You are on page 1of 7

Process Biochemistry 34 (1999) 549 – 555

www.elsevier.com/locate/procbio

Prediction of kLa in yeast broths


Francisco J. Montes, Jacinto Catalán, Miguel A. Galán *
Department of Chemical Engineering, Uni6ersidad de Salamanca, Plaza de los Caidos 1 -5, 37008, Salamanca, Spain

Received 30 July 1998; received in revised form 28 September 1998; accepted 7 October 1998

Abstract

Oxygen transfer rate, can play an important role in the scale-up and economy of many microbial processes and values of the
volumetric oxygen transfer coefficient (kLa) for specific fermentation culture media need to be evaluated and correlated to obtain
the appropriate design tools. The purpose of this work was to determinate the values of kLa in yeast broths (Trigonopsis 6ariabilis)
in a mechanically-stirred, sparger-aerated and baffled reactor, the most common fermenter type, over a wide range of superficial
air velocities impeller rotational speeds and geometric parameters. Three different mixing vessels (2, 5 and 15 litres) were used in
order to consider the effect of the fermenter scale-up on kLa. Once kLa data are obtained, several empirical and theoretical
correlations were used to fit the experimental data. A new correlation is proposed
kLa=3.2·10 − 3

P 0.35
(Usg)0.41

V
based on the power input per unit volume of liquid (P/V) and the superficial gas velocity (Usg). The correlation improves the
prediction of kLa values in culture media with respect to other generic correlations, mainly because early correlations were
developed for strong coalescent and non-coalescent fluids, whereas the medium used in this work and in most the yeast broths
behaves as a typical Newtonian, slightly non-coalescence fluid, due to the moderate presence of mineral salts. © 1999 Elsevier
Science Ltd. All rights reserved.

Keywords: Oxygen transfer; Yeast broths; T. 6ariabilis; Empirical correlation; kLa

Nomenclature Fr Froude number


J width of baffles (m)
a1,a2 empirical coefficients used through K consistency index in the power-law
Eqs. (3) – (7) model of viscosity (Pa * sn)
B constant used in Eq. (9) and num- L length of the blades (m)
ber of blades on impeller M mass of fluid (kg)
C empirical constant used through n flow index in the power-law model
Eqs. (3) – (7) of the viscosity
C1 constant used in Eq. (8) ni number of impellers
D agitator diameter (m) N stirring speed (s − 1)
Di width of turbine blades (m) Np power number in absence of gas
Ds diffusivity of the gas in the liquid Npg power number in presence of gas
(m2 s − 1) Pg aerated power input (W)
Po non-aerated power input (W)
(P/M) power input per unit mass of liquid
* Corresponding author. Tel: + 34-23-294479; fax: + 34-23-
(Eq. (10)) (W kg − 3)
294574. (P/V) power input per unit volume of liq-
E-mail address: magalan@gugu.usal.es (M.A. Galán) uid (W m − 3)

0032-9592/99/$ - see front matter © 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 3 2 - 9 5 9 2 ( 9 8 ) 0 0 1 2 5 - 3
550 F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555

Qg volumetric gas flow rate (m − 3 s − 1) As part of the study of the optimal aerobic growth
Qvvm volumetric gas flow rate related to conditions for the yeast Trigonopsis 6ariabilis, a exten-
the volume of the reactor (s − 1) sive report was made on the rate of oxygen transfer to
R number of baffles the liquid phase. T. 6ariabilis is a potent producer of
Re Reynolds number the enzyme D-amino acid oxidase (DAAO) (D-amino
Sc Schmidt number acid: O2-- oxidoreductase deaminating, (EC 1.4.3.3),
T reactor diameter (m) thus providing a microbial source for large-scale prepa-
Usg superficial air velocity (m s − 1) rations for which the mammalian enzyme from pig
Ut terminal velocity of the bubbles in kidney cannot be fully exploited [2]. This enzyme has
free rise (m s − 1) important applications in the resolution of racemic
VL liquid volume (m3) mixtures of amino acids, production of a-keto acids
W width of the impeller blades (m) and manufacture of Cephalosporin C derivatives [3–
We Weber number 5].
Greek letters The purpose of this work was to determinate values
o gas holdup of kLa in a mechanically-stirred, sparger-aerated and
o power input per mass unit of the baffled reactor, the most common fermenter type, over
fluid (Eq. (8)) (W kg − 1) a wide range of superficial air velocities, impeller rota-
ma apparent viscosity defined by Eq. (9) tional speeds and geometric parameters for the opti-
(Pa s) mum culture medium used in the growth of Trigonopsis
mw viscosity of water (Pa s) 6ariabilis. This culture medium was chosen after com-
rL liquid density (kg m − 3) paring several different media and selecting that which
rg gas density (kg m − 3) optimized the cost of D-amino acid oxidase produced
s gas– liquid surface tension (N m − 1) by T. 6ariabilis [6].
The experimental kLa data were fitted to different
correlations available in the literature [7–9] comparing
the results with those obtained for several researchers
1. Introduction [1,10] under similar experimental conditions.

Oxygen transfer can play an important role in the


scale-up and economy of most aerobic fermentation
processes, since oxygen supply is often the limiting 2. Experimental
factor determining bioreactor productivity. Therefore,
values of volumetric oxygen transfer coefficient (kLa)
for specific culture media need to be obtained and 2.1. Equipment and experimental conditions
correlated by using dimensionless or empirical equa-
tions in order to obtain the appropriate design tools. Three different mechanically-stirred and sparger-aer-
Although the literature describing correlations for ated mixing vessels (2, 5, 15 l), provided with four
kLa in Newtonian and non-Newtonian fluids is exten- baffled to avoid vortices, were used in order to consider
sive [1] there are still considerable problems in predict- the eventual effect of the fermenter scale-up on kLa.
ing kLa values in microbial broths because the The main geometric characteristics of the mixing vessels
coefficients of most of the correlations are considerably are shown in Table 1. kLa data were obtained for a
affected by the geometry of the system, the range of the range of superficial air velocities between 2.5×10 − 3
operating variables and the experimental kLa determi- and 8× 10 − 3 m s − 1 and stirring speeds between 100
nation method. and 700 min − 1.

Table 1
Dimensions of the reactors and its complementsa

Tank volume (1) T Hf D N. Baffles J N. Turbines Blades/turbine L W

2 12.8 17.5 5.2 4 1 2 6 1.4 1


5 16.2 26 6.4 4 1.2 2 6 1.9 1.3
15 21 44 8.4 4 2 3 6 2.4 1.7

a
T is the diameter of the reactor, Hf (cm) is the liquid height in the reactor, D (cm) is the diameter of the agitator, J (cm) is the width of the
baffles, L (cm) is the length of the turbine blades and W (cm) is the width of the blades.
F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555 551

2.2. Culture media and measuring methods for kLa 3. Theoretical considerations

The culture medium under study contained 1% glu- 3.1. kLa correlations
cose (w/w) and DIFCO Yeast Nitrogen Base without
amino acids (6.7 g l − 1) and it was buffered to pH 6. Two different types of empirical correlations have
Different methods for the determination of kLa have been proposed [7,9,10] for the volumetric mass transfer
been described and discussed [10,11]. The sulphite coefficient. One of them is based on dimensional anal-
method has been widely used for the calculation of ysis (Eq. (1))
kLa in stirred vessels [12,13]. This method has come
Sh= f1(Re,Sc,…) (1)
under severe criticism by several researchers [10], be-
cause the reaction rate constant can vary in a un- and the other is based on power input per unit of
known way. Therefore, it can safely be said that volume (Eq. (2))
application of this method should be strongly discour-
kLa= f2(P/V,Usg,…) (2)
aged.
In this work, oxygen absorption and desorption The advantage of the latter approach, which has
methods [14] and a dynamic biological method [15,16] been widely used, lies in its great applicability. Five
were used and compared in order to determine kLa. different forms of Eq. (2) obtained from the literature


The dynamic biological method involves physical oxy- [12,17] were tested.
gen absorption combined with oxygen consumption by a1
P
a cell culture. Temperature (31°C), stirring speed (500 kLa= C· ·U asg2 (3)


min − 1) and air flow rate (1.065 l min − 1) were held V
a1
constant throughout these experiments. P
kLa= C· (4)

 
The results (not shown) indicate no significant dif- V
ference among the values of kLa obtained from the (kLa)V PQg a1
three methods. This result can be explained consider- =C (5)
ni Qg ni
ing that there were no appreciable changes in the main
rheological parameters of this yeast broth with respect (kLa)V
= C·P a1·U asg2 (6)

 
to the culture medium. Therefore, in this work, the T
oxygen desorption method was selected as the stan- a1 a2
P Qg
dard method for measuring kLa because its practical kLa= C· · (7)
simplicity in comparison with the dynamic biological V V
method. The validity for this system of the semi theoretical
Eq. (8) developed by Kawase and Moo-Young [18],
2.3. Electrode response time
    
was tested
o ((9 + 4n)/(10(1 + n)))r 3/5 Usg 1/2
ma − 1/4
The main disadvantage of the calculation of kLa kLa= C1
Ds
(k/r)(1/(2(1 + n)))s 3/5 Ut mw
using the oxygen desorption method is the slow re-
(8)
sponse time of the electrode to changes of oxygen
concentration in the liquid. Response time is defined where ma is the apparent viscosity defined by Calder-
as the time required by the electrode to measure 63% bank and Moo-Young [19], and o is the power input

   n
of the global value of the change in the concentration per unit mass of fluid,
of oxygen [10] and it is related to the diffusion of n/1 − n n − 1
4n
oxygen through the membrane of the electrode. So ma = k BN (9)
that the measures will be sufficiently reliable (error 3n+ 1
B 6%), the response time of the electrode must be For Newtonian fluids, the terminal velocity of bub-
smaller or equal to l/kLa [8] Therefore, an evaluation bles in free rise (Ut) is 0.265 (m s − 1). The constant C1
of the response time of the oxygen electrode used should be calculated using the data for the specific
throughout the experiments. (Ingold Type 82 12/320) surface area. The constant was determined by
was made by measuring the variation of oxygen con- Nishikawa et al. [20] (C1 = 0.675).
centration after the electrode was moved from a satu- The following values obtained from tables of physi-
rated aqueous solution of nitrogen to a saturated cal properties were used for some of the constant
oxygen solution. appearing in Eq. (8):
The results indicate that 63% of oxygen saturation Ds: Oxygen diffusivity in water (31°C): 2.25× 10 − 9
was reached at 9.6 s and 100% was reached at around (m2 s − 1).
20 s. Then, only values of kLa less than 0.1 s − 1 should s: Surface tension (air/water, 31°C): 71.2× 10 − 3 (N
be considered reliable when using this electrode. m − 1).
552 F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555

Table 2
Correlation coefficients, mean errors (from Eq. (17)) and regression coefficients (from Eq. (18)) as result of the least squares fitting of the
experimental kLa data to several correlationsa

Equation type Correlation coefficients Mean error Regression coefficients

C a1 a2 (%) a b R

3 3.2×10−2 0.35 0.42 12.9 0.0012 0.98 0.95


4 3.5×10−2 0.34 — 18.9 0.0056 0.83 0.91
5 12.8 0.31 — 32.0 0.0010 1.01 0.75
6 7.3×10−3 0.4 0.46 28.4 0.0003 1.07 0.83
7 8.2×10−3 0.34 0.22 16.8 0.0040 0.84 0.92
10 7.2×10−3 13/20 1/2 46.0 −0.0200 1.92 0.92

a
Eqs. (4) and (5) do not include a correlation factor for a2, so those values are omitted. The correlation coefficients of Eq. (10) are theoretically
obtained from Eq. (8).

Density and viscosity of the medium were measured NPn = nNP 1 (13)
at 31°C and the results were:
mw: Culture medium viscosity: 0.7808× 10 − 3 (Pa s). In this work, systems with one or two turbines were
r: Culture medium density: 1005 (kg m − 3). used. Power input was calculated on this basis.
Working with a Newtonian fluid, such as a glucose Power dissipated during the agitation of a reactor is
and salts aqueous solution, K = mw (viscosity of the considerably affected by aeration. The presence of bub-
medium), n= 1 and ma =K. Taking, into account all bles, particularly around the turbines, reduces consider-
this values, Eq. (8) becomes: ably the density of the gas–liquid mixture with respect

kLa = (7.210 − 3)
 
P 13/20
U 1/2
sg (10)
to the density of the liquid, with the consequent reduc-
tion of the power consumed. Aeration can reduce the
M power dissipated to one third of the value obtained
without aeration. Most correlations show their results
3.2. Power consumption correlations as a function of the quotient aerated power/non-aerated
power, whose values can range between 0 and 1.
Since all the equations outlined in this work for the Hughmark [24], after a statistic analysis of the avail-
calculation of the kLa use as a correlation term the able data suggests the Eq. (14):
power input during the agitation of the gas – liquid
mixture, it is necessary to evaluate this parameter using Pg
= 0.1

N 2D 4   
− 1/5
Q − 1/4
(14)
relationships available in the literature. Po gDi V 2/3 NV
According to Rushton et al. [21,22], the power input
where the dimensionless correlation factor (Q/NV) is
for the agitation of a non-aerated mixture, Po (W) is
introduced instead of the factor (Q/ND − 3), used previ-
characterized by the dimensionless variable power num-
ously in several correlations [25,26]. On the other hand,
ber (NP 1), following the Eq. (11):
the Weber number of the agitator:
Po = NP 1rLN 3D 5 (11)
We=
 rLN 2D 3  (15)
The power number depends on other dimensionless s
groups such as Reynolds number, Froude number as
is substituted by the Froude number:
 
well as the number of agitator turbines. For a baffled
reactor the vortex is minimized and the influence of N 2D 4
Froude number on the power number is negligible. Fr= (16)
gDi V 2/3
Under these conditions and using only one turbine with
six flat blades (Rushton type), Richards [23] proposed since it reduces the mean absolute deviation to half
the following correlation (Eq. (12)) for the Power num- when compared with the data of Luong and Volesky
ber (NP 1) (one turbine):
        
[26] Bimbinet et al. [27] Michel and Miller [28] and
W L 1.5
J 0.3
B 0.56
R 0.4 Pharamond et al. [29]. The correlation covers a wide
NP 1 =336.5 (12) interval of the different variables and it is dimension-
D D D 6 4
less. Its prediction capacity declines for extreme aera-
While working with a system of n convenient spaced tion values. This equation is recommended in the
turbines (1.5–2 agitator diameters), the corresponding exhaustive study ellaborated by Joshi et al. [9], and it is
power number (NPn ) is n times greater than the value used in this work for the calculation of the reduction of
obtained for one turbine: power input due to aeration.
F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555 553

3.3. Statistical identification of the best model for kLa (7). Table 2 collects the fitting parameters for these
prediction equations.
Table 2 also collects values of the mean errors (%)
In order to choose the most accurate correlation, the calculated using Eq. (17) and the values of a, b and R
mean error (%) of each correlation, defined according obtained using Eq. (18) for all the empirical
to Eq. (17), was calculated: correlations.
From the data shown in Table 2 the following con-
% (( kLaexp −kLa(teor) )/kLaexp) clusions may be drawn:
n
Mean Error (%) = × 100 1. Equation of type (Eq. (3)) provides the lower mean
n error for the set of data (12.9%).
(17)
2. Equation of type (Eq. (3)) provides the closest val-
where n is the number of data. The best correlation ues of a (0.0012), b (0.98) and R (0.95) to the
must minimize the value of this error. optimums.
Another statistical method to identify the best corre- The final equation, including the optimum coeffi-


lation is based on a comparison of the parameters (a cients is:
and b) of the linear fit of Eq. (18), as well as its P 0.35

regression coefficient (R). The best correlation will kLa= 3.2× 10 − 3 (Usg)0.41 (19)
V
provide values of ‘a’ close to zero and of ‘b’ and R
close to 1. Fig. 1 compares the theoretical kLa values obtained
using Eq. (19), and the experimental kLa values. The
kLa(teor) = a+b·(kLa(exp)) (18) upper and lower 20% error limits are also shown.
Around 75% of the data are enclosed in this range.
Figs. 2 and 3 show the experimental kLa data using two
4. Results and discussion different superficial air velocities compared to the em-
pirical correlation obtained in this work (Eq. (19)) and
Experimental data of kLa, P/V and Usg were fitted by to several correlations available in the literature
the least squares method to the correlations Eqs. (3)– [1,7,10,19] for coalescent and non-coalescent fluids. Eq.

Fig. 1. Comparison of the theoretical kLa values obtained from Eq. (19) and the experimental kLa values. The upper and lower 20% error limits
are also shown.
554 F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555

Fig. 2. Experimental kLa data compared to the empirical correlation obtained in this work (Eq. (19)), and compared to several correlations for
coalescent and non-coalescent fluids (Usg = 0.00246 m s − 1).

(19) agrees reasonably well with the same type of with respect to data given by those authors, within the
correlation suggested by Calderbank and Moo-Young ranges of superficial air velocities and power input per
[19], and Van’t Reit [10], for Newtonian coalescent volume unit used. On the other hand, the kLa data from
fluids (C= 2.6×10 − 2, a1 =0.4, a2 =0.5), yielding theo- the correlation and the data from Van’t Reit’s correla-
retical kLa values slightly higher (mean increase of 25%) tion tend to get closer when increasing both correlation

Fig. 3. Experimental kLa data compared to the empirical correlation obtained in this work (Eq. (19)), and compared to several correlations for
coalescent and non-coalescent fluids (Usg = 0.008m s − 1).
F.J. Montes et al. / Process Biochemistry 34 (1999) 549–555 555

factors (Figs. 2 and 3) because the dependence of the Congress Biotechnol 1990:319 – 323.
volumetric mass transfer coefficient due to changes in [5] Dey ES, Flygare S, Mosbach K. Stabilization of D-amino acid
oxidase from yeast Trigonopsis 6ariabilis used for production of
the correlating factors is weaker in the proposed corre- glutaryl-7-aminocephalosporanic acid from cephalosporin C.
lation compared to Van’t Reit’s correlation. Appl Biochem Biotech 1991;27:239 – 50.
It should be noted that, when the culture medium is [6] Montes FJ. PhD. Thesis. University of Salamanca, Spain, 1996.
used, the presence of mineral salts increases the ionic [7] Zlokarnik M. Sorption characteristics for gas – liquid contacting
strength of the solution with respect to a pure water in mixing vessels. Adv Biochem Eng 1978;8:133 – 51.
[8] Van’t Riet K. Thesis, University of Technology, Delft, 1975.
solution, leading to a reduction of the mean size of the
[9] Joshi JB, Pandit AB, Sharma MM. Mechanically agitated gas–
free rising bubble along the reactor and increasing the liquid reactors. Chem Eng Sci 1982;37(6):813– 44.
specific surface area due to coalescent depression. [10] Van’t Riet K. Review of measuring methods and results in
Considering the kLa data obtained in this work, it is nonviscous gas – liquid mass transfer in stirred vessels. Ind Eng
fair to say that the culture medium under study is a Chem Proc Des Dev 1979;18(3):357– 64.
[11] Rainer BW. Determination of the volumetric oxygen transfer
typical Newtonian fluid, with a tiny non-coalescent
coefficient in bioreactors. Chem Biochem Eng Q 1990;4(4):185–
behaviour. This conclusion agrees with other studies 96.
using yeast broths with biomass concentrations lower [12] Moresi M, Patete M. Prediction of kLa in conventional stirred
than 15% (dry W/W) [30,31]. vessels. J Chem Tech Biotechnol 1988;42:197 – 210.
[13] Cho MH, Wang SS. Practical method for estimating oxygen
kinetic and metabolic parameters. Biotech Prog 1990;6:164–7.
[14] Wise WS. Measurement of aeration of culture media. J Gen
5. Conclusions Microbiol 1951;5:167 – 77.
[15] Taguchi H, Humphrey AE. Oxygen transfer in fermentation
New coefficients have been found for a widely used processes. J Ferment Technol 1966;44:881.
type of correlation for the prediction of kLa values in [16] Bandyopadhyay B, Humphrey AE, Taguchi H. Dynamic mea-
yeast culture media. Experimental values of kLa were surement of the volumetric oxygen transfer coefficient in fermen-
tation systems. Biotechnol Bioeng 1967;9:533.
obtained in a mechanically-stirred, sparger-aerated and
[17] Bailey JE, Ollis DF. Biochemical engineering fundamentals, 2nd
baffled reactor over a wide range of superficial air edition. NY: McGraw-Hill, 1986.
velocities, impeller rotational speeds and geometric [18] Kawase Y, Moo-Young M. Volumetric mass transfer coefficients
parameters. The correlation outperforms other generic in aereated stirred tank reactors with Newtonian and non-New-
correlations, mainly because those correlations were tonian media. Chem Eng Res Des 1988;66:284 – 8.
developed for strong coalescent and non-coalescent [19] Calderbank PH, Moo-Young M. Mass transfer efficiency of
distn and gas absorption plate columns. I1. Liquid – phase mass
fluids, whereas the media used in the growth of T. transfer coefficient in sieve-plate columns. Trans Ins Chem Eng
6ariabilis and in most of the yeast behaves as a typical 1961;39:337.
newtonian fluid, with a tick of non-coalescence, due to [20] Nishikawa M, Nakanura M, Hashimoto K. Gas absorption in
the moderate presence of mineral salts. aereated mixing vessels. J Chem Eng Jpn 1981;14:227.
[21] Rushton JH, Costich EW, Everett HJ. Power characteristics of
mixing impellers: Part l. Chem Eng Prog 1950;46:395 –404.
[22] Rushton JH, Costich EW, Everett HJ. Power characteristics of
Acknowledgements mixing impellers: Part 1I. Chem Eng Prog 1950;46:467–76.
[23] Richards IW. Power input to fermenters and similar vessels. Brit
This work was supported by CICYT (Comisión In- Chem Eng 1963;8:158 – 63.
terministerial de Ciencia y Tecnologı́a), Research Pro- [24] Hughmark GA. Power requirements and interfacial area in
gas – liquid turbine agitated systems. Ind Eng Chem Process Des
ject B1092-0304.
Dev 1980;19:638 – 41.
[25] Hassan ITM, Robinson CW. Stirred tank mechanical power
requirements and gas holdup in aereated aqueous phases.
References AICHE J 1977;23:48.
[26] Luong HT, Volesky B. Mechanical power requirements of gas–
[1] Moo Young M, Blanch HW. Design of biochemical reactors: liquid agitated systems. AICHE J 1979;25(5):893.
mass transfer criteria, for simple and complex systems. Adv [27] Bimbinet JJ. M.S. Thesis, Purdue Univ., Lafayette, Indiana,
Biochem Eng 1981;19:1–69. 1959.
[2] Bode R, Birnbaum D. D-Amino acid oxidase, aromatic L-amino [28] Michel BJ, Miller SA. Power requirements of gas – liquid agitated
aminotranferase, and aromatic lactate dehydrogenase from sev- systems. AICHE J 1962;8(2):262 – 6.
eral yeast species: comparison of enzyme activities and enzyme [29] Pharamond JC, Roustan M, Roques H. Détermination de la
specificities. Acta Biotechnol 1987;7(3):221–5. puissance consommee dans une cuve aeree et agitee. Chem Eng
[3] Brodelius P, Nilsson K, Mosbach K. Production of a-Keto Sci 1975;30:907.
acids. Part 1. immobilised cells of Trigonopsis 6ariabilis Contain- [30] Labuza TP, Santos DB, Roop RN. Engineering factors in single-
ing D-amino acid oxidase. Appl Biochem Biotech 1981;6:293 – cell protein production. 1. Fluid properties and concentration of
308. yeast by evaporation. Biotechnol Bioeng 1970;12:123.
[4] Galán MA, Enzymatic separation of a racemic mixture of amino [31] Charles M. Technical aspects of the rheological properties of
acids. In: From genes to bioproducts. Proc. of the 3rd. Spanish microbial cultures. Adv Biochem Eng 1970;8:1.
.

You might also like