Title: Non-stochastic sampling error in quantal analyses for Campylobacter species on poultry products Authors
Submitted to: Analytical and Bioanalytical Chemistry
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: December 12, 2012
Publication Date: February 5, 2013
Repository URL: http://handle.nal.usda.gov/10113/55967
Citation: Irwin, P.L., Reed, S.A., Brewster, J.D., Nguyen, L.T., He, Y. 2013. Non-stochastic sampling error in quantal analyses for Campylobacter species on poultry products. Analytical and Bioanalytical Chemistry. 405(7):2353-2369. Interpretive Summary: The contamination of foods with pathogenic bacteria (e.g., Campylobacter jejuni or C. coli) can lead to food poisoning epidemics. In order to determine the concentration of Campylobacter bacteria in foods it is necessary to be able to quantify as few as a couple of bacteria per sampled volume or mass because the level of these organisms in foods is presumed to be very low. For this purpose we developed a most probable number (MPN) assay using a DNA identification technique, the polymerase chain reaction, to determine if the target organism grew or not (+ or – result). Using this method we found that the cell concentration step was introducing unnecessary error but such error was lessened as the concentration of bacteria dropped below 300 cells per 500g. Thus, such errors should be of little consequence in real samples which have naturally low levels of contamination. Using our method on real samples we were able to quantify 14 to 1226 MPN per 500g of naturally contaminated chicken (skinless chicken pieces) and 18 to 244 MPN per 500g chicken wings, breasts, legs, and thighs (skin on) whereupon about 50% of the 29 total samples tested negative. Four of these samples did have substantially lower Campylobacter levels: 1.24 to 6.11 MPN per 500g. This rsearch provides tools for more accurate quantification of pathogenic bacteria in food.
Technical Abstract: Using primers and fluorescent probes specific for the most common foodborne Campylobacter species (C. jejuni = Cj and C. coli = Cc), we developed a multiplex, most probable number (MPN) assay using quantitative PCR (qPCR) as the determinant for binomial detection: number of p positives out of n = 6 observations each of 4 mL (V) per dilution. Working with thrice frozen-thawed (to minimize native bacteria) chicken washes spiked with known levels of both Cj and Cc, we found that about 20% of the experiments had a significant amount of error in the form of either greater than 25% MPN calculation error ('e) and/or a low apparent recovery rate (R less than 1 = MPN observed ÷ CFU spiked). Assuming such errors were exacerbated by an excessively small n, we examined computer-generated MPN enumeration dats from the standpoint of stochastic sampling error (') and found that such binomial-based assays behave identically to Poisson-based methods (e.g., counting data) except that fewer technical replicates (n) appear to be required for the same number of cells per test volume (µ). This result implies that the qPCR detection-based MPN protocol discussed herein should accurately enumerate atest population with a µ = 1 using n = 6 observation per dilution. For oour protocal this squates to = 8 cells per 400-500g of sampled product. Based on this analysis, the error rate we saw in spiked experiments (where µ >> 1) implied a non-stochastic source. In other experiments we presentevidence that this source was, at least in part, related to the cell concentration step. We also demonstrate that such errors were lessened( from - 20% to 10%) at lower Campylobacter levels (µ = 40) as this would more likely exist in nature. USing this protocol we were able to quantify 14 to 1226 MPN per 450g of naturally contaminated chicken for skinless pieces and 18 to 244 MPN per 450g for chicken wings, breast, leg and thighs (skin on) whereupon about 50% of the 29 samples tested negative. FOur of these chicken wash samples did have substantially lower Campylobacter levels (1.24 to 6.11 MPN per 450g) which might be better enumerated using n = 12 or higher. However, we demonstrated that the limit of this protocol dimishes greatly for n > 6 because one is ever more dilutin the sample, or lessening V, to achieve the requisite n.