Biotechnol. Appl. Biochem. (2005) 42, 227235 (Printed in Great Britain) doi:10.

1042/BA20040197

227

Miniature bioreactors for automated high-throughput bioprocess design (HTBD): reproducibility of parallel fed-batch cultivations with Escherichia coli
Robert Puskeiler*1 , Andreas Kusterer*, Gernot T. John and Dirk Weuster-Botz*
*Lehrstuhl f ur Bioverfahrenstechnik, Technische Universit at M unchen, Boltzmannstrasse 15, 85748 Garching, Germany, and PreSens GmbH, Regensburg, Germany
To verify the reproducibility of cultivations of Escherichia coli in novel millilitre-scale bioreactors, fully automated fed-batch cultivation was performed in seven parallel-operated ml-scale bioreactors with an initial volume of 10 ml/reactor. The process was automatically controlled by a liquid-handling system responsible for glucose feeding, titration and sampling. Atline analysis (carried out externally of the reaction vessel with a short time delay) comprised automated pH and attenuance measurements. The partial pressure of oxygen (pO2 ) was measured online by a novel uorimetric sensor block measuring the uorescence lifetime of uorophors immobilized inside the millilitre-scale bioreactors. Within a process time of 14.6 h, the parallel 1 cultivation yielded a dry cell weight of 36.9 + 0.9 g l . Atline pH measurements were characterized by an S.D. of < 1.1 % throughout the process. Computationaluid-dynamics simulation of single-phase ow yields a mean power input of 21.9 W l1 at an impeller speed of 2800 rev./min corresponding to a power number (NP ) of 3.7.

Introduction
The importance of high-throughput cell-cultivation technology for bioprocess design has been emphasized by many previous studies [1,2]. The generally applied bioprocess design strategy involves shake asks or MTPs (microtitre plates) for batch assays and laboratory-scale stirred tank bioreactors for further process optimization in fed-batch operation mode [3]. By this means, straightforward process development is usually hindered as a result of the reaction engineering limitations of shake asks and MTPs [4,5]. These main limitations are neither oxygen transfer [6,7] nor power input [8], but the limited possibility of controlling pH, conducting fed-batch processes and monitoring important process parameters online. Instrumented shake asks [9] and bubble columns [10] provide pH control and the possibility of performing fed-batch processes. Thus shake asks and bubble columns are appropriate tools for process design, although a parallelization to more than 16 units on the basis

of the applied technology seems difcult. The often-required use of laboratory-scale stirred tank bioreactors for further process optimization is very cost- and labour-intensive and therefore strongly reduces the throughput. Recently, novel approaches for further miniaturizing bioreactors to enable higher throughput and degree of automation have emerged. A stirred tank reactor with a nominal reaction volume of 6 ml was characterized by a maximum oxygen transfer coefcient, kL a, of 0.11 s1 [11]. The bafed reactor is equipped with a miniature sparger and an impeller driven by an electric motor. The incorporation of bre-optic probes enables the monitoring of pH and DO (dissolved oxygen concentration). Cell concentration is estimated from turbidity in the undiluted broth, which will most certainly lead to a limited measuring range depending on the properties of the optical devices. Parallelization of this miniature reactor might prove to be difcult, owing to the connection of the electric motor and the necessity to control the inlet gas ow in the range of ml min1 . To date, no technical solution concerning substrate or titration agent addition has been proposed. Compared with production processes, the maximum reported DCW (dry cell weight) of 1.6 g l1 is low due to the lack of pH control and the restriction to batch operation mode. A second approach consists of an array of eight MTP wells with a nominal volume of 250 l, which are separately aerated via a silicone membrane with electrolytically generated oxygen [3]. The realized OTR (oxygen transfer rate; g l1 h1 ) is comparable, with a kL a of 0.042 s1 . The general low mixing efciency in MTPs [12] was improved by inserting stainless steel balls into the wells. A printed circuit board was attached to the base of the wells containing light-emitting diodes and detectors for attenuance monitoring via light scattering. A linear response up to an attenuance, D, of 4 at a

Key words: aerobic process, automation, bioprocess design, fed-batch cultivation, high throughput, oxygen transfer. Abbreviations used: CFD, computational uid dynamics; CV, coefcient of variation; DO, dissolved oxygen concentration; MTP, microtitre plate; parameters are dened in Table 1. 1 To whom correspondence should be addressed (email r.puskeiler@lrz.tum.de).
C

2005 Portland Press Ltd

228

R. Puskeiler and others

Table 1

Parameters used Denition Substrate concentration Impeller diameter Attenuance at wavelength Dry cell weight Dissolved oxygen concentration Oxygen transfer coefcient Impeller speed Power number Mass ux substrate Oxygen transfer rate Oxygen uptake rate Power Partial pressure oxygen Liquid density Units g l1 m Unitless g l1 Air saturation (%) s1 s1 Unitless g h1 l1 g h1 l1 g h1 l1 W Air saturation (%) kg m3 Figure 1 Cross-section of a bioreactor mounted in the reaction block

Parameter cs DI D DCW DO kL a N NP M OTR OUR P pO2

wavelength of 600 nm was found. pH can be monitored by a solid-state pH-sensor chip mounted on the circuit board. The maximum reported attenuance, D, was 2.4 after a batch cultivation without pH control. A third approach is based on a magnetically driven gasinducing impeller mounted in a bafed bioreactor with a reaction volume of 815 ml [13]. The design as a gas-inducing impeller does not require gas sparging in the reaction volume, thus the individual control of gas inlet uxes in the range of ml min1 is not necessary. The maximum kL a was characterized to be > 0.4 s1 . An automated fed-batch cultivation in such a millilitre-scale bioreactor reached a DCW of 20.5 g l1 in a mineral medium with glucose as the sole carbon source. The growth of Escherichia coli was shown to be equivalent to a reference cultivation in a laboratoryscale stirred-tank bioreactor with a working volume of 3 litres [14]. The present study is the rst to report the application of the latter approach for the parallel cultivation of E. coli, aiming at the verication of the reproducibility of parallel cultivations. Therefore seven gas-inducing bioreactors at a millilitre scale are operated in parallel in a reaction block providing an electromagnetic drive and heat exchangers. The automation of titration, feeding and sampling is realized by a liquid-handling system. An integrated MTP reader enables automated pH and attenuance monitoring. The process is conducted with pH control in a fed-batch operation mode. Moreover, integrated bre-optic sensors in the ml-scale bioreactors allow the parallel monitoring of pO2 [partial pressure of oxygen (percentage air saturation)] with a frequency of six data points/min per reactor. The parameters used in this paper are dened in Table 1.

The lower part of the reaction block is equipped with the magnetic drive and heat exchangers to control the reaction temperature. The upper part of the reaction block contains heat exchangers to enable cooling of the headspace of the bioreactors thus limiting evaporation. The sterile barrier provides the gas inlet and the gas outlet, which simultaneously serves as sampling port. The shafts on which the impellers rotate freely are screwed into the sterile barrier.

Labortechnik, Oberschleissheim, Germany). The gasinducing impellers were machined from polyether ether ketone (PEEK) by a CNC (Computer Numerical Control) milling machine (H + P Labortechnik). The impellers are equipped with samarium/cobalt (SmCo) permanent magnets (Fehrenkemper Magnetsysteme, Lauenau, Germany). A cross-section of one bioreactor mounted in the reaction block is depicted in Figure 1. The dimensions of the bioreactor, the shaft and the gas-inducing impeller are detailed in Figure 2. Reaction block The reaction block (dimensions: length width height, 310 mm 240 mm 140 mm) ts a maximum of 48 bioreactors. The reaction block is equipped with the magnetic drive, heat exchangers and a sterile barrier providing the gas inlet and the gas outlet, which simultaneously serves as sampling port. The shafts are screwed into the sterile barrier. The gas-inducing impeller rotates freely on the shaft. Maximum impeller speed can be set to 4000 rev./min, corresponding to 66.6 Hz. CFD (computational uid dynamics) simulation of hydrodynamics in the millilitre-scale bioreactor The commercial CFD software package CFX-5.7 (Ansys, Otterng, Germany) was used for the modelling of the hydrodynamics of the millilitre-scale bioreactor lled to a liquid height of 40.5 mm corresponding to a liquid volume of 11.2 ml. The code is based on the nite volume technique. The reactor volume is resolved with a total number of 170 000 hybrid grid control volumes. For each of these

Materials and methods

The millilitre-scale bioreactors The bafed bioreactors with a nominal volume of 815 ml were produced from polystyrene by die casting (H + P
C

2005 Portland Press Ltd

Reproducibility in miniature bioreactors

229

(0.1 ml l1 ; Cognis, D usseldorf, Germany) was used. Feeding solution for fed-batch cultivation contained glucose (475 g l1 ), MgSO4 7H2 O (20 g l1 ), EDTA (13.0 mg l1 ), CoCl2 6H2 O (4.15 mg l1 ), MnCl2 4H2 O (24.9 mg l1 ), CuCl2 2H2 O (2.5 mg l1 ), H3 BO3 (5.0 mg l1 ), Na2 MoO4 2H2 O (4.15 mg1 l1 ), zinc acetate dihydrate (21.7 mg l1 ) and ferric citrate (47.6 mg l1 ). Seed culture The 50 ml seed culture in a 500 ml unbafed shake ask was inoculated with 0.067 % of a frozen E. coli cell stock and incubated at 37 C with shaking at 200 min1 for 16 h (Multitron; Infors, Bottmingen, Switzerland) until a DCUV650 of 3.0 (Genesys20; Thermo Electron, Dreieich, Germany) was reached. Millilitre-scale bioreactor culture The millilitre-scale bioreactors were inoculated with 100 % seed culture corresponding to an initial DCUV650 (attenuance in the cuvette photometer at 650 nm) of 3.0. The initial reaction volume was 10 ml. Stirrer speed was initially set to 2000 min1 . The gas ow through the sterile gas cover consisted of oxygen-enriched air. The volumetric uxes of air and oxygen were controlled by thermal mass ow controllers (Brooks Instruments, Veenendaal, The Netherlands). Total gas ow was initially set to 4.8 litres min1 air corresponding to 0.1 litre/min per reactor. During batch cultivation, the stirrer speed was manually increased to maintain pO2 > 20 % air saturation. During fed-batch culture, the volumetric fraction of oxygen was additionally increased up to a total oxygen concentration of 50 % in the inlet gas ow. pH was automatically controlled at 6.8 by intermittent addition of 12.5 % (v/v) NH4 OH. Pipetting precision The pipetting precision of the liquid handler was veried gravimetrically for the pipetting of feed solution, with a volume of 18.7 l corresponding to the highest feed ow of 13.3 g h1 l1 . The density of the feed solution was gravimetrically determined to be 1.19 g cm3 . The test comprised ten pipetting steps per pipette tip. The deviation of the tips from the desired pipetting volume was taken into account for the subsequent pipetting test. Monitoring of cell growth Cell growth was automatically monitored atline in MTPs by measuring the attenuance, DMTP650 (microtitre-plate attenuance at 650 nm), at a wavelength of 650 nm. Samples of 10 l were automatically diluted to 1:20 with PBS (8 g l1 NaCl, 0.2 g l1 KCl, 1.44 g l1 Na2 HPO4 and 0.24 g l1 KH2 PO4 , pH 7.4) by the internal pumps of the MTP reader. DCW was estimated from DMTP650 with a secondorder polynomial function allowing a measuring range of
C

Figure 2 Major dimensions of the ml-scale bioreactor, the shaft and the gas-inducing impeller as they are mounted inside the reaction block All the dimensions shown are in mm.

volumes, a full set of transport equations for mass and momentum is solved in a steady-state model approach. To represent the rotating parts of the reactor, the multiple reference frame technique is used with a rotating inner stirrer volume and a stationary outer vessel volume. Turbulence effects in the uid ow are prescribed using the SST (shear-stress-transport) model. The simulation is performed in a single-phase status using the uid properties of water at ambient temperature. Organism and medium The strain used for cultivation was E. coli K12 (DSM 498), purchased from the German Collection of Micro-organisms and Cell Cultures (Braunschweig, Germany). Seed cultures and bioreactor cultivation of E. coli were carried out in a dened mineral medium of the following composition (adapted from [15]): glucose (15 g l1 ), KH2 PO4 (13.3 g l1 ), (NH4 )2 HPO4 (4.0 g l1 ), MgSO4 7H2 O (1.2 g l1 ), CoCl2 6H2 O (2.5 mg l1 ), MnCl2 4H2 O (15 mg l1 ), CuCl2 2H2 O (1.5 mg l1 ), H3 BO3 (3.0 mg l1 ), Na2 MoO4 2H2 O (2.5 mg l1 ) and zinc acetate dihydrate (13 mg l1 ), citric acid monohydrate (1.86 g l1 ), EDTA (8.4 mg l1 ), ferric citrate (12.5 mg l1 ) and thiamine hydrochloride (4.5 mg l1 ). As the antifoam agent, CLEROL 265

2005 Portland Press Ltd

230

R. Puskeiler and others

undiluted samples between 0.16 and 1.74 DMTP650 units with a mean CV (coefcient of variance), R2 , of 0.995. Taking the dilution factor into account, the resulting DCW measuring range is 3.248.0 g l1 . Reference attenuance measurements (DCUV650 ) were carried out up to a process time of 8.77 h, i.e. the presence of the experimenter. A single beam photometer (Genesys20) with plastic cuvettes of 1 ml volume was used. Measurements were carried out at a wavelength of 650 nm. Monitoring of pH Commercially available MTPs containing pH-sensitive sensor spots (Hydroplate; Presens, Regensburg, Germany) were applied for automated atline pH measurements in an MTP reader (Fluostar Galaxy; bmg labtech, Offenburg, Germany). The hydroplates were calibrated with cultivation medium adjusted to six different pH values covering the measuring range between pH 4 and 9. Monitoring of DO DO was monitored online with uorimetric sensor spots (Presens) immobilized on to the reactor bottom with the silicone rubber compound 692542 (RS Components, Corby, U.K.). The uorescence decay time of the uorophors was measured with a prototype uorescence reader (Presens) containing eight separate light sources and photodiodes integrated into a sensor block tting beneath the reaction block. Measurements were performed with a frequency of six data points/min per reactor. Process control All sampling and feeding steps as well as the MTP movements were programmed with the liquid-handler control software (Gemini; Tecan, Crailsheim, Germany). Sampling of 50 l h1 was carried out by a liquid-handling system (RSP 150; Tecan) with a frequency of 1 h1 during batch growth. An additional sample was taken 30 min after the initial glucose was metabolized. During fed-batch growth, samples were taken each time the mass ow of feed solution was increased. From the process time of 6.9 h until the end of the cultivation, samples were taken with a frequency of 2 h1 . The samples were pipetted into an MTP and then transported to the plate reader (Fluostar Galaxy) that measured pH in undiluted samples and attenuance DMTP650 in 1:20 diluted samples. Communication with the MTP reader was programmed in LabView (version 6.1; National Instruments, Munich, Germany). LabView VIs (virtual instruments) were started from the command line in the liquid-handler control software. Closed loop control for pH was achieved by extracting pH values from the MTP reader data les, processing them in LabView and creating pipetting instructions for the liquid-handler control software that were regularly loaded into the control program. pH
C

was controlled at 6.8 with a simple proportional controller calculating the required amount of base for any single reactor by applying a titration curve that was created in advance. No addition of acid was implemented into the pH-control algorithm. During fed-batch growth, the titration agent was added with a frequency of 3 h1 . After each sampling step, the pH controller increased the base mass ow proportionally according to the measured pH deviation from the set point. After atline measurements, the diluted samples were recovered for manual attenuance measurements in a single beam cuvette photometer (DCUV650 ). The MTP was cleaned in an MTP washer and re-used to receive the subsequent samples. After 8 h of process time, a new hydroplate was used to maintain high reproducibility of pH measurements. After the consumption of the initial glucose at a process time of 4.9 h, feeding was started intermittently with a frequency of 15 h1 corresponding to a feeding interval of 4 min. After 40 and 90 min, the initial feed ow of 4.0 g h1 l1 was increased to 7.5 and 13.3 g l1 h1 respectively. Hence, the highest feeding volume was 18.7 l for each pipetting step.

Results and discussion

Hydrodynamics of the millilitre-scale bioreactor Figure 3 shows the results of the single-phase CFD simulations at an impeller speed of 2800 rev./min. Owing to the rotational movement, the liquid phase is accelerated along the diagonal outward pumping channels, thereby reaching a top speed of > 2.0 m s1 . The strong radial ow induced by the impeller leads to a circulating ow eld above and below the impeller comparable with the ow eld of a single Rushton turbine. As a result of the acceleration of the liquid phase in the diagonal channels, a constant negative pressure builds up in the vertical bottom channel of the impeller as shown in the spatial distribution of the simulated pressure in the liquid phase. The peak negative pressure (< 2000 Pa) is localized around the hollow shaft in the vicinity of the upward pumping diagonal channels. Moreover, negative pressure builds up behind the bafes, thus facilitating the suckingin of gas bubbles when the diagonal channels pass the bafes during the rotational movement. The peak positive pressure (> 2000 Pa) is localized in the zone where the outward pumped liquid encounters the bafes. The zones of highest positive or negative pressure correspond to the zones of highest energy dissipation. Since the integral of the simulated energy dissipation over the liquid volume generally underestimates the power input [16], this variable was calculated by solving the momentum balance of the ow eld. This approach yields a mean power input of 21.9 W l1 for the millilitre-scale bioreactor at 2800 min1 . The singlephase power number NP = 3.7 is obtained by assuming a

2005 Portland Press Ltd

Reproducibility in miniature bioreactors

231

Figure 3 Results of single-phase CFD simulations of axially projected liquid speed and pressure in the millilitre-scale bioreactor for an impeller speed of 2800 rev./min The liquid speed is represented in a vertical cross-section through the centre of the bioreactor. The pressure distribution is represented in a vertical cross-section lying 0.7 mm in front of the centre of the bioreactor in order to display the difference between positive pressure in front of the bafes (right) and negative pressure behind the bafes (left).

liquid density of = 998 kg m3 in the equation: Np = P / N3 D I5 Parallel fed-batch cultivation After three optimization cycles, the accuracy of the eight pipette tips of the liquid-handler was increased from the initial value of 87.3 to 99.9 % with a precision of 3.0 % of 10 subsequent pipetting steps. Optimization was carried out for a volume of 18.7 l of feed solution (results not shown) corresponding to the glucose mass ow of 13.3 g h1 l1 . Results of parallel cultivation are shown in Figures 4, 6 and 8. During batch growth, a maximum mean growth rate 1 () of 0.49 + 0.06 h was reached in the seven, millilitrescale, bioreactors. The initial glucose concentration of 1 15 g l1 led to a nal DCW of 7.2 + 0.5 g l corres+ ponding to a biomass yield (Y XS ) of 0.48 0.03 g g1 . The DCW estimations based on measurements in the MTP reader were characterized by a decreasing CV (Figure 5A). The initially high CV of more than 10 % of the data from the MTP reader is a result of the values corresponding to the lower range of the preliminarily determined measurement range. The CVs of DCW estimations derived from DCUV650 measurements in the single-beam photometer are comparable with the CVs of measurements in the MTP reader at process times above 3 h (Figure 5B). Owing to

intermittent pH control with a frequency of 1 h1 , pH decreases to the value plotted in Figure 6 before the titration agent is added. During batch growth, the reduction in pH/h increases due to the increase in cell concentration. As a result of acetate metabolism at the end of batch growth, pH increases above the set point of 6.8 and reaches 7.05 + 0.05 at a process time of 4.8 h. During batch growth, a mean total base volume (V Base ) of 271 + 7 l was added to the cultures. pO2 during batch growth was manually controlled at 20 % by increasing the impeller speed. After glucose depletion, indicated by a sharp increase in pO2 at a process time of 3.76 + 0.04 h, the main metabolic by-product acetate was consumed by the cells. At the beginning of fed-batch operation, the glucose mass ow was gradually increased in two steps to accommodate the culture to intermittent feeding. At the end of 1 the process, a mean DCW of 36.9 + 0.9 g l was reached in the seven bioreactors (Figure 4). During fed-batch growth, the pH oscillated between 6.85 and 6.36 in all seven reactors due to the intermittent pH control with a frequency of 3 h1 (Figure 6). In all samples taken during fed-batch growth, the CV of pH did not exceed 1.1 % (Figure 7). The mean total base volume added was 553.6 + 31.6 l. From the beginning of fed-batch cultivation, pO2 oscillates, owing to the intermittent feeding (Figure 8). After the addition of glucose to the culture, pO2 decreases as a
C

2005 Portland Press Ltd

232

R. Puskeiler and others

Figure 5 Coefcient of variation of DCW measurements in the seven, millilitre-scale, bioreactors (A) CV of DCW atline measurements of the seven, millilitre-scale, bioreactors in the MTP reader (DMTP650 ). (B) CV of DCW ofine measurements of the seven ml-bioreactors in the single-beam photometer (DCUV650 ).

Figure 4 Parallel, automated, fed-batch cultivation of E. coli K12 in seven, millilitre-scale, bioreactors in a mineral medium with glucose as the sole carbon source After exponential batch growth, a linear, stepwise increasing feeding prole was started. Time course of DCW, substrate concentration (cS ; g l1 ) and substrate mass ow (MS ) are shown. Top to bottom: Reactors numbers 17. 2005 Portland Press Ltd

result of the higher OUR (oxygen uptake rate; g l1 h1 ) of the culture. After glucose depletion, pO2 increases to its initial value. At the maximum glucose mass ow of 13.3 g h1 l1 , a single glucose pulse results in a transitional glucose concentration of 0.89 g l1 . Although this value is higher than the critical concentration of acetate formation of 0.03 g l1 reported for E. coli [17], the culture is able to metabolize the by-product during the feeding interval of 4 min, since no accumulation of acetate could be measured until the end of the process. After 14.6 h, the process started running into oxygen limitation while the stirrer speed and volumetric oxygen fraction in the inlet gas were kept constant. Assuming (i) constant kL a by neglecting composition changes in the culture medium and (ii) negative inuence of increase in cell concentration, constant stirrer speed and volumetric oxygen fraction in the gas inlet ow, guarantees a constant OTR. Nevertheless, during intermittent feeding, the time required to metabolize the added glucose decreases as the cell concentration increases with ongoing process time (Figure 9). The transitional decrease in pO2 concentration after glucose addition is sharper and may thus lead to transitional oxygen limitation at higher cell densities. Monitoring pO2 enables the estimation of the OUR of the culture. With an estimated kL a of 0.22 s1 of the impeller

Reproducibility in miniature bioreactors

233

Figure 7 CV of pH atline measurements of the seven, millilitre-scale, bioreactors in the MTP reader with hydroplates

at 2600 rev./min [18] and a calculated oxygen solubility of 6.09 mg l1 for air saturation at 37 C in the mineral medium [19], the OUR prole depicted in Figure 10(A) can be derived. During the constant feeding with a glucose mass ow of 13.3 g l1 h1 , the mean OUR gradually increases from approx. 4 g l1 h1 to almost 6 g l1 h1 . An increase in oxygen uptake during constant feeding is a result of the increase in maintenance of oxygen demand compared with the oxygen demand due to growth. The lowest pO2 during a feeding interval represents a maximum OUR of the culture. Owing to the intermittent feeding prole, the maximum OUR is more than 2-fold higher than the mean OUR of the culture (Figure 10B). Without taking the highest CV of the rst three DCW measurements in the MTP reader into account, the overall mean CV is 4.4 %. During 14.6 h, a mean DCW of 1 36.9 + 0.9 g l was reached. All relevant process data are summarized in Table 2. Since one of the major aims of bioprocess development is the maximization of cell concentration per process time, we compared the data reported here with the results of Korz et al. [20], who developed a high-cell-density cultivation process for E. coli by applying an exponential feeding prole. Korz et al. reported an E. coli TG1 cell density of 35 g l1 biomass concentration after 22 h of exponential feeding in a mineral medium. However, they started with a lower inoculum density of approx. 0.1 g l1 resulting in a longer batch phase of 12 h. Presuming an equal inoculum density of approx. 1.3 g l1 as in the present study, Korz et al. reached their DCW of 35 g l1 after approx. 15 h, a time span comparable with the process time in the present study. The applied linear feeding strategy was thus able to compete with the high cell density culture developed by Korz et al [20].

Figure 6 Parallel, automated, fed-batch cultivation of E. coli K12 in seven, millilitre-scale bioreactors in a mineral medium with glucose as the sole carbon source

Time course of pH as measured and controlled intermittently. For details on titration frequencies, see the Process control subsection of the Materials and methods section. Top to bottom: reactors numbers 17. 2005 Portland Press Ltd

234

R. Puskeiler and others

Figure 9 Plot of pO2 versus time during one feeding interval of 4 min for three different process times and thus cell concentrations Zero time corresponds to a feeding step at the feed mass ow of 13.3 g l1 h1 . , tP (process time) = 8.6 h (DCW = 21.3 g l1 ); , tP = 10.2 h (DCW = 26.6 g l1 ); , tP = 14.1 h (DCW = 35.8 g l1 ). For clarity, broken lines have been added.

Table 2 Summary of mean process parameters of fed-batch cultivation in seven, millilitre-scale, bioreactors Process phase Batch Parameter DCW (g l1 ) (h1 ) Y XS (g g1 ) pH V Base (l) pO2 increase DCW (g l1 ) V Base (l) Mean value 7.18 0.49 0.48 7.05 271.2 3.76 36.87 553.6 S.D. 0.46 0.06 0.03 0.05 7.0 0.04 0.91 31.6 CV (%) 6.5 12.3 6.5 0.7 2.6 1.1 2.5 5.7

Fed-batch

Figure 8 Parallel, automated, fed-batch cultivation of E. coli K12 in seven, millilitre-scale, bioreactors in a mineral medium with glucose as the sole carbon source 2005 Portland Press Ltd

Time course of impeller speed, oxygen fraction in the inlet gas ow and pO2 . Top to bottom: reactors numbers 17. Impeller [min1 ] means impeller speed in rev./min.

Reproducibility in miniature bioreactors

235

References
1 2 3 4 5 Elmahdi, I., Baganz, F., Dixon, K., Harrop, T., Sugden, D. and Lye, G. J. (2003) Biochem. Eng. J. 16, 299310 Lye, G. J., Ayazi-Shamlou, P., Baganz, F., Dalby, P. A. and Woodley, J. M. (2003) Trends Biotechnol. 21, 2937 Maharbiz, M. M., Holtz, W. J., Howe, R. T. and Keasling, J. D. (2004) Biotechnol. Bioeng. 85, 376381 Weuster-Botz, D., Stevens, S. and Hawrylenko, A. (2002) Biochem. Eng. J. 11, 6972 Duetz, W. A., Ruedi, L., Hermann, R., OConnor, K., B uchs, J. and Witholt, B. (2000) Appl. Environ. Microbiol. 66, 26412646 Hermann, R., Lehmann, M. and B uchs, J. (2003) Biotechnol. Bioeng. 81, 178186 Maier, U., Losen, M. and B uchs, J. (2004) Biochem. Eng. J. 17, 155167 B uchs, J., Maier, U., Milbradt, C. and Zoels, B. (2000) Biotechnol. Bioeng. 68, 589593 Weuster-Botz, D., Altenbach-Rehm, J. and Arnold, M. (2001) Biochem. Eng. J. 7, 163170 Dilsen, S., Paul, W., Herforth, D., Sandgathe, A., Altenbach-Rehm, J., Freudl, R., Wandrey, C. and Weuster-Botz, D. (2001) J. Biotechnol. 88, 7784 Lamping, S. R., Zhang, H., Allen, B. and Shamlou, P. A. (2003) Chem. Eng. Sci. 58, 747758 Weis, S., John, G. T., Klimant, I. and Heinzle, E. (2002) Biotechnol. Prog. 18, 821830 Puskeiler, R., Zacher, H. K. and Weuster-Botz, D. (2003) Eur. Pat. Appl. PCT EP03/14752 Puskeiler, R., Kaufmann, K. and Weuster-Botz, D. (2004) Biotechnol. Bioeng. 89, 512523 Riesenberg, D., Schulz, V., Knorre, W. A., Pohl, H. D., Korz, D., Sanders, E. A., Ross, A. and Deckwer, W. D. (1991) J. Biotechnol. 20, 1728 Gentric, C., Mignon, D., Bousquet, J. and Tanguy, P. A. (2005) Chem. Eng. Sci. 60, 22532272 Vrabel, P., van der Lans, R. G. J. M., van der Schot, F. N., Luyben, K. C. A. M., Xu, B. and Enfors, S. O. (2001) Chem. Eng. J. 84, 463474 Puskeiler, R. (2004) Ph.D. Thesis, Munich University of Technology Schumpe, A. (1993) Chem. Eng. Sci. 48, 153158 Korz, D. J., Rinas, U., Hellmuth, K., Sanders, E. A. and Deckwer, W. D. (1995) J. Biotechnol. 39, 5965

6 7 8 9
Figure 10 OUR in the seven ml-bioreactors (A) Comparison of maximum OUR () with mean OUR (). Estimation is based on a kL a of 0.22 s1 and an oxygen solubility of 6.09 mg l1 at air saturation. (B) The ratio of maximum OUR to mean OUR during intermittent feeding.

10

11 12 13

The higher maximum oxygen demand of a microbial culture during intermittent feeding could be satised by the millilitre-scale bioreactors at an oxygen concentration of 50 % in the inlet gas ow. Our future work will focus on the establishment of a closed-loop control of impeller speed and oxygen concentration in the inlet gas ow, to enable further prolongation of process times and thus maximization of DCW. Moreover, the process control software currently adapted is capable of controlling 48 bioreactors simultaneously.

14 15

16 17

18

Acknowledgments
We gratefully acknowledge the contribution of D. Ortlieb (CFD Consultants, Rottenburg, Germany) in modelling the hydrodynamics of the millilitre-scale bioreactor and the excellent technical assistance of M. Amann. This work was partially supported by the Deutsche Bundesstiftung Umwelt.
19 20

Received 9 December 2004/15 April 2005; accepted 26 April 2005 Published as Immediate Publication 26 April 2005, doi:10.1042/BA20040197

2005 Portland Press Ltd