Escherichia coli's water load affects zebrafish (Danio rerio) behavior.

Traditional physico-chemical sensors are becoming an obsolete tool for environmental quality assessment. Biomonitoring techniques, such as biological early warning systems present the advantage of being sensitivity, fast, non-invasive and ecologically relevant. In this work, we applied a video tracking system, developed with zebrafish (Danio rerio), to detect microbiological contamination in water. Using the fishs' behavior response, the system was able to detect the presence of a non-pathogenic environmental strain of Escherichia coli, at three different levels of contamination: 600, 1800 and 5000 CFU/100 mL (colony forming units/100 mL). Data was collected during 50 min of exposure and analyzed with the artificial neural networks Self-organizing Map and Multi-layer Perceptron. The behavior of exposed fish was more erratic, with pronounced and rapid changes on movement direction and with significant less exploratory activity. The accuracy, sensitivity and specificity values regarding the detection capability (distinction between presence or absence of contamination) ranged from 89 to 100%. Regarding the classification capability (distinction between experimental conditions), the values ranged from 67 to 89%. This research may be a valuable contribution to improve water monitoring and management strategies, by taking as reference the effects on biosensors, without a biased anthropocentric perspective.