Toxicant identification by a luminescent bacterial bioreporter panel: application of pattern classification algorithms.

Genetically engineered microorganisms, tailored to respond by a dose-dependent signal to the presence of toxic chemicals, are a potentially useful tool for environmental monitoring. One manifestation of this approach is based on a panel of luminescent bacterial bioreporters, harboring fusions of the luxCDABE operon to various stress-responsive gene promoters. Such sensors can report by a dose-dependent luminescent signal on the stress sensed by the cells and thus on the presence of toxic compound(s), but they lack the ability to identify the chemicals involved. Here, we demonstrate how the use of a panel of such sensors might offer a solution to this drawback. Five selected Escherichia coli reporter strains harboring fusions of selected gene promoters (grpE, nhoA, oraA, lacZ, and mipA) to luxCDABE were exposed to five model toxicants and to a toxicant-free control in a 40-repetition format. Each of the six treatments activated different promoters to different extents, producing its own unique fingerprint. Two machine learning schemes were challenged with the obtained data set: Bayesian decision theory and the nonparametric nearest-neighbor technique. The Bayesian classifiers performed better and were able to identify the sample's contents within 30 min with an error rate estimate that did not exceed 3% at a 95% confidence level and with zero false negatives. Performance in tap water and wastewater samples was similar. Given the coming of age of whole-cell sensing devices, pattern classification algorithms such as the ones described here offer a step toward the incorporation of reporter cells into future biosensor formats, including whole-cell arrays.