Tatyana Sharpee - US grants

One of the basic requirements for understanding how the brain works is to know what the electrical activity of each neuron "means" with respect to an animal's behavior. The electrical signals are manifested as temporal sequences of very brief electrical impulses, commonly called spikes, which are the currency of the nervous system. Neurons respond to stimuli by spiking, and they communicate with each other by spiking. Interpreting what such spiking patterns encode has been of longstanding scientific interest because it translates into knowing what each neuron does within a circuit. The aim of this project is to go one step further. The goal is to examine how the spike pattern received by one neuron is transformed into a new, output spike pattern. Although much has been learned about what spike sequences encode, the coding transformation which occurs from neuron to neuron has rarely been quantitatively investigated. Only when this recoding of spike patterns is understood can a neuron?s functional role be considered fully characterized.

The research naturally progresses in three stages. First, spike trains will be recorded from connected pairs of neurons responding to naturalistic stimuli. The investigators have chosen to record from neurons in the primate lateral geniculate nucleus because their input, provided from retinal ganglion cells, and their output can be accessed readily. Second, they will apply a novel analytical method to determine the optimal stimulus features encoded by each cell's spike train. This involves a computationally intensive search through the stimulus space to arrive at the stimulus representations which carry the maximum amount of information. By comparing the optimal representations for the input and output spike trains, the feature transformation from one neuron to the next will be revealed. Third, they will test the predictive power of their method by using the found optimal stimuli and systematically degraded stimuli to probe neural responses in a second series of recording experiments. Such a test will serve to validate the computational basis of the feature transformation.

Having chosen a problem of very fundamental interest, the methods being developed will be valuable for studying signal communication and representation in any system of connected neurons. The outcome will provide a basis for comprehending the processing capabilities of neurons with multiple inputs, which are common throughout the brain, and could be applied to engineered systems designed for interpreting visual scenes.

Description (provided by applicant): Vision is an essential part of our well-being. Yet, the neural mechanisms allowing for the stable perception of visual objects under variations in their position are not sufficiently understood. In particular, we currently lack specific knowledge of how neural responses are transformed in the ventral visual pathway where visual object recognition is performed. Here we propose to probe neurons in visual areas V2 and V4 within the ventral pathway with natural stimuli in order to systematically determine their stimulus selectivity preferences. Natural stimuli offer a systematic way to study neural feature selectivity because they elicit robust responses from neurons at all stages of visual processing and because they constitute a stimulus set that is not tied to a particular hypothesis about prevalent neural feature selectivity. We have recently developed statistical methods that allow the characterization of the feature selectivity of neurons that are probed with natural stimuli, and which take into account nonlinear aspects of neural responses. Extending these methods to account for the fact that the responses of neurons in extrastriate visual areas can be tolerant to shifts in stimulus position (aim 1) will allow the systematic characterization of the stimulus preferences of these neurons under natural conditions. Responses of each neuron will be characterized according to the two localized stimulus features that are most relevant for the neural responses. This will provide a direct measure of the selectivity to curved elements in visual scenes thought to occur in V4 responses (aim 2). Finally, we will test whether specific parameters of neural feature selectivity, such as orientation and curvature, determines the directions of increased tolerance to shifts in object position (aim 3). Understanding the neural mechanisms of shape perception in the healthy nervous system is a key element in developing effective therapies for neurological disorders, such as apperceptive agnosia, where visual object recognition is impaired. PUBLIC HEALTH RELEVANCE: This study will elucidate the neural mechanisms responsible for shape perception. These neural processes are disrupted in several neurological disorders affecting vision. Understanding the neural mechanisms of shape perception in the healthy nervous system is a key element of developing effective therapies for neurological disorders where visual object recognition is impaired.

The goals of this CAREER proposal are to help elucidate the principles that make robust object recognition possible. Object recognition is a problem that must be solved by all living organisms, from single-cell organisms to humans. Although the physical signals for recognition based on chemical events, light or sound waves are different, the computational requirements for analyzing these events appear to be similar. Specifically, there are two main properties that any system that mediates robust object recognition must have. The first property is known as "invariance." It endows neurons with a similar response to the same object observed from different viewpoints. The second property is known as "selectivity." Selectivity requires that neurons produce different responses to potentially quite similar objects (such as different faces) even when presented from similar viewpoints. It is straightforward to make detectors that are invariant but not selective or selective but not invariant. The difficulty lies in making detectors that are both selective and invariant.

This CAREER project will develop statistical methods for simultaneously characterizing both the invariance properties of neurons and their selectivity to specific features in the environment. The developed methods will have three distinguishing characteristics. First, it will be possible to recover new types of invariance without any prior assumptions of what the dominant type of invariance is for any given neuron or brain region. Second, they will make it possible to characterize imperfect and approximate types of invariance. Third, the methods will be geared towards stimuli typical of the natural sensory environment that are rich in objects and elicit robust responses from neurons from all stages of sensory processing. These three properties of the developed methods will make it possible to simultaneously study multiple neurons both within and across different regions, without the need to adjust stimuli to a particular neuron or brain region. Application of the developed methods to responses of neurons that mediate visual and auditory object recognition in the brain will help reveal the common principles of sensory processing in the brain and may ultimately lead to improved designs of artificial recognition systems, including sensory prostheses.

This research will be integrated into education and outreach activities involving K-12 students, undergraduate and graduate students. The educational component will help integrate knowledge acquired in computer science, physics, and neuroscience, training a new generation of scientists that are proficient in these disciplines. Outreach to local schools and museums, as well as the creation of an online course will help reach a diverse range of students both locally and worldwide.

This project was developed during a NSF Ideas Lab on "Cracking the Olfactory Code" and is jointly funded by the Chemistry of Life Processes program in the Chemistry Division, the Mathematical Biology program in the Division of Mathematical Sciences, the Physics of Living Systems program in the Physics Division, the Neural Systems Cluster in the Division of Integrative Organismal Systems, the Division of Biological Infrastructure, and the Division of Emerging Frontiers. The sense of smell is essential for maintaining quality of life in humans, and its decline can be an important harbinger of neurodegenerative disease. Moreover, since nearly all animals aside from primates rely on olfaction for most survival functions, understanding chemical sensing has immense practical value, for example, in the control of agricultural pests or in training animals to detect odors relevant for bomb, drug and cancer detection. In spite of its importance, the understanding of olfaction lags far behind the other senses, which is in part due to the lack of understanding of the physical space of odors. The understanding of the neural bases of vision and audition were greatly advanced by investigations of the physical dimensions of visual and auditory stimuli. It is therefore likely that a similar in-depth investigation of odor space - how natural odors occur and the backgrounds against which they must be detected - will reveal a new depth of richness of neural representations of odors in the brain. Insects such as the fruit fly and honey bee are excellent models for this research because of the accessibility of their central nervous systems, because of their ease of use under controlled laboratory conditions, and because of the functional similarity of how odors are processed in insect and mammalian brains. This research will characterize how odor flowers and fruits with respect to behavioral value for honey bees (food) and fruit flies (food and egg laying sites). Further monitoring of neural activity in early and later stage processing in the brain, when combined with computational modeling, will reveal significantly richer neural representations than have heretofore been described. This new understanding stands to have an impact on understanding how healthy brains encode sensations and memories of odors and how brains fail under disease conditions. It will also have an impact on understanding how the sense of smell may be built into engineered devices. Finally, both insects are also of economic importance to agriculture for crop pollination (honey bees) and damage to fruit (fruit flies). The PIs will teach and work with undergraduate, graduate and postdoctoral students and especially recruit students from underrepresented groups in science.

This research will quantitatively characterize the real-world statistics of multi-component natural odor scenes and investigate how they drive behavior and processing in several brain regions. The focus will be on honey bee as well as fruit fly adults and larva as models, where it will be possible to characterize a library of ethologically relevant natural odors associated with a diversity of behavioral outputs. The work will begin by quantitatively characterizing the detailed statistical properties of natural odor scenes in defined ethological contexts. This will build on the rich literature on identified natural odors in insects and mammals. Naturally occurring plant and fruit odor samples from the natural environments of each insect will be collected and chemically analyzed. Nonlinear dimensionality reduction techniques and approaches based on sparse coding will determine the dimensions of odor space that are most salient for behavioral decisions. Such a quantitative deconstruction of the sensory input would be unprecedented in olfactory neuroscience, and should allow the PIs to effectively and comprehensively drive olfactory circuits for the first time. The hypothesis is that the stimulus dimensions that are most behaviorally relevant to the animal will be most efficiently extracted by the olfactory system. Synthetic odor blends will be specially constructed to vary along relevant sensory dimensions, to probe neural codes and adaptive behaviors in the olfactory system. As in research on the visual system, analysis of such evoked neural responses using statistical methods that take into account natural odor statistics will reveal novel olfactory computations and behaviors that have been previously inaccessible. The project will generate datasets of immediate use and importance to scientists in theoretical biology and mathematics, engineering and biology.

What are the relationships between the complex and constantly changing soundscapes that surround us and the electrical activity that represents them in the brain? This project brings together three groups, one experimental and two computational, to address this question at two different levels. First, mechanistic models will be developed, based on known properties of neural networks, to describe how different types of neurons cooperate to represent sounds. Second, the function of these neurons will be characterized by describing the features of sounds that they represent. These two goals will be achieved by combining experimental studies of neural responses to sounds with computational analyses to test candidate mechanisms for how sounds are represented in large cortical circuits. In addition to deeper understanding of auditory perception, this research will provide insights into general principles of cooperation between neurons within a single neural network. As such, the research has implications for understanding representation of signals in other sensory modalities as well as the general principles of neural coding in the brain. The research has a number of potential practical applications, including the design of advanced hearing aids and artificial speech recognition systems. Further, given that the altered balance between excitatory and inhibitory neurons has been implicated in a number of attention deficits and psychiatric disorders, including autism and schizophrenia, the project has potential medical relevance. The outreach component of the project will involve demonstration involving music and speech perception for K-12 students and exhibitions.

Technically, the experimental group will produce recordings of neural responses from the auditory thalamus and cortex in response to pure tones and complex sounds known as tone clouds. The tone clouds have sharp transitions like natural sounds, but with well-controlled spectral and temporal power distributions. Computational components of this project will aim to reproduce neural recordings through analytical modeling and simulations of large scale neural circuits composed of multiple cell types. The experimental and computational results will be matched not only in statistical terms, such as average dynamics of neural responses across the population, but also in terms of specific features of sounds that are encoded by different types of neurons in the network.

This award is cofunded by the Office of International Science and Engineering. Companion projects are being funded by the French National Research Agency (ANR) and the US-Israel Binational Science Foundation (BSF).

Project Summary: Data Science Resource Core The central purpose of the Data Science Resource Core is to construct and implement digital infrastructure that can warehouse, integrate, visualize, and allow interaction with the data products of the three specific aims. This digital infrastructure will be implemented in the form of a multi-tier services-oriented architecture, reusing an open source neuroscience data sharing platform. Where possible, industry standard third party tools and services will be leveraged as a foundation for our work, to avoid rebuilding any components that are already well designed and easy to use. The products of the other Research Projects will be collected via protocols that are amenable to digital representation. These will then be unified via spatial registration in a common coordinate system. Ensuring that the data have appropriate landmarks and that measurements of the spatial extent of key images is a core operation for this endeavor. With data whose content are appropriate for unification, it will subsequently be determined which digital format representation is most appropriate for each data type, with the goal of making them easy to present in an online interface for the consumption of the broader scientific community. Data files with the appropriate format will be made available on cloud storage on the public internet. From here, a user-friendly web-based interface will be designed and implemented on top of a platform for neuroscience data visualization, that itself reuses best-in-class open source visualization software for the web. The interface will ensure that users can easily navigate between individual data sets and also can visualize relationships between data sets across the domains from the other Research Projects. The interface will enable the user to explore data collected about the major interneuron classes, as well as neurons in the major motor pools via mouse click. To encourage broad sharing of the data resources we produce, high quality metadata will be created to include with the data that conveys its provenance. Data products will be enabled to have their own digital object identifiers (DOIs), incorporate compatibility with ORCID ids, and embed RRIDs into the system as necessary. Core data products will further be shared through key federally supported data sharing resources and by applying best practices to data format and dissemination techniques used.

Project Summary: Project 1 ? Modeling Forelimb Motor Circuit Organization and Function This Research Project addresses the dynamics of motor circuits in the cervical spinal cord. Aims in this project will analyze how cervical spinal networks coordinate forelimb movement during rhythmic and non-rhythmic activity. The model will include neuronal elements (excitatory and inhibitory pre-motor neurons and antagonist pools of motor neurons), antagonist pairs of muscles, and mechanical components of the limb. Theoretical and computational techniques will be used to describe how these elements interact, including possible responses to descending brain commands, to generate forelimb activities such as isometric and isotonic movements. The first goal will be to compute how cervical spinal motor neurons fire in response to inputs (based on slice electrophysiology experiments, Projects 2 and 3), and how their firing is converted to muscle force (based on EMG recordings, Project 4). A second goal will be to analyze circuit configurations that achieve maximum control in order and predict new functional cell types that may be present in the spinal cord. Network models composed of motor and pre-motor neurons will then be constructed, defining the patterns of network firing and bursting of antagonistic motor pools, corresponding to elbow and wrist movements and static states. The model will be constrained by connectivity data (Project 2), and informed by theoretical analyses of optimal circuit configurations. Finally, a mechanical model of the limb will be connected to the neuronal model to generate a coherent picture of the transformation from descending input to the cervical spinal cord to limb movements. The dynamical responses of the model will be compared with the behavioral and in vivo electrophysiological data collected in Project 4. Overall, the theoretical analyses performed here, together with experimental investigations from Projects 2-4, will help inform general principles concerning how multiple neuronal types coordinate their response properties to achieve robust control of multiple dynamical variables.

PROJECT SUMMARY ? Integrative Models of Aging Core Aging is a multi-faceted, multi-scale and heterogeneous process. While specific pathways and genes associated with aging are beginning to emerge from computational modeling studies, in the age of high- throughput and single-cell sequencing, proteomic, and imaging analyses, researchers in the aging field must overcome a number of challenges to integrate and interpret these disparate and often massive datasets. Because of this high barrier to entry, biology of aging researchers are only beginning to approach the study of aging using integrated data-driven approaches. To support the mission of the San Diego Nathan Shock Center of Excellence in the Basic Biology of Aging (SD-NSC), the Integrative Models of Aging Core (Modeling Core) will address these challenges by devising strategies for the generation and curation of high-throughput data, developing novel integrative algorithms and models for interrogating specific aspects of aging biology, including the role of cellular heterogeneity, and implementing these models via an accessible online portal. The Modeling Core proposes the following three specific aims. First, the Core will establish sufficient data infrastructure. Integrative modeling requires high quality pre-processed data obtained from diverse high- throughput assays. This Aim will identify and implement the hardware and software resources required to pre- process and store sequencing, proteomics, metabolomics, and imaging data. Second, the Core will implement and develop integrative models of aging that take into account the heterogenicity of the process. Integrative computational models will enable the mapping of diverse pre-processed data inputs into integrated interpretable biological processes. These models can then be used to generate hypotheses that can be tested in the lab, resulting in the generation of additional datasets, which can be used to refine the models in an iterative manner. Work in this aim will consist of: 1) fulfilling requests to apply established techniques for the integrative analysis of data, and 2) developing customized predictive computational models as tools for answering specific questions regarding aging biology. Finally, the Core will provide widespread access to integrative modeling tools. While computational models allow for a systematic understanding of specific aging- related biological processes, widespread access to these models will be needed to maximize impact across NSCs and across the field of aging. This aim is focused on translating the models and theories into interactive tools for use by aging biology researchers.