We applied a systems biology approach to study E. coli metabolic flux dynamics and possible growth limiting factors. Detailed carbon balance and biomass composition analysis were carried out in A-stat and D-stat cultures to examine the dynamic responses of metabolic fluxes and energy metabolism to change of μ and glucose-acetate co-utilisation capability. A simplified MFA was conducted to map the carbon flow through central carbon metabolism.
We detected a two-phase acetate accumulation profile in A-stat which started at μ = 0.27 ± 0.02 h-1 (Figure 1). After linear increase, acetate probably starts to accumulate exponentially because of total repression of acetate consuming enzyme, acetyl-CoA synthetase, by carbon catabolite repression . It became clear from MFA calculations that acetate excretion plays an important role in overall flux patterns and ATP metabolism. Firstly, start of acetate excretion reduces carbon flow from the PTA-ACS cycle to acetyl-CoA and central metabolism triggering reduction of TCA cycle fluxes (Figure 3) that can be also seen by decline of the proportion of CO2 and NADH produced by the TCA cycle (Additional file 3). This was paralleled with induction of PPP, possibly for NADPH regeneration, and reduction of glycolysis (Figure 3). These shifts have been observed in chemostat cultures as well [9, 12–14, 23]. Secondly, disruption of the PTA-ACS node resulting in acetate overflow strongly reduced ATP spilling (non-growth associated ATP production) which declined 32% with increasing of μ (Figure 4). What is more, slope of the overall ATP production rate changed right after overflow switch (Additional file 4). It can be concluded from the latter that acetate metabolism is one of the major regulating factors of central carbon metabolism which is in good agreement with abundance of literature data.
Decrease in ATP spilling (32%) after overflow switch in A-stat was shown by MFA calculations (Figure 4) and decline of the ATP-spending PTA-ACS node throughput by acetate excretion (Figure 2A). This response in energy metabolism was detected in this study since the futile PTA-ACS cycle was included into the model network. Additionally, change in ATP production rate was also seen with increasing μ (Additional file 4). We have to note that the possibility of our ATP calculations being distorted due to carbon imbalance at higher specific growth rates (Figure 5) cannot be excluded which could lead to underestimation of ATP production. However, this seems rather unlikely since CO2 measurement precision is constant for all μ. Furthermore, theoretically no other pathway besides the TCA cycle cannot by far produce enough energy under these carbon imbalance conditions so that decrease of ATP spilling would not be observed. We additionally have to point out that our calculations could also overestimate ATP production since a theoretical ratio for ATP generation efficiency in oxidative phosphorylation (P/O = 2) was chosen which can be higher than the in vivo value. In any case, the P/O ratio per se does not affect the main conclusions of the manuscript since the ratio between ATP spilling and its overall production is independent from the P/O ratio value.
Decrease in ATP spilling (40 mmol/g dry cellular weight (DCW)) might indicate increase of YXS, however, it remained constant in our experiments (Figure 4). This apparent discrepancy between the decrease in ATP spilling and constant YXS (Figure 4) could be explained by the fact that carbon wasting increases from 3 to 11% with rising μ (Figure 6) as follows. As the acetate synthesis/assimilation PTA-ACS is a futile cycle, an equivalent amount of ATP to acetate is concomitantly wasted with production and re-consumption of acetate. Therefore, accumulation of acetate likely triggers a 32% decline of ATP spilling (Figure 4) since re-consumption of acetate (wasting 1 molecule of ATP) decreases with rising μ after overflow switch. This energy save is, however, counteracted by the increase of carbon wasting in the carbon balance from 3 to 11% which results in a constant YXS. However, E. coli might possess additional mechanisms to maintain a constant YXS under increasing carbon wasting conditions during μ increase.
In addition to metabolic flux dynamics, we described novel carbon wasting profiles in E. coli K-12 MG1655 into pyrimidine pathway intermediates orotate, DHO, CBASP, and NAA with rising μ (Figure 2) and under various glucose-acetate co-utilisation capabilities (Table S5 in Additional file 1). Excretion of orotate [18, 19], DHO , CBASP  and NAA  by E. coli has been noted before. Accumulation of the pyrimidine pathway compounds-orotate, DHO and CBASP-can be explained by the E. coli K-12 MG1655 genotype. This specific strain is prone to pyrimidine starvation due to a rph frameshift mutation leading to low pyrE (encodes PyrE protein which catalyses orotate conversion into orotidine-5-phosphate) expression  which could possibly lead to accumulation of precursor molecules which all the latterly mentioned compounds are (Additional file 5). Excretion of a considerable amount of CBASP, DHO, orotate and NAA besides acetate shows that overflow metabolism actually consists of more products than acetate, as generally believed. Detailed by-product measurements enabled us to precisely detect carbon outflow routes for MFA calculations which usually are taken into account predictively either from pyruvate, oxaloacetate, α-ketoglutarate or other potential precursors. For instance, if these product outflows will be excluded from MFA and carbon wasting predicted as pyruvate outflow, pyruvate dehydrogenase, TCA cycle and PEP carboxylase fluxes would deviate by 11, 24 and 60%, respectively, at μ = 0.47 h-1 from the values calculated by our model (Table S6 in Additional file 1). Although the absolute amount of these excreted substances in the carbon balance is not substantial (less than 5%), linking their accumulation dynamics to μ (or metabolic routes) is relevant for acknowledging the potential imbalance between pyrimidine metabolism, carbon re-consumption and ATP spilling.
DHO and CBASP accumulated increasingly up to the start of acetate overflow (Figure 2A). After overflow switch, DHO started to decline whereas orotate and CBASP levelled off suggesting that carbon flow from the PTA-ACS cycle to acetyl-CoA and central metabolism declines indicated by carbon loss to excreted acetate and decreasing TCA cycle flux patterns (Figure 2A & 3). Finally, orotate and CBASP levels started to increase again simultaneously (Figure 2A) with the sharp decrease of cAMP and faster accumulation of acetate (Figure 1). Firstly, this rise could be explained by the high demand for RNA synthesis at higher μ which leads to precursor molecule accumulation because of the low pyrE expression. On the other hand, pyruvate dehydrogenase flux reached its maximum at μ = 0.42 h-1 with concomitant slight increase in glycolysis fluxes that subsequently resulted in accelerated carbon wasting (Figure 3). These observations demonstrate a strong link between overflow metabolism of acetate and carbon wasting into other products.
We validated and quantified, for the first time to our knowledge, excretion of NAA which levels constantly increased with μ in A-stat experiments (Figure 2B). Neither has NAA yet been registered in EcoCyc Ver 15.0  nor there exists a protein catalysing its synthesis for E. coli K-12 MG1655 in KEGG Release 58.0 . Based on homology analysis with the available aspartate N-acetyltransferase protein sequences (human and mouse), we propose that a predicted acetyltransferase YjgM could catalyse the formation of NAA in E. coli (Additional file 6). This is supported by the fact that YjgM is expressed within the studied μ range . We hypothesise that since oxaloacetate is over-produced in the TCA cycle and this excess carbon cannot be shunted towards PPP because of the discovered 'CBASP-DHO-orotate' bottleneck, the excess carbon is excreted as NAA.
To our knowledge, dilution rate dependent carbon wasting profiles in terms of two-substrate (glucose and acetic acid) co-utilisation has not been studied before. We found that carbon wasting patterns are dependent not only on μ as shown by A-stat, but also on glucose-acetate co-utilisation capability (Figure 7). This finding could be useful for bioprocess development since mixed-substrate growth is commonly used there and loss of carbon is unwanted. What is more, overall carbon wasting into the carbon wasting products in the carbon balance was similar (ca 5.5%) under very different maximal glucose-acetate co-utilisation capability values (Figure 7). The latter implies that the quantitative carbon flow through the carbon wasting pathways does not depend on the consumption of additional (to glucose) carbon sources, such as acetate in our case.
Overall carbon wasting in the carbon balance increased with rising μ (Figure 5 & 6). Carbon wasting to other substances than acetate e.g. orotate, DHO, CBASP, NAA may be caused by an imbalance between the supply of precursors, intermediates of central metabolic network and insufficient use of them for the synthesis of biomass monomers and macromolecules which could be the result of energy limitations. Levels of metabolites from the upper part of energy generating glycolysis, fructose-1,6-bisphosphate and glyceraldehyde-3-phosphate, and TCA cycle components increased with the rise in μ , indicating no limits in energy supply at these steps. However, there might be competing futile cycles wasting ATP, and leading to energy limitations, for instance in glycolysis  and the PTA-ACS cycle proposed by us . In addition to the pyrimidine synthesis pathway compounds-orotate, DHO and CBASP-, we observed TCA cycle intermediates α-ketoglutarate and isocitrate accumulation, and decline in intracellular ribose-5-phosphate and erythrose-4-phosphate metabolite levels, which are also intermediates of pyrimidine synthesis . This might result in growth limitation by RNA synthesis and ribosome assembly, especially at higher μ. The latter is supported by the fact about the genotype of E. coli K-12 MG1655 (rph frameshift mutation leading to low pyrE expression) which leads to pyrimidine starvation as described above. The latter proposal is in agreement with RNA concentration measurements which showed that RNA amount increased together with μ until 0.40 h-1 after which it levelled off (Table S1 in Additional file 1). The possible limitation of maximal μ by RNA synthesis and carbon wasting due to the E. coli K-12 MG1655 genotype revealed by this study proposes a way how to increase maximal μ and YXS which is relevant for the biotechnology industry. Furthermore, all the data referred above show that the details of the regulatory mechanisms of cellular growth need further studying using comprehensive systems biology approaches.