Phenotypic library-based microbial source tracking methods: Efficacy in the California collaborative study Valerie J. Harwood, Bruce Wiggins, Charles Hagedorn, R. D. Ellender, Jan Gooch, James Kern, Mansour Samadpour, Annie C. H. Chapman, Brian J. Robinson and Brian C. Thompson Valerie J. Harwood (corresponding author) Department of Biology, SCA 110, University of South Florida, 4202 E. Fowler Ave, Tampa, FL 33620, USA Bruce Wiggins James Madison University, Harrisonburg, VA 22807, USA Charles Hagedorn Annie C. H. Chapman Virginia Polytechnic and State University, Blacksburg, VA 24061, USA R. D. Ellender Brian J. Robinson University of Southern Mississippi, Hattiesburg, MS 39406, USA Jan Gooch Brian C. Thompson NOAA, Charleston, SC 29412, USA James Kern Maptech Inc., Blacksburg, VA 24060, USA Mansour Samadpour Institute for Environmental Health, 8279 Lake City Way NE, Seattle WA 98115, USA ABSTRACT As part of a larger microbial source tracking (MST) study, several laboratories used library-based, phenotypic subtyping techniques to analyse fecal samples from known sources (human, sewage, cattle, dogs and gulls) and blinded water samples that were contaminated with the fecal sources. The methods used included antibiotic resistance analysis (ARA) of fecal streptococci, enterococci, fecal coliforms and E. coli; multiple antibiotic resistance (MAR) and Kirby-Bauer antibiotic susceptibility testing of E. coli; and carbon source utilization for fecal streptococci and E. coli. Libraries comprising phenotypic patterns of indicator bacteria isolated from known fecal sources were used to predict the sources of isolates from water samples that had been seeded with fecal material from the same sources as those used to create the libraries. The accuracy of fecal source identification in the water samples was assessed both with and without a cut-off termed the minimum detectable percentage (MDP). The libraries (∼300 isolates) were not large enough to avoid the artefact of source-independent grouping, but some important conclusions could still be drawn. Use of a MDP decreased the percentage of false-positive source identifications, and had little effect on the high percentage of true-positives in the most accurate libraries. In general, the methods were more prone to false-positive than to false-negative errors. The most accurate method, with a true-positive rate of 100% and a false-positive rate of 39% when analysed with a MDP, was ARA of fecal streptococci. The internal accuracy of the libraries did not correlate with the accuracy of source prediction in water samples, showing that one should not rely solely on parameters such as the average rate of correct classification of a library to indicate its predictive capabilities. Key words | E.coli, Enterococcus, fecal coliform, fecal pollution, indicator organism, water quality INTRODUCTION Water quality indicator organisms approved for use in the US, such as fecal coliforms, Escherichia coli and Entero- coccus spp., are broadly distributed in the feces of various host animals. Their presence therefore provides no information about the source(s) of fecal contamination to waters, confounding efforts such as risk assessment and total maximum daily load (TMDL) assessment. Over 20 years ago, the concept that resistance to antibiotics could aid in determining sources of fecal indicator organ- isms was germinating for enterococci (Kibbey et al. 1978), fecal coliforms (Bell et al. 1983) and E. coli (Krumperman 1983). The proposed mechanism for discrimination was that the exposure of various hosts (i.e. cattle, humans, wild animals) to certain antibiotics varies, and that the selective pressure of antibiotics on microbial populations of the gastrointestinal tract would result in measurable 153 © IWA Publishing 2003 Journal of Water and Health | 01.4 | 2003