Prevalence of antibiotic-resistant E. coli in retail chicken: comparing conventional, organic, kosher, and raised without antibiotics

Retail poultry products are known sources of antibiotic-resistant Escherichia coli, a major human health concern. Consumers have a range of choices for poultry, including conventional, organic, kosher, and raised without antibiotics (RWA) – designations that are perceived to indicate differences in quality and safety. However, whether these categories vary in the frequency of contamination with antibiotic-resistant E. coli is unknown. We examined the occurrence of antibiotic-resistant E. coli on raw chicken marketed as conventional, organic, kosher and RWA. From April – June 2012, we purchased 213 samples of raw chicken from 15 locations in the New York City metropolitan area. We screened E. coli isolates from each sample for resistance to 12 common antibiotics. Although the organic and RWA labels restrict the use of antibiotics, the frequency of antibiotic-resistant E. coli tended to be only slightly lower for RWA, and organic chicken was statistically indistinguishable from conventional products that have no restrictions. Kosher chicken had the highest frequency of antibiotic-resistant E. coli, nearly twice that of conventional products, a result that belies the historical roots of kosher as a means to ensure food safety. These results indicate that production methods influence the frequency of antibiotic-resistant E. coli on poultry products available to consumers. Future research to identify the specific practices that cause the high frequency of antibiotic-resistant E. coli in kosher chicken could promote efforts to reduce consumer exposure to this potential pathogen.

The use of antibiotics in livestock production may pose health risks to humans, as such usage has been correlated with the occurrence of antibiotic-resistant bacteria isolated from human infections 1,2 . Methods of livestock production differ in antibiotic use, and this can influence the frequency of antibiotic-resistant bacteria on retail meats. For example, antibiotic-resistant Escherichia coli has been shown to be less common on poultry raised without antibiotics (RWA) as compared to poultry raised conventionally 3 . Likewise, organic poultry can have lower frequencies of antibiotic-resistant bacteria than poultry raised conventionally 4-10 , although this is not always the case 11-13 . Organic, RWA, and kosher food products supply a growing market niche 14 . Consumers perceive that they offer health benefits [14][15][16][17][18][19][20][21] and are willing to pay a premium for them [22][23][24] . The actual health benefits of organic food are not always clear 25 , and the health benefits of kosher foods are largely anecdotal. Little is known about the frequency of antibiotic-resistant microorganisms on kosher products.
The organic and RWA labels require specific production methods as stipulated in US federal regulations, whereas the kosher label adheres to religious requirements that are regulated privately. The RWA label requires that "livestock have never received antibiotics from birth to harvest" 26 . The United States Department of Agriculture (USDA) organic standard is only slightly less strict, stipulating that "The producer of an organic livestock operation must not sell, label, or represent as organic any animal or edible product derived from any animal treated with antibiotics", but also that "Poultry or edible poultry products must be from poultry that has been under continuous organic management beginning no later than the second day of life" 26,27 . Therefore, injecting antibiotics into eggs or administering them during the first 24 hours of the chick's life will not violate the letter of the USDA organic standard 28,29 . Kosher production differs from organic and RWA in that it is inherently predicated on religious requirements. For kosher meat, the major requirements are that it must be from animals that have split hooves and chew their cud, it must not be mixed with dairy products, and all equipment used must be used exclusively for kosher food 19 . Animals must be slaughtered "humanely", and meat is typically salted to remove blood rapidly, a practice that has been shown to reduce the microbial load 30 . Unlike for organic and RWA, kosher poultry is not regulated by Federal laws but rather by private certification organizations, and thus the specific practices vary 19 .
Here, we compared four major types of poultry-conventional, kosher, organic, and RWA-in order to assess the frequency of contamination with antibiotic-resistant E. coli. We focused on poultry products from a major metropolitan center (the greater New York City area) and products available to typical consumers by studying multiple brands of chicken from multiple stores. Our goal was to compare the frequency of antibiotic-resistant E. coli in these four categories of chicken.

Sample collection
During April-June 2012, raw chicken was purchased from supermarkets, butcher shops, specialty stores, and food distributors in the greater New York City area. A variety of widely available brands were procured in four categories: conventional, kosher, organic and RWA. Within each category of chicken purchased, we collected at least four samples of each brand. Some samples included more than one category (e.g., kosher and organic). Five collections occurred resulting in 213 total samples. Samples were drumsticks or samples from which drumsticks were removed for analysis (all with skin). After purchase, each chicken sample was placed in a labeled, ziplock bag, and placed in a cooler with ice packs. Three coolers with ice packs were shipped overnight to T-Gen North within two days of collection.

Laboratory analyses
Chicken samples arrived at the laboratory in their original packaging and were refrigerated at 4°C until processed. One putative E. coli strain was isolated and screened from each sample using standard methods for assaying for antimicrobial resistance described by the Clinical and Laboratory Standards Institute (CLSI) 31 . The use of one strain per sample enabled efficient testing among a population of chicken samples for differences in the frequency of antibiotic resistance.
One whole drumstick was selected from each package or removed from each whole chicken sample using a sterilized knife. Each sample was transferred aseptically to a Stomacher Bag (VWR, Radon,

Changes from Version 1
We thank the reviewers for their reports. We appreciate Dr. Hanning's comments and suggestions and have revised the manuscript to address each point raised.
The number of samples collected per brand is now indicated in the text. The number of brands per category is listed as "N" in Table 1. We now state, by type of chicken collected, the percentages of isolates positive for some degree of antibiotic resistance in the fifth sentence of the Results section.
We have revised the first few sentences in the Discussion for clarity and to remove potentially inflammatory or unsupported claims. We have added several citations that provide support for our statements, and at the same time, we removed potentially inflammatory adjectives. We also add a statement about one specific estimate of the extent of antibiotic usage for growth promotion. We realize such estimates are controversial, but we feel that these provide important and useful context for readers in the field. For the second sentence in the original, we have followed Dr. Hanning's suggestion to modify the claim by changing "select" to "can select". We appreciate the point that farm-to-farm variability could play a role in our results, and have added a sentence acknowledging this explicitly. Also, the third paragraph of the Discussion discusses the potential for cross-contamination within shared production facilities to influence our results. We note that our design was developed to test for significant effects of type of chicken from the perspective of the consumer making decisions about which chicken to purchase. Thus, while farm-to-farm variability and cross-contamination are important potential sources of variation in antibiotic resistance, incorporating this variance is an important part of our design. Future, more exhaustive surveys could attempt to partition the influence of these factors, but doing so was beyond the scope of the current study.

See referee reports
(i.e., using all the data within conventional, organic, kosher, RWA). For each drug, Microsoft Excel for Mac Version 14.1.0 was used to conduct chi-square tests to determine whether the frequency of resistance varied among categories of chicken: conventional, organic, kosher and RWA.
The total number of drugs and drug classes to which each strain was resistant were enumerated. One-way ANOVA was used to compare the average number of drugs to which strains were resistant among categories, using samples with only one category designation (n=120). This test captures the effect of a consumer's choice whether to purchase chicken in one category over another on the likelihood of exposure to antibiotic-resistant E. coli.
Multi-factor ANOVA was used to test whether trends held across the broader dataset (n=184), including samples with multiple category designations. The collection of samples included adequate replication (>14) for every possible two-way combination of labels (organic & kosher, RWA & organic, and RWA & kosher). Replication for the three-way combination (organic, kosher & RWA) was low (n=5), and all samples were from one brand. To avoid bias, these samples were excluded from the ANOVA. Each of the three labeling categories was included as a factor in three-way ANOVAs (organic, RWA, and kosher, each with two levels), with the number of drugs and drug classes exhibiting resistance as response variables. This tests for the effect of each category and for interactive effects of combining categories.

Results
Across the entire dataset, resistance to cefazolin was most common (41.3%), followed by ampicillin (31.5%), tetracycline (30.4%), and ampicillin sulbactam (19.6%). Some resistance was detected for cefoxitin, (12.5%) and gentamicin (10.9% of strains), but no strain was resistant to amikacin, the other aminoglycoside tested. For the quinolones, some (3.3%) of strains were resistant to nalidixic acid, but none was resistant to ciprofloxacin. Resistance was low (3.3%) for trimethoprim sulfamethoxazole, the one folate pathway inhibitor tested, and was absent for imipenem, the one carbapenem tested. Over half of all strains collected exhibited resistance to one or more antibiotics: 55%, 58%, 60%, and 76% from conventional, RWA, organic, and kosher chicken samples, respectively.
Within categories of chicken purchased, brands did not vary in the extent of antibiotic resistance (Table 1). By contrast, categories of chicken differed in the number of drugs to which strains of E. coli were resistant ( Figure 1). Strains of E. coli isolated from kosher chicken were resistant to more drugs than were strains from the other categories (Tukey's HSD comparisons: kosher vs. conventional, P=0.023; kosher vs. organic, P=0.041; kosher vs. RWA, P=0.002).
These patterns held when analyzing the broader dataset, including the samples with multiple designations. Strains of E. coli isolated from kosher chicken samples were resistant to more drugs compared to the other categories ( Figure 2). Strains of E. coli isolated from samples in the RWA category tended to be resistant to fewer drugs but the difference was not significant versus conventional and organic which did not differ from each other. Six isolates were excluded from further analysis because they were not confirmed as E. coli using the qPCR assay.
Guidelines from the Clinical and Laboratory Standards Institute (CLSI) for disk diffusion methods 31 were used to test each strain for resistance to antibiotics. Some strains did not grow under assay conditions (n=23) and were excluded from further analysis. Twelve antibiotics were tested, representing seven classes of drugs: tetracycline (class, tetracyclines); ampicillin and ampicillin sulbactam (class, penicillins); cefazolin, cefoxitin, and ceftriaxone (class, cephalosporins); gentamicin and amikacin (class, aminoglycosides); nalidixic acid and ciprofloxacin (class, quinolones); trimethoprim sulfamethoxazole (class, folate pathway inhibitors); and imipenem (class, carbapenems) (VWR, Radon, PA). Breakpoint guidelines from the CLSI M100 Tables 2A through 2J for E. coli 31 were used to classify strains into "resistant", "intermediate" or "susceptible"; designations of "intermediate" were lumped with "resistant" for purposes of statistics and inference, a conservative approach with respect to consumer safety.

Statistical analyses
Analysis of variance (ANOVA) was used to test whether antibiotic resistance varied among the brands of chicken sampled, using SYSTAT 13.1. Effects of brand within each category were tested Our finding that brands within categories did not differ significantly in the extent of antibiotic resistant E. coli (Table 1) could arise from the fact that individual brands of chicken obtain product from multiple farms whose production practices may differ, obscuring clear patterns associated with individual brands. Our ability to detect an effect of brand might also be constrained by low statistical power. Our finding that the frequency of antibiotic resistant strains of E. coli on organic poultry did not differ significantly from conventional ( Figure 1 and Figure 2) reflects some past studies in this area that have found no difference in antibiotic resistance between organic and conventional practices 11-13 . Others found that pathogens on organic or RWA poultry products had lower resistance to antibiotics compared to conventional products 4,10,40-43 , which was the trend we  observed for RWA. The distinction between USDA organic from USDA RWA may be important, given that organic chicks can receive antibiotics via in ovo injections and during the first day of life. Previous studies have provided unequivocal evidence that even in ovo injection of antibiotics can affect the susceptibility of the bacteria that contaminate poultry products 2 . With a larger sample, the tendency for E. coli isolated from RWA samples to have lower frequency of antibiotic resistance than other categories (P=0.122; Figure 1 and Figure 2) may emerge as significant.
Cross-contamination is another possible source of antibiotic resistance 44 . Shared facilities for product and slaughter could promote cross-contamination and antibiotic strains could be spread among organism and environments 45,46 . Poultry could then be inadvertently exposed to antibiotic-resistant E. coli. For example, companies with both conventional and organic products may slaughter in the same facilities, promoting cross-contamination. Production facilities that convert from one practice to another could also experience residual contamination, though there is evidence that converting from conventional to organic can reduce frequency of resistance 8 . The identification of possible cross-contamination is outside the scope of this study, but these possibilities would need to be considered when investigating the sources of antibiotic resistance.
The increased resistance of E. coli in kosher chicken compared to conventional was surprising, because, while kosher does not stipulate anything about antibiotic use, kosher is perceived as clean and safe to consume 19 . The higher resistance found in isolates from kosher chicken (Figure 1 and Figure 2), and the distinct antibiotic-resistance profile (Table 2) suggests that use of antibiotics in the kosher production chain is common and that it may be more intensive than use of antibiotics among conventional, organic, or RWA practices. It is not immediately obvious where in the kosher chicken production process antibiotic use might be more prevalent, or where exposure to antibiotic-resistant organisms is more likely. Consumers perceive organic, kosher and RWA products to be healthier 14-21 , though the real health benefits from organic products are unclear 10 , and, to our knowledge, the actual health benefits of kosher have not been assessed. Our findings are consistent with the suggestion that some 'niche market' products, while perceived to be safer, may have higher incidence of foodborne pathogens compared to conventional products 47 .
Our study was limited in geographic and temporal scale, as we focused on the New York metropolitan area over a three-month time period. Yet, the region is large and populous, we focused on the most widely available brands in all categories, and this area particularly offered multiple kosher brands. Our final sample size was limited (n=184) but not atypical for the field 48-51 . Finally, we only assayed for generic E. coli and did not assess virulence or virulence group assignments for each sample. However, E. coli is a useful focal organism because it is widespread and an important potential pathogen.

Conventional
Organic

RWA Organic & RWA
More studies are needed to test whether antibiotic resistance among kosher products is consistently higher than conventional and other categories. Nevertheless, our study offers insight into another area of the food production system increasing the exposure of people to microorganisms that are resistant to antibiotics. In addition to regulation, more consistent surveillance or auditing would add of consumer protection, enabling improved purchase decisions based on price and health benefits guided by meaningful labels.
Author contributions JMM, ARM, JCM, LBP, and BAH designed the study. JMM and ARM collected the samples. KW and HG carried out the laboratory analyses. JMM and BAH analyzed the data. All authors contributed to writing the manuscript.

Competing interests
No competing interests were disclosed. We agree with Dr. Stärk that production methods are likely to impact resistance among generic E.

Grant information
, as our study shows. Nevertheless, the finding is important, because products available to coli consumers differ in resistance profiles, according to production practices. The finding that kosher products have higher frequencies of antibiotic resistant is inconsistent with consumers' E. coli perception of kosher as safer and healthier than other options. For this reason, we think the findings are interesting and relevant to consumers.
Regarding the statistical methods, we appreciate Dr. Stärk's comments and the opportunity to clarify and provide more information.
First, the text describing Table 2 in versions 1 and 2 should read, 'Bold text denotes significant '. No ANOVAs were used for binary differences among categories according to Chi-square test response variables (in this case, resistant or not resistant). For the test of brand effects, we used ANOVA on non-transformed count data, as reported in the text and in Table 1. We also used the Kruskal-Wallis rank-sum test, which is arguably better suited to count data as it does not require the assumption of a normal distribution (though we did not report those tests). With the Kruskal-Wallis tests, we also found no significant effect of brand within chicken categories. (We will include these results in the .) Therefore, we combined brands in our tests comparing next version drug resistance among categories, as shown in Figure 1B. As identified on the vertical axis in Figure 1B, the dependent variable in the ANOVA was the number of drugs to which strains were resistant (non-transformed). The result is identical using the Kruskal-Wallis rank sum test, which also identified a significant effect of category: chi-squared = 9.3891, df = 3, p-value = 0.02454.
We also used a general linear model assuming the Poisson distribution which is typical of count data, an approach that allows multiple mean comparisons using the 'multcomp' package in R. Tukey's post hoc tests from the GLM also supported the findings that E. coli from Kosher chicken samples were resistant to more drugs than conventional (P=0.001), organic (P=0.002), and RWA (P<0.001). Thus, our findings are robust and consistent across a number of statistical models.
We confirm that resistance was defined as a binary variable, as recommended by the Clinical and Laboratory Standards Institute, and as stated in the methods: 'Breakpoint guidelines from the CLSI E. coli M100 Tables 2A through 2J for (reference 31) were used to classify strains into "resistant", "intermediate" or "susceptible"; designations of "intermediate" were lumped with "resistant" for ' purposes of statistics and inference, a conservative approach with respect to consumer safety.
' purposes of statistics and inference, a conservative approach with respect to consumer safety. Thus, our use of a binary variable for resistance is consistent with the most recent standards for evaluating resistance.
Nevertheless, we agree with Dr. Stärk that there is merit in considering resistance as a continuous function, and for this reason we provided the raw (continuous) data used to evaluate resistance (see the data depositied on Figshare). Thus, if new standards develop, or if interested readers wish, the raw continuous data are available for analysis.
We agree with Dr. Stärk that linking the resistance status of the food product to exposure of the animal to antibiotics would provide a stronger basis for causal inference about the source of resistance and identify ways that resistance could be mitigated. However, it would not affect the overall conclusion that production methods yield different frequencies of resistance in , a E. coli conclusion that our data clearly support. Obtaining data about on-farm use of antibiotics in the US is nearly impossible at the current time. We hope that analyses like ours help pave the way to greater openness in the industry with respect to the on-farm use of these powerful drugs, and therefore stronger inference about their impacts on food and people.
no competing interests The study is of interest as this type of data is necessary to fully understand the use of antibiotics in animal production. The manuscript is well written and easy to understand. There are a few points that the authors need to address:

Methods:
Please list the number of brands per category that were sampled as in Table 1. Also, please list the number of samples collected per brand. This information would be helpful and may have some impact on the data.

Results:
The total number of isolates collected per type or brand is not stated. It would be helpful to E. coli know what percentage of each were positive.

Discussion:
The first sentence of the discussion section is very inflammatory to the industry and is not absolutely true. I would suggest refining this greatly or deleting it.
Similarly, the second sentence is quite definitive and implies that antibiotic usage always creates Similarly, the second sentence is quite definitive and implies that antibiotic usage always creates antibiotic resistance which may not be true. I suggest modifying this sentence with a qualifying word such as "may select for" or "can select for".
One variable the authors did not address is the fact that chickens produced by the same brand most likely came from different farms. Because there seems to be a large number of brands sampled, this further adds to the total number of farms that were likely to be sampled. The farm environment does have some impact on the quality of the food. Further, birds from multiple farms may be processed within the same processing plant and this too can impact the microbiological quality of the carcass due to cross-contamination. These are confounding variables that may have impacted the author's data.
I have read this submission. I believe that I have an appropriate level of expertise to confirm that it is of an acceptable scientific standard, however I have significant reservations, as outlined above.
No competing interests were disclosed. We appreciate Dr. Hanning's comments and suggestions and have revised the manuscript to address each point raised.
The number of samples collected per brand is now indicated in the text. The number of brands per category is listed as 'N' in Table 1.
We now state, by type of chicken collected, the percentages of isolates positive for some degree of antibiotic resistance in the fifth sentence of the Results section.
We have revised a few sentences in the Discussion for clarity and to remove potentially inflammatory or unsupported claims. We have added several citations that provide support for our statements, and at the same time, we removed potentially inflammatory adjectives. We also add a statement about one specific estimate of the extent of antibiotic usage for growth promotion. We realize such estimates are controversial, but we feel that these provide important and useful context for readers in the field. For the second sentence in the original, we have followed Dr. Hanning's suggestion to modify the claim by changing ' ' to ' '. select can select We appreciate the point that farm-to-farm variability could play a role in our results, and have added a sentence acknowledging this explicitly. Also, the third paragraph of the Discussion discusses the potential for cross-contamination within shared production facilities to influence our results. We note that our design was developed to test for significant effects of type of chicken from the perspective of the consumer making decisions about which chicken to purchase. Thus, while farm-to-farm variability and cross-contamination are important potential sources of variation in antibiotic resistance, incorporating this variance is an important part of our design. Future, more