Christopher J. Rozell - US grants

DESCRIPTION (provided by applicant): This project aims to achieve a fundamental advance in our understanding of how neural populations process and represent information within sensory cortex. By combining pioneering recording technology with new analytical tools and theoretical frameworks, this research effort will provide the first glimpse at how large numbers of neurons interact within the cortex during the processing of dynamic natural scenes. Silicon polytrodes will be used to record simultaneously from populations of 100+ neurons in cortex. The activity of these populations will be characterized in terms of response precision, sparsity, correlation, and LFP coherence. In order to elucidate the causal factors that contribute to stimulus-evoked responses in the cortex, the joint activity and stimuli will be fit with predictive models that attempt to capture the stimulus-response relationships of large neuronal ensembles. Finally, we will attempt to account for these relationships by building functional models that achieve theoretically-motivated information processing objectives for perception and cognition. The project is highly interdisciplinary in nature, combining the expertise of neurophysiologists, theoreticians, and engineers to answer questions that are beyond the scope of any one discipline. Intellectual merit. The question of how the cortex processes and represents sensory information has been the subject of neurophysiological and neuroanatomical investigation for at least four decades. While much has been learned from these efforts, there remain many fundamental, unanswered questions regarding the dynamical properties of neurons and the information processing capabilities of this system. The usual approach of studying single-unit responses to simple stimuli is limited in that it assumes - either explicitly or implicitly - that the system can be understood one component at a time. In a non-linear dynamical system it is difficult to predict how effects observed in isolation will behave when combined. Thus, in order to properly characterize and understand the dynamics of cortical circuits, it is necessary to observe the joint activities of large numbers of simultaneously recorded neurons in response to complex, timevarying signals arising from dynamic natural scenes. This project represents the first-ever attempt to thoroughly examine the joint responses of large numbers of neurons in the cortex during natural vision. Combined with the computational modeling and theoretical developments that will incorporate findings originating from these studies, this project has the potential to fundamentally advance our understanding of how cortical circuits work. Broader impacts. This project will provide research training to two graduate students, one in neuroscience (UC Berkeley) and one in engineering (Georgia Tech), and these studies will constitute the bulk of their Ph.D. theses. Efforts will be made to recruit women and underrepresented minorities into these positions. The methods developed and the results obtained from this study will be incorporated into coursework at UC Berkeley, Georgia Institute of Technology and Montana State University, and data will be made available on the NSF-funded CRCNS datasharing facility. Advancing our understanding of neural circuit dynamics within the cortex could lead to the development of viable therapies for myriad neurological disorders, and it is crucial to the development of neural prostheses. Furthermore, the proposed work will strengthen our infrastructure for further studies of the cortex by pioneering new simultaneous recording techniques and making the data publicly available as part of a CRCNS data sharing project.

This project aims to achieve a fundamental advance in our understanding of how neural populations process and represent information within visual cortex. By combining pioneering recording technology with new analytical tools and theoretical frameworks, this research effort will provide the first glimpse at how large numbers of neurons interact within the cortex during the processing of dynamic natural scenes. Silicon polytrodes will be used to record simultaneously from populations of lOO-i- neurons in primary visual cortex. The activity of these populations will be characterized in terms of response precision, sparsity, correlation, and LFP coherence. In order to elucidate the causal factors that contribute to stimulus-evoked responses in the cortex, the joint activity and stimuli will be fit with predictive models that attempt to capture the stimulus-response relationships of large neuronal ensembles. Finally, we will attempt to account for these relationships by building functional models that achieve theoretically-motivated information processing objectives for perception and cognition. The project is highly interdisciplinary in nature, combining the expertise of neurophyslologists, theoreticians, and engineers to answer questions that are beyond the scope of any one discipline. RELEVANCE (See instructions): The data obtained and models developed in this work will open a new window into the operation of cortical circuits, providing a first glimpse of the simultaneous activity of large numbers of neurons responding to dynamic natural scenes. These new insights will pave the way for the development of neural prosthetic devices (cortical implants) and new forms of treatment for visual disorders.

PROJECT SUMMARY Deep brain stimulation (DBS) of the subcallosal cingulate (SCC) white matter is an emerging new treatment strategy for treatment resistant depression (TRD) with published studies demonstrating sustained long-term antidepressant effects in 40-60% of implanted patients. Converging evidence from positron emission tomography (PET), electroencephalography (EEG) and diffusion tractography (DTI) strongly suggests that DBS mediates its clinical benefits by direct modulation of the SCC--a key hub in an aberrant neural circuit. Despite encouraging sustained long-term effects in this notoriously difficult to treat patient population, randomized controls trials of SCC DBS and other DBS targets for TRD are now on hold as initial results failed to meet predefined clinical endpoints. While this proposal cannot address those failures directly, a clear necessary next step for effective future testing and eventual dissemination of this treatment is the need to develop brain-based biomarkers to guide lead placement and to titrate stimulation parameters during ongoing care. In the absence of such biomarkers to guide DBS use, there will continue to be variability in the implementation of clinical procedures during testing, leading to ambiguous and possibly misleading trial outcomes, and subsequent abandonment of a potentially useful treatment. To overcome these limitations, we propose to develop and test objective methods for reliable device configuration in individuals by optimizing DBS-SCC treatment with respect to human functional anatomy and key electrophysiological variables. We will leverage the capabilities of a novel bi-directional neuromodulation system (Medtronic RC+S) that allows live streaming of oscillatory activity at the site of stimulation to define novel control strategies to guide programming decisions for DBS delivery. Ongoing measurements of SCC local field potentials (LFPs) will be combined with electroencephalography (EEG) and event related potential studies (ERP) performed as part of an experimental clinical trial of subcallosal cingulate DBS for TRD to identify an oscillatory signal that (1) is sensitive to changes in frequency and current parameters at the tractography defined optimal target and (2) tracks with depression state over time. Connectome-based and machine learning approaches will be used to define the most robust network biomarker and its response characteristics. Once defined, the control policy will be tested in a second phase feasibility study where parameters for initial stimulation will be selected based on the depression brain state biomarker and adjustments made to correct drift from the predefined target signal. If successful, the data- driven model and control strategy will enable objective, rational clinical programming of DBS stimulation for depression and provide a new model and approach for target identification, stimulation initiation and long-term monitoring and management of patients receiving this treatment. .

Despite substantial progress characterizing neural responses, it is particularly challenging to determine causal interactions within recurrently connected circuits due to the confounding influence of the interconnections. This proposed project pioneers a nascent field of closed-loop computational neuroscience that enables real-time feedback stimulation during experiments to decouple recurrently connected elements and make stronger causal inferences about their interactions. Specifically, the contributions of this project will include: Aim 1) Using modern unsupervised machine learning methods to fit latent state dynamical system models of population responses under closed-loop stimulation. The developed techniques will be used to clamp firing rate in genetically targeted inhibitory interneurons across S 1 cortical laminae in the mouse to map the causal effect of inhibitory cells on the sensory gain in excitatory cells. Aim 2) Merging and extending tools from network feedback control and causal inference to identify functional connections between network nodes using realistic experimental constraints. These techniques will be used to clamp firing rate in different S1 laminae of the mouse, using distributed perturbations to identify the functional connectivity between microcircuit layers during sensory stimulation. Aim 3) Developing a large-scale computational modeling environment to serve as an in si\ico testbed for the community. Significance: The proposed project changes the de facto standard use of stimulation in experiments to leverage the full power of new recording and s.timulation technology for decoupling recurrently connected variables and making stronger causal inferences. Broader impacts: While the project uses rodent somatosensation as a model system, the results of this project will provide new techniques to study neurologic disorders involving disfunction of recurrent circuits (e.g., epilepsy, Parkinson's disease and depression). The open-source implementations will constitute critical algorithmic infrastructure for closed-loop stimulation experiments. This project will also result in the production of new trainees in an emerging new interdisciplinary field of closed-loop computational neuroscience. RELEVANCE (See instructions): The neural circuits that fail in many neurologic disorders (e.g., epilepsy, Parkinson's disease and depression) are difficult to study because they involve complex feedback loops. This project will develop algorithms that combine measurements and stimulation in real-time to provide powerful new tools to uncover the operating principles of these circuits and change their operation. Discovery in this area can help improve understanding of neurologic disorders and development of new stimulation therapies.

Despite substantial progress characterizing neural responses, it is particularly challenging to determine causal interactions within recurrently connected circuits due to the confounding influence of the interconnections. This proposed project pioneers a nascent field of closed-loop computational neuroscience that enables real-time feedback stimulation during experiments to decouple recurrently connected elements and make stronger causal inferences about their interactions. Specifically, the contributions of this project will include: Aim 1) Using modern unsupervised machine learning methods to fit latent state dynamical system models of population responses under closed-loop stimulation. The developed techniques will be used to clamp firing rate in genetically targeted inhibitory interneurons across S 1 cortical laminae in the mouse to map the causal effect of inhibitory cells on the sensory gain in excitatory cells. Aim 2) Merging and extending tools from network feedback control and causal inference to identify functional connections between network nodes using realistic experimental constraints. These techniques will be used to clamp firing rate in different S1 laminae of the mouse, using distributed perturbations to identify the functional connectivity between microcircuit layers during sensory stimulation. Aim 3) Developing a large-scale computational modeling environment to serve as an in si\ico testbed for the community. Significance: The proposed project changes the de facto standard use of stimulation in experiments to leverage the full power of new recording and s.timulation technology for decoupling recurrently connected variables and making stronger causal inferences. Broader impacts: While the project uses rodent somatosensation as a model system, the results of this project will provide new techniques to study neurologic disorders involving disfunction of recurrent circuits (e.g., epilepsy, Parkinson's disease and depression). The open-source implementations will constitute critical algorithmic infrastructure for closed-loop stimulation experiments. This project will also result in the production of new trainees in an emerging new interdisciplinary field of closed-loop computational neuroscience. RELEVANCE (See instructions): The neural circuits that fail in many neurologic disorders (e.g., epilepsy, Parkinson's disease and depression) are difficult to study because they involve complex feedback loops. This project will develop algorithms that combine measurements and stimulation in real-time to provide powerful new tools to uncover the operating principles of these circuits and change their operation. Discovery in this area can help improve understanding of neurologic disorders and development of new stimulation therapies.