Ratkowsky et al.

(1982, 1983) proposed mathematical models to describe the effect of temperature on

bacterial growth rate. Between the minimum and just below the optimum temperatures for growth the
relationship was linear and given by the equation:

where r is the bacterial growth rate, T is the temperature in degrees Kelvin, b is a regression coefficient
and Tmin is the theoretical minimum temperature for growth of the organism. An extension of the
equation describes bacterial growth rate throughout the entire biokinetic range:

where T is the theoretical maximum temperature for growth and c is a regression coefficient which, in
common with b, enables the model to fit the data well without necessarily having any biological
significance. When T is much lower than T, the contribution of the terms in braces is negligible and eqn
(2) reduces to eqn (1). As T increases to approach T, the term becomes increasingly more important until
eventually it dominates, and the growth rate Tmax 'McMeekin et al. (1987) noted that eqn (1) is a
special case of the temperature function of Bblehradek (1935), which is given by the equation:

where r = development rate, and aw, d and a are constants to be fitted for the particularorganism under
study. This equation is a solution of the differential equation:

This means that the rate at which growth changes with temperature is proportional to the growth rate
and inversely proportional to the difference between T and a. The latter constant is the so-called
'biological zero', the theoretical temperature at which the development rate is zero or the development
time is infinite, and is formally equivalent to T,,, in eqn (1). The BErlehradek equation has been used
extensively to describe temperature dependence of the rate of development or growth of a wide variety
of aquatic animals (McLaren 1965; McLaren et al. 1969). The exponent d varies from organism to
organism but values in the neighbourhood of 2 for growth rate (or -2 for development time) are
common (Cooley & Minns 1978). When d = 2, eqns (3) and (1) are equivalent, as the exponent 2 in eqn
(3) may be taken to the lefthand side of eqn (1) as a square-root. Use of the square-root form is
preferable for fitting the equation to actual data, as the residuals from the fitted equation in this form
tend to be homogeneous with temperature.

In the experiments reported by Ratkowsky et al. (1982, 1983), the conditions were chosen to allow
unrestricted growth at a rate characteristic of each temperature. In many situations, however,
additional constraints such as lowered water activity (aw), which is defined as the ratio of the vapour
pressures of solution to solvent (Troller & Christian 1978), may be placed on growth. Analyses of the
data of Ayerst (1969) on mould growth and germination and the data of Ohye & Christian (1967) for
growth of Clostridium hotulinum suggested that the square-root models may be valid also at reduced
aw values. Therefore it was of interest to test further the general applicability of the square-root models
in the presence of two growth-limiting factors (temperature and water activity adjusted by elevated
NaCl concentrations). Several responses seemed to be possible: the square-root models might not be
adequate to describe growth with an additional constraint, e.g. the response might be non-linear; the
models might be accurate but the predicted biological zero (Tmin or a) would not be constant; or the
models might be accurate and the predicted biological zero would be constant. Examination of the
growth response to temperature/water activity required selection of test organisms capable of growth
over a wide range of NaCl concentrations/water activities. Ecological considerations suggested that such
organisms might be obtained from saline food products such as salted, dried fish. Several bacterial
genera were isolated from this ecosystem and, from these, a halotolerant coccus capable of growth
from 0 to 4.5 molal NaCl was selected. Organisms of this type include important spoilage bacteria on
many salted fish products. Water activity was controlled by the addition of NaCl to a basal growth
medium. Under these conditions it was recognized that the growth response would be influenced by a
specific solute effect in addition to water activity but the conditions chosen represent those controlling
growth in the natural environment of the test organism.

Materials and Methods

ORIGIN AND CHARACTERIZATION OF STRAINS

Strain CM21/3 was isolated from salted and dried chub mackerel prepared at the Research Institute for
Fish Technology, Jakarta, Indonesia, and was representative of the dominant microflora in the brine and
on the fish. This strain was identified as Staphylococcus xylosus (Sanderson 1986). Strain CM21/3 grew
on nutrient media at 37°C in the presence of 0-4.3 mol/l NaCI.

GROWTH CONDITlONS

The effect of temperature on growth rate was examined in a temperature gradient incubator (Toyo
Kagaku Sangis well within the range of linear response to temperature described by eqn (1) and data
from these experiments were used to test the validity of the exponent d = 2 in eqn (3) and to determine
the value of Tmi, over a wide range of water activity. Data in the range 5"-50"C were used to test the
validity of eqn (2).

The basal growth medium contained (g/l in distilled water): Na,HPO,, 7.1; KH2P0,, 2.7; MgSO,. 7H,O,
0.25; glucose, 1.8; yeast extract, 3.0; tryptone, 1.0; proteose peptone, 5.0. The calculated water activity
of the basal medium was 0.996 (Robinson & Stokes 1949).

The aw of 1 1 of the basal medium was adjusted by addition of NaCl as follows: 0.976 (35.06 9); 0.949
(81.82 8); 0.928 (116.98 8); 0.909 (146.10 g); 0.889 (175.35 g); 0.869 (204.6 g); 0.858 (219.2 g); 0.848
(233.8 g); 0.826 (263.0 9). The pH of the medium was adjusted to 7.0 with NaOH before autoclaving at
69 kPa for 20 min.

Inocula for temperature dependence studies were prepared by growing strain CM21/3 in media of
different water activities for 18-72 h. One ml of the exponentially growing culture was inoculated into
temperature equilibrated tubes containing the same medium and growth at each temperature was
determined by nephelometry (Corning, Unigalvo, Essex). The growth was calculated at each
temperature as the reciprocal of the time to reach an optical density level of 0.3 (optical density data).
Optical density data were not used for comparative purposes as time to reach a specified turbidity is
obviously influenced by slight differences in inoculum size. Where comparisons were to be made
between data sets, generation times were calculated as the reciprocal of the specific growth rate at
each temperature (generation time data).

STATISTICAL METHODS

Equations (2) and (3) are non-linear regression models which can be fitted by least squares regression
using a standard technique such as the Gauss-Newton method (Kennedy & Gentle 1980). Initial
estimates of the parameters are required for use with this method. For eqn (2), these were obtained by
employing the method described by Ratkowsky et al. (1983). For eqn (3), by assuming d = 2 and then
taking the square-root of both sides of the equation, initial estimates of a and CI may be obtained by
linear regression. Data obtained for a range of temperatures at each of nine a, values were used to
estimate the best fit value of d, assuming that it and the zero growth temperature parameter a are
constant values, irrespective of a,, but allowing a to have different values for each of the nine a, values.
Tmin and T,,, were obtained from the fitted eqn (2). In addition, T,,, was estimated using a computerized
search procedure.

Results

VALIDATION OF EQNS (1) AND(3)

To test the validity of eqn (1) when NaCl concentration and water activity were also limiting,yo Co. Ltd,
Tokyo, Japan), over the range 5"-30"C or 5"-50°C. The former the validity of the use of the exponent d =
2 in eqn (3) and the hypothesis that Tmin or c1 is constant, strain CM21/3 was examined at salt
concentrations from 0.0 to 4.0 molal (a, 0.996- 0.848) with the temperature range 5"-30"C. The number
of data points ranged from 13 at a, = 0,848 to 27 at a, = 0.889. Using optical density data at each of nine
a, values, eqn (3) was examined to determine the best fitting exponent of d. Non-linear least squares
regression was used to estimate 11 coefficients (i.e. a, d and nine values of a). The exponent d was
estimated to be 1.97 * 0.04. Since this is within one standard error from 2, the use of the square-root
model [eqn (l)] is valid and may be used to describe the effect of temperature on the growth rate of
Staph. xylosus over a wide range of NaQ concentrations/a, values. The T,, values calculated from optical
density and generation time data at each a, are listed in Table 1. It can be seen that the Tmi, values are
relatively constant and there is no consistent trend in change of Tmin with decreasing a,. Data for each
individual a, value were forced to converge to a common Tmin value (Ratkowsky 1983). For optical
density data this was estimated to be 275.7 & 0.1"K

A direct comparison of growth rates at different a, levels can be made only by comparing generation
time data. The calculated common T,, value using these data was estimated to be 275.9 & 0.2"K (Fig. 1).
Growth was slower at a, = 0.996 than at a, = 0.976 and the growth rate declined sequentially at lower a,
values. Generation times at 28°C and values for the constant b for each a, value are listed in Table 2.
Generation times varied from approximately 27 min at a, = 0.976 to 275 min at a, = 0.848.

VALIDATtON OF EQN (2)

The results for the fit of eqn (2) to optical density data obtained at ditr'erent NaCl concentrations/water
activities are shown in Fig. 2 and good fits to measured data were obtained. Values for the constant b
obtained from optical density data at different water activities cannot be compared directly as time to
reach a specified optical density level obviously depends on inoculum size. Although the range of growth
rates decreased with a,, the notional temperatures ( Tmi, and T,,, , T',,,) calculated from eqn (2)
remained relatively constant (Table 3). The more variable estimates of T,,, result from more limited data
in the T,,, to T',,, region. In this area rates tend to decrease rapidly with increasing temperature. Further
data are required before a definitive statement can be made about the constancy of T,,, with changing
NaCl concentration/water activity.

MODEL FOR COMBINED EFFECTS OF TEMPERATURE AND WATER ACTIVITY

Figure 3 is a plot of b2 us a,, from which it is seen that for the range a, 0.97M.848, the relationship is
approximately linear. As b alone in eqn (1) changes with water activity, growth as a function of
temperature (between the minimum and optimum values) and water activity may be modelled by
replacing b by its relationship with water activity. From eqn (l), b2 is proportional to the rate r, so a
linear relationship between bZ and salt concentration/ water activity is equivalent to a linear
relationship between rate and a,. This is in agreement with the observations of Troller & Christian (1978)
who noted that below the optimum a, growth often declines linearly. Extrapolating the fitted straight
line of Fig. 3 gives a minimum a, of 0.838 (corresponding to a zero rate). When this relationship is
substituted for b in eqn (I), the expression: (5) is obtained. This equation may be used for predictive
purposes at a, levels from 0.848 to the optimum (0.976) and at temperatures from the minimum
observed temperature to 28°C (301.2"K) at each salt cuncentratiunjwater activity. The mean absolute
deviation between predicted and experimental values of r was 17.1%.

Discussion

The models developed by Ratkowsky et ui. (1982, 1983) described the effect of temperature on bacterial
growth rate under conditions where temperature was the only limiting factor. The models appeared to
have general applicability and could be used to describe the temperature relations of psychrophiles,
psychrotrophs, mesophiles and thermophiles. This work has demonstrated that the square-root models
may be used with confidence to predict the effect of temperature and salt concentration/water activity
on the growth rate of Staph. xylosus strain CM21/3. Ratkowsky et al. (1982) suggested that Tmin might
be an intrinsic property of an organism and the data presented support this contention for Staph.
xylosus strain CM21/3. A clear distinction must be drawn between Tmin [eqn (l)], which is a notional
temperature, and the minimum temperature for growth observed during the time of the experiment.
No attempt was made in these experiments to define accurately the minimum temperature at which
growth could be observed at each water activity. However, the minimum temperature at which growth
was observed generally increased as the a, decreased (Fig. 1). This trend is consistent with reports from
the literature. Thus at a, 0.996, 27 values were determined over the temperature range 6.7"-29.2"C,
whilst at a, 0.848, only 13 values were determined from 12.6"- 28.5"C. At a, 0.828, slight and
inconsistent growth was observed at approximately 35°C but cells from this culture used to inoculate
tubes of the same medium held at temperatures from 5" to 28°C failed to grow.

Ingraham et al. (1983) noted the discrepancy between Tmin (35°C) and the observed temperature for
growth (8°C) for Escherichia coli and suggested that in the extreme low range the square-root model
may not be valid. However, despite the increasing distance between Tmin and the observed minimum
temperature for growth of strain CM21/3 with decreasing a,, the square-root model accurately
predicted the growth response at temperatures above the minimum observed temperature for growth
at each a, value (Fig. 1).

Although the Arrhenius equation has proven to be useful for modelling the relationship between the
reaction rate constant and temperature for simple reactions, it and its extensions, such as collision
theory and the theory of absolute reaction rates, are 'still not capable of yielding satisfactory results for
more complex reactions such as those occurring in biochemical systems' (Bray & White 1966). For
bacterial growth, a plot of the logarithm of the growth rate constant against reciprocal absolute
temperature, instead of being a straight line, is a curve concave downwards towards the I/T axis
(Johnson et al. 1974).

A link between the square-root model [eqn (l)] and the Arrhenius equation may be provided as follows.
Taking the natural logarithm of both sides of eqn (1) and differentiating with respect to temperature
yields: (6) The Arrhenius equation in its differential form is written : (7) where E is the apparent
activation energy Combining eqns (6) and (7) leads to the following relationship between E and
temperature: Thus the differential forms of the Arrhenius and square-root models are identical if E is
related to Tmin according to eqn (8). Figure 1 shows that the effect of changing water activity is to alter
the slope b in the square-root model, but not Tmin; in which case, according to eqn (8), E is constant for
any given temperature.

The Arrhenius equation is often written in the integrated form obtained by assuming a constant
activation energy: r = A, exp( -E,/RT) (9) where A, is the ‘collision factor’ and E, the activation energy.
(The subscript T emphasizes the fact that these quantities are temperature dependent.) Thus with E,
constant at a particular temperature, the reduced growth rate caused by lowered water activity (i.e. by
increased salt concentration) is due to a reduced value of the collision factor A,. The square-root model
and Arrhenius equations can thus be viewed as reconciled; the advantage of the square-root equation
for practical use is the fact that Tmin and b do not change with temperature, whereas E, and A, do. The
same reconciliation can be carried out between the Arrhenius and BElehradek equation. Provided the
exponent d is constant, a substitutes for A, and r substitutes for E,.

In this paper we have demonstrated that the square-root models accurately predict the effect of
temperature and salt concentration/water activity on the growth rate of Staph. xylosus strain CM21/3.
The constancy of the theoretical Tmin value provided a significant advantage in the development of a
combined temperature/ water activity model [eqn (5)]. As only the constant h in eqn (1) varied with salt
concentration/water activity it was possible simply to extend eqn (1) by introducing a term to describe
the variation of b with water activity.

We have shown also that a strain of Staph. saprophyticus CM14/7 responds in a similar manner to
temperature and water activity constraints (unpublished data). Preliminary observations on strains of
Halobacterium spp. indicate that eqns (1) and (2) are also applicable to these organisms. Further work is
planned to examine the hypothesis that simple models may be developed for any set of restrictive
conditions involving temperature as one factor.