Linking spatially distributed biogeochemical data with a two-host life-cycle pathogen:A model of whirling disease dynamics in salmonid fishes in the Intermountain West
Abstract: Fish diseases are often caused by waterborne parasites, making them ideal systems for modeling the non-linear relationships between biogeochemical features and disease dynamics. Myxobolus cerebralis, the causative agent of whirling disease, has been a major contributor to the loss of wild rainbow trout populations in numerous streams within the Intermountain West (Colorado, Idaho, Montana, Utah, Wyoming). The parasite alternates between an invertebrate and vertebrate host, being transmitted between the sediment feeding worm T.Tubifex and salmonid fishes. A greater understanding of the linkage between biological stream integrity, geomorphic features, water quality parameters and whirling disease risk is needed to improve current management techniques. Biodiversity and abundance of the worm communities are influenced by biogeochemical features and linked to disease severity in fish. We collected and identified ~700 worms from eight sites using molecular genetic probes and a taxonomic key. Additionally, ~1700 worms were identified using only a taxonomic key. Our work examines the links between worm community structure and biogeochemical features. We use a modified Self-Organizing-Map (SOM), which is a non-parametric clustering method based on an artificial neural network (ANN). Clustering methods are particularly attractive for exploratory data analyses because they do not require either the target number of groupings or the data structure be specified at the outset. ANN clustering methods have been shown to be more robust and to account for more data variability than traditional methods when applied to clustering geo-hydrochemical and microbiological datasets. The SOM highlights spatial variation of worm community structure between sites; and is used in tandem with expert knowledge (Lamb and Kerans) of local worm communities and a Madison River, MT physiochemical dataset (GIS-derived layers, water quality parameters). We iteratively clustered the physiochemical data and then compared the resulting groups to site-specific worm community structures. The SOM mined patterns from this highly dimensional data and produced 2-D visualizations of the data clusters. This process, in concert with iterative feedback with stream ecologists, led to the adaptation of new nonlinear relations and suggests new subsets of input parameters that guide the next round of SOM simulations, expand the pool of concepts, hone existing hypotheses, generate new hypotheses, and so on. The methodologies developed here helped mine the relationship between dominant biogeochemical features and the distribution of an alternative host of a vertebrate disease. This collaboration between modelers, field ecologists and geneticists will prove useful in guiding future data gathering and modeling efforts. (i.e., identifying missing data gaps and sampling frequency), and will enable more effective, high-volume hypothesis generation that, in turn, will better guide complex experimental designs providing integrated understanding of disease dynamics.
[edit database entry]