Microbial cell factories have become a vital tool for producing a wide range of valuable compounds, including recombinant proteins, bulk and fuel chemicals, pharmaceuticals, medical foods and dietary supplements. One such example is human milk oligosaccharides (HMOs): a family of structurally diverse unconjugated glycans that, after lactose and fat, collectively compose the third largest component of mothers’ milk (Bode, 2012). HMOs are uniquely diverse and abundant in human milk compared to that of domesticated animals (Urashima et al., 2013) and are of particular interest due to their beneficial impact on infants’ and adults’ health. Clinical and preclinical data demonstrate that HMOs support gastrointestinal health, immune function and cognitive development (recently reviewed by Hill May 2021; Sprenger et al., 2022; Dinleyici et al., 2023). They have therefore been introduced into both infant formula as well as into dietary supplementation products for adults (Bych et al., 2019; Lu, Mosleh, and Abbaspourrad, 2021; Pérez-Escalante et al., 2022). The most abundant HMO in human milk, 2’-O-fucosyllactose (2’FL), is currently produced on a commercial scale by fermentation using predominantly E. coli as production host (Ammann, 2017; Vandenplas et al., 2018; Bych et al., 2019). These fermentations are fully aerobic and employ a fed-batch feeding strategy with carbon limitation to avoid overflow metabolism, reduce the oxygen transfer rate demand and to maximize the carbon source utilization (using either glucose, sucrose or glycerol as carbon source) (Krause, Neubauer, and Neubauer, 2016). Unlike batch growth, where cells have unlimited carbon source availability, a fed-batch regulates the availability such that the respiratory metabolism of the cells and oxygen transfer rate capacity of the process are not overwhelmed (Enfors and Häggström, 1999; Srivastava and Gupta, 2011). This will otherwise result in overflow metabolism or anaerobic metabolism and the formation of undesirable metabolic by-products, predominantly in the form of acetate in E. coli (Krause et al., 2016; Millard et al., 2021; Li et al., 2022). Not only does acetate production constitute a wasteful use of the carbon source, but acetate also has a toxic effect that hinders growth (Pinhal et al., 2019), decreases the stability of intracellular proteins and limits product yields and titers (Eiteman and Altman, 2006; De Mey, De Maeseneire, et al., 2007). Such impacts have been observed in E. coli, even at relatively low acetate concentrations (Roe et al., 2002; Warnecke and Gill, 2005). Despite using a carbon-limited mode of operation during manufacturing, several circumstances in the industrial setting can lead to exposure to elevated sugar concentrations and induce overflow metabolism. The longer mixing times encountered in large-scale commercial vessels lead to the formation of local zones with significantly elevated concentrations around the feed inlet points and concomitant localized acetate formation (Larsson et al., 1996; Enfors and Häggström, 1999; Enfors et al., 2001; Srivastava and Gupta, 2011). Limitations to the feed control system can also cause unintended exposure to elevated sugar concentrations (Larsson et al., 1996). Finally, if cells grow slower than expected, the feed profile can exceed the threshold of the specific sugar uptake rate that induces overflow metabolism. Consequently, a slight mismatch between growth and feed rate can lead to a negative spiral of growth inhibition and acetate accumulation that quickly result in a lost fermentation batch (Chrapkova et al., 2023). This phenomenon has been observed multiple times with different HMO producing E. coli strains at the dsm-firmenich manufacturing site in Esbjerg. It is therefore desirable to have a production strain that is resistant to sugar exposure as it leads to more robust processes. Various genetic engineering strategies have been explored to achieve this. However, while mixed-acid metabolism under anaerobic conditions is well-understood (Clark, 1989; Förster and Gescher, 2014), the phenomenon of overflow metabolism remains more enigmatic despite decades of research. This has increased the difficulty in devising a successful metabolic engineering strategy. Several hypotheses have been proposed to explain the occurrence of acetate overflow metabolism despite it being less energy-efficient compared to respiration. This includes more cost-efficient proteome allocation in fermentation compared to respiration during faster growth rates (Basan et al., 2015), limited capacity of respiration resulting from space limitation in the plasma membrane where the electron transport chain is located (Andersen and von Meyenburg, 1980; Zhuang, Vemuri, and Mahadevan, 2011) and saturation of the tricarboxylic acid cycle (TCA cycle) and/or electron transport chain with sugar influx that surpasses the cells’ biosynthetic demand (De Maeseneire et al., 2006; Vemuri et al., 2006). Several studies have reported on engineered E. coli strains with a reduced (Farmer and Liao, 1997; Lin et al., 2006; De Mey, Lequeux, et al., 2007; Lara et al., 2008; Parimi et al., 2017) or abolished (Fuentes et al., 2013) acetate overflow during unrestricted growth. However, unrestricted growth produces a very different physiology compared to carbon limited growth. In addition, industrial strains are often designed such that the expression of the production machinery is coupled to changes in the extracellular environment or growth (reviewed in (Zhu et al., 2012)). A typical example is the use of promoters that are catabolically repressed during non-limited growth but fully expressed and only fully burdening the cell machinery in the carbon-limited production phase (Srivastava and Gupta, 2011; Maury et al., 2018; Pedersen and Papadakis, 2019). Since cells in this phase are in a very different physiological state, with a different metabolic burden, transitioning from carbon limitation to carbon excess can lead to a very different physiological reaction than what can be predicted from data acquired from cultures acclimatized to unrestricted growth. It is therefore useful to include such experiments when evaluating a strain’s robustness.

This study evaluated how three different metabolic engineering strategies previously shown to reduce acetate accumulation in E. coli affected the acetate formation, biomass growth and 2’FL production during either unrestricted growth on glucose or in a glucose-limited process exposed to glucose shock. The three strategies are described below and in Figure 1:

Strategy 1 - Alterations of direct pathways to and from acetate (Dittrich et al., 2008; Phue et al., 2010; Tao, Cheng, and Kopatsis, 2012). This included the deletion of poxB—encoding pyruvate oxidase that catalyses the direct conversion of pyruvate to acetate, deletion of pta—encoding phosphate acetyltransferase, the first enzyme in the acetyl-CoA to acetate pathway, deletions of adhE and adhP, encoding two alcohol dehydrogenases involved in ethanol formation from acetyl-CoA and the concomitant oxidation of 2 NADH molecules. While not directly involved in acetate formation, NAD⁺ is needed for acetyl-CoA formation from pyruvate, which could potentially become a limiting factor under conditions when NADH oxidation via oxidative phosphorylation is a bottleneck, such as during overflow metabolism. Reduction of alternative ways to oxidize NADH could thus reduce the acetyl-CoA formation rate and thereby acetate formation. Finally, overexpression of acs encoding acetyl-CoA synthetase, a high affinity ATP driven acetate scavenging enzyme, that converts acetate into acetyl-CoA (Lin et al., 2006).

Strategy 2 - Mutations that channel the metabolic flux away from the acetate precursors pyruvate and acetyl-CoA and into the TCA cycle (De Maeseneire et al., 2006; De Mey, Lequeux, et al., 2007; Zhu et al., 2019). This included overexpression of gltA encoding citrate synthase, which constitutes the main entry point into the TCA cycle, converting acetyl-CoA and oxaloacetate into citric acid, overexpression of ppc that encodes phosphoenolpyruvate carboxylase that is the second entry point into the TCA cycle converting PEP and CO₂ into oxaloacetate and which also reduces the availability of PEP which is driving the glucose uptake via the PTS system, deletion of pckA encoding phosphoenolpyruvate carboxykinase, catalyzing the reverse reaction from oxaloacetate into PEP, deletion of iclR, a negative regulator for the glyoxylate bypass. Deletion of iclR has been shown to increase the flux through the glyoxylate shunt and increase the overall TCA capacity (Waegeman et al., 2011).

Strategy 3 - Reducing the maximum glucose uptake rate below the threshold that triggers overflow metabolism (Chen et al., 1997; Cho et al., 2005; De Anda et al., 2006). This was attempted by the deletion of the enzyme IIBC components in the PTS system for glucose (encoded by ptsG) and N-acetylglucosamine (encoded by nagE), which is also able to transport glucose and contributes to growth in ΔptsG genotypes (Fuentes et al., 2013).

These genetic modifications and their combinations were introduced into an industrial 2’FL producing strain (C1) with a mixed-acid negative phenotype that prevent anaerobic metabolism due to an imposed redox imbalance (ΔpflB, ΔldhA and ΔfocA genotype). Together these knockouts prevent the E. coli cells from passing the pyruvate to acetyl-CoA node without concomitant NAD⁺ to NADH reduction, and from oxidizing any produced NADH by converting pyruvate into lactate as terminal electron acceptor. Therefore, without oxygen available as electron acceptor, these cells quickly run low on NAD⁺ which halts the metabolism (Clark et al., 1988; Mat-Jan and Clark, 1989; Lara et al., 2006; Wu et al., 2007; Singh, Lynch, and Gill, 2009). After an initial screening in microplates, selected strains were tested in bioreactors in a high-cell density, glucose-limited fed-batch process and challenged by a large bolus addition of glucose to trigger E. coli’s overflow metabolism.

The tendency of E. coli to revert to overflow metabolism has long been a challenge for biotechnological applications and it has therefore remained an attractive target for metabolic engineering (reviewed by Eiteman and Altman, 2006; De Mey; De Maeseneire et al., 2006) Many strategies have been proposed and implemented, but with the focus on the effect during prolonged batch growth. Here we compare the effect of three distinct approaches and some combinations thereof in a commercial production strain: 1) alterations to acetate pathways, 2) increased flux towards the TCA cycle, and 3) reduced glucose uptake rate. The strain used for the study redirects a large proportion of the glucose carbon source towards 2’FL production, which exhorts an additional metabolic load adding to the overall stress level, something not typically taken into account in the published studies. In addition to batch growth, we also show the effect of the genetic modifications when transitioning from carbon limited to unrestricted growth, something hitherto unreported. This is especially relevant in industrial fed-batch and continuous fermentations which are utilizing carbon limitation but are exposed to local sugar gradients at the feed inlet zones or could be exposed to sudden increases in sugar concentration due to constraints in the manufacturing setup (Larsson et al., 1996).

To ensure that the genetic modifications did not compromise the general performance of the strains, we first screened them in a microtiter plate assay. The only strains that showed a significant decrease in 2’FL production were the ones with the ptsG deletion (Table 1). The iclR deletion strains, on the other hand, exhibited an enhanced 2’FL yield. However, these results were not confirmed in subsequent fed-batch fermentations with selected strains under glucose-limited conditions. Neither the ptsG nor the iclR deletion had any measurable effect on growth or 2’FL yield in the bioreactors under controlled fermentation conditions. We attribute the discrepancy between the microtiter plate and the bioreactor experiments to two factors: the inoculation ratio and the pH control. The strains carrying the ptsG deletion had a lower initial biomass in the microtiter plate assay due to their reduced growth rate in the previous step and the lack of OD₆₀₀ normalization in the inoculation step. This could have influenced the outcome of the assay. The iclR deletion strains on the other hand might have been less impacted by the lack of pH control in the microtiter plate assay compared to the other strains. While the strains mixed-acid negative phenotype is very effective during anaerobic conditions (Clark et al., 1988; Mat-Jan and Clark, 1989; Singh, Lynch, and Gill, 2009), we have observed that acid formation occur under microaerobic conditions, probably due to restoration of some of the redox imbalance via respiration while still maintaining a fermentative metabolism. The microtiter experiments turn microaerobic over time and due to the lack of a pH controller system the broth therefore acidify over time. This eventually leads to growth inhibition and cessation of product formation. The iclR deletion mutants all showed reduced glutamate accumulation during the glucose stress test (Supplementary Figure S3). If the deletion also affected acid formation in the microplate assay it could have delayed the acidification of the medium and thereby prolonged the productive phase and increased the 2’FL titer.

In contrast to the fed-batch test, the batch tests produced marked differences in both growth rate and acetate yield for some of the genetic modifications. This was seen in both the microtiter assay and the bioreactors. There were two combinations with especially noteworthy synergistic effects. The first was the combined disruption of pta and poxB. In the microtiter experiment, the disruption of the acetate pathway from acetyl-CoA by pta deletion reduced the growth rate and led to an increase in acetate formation (+12%), while the disruption of the acetate pathway from pyruvate via poxB deletion only resulted in a minor reduction in acetate accumulation (−23%). However, together they had a marked synergistic effect and produced a five-fold reduction (−80%). The second was the combination of gltA overexpression and iclR deletion. Overexpression of the gltA gene has previously been shown to effectively eliminate acetate production by intensifying the flux through the TCA cycle (De Maeseneire et al., 2006) and to increase anaplerotic reactions (Im et al., 2020). Although gltA overexpression did not reduce acetate to such an extent in our experiments (−13% and −44% in microplate assay and bioreactor, respectively), its overexpression together with the increased uptake capacity in the TCA cycle with the iclR deletion (Waegeman et al., 2011), which by itself had a negligible effect, produced the overall biggest reduction in acetate accumulation of any strain in the bioreactor batch phase (−80%). In contrast, the combination of the iclR deletion with the pta and poxB deletions did not produce a significant difference. The effect of ΔiclR indicated that the capacity of the TCA cycle had become a bottleneck when the entry point (gltA) was overexpressed. By creating a shortcut in the glyoxylate shunt, this limitation was relieved and the overall capacity from acetyl-CoA to isocitrate increased and more flux was channelled away from acetate formation (Figure 1). The underlying reason could either be a direct rate limitation in one of the enzymatic steps in the TCA cycle, or a limitation in respiration and its ability to oxidize NADH/FADH₂, a demand that is lessened if the glyoxylate shunt is increased as fewer NADH equivalents are produced (Waegeman et al., 2011).

The glucose pulse addition experiment constituted a stress test with a direct transition from glucose limited growth with its characteristic physiology to unrestricted growth - without any adaptation time. Although none of the modifications eliminated overflow metabolism from the freely available glucose, strains with specific genetic modifications accumulated much less acetate and showed an increased robustness. The reduced acetate formation rate in these strains seems to have prevented the accumulation from reaching a critical level, which would otherwise cause a viscous cycle of ever-increasing acetate production and growth inhibition (Millard et al., 2021). In contrast, cultures lacking these modifications kept accumulating acetate until they ceased to grow and produce (Figure 2). Furthermore, all strains that survived the glucose pulse showed an increase in pH, indicating the reassimilation of the accumulated acetate (Figure 3). These findings reflect a previous study that demonstrated that engineered E. coli strains with a reduced acetate phenotype can reduce fermentation failures caused by acetate accumulation (Chong et al., 2013).

Interestingly, in the bioreactor batch experiment the TCA8 strain with gltA overexpression and iclR deletion accumulated significantly less acetate than the Δpta ΔpoxB strains AP6 and APTCA2. This relation completely shifted in the pulse experiment where the AP6 and APTCA2 strains accumulated much less acetate than TCA8. This discrepancy shows the importance of considering the prior physiological state when trying to predict a reaction to a particular change. In this case, it meant that learnings of how particular genetic modifications affected acetate formation during unrestricted growth in the batch phase could not be directly applied to the response to a glucose pulse in a fed-batch phase - even if the pulse led to unrestricted growth. When cells are exposed to such a sudden change, they do not have time to adapt, and therefore their previous physiological state will influence how they react (Vasilakou et al., 2020). The difference in physiology in carbon-limited versus unrestricted growth carried over and made the cultures react differently than what would be expected from the batch experiments alone.

Inactivation of the glucose component of PTS by deletion of ptsG to reduce the phosphoenolpyruvate (PEP)-dependent glucose transport rate was expected to be a very efficient way to reduce acetate overflow metabolism based on previous reports (Sigüenza et al., 1999; De Mey, De Maeseneire, et al., 2007; Lara et al., 2008; Fuentes et al., 2013). E. coli is known to possess several alternative transporters for glucose uptake (Boos and Howard, 1998; Hua et al., 2004), including alternative PTS system for mannose (encoded by manXYZ) and GlcNAc (encoded by nagE) that can take up glucose at a reduced rate (Fuentes et al., 2013; Crigler et al., 2018), as well as non-PTS systems such as galactose permease (encoded by galP), galactose ABC transporter (encoded by mglBAC) or maltose ABC transporter (encoded by malEFG) (Boos and Howard, 1998; Hua et al., 2004; Fuentes et al., 2013; Crigler et al., 2018). Since these alternative transporters operate at a lower rate, they should avoid overloading the TCA cycle or respiratory capacity and thereby prevent, or at least reduce, the accumulation of acetate from overflow metabolism. Surprisingly, the deletion of ptsG did not reduce the acetate yield in the microplate experiment or during batch growth in the bioreactor, and it did not reduce acetate and prevent culture failure during the stress test with the glucose pulse. A lower glucose consumption rate was observed in the mutant, indicating a reduced glycolytic flux and a successful knockout. In theory, this should have resulted in a low acetate overflow phenotype since the TCA cycle and the respiratory capacity had not been changed. This outcome, which is both counterintuitive and different compared to previous studies, is not straightforward to explain. A full genome sequencing could reveal if there were any unintended changes that was introduced during the genetic engineering work, but this was outside the scope of the work.

An alternative strategy to further reduce the flow to pyruvate (and, therefore, acetate) could be to completely knockout the PTS system and overexpress an alternative glucose transporter such as galactose permease (De Anda et al., 2006) or the maltose ABC transporter (Carreón-Rodríguez et al., 2023). In addition to slowing down the glucose uptake rate, non-PTS sugar transporters avoid the conversion of PEP to pyruvate during glucose transport (De Anda et al., 2006), which is a precursor for acetate. Such an approach was able to restore growth while maintaining reduced acetate secretion when galP was overexpressed in a ptsG mutant (De Anda et al., 2006). Though, it remains to be seen if such a strategy would be successful with an industrial 2’FL producing strain.

A recent study showed that overflow metabolism confers a competitive advantage over complete respiration (Rabbers et al., 2022) suggesting an evolutionary role for acetate cycling. Complete elimination of acetate formation may therefore inherently result in trade-offs, such as slower growth, the accumulation of other by-products, or sensitivity to other process conditions that only a very extensive rewiring of the E. coli metabolism might be able to solve.

The findings herein show that blocking the major acetate pathways and increasing the flux towards the TCA cycle can contribute to a reduced acetate phenotype for both regular batch cultures and for glucose limited cultures transitioning to non-limited growth. The genetic modifications were shown to enhance cell robustness, enabling an industrial strain to better tolerate processes exposed to dramatic changes in glucose concentration. The best performers in terms of 2’FL yield after a glucose shock were the strains that overexpressed gltA, and the ones carrying the combined pta, poxB deletions. The iclR deletion, while not effective on its own, reduced acetate formation when combined with gltA overexpression, and it counteracted the tendency of glutamate accumulation caused by the gltA overexpression. Future work for the development of glucose gradient resistant production strains could explore these two strategies, as well as the combination of them (Δpta ΔpoxB with OEgltA and ΔiclR). It is also recommended to follow up with a study of the impact of repeated stress, such as in a pulse-fed scale-down reactor (Vasilakou et al., 2020), as this could have a different effect on the physiology and possibly differentiate the performance of the strains. Finally, as an attempt to understand why the ptsG deletion was not effective in reducing acetate formation in this work, a comparison of the metabolic and transcriptomic expression profiles of a ΔptsG strain with a low acetate phenotype as reported in previous studies would be interesting.