Skip to main content
NCBI home page
Microbial Biotechnology logoLink to Microbial Biotechnology
. 2024 Nov 15;17(11):e70051. doi: 10.1111/1751-7915.70051

Combined oxygen and glucose oscillations distinctly change the transcriptional and physiological state of Escherichia coli

Jonas Bafna‐Rührer 1,✉, Jean V Orth 1, Suresh Sudarsan 1,✉
PMCID: PMC11568247  PMID: 39548707

Abstract

Escherichia coli, a common microbial host for industrial bioproduction, experiences a highly dynamic environment in industrial‐scale bioreactors due to significant glucose and oxygen gradients. In this study, we mimic the combined gradients of glucose and oxygen in high‐throughput bioreactors to study the transcriptional response of E. coli to industrial‐scale conditions. Under oscillating oxygen conditions, E. coli formed less biomass and accumulated the anaerobic by‐product acetate. With respect to oxygen‐responsive genes, we found that genes of the TCA cycle and of different electron transport chain complexes were differentially expressed. A global analysis of the expression data revealed that oxygen oscillations had caused a transition towards a catabolite‐repressed state and upregulation of several stress‐related regulatory programs. Interestingly, the transcriptional changes persisted after oxygen limitation stopped. In contrast, the changes we observed due to glucose starvation, such as induction of the stringent response, were primarily transient. Most importantly, we found that effects of combined oxygen and glucose oscillations were distinct from the ones of oxygen and substrate oscillations alone, suggesting an important interplay between the different metabolic regimes in industrial‐scale bioreactors.

Understanding microbial response to industrial‐scale conditions is imperative to achieve overall process robustness at industrial‐scale. Here, we studied E. coli's response to changing glucose and oxygen levels. We report lower biomass, higher acetate production, and key regulatory mechanisms that were altered under repeated oxygen limitation conditions. Combined glucose and oxygen shifts caused unique responses, highlighting the complexity of microbial behaviour in industrial‐scale production environments.

graphic file with name MBT2-17-e70051-g002.jpg

INTRODUCTION

Industrial fermentations employ carbon‐limited fed‐batch strategies to avoid undesired phenomena such as oxygen limitation, overflow metabolism, overheating of the culture, and catabolite repression (Neubauer et al., 1995). However, due to slow mixing times of industrial‐scale bioreactors, typically in the range of 1–4 min (Enfors et al., 2001; Lara et al., 2006; Vrábel et al., 2000), the feeding of a limiting substrate, usually at the top of the bioreactor, results in the formation of substrate gradients, especially at high‐cell‐density. Additionally, due to the oxidation of substrate in aerobic processes, dissolved oxygen gradients (high at the bottom, low at the top) opposite to the ones of substrate (high at the top, low at the bottom) are formed. Oxygen gradients are further intensified due to the difference of hydrostatic pressure in large bioreactors (Lara et al., 2006). Microbial cells that travel through industrial‐scale bioreactors can experience a highly dynamic environment depending on their current location, with sufficient substrate yet limiting oxygen at the top and complete substrate depletion causing starvation at the bottom of the reactor. When oxygen supply is limited the full oxidation of all available substrate (most commonly glucose) is not possible. Under these conditions, Escherichia coli, a common bacterial host to produce biomolecules and biochemicals, switches to a partially anaerobic metabolism, converting glucose into acetate and other fermentative by‐products (Anane et al., 2019; Salmon et al., 2005; Xu et al., 1999). When faced with substrate starvation, E. coli initiates the stringent response as a defensive strategy (Hardiman et al., 2007). Not only the different metabolic regimes per se but also the frequent switching between them can impact E. coli's growth physiology and transcriptional regulation, which ultimately can cause a decrease in growth and productivity of the industrial fermentation process (Enfors et al., 2001; Neubauer et al., 1995; Ziegler et al., 2021).

In recent years, RNA‐sequencing (RNA‐seq) has emerged as a key technology to understand the transcriptional regulation mechanisms of microbes under relevant bioprocess conditions. Gene expression data is typically analysed by differentially expressed genes (DEGs), which often yields a large number of affected genes, making the deduction of gene regulation mechanisms a complex task. As an alternative to DEG analysis and manually annotated Regulons, i.e., groups of genes thought to be under the control of the same transcription factor, a novel approach applied independent component analysis (ICA) on a compendium of high‐quality RNA‐seq datasets from E. coli grown under different conditions (Lamoureux et al., 2023; Sastry et al., 2019). As a result, independent components consisting of genes that are independently modulated in E. coli were identified (Rychel et al., 2021). These independent components are called iModulons. Regulons and iModulons both describe groups of coregulated genes (Lim et al., 2022; Rychel et al., 2020; Sastry et al., 2019). However, iModulons have weights associated with each gene and activity levels associated with each RNA‐seq sample (Lamoureux et al., 2023; Sastry et al., 2019). This allows to calculate iModulon activities for new RNA‐seq samples based on the iModulon gene weights and to quantitatively compare multiple RNA‐seq samples on the level of the transcriptional regulatory network (TRN) (ca. 200 iModulons) instead of the level of gene expression (ca. 4000 single genes) (Lim et al., 2022; Rychel et al., 2020, 2021). In the latest version of the Precision RNA‐seq Expression Compendium for Independent Signal Exploration, which consists of more than 1000 E. coli RNA‐sequencing samples (PRECISE‐1K), in total, 201 iModulons and their activity levels across the compendium were identified (Lamoureux et al., 2023). For the new RNA‐seq data in this study, the activity levels of these 201 iModulons were inferred from PRECISE‐1K, enabling a systems‐level view of the affected biological and regulatory functions of the TRN under the tested conditions. Recently, iModulon analysis has been successfully applied to provide a systems‐level understanding of the TRN response of E. coli and Yarrowia lipolytica when subjected to repeated glucose oscillations (Bafna‐Rührer et al., 2024) and industrial O2 and CO2 stresses (Kerssemakers et al., 2024).

To predict the process performance and study microbial physiology under industrial‐scale conditions, scale‐down simulators that mimic relevant process phenomena in lab‐scale experiments are employed (Neubauer & Junne, 2010). Scale‐down simulators typically rely either on the spatial or the temporal separation of different metabolic regimes. Spatial separation of the metabolic regimes of interest is achieved by multi‐compartment scale‐down simulators such as the stirred tank reactor (STR)—plug flow reactor (PFR) setup, while temporal separation can be achieved by oscillating one or more process control parameters (e.g., feed, aeration) (Neubauer & Junne, 2010). Several studies have applied scale‐down simulators to study the effects of substrate (Brand et al., 2018; Löffler et al., 2016; Neubauer et al., 1995; Vasilakou et al., 2020) or combined substrate and oxygen gradients (Anane et al., 2019; Enfors et al., 2001; Janoska et al., 2022). However, conflicting results obtained from different scale‐down setups have highlighted that a detailed understanding of the industrial‐scale bioreactor is essential to design realistic scale‐down experiments that provide representative results (Wang et al., 2018). Model representations of industrial‐scale bioreactors can help to obtain good predictions of the metabolic regimes encountered by microbial cells and guide the design of realistic scale‐down experiments (Haringa et al., 2016; Kuschel & Takors, 2020; Wei et al., 2023). Compartment models present a less computationally expensive alternative to computational fluid dynamic (CFD) models (Tajsoleiman et al., 2019). They have recently been used to simulate the formation of mixing gradient during fed‐batch fermentations with changing volume (Nadal‐Rey et al., 2021) and can also be used for the rational design of scale‐down experiments (Kerssemakers et al., 2023).

In this study, we demonstrated the design of scale‐down experiments from compartment model simulations of a representative process scenario to examine the effects of combined substrate and oxygen gradients on E. coli. In the oscillation‐based scale‐down experiments, we also investigated the isolated effects of substrate and oxygen oscillations, testing if the observed effects of combined oscillations are due to either substrate or oxygen oscillations. We applied DEG and iModulon analysis to RNA‐seq data from the scale‐down experiments to study E. coli's TRN response to substrate and oxygen oscillations.

RESULTS

Oxygen oscillations lower the biomass yield of E. coli

To investigate the effects of combined substrate (here glucose) and oxygen gradients, we chose a representative industrial‐scale aerobic fed‐batch fermentation resulting in both anaerobic metabolism at the top of the reactor and substrate starvation at the bottom of the reactor. We used a compartment model of a 90 m3 stirred tank bioreactor with 4 Rushton turbines (Nadal‐Rey et al., 2021) and combined it with an adapted black box kinetic metabolic model (Xu et al., 1999) to simulate substrate and oxygen gradients (Figure 1A,B). The glucose concentration decreased rapidly from the feeding inlet at the top of the reactor, where the concentration was estimated to be 0.12 g L−1 and decreased to concentrations below the assumed K S value of 0.05 g L−1 after ca. 0.5 m (Figure 1A). Based on the simulation, the formation of three metabolic regimes, i.e., substrate starvation, substrate limitation, and oxygen limitation, was predicted (Figure 1C). While the mid‐section was in the substrate limitation regime, the top part of the bioreactor exhibited oxygen limitation, and the bottom part was in substrate starvation due to improper mixing of the glucose feed. Frequent transitions between different metabolic regimes, as experienced by cells travelling in the bioreactor, were mimicked with a scale‐down design based on oscillations of glucose feeding and aeration (Figure 1D).

FIGURE 1.

FIGURE 1

Open in a new tab

Compartment‐model‐based scale‐down experiments reveal reduced carbon efficiency under oxygen limitations. Left panel: Metabolic regimes in a 90 m3 bioreactor. (A) substrate and (B) dissolved oxygen concentration gradients throughout the bioreactor. Concentration values represent average values at the bioreactor height. (C) The concentration gradients translate into a metabolic regime distribution. Arrow points to feed inlet point. For detailed description of E. coli metabolic regimes, refer to Figure S1, Table S2. (D) Scaling‐down of metabolic regimes by oscillating the process control parameters feeding and aeration. (E) To isolate the effect of substrate gradients, a scale‐down only based on feeding oscillations was designed. (F) To isolate the effect of oxygen gradients, a scale‐down only based on aeration oscillations was designed. Right panel: Fermentation results of E. coli wild‐type cultures tested under control conditions (no oscillations), substrate oscillations (S), oxygen oscillations (O), and combined substrate and oxygen oscillations (S + O). (G) Net amounts of formed biomass, (I) formed carbon dioxide, and (K) formed acetate are shown in carbon moles (molC) over time of the scale‐down feeding phase. (H) Yields of biomass YSX, (J) carbon dioxide YSC, and (L) acetate YSA on consumed substrate (glucose) measured at the end of each fermentation are shown for all four fermentation conditions. Error bars show standard deviation of biological triplicates.

To isolate the effects of substrate and oxygen gradients, two additional oscillation profiles were designed, where the metabolic regime switches between substrate limitation and starvation (Figure 1E) and between oxygen and substrate limitation (Figure 1F). During the scale‐down phase of the fed‐batch fermentations, the oscillation profiles were repeated continuously every 10 min. The control parameters used to create stimuli of glucose and/or oxygen (feed rate, airflow, and oxygen flow), as well as the dissolved oxygen (DO) signal in response to the stimuli, are shown for a snapshot time window in Supporting Information (Figure S5–S8). Glucose concentrations were only measured with a focus on the long‐term trend, while frequent sampling to resolve concentrations during a single stimulus was neglected. In the cultures with glucose stimuli, the glucose concentrations during a single stimulus were low and around the KS value (here around 0.05 g L−1) because the feed rate during a glucose pulse unlikely caused excessive glucose accumulation. For the cultures with combined oxygen and substrate stimuli, the glucose concentration likely also remained at limiting concentration levels around KS since DO increased almost immediately after feeding was switched off, indicating a fast consumption of the residual glucose (Figure S8). Compared to the control (non‐oscillating) cultures, where 0.56 ± 8.6 × 10−3 molC biomass was formed, substrate (S) oscillation cultures showed a slightly lower formed biomass (0.53 ± 4.7 × 10−3 molC) at the end of the cultivation (Figure 1G).

Both oxygen (O) oscillation and combined substrate and oxygen (S + O) oscillation cultures showed a substantial decline in formed biomass (0.11 ± 4.8 × 10−3 molC and 0.29 ± 7.8 × 10−3 molC, respectively), with the O oscillation cultures exhibiting a complete growth arrest. The calculated values of the biomass yield on consumed substrate YSX (Figure 1H) indicated the carbon efficiency of E. coli in the different culture types, that is, the relative fraction of carbon available from glucose converted into biomass. E. coli formed 0.53 ± 4.3 × 10−3 molC carbon dioxide (CO2) under repeated substrate oscillations, compared to 0.47 ± 3.1 × 10−3 molC in control cultures (Figure 1I). The highest overall CO2 formation was observed under combined substrate and oxygen oscillations (0.71 ± 23.1 × 10−3 molC). E. coli formed the least amount of CO2 during the scale‐down phase when subjected to repeated oxygen oscillations (0.33 ± 1.8 × 10−3 molC). However, due to substrate accumulation in the culture broth (Figure S11), the CO2 yield on consumed substrate YSC under oxygen oscillations was similar to the control and substrate oscillation cultures (Figure 1J). The major byproduct acetate of anaerobic/overflow metabolism in E. coli, accumulated to 0.3 ± 5.3 × 10−3 molC under repeated oxygen and to 0.07 ± 8.0 × 10−3 molC under combined oxygen and substrate oscillations (Figure 1K). At the end of the cultivation, acetate amounted to about 40% of all carbon products in O oscillation cultures (Figure 1L).

Oxygen oscillations significantly affect gene expression in the central carbon metabolism and electron transport chain of E. coli

To identify overall trends of gene expression, we performed principal component analysis (PCA) with the logTPM values of the RNA‐seq data of this study (Figure 2C). The first four principal components revealed the major contrasts of gene expression in our RNA‐seq data in descending order of intensity. Oxygen oscillations appeared to have the dominant effect on gene expression within the dataset of this study. The first principal component (explained variance 66.8%) majorly separated samples from the oxygen oscillation cultures from other samples. Note that the sampling timepoint (S1, S2) during one oscillation cycle did not impact gene expression in cultures of repeated oxygen oscillations. According to principal component 2 (PC2), combined substrate and oxygen oscillations resulted in gene expression distinct from the other culture types. PC3 (explained variance 6.8%) mostly explained the difference in gene expression between the sample timepoints S1 (substrate limitation; “feast”) and S2 (substrate starvation; “famine”) of S oscillation cultures. PC4 (explained variance 1.2%) predominantly separated the samples of S oscillation cultures from the control culture samples.

FIGURE 2.

FIGURE 2

Open in a new tab

Gene expression analysis of E. coli's response to substrate and oxygen oscillations. (A) Differential gene expression mapped against central carbon metabolism and (B) electron transport chain. The colour of the boxes indicates down (blue) or upregulation (red) in Log2 fold changes. White boxes indicate no significant differential gene expression. For each gene, boxes represent the contrasts for which differential gene expression was calculated. Contrasts from left to right: S + O oscillation versus control cultures, O oscillation versus control cultures, S oscillation (feast sample) versus control cultures, S oscillation (famine sample) versus control cultures, S oscillation cultures famine vs. feast samples. (C) Principal component analysis (PCA) of logTPM gene expression data, showing the pair‐wise comparisons of principal components. The principal components PC1‐PC4 were retained as significant based on the “elbow” of the screen test. The timepoints S1 and S2 indicate different sampling times during one oscillation cycle of S and O oscillation cultures. In O oscillation cultures: S1, oxygen limitation; S2, oxygen excess/substrate limitation. In S oscillation cultures: S1, substrate limitation (“feast”); S2, substrate starvation (“famine”).

To understand the changes in transcription caused by an oscillatory environment, RNA‐seq samples were taken at the end of the scale‐down and control cultivations. Differentially expressed genes (DEGs) between five different groups of samples, that is, contrasts, were analysed. The contrasts were (1) combined substrate and oxygen (S + O) oscillation versus control cultures, (2) oxygen (O) oscillation versus control cultures, (3) substrate (S) oscillation culture samples taken during the limitation regime (feast, S1) versus control cultures, (4) S oscillation culture samples taken during the starvation regime (famine, S2) versus control cultures, and (5) famine (S2) versus feast (S1) samples of the S oscillation cultures. Contrasts (1–4) reveal the long‐term response of the gene expression under scale‐down conditions, while contrast (5) reveals the transient change of gene expression during short‐term oscillations of substrate availability. Significant changes observed in the expression of genes in central carbon metabolism (Figure 2A) and the electron transport chain (Figure 2B) were presented. No genes in the central carbon metabolism and the electron transport chain were found to be differentially expressed between the feast samples and control samples (Figure 2A,B). In O and S + O oscillation cultures, glk (cytoplasmic glucokinase) was upregulated compared to the control cultures (+3.9 and +1.2, respectively; values represent Log2 fold changes). Notably, the genes pgl (−4.4), rpiA (−2.5), tktA (−4.5), tktB (−4.8), and talA (−4.9) of the pentose phosphate pathway (PPP) were downregulated under oxygen oscillations, while gnd (+1.5) was upregulated. In the upper glycolysis, the genes pfkB (+1.2 and + 1.5, respectively) and fbaA (+2.9 and + 1.0, respectively) were upregulated in O and S + O oscillation cultures, and pgi (−2.7), tpiA (−2.6), and fbaA (−2.9) were downregulated in O oscillation cultures. The expression of genes of the tricarboxylic acid (TCA) cycle was significantly impacted by oxygen and combined substrate and oxygen oscillations. In particular, genes expressing the four subunits of the succinate dehydrogenase, sdhABCD, were downregulated (−3.7, −3.7, −3.4, −3.4 in O oscillation cultures, and −2.7, −2.8, −2.2, −2.6 in S + O oscillation cultures). Additionally, O and S + O oscillations resulted in the downregulation of the genes sucAB encoding the subunits of the α‐ketoglutarate dehydrogenase (−3.1, −1.9 in O oscillation cultures, and −2.6, −2.4 in S + O oscillation cultures), and the genes sucCD encoding the subunits of succinyl‐CoA synthetase (−1.7, −1.3 in O oscillation cultures, and −2.7, −2.8 in S + O oscillation cultures). Most genes related to the conversion of pyruvate to acetyl‐CoA were upregulated in O oscillation and S + O oscillation cultures, such as aceE (+4.5, and + 1.9 respectively), aceF (+4.4, and +1.6 respectively), and lpd (+2.8, and + 1.3 respectively).

Differential gene expression levels were mapped to genes of the electron transport chain (ETC) (Figure 2B). The genes appB and appC, encoding subunits I and II of E. coli cytochrome bd‐II, were downregulated in O oscillation cultures (both with a log2 fold change of −1.2). In both O and S + O oscillation cultures, genes encoding the subunits I and II of the cytochrome bd‐I complex, cydA (+4.2, and +1.2 respectively) and cydB (+4.2, and +1.2 respectively), were upregulated. The genes cyoABCDE of the cytochrome bo complex were downregulated in O oscillation cultures (−2.2, −1.6, −1.6, −1.8, and −2.3), while the genes cyoCDE were found upregulated during the famine phase compared to control cultures and the feast phase (+1.3, +1.4, and +1.1 compared to control, +1.7, +1.7, and +1.2 compared to feast phase). The gene ndh, expressing the NADH:quinone oxidoreductase II, was upregulated in O oscillation cultures (+3.0) and downregulated between famine and feast samples of the S oscillation cultures (−1.4). Genes of the nuo operon (NADH:quinone oxidoreductase) were generally downregulated in O and S + O oscillation cultures. The genes nuoE, nuoF, nuoM, and nuoN were significantly upregulated between famine and feast samples of the S oscillation cultures.

iModulons analysis reveals functional clusters of E. coli's transcriptional regulation

To understand system‐wide changes in gene expression, we inferred the iModulon activities of our RNA‐seq dataset from PRECISE‐1K (Lamoureux et al., 2023; Sastry et al., 2019). We applied differential iModulon activity (DiMA) analysis to identify biological functions within the TRN that had been significantly affected by oscillating substrate and/or oxygen conditions. In Supporting Information, Figure S12 and Table S3 offer a detailed description of the DiMA analysis and the biological functions of differentially activated iModulons. To support the global transcriptional analysis, the relative fraction of the explained variance for each iModulon in our dataset was calculated and ranked (Figure 3A). The iModulon gcvB, which was highly differentially activated between O oscillation and control cultures (Figure S12b), accounted for the largest share of explained variance. The activities of all differentially activated iModulons were clustered to identify groups of similarly activated iModulons among our samples (Figure 3B). Three major RNA‐seq sample groups were separated by clustering. The substrate oscillation and control culture samples appeared to be closely related, while S + O oscillation culture samples and O oscillation culture samples formed two separate clusters.

FIGURE 3.

FIGURE 3

Open in a new tab

Systems view on gene expression based on independent components (iModulons). (A) iModulons with the highest explained variance within the dataset of this study were identified. (B) Analysis of differential iModulon activities (DiMAs) was performed (Figure S12 and Table S3). All differentially activated iModulons were clustered according to their z‐score normalized activities. Red lines inside the dendrograms indicate manually selected distance thresholds (rows threshold = 3, columns threshold = 3). (C) iModulons RpoS, and Translation of the fear versus greed trade‐off. The colour bar shows annotated (if present) growth rates of PRECISE‐1K samples. Grey data points show other iModulon activities from PRECISE‐1K. Arrow points to PRECISE‐1K samples in LB + 5% ethanol medium. (D) The activity of the ppGpp iModulon is a biomarker of the stringent response. Grey box plots show ppGpp activity levels in PRECISE‐1K. Activities from batch and fed‐batch samples in PRECISE‐1K are shown separately. (E) The phase portrait between ArcA and Acetate shows the switch between anaerobic growth and growth on acetate. Arrow annotations point to relevant PRECISE‐1K samples as reference.

Three main clusters of iModulons were identified. In the left part of the cluster map were iModulons that were relatively higher activated in O oscillation cultures. Part of this cluster was the iModulon gcvB (related to amino acid metabolism), which showed the largest variance in our dataset (Figure 3A). Other relevant iModulons of the left cluster in Figure 3B were the envelope‐stress‐related iModulons EvgA, Membrane, and Osmoprotectant, as well as iModulons related to other stress functions Biofilm, Efflux Pump, and Salicylic Acid. The right cluster consisted of iModulons that were relatively higher activated in samples from the control and the S oscillation cultures (Figure 3B). Characteristic iModulons of this cluster were the iModulons Crp‐1 and Crp‐2, which capture Crp‐dependent gene regulation that is part of carbon metabolism. Furthermore, both curli‐related iModulons, Curli‐1/−2, and the carbon metabolism related iModulons Acetate and Glycolate were relatively higher activated in control and the S oscillation cultures. The iModulons ppGpp, RpoS, and Translation in the middle cluster showed relatively low activities in samples from O oscillation cultures compared to other samples. Notably, the iModulons ppGpp and Translation of this cluster were also differentially activated between S1 and S2 samples from S oscillation cultures (Figure S12c), pointing to a transient change of these iModulons during feast‐famine oscillations. This was also the case for the iModulons Arginine and UC‐8, which were part of the right cluster (Figure 3B).

iModulon phase portraits reveal global transcriptional trade‐offs

Phase portraits can reveal relationships between two iModulons across large datasets such as PRECISE‐1K and are a useful tool to analyse and compare new data in the context of other RNA‐seq data collected under a large variety of conditions. We focused on two phase portraits relevant to this study, the so‐called fear‐greed phase portrait between the iModulons Translation and RpoS (Figure 3C) and the phase portrait between ArcA (aerobicity) and Acetate (growth on anaerobic byproduct acetate) (Figure 3E). The fear‐greed phase portrait was previously described (Sastry et al., 2019) and shows the trade‐off between growth‐related functions, such as Translation (greed), and stress‐related functions, captured by the RpoS iModulon (fear). Samples from PRECISE‐1K exhibited a negative correlation between Translation and RpoS (Figure 3C), meaning that high activities of Translation generally occur at lower RpoS activities and vice versa. The fear‐greed trade‐off was supported by the annotated growth rates of PRECISE‐1K samples in Figure 3C, in that E. coli cultures growing at higher growth rates also generally exhibited higher Translation activities, and cultures with a high RpoS activity tended to grow slower.

In this study, the RNA‐seq samples of the control, S oscillation, and S + O oscillation cultures exhibited low Translation and high RpoS activities, showing a fear phenotype. Surprisingly, samples of the O oscillation cultures showed low Translation activities and low RpoS activities, breaking the trade‐off relationship between the two iModulons. A similar phenotype was only observed in PRECISE‐1K samples grown in LB media supplemented with 5% ethanol. The iModulon ppGpp, capturing genes related to the stringent response, is closely related to the iModulon RpoS and the fear‐greed trade‐off. In PRECISE‐1K, the activity of ppGpp decreased with the onset of the stringent response in carbon‐limited high‐cell‐density fed‐batch samples compared to samples from the same cultures taken during the batch phase (Figure 3D). The RNA‐seq samples of this study from the control, S oscillation, and S + O oscillation cultures showed ppGpp activity levels in line with carbon‐limited fed‐batch samples from PRECISE‐1K. Samples from the O oscillation cultures showed especially low ppGpp activity levels compared to the overall ppGpp activity levels observed across PRECISE‐1K.

The phase portrait between the iModulons ArcA and Acetate was highly relevant for this study. ArcA captures the energy metabolism during anaerobiosis and consists of aerobic growth genes that are repressed during anaerobic growth. The iModulon Acetate captures the function of acetate uptake and catabolism. The phase portrait (Figure 3E) showed a switching characteristic between the ArcA and Acetate iModulons. Higher Acetate iModulon activities were only observed at neutral ArcA iModulon activities across PRECISE‐1K. Samples in PRECISE‐1K taken from high‐cell‐density cultures during the batch phase (carbon unlimited, aerobic growth) showed neutral ArcA and Acetate iModulon activities. Samples from the same cultures taken during the feeding phase (carbon limited, aerobic growth) showed an increase of Acetate iModulon activities. PRECISE‐1K samples from batch cultures grown on acetate as carbon source showed similar iModulon activities on the phase portrait. The RNA‐seq samples of this study from the control and S oscillation cultures (carbon limited, aerobic growth) grouped together with the high‐density feed and acetate growth samples, while S + O oscillation samples were close to the batch samples on the phase portrait. Samples from the O oscillation cultures also showed neutral Acetate iModulon activities but had additionally moved towards lower ArcA activities, indicating a more anaerobic phenotype.

DISCUSSION

In this study, we applied oscillation‐based scale‐down methods to elucidate the metabolic and transcriptional response of E. coli to industrial‐scale oxygen and substrate gradients. Depending on the type of oscillations, we found that biomass, CO2, and acetate formation were significantly affected. Additionally, from global transcriptional analysis, we found that oxygen oscillations altered the transcriptional state of E. coli much more severely in comparison to the oscillations of glucose.

Mixing gradients of glucose and oxygen were simulated with a bioreactor compartment model (Nadal‐Rey et al., 2021) (Figure 1A–C). The results showed that due to insufficient mixing and aeration of the reactor, E. coli experienced anaerobic metabolism at the top of the reactor and glucose starvation at the bottom of the reactor. Similar results were previously obtained from CFD simulations of partially oxygen‐limited fed‐batch processes (Kuschel & Takors, 2020; Wei et al., 2023). To accurately mimic the relative fraction and intensity of the predicted metabolic regimes in lab‐scale fermentations, our scale‐down design was based on metabolic regime transitions that were periodically repeated in cycles of 10 min. These relatively long cycle times were chosen as a technical compromise to create reproducible cycles of oxygen availability (DO >0%) and oxygen limitation (DO = 0%). Similar to other single‐compartment scale‐down methods reported in literature (Anane et al., 2019; Vasilakou et al., 2020), this scale‐down design neglected realistic transition patterns between metabolic regimes, as well as the residence time of these transitions. Studies that simulated Lagrangian cell trajectories (called lifelines) in a bioreactor were able to identify the residence time distribution of the different metabolic regime transitions (Haringa et al., 2016; Kuschel & Takors, 2020; Wei et al., 2023). For example, six regime transition patterns were identified with residence times in the range of tens of seconds in a Penicillin chrysogenum fermentation with combined oxygen and glucose gradients (Wei et al., 2023). The stochastic tracking of cell lifelines in compartment models was demonstrated recently (Haringa et al., 2022) and could potentially support the design of more accurate compartment‐model‐based single‐compartment scale‐down experiments in the future. Such scale‐down design approaches for more accurate single‐compartment scale‐downs exist in theory (Wei et al., 2023) but have yet to be implemented experimentally. Based on our scale‐down design, combined substrate, and oxygen (S + O) oscillation experiments (Figure 1D), as well as experiments oscillating only substrate (S) (Figure 1E) and oxygen (O) (Figure 1F), were performed to investigate the effects of different metabolic regime transitions.

Repeated glucose oscillations did not result in a significant decline in biomass formation (Figure 1G); however, a slight increase in carbon dioxide formation was observed (Figure 1I). In contrast, oscillating oxygen availability significantly impacted growth and the accumulation of acetate (Figure 1G–L). The accumulation of acetate and decreased biomass formation in bioreactors with oxygen limitation regimes was previously observed (Enfors et al., 2001; Neubauer et al., 1995). Especially in our O oscillation cultures, the high acetate accumulation (45.4 g L−1 at the end of the fermentation) likely caused a complete growth arrest (Kleman & Strohl, 1994; Wolfe, 2005; Xu et al., 1999). In S + O oscillation cultures, the formation of carbon dioxide (Figure 1I,J), as well as the consumption of oxygen (Figure S9), was increased compared to all other cultures. Although acetate accumulated towards the end of the fermentation in these cultures, higher carbon dioxide and lower biomass formation can likely be explained by the formation of acetate during the oxygen limitation regime and the subsequent partial re‐assimilation of acetate during the starvation regime at a lower biomass yield (Enfors et al., 2001; Neubauer et al., 1995; Valgepea et al., 2010). The O and S + O oscillation experiments were designed to experience the same degree of anaerobic metabolism, defined by the amount of glucose supplied during oxygen limitation (Figure S4). Nevertheless, acetate accumulation was much higher in O oscillation cultures compared to S + O oscillation cultures, where the oxygen‐limited regime was followed by the starvation regime without glucose feeding. This may have relieved catabolite repression and thereby helped a faster re‐assimilation of acetate in S + O oscillation cultures (Enjalbert et al., 2017; Valgepea et al., 2010).

Differential gene expression analysis revealed that oxygen oscillations (both isolated and combined with glucose oscillations) majorly impacted the gene expression related to the central carbon metabolism (Figure 2A) and the electron transport chain (Figure 2B). In O oscillation cultures, the genes pgl, rpiA, tktA, tktB, and talA of the pentose phosphate pathway (PPP) were downregulated. Considering the growth arrest in O oscillation cultures, downregulation of PPP genes can be assumed to be due to the strongly reduced demand for biomass precursor metabolites (Hardiman et al., 2007). The downregulation of the TCA cycle genes sucABCD and sdhCDA can be attributed to ArcA and Fnr‐dependent repression under anaerobic conditions (Park et al., 1995, 1997). We observed that the genes cyoABCE transcribing the low‐affinity/high‐efficiency cytochrome bo oxidase complex were downregulated (in O oscillation cultures), while the genes cydAB transcribing the high‐affinity/low‐efficiency cytochrome bd‐I complex, which is the dominant oxidase in E. coli under microaerobic conditions (von Wulffen et al., 2017), were upregulated (in both S + O and O oscillation cultures). The downregulation of cyo genes and upregulation of cyd genes was in line with the findings of other studies (Tseng et al., 1996; von Wulffen et al., 2017). Additionally, in O oscillation cultures, the gene ndh, which transcribes the NADH: quinone oxidoreductase II (NDH‐2), was upregulated, while nuo genes transcribing the subunits of the NADH: quinone oxidoreductase I (NDH‐1) were downregulated in S + O and O oscillation cultures. NDH‐1 is one of the largest protein complexes in the membrane of E. coli and is preferred under anaerobic conditions, contributing to the proton motive force (PMF) (Price & Driessen, 2010; Unden & Bongaerts, 1997; Young et al., 1978). In contrast, NDH‐II does not generate PMF and is preferred under aerobic conditions (Anand et al., 2022; Unden & Bongaerts, 1997). Consequently, (von Wulffen et al., 2017) measured upregulation of the ndh operon in E. coli shortly after switching from anaerobic to aerobic conditions. With that in mind, our results seem contradictory, however, one must note that while (von Wulffen et al., 2017) investigated the effect of a single switch between prolonged anaerobic growth to aerobic growth, this study investigates E. coli under frequently changing aerobic/anaerobic conditions. While NDH‐1 is preferred by E. coli under constant anaerobic conditions, we propose that under repeated oxygen oscillations tested in this study, a downregulated expression of NDH‐1 occurred because cells were unable to allocate the energy and resources required for the biosynthesis of the large protein complex. In addition to the DEG analysis, the use of PCA further supported the observation that oxygen oscillations had the dominant effect on E. coli's transcriptional regulation in our scale‐down experiments (Figure 2C).

We applied iModulon analysis to get a deeper understanding of what regulatory programs had been most affected by the scale‐down conditions. Of particular interest was the iModulon gcvB, which showed an exceptionally high activity in O oscillation cultures and was the iModulon that contributed the most to the explained variance within our dataset (Figure 3A). GcvB has been linked to the inhibition of energy‐demanding amino acid uptake and biosynthesis (Lalaouna et al., 2019), as well as the repression of the global regulators Lrp (Lee & Gottesman, 2016), PhoP (Coornaert et al., 2013), and CsgD (Jørgensen et al., 2012). Therefore, GcvB may help E. coli to save energy and cope with the stress caused by repeated oxygen oscillations. The third cluster of differentially activated iModulons (Figure 3B) consisted of iModulons that mostly indicate carbon‐limited fed‐batch growth (Figure S13). The iModulons of this cluster were generally less activated in O and S + O oscillation cultures, thus, these cultures had shifted away from a typical carbon‐limited phenotype. Furthermore, many differentially activated iModulons related to stress functions, such as acid stress response, biofilm formation, and multi‐drug resistance, were part of the first iModulon cluster (Figure 3B), which consisted of iModulons activated specifically in O oscillation cultures. With respect to S oscillation cultures, we did not identify significant DiMAs compared to control cultures that may have indicated any long‐term transcriptional transitions in response to repeated glucose starvation. However, the DiMAs between feast (S1) and famine (S2) samples from S oscillation cultures highlighted a transient transcriptional response of E. coli to changing glucose levels (Figure S12c). Here, the downregulation of the iModulon ppGpp was of particular interest, indicating the onset of the stringent response under glucose starvation conditions, which was also previously observed during cycles of glucose starvation by (Löffler et al., 2016) in STR‐PFR scale‐down experiments.

We analysed iModulon phase portraits to interpret our gene expression data in the context of global transcriptional regulation patterns in E. coli (Figure 3C–E). Samples from O oscillation cultures deviated from the trend of the so‐called fear‐greed trade‐off (Sastry et al., 2019) by exhibiting low RpoS and Translation activities at the same time. This, however, seemed plausible since (a) the low Translation activity was in line with the observed growth arrest (Figure 1G) and (b) lower RpoS levels have been previously associated with elevated acetyl‐CoA levels and oxygen limitation (Battesti et al., 2015; King & Ferenci, 2005). The new expression data presented in this work motivates to further expand the condition space of the PRECISE‐1K compendium. The ppGpp‐mediated stringent response, which is captured by the ppGpp iModulon and is involved in the regulation of RpoS (Hardiman et al., 2007), showed particularly low activities in O oscillation cultures (Figure 3D). Low ppGpp iModulon activities indicate an elevated stringent response (Figure 3D). Hence, we speculate that the deprivation of biomass building blocks, such as amino acids, which likely coincided with the complete growth arrest in these cultures, resulted in a strong activation of the stringent response. The ArcA‐Acetate phase portrait showed that under oxygen oscillations (both O and S + O oscillation cultures), the TRN had transitioned towards anaerobic metabolism, indicated by low Acetate iModulon activities (Figure 3E). In the case of O oscillation cultures, the TRN had adapted towards a more anaerobic state with lower ArcA iModulon activities, which indicated an ArcA‐dependent repression of the TCA cycle (Brown et al., 2022; Perrenoud & Sauer, 2005; Salmon et al., 2005).

Overall, the findings of our study show that oxygen gradients or oscillations significantly change the physiological and transcriptional state of E. coli. Interestingly, we found that the combined oscillations of substrate and oxygen resulted in a growth physiology and transcriptional regulation that was distinct from the ones observed under isolated substrate and oxygen oscillations. Therefore, scale‐down simulators employed for the development of industrial fermentation processes where both meaningful substrate and oxygen are expected should mimic both gradients at the same time rather than individually. Strain engineering strategies to limit the formation of fermentative byproducts should be applied to limit the accumulation of these byproducts and thereby ensure robust E. coli phenotypes under industrial‐scale fermentation conditions, so‐called fermenterphiles (Straathof et al., 2019). For example, the deletion of arcA was shown to reduce acetate build‐up in E. coli under transient and prolonged oxygen limitation (Veeravalli et al., 2018); hence, knocking out arcA should be considered as an engineering target for a fermenterphile strain design. (Gecse et al., 2024) investigated and reported that to increase robustness of 2'‐O‐fucosyllactose producing E. coli strain, blocking major acetate production pathways by deleting the genes pta and poxB was more efficient for carbon‐limited cultures exposed to sudden glucose shock in comparison to increasing TCA cycle flux by overexpressing the gltA gene when tested under industrially relevant conditions. In contrast to strain engineering strategies, (Sandberg et al., 2017) applied adaptive laboratory evolution approaches to select cultures that are robust to handle dynamic nutrient fluctuations by evolving batch cultures of E. coli with alternate glucose and acetate as substrates. Genome sequencing revealed that alternating with glucose and acetate resulted in development of specialist subpopulations, which might be beneficial for the overall culture performance of a process with glucose and acetate both present as substrates. Further, in cases where the effects of temporary oxygen limitation would be detrimental to the economic success of the production process, strategies to avoid oxygen limitation zones, such as optimized and/or multiple feeding points, are likely needed (Haringa et al., 2018). In addition, process intensification strategies to increase the oxygen transfer rate by using a special injection nozzle for aeration, creating smaller bubble sizes, and using pure oxygen could be applied to combat oxygen limitation (Noorman et al., 2018). We observed that in cultures with oxygen oscillations, the TRN of E. coli transitioned towards a non‐carbon‐limited state. This should be considered during the development of fed‐batch processes where oxygen gradients can be expected at industrial‐scale, since non‐scale‐down fed‐batch fermentations at lab‐scale would not result in such transcriptional adaptation.

EXPERIMENTAL PROCEDURES

Bacterial strain and seed cultures

Escherichia coli K‐12 MG1655 ΔfhuA was used for all fermentations in this study. A fhuA mutant E. coli strain was used to avoid phage contaminations in our laboratory (Lewis et al., 2023). An MG1655 strain was selected to be able to compare our findings from RNA‐sequencing with other published data obtained from MG1655 strains. These are highly represented in the RNA‐sequencing data compendium PRECISE‐1K, which served as a reference for this study. The first seed culture was prepared in culture tubes by inoculating 5 mL seed medium with 0.1 mL of a 15% glycerol stock (stored at −70°C) and incubating for 8 h at 30°C and 225 rpm shaking speed in a Kühner shaking incubator with a shaking diameter of 50 mm. A second seed culture was started from the first seed culture (5 mL grown seed culture into 160 mL fresh seed medium) and incubated in a Sartorius Ambr® 250 microbial culture vessel at 30°C, 1 vvm aeration, and 1000 rpm stirring speed. The seed medium contained (per litre) 2 g glucose, 6.8 g Na2HPO4, 3 g KH2PO4, 0.5 g NaCl, 1 g NH4Cl, 0.49 g MgSO4 × 7H2O, 16.95 mg CaCl2 × 2H2O, 7.5 mg Na2 EDTA × 2H2O, 2.25 mg ZnSO4 × 7H2O, 0.035 mg MnCl2 × 4 H2O, 0.15 mg CoCl2 × 6H2O, 0.1 mg CuSO4 × 2H2O, 0.2 mg Na2MoO4 × 2H2O, 1.5 mg FeSO4 × 7H2O, 0.5 mg H3BO3, 0.05 mg KI, 10 μg Pyridoxine HCl, 5 μg Thiamine HCl, 5 μg Riboflavin, 5 μg Nicotinic acid, 5 μg Calcium D‐(+)‐pantothenate, 5 μg p‐Aminobenzoic acid, 5 μg thioctic acid, 2 μg Biotin, 2 μg Folic acid, 0.1 μg Vitamin B12.

Fed‐batch cultures

Fed‐batch cultures were inoculated with 5 mL of the second seed culture grown in an Ambr® 250 microbial culture vessel. First, E. coli was grown in batch cultures until the depletion of glucose. The batch medium contained (per litre) 20 g glucose, 10 g (NH4)2SO4, 15 g KH2PO4, 0.49 g MgSO4, 0.1 g Thiamine HCl, 1 mL trace metal stock solution, 1 mL iron stock solution, 0.5 mL Antifoam 204. Then, feeding was switched on, and the fed‐batch phase was started. The feed medium contained (per litre) 500 g glucose, 15 g KH2PO4, 13.7 g MgSO4 × 7H2O, 5 mL trace metal stock solution, 5 mL iron stock solution. Trace metal stock solution contained (per litre) 0.33 g CuSO4 × 5H2O, 0.54 g CoCl2 × 6H2O, 2 g CaCl2 × 2H2O, 0.41 g MnSO4 × H2O, 0.3 g Na2MoO4 × 2H2O, 6.3 g ZnSO4 × 7H2O. Iron stock solution contained (per litre) 30.5 g FeCl3 × 6H2O, 250 g Citric acid × H2O. To increase the biomass concentration, all cultures were first grown for 8.8 h at an exponential feeding rate of 0.097 h−1. Subsequently, the scale‐down feeding phase was started. With a feed rate F = D × V, where V is the liquid culture volume (estimated by Ambr® 250 system), and the dilution rate D was kept constant (D = 8.68 × 10−3 h−1). Note that in the case of our fed‐batch process, D is not equal to the specific growth rate μ. Calculated values of μ over time are presented in Supporting Information (Figure S10). For the scale‐down phase, process control was switched to one of the four different scale‐down protocols: (1) combined oxygen and substrate (S + O) oscillations, (2) oxygen (O) oscillations, (3) substrate (S) oscillations, (4) no oscillations (control). Each process protocol, except for control experiments, consisted of continuously repeating cycles with a duration of 10 min. The S + O oscillation cycles started with feeding and aeration switched on (substrate limitation regime). The aeration was switched off after 2.63 min (overflow regime). After 3.8 min, aeration was switched back on, and feeding was stopped (substrate starvation regime). The O oscillation cycles started with feeding switched on, and aeration switched off (overflow regime). After 3.08 min, aeration was switched on (substrate limitation regime). The S oscillation cycles started with feeding and aeration switched on (substrate limitation regime). After 3.8 min, feeding was switched off (substrate starvation regime). Feed rates during the cycles were adjusted such that the average feed rate and the total feeding volume were always equal in all cultures.

Bioreactor compartment modelling

The bioreactor compartment model of a 90 m3 stirred tank reactor with four Rushton turbines by (Nadal‐Rey et al., 2021), together with an adapted version of the overflow kinetic black box model by (Xu et al., 1999), was used to simulate industrial‐scale mixing gradients (see Data S1 for more details). Ordinary differential equations were solved in Python with the odeint solver from the submodule integrate of scipy (Virtanen et al., 2020). The compartment model simulations represented a process snapshot, assuming a constant volume at the maximum filling height, a constant biomass concentration of 45 g L−1, and a glucose feeding rate of 350 kg h−1 at the top of the reactor. A detailed description of the metabolic model adaptation, as well as the derivation of the compartment‐model‐based scale‐down design, is presented in Figures S1–S3, Tables S1 and S2.

Analytical methods

To determine cell dry weight (CDW), 2 mL sample tubes (Eppendorf, Germany) were dried at 80°C for 48 h, then cooled down to room temperature, and subsequently weighed empty on a AX224/E precision scale (OHAUS, Switzerland). Then, 1 mL of the culture broth was added to each tube. After centrifuging for 5 min at 13000 rpm and 4°C, the supernatant was discarded. The centrifugation step was repeated after washing the cell pellet with saline (9 g L−1 NaCl). The washed cell pellets were dried at 80°C for 48 h, and the tubes were weighed again. CDW was determined from the difference between empty and full weight. For the measurement of glucose, acetic acid, and pyruvic acid, the supernatant of culture broth samples was diluted 5‐fold in 9 mM sulfuric acid. Diluted samples were filtered and injected into the UltiMate 3000 HPLC (Thermo Scientific, USA) with 9 mM sulfuric acid at a flow rate of 0.7 mL min−1 as mobile phase, and the Rezex ROA‐Organic Acid H+ (8%) column (Phenomenex, USA) as stationary phase. The quantification of the compounds was based on the refractive index (RI) chromatograms.

RNA‐seq and iModulon analysis

At the end of each fed‐batch culture, 0.5 mL of the culture broth was added to 6 mL of Qiagen RNA‐protect Bacteria Reagent with the Ambr® 250 liquid handling system and mixed by pipetting. The solution was centrifuged, the supernatant was removed, and the cell pellets were stored at −70°C. For total RNA extraction, the RNeasy Mini Kit (Qiagen) was used per vendor protocol. Ribosomal RNA removal, library preparation, and sequencing were performed by an external research partner (BGI Europe A/S, Denmark). Processing of RNA‐sequencing reads was performed as previously described (github.com/avsastry/modulome‐workflow) (Sastry et al., 2019). Differential gene expression analysis was performed with DESeq2 in R (Love et al., 2014). DEGs with a log2 fold‐change of ±1 and an adjusted P value of ≤10−5 were considered significantly up−/downregulated. iModulon activities were inferred from the public E. coli iModulon database PRECISE‐1K (github.com/SBRG/precise1k) as previously described (github.com/SBRG/precise1k‐analyze) (Lamoureux et al., 2023). Differential iModulon activities (DiMAs) were calculated according to a previously described procedure, and pymodulon (pymodulon.readthedocs.io) and iModulonDB were used to analyse and visualize iModulon data (Rychel et al., 2021; Sastry et al., 2019). Principal component analysis (PCA) was performed in R based on z‐score normalized togTPM gene expression levels using the software package PCAtools (github.com/kevinblighe/PCAtools).

AUTHOR CONTRIBUTIONS

Jonas Bafna‐Rührer: Conceptualization; data curation; formal analysis; methodology; software; visualization; writing – original draft; writing – review and editing. Jean V. Orth: Conceptualization; data curation; investigation; methodology; software; writing – original draft; writing – review and editing. Suresh Sudarsan: Project administration; resources; supervision; writing – review and editing.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

Supporting information

Data S1.

MBT2-17-e70051-s001.docx (774.2KB, docx)

ACKNOWLEDGEMENTS

We thank Süleyman Øzmerih for the processing of raw sequencing reads. We thank Daniel Zielinski for providing feedback on the manuscript. We thank Lei Yang for providing the bacterial strain used in this study. We thank Felix Beulig and Emre Özdemir for their input on clustering and visualization of iModulon data. We thank Gisela Nadal Rey for sharing the computer code of the bioreactor compartment model. We thank Helena Junicke, Pablo Iván Nikel, and Andreas Worberg for assisting with project supervision. This work was funded by The Novo Nordisk Foundation, grant number NNF20CC0035580, NNF10CC1016517, and within the framework of the Fermentation‐based Biomanufacturing Initiative (FBM), grant number NNF17SA0031362.

Bafna‐Rührer, J. , Orth, J.V. & Sudarsan, S. (2024) Combined oxygen and glucose oscillations distinctly change the transcriptional and physiological state of Escherichia coli . Microbial Biotechnology, 17, e70051. Available from: 10.1111/1751-7915.70051

Funding: This study was supported by Novo Nordisk Fonden (NNF10CC1016517, NNF17SA0031362, NNF20CC0035580).

Contributor Information

Jonas Bafna‐Rührer, Email: jonruh@biosustain.dtu.dk.

Suresh Sudarsan, Email: sursud@biosustain.dtu.dk.

DATA AVAILABILITY STATEMENT

Original data and code generated in this study are available from https://zenodo.org/records/10642598. Large data files are available separately at https://doi.org/10.6084/m9.figshare.25105811.v2. Raw sequencing reads are published in the BioProject database under accession number PRJNA1075344 (https://www.ncbi.nlm.nih.gov/bioproject/1075344).

REFERENCES

  1. Anand, A. , Patel, A. , Chen, K. , Olson, C.A. , Phaneuf, P.V. , Lamoureux, C. et al. (2022) Laboratory evolution of synthetic electron transport system variants reveals a larger metabolic respiratory system and its plasticity. Nature Communications, 13(1), 3682. Available from: 10.1038/s41467-022-30877-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Anane, E. , Sawatzki, A. , Neubauer, P. & Cruz‐Bournazou, M.N. (2019) Modelling concentration gradients in fed‐batch cultivations of E. coli—towards the flexible design of scale‐down experiments. Journal of Chemical Technology and Biotechnology, 94, 516–526. Available from: 10.1002/jctb.5798 [DOI] [Google Scholar]
  3. Bafna‐Rührer, J. , Bhutada, Y.D. , Orth, J.V. , Øzmerih, S. , Yang, L. , Zielinski, D. et al. (2024) Repeated glucose oscillations in high‐cell‐density cultures influence stress‐related functions of Escherichia coli . PNAS Nexus, 3, 1–13. Available from: 10.1093/pnasnexus/pgae376 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Battesti, A. , Majdalani, N. & Gottesman, S. (2015) Stress sigma factor RpoS degradation and translation are sensitive to the state of central metabolism. Proceedings of the National Academy of Sciences of the United States of America, 112, 5159–5164. Available from: 10.1073/pnas.1504639112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Brand, E. , Junne, S. , Anane, E. , Cruz‐Bournazou, M.N. & Neubauer, P. (2018) Importance of the cultivation history for the response of Escherichia coli to oscillations in scale‐down experiments. Bioprocess and Biosystems Engineering, 41, 1305–1313. Available from: 10.1007/s00449-018-1958-4 [DOI] [PubMed] [Google Scholar]
  6. Brown, A.N. , Anderson, M.T. , Bachman, M.A. & Mobley, H.L.T. (2022) The ArcAB two‐component system: function in metabolism, redox control, and infection. Microbiology and Molecular Biology Reviews, 86, e0011021. Available from: 10.1128/mmbr.00110-21 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Coornaert, A. , Chiaruttini, C. , Springer, M. & Guillier, M. (2013) Post‐transcriptional control of the Escherichia coli PhoQ‐PhoP two‐component system by multiple sRNAs involves a novel pairing region of GcvB. PLoS Genetics, 9, 12–14. Available from: 10.1371/journal.pgen.1003156 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Enfors, S.O. , Jahic, M. , Rozkov, A. , Xu, B. , Hecker, M. , Jürgen, B. et al. (2001) Physiological responses to mixing in large scale bioreactors. Journal of Biotechnology, 85, 175–185. Available from: 10.1016/S0168-1656(00)00365-5 [DOI] [PubMed] [Google Scholar]
  9. Enjalbert, B. , Millard, P. , Dinclaux, M. , Portais, J.C. & Létisse, F. (2017) Acetate fluxes in Escherichia coli are determined by the thermodynamic control of the Pta‐AckA pathway. Scientific Reports, 7, 1–11. Available from: 10.1038/srep42135 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Gecse, G. , Labunskaite, R. , Pedersen, M. , Kilstrup, M. & Johanson, T. (2024) Minimizing acetate formation from overflow metabolism in Escherichia coli: comparison of genetic engineering strategies to improve robustness toward sugar gradients in large‐scale fermentation processes. Frontiers in Bioengineering and Biotechnology, 12, 1–14. Available from: 10.3389/fbioe.2024.1339054 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Hardiman, T. , Lemuth, K. , Keller, M.A. , Reuss, M. & Siemann‐Herzberg, M. (2007) Topology of the global regulatory network of carbon limitation in Escherichia coli . Journal of Biotechnology, 132, 359–374. Available from: 10.1016/j.jbiotec.2007.08.029 [DOI] [PubMed] [Google Scholar]
  12. Haringa, C. , Tang, W. , Deshmukh, A.T. , Xia, J. , Reuss, M. , Heijnen, J.J. et al. (2016) Euler‐Lagrange computational fluid dynamics for (bio)reactor scale down: an analysis of organism lifelines. Engineering in Life Sciences, 16, 652–663. Available from: 10.1002/elsc.201600061 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Haringa, C. , Tang, W. & Noorman, H.J. (2022) Stochastic parcel tracking in an Euler–Lagrange compartment model for fast simulation of fermentation processes. Biotechnology and Bioengineering, 119, 1849–1860. Available from: 10.1002/bit.28094 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Haringa, C. , Tang, W. , Wang, G. , Deshmukh, A.T. , van Winden, W.A. , Chu, J. et al. (2018) Computational fluid dynamics simulation of an industrial P. chrysogenum fermentation with a coupled 9‐pool metabolic model: towards rational scale‐down and design optimization. Chemical Engineering Science, 175, 12–24. Available from: 10.1016/j.ces.2017.09.020 [DOI] [Google Scholar]
  15. Janoska, A. , Buijs, J. & van Gulik, W.M. (2022) Predicting the influence of combined oxygen and glucose gradients based on scale‐down and modelling approaches for the scale‐up of penicillin fermentations. Process Biochemistry, 124, 100–112. Available from: 10.1016/j.procbio.2022.11.006 [DOI] [Google Scholar]
  16. Jørgensen, M.G. , Nielsen, J.S. , Boysen, A. , Franch, T. , Møller‐Jensen, J. & Valentin‐Hansen, P. (2012) Small regulatory RNAs control the multi‐cellular adhesive lifestyle of Escherichia coli . Molecular Microbiology, 84, 36–50. Available from: 10.1111/j.1365-2958.2012.07976.x [DOI] [PubMed] [Google Scholar]
  17. Kerssemakers, A.A.J. , Krishnan, J. , Rychel, K. , Zielinski, D.C. , Palsson, B.O. & Sudarsan, S. (2024) Deciphering the transcriptional regulatory network of Yarrowia lipolytica using machine learning.
  18. Kerssemakers, A.A.J. , Øzmerih, S. , Sin, G. & Sudarsan, S. (2023) Dynamic interplay between O2 availability, growth rates, and the transcriptome of Yarrowia lipolytica . Fermentation, 9, 1–18. Available from: 10.3390/fermentation9010074 [DOI] [Google Scholar]
  19. King, T. & Ferenci, T. (2005) Divergent roles of RpoS in Escherichia coli under aerobic and anaerobic conditions. FEMS Microbiology Letters, 244, 323–327. Available from: 10.1016/j.femsle.2005.02.002 [DOI] [PubMed] [Google Scholar]
  20. Kleman, G.L. & Strohl, W.R. (1994) Acetate metabolism by Escherichia coli in high‐cell‐density fermentation. Applied and Environmental Microbiology, 60, 3952–3958. Available from: 10.1128/aem.60.11.3952-3958.1994 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kuschel, M. & Takors, R. (2020) Simulated oxygen and glucose gradients as a prerequisite for predicting industrial scale performance a priori. Biotechnology and Bioengineering, 117, 2760–2770. Available from: 10.1002/bit.27457 [DOI] [PubMed] [Google Scholar]
  22. Lalaouna, D. , Eyraud, A. , Devinck, A. , Prévost, K. & Massé, E. (2019) GcvB small RNA uses two distinct seed regions to regulate an extensive targetome. Molecular Microbiology, 111, 473–486. Available from: 10.1111/mmi.14168 [DOI] [PubMed] [Google Scholar]
  23. Lamoureux, C.R. , Decker, K.T. , Sastry, A.V. , Rychel, K. , Gao, Y. , McConn, J.L. et al. (2023) A multi‐scale expression and regulation knowledge base for Escherichia coli . Nucleic Acids Research, 51, 10176–10193. Available from: 10.1093/nar/gkad750 [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lara, A.R. , Galindo, E. , Ramírez, O.T. & Palomares, L.A. (2006) Living with heterogeneities in bioreactors: understanding the effects of environmental gradients on cells. Molecular Biotechnology, 34, 355–381. Available from: 10.1385/MB:34:3:355 [DOI] [PubMed] [Google Scholar]
  25. Lee, H.J. & Gottesman, S. (2016) SRNA roles in regulating transcriptional regulators: Lrp and SoxS regulation by sRNAs. Nucleic Acids Research, 44, 6907–6923. Available from: 10.1093/nar/gkw358 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Lewis, J.M. , Janda, K.E. , Kotter, D.B. , Grose, J.H. & McCleary, W.R. (2023) Characterization of the attachment of three new coliphages onto the Ferrichrome Transporter FhuA. Journal of Virology, 97, e0066723. Available from: 10.1128/jvi.00667-23 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Lim, H.G. , Rychel, K. , Sastry, A.V. , Bentley, G.J. , Mueller, J. , Schindel, H.S. et al. (2022) Machine‐learning from Pseudomonas putida KT2440 transcriptomes reveals its transcriptional regulatory network. Metabolic Engineering, 72, 297–310. Available from: 10.1016/j.ymben.2022.04.004 [DOI] [PubMed] [Google Scholar]
  28. Löffler, M. , Simen, J.D. , Jäger, G. , Schäferhoff, K. , Freund, A. & Takors, R. (2016) Engineering E. coli for large‐scale production—strategies considering ATP expenses and transcriptional responses. Metabolic Engineering, 38, 73–85. Available from: 10.1016/j.ymben.2016.06.008 [DOI] [PubMed] [Google Scholar]
  29. Love, M.I. , Huber, W. & Anders, S. (2014) Moderated estimation of fold change and dispersion for RNA‐seq data with DESeq2. Genome Biology, 15, 1–21. Available from: 10.1186/s13059-014-0550-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Nadal‐Rey, G. , McClure, D.D. , Kavanagh, J.M. , Cassells, B. , Cornelissen, S. , Fletcher, D.F. et al. (2021) Development of dynamic compartment models for industrial aerobic fed‐batch fermentation processes. Chemical Engineering Journal, 420, 130402. Available from: 10.1016/j.cej.2021.130402 [DOI] [Google Scholar]
  31. Neubauer, P. , Häggström, L. & Enfors, S.‐O. (1995) Influence of substrate oscillations on acetate formation and growth yield in Escherichia coli glucose limited fed‐batch cultivations. Biotechnology and Bioengineering, 47, 139–146. [DOI] [PubMed] [Google Scholar]
  32. Neubauer, P. & Junne, S. (2010) Scale‐down simulators for metabolic analysis of large‐scale bioprocesses. Current Opinion in Biotechnology, 21, 114–121. Available from: 10.1016/j.copbio.2010.02.001 [DOI] [PubMed] [Google Scholar]
  33. Noorman, H.J. , Van Winden, W. , Heijnen, J.J. & Van Der Lans, R.G.J.M. (2018) CHAPTER 1: intensified fermentation processes and equipment.
  34. Park, S.J. , Chao, G. & Gunsalus, R.P. (1997) Aerobic regulation of the sucABCD genes of Escherichia coli, which encode α‐ketoglutarate dehydrogenase and succinyl coenzyme A synthetase: roles of ArcA, Fnr, and the upstream sdhCDAB promoter. Journal of Bacteriology, 179, 4138–4142. Available from: 10.1128/jb.179.13.4138-4142.1997 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Park, S.‐J. , Tseng, C.‐P. & Gunsalus, R.P. (1995) Regulation of succinate dehydrogenase sdhCDAB operon expression in Escherichia coli in response to carbon supply and anaerobiosis: role of ArcA and Fnr. Molecular Microbiology, 15, 473–482. Available from: 10.1111/j.1365-2958.1995.tb02261.x [DOI] [PubMed] [Google Scholar]
  36. Perrenoud, A. & Sauer, U. (2005) Impact of global transcriptional regulation by ArcA, ArcB, Cra, Crp, Cya, Fnr, and Mlc on glucose catabolism in Escherichia coli . Journal of Bacteriology, 187, 3171–3179. Available from: 10.1128/JB.187.9.3171-3179.2005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Price, C.E. & Driessen, A.J.M. (2010) Biogenesis of membrane bound respiratory complexes in Escherichia coli . Biochimica et Biophysica Acta, Molecular Cell Research, 1803, 748–766. Available from: 10.1016/j.bbamcr.2010.01.019 [DOI] [PubMed] [Google Scholar]
  38. Rychel, K. , Decker, K. , Sastry, A.V. , Phaneuf, P.V. , Poudel, S. & Palsson, B.O. (2021) IModulonDB: a knowledgebase of microbial transcriptional regulation derived from machine learning. Nucleic Acids Research, 49, D112–D120. Available from: 10.1093/nar/gkaa810 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Rychel, K. , Sastry, A.V. & Palsson, B.O. (2020) Machine learning uncovers independently regulated modules in the Bacillus subtilis transcriptome. Nature Communications, 11(1), 6338. Available from: 10.1038/s41467-020-20153-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Salmon, K.A. , Hung, S.P. , Steffen, N.R. , Krupp, R. , Baldi, P. , Hatfield, G.W. et al. (2005) Global gene expression profiling in Escherichia coli K12: effects of oxygen availability and ArcA. Journal of Biological Chemistry, 280, 15084–15096. Available from: 10.1074/jbc.M414030200 [DOI] [PubMed] [Google Scholar]
  41. Sandberg, T.E. , Lloyd, C.J. , Palsson, B.O. & Feist, A.M. (2017) Laboratory evolution to alternating substrate environments yields distinct phenotypic and genetic adaptive strategies (editor) Kivisaar M. Applied and Environmental Microbiology, 83, e00410–e00417. Available from: 10.1128/AEM.00410-17 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Sastry, A.V. , Gao, Y. , Szubin, R. , Hefner, Y. , Xu, S. , Kim, D. et al. (2019) The Escherichia coli transcriptome mostly consists of independently regulated modules. Nature Communications, 10, 1–14. Available from: 10.1038/s41467-019-13483-w [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Straathof, A.J.J. , Wahl, S.A. , Benjamin, K.R. , Takors, R. , Wierckx, N. & Noorman, H.J. (2019) Grand research challenges for sustainable industrial biotechnology. Trends in Biotechnology, 37, 1042–1050. Available from: 10.1016/j.tibtech.2019.04.002 [DOI] [PubMed] [Google Scholar]
  44. Tajsoleiman, T. , Spann, R. , Bach, C. , Gernaey, K.V. , Huusom, J.K. & Krühne, U. (2019) A CFD based automatic method for compartment model development. Computers and Chemical Engineering, 123, 236–245. Available from: 10.1016/j.compchemeng.2018.12.015 [DOI] [Google Scholar]
  45. Tseng, C.P. , Albrecht, J. & Gunsalus, R.P. (1996) Effect of microaerophilic cell growth conditions on expression of the aerobic (cyoABCDE and cydAB) and anaerobic (narGHJI, frdABCD, and dmsABC) respiratory pathway genes in Escherichia coli . Journal of Bacteriology, 178, 1094–1098. Available from: 10.1128/jb.178.4.1094-1098.1996 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Unden, G. & Bongaerts, J. (1997) Alternative respiratory pathways of Escherichia coli: energetics and transcriptional regulation in response to electron acceptors. Biochimica et Biophysica Acta ‐ Bioenergetics, 1320, 217–234. Available from: 10.1016/S0005-2728(97)00034-0 [DOI] [PubMed] [Google Scholar]
  47. Valgepea, K. , Adamberg, K. , Nahku, R. , Lahtvee, P.‐J.J. , Arike, L. & Vilu, R. (2010) Systems biology approach reveals that overflow metabolism of acetate in Escherichia coli is triggered by carbon catabolite repression of acetyl‐CoA synthetase. BMC Systems Biology, 4, 166. Available from: 10.1186/1752-0509-4-166 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Vasilakou, E. , van Loosdrecht, M.C.M. & Wahl, A.S. (2020) Escherichia coli metabolism under short‐term repetitive substrate dynamics: adaptation and trade‐offs. Microbial Cell Factories, 19, 1–9. Available from: 10.1101/2020.03.14.982140 [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Veeravalli, K. , Schindler, T. , Dong, E. , Yamada, M. , Hamilton, R. & Laird, M.W. (2018) Strain engineering to reduce acetate accumulation during microaerobic growth conditions in Escherichia coli . Biotechnology Progress, 34, 303–314. Available from: 10.1002/btpr.2592 [DOI] [PubMed] [Google Scholar]
  50. Virtanen, P. , Gommers, R. , Oliphant, T.E. , Haberland, M. , Reddy, T. , Cournapeau, D. et al. (2020) SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature Methods, 17, 261–272. Available from: 10.1038/s41592-019-0686-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. von Wulffen, J. , Ulmer, A. , Jäger, G. , Sawodny, O. & Feuer, R. (2017) Rapid sampling of Escherichia coli after changing oxygen conditions reveals transcriptional dynamics. Genes, 8, 90. Available from: 10.3390/genes8030090 [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Vrábel, P. , Van Der Lans, R.G.J.M. , Luyben, K.C.A.M. , Boon, L. & Nienow, A.W. (2000) Mixing in large‐scale vessels stirred with multiple radial or radial and axial up‐pumping impellers: modelling and measurements. Chemical Engineering Science, 55, 5881–5896. Available from: 10.1016/S0009-2509(00)00175-5 [DOI] [Google Scholar]
  53. Wang, G. , Zhao, J. , Haringa, C. , Tang, W. , Xia, J. , Chu, J. et al. (2018) Comparative performance of different scale‐down simulators of substrate gradients in Penicillium chrysogenum cultures: the need of a biological systems response analysis. Microbial Biotechnology, 11, 486–497. Available from: 10.1111/1751-7915.13046 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Wei, P. , Haringa, C. , Portela, L.M. & Noorman, H.J. (2023) Metabolic‐fluid dynamics model construction and scale‐down design for an industrial Penicillin chrysogenum fermentation with combined dissolved oxygen and glucose concentration dynamics. Chemical Engineering Science, 276, 118770. Available from: 10.1016/j.ces.2023.118770 [DOI] [Google Scholar]
  55. Wolfe, A.J. (2005) The acetate switch. Microbiology and Molecular Biology Reviews, 69, 12–50. Available from: 10.1128/MMBR.69.1.12 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Xu, B. , Jahic, M. & Enfors, S.O. (1999) Modeling of overflow metabolism in batch and fed‐batch cultures of Escherichia coli . Biotechnology Progress, 15, 81–90. Available from: 10.1021/bp9801087 [DOI] [PubMed] [Google Scholar]
  57. Young, I.G. , Jaworowski, A. & Poulis, M.I. (1978) Amplification of the respiratory NADH dehydrogenase of Escherichia coli by gene cloning. Gene, 4, 25–36. Available from: 10.1016/0378-1119(78)90012-4 [DOI] [PubMed] [Google Scholar]
  58. Ziegler, M. , Zieringer, J. , Döring, C.L. , Paul, L. , Schaal, C. & Takors, R. (2021) Engineering of a robust Escherichia coli chassis and exploitation for large‐scale production processes. Metabolic Engineering, 67, 75–87. Available from: 10.1016/j.ymben.2021.05.011 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Data S1.

MBT2-17-e70051-s001.docx (774.2KB, docx)

Data Availability Statement

Original data and code generated in this study are available from https://zenodo.org/records/10642598. Large data files are available separately at https://doi.org/10.6084/m9.figshare.25105811.v2. Raw sequencing reads are published in the BioProject database under accession number PRJNA1075344 (https://www.ncbi.nlm.nih.gov/bioproject/1075344).

Articles from Microbial Biotechnology are provided here courtesy of Wiley

RESOURCES