Kinetic Model for Nitrogen-Limited Wine
Fermentations
Amanda C. Cramer,1 Sophocles Vlassides,1
David E. Block1,2
1
Department of Viticulture and Enology, University of California, One
Shields Avenue, Davis, California 95616; telephone: (530) 754-6046; fax: (530)
752-0382; e-mail: deblock@ucdavis.edu
2
Department of Chemical Engineering and Materials Science, University of
California, One Shields Avenue, Davis, California 95616
Received 19 September 2000; accepted 4 September 2001
Abstract: A physical and mathematical model for wine
fermentation kinetics has been developed to predict
sugar utilization curves based on experimental data from
wine fermentations with various initial nitrogen and
sugar concentrations in the juice. The model is based on:
(1) yeast cell growth limited by nitrogen; (2) sugar utili-
zation rates and ethanol production rates proportional
solely to the number of viable cells; and (3) a death rate
for cells proportional to alcohol content. All but one
parameter in the model can be estimated from existing
data. However, experiments to ®nd this ®nal parameter,
a constant describing cell death, indicate that cell death
may not be the critical factor in determining fermenta-
tion kinetics as cell viability remains signi®cant until
sugar utilization has ceased. The model, nevertheless,
predicts a transition from normal to sluggish to stuck
fermentations as initial nitrogen levels decrease. It also
predicts that fermentations with high initial Brix levels
may go to completion when supplemented with nitro-
gen in the form of ammonia. Therefore, we hypothesize
that the model is valid but that ethanol causes the yeast
cells to become inactive while remaining viable. Exper-
imental veri®cation of the model has been performed
using ¯ask-scale experiments. The model has also been
used to evaluate the possibility of using nitrogen or vi-
able cell additions to avoid or correct problem (i.e.,
sluggish or stuck) fermentations. ã 2002 John Wiley &
Sons, Inc. Biotechnol Bioeng 77: 49±60, 2002.
Keywords: wine; mathematical model; stuck fermen-
tations; ethanol
INTRODUCTION
During the fermentation of grape juice to wine, the
sugar present in the juice is converted to ethanol by
yeast that is either present in the fermentor or inoculated
into the fermentor. Typically, yeast will fully utilize the
sugar present in the juice in seven to 10 days (Bisson,
1999). However, under certain circumstances, fermen-
tations will take signi®cantly longer to ®nish (i.e., slug-
gish fermentations) or will leave residual sugar greater
than 0.4% w/v (i.e., stuck fermentations). These latter
Correspondence to: David E. Block, Department of Viticulture and
Enology, University of California, One Shields Avenue, Davis, CA
95616.
ã 2002 John Wiley & Sons, Inc.
DOI 10.1002/bit.10133
types of abnormal fermentation kinetics can be a serious
problem in an industrial setting as they can lead to un-
scheduled loss of tank capacity due to extended pro-
cessing time and the potential for microbial instability of
the ®nal product due to residual sugar. The ability to
predict the fermentation kinetics or Brix curves prior to
inoculation based solely on grape juice characteristics
and intended processing would, therefore, be a useful
tool to develop.
Several physical and mathematical models for wine
fermentations have been developed and reported in the
literature over the last 40 years. These models have re-
cently been reviewed by Marin (1999). One of the ®rst
comprehensive kinetic models for wine fermentations
was reported by Boulton (1980). This mechanistic model
included the in¯uence of glucose and fructose levels,
ethanol levels, and temperature on sugar utilization
rates and captured the general trends found in practice.
Caro et al. (1991) used a similar mechanistic model to
describe sugar utilization but also included sugar utili-
zation pathways other than ethanol production (i.e.,
mainly respiration) in order to address the mismatch in
ethanol concentration found in previous models. Several
more empirical (Aiba et al., 1968; Bovee et al., 1984;
Giovanelli et al., 1996; Holzberg et al., 1967; Lopez and
Secanell, 1992) or non-parametric (e.g., neural network)
models (Insa et al., 1995) have also been published.
The mechanistic models developed have typically used
growth expressions dependent on sugar concentration in
a Monod-type relationship. In this manner, these mod-
els have indirectly linked cell growth and ethanol pro-
duction. However, there are clearly cases where yeast
cell growth is complete prior to signi®cant utilization of
sugar. Therefore, another juice component must be the
growth-limiting nutrient. Benzenger and Navarro (1988)
have reported a model incorporating the e€ects of initial
nitrogen level. However, the mechanistic expressions
examined were not satisfactory for ®tting their kinetic
data, and therefore, an empirical model was developed
instead. Mathematical models with other limiting nu-
trients have not been reported. Various models have
also dealt di€erently with the e€ect of alcohol on fer-
mentation productivity and cell viability. These include
expressions in the form of growth inhibition (Aiba et al.,
1968; Holzberg et al., 1967; Marin, 1999; Mota et al.,
1984) and also in the form of speci®c expressions for cell
death (Boulton, 1980; Caro et al., 1991). Incorporating
the e€ect of ethanol in some form into a model is im-
portant for understanding and predicting the occurrence
of problem fermentations. A model that can predict the
transition from normal to sluggish or stuck fermentation
kinetics would be extremely useful in determining a
basic mechanism for this important transition. Previous
mechanistic models have not addressed this issue.
Factors leading to sluggish and stuck fermentations
have been extensively studied. Comprehensive reviews
of this topic have recently been published by Bisson
(1999) and by Alexandre and Charpentier (1998). These
factors include low nitrogen and oxygen (and other
nutrient) levels, high sugar levels, high ethanol levels and
ethanol toxicity, extremes in temperature and pH, and
the presence of toxins. While the causes of problem
fermentations have been well documented, the basic
mechanism that results in the cessation of the conversion
of sugar to ethanol is not well understood. It has been
shown that the process of converting sugar to ethanol is
controlled at the membrane transport step (Bisson,
1999), but how the signal to shut down transport is
transmitted and the actual mechanism through which
the transport proteins are inactivated or down-regulated
have not been established.
Perhaps the most studied cause of sluggish and stuck
fermentations is low initial nitrogen levels. Nitrogen in
grape juice is made up of an ammonia component and a
more complex amino-acid-based nitrogen component. It
is clear from the literature that low levels of nitrogen are
likely to lead to problem fermentations (Bely et al.,
1990, 1994; Ingledew and Kunkee, 1985; Jiranek et al.,
1995; Monteiro and Bisson, 1991; Sablayrolles et al.,
1996; Spayd et al., 1995). However, some researchers
have linked low nitrogen to low cellular activity (Bely et
al., 1990, 1994) and others have associated this condi-
tion with low resultant biomass concentrations
(Monteiro and Bisson, 1991; Spayd et al., 1995). Addi-
tion of nitrogen in the form of diammonium phosphate
(DAP) can alleviate these problems in cases in which the
initial nitrogen level is low but not in cases where the
nitrogen level in the juice is already high (Bely et al.,
1990). Several groups have reported a transition point of
140 mg N/L (Bely et al., 1990; Ingledew and Kunkee,
1985), which may correspond to the point at which the
biomass no longer increases with increasing initial ni-
trogen in some juices. Others have shown that nitrogen
addition throughout the course of fermentation is also
e€ective to varying degrees in assuring rapid completion
of sugar utilization, especially when the initial nitrogen
level is low (Bely et al., 1990). Jiranek et al. (1995) have
examined speci®c components of the nitrogen typically
50 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002
available in juice to ®nd if speci®c amino acids might be
limiting fermentation in certain cases. They found that
amino acids could be grouped into three categories
based on utilization pattern with one group (including
arginine) preferentially depleted from the medium.
While the detailed mechanism of regulation based on
nitrogen components of juice has not been established, it
is clear that nitrogen can play a key role in determining
both the rate and extent of normal and problem fer-
mentations.
Therefore in this paper, we focus on the development
of an unstructured mechanistic model for wine fermen-
tation kinetics based on nitrogen as a growth-limiting
nutrient. We then use the model developed to predict the
role of nitrogen and sugar levels in determining fer-
mentation kinetics. Next, the predictions of the model
are compared to experimental data from small-scale
fermentations, and the information used to rede®ne the
model. Finally, the model is used to predict strategies for
intervening in fermentations that are destined to become
sluggish or stuck.
MATERIALS AND METHODS
Small-Scale Wine Fermentations
Fermentations were conducted using 400 mL total vol-
ume in 500-mL Erlenmeyer ¯asks equipped with fer-
mentation air locks. The juice used for these experiments
was Chardonnay juice, which was either undiluted or
diluted. For diluted juices, diammonium phosphate
(DAP) and sugar were added to vary the concentrations
of these nutrients. The sugar added was 50% glucose and
50% fructose. Fermentations were inoculated with active
dry Premier Cuvee yeast (Red Star, Milwaukee, WI) at
approximately 0.25 g dry weight/L that had been rehy-
drated at 0.1 g/mL for 30 min with stirring according to
the manufacturer's recommendations. Fermentations
were maintained at 23 ‹ 2°C and were minimally agi-
tated at 50 rpm on a platform shaker. Twelve fermen-
tations were initially completed to verify the model
developed. Samples were taken periodically for analysis
of cell number and viability, as well as of sugar and
nitrogen concentrations. These samples were taken two
to three times per day during periods of rapid cell
growth or sugar utilization and less frequently later in
the fermentation. Prior to sampling, ¯asks were mixed
by vigorous swirling in order to suspend all solids.
Total and Viable Cell Concentration
Total and viable cell numbers were estimated micro-
scopically using a hemacytometer. Cells (50 lL) were
added to 50 lL of methylene blue solution (0.4%
methylene blue, 10% ethanol [95%], 0.4 M KH2PO4) and
mixed. Blue cells were counted as dead cells, while cells
without obvious color were counted as live cells. Total
solids concentration was evaluated using optical density
at 600 nm (OD600) using a Milton Roy Spectronic 601
spectrophotometer (Rochester, NY) against a water
blank. Dilutions were made as necessary in order to
keep the optical density below 0.4 AU.
Assays for Sugar and Nitrogen Concentration
For all samples, individual sugars and organic acids
were measured using a Hewlett-Packard HP1100 HPLC
with two 30-cm Aminex HPX-87H columns (Bio-Rad,
Hercules, CA), a mobile phase of 0.001 M sulfuric acid
at a ¯ow rate of 0.6 ml/min, a column temperature of
55°C, and a refractive index detector. In addition, am-
monia was measured using an enzymatic assay (Sigma
Chemical Co., St. Louis, MO), and a-amino nitrogen
was measured using the method of Dukes and Butzke
(1998).
Model Simulations
Simulations were carried out using the numerical inte-
gration feature of Mathematica (Wolfram Research,
Champaign, IL) on a 200 MHz or 333 MHz personal
computer. Model parameters were chosen as described
in the text.
MODEL DEVELOPMENT AND PARAMETER
ESTIMATION
Model Development
In the process of measuring nitrogen levels and cell
concentration in wine fermentations prior to the current
studies, we observed that exhaustion or near exhaustion
of nitrogen seemed to correspond in time with the ces-
sation of exponential cell growth and the initiation of
signi®cant ethanol production. For this reason, we be-
gan to develop a model that is based on nitrogen-limited
growth with completely non-growth-associated produc-
tion of ethanol. Four di€erential equations were derived
to describe the kinetic behavior of the concentrations of
viable yeast cells, total nitrogen, sugar (including both
glucose and fructose), and ethanol.
First, an expression was derived for viable cell con-
centration, Xv:
dXv
ˆ lXv kd Xv ; …1†
dt
where l is in the form of the Monod relationship with
respect to the total nitrogen concentration:
lmax N
lˆ ; …2†
KN ‡ N
and the death constant, kd is linearly related to the
ethanol concentration, E, at any given time in the fer-
mentation:
CRAMER ET AL.: MODEL FOR WINE FERMENTATION KINETICS
kd ˆ k0d E: …3†
Unlike previous models, the growth rate in this model is
not a function of the sugar concentration as we have
observed that the sugar concentration does not change
appreciably until the end of the exponential phase of
growth and entry into stationary phase. In fact, most
sugar is utilized after this point, and therefore, it is not
the limiting nutrient.
Total nitrogen concentration, which would include
nitrogen measured as ammonia and as a-amino nitrogen,
is used proportional to cell growth as shown in Eq. (4):
dN lXv
ˆ ; …4†
dt YX=N
where l has the same functionality as in Eq. (2), and YX/N
is a yield coecient of biomass on nitrogen.
As we have observed sugar utilization and ethanol
production to commence just prior to the cessation of
exponential cell growth, ethanol production in this
model is viewed as a completely non-growth-associated
(Luedeking and Piret, 1959) stoichiometric bioconver-
sion of sugar to ethanol:
dE
ˆ bXv ; …5†
dt
dS bXv
ˆ ; …6†
dt YE=S
where YE/S is the stoichiometric coecient describing
the formation of ethanol from sugar, and b is a function
of the sugar level, S, as in Eq. (7):
bmax S
bˆ : …7†
KS ‡ S
This form of the speci®c ethanol production rate or
sugar utilization rate is consistent with previous obser-
vations that sugar transport in Saccharomyces cerevisiae
is by facilitated di€usion and can be described by this
type of relationship (Coons et al., 1995). Utilization of
glucose and fructose, the main sugars in grape juice, is
not distinguished in this model.
It should be noted here that while a completely non-
growth-associated model for ethanol formation does not
preclude formation during growth, it does make ethanol
formation proportional to biomass concentration, which
is highest at the completion of cell growth. While it is
possible that a small growth-associated component of
ethanol production may actually be present, no signi®-
cant relationship was found in our data between the
speci®c ethanol production rate, (1/X)dP/dt, and the
speci®c growth rate, l. This type of relationship would be
expected for growth-associated product formation (Lu-
edeking and Piret, 1959), and thus, inclusion of a growth-
associated term in the model was not justi®ed in this case.
51
Table I. Model parameters used for initial simulations.
Parameter Value
)1
lmax 0.16 h
KN 0.010 g N/L
YX/N 31 g biomass/g nitrogen
bmax 0.3 g ethanol g biomass)1 h)1
KS 10 g sugar/L
YE/S 0.47 g ethanol/g sugar
k0d 0.0001 L/g ethanol h
Parameter Evaluation
The model described in the previous section contains the
seven constants shown in Table I. Of these seven con-
stants, six of them can be easily estimated from data in
the literature and from our own experimental data.
These are lmax, KN, YX/N, bmax, KS, and YE/S. The
maximum speci®c growth rate can be estimated from
slope of a semi-log plot of viable cell concentration
versus time. The yield coecients, YX/N and YE/S, can be
estimated from the ratios of the speci®c components
consumed or produced during the course of a batch
fermentation. The saturation constants, KN and KS,
were estimated by observing the level of nitrogen or
sugar at which the rate began to decrease signi®cantly.
While these parameters are actually de®ned as the nu-
trient concentration for which the rate is half of maxi-
mum, the model predictions are not sensitive enough to
these parameters to justify a more precise measurement
of these values. For instance, the value of KN can be
varied from 5 mg N/L to 20 mg N/L with little e€ect on
the predicted sugar utilization curves. Finally, the
maximum speci®c sugar utilization rate was calculated
from the linear portion of the sigmoidal sugar utilization
curves. The values chosen for the initial simulations
using the model developed are shown in Table I. The
value of the death constant, k0d , is not easily calculated
from existing data as little previous work has been
published on cell viability as a function of time in wine
fermentations. Therefore, the order of magnitude of this
parameter (0.0001) was originally chosen based on rea-
sonable prediction of normal fermentation kinetics.
In order to evaluate the general predictive capacity of
the model developed, normal fermentation kinetics were
predicted since the shape of these curves has been well
established experimentally. To examine normal fer-
mentation kinetics, typical initial total nitrogen, cell,
and sugar concentrations were used as initial conditions.
These were 150 mg N/L, 0.1 g cells/L, and 240 g sugar/
L, respectively. As can be seen in Figure 1, the model
predicts that viable cell concentration reaches a maxi-
mum as nitrogen is depleted. Ethanol production begins
just prior to this point, and sugar utilization begins at
approximately the same time. It can be noted here that,
even though this model is based on completely non-
growth-associated formation of ethanol, the model
predicts that approximately 10% of the sugar has been
52 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002
Figure 1. Simulation of wine fermentation kinetics. The kinetics of
yeast growth in grape juice were simulated using the model described
in the text and the model parameters shown in Table I. Trends for
biomass, total nitrogen, sugar, and ethanol concentrations are shown.
utilized by the end of exponential cell growth. The shape
of the predicted sugar utilization curve is typical of most
normal wine fermentations (Boulton et al., 1996).
RESULTS
Experimental Veri®cation of Model Predictions
for Normal Fermentations and Estimation
of the Death Constant
In order to verify the relevance and predictive capacity
of the model developed for wine fermentation kinetics
experimentally, it was important to obtain an indepen-
dent measure of the death constant, k0d , to compare with
the order of magnitude estimate discussed in the previ-
ous section. To do this, Eq. (1) was used in the limiting
case of exhausted nitrogen and sugar, assuming that k0d
is constant throughout the fermentation. For this case,
the cell growth term is negligible, and the ethanol con-
centration is a constant. Therefore, the death constant
can be calculated from experimental data by ®nding the
slope of the line formed by plotting the ln(Xv) versus
time. To generate experimental data to perform this
calculation, small-scale ¯ask experiments (400 mL of
juice in a 500-mL Erlenmeyer ¯ask) were inoculated
with Premier Cuvee yeast and cell viability followed
throughout the fermentation as described in the Mate-
rials and Methods. Initial sugar and nitrogen concen-
trations were varied in these experiments (eight
combinations) in order to achieve a range of initial vi-
able cell and ®nal ethanol concentrations. This was ac-
complished by using 300 mL of juice diluted with water
supplemented with the appropriate amounts of sugar
and DAP. The mean value for k0d calculated in this
manner was 0.00005 L/g ethanol h, one half of the order
of magnitude estimate established in the previous sec-
tion. Figure 2a illustrates data from a typical experi-
mental ¯ask fermentation with ``normal'' initial nitrogen
and sugar concentrations of 168 mg/L and 229 g/L, re-

Figure 2. Fitting experimental fermentation data and determination
of kd0 . Experimental data from 400-mL fermentations performed in
500-mL Erlenmeyer ¯asks are ®t with the model. (a) kd0 ˆ 0:00005 is
used as this is the value of this parameter estimated independently
from experimental data. (b) Same experimental data is shown, but with
the model ®t using a kd0 ˆ 0:00022. ``Viable Biomass'' is estimated from
total cell numbers, viable cell fraction, and correlation with initial yeast
dry weight added at inoculum.
spectively. The curves on this plot represent the model ®t
of the data using the calculated death constant along
with the other parameter values found in Table I. As can
be seen from this plot, the prediction of nitrogen ®ts the
experimental data well. However, the sugar utilization
curve only ®ts the data well up to approximately 60 h,
and the viable cell concentration curve only ®ts well up
to the point of nitrogen exhaustion. After the maximum
viable cell concentration is reached, the model predic-
tion loosely passes through the experimental data, but
does not exhibit the same general behavior. This is due
mainly to the observation that, experimentally, the
number of viable cells remains high throughout the
fermentation, and only decreases after signi®cant sugar
utilization has ceased. To our knowledge, this observa-
tion has not been reported before, likely because of
sparse data collection, but is consistently observed in
our fermentations and in those of others (L.F. Bisson,
personal communication). In fact, the number of viable
cells is quite high even in the case of sluggish or stuck
fermentations until after signi®cant sugar utilization has
CRAMER ET AL.: MODEL FOR WINE FERMENTATION KINETICS
ceased (Schreiber and Block, 2001). This point is further
illustrated in Figure 3 in which viable cell concentration
and sugar utilization curves are shown for six fermen-
tations including ones exhibiting ``normal'' kinetics,
such as Figure 3a and Figure 3d, and several that have
become sluggish or stuck due to low nitrogen (Fig. 3b,c)
or high initial sugar concentration (Fig. 3e,f).
In order to ®nd whether another value of the death
constant would give a better ®t of the experimental data
(especially the sugar utilization curve), the value of k0d
was systematically varied and the predicted kinetics
evaluated in terms of the ¯ask fermentation data. Using
a value of k0d ˆ 0:00022, the ®t of the experimental sugar
data was excellent as evidenced by the plot in Figure 2b.
In this plot, the same experimental data shown in Figure
2a is ®t with the model using this new value of k0d . It is
clear that the ®t of sugar concentration at later times in
the fermentation is improved, while the ®t of the viable
cell concentration during exponential growth and ni-
trogen concentration are largely unchanged, as expected
from the model. The trend in viable cell concentration
after the initial growth phase does not ®t the experi-
mental at all, as the number of viable cells does not
decrease dramatically until after sugar exhaustion.
Given the predictive capability of the model for sugar
utilization and the mismatch of the prediction and ex-
perimental data for cell viability, two explanations are
possible. First, it is possible that the model is not physi-
cally accurate and that the ®t of the parameters other than
cell viability is by chance. The second possible explanation
is that the mathematical form of the model is accurate, but
that Xv does not represent viable cell concentration, but
rather the concentration of cells actively converting sugar
into ethanol. In other words, the cells may maintain via-
bility in the presence of ethanol, but not their ability to
produce more ethanol. From the data presented below, we
feel that the latter scenario is more likely.
Effect of Nitrogen on Fermentation Kinetics
Nitrogen-limited growth is a key part of the model
outlined here. Therefore, it is important to evaluate
experimentally, the e€ect of increasing or decreasing
nitrogen on maximum cell concentration and sugar
utilization rates. To do this, a series of ¯ask fermenta-
tions were carried out in which the initial nitrogen levels
were adjusted to the desired concentration by adding
DAP to juice diluted one part juice to three parts water.
In all cases, sugar (50% glucose and 50% fructose) was
added to bring the diluted juice back to the original
sugar concentration. In this way, all initial juices were
identical, except for nitrogen concentration.
Figure 4 shows the e€ect of increasing nitrogen on
total cell concentration during the fermentation. The
experimental data demonstrate that a proportional in-
crease in the maximum cell concentration occurs with
53
Figure 3. Cell viability measurements during normal, sluggish, and stuck fermentations. Each of the above graphs shows a sugar utilization curve
(closed circles) with the associated measurements of viable cell number (open circles). The graph shown in (a) illustrates these curves for a ``normal''
fermentation starting with approximately 220 g sugar/L and 230 mg N/L. The fermentations represented in (b) and (c) also have an initial sugar
concentration of approximately 220 g/L, but have initial nitrogen concentrations of 135 and 60 mg N/L, respectively. The graphs in (d), (e), and (f)
represent data from three fermentations all containing approximately 170 mg N/L, but initial sugar concentrations of 229, 263, and 291 g/L,
respectively.

increasing initial nitrogen up to approximately 140 mg/
L. After this point, nitrogen is no longer the limiting
nutrient in the juices used, so that maximum cell con-
centration no longer increases at high nitrogen levels.
The model developed here predicts the increase in cell
concentration, but does not account for the saturation
behavior at higher nitrogen concentrations. Circum-
stantial evidence for the presence of other limiting
nutrients in the diluted juice can be seen in that undi-
luted juice sustains a higher cell population than the
diluted juice with equivalent sugar and nitrogen added
back (data not shown). This e€ect of increased cell
concentration at higher initial nitrogen was con®rmed at
the 20-L carboy and 200-L barrel scales (data not
shown). However, for the barrel fermentations, the juice
was only diluted to 40% of the original concentration

54 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002


Figure 4. Prediction of the e€ect of nitrogen on the total biomass
concentration. Experimental ¯ask fermentations were performed with
varying amounts of initial nitrogen, and the biomass concentration
followed using optical density and an experimentally determined
conversion factor. Experimental points are represented by symbols.
The simulations, represented by the lines on this plot, were carried out
with the same initial conditions as the actual ¯asks. Model parameters
for these simulations are given in the text.
and the saturation e€ect of increased nitrogen was not
as pronounced, even at a level of 285 mg N/L. There-
fore, it is likely that dilution of the juice arti®cially
lowers the concentration of factors other than nitrogen
and sugar to a point where they become limiting,
whereas at normal (i.e., undiluted juice) levels of these
factors, nitrogen is still limiting growth at a level con-
siderably higher than 140 mg/L.
Another prediction of the model is that decreasing
nitrogen levels at a constant initial sugar level will lead
to a transition from normal fermentation kinetics to
sluggish or stuck fermentations. We tested this predic-
tion experimentally using ¯ask fermentations with di-
luted juice and adjusted nitrogen and sugar levels. As
can be seen in Figure 5, decreasing nitrogen from the
juice nitrogen level of 230 to 138 mg/L decreases the
sugar utilization rate observed. By 60 mg/L, the fer-
mentations are exhibiting kinetics that would be con-
sidered sluggish, and at 15 mg/L the fermentation is
stuck. From an interpretation of the model, this means
that the lower nitrogen is leading to fewer cells. By
starting sugar utilization with fewer cells, there is a
higher probability of all cells dying or becoming inac-
tivated prior to sugar exhaustion. These kinds of tran-
sitions are predicted by the model, though the ®t using
the parameters discussed above does not predict the
experimentally determined minimum nitrogen concen-
tration necessary for completion of the fermentation. In
fact, in order to achieve the model ®t shown in Figures 4
and 5, the YX/N was increased to 50 g biomass/g nitro-
gen, the bmax was decreased to 0.12 h)1, and the k0d was
decreased to 0.00013. As these data were actually col-
CRAMER ET AL.: MODEL FOR WINE FERMENTATION KINETICS
Figure 5. Prediction of the e€ect of nitrogen on sugar utilization
rates. Experimental ¯ask fermentations were performed with varying
amounts of initial nitrogen, and the sugar utilization curves followed
using an HPLC assay for glucose and fructose. Experimental points
are represented by symbols. The simulations, represented by the lines
on this plot, were carried out with the same initial conditions as the
actual ¯asks. Model parameters for these simulations are given in the
text.
lected in a separate experiment than those collected in
the previous section, it is likely that these changes are
due to slightly di€erent experimental conditions. The
di€erence in YX/N may be due to a variation in ammo-
nia/alpha-amino nitrogen ratio or di€erent initial con-
centrations of oxygen, vitamins, fatty acids, or other
trace nutrients that could have been caused by changes
in the di€erent juice dilutions used in these experiments.
However, the calculated YX/N values are on the same
order as those calculated from similar nitrogen-limited
fermentation data (Salmon, 1989). A lower temperature
could explain variation in bmax and k0d as the experi-
ments were conducted at room temperature.
Over the range of initial nitrogen levels examined in
these studies, the maximum viable cell concentration
increases proportionally to increases in nitrogen levels
up to approximately 140 mg/L. Above this value, cell
concentration no longer increases appreciably, presum-
ably as some other nutrient has become limiting (i.e.
nitrogen is in excess).
Effect of High Initial Sugar Concentration
on Fermentation Kinetics
In order to further study the predictive capability of the
model, two ¯ask experiments were conducted at high
initial sugar concentrations. First, a series of ¯ask fer-
mentations was performed varying the initial sugar
concentration from 163 to 291 g/L, all at a nitrogen level
of 167 ‹ 5 mg N/L. The sugar utilization curves for
these fermentations are shown in Figure 6. As can be
seen in this plot, the sugar utilization rates late in the
fermentation slow down as the initial sugar in the juice
55
Figure 6. Prediction of the e€ect of increasing sugar concentrations
on fermentation kinetics. Experimental ¯ask fermentations were per-
formed with varying amounts of initial nitrogen and the sugar utili-
zation curves followed using an HPLC assay for glucose and fructose.
Experimental points are represented by symbols as shown in the leg-
end. The lines represent model predictions using the parameters shown
in Table I with a kd0 of 0.00022.

increases. The maximum sugar utilization rates do not,

however, change signi®cantly. These observations are
consistent with the model as all fermentations would
basically have the same cell concentration, but those
with higher sugar would be exposed to a higher ethanol
concentration for an extended period of time. The model
predicts these curves well, including the transition from
normal kinetics to sluggish or stuck kinetic patterns,
using a k0d ˆ 0:00022 and the other model parameters
used in Figure 2. In fact, these experiments were per-
formed in parallel using the same base juice and dilu-
tions.
In the second experiment, fermentations using juices
containing higher sugar concentrations (263 and 290 g/
L) that had caused problem fermentations in the previ-
ous experiment were initiated with increased N con-
centrations. From Figure 7, it is obvious that the model
over-predicts the e€ect of additional nitrogen. The rea-
son for this, as can be seen in Figure 4, is that cell
concentration is no longer increasing with initial nitro-
gen at such high initial nitrogen levels as the model
would predict. Therefore, the sugar utilization rates do
not change signi®cantly with varying nitrogen concen-
tration. However, the minimum sugar level reached by
300 h does decrease slightly at higher initial nitrogen
levels, indicating a potential physiological or metabolic
change in the cells that is occurring along with any
changes in cell concentration.
Predicting Conditions for Restarting
or Completing Problem Fermentations
Previous experimental studies have shown that additions
of DAP to a nitrogen-de®cient fermentation at various
times during the fermentation can allow the fermenta-
Figure 7. E€ects of high initial nitrogen concentration on completion
of fermentations with high initial sugar concentrations. These plots
demonstrate the experimental and simulated e€ects of adding nitrogen
to fermentations with 263 g sugar/L ([N] = 163, 261, and 335 mg N/L)
(a) and 290 g sugar/L ([N] = 163 and 356 mg N/L) (b). The experi-
mental points are shown as symbols and were derived from ¯ask fer-
mentations. The lines represent the simulations with the same initial
conditions as the actual ¯asks.
tion to ®nish more rapidly (Bely et al., 1990). In order to
evaluate the ability of the model to predict this recovery,
we simulated a nitrogen-limited fermentation using the
initial conditions and parameters used in Figure 1, but
with a lower initial nitrogen content of 50 mg N/L. The
model predicted that this fermentation would stick at
about 32 g sugar/L. Using the model, we were also able
to predict that at least 27 mg N/L would have to be
added in order to avoid an extended period of sugar
utilization. For a qualitative comparison to previous
work, additions of 63 mg N/L (300 mg DAP/L) at 0, 20,
60, 120, 360, and 480 h after inoculation were simulated
and evaluated for their potential in avoiding a problem
fermentation using the model. As can be seen in Figure
8a, the model predicts that all additions of DAP will
result in successful completion of the fermentation. For
comparison to previous work, Figure 8b shows the
sugar utilization rate (as calculated from a simple two-

56 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002


Figure 8. Simulation of additions of DAP to a problem fermentation
at various points during the fermentation. A base fermentation con-
taining 50 mg N/L is used. The simulation of this set of conditions
shows that this fermentation will stick with approximately 32 g sugar/
L. By simulating the addition of DAP at various time points, the
fermentation goes to completion. This is demonstrated by examining
the predicted total sugar curves (a) and the instantaneous sugar utili-
zation rate (b). The lines in (b) correspond to the same lines labeled in
(a).
point forward di€erence method using the predicted
sugar concentration) for each of the cases. These results
agree well qualitatively with the experimental results
reported in the literature (Bely et al., 1990).
We also used the same case study to evaluate the
possibility of using viable cell additions to the nutrient
de®cient fermentation to allow it to continue to com-
pletion. The model predicts that adding viable cells at a
minimum of 0.85 g viable cells/L at any point in the
fermentation will allow complete sugar utilization.
However, as can be seen in Table II, waiting until later
in the fermentation to add the viable cells will lengthen
the total amount of time necessary to complete the fer-
mentation even though the time post-addition decreases.
These data are interesting not only in their prediction
that earlier cell addition is bene®cial (i.e., addition of a
larger inoculum initially can save weeks in fermentation
time) but also that a very large population of cells (more
than three times the original inoculum) is needed to
overcome the nutrient de®ciency.
CRAMER ET AL.: MODEL FOR WINE FERMENTATION KINETICS
Table II. Simulation of viable cell additions for avoiding stuck fer-
mentations.
Time of cell Post-addition Total time to
additiona (h) time to drynessb (h) drynessb (h)
0 460 460
20 460 480
60 430 490
120 380 500
360 290 650
480 270 750
a
Each cell addition consists of 0.85 g viable biomass/L.
b
Dryness is de®ned as less than 4.0 g sugar/L.
DISCUSSION
The model that we have developed for yeast fermenta-
tions in nitrogen-limited grape juice is a mechanistic
model based on nitrogen as the growth-limiting nutrient.
In the model, ethanol is produced in a completely non-
growth-associated manner, and therefore, sugar utiliza-
tion is also uncoupled from cell growth. Cell viability, or
at least the capacity to convert sugar to ethanol, is re-
duced in a time-dependent manner with a rate constant
proportional to the concentration of ethanol. The
mathematical model derived from this physical model is
composed of four non-linear, coupled ordinary di€er-
ential equations that can be readily solved using nu-
merical integration. As shown above, the predictions of
kinetics from this model have been quite good, both for
normal fermentation kinetics under nitrogen limitation
as well as for predictions of transitions to sluggish or
stuck fermentations. It predicts the experimentally ob-
served e€ects of low nitrogen and high sugar concen-
trations. It also seems to predict the e€ect of nitrogen
additions later in fermentations otherwise destined to be
incomplete. Because of this predictive capability, this
model has the potential to become a useful tool for ex-
amining the basic causes of sluggish and stuck fermen-
tations in terms of a theoretical framework. It also has
potential as a tool for determining when certain juice
characteristics (e.g., carbon/nitrogen ratios) are likely to
result in problem fermentations and how to address the
predicted problem before it is too late.
Other models for wine fermentations have been de-
veloped previously with various levels of success in
predicting fermentation kinetics. However, the advan-
tage of the current model is that it is mechanistic in
nature and relatively simple with only seven parameters,
six of which can be estimated from data generated in our
laboratory or reported in the literature. The ®nal model
parameter, k0d , could not be estimated from previous
data, as comprehensive data on cell viability was not
available in the literature for the conditions studied here.
However, we were able to generate data to calculate this
parameter as well. Even with this low level of com-
plexity, our model seems to have good predictive ca-
pacity as compared with many more complex models
57
and does not contain multiple parameters that can be
®tted but not veri®ed experimentally. This is also the
®rst time, to our knowledge, that a mechanistic model
has been used to predict conditions leading to sluggish
or stuck fermentations which are a critical problem for
the wine industry. Previous attempts to predict sluggish
or stuck fermentations have utilized non-mechanistic
models or correlations (Giovanelli et al., 1996; Insa et al.,
1995), which, while useful in a practical setting, yield less
information toward a basic understanding of the un-
derlying phenomena.
This model does not, however, include all of the fea-
tures of some previous models reported in the literature
(Boulton, 1980; Caro et al., 1991). Most notably, the
current model assumes an isothermal fermentation that
is rarely observed in practice, especially at a larger scale.
Since the experiments described here were carried out at
a single temperature in small ¯asks, this assumption may
not be critical for the current studies. Incorporation of
non-isothermal fermentation conditions would be an
important, though non-trivial, extension of this work.
Inclusion of temperature e€ects will be a complex task
as many of the parameters in the current model (most
notably lmax, k0d , and bmax) will all be functions of
temperature and each function will have to be found
independently. Though for the immediate purpose of
investigating the role of nitrogen in normal and problem
fermentations the assumption of an isothermal process
is adequate, addition of temperature e€ects would pro-
vide even more insight as long as the parameters added
to the model could be veri®ed experimentally in some
manner.
In addition to the e€ects of temperature, substrate
(i.e., sugar) inhibition, substrate competition, and
product inhibition are not speci®cally included in the
viable cell growth term as in some previous models
(Aiba et al., 1968; Holzberg et al., 1967). As we have
found that minimal sugar is used for cell growth com-
pared to that converted to alcohol, sugar concentration
is not critical to cell growth in our model. Only nitrogen
concentration is important in determining ®nal cell
concentration and growth rate. Ethanol inhibition is
accounted for in the term that describes cell inactivation
or death and, therefore, does not appear in the expres-
sion for the speci®c growth rate. This is a good as-
sumption, especially as the ethanol level remains low
prior to the cessation of growth in the model.
In this model, the death constant was chosen arbi-
trarily to be a linear function of ethanol concentration,
since this is the simplest starting point for a relationship
between ethanol and cell death or inactivation. Leao
and Van Uden had previously found that the entropy of
activation of thermal cell death is linearly related to
ethanol concentration for concentrations above 0.3 M
(13.8 g/L) which corresponds to most of a typical wine
fermentation (Leao and Van Uden, 1982). This rela-
tionship would actually indicate that the death constant
58 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002
is an exponential function of ethanol concentration.
However, this work was focused on thermal death, and
it is not clear that this is applicable to death or inacti-
vation by ethanol, as the physical mechanisms being
described are likely to be di€erent. The death constant
may also be a function of the physiological state of the
cells and extracellular environment (e.g., fatty acid or
sugar concentration) (Thomas et al., 1978; Thomas and
Rose, 1979). It will be important, therefore, to re-ex-
amine the exact form of this term in the future, espe-
cially as reliable experimental methods become available
for evaluating the degree of cell inactivation.
During the course of the model development and
experimentation, it became clear that cell death may not
be a critical factor in determining the kinetics of wine
fermentations. This is clear for two reasons. First, the
model developed ®ts the cell concentration, nitrogen,
and ethanol concentration kinetics well, as well as pre-
dicting conditions favoring incomplete fermentation.
However, the death constant, k0d , calculated using cell
death (k0d ˆ 0:00005) is not the same as the one that best
®ts the data (k0d ˆ 0:00022). Second, no large loss in
viable cell number was observed experimentally until
sugar utilization ceased, even in cases where fermenta-
tions were sluggish or stuck (an observation that we
have not seen reported in the literature). One explana-
tion for this observation may be that a slowing of sugar
transport may lead to a decrease in viability, rather than
the converse (i.e., lower viability leading to a decrease in
sugar transport). Therefore, the form of the model may
be correct, but Xv must represent a quantity other than
``viable'' cells. It is likely that this quantity represents the
population of cells that are actively transporting sugar
and converting it to ethanol. Decreases in this popula-
tion could be caused by genetically regulated cell be-
havior or simply by some physical change in membrane
or transport protein conformation brought about by the
presence of ethanol. More experimentation will be nec-
essary to distinguish between these two mechanisms.
The interpretation and calculation of the death con-
stant in this model relies on viability estimates using
methylene blue staining. While this method has been
used for many years to determine viability in wine fer-
mentations (Boulton et al., 1996), several studies with
brewer's yeast have found that methylene blue staining
may overestimate the live cell concentration, especially at
low viabilities (Chilver et al., 1978; King et al., 1981;
McCaig, 1990). However, we (Schreiber and Block,
2001) and others (Morrisey, 1998) have found the same
extended viability in problem fermentations using direct
colony counts on plates as well. In the studies completed
by Schreiber and Block (2001), colony counts did, in fact,
yield lower estimates of viable cell number than meth-
ylene blue staining, especially early in the fermentation.
However, neither measure of viable cell number (i.e.,
methylene blue staining and colony counts) decreased
signi®cantly until appreciable sugar utilization ceased.
 The role of juice nitrogen levels is critical to this model.
Initial nitrogen concentrations determine the maximum
cell concentration achieved, and thus the sugar utiliza-
tion rate assuming a constant sugar utilization rate per
cell. Faster sugar utilization rates will increase the like-
lihood that the sugar will be fully utilized prior to all cells
becoming inactive from exposure to ethanol. The model
seems to be able to explain the experimental observations
based only on this mechanism, without having to include
a nitrogen-caused increase in activity per cell. The model
correctly predicts an increase in the maximum viable cell
concentration with increasing initial nitrogen concen-
tration, as well as a transition to sluggish and stuck fer-
mentations as initial nitrogen concentration is decreased.
The only notable discrepancy between the model pre-
diction and experimentation observed is that additional
nitrogen did not allow high speci®c gravity wine fer-
mentations to complete. This type of behavior has been
observed, however, in other high-speci®c gravity fer-
mentations that lead to high ethanol concentration
(Thomas et al., 1996). It was most likely not observed
here because at high nitrogen concentrations, nitrogen is
no longer the limiting nutrient.
We have purposely referred to the nitrogen present as
one homogeneous source. Nitrogen is actually present in
grape juice in the form of ammonia and more complex
alpha-amino nitrogen. In the cases that we studied, all of
the ammonia is used during cell growth, as well as most
of the alpha-amino nitrogen pool. Some residual com-
plex nitrogen remained after the initial growth phase. It
is not clear from the results here what ``pool'' of nitro-
gen constitutes the growth-limiting nutrient that should
be included as ``N'' in the model. This pool may include
speci®c individual amino acids or even small peptides,
along with the readily used ammonia, as reported by
Jiranek et al. for synthetic juice medium (Jiranek et al.,
1995). It would be important to know the pool con-
stituents, as ammonia, by itself, will not substitute for
some component(s) of the nitrogen pool, and parame-
ters such as YX/N and KN are likely dependent on the
type of nitrogen present in the fermentation.
Finally, the model developed here has proven suc-
cessful in at least qualitatively predicting the outcome of
later nitrogen additions to fermentations that are des-
tined to become sluggish or stuck. The predictions are
similar to those found in the literature for this type of
addition to a series of model grape juice fermentations
(Bely et al., 1990). The model could also be used to
predict the minimum nitrogen and viable cell concen-
tration additions that would be needed to achieve a
complete fermentation. This type of prediction could be
very useful in industry for trouble-shooting and cor-
recting problem fermentations.
While the model developed concentrates on the e€ects
of nitrogen on fermentation kinetics, other factors have
been identi®ed that are instrumental in determining
fermentation kinetics. Factors such as yeast strain, ini-
CRAMER ET AL.: MODEL FOR WINE FERMENTATION KINETICS
tial dissolved oxygen levels, fatty acid concentration,
pH, and juice type have also been studied for their e€ect
on fermentation kinetics. It will be interesting to use the
model developed here as a framework for examining
each of these factors, especially in their ability to mod-
ulate the maximum cell concentration and maximum
speci®c activity achieved for any given combination of
initial nitrogen and sugar level.
The authors acknowledge useful discussions with R. Boulton,
L. Bisson, R. Kunkee, C. Butzke, and J. Schreiber and
technical assistance from E. Hill, A. Osganian, J. Rocha, J.
Ferrier, D. Bone, and M. Monticelli. Support for this project
was provided by University of California New Faculty
Startup Funds and the American Vineyard Foundation. S.V.
was supported in part by a Fullbright Scholarship.
NOMENCLATURE
b speci®c ethanol production rate, g ethanol g biomass)1 h)1
bmax maximum speci®c ethanol production rate, g ethanol
g biomass)1 h)1
l speci®c growth rate, h)1
lmax maximum speci®c growth rate, h)1
KN Monod constant for nitrogen, g nitrogen L)1
KS Michaelis±Menten-type constant for sugar, g sugar L)1
YX/N stoichiometric yield coecient of biomass on nitrogen,
g biomass g nitrogen)1
YE/S stoichiometric yield coecient of ethanol on sugar,
g ethanol g sugar)1
kd death constant (ethanol-dependent), h)1
k0d death constant (ethanol-independent), L g ethanol)1 h)1
S total sugar (glucose and fructose) concentration,
g total sugar L)1
References
Aiba S, Shoda M, Nagatani M. 1968. Kinetics of product inhibition in
alcohol fermentation. Biotechnol Bioeng 10:845±864.
Alexandre H, Charpentier C. 1998. Biochemical aspects of stuck and
sluggish fermentation in grape must. J Ind Microbiol Biotech
20:20±27.
Bely M, Sablayrolles JM, Barre P. 1990. Automatic detection of as-
similable nitrogen de®ciencies during alcoholic fermentation in
oenological conditions. J Ferm Bioeng 70:246±252.
Bely M, Salmon JM, Barre P. 1994. Assimilable nitrogen addition and
hexose transport system activity during enological fermentation. J.
Inst Brew 100:279±282.
Benzenger MC, Navarro JM. 1988. Alcoholic fermentation: model
accounting for initial nitrogen in¯uence. Biotechnol Bioeng
31:747±749.
Bisson LF. 1999. Stuck and sluggish fermentations. Am J Enol Vitic
50:107±119.
Boulton R. 1980. The prediction of fermentation behavior by a kinetic
model. Am J Enol Vitic 31:40±45.
Boulton RB, Singleton VL, Bisson LF, Kunkee RE. 1996. Principles
and practices of winemaking. New York: Chapman & Hall.
Bovee JP, Strehaiano P, Goma G, Sevely Y. 1984. Alcoholic fermen-
tation: modelling based on sole substrate and product measure-
ment. Biotechnol Bioeng 26:328±334.
Caro I, Perez L, Cantero D. 1991. Development of a kinetic model for
the alcoholic fermentation of must. Biotechnol Bioeng 38:742±748.
Chilver MJ, Harrison J, Webb TJB. 1978. Use of immuno¯uorescence
and viability stains in quality control. ASBC 36:13±18.
59
Coons DM, Boulton RB, Bisson LF. 1995. Computer-assisted non-
linear regression analysis of the multicomponent glucose uptake
kinetics of Saccharomyces cerevisiae. J Bacteriol 177:3251±3258.
Dukes BC, Butzke CE. 1998. Rapid determination of primary amino
acids in grape juice using an o-phthaldialdehyde/N-acetyl-L-cy-
steine spectrophotometric assay. Am J Enol Vitic 49:125±134.
Giovanelli G, Peri C, Parravicini E. 1996. Kinetics of grape juice fer-
mentation under aerobic and anaerobic conditions. Am J Enol
Viticulture 47:429±434.
Holzberg I, Finn RK, Steinkraus KH. 1967. A kinetic study of the
alcoholic fermentation of grape juice. Biotechnol Bioeng 9:413±427.
Ingledew WM, Kunkee RE. 1985. Factors in¯uencing sluggish fer-
mentations of grape juice. Am J Enol Vitic 36:65±75.
Insa G, Sablayrolles JM, Douzal V. 1995. Alcoholic fermentation
under oenological conditions. Bioprocess Eng 13:171±176.
Jiranek V, Langridge P, Henschke PA. 1995. Amino acid and am-
monium utilization by Saccharomyces cerevisiae wine yeasts from
a chemically de®ned medium. Am J Enol Vitic 46:75±83.
King LM, Schisler DO, Ruocco JJ. 1981. Epi¯uorescent method for
detection of nonviable yeast. ASBC J 39:52±55.
Leao C, Van Uden N. 1982. E€ects of ethanol and other alkanols on
the kinetics and the activation parameters of thermal death in
Saccharomyces cerevisiae. Biotechnol Bioeng 24:1581±1590.
Lopez A, Secanell P. 1992. A simple mathematical empirical model for
estimating the rate of heat generation during fermentation in
white-wine making. Int J Refrig 15:276±280.
Luedeking R, Piret EL. 1959. Kinetic study of the lactic acid fer-
mentation. Batch process at controlled pH. J Biochem Microbiol
Technol Eng 1:393±412.
Marin R. 1999. Alcoholic fermentation modelling: current state and
perspectives. Am J Enol Vitic 50:166±178.
McCaig R. 1990. Evaluation of the ¯uorescent dye l-anilino-8-naph-
thalenesulfonic acid for yeast viability determination. ASBC J
48:22±25.
60 BIOTECHNOLOGY AND BIOENGINEERING, VOL. 77, NO. 1, JANUARY 5, 2002
Monteiro FF, Bisson LF. 1991. Biological assay of nitrogen content of
grape juice and prediction of sluggish fermentations. Am J Enol
Vitic 42:47±57.
Morrisey K. 1998. Role of limiting long-chain unsaturated fatty acids
in stuck fermentations. M.S. Thesis. University of California,
Davis. 98 pp.
Mota M, Strehaiano P, Goma G. 1984. Studies on conjugate e€ects of
substrate (glucose) and product (ethanol) on cell growth kinetics
during fermentation of di€erent yeast strains. J Inst Brew 90:359±
362.
Sablayrolles JM, Dubois C, Manginot C, Roustan JL, Barre P. 1996.
E€ectiveness of combined ammoniacal nitrogen and oxygen ad-
ditions for completion of sluggish and stuck wine fermentations. J
Ferm Bioeng 82:377±381.
Salmon JM. 1989. E€ect of sugar transport inactivation in Sacchar-
omyces cerevisiae on sluggish and stuck enological fermentations.
Appl Environ Microbiol 55:953±958.
Schreiber JA, Block DE. 2001. Cell viability remains signi®cant in wine
fermentations until sugar utilization has ceased (manuscript in
preparation).
Spayd SE, Nagel SW, Edwards CG. 1995. Yeast growth in Riesling
juice as a€ected vineyard nitrogen fertilization. Am J Enol Vitic
46:49±55.
Thomas DS, Hossack JA, Rose AH. 1978. Plasma-membrane lipid
composition and ethanol tolerance in Saccharomyces cerevisiae.
Arch Microbiol 117:239±245.
Thomas DS, Rose AH. 1979. Inhibitory e€ect of ethanol on growth
and solute accumulation by Saccharomyces cerevisiae as a€ected
by plasma-membrane lipid composition. Arch Microbiol 122:49±
55.
Thomas KC, Hynes SH, Ingledew WM. 1996. E€ect of nitrogen lim-
itation on synthesis of enzymes in Saccharomyces cerevisiae during
fermentation of high concentration of carbohydrates. Biotechnol
Lett 18:1165±1168.