Samuel J. Sober - US grants

A critical unsolved problem in neuroscience is to understand how learning is implemented by the nervous system. However, despite years of experimental study, our understanding of how patterns of electrical activity in the brain, and in the muscles that implement behavior, is still rudimentary. A critical gap therefore exists in our understanding of the biological and computational underpinnings of motor learning. Songbirds provide a physiologically accessible model system in which to investigate how motor memories are formed during development and executed in adulthood. The proposed studies exploit the strength of the songbird model to test a novel theory of motor learning, linking changes in the patterns of electrical activity in neurons and muscle fibers to the improvements in motor performance that typify skill learning. Our long-term goal is to fully understand how neural codes change during learning. The objective of the proposed experiments is to understand the adaptive control of a vocal acoustics in birdsong. Our central hypothesis is that that the fundamental computation underlying motor learning is the brain's search for the precisely-timed spike patterns that exploit muscle biomechanics to achieve the desired behavior. This hypothesis emerges from our previous work demonstrating that in adult songbirds, which have fully learned their songs, both neurons and vocal muscle fibers employ a spike-timing based code (rather than a rate-based code) to control vocal behavior. These results, combined with other data showing that the statistics of neural activity change dramatically as a bird first learns to sing, suggest that the key change underlying vocal skill learning is a developmental transition from a rate-based spike code to a timing-based code. Employing a number of novel experimental and computational tools that we have developed, we will test our hypothesis in three specific aims. In the first aim, we will compare the spiking code employed by neurons in the motor cortex before and after young birds learn to sing, providing insight into whether and how the brain's neural strategies for motor control change during learning. In the second aim, we will analyze spiking data from muscle fibers over the same interval, allowing us to determine how muscle codes change with learning, and, using innovative in vitro and ex vivo approaches, quantify how changes in muscle activation improve motor performance in the face of developmental changes in biomechanics. In the third aim, we will record spiking activity from individual motor units continuously while adult birds perform a rapid motor learning task, revealing the changes that occur within a single unit during learning.

DESCRIPTION (provided by applicant): A central goal of neuroscience is to understand how learning algorithms are implemented by neurons and muscles. However, despite decades of psychophysical studies in humans, our understanding of how motor learning is implemented physiologically is rudimentary. A critical gap therefore exists between psychophysical models of learning and the physiological changes in the motor program that reshape behavior. Songbirds provide a physiologically accessible model system in which to investigate behavioral plasticity. However, song learning has previously been studied on timescales too long to allow single-neuron recordings, making it impossible to identify the changes in neural activity that underlie learning. Furthermore, the functions of the song muscles themselves are poorly understood, limiting our understanding of how vocal muscles and the neurons that activate them control behaviorally important acoustic parameters. The proposed experiments overcome these obstacles by combining behavioral and computational approaches drawn from human motor psychophysics with the neurophysiological accessibility of the songbird system, linking learning algorithms to neurons and muscles. Our long-term goal is to understand how the brain controls and modifies vocal output as an animal acquires vocal behaviors and maintains vocal performance throughout its lifetime. The objective of the proposed experiments is to reveal how a single acoustic parameter - fundamental frequency (pitch) - is modified during short-term vocal error correction. Our central hypothesis is that pitch learning depends strongly on the statistics of prior sensorimotor experience, that vocal muscles exert bidirectional influence on pitch across different vocal gestures (song syllables), and that pitch learning is implemented by altering the spike content of bursts fired by neurons in a forebrain premotor nucleus. Drawing on significant quantities of preliminary data, three specific aims will test this hypothesis. The first aim will challenge current theories of vocal learning by using manipulations of auditory feedback to drive adaptive pitch changes in singing birds. The second specific aim will quantify the functions of individual vocal muscles and reveal how muscle activity changes during learning by combining precisely-timed muscle stimulation, behavioral manipulations, and EMG recordings. The third aim will (for the first time) define the changes neural activity that underlie vocal learning by recording from single neurons during a rapid vocal learning paradigm, identifying a locus of vocal motor plasticity and establishing the songbird as one of the only available systems for studying changes in neural activity during online learning. This approach is innovative because it allows us to detect changes in motor command signals online during learning, providing a critical link between behavioral and physiological approaches to motor learning. These studies are significant because a better understanding of the mechanisms of sensorimotor learning could aid in the design of rehabilitative strategies that exploit the plasticity of complex behavio.

A central question in neuroscience is how the brain detects and corrects errors in behavior. This type of error correction is essential for the learning and performance of complex tasks such as speaking and using tools, and deficits of error correction resulting from neurological diseases can have devastating effects on the ability of an individual to interact with the world. This research will examine the mechanisms by which the brain corrects errors in vocal behavior. By building mathematical models of how the brain detects and eliminates errors in performance and by directly recording the electrical activity of individual brain cells, these studies will provide novel insights into this key aspect of brain function. Additionally, the investigators will develop course materials based on their research to be used to enhance science education in underserved communities in the Atlanta area. The benefit to society of this work is therefore twofold: it will reveal previously unknown mechanisms of brain function and will improve the general public's understanding of neurobiology.

The proposed experiments combine a range of experimental approaches to test a simple model of how the brain rapidly corrects errors in behavior. The central hypothesis is that the brain is able to rapidly correct errors in vocal behavior because brain cells that control the vocal output are affected by the sensory (auditory) feedback arising from vocal output. By collecting electrical recordings of single brain cells during both vocal behavior and auditory stimulation, experiments in Aim 1 will compare the sensory tuning (the tuning for the pitch of a recently-heard syllable) and motor tuning (effect on the pitch of vocal output) of neurons that control behavior. Studies in Aim 2 will use lesions and reversible inactivations to selectively eliminate auditory input to premotor neurons, thereby facilitating tests of the hypothesis that such sensory inputs are required for rapid online error correction.

A central goal of neuroscience is to understand how learning is implemented by the nervous system. However, despite years of studies in animals and humans, our understanding of both the computational basis of learning and its implementation by the brain is still rudimentary. A critical gap therefore exists between the large amount of behavioral and neural data that has been collected during learning and a mathematical and biological understanding of the rules governing motor plasticity. This proposal will develop a unified mathematical theory for understanding how the brain learns complex skills. The theoretical framework will be implemented in software and will be applicable to and validated on a wide variety of sensorimotor data. The primary experimental validation system will be songbirds, which provide a physiologically accessible model system to investigate sensorimotor learning. Our objective in the songbird system is to understand sensorimotor learning of a single acoustic parameter ? fundamental frequency (pitch) ? which is known to be precisely regulated by the songbird brain. Our central hypothesis is that learning is implemented as a Bayesian inference, and that the stochastic sampling of motor commands from the current Bayesian a priori distribution of outputs is coordinated by a network of neurons in the forebrain. Drawing on a large quantity of both theoretical and experimental results, two specific aims will test this hypothesis. The first aim will introduce an innovative new class of computational model in which the brain uses an iterative process of Bayesian inference to reshape behavior in response to sensory feedback. The models will be validated using population-averaged animal behavior. The second aim will analyze data recorded from individual animals and single neurons in behaving animals to identify the biological mechanisms underlying sensorimotor learning. Throughout, we will design, test, and make public software that will allow other members of the community to apply our novel tools to their own data. Our approach is innovative because it will provide a unified framework for understanding the results of a wide variety of behavioral and neural studies across both tasks and species. These studies are significant because a better understanding of the mechanisms underlying sensorimotor learning could aid in the design of rehabilitative strategies that exploit the plasticity of complex behavior.

A central goal of neuroscience is to understand how patterns of neural activity in the brain control behavior. In principle, neurons can encode information via their firing rates, the precise timing of their spikes, or both. However, nearly all prior studies of motor systems have relied exclusively on spike rates to investigate how the brain controls behavior. We recently demonstrated that in songbird vocal motor cortex, spike timing ? down to millisecond-scale differences in spike patterning ? is far more informative about upcoming behavior than is spike rate. Although this suggests that variations in cortical spike timing could modulate behavior, it is unknown whether precise cortical spike timing drives a similarly precise code downstream in motor neurons and muscle tissue, or indeed whether variations in motor neuron spike timing are capable of modifying behavior. The proposed experiments will combine innovative behavioral, physiological, and computational techniques to understand how the nervous system uses precisely timed patterns of electrical activity to regulate the acoustics of vocal output in songbirds. Our long-term goal is to understand how spiking activity in the brain controls behavior. The objective of this proposal is to determine which properties of motor spiking drive variations in behavior in control of vocalizations in songbirds. Our central hypothesis is that the brain controls behavior by precisely (down to a precision of a few milliseconds) modulating spike timing. This hypothesis will be tested in three specific Aims. In Aim 1, we will use a newly developed electrode system to study spiking activity from muscle tissue (i.e., the spikes of individual motor units, each of which consists muscle fibers innervated by a single motor neuron) in vocalizing songbirds to determine the timescale of neuromuscular control. In Aim 2, we will use innovative in vivo, in vitro, and ex vivo techniques to determine whether spike-timing differences observed in muscle fibers affect motor output. In Aim 3, we will examine how precise firing patterns are coordinated across multiple muscles. All three Aims will tightly integrate experimental studies and computational analyses to identify specific spike timing patterns that most strongly influence behavior and to generate and test hypotheses about the biomechanical bases of precise motor control. The rationale for these studies is that they will upend long-held notions that cortical motor control is based solely on spike rate codes and establish a broadly applicable framework for analyzing timing-based spike codes both within and beyond the motor system. Furthermore, our findings and techniques may significantly contribute to the improvement of neural prosthetic devices by showing how decoding algorithms might make use of spike timing in addition to the rate information commonly used in current approaches.

Understanding how the brain learns and controls complex skills is one of the crucial problems of neuroscience. In both humans and animals, "motor exploration" - the process by which the brain generates variations in behavior in order to learn - is critical for both the initial learning of a skill and for the recovery of motor function following neurological injuries. In prior research, motor exploration has been assumed to be a random process. In the proposed research, we will evaluate the hypothesis that motor exploration is not a random process, but rather a systematic exploration, of many patterns of brain activity that might be used to achieve the desired output. Understanding how motor exploration is implemented biologically would not only answer a key basic question in neuroscience but would also open doors for creation of new approaches to rehabilitative therapies.

We will achieve these goals by collecting data from the vocal control system in songbirds and by creating novel mathematical tools that can be applied to data from any species. Vocal learning in songbirds shares many behavioral similarities to speech learning, providing one of the few non-human systems in which to examine how the brain learns and controls the sound signals that allow individuals to communicate with each other. Our experimental and computational studies will reveal how the nervous system coordinates the precisely-timed patterns of electrical activity that allow songbirds to explore different vocal outputs in order to produce the desired sound. These studies will, enhance our understanding of complex behavior, create novel analysis techniques for use by other researchers. Additionally, as part of this award we will engage with local middle school teachers to create educational resources for enhancing public education in neuroscience, create training opportunities for graduate students from under-represented groups, and write a textbook on how mathematical tools arising from physics can best be applied to biological problems.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

A crucial problem in motor neuroscience is to understand how muscles are activated to produce normal or pathological motor behavior. However, our understanding of how individual motor units (the muscle fibers activated by a single motor neuron) pattern their spiking to precisely control behavior is poor due to the limitations of current electromyographic (EMG) methods, which include fine wires inserted into muscles and, more recently, cutaneous EMG arrays used to record individual motor units, particularly in human subjects. However, both wire-based and cutaneous arrays face crucial limitations. Notably, they cannot record the very small and/or deep muscles that mediate fine motor control, due to tissue damage induced by wire insertion and by surface array's inability to monitor signals from deeper muscles. Furthermore, surface arrays have limited use outside of tightly-controlled (e.g. isometric force) tasks in humans, are not appropriate for long-term use in natural behaviors or in rehabilitative contexts, and perform poorly in animal models. Finally, wire electrodes typically provide only bulk multiunit recording (rather than single-unit spike trains). In this cross-disciplinary proposal between an electrophysiologist (Dr. Sober) and an electrical engineer (Dr. Bakir), we propose to fill this gap by creating a new generation of micro-scale, high channel-count EMG arrays to record massively parallel single-unit data across many muscles during natural behaviors.