UC Berkeley

Motor actions constitute the way in which we interact with the world, and are driven by millions of neurons in our distributed motor system. Studying how patterns of activity in motor cortical populations of neurons give rise to withholding movement, generating fast and accurate movements, and generating sequences of movements is the topic of this thesis. One challenge in studying how patterns of population activity support features of movement is that population patterns are difficult to manipulate in experiments with current neuroscience techniques. Current methods allow for anatomically-specific and cell- specific activation or inhibition, but do not, for example, allow for the careful manipulation of correlated versus uncorrelated activity pattern. We turn to closed-loop brain-machine interfaces as tools to perturb population activity patterns and to study the consequences of these perturbations on motor behavior.

The first part of this thesis focuses testing how tightly linked a specific feature of motor cortical local field potential signals are with withholding of movement. We use a non- human primate model system where subjects learn to control this neural feature, termed beta band oscillations, through a closed-loop brain-machine interface. Subjects perform tasks where they volitionally bring their internal beta band oscillatory state to a specified level, and immediately afterwards perform a motor task. The sequential task design allows for testing how tightly linked beta band oscillations are to movement onset, more so than can be claimed by correlational studies. We use a similar approach to investigate the relationship between beta band oscillations and movement in parkinsonian subjects.

Sequential task designs shed light on how a specific neural signal contributes to a feature of natural movement. Studying the complete link between cortical neural signals and natural movements, however, is challenging given i) experimenters’ limited access to neural signals driving movement ii) the number of non-linearities in the neural to movement map, iii) the challenge in fully capturing a complete picture of natural movements. One approach to simplifying the problem of studying sensorimotor control is to study control of a fully characterized virtual plant, such as a 2D velocity-controlled cursor, that is controlled neurally through an experimenter-defined transform and with observed neural activity patterns. Such systems have the advantage of allowing the experimenter to define mathematically which types of population activity influence the movement of the plant. We take advantage of this feature to investigate how different decompositions of population activity support fast versus accurate movements. Finally, we use this system to study principles of how neural population activity is generated for different orderings of cursor movements, or action sequences. We find that for action sequences that are constituted from the same commands but in a different ordering, subjects have different ways of generating the same command. We test how large these differences are by decomposing the population activity that updates the movement of the plant and assess how cursor movements are influenced. With this approach, we describe a model of how neural activity is generated that captures a majority of the neural variance observed across different action sequences.

As the motor systems neuroscience field increasingly collects simultaneously acquired population neural activity, hypotheses about how features of the population support movement will continue to emerge. Testing these hypotheses will require manipulation of possibly abstract population decompositions, a challenging feat to do precisely with current stimulation methods, but possible with closed-loop brain-machine interfaces.