RESEARCH Learning, Memory and Biological Olfaction Animals are programmed to solve certain problems by learning predictive relationships among stimuli and storing information about those predictive relationships in a form that is stable for periods ranging from one to 7x105 hours. Remarkably complex associative learning has evolved in animals with neural circuits highly suited to biophysical analysis, such as the terrestrial garden slug Limax maximus. We study odor learning and odor information processing in Limax, as this animal displays robust and reliable one-trial odor conditioning and a variety of higher-order learning modes during olfactory learning. The central circuit which stores odor memories has oscillatory dynamics of its local field potential and propagates activity waves along its apical-basal axis. This dynamics arises from a network of coupled neuronal oscillators that have a gradient of excitability along the apical-basal axis. Oscillatory dynamics is widespread in mammalian cortical networks during sensory processing and motor command generation. Thus we study both synaptic events during learning and the computational function of oscillatory dynamics in the Limax olfactory circuit to shed light on the roles of learning and oscillations during cortical processing. Imaging odor memory formation in vitro Our previous imaging studies of odor responses in the odor memory circuit used naïve brains. We now have a procedure to train a naïve nose-brain preparation in vitro by pairing nerve shock with odor application. We measure action potential generation by identified withdrawal neurons as the output measure. Before conditioning an attractive odor does not activate withdrawal motoneurons while after pairing the odor with nerve shock the odor strongly activates the identified withdrawal motoneurons. We image the odor memory circuit after staining its neurons with a voltage-sensitive dye. We record a series of images with a CCD camera and analyze the patterns of neural activity in response to odor stimulation before and after conditioning. Odor application initially causes a collapse of the apical-basal phase gradient, which in our model of odor memory formation is a necessary precondition for synaptic modification. This can now be tested. Learning-specific dye uptake If a slug is given one-trial odor conditioning and then injected with the highly fluorescent dye Lucifer yellow, a band of neurons is found in the odor memory storage circuit containing Lucifer yellow in membrane bound vesicles. Animals given odor exposure alone or unpaired applications of odor and the aversive stimulus do not show neuronal uptake of Lucifer yellow in the odor memory circuit. We are performing imaging experiments to test the hypothesis that the Lucifer yellow containing neurons found after odor conditioning store the odor memory in the strengths of their synaptic connections. We also train the naïve nose-brain preparation in vitro by pairing nerve shock and odor application while applying drugs selectively to the odor learning circuit. This in vitro training technique with drug application restricted to the odor learning circuit allows us to block long-term memory formation and determine if dye uptake is also blocked. We also want to determine if learning-specific Lucifer yellow uptake occurs in mammalian systems. Odor modulated neurogenesis Mammals and mollusks add new olfactory receptors throughout life and new olfactory interneurons until adulthood, yet their olfactory systems appear to maintain a constant input-output relation. The odor memory storage circuit in Limax hatches with a zone of active neurogenesis that produces 80% of the neurons found in the adult circuit after hatching. We explore how odor experience and odor learning affect neurogenesis by labeling dividing neurons with bromodeoxyuridine (BrdU) and giving slugs varied odor experiences before developing the BrdU label immunocytochemically. The zone of neurogenesis at hatching is at the most apical position in the odor learning circuit, where activity waves originate. New neurons are added only on the apical side of the band of neurogenesis. Removal of one nose retards neurogenesis until the nose regenerates. We plan to image neurogenesis with 2-photon laser-scanning microscopy using fluorescent nucleotides applied to an in vitro nose-brain preparation which can learn odor-shock associations while the zone of neurogenesis is being imaged. Odor learning-activated gene expression Professor Yutaka Kirino and his collaborators at the University of Tokyo have just described (Genes To Cells 6:43, 2001) the activation by one-trial odor conditioning in Limax of a gene coding for an extracellular matrix protein. The gene is expressed in neurons of the odor memory storage circuit and makes a protein that is homologous to proteins found in zebrafish, mice and human. The gene product is secreted into extracellular space and may stabilize connections between neurons in the odor learning circuit. In collaboration with the Kirino laboratory we plan to assess the effect of the new learning-specific extracellular matrix protein on connections between odor memory storage neurons cultured in vitro and on long-term memory formation by the isolated nose-brain preparation trained in vitro. Electronic Olfaction My colleagues and I study the acquisition of sensory information and the use of that information to learn stimulus associations and to guide movement decisions aimed at localizing stimulus sources. We focus on olfaction and its integration with fluid dynamic (wind) information to allow a mobile robot to recognize and localize an odor source. With H. E. Katz (Bell Laboratories) and A. Dodabalapur ( University of Texas, Austin) we are developing a new generation of odor sensors based on plastic transistors. With J. Sturm (Princeton University) we aim to make a chip with 100 different sensors and preprocessing circuitry in an area small enough to be encompassed by the volume of a single sniff. The optimal use of the sensor array input patterns is being explored in two ways. First, new pattern recognition algorithms have been developed by J. J. Hopfield (Princeton University) that require inputs from a large input array with diverse binding constants for odors, such as our new 100 transistor odor sensor array. Second, we want to test predictions from new theoretical work by E. Balkovsky and B. Shraiman (Rutgers University) who developed algorithms that consider how fluid dynamic flow downwind of an odor source affects the utility of odor sampling and movement decisions aimed at localizing the odor source. While our new enose front end is being developed, with D. D. Lee (University of Pennsylvania) we have started work on odor localization using a small commercial electronic nose (enose). This battery-powered enose is mounted on a mobile robot that provides a platform for testing the efficacy of sampling strategies and movement decisions in a real device. The enose robot is being fitted with a wind sensor and new work will explore the best way to integrate odor sampling and wind sensing for odor object localization. In previous work D. D. Lee and S. Seung (MIT) explored the use of visual and acoustic inputs to a machine learning algorithm for visual target localization. New work will explore integration of the visual and acoustic information with odor and wind information. The central processing circuitry integrating these inputs must learn to distinguish background odors and other stimulus patterns from a target stimulus configuration, which will also be a focus of new work. We also study the odor sampling strategies and olfactory interneuron (mitral cell) responses of mice trained to identify and categorize odor stimuli in the presence of background masking odors that are strong and similar to the target odor. Mitral cells respond to both odor and non-odor cues when studied in the intact mouse, which suggests new ways of weighting and integrating olfactory and non-olfactory cues to achieve optimal signal detection in artificial systems. The following questions motivate current work in the Gelperin laboratory. 1. How does odor experience, particularly odor learning, affect the rate and extent of neurogenesis in the Limax olfactory analyzer? 2. Does the learning-gene-activated peptide of the Limax olfactory analyzer stabilize synapses between cultured olfactory interneurons? 3. Does learning-specific labeling of olfactory interneurons with Lucifer yellow reveal the neurons encoding an odor memory? 4. Do gaseous neurotransmitters like nitric oxide and carbon monoxide play an essential role in olfactory oscillations and odor learning? 5. What type of neural model accounts for the finding that removing 80% of the rodent olfactory bulb does not degrade odor recognition or learning? 6. How are the responses of mitral cells in vivo modified by odor learning? 7. What odor sampling and pattern recognition algorithms allow an autonomous robot equipped with an electronic nose to find odor sources? 8. Do organic thin film transistors have sufficiently diverse and robust odor responses to provide inputs for a new generation of electronic nose? 9. How can an enose chip containing 100 organic transistor sensors be fabricated so that the odor volume of a single sniff activates all receptors?