Validating markerless pose estimation with 3D X-ray radiography.

To reveal the neurophysiological underpinnings of natural movement, neural recordings must be paired with accurate tracking of limbs and postures. Here, we evaluated the accuracy of DeepLabCut (DLC), a deep learning markerless motion capture approach, by comparing it with a 3D X-ray video radiography system that tracks markers placed under the skin (XROMM). We recorded behavioral data simultaneously with XROMM and RGB video as marmosets foraged and reconstructed 3D kinematics in a common coordinate system. We used the toolkit Anipose to filter and triangulate DLC trajectories of 11 markers on the forelimb and torso and found a low median error (0.228 cm) between the two modalities corresponding to 2.0% of the range of motion. For studies allowing this relatively small error, DLC and similar markerless pose estimation tools enable the study of increasingly naturalistic behaviors in many fields including non-human primate motor control.

Neural oscillations reflect latent learning states underlying dual-context sensorimotor adaptation.

Recent studies have suggested that individuals can form multiple motor memories when simultaneously adapting to multiple, but oppositely-oriented perturbations. These findings predict that individuals detect the change in learning context, allowing the selective initialization and update of motor memories. However, previous electrophysiological studies of sensorimotor adaptation have not identified a neural mechanism supporting the detection of a context switch and adaptation to separate contexts. Here, we tested the hypothesis that such a mechanism is identifiable through neural oscillations measured through EEG. Human participants learned to manipulate an object in two opposite contexts (mass distribution). This task was designed based on previous work showing that people can adapt to both contexts. We found that sensorimotor α and ß, and medial frontal θ frequency bands all exhibited different response patterns with respect to the error in each context. To determine whether any frequency's responses to error were distinctly related to a switch in context, we predicted single-trial EEG data from a computational learning model that can adapt to multiple contexts simultaneously based on a switching mechanism. This analysis revealed that only medial frontal θ was predicted by a component of the model state that adapts to errors based on a context switch. In contrast, α and ß were predicted by a model state that was updated from performance errors independent of the context. These findings provide novel evidence showing that sensorimotor and medial frontal oscillations are predicted by different adaptation processes, and that changes in medial frontal activity may indicate the formation of motor memories by responding to changes in learning context.

A dynamic subset of network interactions underlies tuning to natural movements in marmoset sensorimotor cortex.

Mechanisms of computation in sensorimotor cortex must be flexible and robust to support skilled motor behavior. Patterns of neuronal coactivity emerge as a result of computational processes. Pairwise spike-time statistical relationships, across the population, can be summarized as a functional network (FN) which retains single-unit properties. We record populations of single-unit neural activity in forelimb sensorimotor cortex during prey-capture and spontaneous behavior and use an encoding model incorporating kinematic trajectories and network features to predict single-unit activity during forelimb movements. The contribution of network features depends on structured connectivity within strongly connected functional groups. We identify a context-specific functional group that is highly tuned to kinematics and reorganizes its connectivity between spontaneous and prey-capture movements. In the remaining context-invariant group, interactions are comparatively stable across behaviors and units are less tuned to kinematics. This suggests different roles in producing natural forelimb movements and contextualizes single-unit tuning properties within population dynamics.

Microstimulation of human somatosensory cortex evokes task-dependent, spatially patterned responses in motor cortex.

The primary motor (M1) and somatosensory (S1) cortices play critical roles in motor control but the signaling between these structures is poorly understood. To fill this gap, we recorded - in three participants in an ongoing human clinical trial (NCT01894802) for people with paralyzed hands - the responses evoked in the hand and arm representations of M1 during intracortical microstimulation (ICMS) in the hand representation of S1. We found that ICMS of S1 activated some M1 neurons at short, fixed latencies consistent with monosynaptic activation. Additionally, most of the ICMS-evoked responses in M1 were more variable in time, suggesting indirect effects of stimulation. The spatial pattern of M1 activation varied systematically: S1 electrodes that elicited percepts in a finger preferentially activated M1 neurons excited during that finger's movement. Moreover, the indirect effects of S1 ICMS on M1 were context dependent, such that the magnitude and even sign relative to baseline varied across tasks. We tested the implications of these effects for brain-control of a virtual hand, in which ICMS conveyed tactile feedback. While ICMS-evoked activation of M1 disrupted decoder performance, this disruption was minimized using biomimetic stimulation, which emphasizes contact transients at the onset and offset of grasp, and reduces sustained stimulation.

Unexpected complexity of everyday manual behaviors.

How does the brain control an effector as complex and versatile as the hand? One possibility is that neural control is simplified by limiting the space of hand movements. Indeed, hand kinematics can be largely described within 8 to 10 dimensions. This oft replicated finding has been construed as evidence that hand postures are confined to this subspace. A prediction from this hypothesis is that dimensions outside of this subspace reflect noise. To address this question, we track the hand of human participants as they perform two tasks-grasping and signing in American Sign Language. We apply multiple dimension reduction techniques and replicate the finding that most postural variance falls within a reduced subspace. However, we show that dimensions outside of this subspace are highly structured and task dependent, suggesting they too are under volitional control. We propose that hand control occupies a higher dimensional space than previously considered.