Escherichia coli BL21 adapts its central carbon metabolism to recombinant protein production and oxygen limitation
Divyata Vilas Rane, Karen Lund Haaland, Åse Refsnes, Hege Skuggedal, Stinus Reklev Øverbø, Per Bruheim, Laura García-Calvo
TL;DR
This study explores how E. coli BL21 adapts its metabolism under stress from protein production and low oxygen, revealing how it maintains stability.
Contribution
The study reveals how E. coli BL21 adapts its central metabolism to both recombinant protein production and oxygen limitation.
Findings
Strains with higher plasmid copy numbers showed more pronounced growth retardation and metabolic changes.
Oxygen limitation caused lower metabolic activity but maintained energy and redox balance without affecting mCherry production.
Both strains showed similar responses to oxygen limitation, with significant central metabolite pool adaptations.
Abstract
High-yielding recombinant protein expression systems often face challenges due to the metabolic burden caused by the competition for cellular resources, resulting in reduced growth and, hence, limiting their industrial applicability. Furthermore, industrial recombinant protein production is also affected by the occurrence of oxygen gradients, which is a prevalent issue in large-scale bioreactors. These gradients create a heterogeneous environment in the bioreactor, which affects cell growth and metabolism, having severe consequences on the process performance. Both these factors alter cellular physiology and metabolism, thereby affecting recombinant protein yields. Understanding metabolic adaptations to these stress conditions is crucial for uncovering the underlying cellular mechanisms, which can direct further optimization of the recombinant strains. In this study, we aimed to explore…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5- —NTNU Norwegian University of Science and Technology (incl St. Olavs Hospital - Trondheim University Hospital)
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsMicrobial Metabolic Engineering and Bioproduction · Viral Infectious Diseases and Gene Expression in Insects · Protein purification and stability
Background
Recombinant protein production (RPP) has played a critical role in industrial biotechnology by enabling the cost-effective and scalable production of complex proteins, including therapeutic agents like insulin, monoclonal antibodies, and vaccines, as well as industrial enzymes used in detergents, food processing, and biofuel production. Innovations in synthetic biology, metabolic engineering, implementation of CRISPR-Cas9 technology, and AI-driven process optimization have further enhanced the efficiency of recombinant processes, improving protein yield and scalability [1, 2].
While designing the recombinant processes, the choice of host expression system plays a critical role, determined by the complexity of recombinant proteins, desired yield and scalability, cost-effectiveness, and downstream processing requirements. Among many expression hosts, Escherichia coli is the preferred choice for industrial recombinant protein production due to its rapid growth, ease of genetic manipulation, well-characterized genome, and cost-effective cultivation [3]. Furthermore, E. coli systems benefit from a vast toolbox of plasmids, promoters, and regulatory elements that allow precise control of recombinant protein expression [4]. A variety of protein expression strains developed for E. coli, including commonly used protease-free strains BL21 (DE3) and expression platforms, such as the T7 promoter system, enable robust and high-yield production of recombinant proteins with fine-tuned control of expression [5, 6]. Since the first commercial success in the bioproduction of human insulin, E. coli has been widely utilized for the industrial production of many recombinant proteins, including hormones, enzymes, and antibodies [7].
Despite the advantages, recombinant protein production in E. coli is influenced by various factors that impact the yield and functionality, such as the nature of the expression systems [8], induction conditions [9], cultivation conditions [10, 11] and codon usage bias [12]. Choice of key regulatory elements which influence gene expression, such as replicon and promoter system, is essential to ensure efficient and scalable recombinant protein production meeting industrial demands [13, 14]. However, high-yield expression systems are often accompanied by significant levels of metabolic burden caused by an additional demand for cellular resources for recombinant plasmid maintenance and expression [15]. This metabolic burden is characterized by physiological alterations in the host system, including reduced growth [16] and downregulation of essential metabolic pathways (reviewed in [17, 18]). The extent of the metabolic burden is determined by the properties of the expression vector, including promoter strength, plasmid copy number (PCN), plasmid size, etc. (reviewed in [19]). Increasing plasmid number has a linear relationship with the consequences of the metabolic burden. A careful optimization of recombinant protein expression is required to achieve a balance between desired levels of protein production and subsequent metabolic burden [20].
In addition to the metabolic burden, industrial production of recombinant proteins can be influenced by the scale-up conditions. Successful scale-up relies on understanding key bioreactor parameters, including mixing, oxygen transfer, heat transfer, and shear forces, which are vital for efficient cell growth and protein production [21]. A key challenge in scaling up is the emergence of concentration and mass transfer gradients at high cell densities, particularly near feed additions (for a fed-batch process), sparger, or where controlling agents like acids or bases are introduced. These spatial gradients, involving dissolved gases, pH, substrate concentrations, and shear rates, arise due to insufficient mixing, a common limitation in traditional scale-up methods and large-scale bioreactor designs [22]. Cells in such heterogeneous environments are exposed to fluctuating conditions as they move through different zones of the bioreactor, leading to variations in metabolism, reduced product yields, and compromised product quality [23].
Dissolved oxygen (DO) gradients are a major issue in industrial bioreactors, caused by slower mixing times than the time required for cellular uptake. Other factors also contribute to the DO gradients, including bioreactor design, pressure difference across the reactor, and variable oxygen consumption by cells, often caused by zones with higher respiration levels in substrate-rich regions [24]. Such gradients affect the metabolism of facultative anaerobic host systems like E. coli, resulting in a rapid switch to anaerobic fermentative pathways, which causes the accumulation of mixed acids as byproducts, such as pyruvate and acetate, at the expense of biomass and/or product production [25]. This is detrimental to process performance as lower growth rates can reduce the overall volumetric productivity of recombinant proteins.
Limited oxygen availability is reported to influence recombinant protein production and reduce the overall yield [25, 26]. The metabolic adaptations to sustain oxygen fluctuations, mainly involving the changes in central carbon metabolism and fluxes, can result in alterations in amino acid synthesis, hence affecting the host’s capacity and efficiency of recombinant protein production [27–29]. Plasmid loss during non-homogeneous oxygen supply has been a commonly observed phenomenon for different host species, including E. coli [30] and Saccharomyces cerevisiae [31]. Though microaerobic conditions are linked to reduced RPP, several studies report an increased production of recombinant proteins upon oxygen limitation [32, 33]. Therefore, it is difficult to draw a general conclusion to explain the effect of oxygen limitation on recombinant protein production.
Both metabolic burden and microaerobic conditions are highly prevalent in industrial settings, also affecting the process performance, but these have not been extensively studied in the context of industrial cultivations. With the development of omics technologies, high-throughput evaluation of metabolic networks is possible, enabling the assessment of metabolic stresses. Several studies have separately addressed the consequences of metabolic burden [34–36], and microaerobic conditions [37, 38] by the modulation of host metabolic pathways. In this study, we address both of these metabolic stresses by deep phenotyping approaches, investigating the implications on central carbon metabolism. We study the effect of oxygen-limiting conditions on the recombinant protein-producing E. coli strains with low and medium plasmid copy numbers, hence also influenced by the metabolic burden caused by recombinant protein production. We implement several targeted approaches to investigate central carbon pathways, along with the investigation of the energy state and redox state of the cells [39]. These recombinant strains and the metabolic burden associated with these strains have been discussed in our previous publication [40]. Here, we extend this study to also investigate the effect of microaerobic conditions on two strains with varying plasmid copy numbers.
Materials and methods
Microbial strains and cultivation conditions
Escherichia coli wild-type strain BL21 (New England Biolabs, Irving, MA, United States), with genotype fhuA2 [lon] ompT gal [dcm] ΔhsdS, was used in this study as a host for RPP. The recombinant strains, A2-mCh and A3-mCh, were obtained by transforming competent cells of E. coli BL21 with the pVB expression vector with different copy numbers (approx. size 8.9 kbp) and ampicillin selection marker. These recombinant strains were provided by Vectron Biosolutions AS (Trondheim, Norway). The construct and characteristics of the strains are discussed in our previous study and with an estimated copy number of 20 and 40 for the A2mCh and A3mCh strains, respectively [40]. These strains were maintained as 16% (v/v) glycerol stocks (24,387.292, VWR) and stored at −80 °C.
The inoculum for bioreactor experiments was obtained from preliminary cultures prepared before the bioreactor cultivation. The primary precultures were prepared by inoculating 75 µL of glycerol stocks into 50 mL rich LB medium, as explained previously [40], and incubated at 30 °C and 200 rpm for ~9 h. For secondary precultures, 200 µL of the primary preculture was transferred into 100 mL of secondary medium in 500 mL baffled flasks. The secondary medium consisted of 100 mL 10× mineral medium salt solution, 0.25 g MgSO_4_‧7H_2_O (M5921, Sigma-Aldrich), 10 g glucose (101176K, VWR), 2 mL trace element solution, and 2 mL of 50 mg L^−1^ cobalt stock solution (CoCl_2_,6H_2_O, C8661, Sigma-Aldrich) in 1 L of Milli-Q (MQ; 18.2 MΩ cm) H_2_O. The composition of the mineral medium salt solution and trace element solution is explained elsewhere [41]. These cultures were incubated at 37 °C and 200 rpm for 13 ± 1 h. 100 μg mL^−1^ ampicillin (A0839.0100, Panreac AppliChem, Darmstadt, Germany) was added to both the precultures to maintain selection pressure.
Bioreactor cultivations were carried out in autoclavable bench-top 1 L glass stirred-tank bioreactors (Applikon Biotechnology, Delft, Netherlands) controlled by my-Control units (Z310210011, Applikon Biotechnology). The reactors were equipped with pH (AppliSens pH sensors, Z001023551, Applikon Biotechnology) and dissolved oxygen (DO, AppliSens Low Drift DO_2_-sensors, Z010023525, Applikon Biotechnology) sensors for online measurements. A salt solution containing 5 g L^−1^ NH_4_Cl (A9434, Sigma-Aldrich), 2 g L^−1^ K_2_HPO_4_ (P8281, Sigma-Aldrich), and 0.5 g L^−1^ NaCl prepared in 0.7 L MQ H_2_O was autoclaved in bioreactor before addition of 0.74 g L^−1^ MgSO_4_‧7H_2_O, 2 mL L^−1^ trace element solution, 2 mL L^−1^ cobalt solution, and glucose to a final concentration of 20 g L^−1^ in aseptic conditions. The reactor was later supplemented with 100 μg mL^−1^ ampicillin, and autoclaved Milli-Q H_2_O was added with a sterile pipette to make up the volume to 0.9 L. 200 µL of antifoam (ADEKA NOL LG-109, Adeka Europe GmbH, 10% w/w) was added before inoculation to avoid foaming during the run.
The bioreactors were inoculated with the secondary preculture at the late exponential growth phase to achieve an initial OD_600_ of ~0.1. Bioreactor cultivations were carried out at 30 °C and pH 7.0, controlled automatically by the addition of 4 M NaOH (28,244.295, VWR). Air was sparged at a flow rate of 600 mL/min, and continuous measurements of airflow and exhaust gases were taken using an in-line mass spectrometer (Prima BT Bench Top Process Mass Spectrometer, ThermoFisher Scientific, Waltham, MA, United States). Aerobic growth was obtained by keeping DO levels 40% and above with an agitation cascade control ranging from 200 to 800 rpm. For the oxygen limitation phase, the DO-agitation cascade was turned off, and constant agitation at 400 rpm was carried out throughout the cultivation. The processes were monitored and controlled using BioXpert® 2 software (Applikon Biotechnology).
For recombinant protein production, the cultures were induced by the addition of 1 mM m-toluate when the biomass reached OD_600_ = (0.6 ± 0.1). After the onset of protein production, samples were taken at regular intervals to determine growth by measuring OD_600_ (V-1200 Spectrophotometer, VWR) and m-Cherry production by fluorescence using a microplate reader (Tecan Spark® 20M, Männedorf, Switzerland). The measurement parameters were obtained as mentioned previously [40].
During the exponential growth phase in both aerobic and microaerobic conditions, samples were collected for exometabolome and endometabolome analyses in at least technical triplicate. Samples were also collected for cell dry weight (CDW) analysis by fast-filtration through pre-dried and pre-weighed 0.45 μm, hydrophilic PVDF membrane disc filters (47 mm diameter, HVLP04700, Merck Millipore, Darmstadt, Germany). The filters were then dried at 110 °C until a constant weight was achieved (~30 h).
Quantification of extracellular metabolites
Samples for extracellular analysis were centrifuged at 4500 rcf for 5 min at 4 °C. Supernatants were separated from pellets in clean microtubes. Both pellets and supernatants were snap-frozen in liquid nitrogen (LN_2_) and stored at −80 °C until analysis. On the day of analysis, samples were thawed to room temperature and filtered using syringe filters (0.2 μm, 13 mm diameter, 514-0068, VWR). For quantification of extracellular organic acids and glucose in fermentation supernatants and media samples, high-performance liquid chromatography (HPLC) (Agilent Technologies, Santa Clara, CA, USA) with a refractive index (RI) and a UV/Vis detector was employed. The column details and method parameters were as mentioned by Rane et al. [42]. The external standard (ESTD) mixture was prepared from commercial standards of glucose (D-(+)-Glucose, G8270, Sigma-Aldrich), organic acids (namely acetic acid (1.00063.1011, Supelco, Bellefonte, PA, USA), lactic acid (L1500, Sigma-Aldrich), succinic acid (S3674, Sigma-Aldrich), fumaric acid (47,910, Sigma-Aldrich), formic acid (84,865.260, VWR), pyruvic acid (P8574, Sigma-Aldrich)), and ethanol (20,821.310, VWR). The ESTD mix was further diluted to obtain a calibration curve, used for interpolation of concentrations in the samples.
Quantification of mCherry production by western blot
mCherry requires molecular oxygen to mature into a fluorescent protein. Therefore, fluorescence could not be used to quantify mCherry production after oxygen limitation, and Western blot was performed to quantify mCherry production for samples taken during oxygen-limited conditions. Pre-weighed cell pellets were lysed by incubating them with lysis buffer (5 µL mg^−1^ biomass wet weight, CelLytic B 2× concentrate, B7310, Sigma-Aldrich) supplemented with 50 units mL^−1^ benzonase nuclease (E1014, Sigma-Aldrich) for 30 min at room temperature with shaking at 100 rpm. The soluble and insoluble fractions were then separated by centrifugation at 10,000 rcf for 5 min, and the insoluble fractions were then resuspended in an SDS-running buffer (#1,610,772, 1× Tris/Glycine/SDS, Bio-Rad, California, USA). Extracts were then further diluted based on the protein concentration, and loading dye (1×) was added (#1,610,747, 4× Laemmli Sample Buffer, Bio-Rad). Extracts were subsequently boiled at 95 °C for 5 min and loaded onto a 26-well pre-cast SDS-gel (5,671,045, 12% Criterion™ TGX™ Precast Midi Protein Gel, 26 well, Bio-Rad) along with molecular weight marker Precision Plus Protein™ Dual Color (#1,610,374, Bio-Rad). Gel electrophoresis was run at 200 V for 30–45 min, after which the gels were carefully removed from cassettes and rinsed with MQ-H_2_O. After this step, gels were either stained with Coomassie Brilliant Blue (#1,610,787, R-250, Bio-Rad, 1,610,436) for visualization on ChemiDoc™ (ChemiDoc XRS + system, Bio-Rad, 1,708,265) or taken further to western blotting for quantification of mCherry in cell extracts.
For western blot, gels after electrophoresis were transferred to membranes (Trans-Blot Turbo Midi 0.2 µm PVDF Transfer Packs, #1,704,157, Bio-Rad) for 7 min using Trans-Blot Turbo® transfer System (Bio-Rad). Further steps, including blocking, antibody binding, and intermediate washes, were performed using the iBind™ Flex Western Device (Invitrogen, Thermo Fisher Scientific, MA, USA) with iBind™ Flex Cards (Invitrogen). Primary antibody (1:1000 solution of Rabbit anti-mCherry, ab167453, Abcam Limited, Cambridge, UK), secondary antibody (1:100,000 solution of Anti-IgG F(c) Goat Polyclonal Antibody, IMMRIR2226, ImmunoReagents, NC, USA), and wash solution (1× iBind™ Flex Solution SLF2020, Invitrogen) were added sequentially as per the instructions on the manual. After the run was complete, the membrane was washed with MQ-H_2_O and revealed using 3,3′,5,5′-tetramethylbenzidine (TMB, T0565, Sigma-Aldrich). The membrane was washed again with MQ-H_2_O to remove excess TMB and imaged using ChemiDoc™. mCherry quantification was performed using a calibration curve prepared from a commercial mCherry standard (4993, BioVision, CA, USA), ranging from 0.1 to 8 ng µL^−1^.
LC–MS/MS-based quantification of intracellular metabolites
Quantification of central carbon metabolites and phosphorylated metabolites
Ion chromatography-tandem mass spectrometry (IC-MS/MS) analysis was carried out for the intermediates in glycolysis, tricarboxylic acid (TCA) cycle, and pentose phosphate pathway (PPP), as well as phosphorylated metabolites including nucleoside phosphates and deoxynucleoside phosphates. The samples were collected by a fast-filtration protocol and processed for metabolite extraction by solvent extraction and lyophilization as detailed previously [40, 41].
The lyophilized extracts were redissolved in 0.5 mL MS-grade H_2_O (83,645.32, VWR) and spin-filtered using 10 kDa cutoff spin filters (516-0230P, VWR) at 20,817 rcf, 4 °C for 10 min. Samples were analyzed by an IC system (Dionex Integrion HPLC, Thermo Fisher Scientific) coupled with a Xevo TQ-XS triple quadrupole mass spectrometer (Waters, Milford, MA, USA). Samples were kept at 10 °C in a thermostatted autosampler (Dionex AS-AP, Thermo Fisher Scientific). The separation was carried out on Dionex™ IonPac™ AS11-HC IC Column (Thermo Fisher Scientific) with (2 mm × 250 mm) anion exchange column (052961, Thermo Fisher Scientific) and (2 mm × 50 mm) guard column (052963, Thermo Fisher Scientific) with injection volume 25 μL in a thermostatted column compartment kept at 40 °C. The suppressor was set to 30 mA, and electrochemical regeneration of the suppressor was carried out by the external flow of MQ-H_2_O at 0.2 ml/min. The KOH gradient was modified from [43, 44]. The following KOH gradient was used: 0–0.1 min: 15 mM, 0.1–11 min: 15–30 mM, 11–18 min: 30–100 mM, 18–28 min: 100 mM, 28–28.5 min: 100–15 mM, 33 min: stop. The MS settings and MRM transitions were the same as reported elsewhere [43, 44], while retention time (RT) windows for each analyte were adjusted to ±2 min of the expected RT. For isotope dilution (ID) correction, samples and calibration standards were spiked with ^13^C-labeled cell extract from Saccharomyces cerevisiae as described by Stafsnes et al. [44].
Quantification of amino acids
Reconstituted lyophilized extracts (described in Sect. 2.4.1) (left over from a previous paper?) were further taken for amino acid analysis by reverse phase chromatography-tandem mass spectrometry (RP MS/MS). ID correction, derivatization, and sample analysis were performed as described previously [39].
Quantification of pyridine nucleotides and their precursors
Snap-frozen cell pellets (as mentioned in Sect. 2.2) (maybe have to add a complete sentence describing the sampling?) were used for the quantification of intracellular pyridine nucleotides, including reduced and oxidized forms of NAD^+^ (Nicotinamide adenine dinucleotide) and NADP^+^ (Nicotinamide adenine dinucleotide phosphate), as well as metabolites involved in NAD^+^ biosynthesis as described previously [42]. The analysis was performed by zwitterionic hydrophilic interaction liquid chromatography (zic-HILIC)–tandem mass spectrometry.
Data processing, statistical analysis, and visualization
Cultivation data obtained from bioreactor cultivations were plotted using Matplotlib v. 3.3.4 [45]. The specific growth rate (μ, h⁻^1^) was determined as the slope of a semi-logarithmic plot of OD_600_ against time. The specific production rate (q, g g⁻^1^ CDW h⁻^1^) for extracellular acids and mCherry was calculated by dividing the respective volumetric production rate (r, g L⁻^1^ h⁻^1^) by the average cell dry weight (CDW, g L⁻^1^) between two sampling points. Average biomass calculated using the exponential function (X(t) = X_0_e^μt^, where X(t) is biomass at time t and X_0_ is initial biomass) was used for calculations performed for the exponential growth phase (before oxygen limitation). The yield (g g⁻^1^ glucose) was determined by dividing the amount of biomass, acids, or mCherry produced (g L⁻^1^) by the glucose consumed (g L⁻^1^) between the time points. The glucose uptake rate (g g⁻^1^ CDW h⁻^1^) was calculated by dividing the glucose consumption rate (g L⁻^1^ h⁻^1^) by the biomass concentration (g CDW L⁻^1^). The specific CO₂ production rate (pCO₂, mmol g⁻^1^ CDW h⁻^1^) was calculated by dividing the total amount of CO₂ released (mmol L⁻^1^) during a given time interval by the average cell dry weight (CDW, g L⁻^1^).
Mass spectrometric data were analyzed and processed using the TargetLynx application within the MassLynx 4 software (Waters). Absolute intracellular concentrations (nmol g^−1^ CDW) of metabolites were estimated by interpolation from calibration curves obtained from analytical grade standards (Santa-Cruz Biotechnology, Dallas, Texas, United States; Sigma-Aldrich) and also corrected for isotope dilution as described by Røst et al. [39]. Potential outliers within technical replicates were identified and removed using two-tailed Dixon’s Q test with a 95% confidence interval [46]. The open-source web platform MetaboAnalyst 6.0 [47] was used for statistical analysis and visualization of data by principal component analysis (PCA). Before PCA, missing values were replaced with 1/5 of the lowest measured value for each metabolite. The data were then mean-centered and autoscaled by dividing by the standard deviation within each group. Statistical significance was determined by two-tailed t-test assuming equal variances for pairwise comparisons and one-way ANOVA with Tukey’s multiple comparison tests. p-value cutoff of 0.05 was used to determine the statistical significance within different groups and was adjusted for the false discovery rate (FDR). Statistical calculations and plots were made using Graphpad Prism 10. Log_2_ fold changes in average metabolite concentrations were mapped onto central carbon metabolic pathways using the Omix editor and metabolic network modeling tool [48]. Adenylate energy charge ratio was calculated during each cultivation condition by the formula (ATP + 0.5 ADP)/(AMP + ADP + ATP) [49].
Results
The two E. coli BL21-based strains, A2-mCh and A3-mCh, carrying XylS/Pm vectors with low and medium plasmid copy numbers (approximately 20 and 40, respectively), as explained elsewhere [40, 50–53], were cultivated at constant pH in bench-top bioreactors (experimental design in Fig. 1) and mCherry production was induced at an optical density (OD_600_) of ~0.6 by adding the inducer m-toluate. For both strains, cultures were grown aerobically until OD_600_ ~ 2.0, after which oxygen limitation was introduced by lowering the stirrer speed in the bioreactors while keeping the air flow rate constant. Continuous monitoring of dissolved oxygen (DO) confirmed the limitation. A minimum of two biological replicates (i.e., independent bioreactor cultivations) were run for all conditions, and one representative replica is presented in detail in this report, while cultivation plots of the other replicas are included as supplementary data (Additional file 1, Figure S1). Samples were taken for CDW, and for analysis of extracellular byproducts and intracellular metabolites during the cultivation. Sampling points differ for the two strains, with 4 sampling points (T1–T4) for A2-mCh and 2 sampling points (T1–T2) for A3-mCh. T1 for both strains represents aerobic growth. During cultivations with oxygen limitation, for A2-mCh, sampling points T2–T4 were taken during oxygen-limited conditions. As the medium PCN strain, A3-mCh, exhibited a strong effect of metabolic burden and a slower growth rate, it was sampled only once (T2) during oxygen-limited conditions for endometabolome analysis.Fig. 1. Cultivation scheme for E. coli BL21 recombinant strains, A2-mCh (Low PCN) and A3-mCh (Medium PCN), followed by sampling and LC–MS/MS analysis (Figure created using https://BioRender.com)
Growth parameters of two recombinant strains during batch cultivations
Low PCN strain, A2-mCh
The induction of mCherry production did not have an immediate effect on the growth rate of A2-mCh, but a 21% reduction (from 0.42 to 0.33 h^−1^) was observed ~2 h after induction at the same time as the onset of mCherry detection in the culture (Fig. 2a). Furthermore, when oxygen-limited conditions were introduced after the recombinant protein induction, the oxygen limitation was found to have an immediate and much stronger negative impact than the induction and resulted in around two-third decrease (0.42 to 0.15 h^−1^) in the growth rate (Fig. 2b). However, mCherry production was maintained regarding both specific productivity (g mCherry g^−1^ CDW h^−1^) and yield (g mCherry g^−1^ glucose) during the oxygen-limited conditions (Table 1). As expected, the oxygen limitation caused an intracellular metabolic imbalance resulting in high acetate production (15% of glucose was excreted as acetate) and a large decrease in the cell mass yield (Table 1).Fig. 2. Growth profiles of E. coli BL21 in benchtop bioreactors, including cultivation data and sampling points. Each figure represents one of the biological replicates, and specific growth rate, µ (h^−1^) values are indicated for a particular period with ±SD (standard deviation calculated between µ values of a minimum of two biological replicates). The sampling points represent sampling for exometabolome and endometabolome analysis. mCherry production is indicated by fluorescence and western blot (mg L^−1^). a Cultivation of A2-mCh (low PCN strain) with recombinant protein induction and no oxygen limitation (Control cultivation, A2-mCh_Aerobic_ind); b Cultivation of A2-mCh with recombinant protein induction and oxygen limitation (A2-mCh_O_2_-Lim); c Cultivation of A3-mCh (Medium PCN strain) with recombinant protein induction and no oxygen limitation (A3-mCh_Aerobic_ind); d Cultivation of A3-mCh with recombinant protein induction and oxygen limitation (A3-mCh_O_2_-Lim). : Western blots corresponding to Fig. 2a, b, d are included in Additional file 1, Figure S2. For A3-mCh_O_2_-Lim cultivation (Fig. 2d), the sampling for western blot quantification was performed at five sampling points (t1, t2, t3, t4, and t5)Table 1. Cultivation parameters (specific growth rate, glucose uptake rate, specific CO_2_ production rate, and biomass yield), and production of extracellular acetic acid and mCherry (as determined from specific production rates and yield) for the two E. coli BL21 recombinant strains, A2-mCh and A3-mCh, during different cultivation conditionsStrain/conditionA2-mChA3-mCh**Aerobic_indO_2_-limNo_indAerobic_ indO_2_-limSpecific growth rate (h⁻^1^)0.42 ± 0.040.15 ± 0.010.37 ± 0.040.08 ± 0.030.07 ± 0.03Glucose uptake rate (g g⁻^1^ CDW h⁻^1^)0.73 ± 0.060.67 ± 0.010.87 ± 0.10.29 ± 0.010.25 ± 0.07CO_2_ production rate (mmol g⁻^1^ CDW h^−1^)7.8 ± 0.164.60 ± 0.357.3 ± 0.653.3 ± 0.281.78 ± 0.12Biomass yield (g CDW g^−1^ glucose)0.45 ± 0.110.30 ± 0.110.45 ± 0.080.42 ± 0.040.24 ± 0.07Exometabolome and mCherry production Specific production rate Acetic acid (g g^−1^ CDW h^−1^)0.02 ± 0.00.09 ± 0.010.02 ± 0.00.02 ± 0.010.06 ± 0.02 mCherry (mCh) (g g^−1^ CDW h^−1^)0.02 ± 0.00.02 ± 0.0–0.02 ± 0.010.05 ± 0.03 Yield Acetic acid (g g^−1^ glucose)0.03 ± 0.00.15 ± 0.020.02 ± 0.00.06 ± 0.020.25 ± 0.03 mCherry (mCh) (g g^−1^ glucose)0.02 ± 0.00.03 ± 0.0–0.09 ± 0.030.18 ± 0.07Negligible production of formic acid and lactic acid was observed for both strains under O₂-limited conditionsCumulative values from sampling points T1–T4 were calculated for A2-mCh_Aerobic_ind (Fig. 2a) and A2-mCh_O_2_-Lim (Fig. 2b)**Cumulative values from T1–T2 were calculated for A3-mCh_No_ind (Figure S1e, Additional file 1), A3-mCh_Aerobic_ind (Fig. 2c), and A3-mCh_O_2_-Lim (Fig. 2d)The values for growth parameters are represented as mean ± SD calculated from at least two biological replicates. For exo-metabolome and mCherry production, specific production rates and yield are calculated based on mean ± SD, based on technical replicates obtained from the cultivations shown in Fig. 2Glucose uptake rate, CO₂ production rate, specific production rate, and yield were calculated as described in “Materials and methods”
Medium PCN strain, A3-mCh
In contrast to A2-mCh, induction of mCherry production in the medium PCN strain, A3-mCh, had a much stronger negative impact, as over an 80% reduction (from µ = 0.35 h^−1^ to 0.06 h^−1^) in the growth rate was observed after the addition of the inducer (Fig. 2c). This can be compared with a cultivation profile without recombinant protein induction (Additional file 1, Figure S1e), where such reduction in growth rate was not seen. An unexpected secondary growth phase, approximately 10 h after induction, was observed late in the cultivation with recombinant protein induction and no oxygen limitation (Fig. 2c). This coincides with a sharp decrease in mCherry production. Plasmid loss is the most likely cause of this, but was not further investigated since it was beyond the scope of the study.
The induced culture continued to grow at the same rate upon introducing the oxygen-limiting conditions (Fig. 2d). Interestingly, mCherry production increased compared to the preceding induced and aerobic conditions (Fig. 2c). Specific productivity slightly increased from 0.02 to 0.05 g mCh g^−1^ CDW h^−1^, and the yield on glucose basis doubled to 0.18 g mCh g^−1^ glucose (Table 1). This is almost twenty percent of the glucose consumption in this O_2_-limited phase and six times higher than the A2-mCh strain, where only a slight increase in yield was observed in the oxygen-limited phase (Table 1). A comparatively higher yield of acetic acid per g of glucose was also observed for A3-mCh in the oxygen-limited phase (Table 1).
The growth and cultivation data presented in Fig. 2 and Table 1 indicate quite different intracellular metabolic adaptations to the induction of recombinant protein production and to oxygen limitation in the two tested strains. A2-mCh comparatively showed much higher glucose consumption and CO_2_ evolution during oxygen-limited conditions. Therefore, intracellular metabolic profiling of these strains is important to characterize the metabolic adaptations to the different plasmid copy numbers and oxygen limitation.
Mass spectrometry-based metabolite profiling
Three LC–MS methods were used to quantify the primary metabolite pools: an IC-MS/MS method for glycolytic, PPP, TCA cycle intermediates, and nucleoside phosphates; an RP LC–MS/MS method using phenyl isothiocyanate derivatization for amino acids; and a Zic-HILIC LC–MS/MS method for NAD^+^ and NAD^+^ -associated metabolites. A2-mCh with its four sampling points (one, T1, before, and three, T2-T4, during oxygen limitation) (Fig. 2a, b) is discussed first, followed by the A3-mCh cultivations with two sampling points (T1 before and T2 after oxygen limitation) (Fig. 2c, d). Since the MS methods are quantitative, the results can be presented as absolute concentrations with units nmol g^−1^ CDW.
Low PCN strain, A2-mCh
Some metabolites were found to be high abundant in A2-mCh, such as glucose-6-phosphate (G6P) and fructose-1,6-biphosphate (F1,6BP) in the glycolytic pathway, Succinate (Suc) in TCA, adenosine triphosphate (ATP) in the nucleoside phosphate pool, NAD^+^ and glutamic acid (Glu) among the pyridine metabolites and amino acids, respectively (Fig. 3a). The heatmap time series of the absolute concentration does not display the needed resolution to indicate changes in concentration as the cultures run into oxygen limitation (right vs. left panel in Fig. 3a), but plotting the log_2_ ratio of oxygen limitation (sampling points T2–4) vs. before limitation (T1) clarifies the quite large increases in the concentration of some amino acids (valine (Val), glutamic acid (Glu), glutamine (Gln), and asparagine (Asn)) and organic acids in TCA when oxygen limitation is enforced, especially at the last time point T4 (Fig. 3b).Fig. 3a Heatmap representing endometabolome absolute concentrations (nmol g^−1^ CDW) in E. coli BL21 low PCN strain, A2-mCh, during cultivation without oxygen limitation (Aerobic_ind, Fig. 2a) and oxygen-limited (O_2_-Lim, Fig. 2b) conditions; b Heatmap representing log_2_ fold change in metabolite concentrations at sampling times T2, T3, and T4 (in oxygen-limited phase) vs. sampling point T1 (in aerobic phase) during oxygen-limited (A2-mCh_O_2_-Lim, Fig. 2b) cultivation of E. coli BL21 low PCN strain, A2-mCh. Metabolite abbreviations are listed in Additional file 2, Table S1. Significance levels calculated using two-tailed T-tests assuming equal variances are marked (*p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001 and ****p ≤ 0.0001)
As also seen from the PCA plot (Additional file 1, Figure S3) representing four sampling points in oxygen-limited cultivation, data points for T4 are separated from those of the other sampling points in the oxygen limitation phase, T2 and T3, along the PC2 axis, indicating a strong effect of oxygen limitation at T4. A comparison at the last time point T4 between the oxygen-limited cultivation vs. the non-limited control cultivation also confirms a strong accumulation of TCA intermediates and amino acids as a response to the oxygen limitation (Fig. 4).Fig. 4. Schematic overview of the central carbon metabolome representing log_2_ fold change in metabolite concentrations at sampling times T4 during ‘O_2_-Lim’ cultivation (Fig. 2b) vs. T4 in ‘Aerobic_ind’ cultivation (Fig. 2a) for E. coli BL21 low PCN strain, A2-mCh. The plot was prepared using the Omix Visualization tool [48]. The figure represents one of the biological replicates for each cultivation. Metabolites in gray were not detected by the analytical method or were below LOQ. Metabolite abbreviations are listed in Additional file 2, Table S1
Medium PCN strain, A3-mCh
The intracellular metabolic concentrations for the medium PCN, A3-mCh strain were in the same ranges as observed for A2-mCh (Additional file 1, Figure S4). The reduction in the growth and glucose consumption in A3-mCh after recombinant protein induction was accompanied by an overall decrease in the primary metabolite pools (Fig. 5a). A moderate decrease was observed for the glycolytic, PPP, and nucleoside phosphate pools at the induced lower growth rate, while the adjustments in TCA were interesting, with the concentrations of Suc increasing and Fum and Mal decreasing (Fig. 5a). Similar to A2-mCh (Fig. 4), an accumulation of intermediates in the TCA cycle (mainly Cit, and α-KG), and amino acids (mainly Val, Gln, Asn, and Met), was observed in A3-mCh upon oxygen limitation (Fig. 5b). In addition, intermediates from PPP (X5P, R5P, RL5P, and 6PG) were also accumulated during the oxygen-limited phase.Fig. 5. Schematic overview of the central carbon metabolome representing log_2_ fold changes in metabolite concentrations of E. coli BL21 medium PCN strain, A3-mCh. The plots were prepared using the Omix Visualization tool [48]. a T1 in cultivation with induction (A3-mCh_Aerobic_ind, Fig. 2c) vs. T1 in cultivation without induction (A3-mCh_No_ind, cultivation carried out without the addition of inducer (Additional File 1, Figure S1e); b T2 (representing O_2_-lim phase) vs. T1 (representing aerobic phase) in Cultivation with O_2_-limitation (A3-mCh_O_2_-Lim, Fig. 2d). The figures represent one of the biological replicates for each cultivation. Metabolites in gray were not detected by the analytical method or were below LOQ. Metabolite abbreviations are listed in Additional file 2, Table S1. Volcano plots with statistically significant features are provided as Additional file 1, Figure S5
A3-mCh, after RPP induction, was found to contain lower pools of most metabolites [except PRPP and some amino acids (Trp and Gln)] than A2-mCh when induced for RPP (Additional file 1, Figure S6a). However, in the O_2_-limited state, it showed higher pools of intermediates from PPP and lower glycolytic pathway, in comparison to A2-mCh (Additional file 1, Figure S6b).
Energy charge and redox ratios
The quantification of adenylate energy charge and redox ratio is important to get insights into the cellular metabolic state and metabolic balance in stress conditions such as metabolic burden and microaerobic environments. Since the endometabolome concentrations were determined quantitatively, it was possible to determine these ratios. The energy charge was maintained at the same level for both strains and all conditions (Additional file 3, Table S2), which indicates that E. coli has robust mechanisms to counteract changes in the growth conditions and intracellular metabolic burden due to forced heterologous protein production.
In A2-mCh, the NADH/NAD⁺ ratio remained stable regardless of oxygen limitation, while the NADPH/NADP⁺ ratio showed a slight decrease in the late oxygen-limitation phase (Additional file 3, Table S2, Figure S7a). Similarly, in A3-mCh cultivation, the NADH/NAD⁺ ratio remained unchanged after induction and oxygen limitation. However, the NADPH/NADP⁺ ratio increased upon RPP induction and showed a decrease under oxygen limitation (Additional file 3, Table S2, Figure S7b).
Discussion
In this study, two different stress conditions were introduced in recombinant strains of E. coli: (i) induction of recombinant protein production, and (ii) limiting oxygen availability in the bioreactor, and both the strains, a low PCN A2-mCh and medium PCN A3-mCh, behaved differently in response to induction but had a similar core metabolic response to the oxygen limitation. Interpreting metabolome data remains challenging, as metabolite pool changes can result from altered fluxes rather than their accumulation or depletion [54]. A change in a metabolite pool during a perturbation does not directly reveal how carbon flux through that part of the metabolic network is affected, as it can be increased or decreased in the same or reverse direction to the changes in the metabolite fluxes. But, regardless, any change in metabolite pools indicates that the metabolism is adjusted around a particular part of the metabolic network as a response to a perturbation. Therefore, metabolite pool data needs to be interpreted in correlation with complementary data sets [55, 56]. Bioreactors are a powerful platform to study these dynamics, as they provide a closed and controlled system where continuous changes such as CO_2_ production and O_2_ consumption rates can be measured. Such data, along with the substrate consumption and biomass/ byproduct formation rates, altogether the major uptake and excretion rates, provide a strong framework to assess the intracellular metabolic changes.
Metabolic burden in medium PCN strain after recombinant protein induction
One key observation of this study is that the medium copy number strain A3-mCh showed an 80% reduction in the growth rate after induction, which was not observed for the low PCN strain (A2-mCh). The large drop in the growth rate after induction in A3-mCh indicates a strong metabolic burden in this medium PCN strain. Overall metabolic activity was found to be turned down as indicated by a reduction in CO_2_ production rates (Table 1) and drop in endometabolome pools after RPP induction (Fig. 5a). The metabolic burden caused due to medium PCN is well-known and well-studied, with growth reduction being the most prominent phenotypical response [16, 57]. The cellular stress response is triggered, downregulating ribosomal machinery and biosynthesis pathways, which further inhibits cell growth [58].
Several studies have reported alterations in the gene expression levels for major catabolic pathways as a consequence of the metabolic burden [59, 60]. Studies on the metabolic burden in BL21 and K-12 strains of E. coli have reported the accumulation of precursor and energy molecules, mainly caused by the overflux through catabolic pathways to mitigate the increasing precursor demands for RPP [61, 62]. We did not observe an accumulation of energy molecules and precursors. The energy charge ratios and redox ratios were found to be unchanged even after RPP induction, indicating that cellular homeostasis was maintained. Biomass production was compromised to achieve the increasing demands of precursors for RPP and maintain cellular homeostasis, as also seen in other studies involving medium PCN strains [63].
Accumulation of extracellular and intracellular metabolites during oxygen limitation
Both the recombinant strains showed extracellular secretion of acetic acid as a product of mixed acid fermentation under microaerobic conditions*. E. coli* BL21 is a low producer of acetate under aerobic fermentation [64]; however, it accumulates acetate upon oxygen limitation as a result of fermentative metabolism [65, 66]. Further adjustments to oxygen limitation include lowering cell mass yields while keeping relatively high glucose consumption to maintain ATP production through substrate-level phosphorylation when oxidative phosphorylation is strangled. The current study presents new information about the intracellular metabolic response to oxygen limitation with a focus on central carbon pathways. Acetic acid is produced from pyruvate via acetyl-CoA, and due to the tight regulations between these pathways, disturbances in the neighboring metabolite pools can be expected. We observed mainly the accumulation of TCA and lower glycolytic pools in both recombinant strains in response to oxygen limitation. Since oxygen limitation restricts electron transport and NAD⁺ regeneration from NADH, the flux through the TCA cycle is expected to decrease, leading to an accumulation of TCA intermediates and, thus, an inverse response to the flux change [67]. These disturbances can subsequently lead to the accumulation of lower glycolytic metabolite pools.
For A3-mCh, the glucose uptake was reduced after RPP induction, and it further remained the same when microaerobic conditions were introduced. It could be said that the glycolytic flux also remained the same for A3-mCh even after oxygen limitation; however, this can’t be confirmed completely since the flux split ratio between glycolysis and PPP is unknown. Increased flux through PPP is highly unlikely to occur, since regeneration of NADP^+^ from the oxidative part of PPP is dependent on active biosynthesis of amino acids, fatty acids, and the non-oxidative part of PPP just feeds the flux back to glycolysis [68, 69]. Thus, one can assume that the glycolytic flux from glucose to pyruvate was at the same levels in A3-mCh in both aerobic and microaerobic conditions, but that the metabolic pools at the lower part of glycolysis accumulated due to constraints in TCA. In this case, the metabolic solution for the cell to maintain the redox homeostasis was acetic acid excretion.
E. coli maintains energetic homeostasis during protein production and microaerobic growth
Interestingly, the energetic state appeared homeostatic, as the energy charge was maintained for both strains throughout RPP induction and oxygen limitation. Several studies have previously reported that RPP is not limited by the adenylate energy charge [40, 61, 62]. Among redox ratios, NADH/NAD^+^ remained unaffected by both oxygen limitation and RPP, while the NADPH/NADP^+^ ratio was significantly altered by RPP for the medium PCN strain, A3-mCh upon induction.
No immediate effect of oxygen limitation was observed on the NADPH/NADP^+^ ratio in both strains (e.g., at T2), but A2-mCh showed a decline during later phases of oxygen limitation (Additional file 3, Figure S7a). This was accompanied by a decline in the concentrations of both NADP^+^ and NADPH (Fig. 3b). This, in turn, underlines the low anabolic activity and lower demand for the anabolic cofactors upon oxygen limitation, also leading to lower growth. Since both A2-mCh and A3-mCh maintained energy and redox homeostasis, and the amino acid pools were increased and not depleted, oxygen limitation would not be directly detrimental to recombinant protein production.
Accordingly, we observed that mCherry production was maintained at the same level in A2-mCh also during the oxygen-limited phase and was rather increased by both specific production rate and yield in A3-mCh (Table 1). Improved recombinant plasmid yields have been observed previously upon oxygen limitation [70, 71], attributed to the altered activity of DNA topoisomerases due to the limiting oxygen conditions [72, 73]. Inspired by the higher protein yields in microaerobic conditions, various studies have implemented microaerobic conditions for achieving growth rate control and, consequently, more economical protein production [74, 75].
Hence, even though changes in endometabolomic pools were observed in response to oxygen limitation (and, also, alterations in RPP in the case of the medium PCN strain), the strains were able to adapt to the changes and maintain the redox and energy homeostasis, continuing to produce recombinant proteins even in challenging microaerobic conditions. E. coli has a rapid transcriptional response to the surrounding dissolved oxygen conditions, resulting in tight regulation of the central carbon metabolome [25]. Upon comparison between the strains, the low PCN strain was found to be a suitable host for industrial protein production. However, the medium PCN strain is not favorable due to the drastic effect of the plasmid metabolic burden on the growth, which may lead to plasmid loss in the later growth stages and subsequent reduction in RPP. Hence, the choice and design of recombinant plasmids, especially promoter and plasmid copy number, are important factors to consider concerning the metabolic burden while constructing recombinant strains for industrial use, as oxygen gradients are more prevalent in industrial bioreactors.
Conclusion
The metabolic response of two recombinant E. coli strains was studied for microaerobic conditions as well as recombinant protein production. Both strains showed distinct responses to RPP but quite similar metabolic responses to the microaerobic conditions when coupled with RPP. Despite the adjustments in the endometabolome, the tested E. coli strains showed robustness in terms of maintaining cellular homeostasis, which was attained at the expense of a switch to mixed acid fermentation pathways. Production of mCherry, however, was not found to be affected by microaerobic conditions, which indicate that these E. coli strains prioritize recombinant protein production over biomass production.
Hence, in this study, we have successfully used targeted metabolic profiling methods to provide an overview of E. coli metabolic pathways under stressful conditions. This can be used as a base study to identify the key metabolites influencing recombinant protein production under these conditions, opening doors for pathway engineering studies and tailored bioprocesses.
Supplementary Information
Supplementary material 1: Figure S1. Cultivation plots for A2-mCh and A3-mCh (Biological replicates). Growth profiles of E. coli BL21 in benchtop bioreactors, including cultivation data and sampling points. The sampling points represent sampling for exometabolome and endometabolome analysis. mCherry production is indicated by both fluorescence and western blot (mg L^-1^). (a) Cultivation of A2-mCh (low PCN strain) with recombinant protein induction and no oxygen limitation (Control cultivation, A2-mCh_Aerobic_ind); (b) Cultivation of A2-mCh with recombinant protein induction and oxygen limitation (A2-mCh_O_2_-Lim); (c) Cultivation of A3-mCh (medium PCN strain) with recombinant protein induction and no oxygen limitation (A3-mCh_Aerobic_ind)); (b) Cultivation of A3-mCh with recombinant protein induction and oxygen limitation (A3-mCh_O_2_-Lim); (e): Cultivation of A3-mCh without adding inducer for production of recombinant protein, mCherrry (A3-mCh_No_ind). Figure S2. Western blot images used for quantification of mCherry. Samples were loaded in duplicates, and two dilutions were performed for each sample. The quantification was performed using a calibration curve prepared from the mCherry standard (mCherry std.). (a) Samples from A2-mCh_Aerobic_ind cultivation (Samples diluted 1:500 and 1:1000 times and loaded in alternate wells); (b) Samples from A2-mCh_O_2_-Lim cultivation (Samples diluted 1:1000 and 1:5000 times and loaded in alternate wells); (c) Samples from A3-mCh_O_2_-Lim cultivation (Soluble and insoluble fractions were diluted 1:500 and 1:1000 times, respectively, and loaded in alternate wells). Figure S3. Combined 2D scores plot and biplot of principal component analysis (PCA) displaying principal components of metabolite concentrations in *E. coli *BL21 A2-mCh during oxygen limiting cultivation (A2-mCh_O_2_-Lim) over four sampling points T1, T2, T3, and T4, where T1 represents the aerobic phase and T2, T3, and T4 are sampled during oxygen limitation (Figure 2 (b)). Figure S4. Heatmap representing endometabolome absolute concentrations (nmol g^-1^ CDW) for *E. coli *BL21 medium PCN strain, A3-mCh during three cultivations: (i) Without Induction of recombinant protein production (No_ind), (ii) without oxygen limitation (fully aerobic, Aerobic_ind), and (iii) with oxygen-limited (O_2_-Lim) condition. Metabolite abbreviations are listed in Additional file 2. Figure S5. Volcano plots showing metabolites with significant log_2_ fold variation in A3-mCh: (a) after RPP induction and (b) during O_2_-limitation. Figure S6. Schematic overview of the central carbon metabolome representing log_2_ fold change in metabolite concentrations for *E. coli *BL21 medium PCN strain, A3-mCh, in comparison with the lower PCN strain, A2-mCh. The plots are prepared using the Omix Visualization tool [48]. (a) T1 in A3-mCh Aerobic_ind cultivation (Figure 2(d)) vs. T1 in A2-mCh O_2_-Lim cultivation (Figure 2(b)); (b) T2 in A3-mCh O_2_-Lim cultivation (Figure 2(d)) vs. T4 in A2-mCh O_2_-Lim cultivation (Figure 2(b)). Supplementary material 2: Table S1. Abbreviations of metabolites used for metabolite profiling data, sorted based on metabolite class and listed with abbreviation and CAS number. Supplementary material 3: Table S2. Energy charge ratios and redox ratios during A2-mCh and A3-mCh cultivations. The data represent average values obtained from at least two biological replicates. Figure S7. (a) Redox ratios during A2-mCh (Low PCN) O_2_-limited cultivation (A2-mCh_O_2_-Lim, Figure 2 (b)). T1 represents aerobic cultivation. Sampling points, T2, T3, and T4 represent microaerobic cultivation. (b) Redox ratios for A3-mCh (Medium PCN) during no induction, RPP-induced_aerobic, and O_2_-limited phases (as explained in Table S2). The data represent average values obtained from at least two biological replicates. Statistical significance was determined by two-tailed t-test assuming equal variances for pairwise comparisons and one-way ANOVA with Tukey’s multiple comparison tests. Supplementary material 4: Table S3: Absolute metabolite concentrations (nmol g^-1^ CDW) during different cultivation conditions for two *E. coli *BL21 recombinant strains.