Determining bad actors: A linear mixed effects model approach to elucidate behavioral toxicity of metal mixtures in drinking water

Abstract

Mixtures of chemical contaminants can pose a significant health risk to humans and wildlife, even at levels considered safe for each individual chemical. There is a critical need to develop statistical methods to evaluate the drivers of toxic effects in chemical mixtures (i.e., bad actors) from exposure studies. Here, we develop a hierarchical modeling framework to disentangle the toxicity of complex metal mixtures from a screening study of 92 drinking well water samples containing multiple metal elements from Maine and New Hampshire, USA. In order to screen for neurodevelopmental impacts from exposure to these drinking water samples, we use a larval zebrafish (Danio rerio) behavioral assay. Zebrafish are an advantageous toxicological model organism due to combining the complexity of a vertebrate organism and higher-throughput exposure methods. We formulate a linear mixed modeling approach that captures intrinsic complexity in a common larval behavioral assay in order to improve its sensitivity and rigor and identify drivers of behavioral toxicity from the metal mixtures within the drinking water samples. Our analysis identifies lead (Pb), cadmium (Cd), nickel (Ni), copper (Cu), barium (Ba), and uranium (U) as metals that consistently impact larval locomotor activity, individually and across nine pairs of those metals. Our model also elucidates three distinct clusters of metal mixture components that drive behavioral effects: (Ba:Cu:U), (Ni:Pb:U), (Ba:Pb:U). Having identified a set of “bad-actor” metals from the water samples, we conduct exposure experiments for each individual metal (Pb, Cd, Ni, Cu, and Ba) at levels considered safe by the US Environmental Protection Agency drinking water regulatory limits and validate Pb, Ni, Cu, and Ba as behavioral toxicants at these concentrations. Collectively, our modeling approach estimates the impact of metal elements on a complex behavioral outcome in a statistically robust manner and establishes an approach to capture “bad actors” and key chemical interactions in a complex mixture.