Recombinant protein production‐associated metabolic burden reflects anabolic constraints and reveals similarities to a carbon overfeeding response

A comparison of the metabolic response of Escherichia coli BL21 (DE3) towards the production of human basic fibroblast growth factor (hFGF‐2) or towards carbon overfeeding revealed similarities which point to constraints in anabolic pathways. Contrary to expectations, neither energy generation (e.g., ATP) nor provision of precursor molecules for nucleotides (e.g., uracil) and amino acids (e.g., pyruvate, glutamate) limit host cell and plasmid‐encoded functions. Growth inhibition is assumed to occur when hampered anabolic capacities do not match with the ongoing and overwhelming carbon catabolism. Excessive carbon uptake leads to by‐product secretion, for example, pyruvate, acetate, glutamate, and energy spillage, for example, accumulation and degradation of adenine nucleotides with concomitant accumulation of extracellular hypoxanthine. The cellular response towards compromised anabolic capacities involves downregulation of cAMP formation, presumably responsible for subsequently better‐controlled glucose uptake and resultant accumulation of glucose in the culture medium. Growth inhibition is neglectable under conditions of reduced carbon availability when hampered anabolic capacities also match with catabolic carbon processing. The growth inhibitory effect with accompanying energy spillage, respectively, hypoxanthine secretion and cessation of cAMP formation is not unique to the production of hFGF‐2 but observed during the production of other proteins and also during overexpression of genes without transcript translation.


| INTRODUCTION
High-level recombinant protein production in Escherichia coli can lead to enhanced acetate secretion and growth inhibition, a phenomenon frequently described as "metabolic burden." The main reason for the protein production associated metabolic burden has been attributed to drainage of precursors and energy away from host cell maintenance and growth towards recombinant gene/plasmid related functions. Enhanced acetate formation and the resultant growth inhibition is also observed when E. coli is growing under nutrient excess conditions, a phenomenon generally known as "carbon overflow" metabolism. For both phenomena, energetic constraints have been held responsible, primarily the differences in proteome cost of energy biogenesis by respiration and fermentation (Basan et al., 2015;Mori, Marinari, & De, 2019;Zeng & Yang, 2019a, 2019b. In this context, overproduction of "useless" proteins has been considered to amplify the problem of carbon overflow metabolism by reducing the fraction of the host cell proteome available for energy production and biomass synthesis. These considerations imply that cells producing recombinant proteins may experience shortages in energy and biomass precursors. Protein synthesis is, compared to the synthesis of the other cellular macromolecules, a more energy-demanding process (Kaleta, Schäuble, Rinas, & Schuster, 2013;Stouthamer, 1973) in particular if energy requirements for protein foldingassociated purposes have to be taken into account (Corrales & Fersht, 1996;Sharma, De los, Christen, Lustig, & Goloubinoff, 2010;Szabo et al., 1994). Moreover, recombinant-protein producing cells require (deoxy)ribonucleotides for plasmid DNA synthesis and recombinant gene transcription as well as amino acids for the production of the recombinant protein potentially leading to a shortage of these precursors for host cell proliferation.
The most common E. coli-based expression system is the bacteriophage T7 RNA polymerase and T7 promoter controlled gene expression combination. Isopropyl β-D-1-thiogalactopyranoside (IPTG) induction of the chromosomally encoded T7 polymerase leads to high-level expression of T7 promoter-controlled genes and correspondingly high recombinant protein titers (Studier & Moffatt, 1986). Usually, production of recombinant proteins is carried out in E. coli B strains which are more robust with a lower tendency to form acetate compared to E. coli K strains (Shiloach & Rinas, 2009;Shiloach, Kaufmann, Guillard, & Fass, 1996). Most commonly, E. coli BL21 (DE3) is employed as it already contains the chromosomally encoded bacteriophage T7 RNA polymerase gene. Recently, it has been shown that even recombinant gene transcription without translation leads to growth inhibition in the E. coli BL21 (DE3) T7based expression system (Li & Rinas, 2020;Mittal, Brindle, Stephen, Plotkin, & Kudla, 2018) challenging the hypothesis that energy constraints are the major factors leading to the metabolic burden.
In this contribution, we analyze the impact of T7-promotercontrolled recombinant protein production on E. coli BL21 (DE3) metabolism in fed-batch cultures with special emphasis on the changes in glucose uptake and by-product formation (e.g., acetate, pyruvate, and glutamate) as well as on the changes in the profiles of energy-rich nucleotides (e.g., ATP) including their degradation products and regulatory molecules such as (p)ppGpp and cAMP. Moreover, we also follow the profiles of these compounds in fed-batch cultures in the absence of induced protein production but in response to carbon overfeeding.

| Medium and cultivation conditions
The composition of Luria-Bertani (LB) broth was as follows: 10 g L −1 tryptone, 5 g L −1 yeast extract, and 5 g L −1 NaCl. The pH was adjusted to pH 7 with NaOH before autoclaving. For solidification, 15 g L −1 agar was added. The composition of the defined glucosesupplemented mineral salt media for batch (shake flask and bioreactor) and fed-batch bioreactor cultures were essentially as described previously (Korz, Rinas, Hellmuth, Sanders, & Deckwer, 1995) with slight modifications. The composition of the batch medium for bioreactor cultures was as follows: 1.65 g L −1 glucose·H 2 O, 1.2 g L −1 MgSO 4 ·7H 2 O, 2.5 g L −1 (NH 4 ) 2 HPO 4 , 13.3 g L −1 KH 2 PO 4 , 1.86 g L −1 citric acid·H 2 O, 100.8 mg L −1 Fe(III) citrate, 2.1 mg L −1 Na 2 MoO 4 ·2H 2 O, 2.5 mg L −1 CoCl 2 ·6H 2 O, 15 mg L −1 MnCl 2 ·4H 2 O, 1.2 mg L −1 CuCl 2 , 3 mg L −1 H 3 BO 3 , 33.8 mg L −1 Zn(CH 3 COOH) 2 ·2- Evonik) was added to the batch medium for the bioreactor cultivation. Glucose·H 2 O (12 g L −1 ) was used for the defined non-inducing broth (DNB) in shake-flask precultures. The pH was adjusted to pH 6.8 using NaOH before autoclaving. The composition of the feeding solution was as follows: 110 g L −1 glucose·H 2 O, 7.55 g L −1 MgSO 4 ·7H 2 O, 40 mg L −1 Fe(III) citrate, 4 mg L −1 Na 2 MoO 4 ·2H 2 O, 4 mg L −1 CoCl 2 ·6H 2 O, 23.5 mg L −1 MnCl 2 ·4H 2 O, 1.81 mg L −1 CuCl 2 , 4.7 mg L −1 H 3 BO 3 , 16 mg L −1 Zn(CH 3 COOH) 2 ·2H 2 O, 16.55 mg L −1 Na 2 -EDTA·2H 2 O. Glucose·H 2 O (330 g L −1 ) was used in the feeding solution for the carbon overfeeding experiment. The feeding solution was sterilized by filtration through a 0.22 μm filter. Details of medium preparation are given elsewhere (Li, Nimtz, & Rinas, 2014). For plasmid maintenance, 50 mg L −1 kanamycin was added to the medium. Precultures were prepared as follows: glycerol stocks of recombinant E. coli were streaked on LB agar plates and incubated overnight at 37°C. A single colony from an LB agar plate was transferred to LB medium. The cultures were shaken at 30°C for 6 h to inoculate the DNB overnight pre-culture at 30°C with a starting OD 600 of 0.005. Shake flask cultivations were carried out using 100 or 500 ml Erlenmeyer flasks containing 10 or 50 ml medium, respectively, at 30°C and 250 rpm using a shaker with an amplitude of 5 cm. DNB pre-cultures were used to inoculate the bioreactor main cultures with a starting OD 600 of 0.2. Bioreactor cultivations were carried out in a 2 L stainless steel bioreactor (B. Braun Biotech International) containing 1.5 L initial batch medium. During bioreactor cultivations, temperature and aeration were set at 30°C and 1 vvm, respectively. Dissolved oxygen (DO) was automatically controlled at a level greater than 40% by changing agitation speed (300-1100 rpm). For the maintenance of the DO level in the carbon overfeeding experiment, the inlet air was slowly enriched with pure oxygen after the agitation speed reached the maximum. The pH was automatically controlled at pH 6.8 by the addition of ammonia solution (25%). After consumption of the initial glucose indicated by an increase of the dissolved oxygen concentration, the fed-batch phase was started using an exponentially increasing feeding rate. Feeding was carried out essentially as described previously (Korz et al., 1995) to maintain a predetermined specific growth rate of 0.35 h −1 (μ set = 0.35 h −1 ). When the optical density reachedOD 600 of 10, the cultivations were either induced by the addition of 1 mM IPTG or for the carbon overfeeding experiment, the predetermined specific growth rate was increased from 0.35 h −1 (μ set = 0.35 h −1 ) to 0.70 h −1 (μ set = 0.70 h −1 ). For reduced carbon feeding, the predetermined specific growth rate was set to μ set = 0.12 h −1 during the entire fed-batch phase of the cultivation.

| Sampling procedure
To minimize the time between sampling from the bioreactor and the inactivation of metabolic activities, a self-made fast sampling port was installed at the side of the bioreactor. The cultivation broth and outside environment were separated by a bromobutyl rubber septum (Sartorius Stedim Biotech) and a 2 mm silicone membrane (thickness, 2 mm). For the analysis of intracellular metabolites (ATP, ADP, AMP, cAMP, GTP, and (p)ppGpp), a perchloric acid (PCA) extraction method was utilized to ensure immediate cell lysis and quenching of all metabolic activities. A pre-weighted 2 ml disposable syringe was filled with approx. 0.6 g 35% PCA (wt/wt, about 5.8 M) containing 80 mM Na 2 -EDTA as described previously (Cserjan-Puschmann, Kramer, Duerrschmid, Striedner, & Bayer, 1999). To prevent precipitation of Na 2 -EDTA in 35% PCA, this acidic solution was kept at 37°C. Through the self-made fast sampling port, about 1 ml sample was quickly taken from the bioreactor by a hypodermic needle (Ф0.8 × 120 mm; B. Braun Melsungen AG) to the preweighted 2 ml disposable syringe containing 35% PCA and 80 mM Na 2 -EDTA. The acidified sample was vortexed for 10 min for complete cell disruption. After vortexing, the sample was transferred to a pre-weighed 15 ml conical centrifuge tube. The sample was neutralized to about pH 7.4 by a base (4 M KOH, 2% formaldehyde) in an ice bath. The cell debris and insoluble KClO 4 were removed by centrifugation at 17,000g and 3°C for 5 min. Finally, the supernatant was aliquoted into 1.5 ml centrifuge tubes, frozen in liquid nitrogen, and stored at −80°C. Analysis of high energy compounds (e.g., ATP and GTP) was performed as quickly as possible after the end of the cultivation. For the determination of the sample volume, at each step, the weight of the sample together with the syringe or centrifuge tube was measured assuming a sample density of 1.016 g ml −1 . For the determination of glucose and extracellular metabolites (acetate, pyruvate, glutamate, uracil, hypoxanthine, and extracellular cAMP), about 1 ml sample was taken from the bioreactor employing the same method as described above using an empty 2 ml disposable syringe. The cultivation broth was immediately centrifuged at 17,000g and 3°C for 3 min. The supernatant was filtrated through a 0.2 μm membrane, aliquoted to 1.5 ml centrifuge tubes, and stored at −80°C. The cell pellet was also stored at −80°C.

| HPLC analysis
Glucose, acetate, and pyruvate were quantified using a highperformance liquid chromatography system (HPLC; Hitachi) with a built-in diode-array detector (DAD) and an L-7490 refractive index (RI) detector (Merck) using an Aminex HPX-87H column (Bio-Rad) with a Carbo-H SecurityGuard™ cartridge (4 mm x 3.0 mm; Phenomenex). Agilent OpenLAB CDS Software (Agilent) was employed for system control and data acquisition. Before HPLC injection, all samples were filtered through a 0.2-µm syringe filter, and a 20 μl culture supernatant were used for injection (Millex GV; Millipore). Operating conditions were as follows: temperature 60°C, flow rate of 0.6 ml min −1 , isocratic mobile phase 20 mM H 2 SO 4 . The DAD at absorbance 254 nm was employed for the detection of pyruvate and the L-7490 RI detector for glucose and acetate.
Glutamate was quantified using an HPLC system (Hitachi) with an ALIAS™ autosampler (Spark Holland), an RF-20A fluorescence detector (Shimadzu) and a 3.5 µm Zorbax Eclipse Plus C18 column (4.6 mm × 150 mm; Agilent) connected with a C-18 Supelguard Cartridge (4 mm × 3.0 mm; Phenomenex). Clarity VA Chromatography Software (DataApex) was used for system control and data acquisition. Culture supernatants (50 μl) were mixed with 200 μl icecold methanol (−20°C) and stored overnight at −20°C. Precipitates were removed by centrifugation at 17,000g and 3°C for 20 min. The supernatant (50 μl) was alkalized by the addition of 50 μl cold sodium borate buffer (400 mM H 3 BO 3 , pH 10 by NaOH at 4°C). OPA/FMOC derivatization of amino acids in this solution (15 μl) was carried out in the ALIAS autosampler. To each vial containing a 15 μl sample, 20 μl OPA reagent (100 mg o-phthalaldehyde and 65 μl 3-mercaptopropionic acid dissolved in 5 ml methanol and 5 ml sodium borate buffer, 400 mM H 3 BO 3 , pH 10 by NaOH) were added and mixed. Following this, 15 μl FMOC reagent (9 mg 9-fluorenylmethoxycarbonyl chloride dissolved in 10 ml acetonitrile) was added and mixed. Lastly, 25 μl injection diluent (10 ml of buffer A [40 mM NaH 2 PO 4 , pH 7.8, 5 mM NaN 3 ] supplemented with 40 μl concentrated H 3 PO 4 ) were added and mixed. From this OPA/FMOC-derivatized sample mixture 15 μl were used for injection. The operating conditions were as follows: temperature 40°C and a flow rate of 1.5 ml min −1 . The mobile phases consisted of two eluents: buffer A (40 mM NaH 2 PO 4 , pH 7.8, 5 mM NaN 3 ) and buffer B (45% methanol, 45% acetonitrile, and 10% H 2 O). The following gradient was used: 4%-57% buffer B for 20 min, 100% buffer B for 3.5 min, and then equilibration at 4% buffer B for 5.5 min to restore the initial condition. RF-20A fluorescence detector (Shimadzu) was used for the detection of derivatized amino acids. The signal program was as follows: 0-17.5 min excitation at 330 nm and emission at 420 nm; the signal was zeroed at 0.01 min. Excitation from 17.5-25 min was set at 266 nm and emission at 305 nm; the signal was re-zeroed at 18 min.
Quantitative analyses of nucleotides and nucleobases (cAMP, ATP, ADP, AMP, GTP, (p)ppGpp, uracil, and hypoxanthine) were performed using the HPLC (HITACHI, Japan) with a built-in diodearray detector (DAD) and a reversed-phase Gemini® 5 µm C18 110 Å column (Phenomenex) connected with a C-18 Supelguard™ Cartridge (4 mm × 3.0 mm; Phenomenex). The operating conditions were as follows: temperature 30°C, flow rate of 1 ml min −1 and DAD at absorbance 254 nm. The mobile phases consisted of two eluents: buffer A (20 mM ammonium acetate, pH 5.9, 4 mM tetrabutylammonium hydroxide, 0.1 mM NaN 3 ) and buffer B (40% buffer A and 60% acetonitrile). The following gradient was used: 15% buffer B for 12 min, 15%-40% buffer B for 36 min, 40%-100% buffer B for 15 min, 100% buffer B for 12 min, 100%-15% buffer B for 2 min, and then equilibration was done at 15% buffer B for 10 min to restore the initial conditions. After about 10 injections, the column was washed with 80% acetonitrile for 1 h. The injection volume was 40 μl for PCA extraction samples and 20 μl for cultivation supernatants. To avoid degradation of ATP and GTP, PCA extraction samples were analyzed immediately after removal from the −80°C freezer.
HPLC standards GTP and pppGpp were from Jena Bioscience GmbH. All other standards were purchased from Sigma-Aldrich.

| Other analytical procedures and calculations
Cell growth was monitored by measurement of the absorbance at 600 nm (OD 600 ). Biomass was determined using the standard curve of OD 600 versus dry cell mass. One unit of OD 600 corresponds to 0.39 g dry cell mass L −1 for E. coli BL21 (DE3). Off-gas analysis was performed in bioreactor cultures using the BlueInOne gas analyzer (BlueSens). The carbon dioxide and oxygen transfer rates were calculated as described previously (Kayser et al., 2005). The dissolved oxygen concentration was determined using a polarographic dis-

| RESULTS
Reports on enhanced acetate formation and growth inhibition in response to recombinant protein production but also as a more general phenomenon observed during carbon overflow metabolism prompted us to compare the recombinant protein productionassociated metabolic response with the response towards carbon overfeeding. For this purpose, cells were grown in a fed-batch procedure supporting a growth rate of μ set = 0.35 h −1 . After reaching an optical density of OD 600 = 10 (=3.9 g L −1 dry cell mass), the cells were either induced to produce the recombinant protein or exposed to carbon overfeeding by suddenly raising the exponential glucose feeding rate in such a way that it could support a growth rate of μ set = 0.7 h −1 . The maximum growth rate of E. coli BL21 (DE3) is μ max = 0.52 h −1 during growth at 30°C using this medium, thus the carbon overfeeding conditions will certainly overwhelm the balanced carbon processing capacities of the cells. The initial situation before induction of protein production or before starting carbon overfeeding was in both types of experiments identical.
3.1 | Growth, glucose uptake, and primary metabolite formation (CO 2 , acetate, pyruvate, and glutamate) When cells are induced to produce hFGF-2 in fed-batch culture (μ set = 0.35 h −1 ) an obvious metabolic response became apparent approximately one and a half-hour after induction through a decline in the growth rate and the respiratory activity ( Figure 1a). Before this decline, pyruvate started to accumulate in the culture medium followed by accumulation of acetate and glutamate ( Figure 1b). Interestingly, E. coli BL21 (DE3) appears to be able to counteract the reduced balanced glucose processing capacities by reducing the glucose uptake rate. When cells were exposed to carbon overfeeding by increasing the exponential glucose feeding rate (not a glucose pulse!), cells instantaneously and strongly increased respiration for approximately 5 min followed by a period of almost constant respiratory activity of approximately 45 min, which was subsequently succeeded by a uniform but a more moderate exponential increase in the oxygen uptake and carbon dioxide formation rates (Figure 2a and inset therein). After initiation of carbon overfeeding, cell growth continued at the higher specific growth rate of μ = 0.45 h −1 but considerably lower than the growth rate aimed for (μ set = 0.7 h −1 ) and also lower than the maximum growth rate possible (μ max = 0.52 h −1 ; Figure 2a, inset). After initiation of carbon overfeeding, accumulation of glucose, pyruvate, acetate, and glutamate occurred (Figure 2a,b). It was again notable that accumulation of pyruvate started before the accumulation of acetate. Again, accumulation of glucose demonstrates that E. coli BL21 (DE3) appears to be capable to control the glucose uptake, keeping the acetate concentration well below 1 g L −1 . However, the immediate response to the higher glucose availability was a  Figure 1). The thin and thick arrows, respectively, indicate the end of the batch phase and the time point of initiation of carbon overfeeding. cAMP, cyclic adenosine monophosphate; CTR, carbon dioxide transfer rate; hFGF-2, human basic fibroblast growth factor-2; IPTG, isopropyl β-D-1-thiogalactopyranoside [Color figure can be viewed at wileyonlinelibrary.com] respire and grow in a seemingly balanced state although at rates lower than possible.

| Energy-rich compounds (ATP and GTP)
ATP and GTP are not only precursors for RNA synthesis but also serve as energy currency with ATP as the most important compound for energy exchange. As protein synthesis is an energy-costly process, the analysis of these compounds will give valuable hints if energy limitation could be a major factor of the protein productionassociated metabolic burden.
The analysis of the intracellular ATP content did not reveal any decrease for the first two hours after induction of hFGF-2 synthesis ( Figure 1c). In contrast, there was a transient increase in ATP, and decreasing ATP levels were first observed after the decline of the respiratory activity. As the majority of hFGF-2 is produced during the first two hours (see Figure 1a, inset), energetic restrictions are unlikely to contribute to the recombinant protein productionassociated metabolic burden. These data also indicate that the later observed ATP decline is the result but not the reason the recombinant protein production-associated metabolic perturbations.
The time-course data of GTP revealed a similar trend, confirming the above conclusions.
The time-course data of the energy-rich compounds ATP and GTP after exposure to carbon overfeeding did not reveal any obvious short-term increase but a more general decline after initiation of overfeeding conditions (Figure 2c). This decline was more obvious for ATP than for GTP. However, it should be noted that no samples were taken for ATP and GTP analysis during the first initial burst phase of respiratory activity (5 min) after initiation of carbon overfeeding.

| Degradation products of purine metabolism (ADP, AMP, hypoxanthine)
The initial degradation products of ATP are the lower energycontaining nucleotides ADP and AMP. Further degradation leads to the nucleobase adenine, which is further degraded to hypoxanthine (Leung & Schramm, 1980). Hypoxanthine formation is not only observed in bacteria but also in other organisms including humans, where it is considered as a general stress indicator (Mellon et al., 2019;Pechlivanis et al., 2015).
Interestingly, the sum of all AXP increased in response to hFGF-2 production, pointing to de novo synthesis of AXP nucleotides and not only to regenerative ATP turnover from ADP/AMP (Figure 1d).
Moreover, accumulation of hypoxanthine occurred in response to recombinant protein synthesis, suggesting enhanced degradation of AXP nucleotides (Figure 1d). ATP did not decrease at the onset of hypoxanthine secretion, in contrast, ATP still increased when hypoxanthine started to accumulate in the culture medium ( Figure 1c,d), indicating that hampered ATP utilization may induce AXP degradation pathways.
In the culture exposed to carbon overfeeding the sum of all AXP nucleotides did not increase, however, hypoxanthine secretion was immediately observed after the initiation of carbon overfeeding, presumably also indicating insufficient demand for adenosine nucleotides ( Figure 1d). Altogether, these findings suggest that ATP homeostasis is particularly hampered in cells induced to produce recombinant proteins but also in cells exposed to carbon overfeeding.

| Regulatory molecules (cAMP and (p)ppGpp)
Induced recombinant protein production but also carbon overfeeding has an impact on glucose metabolism. Thus, it is of interest to follow the concentration of the most important regulatory molecule of carbon metabolism, cyclic AMP (cAMP). cAMP increases instantaneously when cells encounter glucose starvation, for example, during entry into a stationary phase, with the majority being excreted into the medium (Buettner, Spitz, & Rickenberg, 1973;Makman & Sutherland, 1965). Moreover, (p)ppGpp, nutrient stress but also more general stress-sensing molecule (Dalebroux & Swanson, 2012), was followed after induced synthesis of hFGF-2 or carbon overfeeding.
At the end of the batch phase and during the fed-batch phase, extracellular cAMP concentrations gradually increased (Figures 1e   and 2e). When production of hFGF-2 was induced, cAMP concentrations continued to increase for approx. one and a half-hour more and then slowly declined afterwards (Figure 1e). When cells were exposed to carbon overfeeding, extracellular cAMP concentrations immediately stopped increasing for about one hour and subsequently resumed to increase at a higher rate (Figure 2e).
Analysis of the cellular (p)ppGpp content revealed an increase after induction of hFGF-2 synthesis (Figure 1e) but no significant change in the culture that was exposed to carbon overfeeding (Figure 2e).
Analysis of the culture medium revealed that extracellular uracil concentrations increased during the production of hFGF-2 in the fedbatch culture (Figure 1f). However, extracellular uracil concentrations also increased in the fed-batch phase of the nonproducing control culture although less pronounced. In the culture exposed to carbon overfeeding uracil accumulated even more rapidly (Figure 2f).
These data indicate that the growth inhibitory effect of recombinant protein production or carbon overfeeding leads to degradation of RNA and thus also indicates that there are presumably no limitations to the availability of RNA precursor metabolites.
3.6 | What happens when carbon feeding is reduced during hFGF-2 production?
Usually production of hFGF-2 but also of other proteins is carried out at lower feeding rates in fed-batch cultures to circumvent the problem of growth inhibition and acetate formation . To analyze the cellular response towards production of hFGF-2 at reduced carbon feeding, cells were grown in a fed-batch procedure supporting a growth rate of μ set = 0.12 h −1 (instead of μ set = 0.35 h −1 ).
The analyses revealed a neglectable growth inhibition in the hFGF-2 producing culture compared to the induced control culture not carrying any plasmid (Figure 3a). Glucose accumulation was not observed. Moreover, hFGF-2 was produced at a lower rate and almost exclusively as a soluble protein (see inset in Figure 3a). Acetate excretion was marginal and comparable to the control culture, accumulation of pyruvate was not observed (Figure 3b). However, the ATP also increased during production of hFGF-2 at conditions of slow carbon feeding (Figure 3b). Simultaneous to the ATP increase, accumulation of hypoxanthine occurred (Figure 3c), (p)ppGpp increased, and the extracellular accumulation of cAMP ceased ( Figure 3d). These data show that the basic cause for the metabolic burden is still present at slow carbon feeding, but the cells are able to better balance catabolic and anabolic metabolism.
3.7 | Is the phenomenon unique to the production of hFGF-2?
Production of hFGF-2 leads to a strong growth inhibition when production was carried out in fed-batch cultures using a feeding protocol that supports growth at μ set = 0.35 h −1 . Thus, it was tested if the production of other proteins using the T7-based expression system and the described fed-batch conditions were causing similar responses.
Production of GFP and GST-GFP also caused growth inhibition but less severe compared to the inhibition resulting from hFGF-2 production ( Figure 4a). In addition, glucose accumulated in all producing cultures, the more severe the growth inhibition, the earlier It has been proposed that respiratory energy generation might be a growth-limiting bottleneck (Anderson & von Meyenburg, 1980) and more recently, the high energetic costs for synthesizing the enzymatic inventory for respiratory energy generation have been held responsible for restricting growth, leading to carbon overflow metabolism with the well-known consequences of acetate formation and resultant growth inhibition (Basan et al., 2015;Zeng & Yang, 2019a). Our data do not indicate restrictions in respiratory energy generation. In contrast, respiratory activity soars instantaneously for a short period (approximately 5 min) in response to carbon overfeeding followed by an adaptation period until a new (pseudo) steady state is reached (Figure 2a). In the hFGF-2 producing culture, ATP levels still increase after the decline in respiratory activity until they finally start to decrease (Figure 1). Altogether, these data do not show limitations in respiratory ATP generation but suggest that cells which are not able to utilize the generated ATP reduce respiratory activity trying to keep a balanced energetic status within the cell.
The delayed arrest of cAMP accumulation upon induction signals that the recombinant cells are running into a metabolic status of too much carbon inflow, which cannot be processed adequately in anabolic pathways. In the culture exposed to carbon overfeeding, cAMP accumulation immediately stops after increasing the glucose feed signaling carbon excess conditions. In the protein-producing culture as well as in the culture exposed to carbon overfeeding, cessation of cAMP accumulation in the culture broth coincides with the onset of glucose accumulation in the culture medium (Figures 1 and 2). Excessive acetate formation, usually observed in E. coli K strains (Shiloach & Rinas, 2009;Shiloach et al., 1996), does not occur presumably because E. coli BL21 is capable of better adjusting glucose uptake to anabolic consumption. In the culture exposed to carbon overfeeding, cAMP accumulation resumes one hour after initiation of carbon overfeeding, (Figure 2) but in the hFGF-2 producing culture as well as in the other induced recombinant cultures, arrest of cAMP accumulation is observed for the remainder of the cultivation indicating ongoing sensing of carbon overflow conditions (Figures 1   and 4). Thus, the characteristic pattern of extracellular cAMP concentrations in fed-batch cultures can be used as a reliable reporter of fitness of protein-producing cells (see also Lin et al., 2004) signaling their reduced capacities for balanced carbon processing.
In addition, detection of increasing extracellular hypoxanthine concentrations can be used as an indicator of activation of adenine nucleotide degradation pathways initially as a result of insufficient ATP consumption. Accumulation of uracil also points to stressful conditions but extracellular uracil also increases during regular entry into the stationary phase when ribosome degradation occurs  and may only serve as an additional indicator of stress during recombinant protein production. Accumulation of (p)ppGpp, as a general stress indicator, can also signal stress conditions during recombinant protein production, however there are conflicting results on (p)ppGpp levels as an informative reporter on the fitness of protein-producing cells (Lin et al., 2004).
Growth inhibition is often enhanced when cells are producing under conditions supporting rapid growth. When production of hFGF-2 is carried out at slow carbon feeding conditions, growth inhibition is neglectable and accumulation of glucose is not observed (Figure 3).
However, production of hFGF-2 at slow carbon feeding also leads to ATP accumulation and hypoxanthine secretion as well as to cessation of cAMP accumulation. Thus, cells still appear to sense these conditions as "carbon excess conditions" however, at this reduced carbon inflow, the anabolic capacities of the cells do not appear to be overloaded.
The metabolic response is also not unique to the production of hFGF-2 but also observed during the production of other proteins. It even occurs during induction of recombinant gene transcription without recombinant protein synthesis and can also occur when short transcripts are made from "empty" expression vectors. The onset and extent of the metabolic response, however, depends on the specific expression system: the more severe the growth inhibition, the earlier the onset of hypoxanthine accumulation, cessation of cAMP accumulation, and start of glucose accumulation.

| CONCLUSIONS
In conclusion, the growth inhibitory metabolic response towards recombinant protein production is delayed, persistent, and based on constraints in anabolic pathways. Contrary to expectations, neither energy generation nor provision of precursor metabolites for LI AND RINAS | 103 nucleotides and amino acids limits host cell and plasmid-encoded functions. In contrast, accumulation of catabolic by-products, for example, pyruvate, acetate, and glutamate and energy spillage, for example, accumulation and degradation of adenine nucleotides with concomitant accumulation of extracellular hypoxanthine reflect overwhelming carbon catabolism and insufficient anabolic utilization.
The cellular response towards insufficient anabolic carbon utilization involves downregulation of cAMP formation, presumably responsible for subsequently better-controlled glucose uptake and the resultant accumulation of glucose in the culture medium. Growth inhibition is neglectable when protein production is induced at conditions supporting slower growth through reduced carbon availability. At these conditions, the reduced carbon supply does not overburden the compromised anabolic capacities of producing cells. The growth inhibitory effect with accompanying energy spillage, respectively, hypoxanthine secretion and cessation of cAMP formation is observed in response to the production of various proteins but also during overexpression of genes without transcript translation, and even when short transcripts are formed from "empty" expression vectors.

ACKNOWLEDGMENTS
Partial financial support was received from the German Ministry of Education and Research (BMBF) through the FORSYS-Partner program (grant FKZ 0315285) and from the German Research Council (DFG) through the Cluster of Excellence "Rebirth" EXC62. Open access funding enabled and organized by Projekt DEAL.

CONFLICT OF INTERESTS
The authors declare that there are no conflict of interests.

AUTHOR CONTRIBUTIONS
Zhaopeng Li carried out the experiments and performed the data analysis and prepared a first draft of the manuscript. Ursula Rinas directed the study and prepared the final manuscript. All authors read and approved the final manuscript.