Cerebellar learning of accurate predictive control for fast-reaching movements

Long conduction delays in the nervous system prevent the accurate control of movements by feedback control alone. We present a new, biologically plausible cerebellar model to study how fast arm movements can be executed in spite of these delays. To provide a realistic test-bed of the cerebellar neural model, we embed the cerebellar network in a simulated biological motor system comprising a spinal cord model and a six-muscle two-dimensional arm model. We argue that if the trajectory errors are detected at the spinal cord level, memory traces in the cerebellum can solve the temporal mismatch problem between efferent motor commands and delayed error signals. Moreover, learning is made stable by the inclusion of the cerebello-nucleo-olivary loop in the model. It is shown that the cerebellar network implements a nonlinear predictive regulator by learning part of the inverse dynamics of the plant and spinal circuit. After learning, fast accurate reaching movements can be generated. Received: 8 February 1999 /Accepted in revised form: 7 August 1999     

Expression of the bilateral deficit during reflexively evoked contractions.

During maximal contractions, the sum of forces exerted by homonymous muscles unilaterally is typically larger than the sum of forces exerted by the same muscles bilaterally. This phenomenon is known as the bilateral deficit (BLD), and it is suggested that this deficit is due to neural inhibition. It remains unclear, however, whether such inhibition is mediated by supraspinal mechanisms or by reflex pathways at the level of spinal cord. To further study the origin of likely neural influences, we tested for the presence of BLD under the condition of reflexive force generation. Force output and integrated electromyogram (iEMG) (quadriceps femoris) were measured in 17 male participants after initiation of the myotatic patellar reflex under unilateral and bilateral conditions. A significant BLD of 9.26 +/- 1.19 (P = 0.004) and 16.76 +/- 4.69% (P = 0.001) was found for force and iEMG, respectively. However, because similar findings were not evident during maximal isometric knee extensions, it is difficult to predict the contribution of a spinal mechanism to the BLD under the condition of maximal voluntary activation.     

Self-organizing effects of spontaneous neural activity on the development of spinal locomotor circuits in vertebrates

Presented in this paper is a neural network model that can be used to investigate the possible self-organizing mechanisms occurring during the early ontogeny of spinal neural circuits in the vertebrate motor system. The neural circuit is composed of multiple types of neurons which correspond to motorneurons, Renshaw cells and a hypothetical class of interneurons. While the connectivity of this circuit is genetically predetermined, the efficacies of these connections – the synaptic s trengths – evolve in accordance with activity-dependent mechanisms which are initiated by the intrinsic, autonomous activity present in the developing spinal cord. Using Oja's rule, a modified Hebbian learning scheme for adjusting the values of the connections, the network stably self-organizes developing, in the process, reciprocally activated motorneuron pools analogous to those which exist in vivo. Received: 30 December 1996 / Accepted in revised form: 20 June 1997     

Functional identification of interneurons responsible for left-right coordination of hindlimbs in mammals

Local neuronal networks that are responsible for walking are poorly characterized in mammals. Using an innovative approach to identify interneuron inputs onto motorneuron populations in a neonatal rodent spinal cord preparation, we have investigated the network responsible for left-right coordination of the hindlimbs. We demonstrate how commissural interneurons (CINs), whose axons traverse the midline to innervate contralateral neurons, are organized such that distinct flexor and extensor centers in the rostral lumbar spinal cord define activity in both flexor and extensor caudal motor pools. In addition, the nature of some connections are reconfigured on switching from rest to locomotion via a mechanism that might be associated with synaptic plasticity in the spinal cord. These results from identified pattern-generating interneurons demonstrate how interneuron populations create an effective network to underlie behavior in mammals.     

The Beneficial Effects of Treadmill Step Training on Activity-Dependent Synaptic and Cellular Plasticity Markers After Complete Spinal Cord Injury

Several studies have shown that treadmill training improves neurological outcomes and promotes plasticity in lumbar spinal cord of spinal animals. The morphological and biochemical mechanisms underlying these phenomena remain unclear. The purpose of this study was to provide evidence of activity-dependent plasticity in spinal cord segment (L5) below a complete spinal cord transection (SCT) at T8-9 in rats in which the lower spinal cord segments have been fully separated from supraspinal control and that subsequently underwent treadmill step training. Five days after SCT, spinal animals started a step-training program on a treadmill with partial body weight support and manual step help. Hindlimb movements were evaluated over time and scored on the basis of the open-field BBB scale and were significantly improved at post-injury weeks 8 and 10 in trained spinal animals. Treadmill training also showed normalization of withdrawal reflex in trained spinal animals, which was significantly different from the untrained animals at post-injury weeks 8 and 10. Additionally, compared to controls, spinal rats had alpha motoneuronal soma size atrophy and reduced synaptophysin protein expression and Na(+), K(+)-ATPase activity in lumbar spinal cord. Step-trained rats had motoneuronal soma size, synaptophysin expression and Na(+), K(+)-ATPase activity similar to control animals. These findings suggest that treadmill step training can promote activity-dependent neural plasticity in lumbar spinal cord, which may lead to neurological improvements without supraspinal descending control after complete spinal cord injury.     

An unsupervised neural network model for the development of reflex co-ordination

In this paper, we present a model for the development of connections between muscle afferents and motoneurones in the human spinal cord. The model consists of a limb with six muscles, one motoneurone pool, one pooled (Ia-like) afferent for each muscle and a central programme generator. The weights of the connections between the afferents and the motoneurone pools are adapted during centrally induced movements of the limb. The connections between the afferents and the motoneurone pools adapt in a hebbian way, using only local information present at the synapses. This neural network is tested in two examples of a limb with two degrees of freedom and six muscles. Despite the simplifications, the model predicts the pattern of autogenic and heterogenic monosynaptic reflexes quite realistically.     

Dynamic Neural Network Models of the Premotoneuronal Circuitry Controlling Wrist Movements in Primates

Dynamic recurrent neural networks were derived to simulate neuronal populations generating bidirectional wrist movements in the monkey. The models incorporate anatomical connections of cortical and rubral neurons, muscle afferents, segmental interneurons and motoneurons; they also incorporate the response profiles of four populations of neurons observed in behaving monkeys. The networks were derived by gradient descent algorithms to generate the eight characteristic patterns of motor unit activations observed during alternating flexion-extension wrist movements. The resulting model generated the appropriate input-output transforms and developed connection strengths resembling those in physiological pathways. We found that this network could be further trained to simulate additional tasks, such as experimentally observed reflex responses to limb perturbations that stretched or shortened the active muscles, and scaling of response amplitudes in proportion to inputs. In the final comprehensive network, motor units are driven by the combined activity of cortical, rubral, spinal and afferent units during step tracking and perturbations.The model displayed many emergent properties corresponding to physiological characteristics. The resulting neural network provides a working model of premotoneuronal circuitry and elucidates the neural mechanisms controlling motoneuron activity. It also predicts several features to be experimentally tested, for example the consequences of eliminating inhibitory connections in cortex and red nucleus. It also reveals that co-contraction can be achieved by simultaneous activation of the flexor and extensor circuits without invoking features specific to co-contraction.     

Monosynaptic rabies virus reveals premotor network organization and synaptic specificity of cholinergic partition cells

Movement is the behavioral output of neuronal activity in the spinal cord. Motor neurons are grouped into motor neuron pools, the functional units innervating individual muscles. Here we establish an anatomical rabies virus-based connectivity assay in early postnatal mice. We employ it to study the connectivity scheme of premotor neurons, the neuronal cohorts monosynaptically connected to motor neurons, unveiling three aspects of organization. First, motor neuron pools are connected to segmentally widely distributed yet stereotypic interneuron populations, differing for pools innervating functionally distinct muscles. Second, depending on subpopulation identity, interneurons take on local or segmentally distributed positions. Third, cholinergic partition cells involved in the regulation of motor neuron excitability segregate into ipsilaterally and bilaterally projecting populations, the latter exhibiting preferential connections to functionally equivalent motor neuron pools bilaterally. Our study visualizes the widespread yet precise nature of the connectivity matrix for premotor interneurons and reveals exquisite synaptic specificity for bilaterally projecting cholinergic partition cells.     

Spinal and spino-bulbo-spinal neuronal mechanisms of transmission of somatomotor and visceromotor reflexes in the thoracic spinal cord

Early (spinal) and late (spino-bulbo-spinal) responses of interneurons in segments T9–10 to stimulation of the splanchnic and intercostal nerves and the dorso-lateral and ventral funiculi of the spinal cord (at the C3 level) were investigated in experiments on cats anesthetized with chloralose. The experiments showed that interneurons activated by spinal and spino-bulbo-spinal mechanisms differ in their distribution in the dorso-ventral plane of the spinal cord. Cells of layers I–V were excited by spinal pathways only, but those of layers VII and VIII by both spinal and spino-bulbo-spinal or only by the latter. Spino-bulbo-spinal effects were evoked in interneurons by both somatic and visceral afferent waves. A conditioning spino-bulbo-spinal wave evoked deep and prolonged inhibition of late activity induced by somatic or visceral afferent impulses. Early (spinal) activity was inhibited only partially under these circumstances. This inhibition was shown to take place with the participation of supraspinal structures. The possible types of spinal and supraspinal mechanisms of inhibition of early and late activity in spinal neurons are discussed.Institute of Normal and Pathological Physiology, Slovak Academy of Sciences, Bratislava, Czechoslovakia. A. A. Bogomolets Institute of Physiology, Academy of Sciences of the Ukrainian SSR, Kiev, USSR. Translated from Neirofiziologiya, Vol. 5, No. 4, pp. 392–400, July–August, 1973.     

Simultaneous Brain–Cervical Cord fMRI Reveals Intrinsic Spinal Cord Plasticity during Motor Sequence Learning

The spinal cord participates in the execution of skilled movements by translating high-level cerebral motor representations into musculotopic commands. Yet, the extent to which motor skill acquisition relies on intrinsic spinal cord processes remains unknown. To date, attempts to address this question were limited by difficulties in separating spinal local effects from supraspinal influences through traditional electrophysiological and neuroimaging methods. Here, for the first time, we provide evidence for local learning-induced plasticity in intact human spinal cord through simultaneous functional magnetic resonance imaging of the brain and spinal cord during motor sequence learning. Specifically, we show learning-related modulation of activity in the C6–C8 spinal region, which is independent from that of related supraspinal sensorimotor structures. Moreover, a brain–spinal cord functional connectivity analysis demonstrates that the initial linear relationship between the spinal cord and sensorimotor cortex gradually fades away over the course of motor sequence learning, while the connectivity between spinal activity and cerebellum gains strength. These data suggest that the spinal cord not only constitutes an active functional component of the human motor learning network but also contributes distinctively from the brain to the learning process. The present findings open new avenues for rehabilitation of patients with spinal cord injuries, as they demonstrate that this part of the central nervous system is much more plastic than assumed before. Yet, the neurophysiological mechanisms underlying this intrinsic functional plasticity in the spinal cord warrant further investigations.     

Coordinated force production in multi-finger tasks: finger interaction and neural network modeling

During maximal voluntary contraction (MVC) with several fingers, the following three phenomena are observed: (1) the total force produced by all the involved fingers is shared among the fingers in a specific manner (sharing); (2) the force produced by a given finger in a multi-finger task is smaller than the force generated by this finger in a single-finger task (force deficit); (3) the fingers that are not required to produce any force by instruction are involuntary activated (enslaving). We studied involuntary force production by individual fingers (enslaving effects, EE) during tasks when (an)other finger(s) of the hand generated maximal voluntary pressing force in isometric conditions. The subjects (n = 10) were instructed to press as hard as possible on the force sensors with one, two, three and four fingers acting in parallel in all possible combinations. The EE were (A) large, the slave fingers always producing a force ranging from 10.9% to 54.7% of the maximal force produced by the finger in the single-finger task; (B) nearly symmetrical; (C) larger for the neighboring fingers; and (D) non-additive. In most cases, the EE from two or three fingers were smaller than the EE from at least one finger (this phenomenon was coined occlusion). The occlusion cannot be explained only by anatomical musculo-tendinous connections. Therefore, neural factors contribute substantially to the EE. A neural network model that accounts for all the three effects has been developed. The model consists of three layers: the input layer that models a central neural drive; the hidden layer modeling transformation of the central drive into an input signal to the muscles serving several fingers simultaneously (multi-digit muscles); and the output layer representing finger force output. The output of the hidden layer is set inversely proportional to the number of fingers involved. In addition, direct connections between the input and output layers represent signals to the hand muscles serving individual fingers (uni-digit muscles). The network was validated using three different training sets. Single digit muscles contributed from 25% to 50% of the total finger force. The master matrix and the enslaving matrix were computed; they characterize the ability of a given finger to enslave other fingers and its ability to be enslaved. Overall, the neural network modeling suggests that no direct correspondence exists between neural command to an individual finger and finger force. To produce a desired finger force, a command sent to an intended finger should be scaled in accordance with the commands sent to the other fingers. Received: 17 October 1997 / Accepted in revised form: 12 May 1998     

CTCF regulates ataxin-7 expression through promotion of a convergently transcribed, antisense noncoding RNA

Neural networks in the spinal cord control two basic features of locomotor movements: rhythm generation and pattern generation. Rhythm generation is generally considered to be dependent on glutamatergic excitatory neurons. Pattern generation involves neural circuits controlling left-right alternation, which has been described in great detail, and flexor-extensor alternation, which remains poorly understood. Here, we use a mouse model in which glutamatergic neurotransmission has been ablated in the locomotor region of the spinal cord. The isolated in?vitro spinal cord from these mice produces locomotor-like activity-when stimulated with neuroactive substances-with prominent flexor-extensor alternation. Under these conditions, unlike in control mice, networks of inhibitory interneurons generate the rhythmic activity. In the absence of glutamatergic synaptic transmission, the flexor-extensor alternation appears to be generated by Ia inhibitory interneurons, which mediate reciprocal inhibition from muscle proprioceptors to antagonist motor neurons. Our study defines a minimal inhibitory network that is needed to produce flexor-extensor alternation during locomotion.     

A neural network model for the intersensory coordination involved in goal-directed movements

A neural network model for a sensorimotor system, which was developed to simulate oriented movements in man, is presented. It is composed of a formal neural network comprising two layers: a sensory layer receiving and processing sensory inputs, and a motor layer driving a simulated arm. The sensory layer is an extension of the topological network previously proposed by Kohonen (1984). Two kinds of sensory modality, proprioceptive and exteroceptive, are used to define the arm position. Each sensory cell receives proprioceptive inputs provided by each arm-joint together with the exteroceptive inputs. This sensory layer is therefore a kind of associative layer which integrates two separate sensory signals relating to movement coding. It is connected to the motor layer by means of adaptive synapses which provide a physical link between a motor activity and its sensory consequences. After a learning period, the spatial map which emerges in the sensory layer clearly depends on the sensory inputs and an associative map of both the arm and the extra-personal space is built up if proprioceptive and exteroceptive signals are processed together. The sensorimotor transformations occuring in the junctions linking the sensory and motor layers are organized in such a manner that the simulated arm becomes able to reach towards and track a target in extra-personal space. Proprioception serves to determine the final arm posture adopted and to correct the ongoing movement in cases where changes in the target location occur. With a view to developing a sensorimotor control system with more realistic salient features, a robotic model was coupled with the formal neural network. This robotic implementation of our model shows the capacity of formal neural networks to control the displacement of mechanical devices.     

Intersegmental transfer of sensory signals in the stick insect leg muscle control system

Intersegmental coordination during locomotion in legged animals arises from mechanical couplings and the exchange of neuronal information between legs. Here, the information flow from a single leg sense organ of the stick insect Cuniculina impigra onto motoneurons and interneurons of other legs was investigated. The femoral chordotonal organ (fCO) of the right middle leg, which measures posture and movement of the femur-tibia joint, was stimulated, and the responses of the tibial motoneuron pools of the other legs were recorded. In resting animals, fCO signals did not affect motoneuronal activity in neighboring legs. When the locomotor system was activated and antagonistic motoneurons were bursting in alternation, fCO stimuli facilitated transitions from flexor to extensor activity and vice versa in the contralateral leg. Following pharmacological treatment with picrotoxin, a blocker of GABA-ergic inhibition, the tibial motoneurons of all legs showed specific responses to signals from the middle leg fCO. For the contralateral middle leg we show that fCO signals encoding velocity and position of the tibia were processed by those identified local premotor nonspiking interneurons known to contribute to posture and movement control during standing and voluntary leg movements. Interneurons received both excitatory and inhibitory inputs, so that the response of some interneurons supported the motoneuronal output, while others opposed it. Our results demonstrate that sensory information from the fCO specifically affects the motoneuronal activity of other legs and that the layer of premotor nonspiking interneurons is a site of interaction between local proprioceptive sensory signals and proprioceptive signals from other legs.     

Simulation system of spinal cord motor nuclei and associated nerves and muscles,in a Web-based architecture

A Web-based simulation system of the spinal cord circuitry responsible for muscle control is described. The simulator employs two-compartment motoneuron models for S, FR and FF types, with synaptic inputs acting through conductance variations. Four motoneuron pools with their associated interneurons are represented in the simulator, with the possibility of inclusion of more than 2,000 neurons and 2,000,000 synapses. Each motoneuron action potential is followed, after a conduction delay, by a motor unit potential and a motor unit twitch. The sums of all motor unit potentials and twitches result in the electromyogram (EMG), and the muscle force, respectively. Inputs to the motoneuron pool come from populations of interneurons (Ia reciprocal inhibitory interneurons, Ib interneurons, and Renshaw cells) and from stochastic point processes associated with descending tracts. To simulate human electrophysiological experiments, the simulator incorporates external nerve stimulation with orthodromic and antidromic propagation. This provides the mechanisms for reflex generation and activation of spinal neuronal circuits that modulate the activity of another motoneuron pool (e.g., by reciprocal inhibition). The generation of the H-reflex by the Ia-motoneuron pool system and its modulation by spinal cord interneurons is included in the simulation system. Studies with the simulator may include the statistics of individual motoneuron or interneuron spike trains or the collective effect of a motor nucleus on the dynamics of muscle force control. Properties associated with motor-unit recruitment, motor-unit synchronization, recurrent inhibition and reciprocal inhibition may be investigated.     

Role of biomechanics and muscle activation strategy in the production of endpoint force patterns in the cat hindlimb

We used a musculoskeletal model of the cat hindlimb to compare the patterns of endpoint forces generated by all possible combination of 12 hindlimb muscles under three different muscle activation rules: homogeneous activation of muscles based on uniform activation levels, homogeneous activation of muscles based on uniform (normalized) force production, and activation based on the topography of spinal motoneuron pools. Force patterns were compared with the patterns obtained experimentally by microstimulation of the lumbar spinal cord in spinal intact cats. Magnitude and orientation of the force patterns were compared, as well as the proportion of the types found, and the proportions of patterns exhibiting points of zero force (equilibrium points). The force patterns obtained with the homogenous activation and motoneuron topography models were quite similar to those measured experimentally, with the differences being larger for the patterns from the normalized endpoint forces model. Differences in the proportions of types of force patterns between the three models and the experimental results were significant for each model. Both homogeneous activation and normalized endpoint force models produced similar proportions of equilibrium points as found experimentally. The results suggest that muscle biomechanics play an important role in limiting the number of endpoint force pattern types, and that muscle combinations activated at similar levels reproduced best the experimental results obtained with intraspinal microstimulation.     

Traveling-wave pattern generator controls movement and organization of sensory feedback in a spinal cord model

 A traveling wave in a two-dimensional spinal cord model constitutes a stable pattern generator for quadruped gaits. In the context of the somatotopic organization of the spinal cord, this pattern generator is sufficient to generate stable locomotive limb trajectories. The elastic properties of muscles alone, providing linear negative feedback, are sufficient to stabilize stance and locomotion in the presence of perturbative forces. We further show that such a pattern generator is capable of organizing sensory processing in the spinal cord. A single-layer perceptron was trained to associate the sensory feedback from the limb (coding force, length, and change of length for each muscle) with the two-dimensional activity profile of the traveling wave. This resulted in a well-defined spatial organization of the connections within the spinal network along a rostrocaudal axis. The spinal network driven by peripheral afferents alone supported autonomous locomotion in the positive feedback mode, whereas in the negative feedback mode stance was stabilized in response to perturbations. Systematic variation of a parameter representing the effect of gamma-motor neurons on muscle spindle activity in our model led to a corresponding shift of limb position during stance and locomotion, resulting in a systematic displacement alteration of foot positions. Received: 30 July 2001 / Accepted in revised form: 17 April 2002 Correspondence to: A. Kaske (e-mails: alexander.kaske@mtc.ki.se, alexander.kaske@vglab.com)     

Lack of evidence for cell death among avian spinal cord interneurons during normal development and following removal of targets and afferents.

Chick embryos and posthatched chicks were examined at several ages for the presence of pyknotic interneurons in the lumbar spinal cord. Because no pyknotic interneurons were found, direct cell counts of healthy interneurons were carried out and a comparison made between early- and late-stage embryos and hatchlings. There was no decrease in the number of interneurons in the ventral intermediate gray matter of the spinal cord between embryonic day (E) 8 and 2 weeks posthatching (PH) or in the dorsal horn between E10 and 2 weeks PH. To study whether interneuron survival is regulated by targets or afferents, a situation known to exist in other developing neural populations, early embryos were subjected to (1) removal of one limb, resulting in the loss of lateral motor column motoneurons and dorsal root ganglion sensory afferents; (2) transection of the thoracic spinal cord, thereby removing both descending afferents and rostral targets of spinal interneurons, or (3) a combination of the two operations. No reductions in interneuron numbers were found as a result of these operations. Furthermore, morphometric analysis also revealed no change in neuronal size following these experimental manipulations. By contrast, there was a slight decrease in the total area of spinal gray matter that was most prominent in the dorsal region following limb bud removal. Our results indicate (1) that spinal interneurons fail to exhibit the massive naturally occurring death of postmitotic neurons that has been observed for several other populations of spinal neurons, and (2) spinal interneurons appear to be relatively resistant to induced cell death following the removal of substantial numbers of afferent inputs and targets.     

Intersegmental coordination in the lamprey: simulations using a network model without segmental boundaries

Swimming in vertebrates such as eel and lamprey involves the coordination of alternating left and right activity in each segment. Forward swimming is achieved by a lag between the onset of activity in consecutive segments rostrocaudally along the spinal cord. The intersegmental phase lag is approximately 1% of the cycle duration per segment and is independent of the swimming frequency. Since the lamprey has approximately 100 spinal segments, at any given time one wave of activity is propagated along the body. Most previous simulations of intersegmental coordination in the lamprey have treated the cord as a chain of coupled oscillators or well-defined segments. Here a network model without segmental boundaries is described which can produce coordinated activity with a phase lag. This ‘continuous’ pattern-generating network is composed of a column of 420 excitatory interneurons (E1 to E420) and 300 inhibitory interneurons (C1 to C300) on each half of the simulated spinal cord. The interneurons are distributed evenly along the simulated spinal cord, and their connectivity is chosen to reflect the behavior of the intact animal and what is known about the length and strength of the synaptic connections. For example, E100 connects to all interneurons between E51 and E149, but at varying synaptic strengths, while E101 connects to all interneurons between E52 and E150. This unsegmented E-C network generates a motor pattern that is sampled by output elements similar to motoneurons (M cells), which are arranged along the cell column so that they receive input from seven E and five C interneurons. The M cells thus represent the summed excitatory and inhibitory input at different points along the simulated spinal cord and can be regarded as representing the ventral root output to the myotomes along the spinal cord. E and C interneurons have five simulated compartments and Hodgkin-Huxley based dynamics. The simulated network produces rhythmic output over a wide range of frequencies (1–11 Hz) with a phase lag constant over most of the length, with the exception of the ‘cut’ ends due to reduced synaptic input. As the inhibitory C interneurons in the simulation have more extensive caudal than rostral projections, the output of the simulation has positive phase lags, as occurs in forward swimming. However, unlike the biological network, phase lags in the simulation increase significantly with burst frequency, from 0.5% to 2.3% over the range of frequencies of the simulation. Local rostral or caudal increases in excitatory drive in the simulated network are sufficient to produce motor patterns with increased or decreased phase lags, respectively. Received: 15 December 1995 / Accepted in revised form: 17 September 1996