~V
1237_C10.fm Page 303 Wednesday, November 12, 2003 1:11 PM
10
Models —
What Comes after the
Next Generation?
Donald W. Schaffner
CONTENTS
10.1 Introduction
10.2 Cross-Contamination
10.2.1 Issues of Concern
10.2.1.1 Statistical Treatment
10.2.1.2 Additive Nature of Cross-Contamination
10.2.1.3 Factors Influencing Transfer Rate
10.2.1.4 Types of Transfers
10.2.1.5 Multiple Transfers and Complexity
10.2.2 Cross-Contamination Summary
10.3 Inoculum Size Modeling
10.3.1 Clostridium botulinum
10.3.2 Nonspore-Forming Bacteria
10.3.3 Inoculum Size Summary
10.4 Cross-Contamination and Inoculum Size
10.5 Summary
References
10.1 INTRODUCTION
This chapter will highlight two separate and generally unrelated areas of predictive
food microbiology: models for cross-contamination and inoculum size (or models
that consider the initial number of organisms present). These two classes of models
are being included together here, because they represent some of the newer areas
of predictive modeling that are less well developed compared to the more well-
known and established research areas such as growth and inactivation modeling.
This chapter will summarize the current state of research in these two rapidly
evolving areas and will conclude with a short example describing preliminary inves-
tigations into the integration of these two fields of study.
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 304 Wednesday, November 12, 2003 1:11 PM
10.2 CROSS-CONTAMINATION
Predictive models for microbial behavior have traditionally focused on describing
increasing concentrations (as a result of multiplication) and decreasing concentra-
tions (as a result of cell death). A number of lines of inquiry have pointed out the
need for a third class of models that may be required in some cases: cross-contam-
ination models. The three lines of inquiry all come from the interaction of risk
assessment and epidemiology, and are related to three very different microorganisms:
Listeria monocytogenes, Campylobacter jejuni, and food-borne viruses.
L. monocytogenes is a psychrotrophic pathogen that can cause mild illness in
healthy adults and spontaneous abortion in pregnant women. The organism is easily
destroyed by heating but readily recontaminates the cooked product prior to pack-
aging. 25 While this recontamination is known to take place, (and can contribute to
significant disease outbreaks) very few mathematical models are available for micro-
bial risk assessors to use.
The situation with Campylobacter is slightly different. In this case, the organism
is known to cause cross-contamination in a significant number of cases 1 but the
means by which this occurs is not clear. 11 Risk assessment models for Campylobac-
teriosis have incorporated cross-contamination events during final preparation in a
kitchen environment, but as with Listeria risk assessment models, few suitable
models are available.
Finally, it is known that a number of foodborne disease agents (primarily viruses)
can contaminate foods, and in many cases the source of the agent has been an ill
food worker. 12 While quantitative microbial risk assessments have yet to address ill
workers, hand-to-food and other cross-contamination rates using a nonpathogenic
surrogate have been calculated with sufficient detail to be suitable for risk assessment. 7
10.2.1 Issues of Concern
There are a number of issues of particular concern that are important in modeling
cross-contamination — issues that are unique to these sorts of models as compared
to the traditional growth and decline models: appropriate statistical treatment of data,
transfer from one location to another, factors influencing transfer, and the possibility
of multiple transfers.
1 0.2.1 .1 Statistical Treatment
Data on cross-contamination are typically presented as percent transfer, as shown
below:
CFU on target , _ _
— * 100 = percent transfer
CFU on source
The problem arises when multiple observations of the same conditions are to
be combined and reported as an average. It has been shown that when large numbers
of observations are made, the distribution of "percent transfer rates" is distinctly
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 305 Wednesday, November 12, 2003 1:11 PM
"i 1 1 r
25 50 75 100
Percent Transfer
-2-1 1 2
Log Percent Transfer
FIGURE 10.1 Distribution of percent transfer data with linear (left panel) and logarithmic
transformation (right panel).
nonnormal. 7 This point is made clearer when the data are presented visually. Figure
10.1 shows the same data set plotted as number of observations vs. percent transfer
(left panel) and number of observations vs. log 10 percent transfer (right panel). This
same pattern is borne out for many different types of surface-to-surface transfers. 710
Since percent transfer is nonnormally distributed, but log 10 percent transfer is approx-
imately normally distributed, this means data should be log transformed before
averages are calculated. This apparently subtle distinction has important conse-
quences as illustrated below.
Let us assume we have two observed transfer rates of 5 and 50%. If the mean
is calculated arithmetically:
(50 + 5)/2 = 27.5% transfer
Alternatively, using the statistically appropriate log 10 transformed rates leads to
a more complex series of calculations:
(log 0.05 + log 0.50)/2 = (-1.30103 + -0.30105)/2 = -0.80103
Then this number should be converted back to the untransformed percent scale:
1O-0.80103 = 15>8% trans fer
So this simple difference leads to a calculated transfer rate that is over half that
obtained when the statistically incorrect method is used.
10.2.1.2 Additive Nature of Cross-Contamination
Another key feature of any cross-contamination model is that some consideration
should be made of both the source of the contaminant and the destination. If the
source is a contaminated surface, then the number of organisms on that surface must
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 306 Wednesday, November 12, 2003 1:11 PM
be known, since (at least in theory) a more contaminated source will yield a more
contaminated destination. Care must be taken to ensure that the number of organisms
to be added to an already contaminated item is added in a numerically correct manner.
For example, most calculations in a microbial risk assessment will be log 10 CFU
increases or decreases. This would not be correct in cross-contamination. If a food
contained 10 organisms and 100 were added, this would not be a two log 10 increase,
1 + 2 = 3 or 1000 organisms, but 10 + 100 =110 organisms.
10.2.1.3 Factors Influencing Transfer Rate
The next issue that should be addressed would be a consideration of the factors
influencing transfer rate. There are a whole host of factors that may influence transfer,
like source (air, liquid, or solid), the pressure applied (for solid-to-solid transfers),
menstrum effects, contact time, number of organisms present, and surface charac-
teristics. Some of these factors have been investigated for transfer from hands in a
healthcare setting, but little data applicable to food systems have been published.
What has been published in the food and healthcare literature has not been systematic
or comprehensive. 21
1 0.2.1 .4 Types of Transfers
A recent review 10 describes some of the currently available models for recontami-
nation via air, via processing equipment (i.e., biofilms), or via hand contact. These
authors point out that not many available models are directly applicable to the food
industry as most models are developed for aquatic or environmental systems. In
some cases currently available models are contradictory or incompatible. For exam-
ple, competing air recontamination models assume that when the concentration in
the air increases linearly, the concentration in the product increases either linearly 28
or quadratically. 23 The implications of such assumptions obviously have a critical
impact on model predictions.
Models for recontamination that consider the effect of biofilms are quite well
developed, not because of their importance in food processing, but because of their
application in wastewater treatment. Biofilm models can be one-dimensional, or
multidimensional, 16 but the key feature of any biofilm model used for food recon-
tamination is not its dimensionality, but its ability to consider attachment, growth,
and detachment averaged over the food contact surface. 10 Some biofilm models
appropriate for use in food systems have been developed. 31
den Antrekker et al. 10 conclude their review by proposing a schematic for a
general contamination model that is suitable for modeling recontamination via air,
via surfaces, or via hands. This model uses a source, an intermediate phase, and a
product, with transfer rates between source and intermediate phases and intermediate
and product phases that govern the overall transfer to the product.
10.2.1.5 Multiple Transfers and Complexity
The last issue of concern is the modeling of multiple transfers. The food preparation
or handling environment may be such that multiple transfers between many different
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 307 Wednesday, November 12, 2003 1:11 PM
environments may occur. In a processing plant it may be from air to surface, surface
to product, product to clean surface, and finally surface to clean product. In a food
service environment it may be from a contaminated surface to workers' hands, and
then from hands to food. In a home environment it may be from a contaminated
food product to a surface, and then from that surface to another food.
Each of these simple examples would require a series of calculations, using
models that generally do not yet exist. At the same time, it should be realized that
the real world is considerably more complex than these simple examples show, and
that there may be literally dozens of cross-contamination possibilities in even a
simple food process or meal preparation.
10.2.2 Cross-Contamination Summary
The development of mathematical models suitable for describing cross-contamina-
tion events in food production, processing, and preparation is still in its infancy, and
might be likened to the general state of predictive food microbiology in the 1980s.
The past two decades have seen many improvements in the general state of the art
of predictive food microbiology, and there is no reason to believe the next two
decades would not experience similar improvements in cross-contamination model-
ing. This field has attracted the interest of a number of research groups around the
world, and the beginnings of a comprehensive body of work are beginning to emerge.
10.3 INOCULUM SIZE MODELING
Predictive models have traditionally been developed using starting bacterial concen-
tration that may be quite high relative to the levels found in some foods. Modelers
developing models in this way were quite justified in their choice of this approach.
In some cases, inoculum size does not have a significant effect on the response to
be modeled (i.e., growth rate). 5 High initial inoculum size also represents a conser-
vative worst-case approach to modeling, and these high starting concentrations
helped to assure repeatability and simplified some of the considerations about micro-
bial variability, often called "biovariability."
Despite the logic seen in this worst-case approach, modelers have always sought
to improve their models by making them more realistic and representative of real-
world conditions. Also, as predictive food microbiologists' modeling tools and
abilities have improved, their ability to handle more complex models and modeling
techniques have improved concomitantly. As part of this evolutionary improvement
in modeling ability, some modelers have sought to address this shortcoming by
developing models that take initial microbial concentrations into consideration.
10.3.1 Clostridium Botulinum
One of the earliest examples of a predictive model that explicitly acknowledged the
influence of inoculum size were models developed for C. botulinum. nM The authors'
specific objective in this case was to develop models capable of predicting the
probability of toxin formation from a single C. botulinum spore. Later research in
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 308 Wednesday, November 12, 2003 1:11 PM
this same lab also showed an inoculum size effect in fish 3 and poultry 19 systems
inoculated with C. botulinum.
The importance of inoculum size for models for C. botulinum has since been
well documented in the literature. 17 ' 26 ' 2729 ' 30 These C. botulinum models were devel-
oped as an aid to the food industry, but also served to point the way towards a more
mechanistic understanding of populations of C. botulinum spores. The model for
nonproteolytic C. botulinum developed by Whiting and Oriente, 26 for example,
showed that not only did lower population of spores exhibit longer time-to-turbidity,
the variability around that time increased markedly with decreasing inoculum size
as well. This same effect was also seen with proteolytic C. botulinum spores. 27 These
results supported the observations made by others conducting microscopy studies
that there was a marked variability seen in the germination, outgrowth, and lag time
in individual C. botulinum spores observed directly. 4 These apparent interactions
between spores of C. botulinum have also been demonstrated using computer sim-
ulation of different inoculum sizes 30 and seem to be caused by the release of a
signaling molecule into the culture media by germinating spores.
This inoculum phenomenon does not appear to be unique to C. botulinum, as
it has also been observed in Bacillus cereus 9 and Bacillus megaterium 6 using direct
microscopic observation and in Bacillus stearothermophilus 15 indirectly by time to
turbidity.
10.3.2 Nonspore-Forming Bacteria
The effect has also been seen in models for nonspore-forming bacteria, like Bro-
chothrix thermosphacta, ls as well as in nonmodeling research with Salmonella 8 and
L. monocytogenes. 2,2224 These publications show that the effect is most pronounced
when cells are stressed 224 or cultured in inhospitable environments. 8 The response
in vegetative cells has been interpreted by some as being due to death of a proportion
of cells in the inoculum rather than communication, as appears to be the case with
C. botulinum, 22 although others have shown that addition of spent medium from a
stationary -phase culture reduces the variability and length of lag times. 24
10.3.3 Inoculum Size Summary
Clearly modelers have moved beyond predictive models developed using high initial
bacterial concentrations to models using a range of contamination levels. Most of
the effort in this area has focused on models for spore-forming organisms, specifi-
cally C. botulinum. A limited amount of work has also been done with other spore-
forming organisms and vegetative cells. It appears that inoculum size has the most
dramatic effect on the lag time (for vegetative cells) or germination, outgrowth, and
lag time (for spores).
10.4 CROSS-CONTAMINATION AND INOCULUM SIZE
While the two subjects of this chapter do not appear to have much in common,
except for both being aspects of modeling on the "cutting edge," there appears to
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 309 Wednesday, November 12, 2003 1:11 PM
S-H
c/3
c
o
Oh
00
O
•
•
•
• • •
•
• •
•
• •
-
• •
• - -
• •
•
• * %
•
-
w
•
• •
•
• •
*
1
•
I i
•
I
•
•
1
3
2
1
-1
-2
-3
12 3 4 5 6 7
Inoculum Size or Initial Concentration
Log 10 CFU/hand
FIGURE 10.2 Percent transfer as a function of inoculum size.
be some evidence that inoculum size is not only important in traditional growth
modeling, but that it may also be an important consideration in modeling cross-
contamination. Figure 10.2 presents a reanalysis of a portion of the data originally
published by Chen et al. 7 for cross-contamination rate from bare hands contaminated
with Enterobacter aerogenes to lettuce. While the r 2 value (0.49) and a visual
inspection of the plot show that the correlation is not ideal, the effect is highly
significant (p > 10~ 10 ). The regression model for the plot indicates that changing the
starting concentration by 1.5 log 10 CFU will change the log 10 percent transfer rate
by about 1 (i.e., from 10 to 1%). It is also interesting to note that the relationship
between starting concentration and log 10 percent transfer rate is an inverse one, so
that as the starting concentration decreases, the transfer rate increases. This could
have very profound food safety consequences since low levels of pathogens would
have a correspondingly greater ability to transfer. This simple example has been
used to illustrate the exciting and complex nature of predictive food microbiology
at the expanding edges of the discipline.
10.5 SUMMARY
Models for cross-contamination and inoculum size represent areas at the expanding
edge of predictive food microbiology. Both areas characterize modelers' attempts
to make models more useful and representative of microbial behaviors seen in the
real world. Both areas continue to present mathematical, statistical, and method-
ological challenges to those working in the field.
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 310 Wednesday, November 12, 2003 1:11 PM
REFERENCES
1 . Preliminary FoodNet data on the incidence of foodborne illnesses — selected sites,
US, 2001, Mor. Mortal. Wkly. Rep., 51, 325, 2002.
2. Augustin, J.C., Brouillaud-Delattre, A., Rosso, L., and Carlier, V., Significance of
inoculum size in the lag time of Listeria monocytogenes, Appl. Environ. Microbiol.,
66, 1706, 2000.
3. Baker, D.A. and Genigeorgis, C.A., Predicting the safe storage of fresh fish under
modified atmospheres with respect to Clostridium botulinum toxigenesis by modeling
length of the lag phase, /. Food Prot., 53, 131, 1990.
4. Billon, C.M.P., McKirgan, C.J., McClure, P.J., and Adair, C, The effect of temperature
on the germination of single spores of Clostridium botulinum 62A, /. Appl. Microbiol.,
82, 48, 1997.
5. Buchanan, R.L., Smith, J.L., McColgan, C, Marmer, B.S., Golden, M., and Dell, B.,
Response surface models for the effects of temperature, pH, sodium chloride, and
sodium nitrite on the aerobic and anaerobic growth of Staphylococcus aureus 196E,
/. FoodSaf., 13, 159, 1993.
6. Caipo, M.L., Duffy, S., Zhao, L., and Schaffner, D.W., Bacillus megaterium spore
germination is influenced by inoculum size, J. Appl. Microbiol., 92, 879, 2002.
7. Chen, Y., Jackson, K.M., Chea, F.P., and Schaffner, D.W., Quantification and vari-
ability analysis of bacterial cross contamination rates in common foodservice tasks,
J. Food Prot, 64, 72, 2001.
8. Cogan, T.A., Domingue, G., Lappin-Scott, H.M., Benson, C.E., Woodward, M.J., and
Humphrey, T.J., Growth of Salmonella enteritidis in artificially contaminated eggs:
the effects of inoculum size and suspending media, Int. J. Food Microbiol., 70, 131,
2001.
9. Coote, P.J., Billon, C.M.P., Pennell, S., McClure, P.J., Ferdinando, D.P., and Cole,
M.B., The use of confocal scanning laser microscopy (CSLM) to study the germina-
tion of individual spores of Bacillus cereus, J. Microbiol. Methods, 21, 193, 1995.
10. den Antrekker, E., Boom, R.M., Zwietering, M.H., and vanSchothorst, M., Quanti-
fying recontamination through factory environments — a review, Int. J. Food Micro-
biol., 80, 117, 2002.
11. Frost, J. A., Current epidemiological issues in human campylobacteriosis, /. Appl.
Microbiol., 90, 85S, 2001.
1 2. Guzewich, J.J. and Ross, M.P., White paper, Section one: a literature review pertaining
to foodborne disease outbreaks caused by food workers, 1975-1998,
http://vm.cfsan.fda.gov/~ear/rterisk.html, 1999.
13. Jensen, M.J., Genigeorgis, C.A., and Lindroth, S., Probability of growth of Clostrid-
ium botulinum as affected by strain, cell and serologic type, inoculum size and
temperature and time of incubation in a model system, /. Food Saf., 8, 109, 1987.
14. Lindroth, S.E. and Genigeorgis, C.A., Probability of growth and toxin production by
non proteolytic Clostridium botulinum in rockfish stored under modified atmospheres,
Int. J. Food Microbiol., 3, 167, 1986.
15. Llaudes, M., Zhao, L., Duffy, S., and Schaffner, D.W., Simulation and modeling of
the effect of small inoculum size on the time to spoilage by Bacillus stearothermo-
philus, Food Microbiol, 18, 395, 2001.
16. Lu, C, Biswas, P., and Clark, R.M., Simultaneous transport of substrates, disinfectants
and microorganisms in water pipes, Water Res., 29, 881, 1995.
2004 by Robin C. McKellar and Xuewen Lu
1237_C10.fm Page 311 Wednesday, November 12, 2003 1:11 PM
17. Lund, B.M., Quantification of factors affecting the probability of development of
pathogenic bacteria, in particular Clostridium botulinum, in foods, J. Ind. Microbiol.,
12, 144, 1993.
18. Masana, M.O. and Baranyi, J., Growth/no growth interface of Brochothrix thermo-
sphacta as a function of pH and water activity, Food Microbiol, 17, 485, 2000.
19. Meng, J. and Genigeorgis, C.A., Modeling lag phase of nonproteolytic Clostridium
botulinum toxigenesis in cooked turkey and chicken breast as affected by temperature,
sodium lactate, sodium chloride and spore inoculum, Int. J. Food Microbiol, 19, 109,
1993.
20. Montville, R., Chen, Y., and Schaffner, D.W., Glove barriers to bacterial cross-
contamination between hands to food, /. Food Prot., 64, 845, 2001.
21. Montville, R., Chen, Y.H., and Schaffner, D.W., Risk assessment of hand washing
efficacy using literature and experimental data, Int. J. Food Microbiol, 73, 305, 2002.
22. Pascual, C, Robinson, T.P., Ocio, M.J., Aboaba, O.O., and Mackey, B.M., The effect
of inoculum size and sublethal injury on the ability of Listeria monocytogenes to
initiate growth under suboptimal conditions, Lett. Appl. Microbiol, 33, 357, 2001.
23. Radmore, K., Holzapfel, W.H., and Luck, H., Proposed guidelines for maximum
acceptable air-borne microorganism levels in dairy processing and packaging plants,
Int. J. Food Microbiol, 6, 91, 1988.
24. Robinson, T.P., Aboaba, O.O. , Kaloti, A., Ocio, M.J., Baranyi, J., and Mackey, B.M.,
The effect of inoculum size on the lag phase of Listeria monocytogenes, Int. J. Food
Microbiol, 70, 163, 2001.
25. Tompkin, R.B., Control of Listeria monocytogenes in the food-processing environ-
ment, /. Food Prot., 65, 709, 2002.
26. Whiting, R.C. and Oriente, J.C., Time- to- turbidity model for non-proteolytic type B
Clostridium botulinum, Int. J. Food Microbiol, 36, 49, 1997.
27. Whiting, R.C. and Strobaugh, T.P., Expansion of the time-to-turbidity model for
proteolytic Clostridium botulinum to include spore numbers, Food Microbiol, 15,
449, 1998.
28. Whyte, W., Sterility assurance and models for assessing airborne bacterial contami-
nation, /. Parenter. Sci. Technol, 40, 188, 1986.
29. Zhao, L., Montville, T.J., and Schaffner, D.W., Time-to-detection, percent-growth-
positive and maximum growth rate models for Clostridium botulinum 56A at multiple
temperatures, Int. J. Food Microbiol, 11, 187, 2002.
30. Zhao, L., Montville, T.J., and Schaffner, D.W., Computer simulation of Clostridium
botulinum 56A behavior at low spore concentrations, Appl. Environ. Microbiol, 69,
845, 2003.
31. Zwietering, M.H. and Hasting, A.P.M., Modelling the hygienic processing of foods
— a global process overview, Trans IChemE part C, 75, 159, 1997.
2004 by Robin C. McKellar and Xuewen Lu