A recently described mini-scale chemostat setup was used for parallel operation of 12 chemostats 18. Briefly, bacteria were grown in sealed 17 ml Hungate tubes. To avoid stepwise feeding by dripping medium, the feed needle was placed directly in the culture liquid. The culture volume was kept constant at 10 ml by level control with a second needle placed at the desired level to suck off excessive culture broth. The medium removal pump was set to a speed of two culture volumes per minute, thereby also removing air from the headspace. The resulting underpressure caused air influx through a third needle, which was placed at the bottom of the reactor. The rising bubbles both aerated and mixed the cultures.

Due to sedimentation in the mini-scale reactors, B. amyloliquefaciens was the only species that was cultivated in 500 ml bioreactors (Infors AG, Switzerland). The cultivation volume of 250 ml was kept constant by level control. Throughout the cultivation, active mixing was achieved with a magnetic stirrer, and aeration was kept constant at two volume of air per volume of culture and minute.

Species used in this study are listed in Table 1. Frozen glycerol cultures were used to inoculate 5 ml Luria-Bertani (containing per liter: 5 g yeast extract, 10 g NaCl, 10 g tryptone) precultures, supplemented with antibiotics where necessary. From these LB precultures, we inoculated (1 to 500 diluted) 5 ml M9 minimal medium with 5 g/l glucose. From these overnight precultures, aerated Hungate tubes were inoculated at a ratio of 1 to 10. First, batch growth was allowed to proceed for 4 to 8 hours, and cultures were analyzed after continuous dilution of the medium for at least 6 volume changes at 37°C. For each steady state, a new chemostat culture was started from frozen stocks.

The M9 minimal medium was composed of the following components (per liter final volume): 5.64 g Na₂HPO₄, 3 g KH₂PO₄, 0.5 g NaCl, 1 g NH₄Cl, 0.246 g MgSO₄·7H₂O, 0.014 g CaCl₂, at pH 7.4 and 10 ml trace element solution containing (per liter) 1.35 g FeCl₂6H₂O, 0.1 g MnCl₂·H₂O, 0.17 g ZnCl₂, 0.043 g CuCl₂·2H₂O, 0.06 g CoCl₂·6H₂O, 0.06 g Na₂MoO₄·2H₂O 19. Filter-sterilized glucose was added to a final concentration of 1 g per liter. For ¹³C-labeling experiments, glucose was added either as a mixture of 20% (wt/wt) U-¹³C-labeled isotope isomer (99%; Cambridge Isotope Laboratories, Andover, MA) and 80% (wt/wt) natural glucose or as 100% 1-¹³C-labeled isotope isomer (99%; Cambridge, Isotope Laboratories, Andover, MA). Labeled glucose was used throughout the experiment, including the batch phase.

Batch experiments were done at 250 rpm in 500 ml baffled shake flask with a culture volume of 50 ml and 5 g/L glucose for non-labeled experiments and 30 ml volume and 3 g/L glucose for labeling experiments. The medium composition and preculturing were identical to the chemostat experiments. For B. subtilis natto and B. pumilus, biotin and adenine were added from sterile stock solutions to final concentrations of 0.1 μg/ml and 20 μg/ml, respectively.

Cell growth was monitored by determining optical density at 600 nm (OD₆₀₀). Glucose and acetate concentrations in culture supernatants were determined by using refractive index (RI) and UV detectors, respectively, on a HPLC system (Agilent/Hewlett Packard Series 1100) with an Aminex HPX-87H column (Biorad, Hercules, CA).

For continuous cultures, all physiological parameters were determined during steady state between 5 to 7 volume changes after inoculation. Since dilution and thus growth rate are constant in chemostat cultures, consumption and production rates were determined from the differences between substrate (S) and product (P) concentrations in the feed medium and culture supernatant. The relationship q_{S(or P)} = ΔS (or P) (D/X) we calculated specific production and consumption rates, where X is the biomass concentration. For determination of maintenance energy coefficients glucose consumption rates (q_glc) were plotted against dilution rates. The maintenance energy coefficients were determined as the y-axis intercept of the weighted least square (WLS) regression line using the SPSS statistical software package (SPSS Inc. Chicago, IL).

In batch culture, the growth rate (μ) was determined as the coefficient of the log-linear regression of OD₆₀₀ versus time. The biomass yield on the substrate (Y_X/S) was determined as the coefficient of a linear regression of biomass concentration versus substrate concentration during the exponential phase. The specific substrate consumption rate was determined as the coefficient of a linear regression of substrate concentrations versus biomass concentrations multiplied by μ. The same relationship holds for the specific rate of formation of (by-)products.

To calculate specific biomass yields, maintenance coefficients, consumption and production rates, a correlation factor for cellular dry weight (CDW) to OD₆₀₀ was used. To determine the cellular dry weight, 10 ml culture broth was transferred into preweighted 15 ml glass tubes and centrifuged for 10 min at 3000 g at 4°C. The pellets were washed twice with 0.9% NaCl and dried at 105°C for 24 h to constant weight. Except for the spo0A mutant, a single correlation factor (gCDW/OD) was used for all dilution rates (D) of a given species: i.e., B. subtilis wild type 0.48, B. subtilis sigE 0.41, B. licheniformis T218a 0.55, B. licheniformis T380B 0.48, B. subtilis natto 0.54, B. pumilus 0.61 and B. amyloliquefaciens 0.36. For the B. subtilis spo0A mutant the CDW-to-OD correlation varied with the growth rate (data not shown), hence we determined the correlation factors for this mutant at dilution rates of 0.05, 0.1, 0.2 and 0.4 h^-1 from mini-scale cultivations. These values were confirmed in a 1 l working volume stirred tank reactor (data not shown). The resulting correlation factors for the spo0A mutant were 0.39 gCDW/OD for D of 0.05 h^-1, 0.51 gCDW/OD for D of 0.2 to 0.3 h^-1 and 0.44 gCDW/OD for D of 0.4 to 0.5 h^-1.

Samples for gas chromatography-mass spectrometry (GC-MS) analysis were prepared as described previously 2021. Briefly, biomass from ¹³C-labeled chemostat cultures was harvested after stable OD₆₀₀ for at least 2 volume changes (at least 6 volume changes in total). For shake flask experiments, cells were harvested during mid-exponential growth at an OD₆₀₀ of 1–1.5. Cell pellets were hydrolyzed in 6 M HCl at 105°C in sealed microtubes overnight. Hydrolyzates were then dried under a constant air stream at 60°C. Derivatization was carried out at 85°C in 30 μl dimethylformamide (Fluka, Switzerland) and 30 μl N-(tert-butyldimethylsilyl)-N-methyl-trifluoroacetamide with 1% (v/v) tert-butyldimethylchlorosilane (Fluka, Switzerland) for 60 min. Derivatized amino acids were analyzed on a series 8000 GC, combined with an MD 800 mass spectrometer (Fisons Instruments, Beverly, MA). The GC-MS-derived mass isotope distributions were then analyzed using the software FiatFlux 22. Briefly, from the mass distributions of the amino acids the ¹³C-labeling patterns of their related precursor molecules in central metabolism were inferred. A set of probabilistic equations and the mass distributions of selected amino acid fragments were then combined to calculate the relative contribution of converging fluxes to a given metabolite pool 21.

Intracellular fluxes were estimated by fitting a flux distribution to the above flux ratios and quantitative physiological data within a stoichiometric model described by 23 using the software FiatFlux 22. Reaction reversibilities were chosen according to 23. The reaction matrix contained 24 unknown fluxes and 21 metabolite balances, including balances for glucose, acetate, CO₂, O₂, and the cofactors NADH and NADPH. Precursor requirements for biomass formation were taken from 23. To solve this under-determined system of linear equations, five additional constraints in the form of the above calculated flux ratios were used, i.e. serine derived through glycolysis, oxaloacetate originating from pyruvate, phosphoenolpyruvate (PEP) originating from oxaloacetate, upper and lower bound of pyruvate originating from malate and PEP derived through the PP pathway. The sum of the weighed square residuals of the constraints form both metabolite balances and flux ratios was minimized using the MATLAB (The Mathworks) function fmincon and the residuals were weighed by dividing through the experimental error 24. The computation was repeated at least five times with randomly chosen initial flux distributions to ensure identification of the global minimum.

A desirable physiological characteristic of cell factories is rapid and fully respiratory growth. For riboflavin production, in particular, an active PP pathway is expected to be a second relevant criterion because the product is primarily synthesized from pentose units. To investigate the potential suitability of B. subtilis 168 and four closely related bacilli as cell factories for biochemicals production, we determined their physiological parameters and fluxes during exponential growth in glucose batch cultures. The subspecies B. subtilis natto is phylogenetically the closest relative to B. subtilis, followed by B. amyloliquefaciens and B. licheniformis, while B. pumilus is somewhat more distal 25.

The ¹³C-flux data obtained for B. subtilis 168 compare favorably with previously reported data from other B. subtilis strains 26 and also with reported metabolic flux ratios for the type strain 168 27. To extend flux analysis to other bacilli, we first verified their network topologies by comparison with the KEGG database 28. All four genomes contained the genes for the reactions of the B. subtilis network, with the exception of an absent 6-phosphogluconate dehydrogenase in B. pumilus (Additional file 1). Since our labeling experiments with both [U-¹³C] and [1-¹³C] glucose demonstrated significant PP pathway fluxes also in B. pumilus (Figure 1), we assumed the presence of a 6-phosphogluconate dehydrogenase.

None of the investigated bacilli contain genes encoding enzymes of the Entner-Doudoroff pathway and absence of this pathway was also confirmed from the obtained ¹³C-labeling patterns (data not shown). In addition to the B. subtilis network, the B. licheniformis genome contains the two glyoxylate shunt-encoding genes, which enables B. licheniformis to grow on two-carbon units such as acetate or 2,3-butanediol 29. For the experiments shown here, however, we verified the absence of in vivo glyoxylate shunt fluxes from the calculated fraction of labeled CO₂ 18. This result is consistent with the normally observed glucose repression of this shunt 30.

During unlimited growth on glucose, intracellular fluxes (Figure 1) and physiology (Table 2) varied significantly among the investigated bacilli. The growth physiology of B. subtilis natto was similar to B. subtilis, with the obvious difference of elevated acetate secretion and a concomitantly lower TCA cycle flux. B. amyloliquefaciens and B. pumilus both grew rather slowly and were the most distinct from the others. B. amyloliquefaciens has the lowest biomass yield and low relative flux through the PP pathway. Moreover, its overflow metabolism is entirely different, secreting massive amounts of pyruvate instead of acetate. B. pumilus exhibited mostly respiratory metabolism with high relative TCA cycle fluxes. As a peculiarity, it has unusually high fluxes through the pyruvate shunt from malate to pyruvate and on to oxaloacetate that are indirectly inferred from the ¹³C data.

The two B. licheniformis strains are derivatives of the parent strain T5, a producer of thermostable alpha-amylase 3132. Specifically, strain T380B is a sporulation-deficient, direct descendant of T218a that was obtained after multiple rounds of classical mutagenesis and screening for traits that improved alpha-amylase production. Under the investigated conditions, however, less than 10 μg/ml protein was secreted in the medium, which did not significantly affect the mass balances. Both strains grow at intermediate specific growth rates with relatively high biomass yields (Table 2). In terms of intracellular fluxes, both B. licheniformis strains have about half of the TCA cycle flux of B. subtilis.

In addition to B. subtilis wild type 168, we analyzed also the otherwise isogenic B. subtilis 168 mutants spo0A and sigE. In particular, we were interested whether the frequently used spo0A mutation has unfavorable effects on metabolism through one of its many pleiotrophic effects rather than through the intended block of sporulation. As control for block of sporulation we used the sigE mutant that cannot enter stage III of sporulation 12. The phenotypic differences of B. subtilis wild type and spo0A compared to earlier work 27 is mainly due to different cultivation systems that lead to higher growth rates in the present experiments and possibly also to differences in strain background. While the sigE mutation is phenotypically silent under this condition, the spo0A mutant has a significantly increased glucose consumption and acetate formation rate (Table 2). The consequence is a strongly induced overflow metabolism and hence a reduced yield of biomass (Table 2), indicating spo0A as a suboptimal choice to suppress sporulation.

Since most industrial processes are based on carbon source-limited fed-batches at low growth rates, we quantified growth rate dependent physiology and fluxes in 10-ml glucose-limited chemostat cultures 18. In contrast to the discontinuously decreasing TCA cycle flux with decreasing growth rate that was described for glucose-limited E. coli chemostat cultures 18, all ratios of intracellular fluxes in B. subtilis and its sporulation mutants changed continuously or remained constant with decreasing growth rates (Figure 2). While the anaplerotic flux ratio of oxaloacetate derived from pyruvate remained almost invariant, the proportion of glycolytic versus PP pathway (serine derived through glycolysis) and the gluconeogenic flux through the PEP carboxykinase (PEP originating from oxaloacetate) increased continuously with decreasing growth rate (Figure 2). The latter result is consistent with the notion of decreasing catabolite repression at the severe glucose limitation at low dilution rates 33.

Absolute in vivo fluxes were then quantified by integrating the determined extracellular fluxes (Additional file 4) and the intracellular flux ratios (Additional file 2) by ¹³C-constrained flux analysis 24. In contrast to the flux distribution during unrestricted growth on glucose (Figure 1A) glucose-limited chemostat cultures of B. subtilis show i) no overflow metabolism, ii) about doubled relative TCA cycle flux that remained stable over different dilution rates, and iii) significant gluconeogenic flux through the PEP carboxykinase at the lowest dilution rate (Figure 3A).

The other bacilli exhibited generally rather similar trends in their flux ratios as seen for B. subtilis (Figure 2A and 2B). The sole exception was the in vivo PEP carboxykinase activity (PEP from oxaloacetate) of B. licheniformis. While this flux ratio was high at low dilution rates but absent at higher values in all other investigated bacilli, we observed a significant but constant fraction of PEP molecules originating from oxaloacetate in both B. licheniformis strains. At the representative dilution rate of 0.2 h^-1, we then calculated flux distributions for both B. licheniformis strains and B. subtilis natto (Figure 3B). The main differences to B. subtilis were the rather low respiratory TCA cycle fluxes in all three species.

Since the distribution of fluxes differed significantly in the investigated species, we were interested whether their catabolic NADPH formation matched the anabolic demand. For this purpose, we quantified NADPH formation from the previously quantified carbon fluxes through the NADPH-dependent reactions catalyzed by the oxidative PP pathway and the isocitrate dehydrogenase. The anabolic demand of NADPH was directly quantified from the known biochemical requirements of NADPH for growth-dependent macromolecules biosynthesis 34. Within the resolution of the analysis most B. subtilis strains exhibited balanced NADPH production and consumption during unrestricted growth on glucose (Figure 4A). Under glucose limitation however, there was clearly a catabolic overproduction for B. subtilis wild type (Figure 4B) and the two mutants (data not shown), as was reported earlier 35.

The NADPH balance was significantly different in the other bacilli. Firstly, the summed catabolic NADPH formation of both B. licheniformis strains was clearly insufficient to match the anabolic demand in batch and in one case also in chemostat culture (Figure 4A and 4C). This consistent catabolic underproduction was primarily caused by the rather low TCA cycle fluxes in B. licheniformis. Secondly, no species appeared to exhibit a catabolic NADPH overproduction like B. subtilis under any of the investigated conditions (data partially shown in Figure 4A and 4C). The sole exception might be the batch-grown B. pumilus due to its exceptionally high malic enzyme flux (Figure 4A). Since it is unclear which of the two malic enzymes with different cofactor specificities was active, however, this result is not conclusive.

Finally, to assess the bioenergetic suitability of the different species as biochemicals production host, we determined their maintenance metabolism; i.e. the amount of energy required to maintain cellular homeostasis in the absence of growth. Typically this non-growth-associated maintenance energy coefficient is quantified by applying Pirt's chemostat model 4:

q g l c = μ Y g l c max ⁡ + m g l c MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xI8qiVKYPFjYdHaVhbbf9v8qqaqFr0xc9vqFj0dXdbba91qpepeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemyCae3aaSbaaSqaaiabdEgaNjabdYgaSjabdogaJbqabaGccqGH9aqpjuaGdaWcaaqaaiabeY7aTbqaaiabdMfaznaaDaaabaGaem4zaCMaemiBaWMaem4yamgabaGagiyBa0MaeiyyaeMaeiiEaGhaaaaakiabgUcaRiabd2gaTnaaBaaaleaacqWGNbWzcqWGSbaBcqWGJbWyaeqaaaaa@4531@

where q_glc is the specific glucose consumption rate, m_glc the maintenance energy coefficient and Yglcmax⁡ MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacPC6xNi=xH8viVGI8Gi=hEeeu0xXdbba9frFj0xb9qqpG0dXdb9aspeI8k8fiI+fsY=rqGqVepae9pg0db9vqaiVgFr0xfr=xfr=xc9adbaqaaeGaciGaaiaabeqaaeqabiWaaaGcbaGaemywaK1aa0baaSqaaiabdEgaNjabdYgaSjabdogaJbqaaiGbc2gaTjabcggaHjabcIha4baaaaa@356A@ the maximum molar growth yield. Using the physiological data from the above described chemostat cultures, we found, consistent with the Pirt model, a linear dependency of the glucose consumption rate over the whole range of tested dilution rates for B. subtilis (Figure 5), B. licheniformis and B. amyloliquefaciens (Figure 6). For the B. subtilis spo0A mutant we determined growth rate specific OD₆₀₀ to cell dry weight conversion factors, because they did not remain constant, in contrast to the other species.

For these bacilli with linear dependencies of the specific glucose consumption rate with dilution rate, the maintenance energy coefficients were determined as the intercept of the weighted least square regression line with the y-axis (Table 3). The determined coefficient of 0.39 mmol g(cdw)^-1 h^-1 for B. subtilis wild type 168 compares favorably to the previously reported 0.44 mmol g(cdw)^-1 h^-1 of the related B. subtilis wild type 1012 9. The spo0A mutant, but not the sigE mutant, exhibited an increased maintenance coefficient of 0.49 mmol g(cdw)^-1 h^-1, which is qualitatively consistent with the previously reported high maintenance energy coefficient of riboflavin-producing spo0A mutants of B. subtilis 936. Together, these results provide strong evidence that the spo0A mutation has a stimulating and biotechnologically undesirable effect on maintenance metabolism.

Both B. licheniformis strains and B. amyloliquefaciens exhibited lower maintenance energy coefficients than B. subtilis wild type (Table 3). In most cases, however, the difference was minor and not statistically significant. The sole exception with significantly reduced maintenance energy requirements was B. licheniformis T380B, despite a large confidence interval. For B. pumilus and B. subtilis natto, the maintenance coefficients could not be determined because their specific glucose consumption rates did not show a linear relationship with dilution rate (Figure 6). The reasons are strong fluctuations in their steady state biomass concentrations, which, at least for B. pumilus, are related to dilution rates close to the maximum specific growth rate (Table 2).