Electronic Journal of Polish Agricultural Universities (EJPAU) founded by all Polish Agriculture Universities presents original papers and review articles relevant to all aspects of agricultural sciences. It is target for persons working both in science and industry,regulatory agencies or teaching in agricultural sector. Covered by IFIS Publishing (Food Science and Technology Abstracts), ELSEVIER Science - Food Science and Technology Program, CAS USA (Chemical Abstracts), CABI Publishing UK and ALPSP (Association of Learned and Professional Society Publisher - full membership). Presented in the Master List of Thomson ISI.
Volume 9
Issue 4
Food Science and Technology
Available Online: http://www.ejpau.media.pl/volume9/issue4/art-42.html


Paweł Kitzman
Department of Quality Control and Standardization, Meat and Fat Research Institute, Warsaw, Poland



An instrumental method was elaborated to calculate growth curves of two bacterial strains, Escherichia coli and Weisella viridescens, metabolically active and growing in the same mixed culture. The method allows for: quantitative differentiation of above mentioned strains in any time points in the growth curves, estimation of metabolically active counts of bacteria in food samples according to “detection time” principle or to direct investigation of mutual influence of metabolically active strains to dynamics of their growth. Above method relays on spectro-photometric examination of mixed cultures grown in a medium containing rezazurin. Many measurements of absorbance are taken during growth in mixed culture with use of many wave lengths and obtained data (spectro-photometric spectrums) are analyzed with use of rough-sets exploration system and neural network, thus very precise bacterial counts are calculated in effect.

Key words: Escherichia coli, Weisella viridescens, mixed culture, rough-sets, neural network.


In predictive microbiology, a discipline aiming at elaboration truly working in practice mathematical models describing behavior of microorganisms in food environment, little concern was paid so far to important role of concomitant microorganisms, acting as environmental factors on microorganisms of interest. In meat environment growth of a single strain is being molded not only by simple environmental factors of chemical or physical nature as NaCl, NaNO2, pH, temperature, but also by metabolites produced by microorganisms, especially strange metabolites coming from other microorganisms present in the same environment. Automated turbidimetry, colorymetry or conductometry are commode research tools (screening methods) to trace bacterial growth in a liquid medium. “In vitro” experiments are frequently used as screening methods, serving a lot of useful digital data for subsequent mathematical modeling. Because of technical reasons (problem in distinguishing individual microorganisms) above mentioned automatic methods are of no use if one has to trace the growth of two or more microorganisms growing together in the same medium.

On the other hand, estimation of bacterial counts by routine plate count method has several disadvantages. It is labor-consuming and time-consuming (several days of incubation). Every examination of bacterial count in food product starts with extraction of microorganisms from the food environment. In the case of meat products, where some inhibitors of microorganisms growth exist (for instance NaCl and/or NaNO2), extraction and serial dilution causes washout of microorganisms from this chemicals. It should be considered, that some microorganisms in meat product, could be inhibited by some chemical additives, but after transfering to the growth medium, not containing inhibitory substances, they could proliferate. So the number of colonies that should be proportional to living bacterial counts in examined sample is not proportional in fact, and some microorganisms that would never grow in the product grow in the microbiological medium. These microorganisms are not metabolically active in the product, but they appear metabolically active in microbial bed. To avoid this problem, the addition of the same amounts of inhibitors to microbial bed as are present in the investigated sample seems to be reasonable.

The other disadvantage of routine microbiological methods comes from the fact, that in many food products we meet mixed population instead of pure culture of microorganisms. To have in the output of the quantitative method counts of separate strains or types of microorganisms, microbiological media containing selective chemical agents are in use and/or the addition of differentiating chemicals are also applied, causing different color or other kind of appearance in colonies. This kind of selective/differentiating media are neither in 100% selective, nor in 100% differentiating, so the counts are only having conventional meaning.

Optical or electrical instrumental method are acting according to “detection time” principle, relying on finding time corresponding to single point on the growth curve, representing presume signal level. This principle is correct in the case of monocultures, giving growth curves in standard repeatable conditions [6, 8]. In the case of mixed microflora “detection time” principle is not accepted [7, 9], because we are dealing with practically infinite number of various shapes of growth curves.

All these drawbacks and disadvantages of conventional method above mentioned, may be avoided with a new method (approach) described in the following paper.


Microorganisms. Two bacterial strains from American Type Culture Collection were used:

Growth media. For cultivation of E. coli before experiments in 25°C (tube cultures preparation; two consecutive passages from frozen stock in liquid nitrogen) TSB liquid medium (tryptic Soy Broth – Difco) was used. Enumeration with plate count method (CFU/mL estimation, surface-plating) in 30°C was performed on TSA (Tryptic Soy Agar – Difco). W. viridescens was cultivated in 25°C before experimens in the same way as E. coli but in MRS-broth (Man, Rogosa, Sharp broth – Merck) and enumerated on MRS-agar (Merck).

Instrument Spectramax 340 (Molecular Devices – USA) was used as automatic turbidimetric analyzer. Spectramax is a specialized ELISA – reader designed for kinetic experiments, with the possibility to preserve sterile conditions in microtiter plate (96 wells) during long lasting experiments. It is supplied with incubator having the possibility to heat plate cover +1°C higher the bottom part of the plate containing liquids, to avoid dropping water on the inner surface of the cover. It is also supplemented with a shaker. All functions of the instrument can be U-controlled from computer. In kinetic experiments simultaneous absorbance measurements with six free chosen wave lengths (from 350 nm to 750 nm with a step of 1 nm) can be done. If external incubator is used (not connected to the instrument) the microtiter plate must be transfered manually to incubator before measurements. In external incubator mode of action full scan within 350 nm to 750 nm in each time point was performed.

The main concept in experiments planning was to measure absorbance in a liquid medium inoculated, with the addition of rezazurin (0.0005%). The waves lengths from 350 to 750 nm in a step of 10 nm (41 wave lengths in sum) were chosen. Simultaneous measurements (full scans) in every time point of the growth curves were performed. Dependently on the number of active bacteria in a culture, different spectra are obtained for each of two strains. Wavy lines representing spectra, taken for different counts of bacteria for one strain are not parallel. They cross in several points. These points can be considered as relatively stable. Relatively – because the points are not exactly the geometric points, they are rather small regions in the curves. Some other small regions in the curves can be considered as relatively variable (the regions where changes in bacteria counts result in grater changes of absorbance values than in other regions). For two investigated strains relatively stable and relatively variable regions are different, so in mixed cultures the variable proportions of strains gives complex phenomenon of four types of characteristic regions. Beyond the characteristic regions other informative regions may also exist. For instance higher total counts of both strains (the sum of them) display higher absorbance in region close to wave length 350 nm.

Course of experiments. 24 h tube cultures (5 mL) in 25°C of E. coli and W. viridescens from the third passage, after defrosting the permanents, were prepared before the experiments.

The experiments were of two types. First type of experiments was performed to find the model of interrelations between absorbance values and counts of two strains E. coli and W. viridescens growing in the same liquid medium. In this case separate tube cultures of two strains in TSB and MRS broths (dependently of the strain used) with the addition of rezazurin were cultivated in 25°C. Cultivation was stopped after various periods of incubation by frosting them in – 18°C, thus frozen cultures containing various amounts of bacteria were obtained. Changes in color and turbidity were perceptible long enough after incubation period. Just before frosting, number of bacteria was evaluated by plate counts method. Before experiment, cultures were defrosted and mixed (E. coli with W. viridescens) in 1:1 proportions. The mixtures consisted of TSB and MRS broths in 1:1 proportions presenting different states of optical changes proportional to optical changes caused by living bacteria before frosting. The optical changes were of course grater if the density of cultures were higher. The mixtures were prepared in such a way, that various density combinations of strains were represented in them. 564 combinations (proportions) of E. coli and W. viridescens in 4 replicates were scanned. The survival of bacterial cells after defrosting had no significance, because mixtures were only scanned in Spectramax instrument and the instrument “reacted” only to optical changes, not to viability of bacteria. The obtained data were then mathematically processed to calculate proper model of relations between absorbance values and counts of two bacterial strains.

The second type of experiments was the growing experiments performed to validate model in practice, it means to check if the two separate growth curves are possible to obtain from mixed cultures, eventually to valuate how precise the quantitative differentiation of growing bacteria belonging to various strains is.

Calculations in the first type of experiments. Initially Rough Set Exploration System ver. 2.1 (Institute of Mathematics, Warsaw University, Poland), [1, 2, 3, 4, 5] was applied to find the relation between absorbance data and the number of bacteria corresponding to them. Typical procedure was performed consisting of several steps:

  1. Discretization step consisting of two procedures:

  2. – searching for a set of cuts in the best way reflecting the structure in the data,
    – creation the set of discrete values.
  3. Calculation of dynamic reducts. (Reduct for an information system is a subset of attributes which preserves all discernibility information from the information system, and none of its proper subset has this ability [10].)

  4. Calculation of logic rules (logic products implications leading to values in one output attribute).

  5. Classification procedure with generation of confusion matrix containing accuracy and coverage criterions. This step tests the proximity of obtained model to proper classification of output attribute.

  6. The procedure of discretization was repeated with test data set.

  7. The classification procedure was repeated for test data with use of rules calculated previously.

Next the neural network aiNet ver. 1.25, AMSES, Slovenia was used to find more precise model of relation between absorbance data and bacterial counts. In the first model calculated with Rough Set Exploration System the discretization procedure appointed these wave lengths, which are the most significant in quantitative differentiation of two bacterial strains. This set of wave lengths was used in farther neural modeling. Outliers in the plot representing predicted values versus real values were rejected before final fitting penalty coefficient in neural models. The cause of outliers appearing was probably associated with imprecise of instrument.

After obtaining the most valuable neural model some growth experiments were performed showing the possibility of quantitative differentiation of two strains and differences in bacteriostatic activity of tri-poliphosphate. The growth medium (a mixture of MRS medium and TSB medium 1:1) contained tri-poliphosphate (Zakłady Chemiczne Alwernia S.A., Poland), the common meat additive, the component of curing mixtures on several levels: 0.3%, 0.5%, 0.8% and 1%.

Calculations in the second type of experiments. The growth curves were plotted in Cartesian co-ordinate system. The logarithmic values of bacterial counts referring to metabolically active bacteria were calculated from neural model obtained in previous experiment.

Reasoning about estimation the bacterial counts of metabolically active bacteria. On the ground of presented method of quantitative differentiation of metabolically active microorganisms lies the following theory. If we have in a meat product chemical environment enabling the growth of microorganisms in the temperature condition allowing for that, microorganisms could grow with a speed dependent of the contents of chemicals displaying bacteriostatic activity. To estimate the microbiological shelf-life of the product we usually investigate the number of microorganisms during certain time-period with microbiological techniques. The number of microorganisms is tested with use of microbiological beds not containing these chemical compounds that have acted inhibitory in meat product. So the number of microorganisms estimated by colony count technique does not reflect the number of these microorganisms in the product that could really proliferate during storage. The number of microorganisms estimated by plate count technique is high, or we are investigating even not those microorganisms that consist a shelf-life problem. To avoid only this problem it is enough to ad to the solid medium proper amounts of substances – the same and in the same quantities as in the product. After extraction of microorganisms and immediate seeding the number of microorganisms should approximate the metabolically active microorganisms in the product. But another important problem exists. The population of microorganisms in the product, for instance meat product, consists of various strains. These strains influence one to each other in the product and manifest the activity close to bacteriostatic activity of chemical additives. If we were using the so called selective medium we are investigating only in one narrow group of microorganisms, so the chance for proper reflection of realistic conditions by using conventional microbiological methods are rather small. The difference between the number of metabolically active microorganisms and conventional bacterial counts is presented in Figures 1, 2 and Table 1.

Fig. 1. The sequence of colony forming units enumerations, before and after hypothetic division of microbial cells, partially being inhibited

Fig. 2. The influence of inhibitory ability of environment to number of microorganisms

Table 1. Hypothetic growth of microorganisms under foundation that 1/3 population is being inhibited in every time t

Time t (conventional units)

The whole microbial population counted with plate count method in time t (sum of microorganisms inhibited in instant t0 and active in a given instant t)

Microorganisms inhibited in time t (sum of inhibited microorganisms from instant t0 to instant t)

Population of microorganisms active in time t





























In Figure 1 we can see two stage cycle of estimation microbial counts by conventional plate counts method, so we have two counts of microorganisms obtained in time sequence. The first stage of estimation in time t0 refers to any time point during exponential growth of microorganisms, where N0 is the sum of metabolically active microorganisms represented by N0-active and inhibited microorganisms - N0-inhibited in food product. On solid microbial bed even those microorganisms that were inhibited in food product are able to proliferate and form colonies due to washing up procedure resulting from extraction and dilution. After one hypothetical partitioning (vegetative way of reproduction) in food product we have N1 counts of microorganisms in time t1 as a result of multiplication only N0-active microorganisms from the first stage t0. In spite that N1 counts of microorganisms could be split again to metabolically active microorganisms N1-active and N1-inhibited they are estimated as a sum on solid microbiological bed.

In Table 1 short simulation of multiplication occurrence of three kinds of microbial populations (the whole population, populations of inhibited and active microorganisms) is presented. In every time period 1/3 of the whole population is inhibited. On the base of this table Figure 2 was plotted. In logarithmic scale line representing metabolically active population is straight and this follows the overall theory of microbial exponential phase of growth. Counts of other kind of microbial populations resemble parabolas. This could be the reason of why the growth curves obtained by approximation experimental data obtained with use of conventional technique never show straight line section in exponential growth phase, as it is expected from theoretical reasons, but rather a curve with an inflection point.


Calculations with use of Rough Set Exploration System. The Rough set technique used to create models for classification problem relying on quantitative differentiation of E. coli and W. viridescens brought the following results. Separate models for both strains (but growing together in the same medium) were obtained on a large population of data consisting absorbance measurements performed in various populations mixtures. In the case of E. coli and test data consisting of 71 cases 7% of erroneous classifications were obtained and in the case of W. viridescens 4.2%. Because Rough Set system operates on discrete values it is impossible to tell in anambiquous way how big were the absolute errors. Rough Set Exploration System was useful in the phase of discretization as a tool to rejection 8 wave lengths amongst 41 for farther processing.

Calculations with use of neural network. Neural network generated with aiNet 1.25 software were trained with data after rejection of 8 wave lengths by Rough Set Exploration System and then tested with the smaller data base coming from the same series of experiments. In this case determination coefficients adjusted to the degree of freedom for both strains were extremely high (1,0000). The regression line ideally coincided with points.

Gowth experiments. The results of the growth experiments of E. coli and W. viridescens grown in the same liquid medium are shown in Figures 3, 4, 5 and 6.

Fig. 3. The growth of E. coli and W. viridescens in medium containing 0.3% of “Almina” preparation (tri-poliphosphate)

Fig. 4. The growth of E. coli and W. viridescens in medium containing 0.5% of “Almina” preparation (tri-poliphosphate)

Fig. 5. The growth of E. coli and W. viridescens in medium containing 0.8% of “Almina” preparation (tri-poliphosphate)

Fig. 6. The growth of E. coli and W. viridescens in medium containing 1.0% of “Almina” preparation (tri-poliphosphate)

The results show that tripoliphosphate does not influence the growth of metabolically active E. coli and inhibates W. viridescens. It should be remembered that the CFU/g units placed on ordinates refer only to metabolically active bacteria and the growth lines represent only increase in activity of metabolism of them. The full inhibition shown in the case of W. viridescens ends up abruptely, and after some time the metabolic activity increases vitally. The shapes of curves for W. viridescens have the evident angle in the bottom. This could not be possible if the CFU refer to total count estimated with conventional plate method and is possible if we are consider the intensity of metabolism. The metabolism (considered as activity of enzymes) could be stopped immediately and immediately started up. The sequence of figures 3, 4, 5 and 6 show that the longer the inhibition stage was kept the more sudden increase of metabolism follows after. From the above figures no conclusion about lethal activity of polyphosphate against E. coli and W. viridescens could be drawn, because it is not noticeable. It is also not seen that E. coli and W. viridescens influences each other.


Elaborated instrumental method with use of automatic analyzer SpactraMax 340 and microbiological liquid medium with the addition of resasurine very precisely differentiate quantitatively two bacterial strains E. coli and W. viridescens.

Tripoliphosphate influences inhibitory to W. viridescens and it is only the inhibitory activity toward bacterial metabolism not having any impact to viable count of bacteria.

The longer the period of inhibition of W. viridescens (dependent of the tripoliphosphate concentration) the more vital is the metabolic return.

E. coli is completely metabolically resistant to tripoliphosphate activity.

There is no perceptible mutual influence of E. coli and W. viridescens in the domain of their metabolism.

The elaborated method could be useful in:


  1. Bazan J.G., A comparison of dynamic and non-dynamic Rough Set Methods for extracting laws from decision table. In: L. Polkowski, A. Skowron (eds.), Rough Sets in knowledge discovery. Physica Verlag, Heidelberg, 1998.

  2. Bazan J.G., Metody wnioskowań aproksymacyjnych dla syntezy algorytmów decyzyjnych [Methods of approximation inferences for decision algorithms’ synthesis]. Ph. D. thesis, Warsaw University [in Polish].

  3. Bazan J.G, Nguyen S. H., Nguyen T.T., Skowron A., Stepaniuk J., 1998. Decision rules synthesis for object classification. In: E. Orawska (ed.), Incomplete information: Rough Set Analysis. Physica Verlag, Heidelberg.

  4. Bazan J., Nguyen H.S., Nguyen S.H., Synak P., Wróblewski J., 2000. Rough set algorithms in classification problem. In L. Polkowski, S. Tsumoto, T. Lin (eds.). Rough Set Methods and Applications. Physica Verlag, Heidelberg New York.

  5. Bazan J., Szczuka M., Wróblewski J., 2002. A New version of Rough Set Exploration System. Lecture notes in artificial inteligence 2475. Springer Verlag, Berlin.

  6. Dumont J., Menard C., Hevre B., Minier B., 1994. Analytical procedure for use conductance measurement to estimate Escherichia coli in shelfish. J. Appl. Bacteriol. 77, 296-302.

  7. Dumont L.E., Slabyj B.M., 1994. Impedimetric estimation of bacterial load in commercial carrageenan. Food Microbiol. 11, 375-383.

  8. Manninen M., Wirtanen G., Ahvenainen R., Nattila T., 1990. Automated turbidometry in assessing the bacterial content of food products inoculated with pathogens. Lenens. Wiss. U. Technol.23, 20-24.

  9. Mattila T., 1987. Automated turbidometry – A method for enumeration bacteria in food samples. J. Food Prot. 50, 640-642.

  10. Pawlak Z., 1991. Rough sets: Theoretical aspects of reasoning about data. Kluwer, Dordrecht.


Accepted for print: 11.12.2006

Paweł Kitzman
Department of Quality Control and Standardization,
Meat and Fat Research Institute, Warsaw, Poland
Jubilerska 4, 04-190 Warsaw, Poland
phone: +48 22 5097014
email: pawel.kitzman@ipmt.waw.pl

Responses to this article, comments are invited and should be submitted within three months of the publication of the article. If accepted for publication, they will be published in the chapter headed 'Discussions' and hyperlinked to the article.