Intellectual merit: A combined experimental-computational approach will be used to understand the differences between unsupervised and supervised learning using olfactory processing as a model system. It is becoming increasingly clear that different forms of plasticity are distributed across different layers of processing during olfactory learning. The honey bee will be used as a model for studying these different forms of plasticity in behavioral experiments involving different kinds of conditioning experience combined with simultaneous multiunit recordings of the outputs of two sequential layers of olfactory processing - the Antennal Lobe and Mushroom Body. Combining behavior with electrophysiology in the same animals allows for investigation of how different forms of plasticity interact with the transformation of spatiotemporal 'transient'activity patterns in the AL to sparse, spatial representations in the MB. Our group also has developed models of the AL and MB that produce realistic spiking activity patterns. In the proposal these models will continue to be extended to investigate how unsupervised and supervised (reinforcement-based) learning may be implemented in the AL and MB. For example, recent immunohistochemical analyses by one of our laboratories of the reinforcement pathway in the honey bee brain have revealed that reinforcement targets inhibitory pathways both in the AL and MB. These inhibitory pathways are important for generating transient activity patterns in the AL and the sparse coding in the MB. This information will be used to implement plasticity in the computational models. Although these models provide predictions and guidance they are still far from a clear convergence with electrophysiological and behavioral data. The work proposed in this proposal is aimed toward development of an integrative view in which new data will refine development of the models, and the models will help to guide experimental approaches. For example, spike time dependent plasticity and Hebbian learning are dominant when there is no reward and non-supervised learning occurs. On the other hand, in the honey bee octopamine release from a well-established reward pathway is critical for reinforcement learning. Our preliminary models show that Hebbian type of learning increases the similarity of correlated patterns while it separates dissimilar patterns. On the other hand, reinforcement learning operates against the Hebbian-type of rules by increasing correlations between dissimilar patterns or separating highly correlated encoded signals. The current proposal focuses on empirically and computationally testing what underlying biophysical mechanisms can lead to such observations. Future directions: Smith's laboratory has developed a means to disrupt the octopamine-mediated reward pathway in the honey bee brain (Farooqui et al 2003). In the future, this study can be extended to include targeted disruption of this and other pathways important for plasticity to test the predictions of the modeling. Enhancing public health: At least 3,000,000 Americans suffer from chemosensory disorders and that is likely to grow as the aging segment of the population increases. Olfactory impairment leads to inability to detect hazards such as natural gas and spoiled food. Olfaction is also an important early signal of the onset of neurodegenerative diseases such as Parkinson's Disease. Broader impacts: It is expected that the outcome of this research will also improve the development of algorithms for processing sensor arrays of metaoxide sensors for volatile odor compound identification that is currently funded in one of our groups through a MURI award from the Navy. The goal of the MURI project is to develop biomimetic algorithms for odor localization in robots. Furthermore, behavioral work with honey bees is inexpensive and easy to learn. One of our laboratories continuously involves undergraduate students as researchers through an HHMI funded undergraduate research program. The researchers are frequently coauthors on publications, as was the case for two students on a recent publication (Fernandez et al, 2009, J Neuroscience 29, 10191-202). Finally, one PI and the graduate student employed on this award will be involved in outreach programs at Arizona State University (sols.asu.edu/community_outreach/index.php). This research will also involve minority undergraduate students (Bazhenov lab). All of the software for running the models will be shared with other researchers and will be available through the internet.
Showing the most recent 10 out of 20 publications
