All strains, plasmids and oligonucleotids used in this study are listed in Table 1. The C. glutamicum WT as well as the single deletion mutants Δpyc, ΔmalE, Δpck are from Blombach et al. (2013). The double deletion mutant Δppc Δpyc is described in Schwentner et al. (2018). The three remaining double deletion mutants Δpyc Δodx, Δppc ΔmalE and Δpck ΔmalE were constructed by chromosomal inactivation of the ODx gene odx in C. glutamicum Δpyc, as well as the ME gene malE in C. glutamicum Δpck and C. glutamicum Δppc using the plasmids pK19mobsacB-Δodx and pK19mobsacB-ΔmalE. Isolation of plasmids from Escherichia coli was performed as described elsewhere (Eikmanns et al., 1994). Plasmid DNA transfer into C. glutamicum was carried out by electroporation and recombinant strains were selected on Luria–Bertani Brain Heart Infusion agar plates containing appropriate concentrations of kanamycin (50 μg mL⁻¹) (Van Der Rest et al., 1999). The replacement at the chromosomal locus was verified by colony-PCR using primers odxfow/odxrev and Co-malE1/Co-malE1, respectively (Klaffl and Eikmanns, 2010; Blombach et al., 2011). Subsequently, shaking flasks with CGXII preculture medium containing 1% d-glucose (w v⁻¹) were inoculated with the corresponding strains (Unthan et al., 2014). Each culture was harvested during the late exponential phase following centrifugation at 4,500 rpm for 10 min, resuspension in sterile saline and finally in 20% (v v⁻¹) glycerol solution in sterile 0.9% (w v⁻¹) NaCl in distilled water. From this solution cryo stocks were prepared and immediately stored at −80°C. The Δppc Δpyc mutant was able to grow on acetate, from which a corresponding cryo-culture was produced.

C. glutamicum Δppc Δpyc was grown in microtiter plates in a BioLector parallel cultivation system (m2p-labs). Flowerplates with optodes for optical pH and dissolved oxygen (DO) measurements were employed. A cryo-culture for inoculation was washed once with sterile saline and resuspended in d-glucose-free CGXII medium. From this inoculum 50 μL were transferred into each well containing 950 μL 1% d-glucose medium. The plates were sterilely sealed and incubated at 30°C at 1,300 rpm.

For whole-genome sequencing, 200 μL of a well-inoculated at OD_init = 1 (see Supplementary Figure 1) and after growth has ceased was used to inoculate a subsequent shaking flask culture, from which a cryo-culture was produced. From this cryo-culture a sample was generated for whole-genome sequencing using the Illumina platform followed by sequence analysis as described elsewhere (Kranz et al., 2017).

C. glutamicum deletion strains were cultivated in a DASGIP parallel fermentation system (Eppendorf). Bioreactor cultivations of C. glutamicum strains were carried out with defined CGXII medium containing 1% d-glucose and 0.1% undiluted Antifoam 204 but no 3-(N-Morpholino)propanesulfonic acid (MOPS) buffer (Unthan et al., 2014). Bioreactors were inoculated from a preculture in CGXII medium buffered with 42 g L⁻¹ MOPS at pH 7 which was inoculated directly from a cryo-culture of each strain in 80% 0.9% NaCl/20% glycerol (v v⁻¹) stored at −80°C. During bioreactor cultivations DO levels were maintained above 30% by adjusting stirrer speed and oxygen content of the inlet air. The gassing rate was set to 1 vvm and the pH was maintained at pH 7 by feeding either 4 M NaOH or 4 M HCl. The cultivation temperature was 30°C.

For quantitative metabolomics, samples from bioreactor cultivations were drawn into a syringe in technical duplicate at two time points yielding a total of four technical replicates per strain. These time points correspond to target BV concentrations of 5 μL mL⁻¹ (OD₆₀₀ = 6.3) and 10 μL mL⁻¹ (OD₆₀₀ = 12.5), covering the mid-exponential phase (cmp. Supplementary Figure 2). The actual BV concentration in each sample was measured after sampling and used to calculate the extraction volume.

For untargeted proteomics, sampling was performed directly after all quenching samples for the metabolome analysis had been taken. From each reactor samples were drawn in technical quintuplicate by centrifuging 10 mL of culture broth for each replicate (10 min, 4500 rpm, GS-15R Centrifuge, Beckman Coulter). After the supernatant was decanted, the biomass pellets were immediately placed in aluminum racks at −20°C.

For biovolume (BV) measurements, cultivation samples were diluted 1:200 or 1:2,000 depending on the biomass concentration in 10 mL CASYton buffer (OMNI Life Science GmbH). The size distribution of the sample was determined by the MultiSizer3 Coulter Counter (Beckman Coulter) and the biovolume was computed by calculating the first moment of the distribution, assuming a spherical shape of the measured cells. Cell dry weight (CDW) was determined by centrifuging 2 mL of a bioreactor sample in a pre-dried and pre-weighted Eppendorf tube at 13,000 rpm for 7 min. The cells were washed in 1 mL 0.9% (w v⁻¹) NaCl by resuspension and renewed centrifugation. After decanting the supernatant, the pellets were dried for at least 2 days at 80°C.

A correlation between BV in μL mL⁻¹ and CDW in g L⁻¹ was derived for the Δpyc Δodx, Δppc ΔmalE and WT strain (see Supplementary Figure 3) and then applied to calculate the CDW for all cultivations assuming a standard deviation of 5%.

Specific rates for biomass growth (μ) and d-glucose uptake (π_GLC) were estimated using a model-based approach and process data from the exponential phases of corresponding batch cultivations. In short, the remaining d-glucose concentration c_GLC(t) at any given time point t is the integral over the volumetric uptake rate π_GLC,vol(t) given in mmol L⁻¹ h⁻¹:

Assuming a constant d-glucose uptake from exponentially growing cells, it holds:

where X(t)=X0·eμ·t denotes the biomass concentration in μL mL⁻¹ or g L⁻¹, depending on whether the biomass signal is BV or CDW. Inserting Equations (2) into (1) and carrying out the integration yields:

with model parameters X₀, μ, π_GLC, and C, where the latter absorbs the integration constant and the initial substrate concentration. Equations (2) and (3) were jointly fitted to the experimentally observed time courses of biomass and substrate concentration, respectively. The end of the exponential phase was judged by the peak in CO₂ volume fraction in the exhaust gas stream (see Supplementary Figure 2).

The fitting procedure was carried out using a sequential quadratic programming optimization routine from MATLAB (Mathworks Inc., R2019b). The estimation of confidence intervals was based on a parametric Monte Carlo bootstrapping approach from literature (Dalman et al., 2013). In short, every available measurement was perturbed independently within the normal distribution described by the measurement value and its standard deviation of the respective measurement. The perturbation operation consisted in sampling from the normal distribution of each measurement point and inserting the sample as measurements. The above-described variance-weighted least-square fit was then re-performed. The sample was generated as a Latin-Hypercube sample using the lhs-function of MATLAB and comprised 1,000 samples if not otherwise stated. The upper and lower confidence bounds were then derived as the α-th and (1-α)-th percentile of the parameter sample obtained from 1000 least-squared fits. If not otherwise stated α is set to 0.25, the confidence interval being the Interquartile Range (IQR).

The total CO₂ formation rate π_CO2,tot(t) given in mol h⁻¹ was calculated from balancing the gas phase of the bioreactor as:

where Φ denotes the volume fraction of the gas species in question in vol% and the superscripts α and ω denote the inlet and outlet concentrations, respectively. F denotes the inlet air flow in m³ h⁻¹. The biomass-specific carbon dioxide formation rate π_CO2(t) given in mmol gCDW-1 h⁻¹ or mmol mLBV-1 h⁻¹, respectively, was obtained by dividing π_CO2,tot(t) by X(t) at time point t.

The complete set of extracellular rate estimates for all independent bioreactor cultivation experiments can be found in Supplementary Table 1.

The metabolome samples of all strains were spiked with identical internal standard and were measured in one acquisition batch on the LC-ESI-QqQ MS system. Organic acids and unstable sugar phosphates were measured within 24 h from the extraction of samples. Quenching, cell separation, cell extraction, isotope dilution mass spectrometry and metabolite leakage correction were performed according to previous protocols (Paczia et al., 2012; Tillack et al., 2012).

For quantification of organic acids, samples were separated using a synergy hydro C18 reversed phase column (Phenomenex) on an Agilent 1200 chromatography system (Agilent Technologies). The HPLC column outlet was coupled to a QqQ MS device (API 4000, AB Sciex) equipped with a TurboSpray ion source in negative ionization mode. The elution was isocratic at 84% buffer A and 16% buffer B at a flowrate of 0.45 mL min⁻¹ at 20°C. The eluents were as follows: Buffer A: 10 mM tributylamine, 15 mM acetic acid, pH 4.95; Buffer B: methanol. MS parameters were as follows: CAD (collision gas pressure): 5, CUR (curtain gas flow): 30, GS1 (nebulizer gas flow): 70, GS2 (turbo heater gas flow): 70, IS (electrospray voltage): −4,500 V, TEM (heater gas temperature): 650°C, entrance potential: −10 eV. Injection volume was 10 μL.

For quantification of amino acids, samples were separated using Luna SCX cation exchange column at 60°C on a JASCO HPLC system. The following buffers were employed: Buffer A: 5% acetic acid, B: 15 mM ammonium acetate. The applied elution gradient can be found in Supplementary Table 2. Injection volume was 10 μL.

For quantification of sugar and nucleoside phosphates, samples were separated on a synergy hydro C18 reversed phase column at 40°C. Eluent were Buffer A: 10 mM tributylamine, 15 mM acetic acid, pH 4.95 and Buffer B: methanol. The HPLC system, the QqQ MS device and its MS parameter settings were the same as for the LC-MS/MS method for organic acids. For quantification the gradient of Supplementary Table 3 was applied. Injection volume was 10 μL.

To populate our C. glutamicum ion library, separate IDA acquisitions of samples of C. glutamicum WT cultivated as described above but with different carbon sources were performed. In each cultivation the carbon source was either 55 mM d-glucose, 111 mM sodium pyruvate, 83 mM disodium-L-malate, 55 mM sodium citrate or 48 mM sodium benzoate. Each sample was lysed and 100 μg protein thereof digested and processed as described elsewhere (Voges and Noack, 2012).

These samples were separated on a Agilent 1260 Infinity HPLC system (Agilent Technologies) equipped with a 150 ^* 2.1 mm Ascentis Express Peptide ES-C18 column with 2.7 μm particle size and an appropriate 5 ^* 0.3 mm Acclaim PepMap Trap Cartridge (Thermo Scientific) which were both maintained at 25°C and a flow rate of 200 μL min⁻¹. For LC separation 0.1% formic acid in LC-MS grade water (v v⁻¹) was used as buffer A, whereas buffer B was 0.1% formic acid in LC-MS grade acetonitrile (v v⁻¹). Before each injection, the column was equilibrated for 12 min at 97% A. After 20 μL were injected, the gradient of Supplementary Table 4 was applied. The LC-eluent was coupled to an ESI-QqTOF MS (TripleTOF 6600, AB Sciex) equipped with an DuoSpray ion source. The data acquisition was performed using Analyst TF 1.8 (AB Sciex). An information-dependent acquisition was performed on each injection during which all ions with m/z >300, charge state 2–4 and above intensity of 150 were selected for fragmentation.

The acquired MS2 spectra of the C. glutamicum digests were searched against a FASTA database of C. glutamicum ATCC 13032 (GenBank assembly accession: GCA_000196335.1) using ProteinPilot software 5.0 (AB Sciex) employing default probabilities for biological modifications. The confidently identified peptides of each injection were assembled into a library covering 1727 ORFs. This library incorporates the peptide confidence after identification, peptide (precursor) intensity in the MS1 scan from the IDA acquisition, fragment ion intensities and the observed peptide retention time.

Starting from an IDA acquisition of each organism variable SWATH windows were calculated using the SWATH Variable Window Calculator 1.0 (AB Sciex). Using these windows, a SWATH acquisition method was set up, which employed the same chromatographic gradient and ion source setting as the IDA acquisition. The window width and CE ramp parameters can be found in Supplementary Table 5. For SWATH acquisition a digest of each sample was prepared according to the protocol of Voges and Noack, which involves mixing 50 μg unlabeled sample protein and 50 μg internal standard from a separate cultivation of C. glutamicum with (¹⁵NH₄)₂SO₄ (Voges and Noack, 2012). Ten microliter of each sample was injected.

SWATH data processing was performed using the MS/MS^all SWATH Acquisition MicroApp in PeakView 2.2 (AB Sciex). The ion library from above was imported from ProteinPilot into this app by excluding shared peptides but not modified ones. From the imported ion library the ten most intense peptides were selected for quantification provided they had a peptide confidence of >96%. The intensity selection is based on the MS1 survey scan intensity of each peptide in the IDA runs used to build the ion library. If <10 peptides fulfilled the above criterion for a protein, only the available peptides were quantified.

For each peptide group the 12 most intense fragment ion traces were chosen by the SWATH processing algorithm. This algorithm favors the most intense fragment ion traces from the library spectrum whose m/z value lies above the Q1 window of their precursor ion. For each fragment ion, within 5 min around the expected retention time, an unlabeled mass trace was extracted from the SWATH spectra within ±15 ppm of its monoisotopic mass, whereas the labeled mass trace was extracted within ±15 ppm of its fully ¹⁵N-labeled isotopologue. All transitions of one peptide were assembled into a so called peak group which was scored for congruency with the ion library. The false discovery rate was set to 0.1%. The finished processing session was saved as MarkerView file (.mrkvw extension), which was opened in MarkerView 1.3.1 (AB Sciex) for estimation of fold-changes (ratio of means) of protein levels between mutant and control.

The concentration of carbon, nitrogen and sulfur in biomass were determined at the central Analytical core facility of Forschungszentrum Jülich (ZEA-3). Biomass samples from the early stationary phase were processed following the same procedure as for CDW content determination. The dried biomass pellet was ground to a fine powder in a pre-dried mortar using a pre-dried pestle and sent in a sealed container to ZEA-3.

In our previous flux identifiability analysis focusing on the anaplerotic node in C. glutamicum those metabolic network structures were identified that are structurally identifiable under defined d-glucose conditions (Kappelmann et al., 2016). These structures are based on deletions of specific genes encoding for anaplerotic reactions. In this study, we comprehensively analyzed a set of mutant strains for which a unique solution using ¹³C-MFA theoretically exists as well as mutants that are non-flux-identifiable strains.

All selected deletion mutants were able to grow on defined CGXII medium with d-glucose as sole carbon and energy source, except for strain C. glutamicum Δppc Δpyc that is deficient in PEPCx and PCx activity (Table 1). While this mutant was able to grow on acetate without lag-phase no biomass formation could be monitored within 48 h cultivation on d-glucose and this observation is concordant with Peters-Wendisch et al. (1998). However, prolonged incubation of C. glutamicum Δppc Δpyc under defined d-glucose conditions in a microbioreactor setup and at different inoculum sizes always resulted in the onset of growth after 75 h or later, mainly after 100 h (Supplementary Figure 1).

The whole mutant library was then cultivated under controlled bioreactor conditions on defined CGXII medium with 1% d-glucose. Specific growth and substrate uptake rates were estimated using a model-based approach (see Materials and Methods section). The d-glucose uptake rate of 4.82 mmol gCDW-1h-1 and the specific growth rate of 0.45 h⁻¹ for C. glutamicum WT agree well with literature (Buchholz et al., 2014). Based on the determined confidence intervals discernible growth phenotypes can be identified among the set of deletion mutants (Figure 2A and Table 2).

First, the C. glutamicum Δpck mutant showed a lower growth rate although PEPCk catalyzes a gluconeogenetic reaction, not needed under glycolytic conditions to supply biomass precursors. However, the observed growth defect clearly indicates its activity under glycolytic conditions. This result was also obtained in the study of Riedel et al. with the comparable genotype (Table 2) and in the study of Petersen et al. (2001), who inferred PEPCk activity in a C. glutamicum l-lysine producer from ¹³C-labeling data. Since the carbon flow catalyzed by PEPCk is of opposite direction compared to the overall carbon flow under glycolytic conditions, its catalytic effect must be of indirect nature. Strangely, the double deletion mutant C. glutamicum Δpck ΔmalE showed a restored growth phenotype, with a growth rate equivalent to the WT.

Deletion mutants comprising a deletion in PEPCx do not show a significantly altered growth phenotype as long as PCx is still active (Figure 2A and Table 2). The stoichiometry of the reaction catalyzed by PEPCx is identical to the reaction sequence of pyruvate kinase (PK) and PCx, which apparently fully compensates for the missing carboxylation activity of PEPCx. On the other hand, mutant strains C. glutamicum Δpyc and C. glutamicum Δpyc Δodx do exhibit a growth phenotype. This finding underscores the role of PCx as most important anaplerotic reaction under aerobic conditions and suggests that its catalyzed flux is most likely greater than that of PEPCx in the WT under standard d-glucose conditions. Noteworthy, a C. glutamicum mutant with single deletion of the odx gene was shown to grow equally well as the wild type (Klaffl and Eikmanns, 2010).

The evolved C. glutamicum Δppc Δpyc strain, missing both carboxylation activities shows a greatly reduced growth rate of 0.27 h⁻¹ (Figure 2A and Table 2). In the absence of both PEPCx and PCx, three other and different anaplerotic activities can possibly substitute the anaplerotic activity of PEPCx and PCx: First, ME may catalyze the carboxylation of pyruvate to malate in an NADPH-dependent manner. This appears plausible since ME was found to catalyze this reaction sequence in in vitro assays (Cocaign-Bousquet et al., 1996; Gourdon et al., 2000). Moreover, it could be experimentally shown that a NADPH-dependent ME from E. coli can act as sole anaplerotic enzyme in Saccharomyces cerevisiae sustaining a growth rate of 0.06 h⁻¹ (Zelle et al., 2011). Second, PEPCk may act in reverse direction from phosphoenolpyruvate to oxaloacetate. This hypothesis is rather unlikely as this reaction directionality would couple anaplerotic carboxylation to the substrate level phosphorylation, generating one GTP molecule. Since the reaction from oxaloacetate to phosphoenolpyruvate is coupled to the hydrolysis of one GTP molecule, it can be expected to be favored. A third alternative represents the glyoxylate shunt, which exclusively fulfills the anaplerotic function in C. glutamicum under growth on acetate (Wendisch et al., 1997).

Interestingly, the C. glutamicum Δppc Δpyc mutant also showed an altered ratio of specific glucose uptake and growth rate in comparison to other mutants and the WT (Figure 2A). The observation that less biomass was formed per unit uptake rate raised the question as to where the excess carbon atoms end up. To answer this question, a carbon balance was performed. Contrary to the conventional carbon balancing approach, consisting in the quantification of the total carbon recovery in biomass and exhaust gas by the time all substrate has been consumed (Buchholz et al., 2014), we calculated the instantaneous carbon balance Θ as:

where μ denotes the specific growth rate given in h⁻¹ (here assumed to be constant for the considered exponentially growing cells), ω_C denotes the mass fraction of carbon in the biomass in g_C gCDW-1, M_C denotes the molecular weight of carbon in g mmol⁻¹, and π_GLC as well as π_CO2 denote the specific d-glucose uptake and carbon dioxide formation rates as derived from Equations (2) to (4).

Equation (5) balances the specific rates of carbon uptake and carbon flow into sinks at any given time. Here biomass and CO₂ formation are the only considered carbon sinks (any other by-product formation could be excluded for all strains under investigation). The quantity ω_C has been reported several times in the literature: Marx et al. (1996) reported 0.408 g_C gCDW-1 for C. glutamicum MH20-22B, a l-lysine producer strain, determined in freeze-dried biomass with a CHNS elemental analyzer. More recently, Buchholz et al. (2014) reported a value of 0.514 gC gCDW-1 for C. glutamicum ATCC 13032, which was determined by separately quantifying the total carbon in liquid bioreactor samples (supernatant and biomass) and the total inorganic carbon (total dissolved carbon) in the supernatant. In this work, we experimentally quantified ω_C for wild-type C. glutamicum to be around 0.4–0.42 gC gCDW-1. The uncertainty of this parameter notwithstanding, its value was assumed to be 0.45 gC gCDW-1 for all following calculations. In face of the apparent uncertainty, its standard error was assumed to be 0.05 gC gCDW-1. The arising interval covers all available literature information on this quantity. The uncertainty in derived quantities thereof can be computed by Gaussian error propagation.

From Figure 2B it becomes apparent that the evolved Δppc Δpyc strain grows with a higher relative CO₂-formation rate with respect to the uptake rate. At the same time, the sum of the specific rates at which carbon flows into sinks amounts to the same relative value with respect to the carbon uptake rate as in other strains. Therefore, it seems plausible that a higher decarboxylation activity explains the lower relative growth rate in the C. glutamicum Δppc Δpyc mutant. This higher relative CO₂-formation rate further substantiates the hypothesis that an altered ratio of ICL and ICD activity involves the glyoxylate shunt as anaplerotic reaction sequence in this mutant. Exclusive anaplerotic activity through the glyoxylate shunt would release two equivalents CO₂ per C4-body of oxaloacetate formed, instead of fixing one CO₂ as in the case of alternative ME or reversible PEPCk activities.

One result holds true irrespective of the genetic background: The recovery of carbon at any given time in the reactions that act as carbon sinks amounts to 80% of the carbon equivalent of the uptake rate (Figure 2B and Table 2). The non-closed instantaneous carbon balance may hint to extensive by-product formation or indicate systematically biased extracellular rates. The former is unlikely since C. glutamicum WT is known to produce only minor by-products under aerobic conditions (the DO was maintained at 30%). Notwithstanding, the genetic alterations may induce a more extensive overflow metabolism in some deletion mutants. However, all organic acids, sugar phosphates and amino acids in the culture supernatant measured by targeted LC-MS/MS account for a total of 274–786 μmol_C gCDW-1, depending on the strain. Therefore, the exometabolome can be neglected as carbon sink since the gap in the balance of specific rates amounts to several mmol_C gCDW-1 h⁻¹ (Figure 2B).

Remarkably, the value of 80% for the instantaneous carbon balance matches the determined carbon balance closure in the study of Buchholz et al. (2014). This study convincingly showed that the gap in carbon balance can be traced back to the fact that part of the CO₂-production at any given time dissolves as HCO3- in the culture broth. This share is not recovered as gaseous CO₂ at the detector, leading to an underestimation of the CO₂-production rate.

To gain further insight into the metabolism of each mutant, untargeted proteome and targeted metabolome analyses were performed. To ensure comparability, eight strains were cultivated in parallel and subjected to identical and isochronous sample processing in subsequent steps (see section Materials and Methods for detailed descriptions). In total, we analyzed 1199 cytosolic proteins and 48 metabolites of central metabolism.

Since the Δppc mutant showed no altered phenotype and has been thoroughly characterized before, it was omitted from the set of strains to be analyzed. The evolved strain C. glutamicum Δppc Δpyc showed significant changes in specific proteins and metabolites, which will be discussed separately in the next section.

Further differentially expressed proteins were found in C. glutamicum Δpck and C. glutamicum Δpyc (Figure 3). In none of the other tested deletion mutants significantly changed protein abundances [p < 0.05, |Log2(fold change)| ≥ 0.5] were found (data not shown).

Only one protein encoded by cybD (cg0282) and which might be involved in stress response was found to be up-regulated in the C. glutamicum Δpck mutant (Table 3). In C. glutamicum Δpyc the enzyme quinolinate synthase A encoded by the nadA gene (cg1216) was up-regulated. This enzyme catalyzes the condensation of iminoaspartate with dihydroxyacetone phosphate to form quinolinate, and represents the second step of the de novo synthesis of NAD⁺. The latter starts from l-aspartate and up-regulation of this enzyme could be a cellular response to the limited availability of this amino acid following the inactivation of PCx and to ensure sufficient NAD⁺ supply. In addition, the putative transcriptional regulator (cg0787) and the 50S ribosomal protein L36 encoded by the rpmJ gene (cg2791) were found to be down-regulated in C. glutamicum Δpyc.

Figure 4 shows intracellular and extracellular concentrations of selected metabolites. The l-aspartate pool is most closely correlated with the growth rate, i.e., the lowest concentrations were observed for all mutants with reduced growth rate. Apparently, its supply appears to be limiting the growth as the restored growth rate of the Δpck ΔmalE mutant in comparison to the single deletion strain Δpck goes hand in hand with increased l-aspartate supply. The concentration pattern of l-aspartate seems to reflect itself in the l-homoserine pool, which is derived from the former through three intermediate reaction steps, consuming two NAD(P)H molecules and one ATP molecule. In contradistinction to l-aspartate, however, l-homoserine is clearly higher concentrated in the Δpck mutant. Since the reaction sequence between both intermediates is redox-dependent, one may be tempted to attribute this observation to redox balancing. The biosynthesis of l-glutamate from α-ketoglutarate and of l-proline from l-glutamate are also redox-dependent. Conspicuously, both pools are also higher concentrated in this strain (Figure 4).

Moreover, we analyzed intracellular levels of NADPH and NADH alongside their oxidized analogs. No significant difference was detected in the reduced forms of these co-factors (data not shown). However, the mean of relative standard deviation over all mutants for NADH and NADPH concentrations amounts to 46% and 48%, respectively. This high technical error, impeding a precise quantification, cannot be traced back to inaccuracies in pipetting or BV concentration measurements since these factors also apply to all other metabolite quantifications in the same sample. Since the mean relative standard deviation for other metabolite pools was below 10% these factors appear to have been controlled quite well. The most likely reason for the observed coefficients of variation is metabolite instability. The redox equivalents are known to be quite sensitive to oxidation and degradation (Siegel et al., 2014). Slightly different temperature time courses, residual enzymatic activity in metabolite extracts and oxidation most likely account for the observed differences. Therefore, no accurate conclusion about the redox state in each mutant could be drawn. Nonetheless, the fact that the removal of a redox-dependent enzyme like ME restores the growth rate of the C. glutamicum Δpck mutant suggests an involvement of the redox balance mediating some of the observed changes in metabolite pools.

Whole-genome sequencing of evolved C. glutamicum Δppc Δpyc confirmed the absence of genes ppc and pyc across the cell population (Table 4), excluding any growth contamination effect. Two insertions and some SNPs (see Supplementary Table 6) were detected in the coding region for ICD. Moreover, ICL (cg2560), MS (cg2559), and cis-aconitase (ACN, cg1737) are highly up-regulated in C. glutamicum Δppc Δpyc (Figure 3 and Table 3) and only in this mutant intracellular and extracellular accumulation of glyoxylate was detected (Figure 4).

These findings are in agreement with the study of Schwentner et al. and point to a redirection of carbon flux in our evolved strain from the oxidative decarboxylation branch of the TCA cycle into the glyoxylate shunt (Schwentner et al., 2018). In the absence of C3-carboxylation activity at the anaplerotic node, accumulation of the substrate pools phosphoenolpyruvate and pyruvate can be expected. Indeed, phosphoenolpyruvate was significantly higher concentrated in the evolved Δppc Δpyc strain, while the intracellular pyruvate pool remained unaffected (Figure 4). However, metabolite levels of l-alanine and l-valine, which are directly derived from pyruvate, were strongly increased and this finding is also consistent with previous data (Schwentner et al., 2018). The lack of statistical significance of strain differences in the pyruvate pool is most likely due to higher technical errors during metabolite quantification. It is well-established that the accurate quantification of organic acids in cell extracts represents a veritable challenge (Zimmermann et al., 2014).

Moreover, the metabolite pool of l-glycine shows one of the most significant concentration changes, clearly distinguishing the evolved Δppc Δpyc strain. l-glycine, in turn, is derived from l-serine, which also showed an increased concentration (Figure 4). It appears that the missing carboxylation rate cannot be matched by the pyruvate dehydrogenase activity in the Δppc Δpyc mutant for substrate pools like phosphoenolpyruvate as well as amino acids derived from the lower glycolytic intermediates appear to accumulate intracellularly. This would also explain why the phosphoglycerate dehydrogenase encoded by serA (cg1451) was found to be significantly down-regulated in this mutant (Table 3). The enzyme catalyzes the first step in the biosynthesis of l-glycine, l-serine and l-cysteine and its down-regulation could be the cellular response to the higher availability of 3-phosphoglycerate.

In terms of glyoxylate shunt regulation, Wendisch et al. suggested that the carbon-source dependent regulation of this pathway is mediated by intracellular acetyl-CoA concentrations (Wendisch et al., 1997). However, intracellular acetyl-CoA concentrations did not vary significantly with respect to strain background (Figure 4), and therefore this hypothesis could not be validated with the made intracellular measurements. Unfortunately, acetyl-CoA measurements are notoriously error-prone due to the high instability of thioesthers. Though taking strenuous efforts to keep the sample below−20°C and immediate analysis after extraction, we still obtained a highly variable signal within each treatment group.

The corresponding genes aceA and aceB of ICL and MS, respectively, are thought to be repressed by the regulator protein RamB (cg0444) under glycolytic conditions (Auchter et al., 2011). Moreover, it has been established that RamA (cg2831) acts as transcriptional activator of both genes when acetate is the carbon source. However, no significant change in the abundances of RamA and RamB were found in Δppc Δpyc mutant compared to the wild type. The fact that a ΔramA mutant is not able to grow on acetate suggests that RamA is strictly required as transcriptional activator for increased expression of at least one of the essential enzymes of acetate assimilation, which are acetate kinase (AK) encoded by ackA (cg3047), phosphotransacetylase (PTA) encoded by pta (cg3048), ICL and MS (Cramer and Eikmanns, 2007). Indeed, AK and PTA are also significantly up-regulated in the Δppc Δpyc mutant (Table 3).

Most interestingly, this strain also shows an up-regulation of the three enzymes of the methylcitrate cycle in C. glutamicum (Claes et al., 2002), namely 2-methylcitrate synthase (PrpC2) encoded by prpC2 (cg0762), 2-methylcitrate dehydratase (PrpD2) encoded by prpD2 (cg0759) and 2-methylcitrate lyase (PrpB2) encoded by prpB2 (cg0760) (Table 3). Together with the activities of AK and PTA this cycle is known to be the predominant route for the degradation of propionate into pyruvate and succinate (Figure 1). In the presence of propionate all three genes are transcriptionally activated by the regulator PrpR (Plassmeier et al., 2012). Under the applied d-glucose conditions, however, no change in the PrpR abundance was detected. Moreover, it was found that the prpDBC2 operon is transcriptionally activated by RamA, but not affected by RamB (Auchter et al., 2011). This makes the conclusion compelling that the concentration of at least one, still unknown, metabolite effector binding to RamA was altered, leading to the activation of the aforementioned genes. Following the strong up-regulation of the prpDBC2 cluster in combination with the glyoxylate shunt it might be speculated that the surplus of pyruvate (from PEPCx and PCx inactivation) and succinate (from ICL activity) is channeled into the methylcitrate cycle, operating in the reverse direction. This route would represent a by-pass of the TCA cycle reactions succinate: quinone oxidoreductase (SQO), fumarase (FUM) and malate: quinone oxidoreductase (MQO) to provide oxaloacetate as biomass precursor (Figure 1). From a thermodynamic point of view, the PrpB2 reaction in direction of 2-methylisocitrate is favored (Δ_rG^′0 = −9.3 kJ/mol) in comparison to the SQO reaction leading to fumarate (Δ_rG^′0 = 14.5 kJ/mol). However, the potentially last step of the methylcitrate cycle catalyzed by PrpC2 is thermodynamically very unfavorable (Δ_rG^′0 = 39.2 kJ/mol), but might still work under in vivo conditions due to very low concentrations of oxaloacetate (<100 nM of lower detection limit of applied LC-QqQ MS). Indeed, Plassmeier et al. showed that the prpDBC2 operon is also expressed under normal cell growth conditions in wild-type C. glutamicum, independent of the application of propionate as carbon source (Plassmeier et al., 2007). Moreover, a high up-regulation of the prpDBC2 operon was found in a C. glutamicum strain engineered for overproduction of l-isoleucine (Ma et al., 2018). Here it was speculated that this up-regulation was due to a high intracellular formation of propionyl-CoA, which was derived from the intermediate 2-ketobutyrate of the l-isoleucine biosynthesis. The resulting propionyl-CoA was then further converted into a polyhydroxyalkanoate by heterologous expression of the phaCAB gene cluster. In our case, however, we can exclude propionyl-CoA formation from 2-ketobutyrate because l-aspartate is limiting. Additionally, it remains open what happens with propionyl-CoA as the second product of the PrpC2 reaction when running in reverse direction (Figure 1). Therefore, further investigations on the interplay between the different pathways are required to substantiate our hypothesis.

Moreover, we found a SNP in the intergenic region between cg3314 and cg3315, four nucleotides upstream from the translation start of MalR (malR, cg3315). This protein has been originally identified as repressor of the malE gene in the study of Krause et al. (2012) and was recently reported to bind to several other loci (Hünnefeld et al., 2019). Since the mutation occurred outside the coding sequence of this regulator, only expression changes of the gene are able to affect metabolism. Unfortunately, the specific protein data for MalR was not very accurate (p > 0.5) to enable any conclusion, but ME appeared to be down-regulated in the evolved Δppc Δpyc strain when grown on d-glucose as sole carbon and energy source (0.54-fold, p < 4.73e-10). Since purified ME of C. glutamicum was shown to carboxylate pyruvate in in vitro assays with an apparent K_m constant of 13.8 mM (Gourdon et al., 2000) it was speculated that ME serves as anaplerotic enzyme under circumstances when PEPCx and PCx activities are absent and pyruvate availability is still ensured through running glycolysis. In our case, however, the amount of ME was significantly reduced with respect to the wild type and its intracellular substrate pool of pyruvate is not significantly altered and well below the K_m value (Figure 4A). Therefore, it can be excluded that ME catalyzes a flux compensating for the missing PCx and PEPCx activity in the evolved Δppc Δpyc strain.

Finally, a deletion was detected in the mctC gene (cg0953), which has been shown to be essential in C. glutamicum for the uptake of pyruvate (Jolkver et al., 2009). Based on this, it can be hypothesized that the genetic alteration in this transporter may lead to a decreased uptake of pyruvate from the medium, which would explain the extracellular accumulation of pyruvate exclusively found in the Δppc Δpyc mutant (Figure 4B).