Predictive Model for Growth of Bacillus cereus at

Temperatures Applicable to Cooling of Cooked

Pasta
Vijay K. Juneja , Chase E. Golden, Abhinav Mishra, Mark A. Harrison, and Tim B. Mohr

Abstract: A model was developed to predict the growth of Bacillus cereus from spores during cooling of cooked pasta.
Cooked pasta was inoculated with a cocktail of four strains of heat-shocked (80 °C/10 min) B. cereus spores to obtain a
final spore concentration of approximately 2 log CFU/g. Thereafter, growth was determined at isothermal temperatures
starting at 10 °C and every three degrees up to 49 °C. Samples were removed periodically and plated on mannitol egg yolk
polymyxin agar. The plates were incubated for 24 hr at 30 °C. Baranyi, Huang, and modified Gompertz primary growth
models were used to fit growth data. The modified Ratkowsky secondary model was used to fit growth rates determined
by the primary growth models with respect to temperature. All three primary models fitted the growth data well. The
modified Ratkowsky secondary model adequately fit growth rates generated by the three primary models (R2 values
ranging from 0.96 to 0.98). After acceptable prediction zone (APZ) validation and goodness of fit statistical analyses, it
was determined that the Baranyi primary growth model was best suited for these data. For both single-rate exponential
cooling and biphasic linear cooling model validation, all Baranyi model predictions (n = 24 and 28, respectively) fell
within the APZ (−1.0 to 0.5 log CFU/g). The model will assist institutional food service settings to determine the safety
of cooked pasta subjected to longer cooling times or stored at improper temperatures.

Keywords: Bacillus cereus, Baranyi model, cooling, pasta, predictive modeling

Practical Application: Predictive model can be used to estimate extent of microbial growth during cooling of cooked
pasta and in designing HACCP program and setting of critical control levels. Retail food industry would need fewer
challenge studies to validate the safety of their products. The model will provide regulatory agencies and food industry
with an objective means of assessing the microbial risk and ensuring that the public is not at risk of acquiring food
poisoning.

Food Microbiology &

Introduction
Bacillus cereus is a ubiquitous, spore-forming, gram-positive,
motile, facultative anaerobe that grows well under aerobic con-
ditions (Granum, 1994). The pathogen is a common food con-
taminant and has been isolated from a variety of foods including
meat and meat products, milk and dairy products, raw and pureed
vegetables, spices, cereals, rice, pasta, noodles, and so on (Eglezos,
Huang, Dykes, & Fegan, 2010; Fangio, Roura, & Fritz, 2010;
Granum, 2005; Kramer & Gilbert, 1989; Wijnands, Dufrenne,
Rombouts, in’t Veld, & van Leusden, 2006). B. cereus is an etiolog-
ical agent of two distinct forms of foodborne illnesses, self-limiting
(24 to 48 hr) gastrointestinal illness, that is, emetic and diarrheal
(Granum, 1994; Lee, Chang, Kim, Kim, & Park, 2006). Although
diarrheal disease outbreaks are primarily linked to protein-rich
foods, emetic food poisoning is usually associated with the con-
sumption of starchy foods, including pasta.
JFDS-2018-1626 Submitted 10/10/2018, Accepted 12/27/2018. Author Juneja
is with U.S. Dept. of Agriculture, Agricultural Research Service, Eastern Regional
Research Center, 600 East Mermaid Lane, Wyndmoor, PA, 19038, USA. Authors
Golden, Mishra, and Harrison are with Department of Food Science & Technology,
Univ. of Georgia, Athens, GA, 30602, USA. Author Mohr is with U.S. Dept.
of Agriculture, Food Safety and Inspection Service,, Office of Public Health Science,
Science Staff, 530 Center Street, NE, Suite 401, Salem, OR, 97301, USA. Direct
inquiries to Juneja (E-mail: vijay.juneja@ars.usda.gov).
Safety
Heat resistant B. cereus spores in pasta can survive the time
and temperature used to cook pasta in food-service operations.
The surviving heat-activated spores pose a public health risk due
to their potential for subsequent germination, outgrowth, and
multiplication in pasta, especially when the rate and extent of
chilling is not sufficient. Also, cooked pasta can be unsafe for
human consumption if not stored, transported, and distributed
under proper refrigeration, because there is a likelihood of B.
cereus spore germination and vegetative cell growth. Levels ࣙ5
log CFU/g of B. cereus are a public health hazard and may cause
food poisoning. Foods with population levels between 3 and 5
log CFU/g are considered borderline and carry moderate risk of
causing foodborne illness (Health Protection Agency, 2009).
Several predictive microbial models have been developed to es-
timate the growth of foodborne pathogens such as Clostridium
botulinum and Clostridium perfringens, at temperatures applicable to
the cooling of cooked products. Some of these models are avail-
able on the United States Dept. of Agriculture (USDA), Agri-
cultural Research Service (ARS), Pathogen Modeling Program
(PMP; https://pmp.errc.ars.usda.gov/PMPonline.aspx). There are
also some models for predicting the growth of B. cereus available on
the USDA-ARS-PMP. The predictions from these latter models
are only applicable at static temperature conditions and the data
for these models were collected in liquid medium. Accordingly,
a dynamic model to estimate growth of B. cereus from spores in
beans and rice was recently published (Juneja et al., 2019; Juneja,

C 2019 Institute of Food Technologists

 R

doi: 10.1111/1750-3841.14448 Vol. 0, Iss. 0, 2019 r Journal of Food Science 1

Further reproduction without permission is prohibited
 Predictive model for growth of bacillus cereus in pasta . . .

Mishra, & Pradhan, 2018). This study was aimed to develop a range. For each temperature, a bag of pasta was pulled at predeter-
model that can be used to predict the growth of B. cereus from mined time intervals. The sampling frequency depended on the
spores during a cooling period for cooked pasta. Because of the incubation temperature and were based on predictions from the
potential health hazards in cooling cooked foods, this model could ARS-PMP, for example, 7 days at 10 °C; 3 to 24 hr at 49 °C, and
assist processors to determine the safety of cooked pasta subjected the total incubation time ranged from 35 days at 10 °C to 96 hr at
to longer cooling times or stored at improper temperatures. 49 °C. Growth experiments were independently conducted three
times for each temperature.
Materials and Methods
Growth of B. cereus at dynamic cooling temperatures
Bacterial strains Both single-rate exponential and biphasic linear cooling exper-
B. cereus strains were obtained from the Eastern Regional Re- iments were performed by storing pasta sample bags in a circulat-
search Center (Wyndmoor, PA, USA) culture collection. Two ing water bath (NESLAB RTE-221, NESLAB Instruments, Inc.,
emetic strains (NCTC 11143[4810/72] and Mac 1) isolated from Newington, NH, USA) and a programmable incubator (Binder,
cooked rice and macaroni and cheese, respectively, and two diar- Model HK 63-UL, Binder, GmbH, Tuttlingen, Germany), re-
rheal strains (935A/74 and Brad 1) isolated from turkey loaf and spectively, with continuously changing temperatures from 54.5 to
canned soup, respectively, were maintained at −20 °C as sporu- 7.2 °C. For single-rate exponential cooling experiments, a pro-
lated stock cultures in sterile distilled water. grammable circulating water bath was set to reduce temperature
from 54.5 to 7.2 °C according to target cooling times of 6, 9, 12,
Preparation of spore suspension 15, 18, or 21 hr. For biphasic linear cooling experiments, a pro-
B. cereus spores were produced according to the methodology grammable incubator was set to cool for 6.5 hr total by reducing
described earlier (Juneja et al., 2018) and were harvested using temperature from 54.5 to 7.2 °C according to the following seven
sterile distilled water and sterile plastic cell spreaders when at least cooling conditions:
>90% of the cells had sporulated, as determined by a micro-
scopic examination. The pooled spore suspensions from 10 plates (a) 54.5 to 27 °C in 6.5 hr
were washed twice with sterile distilled water by centrifugation at (b) 54.5 to 27 °C in 5 hr + 27 to 7.2 °C in 1.5 hr
10,000 × g for 15 min at 4 °C. The population of B. cereus (6 to (c) 54.5 to 27 °C in 4 hr + 27 to 7.2 °C in 2.5 hr
7 log10 CFU/mL) for each strain spore suspension was subjected (d) 54.5 to 27 °C in 3 hr + 27 to 7.2 °C in 3.5 hr
to heat-shock (80 °C/10 min) before plating appropriate 10-fold (e) 54.5 to 27 °C in 2.5 hr + 27 to 7.2 °C in 4 hr
serial dilutions in peptone water (0.1% [w/v]) onto mannitol egg (f) 54.5 to 27 °C in 2 hr + 27 to 7.2 °C in 4.5 hr
yolk polymyxin agar (MYP agar; Criterion Hardy Diagnostics, (g) 54.5 to 27 °C in 1.5 hr + 27 to 7.2 °C in 5 hr
Santa Maria, CA, USA), and counting colonies after incubating
plates for 24 hr at 30 °C. An equal number of spores from each A wireless thermocouple data logger (Madge Tech 4; Madge
suspension was combined to create a four-strain spore cocktail that Tech, Inc., Warner, NH, USA) was used to record the tempera-
Food Microbiology &

was stored at 4 °C until used and heat-shocked just before being ture of all dynamic cooling experiments. Two independent trials,
used. as defined by a new batch of cooked pasta, were performed for
Safety

both single-rate exponential and biphasic linear cooling. Dupli-

Preparation and inoculation of pasta cate pasta bags were pulled at specific time intervals and sampled
Quick cook 3-min rotini pasta was purchased from a local su- for single-rate exponential cooling experiments. For biphasic lin-
permarket. The pasta was cooked by bringing 5-qt of water to a ear cooling experiments, duplicate pasta bags were removed and
rapid boil before adding the full box of pasta (16-oz). The pasta sampled immediately before and after cooling. The data collected
was stirred, returned to a rapid boil, and cooked for 3 min. The were used to validate the predictive microbial model developed in
pasta was also stirred occasionally while cooking and then drained this study.
well after cooking.
Five grams of cooked pasta at room-temperature was distributed Microbiological analysis
into sterile filter bags (Whirl-Pak bags, 7-oz [207-mL] capacity; All pasta sample bags from isothermal and dynamic temperature
3.0 mL [0.076 mm] thick; product no. B01385; Nasco, Ft. Atkin- experiments were enumerated for total germinated B. cereus popu-
son, WI, USA) and inoculated with 100 μL of heat-shocked lation. Pasta samples were combined with 5 mL of sterile peptone
(80 °C/10 min) B. cereus spore cocktail, yielding a final concentra- water (0.1% [w/v]) and mixed in a small laboratory blender (Min-
tion of approximately 2.0 log CFU/g. The cocktail was uniformly iMix 100; Interscience, Rockland, MA, USA) for 2 min. The
combined with the pasta by shaking and massaging the bags thor- slurry was serially diluted and 0.1-mL aliquots were spread plated
oughly for approximately 2 min. The bags were then carefully in duplicate onto MYP agar plates. After plates were incubated
flattened using fingers to remove most of the air without distort- for 24 hr at 30 °C, colonies were counted and the means of cell
ing the pasta and were closed by twice folding over the aluminum populations from four platings of each sampling point were calcu-
wire ends. Noninoculated pasta was also weighed into sterile filter lated. The total population of B. cereus at each sampling time was
bags and used as controls. recorded as log colony-forming units per gram (log CFU/g) of
pasta. Negative controls (noninoculated pasta samples) were used
Growth of B. cereus in pasta at isothermal temperatures to confirm that naturally occurring B. cereus were not present.
Sample bags with spore-inoculated pasta were stored in a Lab-
Line Imperial III incubator (Labline Instruments, Inc., Melrose Primary models
Park, IL, USA) at isothermal temperatures starting at 10 °C and Three primary growth (Baranyi, Huang, and modified Gom-
every 3° up to 49 °C, representing an entire biokinetic growth pertz) models were used to fit the growth data of B. cereus in

2 Journal of Food Science r Vol. 0, Iss. 0, 2019

Predictive model for growth of bacillus cereus in pasta . . .

cooked pasta. These primary models were fitted to the experimen- formula (Gibson, Bratchell, & Roberts, 1988; Zwietering, Jon-
tal growth data using MATLAB (Version R2017a; Mathworks, genburger, Rombouts, & Van’t Riet, 1990):
Natick, MA, USA) nonlinear least squares curve fitting toolbox
with trust-region algorithm. Growth data were transformed to BC
μmax = (6)
loge CFU/g units prior to model fitting. e
The Baranyi model (Baranyi & Roberts, 1994) was used to fit
the isothermal bacterial growth curves: The exact calculation for lag phase from Equation (5) was ex-
pressed by McMeekin, Olley, and Ross (1993):
 
e μmax F (t ) − 1
yt = y0 + μmax F (t ) − ln 1 + (1) 1
e ymax −y0 λ = M − (1 − exp (1 − exp (B M))) (7)
B
where Growth data experiments were performed three times per
1   isothermal condition tested. All three primary models were sepa-
F (t ) = t + ln e −vt + e −h 0 − e (−vt −h 0 )
(2) rately fitted to each replicate at each temperature, and parameter
v
values were averaged across the replicates at each temperature. All
yt represents the cell concentration in ln CFU/g at time t; y0 rep- three primary models were able to be fitted to growth data from
resents the initial cell concentration in ln CFU/g; ymax represents isothermal growth conditions 13 to 46 °C, with two exceptions.
the maximum cell concentration in ln CFU/g; μmax is the max- At 22 °C, all three models were only able to be fitted to one repli-
imum specific growth rate in ln CFU/h; v is the rate of increase cate, and at 13 °C, the Huang model was only able to be fitted to
in the limiting substrate, assumed to be equal to μmax ; h 0 is equal two replicates.
to μmax λ; λ is the duration of the lag phase in hours.
The parameter h0 is recommended to be constant when the Secondary models
pre-inoculation history of the bacterial cells is identical (Baranyi To characterize the effect of temperature on growth parame-
& Roberts, 1994). Because of this, initially, four parameters of the ters estimated by the three primary growth models, average es-
Baranyi model were estimated: y0 , ymax , μmax , h0 . The average h0 timated μmax and λ values for each model were plotted with
value, h 0 , was then calculated and the other three parameters were respect to isothermal temperature profile, respectively. The modi-
estimated again with the h0 term in Equation (1) and (2) fixed at fied Ratkowsky secondary model (Ratkowsky, Lowry, McMeekin,
h 0 (Juneja et al., 2007). Stokes, & Chandler, 1983) was fitted to the average maximum spe-
The Huang model (Huang, 2008) was also used to characterize cific growth rates (μmax ) for each primary model:
growth of B. cereus in cooked pasta. It is characterized by the 
following equations: μmax = a (T − Tmin )2 1 − exp [b (T − Tmax )] (8)
  
yt = y0 + ymax − ln e y0 + e ymax − e y0 e μmax B(t ) (3) where T represents the temperature in °C; Tmin and Tmax are

Food Microbiology &

theoretical growth limits in °C; a and b are regression coefficients.
where The hyperbolic function described by Zwietering, De Koos,

Safety
Hasenack, De Witt, and Van’t Riet (1991) was fitted to lag times
1 1 + e −α(t −λ)
B(t ) = t + ln (4) with respect to temperature for each primary model:
α 1 + e αλ
p
λ = e T−q (9)
yt represents the cell concentration in ln CFU/g at time t; y0 rep-
resents the initial cell concentration in ln CFU/g; ymax represents where T represents the temperature in °C; p is a parameter that
the maximum cell concentration in ln CFU/g; μmax is the max- accounts for the decrease in lag time as temperature increases; q
imum specific growth rate in ln CFU/h; λ is the duration of the represents the temperature (°C) at which lag time is infinite.
lag phase in hours. An α value of four was used as recommended Secondary models were fitted using MATLAB (Version
by Huang (2013). R2017a; Mathworks) nonlinear least squares curve fitting tool-
The modified Gompertz model is an empirical growth model box with trust-region algorithm.
represented by the following equation (Gibson, Bratchell, &
Roberts, 1987):
Goodness of fit statistics
yt = A + C exp (− exp (−B (t − M))) Goodness of fit statistics of the primary and secondary models


e ·μ  were used to compare the performance of each model. Perfor-
max
= A + C exp −exp − (t − λ) + 1 (5) mance was assessed using accuracy and bias factor (Ross, 1996),
C root mean square error (RMSE; Chai & Draxler, 2014), sum of
squared errors of prediction (SSE), and coefficient of determina-
where yt represents the cell concentration in ln CFU/g at time t;
tion (R2 ). The statistics are represented by the following equations:
A represents the asymptotic log count as t decreases indefinitely,
equal to ln (xmax /x0 ); B represents the growth rate at M, where M
 
is the time at which the growth rate is at a maximum in hours; 
log

GTpredicted

GTobserved
C is the asymptotic log count as t increases indefinitely; λ is the Accuracy factor (Af ) = 10 n (10)
duration of the lag phase in hours; μmax is the maximum specific
growth rate in ln CFU/hr.
 
 GTpredicted
The maximum specific growth rate was estimated from the log GT
observed
modified Gompertz formula (Equation (5)) with the following Bias factor (Bf ) = 10 n (11)

Vol. 0, Iss. 0, 2019 r Journal of Food Science 3

 Predictive model for growth of bacillus cereus in pasta . . .


1 n (PE) for each observation with log10 CFU/g units. A PE of 0 was
RMSE = e2 (12) considered a perfect prediction, although positive PE values were
n i =1 i
considered fail-dangerous and negative PE values were consid-
ered fail-safe. Prediction error values between −1.0 and 0.5 were
n
SSE = e2 (13) considered “acceptable” (Mishra, Guo, Buchanan, Schaffner, &
i =1 i Pradhan, 2017).
where GTpredicted represents the predicted generation time (hr),
GTobserved represents the observed generation time (hr), n repre- Results and Discussion
sents sample size, and ei represents model prediction error at the
ith term given by the predicted value minus the observed value. Primary models
Naturally occurring B. cereus spores were not found in uninoc-
Tertiary models ulated pasta samples. B. cereus growth from spores in cooked pasta
was not observed at 10 and 49 °C during a storage period of 28
The differential growth model proposed by Baranyi and Roberts
and 4 days, respectively. B. cereus growth data in cooked pasta from
(1994) can be expressed by the following two first-order differen-
13 to 46 °C were well fitted by the three primary growth mod-
tial equations:
els used in this study (Table 1). Representative estimated growth
dy 1   curves of the models are compared to observed growth values in
= μmax (T(t )) 1 − e (y(t )−ymax )
(14) Figure 2. No or limited growth of B. cereus was observed in pasta
dt 1 + e −Q(t )
samples stored at 10 and 49 °C. At 10 °C, the bacterial population
increased by an average (n = 3) of 0.25 ± 0.09 log CFU/g after 35
dQ days of storage. At 49 °C, the B. cereus population increased by an
= μmax (T(t )) (15) average (n = 3) of 0.13 ± 0.17 log CFU/g after 96 hr of storage.
dt
The Huang model (Equation (3) and (4)) was fitted to growth
where y(0) = y0 and Q(0) = ln(Q0 ) are the initial conditions. data with α = 4 as recommended by Huang (2013). Originally,
These equations were solved by utilizing the fourth-order Runge– α = 25 was recommended to allow the growth curve to quickly
Kutta method in MATLAB. Once solved, the equations provided and smoothly transition from stationary to exponential phase and
a prediction of microbial growth based on dynamic temperature accurately predict lag phase duration (LPD; Huang, 2008, 2010).
profiles displayed in Figure 1. This value was arbitrarily chosen and lag times >28 hr were re-
quired to be converted to days in order for the model to function
Validation properly. It was found that with α = 4, lag times up to 7.3 days
Acceptable prediction zone (APZ) analysis (Oscar, 2005) was (175 hr) could be analyzed properly by the Huang model with-
performed to assess the growth predictions made by the Baranyi out significantly affecting the capability of accurately predicting
model for each dynamic temperature profile. Predicted values were the other growth parameters (for example, μmax ; Huang, 2013).
subtracted from observed values to generate the prediction error This was an especially important consideration in the 13 °C is
Food Microbiology &
Safety

60 60
6.5 h
6h
5h+1.5h
50 9h 50 4h+2.5h
12 h 3h+3.5h
Temperature (oC)

Temperature ( oC)

15 h 2.5h+4h
40 40 2h+4.5h
18 h
1.5h+5h
21 h
30 30

20 20

10 10

0 0
0 3 6 9 12 15 18 21 24 0 1 2 3 4 5 6 7
Time (h) Time (h)

Figure 1–Cooling profiles for (A) single-rate exponential cooling, and (B) biphasic linear cooling rate experiments.

Table 1–Modified Ratkowsky model parameters and goodness of fit statistics for each primary growth model used in this study.

Primary model aa ba Tmin b Tmax b RMSEc R2 d

Baranyi 0.0040 (0.0016, 0.006541) 0.0821 (0.0146, 0.1497) 8.36 (5.05, 11.66) 49.92 (48.39, 51.46) 0.075 0.994
Gompertz 0.0048 (−0.0004, 0.0100) 0.0716 (−0.0212, 0.1644) 7.52 (2.02, 13.02) 48.81 (47.10, 50.52) 0.140 0.978
Huang 0.0051 (−0.0032, 0.0134) 0.0616 (−0.0590, 0.1822) 8.99 (2.20, 15.79) 49.11 (46.63, 51.59) 0.160 0.966
The 95% confidence bounds are presented for each estimated parameter value.
a
a and b are regression coefficients.
b
Tmin and Tmax are theoretical growth limits in °C.
c
RMSE is the root mean square error.
d 2
R is the coefficient of determination.

4 Journal of Food Science r Vol. 0, Iss. 0, 2019

Predictive model for growth of bacillus cereus in pasta . . .

o
46 C o
10 10
43 C Figure 2–Observed growth of B. cereus in cooked pasta
stored at different temperatures and fitted primary
8 8 growth models.
log CFU/g

log CFU/g
6 6

4 4
Observed Observed
2 Baranyi Baranyi
2
Gompertz Gompertz
Huang Huang
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (h) Time (h)

o o
40 C 37 C
10 10

8 8

log CFU/g
log CFU/g

6 6

4 4
Observed Observed
2 Baranyi 2 Baranyi
Gompertz Gompertz
Huang Huang
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (h) Time (h)

34 oC
10 31 oC
10

8 8
log CFU/g

6
log CFU/g

4 4
Observed Observed
2 Baranyi Baranyi
2
Gompertz Gompertz
Huang Huang
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (h) Time (h)
o
28 oC 25 C
10 10

Food Microbiology &

8 8
log CFU/g
log CFU/g

6 6

Safety
4 4
Observed Observed
Baranyi 2 Baranyi
2
Gompertz Gompertz
Huang Huang
0 0
0 5 10 15 20 25 0 5 10 15 20 25
Time (h) Time (h)

o
22 C 19 oC
10 10

8 8
log CFU/g
log CFU/g

6 6

4 4
Observed Observed
Baranyi 2 Baranyi
2
Gompertz Gompertz
Huang Huang
0 0
0 5 10 15 20 25 30 0 10 20 30 40 50
Time (h) Time (h)

o
16 C 13 oC
10 10

8 8
log CFU/g
log CFU/g

6 6

4 4
Observed Observed
2 Baranyi 2 Baranyi
Gompertz Gompertz
Huang Huang
0 0
0 10 20 30 40 50 60 70 80 90 100 0 20 40 60 80 100 120 140 160 180
Time (h) Time (h)

Vol. 0, Iss. 0, 2019 r Journal of Food Science 5

 Predictive model for growth of bacillus cereus in pasta . . .
1 8
0.995
0.99
R2
0.985
0.98
Baranyi
0.975 Gompertz
Huang
0.97
13 16 19 22 25 28 31 34 37 40 43 46
o
Temperature ( C)
0.8
RMSE 0.6
0.4
0.2
0
13 16 19 22 25 28 31 34 37 40 43 46
Figure 3–Mean (n = 3) goodness of fit statistics for each primary growth model’s performance at each isothermal growth temperature.
othermal growth curve (Figure 2), which had an estimated lag
time of approximately 42 hr.
Figure 3 displays average fit statistics, graphically, for the three
primary growth models over the plotted temperatures. All three
models performed exceptionally well (R2 > 0.97, SSE < 8.0,
RMSE < 1.1), especially at intermediate temperatures (22 to
37 °C). This is likely due to the fact that B. cereus has an ideal
Food Microbiology &
growth temperature range in this temperature interval (El-Arabi &
Griffiths, 2013), making growth less variable and more predictable.
Safety
The modified Gompertz model performed best at extreme
temperatures.
All three primary growth models used in this study have shown
acceptable ability at analyzing bacterial growth. The Baranyi and
Huang models are mechanistic models that attempt to account for
biological factors that bacteria encounter during adaptation and
growth (Baranyi & Roberts, 1994; Huang, 2008). This allows the
models to better characterize the transition from lag phase to ex-
ponential phase. For these reasons, mechanistic models are often
preferred when predicting bacterial growth in dynamic condi-
tions. Empirical models, such as the modified Gompertz model,
struggle with growth data with no clear stationary phase (Juneja,
Melendres, Huang, Subbiah, & Thippareddi, 2009). Due to the
fact that the Baranyi model performed well in primary modeling
and that it is well equipped to predict bacterial growth in dy-
namic conditions, it was chosen for tertiary modeling (Table 1).
Although the modified Gompertz and Huang models performed
well under isothermal conditions, they are not designed to predict
growth in dynamic conditions, so they were not considered for
tertiary modeling (Table 1).
The Baranyi primary growth model contains a term, h0 , that
accounts for the physiological state of bacteria during growth
(Baranyi & Roberts, 1994). The model assumes that this value is
constant for a given bacterium in a given growth substrate. To
account for this, the Baranyi model was fitted to each isothermal
growth profile, and h0 was estimated for each. Estimated h0 values
6 Journal of Food Science r Vol. 0, Iss. 0, 2019
Baranyi
Gompertz
6 Huang
SSE
4
2
0
13 16 19 22 25 28 31 34 37 40 43 46
Temperature (o C)
Baranyi
Gompertz
Huang
Temperature (o C)
ranged from 0.51 to 7.26 with an average, h 0 , of 3.46. The Baranyi
model was then re-fitted to each isothermal growth profile with
the h0 term fixed at 3.46. Values for h0 are highly dependent on the
type of organism being studied and the growth medium (Juneja
et al., 2007). Baranyi, Robinson, Kaloti, and Mackey (1995) re-
ported an h 0 of 3.20 for B. thermosphacta grown in broth. Juneja
et al. (2018) reported an h 0 of 4.80 for B. cereus grown in cooked
beans. In another study (Juneja et al., 2019), an h 0 of 4.80 was
reported for B. cereus in cooked rice.
Secondary models
Table 1 shows the estimated parameters and goodness of fit
statistics of the modified Ratkowsky secondary model for each
primary growth model. For all three primary models, the R2 values
were greater than 0.96, with the Baranyi model having the high-
est at 0.99. The modified Ratkowsky best fitted the μmax values
generated by the Baranyi model with respect to RMSE as well.
Although the Baranyi model was the only primary growth model
used for dynamic (nonisothermal temperature) modeling in this
study, the parameters of the modified Ratkowsky model are in-
cluded for the other two models because of their extensive use
in predictive food microbiology and online modeling databases
such as ComBase Predictor, USDA PMP, and USDA Integrated
Pathogen Modeling Program (IPMP).
The minimum and maximum temperatures for growth (Tmin
and Tmax ) predicted by the modified Ratkowsky model for each
primary model are displayed in Table 1. These temperatures are
theoretical cutoff temperatures, below or above, growth would
not occur (Zwietering, De Wit, & Notermans, 1996). Pre-
dicted Tmin and Tmax ranged from 7.52 to 8.99 °C and 48.81 to
49.92 °C, respectively. These values are similar to those predicted
by the primary growth models for B. cereus in cooked beans and
rice (Juneja et al., 2018, 2019). Experimental growth in cooked
pasta only occurred between 13 to 46 °C. Theoretical Tmin and
Tmax values are commonly lower and higher, respectively, than
Predictive model for growth of bacillus cereus in pasta . . .

Table 2–Hyperbolic function parameters and goodness of fit statistics for each primary growth model used in this study.

Primary model pa qb RMSEc R2d

Baranyi 28.79 (25.35, 32.24) 4.71 (3.69, 5.73) 0.691 0.995
Gompertz 44.62 (31.18, 58.07) 1.15 (-2.56, 4.86) 2.695 0.986
Huang 32.28 (25.79, 38.77) 4.41 (2.63, 6.18) 1.518 0.986
The 95% confidence bounds are presented for each estimated parameter value.
a
p is a parameter that accounts for the decrease in lag time as temperature increases.
b
q represents the temperature (°C) at which lag time is infinite.
c
RMSE is the root mean square error.
d 2
R is the coefficient of determination.

Table 3–Concentration of B. cereus in cooked pasta before and after each dynamic cooling profile.

Final concentration observed

Initial concentration (log CFU/g, (log CFU/g, mean ± standard
Cooling profile mean ± standard deviation) deviation)
Exponential (54.5 to 7.2 °C) 6 hr 2.09 ± 0.07 2.48 ± 0.08
9 hr 2.53 ± 0.12 3.67 ± 0.10
12 hr 2.41 ± 0.09 4.66 ± 0.16
15 hr 2.00 ± 0.07 5.18 ± 0.27
18 hr 2.55 ± 0.15 6.00 ± 0.09
21 hr 2.41 ± 0.10 6.96 ± 0.14
Biphasic linear 54.5 to 27 °C (6.5 hr) 2.36 ± 0.06 4.61 ± 0.21
54.5 to 27 °C (5 hr) + 27 to 7.2 °C (1.5 hr) 2.16 ± 0.18 3.52 ± 0.35
54.5 to 27 °C (4 hr) + 27 to 7.2 °C (2.5 hr) 2.34 ± 0.18 3.55 ± 0.08
54.5 to 27 °C (3 hr) + 27 to 7.2 °C (3.5 hr) 2.42 ± 0.05 3.16 ± 0.19
54.5 to 27 °C (2.5 hr) + 27 to 7.2 °C(4 hr) 2.37 ± 0.06 3.12 ± 0.19
54.5 to 27 °C (2 hr) + 27 to 7.2 °C (4.5 hr) 2.72 ± 0.40 2.86 ± 0.08
54.5 to 27 °C (1.5 hr) + 27 to 7.2 °C (5 hr) 2.45 ± 0.17 2.75 ± 0.14

the true minimum and maximum temperatures supporting a bac- temperature range. These observations of high growth are ex-
terium’s growth (McKellar & Delaquis, 2011). pected because the ideal temperature range for B. cereus is 30 to
A hyperbolic function was utilized to predict the effect of tem- 40 °C (El-Arabi & Griffiths, 2013).
perature on LPD for each primary growth model. The predicted For single-rate exponential cooling profiles, a Baranyi model

Food Microbiology &

function parameters and goodness of fit statistics are displayed in estimated growth curve was compared to observed growth values
Table 2. The hyperbolic function fitted LPD values generated by over the course of each exponential cooling profile (Figure 4).
all three primary models well, with R2 values >0.98 and RMSE

values <2.7. The goodness of fit parameters corresponding to the
Baranyi model were better than those of modified Gompertz and
Huang models (Table 2).
In predictive microbiology, it is challenging to predict LPD
because of its strong dependence on the temperature history of
the bacteria being observed (Mishra et al., 2017). In this study,
B. cereus spores were heat shocked at 80 °C for 10 min prior to
experimentation. Because of this, the presented models might not
be accurate at predicting LPD of bacteria in cooked pasta that did
not encounter the same temperature history (Juneja et al., 2018).
This problem does not arise for maximum specific growth rates
predicted as part of this study.
Validation using dynamic temperature profiles
A tertiary model generated by the Baranyi model was validated
using single rate (exponential) and biphasic linear cooling profiles
(Figure 1). Experimental growth results for each cooling profile
are displayed in Table 3. For the exponential cooling profiles, bac-
terial growth (log10 CFU/g) increased as cooling time increased.
Pasta cooled over 21 hr contained an average of 4.48 log10 CFU/g
more B. cereus than pasta cooled over 6 hr. For biphasic linear
cooling profiles, growth increased with longer food-holding time
in the 54.5 to 27 °C temperature range. Pasta that spent 6.5 hr in
that temperature range contained on average 1.86 log10 CFU/g
more B. cereus than pasta with 1.5 hr spent in the 54.5 to 27 °C
Safety
Predicted and observed generation time values were compared
for each exponential cooling profile, and accuracy and bias factor
statistics were generated. The accuracy factor and bias factor for
the single rate exponential cooling validation experiment were
1.06 and 0.98, respectively. A perfect predictive model would ex-
hibit accuracy and bias factors that both equal 1, but this is rarely
seen. Accuracy factor is expected to increase by up to 0.10 to 0.15
for every additional variable introduced to the model (Ross, Dal-
gaard, & Tienungoon, 2000), so with one variable in the proposed
dynamic model, accuracy factor values up 1.10 to 1.15 would in-
dicate “acceptable” model performance. A bias factor range of
0.70 to 1.15 has been recommended to indicate “acceptable”
model prediction performance (Ross, 1996; Ross et al., 2000).
The accuracy and bias factor results presented above show good
performance of the dynamic model. The bias factor results indi-
cate that the generated dynamic model showed a slight tendency
to over-predict growth, due to the bias factor range being less
than 1.
Equation (14) and (15) were solved according to the different
exponential and biphasic linear cooling profiles to estimate bacte-
rial growth after cooling. Predicted results were compared to final
observed growth numbers in Table 3 and prediction errors were
plotted for each dynamic cooling profile (Figure 5). For exponen-
tial cooling profiles, all 24 observations fell within the APZ of −1
to 0.5. For the biphasic linear cooling profiles, all 28 observations
fell within the APZ. These results further validate the generated
tertiary model.

Vol. 0, Iss. 0, 2019 r Journal of Food Science 7

 Predictive model for growth of bacillus cereus in pasta . . .

6h 9h
7 60 7 60
Predicted Predicted
Observed 50 Observed
6 50

B. cereus (Log CFU/g)

6

B. cereus (Log CFU/g)

Temperature Temperature

Temperature ( C)

Temperature ( C)
40

o
40

o
5 5
30 30
4
4
20 20
3
10 3
10
2
0 2 0
0 1 2 3 4 5 6 0 2 4 6 8
Time (h) Time (h)

12 h 15 h
7 60 7 60
Predicted Predicted
Observed 50 Observed 50
6 6
B. cereus (Log CFU/g)

B. cereus (Log CFU/g)

Temperature Temperature

Temperature ( C)

Temperature ( C)
40 40

o
5 5
30 30
4 4
20 20

3 3
10 10

2
2 0 0
0 2 4 6 8 10 12 0 5 10 15
Time (h) Time (h)

18 h 21 h
7 60 8 60

50 7 50
6
B. cereus (Log CFU/g)

B. cereus (Log CFU/g)

Temperature ( C)

Temperature ( C)
40 6 40
o

o
5 Predicted Predicted
Observed 30 5 Observed 30
4 Temperature Temperature
20 4 20
Food Microbiology &

3
10 3 10
Safety

2 0 2 0
0 5 10 15 0 5 10 15 20 25
Time (h) Time (h)

Figure 4–Baranyi-predicted and mean (n = 4) observed B. cereus growth behavior during the single-rate exponential cooling experiment. The observed
temperature during each experiment is illustrated with a dashed line.

0.5 0.5
Prediction error

Prediction error

0 0

-0.5 -0.5

-1 -1

6h 9h 12h 15h 18h 21h 6.5h 5+1.5h 4.+2.5h 3+3.5h 2.5+4h 2+4.5h 1.5+5h
Exponential cooling profile Biphasic linear cooling profile

Figure 5–Acceptable prediction zone analysis (APZ) for (A) single-rate exponential cooling, and (B) biphasic linear cooling. X-axis labels refer to
different cooling profiles mentioned in Figure 1.

Conclusions growth conditions. The presented model will allow the retail food
The Baranyi primary model and modified Ratkowsky sec- industry to determine proper cooling steps of cooked pasta to pre-
ondary model provided a dynamic model that was capable of vent outgrowth of B. cereus. It will also allow users to accurately
accurately predicting B. cereus growth in cooked pasta in dynamic predict B. cereus growth during unforeseen processing temperature

8 Journal of Food Science r Vol. 0, Iss. 0, 2019

Predictive model for growth of bacillus cereus in pasta . . .
deviations. The presented model was generated to predict B. cereus
growth in pasta that does not contain any other ingredients (for
example, spice, vegetables, and so on), so further studies that de-
termine the behavior of B. cereus in pasta supplemented with other
ingredients need to be conducted to supply an adequate predictive
model for pasta products that contain multiple ingredients.
References
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., Robinson, T., Kaloti, A., & Mackey, B. (1995). Predicting growth of Brochothrix
thermosphacta at changing temperature. International Journal of Food Microbiology, 27, 61–75.
Chai, T., & Draxler, R. R. (2014). Root mean square error (RMSE) or mean absolute error
(MAE)? Arguments against avoiding RMSE in the literature. Geoscientific Model Development,
7, 1247–1250.
Eglezos, S., Huang, B. X., Dykes, G. A., & Fegan, N. (2010). The prevalence and concentration
of Bacillus cereus in retail food products in Brisbane, Australia. Foodborne Pathogen and Disease,
7, 867–870.
El-Arabi, T. F., & Griffiths, M. W. (2013). Bacillus cereus. In M. E. Potter & J. G. Morris (Eds.),
Foodborne Infections and Intoxications (4th ed., pp. 401–407). Amsterdam: Academic Press.
Fangio, M. F., Roura, S. I., & Fritz, R. (2010). Isolation and identification of Bacillus spp. and
related genera from different starchy foods. Journal of Food Science, 75, M218−M221.
Granum, P. E. (1994). Bacillus cereus and its toxins. Journal of Applied Bacteriology, 76(Sppl.),
S61−S66.
Granum, P. E. (2005). Bacillus cereus. In P. M. Fratamico, A. K. Bhunia, & J. L. Smith (Eds.),
Foodborne pathogens: Microbiology and molecular biology (pp. 409–419). Norfolk, UK: Caister
Academic Press.
Gibson, A. M., Bratchell, N., & Roberts, T. (1987). The effect of sodium chloride and tem-
perature on the rate and extent of growth of Clostridium botulinum type A in pasteurized pork
slurry. Journal of Applied Microbiology, 62, 479–490.
Gibson, A. M., Bratchell, N., & Roberts, T. (1988). Predicting microbial growth: Growth
responses of salmonellae in a laboratory medium as affected by pH, sodium chloride and
storage temperature. International Journal of Food Microbiology, 6, 155–178.
Health Protection Agency. (2009). Guidelines for assessing the microbiological safety of ready-to-eat
foods placed on a market. London: Health Protection Agency.
Huang, L. (2008). Growth kinetics of Listeria monocytogenes in broth and beef frankfurters—
determination of lag phase duration and exponential growth rate under isothermal conditions.
Journal of Food Science, 73(5), E235−E242.
Huang, L. (2010). Growth kinetics of Escherichia coli O157:H7 in mechanically-tenderized beef.
International Journal of Food Microbiology, 140, 40–48.
Huang, L. (2013). Optimization of a new mathematical model for bacterial growth. Food Control,
32, 283–288.
Juneja, V. K., Melendres, M. V., Huang, L., Gumudavelli, V., Subbiah, J., & Thippareddi,
H. (2007). Modeling the effect of temperature on growth of Salmonella in chicken. Food
Microbiology, 24, 328–335.
Juneja, V. K., Melendres, M. V., Huang, L., Subbiah, J., & Thippareddi, H. (2009). Mathematical
modeling of growth of Salmonella in raw ground beef under isothermal conditions from 10 to
45 °C. International Journal of Food Microbiology, 131, 106–111.
Juneja, V. K., Mishra, A., & Pradhan, A. K. (2018). Dynamic predictive model for growth of
Bacillus cereus from spores in cooked beans. Journal of Food Protection, 81, 308–315.
Juneja, V. K., Golden, C. E., Mishra, A., Harrison, M. A., Mohr, T., & Silverman, M. (2019).
Predictive model for growth of Bacillus cereus during cooling of cooked rice. International
Journal of Food Microbiology, 290, 49–58.
Kramer, J. M., & Gilbert, R. J. (1989). Bacillus cereus and other Bacillus species. In M. P. Doyle
(Ed.), Foodborne bacterial pathogens (pp. 21−70). New York: Marcel Dekker, Inc.
Lee, N. A., Chang, H. G., Kim, H. P., Kim, H. S., & Park, J. H. (2006). Toxicity of 5 Bacillus
cereus enterotoxins in human cell lines and mice. Food Science & Biotechnology, 15, 458–461.
McKellar, R. C., & Delaquis, P. (2011). Development of a dynamic growth–death model for
Escherichia coli O157:H7 in minimally processed leafy green vegetables. International Journal of
Food Microbiology, 151, 7–14.
McMeekin, T. A., Olley, J., & Ross, T. (1993). Predictive microbiology: Theory and application.
Taunton, UK: John Wiley & Sons Ltd.
Mishra, A., Guo, M., Buchanan, R. L., Schaffner, D. W., & Pradhan, A. K. (2017). Development
of growth and survival models for Salmonella and Listeria monocytogenes during non-isothermal
time-temperature profiles in leafy greens. Food Control, 71, 32–41.
Oscar, T. E. (2005). Validation of lag time and growth rate models for Salmonella Typhimurium:
Acceptable prediction zone method. Journal of Food Science, 70(2), M129−M137.
Ratkowsky, D., Lowry, R., McMeekin, T., Stokes, A., & Chandler, R. (1983). Model for
bacterial culture growth rate throughout the entire biokinetic temperature range. Journal of
Bacteriology, 154, 1222–1226.
Ross, T. (1996). Indices for performance evaluation of predictive models in food microbiology.
Journal of Applied Microbiology, 81, 501–508.
Ross, T., Dalgaard, P., & Tienungoon, S. (2000). Predictive modelling of the growth and survival
of Listeria in fishery products. International Journal of Food Microbiology, 62, 231–245.
Wijnands, L. M., Dufrenne, J. B., Rombouts, F. M., in’t Veld, P. H., & van Leusden, F. M. (2006).
Prevalence of potentially pathogencic Bacillus cereus in food commodities in The Netherlands.
Journal of Food Protection, 69, 2587–2594.
Zwietering, M., De Koos, J., Hasenack, B., De Witt, J., & Van’t Riet, K. (1991). Modeling
of bacterial growth as a function of temperature. Applied and Environmental Microbiology, 57,
1094–1101.
Zwietering, M., De Wit, J., & Notermans, S. (1996). Application of predictive microbiology
to estimate the number of Bacillus cereus in pasteurised milk at the point of consumption.
International Journal of Food Microbiology, 30, 55−70.
Zwietering, M., Jongenburger, I., Rombouts, F., & Van’t Riet, K. (1990). Modeling of the
bacterial growth curve. Applied and Environmental Microbiology, 56, 1875–1881.
Food Microbiology &
Safety
Vol. 0, Iss. 0, 2019 r Journal of Food Science 9