Major remaining gaps in models of sensorimotor systems

AngoraPy: A Python toolkit for modeling anthropomorphic goal-driven sensorimotor systems

Goal-driven deep learning increasingly supplements classical modeling approaches in computational neuroscience. The strength of deep neural networks as models of the brain lies in their ability to autonomously learn the connectivity required to solve complex and ecologically valid tasks, obviating the need for hand-engineered or hypothesis-driven connectivity patterns. Consequently, goal-driven models can generate hypotheses about the neurocomputations underlying cortical processing that are grounded in macro- and mesoscopic anatomical properties of the network's biological counterpart. Whereas, goal-driven modeling is already becoming prevalent in the neuroscience of perception, its application to the sensorimotor domain is currently hampered by the complexity of the methods required to train models comprising the closed sensation-action loop. This paper describes AngoraPy, a Python library that mitigates this obstacle by providing researchers with the tools necessary to train complex recurrent convolutional neural networks that model the human sensorimotor system. To make the technical details of this toolkit more approachable, an illustrative example that trains a recurrent toy model on in-hand object manipulation accompanies the theoretical remarks. An extensive benchmark on various classical, 3D robotic, and anthropomorphic control tasks demonstrates AngoraPy's general applicability to a wide range of tasks. Together with its ability to adaptively handle custom architectures, the flexibility of this toolkit demonstrates its power for goal-driven sensorimotor modeling.

Diversified physiological sensory input connectivity questions the existence of distinct classes of spinal interneurons

The spinal cord is engaged in all forms of motor performance but its functions are far from understood. Because network connectivity defines function, we explored the connectivity of muscular, tendon, and tactile sensory inputs among a wide population of spinal interneurons in the lower cervical segments. Using low noise intracellular whole cell recordings in the decerebrated, non-anesthetized cat in vivo, we could define mono-, di-, and trisynaptic inputs as well as the weights of each input. Whereas each neuron had a highly specific input, and each indirect input could moreover be explained by inputs in other recorded neurons, we unexpectedly also found the input connectivity of the spinal interneuron population to form a continuum. Our data hence contrasts with the currently widespread notion of distinct classes of interneurons. We argue that this suggested diversified physiological connectivity, which likely requires a major component of circuitry learning, implies a more flexible functionality.

Computational Modeling: Human Dynamic Model

Improvements in quantitative measurements of human physical activity are proving extraordinarily useful for studying the underlying musculoskeletal system. Dynamic models of human movement support clinical efforts to analyze, rehabilitate injuries. They are also used in biomechanics to understand and diagnose motor pathologies, find new motor strategies that decrease the risk of injury, and predict potential problems from a particular procedure. In addition, they provide valuable constraints for understanding neural circuits. This paper describes a physics-based movement analysis method for analyzing and simulating bipedal humanoid movements. The model includes the major body segments and joints to report human movements' energetic components. Its 48 degrees of freedom strike a balance between very detailed models that include muscle models and straightforward two-dimensional models. It has sufficient accuracy to analyze and synthesize movements captured in real-time interactive applications, such as psychophysics experiments using virtual reality or human-in-the-loop teleoperation of a simulated robotic system. The dynamic model is fast and robust while still providing results sufficiently accurate to be used to animate a humanoid character. It can also estimate internal joint forces used during a movement to create effort-contingent stimuli and support controlled experiments to measure the dynamics generating human behaviors systematically. The paper describes the innovative features that allow the model to integrate its dynamic equations accurately and illustrates its performance and accuracy with demonstrations. The model has a two-foot stance ability, capable of generating results comparable with an experiment done with subjects, and illustrates the uncontrolled manifold concept. Additionally, the model's facility to capture large energetic databases opens new possibilities for theorizing as to human movement function. The model is freely available.

Proprioceptive Recognition with Artificial Neural Networks Based on Organizations of Spinocerebellar Tract and Cerebellum

Muscle kinematics and kinetics are nonlinearly encoded by proprioceptors, and the changes in muscle length and velocity are integrated into Ia afferent. Besides, proprioceptive signals from multiple muscles are probably mixed in afferent pathways, which all lead to difficulties in proprioceptive recognition for the cerebellum. In this study, the artificial neural networks, whose organizations are biologically based on the spinocerebellar tract and cerebellum, are utilized to decode the proprioceptive signals. Consistent with the controversy of the proprioceptive division in the dorsal spinocerebellar tract, the spinocerebellar tract networks incorporated two distinct inferences, (1) the centralized networks, which mixed Ia, II, and Ib and processed them together; (2) the decentralized networks, which processed Ia, II, and Ib afferents separately. The cerebellar networks were based on the Marr-Albus model to recognize the kinematic states. The networks were trained by a specific movement, and the trained networks were subsequently required to predict kinematic states of six movements. The results demonstrated that the centralized networks, which were more consistent with the physiological findings in recent years, had better recognition accuracy than the decentralized networks, and the networks were still effective even when proprioceptive afferents from multiple muscles were integrated.

Synergies in coordination: a comprehensive overview of neural, computational, and behavioral approaches

At face value, the term "synergy" provides a unifying concept within a fractured field that encompasses complementary neural, computational, and behavioral approaches. However, the term is not used synonymously by different researchers but has substantially different meanings depending on the research approach. With so many operational definitions for the one term, it becomes difficult to use as either a descriptive or explanatory concept, yet it remains pervasive and apparently indispensable. Here we provide a summary of different approaches that invoke synergies in a descriptive or explanatory context, summarizing progress, not within the one approach, but across the theoretical landscape. Bernstein's framework of flexible hierarchical control may provide a unifying framework here, since it can incorporate divergent ideas about synergies. In the current motor control literature, synergy may refer to conceptually different processes that could potentially operate in parallel, across different levels within the same hierarchical control scheme. There is evidence for the concurrent existence of synergies with different features, both "hard-wired" and "soft-wired," and task independent and task dependent. By providing a comprehensive overview of the multifaceted ideas about synergies, our goal is to move away from the compartmentalization and narrow the focus on one level and promote a broader perspective on the control and coordination of movement.

Real-time slacking as a default mode of grip force control: implications for force minimization and personal grip force variation

During trial-to-trial movement adaptation, the motor system systematically reduces extraneous muscle forces when kinematic errors experienced on previous movements are small, a phenomenon termed "slacking." There is also growing evidence that the motor system slacks continuously (i.e., in real-time) during arm movement or grip force control, but the initiation of this slacking is not well-characterized, obfuscating its physiological cause. Here, we addressed this issue by asking participants ( n = 32) to track discrete force targets presented visually using isometric grip force, then applying a brief, subtle error-clamp to that visual feedback on random trials. Participants reduced their force in an exponential fashion, on these error-clamp trials, except when the target force was <10% maximum voluntary contraction (MVC). This force drift began <250 ms after the onset of the error-clamp, consistent with slacking being an ongoing process unmasked immediately after the motor system finished reacting to the last veridical feedback. Above 10% MVC, the slacking rate increased linearly with grip force magnitude. Grip force variation was approximately 50-100% higher with veridical feedback, largely due to heightened signal power at ~1 Hz, the band of visuomotor feedback control. Finally, the slacking rate measured for each participant during error-clamp trials correlated with their force variation during control trials. That is, participants who slacked more had greater force variation. These results suggest that real-time slacking continuously reduces grip force until visual error prompts correction. Whereas such slacking is suited for force minimization, it may also account for ~30% of the variability in personal grip force variation. NEW & NOTEWORTHY We provide evidence that a form of slacking continuously conditions real-time grip force production. This slacking is well-suited to promote efficiency but is expected to increase force variation by triggering additional feedback corrections. Moreover, we show that the rate at which a person slacks is substantially correlated with the variation of their grip force. In combination, at the neurophysiological level, our results suggest slacking is caused by one or more relatively smooth neural adaptations.

A Proposed Circuit Computation in Basal Ganglia: History-Dependent Gain

In this Scientific Perspectives we first review the recent advances in our understanding of the functional architecture of basal ganglia circuits. Then we argue that these data can best be explained by a model in which basal ganglia act to control the gain of movement kinematics to shape performance based on prior experience, which we refer to as a history-dependent gain computation. Finally, we discuss how insights from the history-dependent gain model might translate from the bench to the bedside, primarily the implications for the design of adaptive deep brain stimulation. Thus, we explicate the key empirical and conceptual support for a normative, computational model with substantial explanatory power for the broad role of basal ganglia circuits in health and disease. © 2018 The Authors. Movement Disorders published by Wiley Periodicals, Inc. on behalf of International Parkinson and Movement Disorder Society.

Cardinal features of involuntary force variability can arise from the closed-loop control of viscoelastic afferented muscles

Involuntary force variability below 15 Hz arises from, and is influenced by, many factors including descending neural drive, proprioceptive feedback, and mechanical properties of muscles and tendons. However, their potential interactions that give rise to the well-structured spectrum of involuntary force variability are not well understood due to a lack of experimental techniques. Here, we investigated the generation, modulation, and interactions among different sources of force variability using a physiologically-grounded closed-loop simulation of an afferented muscle model. The closed-loop simulation included a musculotendon model, muscle spindle, Golgi tendon organ (GTO), and a tracking controller which enabled target-guided force tracking. We demonstrate that closed-loop control of an afferented musculotendon suffices to replicate and explain surprisingly many cardinal features of involuntary force variability. Specifically, we present 1) a potential origin of low-frequency force variability associated with co-modulation of motor unit firing rates (i.e.,‘common drive’), 2) an in-depth characterization of how proprioceptive feedback pathways suffice to generate 5-12 Hz physiological tremor, and 3) evidence that modulation of those feedback pathways (i.e., presynaptic inhibition of Ia and Ib afferents, and spindle sensitivity via fusimotor drive) influence the full spectrum of force variability. These results highlight the previously underestimated importance of closed-loop neuromechanical interactions in explaining involuntary force variability during voluntary ‘isometric’ force control. Furthermore, these results provide the basis for a unifying theory that relates spinal circuitry to various manifestations of altered involuntary force variability in fatigue, aging and neurological disease.

Involuntary fluctuations in muscle force are an unavoidable consequence of human motor control and underlie movement execution errors. Amplification and distortion of involuntary force variability are common phenomena found in various neurological conditions and in fatigue. However, the underlying mechanisms for this are often unclear. We investigated the generation and modulation of involuntary force variability arising from different sources, as well as their interactions. We used a closed-loop simulation which included a physiologically-grounded model of an afferented musculotendon and an error-controller. We show that interactions among neural noise, musculotendon mechanics, proprioceptive feedback, and error correction are critical components of force control, and by taking these into account, our model was able to both replicate and explain many cardinal features of involuntary force variability previously reported experimentally. Also, our results suggest previously unrecognized pathways through which force variability may be altered in fatigue and in certain neurological diseases. Finally, we emphasize the potential for important clinical and scientific information to be extracted from relatively simple, non-invasive measurements of force.

Loss of Shp2 within radial glia is associated with cerebral cortical dysplasia, glial defects of cerebellum and impaired sensory‑motor development in newborn mice

Radial glia are key neural progenitors involved in the development of the central nervous system. Tyrosine-protein phosphatase non‑receptor type 11 (Shp2) is a widely expressed intracellular enzyme with multiple cellular functions. Previous studies have revealed the critical role of Shp2 in a variety of neural cell types; however, further investigation into the function of Shp2 within radial glia is required. In the present study, a conditional knockout mouse was generated using a human glial fibrillary acidic protein (hGFAP)‑Cre driver, in which the Shp2 genes were deleted within radial glia. Loss of Shp2 within radial glia was associated with developmental retardation, postnatal lethality, reduced brain size and thinner cerebral cortices in newborn mice. Deletion of Shp2 also led to an increase in gliogenesis, a reduction in neural genesis and extracellular signal‑regulated kinase signaling within the cerebral cortex. Furthermore, glial cell defects within the cerebellum of Shp2 mutants were observed, with abnormal granular cell retention and glial cell alignment in the external granular layer. In addition, Shp2 mutants exhibited impaired sensory‑motor development. The results of the present study suggested that Shp2 may have an important role within radial glia, and regulate cerebral cortical and cerebellar development in newborn mice.

Muscle and Limb Mechanics

Understanding of the musculoskeletal system has evolved from the collection of individual phenomena in highly selected experimental preparations under highly controlled and often unphysiological conditions. At the systems level, it is now possible to construct complete and reasonably accurate models of the kinetics and energetics of realistic muscles and to combine them to understand the dynamics of complete musculoskeletal systems performing natural behaviors. At the reductionist level, it is possible to relate most of the individual phenomena to the anatomical structures and biochemical processes that account for them. Two large challenges remain. At a systems level, neuroscience must now account for how the nervous system learns to exploit the many complex features that evolution has incorporated into muscle and limb mechanics. At a reductionist level, medicine must now account for the many forms of pathology and disability that arise from the many diseases and injuries to which this highly evolved system is inevitably prone. © 2017 American Physiological Society. Compr Physiol 7:429-462, 2017.

Neural Network Model of Vestibular Nuclei Reaction to Onset of Vestibular Prosthetic Stimulation

The vestibular system incorporates multiple sensory pathways to provide crucial information about head and body motion. Damage to the semicircular canals, the peripheral vestibular organs that sense rotational velocities of the head, can severely degrade the ability to perform activities of daily life. Vestibular prosthetics address this problem by using stimulating electrodes that can trigger primary vestibular afferents to modulate their firing rates, thus encoding head movement. These prostheses have been demonstrated chronically in multiple animal models and acutely tested in short-duration trials within the clinic in humans. However, mainly, due to limited opportunities to fully characterize stimulation parameters, there is a lack of understanding of “optimal” stimulation configurations for humans. Here, we model possible adaptive plasticity in the vestibular pathway. Specifically, this model highlights the influence of adaptation of synaptic strengths and offsets in the vestibular nuclei to compensate for the initial activation of the prosthetic. By changing the synaptic strengths, the model is able to replicate the clinical observation that erroneous eye movements are attenuated within 30 minutes without any change to the prosthetic stimulation rate. Although our model was only built to match this time point, we further examined how it affected subsequent pulse rate modulation (PRM) and pulse amplitude modulation (PAM). PAM was more effective than PRM for nearly all stimulation configurations during these acute tests. Two non-intuitive relationships highlighted by our model explain this performance discrepancy. Specifically, the attenuation of synaptic strengths for afferents stimulated during baseline adaptation and the discontinuity between baseline and residual firing rates both disproportionally boost PAM. Comodulation of pulse rate and amplitude has been experimentally shown to induce both excitatory and inhibitory eye movements even at high baseline stimulation rates. We also modeled comodulation and found synergistic combinations of stimulation parameters to achieve equivalent output to only amplitude modulation. This may be an important strategy to reduce current spread and misalignment. The model outputs reflected observed trends in clinical testing and aspects of existing vestibular prosthetic literature. Importantly, the model provided insight to efficiently explore the stimulation parameter space, which was helpful, given limited available patient time.

Neural Data-Driven Musculoskeletal Modeling for Personalized Neurorehabilitation Technologies

OBJECTIVES

The development of neurorehabilitation technologies requires the profound understanding of the mechanisms underlying an individual's motor ability and impairment. A major factor limiting this understanding is the difficulty of bridging between events taking place at the neurophysiologic level (i.e., motor neuron firings) with those emerging at the musculoskeletal level (i.e. joint actuation), in vivo in the intact moving human. This review presents emerging model-based methodologies for filling this gap that are promising for developing clinically viable technologies.

METHODS

We provide a design overview of musculoskeletal modeling formulations driven by recordings of neuromuscular activity with a critical comparison to alternative model-free approaches in the context of neurorehabilitation technologies. We present advanced electromyography-based techniques for interfacing with the human nervous system and model-based techniques for translating the extracted neural information into estimates of motor function.

RESULTS

We introduce representative application areas where modeling is relevant for accessing neuromuscular variables that could not be measured experimentally. We then show how these variables are used for designing personalized rehabilitation interventions, biologically inspired limbs, and human-machine interfaces.

CONCLUSION

The ability of using electrophysiological recordings to inform biomechanical models enables accessing a broader range of neuromechanical variables than analyzing electrophysiological data or movement data individually. This enables understanding the neuromechanical interplay underlying in vivo movement function, pathology, and robot-assisted motor recovery.

SIGNIFICANCE

Filling the gap between our understandings of movement neural and mechanical functions is central for addressing one of the major challenges in neurorehabilitation: personalizing current technologies and interventions to an individual's anatomy and impairment.

Restoring Behavior via Inverse Neurocontroller in a Lesioned Cortical Spiking Model Driving a Virtual Arm

Neural stimulation can be used as a tool to elicit natural sensations or behaviors by modulating neural activity. This can be potentially used to mitigate the damage of brain lesions or neural disorders. However, in order to obtain the optimal stimulation sequences, it is necessary to develop neural control methods, for example by constructing an inverse model of the target system. For real brains, this can be very challenging, and often unfeasible, as it requires repeatedly stimulating the neural system to obtain enough probing data, and depends on an unwarranted assumption of stationarity. By contrast, detailed brain simulations may provide an alternative testbed for understanding the interactions between ongoing neural activity and external stimulation. Unlike real brains, the artificial system can be probed extensively and precisely, and detailed output information is readily available. Here we employed a spiking network model of sensorimotor cortex trained to drive a realistic virtual musculoskeletal arm to reach a target. The network was then perturbed, in order to simulate a lesion, by either silencing neurons or removing synaptic connections. All lesions led to significant behvaioral impairments during the reaching task. The remaining cells were then systematically probed with a set of single and multiple-cell stimulations, and results were used to build an inverse model of the neural system. The inverse model was constructed using a kernel adaptive filtering method, and was used to predict the neural stimulation pattern required to recover the pre-lesion neural activity. Applying the derived neurostimulation to the lesioned network improved the reaching behavior performance. This work proposes a novel neurocontrol method, and provides theoretical groundwork on the use biomimetic brain models to develop and evaluate neurocontrollers that restore the function of damaged brain regions and the corresponding motor behaviors.

Suboptimal Muscle Synergy Activation Patterns Generalize their Motor Function across Postures

We used a musculoskeletal model to investigate the possible biomechanical and neural bases of using consistent muscle synergy patterns to produce functional motor outputs across different biomechanical conditions, which we define as generalizability. Experimental studies in cats demonstrate that the same muscle synergies are used during reactive postural responses at widely varying configurations, producing similarly-oriented endpoint force vectors with respect to the limb axis. However, whether generalizability across postures arises due to similar biomechanical properties or to neural selection of a particular muscle activation pattern has not been explicitly tested. Here, we used a detailed cat hindlimb model to explore the set of feasible muscle activation patterns that produce experimental synergy force vectors at a target posture, and tested their generalizability by applying them to different test postures. We used three methods to select candidate muscle activation patterns: (1) randomly-selected feasible muscle activation patterns, (2) optimal muscle activation patterns minimizing muscle effort at a given posture, and (3) generalizable muscle activation patterns that explicitly minimized deviations from experimentally-identified synergy force vectors across all postures. Generalizability was measured by the deviation between the simulated force direction of the candidate muscle activation pattern and the experimental synergy force vectors at the test postures. Force angle deviations were the greatest for the randomly selected feasible muscle activation patterns (e.g., >100°), intermediate for effort-wise optimal muscle activation patterns (e.g., ~20°), and smallest for generalizable muscle activation patterns (e.g., <5°). Generalizable muscle activation patterns were suboptimal in terms of effort, often exceeding 50% of the maximum possible effort (cf. ~5% in minimum-effort muscle activation patterns). The feasible muscle activation ranges of individual muscles associated with producing a specific synergy force vector was reduced by ~45% when generalizability requirements were imposed. Muscles recruited in the generalizable muscle activation patterns had less sensitive torque-producing characteristics to changes in postures. We conclude that generalization of function across postures does not arise from limb biomechanics or a single optimality criterion. Muscle synergies may reflect acquired motor solutions globally tuned for generalizability across biomechanical contexts, facilitating rapid motor adaptation.