International Journal of Food Microbiology 105 (2005) 333 345

www.elsevier.com/locate/ijfoodmicro

Predictive modelling and validation of Listeria innocua growth

at superatmospheric oxygen and carbon dioxide concentrations
S. Geysen a,*, B.E. Verlinden a, A.H. Geeraerd b, J.F. Van Impe b,
C.W. Michiels c, B.M. Nicola a
b

a
Flanders Centre/Laboratory of Postharvest Technology, Katholieke Universiteit Leuven, Willem de Croylaan 42, B-3001 Leuven, Belgium
BioTeC-Bioprocess Technology and Control, Department of Chemical Engineering, Katholieke Universiteit Leuven, Willem de Croylaan 46,
B-3001 Leuven, Belgium
c
Laboratory for Food Microbiology, Katholieke Universiteit Leuven, Kasteelpark Arenberg 23, B-3001 Leuven, Belgium

Received 18 December 2004; received in revised form 15 April 2005; accepted 21 April 2005

Abstract
The effect of superatmospheric oxygen and carbon dioxide concentrations on the growth of Listeria innocua, which was
used as a model organism for the pathogen Listeria monocytogenes, was evaluated. The bacteria were grown on a nutrient agar
surface at 7 8C. Three carbon dioxide levels (0%, 12.5% and 25%) were combined with different levels of high oxygen
concentrations (above 20%) based on a mixture design. The applied oxygen concentrations did not significantly influence the
growth. High CO2 concentrations, on the contrary, reduced the maximum specific growth rate and prolonged the lag time.
An overall model to describe the growth of L. innocua under high carbon dioxide conditions was constructed based on nine
growth experiments, using a weighted one-step regression procedure. The influence of carbon dioxide on lag time and
maximum specific growth rate was described using Ratkowsky-type models and inserted in the Baranyi equation. The
model described the growth very well. To assess the validity of the model, 14 additional experiments were carried out.
There was a good correlation of the model predictions and observed validation data.
D 2005 Elsevier B.V. All rights reserved.
Keywords: Listeria innocua; Superatmospheric oxygen; Superatmospheric carbon dioxide; Modified atmosphere; Pathogen; Predictive modelling; Model validation

1. Introduction
Consumers increasingly require fresh, natural food
products which are ready to use. Modified atmosphere
* Corresponding author. Tel.: +32 16 32 81 53; fax: +32 16 32 29
55.
E-mail address: Sabine.Geysen@biw.kuleuven.be (S. Geysen).
0168-1605/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijfoodmicro.2005.04.015

packaging (MAP) is a successful preservation technique which can fulfil this demand for a wide variety of
products, including raw and cooked meats, fish, fresh
pasta, fruit and vegetables (Phillips, 1996; Jayas and
Jeyamkondan, 2002). Generally, combinations of oxygen, carbon dioxide and nitrogen are used in MAP.
Carbon dioxide is the major anti-microbial factor of
MAP. Nitrogen is used to prevent pack collapse (Phil-

334

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

lips, 1996) and to replace oxygen in order to prevent

rancidity of meat and inhibit the growth of aerobic
organisms (Farber, 1991). Oxygen has several effects
on foods. Low levels are typically applied to reduce
respiration and the associated quality loss of vegetable
and fruit produce. High-oxygen modified atmosphere
packaging systems have been known to the meat industry for many decades (Jayas and Jeyamkondan, 2002).
They stabilize the cherry red colour of meat (large parts
or cuts) by maintaining the myoglobin in the oxygenated form (oxymyoglobin) (Gill, 1996). The application of high oxygen concentrations has been proposed
as an alternative to low-oxygen MAP for fruits and
vegetables (Day, 1996). Jacxsens et al. (2001b) stated
that high oxygen concentrations are particularly effective for ready-to-eat vegetables that are sensitive to
enzymatic browning and spoilage by yeasts.
MAP can thus be used to prolong the shelf-life and
to preserve the microbial and/or sensorial quality of
the product. Besides these positive effects, modified
atmosphere packaged foods can imply a potential
health risk. It is known that MAP can favour the
growth of some pathogens due to the elimination of
natural competitors and the associated prolongation of
the shelf life (Bennik et al., 1995; Hugas et al., 1998;
Francis et al., 1999). Pathogens that survive at low
temperatures and anoxic situations are of special interest (Francis et al., 1999). One of these pathogens is the
psychotrophic, facultative anaerobic bacterium Listeria monocytogenes, which can be found in a variety
of foods and causes listeriosis.
As compared to the numerous studies on the effect
of high carbon dioxidelow oxygen atmospheres, information about the effect of high-oxygen atmospheres
on Listeria growth is relatively scarce. When investigated on produce, the effect of high-oxygen atmosphere packaging often is controversial. In meat
products, high-oxygen atmosphere can have a different
effect on the growth of Listeria, depending on the type
of meat. This is possibly due to the difference in pH,
water activity and/or endogenous flora. For example,
Mano et al. (1995) have not detected the growth of
Listeria spp. in pork at either 1 8C or 7 8C, or in turkey
at 18C packaged in high-oxygen atmospheres (80% O2
plus 20% CO2 and 60% O2 plus 40% CO2), whereas
growth was observed in turkey under these atmospheres at 7 8C. Hugas et al. (1998) studied the effect
of 80% O2 plus 20% CO2 on raw minced pork, poultry

breasts and cooked pork. Inhibition of Listeria growth

could not be achieved, although the atmosphere had a
retarding influence depending on the type of meat. On
fresh processed mixed salads, packaging in a low or a
high barrier film with an initial oxygen concentration
of 95% did not affect the growth of L. monocytogenes
at 4 8C as compared to conventional MAP conditions
(Allende et al., 2002). In all conditions, the pathogen
only survived on the salads, without any growth.
Growth of L. monocytogenes was observed on the
green top of strawberry fruit which was high-oxygen
modified atmosphere packaged and stored at 7 8C
(Jacxsens et al., 2001a).
To avoid variations in gas atmosphere, in food
compositions and presence of endogenous microorganisms, high-oxygen modified atmosphere experiments with pure Listeria cultures on solid surface
laboratory medium have been performed (Ogihara et
al., 1993; Amanatidou and Smid, 1999; Jacxsens et
al., 2001b; Van der Steen et al., 2003). In all cases,
only a few gas combinations were tested and data
analysis was restricted to primary modelling. To our
knowledge, no models have been developed for the
effect of high oxygen and carbon dioxide atmospheres
on the growth of Listeria spp.
In the current study, Listeria innocua growth under
superatmospheric oxygen and carbon dioxide conditions was examined under laboratory conditions and a
predictive growth model was developed. A solid surface system was used, since many products are mainly
contaminated at the surface (Bennik et al., 1995).
Growth was studied at a temperature of 7 8C, being
the maximum allowed temperature for storage of
refrigerated products in Belgium (Anonymous,
1982). L. innocua was used as a model organism for
the pathogenic L. monocytogenes.

2. Materials and methods

2.1. Micro-organism and inoculation of agar plates
L. innocua Seeliger 1983 strain CIP 80.12 (isolated
from faeces), obtained from the Institut Pasteur (Paris,
France), was used.
Bacteria were stored at
80 8C in nutrient broth
(NB, Oxoid, Hampshire, England) supplemented with
25% glycerol. Two subsequent subcultures were

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

335

a temperature of 7 8C. Pure gases (oxygen, carbon

dioxide and nitrogen) were mixed using mass-flow
controllers (model 5850S, Brooks Instrument, The
Netherlands). The obtained gas mixtures were humidified by passage through a water bottle. The humidified gas was then introduced into a series of jars
containing the agar plates at a flow rate of 10 l/h. Gas
concentrations inside the jars were checked using a
micro-GC (Varian-Chrompack, Bergen op Zoom, The
Netherlands) and were found to be constant during the
experiment and identical in all jars within a series.
The tested gas conditions were selected using a
mixture design (Cornell, 1981). The principle of mixture theory is that the sum of the proportions of the
compounds is 1. In our case, the sum of the O2, CO2
and N2 proportions is equal to 1. Mixtures with three
compounds can be presented in a ternary diagram, the
corners of the triangle representing pure gases. In the
middle of the equilateral triangle, the three gases are
present in equal proportions. An experimental region
with the following constraints was chosen: CO2
between 0% and 25% and O2 between 20% and
100%, resulting in a subregion of the whole factor
space (see Fig. 1). Carbon dioxide concentrations
higher than 25% were not included because of their
detrimental effect on quality for the majority of vegetable products (Herner, 1987). A total of nine gas
combinations were selected as based on the mixture

grown in NB at 37 8C until stationary phase, for 24 h

and 21 h, respectively. The second subculture was then
kept at 7 8C for 4 h to allow cold adaptation. From the
second subculture, a dilution series was made in PPS
(Peptone Physiological Salt containing 8.5 g/l NaCl
and 1 g/l bacteriological peptone [L37, Oxoid]). 100 Al
from the appropriate dilution was spread on nutrient
agar plates with a diameter of 9 cm to obtain a population density of 103104 cfu cm
2 agar plate. Nutrient
agar was chosen for these experiments because it is
less nutritious than most other commonly used nonselective growth media such as tryptic soy and brain
heart infusion agar, to represent food products with
poor nutritional quality like vegetables. The growth
ability of the selected L. innocua strain on nutrient agar
at 7 8C under ambient atmosphere was confirmed in
preliminary tests (data not shown). Petri dishes with
ventilation ribs were used to ensure good gas
exchange.
2.2. Experimental set-up
Inoculated agar plates were incubated in glass jars
in a cool room set at 7 8C. Temperature was monitored
every 30 min with temperature loggers (Escort Junior,
Techninnovators, New Zealand). The actual temperature in the cool room was 6.7 F 0.3 8C. Inoculated
agar plates at room temperature needed 1.5 h to reach

O2
(100, 0, 0)
(87.5, 12.5, 0)
(75, 25, 0)
(60, 0, 40)

O2 between 20 and 100 %

(53.75, 12.5, 33.75)

(47.5, 25, 27.5)

(20, 25, 55)

CO2

(20, 0, 80)
(20, 12.5, 67.5)

N2
CO2 between 0 and 25 %

Fig. 1. Mixture design for the selection of tested gas conditions. The region of interest is marked in grey and is limited in oxygen between 20%
and 100% and carbon dioxide between 0% and 25%. The tested gas mixtures for identification of the models are marked with crosses (). The
corresponding gas concentrations are between brackets in the following order: % O2, % CO2, % N2.

336

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

design (% O2:% CO2:% N2): (100:0:0); (87.5:12.5:0);

(75:25:0); (60:0:40); (53.75:12.5:33.75); (47.5:25:
27.5); (20:0:80); (20:12.5:67.5); and (20:25:55) (Fig.
1). A mixture design was chosen because classic
designs such as, for example, a full factorial design,
are less suitable when applying oxygen concentrations
up to 100%. For example, an oxygen concentration of
90% can only be combined with carbon dioxide levels
of 10% or lower. In the present study, mixture theory
was only used to define a suitable design and was not
used in further evaluation of the data because nitrogen
has no effect on microbial growth, thus only leaving
oxygen and carbon dioxide to be considered.

Eqs. (1)(3) was used. The estimate for lag time (k)
was restricted to give a positive outcome. The base 10
logarithm of the cell numbers, N(t), is given by
N t N0

N1 t
N2 t


ln10 ln10

where



N1 t lmax t ln e
lmax t
e
lmax tk e
lmax k
2
and





N2 t ln 1 10N0
Nmax elmax t
k
e
lmax k :

2.3. Enumeration of viable cells

On each sampling time, three replicate plates per
gas condition were analysed. The content of each agar
plate was emptied aseptically in a stomacher bag. PPS
was added until a total volume of 100 ml was reached.
The sample then was homogenized in a masticator
(IUL Masticator Basic 0470, IUL S.A., Spain).
Appropriate dilutions were plated on nutrient agar
and colonies were counted after 48 h of incubation
at 37 8C. Eight to fourteen sampling times per gas
condition were chosen, based on the results of a
preliminary experiment (results not shown), in order
to obtain data in all stages of microbial growth. A
count of appropriate dilutions of the inoculum suspension used was carried out to calculate the initial
population density in the plates.
2.4. Model development
We used three steps to obtain a model for the
influence of time, oxygen and carbon dioxide on the
growth of L. innocua. The first two steps correspond
to the standard two-step method that is mostly used in
predictive microbiology, as described in, for example,
Whiting (1995). For primary modelling, the growth
data were analysed with the currently widely used
Baranyi equation (Baranyi and Roberts, 1994). To
fit the model to the data, a program was written in
MatlabR (The Mathworks, Inc., Natick, USA) using
the Optimization Toolbox. A reparameterised model
as it is applied in the program MicroFit version 1.0
(developed by the Institute of Food Research, http://
www.ifrn.bbsrc.ac.uk/MicroFit/) and described in

3
The parameters of the model are N 0 and N max (the
base 10 logarithm of the initial and final cell numbers
in log(cfu cm
1)), l max (the specific growth rate in
h
1) and k (the lag time in h). Note that the cell
numbers are given in base 10 logarithms, but that
l max and k are estimated as natural logarithms.
The parameters and their corresponding approximate 95% confidence limits were estimated. In the
second stage, the estimates for l max and k were fitted
by equations describing the effect of the gas condition
on l max or k. The variances of l max and k were
analysed according to Zwietering et al. (1994) and
were found to be independent of the magnitude of
l max and k. Consequently, no variance-stabilising
transformations were applied.
The estimates for l max were fitted by a Ratkowskytype model (Ratkowsky et al., 1982) (Eq. (4)) which
describes the effect of CO2 on l max.
lmax mCO2

max

CO2 :

The constants m and CO2 max are regression

parameters.
The effect of CO2 on the lag phase was also
described by a Ratkowsky-type model (Eq. (5)).
p
k l CO2
CO2 min
5
The constants l and CO2 min are regression parameters.
According to Ross (1993), the parameters CO2 max
and CO2 min are the dnotionalT maximum and minimum CO2 concentration. This means that they do not
have to be interpreted as a dtrueT maximum and mini-

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

mum, but as the CO2 concentration at which the given

parameter (l max or k) becomes 0.
After identification of the secondary models, the
expressions for l max and k were inserted in the Baranyi equation, which results in an doverall modelT as it
will be called below. The insertion of expression (4)
for l max and (5) for k in the Baranyi equation (Eq. (1))
results in the overall model. In this way, we obtain a
non-linear equation with six parameters.
These parameters were re-estimated, using the
parameter estimates of the inserted secondary models
as the initial starting point for the non-linear search
procedure. This one-step regression is applied to
prevent the accumulation of fitting errors (see, for
example, Willocx, 1995; Fernandez et al., 2002;
Valdramidis et al., 2005). The precision of the estimates is increased because the method uses all the
raw growth data, whereas a standard secondary
model is built on estimates resulting from primary
modelling. The one-step regression method should,
in general, result in a minimum overall deviation
between experimental and predicted values.
The application of one-step regression assumes that
the underlying errors have homogeneous variances.
To determine the variation of the errors of the N data
as a function of the gas condition, a Levenes test
(Levene, 1960) on the residuals resulting from the fit
of Eq. (1) was performed. When errors are found to
have non-homogeneous variances, a weighted sum of
squared errors (WSSE) criterion needs to be implemented to fit the overall model equation on the data.
WSSE

n
X
i1

wi

mi
X

logNoverall xi ; tij
logNijcount 2

j1

6
with
wi

1
MSEi

n is the number of gas conditions, m i is the number of

measurement points of the ith gas condition. N overall
(x i ,t ij ) is the bacterial count at the ith specific gas
condition (x i ) and time (t ij ) predicted with the overall
model and N ijcount is the observed bacterial count at the
ith specific gas condition and time t ij . MSEi is the
mean squared error resulting from the fit of Eq. (1) on
the data of the ith specific gas condition.

337

For linear regression, the GLM procedure of the

software package SAS/STATR version 8.2 (SAS
Institute Inc., Cary, NC, United States) was applied.
The (weighted) least-squares optimisation is carried
out using the Matlab least-squares non-linear optimisation routine (lsqnonlin) with the LevenbergMarquardt method. For primary as well as for overall
modelling, the solution for lag time was restricted to
a positive value. The adjusted R 2 and Root Mean
Squared Error (RMSE) were calculated. To assess
normality of the residual values, normal probability
plots were made and evaluated in SAS/STAT.
2.5. Validation of the model
To assess the validity of the overall model, two
additional sets of validation experiments were carried
out at gas conditions in the region of interest. The first
set of five validation experiments was done with an
inoculum density of 3.5 log cfu cm
2. The tested gas
conditions were (% O2:% CO2:% N2): (87.5:12.5:0);
(75:25:0); (53.75:12.5:33.75); (47.5:25:27.5); and
(20:25:55). The second set consisted of nine validation experiments, where the inoculum density was 4.6
log cfu cm
2. The gas conditions were identical to
those used for model identification. The experimental
method was the same as explained previously.
The validation of the overall model was quantified
by determining, firstly, the adjusted R 2, the RMSE (in
order to use the same measures as during model
identification) and, secondly, the Bias and Accuracy
factors. The latter were calculated in two different
ways. Firstly, they were calculated using the natural
logarithm of l max as described by Baranyi et al.
(1999) and are shown in Eq. (8) (Accuracy factor
A l ) and Eq. (9) (Bias factor B l ).
"s #



2
Pn


i1 ln lmax; overall xi
ln lmax; primary xi
Al exp
n

8
" Pn


i1 ln lmax;

Bl exp

overall xi

ln lmax;

 #

primary xi

9
n is the number of validation gas conditions,
l max, overall (x i ) is the maximum specific growth
rate at the ith specific validation gas condition (x i )

338

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

predicted with the overall model and l max, primary (x i )

is the maximum specific growth rate at the ith
specific gas condition (x i ) resulting from the fit of
the primary model to the validation data.
Secondly, the Bias and Accuracy factors were calculated on all count data (not averages) as shown in
Eqs. (10) and (11).
Acount

Cholesky decomposition matrix with a vector from

which each element is randomly chosen from a standard normal distribution yields a vector of deviation.
Summing the deviation vector with the mean estimated
parameter vector yields a new parameter vector. This
was repeated 5000 times yielding a parameter vector

"s
#
Pn Pmi
count 2

j1 logNoverall xi ; tij
logNij

i1
Pn
exp
i1 mi

(a)
7

10
Bcount exp

i1

Pmi

j1

logNoverall xi ; tij
logNijcount

Pn
i1 mi

11
5

n is the number of validation gas conditions, m i is the

number of measurement points of the ith validation
gas condition, N overall (x i ,t ij ) is the bacterial count at a
specific gas condition (x i ) and time (t ij ) predicted with
the overall model (using 3.5 log cfu cm
2 and 4.6 log
cfu cm
2 as the value for N 0 for the first and second
set of validation experiments, respectively) and N ijcount
is the observed bacterial count at the ith specific gas
condition (x i ) and time (t ij ).
The percent discrepancy (%D) and percent bias
(%B) are calculated as described in Baranyi et al.
(1999) as shown in Eqs. (12) and (13), respectively.

(b)
7

Log cfu cm-2

" Pn

6
5
4

%D A
1100%

12

%B sgnLnBexpjLnBj
1100%

13

Sgn(LnB) = +1 if LnB N 0, denoting that the model

predicts faster growth than the observations (fail
safe), 0 if LnB = 0, denoting a perfect prediction of
the data, and
1 if LnB b 0, denoting that the model
predicts slower growth than the observations (fail
dangerous).
Percentages discrepancy and bias calculated using
A count and B count will be denoted by %D count and
%B count. %D and %B calculated using A l and B l
will be denoted by %D l and %B l .
The 95% confidence limits of the overall model
were estimated for each validation gas condition
using a Monte Carlo procedure written in MatlabR 6
(Rubinstein, 1981). To generate samples of a vector of
correlated Gaussian random parameters, a Cholesky
decomposition of the variancecovariance matrix of
the fitted parameters was carried out. Multiplying the

(c)
7
6
5
4

100 200 300 400 500 600 700

Time (hours)

Fig. 2. Growth of L. innocua at 7 8C under atmospheres with 0%

CO2 (a), 12.5% CO2 (b) and 25% CO2 (c). For (ac), 5 and dotted
lines denote average bacteria counts and fitted Baranyi curves at
oxygen concentrations of 100%, 87.5% and 75%, respectively. 
and dashed lines denote average bacteria counts and Baranyi curves
at 60%, 53.75% and 47.5% O2, respectively. Average counts and
fitted curves at 20% oxygen are marked with E and solid lines.

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

sample with the same covariance structure as the fitted

parameter values. The model equation was calculated
in 100 time points per inserted parameter set resulting
in a 5000  100 matrix for each of the validation gas
conditions. At this stage also, the variation of the
measurements was included. To do this, 5000  100
measurement errors were randomly sampled from a
normal distribution with mean zero and a variance
equal to the mean squared error of the model fit and
added. Per time point, the calculated values were
sorted. The lower and higher 95% prediction limit,
respectively, of each time point was determined by
the 126th and 4875th element of the sorted row. The
interval as we calculated it here is the 95% confidence
interval for a new prediction or briefly the 95% prediction interval, because both the confidence limits for the
parameters and the variation of the measurements were
included.

339

trend was not statistically significant. The time to reach

N max was mainly influenced by the carbon dioxide
concentration in the gas mixture. In the absence of
CO2, N max was reached after approximately 150 h
(Fig. 2a). Bacteria stored under a gas atmosphere
with 12.5% CO2 reached the maximum population
density after approximately 200 h (Fig. 2b). At atmospheres with 25% CO2, the bacteria needed 300 h up to
600 h to reach N max (Fig. 2c).
The observed retardation at elevated carbon dio
xide concentrations was due to a prolongation of the
lag time combined with a decreased maximum specific growth rate. Lag time (k), maximum specific
growth rate (l max), maximum population density
(N max) and their corresponding 95% confidence limits
were estimated using the Baranyi equation and are
listed together with the values for MSE in Table 1.
3.2. Secondary growth models

3. Results
3.1. Primary growth models
Growth curves for L. innocua were obtained for all
nine gas combinations (Fig. 2). Oxygen and carbon
dioxide concentration did not significantly influence
the population density in the stationary phase (N max)
which was between 6.2 and 7.4 log cfu cm
2 for all
conditions. Although in the gas conditions with 0%
CO2 (Fig. 2a) there seems to be an influence of the gas
atmosphere on the maximum population density, this

First, a linear regression was performed on the

estimates for l max and k that were obtained through
primary modelling. Carbon dioxide was found to have
a significant effect on both growth parameters, whereas
oxygen concentrations did not. Subsequently, secondary models to describe the effect of carbon dioxide on
l max and k were defined (respectively, Eqs. (4) and
(5)). The parameter estimates for both secondary
models are given in Table 2. The adjusted R 2 and
MSE for the lag time model were 0.87 and 310.9,
respectively. For the l max model, the goodness-of-fit
parameters equalled 0.78 (R 2adj) and 0.0001 (MSE). A

Table 1
Estimated parameters and corresponding approximate 95% confidence limits of lag time (k), maximum specific growth rate (l max) and
maximum population density (N max) for the growth of L. innocua at 7 8C under different high-oxygen atmospheres, according to the Baranyi
model (Baranyi and Roberts, 1994)
Oxygen
(%)

Carbon
dioxide (%)

k (h)

l max (h
1)

N max
(log cfu cm
2)

MSE

20
60
100
20
53.75
87.5
20
47.5
75

0
0
0
12.5
12.5
12.5
25
25
25

0F8
2 F 29
0F8
57 F 18
74 F 21
12 F 25
61 F 58
67 F 46
56 F 58

0.063 F 0.005
0.055 F 0.019
0.066 F 0.006
0.057 F 0.010
0.056 F 0.012
0.034 F 0.004
0.025 F 0.007
0.019 F 0.003
0.018 F 0.003

7.43 F 0.06
6.22 F 0.17
6.79 F 0.06
6.99 F 0.41
6.50 F 0.09
6.61 F 0.07
6.35 F 0.14
6.69 F 0.16
6.39 F 0.16

0.011
0.090
0.011
0.060
0.052
0.022
0.093
0.054
0.058

The mean squared error (MSE) for each of the gas conditions used to calculate the weights in the overall model fitting procedure are also given.

340

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

Table 2
Parameter estimates and corresponding approximate 95% confidence intervals for the secondary model and the overall model
Parameter

Secondary model estimates F approximate

95% confidence limits

Overall model estimates F approximate

95% confidence limits

N 0 (log cfu cm
2)

N max (log cfu cm
2)
l (h2/%)
CO2 min (%)
m (l/h%)
CO2 max (%)

3.69 F 0.02
6.85 F 0.01
118 F 15

0.007 F 0.04
0.00204 F 0.00005
31.5 F 0.2

160 F 99
0.0 F 0.2
0.0016 F 0.0008
39 F 14

graphical representation of the models and parameter

estimates is shown in Fig. 3.
3.3. Overall model
Looking back to Fig. 2, it is apparent that the variation on the experimental data seems to be dependent on
the gas condition applied (as also quantified through
the MSE values reported in Table 1). This was confirmed by a Levenes test, and consequently a weighted
least squares optimisation procedure was carried out.
The growth parameters N 0 and N max were considered
independent of the oxygen and carbon dioxide concentration and were initialised at 3.5 and 6.5 log cfu cm
2,
respectively. For the other parameters, the obtained
parameter estimates from secondary modelling were
80

60
40
20
0
0.16

used to initialise the non-linear least-squares search

procedure for the overall model. In Table 2, the parameter estimates for the overall model are given. The
adjusted R 2 was 0.92. The RMSE was 0.34. The residuals were distributed around zero and the normal
probability plots did not show any abnormalities
(data not shown).
3.4. Validation of the model
The overall model was validated using the results of
two sets of validation experiments. The adjusted R 2
(R 2adj), RMSE and the Bias and Accuracy factors are
given in Table 3. R 2adj was equal to 0.81. The RMSE
value was 0.52. The small % B count (
2%) means that
the observed counts were close to the model predictions. B count was slightly smaller than 1 (or % B count
was negative), which means that the overall model
underestimated the growth of L. innocua. However,
even if data points do not perfectly match the model
prediction (which has to be interpreted as the average
response of the bacterial cells), they can still be situated
in the 95% prediction region of the model. The upper
and lower 95% prediction boundaries and measured
values for the two validation experiments are represented in Figs. 4 and 5, respectively. Of a total of 382

max

0.12
Table 3
Validation parameters of the overall model for growth of L. innocua
at 7 8C

0.08
0.04
0
0

10

20

30

CO2 concentration
Fig. 3. Secondary growth models and observed values for l max and
k used for model identification.

Validation parameter

Value

R 2adjusted

0.81
0.52
0.98
1.68
0.70
1.98

RMSE
B count (% B count)
A count (% D count)
B l (% B l )
A l (% D l )

(
2%)
(68%)
(
43%)
(98%)

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

Log cfu/cm2

10
9
8
7
6
5
4
3

53.75% O + 12.5% CO

87.5% O + 12.5% CO

9
8
7
6
5
4
3

20% O + 25% CO

47.5% O + 25% CO

9
8
7
6
5
4
3

75% O2 + 25% CO2

100

200

300

100

200

341

300

400

500

600

Time (hours)

400

500

600

Time (hours)
Fig. 4. Results of the first set of validation experiments shown per gas condition (circles). The inoculum density was 3.5 log cfu cm
2. The lines
represent estimated lower and upper 95% prediction limits for the overall model as calculated using a Monte Carlo simulation.

validation points, 51 points (13%) fall outside the 95%

prediction region.
3.5. Application of the model
The overall growth model for L. innocua was combined with a growth model for the spoilage bacterium
Pseudomonas fluorescens at superatmospheric oxygen
and carbon dioxide concentrations (Geysen et al.,
2005) to calculate a drisk areaT. dRisk areasT were
defined by Devlieghere et al. (2001) as combinations
of factors at which food pathogens can develop to an
unacceptable level before spoilage occurs. As a theoretical example, a risk area will be calculated for freshcut lettuce. According to Nguyen-the and Carlin
(1994), it was assumed that the initial contamination
level for P. fluorescens is 103 cfu g
1 and decay related
with spoilage occurs at a concentration of 107 cfu g
1.

Initial contamination of L. innocua was assumed to be

1 cfu g
1 and was allowed to reach 102 cfu g
1 at the
end of the shelf life as proposed by several authors
(e.g., Nguyen-the and Carlin, 1994; Chen et al., 2003)
and according to guideline SANCO/4198/2001 Rev. 15
of the Commission of the European Communities and
for L. monocytogenes. The resulting drisk areaT plot is
shown in Fig. 6. The combinations of oxygen and
carbon dioxide that imply a potential health risk are
represented in grey.

4. Discussion
The data presented in this paper show that high
carbon dioxide concentrations had a retarding effect
on L. innocua growth at 7 8C. No effect of superatmospheric oxygen concentrations was found on the

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

Log cfu/cm2

342

10
9
8
7
6
5
4
3
10
9
8
7
6
5
4
3
10
9
8
7
6
5
4
3
0

20% O2 + 0% CO2

20% O2 + 12.5% CO2

20% O2 + 25% CO2

60% O2 + 0% CO2

53.75% O2 + 12.5% CO2

47.5% O2 + 25% CO2

100% O2 + 0% CO2

87.5% O2 + 12.5% CO2

75% O2 +Time(hours)
25% CO2

100 200 300 400 500 600

100 200 300 400 500 600

100 200 300 400 500 600

Time (hours)
Fig. 5. Results of the second set of validation experiments shown per gas condition (circles). The inoculum density was 4.6 log cfu cm
2. The
lines represent estimated lower and upper 95% prediction limits for the overall model as calculated using a Monte Carlo simulation.

growth of L. innocua. These results were in agreement

with the results of Ogihara et al. (1993) and Day (2001)
who found that in vitro growth of L. monocytogenes
was inhibited due to high levels of carbon dioxide and
not of high oxygen levels in the applied gas mixtures.
In our experiments, the effect of CO2 on L. innocua
was a prolongation of the lag time and a reduction of the
maximum specific growth rate. Bacteria kept at gas
atmospheres without carbon dioxide did not have a
significant lag phase, whereas under gas atmospheres
with 25% CO2 lag times of approximately 60 h were
found. The maximum specific growth rate when no
carbon dioxide was applied was 0.06 h
1 and
decreased to 0.02 h
1 with application of 25% CO2.
The maximum population density was not affected by
the carbon dioxide levels used in our experiments.
Different authors previously examined the effect of
carbon dioxide (in combination with other factors as,

e.g., pH and temperature) on L. monocytogenes in

laboratory medium (Bennik et al., 1995; Farber et
al., 1996; Fernandez et al., 1997; Devlieghere et al.,
2001). Bennik et al. (1995) found a decrease of N max
of 0.68 log cfu cm
2 maximally at 50% CO2, whereas
Farber et al. (1996) and Fernandez et al. (1997) did
not find an effect on N max of carbon dioxide concentrations up to 90% and 100%. This is in agreement
with our results where no effect on the maximum
population density was found.
Published studies on the effect of carbon dioxide
on the lag time of Listeria show inconsistencies.
Farber et al. (1996) and Devlieghere et al. (2001)
reported a prolonged lag time when applying carbon
dioxide. However, Bennik et al. (1995) found no
effect of CO2 on the lag time of L. monocytogenes.
As lag times are not only influenced by the environmental conditions during the experiment, but also by

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

Fig. 6. L. innocua risk area when assuming an initial and final

contamination of P. fluorescens of 3 and 7 log cfu g
1, respectively.
L. innocua initial and final contamination was 1 and 100 cfu g
1,
respectively. At combinations of O2 and CO2 situated in the shaded
region, L. innocua levels were higher than 100 cfu g
1 at the end of
the shelf life.

the physiological status of the cells at the start of the

experiment, the inconsistencies in lag time are probably caused by a different history of the cells, due to
different growth conditions of the inoculum (Swinnen
et al., 2004).
The most consistent effect of carbon dioxide on
Listeria is the effect on the maximum specific growth
rate. In comparison to earlier reported influence of
CO2 on l max of L. monocytogenes (Bennik et al.,
1995; Farber et al., 1996; Fernandez et al., 1997;
Devlieghere et al., 2001), the effect on L. innocua
in our experiments was stronger. Whereas the addition
of 25% CO2 resulted in more than halving the l max in
our experiments, the same effect was only accomplished with 50% CO2 or more based on the results
described in the literature. This difference can be due
to the fact that we used a relatively poor growth
medium in comparison to the other authors (Nutrient
Agar instead of Brain Heart Infusion Agar/Broth or
Tryptic Soy Broth) or because of a difference in
response between L. innocua and L. monocytogenes.
However, several studies have demonstrated that the
behaviour of both organisms is comparable as affected
by temperature, acidification and modified atmosphere (Hugas et al., 1998; Thomas et al., 1999).
Begot et al. (1997) examined differences among 58
strains of L. monocytogenes and 8 strains of L. inno-

343

cua. Large variations in lag times were found between

the strains within one species, whereas the variations
in generation times were less pronounced.
In the current study, an overall model for L. innocua growth under high carbon dioxide modified atmospheres at 7 8C was built. To construct the model, a
done-step regressionT approach was followed in which
the data were weighted according to the variances on
the error of the primary models. This approach
yielded more precise estimates of the parameters by
using all growth data without applying unnecessary
intermediate steps. Indeed, approximate 95% confidence intervals of the lag phases resulting from primary modelling were rather large (see Table 1). Also,
a high correlation between k and l max could be seen
from the correlation matrices (results not shown).
These were indications that k could not be estimated
well per experiment separately. In this case, the uncertainty on the parameter estimates was expected to
decrease when applying a one-step regression procedure where all the information over the different gas
conditions was considered in one single optimisation
procedure. From Table 2, it can indeed be seen that
the approximate 95% confidence intervals on the
estimates of the overall model are largely reduced as
compared to the secondary model. The overall model
had a good predictive quality based on goodness-of-fit
parameters. Secondly, validation experiments were
carried out. The overall model slightly underestimated
the bacterial growth. This was reflected by a Bias
factor smaller than one. A Monte Carlo simulation
was carried out in order to estimate the 95% prediction limits of the model. While one would expect 5%
of the validation points to fall outside this prediction
region, we found a higher ratio of 13% of the validation points outside the prediction region. A possible
explanation is that validation data points obtained at
different time points from a single experiment are not
totally independent, an assumption which was made
for the Monte Carlo simulation of the prediction
boundaries. Once a growth curve is evolving somewhat slower or somewhat faster, it will keep on evolving slower or faster for the remaining of the
experiment. Also, it has to be stressed that the validation was a primary validation. This validation gives
the primary bias and primary error of the model and is
a first step in the validation of a model according to
Pin et al. (1999). For spoilage micro-organisms, the

344

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

next steps are validation tests in inoculated sterile

foods and in foods with the natural microflora (Pin
et al., 1999). For pathogenic micro-organisms, like L.
innocua as a model organism for L. monocytogenes,
an additional intermediate step exists, namely, validation with inoculated pathogens in foods with natural
(spoilage) microflora (Miconnet et al., in press).
A drisk areaT was calculated by combining the
overall growth model for L. innocua with the growth
model for P. fluorescens under high oxygen and carbon dioxide conditions developed by Geysen et al.
(2005). From Fig. 6, it can be seen that the application
of oxygen concentrations above 40% imply a potential health risk when combined with carbon dioxide
concentrations below 20%. Furthermore, even when
applying oxygen concentrations below 40%, at least
2% carbon dioxide should be added to the atmosphere
in order to provide a safe storage atmosphere. This is
due to the fact that the growth of P. fluorescens was
retarded by both oxygen and carbon dioxide, whereas
only carbon dioxide had an effect on the growth of L.
innocua. Safe application of oxygen concentrations
up to 60% is possible when combined with 25% CO2.
The authors wish to stress that in vitro models were
used to calculate the risk area. Also, interactions
between the studied micro-organisms were not
included. Hence, the risk area should be interpreted
as a theoretical example and may not be transferred to
practical situations as such.

5. Conclusions
The effect of superatmospheric oxygen and carbon
dioxide concentrations on the growth of L. innocua
was evaluated. Bacterial growth was not significantly
influenced by high oxygen concentrations. Carbon
dioxide had a prolonging effect on lag time and
reduced the maximum specific growth rate.
An overall model to describe growth of L. innocua
under high carbon dioxide conditions was constructed
using a one-step regression procedure. Here, the influence of carbon dioxide on lag time and maximum
specific growth rate was described using a Ratkowsky-type model inserted in the Baranyi equation.
Primary validation of the model was carried out.
There was a good correlation of the model predictions
and observed growth data.

Acknowledgements
This research has been supported by the Belgian
Ministry of Small Enterprises, Traders and Agriculture as part of the project S-6056 and by the Institute
for the Promotion of Innovation by Science and Technology in Flanders (IWT-Vlaanderen, project CO020803). A. Geeraerd is a Postdoctoral Fellow with
the Fund for Scientific Research-Flanders (Belgium)
(F.W.O.-Vlaanderen).

References
Allende, A., Jacxsens, L., Devlieghere, F., Debevere, J., Artes, F.,
2002. Effect of superatmospheric oxygen packaging on sensorial quality, spoilage, and Listeria monocytogenes and Aeromonas caviae growth in fresh processed mixed salads. Journal of
Food Protection 65, 1565 1573.
Amanatidou, A., Smid, E.J.G.L.G.M., 1999. Effect of elevated
oxygen and carbon dioxide on the surface growth of vegetable-associated micro-organisms. Journal of Applied Microbiology 86, 429 438.
Anonymous, 1982. Koninklijk besluit van 4 februari 1980 betreffende het in de handel brengen van te koelen voedingsmiddelen.
De warenwetgeving. Verzamelingen van reglementen betreffende voedingsmiddelen en andere consumptieproducten. Die
Keure, Brugge, Belgium, pp. 13 14.
Baranyi, J., Roberts, T.A., 1994. A dynamic approach to predicting
bacterial growth in food. International Journal of Food Microbiology 23, 277 294.
Baranyi, J., Pin, C., Ross, T., 1999. Validating and comparing
predictive models. International Journal of Food Microbiology
48, 159 166.
Begot, C., Lebert, L., Lebert, A., 1997. Variability of the response of
66 Listeria monocytogenes and Listeria innocua strains to
different growth conditions. Food Microbiology 14, 403 412.
Bennik, M.H.J., Smid, E.J., Rombouts, F.M., Gorris, L.G.M., 1995.
Growth of psychrotrophic foodborne pathogens in a solid surface model system under the influence of carbon dioxide and
oxygen. Food Microbiology 12, 509 519.
Chen, Y., Ross, W.H., Scott, V.N., Gombas, D.E., 2003. Listeria
monocytogenes: low levels equal low risk. Journal of Food
Protection 66 (4), 570 577.
Cornell, J.A., 1981. Experiments with Mixtures. Designs, Models
and the Analysis of Mixture Data. John Wiley and Sons, New
York, p. 305.
Day, B.P.F., 1996. High oxygen modified atmosphere packaging for
fresh prepared produce. Postharvest News and Information 7,
31 34.
Day, B.P.F., 2001. Fresh prepared produce: GMP for high oxygen
MAP and non-sulphite dipping. Guidance 31 (75 pp.).
Devlieghere, F., Geeraerd, A.H., Versyck, K.J., Vandewaetere, B.,
Van Impe, J., Debevere, J., 2001. Growth of Listeria monocy-

S. Geysen et al. / International Journal of Food Microbiology 105 (2005) 333345

togenes in modified atmosphere packed cooked meat products:
a predictive model. Food Microbiology 18, 53 66.
Farber, J.M., 1991. Microbiological aspects of modified-atmosphere
packaging technologya review. Journal of Food Protection 54,
58 70.
Farber, J.M., Cai, Y., Ross, W.H., 1996. Predictive modeling of the
growth of Listeria monocytogenes in CO2 environments. International Journal of Food Microbiology 32, 133 144.
Fernandez, P.S., George, S.M., Sills, C.C., Peck, M.W., 1997.
Predictive model of the effect of CO2, pH, temperature and
NaCl on the growth of Listeria monocytogenes. International
Journal of Food Microbiology 37, 37 45.
Fernandez, A., Collado, J., Cunha, L.M., Ocio, M.J., Martnez, A.,
2002. Empirical model building based on Weibull distribution to
describe the joint effect of pH and temperature on the thermal
resistance of Bacillus cereus in vegetative substrate. International Journal of Food Microbiology 77, 147 153.
Francis, G.A., Thomas, C., OBeirne, D., 1999. The microbiological
safety of minimally processed vegetables. International Journal
of Food Science and Technology 34, 1 22.
Geysen, S., Geeraerd, A.H., Verlinden, B.E., Michiels, C.W., Van
Impe, J.F., Nicola, B.M., 2005. Predictive modelling and validation of Pseudomonas fluorescens growth at superatmospheric
oxygen and carbon dioxide concentrations. Food Microbiology
22, 149 158.
Gill, C.O., 1996. Extending the storage life of raw chilled meats.
Meat Science 43, S99 S109.
Herner, R.C., 1987. High CO2 effects on plant organs. In: Weichmann, J. (Ed.), Postharvest Physiology of Vegetables. Marcel
Dekker, Inc., New York, USA, pp. 239 253.
Hugas, M., Pages, F., Garriga, M., Monfort, J.M., 1998. Application
of the bacteriocinogenic Lactobacillus sakei CTC494 to prevent
growth of Listeria in fresh and cooked meat products packed
with different atmospheres. Food Microbiology 15, 639 650.
Jacxsens, L., Devlieghere, F., Debevere, J., 2001a. Survival of acidresistant pathogens on packaged strawberry fruits. Poster presentation at the Sixth Conference on Food Microbiology. Lie`ge
University, Belgium.
Jacxsens, L., Devlieghere, F., Van der Steen, C., Debevere, J.,
2001b. Effect of high oxygen modified atmosphere packaging
on microbial growth and sensorial qualities of fresh-cut produce.
International Journal of Food Microbiology 71, 197 210.
Jayas, D.S., Jeyamkondan, S., 2002. Modified atmosphere storage
of grains meats fruits and vegetables. Biosystems Engineering
82, 235 251.
Levene, H., 1960. Robust tests for equality of variances. In: Olkin,
I. (Ed.), Contributions to Probability and Statistics. Stanford
University Press, California, USA, pp. 278 292.
Mano, S.B., Garcia De Fernando, G.D., Lopez-Galvez, D., Selgas,
M.D., Garcia, M.L., Cambero, M.I., Ordonez, J.A., 1995.
Growth/survival of natural flora and Listeria monocytogenes
on refrigerated uncooked pork and turkey packaged under mo
dified atmospheres. Journal of Food Safety 15, 305 319.
Miconnet, N., Geeraerd, A.H., Van Impe, J.F., Rosso, L., Cornu, M.,
in press. Reflections on the use of robust and least-squares non-

345

linear regression to model challenge tests conducted in/on food

products. International Journal of Food Microbiology.
Nguyen-the, C., Carlin, F., 1994. The microbiology of minimally
processed fresh fruits and vegetables. Critical Reviews in Food
Science and Nutrition 34 (4), 371 401.
Ogihara, H., Kanie, M., Yano, N., Haruta, M., 1993. Effect of
carbon-dioxide, oxygen, and their gas-mixture on the growth
of some food-borne pathogens and spoilage bacteria in modified
atmosphere package of food. Journal of Food Hygienic Society
of Japan 34, 283 288.
Phillips, C.A., 1996. Review: modified atmosphere packaging and
its effects on the microbiological quality and safety of produce. International Journal of Food Science and Technology 31,
463 479.
Pin, C., Sutherland, J.P., Baranyi, J., 1999. Validating predictive
models of food spoilage organisms. Journal of Applied Microbiology 87, 491 499.
Ratkowsky, D.A., Olley, J., McMeekin, T.A., Ball, A., 1982. Relationship between temperature and growth rate of bacterial cultures. Journal of Bacteriology 149 (1), 1 5.
Ross, T., 1993. Belehra`dek-type models. Journal of Industrial
Microbiology 12, 180 189.
Rubinstein, R.Y., 1981. Simulation and the Monte Carlo Method.
John Wiley and Sons, New York, p. 278.
Swinnen, I.A.M., Bernaerts, K., Dens, E.J.J., Geeraerd, A.H., Van
Impe, J.F., 2004. Predictive modelling of the microbial lag
phase: a review. International Journal of Food Microbiology
94 (2), 137 159.
Thomas, C., Prior, O., OBeirne, D., 1999. Survival and growth of
Listeria species in a model ready-to-use vegetable product
containing raw and cooked ingredients as affected by storage
temperature and acidification. International Journal of Food
Science and Technology 34, 317 324.
Valdramidis, V.P., Belaubre, N., Zuniga, R., Foster, A.M., Havet,
M., Geeraerd, A.H., Swain, M.J., Bernaerts, K., Van Impe, J.F.,
Kondjoyan, A., 2005. Development of predictive modelling
approaches for surface temperature and associated microbiological inactivation during hot air decontamination. International
Journal of Food Microbiology 100, 261 274.
Van der Steen, C., Devlieghere, F., Debevere, J., 2003. High oxygen
concentration in combination with elevated carbon dioxide to
affect growth of fresh-cut produce micro-organisms. In: Verlinden, B.E., Nicola, B.M., De Baerdemaeker, J. (Eds.), Acta
Horticulturae: Proceedings of the International Conference Postharvest Unlimited. ISHS, Leuven, Belgium, pp. 141 147.
Whiting, R.C., 1995. Microbial modeling in foods. Critical Reviews
in Food Science and Nutrition 35, 467 494.
Willocx, F., 1995. Evolution of microbial and visual quality of
minimally processed foods: a case study on the product life
cycle of cut endive. PhD thesis, Katolieke Universiteit Leuven,
p. 228.
Zwietering, M.H., Cuppers, H.G.A.M., De Wit, J.C., Van t Riet, K.,
1994. Evaluation of data transformations and validation of a
model for the effect of temperature on bacterial growth. Applied
and Environmental Microbiology 60, 195 203.