Heterogeneous oxygen availability affects the titer and topology but not the fidelity of plasmid DNA produced by Escherichia coli

The production of plasmid DNA (pDNA) of sufficient quality for transfection to cell cultures and therapeutic purposes requires efficient and robust culture performance. Poor mass transfer can result in oxygen limitations of cultures in shake flasks [1–3], single use and large-scale bioreactors [4, 5]. In particular, dissolved oxygen tension (DOT) gradients can easily occur in large-scale bioreactors due to the limited mass transfer capacity, high oxygen demand and relatively high mixing times in such vessels [5]. Oxygen limitation strongly influences the physiology of E. coli, causing a metabolic rearrangement from aerobic to fermentative metabolism and leading to the accumulation of partially oxidized byproducts like organic acids and ethanol. The production of byproducts results in a decrease in the growth rate and yield of products and biomass [5]. However, less is known about the effects of oxygen on pDNA production. It has been reported that transient exposure (1.6 h) to oxygen limitation diminishes the fraction of plasmid-bearing cells in a batch culture [6]. Namdev et al. [7] reported that oxygen fluctuations created by intermittent aeration did not lead to plasmid segregation but caused a decrease in the plasmid copy number. According to another report, microaerobic conditions (DOT = 5% air sat.) did not cause plasmid loss and even increased the pDNA titer compared with aerobic conditions [8]. The effects of oxygen availability on the quality of the produced pDNA have not been properly characterized. For instance, the plasmid sequence fidelity should be verified during the production steps [9], while the pDNA topology is a quality factor required for therapeutic use [9] and transfection to cells in culture [10]. Cortassa and Aon [11] showed that plasmid supercoiling increased 6 min after shifting E. coli from aerobic to anaerobic conditions, and the supercoiling was maximum after 20 min of the shift. This was attributed to a decrease in the DNA relaxing activity of topoisomerase I and the constant activity of gyrase after the shift to anaerobiosis.

The above mentioned reports focused on one-step shift anaerobiosis or constant microaerobic conditions, which are not representative of large-scale cultures. In the present work, the effect of constant and oscillating DOT on the growth of E. coli and pDNA production was studied. Fully aerobic (DOT ≥30% air sat.) and microaerobic (DOT = 3% air sat.) conditions were used. Heterogeneous DOT conditions were simulated by the cyclic variation of the stirring speed, which resulted in DOT oscillations ranging from 100 to 0% air sat. during batch cultures. To obtain insight into the metabolic effect of the growth conditions and to better understand the cellular response, flux balance analysis was performed. The pDNA supercoiled fraction (SCF) was analyzed by agarose gels and microfluidics chips. To assess the fidelity, the produced plasmid was sequenced using Illumina Technologies.

Bioreactor scale-up often results in lower global mixing efficiency than bench-scale bioreactors. As a consequence, substrates and dissolved gases are heterogeneously distributed. The physiological adaptation of cells to heterogeneous conditions has been intensively studied yet it is far from being well understood. More studies focusing on the quality of a given product are needed to assess the process robustness. One of the key parameters affecting the physiology of E. coli is the availability of oxygen because it causes important metabolic rearrangements. While several previous studies have focused on the impact of oxygen availability on E. coli, only a few have studied the relevance of this parameter on product quality. For example, it has been demonstrated that oxygen limitation contributes to the synthesis of the non-canonical amino acid norvaline by E. coli [21, 22]. This may affect the quality of the recombinant protein produced. In the present study, DOT heterogeneities were simulated by transitions in the stirring rate, (similar to previous studies by Brognaux et al. [23], and Han et al., [24]) and were compared with cultures under constant oxygen availability. Specifically, we measured the plasmid production capabilities of E. coli in three scenarios in relation to the oxygen availability, aerobic, microaerobic and oscillatory DOT conditions. The observed accumulation of mixed acid fermentation products and variations in μ and Y _x/s with respect to aerobic cultures are well known consequences of microaerobic conditions. Under microaerobic conditions, the observed decrease of \( {q}_{O_2} \)and \( {q}_{CO_2} \) is evidence of the oxygen limitation at DOT = 3% air sat. By contrast, the oscillatory transit of the cell between aerobicity and oxygen limitation lead to a sharp drop in \( {q}_{O_2} \)and \( {q}_{CO_2} \) but an RQ greater than 1, indicating the prevalence of a fermentative rather than a respiratory catabolism of the glucose. This brought duplication of acetate production compared with the microaerobic case, a strategy used by the cell to maximize ATP production. This may occur due to very fast accumulation of organic acids when oxygen is totally depleted, as reported previously [5]. On the other hand, high-speed agitation periods increased DOT by enhancing oxygen transfer and quickly restored respiratory metabolism. Nonetheless, the fermentative enzymes remained inactive in the presence of oxygen. Carbon and redox balances for cultures under oscillatory DOT did not close as satisfactorily as those for aerobic and microaerobic cultures (Table 1). It is likely that some metabolites not measured, such as pyruvate or CO₂ in the liquid phase or ethanol in the gas phase, affected the overall balance.

The estimated metabolic fluxes under microaerobic conditions were consistent with those previously reported using E. coli W3110 and similar growth conditions [25]. The estimated lower flux to the pentose phosphate pathway and TCA cycle agrees with the observed reduction in the growth rate and biomass synthesis compared with that in aerobic cultures. Despite such lower fluxes (closely related to DNA building block synthesis and energy generation), the values of Y _p/x were greater under microaerobic than under aerobic conditions. This effect on Y _p/x coincides with previous reports, which attributed such an increase to a reduction in the growth rate under microaerobic conditions [9]. Therefore, the ATP generation from acetate synthesis could play an important role in the synthesis of pDNA under oxygen limitation. Glyoxylate bypass was estimated to be active under aerobic conditions, coincident with previous studies using the strain W3110 [26]. Furthermore, fluxes through the glyolxylate bypass decreased upon oxygen limitation in accordance to experimental evidence [27], rendering glyoxylate bypass completely inoperative under oscillated DOT (Fig. 4).

To overcome the lack of a (quasi) steady-state under oscillating DOT, we used exhaust gas monitoring to estimate O₂ and CO₂ fluxes to obtain feasible solutions for maximizing biomass at each time. The results of this time-point approach yielded reasonable changes in the fermentation metabolite production rates. As expected, byproducts were low at periods of high O₂ flux and vice versa. Acetate flux was less accurately predicted, and formate formation was inconsistent because almost all times were different from zero. Fluxes through the cytochromes bo and bd were also evaluated. Cytochrome bo responded fairly well to O₂ flux changes, but not cytochrome bd, which behaved more like an extra supply rather than a backup for respiratory activity that was expected to switch on once cytochrome bo flux was low [28, 29]. We did not incorporate a regulatory approach into our model to account for oxygen limitation inhibition for cytochrome bo and pyruvate formate lyase fluxes (as in [20]). Therefore, the discrepancies found only reflect the limitations of the model and strategy used to evaluate it, but they were still acceptable due to the good results obtained by classical FBA and a reduced metabolic network. Undoubtedly, refinement of the tools and of the model are needed to overcome such limitations. Moreover, the obtained results can guide the experimental effort and help to identify targets for cell engineering. Eliminating some of the fermentative pathways could be viable if respiratory activity could be sustained for prolonged O₂ limitation [25].

The production of pDNA was nearly unaffected by a constant low DOT. This opens the possibility for the microaerobic production of pDNA, which may be viable if strains with a lower accumulation of by-products and increased respiratory capacity are available [25]. By contrast, under oscillatory oxygen limitation, pDNA production was lower. However, such a decrease was not at the same level because it was reported using anaerobic shifts [7]. Nonetheless, these differences may be the result of using different E. coli K-12 derivative strains and plasmids. Hopkins and coworkers [7] used the E. coli AB1157 strain to replicate an RK6 replicon plasmid. Namdev and coworkers [8] used the E. coli DH5α strain, a highly mutated background commonly used for pDNA amplification, to replicate a pUC replicon plasmid. In our work, we employed a much less mutated background to also amplify a pUC replicon plasmid. This opens the question of whether such discrepancies have any relation to the plasmid-host interactions.

The pDNA topology was strongly affected by DOT oscillations. In the model strain, the degree of pDNA supercoiling is mainly determined by the activities of topoisomerase I (an IA type topoisomerase) and gyrase (a IIA topoisomerase) [30, 31]. Because gyrase introduces negative supercoils at the expense of ATP hydrolysis, its activity is connected to the [ATP]/[ADP] ratio. Furthermore, it has been demonstrated that the ratio correlates with the degree of pDNA supercoiling [32, 33]. Hsieh and coworkers [32] found that, upon an anaerobic shift, the adenylate pool rapidly drops and concomitantly relaxes DNA in less than 15 min, rendering pDNA heterogeneity. In transitions from anaerobiosis to aerobiosis, the plasmid becomes heterogeneous and relaxed within 5 min. Nonetheless, it becomes homogeneous but with a lower degree of negative supercoiling than before the shift after 15 min. Although the [ATP]/[ADP] ratio is expected to be lower under the microaerobic than under the aerobic conditions studied in the present report, SCF was similar between them. The energy metabolism is more affected under oscillating DOT. Accordingly, a strong decrease in the SCF was observed for these cultures, as well as higher heterogeneity evidenced by chip electrophoresis. A detailed analysis of the supercoiling density may be very useful to better understand this fact and to design pDNA backbones more robustly for industrial applications [34].

The impact of the process conditions on pDNA sequence, which can be a major issue, is still poorly documented. For instance, a recent study reported the contamination by 1.3 kb IS2 in a pDNA vaccine candidate, with pVAX1 as the backbone, between the kanamycin resistance cassette and BGH polyA signal [35]. The frequency at which such undesirable events could be triggered relies on the presence of switch sites of GC and AT skews (ssGCs and ssATs) in pDNA, as well as on the time, cell density and mode of cultivation [36]. Complete sequencing of the plasmids from the master cell bank and from the conditions assessed in this work demonstrated that DOT availability did not altered plasmid sequence fidelity. The nucleotide changes detected in plasmids from all conditions were the same and might were already present in the pVAX1 plasmid extracted from the master cell bank. The changes occurred with major incidence around the origin of replication and kanamycin resistance cassette. These results should be taken cautiously because the number of base changes detected fall within the error rate of the technique [37]. Dedicated experiments to better clarify this issue and its underlying mechanisms should be performed.

Estimation of the metabolic fluxes could be useful for the design of strains that better cope with environmental limitations. The network used in the present work was used to estimate fluxes that could lead to maximum biomass synthesis. Eliminating the flux through the pyruvate dehydrogenase complex was found to increase the flux through the pentose phosphate pathway (PPP) and thus the synthesis of ribose-5-phosphate, as evidenced by the Y_r5p (Table 3). This could allocate more building blocks for DNA synthesis, while the AcCoA would be synthesized mainly from PEP. Increasing the flux to the PPP efficiently reduced the metabolic burden in aerobic cultures of E. coli bearing a high copy number plasmid [38]. Together with the mutations of the glyoxylate shunt (ICL and MALS) elimitaing the flux through the PDH further increased the predicted Y_r5p. The underlying rationale of these results might be to eliminate as much as possible the redundant enzymatic activities that scavenge ATP, NADH or NADPH. Since the network used for these simulations did not include specific reaction for pDNA synthesis, it is not possible to estimate whether pDNA yields would increase in the proposed mutants.

Microaerobic conditions shifted the metabolism of E. coli to a predominantly fermentative mode. This was more evident in the case of oscillatory DOT. However, microaerobic conditions had little effect on pDNA production and quality. Therefore, the microaerobic production of pDNA may be a viable alternative, provided that engineered cells are available with a better metabolic performance. The estimated TCA and pentose phosphate pathway fluxes can help to identify cell engineering targets for such a task. Our results suggest that DOT heterogeneities may increase the topological heterogeneity pDNA and that oxygen availability does not influence pDNA sequence. This should be considered during bioreactor operation at several scales and for better strain and vector design.