Transcript What is modelled?

Predictive microbiology
Tom Ross
Food Safety Centre, University of Tasmania and
International Commission on Microbiological Specifications for Foods
A question…
• if I left a piece of chicken at 10°C for 6
hours, would that allow Salmonella to grow to
dangerous levels?
• would the shelf life be greatly reduced?
Seafast Symposium, Bogor, December, 2009
‘new’ microbial food safety management
• science-based
• ‘farm-to-fork’
• relies on being able to estimate changes in numbers
of pathogens from farm-to-fork, i.e.
• is quantitative
Seafast Symposium, Bogor, December, 2009
The “ICMSF Equation”
Ho - ∑R + ∑I ≤ PO / FSO
Initial Contamination level less
the sum of reductions (e.g. dilution, inactivation) plus
the sum of increases (e.g. recontamination, growth)
should not exceed
the Performance Objective (or Food Safety Objective)
Seafast Symposium, Bogor, December, 2009
HACCP
HACCP, complemented by GMP, is almost universally
endorsed as the most rational approach to the
production of safe food
Sooner or later, if you do HACCP properly, you end up
asking some hard questions . . .
Seafast Symposium, Bogor, December, 2009
HACCP: setting critical limits
How much control is needed? e.g. what are:
– the critical times and temperatures of processes or steps
– appropriate product formulations for desired safety (and shelf life)
– storage and packaging needs
that are required to achieve control?
… and, if control is lost
– how much did the risk increase?
– could control be ‘regained’ and, if so,
– how much reprocessing/storage would be required to return
quality/safety to an acceptable level?
Seafast Symposium, Bogor, December, 2009
managing microbial food safety and quality
• effects of microorganisms are related to their number
– risk of illness increases with number of pathogens ingested
– quality decreases as number of spoilage organisms
increases
• we need to know about numbers of microorganisms
in the food, and how they change over time
Seafast Symposium, Bogor, December, 2009
microbial ecology of foods
microbes in foods can:
•
•
•
grow,
survive,
die
but these processes are not instantaneous and the amount of
growth or death, or whether survival occurs, depend on:
• food composition and additives,
• other microbes in the food,
• processing steps,
• storage conditions, etc. and
• time
Seafast Symposium, Bogor, December, 2009
microbial ecology of foods is predictable
• collectively, responses to these factors constitute the
ecology of the microorganism in the food
• the interactions and effects can be complex but are
predictable, and can be described and quantified
• this is the domain of predictive microbiology
Seafast Symposium, Bogor, December, 2009
Overview
• what is predictive microbiology?
• what can it do?
• what can’t it do ?
• where are the resources?
• how is it being used?
Predictive Microbiology - concepts
• microorganisms react reproducibly to environmental
conditions
– the fundamental premise is that microorganisms can’t think, so
that they behave reproducibly (or “predictably”) in ways
dictated by their environment.
thus
– if we can measure their environment, we can predict what they
will do and how quickly they will do it.
Seafast Symposium, Bogor, December, 2009
Predictive Microbiology - concepts
• in foods, there is a small number of environmental factors
that determine microbial growth rate, namely:
– temperature
– pH
– water activity
• for some foods this works, but for processed foods its
probably an oversimplification, so
– other factors sometimes need to be considered: e.g. organic
acid type and level, nitrite, gaseous atmosphere, smoke
compounds, other microbes in the food
Seafast Symposium, Bogor, December, 2009
Predictive Microbiology - concepts
i.e., it is assumed that the actual food is less important
than the physico-chemical properties of the environment
(i.e. the food and its storage conditions), so long as
basic (microbial) nutritional needs are met and nutrients
are non-limiting
Seafast Symposium, Bogor, December, 2009
Predictive Microbiology - concepts
its also assumed that death rate is affected by
physicochemical conditions in the food, but normally
death rate is most strongly governed by the treatment,
e.g.,
• high temperature
• pressure
• irradiation (UV, gamma etc.)
• electric field strength
Seafast Symposium, Bogor, December, 2009
Building predictive models
How predictive models are made
• based on measurements of changes in microbial
numbers over time and environmental conditions
• data can be from
– deliberately designed studies
– “data mining”
– studies in broths, or in foods
Seafast Symposium, Bogor, December, 2009
How models are made
• data are analysed and patterns of response are identified
• these are expressed in the form of mathematical relationships
• the relationships are turned into equations by finding the best
values of the parameters to describe individual sets of data, i.e
specific to a particular organism - this is the process of ‘model
fitting’
• performance of the model is then evaluated and, if necessary,
model revised or new models constructed
• equations are incorporated into ‘user-friendly’ software
Seafast Symposium, Bogor, December, 2009
10
9
log(cell concentration)
8
7
6
5
4
3
2
increasing environmental stringency
1
0
0
50
100
150
time
Seafast Symposium, Bogor, December, 2009
Uses and limitations
What can we ‘predict’
• amount of microbial growth after time, (from temperature
and product formulation; includes lag time, growth rate)
• reduction in microbial numbers over time, from
knowledge of treatment conditions and product
formulations (includes delay, death rate)
• probability of growth/toxin production
– stability of foods (absolute or within defined time)
Seafast Symposium, Bogor, December, 2009
What is modelled?
• growth rates
– bacteria
– yeasts and moulds
• inactivation (death) rates
–
–
–
–
–
bacteria
yeasts and moulds
viruses
protozoa
microbial toxins?
• probability of growth/toxin formation
– bacteria
– yeasts and moulds
– micro-algae*
Seafast Symposium, Bogor, December, 2009
Example of model performance: E. coli growth
under fluctuating temperature and water activity
8.0
40
35
7.0
6.0
25
5.0
20
15
4.0
10
Observed E. coli Growth
Predicted E. coli Growth
aw
Temperature Profile
3.0
Temperature [°C]
E. coli [LogCFU.ml
-
30
5
2.0
0
0
5
10
15
20
25
30
35
Time (hours)
Seafast Symposium, Bogor, December, 2009
uses of predictive microbiology
models
• “reactive” (assessing what we have)
– identifying CCPs (in Food Safety Programs)
– assessment of food safety implications of a loss of “control”
– assessment of equivalence of processes
– risk assessment/risk management decisions
• “pro-active” (identifying what we could do…)
– product and process design to meet objectives (e.g. current
consumer expectations with safety)
– i.e. “innovation”
Seafast Symposium, Bogor, December, 2009
limitations (i)
models
– don’t tell us whether the pathogen is present
– don’t tell us how many there were to start with
Seafast Symposium, Bogor, December, 2009
limitations (ii)
models:
• often don’t indicate level of confidence that users
should have in the prediction, (or the range of
variability that could be expected)
• don’t usually indicate the limits of their application
Seafast Symposium, Bogor, December, 2009
reasons for limitations:
variability/system complexity
• don’t know which strain
– significant differences between some strains of some pathogens
• don’t always really know the environment
– micro-environments can exist around the cell
• i.e. a problem of not always having enough relevant data to make an
accurate prediction
• nonetheless, in many situations appropriate models can
perform very well
Seafast Symposium, Bogor, December, 2009
where models work well
• defined, controlled systems with few variables
• predicting the relative effects of change
Seafast Symposium, Bogor, December, 2009
“state of the art”
• used by industry
– HACCP, product/process design
• beginning to be used by regulators
• e.g., “Refrigeration Index” in Australia
• often a key part of microbial food safety risk assessment
• several large internet-accessible databases and tools
– e.g. Pathogen Modelling Program, ComBase, Symprevius, SSSP
Seafast Symposium, Bogor, December, 2009
Predictive microbiology resources
Roberts and Jarvis (1983)
• in proposing the concept of predictive microbiology, advocated a
more systematic and cooperative approach to food safety
microbiology within which:
‘the growth responses of the microbes of concern would be
modelled with respect to the main controlling factors such as
temperature, pH and aw’ to generate models that would “enable
predictions of quality and safety to be made speedily with
considerable financial benefit.”
Seafast Symposium, Bogor, December, 2009
a model
Seafast Symposium, Bogor, December, 2009
models and databases: on-line
ComBase (database)
(http://www.combase.cc)
ComBase Predictor (models)
(http://www.combase.cc)
Pathogen Modeling Program (on-line)
(http://pmp.arserrc.gov/PMPOnline.aspx)
Seafood Spoilage Predictor
(http://www.dfu.min.dk/micro/sssp/Home/Home.aspx)
Refrigeration Index
([email protected]; http://www.mla.com.au)
Seafast Symposium, Bogor, December, 2009
ComBase
www.combase.cc
• large, searchable, database of microbiological raw data
• still growing, users can add data
• web-based, free access
• integrates “Food Micromodel” and “Pathogen Modeling
Program” data, and many more
• includes new models in “ComBase Predictor”
Seafast Symposium, Bogor, December, 2009
Data in ComBase
• >40,000 records on growth and survival of pathogens and
spoilage organisms
– ~28,000 records on pathogens
– ~4,000 on spoilage organisms, including
– ‘total spoilage bacteria’ (346)
– ‘bacillus spoilage bacteria’ (65)
– Brocothrix thermosphacta (741)
– enterobacteriaceae (338)
– lactic acid bacteria (701)
– Shewenella putrefasciens (57)
– “spoilage yeast” (44)
– ~22,000 full log-count curves
– ~10,000 growth/death rates
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
Pathogen Modeling Program
FREE DOWNLOAD: http://portal.arserrc.gov/
Seafast Symposium, Bogor, December, 2009
Pathogen Modeling Program
pmp.arserrc.gov/PMPOnline.aspx
• USDA program
• can also be downloaded
• suite of models for
– various pathogen
– growth
– death by various treatments
• part of the predictive microbiology information portal - an on-line
predictive microbiology resource:
portal.arserrc.gov/
Seafast Symposium, Bogor, December, 2009
seafood spoilage and safety predictor
www.dfu.min.dk/micro/sssp/
• predicts growth of bacteria in different fresh and
lightly preserved seafoods
• allows prediction of:
–
–
–
–
rates of spoilage of seafood
shelf life of various seafoods
effect of fluctuating conditions
simultaneous growth of Listeria monocytogenes (a
pathogen) and spoilage bacteria in cold-smoked salmon
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
“refrigeration index”
www.mla.com.au *
• Australian product with regulatory approval for use
under Australian Export Meat Orders
• predicts growth of E. coli (as an indicator of safe
temperature control) from continuous temperature
history using the idea of time-temperature function
integration
* before downloading please contact Mr. Ian Jenson
[email protected]
Seafast Symposium, Bogor, December, 2009
Seafast Symposium, Bogor, December, 2009
‘home made’ software
• there are many more models in the published
literature
• relatively easy to translate these into user-friendly
software tools using spreadsheet software
Seafast Symposium, Bogor, December, 2009
summary
• best microbial food safety/quality systems rely on knowledge of
microbial ecology, not testing
• predictive microbiology models provide condensed,
quantitative, expert knowledge
• models provide ‘decision support’ for many practical
problems/questions and/or an alternative/adjunct to
microbiological testing
• predictive microbiology models are now being used by industry
and regulators to improve productivity and food safety
Seafast Symposium, Bogor, December, 2009
Summary
• correct use of models requires microbiology understanding and
basic mathematical skills but that knowledge is critical to
appropriate application
• users should be aware of the current limits of models
– both in terms of range of application and confidence intervals on
model predictions
Seafast Symposium, Bogor, December, 2009
thank you for your attention,
and for your questions and comments
Seafast Symposium, Bogor, December, 2009