Use of site occupancy models to estimate prevalence of Myxobolus cerebralis infection in trout

Empirical estimates of pathogen prevalence in samples of fish may underestimate true prevalence because available detection techniques are incapable of perfect detection. Trout of several species were collected from enzootic (Myxobolus cerebralis, causative agent in whirling disease) habitats, and individual fish were examined for presence of the parasite two or six times by one of four methods: pepsin-trypsin digest (brown trout Salmo trutta), plankton centrifuge (brown trout), polymerase chain reaction (rainbow trout Oncorhynchus mykiss), or histopathology (brook trout Salvelinus fontinalis). The presence-absence data were modeled for prevalence of infection (psi) and probability of detection (p) of the parasite via occupancy models that accounted for imperfect detection of the organism. Based on estimates from the most-supported model for comparison, two myxospore concentration methods underestimated prevalence by about 12% for whole-head results and 34% for the expected value of half-head analysis. Polymerase chain reaction and histopathology gave virtually the same prevalence estimates for whole-head results as the best models but underestimated prevalence by about 6% and 12%, respectively, for the expected value of half-head analysis. The probability of detecting the parasite in a single survey of a fish head, conditional on the parasite's presence, was 0.66 for myxospore concentration methods, 0.81 for histopathology, and 1.0 (left halves) or 0.89 (right halves) for polymerase chain reaction. The occupancy models used in this study may be extended to large-scale monitoring of M. cerebralis to estimate expansion or contraction of the parasite's range over time.