Cultivation of microbial cultures at high cell density in microbioreactor fermentations to increase protein yield

Microbioreactors have been proven to be a useful tool in high‐throughput applications, such as clone screening, synthetic library testing, and media optimization. Most were designed for low cell density applications, where the optical density of the cultures typically does not exceed an OD600 of 10. In microbial applications, where protein is to be expressed, such a scale is not sufficient to produce material for extensive target molecule testing. Here, we present a method for growing high‐cell density Escherichia coli cultures in milliliter‐scale bioreactors, to produce milligram quantities of target protein. We used a micro‐Matrix system with a starting volume of 3 ml per culture. A combination of defined medium, a fed‐batch feeding strategy at low temperature, and an advanced self‐adapting control algorithm achieved up to 0.7 g of wet cell weight (WCW) in a 5.7 ml final culture volume, which corresponds to 123 g/L WCW. This translates to an estimated protein yield of 1150 mg of target protein per liter final volume.


INTRODUCTION
The first commercial laboratory bioreactors had working volumes of 20 L. 1 This volume has significantly decreased over the last decades. Instruments like the Sartorius Ambr250 and the DasBox have working volumes of 0.3 L. These benchtop instruments have similar components, being stirred tank bioreactors with multiple feed and gas lines sharing similar geometry and mixing properties, which makes them ideal systems for the development of production-scale processes. In the last 10 years, microbioreactors have become available, for scaled down applications using milliliter volumes. The micro-Matrix (Applikon) has recently been shown to be a useful scale-down tool for the culturing of several microbial strains. 2 At the present working volumes used in microbioreactors, it is hard to maintain correlations with productionscale vessels. These bioreactors are therefore limited to high-throughput screening: host selection, synthetic biology elements screening, clone selection, and parallel media optimization for batch cultivations. [3][4][5] Their main limiting factor is the somewhat inefficient oxygen transfer, which limits their use to low cell density applications, such as eukaryotic cell cultures or short microbial fermentations operated in batch mode. The highest cell density reported so far for such reactors is a dry cell weight (DCW) of 30 g/L, or approximately 90-120 g/L of wet cell weight (WCW). 6,7 This is lower than the 200-300 g/L WCW readily achieved with benchtop tank bioreactors. 8 Developing a means of high-cell density fermentation in a microbioreactor could be beneficial, as it may allow faster process development, but also for the simultaneous generation of multiple target proteins for screening and analysis, if small milligram quantities of material are required.
Here, we present an adaptation of a high-cell density protocol ( Figure 1) to the micro-Matrix microbioreactor. The micro-Matrix makes use of a 24-well cassette, with integrated sensors for real-time monitoring of pH, temperature, and percentage dissolved oxygen (%DO), that is mounted on an orbital shaking platform within a temperature-controlled hood. Each well is essentially a stand-alone bioreactor fed by four gas and one liquid addition lines. Our protocol for culturing E. coli expressing a recombinant protein in a final volume of 5.7 ml achieves protein yields, per WCW, similar to that achieved with benchtop fermenters.

Strains and growth medium
The E. coli BL21(DE3) strain was used for fermentations. Cells were transformed with a pET-derived plasmid carrying a codon-optimised gene of Clostridium botulinum toxin A under the control of a T7 promoter. 9 The toxin contained two substitutions, E224Q and H227Y, that rendered it enzymatically inacti ve. 10,11 The defined medium used in fermentations was the modified mineral medium from Korz et al. 12 Trace metals were filter-sterilized and added after autoclaving. The growth medium contained 5 g/L glycerol in the seed and the micro-Matrix fermentations, whereas 30 g/L glycerol was used in the bench-scale medium. All media were autoclaved and added to the reactors just before the start of each fermentation.

Bench-scale bioreactor cultivation
The fermentations were performed using the Eppendorf DasBox (0.3 L working volume) and the Eppendorf BioFlo 120 (1 and 3 L working volume). The fermentation process employed a two-phase (batch and expression) strategy in defined medium, with protein expression induced upon carbon depletion by 1.8 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Carbon depletion was detected as a sharp spike in DO. The process was controlled by DasWare 5.0 software using a customised script. 13 The temperature was set to 37 • C during the batch phase and to 21 • C during the protein expression phase. The pH was maintained at 6.7 by supplying 25%-28% (v/v) ammonium hydroxide. The oxygen saturation was maintained at 30% throughout the fermentation. The media was inoculated with 2% of the shake-flask starter culture. The expression phase lasted 24 h. The overall fermentation time was 38-40 h.

Microbioreactor cultivation
Fermentation experiments were performed over 32-48 h. Each well of a micro-Matrix cassette contained 3 ml of defined medium, which was inoculated with 300 μl of the starter culture. The cassette was installed on the orbital shaker and the control assembly, as well as the liquid feed module, attached as described in the manufacturer's operation manual. The liquid feed module was filled with 50 ml of defined media containing glycerol in the range of 35%-70% (v/v) glycerol. The glycerol concentration used depended on the parameter being tested (see Results). The liquid feed also contained 10 mM IPTG, which induced protein expression once feeding started. For upward pH control, the ammonium pressure vessel contained 200 ml of 20% (v/v) ammonium hydroxide, which supplied ammonium gas to each well. The pressure vessel and the liquid feed module were under 2 barg of pressure. Cultures were aerated with compressed air, which was delivered to each well at 0.3 barg. The orbital shaker rotated at 350-400 rpm, depending on parameter being tested (see Results), and the cassette temperature was maintained at 37 • C during the batch phase and at 21 • C during the protein expression phase. The settings for the micro-Matrix instrument can be found in the Supporting Material.

Measuring culture optical density and cell paste wet weight
Culture optical densities were measured using an Ultraspec 10-cell density meter from Amersham Biosciences. The WCW of individual bioreactors was determined by aspirating 1 ml of the culture into a weighed tube. The samples were centrifuged at 29,000 g for 3 min and the supernatant discarded. The WCW is the measured weight with the weight of the tube subtracted. The results were converted to grams/liter WCW. The WCW in the 24-well micro-Matrix cassettes was determined by transferring the culture from the cassette to a 24-well plate and centrifuging that plate for 20 min at 3578 g using a swing bucket rotor. The supernatant was decanted, and the entire plate weighed against an empty plate. The average cell paste weight was calculated by dividing the weight by 24.

Estimating target protein expression
Cultures were sampled after fermentation to assess protein expression. A 10 μl volume of a culture was diluted with 20 μl water and then mixed with 30 μl of 2× Tris-glycine SDS sample buffer (Novex). Ten microliters of each sample was loaded per well of a 10-well 4%-12% NuPAGE Bis-Tris precast gel of 1 mm thickness (Invitrogen). The BenchMark Protein Ladder (Invitrogen) was used as molecular weight standard, and the gels were stained with Simple Blue Safe Stain (Thermo Fisher). The expression levels of the target protein were estimated by densitometry of bands in the polyacrylamide gels using GeneTools (Syngene). The expression of botulinum toxin was measured relative to the total protein content. Total protein content is the sum of all proteins in one lane of the gel. The densitometry measurements of each band in a lane were summed and then the botulinum toxin band densitometry value was divided by this total value to arrive at the percentage target gene expression in each culture. Total protein concentration in cleared lysate was measured spectrophotometrically using absorbance at 280 nm.

RESULTS AND DISCUSSION
Fermentations in the DasBox and the BioFlo 120 were used as references for all micro-scale fermentation experiments. The results were highly reproducible across all bioreactor sizes. The %DO monitored over time for the 0.3, 1, and 3 L fermentations were almost identical. The WCW of each was very similar, between 155 and 165 g/L, and the percentage of expressed target protein ranged from 11% to 12% (Table 1). This corresponds to an estimated yield of approximately 1020-1100 mg/L of target protein from each fermentation ( Table 1). These data were used as a benchmark for the micro-Matrix high-cell density fermentation.
The aim was to achieve a WCW of 150 g/L and an estimated target expression of 1000 mg/L. It was not possible to directly transfer the fermentation protocol from the bench-scale bioreactors to the micro-Matrix system, because of significant design differences. For example, there are four gas lines in the micro-Matrix and one liquid addition line. The larger bioreactors have at least three liquid addition lines, which makes it possible to maintain the pH, feed and deliver inducing agent to the culture separately. In the micro-Matrix, the pH is controlled by the addition of gaseous NH 3 evaporated from an ammonium hydroxide solution. The gas is supplied through one of the gas lines. The single liquid addition line can deliver only one liquid. To overcome this limitation, we added 10 mM IPTG to the glycerol feed, so that we could both feed and induce the culture from the single liquid addition line. Protein expression is therefore induced gradually during fermentation, as the IPTG concentration increases over time.
Furthermore, preliminary experiments showed that the manufacturer's suggested parameters for microbial fermentations were meant to be used for relatively short low-cell density fermentations in complex media (Luria-Bertani or Terrific Broth) that do not demand high aeration rates or strict pH control. The default settings of the pH control cascade (Proportional, Integral, Differential gains [PID] set to 250) were too weak to maintain a steady pH at the desired value of 6.7 (Figure 2). In some wells, the pH reached as low as 5. Increasing the PID in steps to 1000, 5000, and then to its maximum of 9999 led to better pH control. The latter value was used in all subsequent experiments.
It was also difficult to maintain sufficient aeration at the recommended orbital speeds of 300-350 rpm and Target protein expression (%) 12 11 11 Estimated volumetric yield of target protein (mg/L) 1100 1020 1090 The discrepancies in WCW per final volume are possibly caused by the different condenser efficiencies of the exhaust cooling system.

F I G U R E 2
The pH control during fermentation in the micro-Matrix. Each line represents the pH in a well of the 24-well cassette. The pH was set to be maintained at pH 6.7. The P (proportional gain) coefficient in PID setup of the pH cascade was changed stepwise from the default 250 to 1000, then 5000, and finally to the maximum value of 9999. The three curves with pH values between 7.5 and 8 in the PID 9999 section were of cultures that did not survive the preceding lower pH environment.

F I G U R E 3
The percentage dissolved oxygen (%DO) during fermentation in the micro-Matrix. Arrow 1 indicates where the system had tried to maintain the %DO at a setpoint of 30%. Arrow 2 indicates the period of microaerobic fermentation, during which the air mass transfer was insufficient. Arrow 3 indicates the stage at which the carbon source was depleted, growth slowed, and a spike in %DO was detected. Arrow 4 indicates wells in which the cultures continue growing at the set %DO of 30%.
an aeration rate of 9 ml/min per well (approximately 3 volumes/volume/minute [vvm]). These settings led to several hours of microaerobic fermentation during the batch phase, which can lead to lower WCW (Figure 3).

F I G U R E 4
Screening for the optimal glycerol concentration. 5 g/L glycerol was depleted after reaching the maximum air transfer rate. There is a significant anaerobic phase at the other concentrations.
The integrated DO-spike detecting algorithm proved useful for switching between the batch phase to the expression phase. The agitation rate was set to the maximum 400 rpm, and three gas lines were used to boost the aeration from 9 to 27 ml/min per well (effectively 9 vvm). These settings led to the same oxygen transfer rate as in the DasBox system, which was set to approximately 800-1000 rpm agitation and 1 vvm aeration (which is far from its maximum capacity).
In the micro-Matrix, gases are supplied to the enclosed space above the medium, as opposed to a submerged sparger that bubbles gas through the medium. Despite the high aeration rates that were applied to the cultures, there was little evidence of evaporation. This may be due to the low temperature (21 • C) that was used for most of the fermentation or a combination of that, and a low exhaust gas water vapour content (as the aeration is not bubbled through the medium) and additional humidity supplied through the ammonium solution used to control the pH.
The stage of microaerobic fermentation ("anaerobic pit") suggested the original carbon source concentration may have been too high. Different glycerol concentrations were assessed to identify an ideal carbon source concentration. Glycerol at 5, 10, 20, and 30 g/L was supplied to cultures in the starting medium, one concentration per row of the 24-well cassette. The row containing glycerol at 5 g/L was the optimal concentration ( Figure 4). The failure rate of bioreactors per plate was 4.16%. We next investigated two different feeding strategies. One was a constant-rate and the other a DO-dependent feeding strategy, also known as DO-stat. 14 Due to big differences between the fermentation systems, the main aim was not to mimic the original protocol, but to maximize the biomass yield. Different constant feeds were tested along with the DO-dependent feeding: 300, 400, 500, and 600 nl/min constant feeding were tested and 0.5, 1, 2, and 4 μl DO-dependent feeding. In the latter, 0.5-4 μl of the feed is added if the %DO exceeds 30%. The higher rates were favorable to biomass yields ( Figure 5, Table 2). The DO-dependent strategy showed an exponential trend, which correlates with the exponential growth of bacterial cells ( Figure 5A). Oxygen saturation at 300 and 400 nl/min constant feed rates were insufficient as the %DO is higher than the setpoint of 30% ( Figure 5B). The 600 nl/min rate is too high for the first third of the expression phase, subjecting the culture to anaerobic conditions. This comparison of cultivation strategies and the final yields led us to several conclusions. The constant feeding was extremely easy to implement by triggering it when the temperature of the shake plate drops below 35 • C. A final OD of around 60 was achieved, the WCW per well reached 0.5 g, and the expression level of the target protein was 15%-16% (Table 2). However, the fact that the constant feed strategy relies on a temperature drop as a trigger, means that this cannot be controlled for each well individually, as the entire cassette is in a hood and the temperature would have to be dropped for all the wells simultaneously. The best results of the DO-dependent feed were similar. However, the constant feed does not meet the cultures' demands: it is too high at the beginning and insufficient at the late stages of the expression phase ( Figure 5B). This leads to an anaerobic condition at the beginning of the expression phase, which would be detrimental to cell growth and protein expression. DO-dependent feeding allows per-well control, as it can respond to the cultures growth and self-adjusts to the actual air intake ( Figure 5C). As the culture never becomes anaerobic, the results of the fermentation are more consistent and the target protein quality is not compromised. Per-well individual feeding would allow testing of several expression constructs that may differ in growth rate without underfeeding or overfeeding them. The only drawback of the DO-dependent feeding strategy is a more complicated set up of the micro-Matrix.
The system allows two different approaches to DOdependent feeding. The first one is simpler and relies on the DO threshold, that is, the system adds feed every time the DO is higher than the threshold (30%). It demands very precise timings of the glycerol depletion detection (DO spike), which serves as a trigger for phase switching. In our experiments, the time to the DO spike was always different and was between 4 and 7 h from the start of cultivation. If the feeding delay is too short, the culture will be induced before the DO spike at lower cell density. If it is too long, the induction will occur after the DO spike, which means the culture will starve. Both are detrimental for the cell paste and target protein yields.
The second approach uses the integrated DO-spike detection algorithm so that the system adds feed when the specified DO-spike is detected. While the first approach works with absolute values of dissolved oxygen, the second approach uses relative values. For example, if the specified value of an "up spike" is 25%, the system will add feed solution every time the difference between the actual DO value and that previously measured is higher than 25%. This means it detects the first DO spike caused by the depletion of batch-phase glycerol in the medium and then is triggered by any further minor DO spikes. It is F I G U R E 5 Assessment of different feeding strategies. (A) The amount of feed added to the well during fermentation. The DO-dependent strategy shows an exponential trend, which correlates with the exponential growth of bacterial cells. (B) Oxygen saturation at different constant feed rates. 300 and 400 nl/min are insufficient as the percentage dissolved oxygen (%DO) is higher than the setpoint of 30%. 600 nl/min is too high for the first third of expression phase, subjecting the culture to anaerobic conditions. (C) Oxygen saturation of the best DO-dependent feed rate, which was 4 μl rate. The feeding adapts to the oxygen consumption of the culture.
important that the feed addition is large enough to cause the subsequent DO spikes. The tested 0.5, 1, and 2 μl additions were too small to pass the DO detection threshold of 25% (Table 2); however, the 4 μl addition did ( Figure 5A). The results for that addition volume were close to the benchmark set using the bench-scale reactors. The cell-paste yield of 0.5 g from one well with the final volume 4.63 ml corresponds to 110 g/L WCW (Table 2).

F I G U R E 6
The percentage dissolved oxygen (%DO) trend of the final protocol (DO spike-based feeding). Note that the DO-spike detection was set to 10% for better utilization of available oxygen.
The overall volumetric yield was 840 mg/L, with a protein expression level of 16%. Due to high viscosity of the 70% glycerol feed medium, we experienced frequent blockages of feeding lines. To improve the robustness of the protocol, the feed medium viscosity was decreased by halving the glycerol concentration to 35%, which also reduced the IPTG concentration to 5 mM. To maintain the same addition of nutrients and IPTG, the feed addition volume was doubled to 8 μl.
Further protocol improvements aimed at improving oxygen availability. The final method used 8 μl feed additions of the 35% glycerol and 5 mM IPTG feed. The DO-dependent feed was used with the DO-spike detection algorithm set to an "up spike" of 10% to increase oxygen availability ( Figure 6). The Induction phase was set to 40 h. This achieved the best results: the final volume was 5.7 ml, there was 0.7 g cell paste per well, and 16% target protein expression. This corresponds to 123 g/L WCW per final volume, which is slightly below the set target of 150 g/L. However, the medium in micro-Matrix was more diluted with the feed as the feeding solution was two times less concentrated and the feeding time was 40 h instead of the 24 h used in the bench-scale bioreactors. If compared to the initial volume, the cell-paste yield for bench-scale bioreactors was 185-220 g/L, while for micro-Matrix, adjusting the WCW to media volume, it was 200-230 g/L. Despite of the heavy dilution, the estimated volumetric yield reached 1150 mg of target protein per 1 L final volume, which is within the range of the benchtop bioreactors (1000-1100 mg/L). More importantly, it is possible to isolate up to 5 mg of target protein from one well, provided all the target protein is soluble.

CONCLUSION
We report the successful adaptation of high-cell density protocol to a microbioreactor fermentation system. The protocol originally developed for bench-scale bioreactors was altered to fit the constraints of the micro-Matrix system. The new scale-down protocol utilizes the full capacity of the instrument: all the available air lines are used; the single liquid addition line combines feeding and induction. A longer expression phase increases both biomass and target protein yield. A self-adapting DOdependent feeding method controls phase switching and adds more flexibility to the process.

A C K N O W L E D G M E N T S
The authors thank Karen Bunting from Ipsen for her critical reading of the manuscript. This study was sponsored by Ipsen.

C O N F L I C T O F I N T E R E S T Matthias Torsten Ehebauer is an employee and Stanislav
Pepeliaev was an employee of Ipsen.