You are on page 1of 15

International Journal of Food Microbiology 99 (2005) 157 – 171

www.elsevier.com/locate/ijfoodmicro

The mathematical properties of the quasi-chemical model for


microorganism growth–death kinetics in foods
E.W. Ross, I.A. Taub, C.J. Doona*, F.E. Feeherry, K. Kustin
US Army RDECOM, Natick Soldier Center, Combat Feeding Directorate, Combat Feeding Innovative Science Team, Kansas St.,
Natick, MA 01760-5018, USA
Received 16 March 2004; received in revised form 5 July 2004; accepted 26 July 2004

Abstract

Knowledge of the mathematical properties of the quasi-chemical model [Taub, Feeherry, Ross, Kustin, Doona, 2003. A
quasi-chemical kinetics model for the growth and death of Staphylococcus aureus in intermediate moisture bread. J. Food Sci.
68 (8), 2530–2537], which is used to characterize and predict microbial growth–death kinetics in foods, is important for its
applications in predictive microbiology. The model consists of a system of four ordinary differential equations (ODEs), which
govern the temporal dependence of the bacterial life cycle (the lag, exponential growth, stationary, and death phases,
respectively). The ODE system derives from a hypothetical four-step reaction scheme that postulates the activity of a critical
intermediate as an antagonist to growth (perhaps through a quorum sensing biomechanism). The general behavior of the
solutions to the ODEs is illustrated by several examples. In instances when explicit mathematical solutions to these ODEs are
not obtainable, mathematical approximations are used to find solutions that are helpful in evaluating growth in the early stages
and again near the end of the process. Useful solutions for the ODE system are also obtained in the case where the rate of
antagonist formation is small. The examples and the approximate solutions provide guidance in the parameter estimation that
must be done when fitting the model to data. The general behavior of the solutions is illustrated by examples, and the MATLAB
programs with worked examples are included in the appendices for use by predictive microbiologists for data collected
independently.
D 2004 Elsevier B.V. All rights reserved.

Keywords: Predictive food microbiology; Quasi-chemical model; Chemical kinetics; Differential equation system; Staphylococcus aureus;
Escherichia coli; Hurdles; High pressure processing

1. Introduction

The quasi-chemical model for microbial growth–


* Corresponding author. Tel.: +1 508 233 5083; fax: +1 508
233 5200. death kinetics (Taub et al., 2003) is a unique
E-mail address: christopher.doona@natick.army.mil mathematical tool for predicting microbial kinetics
(C.J. Doona). in food substrates. The model was developed in
0168-1605/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijfoodmicro.2004.07.019
158 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

concert with the study of the growth kinetics of a models based on chemical reaction mechanisms, and
cocktail of Staphylococcus aureus strains in inter- this approach has been used effectively in under-
mediate moisture (IM) military bread (Feeherry et al., standing nonlinear chemical dynamics such as oscil-
2003) with variations in water activity (a w, as adjusted lating chemical reactions (Epstein and Pojman, 1998).
by desorptive and absorptive techniques), pH, and The chemically triggered alternation between inde-
temperature (T). In the latter study, both the logistic pendently living cells and cooperative aggregates of
equation and Gompertz equation successfully fit the the cellular slime mold Dictyostelium discoideum was
growth kinetics data and generated accurate estimates successfully modeled by introducing the chemical
for the lag time (k), maximum growth rate (l), and kinetics concept of autocatalysis (Goldbeter, 1991).
asymptotic maximum (called ratio of growth and Additionally, a predator–prey life cycle that was
denoted as R g). Some limitations with this approach successfully duplicated in a controlled experiment
were encountered. Microbial population data collected utilizing planktonic rotifers feeding on algae in a
after completion of the stationary phase showed a chemostat was also modeled by integrating a bio-
decline, characteristic of the death phase of the chemically based set of ODEs (Fussman et al., 2000).
bacterial life cycle (McMeekin et al., 1993; Bailey The quasi-chemical model postulates an underlying
and Ollis, 1986; Lynch and Poole, 1979). For some reaction scheme that combines autocatalytic growth
data profiles, distinguishing late stationary phase data and negative feedback, presumably in the form of a
from early death phase data was ambiguous, and the hypothetical antagonistic metabolite, which hastens
resulting estimated value of l depended on which the death of the actively multiplying cells. The actual
data was included in the kinetics growth curve. chemical identity of this antagonist and its biomolec-
Microbial kinetics models generally fit either ular mechanism have not yet been confirmed. One
growth or death kinetics (Legan et al., 2002; Baranyi possibility is that the production of low levels of a
and Roberts, 2000; McMeekin et al., 1997; Whiting, diffusible metabolite by the bacterial cells acts as an
1995). Some alternative models have been proposed intercellular signalling molecule and inhibits growth
to account for microbial growth–decline data, such as in the cell density-dependent gene expression phe-
the three-step enzyme kinetics model (Whiting and nomenon commonly referred to as quorum sensing
Cygnarowicz-Provost, 1992). Other models combine (Smith et al., 2004; England et al., 1999; Novick,
expressions for growth and death into a single 1999; Bassler, 1999). Another possible negative
equation, such as the logistic model with a super- feedback mechanism would involve the accumulation
imposed Fermi term (Peleg, 1996), the Jones and of a waste product such as lactic acid (Vereecken et
Walker model (Jones and Walker, 1993; Jones et al., al., 2003). A mechanism involving nutrient depletion
1994), the Churchill model (Membré et al., 1997), and (Leroy and De Vuyst, 2004) is less likely for
a model combining two Baranyi-type equations for describing the growth of a pathogen such as S. aureus
yeast fermentation in a model wine system (Del in the nutrient-rich environment of a food product
Nobile et al., 2003). The Baranyi equation is a non- (Van Impe et al., 1999; McMeekin et al., 1993; Lynch
autonomous differential equation (Baranyi et al., and Poole, 1979).
1993; Baranyi and Roberts, 1994) that is used to From the proposed hypothetical mechanism, one
model microbial growth kinetics in foods (Membré et can establish a set of kinetics rate equations that
al., 1999), and whose mathematics have been studied form the basis for a system of ordinary differential
extensively (Baranyi and Roberts, 1995). equations (ODEs) that imbue the quasi-chemical
The quasi-chemical model combines the concepts model with mathematical attributes. These attributes
of chemical kinetics and predictive modeling in foods. of the quasi-chemical model were originally used
Applying chemical kinetics to the macroscopic (Taub et al., 2003) to fit growth–death kinetics data
behaviors of microbial populations requires a much- and death-only kinetics for S. aureus in bread as
simplified view of the network of biochemical functions of a w, pH, and temperature. Statistical
reactions occurring in the microorganism (Bray and analysis of the results from the quasi-chemical
White, 1966; Hinshelwood, 1953). Insight into these model produced mathematical relationships among
complex metabolic processes can be gained from the parameters that were used to delineate a growth/
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 159

no-growth boundary in the pH–a w plane. An in- Table 1


teresting comparison showed that the quasi-chemical Summary of mechanism and equations for the quasi-chemical
model
model estimated values of l for the growth–death
and growth-only data different from the value Reaction Rate ODEs
step equations
estimated with the Gompertz function (for the
MYM* v 1=k 1M dM =dt ¼ ð  v1 Þ ¼  k 1 M ð1Þ
growth-only data). More recently, the quasi-chemical
model has been generalized (Feeherry et al., 2004) M*Y2M*+A v 2=k 2M* dM 4=dt ¼ ðv1 þ v2  v3  v4 Þ
to characterize the growth–death kinetics for other ¼ k1 M þ M 4ðG  eAÞ ð2Þ
pathogens (Escherichia coli and Listeria monocyto- M*+AYD v 3=k 3M*A dA=dt ¼ ðv2  v3 Þ ¼ M 4ðk2  eAÞ ð3Þ
genes) in foods (IM turkey meat, ham, and cheese)
M*YD v 4=k 4M* dD=dt ¼ ðv3  v4 Þ ¼ M 4ðk4 þ eAÞ ð4Þ
under various experimental conditions (a w, pH,
U ðtÞ ¼ M ðtÞ þ M4ðtÞ ð5Þ
temperature, and lactate concentration). Additionally,
the quasi-chemical model was successfully tested to
account for the inactivation kinetics of E. coli in
model foods using the emerging non-thermal tech- The net natural growth rate, G ( Guk 2–k 4), depicts the
nology of high pressure processing (Feeherry et al., excess of the growth rate over the natural death rate.
2004). Presently, we explore the mathematical The parameter e relates to k 3 by e=hk 3, in which the
properties of the quasi-chemical kinetics model that scaling factor h (defined as h=e/k 3) is assigned the
make it a useful tool for predictive microbiology. value 109 (and consequently (0Vebb1) to facilitate
We examine the mathematical relationships among numerical solutions of the ODEs. The initial con-
the parameters that give the quasi-chemical model ditions at t=0 are as follows: M equals I (the inoculum
the ability to fit growth–death kinetics and (linear level, Ic103–104), and M*=A=D=0. Changes in the
and nonlinear) death kinetics satisfactorily. In microbial population size are represented by the
Appendix A, we list the MATLAB programs that quantity U=M+M* (in units of CFU/mL) and are
can be used to fit data by this model, and Appendix B derived from microbiological plate counts that do not
describes a worked example of a typical data set using distinguish lag phase (M) and growth phase (M*) cells
these programs. The authors would be pleased to (see Table 1, Eq. (5)). Fig. 1 (Taub et al., 2003) depicts
provide electronic versions of these programs upon the behavior of the individual entities in the quasi-
request. chemical mechanism (M, M*, A, D, and U) from the
fit of a typical set of growth–death kinetics data.

2. Material and methods


3. Results
2.1. The quasi-chemical model
3.1. Properties of the model
Table 1 briefly summarizes the theoretical basis of
the quasi-chemical model; namely, the hypothetical The growth–death profile of U(t) shown in Fig. 1
four-step reaction scheme, and the corresponding rate is based on a non-zero value of k 3 (k 3=100) and a
equations and system of ODEs (Taub et al., 2003). positive value of G (k 2–k 4=2). When the relative
The concentration of lag phase cells is denoted by M, values of the rate constants are varied, different
growth phase cells by M*, antagonist by A, and dead patterns for U as a function of time are observed.
cells by D. The concentrations and rate constants (k 1 Fig. 2 displays U(t) with GN0 and fixed values of
through k 4) have only non-negative values. The k 1=1, k 2=4, and k 4=2, while k 3 is varied incrementally
procedures for making initial estimates for these from 0 to 1000. With values of k 3N0, the calculated
parameters and determining quantities such as max- function for U (and also for M*—see Fig. 1) as a
imum logarithmic growth rate (l), lag time (k), function of time depicts growth to a maximum
maximum population size (called the ratio of growth followed by a decline. As k 3 becomes larger, the
and denoted R g) are described in subsequent sections. calculated U(t) function becomes more pulse-like,
160 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

Fig. 1. Predicted plot of individual functions M, M*, A, D, and U versus time with k 1=1, k 2=4, k 3=100, and k 4=2.

reaching its maximum earlier and at a lower value. The ODE system (Table 1) is generally nonlinear
When k 3=0, the function U(t) shows limitless, nearly due mainly to the interaction between the variables
linear growth, which was not observed experimentally. M* and A in Eqs. (2) and (3), and solutions cannot

Fig. 2. Predicted effect of non-zero values of k 3 on growth–death behavior exhibited by U(t) with {k 1=1; k 2=4; k 3=0, 1, 10, 100, 1000; and
k 4=2}.
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 161

always be expressed by exact formulas. An exception decreases. Moreover, the quantity [1e/(t )]V1,
is the solution to Eq. (1) which when inserted into Eq. (7) leads to

M ¼ Ieðk1 tÞ ð6Þ Aðt ÞVðk2 =eÞ ¼ k2 =ðhk3 Þ: ð8Þ

that shows that M undergoes an exponential decline Consequently, A(t) increases steadily from zero as
from I toward zero over time. The variable D is absent time increases, but its value remains below the upper
from Eqs. (1)–(3) and can be calculated from Eq. (4), bound of k 2/hk 3 (see Fig. 1). The relationship
after M* and A have been determined (see Table 1). between A and M*, Eq. (7), establishes an essential
Since D does not influence the other variables, and M feature of the ODE system (Table 1) when k 3p0. In
is known if k 1 is known, the relationship between M* the special case of k 3=0, this upper bound becomes
and A can be determined in the following way. First, irrelevant.
assume that M*(t) is known, then solve Eq. (3) by The interrelationship between A and M* in
elementary means (variation of parameters) according response to variations in the values of the rate
to Eq. (7) constants and their relative proportions can be seen
in phase-plane plots. Figs. 3 and 4 show trajectories in
h i the (M*, A) phase plane using values of A and M*
Aðt Þ ¼ ðk2 =eÞ 1  e/ðtÞ obtained by numerical solution of Eqs. (1)–(4) and
Z s¼t
representative values assigned to the k’s. The arrows
ðin which /ðt Þ ¼ e M 4ðsÞdsz0Þ ð7Þ in Figs. 3 and 4 indicate the direction of the respective
s¼0 trajectories as t increases from 0.1 to 10. In Fig. 3, the
value of the rate constants are set {k 1=1; k 2=2, 3, 4,
Eq. (7) is not a solution for A, since M* is not known, and 5; k 3=0; and k 4=4} such that k 3=0 and G ranges
but rather an integral relationship between A and from negative to positive values. In these cases with
M *. Differentiating Eq. (7) leads to dA /dt = k 3=0 and GN0, the model projects unrestrained
k 2M*(t)e/(t )N0, which implies that A(t) never growth of M* and A. With k 3=0 and Gb0, A increases

Fig. 3. Phase-plane plot [A, M*] showing trajectories interrelating A and M* with various values of G (=k 2k 4) when k 3=0.
162 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

Fig. 4. Phase-plane plot [A, M*] showing trajectories for various values of G when k 3=100.

toward a limiting value, and M* changes only slightly, Eq. (9) is satisfied when M=0, Eqs. (11) and (12) are
then decreases. Fig. 4 illustrates phase-plane plots satisfied when M*=0, and Eq. (10) is satisfied when
with a set of rate constant values {k 1=1; k 2=2, 3, 4, M=M*=0. The lone critical point in this system occurs
and 5; k 3=100; and k 4=4} such that k 3N0 and G with M=M*=0 and is reached only in the limit as
ranges from negative to positive values. With k 3=100 t Yl. The values of D(t) and A(t), as mentioned
and GN0, the phase trajectories do not exhibit the previously for Eqs. (3) and (4), increase steadily
limitless growth seen in Fig. 3. Instead, both M* and towards constant values as t Yl (see Fig. 1, for
A increase initially, then M* eventually reaches a which k 3N0 and GN0).
maximum and later declines, while A tends continu-
ously toward a limiting maximum value. With k 3=100 3.3. Behavior near t =0
and Gb0, the trajectory behaves similarly to its k 3=0
and Gb0 counterpart (Fig. 3). These results indicate Approximate solutions of the following form are
that non-zero values of k 3 and positive values of G are useful near the beginning of the process (when tc0)
needed to produce growth–death kinetics. and can be found by a Taylor series expansion about
t =0:
3.2. Behavior as t Yl
 
k 2 t2
The critical points of the system (i.e. the points at M cI 1  k1 t þ 1 ð13Þ
2
which the system as a whole is stationary) must satisfy
the following conditions
 
dM =dt ¼ 0 ¼  k1 M ð9Þ ðG  k1 Þt 2
M 4ck1 I t þ ð14Þ
2
dM 4=dt ¼ 0 ¼ k1 M þ M 4ðG  eAÞ ð10Þ
dA=dt ¼ 0 ¼ M 4ðk2  eAÞ ð11Þ
k1 k2 It 2
dD=dt ¼ 0 ¼ M 4ðk4 þ eAÞ ð12Þ Ac ð15Þ
2
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 163

As described in the previous section, k 3 exerts little


k1 k4 =t 2
Dc ð16Þ influence in the early stages, and the approximate
2 solutions for k 3=0 (Eqs. (18)–(22)) can be used.
  
Gk1 t 2
U cI 1 þ ð17Þ U cM 4c k1 I=mÞeGt and lnðU Þclnð k1 I=mÞ þ Gt
2

Eq. (14) shows that M* increases initially from 0 When k 3=0, Eqs. (2) and (3) lead to the equation
approximately linearly, while A and D increase more dU =dt ¼ dM =dt þ dM4=dt ¼ GM 4
slowly and in a quadratic fashion according to the
ratio ¼ Gk1 If ðt; mÞeGt

A k2 Thus, dU/dt has the same sign as G, which indicates


c that U always increases with t if GN0, and corre-
D k4
spondingly U always decreases if Gb0. Growth–
U increases if GN0, and decreases if Gb0. In all cases decline kinetics (Taub et al., 2003) cannot be
at t=0, dU/dt=0, and the model predicts that U(t) predicted by the formulas derived with k 3=0 in Eqs.
always exhibits a non-zero value for the initial lag, (18)–(22). Relatively large values of k 3 are needed for
although its duration may be immeasurably small. the quasi-chemical model to depict the growth–death
None of these approximations is affected by k 3. At phenomenon characteristic of the microbial life cycle.
sufficiently early times, solutions with k 3=0 are there- The relation of the maximum (logarithmic) growth
fore acceptable approximations to the true solutions. rate (l) and G is consistent with these findings. Let

3.4. Approximation when k 3=0 L ¼ Lðt Þ ¼ log10 fU ðt Þg; then Lclog10 ð k1 I=mÞ þ BGt

in which B=log10(e)=0.4343, and the slope of L is


In examples with k 3=0, the nonlinear terms approximately constant,
disappear from the ODE system, and the effect of A
on the microbial kinetics is negligible. In this case, dL=dtcBG
exact solutions for M*, A, and U can be found by
elementary means as follows: lumaxf dL=dt g ð23Þ
Since dL/dt is approximately constant, this equation
M 4 ¼ k1 If ðt; mÞeGt ð18Þ
yields the relation
  
U ¼I emt þ k1 f ðt; mÞgeGt ¼ I ek 1t þ k1 f ðt; mÞeGt lcBG ð24Þ
ð19Þ Fig. 5 shows a plot of l versus G using previously
reported values obtained from modeling the growth–
A ¼ ð Ik1 k2 =mÞf f ðt;  GÞ  f ðt; k1 Þg ð20Þ death kinetics of S. aureus in IM bread at various
conditions of a w, pH, and temperature (Taub et al.,
with the definitions
2003). The straight-line graph of Eq. (24) shows that
m ¼ G þ k1 ¼ k1 þ k2  k4 ð21Þ the latter is a good predictor of this relationship,
except possibly when |G| is near zero. This estimate is
consistent with the conclusion that l usually occurs
f ðt; mÞ ¼ ð1  emt Þ=m ð22Þ
early enough in the process to justify the application
A special case occurs when k 1 is very large. These of the approximate solutions obtained for k 3=0.
conditions are uncommon and require special circum-
stances that will be discussed later. Specifically, when 3.5. The death-only case
k 1NN|G|, the value of mck 1NN1, and accordingly
f(t,m)c1/m. After a brief time increment elapses, The quasi-chemical model also fits death-only
MbbM* and UcM*. kinetics observed as a result of bno-growthQ con-
164 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

Fig. 5. Linear relationship of l and G (values obtained from fitting with the quasi-chemical model) in accordance with Eq. (24) (l=BG).

ditions of a w, pH, and T (Taub et al., 2003), or from (especially those involving high pressures), the plot of
the application of high hydrostatic pressure (Feeherry log(U) versus t is approximately linear, and Eqs. (18)–
et al., 2004). The fitting procedure in these cases leads (22) show that [ G (fast set)ck 1(slow set)]. Although the
to two entirely different sets of parameters (called the functions U are nearly identical in both sets, the
bfast setQ and the bslow setQ) that fit the data equally different parameters calculate very distinct temporal
well. For example, the fitted lines calculated by the functions for M, M*, A, and D (Figs. 6 and 7). The
bfastQ and bslowQ sets for the high pressure ( P=50,000 most notable differences are the rapid decay of M and
psi) inactivation of E. coli in whey protein (Feeherry rapid production of M* in the fast set (Fig. 6), in
et al., 2004) are nearly identical, and the correspond- comparison to the slow decay of M and the slow rise–
ing estimates of l are also very similar (lc0.12, fall kinetics of M* generated by the slow set (Fig. 7).
see Table 2). With the slow set, the production of A is also
Both the bfastQ and bslowQ parameter sets have significantly delayed and the evolution of D is slightly
negligible values for k 3, and solutions are approxi- attenuated.
mated using Eqs. (18)–(22). The bfast setQ features
large values of k 1 (k 1NN1), and M*NNM shortly after 3.6. Model fitting and parameter estimation
t=0, while the bslow setQ has small values of k 1 and
M*bbM. In cases with negligibly small lag times Nonlinear least-squares analysis was used to
obtain best-fit values of the model parameters (rate
constants k 1 through k 4) by solving numerically the
ODE system for the four state variables M, M*, A,
Table 2 and D which lead to the estimate of the measure-
bFast setQ and bslow set Q from fitting microbial inactivation data
ment variable, U(t). This procedure was carried out
k1 k2 k3 k4 G l using MATLAB software, specifically the stiff
bFastQ 34.30 0.96 0 10.25 0.282 0.12 differential equation solver ODE15s, in conjunction
bSlowQ 0.282 0.70 0 19.31 18.62 0.12 with the nonlinear regression analysis program
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 165

Fig. 6. Plots predicting M, M*, A, D, and U versus time for death kinetics using bfast setQ of parameters.

LSQNONLIN. These programs (see Appendices A squares fitting procedure worked well for data that
and B) require initial estimates for the k’s and were not too noisy, but the goodness of fit depended
iteratively seek the values of k’s that generate the on fortuitous initial estimates of the k’s. Like many
smallest sum of squared errors. Generally, the least- nonlinear searching programs, LSQNONLIN may fail

Fig. 7. Plots predicting M, M*, A, D, and U versus time for death kinetics using bslow setQ of parameters.
166 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

if the initial choices are bad. Also, the initial estimates 4. Discussion
can appear to influence the final results if the data are
irregular or too sparse. While no general procedure The current models used most commonly for
will help in all cases, it is possible to state the microorganism kinetics (Gompertz or logistic
following rules of thumb using the information about equations) tend to focus exclusively on either
the mathematical properties of the model described growth or inactivation, and generally ignore the
above: death phase when it succeeds growth. The quasi-
chemical model (Taub et al., 2003), like some
(A) If the data appear to start by growing, choose reported alternatives, has the capacity to model
initial estimates having k 2Nk 4, and choose k 2bk 4 continuously growth–death kinetics and incorpo-
if the data decline with time. rates death data in the estimation of the growth
(B) Increasing k 1 generally speeds up the process. kinetics parameters. This feature is advantageous,
(C) If there is enough data near the start of the when there is a paucity or irregularity in the
process, Eqs. (13) and (17) may suggest an growth data, and also eliminates the need for
initial estimate of G. subjectivity when determining which stationary
(D) In program SD in Appendix A, the line that phase data constitute the endpoints of the growth
begins LB=[. . . allows the user to specify lower curve.
and upper bounds in the search for parameter The quasi-chemical model consists of four
values. For example, if we think that k 1c2, we autonomous ODEs for four variables (M, M*, A,
can simplify the search by setting LB=[1.9 1 1 and D) which together specify the system and the
1]; UB=[2.1 1 1 1] on that line. predicted microorganism concentration, U(t). Even
this rudimentary representation of the microbial
The value of l was estimated by numerically life cycle provides some insight into the metabolic
differentiating L(t)=log10(U(t)) to find its slope and activities governing the closed-system behavior of
searching for the maximum of that slope. In cases the bacterial cell. Specifically, the model predicts
where only death occurred, l was determined from (at least) two different pathways to death, one that
the minimum slope of L(t) as a negative growth occurs in no-growth conditions and another that
rate. If the time at which l occurs is defined as t l , prevails after a period of robust growth through a
then the lag (denoted k) is estimated from the mechanism involving negative feedback. Bacteria
formula are known to produce substances that inhibit
 growth, some in the form of diffusible molecules
k ¼ tl  Ll  Lo =l: that act in a cell density-dependent way to
influence specific gene expression, called quorum
where sensing (Smith et al., 2004; England et al., 1999;
 Novick, 1999; Bassler, 1999), and others in the
Lo ¼ Lð0Þ and Ll ¼ L tl form of organic acids (Vereecken et al., 2003).
We do not present experimental evidence to
The ratio of growth (R g) characterizes the extent of confirm the presence, identity, or mechanism of
multiplication of the initial organism concentration the hypothetical antagonist in the quasi-chemical
(asymptotic maximum). The quantity R g is model that is proposed to act in a manner
expressed by the relation analogous to these known inhibitors. The present
results do not contradict the hypothesis that there
Rg ¼ log10 fmax½U ðt Þ =U ð0Þg ð25Þ is more than one pathway to death of the
organism, and that metabolic by-products influence
The quantity R g applies only to situations involving the rate of the alternate pathways.
growth, and R g has the value of zero, in cases when The quasi-chemical model is an ODE model
only death kinetics are observed (bno-growthQ based on the simple premise that the mathematical
conditions). model comprises one or more differential equa-
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 167

tions. Presently, the ODE model contains four ization of the death rate as (l), but other
parameters (k 1 through k 4) that specify the model definitions are possible and could be a subject
behavior and are estimated by fitting the data (U). for future study. In cases where only growth
The ODEs are generally nonlinear due to the kinetics data are collected, a cursory exploration
interaction of M* and A in the proposed mecha- of existing data suggested that in the fitting
nism, and this relationship is critical for reproduc- process the resulting estimates of the k’s might
ing growth–death kinetics. Solutions to the ODEs sometimes depend on the initial estimates of these
obtained through various approximations show that parameters. However, as in the death-only case,
the behavior of U depends principally on certain the changes in the k’s did not appear to affect the
values for the quantities k 3 and G. Specifically, estimates of U or l significantly.
only if k 3N0 and GN0, will U(t) show growth–
decline behavior. If k 3z0 and Gb0, then only
death occurs. The value of l is linearly related to Acknowledgements
G (see Eq. (24) or Fig. 7), but k and R g are not
closely associated with G or the k’s. The parameter k 1 The authors wish to express their gratitude to
affects primarily the overall time-scale of the process, Mr. James Doona (Mathematician), Dr. Jack Briggs
and the relation between M* and A is not directly (Senior Food Technologist), and Dr. Fred Allen
affected by changing k 1 (Eq. (7)). The approximations (Chemical Engineer) for providing helpful discus-
made when k 3=0 are influenced by the quantity m, sions and insight into issues regarding mathe-
which depends on k 1 (Eq. (21)). If k 3=0, the model matics, fermented food products and bioprocess
predicts that U grows without limit when GN0, and engineering.
declines when Gb0. None of the data exhibited
unlimited growth.
The quasi-chemical model adequately character- Appendix A. MATLAB programs for fitting the
izes the death-only case (when Gb0) in the sense ODE model
that it can satisfactorily fit the U(t) data and
estimate a value for l. In its present form, the These programs for fitting the data by the ODE
model lacks uniqueness, since at least two sets of model consist of a MATLAB script, SD, that
parameters may provide equally good fits of the invokes the nonlinear least-squares function
data. In fact, all sufficiently large values of k 1 in LSQNONLIN which calls the function, ED, to
the bfast setQ that induce M to become negligible assist in the fitting. ED uses MATLAB program
quickly will also produce nearly identical values ODE15S to solve the ODE system which is defined
of U(t). Eq. (19) justifies this statement; the in the function FD. The separate MATLAB script
relationship mck 1NN1 implies k 1f(t,m)c1, and PredODE simply calculates the solutions, M, M*,
UcIeGt (i.e. U depends only on G and not k 1). For A, D, and U, given the k values and the initial
some purposes, this non-uniqueness is not detri- variable values; it does not involve any fitting of
mental. It suggests that the model in its current data but does solve the ODE system. Each of these
form may be over-parameterized for death-only programs must be stored in a file having the same
kinetics, but also leaves open the question of name as the program, e.g. SD must be stored in a
which parameter set more closely resembles the file named SD.m. The percent symbol, %, in the
true dynamics. Both parameter sets are equally programs denotes that the remaining text on that
valid in terms of mathematics, and more exper- line is merely commentary. In these programs, the
imental data would be needed to distinguish which variables M, M*, A, and D are re-named in matrix
is more appropriate from a microbiological per- fashion as follows:
spective. This uncertainty does not affect the
determination of U(t) or l, although there is M ¼ x1 ¼ xð :; 1Þ; M * ¼ x2 ¼ xð :; 2Þ; A ¼ x3
ambiguity in the definition of l when Gb0. The
present definition leads to quantitative character- ¼ xð :; 3Þ; D ¼ x4 ¼ xð :; 4Þ:
168 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 169

Appendix B. Output of comparison calculations Next is the command, SD, to execute the script SD.
This is followed by output from SD. The first block
Below we show the input and output for a of output gives information about the intermediate
MATLAB fitting of a typical data set. First, we steps in the repetitive process that LSQNONLIN
display the results for the ODE model. The uses. Generally, f(x) and the Norm of Step should
solicitation symbol, NN, displayed by MATLAB, both decrease, but the latter may occasionally
indicates that we are expected to enter something increase slightly from one iteration to the next. We
from the keyboard. The first three entries are the do not list below all the steps in the iteration, but
input, which defines the quantities td, xd, the time only the first few and last few. The output may
and concentration, i.e. the test data that we are trying depend slightly on the computer and operating
to fit, and k, our initial guesses for the parameters. system on which this is run.
170 E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171

After running SD, we usually run PredODE, which References


calculates l, k, and R g and makes a smoother graph of
the predicted U than SD does. Bailey, J.E., Ollis, D.F., 1986. Biochemical Engineering Funda-
mentals. McGraw-Hill, New York.
Baranyi, J., Roberts, T.A., 1994. A dynamic approach to predicting
bacterial growth in food. Int. J. Food Microbiol. 23, 277 – 294.
Baranyi, J., Roberts, T.A., 1995. Mathematics of predictive food
microbiology. Int. J. Food Microbiol. 25, 199 – 218.
Baranyi, J., Roberts, T.A., 2000. Principles and applications of
predictive modeling of the effects of preservative factors on
microorganisms. In: Lund, B.M., Baird-Parker, T.C., Gould,
G.W. (Eds.), The Microbiological Safety and Quality of Food,
vol. 1. Aspen Publishers, Gaithersburg, pp. 342 – 358.
Baranyi, J., Roberts, T.A., McClure, P.J., 1993. A non-autonomous
differential equation to model bacterial growth. Food Microbiol.
10, 43 – 59.
Bassler, B.L., 1999. How bacteria talk to each other: regulation of
gene expression by quorum sensing. Curr. Opin. Microbiol. 2
(6), 582 – 587.
E.W. Ross et al. / International Journal of Food Microbiology 99 (2005) 157–171 171

Bray, H.G., White, K., 1966. Kinetics and Thermodynamics in McMeekin, T.A., Brown, J., Krist, K., Miles, D., Neumeyer, K.,
Biochemistry. Academic Press, New York, pp. 364 – 404. Nichols, D.S., Olley, J., Presser, K., Ratkowsky, D.A., Ross,
Del Nobile, M.A., D’Amato, D., Altieri, C., Corbo, M.R., T., Salter, M., Soontranon, S., 1997. Quantitative Micro-
Sinigaglia, M., 2003. Modeling the yeast growth-cycle in a biology: a basis for food safety. Emerg. Infect. Dis. 3 (4),
model wine system. J. Food Sci. 68 (6), 2080 – 2085. 541 – 549.
England, R.R., Hobbs, G., Bainton, N.J., Roberts, D.M., 1999. Membré, J.M., Thurette, J., Catteau, M., 1997. Modeling the
Microbial Signaling and Communication, 57th Symposium of growth, survival, and death of Listeria monocytogenes. J. Appl.
the Society for General Microbiology. Cambridge University Microbiol. 82, 345 – 350.
Press, Cambridge. Membré, J.M., Kubaczka, M., Chéné, C., 1999. Combined effects
Epstein, I.R., Pojman, J.A., 1998. An Introduction to Nonlinear of pH and sugar on growth rate of Zygosaccharomyces rouxii, a
Chemical Dynamics. Oxford University Press, New York. bakery product spoilage yeast. Appl. Environ. Microbiol. 65
Feeherry, F.E., Doona, C.J., Taub, I.A., 2003. Effect of water (11), 4921 – 4925.
activity on the growth kinetics of Staphylococcus aureus in Novick, R.P., 1999. Regulation of pathogenicity in Staphylococcus
ground bread crumb. J. Food Sci. 68 (3), 982 – 987. aureus by a peptide-based density-sensing system. In: Dunny,
Feeherry, F.E., Ross, E.W., Doona, C.J., 2004. A quasi-chemical G.M., Winans, S.C. (Eds.), Cell–Cell Signaling in Bacteria.
model for the growth and death of microorganisms in foods by ASM Press, Washington, DC, pp. 129 – 146.
non-thermal and high pressure processing. Int. J. Food Micro- Peleg, M., 1996. A model of microbial growth and decay in a closed
biol. (in press). habitat based on a combined Fermi’s and the logistic equation. J.
Fussman, G.F., Ellner, S.P., Shertzer, K.W., Hairston, N.G., 2000. Sci. Food Agric. 71, 225 – 230.
Crossing the Hopf bifurcation in a live predator–prey system. Smith, J.L., Fratamico, P.M., Novak, J.S., 2004. Quorum sensing: a
Science 290, 1358 – 1360. primer for food microbiologists—review. J. Food Prot. 67 (5),
Goldbeter, A., 1991. Biochemical Oscillations and Cellular 1053 – 1070.
Rhythms. Cambridge University Press, London. Taub, I.A., Feeherry, F.E., Ross, E.W., Kustin, K., Doona, C.J.,
Hinshelwood, C.N., 1953. Autosynthesis. J. Chem. Soc., 1947 – 2003. A quasi-chemical kinetics model for the growth and
1956. death of Staphylococcus aureus in intermediate moisture bread.
Jones, J.E., Walker, S.J., 1993. Advances in modeling microbial J. Food Sci. 68 (8), 2530 – 2537.
growth. J. Ind. Microbiol. 12, 200 – 205. Van Impe, J.F., Versyck, K.J., Vereecken, K.M., Geeraerd,
Jones, J.E., Walker, S.J., Sutherland, J.P., Peck, M.W., Little, A.H., Dens, E.J., Bernaerts, K., 1999. Predictive micro-
C.L., 1994. Mathematical modeling of the growth, survival, biology of structured foods: development of a unifying
and death of Yersinia enterocolitica. Int. J. Food Microbiol. 23, modeling framework and application to microbial interac-
433 – 447. tions and the selection of advanced model building blocks.
Legan, D., Vandeven, M., Stewart, C., Cole, M., 2002. Modeling In: Tuijtelaars, A.C.J., Samson, R.A., Rombouts, F.M., Noter-
the growth, survival, and death of bacterial pathogens in foods. mans, S. (Eds.), Food Microbiology and Food Safety into the
In: de W. Blackburn, C., McClure, P.J. (Eds.), Foodborne Next Millenium. Ponsen & Looyen, Wageningen, Netherlands,
Pathogens: Hazards, Risk Analysis, and Control. Woodhead pp. 913 – 918.
Publishing, Cambridge, pp. 53 – 95. Vereecken, K.M., Devlieghere, F., Bockstaele, A., Debevere, J., Van
Leroy, F., De Vuyst, L., 2004. Growth of the bacteriocin-producing Impe, J.F., 2003. A model for lactic acid-induced inhibition of
Lactobacillus sakei strain CTC 494 in MRS broth is strongly Yersinia enterocolitica in mono- and co-culture with Lactoba-
reduced due to nutrient exhaustion: a nutrient depletion model cillus sakei. Food Microbiol. 20, 701 – 713.
for the growth of lactic acid bacteria. Appl. Environ. Microbiol. Whiting, R.C., 1995. Microbial modeling in foods. Crit. Rev. Food
67 (10), 4407 – 4413. Sci. Nutr. 35 (6), 467 – 494.
Lynch, J.M., Poole, N.J., 1979. Microbial Ecology: A Conceptual Whiting, R.C., Cygnarowicz-Provost, M.L., 1992. A quantitative
Approach. John Wiley & Sons, New York. model for bacterial growth and decline. Food Microbiol. 9,
McMeekin, T.A., Olley, J.N., Ross, T., Ratkowsky, D.A., 1993. 269 – 277.
Predictive Microbiology: Theory and Application. Research
Studies Press, Somerset.