Michael J. Berry - US grants

[unreadable] DESCRIPTION (provided by applicant): The broad goals of this project are to extend and deepen our understanding of the properties of neural circuits. The vertebrate retina is chosen as a model system, because of its ease of experimental access and its complex anatomy. When neurons are hooked together into a circuit, two properties become important: first, the activity of the neurons means something in the context of the animal; second, neurons perform a computation on their inputs. We thus will study how the retina encodes and processes visual information. The specific aims are: 1) perform a functional classification of retinal ganglion cells using information theoretic techniques, 2) acquire a database of natural movie clips and categorize their statistics, 3) build a spike word dictionary to efficiently capture nonlinear processing in the retina, and 4) investigate the generality of retinal adaptation and characterize its effects on the neural code. A detailed knowledge of what image processing occurs in the retina is of fundamental interest to neuroscience and is also important for a retinal prosthesis to be able to restore vision successfully.

Everything that we know about the visual world is built from the sequences of discrete electrical pulses that stream from eye to brain along the roughly one million individual cables that form the optic nerve. Until recently, when scientists wanted to "listen in" on these signals, all they could do was to monitor one cable or cell at a time, getting only a very limited view of the messages being passed through the visual system. In particular, this left us literally and figuratively blind to the possibility that combinations of pulses from neighboring cells could have special meaning. The emergence of experimental techniques for recording the signals from many cells simultaneously, under reasonably natural stimulus conditions, thus offers an unprecedented opportunity to explore the full complexity of the raw data that the brain has to work with as it tries to understand what we see. Indeed, the most recent generation of multi-electrode array technologies achieves a sampling efficiency that should allow recording from nearly all of the cells in a small patch of the retina, providing access to all of the signals that the brain can use in making sense out of the corresponding small patch of the visual world. The goals of this project are to develop in tandem the experimental methods to realize this promise and the theoretical framework for understanding the problem that the brain has to solve in dealing with these data.

Even a relatively small group of twenty cells in the eye can send millions of possible signals to the brain. Just as not all combinations of twenty letters constitute real words or phrases in English, we expect that not all combinations of pulses from nearby cells actually are used. Building on preliminary results, we will develop the mathematical tools required to describe the analog of spelling rules for these neural words. What we intuitively think of as a surprising event presumably is represented by a rare word, and we will push our understanding of the relative frequencies with which the different words are used so that we can identify reliably the signals that the brain must recognize as surprising. In a similar direction we will examine how one word predicts the next, and how the brain could effectively separate the predictable from the surprising. We will compare these theoretical descriptions with what is known about brain's mechanisms for detecting such events. Improving our instruments in parallel with these mathematical developments will allow us to deal not just with twenty cells but with roughly two hundred, where the number of possibilities truly becomes astronomical.

Historically, studies of the retina have provided the conceptual foundations for work on many different areas of the brain, and we thus expect that the results of this project will help understand not just the language that the eye uses in describing the world but also provide a general set of tools for exploring the way in which many cells cooperate to shape our perceptions and thoughts.

DESCRIPTION (provided by applicant): The broad goals of this project are to understand the scope and sophistication of retinal processing. This project builds on preliminary data showing that the retina can quickly learn simple patterns in the visual stimulus, such as a temporal periodicity or a smooth motion trajectory, so that a subset of ganglion cells fires selectively after a violation of the predicted pattern. Such cognitively loaded phenomena are typically thought to arise only in the cortex, but observing them in the retina both enhances our appreciation of the computational capabilities of neural circuits and also allows us to explore the mechanisms responsible, a task that often cannot be carried out in more central neural circuits. The Specific Aims are: 1) exploring the properties of retinal pattern detection with multi-electrode array experiments;2) studying the mechanisms of temporal pattern detection using designed stimuli, pharmacology, and computational models;3) studying the properties of and mechanisms behind the retina's ability to detect motion discontinuities. A detailed knowledge of how the population of retinal ganglion cells represents the visual world is of fundamental interest to neuroscience and is also important for guiding the development of a retinal prosthesis to restore vision successfully.

DESCRIPTION (provided by applicant): The goal of the proposed course, entitled Neurotechnologies for Analysis of Network Dynamics (NAND), is to introduce students with training in physics, mathematics, computer science, and engineering to the theory and practice of modern neuroscience, with special emphasis on modern methods for the analysis of network dynamics in mammalian cortical circuits responsible for behavioral decision making. Recent progress in developing methods for electrical and optical recording from 10s to 100s of neurons simultaneously with cellular resolution is revolutionizing our ability to define neural activity sttes that correspond to periods of memory storage and decision-making in awake, behaving vertebrate animals, including mammals. The goal of this new course is to give students with advanced training in quantitative disciplines outside of neuroscience the theoretical background and practical experience needed to understand and contribute to the ongoing revolution in the analysis of neural network dynamics in anatomically defined neural circuits being used to make decisions based on sensory input, implement those decisions by generation of motor commands, and store information in both short-term and long-tem memory for use in subsequent decision-making. A four-week course is planned. The lectures will introduce the students to: i) the basics of cellular and synaptic physiology; ii) an array of cellular and molecular tools to facilitate network analysis; iii) technology for large scale multi-site recordin; iv) the analysis of large data sets produced by multi-channel recording; v) methods for obtaining multi-site recordings, both electrical and optical, from both animal and human brains; vi) models of information processing in cortical circuits, particularly during reinforcement learning. The laboratory component will provide experience in the recording and analysis of extra- and intracellular recordings, generation and recording of synaptic plasticity in the hippocampus, and multi-site optical and electrical recordings from awake behaving rodents. Students will use optogenetic methods for circuit analysis and learn to make fMRI measurements of activity in human subjects during an information processing task. The capstone of the laboratory is a one-week period devoted to student designed Independent Projects.

In virtually every part of the brain, information about the sensory environment, internal body states, or intended movements is encoded by more than one neuron. This was apparent as early as the nineteenth century from the extensive interconnectivity of nearby neurons and continues to be apparent from numerous measurements of the tuning curves and correlation of nearby neurons. Despite its fundamental importance, population neural codes are poorly understood. In this project, the PI will combine large-scale neural recording methods with state-of-the-art theoretical analyses to study collective phenomena in population neural codes. The project will focus on the retina as a model system, where the quality and completeness of experimental data is the greatest. The project will give us a thorough and clear understanding of what are the "ingredients" needed at the population level to give rise to criticality, which will be important in generating hypotheses about what other regions of the brain might exhibit criticality in their population codes. It will also explore a novel hypothesis about how critical population states give rise to a discrete aspect of the neural code and one that is highly robust to neural noise. This could give us new insight into how we effortlessly divide the sensory world into objects.

The PI has an extensive track record in multi-electrode recording from the vertebrate retina, along with applying maximum entropy models to analyze states of network activity. The PI hypothesizes that the retinal population code has an unusual and nontrivial structure that it analogous to the critical state in physical systems. This structure leads to the definition of a "collective mode" of neural activity, which is a set of neural activity states that groups visual stimuli into discrete classes. These collective modes constitute a novel hypothesis about population neural codes that is qualitatively different from the view of information encoding at the single-neuron level. The proposed projects aims are: understanding the origin of criticality, including studying the role of correlations in the stimulus, using adaptation to test how specifically retinal circuitry is tuned to give rise to criticality, studying the pattern and strength of correlation required for any network to give rise to criticality as well as formulating receptive field models to see what degrees of overlap and functional heterogeneity are required in defining collective modes of neural activity and studying what stimuli they encode and how reliably they are activated by a given visual stimulus. The broader impacts of the proposed work lie along three distinct directions:(i) the potential to develop new concepts about population neural codes that could prove applicable across many different brain regions; (ii) the development and dissemination of software to perform maximum entropy fits to neural data, which could help spur on the research programs of many labs that use these methods to analyze neural populations; (iii) the broadening of our ideas about critical systems in physics to include the kind of asymmetric, intermediate cases found in biology.

This project is being jointly supported by the Physics of Living Systems program in the Division of Physics and the Cellular Dynamics and Function Program in the Division of Integrative Organismal Systems.

? DESCRIPTION (provided by applicant): The broad goals of this project are to explore adaptation and learning in response to repeated presentation of temporal sequences of visual stimuli. These studies are motivated by broader ideas about how local circuits in the neocortex may be carrying out predictive computations on their inputs. Such ideas could lead to new insights into the function of the human cerebral cortex, and hence have many possible medical and technological applications. In the proposal, we describe preliminary data taken in collaboration with the Tank lab at Princeton. We first performed behavioral experiments in which we trained mice to lick in response to a violation of a repeated temporal sequence. This behavior was robust and emerged rapidly, often on the first day of training. Next, we used two-photon imaging to record from neurons in layer 2/3 of mouse V1. We found that neurons rapidly modified their responses during repeated presentation of temporal sequences. Most of the neurons adapted rapidly to the presentation of repeated temporal sequences of oriented spatial images, reaching essentially zero baseline response in several seconds (what we call the `transient response'). Then, these same neurons generate a strong response to a novel spatial image that violates the ongoing temporal sequence. In addition, we find a small subset of neurons (roughly 2 in 100) that produce a sustained response to repeated temporal sequences. This sustained response ramped up over several cycles of the sequence. Then, on a longer time scale, it exhibited some features of learning, such as an anticipatory shift to earlier times s well as a form of pattern completion when presented with a novel image. We propose to extend our studies in two ways: 1) consolidate our behavioral measurements over a larger cohort of animals with identical shaping and training histories; 2) study how these responses change during training in animals engaged in the behavioral task.