You are on page 1of 6

Modeling of Ethanol Fermentation at High

Yeast Concentrations
A. 6. Jarzebski and J. J. Malinowski
Polish Academy of Sciences, Institute of Chemical Engineering, 44- 100
Gliwice, ul. Bahycka 5, Poland

G. Goma
Departement de Genie Biochimique et Alimentaire, lnstitut National des
Sciences Appliquees, UA-CNRS-No544, Toulouse Cedex, France
Accepted for publication February 1, 1989

Three models of ethanol fermentation at high yeast concentrations were developed and verified by comparing
them w i t h experimental data. The conventional approach, neglecting loss of cell viability, proved to be the
least accurate. The models, allowing both for the
growth of viable cells and accumulation of the inactive
yeast, satisfactorily portray the fermentation process at
very high cell concentration and may be used to facilitate process optimization. Analysis of results shows that
both intrinsic and nonintrinsic models provide similar
results for ethanol fermentation.

INTRODUCTION
The growing prospects of ethanol as a fuel and future
chemical feedstock have prompted considerable efforts to
increase effectiveness of the fermentation process. Continuous fermentation markedly reduces costs but low yeast
concentration in the fermentor and the low dilution rates
usually applied seriously limit process productivity. Cysewski and Wilke'.' first proposed yeast cell recycling, and
this increased productivity four times that of a conventional process. Initially, a settling tank or a centrifuge was
used to recycle cells. However, the most recent reports
advocate continuous concentration by means of crossflow filtration both in ultra- and microfiltration
The latter technique gives cell concentrations of up to
200 kg m-3 dry weight3 and even more,537while the deduced maximum cell density of the yeast cultures is ca.
300 kg m-3.739310
At very high yeast cell concentrations, conditions for
growth and metabolism are less favorable due to hindered
access to nutrients, space limitations, and cell interaction.
These facts together with prolonged residence time of cells
lead to kinetics and compositions different from those usually encountered. An adequate mathematical model should
certainly take these factors into account.
In spite of widespread interest in experimental investigation of the process, mathematical modeling of fermentation
at high cell concentrations has attracted relatively little atBiotechnology and Bioengineering, Vol. 34, Pp. 1225-1230 (1989)
0 1989 John Wiley & Sons, Inc.

tention. Lee, Polard, and Coulman" developed a model


for glucose-to-ethanol fermentation at high cell concentrations to evaluate the feasibility of improving fermenter
productivity by using a multiple-stage reactor. This model
uses nonintrinsic concentrations, and the growth rate of
cells is expressed by Monod type kinetics extended to allow both for product and cell inhibition. The rate of
ethanol formation is assumed to be proportional to the rate
of cell growth while the loss of cell viability, which may
be an important factor at high cell concentration^,^'^^^ is totally neglected. A comparison of theoretical predictions
with several experimental points does not provide sufficient evidence to draw any overall conclusions as to the
accuracy of the developed model. Hence, for computer
simulations relevant parameters were assumed from the
range of feasible values rather than those assigned from
experimental data.
More recently, Monb~uquette'~
emphasised the need to
use the intrinsic concentration approach to describe high
cell density bioreactor operation and indicated that the
nonintrinsic approach is suitable when the attainable
biomass volume fraction is less than ca. 0.1. Comparing
corresponding predictions from intrinsic and nonintrinsic
models, Monbouquette found that nonintrinsic models can
lead to gross errors in calculated substrate and product
concentrations, substrate conversions, and volumetric productivity. Standard Monod-type kinetics were extended to
take into account product inhibition but neglected the cell
inhibition effect and possible death phase, similarly in another report."
To summarize, the available literature does not provide
a verified model for fermentation at high cell concentrations. The aim of this communication is to develop and
investigate several approaches for this process by comparing theoretical predictions with recently published experimental data on glucose-to-ethanol fermentation at high
yeast concentration^.^

CCC 0006-3592/89/0901225-06$04.00

FERMENTATION MODELS

(5)

A Conventional Approach
Conventional models for fermentation precesses neglect
loss of cell viability and the appearance of inactive cells.
Since both models developed1 re~ently""~
belong to this category the questions of whether and to what extent such an
approach can successfully portray the process of fermentation at high cell concentrations is of practical interest.
The central point in each fermentation model is the kinetics. Analysis of the concentration profiles' indicated
that chief inhibiting factors are product and cell concentrations. Adopting Monod-type kinetics and extending them
in the simplest possible form to allow for inhibition effect,
the rate of growth is expressed as

i.e., growth ceases if either product concentration P or total cell concentrations X , reach certain maximum values P,
or X,, respectively. Exponents A , and A, take into account
nonlinearity of inhibition effects.
The rate of substrate consumption is1'

and rate of ethanol formation is assumed to have the form


of the Luedeking-Piret equation
=

A Modified Approach
Several authorsL2213
have already noted the loss of cell
viability at high cell concentrations. Analysis of concentration profiles reported recently by Lafforgue and co-worke r ~ indicate
*
that inactive cells can constitute as much as
one-quarter of the total dry biomass and hence their appearance cannot be neglected. More thorough investigation
of this question reveals that the rate of death depends on
the rate of growth of the viable phase and also on its concentration. Taking all this into account, it is assumed that
the total biomass comprises a viable (active) phase X , and
an inactive (dead) phase X,. Consequently, the growth rate
of the viable phase is

(2)

rs = -rx/yx,s - msx,

rp

The system of eqs. (1)-(6) supplemented with a set of values of operational and adjustable parameters constitute the
model of the fermentation at high yeast cell concentrations.

r*/Yxp+ mpxt

(3)

The typical system of fermentation with membrane cell recycle module is shown schematically in Figure 1. The filtering device is assumed perfect and the feed, bleed, and
filtrate streams are denoted. by F, B , and L, respectively.
In the conventional nonintrinsic approach, the final set
of mass balance equations describing operation of the fermentation system is
(4)

where X , = X , + X,.
Similarly, the rate of death may be expressed as
rd

(k,w

+ k2)X"

(8)

where p is the specific growth rate of the viable phase and


the rate of substrate consumption is of the form
rs

(9)

-r*/yx,s - msxv

Since specific ethanol productivity decreases monotonically with cell concentration, the rate of ethanol production
can be described, as done by Mota et al.,I6 by the exponential function
rp = a x , exp( -bX,) .

(10)

In the intrinsic approach, the final set of equations is

-s ,
L

P,

x=o
X

8
*
Figure 1. Fermentation sysiem with membrane filtering module.

1226

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 34, NOVEMBER 1989

S - LS

+ r,V

(13)

while in the traditional nonintrinsic approach, substrate


and product concentrations are given by eqs. (5) and (6)
and equations for active (viable) and inactive phases are as
above. The systems of eqs. (7)-(14) or eqs. (5)-(12) constitute two versions of the modified model of the fermentation at high yeast cell concentrations.

modification of the expression for ethanol production rate


markedly improved ethanol concentration profile but had
little, and in fact adverse, effect on the predictions of yeast
concentration. On the whole, the conventional modelling
approach appears inadequate to portray accurately
the fermentation process at high yeast concentrations.
The same experimental observations of Lafforgue and
co-workers5 are compared in Figure 3 with predictions of
RESULTS AND DISCUSSION
the modified approach. Model parameters estimated numerically to provide the best fit of the curves took the valThe experimental observations of Lafforgue, Malinowski,
ues: p, = 0.24; A , = 0.85; A , = 1.1; K , = 0.5; a =
and Goma' obtained in a fermentation system (fermenter
0 . 5 5 ; b = 0 . 0 0 6 1 ; Y,,, = 0 . 1 2 ; k , = 0 . 2 0 8 ; k , =
and cell separator) with total working volume 2.4 X
2.08 X
and m, = 0.27. Similarly as before P,,, was
m3, S, = 150 kg m-3, and dilution rates D = 0.5 and
assumed to be equal to 90 kg m-3 and X , = 320 kg m-3.
B = 0 are compared in Figure 2 with predictions of the
It is noteworthy that estimated values are in fair agreement
conventional model (solid lines). Model parameters estiwith those reported by previous
and exammated numerically took the values: p, = 0.24 h-I;
ined in the most recent report of Daugulis and Swaine.I9
K , = 0.5; A , = 0.85; A, = 1.1; Y,,, = 0.13; m, = 0.22;
From Figure 3 it may be seen that agreement between preY,,, = 0.4; and m, = 0.13. From results reported previdictions and experiment is good for all profiles and
~ u s l y , P,,,
' ~ was assumed to be equal to 90 kg m-3; from
data collated by Lafforgue, Malinowski, and G ~ m a it, ~ throughout the whole time range. Moreover, the two modified models provide similar results, in particular for yeast
was taken that X , = 320 kg m-3.
concentration. The largest discrepancy was found in subIt is clear from Figure 2 that the conventional model
strate concentration but this was never larger than about
fails to predict ethanol concentration and predicts too rapid
twenty percent. As this agreement remains fair also for
a rise of yeast concentration in the exponential growth
B > 0 both models may be used for simulation purposes
phase. To overcome the first drawback, eq. (3) was reand further analysis of the process.
placed by the expression
From eqs. (6) and (10) and also from eqs. (10) and (14)
r, = d,
exp(-bX,)
(15)
(after minor rearrangements), it may be seen that ethanol
productivity, equal to the product of dilution rate and ethaand predictions obtained for a = 0.43, b = 0.0048, Y , , =
nol abiotic phase concentration, under steady (or quasi0.15, and m, = 0.21, with other parameters as before, are
steady) operation may be expressed as
shown in Figure 2 (dashed lines). It can be seen that the

t (hl
Figure 2.

Comparison of experiment with theoretical predictions of the conventional approach

COMMUNICATIONS TO THE EDITOR

1227

50

20

90

60

s
20

60

100

140

Figure 3. Comparison of experiment with predictions of the modified models: (---)


and (-) nonintrinsic model.

DP = d,
exp(-bX,)

( 16)

It is evident that productivity initially rises with increase in


X,, attains a maximum value, and subsequently falls exponentially but specific ethanol productivity falls monotonically with increase in X,; this agrees with the most recent
observations of Richter and Meyer.20
The viable cell concentration corresponding to maximum productivity is
X"

l/b

maxDP

and for the case considered this value was ca. 164 kg m-3.
The corresponding maximum ethanol productivity was
found to be equal to 33.2 kg m-3 h-I and ethanol exit concentration of ca. 66.4 kg m-3. It may be noted that these
computed and experimental values correspond well at
maximum product concentration (see Fig. 3). Certainly,
optimum conditions for maximum ethanol production can
be achieved for incomplete cell recycle, i.e. when bleed
stream B # 0. A bleed is also necessary to maintain a cell
concentration suitable for undisturbed operation since build
up of cells continues until the fermenter becomes inoperable due to high viscosity.2' The most suitable value of ratio B / F , from the ethanol productivity standpoint, together
with the actual values of ;U, and S can easily be computed
by resolving the system of eqs. (11)-( 13) or eqs. (5), (1l ) ,
and (12), respectively, taking derivatives equal to zero. In
the case analyzed maximum productivity occurred for B / F
of ca. 0.03. Figure 4 portrays predictions of concentrations
X,, P, and S vs. time for B / F ratios 0.01, 0.03, and 0.05
and Figure 5 depicts dependence of ethanol productivity

1228

180

z-

30

intrinsic model

and substrate conversion on the bleeding, obtained from


the nonintrinsic approach. It is noteworthy that the conventional productivity expression DP and the more rigorous
form given by Monbouquette [eq. (15)]14 provide similar
results. The range B / F 2 0.03 appears to offer the most
favorable conditions for practical applications.

CONCLUSIONS
The models, allowing both for the growth of the viable
cells and the appearance and accumulation of the inactive
yeast, can be used to portray fermentation at high cell concentration. The growth of yeast can be adequately modeled
by conventional Monod kinetics supplemented with terms
taking into account cell and product inhibitions which
prove not to depart markedly from a linear relationship.
Recycling of cells inevitably leads to the appearance of an
inactive (dead) cell phase, which increases in proportion to
the rate of growth of viable cells and cell concentration.
Cell inhibiting effects depend on the total concentration of
viable plus dead phases. With an increase in viable cell
concentration, productivity of ethanol initially rises, attains
a maximum value, and subsequently falls while specific
productivity decreases monotonically. For ethanol fermentation, both intrinsic and nonintrinsic approaches give
similar predictions and hence are equally suitable for simulation purposes.

NOMENCLATURE
A , , A Z power indices, eqs. (1) and (7)
a
parameter in eqs. (10) and (15) (kg kg-' h-')

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 34, NOVEMBER 1989

32 0

- B/F=O.Ol

280
24 0

_ _ _ - _ _ _- - - - - - - -

,
zoo
I

- 160

_ , _ , _ _ _ . _ .-.-.-.-.-.-.-

I-

120
80
40

-. -.

-. -,-.

- - - - _ _ -_- - - - _ _ _ _-_- I

- I - . - .

20

60

100
t (hl

140

180

Figure 4. Effect of bleedstream ratio on total biomass, product and substrate concentrations

parameter in eqs. (10) and (15) (kg-' m3)


bleedstream flow rate (m' h-l)
dilution rate (h-')
feedstream flow rate (m3 h-')
substrate saturation constant (kg m-3)
parameter in eq. (8)
parameter in eq. .(8) (h-I)
filtrate stream flow rate (m3 h-I)
constant in eq. (3) (kg kg-' h-')
maintenance coefficient (kg kg-' h-I)

35

product concentration (kg m-')


rate of ethanol production (kg m-3 h-')
rate of substrate consumption (kg K3h-')
rate of yeast cell production (kg m-l h-I)
substrate concentration (kg m-3)
yield of cells based on substrate consumed (kg kg-')
yield of cells based on product formed (kg kg-l)
yeast cell concentration (kg IT-')
volume (m')
specific growth rate (h-l)

31

1.0

0.9

.Ic

I
0.8 -

T' 27
E

Ol

cn"

23

0.7

j
:
19

0.6

15

0.04

0.12

0.0 8

B/F

0.16

0.5
0.2 0

(-I

Figure 5. Effect of bleeding on ethanol productivity and substrate conversion.

COMMUNICATIONS TO THE EDITOR

1229

Subscripts
d
inactive (dead) phase
m
maximum value
v
viable (active) phase
t
total concentration

References
1 . G . R. Cysewski and C. R. Wilke, Biotechnol. Bioeng., 18, 1297
(1976).
2. G. R. Cysewski and C. R. Wilke, Biotechnol. Bioeng., 19, 1125
(1977).
3. B. L. Maiorella, C. R. Wilke, and H. W. Blanch, Adv. Biochem.
Eng., 20, 43 (1981).
4. M. Cheryan and M. A. Mehaia, Biotechnol. Lett., 5 , 519 (1983).
5 . C. Lafforgue, J. J. Malinowski, and G. Goma, Biotechnol. Lett., 9,
347 (1987).
6. H. Hoffman, T. Scheper, and K. Schiigerl, Chem. Eng. J . , 34, 813
(1987).
7. C. W. Lee and H. N. Chang, Biotechnol. Bioeng., 29, 1105 (1987).
8. P. N. Patel, M. A. Mehaia, and M. Cheryan, J . Biotechnol., 5 , 1
(1987).

1230

9. S. D. Inloes, D. P. Taylor, S. N. Cohan, A. S. Michaels, and C. R.


Robertson, Appl. Environ. Microbiol., 46, 264 (1983).
10. E. J. del Rosario, K. J. Lee, and P. L. Rogers, Biotechnol. Bioeng.,
21, 1477 (1979).
11. J . M. Lee, J. F. Pallard, and G. A. Coulman, Biotechnol. Bioeng.,
25, 497 (1983).
12. T. W. Nagodawithana, C. Castellano, and K. H. Steinkraus, Appl.
Microbiol., 28, 383 (1974).
13. T. W. Nagodawithana and K. H. Steinkraus, Appl. Environ. Microbiol., 31, 158 (1976).
14. H. G. Monbouquette, Biotechnol. Bioeng., 29, 1075 (1987).
15. J. A. Roels, Energetics and Kinetics in Biotechnology (Elsevier Biomedical Press, Amsterdam, 1983).
16. M. Mota, C. Lafforgue, P. Strehaiano, and G. Goma, Bioprocess
Eng., 2, 65 (1987).
17. S . Aiba, M . Shoda, and M. Nagatani, J . Ferment. Technol., 47,
1401 (1969).
18. R. L. Fournier, Biotechnol. Bioeng., 28, 1206 (1986).
19. A . J. Daugulis and D. E. Swaine, Biotechnol. Bioeng., 29, 639
(1987).
20. K . Richter and D. Meyer, 8th International Biotechnology Symposium, Poster D-107, Paris, France, 1988.
21. J. J. Malinowski, C. Lafforgue, and G. Goma, J . Ferment. Technol.,
65, 319 (1987).

BIOTECHNOLOGY AND BIOENGINEERING, VOL. 34, NOVEMBER 1989

You might also like