05-03-0096r.qxp
11/14/2005
9:45 AM
Page 2023
Effect of intensity of fecal pat sampling on estimates of Escherichia coli O157 prevalence Alejandro Echeverry, MS; Guy H. Loneragan, BVSc, PhD; Bruce A. Wagner, PhD; Mindy M. Brashears, PhD Objective—To evaluate site-to-site variation within fecal pats from cattle with regard to detection of Escherichia coli O157 and determine the effect on the accuracy of prevalence estimates of assay of multiple samples collected from the same fecal pat. Sample Population—120 freshly voided fecal pats collected from 2 beef feedlots. Procedures—5 samples were systematically collected from each fecal pat and analyzed for E coli O157 via selective preenrichment techniques, immunomagnetic separation, and biochemical tests. Presumptive isolates were definitively identified via agglutination assays and polymerase chain reaction techniques. Best estimators of prevalence were calculated from the distribution of E coli O157–positive samples per pat. Results—Of the 120 fecal pats, 96, 13, 4, 2, 3, and 2 fecal pats had 0, 1, 2, 3, 4, and 5 E coli O157–positive samples, respectively. The greatest estimate of E coli O157 prevalence (20%) was achieved when all 5 samples were assessed; this estimate represented a 2.4fold increase in prevalence, compared with that provided via analysis of 1 sample/pat (8.2%). Compared with assessment of 5 sites/pat, the relative sensitivity of detecting an E coli O157–positive fecal pat via analysis of 1 site/pat was 40.1%. Conclusions and Clinical Relevance—Results suggest that estimates of E coli O157 prevalence derived from sampling of 1 location/pat are likely underestimates of the true prevalence of this pathogen in fecal pats (and by extension, cattle). Additional research is warranted to confirm these results in situations of high and low prevalence and across different feedlots. (Am J Vet Res 2005;66:2023–2027)
E
scherichia coli O157 has emerged as an important food-borne pathogen. Exposure of humans to this bacterium may lead to bloody diarrhea, hemorrhagic colitis, hemolytic uremic syndrome, and death.1,2 Although a wide variety of foods have been implicated in E coli O157–induced disease,3-6 cattle have been Received March 14, 2005. Accepted May 13, 2005. From the Department of Animal and Food Sciences, Texas Tech University, Lubbock, TX 79409 (Echeverry, Brashears); the Feedlot Research Group, Division of Agriculture, College of Agriculture, Nursing and Natural Sciences, West Texas A&M University, Canyon, TX 79016 (Loneragan); and the USDA:Animal Plant Health Inspection Service:Veterinary Services:Centers for Epidemiology and Animal Health, Mail Stop 2E7, 2150 Centre Ave, Fort Collins, CO 80526 (Wagner). Presented orally at the 84th Annual Meeting of the Conference of Research Workers in Animal Diseases, Chicago, November 2003 and at the 91st Annual Meeting of the International Association for Food Protection, Phoenix, August 2004. Address correspondence to Dr. Loneragan. AJVR, Vol 66, No. 12, December 2005
identified as the major animal reservoir of the pathogen.7,8 As such, a better understanding of the ecology and epidemiology of the microorganism in cattle and their environment is needed to appropriately assess and quantify the microbial risk posed by and potential interventions for this pathogen.9-11 Ostensibly, it appears that the prevalence of E coli O157 in cattle has increased over time. In part, this is likely a result of refinement of sampling procedures, changes in sample quantity and type used in analyses, and the use of more sensitive microbiological methods.12-14 Ultimately, studies7,8,15,16 in which sensitive methods have been used have revealed the ubiquitous nature of this organism among cattle populations. Rice et al2 determined that, compared with analysis of fecal material, the sensitivity of detection of this pathogen in artificially inoculated cattle was even greater via collection and analysis of rectoanal mucosal swabs. This greater sensitivity may have been detected because lymphoid tissue just proximal to the rectoanal junction has been proposed as the site of colonization of E coli O157 in cattle.17 As a consequence of this limited and distal site of colonization within the intestinal tract, E coli O157 may not be randomly distributed in fecal material.17 Moreover, Pearce et al18 found evidence of variable distribution of E coli O157 in voided feces of animals kept in pastures. Therefore, it seems possible that misclassification of the shedding status of cattle may occur at times because samples for microbiological purposes are collected from a relatively small proportion of fecal pats (in those studies involving fecal pat sampling). As a consequence, there would be an underestimate of E coli O157 prevalence and a bias toward null in controlled studies that are performed to evaluate strategies to reduce carriage of E coli O157. The objective of the study reported here was to evaluate site-to-site variation within fecal pats from cattle with regard to detection of E coli O157 and determine the effect on the accuracy of prevalence estimates of collection and assay of multiple samples from the same fecal pat. Materials and Methods Sample collection—Freshly voided pen-floor fecal samples were obtained from cattle maintained at 2 feedlots in the Texas Panhandle. Each feedlot was visited on 2 occasions (during the months of July and August), and during each visit, 30 freshly voided fecal pats were chosen for sample collection. Five samples (approx 100 g each) were systematically obtained from each pat while maintaining across-pat consistency in sample collection. Briefly, the 5 samples from individual fecal pats were collected in north to south lines starting on the west side of the pat and progressing east (sites 1 through 5, respectively; Figure 1). Individual samples were placed in sterile prelabeled specimen cups. Separate single2023
05-03-0096r.qxp
11/14/2005
9:45 AM
Page 2024
use plastic spoons were used for each sampling site. Samples were stored on wet ice and transported to the laboratory on the day of sample collection. Microbiological processes began within 2 hours of sample collection. Microbiological methods—Immediately after arrival at the laboratory, 10 g of feces from each sample was transferred for selective enrichment of gram-negative organisms into 90 mL of a broth containing vancomycin (8 mg/mL), cefixime (50 ng/mL), and cefsulodin (10 mg/mL) and incubated at 37oC for 6 hours as described.19 After enrichment, 20 µL of paramagnetic beads coated with anti–E coli O157 antibodya was added to 1-mL aliquots of each sample and incubated for 30 minutes at 22oC. Beads were held against the side of the vial by a magnet and washed 3 times; 50 mL of the bead-bacteria mixture was spread onto sorbitol MacConkey agar plates supplemented with cefixime (0.05 µL/L) and potassium tellurite (50 µL/L) and incubated at 37oC for 18 to 24 hours. As many as 3 typical non–sorbitol-fermenting colonies/plate were streaked for isolation on sorbitol MacConkey agar plates and incubated for approximately 18 to 24 hours at 37oC. Non–sorbitol-fermenting colonies (clear or whitish) were selected and inoculated on MacConkey agar and an E coli O157:H7–selective agar containing 4-methylβ-glucuronide (MUG). After incubation, MUGumbelliferyl-β negative colonies were transferred from the plates into tryptic soy broth, triple sugar iron agar, and MacConkey broth and incubated overnight in aerobic conditions at 37oC. Presumptive E coli O157 colonies (those that were MUG negative, sorbitol negative, lactose positive, indole positive, and triple sugar iron positive) were recovered from the backed tryptic soy broth and confirmed by use of a commercial latex agglutination test.b Serotype confirmation (including genes encoding production of shiga toxins) was made by use of a commercially available polymerase chain reaction system.c Statistical analysis—Data were analyzed by use of a commercially available software package.d Descriptive statistics were generated and reported in graphic or tabular format. Prevalence among sampling sites 1 through 5 was evaluated by use of logistic regression techniques. The outcome vari-
Figure 1—Photograph to illustrate the sites on fecal pats from which samples were collected. The arrow indicates north; numbers identify sample sites (sites 1 and 5 were the west and east margins of the pat). For each of 120 pats, samples were collected in order (1 though 5) by use of the same sampling protocol. 2024
able was a binomial response variable for each sample site. In addition, a contrast was constructed to evaluate the peripheral samples (sites 1 and 5) versus the other samples (sites 2, 3, and 4). It was also of interest to determine whether estimates of prevalence differed when 1 sample/pat had been randomly collected, 2 samples/pat had been randomly collected, and so forth through to collection of 5 samples/pat. The probabilities of fecal pats yielding positive test results were generated from a hypergeometric distribution on the basis of observed data and the various sampling scenarios (1, 2, 3, 4, or 5 random samples/pat). In other words, the probability of 1 or more sample sites being positive for E coli O157 within a pat (fecal pat prevalence) was calculated as 1 minus the probability of 0 sites being positive for E coli O157, given the distribution of positive sample sites and the number of samples per pat collected. The probabilities were used to calculate the expected number of E coli O157–positive fecal samples and, ultimately, expected prevalence estimates. Standard error values were calculated for point estimates by use of binomial distribution. On the basis of prevalence estimates, expected number of E coli O157–positive fecal pats per visit was calculated for each sampling scenario. The expected number of E coli O157–positive samples was modeled as a binomial response variable (the denominator being the number of samples collected/visit), and independent variables evaluated were linear and quadratic terms representing number of samples per pat; feedlot and visit nested with feedlot were modeled as a random variable by use of a randomintercepts, random-slopes approach. The fixed component of the model (effectively averaged over the random effects of feedlot visit) may be written as follows: y = exp(µ + β1xi + β2xi2)/(1 + exp[µ + β1xi + β2xi2]) where y is the expected prevalence, exp is the exponential, : is the intercept, x is the number of fecal pats sampled (i = 1 through 5), β1 is the slope for the linear effect of x, and β2 is the slope for the quadratic effect of x (as indicated by xi2). The data were modeled by use of a freely available macro program in which an events-trials binomial variable served as the dependent variable.e Unexplained variation was partitioned to the various levels of hierarchy (feedlot and visit within feedlot). Sensitivity of detecting E coli O157 in a fecal pat was calculated for sampling scenarios of 1, 2, 3, or 4 sites/pat, relative to sampling all 5 sites/pat. For all models, a value of P ≤ 0.05 was considered significant.
Results Six hundred samples were collected from 120 fecal pats (120 samples from each of the 5 sites within the pats). Overall, E coli O157 was isolated from 49 (8.2%) samples. Escherichia coli O157 was isolated from at least 1 sample from 24 of the 120 (20%) pats. The bacterium was detected in at least 1 sample from 5 (8.3%) and 19 (31.7%) fecal pats collected from feedlots A and B, respectively (Table 1). There was no evidence that prevalence varied with sample site within fecal pats (P = 0.60); E coli O157 was isolated from 14 (11.7 ± 2.9%), 9 (7.5 ± 2.4%), 8 (6.7 ± 2.3%), 8 (6.7 ± 2.3%), and 10 (8.3 ± 2.5%) samples collected at sites 1 through 5, respectively. Similarly, the combined estimate of prevalence obtained from sample sites 1 and 5 did not vary significantly (P = 0.20) from the estimate obtained from sample sites 2, 3, and 4. The distribution of the number of E coli O157–positive samples per fecal pat for each feedlot by AJVR, Vol 66, No. 12, December 2005
05-03-0096r.qxp
11/14/2005
9:45 AM
Page 2025
Table 1—Distribution of number of Escherichia coli O157–positive samples per fecal pat (5 sites/pat) collected during each of 2 visits to 2 beef feedlots. Feedlot A No. of E coli O157–positive samples/pat 0 1 2 3 4 5 Total pats sampled
Feedlot B
Visit 1
Visit 2
Visit 3
Visit 4
Overall
27 2 1 0 0 0 30
28 2 0 0 0 0 30
17 5 3 2 3 0 30
24 4 0 0 0 2 30
96 13 4 2 3 2 120
Table 2—Estimates of E coli O157 prevalence (%) and 95% confidence intervals calculated from data collected from 1, 2, 3, 4, or 5 randomly sampled sites per fecal pat for each of 2 visits to 2 beef feedlots. Feedlot A No. of sites assessed/pat 1 2 3 4 5
Visit 1 2.7 (0.0–8.4) 5.0 (0.0–12.8) 7.0 (0.0–16.1) 8.7 (0.0–18.7) 10.0 (0.0–20.7)
Feedlot B Visit 2
1.3 (0.0–5.4) 2.7 (0.0–8.4) 4.0 (0.0–11.0) 5.3 (0.0–13.4) 6.7 (0.0–15.6)
Visit 3 19.3 (5.2–33.5) 29.7 (13.3–46.0) 35.7 (18.5–52.8) 40.0 (22.5–57.5) 43.3 (25.6–61.1)
Visit 4
Overall
9.3 (0.0–19.7) 12.0 (0.4–23.6) 14.7 (2.0–27.3) 17.3 (3.8–30.9) 20.0 (5.7–34.3)
8.2 (3.3–13.1) 12.3 (6.4–18.2) 15.3 (8.9–21.8) 17.3 (11.0–24.7) 20.0 (12.8–27.2)
Values in parentheses represent the 95% confidence interval.
visit was determined (Table 1). Escherichia coli O157 was not detected in any of the samples collected from 96 of 120 (80%) fecal pats. Overall, E coli O157 was detected in 13 (10.8%), 4 (3.3%), 2 (1.7%), 3 (2.5%), and 2 (1.7%) of the 120 fecal pats in 1, 2, 3, 4, or 5 samples, respectively. Visit-level and overall estimators of prevalence given various sampling scenarios were assessed (Table 2). If 1 sample/pat was randomly selected, the overall best estimate of prevalence was 8.2% (95% confidence interval [CI], 3.3 to 13.1). However, if 2 samples were collected at random per pat, the overall estimate of prevalence increased to 12.3% (95% CI, 6.4 to 18.2). This represented an absolute increase in the estimate of prevalence of 4.1% and a relative increase of 50%. The relative sensitivity of detecting E coli O157 when 1 sample was randomly selected, compared with detection of the pathogen by use of 2 samples, was 66.7% (95% CI, 58.6 to 75.4). When 5 samples/pat were included in the sampling scenario, the estimate of prevalence was 20% (95% CI, 12.8 to 27.2), which is a 2.4-fold increase in prevalence, compared with that determined when just 1 sample/pat was collected. The relative sensitivity of correctly classifying a fecal pat as E coli O157–positive increased with the increasing number of sample sites per pat. Relative to sampling at 5 sites/pat, the fecal pat–level sensitivity of detection was 40.1% (95% CI, 31.3 to 48.9), 61.5% (95% CI, 52.8 to 70.2), 76.5% (95% CI, 68.9 to 84.1), and 86.5% (95% CI, 80.4 to 92.6) for sampling scenarios that included 1, 2, 3, or 4 sites/pat, respectively. The relative sensitivities of including 1 sample/pat, compared with 5 samples/pat, for each of the visits were as follows: 27% (feedlot A, visit 1), 19.4% (feedlot A, visit 2), 44.6% (feedlot B, visit 1), and 46.5% (feedlot B, visit 2). AJVR, Vol 66, No. 12, December 2005
The estimate of prevalence was quadratically associated (P = 0.05) with the number of samples per pat included in the analysis. The fixed component of the model, which is effectively the mean of the randomintercepts, random-slopes model, was as follows: y = exp(–3.24 + 0.57xi – 0.05xi2)/ (1 + exp[–3.24 + 0.57xi – 0.05xi2]) Feedlot accounted for 69.4% of random variation in the model, whereas visit within feedlot accounted for 24.4% of random variation. In a separate model in which the mean prevalence for each visit was forced into the model, feedlot and visit within feedlot accounted for 0.0% and 45.0% of random variation, respectively. Discussion On the basis of results of the present study, it seems probable that animal- or pat-level estimates of E coli O157 prevalence derived via simple fecal pat sampling substantially underestimate the true prevalence. Overall, E coli O157 was isolated from 24 of the 120 (20%) fecal pats. If 1 sample/pat had been randomly collected, the expected prevalence would have been considerably less (ie, 8.2%). With each additional sample collected per pat, the expected prevalence estimate increased. However, the absolute and relative increases in estimates diminished with increasing sample number. This is reflected in the negative quadratic parameter estimate (ie, –0.05). Ideally, an optimum number of samples per pat that maximized prevalence would have been discovered; however, on the basis of the data described herein, the optimum number of samples was 5.7, which was outside the inferential limits of the data. We chose to collect samples from 5 sites 2025
05-03-0096r.qxp
11/14/2005
9:45 AM
Page 2026
within each pat because the volume of fecal material in pats of feedlot cattle made collection of more than five 100-g samples/pat impractical in many instances. It would, however, have been possible to increase the number of samples at the expense of sample volume. If we had done so, it may have been possible to identify the optimum number of samples per pat that maximizes prevalence and falls within the inferential limits of the data. Feedlot and visit within feedlot accounted for 69.4% and 24.4% of the random variation observed in the model, respectively, which indicated that feedlotlevel factors may explain some of the apparent variation in 1-sample sensitivity (relative to 5 samples) detected between the 2 feedlots. Feedlot-level factors that may have contributed to the results are uncertain, and unfortunately, information beyond feedlot location was not collected. Despite this, when mean visit prevalence was forced into the model, the proportion of variation attributable to feedlot was reduced to 0%. Therefore, it is possible that differences in prevalence, and not feedlot per se, may account for most of the variation in 1-sample sensitivity. If so, feedlot-level factors that affect prevalence would be of interest. In the present study, all visits were conducted during July and August, months in which prevalence of E coli O157 is typically greatest. Factors contributing to the observed variation are uncertain but are clearly important in the epidemiology of E coli O157 in feedlots; without doubt, more research needs to be performed to better understand these factors. Despite the unexplained model variation in our study, estimates of prevalence increased with the number of samples per pat for all 4 visits. This suggests that sampling strategies that include multiple samples per pat should result in improvements in prevalence accuracy in most situations. However, it should be noted that the present study was limited in the number of fecal pats from which samples were collected and only 2 feedlots were included. More research is needed to validate these results across a range of E coli O157 prevalence and across feedlots. In cattle, the likely site for long-term colonization with E coli O157 is in lymphoid tissue just proximal to the rectoanal junction.17 One consequence of this distal colonization site within the intestinal tract is that distribution of E coli O157 in fecal material may not be uniform; the distance between the colonization site and the anus would presumably be insufficient to allow thorough homogenous distribution of the pathogen within the fecal material before defecation. Moreover, Naylor et al17 determined that there was a greater concentration of E coli O157 on the fecal pat surface than within the fecal pat. In addition, variation in the concentration of E coli O157 in feces of animals kept on pasture has been identified.18,20 If indeed there is nonuniform distribution of E coli O157 within fecal material of feedlot cattle, it may have accounted for the improved pat-level (and ultimately animal-level) estimates of prevalence when the number of samples per pat assessed was increased. Although we propose that the data of the present study support the findings of other investigators that there is a nonuniform distribu2026
tion of E coli O157 within fecal material, the distribution of E coli O157 was not directly evaluated in our study and can only be inferred. Unfortunately, the semiliquid consistency of feces from cattle fed highconcentrate diets (which are typically used in modern commercial feedlots) made comparison of surface and internal pat material impractical. We did not detect significant variation in E coli O157 prevalence estimates between sampling sites within pats nor did we find evidence that the margins of pats had greater prevalence than other samples. More research is needed to directly evaluate variation in concentration of E coli O157 within feces of feedlot cattle. Sampling 5 sites within a pat offered substantial improvements in prevalence accuracy, compared with estimates based on data from only 1 site. However, increased sampling intensity also substantially increased costs associated with supplies and isolation of the organism. In future studies, the number of samples per pat to be collected should depend on the desire for increased accuracy, number of pats from which samples are to be collected, and financial limitations of the study. a. b. c. d. e.
Dynabeads anti–E coli O157, Dynal Biotech Inc, Brown Deer, Wis. Escherichia coli O157:H7 latex test, REMEL Inc, Lenexa, Kan. BAX detection system, DuPont Qualicon, Wilmington, Del. SAS for Windows, version 8.2, SAS Institute Inc, Cary, NC. GLIMMIX macro for the SAS System, release 8. Available at: www.ftp.sas.com/techsup/download/stat/glmm800.html. Accessed Mar 16, 2003.
References 1. Banatvala N, Griffin PM, Greene KD, et al. The United States National Prospective Hemolytic Uremic Syndrome Study: microbiologic, serologic, clinical, and epidemiologic findings. J Infect Dis 2001;183:1063–1070. 2. Rice DH, Sheng HQ, Wynia SA, et al. Rectoanal mucosal swab culture is more sensitive than fecal culture and distinguishes Escherichia coli O157:H7-colonized cattle and those transiently shedding the same organism. J Clin Microbiol 2003;41:4924–4929. 3. Tuttle J, Gomez T, Doyle MP, et al. Lessons from a large outbreak of Escherichia coli O157:H7 infections: insights into the infectious dose and method of widespread contamination of hamburger patties. Epidemiol Infect 1999;122:185–192. 4. Hilborn ED, Mshar PA, Fiorentino TR, et al. An outbreak of Escherichia coli O157:H7 infections and haemolytic uraemic syndrome associated with consumption of unpasteurized apple cider. Epidemiol Infect 2000;124:31–36. 5. Watanabe Y, Ozasa K, Mermin JH, et al. Factory outbreak of Escherichia coli O157:H7 infection in Japan. Emerg Infect Dis 1999;5:424–428. 6. Sivapalasingam S, Friedman CR, Cohen L, et al. Fresh produce: a growing cause of outbreaks of foodborne illness in the United States, 1973 through 1997. J Food Prot 2004;67:2342–2353. 7. Smith D, Blackford M, Younts S, et al. Ecological relationships between the prevalence of cattle shedding Escherichia coli O157:H7 and characteristics of the cattle or conditions of the feedlot pen. J Food Prot 2001;64:1899–1903. 8. Keen JE, Elder RO. Isolation of shiga-toxigenic Escherichia coli O157 from hide surfaces and the oral cavity of finished beef feedlot cattle. J Am Vet Med Assoc 2002;220:756–763. 9. Elder RO, Keen JE, Siragusa GR, et al. Correlation of enterohemorrhagic Escherichia coli O157 prevalence in feces, hides, and carcasses of beef cattle during processing. Proc Natl Acad Sci U S A 2000;97:2999–3003. 10. Sargeant JM, Sanderson MW, Griffin DD, et al. Factors associated with the presence of Escherichia coli O157 in feedlot-cattle water and feed in the Midwestern USA. Prev Vet Med 2004;66:207–237. AJVR, Vol 66, No. 12, December 2005
05-03-0096r.qxp
11/14/2005
9:45 AM
Page 2027
11. Sargeant JM, Sanderson MW, Smith RA, et al. Associations between management, climate, and Escherichia coli O157 in the faeces of feedlot cattle in the Midwestern USA. Prev Vet Med 2004;66:175–206. 12. Chapman PA, Wright DJ, Siddons CA. A comparison of immunomagnetic separation and direct culture for the isolation of verocytotoxin-producing Escherichia coli O157 from bovine faeces. J Med Microbiol 1994;40:424–427. 13. Sanderson MW, Besser TE, Gay JM, et al. Fecal Escherichia coli O157:H7 shedding patterns of orally inoculated calves. Vet Microbiol 1999;69:199–205. 14. Barkocy-Gallagher GA, Arthur TM, Rivera-Betancourt M, et al. Seasonal prevalence of Shiga toxin-producing Escherichia coli, including O157:H7 and non-O157 serotypes, and Salmonella in commercial beef processing plants. J Food Prot 2003;66:1978–1986. 15. Khaitsa ML, Smith DR, Stoner JA, et al. Incidence, duration, and prevalence of Escherichia coli O157:H7 fecal shedding by feedlot cattle during the finishing period. J Food Prot 2003;66:1972–1977.
AJVR, Vol 66, No. 12, December 2005
16. Sargeant JM, Sanderson MW, Smith RA, et al. Escherichia coli O157 in feedlot cattle feces and water in four major feeder-cattle states in the USA. Prev Vet Med 2003;61:127–135. 17. Naylor SW, Low JC, Besser TE, et al. Lymphoid follicledense mucosa at the terminal rectum is the principal site of colonization of enterohemorrhagic Escherichia coli O157:H7 in the bovine host. Infect Immun 2003;71:1505–1512. 18. Pearce MC, Fenlon D, Low JC, et al. Distribution of Escherichia coli O157 in bovine fecal pats and its impact on estimates of the prevalence of fecal shedding. Appl Environ Microbiol 2004;70:5737–5743. 19. Brashears MM, Galyean ML, Loneragan GH, et al. Prevalence of Escherichia coli O157:H7 and performance by beef feedlot cattle given Lactobacillus direct-fed microbials. J Food Prot 2003;66:748–754. 20. Robinson SE, Brown PE, John Wright E, et al. Heterogeneous distributions of Escherichia coli O157 within naturally infected bovine faecal pats. FEMS Microbiol Lett 2005; 244:291–296.
2027
