on our engineered 4-HC producing strain, we successfully achieved the increased accumulation of SA, and sufficient supply of malonyl-CoA, which jointly leading to the enhanced production of 4-HC. The fluctuation of metabolisms and transcripts of key enzymes were also validated, providing ideas for optimizing the supply of multiple precursors of other synthetic pathways.

3. Discussion

As the next frontier in metabolic engineering and synthetic biology, dynamic regulation has been demonstrated possessing greater potentials to address the imbalanced metabolic flows than static regulation. Additionally, dynamic regulation could make strains more robust within changing conditions and increase the productivity (Holtz et al., 2010). In this study, we established a bifunctional self-regulated network to balance the carbon flow towards multiple precursors departing from a central metabolite node on glycolysis in E. coli. We chose to regulate the 4-HC synthetic pathway due to its requirements for two precursors, SA and malonyl-CoA. Moreover, the two precursors competitively consume carbon flux both from PEP nodes, and even compete for resources with cell growth. We found that the pyruvate generated from the SA pathway could be an alternative to the native pyruvate supply. To examine if coupling the cell growth with production pathways would be able to improve the production that suggested in the previous research, we rewired the metabolic background to couple the SA production with cell growth (Wang et al., 2019; Noda et al., 2016; Noda et al., 2017; Ponce et al., 1998). Although the increased SA was observed, additional problems such as precursor unbalance occurred. To address them, we further employed a SA-responsive biosensor-controlled CRISPRi system to relieve the negative impact of gene knockout. Moreover, such a biosensor system was used to regulate the enzymes in 4-HC biosynthesis, mitigating extra burden brought by the expression of pathway enzymes during the early stage. We ultimately constructed a self-regulated network by combining the dynamic repression system with the dynamic activation system. Significantly, the self-regulated bifunctional network that harbors sgpykF3 could improve the 4-HC biosynthesis to 0.78 g/L, substantially higher than the static regulation group. Additionally, transcriptomic analysis further detected the dynamic changes of genes sdgA and pykF from the transcriptional level, confirming the design logic of our bifunctional strategies.

In summary, this study contributed to an innovative approach to regulate the carbon flux derived from PEP metabolic nodes by balancing the branches. As demonstrated, our dynamic strategy offered an ideal approach to govern the PEP metabolic node, mitigating the adverse effects on cell viability and the limited PEP availability that the approach of disrupting ppc genes could have (Arioli et al., 2007). Our strategy also achieved autonomous carbon flux allocation between cell growth and two biosynthesis precursors, responding to the varying metabolic states by sensing the intermediate. Prospectively, this approach has broader potential to improve the biosynthesis of products originating from the shikimate pathway.

As illustrated in the study, coupling the intermediated-responsive biosensor systems with pathway genes and CRISPRi system exhibited remarkable robustness and achieved the desired response ranges. To meet different regulation requirements, we employed two cognate biosensor variants with different dynamic pattern, PmarO-MarR and I12AII14T-MarR. Both contain same regulator-binding operon but have different promoter sequences with different promoter strength. Specifically, PmarO is a stringent promoter with limited expression strength, which is suitable for controlling the CRISPRi system due to the high repression efficacy of sgRNA. I12AII14T possess high expression strength with leakage, which can be used for controlling downstream pathway genes. Thus, we illustrated the durability and applicability of a single biosensor system with the advances of biosensor engineering. Besides the manipulation used in this study, other strategies can also be adopted to adjust various sensitivity and dynamic properties of the biosensor system. For example, it is possible to engineer the promoter-regulator component, which could be accomplished by manipulating the binding motif between the promoter and the regulator protein. Additionally, it’s worth noting that by selecting sgRNA spacers with different sequence or length, the repression efficacy of CRISPRi system is adjustable, which is beneficial to down regulating genes critical to cell viability. Furthermore, alternative methods such as programmable RNAi and protein degradation can also be integrated into the dynamic regulation system to achieve a wider array of sophisticated dynamic effects (Yang et al., 2018; Hsia et al., 2016; Solomon et al., 2012; Na et al., 2013; Zhou et al., 2011). However, the dynamic regulation approaches may not consistently outperform static regulation methods. In instances when static regulation approaches impose stress on cell viability or fail to address such stresses, dynamic regulation strategies can offer more effective solutions.

4. Methods and Materials

4.1. Bacterial Strains and Chemicals

Bacterial strains used in this study were listed in Supplementary Table 1. Luria-Bertani (LB) medium was chosen as the basis for E. coli inoculation and plasmid propagation. Appropriate antibiotics such as ampicillin, kanamycin, and chloramphenicol were added into the medium at the final concentration of 100μg/mL, 50μg/mL, and 30μg/mL, respectively. The E. coli strain XL1-Blue was used for gene cloning and plasmid construction. The Strain BW25113 containing F’ from XL1-Blue was used as the host strain for the biosynthesis of salicylate and 4-HC. Knockout derivatives of BW25113 F’ were created via P1 phage transduction method (Thomason et al., 2007; Atsumi et al., 2008). M9Y medium containing (per liter): glycerol (20 g), yeast extract (5 g), NH₄Cl (1 g), Na₂HPO₄ (6 g), KH₂PO₄ (3 g), NaCl (0.5 g), MgSO₄·7H₂O (2 mmol), CaCl₂·2H₂O (0.1 mmol), and vitamin B1 (1.0 mg) was used for biosynthesis. Standard chemicals including salicylic acid and 4-HC were purchased from Sigma-Aldrich unless otherwise specified.

DNA Manipulation

All manipulations of DNA were conducted referring to the standard molecular cloning protocols (Gibson et al., 2009). Plasmids used in this work were listed in the Supplementary Table 1. The plasmids of salicylic acid (SA) and 4-hydroxycoumarin (4-HC) synthesis, pZE-EP-APTA and pCS-PS were obtained from our previous study (Shen et al., 2017). Plasmids pHA-MCS (high-copy number) (Jiang et al., 2022), pMK-MCS (medium copy number) and pLC-MCS (low copy number) were employed in this study. Specifically, the plasmid pMK-MCS containing a p15A origin, a kanamycin resistance gene, pLlacO1 promoter, and T1 terminator, was constructed in our lab. And the plasmid pLC-MCS containing a pSC101 origin, a chloramphenicol resistance gene, pLlacO1 promoter, and T1 terminator, was constructed in our lab as well. These plasmids also carry a synthetic multi-cloning site (MCS) that sequentially contains the recognition sites of Acc65I, NdeI, BsrGI, SalI, ClaI, HindIII, NheI, BamHI, and MluI (Lutz and Hermann., 1997). For the characterization of the MarR-PmarO biosensor system, the PmarO promoter harboring green fluorescent gene egfp and the hybrid promoter I12AII14T harboring egfp were amplified from our previous study (Zou et al., 2021). Both were inserted into plasmid pMK-MCS separately by XhoI and SalI, resulting in plasmids pMK-PmarO-eGFP and pMK-I12AII14T-eGFP. To express regulator protein in the same plasmid configurations, lpp promoter harboring marr was amplified from our previous study and inserted into plasmids pMK-PmarO-eGFP and pMK-I12AII14T-eGFP by using SacI and SpeI, respectively (Zou et al., 2021). For the characterization of the self-regulated bifunctional networks, sgRNA variants were designed and constructed into biosensor systems using ApaLI and NheI, resulting in pMK-PmarO-sgRNAs. A red fluorescent protein encoded gene rfp was amplified and constructed into biosensor systems using KpnI and SalI, yielding pMK-I12AII14T-RFP. For the dynamic regulation section, pLlacO1 promoter was substituted by PmarO promoter and I12AII14T promoter in pCS-PS using XhoI and EcoRI, resulting in pCS-PmarO-PS and pCS-I12AII14T-PS. In order to inhibit the expression of pykA, PmarO promoter harboring sgRNA variant sgpykA was inserted into plasmid pLC-MCS using XhoI and NheI, yielding pLC-PmarO-sgpykA. To generate different plasmid configurations of 4-HC synthesis, PmarO promoter harboring sgRNA variants were amplified and constructed into the plasmid pCS-PS using SpeI and SacI, leading to pCS-PS-PmarO-sgRNAs. As the control group, constitutive promoter lpp0.5 harboring marr was constructed in the plasmid pCS-PS-PmarO-sgRNAs using NheI and SacI, resulting in pCS-PS-PmarO-sgRNAs-lpp0.5-marR (see Supplementary Table 2 for a list of all sgRNAs used in this study).

Cultivation Conditions

For the characterization of PmarO-MarR biosensor system and self-regulated bifunctional networks, all transformants were cultured in 3.5 mL of LB medium with appropriate antibiotics at 37 °C with a shaking speed at 270 rpm in the New Brunswick Excella E24 shaker (orbital diameter: 19.1 mm). Cultures of 150 μL were transferred into test tubes containing 3.5 mL fresh LB medium with appropriate antibiotics. The cells were then cultivated at 37 °C with a shaking speed at 270 rpm. Different concentrations of salicylic acid in sodium salt form were added into medium after 1.5 h (when OD₆₀₀ reached approximately 0.4). The samples were collected at the appropriate time point and then subjected to measurement of cell densities (OD₆₀₀) and green fluorescence intensities.

For salicylic acid and 4-hydroxycoumarin biosynthesis, all transformants were cultured in 3.5 mL LB medium with appropriate antibiotics at 37 °C with shaking at 270 rpm for 9 hours. Then 400 μL cultures were transferred into 125-mL baffled flasks containing 20 mL of proper media and cultivated at 30 °C with shaking at 270 rpm for 48 hours. When needed, isopropyl β-D-1-thiogalactopyranoside (IPTG) was added to the medium during culture transfer with a final concentration of 0.5 mM. 1 mL cultures were sampled to measure cell optical density at 600 nm (OD₆₀₀) and analyzed the products by HPLC at the appropriate time point.

Fluorescence Assays

The Synergy HT plate reader from Biotek was used for fluorescence assays. The samples were diluted 10 times (20 μL sample with 180 μL water) and transferred into a black 96-well plate with clear bottom (Corning 3603). The plate was scanned in the Synergy HT (BioTek) plate reader. The eGFP fluorescence intensity was detected by using an excitation filter of 485/20 nm and an emission filter of 528/20 nm. The cell densities (OD₆₀₀) were also measured using this plate reader. The experiments were performed in triplicates, and data are presented as the averages and standard deviations (n = 3).

HPLC Analysis

The standard and samples were dissolved by adding methanol and water, respectively. The mixture was centrifuged at 12000 rpm for 10min. salicylic acid and 4-hydroxycoumarin concentration was analyzed by Agilent HPLC 1260 Infinity II (1260 Infinity II Diode Array Detector WR) with a reverse-phase ZORBAX SB-C18 column. For salicylic acid and 4-hydroxycoumarin detection, 0.1%TFA and methanol were used as the mobile phase at a flow rate of 1 mL/min. The analyzing method was set as follows: 5% methanol from 0 to 2 min, 5%–60% methanol from 2 to 25 min, 60%–90% methanol from 25 to 27 min, 90%–5% methanol form 27–29 min. Salicylic acid and 4-hydroxycoumarin can be detected at 25.4 min and 26.5 min, respectively. Glycerol concentration was analyzed by Dionex Ultimate 3000 (Ultimate 3000 Photodiode Array Detector) with a Coregel-64H column (Transgenomic). 4 mM H2SO4 was used as mobile phase at a flow rate of 0.40 mL/min. The oven temperature was set to 45°C. Glycerol can be detected at 23.1 min.

RT-qPCR Analysis

Cells were collected during different fermentation times from the shaking flask cultures and frozen in liquid nitrogen. Total RNA was purified with the NucleoSpin miRNA Kit (Macherey-Nagel). The synthesis of complementary DNA was conducted using iScript Reverse Transcriptase with 1 mg of total RNA. Quantitative PCR was performed with EvaGreen Supermix and analyzed on a CFx96 Real-Time System (BioRad). Gene rrsA was selected as the internal standard. For strand-specific RT–PCR, gene-specific primers were used to synthesize the cDNA and 20 mg/ml actinomycin D was supplemented into the buffer to ensure strand specificity. The experiments were performed in triplicates, and data are presented as the averages and standard deviations (n = 3).

Statistics

Sample size was not predetermined using any statistical method. For eGFP and RFP regulation assays and shake flask experiments, all data were reported as the mean ± standard deviation of biological triplicates (n = 3) and were presented in the corresponding figure legends. Data analysis was performed using Microsoft Excel. The colonies used for data collection were randomly selected from the agar plates. The investigators were not blinded to allocation during experiments or outcome assessment.

Highlights.

• We established a self-regulated bifunctional network to relieve precursor competitions.

• We employed two variants derived from the same biosensor for distinct dynamic regulation goals.

• Coupled CRISPRi with biosensor, balancing the carbon flux for improved cell growth and production.

• The self-regulated bifunctional network harboring sgpykF3 improved 4-HC production to 0.78g/L.

• Transcriptional changes of sdgA and pykF indicated the design logic of dynamic regulation system.