Developing an Automatically Controlled Feeding Process in an E. coli Fermentation Process for Recombinant Protein Production

June 2, 2010
Luis Ducrey

Luis Ducrey is the head of the technology transfer department at GemaBiotech

Lorena Sabarots

Lorena Sabarots is a research assistant in cell line and bioprocess R&D at GemaBiotech

Raúl Medina

Raúl Medina is a research assistant in cell line and bioprocess R&D at GemaBiotech

Matilde Hermida

Matilde Hermida is a research assistant in cell line and bioprocess R&D at GemaBiotech

Florencia Rogers

Florencia Rogers is the head of cell line and bioprocess R&D at GemaBiotech

Eduardo Orti

Eduardo Orti is the R&D director at GemaBiotech

Julieta Aulicino

Julieta Aulicino in cell line and bioprocess R&D at GemaBiotech

BioPharm International, BioPharm International-06-02-2010, Volume 2010 Supplement, Issue 5

An exponential feeding strategy coupled with automation doubled protein yields while reducing fermentation time by 25%.


A strategy for improved process operational procedures was developed that enabled us to supervise and automatically control a fermentation feeding process in E. coli. Exponential feeding was used so that the cells could be grown at the desired specific growth rate (µ) by programming the bioreactor software. By providing proper nutrient and operating conditions, this strategy allowed us to reach a high cell concentration in less time, thus minimizing operational costs during scale-up and maximizing volumetric productivity and reproducibility.


The US Food and Drug Administration has been putting pressure on manufacturers to exert greater control over product quality by increasing their understanding of the data gathered from manufacturing processes. In particular, the process analytical technology (PAT) initiative calls on manufacturers to achieve those goals by:

  • incorporating new techniques for online monitoring

  • applying online measurements to detect deviations from "in control situations" before the process fails

  • using automatic feedback controls to correct processes if significant deviations from the set-point profiles appear.

Indeed, a key component for the commercial success of a biopharmaceutical product is the ability to carry out large-scale manufacturing in a robust and reproducible fashion.

The design and optimization of a fermentation process plays a key role in achieving high productivity and robustness at the production scale. In this article, we discuss a strategy to improve operational procedures in an Escherichia coli fed-batch fermentation through automated process control. To assess the results, we have compared the automated process, which relies on an exponential feeding strategy, to the most commonly used model of stepwise increase of feeding.

Experimental Design

In fed-batch fermentations, the feeding strategies commonly used for increasing cell concentration include feeding at a constant rate, making stepwise increases in feeding, feeding based on feedback control (based on pH or dissolved oxygen measurements [pH-stat or DO-stat]), and exponential feeding.

Exponential feeding makes use of an empirical model of cell growth to regulate the feeding rate. Ideally, by providing proper nutrient and operating conditions, the cells grow exponentially, achieving a high biomass concentration faster.1

In our model for producing recombinant proteins in E. coli, fermentation is divided into two phases. In the first phase, biomass generation takes place. In the second phase, the cells are devoted to product formation, after protein production has been induced with IPTG. In this fed-batch strategy, all nutrients except for carbon and oxygen are in excess throughout the process. We compared two possible alternatives for process operation with this model.

In the first approach, we combined simple indirect feedback methods (pH-stat and DO-stat) to determine when to start feeding, followed by manual stepwise increases in feeding based on glucose uptake.1,2

In this setup, fermentation began with a short batch phase (with the presence of glucose in the culture media), but once the carbon source was consumed and the dissolved oxygen (DO) or pH level rose above an upper limit as a result of substrate depletion, a loop in the bioreactor programming software started feeding at a fixed speed. Then we increased the speed of the feeding pump manually according to the glucose and oxygen uptake rates, until the end of the fermentation process (Figure 1).

Figure 1

In a second approach, we implemented an exponential feeding program designed according to the following equation:

in which F(t) is the feeding rate at time t (in L/h); F0 is the initial feeding rate (in L/h); uset is the selected specific growth rate (per h); and t is the time (in h). The coefficient F0 is known to be a function of X0 (the initial biomass concentration in the bioreactor), V0 (the initial media volume in the bioreactor), µ (the specific growth rate), and biomass yield.

In this approach, the growth media did not contain glucose and the process began in the exponential fed-batch mode, in which the substrate feeding rate controlled the specific growth rate (Figure 2).

Figure 2

During exponential feeding, cells can be grown at a desired specific growth rate (µset) by programming the bioreactor software for an exponential increase in the feeding rate. In this model, we assumed that the desired specific growth rate µset should be lower than the maximum growth rate, µmax. This is because at uset = umax, if the value of X0 (the initial biomass concentration in the bioreactor) drops below the value used to calculate the initial feeding rate (F0), the cells will not be able to increase their growth rate by consuming the excess glucose in the media, as they are already growing at the maximum specific growth rate. On the other hand, at uset < umax, the system will automatically compensate for a variation in the initial biomass concentration by increasing the growth rate, thus removing the deviation and ensuring a robust process.

Therefore, before the process began, a feed rate profile (µset pre-induction and µset post-induction) was designed according to Equation 1 and programmed into a programmable controller linked to the feed pump of the bioreactor. This was done by generating loops in the bioreactor software to correlate the feeding rate to µset (Figure 3).

Figure 3

In both approaches, we used a biomass probe (Fogale Nanotech, Nimes, France). This instrument allowed us to carry out online monitoring of the growth curves during the batch and fed-batch phases and enabled us to detect deviations from the expected growth before the process failed.

The nutrient composition of the feed media (basic defined mineral salt media with a high glucose concentration) and the fermentation variables such as temperature, pH, and DO were the same in both models. Samples were taken every hour during all fermentation runs. Cell growth also was followed by optical density measurement at 600 nm, and the biomass concentration was determined as dry cell weight (DCW) for further analysis in g/L.


Fermentation Growth Curves

A comparison of the typical growth curves of E. coli fermentations for the two models is shown in Figure 4. The measurements in picoFaraday/cm3 (pF/cm3 ) follow the same pattern as off-line OD and net weight (g/L) measurements (data not shown).

Figure 4a

In both approaches, the cultures remained in the lag phase during the first six hours. In the stepwise model, after the lag phase, the biomass increased and reached a specific growth rate of ~0.6/h (µmax) during the batch phase. Feeding was started once the glucose initially added to the medium was depleted (as determined by pH or a DO peak). At this point, the specific growth rate increased gradually from 0.08/h to 0.15/h in response to a manual increase in the feeding rate based on glucose consumption and biomass increases (Figure 4a).

To ensure good yield with this approach, the operator must regulate the flow of the feed solution to keep glucose levels in the bioreactor below a critical value established for the process (in this case, 1 g/L). It should be taken into consideration that the specific growth rate must not drop below 0.1/h before the point of induction.

In the automatically controlled feeding model, the biomass and feeding rate increased exponentially at µset until the induction point, when IPTG was added to the culture (Figure 4b).

Figure 4b

As seen in Figure 4b, the µset during the exponential feeding period was 0.35/h, and the cells went through at least two generations of growth at that rate until they were induced five hours later. It should be noted that no batch phase was observed, because there was no glucose in the basal media.

IPTG was added at 60 g/L dry cell weight (corresponding to 5–6 pF/cm3 ) and the exponential feeding was programmed at a µset of 0.1/h during the rest of the fed batch period.

In comparing the two fermentation processes, we see that with the stepwise feeding strategy it took 24 h to reach 3 pF/cm3 whereas with the exponential feeding strategy, the system reached 6 pF/cm3 in 18 h. It took longer to reach a high cell density in the stepwise model because the glucose additions were not optimized.

Standard Deviation and Productivity

To compare the protein yield and robustness of the exponential feeding process at different specific growth rates, we repeated the exponential fermentation strategy under the same operating conditions at least four times for each selected pre-induction specific growth rate (0.30/h, 0.33/h, 0.35/h, 0.37/h, and 0.4/h).

We then calculated the standard deviation (SD) in the biomass concentration (as dry cell weight) at induction and harvest (Figure 5), and the concentration of the recombinant protein generated (in g/L) by quantifying Coomassie stained gels.

Figure 5

Our results show that at a µset of 0.37/h, the SD decreased significantly during both induction and harvest compared to the other selected values. It is important to note, however, that except in the case of the specific growth rate of 0.4/h, the SD of the other preselected µ also were acceptable.

At a µset of 0.37/h, we also achieved a two-fold increase in protein production compared (9 g/L versus 4.5 g/L) to the stepwise feeding process (Figure 6). The yield of product per cell mass also was 30% higher in the exponential feeding setup, with a recombinant protein accumulation of 140 mg of protein/mg dry cell weight (data not shown).

Figure 6

It is important to note that despite the high µset of 0.37/h, acetate production value was kept below 800 mg/L during the process, which may explain the protein concentration obtained. 3–6


Because both overfeeding and underfeeding of nutrient in fed-batch operations is detrimental to cell growth and product formation, the development of a suitable feeding strategy is critical in fed-batch cultivation processes at production scale.

Operator-controlled systems are bound to fail because a fermentation process must be carefully formulated and monitored. Because the production of recombinant proteins requires the use of prokaryotic cells which are by nature a fluctuating biological system, it is very important to implement a detailed system for controlling the growth medium, oxygen incorporation, pH, temperature, and glucose additions.

Here, we suggest a robust model to overcome these problems while increasing protein production. In our particular case, productivity was doubled during the exponential feeding period and the elapsed fermentation time was reduced by 25%, enabling the cost-effective production of recombinant proteins.

Because exponential feeding allows cells to grow at a constant specific growth rate, acetate production can be minimized by keeping the cell growth below critical µ values (which are unique for each strain and growth medium). Moreover, exponential feeding allowed us to significantly decrease the standard deviation of the dry cell weight value during induction and harvest, improving batch-to-batch reproducibility.

It is important to note that although the stepwise feeding rate did not show an alarming deviation from the mean, it was controlled by the operator, which is far from ideal (Figure 7).

Figure 7

The automated system based on exponential feeding allowed us to achieve a reproducible, consistent, and operator-independent process. The introduction of minimal changes in the growth rate resulted in significant increases in the quality and quantity of the protein obtained.

These results suggest that to achieve high cell densities and production rates, it is important to have a controlled feeding strategy that minimizes fluctuations while ensuring a complete understanding of the data gathered from the processes.

JULIETA AULICINO is a research scientist and MATILDE HERMIDA, RAÚL MEDINA, and LORENA SABAROTS are research assistants, all in cell line and bioprocess R&D; LUIS DUCREY is the head of the technology transfer department, EDUARDO ORTI is the R&D director, and FLORENCIA ROGERS is the head of cell line and bioprocess R&D, all at GemaBiotech, Buenos Aires, Argentina. +54.11.4825.8029;


1. Kim BS, Lee SC, Sang Yup Lee, Chang YK, Chang HN. High cell density fed-batch cultivation of Escherichia coli using exponential feeding combined with pH-stat. Bioprocess Biosyst Eng. 2004;26:147–50.

2. Choi JH, Keum KC, Lee SY. Production of recombinant proteins by high cell density culture of Escherichia coli. Chem Eng Sci. 2006;61:876–85.

3. Akesson M, Hagander P, Axelsson JP. Avoiding acetate accumulation in Escherichia coli cultures using feedback control of glucose feeding. Biotech Bioeng. 2001;73(3):223–30.

4. Sandén A, Pedersen S. Limiting factors in Escherichia coli fed-batch production of recombinant proteins. Biotech Bioeng. 2002;81(2):158–64.

5. Aristidou AA, San KY, Bennett GN. Modification of central metabolic pathway in Escherichia coli to enhance recombinant protein production through acetate formation. Biotechnol Progress. 1994;11:475–78.

6. Eiteman MA, Altman E. Overcoming acetate in Escherichia coli recombinant protein fermentations. Trends Biotechnol. 2006;24(11):530–6.

Related Content:

Upstream Processing