A second-generation device for automated training and quantitative behavior analyses of molecularly-tractable model organisms

Current efforts in the discovery of novel genes, proteins, or chemical reagents that have interesting, useful, and/or enlightening effects upon living systems are focused on screening approaches: locating valuable reagents by large-scale, parallel, automated examination of candidate molecules present in a combinatorial "library" [157], [158], [159], [160], [161], [162], [163]. A number of academic and commercial pharmaceutical projects have generated genetic, proteomic, or small-molecule (drug) libraries that must be screened to identify compounds of interest to both biomedicine and basic biology. Examples include searches for drugs that inhibit particular enzyme pathways in human disease [164] or proteins involved in specific patterning events in developing embryos [165], [166], [167], [168], [169], [170], [171], [172].

There is an enormous list of potential targets for which screening of libraries would result in medically-valuable reagents; similarly, many biological processes, when perturbed and characterized, can readily be used as assays that might lead to a better understanding of endogenous control mechanisms. The crucial and usually most difficult aspect is the choice of screening method. This requires a tractable yet relevant model system and a degree of automation (to ensure temporal and financial feasibility). Some screens—such as those for tumor suppressant drugs—have been successfully conducted using cell culture assays or unicellular organisms such as bacteria or yeast [173]. However, many targets of interest are relevant only in the context of complex organisms (such as the search for memory-enhancing drugs or gene products which participate in specific behaviors, for example). Large-scale behavioral screens in otherwise popular model systems such as mice are not feasible due to cost constraints. Model organisms such as Xenopus and zebrafish are ideal because they offer complex vertebrate systems with high biomedical relevance as well as being readily amenable to state-of-the-art molecular, cell, and neuro-biology techniques. Screens on multicellular models such as zebrafish [174], [175], [176], [177], [178], [179], [180] have been successful in cell-biological assays, but the lack of a parallelized, automated system for analyzing memory and individual responses to stimuli precludes effectiveness in high-throughput neurobehavioral screens. The worm C. elegans has been utilized for high-throughput automated screens, but this has been in a cell-biological context (measuring death vs. survival), rather than in a behavioral one [181].

One nice counterexample, which illustrates the potential of this approach but is not available in a form generally applicable to other kinds of experiments, was used in a high-throughput fish screen for the perception of auditory stimuli [182]. It is the lack of a scalable, powerful, well-characterized system for parallel investigation of memory and learning in vertebrate organisms that has hampered the commercial screening for new classes of cognitive enhancers or “nootropics” [183].

Automated systems will enable cognitive scientists to ask questions that are currently difficult to address, and will significantly lower the barrier for developmental neuroscience labs embarking on projects that require them to quantitatively characterize the neurobehavioral consequences of alterations in embryonic development pathways. Likewise, pharmaceutical efforts using small animal models to identify compounds or genes with specific neurobiological effects [184], [185], [186], [187] would be able to use scaled-up versions of such automated platforms to perform high-throughput screens for complex behavioral or neurological outcomes. Optical, microfabrication, and computer technology are progressing with increased rapidity and will further potentiate the capabilities of such machine vision and environmental control systems. The advances in basic science and neuromedicine that will result from the widespread availability and increased sophistication of automated behavioral screening are truly exciting to contemplate.

This paper is dedicated to Inna M. Sheiman and colleagues for their pioneering work on memory and its dynamics during metamorphosis. We thank Emily Yuan and Caitlin Hicks for their participation in our efforts to develop an automated behavior analysis platform. We thank Richard Scarlet for his help in all phases of the project and especially the creation of the Excel data analysis worksheet suite, and Mark Bramhall for his programming of the front-end code. We thank Punita Koustubhan and Amber Currier for Xenopus husbandry, Junji Morokuma and Wendy Beane for planaria colony care, and all of the members of the Levin lab for many useful discussions.