Food Safety
Online ISSN : 2187-8404
ISSN-L : 2187-8404
Review (Invited)
Predictive Modeling for Estimation of Bacterial Behavior from Farm to Table
Shigenobu Koseki
Author information
JOURNAL FREE ACCESS FULL-TEXT HTML

2016 Volume 4 Issue 2 Pages 33-44

Details
Abstract

Microbial contamination is inevitable for raw and/or minimally processed ready-to-eat foods. As a consequence of the pathogenic bacterial contamination, the risk of food-borne illness increases during distribution and storage until consumption. Prediction of microbial growth and/or inactivation in/on those foods provides important information for ensuring the microbial food safety. Although numerous predictive models for bacterial growth have been proposed for various microorganisms, this review focuses on the modeling of pathogenic bacterial growth in raw and minimally processed ready-to-eat foods such as fresh-cut produce and raw minced-tuna, a common ingredient for sushi. The growth models described here take into account both the environment temperature and microbial competition in the food matrix. Microbial competition plays a key role in real food environments. Food-based predictive models enable not only to directly estimate the microbial growth on those foods, but also to apply to validation of culture-medium-based predictive models. Furthermore, toward a development of accurate and/or realistic bacterial dose-response models, a model for inactivation of pathogenic bacteria during simulated gastric fluid is also introduced.

1. Introduction

Most processed foods are pasteurized and/or sterilized through thermal processing. During thermal processing, most pathogenic and/or spoilage microorganisms are inactivated and microbial safety of the foods is assured. In contrast, raw and/or minimally processed foods such as fresh-cut produce and/or raw fish are not usually thermally inactivated during processing. There is microbial contamination in/on the raw or minimally processed foods due to the nature of the ingredient. For example, microbial contamination of fresh vegetables is inevitable even if the produce is washed by sanitizers. Raw fish such as sashimi or sushi ingredients also inevitably contain microbial contamination, because there is no microbial killing process during preparation of raw fish. As a consequence, consumers might be exposed by the risk of food-borne illness through consumption of the raw and/or minimally processed foods.

To estimate and predict microbial behavior in raw ready-to-eat foods is key in preventing infection of food-borne illness through consumption of such foods. Mathematical models to predict the microbial responses in those raw ready-to-eat foods would play an important role in management of microbial food safety. Predictive microbiology is a well-established and well-recognized scientific discipline with a burgeoning body of literature1,2,3). The quantitative evaluation of microbial responses in food environments allows us to estimate risk of food-borne illness. However, most of the predictive models developed have been based on microbial responses in liquid culture media. There are normally differences in estimates between prediction of culture media base and those of real food matrix base. Thus, food-based predictive models for microbial responses would be applied directly in risk estimation of consuming those foods. In particular, since raw ready-to-eat foods are directly ingested without any thermal cooking process, prediction of the microbial contamination level of those foods just before consumption will be critical information. In this review, fresh-cut lettuce and raw minced tuna are discussed as sample raw ready-to-eat foods.

Dose-response models play an important role in characterizing the risk in quantitative microbial risk assessment. The currently available dose-response models of microbial risk assessment are derived from different source data as follows: human-feeding volunteer tests, extrapolation of the results of animal experiments to humans4,5), and estimations from epidemiological studies of outbreaks of food-borne illness6). However these approaches to dose-response modeling of low-dose exposures to pathogens rely on multiple assumptions and extrapolations, which reflect the current lack of knowledge regarding the fundamental mechanisms underlying the responses to a dose7). Recently, a new concept, called the Key Events Dose-Response Framework (KEDRF), has been introduced as an alternative approach for developing dose-response models7). Herein, an investigation of survival of pathogens in the upper gastrointestinal tract (stomach) is discussed to apply for the KEDRF approach.

2. Contamination of Fresh Produce and Modeling the Growth Behavior

2.1. Background

Outbreaks of food-borne illnesses related to the consumption of fresh produce have been documented8). In a recent risk prioritization study, leafy green vegetables were identified as the commodity group of highest concern from a microbiological safety perspective9). Escherichia coli O157:H7, Salmonella enterica, and Listeria monocytogenes are among the bacterial pathogens most frequently associated with food-borne disease resulting from the consumption of fresh produce. Fresh produce can be contaminated during growth from many sources, such as soil, water, wild animals, birds, and insects. Following production, processes involving harvesting, washing, cutting, packaging, and shipping could create additional contamination.

Thus, when fresh produce were contaminated with bacterial pathogens for some reasons, we need to know quantitatively how the bacterial pathogens behave on the produce for microbial risk assessment. However, there has been a limited number of reports on the bacterial pathogens behavior on fresh produce compared with those on meat or meat products.

2.2. Related Bacterial Pathogens

Salmonella spp. is the most common cause of disease outbreaks linked to fresh produce, including sprouted seeds, cantaloupe, tomatoes, unpasteurized citrus juices, rocket and other lettuce varieties8). Greene et al. (2008)10) reported that recurrent multistate outbreak of Salmonella Newport associated with tomatoes. In 2002, tomatoes grown and packed on the eastern shore of Virginia contaminated with a pan-susceptible S. Newport strain caused illness in 510 patients in 26 states. In July–November 2005, the same strain caused illness in at least 72 patients in 16 states. Munnocha et al. (2009)11) reported that multi-state outbreak of Salmonella Saintpaul in Australia associated with cantaloupe consumption occurred in 2006. Also, an outbreak of 26 cases of Salmonella Litchfield infection occurred in the states of Western Australia and Queensland between October 2006 and January 200712).

Escherichia coli O157:H7 has also been documented as a cause of produce-related food-borne disease outbreaks. Although fresh produce was not considered a significant vector for the transmission of E. coli O157:H7 until the mid-1990s, a series of outbreaks associated with minimally processed produce clearly showed that contamination can occur9). The largest E. coli O157:H7 outbreak occurred in 1996, when >6,000 school children in Japan were infected with E. coli O157:H7 from white radish seed sprouts13). Between 1993 and the year 2006, 26 reported outbreaks of E. coli O157:H7 infection have been traced to contaminated lettuce and leafy green vegetables14). A recent multi-state outbreak in the USA linked to bagged fresh spinach caused approximately 205 confirmed illnesses, 31 cases of hemolytic uremic syndrome and 3 deaths15). Between 14 September and 20 October 2007, an outbreak of Shiga toxin-producing Escherichia coli (STEC) O157 simultaneously occurred in the Netherlands and Iceland. The most probable cause of this international outbreak was contaminated lettuce, shredded and pre-packed in a Dutch food processing plant16). Multistate outbreak of E. coli O157:H7 infection associated with consumption of packaged spinach in August–September 2006 was also reported17). More recently, multi-nation outbreak in Europe linked to sprout caused approximately 47 deaths by E. coli O10418).

The Gram-positive bacterium Listeria monocytogenes is another food-borne pathogen of both public health and food safety significance. Although substantial literature concerns the isolation, attachment, survival and growth of L. monocytogenes on produce19), only two fresh-cut produce-related listeriosis outbreaks have been reported in the USA20) by 2010. However, multi-state outbreak of listeriosis linked to whole cantaloupes occurred in 2011. The outbreak of listeriosis sickened more than 146 individuals in 28 states. The infection was eventually linked to contaminated cantaloupe, and the outbreak has been blamed for at least 30 deaths21).

2.3. Modeling the Bacterial Behavior In/on Fresh Produce

Modeling the growth and survival of pathogenic and spoilage microorganisms is a basic tool for the prediction of food safety and the microbial deterioration of food products in the food chain3). Although numerous bacterial growth models have been published, few predictive models have been constructed with respect to fresh produce22). These models, however, provide predictions of bacterial growth during constant environmental conditions. A dynamic model has been developed by Baranyi and Roberts23) which can successfully provide predictions of bacterial growth in fluctuating temperature conditions. Shorten et al. (2004)24) applied the Baranyi model to Erwinia carotovora in vegetable juice under conditions of fluctuating temperature.

Recently, more studies have revealed the pathogen behavior on fresh produce. Carrasco et al. (2008)25) showed that that L. monocytogenes grew on iceberg lettuce at 5 °C and 13 °C with increments of 2.66 and 4.85 log cfu/g, respectively, after 14 days under modified atmosphere packaging (MAP) condition. Doering et al. (2009)26) indicated that on cut lettuce and whole leaf spinach that was packaged and stored at 4 °C, E .coli O157:H7 contamination could still be detected after typical handling practices, although populations decreased from initial levels in many cases by at least 1.5 log units. Although E. coli O157:H7 levels decreased on products handled and stored under recommended conditions, survivors persisted. Theofel et al. (2009)27) illustrated that at 20 °C, pre-inoculation culture conditions had little impact on growth of E. coli O157:H7 on cut lettuce. However, the impact of pre-inoculation handling on survival on lettuce plants was less clear. Luo et al. (2010)28) indicated that storage at 5 °C allowed E. coli O157:H7 to survive, but limited its growth, whereas storage at 12 °C facilitated the proliferation of E. coli O157:H7. There was more than 2.0 log units increase in O157 populations on lettuce when held at 12 °C for 3 days. Although there was eventually (10 days) a significant decline in visual quality of lettuce held at 12 °C, the quality of this lettuce was still fully acceptable when E. coli growth reached a statistically significant level. McKellar and Delaquis (2011)29) and McKellar et al. (2012)30) developed a predictive model for growth/die-off of E. coli O157:H7 on lettuce and successfully simulated E. coli O157:H7 behavior on lettuce under static and fluctuating temperature conditions. Sant’Ana et al (2012)31) examined L. monocytogenes and Salmonella growth on various vegetables. They indicated that L. monocytogenes was able to grow better at storage (temperature of 7 °C or 15 °C than Salmonella. Growth of both microorganisms were inhibited in carrots, with greater inhibition on L. monocytogenes. Furthermore, Sant’Ana et al (2012)32) developed a growth model of L. monocytogenes and Salmonella on cut-lettuce.

2.4. A Case Study on Prediction of Pathogen Growth on Lettuce

A case study on pathogen growth on cut-lettuce is discussed as an example33,34). A series of five growth curves for each pathogen on lettuce was produced at temperatures of 5, 10, 15, 20 and 25 °C. Square root analysis was carried out to determine the relationship between µmax (maximum specific growth rate) and the temperature of each bacterium on lettuce. The square root model is described as follows in general35):   

μ max = b ( T T min ) (1)
where T denotes temperature and Tmin is theoretical minimum temperature for growth. The relationship between square-root of µmax and temperature was linear with a high correlation coefficient of linearity (Fig. 1).

Fig. 1.

The relationship between the square root of specific growth rate (µmax) on lettuce and temperature in L. monocytogenes (□), E. coli O157:H7 (○) and Salmonella spp.(●)

The real temperature history of lettuce collected from the farm to the retail store in early October in Japan was used as a case study. The transportation distance was about 400 km and the total time from farm to retail store was 72 h. The temperature of the lettuce just after being harvested was 15 − 17 °C, and three hours later the lettuce was pre-cooled by 5 °C in a vacuum refrigerator. The pre-cooled lettuces were stored at 5 °C in a storehouse for around five hours until they were shipped. Then, trucks without any temperature control apparatus transported the lettuces. Lettuce temperature fluctuated in the range from 3 to 15 °C during transportation. The lettuces were displayed on a showcase set at 7 °C after arrival in the retail store.

The viable counts of L. monocytogenes on lettuce under real temperature conditions are shown in Fig. 2 with growth curves predicted using the Baranyi-Ratkowsky model as shown below23).   

d q d t = μ max q ( t ) (2)
  
d N d t = q ( t ) 1 + q ( t ) × μ max × ( 1 N N max ) × N (3)
where N denotes the bacterial cell concentration (CFU/g) at time t, q is a dimensionless quantity related to the physiological state of the cells, µmax is the maximum specific growth rate (1/h), Nmax represents the maximum population density of the bacteria (CFU/g). The model for µmax was substituted into the above differential equation, and the temperature was allowed to be dependent on time. The system was solved numerically by the fourth-order Runge-Kutta method as a means of obtaining predictions of bacterial concentration during time-dependent temperature fluctuations.

Fig. 2.

The observed growth of L. monocytogenes (●) on lettuce under real temperature history (-----). The growth curves (——) were predicted using the Baranyi-Ratkowsky model. (with permission from Elsevier)

Overall predictions for each pathogen agreed well with observed viable counts. The results indicated that the Baranyi-Ratkowsky model is able to predict the growth of pathogens on lettuce under real temperature history during distribution from the farm to the retail store in most cases. Predicting the growth and behavior of pathogenic bacteria in or on lettuce will help to reduce the microbial risks associated with the consumption of salad vegetables such as lettuce, as well as provide valuable information concerning the shelf life of products to consumers. Since the prediction of pathogenic growth during distribution will serve as proof of the importance of low-temperature management, it is useful to thoroughly investigate all aspects of temperature management for those concerned with the distribution of such products.

3. Contamination of Minced Tuna by Listeria Monocytogenes and the Growth Modeling

3.1. Background

A high incidence of Listeria spp. in ready-to-eat (RTE) seafood from retail stores has been documented due to the ability of L. monocytogenes to grow during refrigerated storage of naturally contaminated products19). A previous study also reported that raw seafood may be contaminated with L. monocytogenes, and that minced tuna in particular was contaminated with L. monocytogenes in 14.3% of Japanese retail stores36). In addition, the cell numbers of L. monocytogenes in minced tuna increased by 102/g at refrigerated temperature within 3 days using an MPN method37). Since raw minced tuna is one of the most popular sushi ingredients in Japan and other parts of the world, the consumption of minced tuna might be fraught with the potential risk of listeriosis food poisoning. Although there has been no report on the outbreak of L. monocytogenes caused by minced tuna so far, the dose-response models for L. monocytogenes illustrated the infection probability 10−6 to 10−1 by ingesting ~106 CFU according to the FAO/WHO report38). If the number of L. monocytogenes increased by 103~104 CFU/g in minced tuna, total ingestion dose per meal might reach by 105~106 CFU. Although total ingestion dose depends on the consumption quantity, the risk is not negligible level. Thus appropriate prediction of L. monocytogenes growth in minced tuna would contribute to safer raw tuna consumption.

3.2. Competition Growth Modeling of L. monocytogenes

Nonpathogenic microorganisms are present in most ready-to-eat foods. The interaction and/or competition of a particular pathogen with the natural flora (NF) has been investigated and modeled by some researchers39). The competitive flora affects the maximum population density (Nmax) of co-cultured pathogenic bacteria. The effect known as the “Jameson effect” implies that the growth of both competing members of the flora is stopped as soon as one has reached its maximum level in the environment40). It is important for realistic prediction of the targeted pathogenic bacteria to take into account the effect of competitive natural members of the flora. Mejlholm and Dalgaard41) developed the interaction model of growth of L. monocytogenes and lactic acid bacteria in lightly preserved seafood such as cold-smoked salmon. The developed model is available in the Seafood Spoilage and Safety Predictor (SSSP ver. 3.1)42). Although this model is quite useful to simulate simultaneous growth of L. monocytogenes and lactic acid bacteria in lightly preserved seafood, the model might not be applicable to the L. monocytogenes strains in minced tuna due to the difference in the characteristics of the food. Furthermore, since minced tuna could be exposed to room temperature in real distribution situations before consumption, the SSSP model may not be applicable because of its temperature range (2 to 15 °C).

Here we show the growth kinetics of L. monocytogenes and NF in minced tuna from refrigeration temperature to room temperature. The inhibiting effect of NF on the growth of L. monocytogenes was also examined to attain a more realistic prediction of L. monocytogenes in minced tuna.

3.3. Model Development Procedure43)

Maximum specific growth rate (µmax), lag time (λ), and maximum population density (Nmax) were modeled as follows. The µmax (1/h) values determined at different temperatures and different inoculum level of natural flora (NF) were fitted to the modified square root model as follows35):   

μ max = a 0 + a 1 × T + a 2 × 10 I C + a 3 × ( T × 10 I C ) (4)
where ai are coefficients to be estimated, T is the temperature (°C) and IC is the inoculum level of NF (log10 CFU/g). Furthermore, the Nmax (log10 CFU/g) was modeled as a multivariate linear model as follows:   
N max = b 0 + b 1 × T + b 2 × I C + b 3 × ( T × I C ) (5)
where bi are coefficients to be estimated and other abbreviations are same as mentioned above.

To simulate the simultaneous growth of L. monocytogenes and NF under different temperatures, the modified Baranyi model incorporating with the effect of interspecies competition as reported by Gimenez and Dalgaard39) was used in this study.   

d q L m d t = μ max L m q L m (6)
  
d L m d t = q L m 1 + q L m × μ max L m × ( 1 L m t L m max ) × ( 1 N F t N F max ) × L m t (7)
  
d q N F d t = μ max N F q N F (8)
  
d N F d t = q N F 1 + q N F × μ max N F × ( 1 N F t N F max ) × ( 1 L m t L m max ) × N F t (9)
where Lmt and NFt denote the bacterial cell concentration (CFU/g) of L. monocytogenes and natural flora at time t, respectively. The qLm and qNF are dimensionless quantities related to the physiological state of the cells, µmax is the maximum specific growth rate (1/h), Lmmax and NFmax represent the Nmax of L. monocytogenes and natural flora, respectively. For the initial value of q0 which is a measure of the initial physiological state of the cells, a geometric mean value for the physiological state parameter α0 was estimated from the constant temperature experimental data. It should be noted that the relationship between lag time (λ) and α0 could be shown as follows23):   
μ max λ = 1 n ( 1 + 1 q 0 ) = 1 n ( α 0 ) (10)

The models for µmax and Nmax along with α0 were substituted into the above differential equations (Eq. 6 to 9), and the temperature allowed to be dependent on time. The system was solved numerically by the fourth-order Runge-Kutta method as a means of obtaining predictions of bacterial concentration.

3.4. Model Performance

High numbers of competitive natural flora reduced the Nmax of L. monocytogenes. Since the relationship between the initial flora level and the Nmax of L. monocytogenes was described as a function, we could incorporate the effect of initial natural flora level on the Nmax of L. monocytogenes into the predictive model (Eq. 7 and 9). This would yield a flexible prediction for the both number of the natural flora and L. monocytogenes. As an example, the model performance was illustrated under fluctuating temperature conditions that assumed a possible handling in a retail store (Fig. 3).

Fig. 3.

Simultaneous prediction of growth of L. monocytogenes and natural flora in minced tuna during storage under fluctuating temperatures. Solid and dashed lines represent the prediction using the developed model for L. monocytogenes and natural flora, respectively. The dotted line represents changes in the temperature during storage. Observed values of L. monocytogenes and natural flora are represented by the symbols ● and ○, respectively (reprint from a previous publication43) with permission from International Association for Food Protection).

4. Modeling Pathogen Survival during Simulated Gastric Digestion

4.1. Background

Survival of pathogens in the upper gastrointestinal tract (stomach) is the first event in food digestion. The number of pathogens that survive the acidic environment of the stomach is influenced by several factors, including the food matrix, the quantity and composition/acidity of foods consumed, and the general level of acidity in the stomach. Gastric juice has been investigated and reviewed as the first barrier against various infectious diseases44). Numerous studies on the effect of low gastric juice pH on the microbial survival in vitro have been reported45,46,47,48). However, since the pH in the stomach during digestion is known to change continuously49,50,51) the previous reports may have overestimated the microbial inactivation effect of gastric fluid44). To account for the effect of alterations in the pH of the gastric fluid, several studies on the inactivation of bacteria under dynamic pH conditions have been conducted52,53,54,55). In addition to the varying pH in the stomach, the effect of food-related factors also needs to be clarified, since the ingested food matrices affect microbial inactivation in the stomach. Recently, Barmpalia-Davis et al.56,57,58) demonstrated the inactivation effect of gastric fluid under dynamic pH conditions on Listeria monocytogenes inoculated into frankfurters. These investigations provide useful information for understanding the mechanism of pathogenic dose-response. In addition to the experimental data of behavior of bacteria in gastric environment, if a mathematical model of bacterial pathogen survival during the gastric digestion process could be developed, a part of the mechanism of a dose-response model could be elucidated.

Herein, a development of a mathematical model of bacterial inactivation kinetics in a gastric environment using simulated gastric fluid (SGF) is described. To correspond to varying pH values in the stomach during the digestion process, a modified logistic differential equation model were examined. The pH dependency of the inactivation rate was investigated and subsequently combined with the differential equation models to obtain a complete inactivation curve. The models developed were validated by a simulated gastric digestion process. The major bacterial pathogen Escherichia coli O157:H7 was used for the model development and validation with actual food causing food poisoning for each pathogen.

4.2. Survival Kinetics Prediction of E. coli O157:H7 during Simulated Digestion Process59)

The inactivation curves of E. coli O157:H7 at various pHs were represented by an exponential death phase (linear portion) followed by tailing (Fig. 4). The inactivation kinetics of E. coli O157:H7 was evidently pH dependent as a lower pH resulted in faster inactivation. From the experimental data, we estimated the maximum inactivation rate (rmax, log10 base) from the exponential death phase (linear portion) on a semi-log plot and then calculated the specific death rate (kmax, ln base). The relationships between pH and the k m a x for E. coli O157:H7 (EC) were closely linear with a correlation coefficient (R2) of >0.97. The relationship was expressed by the following equations:   

k max E C = 0.74 ( p H 2.11 ) (11)

Fig. 4.

Survival kinetics of E. coli O157:H7 in simulated gastric fluid with different pHs. Values represent mean ±standard deviation (n =3). The lines represent fitted lines of the developed modified logistic model. □: pH 2.8, ■: pH 2.6, ▽: pH 2.4, ▼: pH 2.2, ○: pH 2.0, ●: pH 1.8, ◇: pH 1.6, ◆: pH 1.4, △: pH 1.2 (reprint from a previous publication59) with permission from the American Society of Microbiology).

According to the equations, the inactivation rate for E. coli O157:H7 is zero when the pH is 2.11. Although these were theoretically assumed to be maximum pH values, these pH values are almost consistent with the experimental data, which showed no significant change in the bacterial number.

Figure 5 shows a schematic diagram of the experimental apparatus. The gastric digestion process was simulated in a stomacher bag filled with SGF adjusted to pH 1.5 (40 mL, fasting quantity of gastric juice for normal adult)50,60) maintained at 37 °C and the secretion of gastric juices was simulated using a peristaltic pump with peristaltic pump tubing. The food (~80 g) inoculated with E. coli O157:H7 was transferred into a stomacher bag filled with SGF (40 mL).

Fig. 5.

Schematic diagram of the simulated gastric model. The P represents peristaltic pump. (reprint from a previous publication59) with permission from the American Society of Microbiology).

Immediately after adding the cut lettuce to SGF, the pH of the mixture increased to approximately 4 and then gradually decreased to pH 2 over 180 min (Fig. 6a). The reduction of the bacterial numbers in E. coli O157:H7 was not remarkable during digestion process within 180 min, resulting in only a 1.5 log reduction, although the modified logistic model for the survival of E. coli O157:H7 showed a relatively large discrepancy between observed and predicted values (Fig. 6a). In addition, E. coli O157:H7 in the hamburger (Fig. 6b) was examined. While the modified logistic model prediction indicated no reduction in the surviving population of E. coli O157:H7, the observed number of E. coli O157:H7 slightly decreased (0.7 log) during digestion over 180 min. The pH in the simulated stomach was maintained at approximately 3 after 180 min. One of the reasons for the discrepancies could be the difference between the topical pH close to the tube outlet of SGF and the homogeneous pH of the simulated stomach. The pH in the simulated digestion system might be a topically low pH, such as pH 1.5 for fresh SGF. The topical low pH would lead to death of pathogens.

Fig. 6.

Survival of E. coli O157:H7 during simulated gastric process in fresh cut lettuce (a), and in hamburger (b). Dotted and solid lines represent the change of pH during digestion and the modified logistic model prediction of surviving bacterial numbers, respectively. Observed values (●) represent mean ±standard deviation (n =3). (reprint from a previous publication59) with permission from the American Society of Microbiology).

Although the developed model enabled us to predict pathogen inactivation during gastric digestion, its results also suggested that the ingested bacteria in the stomach would barely be inactivated in the real digestion process. The results of this study will provide important information on a part of the mechanism of dose-response of bacterial pathogens.

5. Summary

Herein, we described are examples of modeling the bacterial growth in raw ready-to-eat foods as a consequence of bacterial contamination of the ingredients. The growth models discussed here are described as a function of temperature in each food matrix. Although more global model including various environmental parameters based on culture media could be applied to those ready-to-eat foods, food based models would reflect more real situation. If the concrete data and/or models for the real food are accumulated in literature and/or databases such as ComBase (www.combase.cc), food processors and distributors would be able to estimate not only the risk of food-borne illness but also the risk of spoilage.

Acknowledgements

This study was partly supported by Grants-in-Aid from the Food Safety Commission, Japan (No. 0705) and also by the research fellowships of the Japan Society for the Promotion of Science (JSPS) for Young Scientists.

References
 
© 2016 Food Safety Commission, Cabinet Office, Government of Japan
feedback
Top