Journal of Bioscience and Bioengineering VOL. 107 No. 4, 419 424, 2009 www.elsevier.

com/locate/jbiosc

Estimation of dissolved carbon dioxide stripping in a large bioreactor using model medium
Naoki Matsunaga,1, Kenjiro Kano,1 Yasuyuki Maki,2 and Toshiaki Dobashi2
Bio Process Research and Development Laboratories, Kyowa Hakko Kirin Company, Limited, Hagiwara-cho, Takasaki, Gunma 370-0013, Japan 1 and Department of Chemistry and Chemical Biology, Faculty of Engineering, Gunma University, Tenjin-cho, Kiryu, Gunma 376-8515, Japan 2 Received 17 June 2008; accepted 24 November 2008

Dissolved carbon dioxide (dCO2) accumulation is one of the most serious problems in the scale-up of industrial cell culture. To predict the effects of dCO2 stripping in different culture conditions and at different scales, we examined a method of estimation of dCO2 stripping using a model medium. The operational parameters (e.g., sparging and agitation rate) and the size of the bioreactor (working volume: 80 L, 500 L, 2000 L; aspect ratio: 1.0 1.6) were varied, and the model medium was prepared by adjusting pH, density, viscosity, surface tension, and buffer conditions. dCO2 stripping efficiency was evaluated using the index kLaCO2, which was defined in accordance with the volumetric oxygen transfer coefficient kLa. The model medium exhibited dCO2 stripping behavior similar to real culture medium in all experimental conditions tested. It is expected that the use of the model medium to estimate dCO2 stripping in real cultures will be valuable for determining the culture conditions in bioreactors in scale-up. 2009, The Society for Biotechnology, Japan. All rights reserved.
[Key words: Dissolved carbon dioxide stripping; Scale-up; Bioreactor; Model medium; Gas exchange]

Mammalian cell cultures, such as Chinese hamster ovary (CHO) cells, have been used increasingly to manufacture biomedicines (e.g., antibody medicines). CHO cells are preferred for the mass production of recombinant proteins, since they are ideal hosts with respect to rapid growth rate, high stability and efficient foreign gene expression (1). Over the last decade, many studies promoting cell growth or an increase in the productivity of proteins have been conducted (25). In general, although antibody medicines show low toxicity, large dosages are required for effective therapy. Therefore, it is necessary to establish a high-density culture system in large-scale bioreactors. The scale-up of culture starts from a few liters in the laboratory and progresses through many hundreds of liters up to several thousand liters according to stage of development. In the scale-up of industrial cell cultures, dissolved carbon dioxide (dCO2) build-up is one of the most serious problems. The accumulation of dCO2 affects cell growth, productivity, and the structure of polysaccharide post-translational modifications, among other factors. Gray et al. observed that higher levels of pCO2 (N105 mmHg) resulted in CHO cell growth inhibition and a dramatic reduction in productivity (6). Kimura et al. reported a dose-dependent inhibition of cell growth by elevated pCO2 in the range of 36250 mmHg, while 250 mmHg pCO2 inhibited the specific growth rate of the MT2-1-8 CHO cell line by 30% under constant osmolarity (7). In addition, for CHO-derived recombinant tissue plasminogen activator (tPA) it was reported that the proportion of sialic acids comprising N-glycolylneuraminic acid
Corresponding author. Fax: +81 027 353 2094. E-mail address: naoki.matsunaga@kyowa-kirin.co.jp (N. Matsunaga).

decreased from 2.34.0% at 36 mmHg pCO2 to 1.52.2% at 250 mmHg pCO2 (8). These results suggest the importance of the dCO2 stripping in large-scale cultures. Therefore, dCO2 stripping is a very important operational factor for the scale-up of bioreactors. Conventionally, dCO2 stripping has been performed by surface aeration, sparging air into the medium, and agitation in real cultures. However, dCO2 stripping from the surface becomes ineffective in scale-up, because the liquid surface-to-volume ratio decreases as the size of commercial bioreactors is increased. In addition, the gas sparging and agitation rates are restricted to low levels because animal cells are very sensitive to shear stress and are easily damaged (912). These restrictions induce dCO 2 accumulation in large bioreactors. Moreover, the system of dCO2 stripping from the medium is very complex, because it depends on many features of the bioreactor (e.g., the geometrical conditions characterized by the size and shape of the tank, the position and tip speed of the impeller, the volume of surface aeration, the sparging rate, the sparger pore size, the surface area of the sparger, and the chemical equilibrium of CO2). Therefore, it is difficult to predict analytically the dCO2 stripping rate for each set of operational conditions for each bioreactor. Computational fluid dynamics (CFD) is a method for predicting dCO2 stripping and other characteristic quantities in cell cultures, and can be very useful in some cases (1315). However, it is necessary to select a suitable theoretical formula and determine critical parameters, and it takes a long time to perform the complex calculations. If it can be used to optimize the operating conditions when a culture tank is used, it may be more advantageous than CFD simulation. On the other hand, largescale bioreactors of 1000 L capacity or more for mammalian cell

1389-1723/$ - see front matter 2009, The Society for Biotechnology, Japan. All rights reserved. doi:10.1016/j.jbiosc.2008.11.017

420

MATSUNAGA ET AL.

J. BIOSCI. BIOENG.,

Nomenclature d [dCO2]0 specific gravity [g/mL] ratio of mole of dissolved carbon dioxide at the start of air sparging to mole of total gas dissolved in the medium [%] ratio of mole of dissolved carbon dioxide at a steady state after a long sparging period to mole of total gas dissolved in the medium [%] ratio of mole of dissolved carbon dioxide at sparging time t to mole of total gas dissolved in the medium [%] bubble diameter [m] Etvs Number [] amount of CO2 stripping per sparged [mol L 1] gravity acceleration [m/s2] Henry constant [mol L 1 bar 1] proton concentration [mol L 1] volumetric mass transfer coefficient [h 1] equilibrium constants [] volumetric oxygen transfer coefficient [h 1] volumetric dissolved carbon dioxide transfer coefficient [h 1]

[dCO2]s

k L M n p0 pCO2 Q Re r U V f vvm

[dCO2]t

de E0 f g H [H+] K K1, K2 kLa kLaCO2

rate of mass transfer between bubble and liquid [m/s] liquid depth [mm] Morton Number [] number of liquid drops [] atmospheric pressure [bar] partial pressure of dissolved carbon dioxide [mmHg] bubble volume sparged per unit time to bioreactor [L min 1] Reynolds number [] radius of bubble [mm] rising speed of bubble [m/s] liquid volume [L] terminal velocity of bubble [mm/s] gas volume per liquid volume per min [L min 1 L 1] coefficient due to agitation [] surface tension [dyn/cm] difference between gas density and liquid density [Kg/m3] coefficient of viscosity [Pa/s] Q/V [L min 1 L 1] density [Kg/m3] surface tension [N/m] lifetime of bubble [s]

culture are not as widespread, and it is difficult in practical terms to use real culture medium as a test solvent because of its high cost. Fortunately, different working-volume bioreactors (80 L, 500 L, and 2000 L) are at hand. The use of a model medium is a potential method for estimating dCO2 stripping from real cultures. This method is advantageous because it is simple, rapid, and cheap, and it can be applied to various types of bioreactors. Here, we examined dCO2 stripping by air sparging in bioreactors with a wide range of volumes from 80 L to 2000 L for test liquids of a real culture medium and a model medium with a similar density, surface tension, viscosity and pH, and the same amount of sodium bicarbonate, and water as a control. The results of the experiments confirm that the model medium is useful for the estimation of CO2 stripping in real cultures.
MATERIALS AND METHODS Bioreactors All data for this paper were obtained using 80 L, 500 L, and 2000 L bioreactors with impellers, baffles and a sparger (mesh size: 100 m), as shown in Fig. 1.

The volumes of the bioreactors are the basic working volumes. The bioreactors of 80 L and 500 L had no middle paddle. The aspect ratio (L / D) was in the range of 1.01.6. The liquid surface-to-volume ratios of the bioreactors were 0.020, 0.010, and 0.007 cm 1 for the 80 L, 500 L, and 2000 L bioreactors, respectively. Sensors (temperature, pH, and dCO2) were present in each bioreactor. Preparation of medium and model medium Water was purified by ultrafiltration and deionization. Culture medium was prepared by dissolving a commercial medium (EX-CELL325, SAFC) and sodium bicarbonate in the purified water, in accordance with the manufacturer's protocol. Model medium was prepared by mixing the following reagents in purified water at concentrations determined by trial and error to adjust viscosity, surface tension, and pH to those of the real culture medium: D-glucose, 25 g/L; Pluronic F-68 (PF68), 1 g/L; sodium bicarbonate, 1.6 g/L; and sodium chloride, 7.6 g/L. Measurement of viscosity and surface tension The viscosity of the test liquids was measured twice using an Ubbelohde viscometer at 25 C and the average was calculated. The surface tension of the test liquids was measured using the drop number method. The numbers of 0.5 mL drops of the test liquids flowing through a graduated pipette were counted, and the surface tension of the medium or the model medium

FIG. 1. Configuration of the 80 L, 500 L, and 2000 L bioreactors: 80 L and 500 L bioreactors have only one wide impeller inclined near the bottom, and the 2000 L bioreactor has an impeller not inclined near the bottom and an impeller inclined at the middle. All bioreactors have four baffles. The aspect ratios were from 1.0 to1.6. The liquid depths were 520 mm (80 L), 905 mm (500 L), and 1530 mm (2000 L).

FIG. 2. Experimental setup for the foam stability test.

VOL. 107, 2009

TABLE 1. Comparison of density, surface tension, and viscosity of test liquids Test liquid Water Medium Model medium Density (g/mL) 1.00 1.01 1.01 Surface tension (dyn/cm) 71.8 59.6 58.4 Viscosity (cP) 0.89 0.95 0.96

ESTIMATION OF DCO2 STRIPPING IN A LARGE BIOREACTOR

421

TABLE 2. Foam height measured at different times on the surface of the test liquids Test liquid Temperature (C) 22.9 23.0 22.9 Foam height (cm) 0(s) 0.4 9.0 6.5 15(s) 0.0 5.7 4.5 30(s) 0.0 1.6 1.0 60(s) 0.0 0.9 0.6 120(s) 0.0 0.8 0.6 180(s) 0.0 0.8 0.6

Water Medium Model medium

(2) was obtained from the number of drops (n1 for water, and n2 for the medium or the model medium) and known surface tension of water (1), using the equation: g1 n2 d1 = ; g2 n1 d2 1

where d1 and d2 are the specific gravities of water, and the medium or the model medium, respectively. Test for stability of foam The production and stability of the foam was measured in line with a modified RossMiles method (ISO Standard 696-1975(E)) as follows. The apparatus consisted of two glass vessels placed one above the other, as shown in Fig. 2. Fifty milliliters of each test liquid was poured into the lower vessel (a graduated cylinder) and then 300 mL of the test liquid was added to the upper vessel. The test liquid was allowed to run from the upper vessel through a duct into the test liquid in the lower vessel to yield foam. The time course of the height of the foam was measured. Measurements were obtained three times at an adjusted temperature of about 23 C and the average was calculated. Measurement of bubble diameter The acrylic vessel was filled with the test liquids (medium or water) and was sparged with O2 gas. Bubbles were observed with a high-speed camera and the image was analyzed. The diameters of 100300 bubbles were measured, and the mean value was calculated. This data was obtained by Hitachi, Ltd., as a contract research project. The dCO2 sensor used for this study was Evaluation of dCO2 stripping YSI8500 (YSI) (16). The sensor was calibrated with water saturated by 5% CO2 gas. The bioreactors were filled with the test liquids (medium, model medium, or water) up to the level of the basic working volume, and CO2 gas was sparged into the test liquids until dCO2 reached about 20%. Then, air sparging commenced with agitation under the desired operational conditions. The dCO2 levels in the test liquids were measured continuously at 2428 C. Defined by the following expression, kLaCO2, was used as an index for evaluating the efficiency of dCO2 stripping. kL aCO2 = ln dCO2 0 dCO2 s ln dCO2 t dCO2 s ; t 2

where [dCO2]s, [dCO2]0, and [dCO2]t are dCO2 at a steady state after a long sparging period ( 0), at the beginning of air sparging (t = 0), and at sparging time, t, respectively.

TABLE 3. Comparison of mean bubble diameter (mm) sparged into water and the medium Sparging volume (mL min 1) Sparger A Water Medium Sparger B Water Medium Sparger C Water Medium 40 0.5 (0.015) 0.3 (0.007) 0.9 (0.045) 0.5 (0.009) 2.0 (0.045) 1.4 (0.037) 100 0.5 (0.015) 0.3 (0.006) 1.0 (0.040) 0.6 (0.016) 1.7 (0.028) 1.2 (0.029) 200 0.6 (0.019) 0.3 (0.006) 1.1 (0.039) 0.5 (0.014) 1.8 (0.021) 1.0 (0.012) 400 0.7 (0.026) 1.1 (0.039) 0.5 (0.009) 2.1 (0.036) 1.1 (0.010) 800 1.0 (0.039) 1.6 (0.053) 0.5 (0.011) 2.2 (0.051) 1.0 (0.021) 1500 1.5 (0.072) 1.9 (0.064) 2.2 (0.049) 1.0 (0.013) FIG. 3. kLaCO2 as a function of vvm in bioreactors with volumes of 80 L (a), 500 L (b), and 2000 L (c) at the indicated tip speeds for different test liquids.

The mesh sizes of spargers A, B, and C were 10 m, 40 m, and 100 m, respectively. Figures in parentheses indicate standard error.

422

MATSUNAGA ET AL. RESULTS AND DISCUSSION

J. BIOSCI. BIOENG., is known that the forms of the bubbles (e.g., spherical, wobbling, spherical-cap and skirted) are estimated by a combination of nondimensional numbers of the Etvs Number (E0), Morton Number (M), and Reynolds number (Re) as follows (17). E0 = M= gDq d2 e ; r g l4 Dq ; q2 r3 q de U ; l

Comparison of physical properties of test liquids Table 1 shows the density, surface tension, and viscosity of each test liquid. The values for the medium and model medium were roughly the same; however, the surface tension of water was higher and the viscosity of water was lower than those of the medium and model medium. Table 2 shows the time course of the foam height measured at 01800 s, which is an index of foam yield and foam stability. It suggests that the yield of the foam was very small and the foam disappeared quickly in water (low-stability foams), whereas a large yield of the foam was obtained and it did not breakdown easily in the medium and model medium (high-stability foams). In the sparging test with real bioreactors, we observed that the liquid surface was covered with foam in the medium and model medium, whereas the water surface was not covered with foam. The results suggest that the lifetime of the foam in the medium and model medium is much longer than that in water. Bubbles sparged into the test liquids were observed with a high-speed camera. The shapes of the bubbles in medium were regarded as spheres, although they were slightly distorted in water. It

Re =

where g is the gravity acceleration (m/s2); is the difference between the gas density and liquid density (Kg/m3); de is the bubble diameter (m); is the surface tension (N/m); is the coefficient of viscosity (Pa/s); is the density (Kg/m3); and U is the rising speed of the bubble (m/s). These values were calculated for water and medium as follows: E0 : 0:55 water; 0:17 medium

FIG. 4. Comparison of CO2 in test liquids at the indicated tip speeds in bioreactors with volumes of 80 L (a), 500 L (b), and 2000 L (c).

VOL. 107, 2009 M : 1:7 1011 water; 3:7 1011 medium Re : 1:1 103 water; 5:3 103 medium: From the previously established diagram for the shape of bubbles as a function of these parameters (17), these values indicate that the bubbles sparged in water and medium are spherical, which is consistent with the observations obtained using a high-speed camera. The bubble diameters in the test liquids are given in Table 3. The characteristic behaviors result from the differences in the physical quantities of the medium and model medium compared with water (Table 1). The bubbles are smaller in the medium than in water (Table 3), which is attributed to the lower surface tension of the medium. As shown in Table 1, the surface tension and viscosity of the model medium were successfully adjusted to those of the real culture medium. Therefore, a similar smaller bubble size is expected for the model medium.

ESTIMATION OF DCO2 STRIPPING IN A LARGE BIOREACTOR

423

Comparison of CO2 stripping efficiency of test liquids Figs. 3ac shows the effect of tip speed, vvm (sparging gas volume per liquid volume per min), and bioreactor volume on kLaCO2, with no aeration of the liquid surface of each test liquid. kLaCO2 increased with increasing tip speed and vvm for all the test liquids. On the other hand, the kLaCO2 values of the medium and model medium were lower than that of water when compared at the same conditions. Where the rate of gas transfer between a spherical bubble and liquid is slow, the relationship between the volumetric mass transfer coefficient K and sparging volume per liquid volume per min vvm = Q / V is expressed as follows (18). K= aQ s aaL X ; = V vf 3

where V is the liquid volume; Q is the bubble volume sparged during unit time into the bioreactor; is the lifetime of a bubble; L is the liquid depth; f is the terminal velocity of bubbles, and , a coefficient due to agitation with a = 3k ;, where k is the rate of mass transfer between r a bubble and the medium liquid, is defined as the ratio of the diffusion constant of the gas and the length of the boundary layer surrounding the bubble; and r is the radius of a bubble, and is defined as Q/V (=vvm). To examine the behavior under various conditions and to compare the results, we approximated the plots in Figs. 3ac to straight lines to estimate K / from the slope CO2. CO2 should be an index of intrinsic CO2 stripping efficiency. According to Eq. 3, CO2 should be constant when the terminal velocity of bubbles f, the rate of mass transfer between a bubble and the medium liquid k, the radius of a bubble r, and the depth of the medium L are constant. As shown in Fig. 4, the values of CO2 in the model medium are similar to those of the culture medium in all experimental conditions, and in contrast, those in water are higher. The higher Chemical equilibrium of dCO2 in the test liquids values of CO2 observed in water than in the medium could be attributed to the difference in the chemical equilibrium of CO2. The contribution of this effect on CO2 was estimated as follows. When CO2 gas is dissolved in water, the following chemical equilibrium is reached: CO2 g X CO2 aq CO2 aq + H2 O X H2 CO3 : Further dissociation of H2CO3 into HCO and CO2 is expressed as 3 3 follows (19): H2 CO3 X HCO + H + 3 HCO X CO2 + H + 3 3 K1 = 106:35 K2 = 1010:33

where K1 and K2 are the equilibrium constants under the standard conditions at 25 C and no ionic strength. The constants K1 and K2 are known functions of temperature and salt concentrations. The relative amounts of the components of CO2(aq), H2CO3, HCO, and CO2 3 3 depend on the pH of the solution. In the measurement of kLaCO2, pH of water changed between 4.4 and 4.8. When we assume the pH of water to be 4.5, the concentration of each chemical component is estimated as follows: + 2 H  100 = 98:6k k of H2 CO4 =  2 3 H + + H + K1 + K1 K2
FIG. 5. Total stripping of carbon dioxide per sparged gas volume per unit time at t = 1 h as a function of vvm at the indicated tip speeds in bioreactors with volumes of 80 L (a), 500 L (b), and 2000 L (c).

+ H K1  100 = 1:4k k of HCO3 =  2 + + H + H K1 + K1 K2

424

MATSUNAGA ET AL. References

J. BIOSCI. BIOENG.,

K1 K2 2  100 = 2:1 106 k k of CO3 =  2 H + + H + K1 + K1 K2 where the percentage of H2CO3 is the sum of the percentages of CO2 (aq) and H2CO3. There is little HCO and CO2 in water, and dCO2 measured by the 3 3 sensor in the experiment is almost equal to H2CO3 . In the measurement of kLaCO2, the pH of the medium and model medium changed between 6.3 and 6.6. When we assume the pH of the medium to be 6.5, the concentrations of each component are calculated using the same procedure as follows: % of H2 CO4 = 41:1k 3 k of HCO3 = 58:5k k of CO3 = 8:7 103 k:
2

The mol of CO2 stripping per sparged gas volume is estimated from: f(t) ([dCO2]0 [dCO2]t)p0HV / Qt, where 1 is the percen tage of H2CO3 , p0 is the atmospheric pressure and H is the Henry constant. Fig. 5 shows that the f value at t = 1 h in the medium is on average 1.7 times larger than that of water. The larger value of f of the medium than water could be attributed to the smaller size of bubbles, as shown in Table 3, resulting in greater efficiency, because of its higher surface area / volume. This effect could also be attributed to the larger foam yield and higher foam stability in medium, as shown in Table 2. In the medium, the bubbles were not easily broken but remained on the liquid surface, which resulted in a longer lifetime of bubbles for gas exchange. Conclusion In this study, we attempted to prepare a model medium to estimate CO2 stripping in real cultures by adjusting critical parameters such as pH, coefficient of viscosity, density, surface tension, and the amount of buffer (to reach the same equilibrium conditions of CO2). The index of CO2 stripping, kLaCO2, of the model medium showed behavior similar to the real culture medium, irrespective of the size of the bioreactor in the range between 80 L and 2000 L, and under various operation conditions. It is expected that the use of the model medium will be valuable for estimating dCO2 stripping in various types of bioreactors in scale-up.

1. Han, Y., Liu, X. M., Liu, H., Li, S. C., Wu, B. C., Ye, L. L., Wang, Q. W., and Chen, Z. L.: Cultivation of recombinant Chinese hamster ovary cells grown as suspended aggregates in stirred vessels, J. Biosci. Bioeng., 102, 430435 (2006). 2. Liu, C. H. and Chen, L. H.: Promotion of recombinant macrophage colony stimulating factor production by dimethyl sulfoxide addition in Chinese hamster ovary cells, J. Biosci. Bioeng., 103, 4549 (2007). 3. Becker, E., Florin, L., Pfizenmaier, K., and Kaufmann, H.: An XBP-1 dependent bottle-neck in production of IgG subtype antibodies in chemically defined serumfree Chinese hamster ovary (CHO) fed-batch processes, J. Biotechnol., 135, 217223 (2008). 4. Ng, S. K., Wang, D. I. C., and Yap, M. G. S.: Application of destabilizing sequences on selection marker for improved recombinant protein productivity in CHO-DG44, Metab. Eng., 9, 304316 (2007). 5. Chen, P. and Harcum, S. W.: Effects of amino acid additions on ammonium stressed CHO cells, J. Biotechnol., 117, 277286 (2005). 6. Gray, D., Chen, S., Howarth, W., Inlow, D., and Maiorella, B.: CO2 in large-scale and high-density CHO cell perfusion culture, Cytotechnology, 22, 6578 (1996). 7. Kimura, R. and Miller, W. M.: Effects of elevated pCO2 and/or osmolarity on the growth and recombinant tPA production of CHO cells, Biotechnol. Bioeng., 52, 152160 (1996). 8. Kimura, R. and Miller, W. M.: Glycosylation of CHO-derived recombinant tPA produced under elevated pCO2, Biotechnol. Prog., 13, 311317 (1997). 9. Van der Pol, L. and Tramper, J.: Shear sensitivity of animal cells from a culture medium perspective, Trends Biotechnol., 16, 323328 (1998). 10. Chisti, Y.: Animal-cell damage in sparged bioreactors, Trends Biotechnol., 18, 420432 (2000). 11. Keane, J. K., Ryan, D., and Gray, P. P.: Effect of shear stress on expression of a recombinant protein by Chinese hamster ovary cells, Biotechnol. Bioeng., 81, 211220 (2003). 12. Ma, N., Chalmers, J. J., Aunins, J. G., Zhou, W., and Xie, L.: Quantitative studies of cellbubble interaction and cell damage at different pluronic F-68 and cell concentrations, Biotechnol. Prog., 20, 11831191 (2004). 13. Murakami, S., Watler, P., Ishihara, T., and Yamamoto, S.: Process modeling proposition in biopharmaceutical manufacturing, Pharm. Eng., 25, 818 (2005). 14. Heath, C. and Kiss, R.: Cell culture process development: advances in process engineering, Biotechnol. Prog., 23, 4651 (2007). 15. Humphrey, A.: Shake flask to fermentor: what have we learned? Biotechnol. Prog., 14, 37 (1998). 16. Robert, N. P., Janani, S., Brett, M., Chris, H., and Bert, T. F.: Measurement and control of dissolved carbon dioxide in mammalian cell culture process using an in situ fiber optic chemical sensor, Biotechnol. Prog., 16, 769774 (2000). 17. Clift, R., Grace, J. R., and Weber, M. E.: Bubbles, Drops, and Particles, Dover Publications, 2005. 18. Matsunaga, N., Kano, K., Maki, Y., and Dobashi, T.: Culture Scale-up Studies as Seen from Oxygen Supply and Dissolved Carbon Dioxide Stripping, J. Biosci. Bioeng., 107, (4) 412418 (2009). 19. Morel, F. M. M. and Hering, J. G.: Principles and Applications of Aquatic Chemistry, 2nd ed. John Wiley and Sons Inc., 1993.