OBTENTION AND DATA MANAGEMENT IN PREDICTIVE MICROBIOLOGY MODELS Cris6na L.M. Silva 5th September 2012
CBQF – Centro de Biotecnologia e Química Fina, Escola Superior de Biotecnologia,
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
ObjecFves of food industry
PredicFon of shelf life
Control/monitor the growth of microorganisms
6 Product Design Primary Production Industrial Production Product Transport Logistics Trade Home Storage Consumption Disposal DistribuFon chain
7 Product Design Primary Production Industrial Production Product Transport Logistics Trade Home Storage Consumption Disposal DistribuFon chain
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
The challenge
Microorganisms response depends on:
- Intrinsic factors -‐ Extrinsic factors -‐ System dynamics pH aw others T pH humidity salt gas concentraFon others
• microbial interac6on
• natural strains diversity
• history of ini6al popula6on
• complexity of food structure
• interac6on food/microorganism
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
PredicFve microbiology
The use of mathematical models in the description of
Thermal lethality
-‐ D and z values – Bigelow model
Fc =
∫
tp10 Tc −Trefm zm dt 0 ( )/The idea is not even recent!
mathematics
microbiology
statistics
predictiveModel à mathematical expression
i=1,2,...,n (number of experimental runs/observations) j=1,2,...,v k=1,2,...,p
y
i= f(x
ij,
θ
k) + ε
i 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 1000 2000 3000 4000 5000 x y ymodel yexp θ* Precise ? Accurate ? Minimize differencesprecise and accurate description of observations
model adequacy
quality of model parameters
" knowledge of the process
" process effects on product
" control of process variables
Processes Chemical Physical Food Processes Transport Phenomena • heat • mass • momentum Reaction kinetics Properties Modelling mathematical function
variables parameters data regression schemes experimental design design Criterium validation control optimization Objectives quality safety
"
prediction / simulation
"
development of efficient
processes
contribution to
safety
aplication
sigmoidal behaviour
presence of aggregated microorganisms or sub populations more heat (or other stress factor) resistent
inactivation
0 1 2 3 4 5 6 7 8 9 0 500 1000 1500 2000 time(s) lo g N0 1 2 3 4 5 6 7 8 0 20 40 60 80 100 120 140 0 1 2 3 4 5 6 7 8 0 20 40 60 80 100 120 140 Miller (2004) 52.5 ºC Listeria innocua liquid medium 55 ºC 57.5 ºC 60 ºC 62.5 ºC 65 ºC log N time (min)
0 1 2 3 4 5 6 7 8 0 20 40 60 80 100 120 140 52.5 ºC 55 ºC 57.5 ºC 60 ºC 62.5 ºC 65 ºC log N time (min) Miller (2004) Listeria innocua liquid medium
v primary 0 1 2 3 4 5 6 7 8 0 0.5 1 1.5 2 2.5 Time (min) Log C FU /g kinetics parameters
v primary v secondary 0 1 2 3 4 5 6 7 8 0 0.5 1 1.5 2 2.5 Time (min) Log C FU /g parameters pH aw temperature
v primary
v secondary
v terciary - integration of the previous models - software
0 1 2 3 4 5 6 7 8 0 0,5 1 1,5 2 2,5 Time (min) Log C FU /g parameters pH aw temperature
v primary 0 1 2 3 4 5 6 7 8 9 0 500 1000 1500 2000 logN k logN0 logNres L Time (s) empirical fundamental
N0 number of initial viable spore cells Nres number of residual spore cells k maximum inactivation rate L lag or shoulder
v primary First order ( kt) exp N N = 0 − D t N log logN = 0 − ( k t) (1 F )exp( k t) exp F N N 2 1 1 1 0 = − + − − Cerf (1977) Kamau et al. (1990) ( ) ( ) ( )⎟⎟⎠ ⎞ ⎜⎜ ⎝ ⎛ + − + + = t k exp 1 F 1 2 t k exp 1 F 2 log N N log 2 1 1 1 0
D – decimal reduction time
F1 – fraction of inactivated microorganisms k1 e k2– kinetic constants biphasic 0 1 2 3 4 5 6 7 8 0 2 4 6 8 Time (min) LogC FU /g 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 Time (min) Log C F U /m l
v primary Cole et al. (1993) ( )⎟ ⎠ ⎞ ⎜ ⎝ ⎛ − − + − + = σ w t log λ σ 4 exp 1 α w α N log 0 1 2 3 4 5 6 7 8 0 0.5 1 1.5 2 2.5 Time (min) Log C FU /g ( ) ( ) ( ) ( ) ( )( ( )) ( ) ( ) ⎟⎟⎠ ⎞ ⎜⎜ ⎝ ⎛ − + − + − + − + − + = L t k exp 1 L k exp 1 F 1 L t k exp 1 L k exp 1 F log N N log 2 2 1 1 1 1 0 Whiting & Buchanan (1992) distribution of heat sensibility of microbial populations L – lag or shoulder
v primary Baranyi et al. (1993) ( ) ( ) (t 0) N0 N N t t k dt dN = = − = α β ‘lag’ function ‘tail’ function 0 1 2 3 4 5 6 7 8 9 0 500 1000 1500 2000 tempo (s) log N
v primary Baranyi et al. (1993) ( ) ( ) (t 0) N0 N N t t k dt dN = = − = α β ‘lag’ function ‘tail’ function 0 1 2 3 4 5 6 7 8 9 0 500 1000 1500 2000 tempo (s) log N ( ) ( ) ( ) ( )0 exp( k t) Q 1 0 Q 1 t k exp log N N log max max 0 + − + − = ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ Geeraerd et al. (2000) ( ) Q k dt dQ N Q k k dt dN max Q max − = − =
v primary Gompertz 0 1 2 3 4 5 6 7 8 9 0 500 1000 1500 2000 tempo (s) log N Bhaduri et al (1991) Linton et al. (1995, 1996) Xiong et al. (1999) ( ) ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ ⎟⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜⎜ ⎜ ⎜ ⎜ ⎝ ⎛ + − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − = L t 1 N N log e k exp exp N N log logN logN res 0 res 0 0 Logistic Listeria monocytogenes ( ) (k t - L ) exp 1 c logN + = c – constant
reparameterized for inactivation based in Zwitering (1990)
v secondary
Arrhenius
Davey / Arrhenius modified
“Square-root type models”
Ratkowsky et al. (1982) McMeekin et al. (1987) Adams et al. (1991) McMeekin et al. (1992) ) T b(T k = − min ) a (a ) T b(T k = − min w − wmin ) pH (pH ) T b(T k = − min − min ⎟ ⎠ ⎞ ⎜ ⎝ ⎛ = RT E -exp k k 0 a lnk = lnk0 − RTEa 2 W 4 W 3 2 2 1 0 CT TC C a C a C lnk = + + + + ) pH (pH ) a (a ) T b(T
k = − min w − wmin − min
min – minimal value for growth
⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ref a ref exp - ER T1 T1 k k
v tertiary
softwares
Microbial growth Shelf life prediction
Difficul6es in food processes modelling:
" Dynamic processes
" Complexity and heterogeneity of products
Gompertz ( ) ) time ( d N log d ( ) (L t') 1 dt' N N log ) 1 exp( k exp exp 1 ' t L N N log ) 1 exp( k exp ) 1 exp( k N log N log t 0 res 0 res 0 0 ∫ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ + − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ + − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − =
dynamic situation of temperature
⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ref a ref exp - ER T1 T1 k k ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ref T 1 T 1 b exp a L Gill (2011)
Linear ( ) ) time ( d N log d
dynamic situation of temperature
(T Tref) z 1 ref 10 D D = − − ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ −
∫
=
− PT 0 Zref T T ref T dt 10 D 1 010
N
N
approach by Vieira et al. (2002) Cupuaçu nectar
Case studies:
( ) ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ⎛ + − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − ⎟⎟ ⎠ ⎞ ⎜⎜ ⎝ ⎛ − = L t 1 N N log e k exp exp N N log logN logN res 0 max res 0 0 Square-root! ! " kmax = c (T − d) Arrhenius! ! " " " ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ref a ref max k exp - ER T1 T1 k Williams-Landel-Ferry! ! " " ( ) ( )⎟⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ + = min min T -T b T -T a 10 L Arrhenius! " " " " ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − = ref T 1 T 1 b exp a L
c, d !constants! Kref !reaction rate at Tref! Ea !activation energy! a, b !constants! 1! 2! 4! 3! Gill (2011)
log N! time (s)! 0 1 2 3 4 5 6 7 8 9 0 1000 2000 3000 4000 5000 6000 7000 T=52ºC T=56ºC logN time (s) 0 1 2 3 4 5 6 7 8 9 0 100 200 T=60ºC T=64ºC T=68ºC 0 500 1000 1500 2000 2500 3000 320 325 330 335 340 345 T (K) L ( s ) 0 0,05 0,1 0,15 0,2 0,25 0,3 320 325 330 335 340 345 T (K) kma x ( s -1) 0 1 2 3 4 5 6 7 8 9 0 1000 2000 3000 4000 5000 6000 7000 time (s) log N T=52ºC T=56ºC 0 1 2 3 4 5 6 7 8 9 0 100 200 time (s) log N T=60ºC T=64ºC T=68ºC Gompertz" Two-step" L= f(T)! kmax= f(T)! One-step"
Equations 1 and 4 selected"
Tank with water + L. innocua 15 minutes Weibull model Inumeration OzonaFon
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
How to obtain the data
Experiments in broth at various condiFons (pH, T, aw,
[growth inhibitors], etc.)
InoculaFon studies in foods under various condiFons
" Heuris6c sampling
" Experimental design Minimize variance of:
" predicted response " parameter es.mates
Sampling:
" Regression schemes " Analysis of residuals
(
)
[
]
2 n 1 i k ij i n 1 i 2 i y f x , e SSR∑
∑
= = θ − = = Least-squares methodData analysis:
Mathematical
complexity Adequate description
model parameters
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
Available soMware
à predicts shelf-‐life as well as growth of spoilage and pathogenic bacteria in seafood
à evaluates the effect of constant or fluctuaFng temperature storage condiFons (Dalgaard et al. 2002, 2003, 2008)
à more than 40 models for different bacterial pathogens
à the soMware allows growth or inacFvaFon of pathogens to be predicted for different combinaFons of constant temperature, pH, NaCl/aw and, in some cases, other condiFons such as organic acid type and concentraFon, atmosphere, or nitrate
à includes more than 40.000 curves/data on growth, survival or inacFvaFon of microorganism in foods.
à data has been obtained from the literature or provided by supporFng insFtuFons
à the modelling toolbox within ComBase includes the Combase Predictor (previously Growth Predictor and Food MicroMoodel).
à French decision support system that includes (i) a database with growth and inacFvaFon responses of microorganisms in foods and (ii) predicFve models for growth and inacFvaFon of pathogenic bacteria and some spoilage microorganisms
à predicts the effect of organic acids, temperature, pH and moisture on growth of Listeria monocytogenes in products
predict the growth of
Listeria monocytogenes
and Staphylococcus
aureus on Ready-‐To-‐Eat
meat products as a funcFon of pH and water acFvity
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
ValidaFon studies
Before market introducFon, valida6on has to be
carried out for new or altered products
AMer the concept/prototype
à acceptance tests à refinement of models à final formulaFon à validaFons in challenge test
have to be performed Milkowski (2012)
This is the main difference between free and paied
soMware
A complete challenge test takes aproximately 3
months + evaluaFon of process variaFons +
idenFficaFon of acceptable limits for formulaFon limits à establish a theoreFcal shelf life
Industry saves costs and Fme when using reliable
predicFve micro modelling
Milkowski (2012)
Industry has to perform validaFon studies, for final
verificaFon
The analyFcal values for pH, water acFvity, moisture,
etc are crucial
However, predicFve microbiology does not replace
hygiene measure or Good Manufacturing PracFces à models can not be the only hurdle to pathogens
Milkowski (2012)
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
Acknowledgement by regulaFon
Milkowski (2012) United States: -‐ U.S. 9CFR RegulaFons-‐ 2008 USDA Supplementary Guidance
European Union:
-‐ 2005 & 2010 EU regulaFon
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
The complexity of dynamic condiFons
The greatest modeller’s effort has been given to data obtained
under constant (or staFc) environmental condiFons
From a realisFc point of view this is somehow restricFve, since the
majority of thermal processes occur under Fme-‐varying environmental condiFons, and kineFc parameters obtained under such circumstances may differ from the ones esFmated at staFc
OUTLINE
ObjecFves of food industry
The challenge
PredicFve microbiology
How to obtain the data
Available soMware
ValidaFon studies
Acknowledgement by regulaFon
The complexity of dynamic condiFons
Conclusions
Conclusions
Great progresses in the past 20 years
There are many models available, each with its
benefits and limitaFons
Yet much work has sFll to be developed, parFcularly
kineFc studies under dynamic condiFons
ValidaFon studies have to be carried out for new or
altered products, due to the complexity of the systems
PredicFve microbiology is a powerful tool, but does
not replace hygiene measures or Good Manufacturing PracFces