30th Annual Meeting Program

Chair: Tamar Makin¹

Presenters: Jeff Lockman², Tessa Dekker¹, Asif Ghazanfar³, Daniela Vallentin⁴

Confirmed Discussant: Juan Alvaro Gallego⁵

¹University College London, ²Tulane University, ³Princeton University, ⁴Max Planck Institute for Ornithology, ⁵Imperial College London

The development of some of the most basic motor functions is often protracted. For example, learning to reach to a target begins in utero, accelerates over infancy and is refined across childhood. Which neural processes need to emerge and mature during the course of this prolonged learning? A better understanding of the neural basis of sensorimotor development may be important for understanding and improving motor control across the lifespan. Here we explore a set of fundamental behaviours with varying complexity across early-life development in babies, children, marmoset monkeys and songbirds. We first consider sensorimotor development in humans. We begin with one of the most elementary processes crucial for successful motor control: the development of human infants knowledge about their bodies layout. Jeff Lockman will ask how infants develop the ability to reach to a tactile target, uncovering the development of a sensorimotor reaching map of the body. We then explore sensorimotor integration in school-age children, this time for visually guided actions of the arm and body. Human adults are highly adept at accounting for imprecisions in their perceptual judgements and movements, allowing them to optimally choose successful actions across a wide range of context. Tessa Dekker will demonstrate, using a combination of model-driven neuroimaging and behavioural paradigms, that this is not a trivial ability to acquire – children as old as 10-11 years do not correctly account for the noise in their visual and motor systems during basic actions. Next, we turn to sensorimotor integration during the development of vocal control. Vocal development is a consequence of many interacting factors including the growth of the vocal apparatus, the muscles that innervate it, and the maturation of the nervous system that controls those muscles. With marmoset monkeys as a model system, Asif Ghazanfar will outline the complex interactions between the developing nervous system, developing biomechanics and the social interactions that adjust nervous system function via experience in generating the motor control of vocalisation. Finally, Daniela Vallentin will present key neural mechanisms underlying vocal interactions in songbirds, a motor behaviour which is learnt throughout development. Using intercellular recordings and pharmacological inactivation, Daniela will highlight the role of premotor inhibition in successful learning of socially-coordinated vocalisation. She will also introduce first findings from male nightingales, which learn over 100 different song motifs and are also able to spontaneously imitate a motif of another nightingale. By presenting key mechanisms from multiple disciplines in psychology and behavioural and systems neuroscience, the panel aims at providing a broad overview of current research on sensorimotor development and promoting understanding of commonalities and differences in processes of motor development across different species.

Chair: Devika Narain¹

Presenters: Devika Narain¹, Stephen Lisberger², Nuo Li³

Confirmed Discussant: John Krakauer⁴

¹Erasmus University Medical Center, ²Duke University, ³Baylor College of Medicine, ⁴Johns Hopkins University School of Medicine

The ability to anticipate, deliberate, and correctly time actions is central to motor and cognitive function. Neurons in the frontal, parietal, and motor cortices, which are believed to support such capacities, often exhibit persistent or ramping activity well before movement onset. Such preparatory activity assumes different roles and functions under various behavioral contexts. In this short panel, we will consider recent perspectives on the purpose, origin, and mechanisms underlying preparatory activity in three motor tasks, viz., smooth pursuit eye movements, directed motor planning, and motor timing. Further, we will discuss how preparatory activity subserves movement in these tasks in the light of contrasting theoretical perspectives. Finally, we will present evidence for how neural computations supported by preparatory activity are enabled by both cortical and subcortical influences. The panel shall attempt to integrate these perspectives across various animal models, theoretical views, and anatomical regions to uncover general principles governing preparatory activity associated with movement. First, Stephen Lisberger will provide us with an account of a new role for preparatory activity in enabling smooth pursuit eye movements in monkeys and its link to normative theories of sensorimotor control. In contrast to certain dynamical systems theories, Stephen will provide an alternative view for how preparatory activity in the monkey cortex can evolve without evoking premature eye movements. Second, Nuo Li will present a different perspective for anticipatory activity based on motor planning in rodents. He will elucidate the mechanisms by which cross-cortical and subcortical influences maintain preparatory activity and will link these insights to dynamical systems theory. The third speaker, Devika Narain, will discuss preparatory activity in light of motor timing using in vivo and in silico approaches. Using dynamical systems frameworks, she will examine how anticipatory activity observed in cortical and subcortical populations enables the control of movement timing. Overall, the panel aims to tackle the diversity of function, mechanisms, and theoretical perspectives surrounding preparatory activity for movements. In the discussion, we hope to identify disparities, constraints, and common ground among these views that could help provide general insights for motor preparation.

Sarah Wilterson¹, Andrew Wilterson¹, Carlos Valazquez¹, Samuel McDougle², Jordan Taylor¹

¹Princeton University, ²UC Berkeley

While much attention has been focused on understanding the processes underlying sensorimotor adaptation in highly practiced actions, relatively little attention has been paid to understanding how truly novel sensorimotor mappings are acquired. From the few recent studies that have focused on motor learning de novo, it appears that the learning processes in these tasks depart from those processes observed in adaptation tasks. The patterns associated with de novo learning closely reflect characteristics of reinforcement learning, as opposed to sensory-prediction-error learning processes. To better characterize how sensorimotor mappings are acquired de novo, we modified a task developed by Fermin and colleagues (2010). This task involved moving a virtual cursor across a grid to a goal position. Importantly, the mapping between the keys was arbitrary and unintuitive, defining a novel sensorimotor mapping. This task serves as a model system to study learning and adaptation of de novo sensorimotor mappings. The first goal of our study was to use behavioral manipulations and computational modeling to distinguish between model-free and model-based reinforcement learning processes. Two groups of participants trained to move a cursor between a paired start-end position. One group was provided with instructions – the shortest sequence of key. Following acquisition, both groups were tasked with moving the cursor between novel start-end position pairings. While the group trained with instructions was less able to generalize their learning to the novel start-end position pairings. This suggests that instruction blunted learning of the underlying sensorimotor mapping. In a follow-up experiment, we probed the acquisition time course of the sensorimotor mapping without any instruction. We found that with training, participants were able to generalize and perform progressively further future-state planning (i.e. the number of keypresses required to move from the start to the end position). We modeled these behavioral results using six increasingly complex reinforcement learning models: from simple state-action reinforcement to using the learned action-mapping for planning. We found that a combination of both model-free learning and model-based planning was necessary to account for learning and generalization of the new sensorimotor mapping. The second goal of our study was to show that the newly learned sensorimotor mapping can be adapted to a perturbed environment. To demonstrate this adaptation effect, we shifted the mapping of each finger to the finger immediately adjacent (e.g. the middle finger now does what was initially learned by the ring finger). Participants were able to adapt to this configuration more readily than they could a completely new, random mapping. These findings show that sensorimotor mappings learned de novo are adaptable shortly after acquisition and that adaptation processes are not limited to well-practiced, continuous movements.

Adam Rouse¹

¹University of Kansas Medical Center

While most studies of neurons in primary motor cortex during reaching have focused on their tuning to movement parameters like muscle activity or reach direction, it is known that neurons’ firings rates change relative to baseline for most movements in addition to being tuned to the particular direction of movement. This global change in firing rate for a neuron has been defined within certain tasks as speed-tuning with increased rates for faster movements while more recent studies from our lab and others, however, have described these global firing rate changes as condition-independent activity that occur with all movements. Additionally, rather than a rise and fall in activity time-locked for all neurons, the population activity has a more complex neural trajectory in the neural space with an ordered progression of some neurons leading and others lagging. These descriptions of condition-independent activity and consistent neural trajectories create a challenge for describing the neural encoding of movement beyond single movements when sensorimotor integration of feedback is required for high precision. It is unclear if and how these intrinsic neural dynamics in motor cortex influence the brain’s ability to integrate new sensory information and timing of corrective movements. To explore these questions, our lab had monkeys (Macaca mulatta) perform precision center-out reaching to small targets. In this precision task, the animals often needed to make additional corrective movements after an initial reach if they did not land precisely within the small target. For the reaches with subsequent movements after the initial reach, the movements could be divided into submovements by identifying when multiple speed peaks occurred during a trial. These submovements were observed to have similar bell-shaped velocity profiles as a function of time whether they were during the initial reach or subsequent corrective movements. The neural recordings showed a consistent, cyclic neural trajectory in certain dimensions of the neural space–identified with jPCA dimensionality reduction–that occurred during the initial reach and was repeated when the animal needed to make additional corrective submovements. The timing of peak movement speeds were phase-locked with these cyclic neural trajectories a better predictor of when speed peaks occurred than simply the instantaneous firing rate across the population. Finally, the average change in firing rate for many neurons was nearly as large for the repeated cycles during the smaller amplitude movements as large ones. This indicates the global change in firing rate is not simply a marker for movement speed. Rather it appears to represent a truly condition-independent increase of neural activity related to the execution of each submovement. Suggesting that the encoding of the initial and any corrective movements in this precision reaching task occurred as a series of discrete submovements.

Linda Solstrand Dahlberg¹, Benjamin De Leener¹, Ali Khatibi², Nawal Kinany³, Ovidiu Lungu¹, Shahab Vahdat⁴, Ella Gabitov¹, Veronique Marchand-Pauvert⁵, Julien Doyon¹

¹Montreal Neurological Institute, McGill University, ²University of Birmingham, ³École Polytechnique Fédérale de Lausanne, ⁴University of Florida, ⁵Sorbonne University INSERM

The recognition of the spinal cord as an important structure, not just for mediating sensory and motor signals, but also signal processing, have become more evident with the use of fMRI. Both animal and human studies show that blood-oxygen-level-dependent (BOLD) signal in the spinal cord is reliable, and is associated with activities involving the spinal cord. A previous study from our lab showed learning-induced plasticity in the spinal cord during a motor sequence learning task (MSL) that was found to be independent of contributions from the brain. At the level of the brain, MSL leads to altered resting state functional connectivity in sensorimotor regions that are correlated with task performance, suggesting processes linked to learning and consolidation of a newly learnt motor task may show traces in neural activity also during rest. However, it is not known whether such changes occur at the level of the spinal cord. Here, we present for the first time evidence of movement related adaptation changes during rest in the spinal cord following a motor learning task. Resting state fMRI activity of the cervical spinal cord was recorded in two sessions before, and one session after a MSL task. Resting state connectivity was examined using Regional Homogeneity (ReHo), which tests the synchronicity of a voxels’ time courses with that of its nearest neighbours, thus providing a measure of intrinsic local connectivity. This measure is represented by Kendall’s coefficient of concordance (KCC) values. A cluster of activation at the level of the C6-C7 was found in response to the MSL task, and was used as a region of interest to extract KCC values from each of the three resting state sessions (RS1, RS2 and RS3). KCC values from a region two segments above, and another region two segments below the MSL cluster were also extracted as control. A paired t-tests revealed a significant difference between RS3 and RS2, as well as between RS3 and RS1, but not between RS2 and RS1. A correlation analysis found a significant relationship between the difference in KCC-values in RS3 and RS2 (RS3 – RS2) with MSL performance duration (r=-0.647; p=0.004), indicating that faster performers shows greater increase in synchronicity of activity during resting state in the region that was found to be active during the MSL task. For the first time, we provide evidence that motor learning related changes go beyond the level of the brain, and are evident even at the level of the spinal cord. During rest, intrinsic neural networks in the spinal cord are changed following motor sequence learning. Moreover, the changes are linked to performance of the task, with fast performers showing more coherent local connectivity. This finding challenges our assumptions of the role of the spinal cord, arguing for a more active function than merely a medium between the brain and the body, which may even contribute to learning and acquisition of a new motor skill.

Najja Marshall¹, Joshua Glaser¹, Sean Perkins¹, Larry Abbott¹, John Cunningham¹, Mark Churchland¹

¹Columbia University

Voluntary movement requires communication from cortex to the spinal cord, where motor units (MUs) relay neural drive to muscles. The canonical description of MU control rests upon two foundational tenets. First, cortex cannot control MUs independently; rather, it supplies a shared one-dimensional ‘common drive’ to each muscle’s MU pool. Second, MU firing rates are sigmoidal functions of that common drive. The point where each function rises is determined by MU ‘size’ (number of muscle fibers activated by the motor neuron). Thus, MUs are recruited in a consistent order as force rises. These tenets comprise the decades-old ‘size principle’: MU activities are a prespecified function of a one-dimensional descending command. We re-examined motor unit recruitment using a novel task, multi-channel cortical microstimulation, and intramuscular recordings that isolated multiple single MUs simultaneously. We trained rhesus macaques to generate a variety of force profiles (steps, ramps, sinusoids, and chirps) via isometric muscle contractions. In one set of experiments, forces varied over multiple frequencies (0-3 Hz). In a second set of experiments, forces were generated with the arm in two postures. In a third set of experiments, cortical stimulation during task performance induced small transient increases in force. The activity of multiple nearby single MUs was recorded using eight modified percutaneous electrodes and a custom EMG decomposition algorithm that leverages Bayesian nonparametrics and optimal filtering. The size principle predicts that MU activities are prespecified functions of a 1-D descending command. Thus, when plotting the firing rate of one MU against another, activity should lie on a 1-D manifold. Yet this was not the case during either natural force production or stimulation-induced forces. MU activity lay on the same manifold for static and slowly changing forces, but departed from the manifold during rapidly changing (> 1 Hz) forces. MU activity also lay on different manifolds when a force was generated from the two postures. Finally, cortical stimulation delivered through different electrode channels could recruit MUs independently of one another, such that their activity lay on a distinctive manifold for each stimulation site. We developed a probabilistic latent factor model to assess whether empirical MU activity could be accounted for by a single latent factor, with each MU having a unique, monotonically increasing link function. To infer the most likely distribution of latent variables given the data, and to learn the link functions and other parameters, we used black-box variational inference. We found that single-factor models were insufficient during both natural performance and artificial stimulation. Instead, multiple factors were needed to account for the diversity of MU responses. Thus, contrary to classical predictions, MUs are flexibly controlled to meet task demands.

Supriyo Choudhury¹, Ravi Singh¹, A Shobhana¹, Dwaipayan Sen¹, Sidharth Anand¹, Shantanu Shubham¹, Suparna Gangopadhyay¹, Mark Baker², Hrishikesh Kumar¹, Stuart Baker²

¹Institute of Neurosciences Kolkata, ²Newcastle University

Background: In monkey, reticulospinal connections to hand and forearm muscles are spontaneously strengthened following corticospinal lesion, likely contributing to recovery of function. In healthy humans, pairing auditory clicks with electrical stimulation of a muscle induces plastic changes in motor pathways (probably including the reticulospinal tract), with features reminiscent of spike-timing dependent plasticity. In this study, we tested whether pairing clicks with muscle stimulation could improve hand function in chronic stroke survivors. Methods: Clicks were delivered via a miniature earpiece; electrical stimuli at motor threshold targeted forearm extensor muscles. A wearable electronic device (WD) allowed patients to receive stimulation at home while performing normal daily activities. Ninety-five patients >6 months post-stroke were randomised to three groups: WD with shock paired 12 ms before click; WD with clicks and shocks delivered independently; standard care. Those allocated to the device used it for at least 4 hours per day, every day for 4 weeks. Upper limb function was assessed at baseline, and weeks 2, 4 and 8, using the Action Research Arm Test (ARAT) which has four sub-domains (Grasp, Grip, Pinch and Gross). Results: Severity across the three groups was comparable at baseline. Only the Paired Stimulation Group showed significant improvement in total ARAT (median baseline: 7.5; week 8: 11.5; p=0.019) and the Grasp sub-score (median baseline: 1, week 8: 4; p=0.004). The total number of paired stimulations received over the four weeks of device usage was significantly correlated with the change in ARAT score at visit 3 (Spearman’s ñ=0.53, p=0.013) but not at visit 4 (ñ=0.285, p=0.223). Conclusion: A wearable device delivering paired clicks and shocks over four weeks can produce a small but significant improvement in upper limb function in stroke survivors. Trial registration – CTRI/2018/03/012628.

Marta Russo¹, Dagmar Sternad¹

¹Northeastern University

Numerous studies have shown that postural balance is improved through light touch of a stable surface. Less attention has been paid to the support by an unstable system like a cane, which is often used by frail individuals to reduce the risk of falling. Further, little is known about how another individual provides support via touch, as in physical therapy or dancing. We developed a set of experimental paradigms to examine the effect of mechanical and human support on balance. A first study focused on the role of canes in assisting balance when standing on uneven terrain. Participants (n=16) were instructed to stand on a narrow beam placed on a force platform, supporting themselves by two instrumented canes, one in each hand. Applying different forces on the canes, light touch proved superior in decreasing the sway; increasing the exerted force yielded no further benefit, in fact had a destabilizing effect. The displacement of the hand on the cane handle was proportional to the variability of the sway and the force applied, consistent with the well-known mechanical effect that pushing destabilizes an inverted pendulum. A second study compared postural sway in professional ballet dancers and participants without balance training (n=28). They were asked to balance on the same narrow beam while holding the handle of a robotic arm that provided 3 levels of increasing stability simulating canes of different lengths; for control, the robotic arm could also be stable or fully compliant. Interaction force was measured by the robot together with postural sway, whole-body kinematics and EMG activity. As expected, increased stability of the support decreased the sway; less expected was that ballet dancers displayed larger sway than non-trained participants, indicating that sway may not be an ideal metric for balance ability. Remarkably, even when participants only held canes above ground or when the robotic arm was compliant, postural sway was reduced. In a third experiment, a second person, also standing on a narrow beam, held the same robot handle to provide support to the subject. Novice participants significantly improved their balance when supported by another novice, whereas the sway of professional dancers was minimally affected by the partner. Ongoing analyses examine the nature of this active information transmission in its relation to postural sway. In summary, different from novices, professional ballet dancers did not take advantage of the support provided either by a mechanical device or another individual. However, we cannot claim that they are less stable than non-trained individuals. This unexpected result suggests that the center of pressure, gold-standard measure of postural stability, only reflects effective balance and may not be sufficient to describe the mechanisms engaged in postural control. Further analysis on the direction of the ground-on-foot force may better characterize the observed behavior.

Chair: Vivian Weerdesteyn¹

Presenters: Vivian Weerdesteyn¹, Brian Corneil², Stuart Baker³, Claire Honeycutt⁴

Confirmed Discussant: Timothy Carroll⁵

¹Donders Institute, Radboud University Medical Centre, ²University of Western Ontario, ³Newcastle University, ⁴Arizona State University, ⁵University of Queensland

Traditionally, the reticulospinal pathway is considered to be involved with the control of posture and locomotion. In recent years, however, it has become increasingly evident that this pathway may also contribute to the control of upper and lower extremity movements. Not only have these insights fundamentally challenged our view of the neural control of movement, they have also led to new hypotheses on the potential utility of this pathway to compensate for loss of corticospinal motor control (e.g. following stroke). Yet, many questions still remain on the organization and expression of reticulospinal motor control in health and disease. This panel with researchers from both fundamental and clinical neuroscience will take a step forward in addressing these questions. Specifically, they will present recent experimental evidence from studies in humans and monkeys with and without corticospinal lesions showing that 1) the reticulospinal pathway is implicated in a fast visuomotor network that recruits ultrafast and directionally-appropriate muscle responses to visual stimuli, 2) startling stimuli interacting with this network have the capacity to expedite the ensuing goal-directed movements, and 3) various interventions exist that may harness compensatory reticulospinal motor control following stroke. In this panel, Brian Corneil will discuss how visual stimuli induce directionally-tuned visuomotor responses at latencies of 80-120 ms in neck and upper limb muscles in non-human primates and humans (including people with Parkinson’s disease). He will show how the characteristics of these responses provide evidence for their recruitment through a phylogenetically-conserved fast visuomotor network, involving superior colliculus and the tectoreticulospinal pathway. Vivian Weerdesteyn will discuss how results from studies in healthy individuals and in people with neurological diseases have provided evidence for the involvement of the reticulospinal pathway in startle-induced shortening of reaction times of voluntary movements. She will present new results that show how interactions between startles and the fast visuomotor network may explain previous disparaties in choice reaction tasks. Stuart Baker will discuss his work in non-human primates and humans on plasticity in the reticulospinal system. Reticulospinal connections are strengthened during recovery from corticospinal lesions, but may also be modified by interventions in healthy individuals. Such protocols could be exploited to enhance recovery. Claire Honeycutt will discuss the implication of the reticulospinal system in motor learning and present findings that show how startles can be used for probing the degree of motor learning in young and older individuals. She will also present exciting new findings that show how startles can be used for harnessing the potential of the reticulospinal system for improving motor function after stroke.

Chair: Marco Capogrosso¹

Presenters: Marco Capogrosso¹, Elvira Pirondini², Monica Perez³, John Krakauer⁴

Confirmed Discussant: Solaiman Shokur⁵

¹University of Pittsburgh, ²University of Geneva, ³Northwestern Univeresity, ⁴Johns Hopkins University, ⁵EPFL

Decades of basic research in animal models and in humans has provided evidence that changes in the CNS can lead to behavioral improvements after damage to the motor system after either stroke or spinal cord injury. However, despite this knowledge, we still lack a therapeutic approach that achieves much beyond what can be expected from spontaneous biological recovery. As a result, most rehabilitation approaches emphasize compensatory strategies rather than true restoration. In parallel exciting advancements in neuro-technologies for animal and human research offered new tools to investigate and manipulate the neural activity of specific populations of neurons across the entire nervous system. In consequence, against this somewhat pessimistic backdrop, we will suggest that recent results from new studies in humans and non-human primates using these tools and combining intense behavioral training and electrical stimulation provide reason for optimism. Dr. Elvira Pirondini will open the panel with discussing with a detailed investigation of the neural changes that occur both at single cell and a neural population level in non-human primate undergoing intense training activity after a sub-cortical stroke. She will show with a combination of classic and modern neural population analysis, kinematic data and non-invasive neuroimaging, what are the relevant changes that can be observed in a clinically relevant animal model after a lesion severely impacting the dexterity of the hand. Dr. Marco Capogrosso, will then present his work on electrical stimulation of the cervical spinal cord in monkeys. Specifically, he will discuss how this methodology can influence the behavior of spinal motor circuits controlling the arm and hand and the importance of voluntary interaction with neuromodulation technologies to improve motor control. Dr. Monica Perez, will then shift towards human applications, showing that electrical stimulation can be used to tape into ancestral mechanisms of synaptic plasticity to boost motor recovery in people with spinal cord injury. Finally, Dr. John Krakauer will close the panel, by summarizing the evidence accumulated so far and discussing the very concept of recovery, compensation and plasticity after injury. The open discussion will revolve around an important question: are we ready now to design technologies that can really improve the life of people with motor deficits?

Chair: Brian Horslen¹

Presenters: Brian Horslen¹, Scott Albert², Stella Koutsikou³, Friedl De Groote⁴

¹Emory University, ²Johns Hopkins University, ³University of Kent, ⁴KU Leuven

Not moving is just as important for successful motor control as moving. Maintaining a static posture, where limb or whole-body orientation is held constant, is an active sensorimotor task that has long been used as a paradigm to study basic motor control principles. However, recent research demonstrates there are key differences in neural control, body biomechanics and sensory state that distinguish static postural control from control of dynamic movement. Furthermore, in real-world behavior, people and animals fluidly change between static and dynamic states but the systems and conditions that drive these changes are not always clear. This panel will compare static and dynamic motor control, exploring how postural control can reveal basic tenets of motor control and how neural control, biomechanics, as well as sensory processing and state may differ between tasks to either promote or suppress movement. Friedl De Groote will discuss how complex interactions between neuromechanical principles, such as muscle tone, stiffness and stretch reflexes, can be explored in a “simple” static behavior and then extrapolated to more complex dynamic tasks. Using a blended experimental and computational approach, Dr. De Groote found that joint torques during both reactive standing balance and perturbed walking could be reproduced by center of mass feedback. This suggests as a similar control scheme, with subtle differences in gains, can account for center of mass control in both static and dynamic behaviors. Scott Albert will present a model of upper limb postural control where holding the arm still at a target depends on the movement commands that sent the arm to the target, not the spatial location of the target. Drawing from human and non-human primate behavioral evidence, he will describe a mathematical coupling between reaching and holding commands that persists when the cortex is damaged by stroke. Brian Horslen will explore how changing muscle mechanics cause muscle spindles to operate in high- or low-sensitivity states in non-moving and moving conditions. In a static behavior, like standing in-place, muscle spindles operate in a high-sensitivity state, but increasing movement, making the behavior more dynamic, decreases sensitivity due to reduced muscle stiffness. These data suggest a movement-dependent interaction between muscle mechanics and muscle spindle sensitivity where the amount of sensory information the nervous system receives depends on how much movement occurs prior to stretch. Stella Koutsikou will discuss how and when the nervous system decides to break out of holding still and initiate movement. Drawing evidence from rat and Xenopus laevis tadpole electrophysiological and behavioral experiments, Dr. Koutsikou will discuss how the central nervous system modulates sensory processing of different modalities to either promote freezing or flight-or-fight behaviors, as well as how sensory build-up leads to initiation of locomotion.

David Xing¹, Wilson Truccolo¹, David Borton¹

¹Brown University

Our ability to directly modulate our gait while walking with precise, voluntary adjustments is what allows us to navigate challenging and complex terrains. However, how our nervous system generates the command signals to control movement during unattended locomotion, precise voluntary movement, and the transition in between, is poorly understood. Previous studies have shown that rodents and felines are able to perform basic walking even after motor cortex is shut down, but their ability to reach with their limbs or avoid obstacles is abolished. In non-human primates, lesions of the corticospinal tract led to greater impairment, and slower recovery, of stationary volitional movements compared to unattended locomotion. These prior findings suggest that motor cortex may be differentially engaged in these two types of behaviors. We wanted to investigate whether the relationship between activity patterns in primary motor cortex (M1) and the generated movement is similar across both voluntary movements and locomotion. To directly compare the cortical activity during the two movement modalities, we recorded from M1 in rhesus macaques while they performed locomotion on a treadmill as well as stepping over an incoming obstacle from a stationary, standing position, and stepping over the obstacle while walking on the treadmill. We found large reorganization of the same recorded neural population both in terms of the correlation structure, as well as gait phase tuning and depth of modulation during voluntary control compared to treadmill walking. Additionally, neurons whose spiking activity was well correlated with movement kinematics during obstacle avoidance did not necessarily maintain this correlation during walking. Using a Wiener filter decoder, we were able to accurately decode the kinematics from the neural activity for both types of movements separately, but performance was poor when attempting to generalize the stationary obstacle avoidance decoder to walking or vice-versa. Furthermore, we were able to clearly separate the population activity between the two behaviors using LDA. During obstacle avoidance while walking, the neural activity smoothly transitioned from the region occupied during walking to the region occupied during stationary obstacle avoidance in the LDA dimension. Despite this, using dPCA, we were able to find a low-dimensional neural manifold that was preserved between the two behaviors. The top three components of this preserved manifold accounted for >50% of the total variance in the full-population, and was well separated from the components that discriminated between behaviors. Additionally, the neural activity within this manifold exhibited rotational structure. These findings demonstrate that different patterns of M1 activity are engaged during voluntary movements compared to autonomous locomotion, in agreement with previous M1 inactivation studies, although certain population-level features, such as rotations, are preserved.

Tianwei Wang¹, Yiheng Zhang¹, He Cui¹

¹Institute of Neuroscience, CAS

Whereas motor cortex has been found to encode the kinetics and kinematics of arm movements, its neuronal activity is also influenced by serial actions. Accumulating evidence suggests that movement generation might be achieved via dynamical evolution of neural populations in motor cortex, but it is unclear whether the neural dynamics underlies ballistic movements only, or is involved in more intricate movements, such as sequential reach. Previous studies often presented stimuli sequentially, with intervening delays between motor elements, which might reflect sensory sequence information and split movement generation. In the present study, we recorded neuronal activity in motor cortex while monkeys performed a coherent sequential arm reach task. There were three types of trials: center-out reach, clockwise double reach, and counterclockwise double reach. During the double reach trials, two targets of different shapes were presented simultaneously during a cue period, and after a 500-700 ms memory period, the monkey was required to reach memorized locations in a certain sequence without a compulsive delay between the two reaches. Neuronal activity was recorded from three monkeys, one with a 96-channel Utah array in the anterior bank of the central sulcus, and the others with a single electrode. We separated trials into clusters according to first movement direction, then studied the subsequent movements’ effects on preparatory and execution activity. In contrast to previous findings from the parietal reach region (Baldauf et al. J. Neurosci. 2008) and dorsal parietal area 5 (Li and Cui J. Neurosci. 2013), neurons in motor cortex exhibited heterogeneous firing patterns related to reaching sequences. Many neurons encoded first reaching direction during preparatory and premovement periods, but they were modulated by subsequent movement. In detail, 52% of the 322 sequence recorded neurons and 70% of 74 array recorded neurons showed this modulation in preparatory activity. Unlike some previous studies in which many neurons exhibited selectivity for particular orders of sequentially presented stimuli (i.e., Shanechi et al. Nat Neurosci 2012), most neurons carried information regarding both reaches when stimuli were presented simultaneously. To further illustrate this sequential modulation and its temporal dynamics through motor preparation and execution, we applied principal component analysis (PCA) and linear discriminant analysis (LDA) and found that the initial states for single and sequential movements differed. Furthermore, demixed principal component analysis (dPCA, Kobak et al. eLife 2016) suggests the existence of sequence-related factors, which evolve from the preparatory period through the execution period. In the absence of sequential sensory stimuli, motor cortex still encoded coherent double reach with sequence-related neural states during the preparatory period, suggesting a neural population dynamics underlying complex motor planning.

Kevin Cross¹, Douglas Cook¹, Stephen Scott¹

¹Queen’s University

When moving and interacting in a complex environment, the brain must process and integrate multiple sources of sensory feedback to plan and guide motor actions. During a goal-directed reach, vision provides feedback of the limb and goal, however, how these feedback sources are integrated is unclear. One theory suggests a difference vector is calculated by posterior parietal cortex (PPC) between the limb and goal. The difference vector is then projected to primary motor cortex (M1), which generates motor commands (i.e. control policy) to guide the limb to the goal. This hypothesis predicts that M1 neurons responsive to visual feedback about the limb will also be sensitive to visual feedback about the goal. We explored this hypothesis by recording from M1 while monkeys reached to a goal with/without visual feedback of the hand represented by a cursor. On random trials, we probed feedback responses by applying a jump of the cursor’s position or a jump of the goal’s position. We found 73% of neurons were sensitive to cursor and goal jumps, which was greater than expected by chance (χ2(1,226)=112). There was a strong correlation between a neuron’s activity gains for the cursor and goal jumps (r=0.92). We also found the activity for the cursor and goal jumps resided in the same neural subspace. Our results support the hypothesis that the motor system calculates a difference vector. We conducted a second experiment exploring how information conveyed by different sensory modalities are integrated. A common hypothesis is that proprioceptive and visual feedback are integrated together to estimate the position and motion of the limb in PPC (i.e. state estimation), which is then projected to M1. This hypothesis suggests that M1 neurons responsive to proprioceptive feedback of the limb will also be responsive to visual feedback of the limb, as they arise from the same input to M1. However, M1 receives input from several distinct feedback loops including both parietal and primary somatosensory cortex. M1 may receive proprioceptive and visual feedback independently, and thus, the prediction is that M1 neurons may be selectively responsive to proprioceptive or visual feedback. We explored these hypotheses by having monkeys perform the same reaching task as above. On random trials, we probed feedback responses by applying a mechanical load or a jump of the cursor’s position. We found 56% of neurons were sensitive to mechanical loads and cursor jumps, which was greater than expected by chance (χ2(1,226)=53). There was a strong correlation between a neuron’s activity gains for the mechanical loads and cursor jumps (r=0.77). We also found the activity for the mechanical loads resided in a neural subspace that overlapped by >50% with the subspace in which the cursor jump activity resided. Our results best align with the hypothesis that proprioceptive and visual feedback are integrated by upstream cortical areas such as PPC.

Macauley Breault¹, Jorge González-Martínez², Sridevi Sarma¹

¹Johns Hopkins University, ²University of Pittsburgh

Neuroscientists are faced with a challenge when studying humans because humans, in particular, can dwell on the past which can affect present and future behaviors. For example, a basketball player may miss a routine free throw if they are upset over an earlier foul. Variability is often observed across sequentially cued movements because past outcomes alter a subject’s internal states (i.e., confidence, emotion, or motivation) which influence the way the subject moves. The effects of internal states on behavior tend to be overlooked, as they are not directly measurable and, therefore, difficult to connect back to neural processes. In our study, we attempt to investigate these factors by utilizing state-space models to estimate the internal states from measured behavioral data, and then we map these states to neural activity. We exploit a unique paradigm in which 9 human subjects, implanted with depth electrodes, performed a goal-directed center-out reaching task designed to induce changes in internal states with instructed speeds and random perturbations. For example, a subject attempting to reach a specific target at a specific speed may be forced to stop and reevaluate their trajectory when faced with a perturbation, thus negatively affecting their performance and perhaps their confidence. Neural recordings were collected using StereoElectroEncephaloGraphy (SEEG) technique to obtain Local Field Potential (LFP) activity from nonmotor regions, including limbic and cognitive structures. This makes these data particularly useful as we believe it is here where internal states may be encoded, as opposed to the motor cortex. In this talk, we present how we were able to estimate internal states to (1) predict variability in reaction time and (2) identify brain regions associated with internal states. We identified four state variables that accumulate past evidence on errors, perturbations, and performance and that predict highly variable reaction times in all 9 subjects. This demonstrates the utility of the state-space modeling framework to estimate latent factors that explain variability from measurable data. When we mapped the neural activity represented in the spectral domain to the four internal states, we found regions related to memory and attention. This supports the notion that internal states capture cognitive processes encoded in the brain that ultimately influence behavior.

Sara Hussain¹, Mary Vollmer¹, Jessica Stimely¹, Gina Norato¹, Christoph Zrenner², Ulf Ziemann², Ethan Buch¹, Leonardo Cohen¹

¹National Institutes of Health, ²University of Tübingen

The ability to learn new motor skills and store them in memory is a fundamental feature of human behavior. It has been proposed that the neural mechanisms supporting memory may be rhythmic, such that they vary with oscillatory phase in task-relevant brain regions. However, direct and causal evidence for rhythmic memory processing is lacking. Here, we tested this possibility in the context of motor skill consolidation by delivering closed-loop transcranial magnetic stimulation (TMS) to the primary motor cortex (M1), a crucial node within the motor learning network, immediately after healthy adults learned a new motor skill. On Day 1, all participants practiced an explicit motor sequence learning task in which they repeatedly typed a numerical sequence (4-1-3-2-4) as quickly and as accurately as possible using their non-dominant left hand. After practice, when consolidation is known to occur, active closed-loop TMS was delivered to the contralateral M1 during either mu peak (active-peak group, N=17) or trough (active-trough group, N=17) phases. In a third group, sham closed-loop TMS was delivered to contralateral M1 during random mu phases (sham-random group, N=16). All participants were tested on the skill the next day to evaluate consolidation. We first confirmed that the real-time EEG analysis algorithm used to deliver closed-loop TMS accurately targeted mu peak, trough, and random phases. Indeed, oscillatory phases clustered near 90° in the peak-active group (p<0.001) and near 270° in the trough-active group (p<0.001), while phases were uniformly distributed in the random-sham group (p=0.652). Next, we evaluated differences in skill acquisition and consolidation across groups, measured as changes in the number of correct sequences typed per trial. Although skill acquisition was comparable (p>0.56 for all), overnight skill consolidation differed significantly between groups. Specifically, consolidation was enhanced in the trough-active group relative to the peak-active (p=0.003) and random-sham groups (p=0.001), while it was similar in the peak-active and random-sham groups (p=1.0). The consolidation enhancement was driven by improvements in speed (trough-active vs. peak-active, p=0.003; trough-active vs. random-sham, p=0.001; peak-active vs. random-sham, p=1.0) rather than accuracy (p>0.272 for all). Our results document oscillatory phase-dependent enhancement of human skill memory using closed-loop TMS, establishing the causal contribution of M1 activity during sensorimotor oscillatory trough but not peak phases to skill learning. We conclude that the neural mechanisms supporting skill consolidation fluctuate rhythmically and coherently with the sensorimotor mu rhythm, such that consolidation occurs preferentially during brief windows of opportunity corresponding to optimal oscillatory phases.

Jung Uk Kang¹, Eric Mooshagian¹, Lawrence Snyder¹

¹Washington University School of Medicine

Bimanual coordination is essential to activities such as tying shoelaces or playing a musical instrument. It likely requires communication between brain areas on each side of the brain. Yet, the role of interhemispheric communication in bimanual coordination is poorly understood. The Parietal Reach Region (PRR) encodes early planning of contralateral arm movements. The corpus callosum is the most direct path between the cerebral hemispheres. To test whether the callosum helps support bimanual coordination, we compared neural activity and behavioral performance before and during reversible blockade of the callosal pathways connecting left and right PRR. We first identified white-matter pathways between left and right PRR using in vivo manganese-enhanced magnetic resonance. We found that axons connecting left and right PRR across the callosum were restricted to the splenium. We then reversibly blocked this pathway using focal lidocaine injections into the callosum while an animal planned and then executed unimanual movements to a single target or bimanual movements to either a single target or two different targets. We measured interhemispheric LFP-LFP coherence to get an estimate of interhemispheric communication between left and right PRR, before and during the blockade. Before blockade, we observed three different levels of interhemispheric LFP-LFP coherence during the planning period in the 20-36 Hz frequency range. Coherence was high when an animal planned bimanual movements to a single target; intermediate when an animal planned unimanual movements to a single target; and low when an animal planned bimanual movements to two different targets. There were no effects at other bands. This task- and frequency-dependent modulation of interhemispheric LFP-LFP coherence is consistent with the hypothesis that bimanual coordination involves communication between right and left PRR. During blockade, the difference in LFP-LFP coherence amongst different movement types was abolished. This indicates that the communication takes place via the callosum. Additional findings regarding interhemispheric spike-LFP coherence also support this conclusion. Behaviorally, we observed that movements of the two arms were less synchronous during callosal blockade, consistent with our hypothesis that information carried by callosal PRR pathways supports bimanual coordination. We also observed faster reaction times in both unimanual and bimanual movements during blockade. This suggests that callosal connections facilitate bimanual coordination via a competitive process such that the faster arm slows down to match the slower arm. In conclusion, interhemispheric communication between left and right PRR via the posterior corpus callosum supports sharing of information about the movement plan of each arm in service of bimanual coordination.

Chair: Wolf-Julian Neumann¹

Presenters: Wolf-Julian Neumann¹, Andreea Bostan², Andrew Sharott³, Roxanne Lofredi¹

Confirmed Discussant: Robert Turner²

¹Charité – Universitätsmedizin Berlin, ²University of Pittsburgh, ³University of Oxford

What are the functions of the basal ganglia? Thirty years ago, the answer to this question seemed to be within reach, but lasting efforts in search for a unifying framework were of no avail. Indeed, behavioral correlates of basal ganglia activity seem as diverse as the ever-increasing complexity of their anatomical and molecular circuit characteristics. The basal ganglia network is uniquely positioned to integrate widespread cortical and subcortical information and distribute the result of these computations to a similarly diverse array of cortical and subcortical outputs. Whatever the function of the basal ganglia, it is likely embedded in the ability of this web of long-range synaptic connections to shape motor control and learning across the brain. Conversely, the power of these network level connections can be observed in the synchronous oscillations across cortex, basal ganglia and thalamus that are pathologically altered in basal ganglia- based movement disorders. The present panel reviews recent advances in pathway-specific functional anatomy and physiological mechanisms of basal ganglia dependent motor control. It aims to integrate rodent, non-human primate and human clinical research spanning a variety of research methods including transsynaptic tracing, optogenetics, invasive electrophysiology, neuroimaging and computational modelling. All of the presented studies focus on the synaptic and/or physiological connectivity across the basal ganglia circuit and the resulting implications for motor learning and kinematic control. Research highlights include the overlap of basal ganglia and cerebellar circuits in non-human primates with neuromodulation induced changes in human trial to trial motor improvement, optogenetic manipulation of the basal ganglia receiving thalamus for within day and between day task performance gains and dopamine dependent vigor signals reflected in temporal dynamics in human subthalamic beta and gamma bursts. The presented results serve as case studies to challenge the resilience of influential basal ganglia theories such as the vigorous tutor paradigm, habit formation, reward prediction error signals and energy cost discounting. Our studies suggest that basal ganglia computations result in synaptic modulation of distributed motor networks for invigoration and consolidation with input and output feedback loops prevailing at each stage of the circuit. Dopamine dependent beta and gamma oscillations may subserve communication in these distributed neural populations, resulting in modulation of excitability and vulnerability for synaptic potentiation. In summary, our panel highlights the importance of circuit-level computations for understanding basal ganglia function and dysfunction. As the cortico-basal ganglia circuit increasingly becomes the target for feedback-dependent, adaptive stimulation strategies, such insights have potential to inform and optimize neurotechnology-based treatments for movement disorders.

Chair: Britton Sauerbrei¹

Presenters: Britton Sauerbrei¹, Yifat Prut², Irina Beloozerova³, Robert Turner⁴

Confirmed Discussant: Hansjoerg Scherberger⁵

¹Janelia Research Campus, HHMI, ²The Hebrew University, ³Georgia Tech, ⁴University of Pittsburgh, ⁵German Primate Center

Primary motor cortex controls skilled movements of the limbs, such as reaching, grasping, and visually-guided locomotion over uneven surfaces. Patterns of activity in motor cortex generate movement through spinal and brainstem centers, and these patterns are strongly influenced by the motor thalamus, which in turn receives input from the cerebellum and basal ganglia. Recently, an integrative view of cerebellum-thalamocortical and basal-ganglia-thalamocortical function has begun to emerge from the use of simultaneous recordings in multiple brain regions, of electrical, optogenetic, or pharmacological perturbations in conjunction with recording in downstream areas, and of computational modeling. This panel will focus on how ascending projections from subcortical regions influence thalamic activity, motor cortical dynamics, and, ultimately, movement. Yifat Prut examines neurons in motor cortex that receive input from the cerebellum-thalamocortical pathway during a reaching task in macaques. Using microstimulation of the superior cerebellar peduncle, pharmacology, and waveform criteria, she identifies a population of cerebellar-recipient inhibitory cortical interneurons. This population has weaker directional tuning than pyramidal cells, but is active earlier relative to movement onset. This finding suggests that feedforward inhibition may be a key mechanism by which cerebellar output influences the motor cortex. Robert Turner studies the role of the inhibitory projection from basal ganglia (GPi) to the motor thalamus (VLa) during reaching in macaques. Contrary to the predictions of current theories, VLa is active earlier than GPi, and on average, firing rates in the two areas exhibit increases, rather than reciprocal modulation. Furthermore, simultaneous recordings in the two regions rarely reveal strong correlations in spike timing or trial-to-trial fluctuations. Thus, the GPi-recipient thalamus may be influenced more strongly by inputs outside the basal ganglia, such as cortical feedback. Irina Beloozerova examines the role of thalamic inputs to motor cortex in the cat during locomotion on a flat surface and on a horizontal ladder. She shows that pharmacological inactivation of motor thalamus during locomotion in the cat alters the firing patterns in motor cortex, and observes differential effects on cortical neurons with different receptive fields and conduction velocities. Finally, Britton Sauerbrei studies the contribution of thalamic inputs to motor cortical function during a reaching task in mice. He shows that optogenetic silencing or stimulation of thalamocortical projections disrupts cortical dynamics and arm kinematics, and suggests that cortical pattern generation during movement execution requires temporally-patterned thalamic input. Taken together, the results in this panel highlight the motor thalamus as a critical hub for limb control.

Giacomo Ariani¹, Young Han Kwon¹, Jörn Diedrichsen¹

¹Western University

Movement repetition is an essential component of long-term skill acquisition. However, repetition has also been shown to have immediate benefits on performance, increasing the speed and accuracy of a second execution. Such effects have been reported even for overlearned, simple behaviors such as point-to-point arm reaching. In two tightly related streams of research, we studied repetition effects in the context of complex motor sequences, thereby gaining novel insight into motor planning processes during sequence production. In the first series of behavioral experiments, we used a discrete sequence production (DSP) task in which human volunteers had to perform short sequences of finger movements. In Exp. 1 (N=49), participants were presented with random sequences and we manipulated whether they had to execute the sequence (Go), or not (No-Go), and whether the sequence was repeated on the next trial (Repetition), or not (Switch). We established that repeating a sequence of finger movements led to immediate improvements in speed without associated accuracy costs. The biggest benefit was observed in the middle part of a sequence, suggesting that the repetition effects likely resulted from facilitated online planning – the ability to plan upcoming elements while executing current ones. This claim was further supported by Exp. 2 (N=40), in which we kept a set of sequences fixed allowing participants to develop sequence-specific knowledge: once learning reduced the need for online planning, the benefit of repetition disappeared. Finally, we found that repetition-related improvements only occurred for the trials that had been preceded by sequence production, suggesting that recent movement experience may be required to reap the benefits of repetition: actual practice might be more beneficial to the human sensorimotor system than mental rehearsal for producing short-term performance improvements. In the second, currently ongoing, stream of this research, we are investigating whether these findings generalize beyond discrete sequences of finger movements to sequences of upper limb reaching movements. While finger movements are essentially non-inertial and can be produced discretely, control of the arm is highly inertial, and transitions between reaching movements can be either discrete, or different movement elements can be concatenated into a single, complex arm movement. Thus, the investigation of arm movements sequences provides a useful window into sequencing, and allows for an explicit test of whether the processes of online planning, and the benefit of repetition, qualitatively differ between sequences of discrete movement elements, as compared to the continuous production of a complex arm trajectory. Together, our results will shed light on existing theories of sequence production across tasks and effectors, highlighting the importance of practice for enhancing our ability to link individual sequence elements into skilled sequential behavior.

Benjamin Parrell¹, Caroline Niziolek¹

¹University of Wisconsin-Madison

Although movement variability is often attributed to unwanted noise in the motor system, recent work has demonstrated that variability may be actively controlled. Variability is minimized along task-relevant dimensions, but permitted in less-relevant dimensions of control (uncontrolled manifold hypothesis [UCM], optimal feedback control [OFC]). Conversely, motor learning can selectively increase task-relevant variability, potentially to facilitate future learning (Wu et al. 2014). Variability can also be reduced when needed: participants exposed to a visual perturbation that magnified the horizontal displacement of the hand away from the midline during point-to-point reaching movements reduced their variability in this dimension (Wong et al. 2009). To date, research on regulation of motor variability has relied on relatively simple, laboratory-specific reaching tasks. It is not clear if and how these results translate to complex, well-practiced, real-world tasks or to actions controlled via non-visual sensory feedback. Here, we test how variability is regulated during speech production, a highly over-practiced motor behavior that relies on auditory and somatosensory feedback. Spoken vowels are produced in a two-dimensional space defined by the two lowest resonances of the vocal tract, known as the first (F1) and second (F2) formants. Speakers produce vowels in a distribution around individual-specific targets in this F1/F2 space. In two groups, we alter participants’ speech in real time, perturbing vowel formants either toward or away from these targets (defined as the center of the distribution for a particular vowel on a participant-specific basis). By scaling the error between the vowel formants and their targets, these inward- and outward-pushing perturbations respectively reduce and magnify the perceived variability of speech production. We test how these changes affect variability in produced vowel formants during the perturbation as well as after the perturbation is removed. Participants exposed to the inward-pushing perturbation increased their produced variability both during and after exposure. That is, lower perceived variability “frees” the motor system to be less precise and, surprisingly, leads to lasting changes in precision even when normal feedback is restored. However, the outward-pushing perturbation had no consistent effect on produced variability, contrary to results in reaching. This lack of a variability decrease suggests that at least some skilled movements are already produced at the lower limits of possible variability, as predicted by UCM/OFC. While overall variability did not change, vowel “centering”, a measure of within-trial correction for variability, did increase, suggesting participants became more responsive to errors. No such change was seen in the inward-pushing condition. Together, these results suggest that variability, even in complex tasks such as speech, is actively regulated.

Cristiano Alessandro¹, Filipe Barroso¹, Matthew Tresch¹

¹Northwestern University

Many studies have shown co-variation between muscle activation during behavior. According to one common proposal, this co-variation reflects simplification of task performance by the central nervous system (CNS), so that muscles with similar contributions to task variables are controlled together. Alternatively, this co-variation might reflect regulation of low-level aspects of movements that are common across tasks. Here, we demonstrate that the co-variation between quadriceps muscles in rats reflects regulation of stresses and strains within the knee. We analyzed co-variation patterns in quadriceps muscle activity during locomotion in rats (Alessandro et al. 2019). The three vastii (vastus medialis VM; vastus lateralis VL; and vastus intermedius VI) produce knee extension and so have identical contributions to task performance (Sandercock et al. 2018); rectus femoris (RF) produces an additional hip flexion. Consistent with the proposal that muscle co-variation is related to similarity of muscle actions on task variables, we found that the correlation between the activity of VM and VL was stronger than their correlations with RF. However, correlation between VM and VL was also stronger than their correlation with VI. Since all three vastii have identical actions on task variables, this finding suggests that muscle co-variation is not solely driven by simplification of task performance. Instead, the preferentially strong co-variation between VM and VL supports the control of internal joint stresses: since VM and VL produce opposing mediolateral forces on the patella, their strong positive correlation minimizes the net mediolateral patellar force. Consistent with this interpretation, the correlation between VM and VL was strongest during stance, when the interaction with the ground may increase the risk of patellar instability. Finally, VM-VL correlation was robust to the application of a chronic lateral patellar load: the CNS altered the relative activation of VM and VL to counterbalance the external load (Barroso et al., under review), but maintained the strong correlation between these muscles in order to regulate patellar stability. To probe the neural mechanisms underlying these phenomena, we investigated whether the co-variation between VM and VL was driven by feedback from joint afferents. Inhibition of these afferents by injection of lidocaine into the knee capsule did not alter the correlation between VM and VL, suggesting that the covariation between these muscles is either specified centrally or involves other sensory modalities. Interestingly, lidocaine injections reduced the correlation between VI and both VL and VM, suggesting the existence of multiple mechanisms involved in coordinating vastii muscle activations. These results provide important insights into the neural control of joint integrity and help characterize the neural substrates underlying this control.

Konstantina Kilteni¹, H Henrik Ehrsson¹

¹Karolinska Institutet

Self-generated touch feels less intense and less ticklish than identical externally generated touch. This somatosensory attenuation occurs because the brain predicts the tactile consequences of our self-generated movements. To produce attenuation, the tactile predictions need to be time-locked to the movement, but how the brain maintains this temporal tuning remains unknown. Using a bimanual self-touch paradigm, we demonstrate that people can rapidly unlearn to attenuate touch immediately after their movement and learn to attenuate delayed touch instead, after repeated exposure to a systematic delay between the movement and the resulting touch. The magnitudes of the unlearning and learning effects are correlated and dependent on the number of trials that participants have been exposed to. We further show that delayed touches feel less ticklish and non-delayed touches more ticklish after exposure to the systematic delay. These findings demonstrate that the attenuation of self-generated touch is adaptive. Kilteni et al. eLife 2019; DOI: https://doi.org/10.7554/eLife.42888

Daniela Buchwald¹, Hans Scherberger¹

¹Deutsches Primatenzentrum GmbH

Our hands are important tools that allow us to interact with our surroundings, ranging from the usage of our sense of touch for shape and texture recognition to object grasping. Many different brain areas are involved in the sensory-motor transformation that is required to link sensory information to meaningful movements. However, whether grasp planning activity depends on how object information is acquired and how the cortical areas involved in this process interact has not been studied well. In order to pursue this question, we recorded neural activity from a rhesus macaque (Macaca mulatta) during a delayed-grasping task, in which the animal was instructed to lift various objects that he perceived beforehand either visually or tactually. This allowed us to gain insight in how these two sensory modalities influence grasp planning. While the animal performed this task, we recorded spiking activity from four cortical areas (anterior intraparietal area AIP, premotor area F5, primary motor area M1, and the primary somatosensory area S1) involved in hand grasping. When comparing the activity of these brain areas between visual and tactile trials, we found differences in how the brain plans the required grasps, even though the actual movement was identical. To this end, we paid special attention to the preparatory period of this task, after an object was seen or touched, but before the grasp action was started. First, we checked how many units were significantly tuned to visual and tactile trials, finding differences between both trial types. In order to better quantify the difference between visual and tactile grasp planning activity, we trained a decoder to differentiate between the various objects and task types. During early memory, the decoder could distinguish well between visual and tactile trials, even well after the cue signals have vanished. During late memory, i.e., shortly before the movement start, the decoding performance decreased, but significant distinction between visual and tactile trials was still possible in all brain areas investigated. Only once the actual movement started, decoding accuracy decreased further, since from then on the executed movement and sensory information (darkness and touching the object to lift it) was identical. Together, we demonstrated that grasping movements based on visual and tactile object information were planned differently in these brain areas, even when the resulting movements are the same, indicating that the type of sensory modality used for a particular action planning is relevant for understanding the underlying neuronal processes. Supported by: Deutsche Forschungsgemeinschaft (SFB 889, Project C09).

Michele Tagliabue¹, Theo Morfoisse¹, Gabriela Herrera Altamira¹, Leonardo Angelini², Gilles Clément³, Mathieu Beraneck⁴, Joseph McIntyre⁵

¹Université Paris Descartes, ²HES-SO, ³Lyon Neuroscience Research Center, ⁴CNRS-Université Paris Descartes, ⁵Ikerbasque Science Foundation

Hand movements aimed at grasping an object are controlled on the basis of the perception of its position, shape and dimensions. The aim of this study is to investigate object 3D perception. Human visual 3D perception is flawed by distortions, which are influenced by non-visual factors, such as gravitational vestibular signals. Distinct hypotheses regarding the sensory processing stage at which gravity acts may explain the influence of gravity: 1) a direct effect on the sensory system, 2) a role in shaping the internal representation of space that is used to interpret sensory signals or 3) a role in the ability to reconstruct multiple modality-specific representations of the perceived object. To test these hypotheses, we performed experiments comparing visual versus haptic 3D object perception, and the respective effects of microgravity on these two sensory modalities. If the effect of microgravity on visual and haptic perception are independent, then the first hypothesis is the most likely, because gravity act differently on the two sensory systems. If the effect of microgravity is the same for the two sensory modalities, the second hypothesis should privileged, because the way an alteration of the internal representation of space affect the 3D perception is independent of the sensory channel used to sense the object. The results show that visual and haptic perceptual anisotropies reside in body-centered, and not gravity-centered, planes, suggesting an ego-centric encoding of the information for both sensory modalities. Although coplanar, the perceptual distortions of the two sensory modalities are in opposite directions: depth is visually underestimated, but haptically overestimated, with respect to height and width. Interestingly, microgravity appears to amplify the ‘terrestrial’ distortions of both senses. Through computational modeling, we show that these findings can be parsimoniously predicted only by the third hypothesis,  founded on the idea that gravity facilitates cross-modal transformations of the sensory information. This theory is able to explain not only how gravity can shape intrinsically egocentric processes, but also the unexpected opposite effect of gravity on visual and haptic 3D perception. Overall, these results suggest that the brain reconstructs concurrent, modality-specific internal representations of the 3D objects even when they are sensed through only one sensory channel.

Chair: Neeraj Gandhi¹

Presenters: Michelle Heusser¹, Timothy Darlington², Eilon Vaadia³, Matthew Perich⁴

Confirmed Discussant: Aaron Batista¹

¹University of Pittsburgh, ²Duke University School of Medicine, ³The Hebrew University of Jerusalem, ⁴University of Geneva

How do neural populations differentially represent signals across conditions? How do brain areas flexibly communicate with each other? Recently, neuroscientists have begun to investigate how neural populations dynamically encode multiple signals through neural “subspaces.” A “subspace” is, broadly speaking, formed by population-wide patterns of neural during a particular task or context (typically after performing dimensionality reduction). This session aims to highlight research across brain regions, end effectors, and paradigms employing this technique, with the first two talks focusing on the primate oculomotor system, and the latter two on the skeletomotor system. Each panelist will discuss how subspaces can be found and exploited to study population coding during motor tasks. First, Heusser will characterize the subspaces formed by a population of neurons in the superior colliculus (SC) during the visual and movement epochs of a delayed saccade task. She employs state-space methods to compare delay period activity to these putative visual and motor subspaces and to quantify sensorimotor transformation. Her results uncover a separate subspace formed during the delay period, and show that the activity within this subspace is indicative of the single-trial movement preparation state. Darlington will compare subspaces found in frontal eye fields (FEF) population activity during preparation and initiation of smooth pursuit eye movements. He finds separable subspaces during the preparation and execution periods, suggesting that the population activity undergoes partial reorganization. He then examines individual-neuron and subpopulation structure that might give rise to this low-dimensional separation. Vaadia will present recent work comparing neural activity in primary motor cortex (M1) during arm movement and brain-machine interface (BMI) tasks. In short, M1 activity during these tasks parcellates into orthogonal subspaces, with some representations unique to the effector and others that are shared. He will also demonstrate that the population state during sensorimotor learning is flexible, with M1 neural networks adapting to create neural subspaces corresponding to the current action-perception state. Finally, Perich will examine the dynamics exhibited by neural populations in M1 and somatosensory cortex (S1) during a reach-to-grasp task. He will show that there are distinct components of M1 dynamics that reflect motor execution and sensory feedback, arguing for a communication subspace across these regions. The session will conclude with a discussion of the rationale for subspace-based approaches as well as the implications of these findings for our understanding of population coding during movement.

In order to link perception and action the brain must have a spatially accurate representation of the visual world, so it can generate actions appropriate to the objects it perceives. The only way visual information enters the eye is through the retina, which moves constantly between brief fixations. The retinal location of targets for action is not useful for calculating movements to acquire those targets. Two strategies have been postulated to calculate the accurate location of movement targets: Helmholtz suggested that the brain knows the command to move the eye, and therefore can use that motor command to update the sensory representation. This feedback from the motor system to the sensory system is now known as corollary discharge. Sherrington suggested that the brain can calculate accurate target location if it knows the position of the eye in the world, and the first step in this process is to know the position of the eye in the orbit. He postulated that this signal arose from oculomotor proprioceptors. The lateral intraparietal area (LIP) is a brain region important in choosing targets for saccadic eye movements, and solves the spatial accuracy problem using both Helmholtz’s and Sherrington’s strategies: a rapid, relatively accurate corollary discharge mechanism, and a slower, but more accurate proprioceptive mechanism, which is dependent upon the sensory representation of eye position in Area 3a of the primary somatosensory cortex.