Introduction

Tacrolimus (FK506), a 23-membered macrolide polyketide, is well known for its immunosuppressant activity and can be used to prevent graft rejection after allogeneic kidney, liver and heart transplantations [2, 3, 22, 46]. It has also been successfully applied in the treatment of inflammatory skin diseases, and is being investigated for therapeutic effects in other autoimmune diseases, including rheumatoid arthritis, ulcerative colitis, and Crohn’s disease [1, 5, 23]. However, owing to the low titers obtained to date through microbiological synthesis, the cost of use of tacrolimus in the clinical therapy is very high. Ever since tacrolimus was first isolated from Streptomyces tsukubaensis no. 9993 by the Fujisawa Pharmaceutical Company in 1987, it has attracted increasing attention from a number of research groups, due to the aforementioned pharmaceutical potential and broad clinical applications [10, 24, 27, 34, 36, 37, 50, 61].

The biosynthesis of the tacrolimus macrolactone is mediated by the polyketide synthase (PKS)/non-ribosomal peptide synthetase (NRPS) system, comprising multiple-fatty acid synthase-like domains arranged in sets of modules [16]. Specifically, the biosynthesis of tacrolimus begins with an unusual chorismate-derived unit (4R,5R)-4,5-dihydroxycyclohex-1-enecarboxylic acid (DHCHC) [3]. Subsequently, ten successive extender units (including two malonyl-CoA, five methylmalonyl-CoA, two methoxymalonyl-ACP and a unique allylmalonyl-CoA) are assembled into a polyketide chain by PKS [3, 15, 34]. The linear polyketide chain is subsequently condensed with lysine-derived pipecolate by NRPS and the macrolide ring is formed by cyclization. Finally, the ring is further modified by two post-PKS processing reactions to form the final product (tacrolimus) [9]. Compared with the traditional strategies for tacrolimus overproduction (i.e., random mutagenesis, optimization of culture media and process conditions) [15, 26, 33, 51, 62], the identification of the entire tacrolimus biosynthesis genes cluster [17, 35] and the subsequent elucidation of the full genome sequence [4] enabled the development of more rational strategies for tacrolimus overproduction. For example, the exogenous feeding of tacrolimus precursors in the culture media has been regarded as an efficient and economical strategy to increase the yield. These precursors included methyl oleate [34], vinyl pentanoate [37], shikimate [21], picolinic acid, nicotinic acid and salicyl alcohol [49, 50], amino acids (i.e., lysine, proline, leucine, etc.) [21, 61], and so on. At the same time, the copy number increase of tacrolimus biosynthetic genes and the modification of the specific regulatory genes (fkbN, fkbR, tcs2, tcs7 and allN) were also able to effectively improve tacrolimus production [8, 16, 37, 38, 40, 64]. Additionally, the genome-scale static/dynamic metabolic network model had been applied to search for the bottlenecks of tacrolimus biosynthesis [19, 52] and several potential genetic targets have been predicted that promise to enable an improved yield of tacrolimus.

Metabolomics has been widely applied as a powerful tool to describe the comprehensive metabolic status of a microorganism, as well as to reveal dynamic metabolic responses to exogenous or endogenous disturbances by analyzing the diversity and abundance of small-molecule metabolites [12, 13, 28, 57]. The most common statistical approaches used in metabolomics are principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA). PCA and PLS-DA are used not only to convert metabolomic data into a qualitative visual presentation (scores plot), but also to cluster biological samples into groupings based on similarity or difference [58]. Furthermore, a new approach named weighted correlation network analysis (WGCNA) has been recently proposed to describe the correlation relationships between clusters of highly correlated genes, proteins, metabolites or modules and external conditions or sample traits [42]. This approach can be used to reflect the continuous nature of the underlying co-expression information, and avoids information loss by setting artificial threshold parameters [42, 47, 48, 53, 55, 63]. With the help of appropriate statistical and visualization tools, metabolomic analysis has recently been applied to identify key metabolites or pathways associated with various biochemical products in different Streptomyces species, including the intracellular metabolic responses to salt stress and nutrient depletion in S. coelicolor [28, 57], intracellular amino acid responses of Streptomyces lydicus to different pitching ratios [13], metabolomic profiling of S. lydicus E9 following the addition of glutamate and proline [12], and comparative metabolic profiling of S. hygroscopicus and S. tsukubaensis to enhance rapamycin and tacrolimus production, respectively [61, 65]. These results demonstrated that metabolomics is a valuable tool for analyzing cell metabolism and the intracellular responses to environmental changes. However, even though several exogenous feeding strategies for improving tacrolimus overproduction have been reported, to our best knowledge, the dynamic response mechanisms of intracellular metabolism and the key-limiting steps of the corresponding pathways have not been systematically analyzed to date.

In this study, the intracellular metabolic responses to the enhancement of four synthetic precursor pathways in tacrolimus biosynthesis have been evaluated via the individual exogenous feeding of malonate, lysine, shikimate and citrate. Using metabolomic datasets obtained by gas chromatography-mass spectrometry (GC–MS) and liquid chromatography-mass spectrometry/mass spectrometry (LC–MS/MS), WGCNA was applied to reveal the metabolic response modules and hub metabolites highly association with tacrolimus accumulation during the fermentation time course. These results indicated the pentose phosphate, shikimate and aspartate pathways showed higher positive correlations with tacrolimus production than other precursor pathways during the rapid synthesis phase. Combined with the GSMM model of S. tsukubaensis, we predicted several possible key-limiting steps in the identified metabolic pathways. The best strain (HT-aroC/dapA) produced 128.19 mg/L of tacrolimus, which was 1.64-fold higher than the result obtained with the parent strain under the same conditions.

Materials and methods

Strains, plasmids and culture conditions

The parent strain S. tsukubaensis D852, which was stocked in our laboratory and deposited in China General Microbiological Culture Collection Center with the accession number CGMCC 7180, was used in this work. E. coli JM109 was used to propagate all plasmids according to standard molecular biology procedures [44]. The integrative E. coli–Streptomyces vector pIB139 containing the ermE* promoter (PermE*) was used for gene overexpression in S. tsukubaensis [59]. E. coli ET12567/pUZ8002 was used as the nonmethylating plasmid donor strain for intergeneric conjugation with S. tsukubaensis [25]. E. coli strains were cultured in Luria–Bertani (LB) medium at 37 °C. All strains and plasmids used in this work are listed in the supplementary material (Supplementary Materials A: Table S1).

The spores and seed culture of S. tsukubaensis were performed as the previously described methods of Xia et al. [61] and Martínez-Castro et al. [31], respectively. A total of 1 mL seed culture was transformed into 100 mL fermentation medium of a 500 mL Erlenmeyer flask and inoculated at 28 °C and 220 rpm for 7 days. The batch fermentation medium contained 20.0 g/L glucose, 40 g/L dextrin, 2.5 g/L soybean oil, 1 g/L NaCl, 2 g/L (NH4)2SO4, 1 g/L K2HPO4, 2 g/L CaCO3, 1 g/L MgSO4·7H2O, 0.001 g/L FeSO4·7H2O, 0.001 g/L MnCl2·4H2O, and 0.001 g/L ZnSO4·7H2O, and adjusted to pH 7.0 before autoclaving. Four kinds of metabolites (lysine, sodium citrate, sodium malonate, and sodium shikimate) were first prepared with sterilized water and filtered by 0.22 μm filter. Then, the sterile solutions of these metabolites were added into the medium with the final concentration 0.05 g/L after 30 h cultures, respectively.

Gene cloning, plasmid construction and transformation

All DNA manipulation and transformation of Streptomyces were prepared using the standard protocols [44]. The chorismate synthase (CHS) gene (aroC, STSU_29556) and dihydrodipicolinate synthase (DHDPS) gene (dapA, STSU_08193) were amplified from S. tsukubaensis genomic DNA by PCR amplification using the corresponding primer pairs (Supplementary Materials A: Table S2), respectively. Specifically, the gene aroC and dapA were digested by NdeI-XbaI and then inserted into the same site of pIB139 to obtain pAROC and pDAPA, respectively. For the plasmid construction of two genes co-overexpression, PCR product of aroC was digested with NdeI-HindIII, and cloned into the same sites of pUC18, generating pUC18-C. PCR product of dapA was digested with HindIII-XbaI and cloned into the HindIII-XbaI sites of pUC18-C, generating pUC18-CA. Then, pUC18-CA was digested with NdeI-XbaI and the long fragment containing the two genes was transferred to the same sites of pIB139, yielding pCA. After these plasmids were constructed, they would be transferred into E. coli ET12567/pUZ8002, and separately introduced into S. tsukubaensis through conjugal transfer [25]. Finally, the engineered strains of HT-aroC and HT-dapA would be obtained and the positive exconjugants were verified by PCR amplification and DNA sequencing with the corresponding primer pairs. The positive exconjugant of HT-aroC/dapA was verified by PCR amplification and DNA sequencing with primer pair pIB-F/pIB-R (Supplementary Materials A: Table S2).

Analytical methods

Dry cell weight was determined by washing twice with deionized water after centrifuging at 8000 rpm for 10 min, and then drying at 80 °C to constant weight. The residual total glucose in fermentation broth was quantified using phenol–sulfuric acid method [61]. Inorganic nitrogen (NH4 +-form) was quantified using the indophenol method [56, 60]. Inorganic phosphorus (PO4 3+-form) content was assayed with the ammonium molybdate spectrophotometric method [11, 60]. Tacrolimus was extracted and quantified by the previous described methods [21, 61].

Analysis of in vitro enzyme activities

The samples for enzyme activity measurement were harvested after 72 and 144 h cultures, respectively, and prepared as the reported methods [20]. The CHS activity was measured by monitoring the formation of chorismate at 275 nm as previously described [7]. The enzyme activity of DHDPS was determined as described by Laber et al. [29]. Total protein concentrations were quantified by Bradford assay with a reagent solution (Quick Start Bradford Dye, BioRad). The assays of enzyme activities were normalized to the total protein.

The identification of the key-limiting steps in the significant precursor pathways

Based on the constructed GSMM model of S. tsukubaensis by our group [19], the previous model had been further supplemented and completed, including filling of metabolic gaps, correcting of reaction directions and adding the exchange reactions (Supplementary Materials B). In this work, the identified metabolic pathways, which showed highly positive association with tacrolimus biosynthesis under different exogenous feeding treatments, would be selected to predict the possibly effective targets of enhancing tacrolimus production (Supplementary Materials A: Table S3). After the model was validated under the present cultural condition, the overexpressed targets would be predicted by the previous reported methods [19]. In brief, the intracellular metabolic fluxes could be firstly calculated by flux balance analysis (FBA) algorithm, and then the flux of the candidate reaction would be amplified to some extent (for instance 2.0-fold). Subsequently, a new set of intracellular metabolic fluxes would be calculated by minimization of metabolic adjustment (MOMA) algorithm. In the above process, MOMA employs quadratic programming to identify a point in flux space, which is closest to the initial flux, compatibly with the flux adjustment constraint [30].

The identification of targets was performed by comparing a fraction value, f PH (the ratio of weighted and dimensionless specific growth rate and specific tacrolimus production rate) [6, 19, 32]:

$$f_{\text{PH}} = f_{\text{Biomass}} \times f_{\text{FK506}} = (C_{\text{Biomass,\,model}} /C_{\text{Biomass,\,wild}} ) \times (C_{\text{FK506,\,model}} /C_{\text{FK506,\,wild}} )$$

The target with a higher f PH value would be considered as the better candidate for improving tacrolimus production.

GC–MS and LC–MS/MS analysis

In order to analyze the intracellular metabolic response of S. tsukubaensis on different exogenous feedings, GC–MS and LC–MS/MS were performed for intracellular metabolites analysis. The cells collected at 6, 24, 48 and 72 h after feeding would be immediately quenched by 60% methanol (−40 °C, v/v) after taking from the fermentation broth. Then, the extraction, determination and analysis of intracellular metabolites were performed according to the previous described methods [61]. Succinic d4 (St. Louis, MO, USA) and d-sorbitol-13C6 (St. Louis, MO, USA) would be added in advance and served as the internal standard for GC–MS analysis and LC–MS/MS analysis, respectively. LC–MS/MS measurements of metabolites only included sugar phosphate intermediates of PPP and glycolysis, such as glucose 6-phosphate, fructose 6-phosphate, fructose 1,6-biosphosphate, glyceraldehyde 3-phosphate, phosphoenolpyruvate, erythrose 4-phosphate, ribose 5-phosphate, ribulose 5-phosphate, xylulose 5-phosphate and sedoheptulose 7-phosphate. All the chemicals used in the metabolome isolation, GC–MS and LC–MS/MS analysis were purchased from Sigma-Aldrich.

WGCNA network construction

A metabolic correlation network was established from the GC–MS and LC–MS/MS data sets by calculating weighted Pearson correlation matrices relative to metabolites abundance firstly. And the standard procedure of WGCNA would be performed to create the networks through the previous reports [42, 53]. Briefly, connection strength matrices would be first created from the transformation of weighted correlation matrices by a power function, and then used to calculate topological overlap (TO). The metabolites with highly similar correlation relationships would be grouped into the same modules through hierarchical clustering based on the results of TO. Subsequently, average linkage hierarchical clustering was performed to obtain metabolite dendrograms. The module assignment determined by the Dynamic Tree Cut of WGCNA was distinguished with the color row underneath the dendrograms. The modules with correlation r ≥ 0.5 and statistical significance p value less than 0.05 were extracted for further investigation. Hubs metabolites were screened by high connectivity with other metabolites (>5) in the distinct modules by exporting the WGCNA network to VisANT and Cytoscape for network visualization [42].

Data processing and multivariate statistical analysis

The peak area of each metabolite was normalized to the internal standard succinic acid d4 and cell biomass in GC–MS analysis. Absolute metabolite concentrations determined by LC–MS/MS in extracts were normalized to the biomass and internal standard d-sorbitol-13C6. Both of the normalized metabolites between GC–MS and LC–MS/MS analysis were combined together finally. The relative abundance matrix of the whole identified metabolites at each sampling time point was imported into SIMCA-P package (Ver 11.5; Umetrics, Umea, Sweden) for the multivariate statistical analysis. In this study, five biological replicates were used to perform multivariate analysis for each sample condition, and the experimental data were calculated as the mean value with the error indicated by the standard deviation.

Results

Fermentation properties under different exogenous feedings

The main goal of the various exogenous feeding strategies tested in this study was to enhance precursor synthesis for tacrolimus overproduction, building upon earlier work [3]. It would be more effective and intuitive to understand the dynamic responses of intracellular metabolism by directly disturbing the carbon fluxes of the precursor pathways of tacrolimus biosynthesis and examining the consequences. On the basis of literature research, the exogenous feeding of shikimate, malonate, lysine and citrate were very effective on tacrolimus overproduction by directly enhancing the fluxes in the shikimate pathway, malonyl-CoA synthesis, lysine metabolism and TCA cycle, respectively [3, 15, 54, 61]. Therefore, considering the simple structure and availability of these precursors, all four was finally selected for the exogenous feeding experiments.

Aiming at finding the features of different exogenous feeding strategies, the key fermentation parameters were first compared with the basic culture medium. As shown in Fig. 1, while a similar fermentation tendency was found among all feeding treatments, obvious difference could also be observed in pH values, substrates consumption (total glucose, NH4 + and PO4 3+), biomass and tacrolimus accumulation. For example, all of the cultures undergoing exogenous feedings displayed higher substrate consumption rates than the control (Fig. 1a-c). Cell growth was enhanced significantly by all exogenous feeding treatments compared with the control (Fig. 1e). Furthermore, an earlier starting point of tacrolimus biosynthesis than in the control, at 6 h after commencing exogenous feeding, was found in all treatments (Fig. 1f). The final tacrolimus concentrations in the lysine, malonate, shikimate and citrate feeding strains had reached to 133.49 ± 4.25, 127.32 ± 5.02, 119.73 ± 5.26 and 103.01 ± 3.05 mg/L, respectively, significantly higher than the control (78.26 ± 3.26 mg/L). Moreover, the strains fed with malonate, shikimate and citrate showed higher final biomass concentrations than the control (9.24 ± 0.41 g/L), such as 10.83 ± 0.91, 10.37 ± 1.22 and 10.5 ± 0.8 g/L for malonate, shikimate and citrate, respectively. Interestingly, the lysine feeding strain had a higher biomass concentration during the exponential and early-stable phases, while its final biomass yield (9.17 ± 0.52 g/L) was lower than that of the control. These results suggested that the intracellular metabolism was perturbed obviously and that additional carbon flux was rerouted towards tacrolimus biosynthesis under the exogenous feeding treatments.

Fig. 1
figure 1

The fermentation properties of S. tsukubaensis under different exogenous feeding (malonate, lysine, shikimate and citrate). a total glucose, b inorganic ammonium (NH4 +-form), c inorganic phosphorus (PO4 3+-form), d pH, e dry cell weight, f tacrolimus concentration

Full size image

Differential effects of precursor feedings on metabolomic profiles

In the previous study, we developed an optimized protocol for GC–MS and LC–MS/MS analysis, which enabled to identify more than 98 metabolites from S. tsukubaensis, encompassing sugars, amino acids, organic acids, fatty acids and so on [61]. Using the same analytical method, a total of 93 intracellular metabolites, including 83 metabolites by GC–MS analysis and 10 sugar phosphate by LC–MS/MS analysis (Supplementary Materials A: Table S4), were identified in 100 samples at the different sampling times (i.e., 6, 24, 48 and 72 h after exogenous feeding) for all treatment conditions. PCA score plots of the corresponding metabolomic profiles were compared between the control and all exogenous feeding treatments (Fig. 2).

Fig. 2
figure 2

PCA analysis of GC–MS and LC–MS/MS metabolomic profiles. Score plot of fermentation samples collected at 6, 24, 48, and 72 h after feeding, respectively. The culture medium was supplemented with 0.05 g/L shikimate, citrate, malonate and lysine after 30-h cultivation, respectively

Full size image

Analysis of the score plots revealed the following features: (1) all of the same treated samples at different time points could be obviously clustered together, respectively, and separated with other time point samples completely; (2) the differently treated samples collected at the same time point could also be clustered together clearly; (3) a clear change trend of the metabolomic profiles across the time courses could be found among all exogenous feeding strategies. Interestingly, a remarkable regularity of intracellular metabolism could also be clearly observed along the time course. For example, the differences of metabolomic profiles in different feeding treatments were gradually reduced from 6 to 48 h (in the exponential phase), but an increasing difference reappeared at the stationary phase (at 72 h after feeding). A possible reason may be that cell growth played a dominant role in the exponential phase, while some metabolic differences still existed in all of the samples. Meanwhile, the metabolic differences resulting from the different feeding treatments may not have disappeared, and still remained in a “masked” state beneath the heavy growth-related fluxes until the stationary phase. Although the molecular details are still unknown, the different effects of exogenous feeding strategies merit further investigation.

WGCNA correlation network construction

In order to analyze the intracellular metabolic responses to tacrolimus synthesis among all the exogenous feeding treatments, a WGCNA was constructed to identify the significant metabolic modules and hub metabolites highly associated with tacrolimus biosynthesis across the time courses. Using the GC–MS and LC–MS/MS datasets, 4 (M6h-F1, M6h-F2, M6h-F3 and M6h-F4), 1 (M24h-F1), 1 (M48h-F1) and 2 (M72h-F1 and M72h-F2) distinct metabolic modules were identified successfully at 6, 24, 48 and 72 h after exogenous feedings, respectively (Table 1; Fig. 3).

Table 1 Member metabolites clustered into each of tacrolimus-responsive modules
Full size table
Fig. 3
figure 3

WGCNA of metabolic profiles of S. tsukubaensis highly associated with tacrolimus biosynthesis under different exogenous feedings. Fermentation samples were collected at 6 h (a), 24 h (b), 48 h (c), and 72 h (d) after exogenous feeding, respectively. The heatmap depicts the topological overlap (TO) matrix among 93 metabolites in the analysis. Light color represents low TO and progressively darker red color represents higher TO. Blocks of darker colors along the diagonal are the modules. The metabolites dendrogram and module assignment are also showed along the left side and the top. The modules highly associated with tacrolimus production (r ≥ 0.5 and p value <0.05) are indicated with their metabolites within the modules in different colors, and the correlation coefficients and p values are listed below the modules (color figure online)

Full size image

The metabolites within the modules from different time points showed great differences. For example, the most modules could be found at 6 h after feeding, and several metabolites in these modules (M6h-F1, M6h-F2, M6h-F3 and M6h-F4) were involved in central carbohydrate metabolism and amino acid metabolism. Furthermore, the enhancement of several precursor pathways in tacrolimus biosynthesis was also observed based on their respective specific metabolites, such as pipecolate, malonic acid, shikimate, phosphoenolpyruvate and erythrose 4-phosphate. Compared with the control, an earlier start-point of tacrolimus synthesis was also found in the all feeding treatments (Fig. 1f). Among all the sampling time points, M24h-F1 and M48h-F1 were identified in the exponential phase (at 24 and 48 h after feeding, respectively). A possible reason is that the rapid growth of bacteria might take precedent over all other cellular processes in the exponential phase, including over tacrolimus biosynthesis. In agreement with this idea, sugar phosphate intermediates and organic acid were found in the module M24h-F1 and M48h-F1 (Table 1). During the stationary phase (at 72 h after feeding), analysis of the metabolites within each module (M72h-F1 and M72h-F2) indicated that a part of the precursor synthesis pathways showed a high correlation with tacrolimus production. These included the shikimate pathway (i.e., phosphoenolpyruvate, tryptophan and phenylalanine) and aspartate pathway (i.e., aspartic acid, isoleucine, methionine and threonine). At the main phase of tacrolimus accumulation, the enhancement of shikimate and aspartate pathways would provide additional amounts of important precursors, such as DHCHC, pipecolate and propionyl-CoA.

Hub metabolites possess a high degree of connectivity in biological interaction networks and consequently have high biological importance [63]. In accordance with the results, only the modules in the rapid synthesis phase of tacrolimus production (M48h-F1, M72h-F1 and M72h-F2) were used to further analysis of hub metabolites. A total of 3 (pyruvate, phosphoenolpyruvate and methylmalonate) and 4 (leucine, methionine, tryptophan and phenylalanine) hub metabolites were significantly associated with the modules of 48 and 72 h, respectively (Fig. 4). In the M48h-F1 module, the hub metabolites of pyruvate, phosphoenolpyruvate and methylmalonate were located in the key nodes for DHCHC and methylmalonyl-CoA synthesis, all of which are necessary tacrolimus precursors [3]. Among the hub metabolites in the modules M72h-F1 and M72h-F2, three metabolites (leucine, tryptophan and phenylalanine) had been confirmed to promote tacrolimus production by exogenous feeding [61]. Methionine was first found to show a high connectivity with other metabolites and to have a strong correlation with tacrolimus synthesis. In fact, methionine could transform into acetyl-CoA or propionyl-CoA via its degradation pathway [14], both of which are the important coenzyme A sources in tacrolimus biosynthesis. These identified hub metabolites might not only serve as biomarkers for tacrolimus synthesis, but also specifically indicate that the shikimate and aspartate pathways might be very important for tacrolimus synthesis during the stationary phase.

Fig. 4
figure 4

Hub metabolites and their metabolic profiles as represented by node and edge graph. a Pyruvate, phosphoenolpyruvate and methylmalonate for M48h-F1; b leucine and methionine for M72h-F1; c tryptophan and phenylalanine for M72h-F2

Full size image

Prediction of potential targets based on in silico simulations

In a quasi-steady state metabolic model, there are an infinite number of flux distributions that satisfy the GSMM model so that the subspace of allowed flux distributions needs to be further constrained by boundary conditions to obtain an optimal flux distribution [18]. In this work, total glucose (including dextrin, which can be hydrolyzed by the cells to yield glucose [43]), inorganic nitrogen (NH4 +-form), inorganic phosphorus (PO4 3+-form) and tacrolimus production were calculated as the model constraints based on the existing culture conditions. For instance, the specific uptake rates of glucose, NH4 + and PO4 3+ were set to 1.259, 0.0913 and 0.0476 mmol/g DCW/h, respectively. At the same time, the experimental specific production rate of tacrolimus was determined to be 0.1069 μmol/g DCW/h and was set as the lower bound for the tacrolimus exchange reaction. Using the experimentally determined values, the specific growth rate of GSMM was predicted to be 0.037 h−1, which was basically consistent with the experimental data (0.042 h−1).

According to the analysis of these distinct modules, several metabolites derived from PPP, shikimate and aspartate pathways had been identified as having highly positive correlations with tacrolimus production during its rapid synthesis phase (Table 1). However, the key-limiting steps of these metabolic pathways could not be identified accurately. Thus, to find the key-limiting steps, the above three metabolic pathways were screened to predict the effective overexpression targets using the GSMM model. As shown in Fig. 5, all of the potential targets were identified and ranked by their f PH values. Among all predicted overexpression targets, the most effective targets were found in PPP. These included r56 (Glucose-6-phosphate 1-dehydrogenase, f PH = 30.95), r57 (6-phosphogluconolactonase, f PH = 30.95), r58 (Phosphogluconate dehydrogenase, f PH = 5.96), r70 (Phosphoric-monoester hydrolases, f PH = 2.28), r62 (Phosphoglucomutase, f PH = 2.11) and r64 (Transaldolase, f PH = 1.45). The enhancement of flux in PPP could not only provide more reducing power (NADPH), but may also result in an increased flux of carbon to erythrose-4-phosphate, (a direct precursor of shikimate pathway), both of which are very important for improving tacrolimus production. In previous reports, several targets in PPP, such as glucose-6-phosphate dehydrogenase (r56) [19] and transketolase [52], were shown to play an effective role in improving tacrolimus titer. The next group of targets (r145, 2-dehydro-3-deoxyphosphoheptonate aldolase; r146, 3-dehydroquinate synthase; r149, Shikimate kinase; r151, Chorismate synthase) that were located in the shikimate pathway, had the same f PH values (f PH = 14.86). Overexpressing these targets may improve the pool of DHCHC, which is the starting unit of tacrolimus biosynthesis, by enhancing the shikimate pathway [3, 19]. Meanwhile, previous reports had also confirmed that overexpression of 3-deoxy-d-arabinoheptulosonate-7-phosphate synthase (r145) enhanced tacrolimus production by approximately 50% in S. tsukubaensis D852 [19]. Among the predicted targets in the aspartate branch, those targets in lysine biosynthesis pathway (i.e., r96, dihydrodipicolinate synthase; r97, Dihydrodipicolinate reductase; r98, 2,3,4,5-tetrahydropyridine-2-carboxylate N-succinyltransferase; r99, Succinyldiaminopimelate aminotransferase; r100, Succinyl-diaminopimelate desuccinylase; r101, Diaminopimelate epimerase) had higher f PH values (f PH = 3.36), which would result in an increase of intracellular lysine flux and precursor pool (pipecolate), thus improving the product titer. Similar with lysine biosynthesis, the targets r93 (Aspartate kinase, f PH = 2.39) and r94 (Aspartate-semialdehyde dehydrogenase, f PH = 2.39) in the homoserine biosynthesis branch showed positive effects on tacrolimus overproduction, both of which were also beneficial for directly and indirectly improving the levels of lysine and pipecolate.

Fig. 5
figure 5

The predicted results of the key-limiting targets by GSMM model in S. tsukubaensis. All of the candidate targets were selected from PPP, shikimate and asparate pathways based on the WGCNA analysis. The detailed information of the all predicted targets was as following: r56 glucose-6-phosphate 1-dehydrogenase, r57 6-phosphogluconolactonase, r58 Phosphogluconate dehydrogenase, r62 Phosphoglucomutase, r64 Transaldolase, r70 Phosphoric-monoester hydrolases, r93 Aspartate kinase, r94 Aspartate-semialdehyde dehydrogenase, r96 dihydrodipicolinate synthase, r97 Dihydrodipicolinate reductase, r98 2,3,4,5-tetrahydropyridine-2-carboxylate N-succinyltransferase, r99 Succinyldiaminopimelate aminotransferase, r100 Succinyl-diaminopimelate desuccinylase, r101 Diaminopimelate epimerase, r145 2-dehydro-3-deoxyphosphoheptonate aldolase, r146 3-dehydroquinate synthase, r149 Shikimate kinase, r151 chorismate synthase

Full size image

Effect of the targets overexpression manipulation on tacrolimus biosynthesis

According to the f PH values, the three targets r56, r57 and r58 are all located in the initial step of the pentose phosphate pathway, which directly control the carbon flux into this pathway. Since a previous report had confirmed that overexpressing the target r56 was effective for tacrolimus overproduction [19], these three targets were excluded from the subsequent genetic manipulation study. Similarly, the target reactions r145 and r146 were not been considered as gene modification targets because the target r145 had also been verified before [19]. By contrast, the target r151 catalyzes the chorismate synthesis reaction, and thus may be capable of more directly increasing the chorismate concentration if overexpressed. Therefore, only r151 was selected as a candidate for overexpression, while all four targets in the shikimate pathway had the same f PH value. Among the predicted targets from lysine pathway, the target r96 had been reported to catalyze the limiting step for lysine overproduction [41]. In fact, lysine metabolism was the only source of pipecolate in tacrolimus biosynthesis. Therefore, the increasing of lysine concentration would be benefit for improve tacrolimus yield. The gene corresponding to r96 was, therefore, also considered for overexpression, while the rest of the predicted targets (r97, r98, r100, r99 and r101) had the same f PH values as r96. Finally, the targets from the shikimate pathway (r151) and lysine biosynthesis (r96) were selected for further experimental verification. To our delight, the final tacrolimus production was improved effectively in all the overexpression strains (Fig. 6). Moreover, the results of enzyme activity assays also further confirmed that the corresponding targets were successfully enhanced both in the exponential and the stationary phase (Table 2). Additionally, the different promoting effects of the selected targets on tacrolimus accumulation could also be observed clearly.

Fig. 6
figure 6

The effect of genes overexpression manipulation on tacrolimus production and cell growth

Full size image
Table 2 Specific activity of enzymes by parent strain S. tsukubaensis and recombinants in batch cultures
Full size table

As is shown in Fig. 6, the engineered strain HT-aroC showed high tacrolimus production of up to 109.87 ± 8.26 mg/L, due to its enhanced shikimate pathway. Interestingly, a slight decrease of the final biomass concentration could be observed in the HT-aroC mutant. The DHDPS activity of the strain HT-dapA, which overexpresses the lysine synthesis gene dapA, was clearly increased both at 72 and at 144 h (Table 2). There was no significant change of cell growth (9.24 ± 0.29 g/L) in the engineered mutant, but tacrolimus production was nevertheless enhanced by approximately 23.14% (96.37 ± 4.19 mg/L), compared with the parent strain (78.26 ± 6.22 mg/L) (Fig. 6). According to the metabolomics analysis, a combined synergistic effect of enhancing chorismate and lysine biosynthesis simultaneously had also been performed by multiple genes overexpression manipulation (aroC and dapA together). The CHS and DHDPS activities of HT-aroC/dapA were indeed increased compared with the control, reaching, respectively, 3.5- and 1.77-fold higher values at 144 h (Table 2). Moreover, the HT-aroC/dapA mutant was able to produce 128.19 ± 9.57 mg/L tacrolimus, a 1.64-fold increase over the parent strain. Nevertheless, the growth of the HT-aroC/dapA strain was somewhat affected, leading to a lower final biomass concentration (8.12 ± 0.60 g/L) compared with the parent strain (9.24 ± 0.29 g/L).

Discussion

The higher-value immunosuppressant tacrolimus has exhibited a good therapeutic potential in the clinical treatment of a number of diseases. Owing to the complexity of its biosynthesis, several primary metabolic pathways have significant effects on its production. Therefore, considerable efforts have been made to improve tacrolimus production by various exogenous feeding strategies. However, the response mechanisms of intracellular metabolism under different exogenous feeding strategies has to our best knowledge not been reported before. Furthermore, the key-limiting steps among the many precursor pathways of tacrolimus synthesis still remained unknown, which further limited tacrolimus overproduction.

Metabolomics is considered to be more closely connected to the physiology of organisms than the other “omics” approaches (i.e., transcriptomics and proteomics) because the rate of enzymatic reactions is regulated directly by the concentrations of substrates and products in cells [39, 45]. We, therefore, selected four different exogenous feeding strategies targeting different precursor pathways to decipher the intracellular dynamic metabolic responses during tacrolimus production. Coupled with tacrolimus synthesis, WGCNA analysis identified several distinct metabolic modules and hub metabolites that were tightly associated with tacrolimus synthesis in different culture stages. Starting especially from the post-exponential phase (at 48 h after feeding), tacrolimus accumulated rapidly, so that several precursor synthesis pathways (i.e., methylmalonyl-CoA pathway, PPP pathway and shikimate pathway) quickly started to become responsive to tacrolimus accumulation. Hub metabolites form the module M48h-F1 implied that pyruvate, phosphoenolpyruvate and methylmalonate had a high degree of connectivity during this phase. The metabolic flux node between pyruvate and phosphoenolpyruvate played important role in this key branch of tacrolimus precursors. For example, phosphoenolpyruvate is the starting point of DHCHC synthesis [3], and methylmalonate was the direct precursor of methylmalonyl-CoA [3]. Pyruvate can not only be transformed into methylmalonyl-CoA, but can also control the synthesis of acetyl-CoA in the central carbon metabolism, which in turn is an important node for the synthesis of the important tacrolimus precursor allylmalonyl-CoA [19]. Thus, balancing the flux of this key node would be very important during the rapid phase of tacrolimus synthesis. The analysis of the distinct metabolic modules and hub metabolites at 72 h indicated that the shikimate and aspartate pathways might directly control the final tacrolimus production. In fact, DHCHC, which is the starter unit for tacrolimus synthesis, is derived from shikimate pathway [3]. Isoleucine, methionine and lysine (aspartate pathway) can also be transformed into several necessary precursors, such as pipecolate and propionyl-CoA. Therefore, simultaneously enhancing these two metabolic pathways might further improve tacrolimus production during the stationary phase.

In addition to identifying the crucial metabolic pathways, another major goal of this study was to decipher the possible key-limiting steps in said pathways, and further improve tacrolimus production. Therefore, the GSMM model was applied to predict potential targets in PPP, shikimate and aspartate pathways. Several uncovered targets showed highly positive effects on tacrolimus production (Fig. 5). Following further experimental verification, the final tacrolimus production was improved markedly in all the resulting overexpression strains. Specifically, overexpressing the aroC gene from the shikimate pathway resulted in a higher tacrolimus yield, which indicated that this might be a key-limiting step in tacrolimus biosynthesis. In the lysine synthesis pathway, overexpressing the gene dapA also enhanced tacrolimus production effectively (up to 96.37 ± 4.19 mg/L). Additionally, simultaneously enhancing the flux in the shikimate and lysine pathways (HT-aroC/dapA) resulted in the highest tacrolimus yield (128.19 ± 9.57 mg/L) of all the engineered strains, which was consistent with the metabolomic analysis. These results also showed that the strategy used in this work was effective for tacrolimus overproduction. Nevertheless, other detailed information of intracellular metabolism are still needed to improve and perfect the GSMM of S. tsukubaensis in the further, including metabolic regulation, 13C metabolic analysis, thermodynamic and kinetic information. The addition of these information will be very important for further improving simulation accuracy, and by extension tacrolimus production in the further.

In conclusion, the GC–MS- and LC–MS/MS-based metabolomic analysis and a WGCNA network analysis of S. tsukubaensis under shikimate, citrate, malonate and lysine feeding regimens clearly showed that intracellular responses in the tacrolimus biosynthesis pathway were different, as evidenced by the strengthening of different precursor synthesis pathways under said regimens. Further, the PPP, shikimate and aspartate pathways were found to play crucial roles in tacrolimus biosynthesis. Based on the GSMM predictions, several potential targets were identified as the key-limiting steps in these metabolic pathways. Overexpression of some of the target genes further corroborated these predictions, and the final tacrolimus production in the resulting engineered strain HT-aroC/dapA was improved approximately 1.64-fold compared with the control. These findings, thus, lay the foundation for further improvements of tacrolimus production in the future.