Quantitative Microbial Risk Assessment for Staphylococcus aureus and Staphylococcus Enterotoxin A in Raw Milk

Abstract

A quantitative microbial risk assessment was constructed to determine consumer risk from Staphylococcus aureus and staphylococcal enterotoxin in raw milk. A Monte Carlo simulation model was developed to assess the risk from raw milk consumption using data on levels of S. aureus in milk collected by the University of California-Davis Dairy Food Safety Laboratory from 2,336 California dairies from 2005 to 2008 and using U.S. milk consumption data from the National Health and Nutrition Examination Survey of 2003 and 2004. Four modules were constructed to simulate pathogen growth and staphylococcal enterotoxin A production scenarios to quantify consumer risk levels under various time and temperature storage conditions. The three growth modules predicted that S. aureus levels could surpass the 10^sup 5^ CFU/ml level of concern at the 99.9th or 99.99th percentile of servings and therefore may represent a potential consumer risk. Results obtained from the staphylococcal enterotoxin A production module predicted that exposure at the 99.99th percentile could represent a dose capable of eliciting staphylococcal enterotoxin intoxication in all consumer age groups. This study illustrates the utility of quantitative microbial risk assessments for identifying potential food safety issues. [PUBLICATION ABSTRACT]

Headnote

ABSTRACT

A quantitative microbial risk assessment was constructed to determine consumer risk from Staphylococcus aureus and staphylococcal enterotoxin in raw milk. A Monte Carlo simulation model was developed to assess the risk from raw milk consumption using data on levels of S. aureus in milk collected by the University of California-Davis Dairy Food Safety Laboratory from 2,336 California dairies from 2005 to 2008 and using U.S. milk consumption data from the National Health and Nutrition Examination Survey of 2003 and 2004. Four modules were constructed to simulate pathogen growth and staphylococcal enterotoxin A production scenarios to quantify consumer risk levels under various time and temperature storage conditions. The three growth modules predicted that S. aureus levels could surpass the 10^sup 5^ CFU/ml level of concern at the 99.9th or 99.99th percentile of servings and therefore may represent a potential consumer risk. Results obtained from the staphylococcal enterotoxin A production module predicted that exposure at the 99.99th percentile could represent a dose capable of eliciting staphylococcal enterotoxin intoxication in all consumer age groups. This study illustrates the utility of quantitative microbial risk assessments for identifying potential food safety issues.

Foodborne illness and food safety are issues of increasing importance and concern in the world today. In the United States, an estimated 76 million people per year are afflicted with a food related illness or intoxication (18, 20). In addition to the direct impact on the victims of foodborne illness, these illnesses impart a heavy burden on government regulatory agencies, public health officials, and food manufacturers and producers, who must respond with efforts to reduce the incidence of such outbreaks and cases. Milk and dairy products are sometimes implicated as the sources of illness associated with milk collection and normal processing conditions that allow bacteria present in the dairy cows and the dairy environment to be introduced directly into the milk (2, 15, 23). Once these bacteria are introduced, the highly nutritive milk medium supports rapid microbial growth (14). Consequently, the potential for foodborne illness and intoxication from consumption of milk and dairy products is of concern.

The introduction and widespread use of appropriate harvesting techniques and proper pasteurization methods has successfully resulted in a milk supply that is consistently free from major microbial pathogens (26). Unfortunately, even with the diligent use of pasteurization, postpasteurization contamination and improper temperature storage conditions can occur and have led to foodborne illness throughout the world (5, 14, 18). Although some outbreaks have been associated with pasteurized milk, pasteurization still is considered an extremely effective method for reducing bacterial pathogens in milk, and these outbreak events usually are rare (26). However, not all milk and dairy products are pasteurized, and raw unpasteurized milk is widely consumed in countries throughout the world. Raw milk consumption in the United States has increased recently (6, 14, 23, 26, 29). In raw unpasteurized milk, bacterial pathogens present in the milk at the time of milking and/or introduced during harvesting, storage, and processing can remain in the product and potentially proliferate to numbers capable of eliciting disease when consumed.

One organism of particular interest to milk food safety is Staphylococcus aureus. This facultative anaerobic grampositive bacterium is a major cause of foodborne intoxications and outbreaks throughout the world because of its ubiquity and its ability to persist and grow under various conditions. S. aureus is able to survive and multiply in a variety of food substrates, at a variety of temperatures (7 to 45.6 or 48.5°Q and pH values (4.5 to 9.3), and at water activities of 0.83 to >0.99 (2, 5, 18, 21, 27, 30). Unlike other common foodborne illnesses that require consumption of and infection by viable pathogenic microbial cells, sickness associated with S. aureus occurs as a result of ingestion of numerous heat- and protease-stable staphylococcal enterotoxins (SEs) produced under specific environmental conditions when the population density of the pathogen reaches 105 CFU/ml (12, 18, 22).

Enterotoxins are low-molecular-weight extracellular, superantigenic chemicals that initiate nonspecific T-cell proliferation resulting in severe diarrhea, nausea, vomiting, and abdominal pain 1 to 6 h after consumption of a large enough dose (2, 12, 29). Classic SEs have been given letter designations (A through E) and are implicated in 95% of the food enterotoxin outbreaks in the United States (2, 6, 10, 28). SEA is the primary enterotoxin associated with foodborne illness and has been identified in 77.8% of foodborne enterotoxin-related outbreaks in the United States (2, 6, 10). Although experimental and epidemiologic studies have yet to conclusively delineate the required dose of each of these distinct SEs, in previous SEA-associated outbreaks the dose of SEA causing intoxication may have been as small as 94 ng (12). Recently, additional SEs have been identified: SEG, SEH, SEI, SEJ, SEK, SEL, SEM, SEN, SEO, SEP, SEQ, SER, and SEU (21, 23). Many of these newly discovered enterotoxins are structurally similar to the classic enterotoxins, which suggests that they also may illicit foodborne illness when consumed in large enough doses (23). The significance of these SEs in causing foodborne intoxication remains largely unknown and requires both future research and increased surveillance (2, 3, 23).

SE intoxication is believed to be the second largest cause of foodborne disease in the United States and is estimated to impart an annual economic burden of $1.5 billion (2, 6, 28). S. aureus often is carried by both humans and animals; as a result, it is easily introduced into food directly from food animals and by-products or as a consequence of improper handling and preparation (2, 15, 22). S. aureus is frequently associated with dairy cows and the dairy environment and is commonly the etiologic agent of mastitis, a problematic disease often found in dairy herds (5, 6, 10, 28). S. aureus may be carried by healthy cows and mastitic dairy cows and can easily be shed into the milk during collection (5, 6, 10, 28). Once the milk is contaminated, SEs can be produced when the milk is not cooled quickly and/or is not efficiently pasteurized (21, 25). To encourage proper pasteurization, the U.S. Food and Drug Administration Pasteurized Milk Ordinance of 2005 requires the use of sterile equipment and storage temperatures of ≤7°C for raw milk and ≤10°C for pasteurized milk throughout harvesting, collection, and transportation (34). These regulations are routinely enforced to help assure a safe milk supply, but when these rules are ignored or not followed diligently, conditions capable of allowing pathogen growth and enterotoxin production may occur. Consumption of raw milk may be associated with a greater risk of SE intoxication in comparison with pasteurized milk because no microbiological "kill step" has been implemented to eliminate bacteria introduced during the production and collection process (6, 23).

The trend of increasing raw milk consumption in the United States increases concerns associated with 5. aureus contamination (36, 37). Risk managers, milk producers, and government agencies must be able to assess milk collection and processing conditions and regulations to determine whether current practices used by raw milk producers and retailers are adequate for assuring public health safety. The introduction of hazard analysis and critical control point (HACCP) practices in the food and dairy industries can help these groups address issues of food safety and safe manufacturing processes by identifying steps within production and processing that are crucial to controlling food safety risks (33, 35).

Quantitative microbial risk assessments (QMRAs) for milk and other foods can be used to assist food safety regulators and manufacturers in identifying milk processing, storage, and retail practices that can encourage pathogen introduction and proliferation (16). Once such practices are identified, QMRAs can lead to the development of more effective HACCP programs and good manufacturing practices to ensure adequate consumer protection from bacterial pathogens and toxins in raw milk (16).

QMRAs can be constructed using Monte Carlo sampling of user determined input distributions such as contamination levels and consumption patterns (16, 25, 39). Monte Carlo methods allow representation of both the uncertainty in data inputs and assumptions and the inherent variability associated with these values (16, 25, 39). Uncertainty represents unknowns that can be reduced with further information, whereas variability describes the natural variation associated with an input (16, 39). A complete QMRA has been defined by the Codex Alimentarius to require four vital components: hazard identification, hazard characterization (or dose-response), exposure assessment, and risk characterization (8). When these four topics are appropriately addressed, a QMRA can be developed using computer software capable of performing thousands of random Monte Carlo samplings from the user defined input distributions to represent factors such as potential servings, levels of bacteria, temperature and storage time scenarios, and toxin and bacterial exposure scenarios. The results from these thousands of computer simulations generate the probability of certain risks in a population for different situations and can quickly be generated and analyzed (16, 25, 39).

Past QMRAs have been limited because of the absence of data regarding inputs such as microbiological levels, dose-response studies, and consumption and exposure patterns (19, 40). Because of these significant data gaps, QMRAs have relied upon numerous assumptions to complete the risk assessments and to ultimately estimate the probability of certain risks occurring for a population (19). Even with these uncertain assumptions and identified data gaps, QMRAs are evolving to become powerful tools for researchers, risk managers, health officials, and government agencies to assist in the identification of potential food safety problems and risks to the public. One especially promising use of QMRAs is development of "what if" scenarios to model instances and/or conditions that may create or reduce opportunities for foodborne outbreaks to occur. QMRAs can help guide decisions for changes and improvements in food production and manufacturing processes and ultimately help reduce foodrelated health risks. Once a QMRA has been completed, it can be used to identify areas of scientific research that require further investigation.

The first objective of this study was to determine the risk associated with SE poisoning from the consumption of raw milk. Information used in QMRAs included semiquantitative data of S. aureus levels in raw bulk tank milk, three possible temperature and time growth situations to model pathogen growth, a formula to simulate SEA production within milk in each temperature and time growth scenario, and consumption survey patterns to ultimately model overall consumer exposure. The process is illustrated in Figure 1. The second objective of the study was to demonstrate the utility of QMRAs for guiding the development of regulations and processing conditions to improve overall food safety and reduce health risks for consumers.

MATERIALS AND METHODS

Microbial concentration distributions. 5. aureus levels (CFU per milliliter) were obtained from bulk tank milk data originally collected from various California dairies for surveillance and monitoring of milk microbiological quality and standards. This data set was generously provided by the Dairy Food Safety Laboratory at the University of California-Davis Veterinary Medicine Teaching and Research Center (VMTRC; Tulare, CA). These microbiological assays were performed to assist producers in tracking the microbiological quality of their milk and determining whether their milking and hygienic practices were appropriate and effective. S. aureus levels were used to construct probability distributions representing the levels within raw bulk tank milk samples. These data were collected from 2,336 dairies and 4,820 bulk tanks within California from 2005 to 2008. Samples that did not have a unique creamery label were not included in the count of dairies, although the results of the assays from these samples were included in the construction of the bacterial concentration distributions to maximize the amount of data used in the risk assessment. During this time, 51,963 milk samples were analyzed, and 13,157 (25.3%) of these samples contained S. aureus. Each data point represents 0.01 ml of milk plated by the spiral plating method onto agar containing washed bovine red blood cells and subsequently evaluated and quantified by expert laboratory technicians at the VMTRC to determine the S. aureus level (CFU per milliliter) in that dairy's bulk tank (6). Samples sent into the VMTRC from the originating dairies may have been split into subsamples for microbiological plating at the VMTRC. In these instances, every individual plated aliquot was used to represent a S. aureus level data point to construct the distribution of S. aureus levels for the risk assessment.

Accurate quantification of the bacterial levels for use in a QMRA is complicated by the original purpose of these samples as a means to monitor quality and other bacterial issues in dairy bulk tanks and by the physical limitation imposed by the 0.01 -ml sample volumes. A direct result of the 0.01 -ml limitation is that the maximum bacterial level observed in this data set was limited to and recorded as 40,000+ CFU/ml. The + symbol is an annotation implemented by this laboratory to represent a sample containing greater than 40,000 CFU/ml. Similar annotation was used once the level reached 10,000 CFU/ml (recorded as 10,000+), 20,000 CFU/ ml (recorded as 20,000+), and 30,000 CFU/ml (recorded as 30,000+). These annotations were made to minimize costs to customers and to maximize time and resources at this diagnostic laboratory. In the construction of an ideal QMRA, data for the exact level would be enumerated from assays implementing plating of larger milk aliquots (i.e., 1 ml). The use of 1-ml milk samples would eliminate the currently required assumption that the bacterial cells found in the 0.0 1-ml sample are uniformly distributed in the inferred final output of bacterial level.

Three representative probability distributions of bacterial levels were constructed using @Risk 5.0 (Palisade Corporation, Ithaca, NY) to represent the observed S. aureus levels in raw bulk tank milk. The first distribution is a histogram (Histogram) of the actual data set of bacterial levels, and it allows for random sampling of values that were actually observed from 2005 through July 2008, but it does not allow any bacterial levels greater or less than the observed minimum and maximum values to be generated. The second distribution of S. aureus levels, Histogram2, addressed the data points that were recorded as 10,000+, 20,000+, 30,000+, and 40,000+ CFU/ml by inserting a uniform distribution of bacterial levels within each strata and then constructing an overall histogram of these results. Using this method, every iteration of the risk model generated a slightly different histogram because a random value selected by the program for each of the recorded values in the 10,000 to 19,999, 20,000 to 29,999, 30,000 to 39,999, and 40,000 to 49,999 CFU/ml ranges. This method allows for representation of some of the uncertainty surrounding these semiquantitative data points. These two distributions of bacterial levels are able to represent the observed data set well because only values contained within the limits of the original data set of bacterial levels were sampled during each model iteration. A major drawback to these distributions is that S. aureus levels greater than those observed or allowed by the inherent limitations of the assay cannot be sampled. Because of the possible significance of these extreme unsampled microbiological levels, a 10,000 iteration run using Latin Hypercube sampling of the Histogram distribution was fit using @Risk 5.0, resulting in generation of an inverse Gaussian distribution. The advantage to using this distribution in the risk assessment is that it can generate concentration values that were not actually observed within the limits of the assay while maintaining the overall distribution pattern observed in the original VMTRC bulk tank milk data set.

Each of these distributions has unique advantages and disadvantages for inclusion into the model. The Histogram distribution relies only upon observed CFU per milliliter values and thus is limited to a broad grouping for the collected 10,000+, 20,000+, 30,000+, and 40,000+ CFU/ml values. The Histogram2 distributions addresses the uncertainty issue surrounding the "+" annotations in the Histogram distribution by allowing for CFU per milliliter counts to be generated or sampled within these 10,000 CFU/ml ranges. However, a weakness of the Histogram2 distribution is that it still cannot generate values that exceed the upper limit of the 0.01 ml assay (40,000+ CFU/ml). Because values above 49,999 CFU/ml may occur and could play an important role in causing foodborne illness, the inverse Gaussian distribution was included to allow for S. aureus levels higher than the 40,000 CFU/ml and the 49,999 CFU/ml upper limits in the Histogram and Histogram2 distributions. The inclusion of all three distributions provided a means of viewing and analyzing the impact that each of these assumptions has on the final level of risk output.

Within the risk assessment, a discrete probability of S. aureus presence of 25.3% was used based upon calculation of S. aureus presence in bulk tank milk samples during sampling period of 2005 through July 2008. When S. aureus was found in a sample, the logtransformed value (log CFU per milliliter, converted for convenience) was recorded from each of the three distributions of bacterial levels and used throughout each of the growth and toxin modules created in this risk assessment (Fig. 2). Conversely, there is a 74.7% probability of S. aureus absence in this model, and in those iterations 0 CFU/ml was recorded to calculate the final risk (Fig. 1).

Consumption distributions. To represent potential exposure to 5. aureus and SEA, consumption distributions were constructed to describe normal milk consumption patterns within the United States. The National Health and Examination Survey (NHANES) is a national initiative that began in the 1960s to study the health and nutritional status of U.S. residents of all age groups by examining consumption and lifestyle patterns of groups of sample participants (38). Using the NHANES 2003 and 2004 data, 2 days of consumption data from 10,122 individuals regarding actual servings of nonfat, 1%, 2%, and whole milk were obtained and grouped for this QMRA into the following age categories: <1 year, 1 to 2 years, 3 to 5 years, 6 to 12 years, 13 to 19 years, 20 to 60 years, and 61 to 85 years. Each age group's consumption pattern was then fit using @Risk 5.0's best-fit function to determine the most representative consumption distribution for each age group's observed consumption data (38). Milk serving sizes in the NHANES 2003-2004 data set were recorded in grams, although toxin production and microbial level were recorded and generated in milliliters. The risk model assumed the density of milk to be 1.025 g/ml to standardize milk consumption and bacterial contamination levels.

Growth modules. Using the Histogram, Histogram2, and inverse Gaussian distributions of 5. aureus levels, "what-if ' scenarios were developed to model potential storage times, temperatures, and growth of S. aureus within bulk tanks. Two unique growth formulas were constructed from the Pathogen Modeling Program (PMP) developed by the U.S. Department of Agriculture, Agricultural Research Service, Eastern Regional Research Center (USDA-ARS-ERRC; Wyndmoor, PA) (31) and the Combined Database for Predictive Microbiology (ComBase) developed by the Food Standards Agency (London, UK), the Institute of Food Research (Colney, Norwich, UK), the USDAARS-ERRC, and the Food Safety Centre (Hobart, Australia) (9). The PMP is one of the most well-recognized and used predictive microbiological software programs and allows free and rapid access to data concerning microbial growth under different environmental conditions and on different substrates (4, 17). Nevertheless, there are noted weaknesses associated with the PMP that occasionally could affect its validity and practicality for predictive microbiology of bacteria within food matrices (4). The program does not conveniently allow the user to obtain and/or analyze the data upon which the program calculates growth curves, and the majority of the PMP database is generated from bacterial growth experiments conducted in sterilized laboratory media (4). The PMP growth version included in this QMRA was used to generate growth and lag time data for S. aureus in broth culture at a pH of 6.7. In addition to the PMP data, ComBase growth data also was obtained for S. aureus in previously heated (i.e., cooked, baked, or pasteurized) nonsterilized milk at 100C and in raw milk at 20, 25, and 300C to help address the effects of substrate on bacterial growth rate (Table 1) (9). Because of the ubiquity and popularity of both the PMP and the ComBase database, this QMRA incorporated growth data from each to clearly illustrate the final impact on the risk output calculated from using either database.

The model's growth formulas were obtained by plotting the square root of the maximum PMP or ComBase growth rate versus temperature and by linear regression (Fig. 3) (24, 25). To determine the formula for lag time, a similar plot was constructed based on the square root of the inverse lag time (Fig. 4) (13, 24, 25). For the variable growth module, a Pert temperature distribution with a minimum of 3°C, a most likely (i.e., mode) temperature of 4.5°C, and a maximum of 200C was developed to model potential storage temperatures. A Pert distribution, a form of the Beta distribution, was chosen because it can be used to input expert opinions on the minimum, maximum, and most likely values while placing a greater emphasis on the most likely value rather than on the minimum and maximum values, as is done in the triangle distribution and can lead to an oversampling of values not likely to be observed in our system (40). The maximum temperature of 200C was chosen because it was the upper category observed and/or recorded in the 1999 food temperature evaluation of consumer refrigerators (1). A uniform time distribution of 0, 1, 2, 4, 6, 12, and 24 h was used to simulate potential storage times. Using these time and temperature distributions and the growth formulas, S. aureus growth scenarios under proper and improper storage conditions for raw milk were developed to observe the potential risk associated with up to 24 h of improper storage in the bulk tank or later in the milk production and storage process.

The second growth module, the static growth module, was built from the same temperature distribution as the variable growth module but calculated the S. aureus growth at static time intervals of 0, 1, 2, 4, 6, 12, and 24 h to allow for deterministic time comparisons. This module effectively illustrated the proliferating risk as a function of time in varying instances of temperature abuse or mishandling. The utility of the variable growth module and the static growth module together is that they allow the risk manager and milk producer to observe the ultimate consequences of periods of short or long times and variable temperature conditions and to identify the overall impact that each of these factors can have on the quality and safety of the product.

The final module, the long-term consumer storage module, applied the same growth formulas (PMP and ComBase) as the two other modules but used novel time and temperature distributions to observe the impact that consumer storage practices have on a consumer's associated health risk. A Pearson distribution to describe refrigerator temperatures was developed from a best fit performed by @Risk 5.0 following 10,000 iterations of the histogram data representing actual product temperatures in consumer refrigerators as recorded from the 1999 food temperature evaluation (1). A time distribution of 0 to 336 h (2 weeks) was used to attempt to simulate the practices that raw milk consumers may follow to store milk in the home based upon the stated shelf life of raw milk obtained from a well-known raw milk producer and the presumed average consumer milk storage time of 1 week (7).

In each growth module and every iteration of the model, the PMP and ComBase growth formulas were applied to the value selected from each of the distributions of S. aureus levels when a temperature greater than 7 °C was pulled from each module's appropriate temperature distribution. The lower temperature limit for growth was obtained from a study that included 77 S. aureus strains (8 strains from mastitic cow milk, 61 strains collected by the Swiss food control laboratories from 30 various foods, and 8 strains obtained from the human nasopharyngeal region) grown in brain heart infusion medium in temperature grathent incubators (27). When a temperature below 7°C was selected, the model recorded the original microbial level as the final output. The maximum S. aureus population density of 10^sup 9^ CFU/ml (9 log CFU/ml) was applied throughout the model, preventing the generation of unrealistic microbial populations within milk (25). A population of 10^sup 5^ CFU/ml was used as the threshold of concern because this level has been previously accepted as a proxy for potential enterotoxin risk in past risk assessments and food safety decisions and regulations (25, 32).

SEA production module. Foodborne intoxication from S. aureus results from consumption of preformed enterotoxins. In a study of SEA poisoning from the consumption of chocolate milk in the United States, a dose as small as 94 ng was considered sufficient to elicit intoxication (12). The SEA production module uses the construct of the variable growth module but integrates an additional step to generate a level of SEA in a consumed serving. Modeling of potential SEA concentrations (nanograms per milliliter) produced in milk by a toxin-producing strain of S. aureus was accomplished using a previously published formula that has a zero-order relationship with time and a temperaturedependent rate constant (p = 0.03767/ - 0.559) for SEA production at constant temperatures in milk (14). SEA production was allowed by the risk module only when specific conditions were met regarding microbial population density, temperature, and time outputs of the variable growth module. SEA is not produced when the 5. aureus density is less than 10^sup 5^ CFU/ml (5 log CFU/ml) and at temperatures of less than 15°C (13, 18, 22, 32). Therefore, these conditions had to be concurrently met for the module to produce an output of SEA production (Fig. 5). Additionally, a correction factor was applied to this formula's output to account for the fact that not all S. aureus milk isolates harbor enterotoxin genes (2, 9, 30, 32). A Pert distribution (0.52 minimum, 0.67 most likely, 1 maximum) was used to correct the final SEA concentration generated for the risk model following each iteration. The Pert distribution representing the probability of enterotoxigenic isolates was based upon recent studies focusing on the percentage of raw milk S. aureus isolates harboring these genes (3, 10, 23, 28). The application of this correction factor within the model effectively reduces the total toxin output to account for this enterotoxin gene factor in an attempt to generate more realistic toxin concentrations in raw milk servings. The correction factor was applied following computation of initial SEA concentrations because production of enterotoxin is dependent on the total bacterial cell density, 105 CFU/ml (5 log CFU/ml), regardless of whether all these cells are enterotoxigenic. Cells that do not directly contribute to SEA production still are able to indirectly impact SEA generation by signaling enterotoxigenic strains to commence enterotoxin production (22).

The outputs from each age group's NHANES consumption distribution and SEA production module were combined to determine the number of servings in each consumer age group that have a large enough dose of SEA to cause illness. To combine these outputs, the value generated in each model iteration of the SEA production module (nanograms per milliliter) was multiplied by the serving size consumed generated during that iteration to generate the total nanograms consumed in the serving. The level of toxicologic concern considered was 94 ng of SEA per serving (12).

Monte Carlo simulation. Two Monte Carlo simulations for both the PMP and ComBase growth versions were performed using @Risk 5.0 with settings for 100,000 iterations of Latin Hypercube sampling using the @Risk random number generator and a different starting seed for each simulation. A comprehensive list of the formulas used during each simulation to develop the QMRA is given in Table 2.

RESULTS

Growth modules. The output results from the Histogram, Histogram2, and inverse Gaussian S. aureus distributions in all of the growth modules produced similar values using both the ComBase-based and PMP-based growth formulas. Even though the PMP-based growth formula accounted for the inclusion of lag time, this version's three bacterial distribution outputs tended to be larger (higher log CFU per milliliter) than the ComBase-based growth results. The differences between the S. aureus maximum growth rate for the two growth formulas are most likely due to differences in the growth medium used (broth culture for PMP versus raw or previously heated milk for ComBase). This finding reaffirms the common problem associated with PMP use and its overestimation of growth for microbial growth in food systems and should be considered when PMP modeling is chosen for use in a QMRA (4 ).

The growth modules results are reported with the percentiles of servings containing more than the 5 log CFU/ ml level of concern. Percentiles of servings or doses of toxin are commonly used as outputs from risk assessments as a means of determining an acceptable level of risk. The percentile used often is dependent on the type of risk considered and the population to which the risk pertains. The Environmental Protection Agency (EPA) currently uses quantitative chemical risk modeling to make decisions concerning pesticide regulations (11). The EPA considers exposure to a chemical residue to be acceptable when exposure to the consumer at the 99.9th percentile does not exceed the toxicological level of concern (11). For this microbial risk assessment, conclusions about the safety of milk containing S. aureus and/or SEA also were made on the basis of the 99.9th percentile of exposure.

Variable growth module. This module used a uniform time distribution of 0, 1, 2, 4, 6, 12, and 24 h and the same temperature distribution (Pert: 3, 4.5, 2O0C) as that of the static growth module. The variable growth module calculated similar results using both the ComBase-based and PMP-based growth scenarios (Table 1). The results from the ComBase version of the variable growth module generated levels above 5 log CFU/ml for the 99.99th percentile using the Histogram and Histogram2 distributions of bacterial population levels and slightly below 5 log CFU/ml (Histogram, 4.97 log CFU/ml; Histogram2, 4.82 log CFU/ ml) at the 99.9th percentile. The inverse Gaussian distribution was above the level of concern for both the 99.9th and 99.99th percentiles. The PMP version of the variable growth module produced results at the level of concern for the 99.9th percentile using all distributions: Histogram, 5.31 log CFU/ ml; Histogram2, 5.16 log CFU/ml; and inverse Gaussian, 5.34 log CFU/ml. Although not all of the growth versions and distributions of bacterial levels in the ComBase version generated values at the level of concern for the 99.9th percentiles of servings, all three bacterial level distributions and growth formula versions did report values above the 5 log CFU/ml level of concern for the 99.99th percentiles of servings and samples (Table 3). These results indicate that there is potential for enterotoxin production and presence within both the 99.9th and 99.99th percentiles of milk servings, and thus there may be some risk of food poisoning.

Static growth module. The static growth module was used to highlight the differences between the deterministic times of 0, 1, 2, 4, 6, 12, and 24 h of storage and temperatures ranging from 3 to 20°C (Pert: 3, 4.5, 20°C). The ComBase growth version produced results indicating that milk samples and servings within the 99.9th percentile would contain S. aureus at the level of concern, 5 log CFU/ ml, after approximately 12.5 h of storage for the Histogram and Histogram2 bacterial distributions and at 0 h for the inverse Gaussian bacterial distribution (Table 4). The PMP version of the static growth module produced results similar to those of the ComBase version but required shorter incubation or storage times to reach the level of concern. The PMP version generated values indicating that the 99.9th percentile of samples would contain S. aureus at the level of concern after 9.5 to 10 h using the Histogram and Histogram2, whereas the inverse Gaussian bacterial distributions generated this level of concern at 0 h as calculated in the ComBase version (Table 4). The results from the inverse Gaussian distribution for both growth versions illustrate the impact of including bacterial population values above the 40,000 CFU/ml limit; with this distribution, the model calculated that servings in the 99.9th percentile would be above the critical limit without storage time. This result illustrates the effect that the bacterial distribution can exert on the final results of a QMRA and the importance of analysis to determine how well a chosen distribution represents actual bacterial levels. Overall, the ComBase and PMP outputs from this module indicate how quickly S. aureus can multiply to levels capable of potentially producing SEA. These results clearly indicate the importance of consistent and regulated temperature control of raw milk.

Long-term consumer storage module. Results from this module simulated conditions that could occur within a consumer's refrigerator during normal storage of milk in the home. This module is based upon assumptions about how individuals store and use milk in the United States. However, the behaviors of consumers who purchase and store raw milk may not be represented well by the practices used by consumers who store pasteurized milk. In the future, inclusion of practices used by actual consumers of raw milk would more accurately predict potential raw milk storage and consumption patterns. The results using the ComBase version of this module indicated that servings of milk stored in refrigerators in the 98.5th percentile would have S. aureus levels above the level of concern (5 log CFU/ml) under common storage and consumption conditions (Table 5). Similarly, the PMP version calculated that the level of concern would be breached in servings contained within the 97.5th percentile (Table 5).

SEA production module. Both SEA production modules (ComBase and PMP) produced results indicating that the 99.9th percentile of consumers would not likely be at risk from SEA doses large enough to cause illness under conditions generated by the variable growth module temperature and time distributions. None of the age groups reached the 94-ng dose (Table 6). The requirements of 5. aureus at 5 log CFU/ml and a temperature of ^150C affect the output from the simulation greatly, and the unlikely occurrence of these values selected from each of these distributions plays a significant role in minimizing the anticipated production of SEA. This module effectively demonstrates that generation of SEA doses large enough to cause illness can occur within all ages based on consumption practices for the 99.99th percentile of each age group (Table 6). Based on the 99.9th percentile cutoff frequently assumed to represent a reasonable risk, raw milk servings do not appear to pose a significant health risk from SEA intoxication (11). The ultimate utility of this SEA production module is not necessarily the final risk (dose) output but rather the ability to discern potential doses of SEA and other SEs in raw milk servings once additional information becomes available. With more SE data, a greater understanding of raw milk risk potential can be obtained through the construction of more rigorous QMRAs.

DISCUSSION

This risk assessment incorporated a model capable of determining the probability of consumer exposure to S. aureus and SEA from consumption of raw milk. In addition to the determination of the probabilities, this QMRA also identifies data gaps and/or areas that require further research. The assumptions made within this risk assessment were necessary to complete the QMRA, and now that these assumptions have been identified they can be progressively addressed and modified as new scientific data become available. The inclusion of surveillance data from the VMTRC on microbial levels was a significant improvement over previous QMRAs because the VMTRC provided an extremely large data set on S. aureus levels in raw bulk tank milk. In the future, a microbial assay protocol for complete 5. aureus quantification is needed for samples with loads greater than the current maximum of 40,000 CFU/ml and for samples with levels below the current minimum of 100 CFU/ml. This improved precision would reduce the uncertainty surrounding the population limits and therefore eliminate the need for multiple distributions of bacterial population levels.

Growth curve experiments conducted with raw milk, including associated lag times, should be conducted to take into account the significance of competitive microflora, milk antimicrobials, and other milk components that may affect S. aureus growth and survival (15, 25). Future research on the relationship between growth and enterotoxin production by S. aureus strains also must be conducted to more accurately model toxin production and the probability of illness from S. aureus present in raw milk. Although studies to determine the percentage in milk of S. aureus isolates carrying enterotoxin genes have been performed, additional experiments must be completed to study the expression pattern of those enterotoxin genes when present in milk and in various other environmental settings (6, 21, 23, 28, 30). It is also important to gain a greater understanding of the relationship between dose and response following consumption of each individual SE to develop a more realistic QMRA for S. aureus in raw milk. Currently, toxic levels for even the most common enterotoxin associated with foodborne illness, SEA, range from 94 ng to 1 µg, and virtually nothing is known about the remaining SEs (12, 23). Another important avenue of research would be determination of whether the health status of the consumer (i.e., immunocompromised, pregnant, old, or young) affects the dose of enterotoxin required to cause illness. An improved QMRA of raw milk will require an understanding of both the kinetics of production for each enterotoxin under varying environmental conditions and the dose of each enterotoxin required to cause illness in a variety of individuals (15, 21).

The field of QMRAs has infrequently integrated second order modeling to treat uncertainty and variability separately within the simulation model because of the complexity of such an approach (19, 39). An accurate separation of these two aspects in a risk model can create a higher level of confidence in the risk outputs and assist in identifying remaining uncertain aspects that require more research to increase the accuracy of the QMRA (39). Within the current risk assessment, uncertainty and variability were not separated, and this aspect of the method could be improved upon in the future to provide a more accurate assessment of the health risks associated with raw milk. However, even with these assumptions, data gaps, and the combination of uncertainty and variability, this work provides an example of how to estimate the possible risk a consumer takes when choosing to consume raw milk. This QMRA of S. aureus in raw milk provides new information that can be used to assist consumers, the dairy industry, and public health regulators in making informed decisions about the consumption, regulation, and processing and distribution of raw milk while providing a necessary foundation from which to build future QMRAs to ultimately improve food safety in the United States and throughout the world.

References

REFERENCES

1. Audits International. 1999. Audits/FDA temperature databases. Food Safety Risk Analysis Clearinghouse. Available at: http://foodrisk.org/ exclusives/audits/index.cfm. Accessed 8 July 2008.

2. Balaban, N., and A. Rasooly. 2000. Staphylococcal enterotoxins. Int. J. Food Microbiol. 61:1-10.

3. Bania, J., A. Dabrowska, J. Bystron, K. Korzekwa, J. Chrzanowska, and J. Molenda. 2006. Distribution of newly described enterotoxinlike genes in Staphylococcus aureus from food. Int. J. Food Microbiol. 108:3641.

4. Baranyi, J., and M. L. Tamplin. 2004. ComBase: a common database on microbial responses to food environments. /. Food Prot. 67:19671971.

5. Bennet, R 2005. Staphylococcal enterotoxin and its rapid identification in foods by enzyme-linked immunosorbent assay-based methodology. /. Food Prot. 68:1264-1270.

6. Cenci-Goga, B. T., M. Karama, P. V. Rissitto, R. A. Morgante, and J. S. Cullor. 2003. Enterotoxin production by Staphylococcus aureus isolated from mastitic cows. /. Food Prot. 66:1693-1696.

7. Claravale Dairy. 2008. Frequently asked questions. How long will raw milk last? Available at: http://claravaledairy.com/. Accessed 6 September 2008.

8. Codex Alimentarais. 1999. Principles and guidelines for the conduct of microbial risk assessment. CAC/GL-30. Food and Agriculture Organization, Rome.

9. ComBase. 2008. A combined database for predictive microbiology. Available at: http://www.combase.cc/default.html. Accessed 19 November 2008.

10. Cremonesi, P., M. Luzzana, L. Brasca, S. Morandi, R. Lodi, C. Vimercati, D. Agnellini, G. Caramenti, P. Moroni, and B. Castiglioni. 2005. Development of a multiplex PCR assay for the identification of Staphylococcus aureus enterotoxigenic strains isolated from milk and dairy products. MoI. Cell. Probes 19:299-305.

11. Environmental Protection Agency. 2000. Choosing a percentile of acute dietary exposure as a threshold of regulatory concern. Available at: http://www.epa.gov/oppfeadl/trac/science/trac2b054.pdf. Accessed 13 November 2008.

12. Evenson, M. L., M. Hinds, B. Ward, S. Robert, and M. S. Gergdoll. 1988. Estimation of human dose of staphylococcal enterotoxin A from a large outbreak of staphylococcal food poisoning involving chocolate milk. Int. J. Food Microbiol. 7:311-316.

13. Fu, B., P. S. Taoukis, and T. P. Labuza. 1991. Predictive microbiology for monitoring spoilage of dairy products with timetemperature integrators. /. Food Sci. 56:1209-1215.

14. Fujikawa, H., and S. Morozumi. 2006. Modeling Staphylococcus aureus growth and enterotoxin production in milk. Food Microbiol. 23:260-267.

15. Halpin-Dohnalek, M., I. Magaret, and H. Elmer. 1989. Growth and production of enterotoxin A by Staphylococcus aureus in cream. /. Dairy Sci. 72:2266-2275.

16. Jaykus, L. 1996. The application of quantitative risk assessment to microbiological safety risks. Crit. Rev. Microbiol. 22:279-293.

17. Koseki, S., and I. Seiichiro. 2005. Prediction of pathogen growth on iceberg lettuce under real temperature history during distribution from farm to table. Int. J. Food Microbiol. 104:239-248.

18. Le Loir, Y., F. Baron, and M. Gautier. 2002. Staphylococcus aureus and food poisoning. Genet. MoI. Res. 1 :63-76.

19. Lindqvist, R., S. Sylvén, and I. Vâgsholm. 2002. Quantitative microbial risk assessment exemplified by Staphylococcus aureus in unripened cheese made from raw milk. Int. J. Food Microbiol. 78: 155-170.

20. Mead, P. S., L. Slutsker, V. Diets, L. F. McCaig, J. S. Bresee, C. Shapiro, P. M. Griffen, and R. V. Tauxe. 1999. Food-related illness and death in the United States. Emerg. Infect. Dis. 5:607-625.

21. Morandi, S., M. Brasca, R. Lodi, P. Cremonesi, and B. Castiglioni. 2007. Detection of classical enterotoxins and identification of enterotoxin genes in Staphylococcus aureus from milk and dairy products. Vet. Microbiol. 124:66-72.

22. Novick, R. P. 2003. Autoinduction and signal transduction in the regulation of staphylococcal virulence. MoI. Microbiol. 48: 1429-1449.

23. RaU, V. L. M., F. P. Viers, R. RaU, R. L. Vieitis, A. Fernandez, Jr., J. M G. Candeias, K. F. G. Cardoso, and J. P. Araújo, Jr. 2008. PCR detection of staphylococcal enterotoxin genes in Staphylococcus aureus strains isolated from raw and pasteurized milk. Vet. Microbiol. 132:408413. DOI: 10.1016/j.vetmic.2008.05.011.

24. Ratkowsky, D A., J. OUey, T. A. McMeekin, and A. BaU. 1982. Relationship between temperature and growth rate of bacterial cultures. /. Bacteriol. 149:1-5.

25. Rho, M., and D. W. Schaffner. 2007. Microbial risk assessment of staphylococcal food poisoning in Korean kimbab. Int. J. Food Microbiol. 116:332-338.

26. Schmidt, R. H, and M. P. Davidson. 2008. MUk pasteurization and the consumption of raw milk in the United States. Food Prot. Trends 28:45^7.

27. Schmitt, M., U. Schuler-Schmid, and W. Schmidt-Lorenz. 1990. Temperature limits of growth, TNase and enterotoxin production of Staphylococcus aureus strains isolated from foods. Int. J. Food Microbiol. 11:1-20.

28. Srinivasan, V., A. A. Sawant, B. E. GUlespie, S. J. Headrick, L. Ceasaris, and S. P. OUver. 2006. Prevalence of enterotoxin and toxic shock syndrome toxin genes in Staphylococcus aureus isolated from milk of cows with mastitis. Foodborne Pathog. Dis. 3:274-283.

29. Su, Y., and A. C. Lee Wong. 1997. Current perspectives on detection of staphylococcal enterotoxins. /. Food Prot. 60:195-202.

30. Sutherland, J. P., A. J. BayUss, and T. A. Roberts. 1994. Predictive modelling of growth of Staphylococcus aureus: the effects of temperature, pH and sodium chloride. Int. J. Food Microbiol. 21: 217-236.

31 . U.S. Department of Agriculture, Agricultural Research Service. 2006. Pathogen modeling program, version 7.0. Available at: http://www. ars.usda.gov/naa/errc/mfsru/pmp. Accessed 19 November 2008.

32. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 1992. Foodborne pathogenic microorganisms and natural toxins handbook Available at: http://www.cfsan.fda.gov/-mow/ chap3.html. Accessed 12 September 2008.

33. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 2001. HACCP: a state-of-the-art approach to food safety. Available at: http://www.cfsan.fda.gov/-lrd/bghaccp.html. Accessed 12 November 2008.

34. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 2005. Grade "A" pasteurized milk ordinance (2005 revision). Available at: http://www.cfsanfda.gov/-ear/pmo05-4. html. Accessed 15 December 2008.

35. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 2006. Hazards and controls guide for dairy foods, HACCP. Available at: http://www.cfsan.fda.gov/-ear/daihaz.html. Accessed 12 November 2008.

36. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 2007. Raw milk-associated pubUc health risks. Available at: http://www.cfsan.fdagov/-dms/rawmiUc2.html. Accessed 19 September 2008.

37. U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition. 2007. FDA and CDC remind consumers of the dangers of drinking raw milk. Available at: http://www.fda.gov/bbs/ topics/NEWS/2007/NEW01576.html. Accessed 19 September 2008.

38. U.S. National Center for Health Statistics. 2008. National health and nutrition examination survey, NHANES 2003-2004 data file. AvaUable at httpy/www.cdc.gov/nchs/about/major/nhanes/nhanes2003-2004/ nhanes03_04.htm. Accessed 8 July 2008.

39. Vicari, A. S., A. Mokhatari, R. A. Morales, L. Jaykus, H. C. Frey, B. D. Slenning, and P. Cowen. 2007. Second-order modeling of variability and uncertainty in microbial hazard characterization. /. Food Prot. 70:363-372.

40. Vose, D. 2005. Risk analysis: a quantitative guide, 2nd ed. John WUey & Sons Ltd., Chichester, UK.