Sei sulla pagina 1di 9

The wide range of possible envrionmental factors a food we collect data during a storage period for some mode

some mode of de-


could be exposed to brings . up . several verv important con- terioration or chemical reaction, how can these data be used
sidw;~tionsin setting up a kinetic study o f f d q u a l i t y lms. to make projections about longer storage periods or other
For example, should the food product under study he stored storage conditions such as higher or lower temperature.
in a contrhled environmentaichamber in the package typi- The first attempts to linearize data (i.e., drawing the
cally used for distribution, or should it be stored sealed off straight line as suggested above) were made when chemists
from light and moisture exchange by putting it in a can or jar? began to study the mechanisms of various reactions; their
As an example of the problems inherent in this decision, let approaches have developed over the years into elaborate or
us consider a dehydrated or baked cereal product that would simple models, depending on the degree of sophistication used
be distributed in a semipermeable pouch. Using the typical to detect the various reactants and end products.
synthetic polymer material as the package during the kinetic For example, suppose we have the reaction
study will give the shelf life of the foodlpackage system, not
the food itself. This could confound the data if the moisture AZC (2)
gainhoss effects on the rate,of loss of quality cannot be sepa-
rated f r m other environmental factors. For example, one where A is a reactant; C, the product; and kf, the forward rate
might store the pruduct in an abuse condition at 40°C and 10mo constant. If we use more sophisticated analytical tools we
R H usinv the tvoiral nanrrhuard hox. In this condition the might find out, in fact, that the reaction is really much more
product will d r G u t and there will be a possibility of an in- complex, possibly
creased rate of rancidity development. On the other hand,
room temperature data (e.g., 23% 50%RH) might not show
t instead pick UD moisture.
anv ranciditv since the ~ r o d u cwill
and the powihle development of nonenzymatic browning will In this case A reacts with B to form two products, C and D,
he thc limitinr factor. If the food were stored in iars instead. which can also hack react with a rate constant of kb. The lower
moisture exchgnge would not occur, and the shelf life of the case letters indicate that more than one molecule of each
food at constant reaction phase conditions (e.a.. constant a,) .
com~onentis ~articinatine. The standard eauation defining-
woulci br rlrtermined. ohe ever, the jar could ;etain oil-odors the rate of loss of the reactants or gain of any component (e.g.,
that were developed hut which, if the distribution package see Benson (13)) is
wcrr used, might bass through the parkageur he ahsorbed by
thr r~arkurinrmaterial itself. Thus, storarr in a iar could evrn
lead to a cow& predicted shelf life. such probiems are more
difficult to deal with in practice than in theory. This problem
is eliminated with canned foods since moistu~eexchangedoes
not occur. Moisture loss for frozen foods is more of a problem T o solve the above equation,.which predicts the change in
in the defrost cycle when package ice forms. concentration of any component with time, one needs to know
Another consideration arises from the fact that, in the high the concentrations-of each component as a function of time.
abuse temperature studies, secondary chemical reactions that If this were possible, we would still have only one equation
would occur only slowly at the lower temperatures might now with six un!&owns (kf, kb, a, b , c, d). Thus the situation is
proceed at significant rates. This, of course, could compromise technically impossible to solve. Most kinetics or physical
the predictive power of the data since a different mode of chemistry books such as Benson (13), Wallas (16),or Amdur
deterioration might now become the controlling one. Thus, (17) suggest methods to get around this problem. Basically
one should not decide to test only for one chemical index if one needs to choose conditions such that either the forward
there
~~~-~~
~~ ~~
.
is a notential for multinle reactions to occur. This would
~~~

be true for any dry or semimoist food containing unsaturated


or backward reaction predominates andlor the concentrations
of several species such as B are so high that their change in
l i ~ i d sreducine
. suears. and protein. For example. dehvdrated concentration with time is negligible. Thus, if kb <<< k f and
pbtatbes go r&cG hklow '30-32'~ and de;eldp browning [B] is very large, the loss in [A] or gain in either [C] or [Dl is
(through NEB) above this temperature (I). In cases such as
this, it would he impossible to extrapolate shelf-life data ac-
curatelv from hieh temperature to low temperature unless the
specific effects on the kinetics of the reactibn are understood.
Finally, we have the question of cycling storage temperatures where k / is the ~ s e u d forward
o rate constant. Obviouslv this
versus constant temperature. Foods do not experience con- form of the equation ignores the true mechanism of theieac-
stant temperature during distribution. Should we thus study tion but is useful from a practical standpoint. as we will
the kinetics under variable conditions, or can constant tem- see.
perature conditions he used t o make predictions for variable A food system is very complex. Assuming that, for a given
conditions? Does the temperature cycling itself cause new mode of deterioration, the above assumptions hold, one can
reactions that would introduce errors in identifying the main simply write for the rate of gain of an undesirable quality
factor in deterioration, sometimes called a ''history effect" factor [B] or loss of a desirable quality factor [A] the following
(14)? equations

General Approach to Kinetics of Food Deterioration loss of quality -


Basic Mathematical Approach to the Kinetics gain of undesirable factor
of Deterioration of Food
There once was a chemist from Latta
Who couldn't interpret his data, where k is the pseudo rate constant, n is the order of the re-
So he drew a straight line, action (a curve-fitting parameter), and [A] or [B] are the
Now everything's fine, quality factors measured (e.g., a sensory value such as a he-
Except for the damnable scatter.
donic score).
The ahove limmerirk from Leffler and Grunwald (15) con- The above forms of the equations are used only for curve
cisely states the panicular prohlem facing chemists, especially fitting of the data (as defined by our Latta chemist) and in fact
those who have studied food deterioration for many years. 1f may have nothing to do with the true mechanistic steps in the

350 Journal of Chemical Education


degradation. For example, this simple form has been found 1for up to 50% loss (if this were as far as one collected data)
to apply to the growth or death of microorganisms, a process one would find it hard to distinguish between zero- and
which must proceed by a very complex set of steps. The major first-order. The poorer the analytical precision the greater
caveats in usine these eouations are that the above assumo- would he the chance of error in deciding on reaction order
tions must h o l i (i.e., the back reaction is negligible, and the unless the study were carried out far enough. Fortunately,
concentrations of other species are not limiting) and that all although food quality loss is slow a t room temperature, many
reaction phase condition; (pH, water activity, itemperatwe, food products become unacceptable with only a 20-30%
redox potential, concentration of all other species) must re- change from the initial value. Thus, it may not &en he nec-
main ~ o n s t a n t . ~ a s i c athis
l l ~ means that ihe pseudo rate essary to decide on thr reaction order and the simpler zero-
constant h is varticular to a given food system. order model could he chosen. It should be noted that in~ Firmre
~~ ~
~~~~~~ - ~~~ -

Labuza (1)has outlined in detail the Hpp~icationof these 1the slope of the line is the zero-order rate constant ho {hie
mathematics to the shelf-life testing of foods based on both in Figure 2, using loglo for the ordinate,
temperature and humidity abuse, and Labuza and Kamman
(18) have reviewed some of the computer solutions for analysis
of shelf-life data. Only a brief review of the critical parameters
can he presented here. What should be made clear is that there Thus to get h l one should multiply the value of the slope by
are two general ways to approach shelf-life testing. The most
common method is to select some single abuse condition such
as high temperature, analyze the product two or three times
during some specified storage period, and then extrapolate
the results to normal storage conditions by using some "fudge
factor," perhapsgained by experience with similar foods. The
..
better annroach is to assume that certain nrinci~lesof
r
chemical kinetics apply with respect to temperatureacceler- 0
ation.. ex..
- . the Arrhenius rrlationshiu. ..and w h r e kinetir de- 6 so-
sign to produce a more accurate prediction. a
Y

Reaction Order > 'O-


k
Fortunately, from a data manipulation standpoint, most 2 60-
3
literature data for change in food quality (based either on 0
some chemical reaction, microbial growth, death, or sensory Y
0
value) follow a zero-order (n = 0) or first-order (n = 1) reac-
tion model. Thus integrating eqns. (7) and (8)gives +
Z
40-
3
0
Zero Order L 30-
loss: [A1 = [Aol - kot (9) 4

gain: [Bl = [Bolt kot (10) 20-

First Order 10-


loss: In[A]/[Aol = -kit
or [A] = [A~le-~l'
(11)
(12) 0
I 1
50
I
m e4 200 250 3bo
gain: In[B]/[Bo]=+kit (13) TIME IN DAYS
01 [B] = [ B ~ ] e + ~ l ' (14) Figure 1. Loss of fwd quality as a function of time showing difference between
r e r p and firstkxder reaction. End of shelf life time (t,) occurs when quality
reaches maximum allowable loss value [A,].
If the data are plotted with quality factor on the y-axis versus
time on the x-axis, the above equations determine that the
reaction is zero-order when a straieht line is produced on linear
coordinates and first-order when a straight line is obtained
on semilon -paver.
. . It is possible to determine other orders if
the mechanism is known and the experimental procedure is
very precise. For example, a plot of 1/[A] or .l/[B] versus time
will give a straight line for a second-order reaction such as the
initial vitamin C loss in milk products or infant formula liquid.
Thus, if the order is known, one can extrapolate to the value
of A, or B, (subscripts denotes value at end of shelf life) on
the appropriate plot from only a few data points. Of course
doing this can result in a large error if there is error in the
values of [A] or [B]. Since a straight line can always be drawn
between two points, as we add more data we increase our
ability to predict both outside as well as inside the range of the
data points.
Thus, in Figure 1,a linear plot of quality versus time for a
loss in quality of a food would allow easy extrapolation to the
end of shelf life (t.) since the zero-order data fit a gwd straight
line. However, the first-order data on the same coordinates *:
would be more difficult to extrapolate because of the curvature 10
80 00 1% 00 110
k
300
of the graph. Plotting the first-order data ss in Figure 2 solves TIME H DAYS
this extrapolation problem since it gives us thedesired straight F i w e 2. Loss of fwd quality factor f a a firstqder reaction pldted on semilog
line. w.End of shelf lifetims (ao c m when qualily reaches maximwmallowable
Interestingly, in looking a t the loss-of-quality data in Figure loss value [A.].

Volume 61 Number 4 April 1984 351


2.3. If the actual natural log, of the value of [A] or [B] is least-squares method or a modified-least-squares method
plotted on linear coordinates, then the slope is kl. One simple which is used for loearithmic data (first order) to eliminate
mathematical tool for first-order reactions is to use the time sume of the bias of he log function. In thew a;proaches the
at which the quality decreases to 50% (tllz), then measure of how well the dnta fit a straieht line is r2, the coef-
-1" 501100 =-t0.623 ficient of determination, where r 2 = 1represents a perfect fit.
k l= (16) It should he noted, however, that the r 2 of two points will al-
t ,I2 t112 ways be l, yet one could he far off in any prediction either
An illustration of the use of these eauations and the dif- outside or inside that range. The more data points the greater
ference in zero- versus first-order is shown in Table 1where is the confidence in using the r2 value as a statistical tool. Thii
the losses are calculated based onlv on a zero time value and regression analysis could he used to decide between zero- and
on the point at which there was 50% loss of quality. Thus, first-order.
- ~ ~ - -~ ~
~ ~ ~ -

using eqn. (9) for zero order, The next step then in this approach is to get an estimate of
[Ao] - [A] = -100 - 50 = -
50 quality unit the error ink from these regression calculations. Basically the
ko = - =1 true slope (8) is the measured slope (k) f the confidence
t 50 50 d limits
For a first-order reaction (eqn. (16)) the same data give

where 8 = true slope of regression line, t,lz = Student t value


Here one should note that the units of k for zero order are a t given degree of confidence desired (e.g., 90,95,99%) at
amountltime while that for first order is reciprocal time. One = n - 2 degrees of freedom, n = number of data points, t =
should not mix these k's or compare them absolutely as they time for each data point and
are different. As seen in Table 1, if the analytical error were
f5% one could not really distinguish between the orders using S. = standard error of [A] versus time
data for less than 50% loss.
Statistical Evaluation of Rate of Loss of OUaliW
Many different statistical mathematical approaches can These euuations illustrate that the more data points, esoe-
be used to evaluate the soundness of the amount of quality and cially at longer times which produre prater changes of quality
time data and to get an estimate of the error in the prediction . the smaller will be S, and the smaller the confidenre
from IAol.
o f t , (13, 19, 20). As noted by Lund (21), "Reporting data limits on the right.hand side of eqn. (1:). For n = 3 samples,
without an indication of error is one of the most common oc- the deeree of freedom is @ = 3 - 2 = I and thus the t . . . value
~
currences in the scientific literature today.. . investigators a t 95%confidenceis large (=12.71). If aminimum of 8 samples
have ignorrd the error assoriatrd with a parameter because are taken as suggested by Labuza and Kamman (18), then t,/z
the error is so large that it is meaningless to discuss the mean = 2.45 which gives a reasonable confidence in k and is still
valueof the parameter." What Lund says is quite true and is within practical and economic limits of most experimentation.
an indication of the problrm the "l.atta" rhemist would have Benson (13) has used this analysis to illustrate the effect of
with moat food deterioration data. The prohkm exists because precision on percent error in k (see Table 2). The hetter the
it is not always rronomirally frasihle or physically possible analytical precision and the greater the extent t o which the
to rollect enmgh data to satisfy some statistical model. As- reaction is studied the hetter is the estimate of k. If the pre-
suredly. appliration of statistics is most drsirahle and gives cision is only f5% and the reaction is carried out to only 20%,
one thk fekiing that the data are more useful, but those data the error in k is f35%.
without statistics should not be deemed useless. Prior A serond approarh, which was suggested by Lund (21) as
knowledge of the statistical requirements of a kinetic study wellas being il11atrau.d in kineticstexthx,ks.is t assume that
should help in the experimental design. Modern calculators each data point is an independent experiment with respect
and computer programs make the calculations easy once the to zero time. This is sometimes called point-by-point analysis.
data are collected. Thus
A review of statistical analysis of food deterioration data Xk i -
f l = havg195%confidence limits = - t./zo
(19)
has been made hy Labuza and Kamman (181, Lund (211, n 6
Saguy and Karel (22), Lenz and Lund (23), and Hill and
Griep-er-Block (24),and the reader is referred to these for in- where o is the standard deviation of all nk values and each k
depth analysis. Hasicall" two approarhes ran he used to get is determined using the zero value l&l (or IRol) and the ap-
an twimate of the error in the ratr ronswnt k. which is the key orooriateeuuations (10-15).'l'his method wndsto reduce the
factor used in making predictions for other conditions. ial;e of the 95% confidence limits making one feel that the
The first approach is to use some standard method of linear data are better than they may really be. The value of o comes
regression o n quality versus time data, for example the from the errors in evaluating quality and decreases as the
~rocedure'snrecision aets hetter, The confidence limits also
decrease as ihe numb& of data points is inrreased (propor-
Table 1. Experlrnental Data Manipulation tional t o n - ' % The choire of using this method versus using

Basis:A=IOOatt= Od
A = 50att=50d Table 2. Estlrnate of Percent Error In Reactlon Rate Constant, k
Calculated QualityValues ~.
Analvtical
~~

Time (d) Zero Order First Order Precision -Change in reactant species monitored.-,
(%) 1% 5% 10% 20% 30% 40% 50%

352 Journal of Chemical Education


least sauares is a matter of the oersonal oreference of there- Kinetic Approach to Accelerating Shelf-Life Deterioration
searcher and the statistical c o n h a n t . E'igure 3 shows a plot
of some actual data for the browning reaction during storage Introduction
of nonhygroscopic whey powder (25) illustrating thevisualiy In order to accelerate a shelf-life study of food deterioration
good fit to zero-order kinetics, whereas Figure 4 shows the as well as to get data which can he applied over a broader range
poor fit to first-order kinetics for the 35'C data. In using hoth of conditions, as noted, certain chemical laws can be used.
statistical methods as well as a visually drawn line to statis- With foods, however, there are limits to use of accelerated test
tically analyze the data a t 35°C the following would he conditions. Generally three approaches have been taken: (1)
found concentration acceleration, (2) moisture or humidity accel-
Zero Order: (35W eration iinrreased a,), and (3) temperature arreleration.
Regression analysis r2 = 0.994 With concentration accelsratirm Le., increasing the amount
S. = 0.79 of a reactant) other changes in the reactant phase or other
t & = 2.2 @ 95% CL *=9
6 = (0.288 i 0.017) X 10V unitlday
mechanisms may occur such that one could not use the data
t o predict for the normal concentration range. For example,
Point-by-point 6 = (0.305 -L. 0.015) X 10V unitlday with lipid oxidation of dry food products there is no change
Visual k = 0.3 X in rate above 10%oxygen in the headspace (26). Below this
oxygen level other reaction steps begin to predominate and
Thus, if the initial browning value were 0.01 and the product the rate decreases exponentially as 0 2 decreases.
hecame unacceotable at a value of 0.5. then we would find that The second test acceleration is to increase the wapr content
the time requiied to attain this value would he: of a food which in turn increases its water activitv. This is a
Average Minimum Maximum
well-known phenomenon which can he applied dry foods to
and has been studied in deoth for manv deterioration reac-
Regression 170 d 161d 181d tions (27,28). Generally, wkat is found is that betkeen the
Point-by-point 161 153 169
Visual 163 ... .. . moisture content corres~ondineto the monolaver (a, of about
0.2-0.4) and the moistire content at an a, of"arouid 0.60.8
(about 15-30 g H20/100 g solids) a plot of the rate constant
These results show that the standard linear remession (usine or shelf-life time (t.) on a log scale versus a, gives a straight
95% confidence limits) gives us a range for the end of shelf li& line, as in Figure 5. No mathematical model exists for this
that is not verv different from the ranee that the ~oint-hv- although several have been attempted (29).Thus for each food
point method gives. In addition, the visual average value is a t each temverature a different curve will he found, although,
within the limits of hoth statistical methods. This is auite a in general, the slope of the line shows about a two- to threefold
good fit to the data since t h ~I~rowning
. precision is generally increase in rate for evew 0.1 a, unit increase over the normal
less than f 2%. In the same study the results for loss of the 2040°C range. Of importance, however, is the observation
nutririmal quality of the protein (a first-order reaction) were that as the a, is increased certain reactions that do not occur
not as good because of the method orecision rf 1590).althouah at lower a,'s will predominate, thus giving results that cannot
eight sampling times were take; and the experiment was
carried out to 60%loss. For example, plotting the lysine data
a t 35°C gave a visual prediction of 65 days to reach 50%loss.
The linear regression value r 2 was only 0.89, and the time to
50% loss predicted by linear regression had a range of 48-56
d (average of 52 d). In this case, it should he noted that the
eyeball plot gave a shelf-life time outside of the calculated
regression range.

I . m . 3 . n . a . &
-0 O
. 110 180 0 20 40 60 80 100

D% DAYS
Figure 3. Zerwrder plot of extent of browningof nanhygr-plc whey powder F i w e 4. Firstader plot of extent of bmwning of omhygroscopicwhey powder
held at water activity of 0.44 for three temperawes. held at 35% and a water activw of 0.44 (data from Fig. 3.)

Volume 61 Number 4 April 1984 353


be projected to other conditions. In addition, some reactions, constant k increases. A plot of In k or k on a semilog plot
such as lipid oxidation, increase in rate as one decreases the versus 1IT should yield a straight line, as in Figure 6, since
moisture content below the monolayer due to changes in the
controlline kinetic mechanisms (29).
'~'hegel;bral method toaccelerate ihe riite of deterioratim
of foci1 IS to increase the tenmerature at which thc ~ r o d u r is
i The larger the value of EA,the steeper the slope. Thus, data
stored. Both theoretical andkmpirical models ( I ) have been from studies a t several high temperatures can be used to
used to extrapolate the high temperature data to shelf life at project the shelf life a t lower temperature, subject, of course,
low temperature. These techniques are also common in the to the errors in evaluating k previously discussed. Even though
testing of the shelf life of biologics and other drugs (30-36). the r 2 is usually high, if standard linear regression methods
The difference, however, is that with drugs the extrapolations were applied to the typical three-temperature study, the 95%
are generally easy because only one reaction is occurring, confidence limits of the slope ( E d R ) , and, thus, the error in
whereas with foods there is always the problem that, at higher EA, could still be quite high. For the browning data a t the
temperatures, the reactions which are the limiting mode of three temperatures shown earlier, we would find
deterioration may be different from those a t lower tempera-
tures. Visual plot

Temperature Acceleration for Shelf-Life Testing of a Food: Linear regression r2 = 0.999


Arrhenius Model (Ink versus 11T) +
EA = 29.68 11.48 kcallmole
Studies of the effects of temperature on the increase in In ko = 47.38 + 18.8
chemical reaction rate was first published in the late 1800's Point-by-paint analysis E A = 30.79 + 1.93kcallmale
(37). One of the most accepted models is that of Arrhenius in
which the temperature effect is incorporated into an expo-
*
In ko = 49.24 3.15
nential model of the rate constant in the form
k = hoe-EdRT (20) From this we see that although the three EA'S are close in
value, the linear regression method produces a very large
where ko = pre-exponential factor, E A = activation energy in confidence level for both E Aand ko. NOwonder this kind of
callmole, R = gas constant = 1.987 calloK mole, and T = ah- information is not included in the literature, as Lund has
solute temperature in OK. The theoretical basis for this stated (21). I t has been suggested that one should use the
equation can be found in some of the physical chemistry point-by-point method to get an E, from each individual k
textbooks mentioned previously. The value of E A (activation (18,21). As seen above, this gives an E Asimilar to that pro-
energy) is a measure of the temperature sensitivity of the re- duced by linear regression and reduces the confidence interval
action.. i.e... how much faster it will ao if the temnerature is significantly. This, in turn, would reduce the error limits on
raisrd. Mwt simplr hydrolysis renrtims have a value of 10-20 the value of k projected for a lower temperature. Of course,
krnllmole, lipid oxidatiun hy a free radical mechanism has an the better solution would be to do studies a t several more
E A = 15-25 kcallmole, nonenzymatic bruwning about 20-40 temperatures or even at the desired temperature, but this is
kcallmole. and enzvme and microhiill destruction from 50 to usually not possible with foods both because of cost and time.
150 kcal/&ole. ~ h e b a l u eof EAis very specific for each system With drugs and simple chemical solutions, another approach
and can vary with water activity. Basically, this model suggests is to do a study on a sample which is put through a sequence
that if a molecule has a total energy E 2 EA,then i t has a po- of continuously increasing temperature in some prescribed
tential for reactina which is controlled bv the value of ko. manner (linear or exponential). Many samples are taken over
sometimes called the collision factor. As the temperature in- time, and, with the correct mathematical model, the E Acan
creases, k~ and E A theoretically remain constant, but the
fraction of molecules with E 2 E A increases, thus the rate

I -1
0.
0.2 0.4 0.6 0.11 1.0 es so 3.1 3.2 3.3 M
W01.r AcIIYII~ I/T * K X 103
Figure 5. lilustratiMl of the ettect on increasing water activity (a,) on the shelf Figure 6 . Typical Arrhenius plot of log k versus inverse absolute temperature
life of a food at three differem temperatures ( T , < <
T2 T3). showing projection to lower temperature.

354 Journal of Chemical Education


he determined (30,38). This is usually not possible for food Then, substituting into eqn. (23) for shelf life a t T and T +
shelf-life studies because the temperatures required are 1O0C,we get
usually too high and the physical size of the sample causes heat
transfer limitations. This also would he almost impossible or
illogical to do with the 200-300 lh of food that might he needed and
for a sensory study. It should also be noted that several other
statistical models are available to evaluate the confidence
limits on E A such as a weighted-least-squares (39) and a
nonlinear simplex procedure for least squares (40). where T is in OK.
A convenient tool in transforming the Arrhenius plot is to Thus, the Qlo or slope b of the shelf-life plot will depend on
plot t , rather than k on the y-axis, as shown in Figure la. As the value of the activation energy. Qlo values range from 1.5
was shown by Lahuza (41 ),this derives simply from mathe- to 2 for sensory quality loss in canned foods, 1.5-3 for rau-
matical manipulation of the reaction equations and the cidity, 6 1 0 for browning reactions, and 20-40 for quality loss
Arrhenius relationship for any order. I n addition, since most of some frozen fruit and vegetables ( I ).
food studies are done over a rather narrow temperature range
(20-30°C), a simplification can he made in which log t , is Limitations of the Arrhenius Model
plotted directly against temperature rather than 1/T. Typical Based on the above discussions. it can he seen that there
"shelf-life plots" for over 400 different foods and modes of c(1~1~1 he many limitations besides s~atisticalr r n m in using
deterioration have been made ( I ) . either the Arrhenius ulut or shelf-life nlot to ~ r e d i c shelf t life
Instead of the activation enerm concept, manv researchers a t some lower tempeiature. ~enerall; the prbhlems exist he-
in the food lield have used the d;oapprdach for;ernperature cause somethinz occurs a t hieher temDeratures which does not
acceleration u,here ihe 0," is ihe derreme in shelf life (increase occur a t lower temperatures . . (14). ~ h e s limitations
e include
in rate) for a 1O0C temperature increase. the following
In fact, many have assumed the Qlo to he equal to a factor
1) At high temperatures (235°C) phase changes such as melting
of 2. However, this is erroneous and would apply only to re- of fats might occur. The liquid-phase fat can a d as a reaction
actions a t a certain temperature and E A value. T h e relation- medium or react itself where it would not at the lower temper-
ship is derived directly from the Arrhenius equation where ature.
2) During freezing, reactants are concentrated in the unfrozen
water. This concentration is temperature-dependentand not
accounted for in the typical reaction kinetic approach. The
By using O°C as an arbitrary reference temperature the major effectwould be seen just below the thaw point for frozen
equation for the shelf-life plot may he written foods.
3) Carbohydrates that might he inan amorphous physical state
t8 = t,ere-b(~-~mr)
= toe-br (23) at room temDerature and available for chemical reaction could
crystallize out at higher temperatuw and tht~etorenot react.
where b =slope in ADC,T = "C, t,.f = shelf life a t reference Thia uwld leml ttn ~ ~ ~ e r p r ~ d ~t.frhelf
c t i o ulife at nwm tpmper-
temperature T,.f, and to = shelf life a t 0°C.
4) Generally, many reactions are possible in any given fwd to cause
its deterioration. It may he possible that above some tempera-
ture a reaction with a higher En will .
~~
vredominate while below
that temperature a diffe;ent reaction will lead to loss of shelf
life. Thus, studying only one reaction will lead to significant
error in oredictions.
5) The untrr art~\,itv of a dehydrated loud at constant moisture
rmrent incrcares wirh Wrnprrature. Since some reacrron rates
are a, dependent, the acceleration of deterioration at higher
temperature would be larger than predicted solely hy~the
109 Arrhenius model. This would lead to overprediction of the shelf
life at room temperature. Many models exist for incorporating
Shelf moisture change and temperature change (22) but will not be
L~fe discussed here.
8 Since the solubility of gas decreases as T increases, oxidation

readions might became oxygen limiting. Several test procedures


for shelf life of oils get around this by using oxygen pressures
of 5&200 psi (26).
Other factors causing difficulties in making accurate predic-
tions could include change in paktitioning between the
aqueous and lipid phase, change in pH, protein conformation
changes, and moisture loss. Thus, there generally is an upper
limit of temperature that can he used f& accelkrating &c-
tious; this temperature is about 40% for canned foods (due
in part to possible thermophilic growth which begins a t 40-
109 45'C), 35-45°C for dry foods, 7-10°C for refrigerated products
Shelf and -5'C for frozen foods. Of importance here is that these
Life possible changes suggest that more than three temperatures
he used if one wants to find out if there is a shift in the kinetic
mode of deterioration with temperature, i.e., is the Arrhenius
or shelf-life plot truly linear?
Shelf-Llfe Predictions
Introduction
Figure 7.A, hansfwmed Anhenius plat of lag of shelf-lifeversus inverse tem-
Peralm. 6, a psedPtransformationof log of shelf life versus temperature for Once data are collected to prepare an Arrhenius or shelf-life
narrow temperature ranges. plot, one can then use this information to project the shelf life

Volume 61 Number 4 April 1984 355


a t some other temperature. The question, of course, is "what additive model called the Time/Temperature/Tolerance
temperature?" In fact, in the real world food products go (TTT) method or (2) a kinetic model based on the earlier
through rather variable temperature distrihutions (especially theoretical publications of Hicks (53) and Schwimmer et al.
canned and dry goods) and those distributions will each de- (54). Neither of these models considered the order of the re-
pend on the time of year the product is manufactured, the action, thus implying zero order. Lahnza (41) showed how each
physical position of the package (i.e., in the case, in the pallet, of these models could he derived solely from the kinetics
and where the pallet is in the warehouse, truck, rail car, or presented earlier and developed the equations for hothfirst-
shelf) and, of course, the peculiarities of the weather. In fact, and zero-order reactions.
the true shelf life of a particular food package which exits a The hasic tenet of any model is knowledge of the tempera-
food plant is unknown. It can only he "guesstimated" for a ture distrihution as a functiun of rime whirh is inserted into
particular temperature or temperature distrihution. either a zero- or first-order reaction equation in differential
Even if these things were taken into consideration, i t is form where
possible that temperature fluctuations themselves would
cause a "history" effect in which the rate of deterioration is
a function of the timeltemperature of prior storage. Little is
known about the magnitude of the history effect. Quast and The rate constant for each order takes the form
Karel(42) have found this for rancidity development in potato
Zero
Rate = k = ([A01 - [%I)
chi'pg, and some shelf-life studies of this phenomenon have
been conducted with pasta (43,44,45). In these latter studies, (27)
the question asked was whether one could predict the shelf
life of a product undergoing a knowri temperature fluctuation First
if one knew the temperature distrihution and had data at
constant temperatures in the form of a shelf-life plot. I t had
been the prejudice in the food research field that a product Inserting into eqn. (26) and solving for a single temperature
undergoing a reguldr temperature fluctuation, such as a 251 (Ti), i.e., k is constant then for a given time t
45°C sine wave, deteriorated more rapidly than if stored at Zero
the mean temperature (see Fig. 8) due to some "unknown
mechanism," and thus shelf life studies required this tern:
perature cycling. Most large f o d companies, from the 1950's
through the 1970's, in fact creavd weather rooms which cycled
over various extremes. What was not understood was that the First
increased rate could he merely due to the logrithmic in-
creaseldecrease in reaction rate constant due to a temperature
change. Thus, in fact, if the theoretical k were calculated and
compared to an actual k , one'could get a measure of the For zero order, the ratio of the time a t some given temperature
magnitude of the "history" effect. (t) to the total time the product could last if i t were always at
Fluctuating Temperature Kinetic Modeb that temperature (t.) gives the fraction of shelf life consumed
(f,,) since it is equal to [An] - [A], the amount of the mea-
Early storage studies of the effects of temperature cycling sured quality which has been used up, divided by [Ao] - [&I,
were done on food quality loss in frozen foods (46-48) and on the amount of that aualitv which could he used up. However,
ascorhic acid loss, microbial growth, and enzyme destruction for first order, sinceit involves a log function, the ratio is not
(4Wi2). These studies used one of two models: (1)a simplistic the fraction of shelf life consumed as was erroneously assumed
in the early research. In these studies, i t was assumed that a
distrihution could he divided into a set of constant tempera-
ture regions, and then the fraction consumed was summed up
for each time period.

The early work presented no hard data comparing theory to


actual results, and no statistics or estimates of the errors were
included in the studies. This T T T method became the ac-
cepted practice in the frozen food field, perhaps because of its
simplicity. I t should he noted that, perhaps fortuitously,
frozen foods were used which generally follow zero-order-loss
kinetics and unacceptahility occurs with only a 20-30s change
in some related chemical index (I).
A group at the University of Georgia followed up on the
T T T studies by applying a more elegant model to ascorbic
acid loss, microhial death, and heat denaturation of enzymes.
This work demonstrated that the reaction was faster for a sine
or square wave temperature periodicity than at the time av-
erage temperature; however, they found errors of from 10 to
60% in comparing theoretical rates to actual loss rates. This
was attributed partially to analytical error and in part to a
possible change in Qlo with temperature (they did not use E d .
- 0 The main errors, however, arose from the use of a zero-order
8 (days) mathematical model for first-order reactions. In addition, heat
Figure 8. Comparison between rate of quality 10% for a f w d held at constant transfer limitations were not taken into account when rapid
temperature of 25.35, and 45% and one undergoinga sine wave temperature cycling of temperature was done.
cycle of 25145°C. The food kinetics group at the University of Minnesota has

356 Journal of Chemical Education


Table 3. Results of Universitv of Minnesota Shelf-Life Studies
Actual Rate Predicted Rate
F w d Product Reference Condition Constant Constant

W g o s c o p i c whey P o d r
Nonenzymatic browning
(25) square wave 25/4S°C
a. = 0.44 *
0.019 0.0004 ODld 0.015 ODId
6-DNP lysine loss a, = 0.44 0.018 f 0.001 d-' 0.026 d-'
Relative nuhitianal valve a, = 0.44 0.0123 f 0.0004 d-' 0.0155 d-'
Nonhyqoscopic Whey P O W (25)
Nonenzymatic browning a. = 0.44 0.0069 f 0.0002 ODld 0.0076 OD/d
a. = 0.44 0.0189 f 0.004 dC' 0.0223 d-'
6DNP lysine
Relative nuiritionai value a,= 0.44 0.0116 *
0.001 d-' 0.0153 d-'
pasta
Thiamin (43 square wave 25145'C
a, = 0.44 0.0012 f 0.0002 d-' 0.0016 d-'
a. = 0.54 0.0019 f 0.0003 d-' 0.0023 d-'
a, = 0.65 0.0028 f 0.0002 d-' 0.0037 d-'
Thiamin ( 45) sine wave 25145°C
a, = 0.49 0.0016 f 0.0003 d-' 0.0013 d-'
Lysine Iwslprotein quality 144) square wave 35/55-C
a. = 0.44 0.033 f 0.006 wk-' 0.022 wk-'
a. = 0.65 0.063 f 0.012 wk-' 0.063 wk-'
Browning Square wave 35155°C
a, = 0.44 2.69 f 0.7 ODldl100g 2.15
a. = 0.65 4.00 f 0.8 O D / d / 1 0 0 g 2.93
a-ONP lysine loss* I55) sine wave 25145'C
a, = 0.49 0.0104 wk-' 0.0073 wk-'
Relative nutritional value 0.0118 f 0.0013 wk-' 0.0093 f 0.0013 wk-'
potato chips 15s) sine wave 25145%
peroxide iormatlon a,= 0.11 0.115 f 0.006 PVW 0.1 19 f 0.002 PVld

conducted studies of quality loss for a series of different food In some cases, the actual rate was faster than the predicted
oroducts stored for uo to one vear under square wave and sine rate indicating a positive-historyeffect while in other cases the
wave temperature distributions. Integrated mathematical opposite was true. These studies illustrated that, although one
models based on the Arrhenius eauation were used and the may be able to develop elegant mathematical models based
equations were corrected for reaction order. In addition, sta- on kinetics and may use elegant statistical models to get
tistics were applied to some of the results of these studies straight line ~ l o t sthe
, errors involved in analysis plus the
where enough data were available. The actual measured k was possi%le changes in mechanism with changingtemperature
compared to the predicted k based on constant temperature may preclude one from accurately predicting the quality loss
studies from a t least three temperatures. of a food undergoing a temperature distribution. This does
Using eqn. (26) as the basis, the group developed the fol- not mean we should not utilize kinetic principles nor apply
lowing equations for the theoretical rate constants. statistics, rather these results indicate the need for better
analytical tools and better experimental design, including
Square waue more temnerature measures and samoline times. and the need
toexamiie the Arrhenius plot (or s h h k to determine
if a chanee in kinetics occurs with temoerature. As our "Latta
chemist"may have found, sometimesone should not draw a
Sine waue straight line.
Acknowledgment
This project was supported in part by the University of
Spike w o w Minnesota Agricultural Experiment Station Grant No. 18-78
and NSF Grant No. 7910370.

where k ~ is, the rate constant a t the mean temperature of the


cycle (T, = (T,,,,,, + T1~~~.)12), Z = (Tm+ lO)I(T, + ad in Literature Cited
OK, 2' = (T, +
10)l(T, - no) in O K , a. = amplitude magni- (11 Labuza, T P., "Shelf Life Dating of Foods." Fwd and Nutrition Press, Westpart, CT,
tude =, ,T( - Tol ,J/2, and b = slope of shelf life plot from 1982.
eqn. (25). Thus, for a reaction undergoing cycling for some (2) Waller, G. R.. and Feather. M. S.. (Editor*), "The Maillard Reaction in Fmds and
Nulrition,"Ameriran Chemical Society, Washington. DC. 1983.
given condition, one can use eqns. (10)-(15) by substituting (3) Erikson, C.. (Editor). "Maillard Reactions in Foods." Pelgsmon Press, New Yark.
in them one of the above rate constants (k,, k.i.., ksd fork. 1982.
(4) Friedman, M.. (Editor). "Protein Crm~linking-Nutritional and Medical Conse-
If the cycling pattern is not regular, then eqn. (26) should be quences." Plenum Prees, New Yark, 1977.
used, and the quality loss summed up for each timeltemper- (5) Labuza, T. P., and Saltmarch, M., in "Water Activity: lnfluenee~on Fwd Quality,"
iEdilars: Rakland, L. and Steward, G I AcadernicPrens, New York, 1981.
ature segment as in the T T T method. (61 Labu2a.T. P., C r i l ReuFoad Technol.2.335 (1971).
Table 3 is a summary of the kinetic studies done a t the (71 Gseuls,M. C.. Jr., J. FaodSci.,40,399 (1975).
Universitv of Minnesota. A wide variety of reactions were 181 Gacula. M. C.. Jr.. and Kubsla. J. J.. J. FoodSci.. 40.404 119751.
studied mid both sine wave and square wa;e rycling were used.
The differenct between the predicted and actual reaction rare
chap. 6, p. 173.
constant ranged from as small as 3.5% for lipid oxidation to (12) Meisner,D. F.. BokwsDige~t.27,lW (19531.
as high as 30-44% for lysine loss in pasta and whey powder. (13) Benson. S. W., "Foundstionsaf Chemical Kinelia." McGraw-Hill, New York, 1960.

Volume 61 Number 4 Aoril 1984 357


(34) Wm1fe.A. J.,and Worthingfon,H. E. C.,DrugDeud Comm., 1131,185(1974).
(35) Kirkwmd, T. B. L.. Biamstrics. 33.736 (1977).
(36) Pop8.D. G.,DrvgCosm. lnd., 127151.54 (19301.
(371 Bunker,D.,Acrts. Chem. lies.,?. 195 (19751.
1281 Rnmrr.A. R.. J. Phorm. Pharmacol. 16. l 0 l T 119631.
(17) Amdur, L a n d Hsmmes, G., "Chemieal Kinetics," McCraw-HiU. New York, 1966.
(18) Lab3~za.T.P..and Kammsn, J.,in"ApplieationsofComputerliinFmdResesrchand
Fmd Industry: (Edilac Saguy, I.), Marcel Dekker, N w York. 1983, p.71.
(19) Cochrsn, W. G., and Cox. G. M.. "Experimental Dosigns:2nd ed., J. W h y & Son& (421 Quast, D.,and Karel, M., J. Food Sei., 37,584 11970).
Ine.. NwYork, 1957. (43) Ksmman, J. F , Labuza, T. P.. and Warthesen, J . J., J. FoedSci.. 46,1457 (1981).
(20) h u n d , J., "Modern Elementary Statistics," 3rd ed., Pmnti~e-HallN w York. (44) Labuza,T. P., Bohnrsck, K., and Kim, M. N., C w d Chem., 59,142 (19821.
,007 (45) Riboh.D. L.,and Labula,T.P..JFodPIoe. Preseru. 6.253(1982).
1211 Lund. D..Faod Technol.. 37111.92 119631. (46) GuUchmidt.J.,Lebmam. Wiss. u-Techno1.,7.137 11974).
i22i saw;, 1.i-d K ~ ~ I , M . , F ~ ~ ~ 34121.78
' T ~ c ~(1980).
~~I.. (47) Ysn Arsdel, W. B.. Food Tmhnol., 11.26 (1957).
(23) Lanz, M. K., and Lund, D. B.,Faod Technol., 34[2],51l1980). (48) Olson, R. L., in '"Low Temprature Biology of Foodstuffs," (Editor HaMharne, J.)
(24) Hill. C. G., and Griegcr-Bld, R. A.,Food Technol., 31121.56 (1980) Pergsmon Press. New York, 1968.
(25) Labuza, T. P., and Saltmarch, M., J. Food S c i , 47.92 (19611. (49) Powers, J. J., Lukasziwin, W., Wheeler, R.,snd Dorn%ettc.,T. P.,J FodSci.30.520
1261 Raenanaon. J.0.ar.d Labua.T. P..FoodChern..2. W1 119771. (19651.
(50) Wu, A. C. M., Eitenmiller, R. R., and Vowers, J. J., J. Food Sci., 39,1179 (19741.
(51) Wu, A. C. M., Eitenmiller, R. R., and Powers, J . J., J Food Sci. 40,840 (1975).
(52) Wu,A. C. M.,Eitenmiller,R. %and Powem, J . J., J. FoodSci.,40,1171i1975).
(63) Hicks, E. W., J. Covn Sic. Ind. Re$. iAu3trtralial. 17,111 (19441.
(29) Labuza, T. P., Food Teehnal., 34141.36 (19801. (641 Schwimmer, S., Ingraham, L. L., and Hughes, H. M., Ind. En& Chsm., 27, 1149
(30)Yang, W.,and Roy, S.,Drug Deud Ind. Phorm., 6[6],591(19801. ~.""",.
/IQSiJ

(31) Dauies, 0. L., and Budgeti, D. A., J Phorrn. Phormocol., 82,155 (1980) (55) Chen, J. Y., Bohnsack. K., and Labuza,T. P., J F w d Sci, 48,460 (19831.
(321 Co1e.B. R.,andLasdbsfer,L., J. Phorm.Phormocol.,18.101(1966). (56) Lsbula,T.P.,and Bergguiat, S., J FoodSci,48,712 (19831.
I331 Garrett, E. R., J Amen Phorm. Asaoc., 45Bl,171i1956).

358 Journal of Chemical Education

Potrebbero piacerti anche