Introduction

Feedlot practices were introduced in Argentina in the 90s mainly for economic reasons. Nowadays, this practice still continues and at the same time feedstuff production has globally increased. A recent feed survey indicates that worldwide feed production has exceeded a billion metric tons4; 10.6% of feed production in the Latin American region is represented by Argentina34. Particularly, more than 14 million tons were produced in 2015, while the most recent estimates suggest that around 25 million tons will be produced by 202027. Animal feeds are intended for cattle, dairy cattle, porcine, poultry, aquaculture, rabbit and chinchilla, and more recently buffalo. Moreover, an important growth has been registered in the pet food industry in the last years. In addition to industrial feed production, small farmers also elaborate their own feed. However, in some circumstances, lack of quality assurance of raw material, improper technical production process or inappropriate storage conditions can lead to fungal spoilage development and mycotoxin production. In addition, sometimes due to lack of knowledge or negligence, despite the fact that visible fungal growth is observed, feeds are mixed with non-contaminated feed in order to dilute its contamination, thus economic losses are avoided. Therefore, feed quality decreases and becomes a risk for animals, and consequently for the complete food chain37.

Animal feeds are composed of the necessary nutrients that animals need at every stage of their growth to reach a proper development and maximal growth in a profitable period of time for producers. Several parameters are involved and considered to define the quality of feed pellets such as formulation, durability, starch gelatinization and protein content, particle size, exposure to high pressure steam and proper cooling, among others17. In addition, the final product stability is regulated byaw. Therefore,awis the most critical parameter to be taken into consideration since variations tend to increase fungal proliferation11,29,46.

Fungal contamination in raw materials and animal feed has been widely reported worldwide. In fact, most of them emphasize mycotoxigenic genera, mainlyAlternaria,Aspergillus,FusariumandPenicillium5,16,26,37,42. In addition, based on the findings reported by Hassan et al.26, mixed cereals used as animal feeds are contaminated with high levels of mycotoxigenic fungi and mycotoxins. Moreover, several cases of human and animal intoxication and death caused by food and feedstuff contaminated with mycotoxins were also informed in the recent years^18,21,33,38.

The presence and development of fungi cause changes in food and feed properties involving nutritional and organoleptic quality aspects. Due to their xerophilic physiologyEurotiumspecies are able to grow atawvalues as low as 0.6512,28. Furthermore, these species are also able to produce a large variety of secondary metabolites10,22,32,43. Therefore, it is of relevance to study the behavior ofEurotiumspecies in order to prevent their growth in animal feed, and their potential production of secondary metabolites.

In recent years mathematical modeling has gained importance since they are considered helpful tools applied to avoid fungal growth and increase food shelf-life.

Previously, we have reported a high incidence ofEurotiumspecies in raw materials and animal feedstuff and an initial study of the effect ofawand temperature on the growth ofEurotiumspecies was performed on synthetic media23. However, considering the high incidence and magnitude of damages that this genus can cause, it is more accurate to study the effect of environmental parameters in a more realistic context thus, in this case in a feed matrix.

Therefore, the aims of this work were (i) to study the effect ofawand temperature on the growth ofEurotiumspecies in a feed matrix, and (ii) to obtain predictive mathematical models for each species: the cardinal model of Rosso for secondary kinetic growth, and the probabilistic model for growth/no-growth interface.

Materials and methodsFungal isolates

FourEurotiumspecies previously isolated from rabbit and chinchilla feed were used in the present study.Eurotiumamstelodami,E. chevalieri,E. repensandE. rubrumstrains were previously identified by scanning electron microscopy and DNA sequencing and characterized by secondary metabolite profiling, and growth on synthetic media22,23.

Experimental design and data generation

A full factorial design was used to study the growth of fourEurotiumspecies in a feed matrix. The effect of nineawand five temperatures were evaluated. Theawassayed were in the range 0.71-0.97 (0.71, 0.73, 0.75, 0.79, 0.83, 0.87, 0.91, 0.95, 0.97), and the incubation temperatures were 5, 15, 25, 30 and 37°C.

Additionalawvalues were evaluated for some of the species:E. amstelodamiat 0.77 and 0.81awat 15, 25, 30 and 37°C;E. repensat 0.85 and 0.89awand 5, 15, 25 and 30°C.

The maximumawthat the feed could reach without showing excess of water was 0.97awcorresponding to a moisture content of 60%. Above this value there was surrounding water.

Feedstuff media and water activity adjustment

Commercial a package of rabbit and chinchilla feed was purchased from a local store. The total content was subdivided into 1kg samples, each of which was placed in plastic bags and the total number of bags were placed into the original package made of a double layer of wood paper, and then closed again. Immediately, the bag sample was sent to the Semi-Industrial Irradiation Plant (PISI) in the Ezeiza Atomic Center (Buenos Aires province, Argentina) where it was irradiated with 35kGy of gamma irradiation in order to achieve sterilization. Initial moisture content andawwere 11.06±0.43% (humid basis) and 0.687±0.014, respectively. Then, an adsorption isotherm was built. First, small samples (50-100g) were adjusted to the requiredaw, and then were scaled to 1kg sample size. Each kilogram of sample was put into a 3 liter glass jar previously sterilized in autoclave at 121°C for 15min and dried in a drying oven at 105°C for 3h. Later, appropriate volumes of sterile distilled water were added in aseptic conditions, mixed and stored at 4°C for 7 days with periodic mixing. Moisture content was determined according to the AOAC Official Method6andawwas measured with LabSwift-aw Novasina. Each determination was done by triplicate, jars were vigorously shaken and samples were taken from different points. The feedstuff media were aseptically transferred into the Petri plates, a thin layer of feed was placed into them (10-15g), and previous to inoculation the plates were placed at each final incubation temperature for 24h in order to reach the incubation temperature40.

Inoculum, inoculation and incubation

Heavy sporulating cultures of each strain were obtained after seven days of growth at 25°C on Czapek yeast extract agar with 20% of sucrose (CY20S). Conidia were collected from the plate surface and suspended in 5ml of a water/glycerol solution adjusted to the requiredawlevel and containing 0.05% v/v of Tween 80. Conidia concentration was determined with a Neubauer chamber and adjusted to a final concentration of 10⁵conidia/ml. Petri plates containing the feed substrate were centrally inoculated with 1μl of the conidia suspension using a calibrated loop. Petri plates with the sameawwere enclosed in polyethylene bags with plastic glasses containing water/glycerol solutions of the sameawin order to keep constant the relative humidity inside the bags and plates; water/glycerol solutions were renewed weekly. Incubation temperatures were 5, 15, 25, 30 and 37°C and the experiment had a duration of 40 days and a maximum of 90 days for the most unfavorable conditions. For each combination ofaw*T*species ten plates were used. Additional non-inoculated plates with adjustedawwere placed into the corresponding bags to asses anyawdeviation during the study.

Growth assessment

Plates were observed twice a day with a binocular magnifier until growth was detected. Then, two perpendicular diameters of the colony were measured daily without opening the plates, until the colony reached the edge of the plate or until the end of the experiment.

Growth data treatment and mathematical analysisData treatment

Colony diameters were plotted against time, then linear regression was applied to obtain the growth rate (mm/day) as the slope of the line. All obtained data for each pair of (aw,T) conditions and replicates were transformed to obtain homogeneity of variance. All data were square root-transformed.

Secondary model

Secondary models are used to describe the influence of environmental factors on fungal growth39. These models are of great importance and fundamentally very useful because they can be used to predict the microbiological shelf life of a food product35. In particular, in the Rosso cardinal model all of the studied parameters have a physiological meaning19. The data set obtained from this study was tested with other secondary models (data not shown), however the best fit was obtained with the Rosso cardinal model.

Then, the influence ofawand temperature on fungal growth was studied. The Rosso cardinal model was applied to describe the effect of both environmental parameters on the radial growth rate. Each model performance was assessed by the root mean square error (RMSE) and coefficient of determination (R ²).

The equation describing the model is:

The equation was fitted using the nls function (nonlinear least squares) from the R Stats Package, in R (version 3.4.0).

The parameterawmax was fixed at 0.97 since it was the maximum value that the feed can reach by addition of water without having free water/surrounding water. It is significant to mention that above 0.95awvalues there are notorious changes in feed appearance (pellet disintegration). Moreover, according to Sautour et al.41awmax should be set to one only for non-xerophilic fungi.

Growth boundary model

Probabilistic models are used when the aim is to determine if growth can happen or not under specific conditions19. The information provided by these models is very useful because it allows to predict and avoid fungal (or bacterial) growth or even toxin production.19Then, this model instead of determining growth rate, determines whether or not growth occurs.

The growth/no growth boundaries at differentaw, temperature and time were built according to the linear logistic regression analysis. A value of 1 was assigned when growth was registered and 0 value when no growth occurred. A total of ten replicates were used.

The equation for the logistic regression model is:

wherePis the probability of growth (range 0-1),biare the coefficients to be estimated, t (days) is the incubation time,awis the water activity of the feed medium, and T is the incubation temperature (°C).

The equation was fitted using the glm function (generalized linear models) from the R Stats Package, in R (version 3.4.0).

The predicted growth/no growth boundaries interfaces forp=0.5 were calculated and plotted with Microsoft Excel 2007 Solver and Octave (GNU Project, Version 4.2.1).

The predictive power/capacity and the goodness of fit of the logistic models were assessed by pseudoR ²indices (McFadden and Nagelkerke), percentage of concordance and accuracy3,14.

Performance of probability models at probability level of 0.50 was additionally evaluated by comparison with independent data obtained from studies carried out in analog media1,2,24,30,44.

ResultsGrowth description in feed

Growth ofE. amstelodami,E. chevalieri,E. repensandE. rubrumwas studied at differentawand temperatures.

E. amstelodamigrew in the range 0.77-0.97awat 25 and 30°C. At 15°C the range was 0.81-0.97aw; and at 37°C the range was shorter, from 0.91 to 0.95aw. The minimum growth rate was 0.19mm/day at 0.77awand 25°C, while the maximum growth rate was 18.3mm/day at 0.91awand 30°C. High growth rates were observed in the range 0.87-0.95awin the temperature interval of 15-37°C. Nevertheless, optimalawand temperature were 0.91 and 30°C, respectively.

E. chevalierigrowth was observed in the range 0.79-0.97awfrom 15 to 37°C and at 0.75awat 30°C. The lowest growth rate was 0.13mm/day at 0.75awand 30°C, while the highest growth rate was 23.6mm/day at 0.91awand 30°C. High growth rates ranging from 10.5 to 15.8mm/day were also observed in the intervals of 0.83-0.91awat 25-37°C temperature range.

ForE. repens, growth was registered from 0.91 to 0.97awat 5, 15, 25 and 30°C. Minimum growth rate was 0.29mm/day at 0.97awand 5°C, whereas maximum growth occurred at 0.91awand 25°C (20.16mm/day). At 15 and 30°C, high growth rates also occurred at both 0.91 and 0.95aw(10.2mm/day at 15°C and 16.7mm/day at 30°C; 9.3mm/day at 15°C and 6.3mm/day at 30°C, respectively).

With regard toE. rubrum, growth was observed in the interval of 0.75-0.97aw. At 0.75awonly grew at 25°C, and growth at 5°C only occurred at 0.91aw. From 0.79 to 0.97awgrowth took place at 15, 25 and 30°C. The optimalawvalues ranged from 0.83 to 0.95 at 25 and 30°C with growth rates above 10mm/day, except at 0.95awand 25°C. The maximum growth rate was 15.9mm/day (0.91awat 25°C) and the minimum growth rate was 0.16mm/day (0.75awand 25°C). Noawupper limit of growth was observed.

It has been observed that the rate of growth in all species studied decreased atawabove 0.91 at all the temperatures tested.

Secondary model

Table 1 shows the cardinal values of environmental factors obtained using the secondary cardinal model. The RMSE was calculated to assess the performance of the predictive models; in the calculation no-growth conditions next to the growth ones were included31. In general terms, the models showed a good fit withR ²ranging from 0.885 to 0.950. Nevertheless, the RMSE values were above 1, indicating a poor fit to experimental data. Despite the fact that all the RMSE values were higher than 1, there was high agreement between the parameters estimated by the model and the experimental data. Forawmin,awopt andawmax, the model provided values that fell within the experimental domain.

For three out of four species there was agreement between the experimentalawopt andawmax and those estimated by the model (0.91 and 0.97, respectively). In addition, forTopt andTmax, the model also showed high concordance with experimental data.

With regard toTmin forE. repensandE. rubrum, both species were able to grow at 5°C. The estimated value was below 5°C (2.78°C) forE. repens, whereas forE. rubrum,Tmin was estimated slightly above 5°C.

With respect to optimal growth rate (μopt), there was high agreement for three out of four species. Nevertheless, there was a significant difference forE. repens, where the estimation of the model turned out to be twice as that of the experimental value.

Growth/no-growth interface model

Probabilistic models were built for each of theEurotiumspecies at 7, 15 and 40 days. Table 2 shows the coefficients obtained for each model.

Developed models include linear, quadratic and interaction terms. There were two exceptions since terms were not significant (p-value>0.001):aw*TforE. chevalieri(p-value=0.0079), andaw ²forE. rubrum(p-value=0.452). Despite the fact that coefficients to assess predictive power and goodness of fit were accurate in both cases, the termsaw*Tandaw ²were suppressed in each respective case in order to improve model fit and avoid overfitting (backward elimination of terms).

Figures 1-4 show plots of probability of growth (p=0.50) for temperature andawat 7, 15 and 40 days of incubation for each species. As it can be seen from the figures, growth occurred at almost all suitable conditions during the first 7 days. In general, a high percentage of logistic model agreement with the experimental data was found. The models showed a concordance percentage above 95 and accuracy above 0.929 (Table S1, supplementary material).

Figure 1: Growth/no growth interface ofE. amstelodamiafter 7, 15 and 40 days of incubation in a feed matrix. 

P(0.50): short dashed line: at 7 days; long dashed line: at 15 days; solid line: at 40 days; white square: growth at 7 days; gray square: growth at 15 days; black square: growth at 40 days; cross: no growth. (0,07MB).

Figure 2: Growth/no growth interface ofE. chevalieriafter 7, 15 and 40 days of incubation in a feed matrix. 

P(0.50 short dashed line: at 7 days; long dashed line: at 15 days; solid line: at 40 days; white square: growth at 7 days; gray square: growth at 15 days; black square: growth at 40 days; cross: no growth.

(0,07MB).

P(0.50): short dashed line: at 7 days; long dashed line: at 15 days; solid line: at 40 days; white square: growth at 7 days; gray square: growth at 15 days; black square: growth at 40 days; cross: no growth.

(0,07MB).

Figure 3: Growth/no growth interface ofE. rubrumafter 7, 15 and 40 days of incubation in a feed matrix. 

P(0.50): short dashed line: at 7 days; long dashed line: at 15 days; solid line at 40 days; white square: growth at 7 days; gray square: growth at 15 days; black square: growth at 40 days; cross: no growth.

(0,07MB).

Figures S1-S4 (supplementary material) show plot probability of growth vs. temperature for different values ofawand incubation times of 7, 15 and 40 days. These figures were also constructed by using the equations obtained by the model, and are complementary toFigures 1-4.

It is important to notice that the probability curves are almost centered at the experimental optimal temperature of growth for each species (except forE. repens, whose probability curve is centered at a lower temperature), and show how the probability of growth increases with time. It is worth noting that forE. chevalieriandE. rubrumthis probability increment is notorious when comparing figures for 7 and 40 days respectively (Figs. S2 and S4).

In addition, the validation of the models was also performed with external data from other authors1,2,24,30,44. This comparison is shown in Table S2 (supplementary material), where the probability of growth of the models is shown on the “Pmodel” column. Despite the fact that some disagreements were observed (8 false-positives and 5 false-negatives out of a total of 70 samples), the models predicted acceptably considering that the validation points were located at the domain-model boundaries.

Discussion

Growth and probabilistic models were developed forEurotiumspecies isolated from animal feed. It was found that in most of the evaluated conditions ofawand temperature, growth happened in a shorter period of time than in previously assayed synthetic media (MEA)23. Synthetic media are richer than natural feed and food, the latter provide a more real condition while synthetic media might induce an overestimated growth7. However, some discrepancies were found in the present study since in the feed matrix the temperatures required for growth at 0.75awwere 25 and 30°C (forE. rubrumandE. chevalieri, respectively), and no growth was detected at 15°C, while in synthetic media the lowestawvalue at which growth occurred was 0.75 at 15 and 25°C. In addition, the temperature range of growth in feed was wider than in synthetic media, while theawrange of growth was more limited. Thus, we consider that the most appropriate choice is to perform the analysis in a culture media as close as possible to the original source from which molds were isolated. With regard to this topic, it was determined that the temperature andawat which fungal growth occurred in animal feed were more in accordance with several authors1,13,15,47. However, differences might be due to the nutritional composition of the substrates and geographical origin7,20,36.

In relation to the cardinal models, they provided accurate estimations forawmax andawopt,μopt (except forE. repens),Tmax andTopt for all the four species studied (exceptTopt forE. chevalieri). On the other hand, the parametersawmin andTmin were underestimated in all cases with the exception ofawmin forE. chevalieriwhich was closer to the experimental observation.

Rosso and Robinson39estimatedE. amstelodamiandE. chevalieriradial growth rate,awmin andawopt from external data, which was obtained in a laboratory medium at 25°C and glycerol as substrate to controlaw(among other solutes). These authors determined a radial growth rate of 2.52mm/day, 0.647awmin and 0.974awopt forE. amstelodami; while forE. chevalierithe radial growth rate was 2.664mm/day, 0.705awmin and 0.955awopt. Once again, our findings differ widely since the radial growth rate in feed was about ten times higher than any informed value,awmin was higher than those andawopt was lower than the values reported by the authors. However, Guynot et al.25reported a mean of maximum growth rates estimated by linear regression (at 0.90aw) comparable with the informed growth rates in the present study. In addition, Suhr and Nielsen44reported thatE. repenswas able to grow in rye bread medium at 0.80aw(controlled by glycerol) and exhibited visible growth before 30 days. Moreover, these authors reported that the shortest time for spoilage was observed at the highestaw(0.95) within a week44. In our study,E. repenswas not able to grow below 0.91awin animal feed and all the four species were able to grow at the highestaw(0.97) in less than a week's time.

The RMSE andR ²were the parameters used to evaluate the goodness of fit and the fitting capacity of the models, respectively. The first parameter measures the deviation between observed and predicted values, while the second parameter indicates how useful the explanatory variables are in predicting the response variable9. A good fit is indicated by RMSE values close to zero, while a proper fitting capacity (R ²) is indicated by values close to one.

In the present work, despite the fact of having inoculated molds directly on the feed pellets, the obtainedR ²values were very high (>0.885), comparable to values usually obtained from well-defined synthetic media45. Otherwise, RMSE values were too high to indicate a proper fit (2.139>RMSE>1.614). Similarly, Basak and Guha8informed anR ²range from 0.980 to 0.991 while RMSE ranged from 3.5 to 5.3 in aPenicillium expansumspore germination study.

Probabilistic models developed for eachEurotiumspp. showed agreement between observed and predicted data. This agreement was supported by the high values of the obtained coefficients. The pseudoR ²indices (McFadden and Nagelkerke) were accurate enough to indicate a good fit for each model. In addition, accuracy and concordance percentage for all developed models were also high, ranging from 95.58 to 97.81% and from 0.929 to 0.945, respectively. This agreement was also supported graphically, although probability models failed in some cases. Nevertheless, this kind of information can be usefully applied to preventEurotiumspp. growth in animal feed during storage. As many of the conditions which favored growth occurred during the first 7 days, a safe storage fromE. amstelodamigrowth will be achieved below 0.77awregardless of the temperature (pof growth<0.10). To preventE. chevalierigrowth,awabove 0.79 (pof growth<0.10) should be avoided. In the case ofE. repens awshould be kept below 0.91, and below 0.73awforE. rubrum.

In conclusion, the predictive models applied in the present study were able to satisfactorily describe and predict the effect ofawand temperature on growth and the probability of growth ofEurotiumspecies in a feed matrix. Since these models were constructed in a real feed matrix they can be used as predictive tools to improve food safety, optimize food storage and prevent fungal spoilage. Thereby, important economic losses can be avoided and feed quality can be improved over time.

To our knowledge this is the first report assessing the growth parameters ofEurotiumspecies directly in animal feed.

Conflict of interest

The authors declare that they have no conflicts of interest.

ACKNOWLEDGMENTS

Thisworkwasfinanciallysupportedby Universidad Nacional de Quilmes and Agencia Nacional de Promoción Científica y Tecnológica.

The authors also thank Gonzalo Pacheco for programming assistance.