Spinal circuits for motor learning

The spinal cord facilitates cerebellar upper limb motor learning and control; inputs from neuromusculoskeletal simulation

Complex interactions between brain regions and the spinal cord (SC) govern body motion, which is ultimately driven by muscle activation. Motor planning or learning are mainly conducted at higher brain regions, whilst the SC acts as a brain-muscle gateway and as a motor control centre providing fast reflexes and muscle activity regulation. Thus, higher brain areas need to cope with the SC as an inherent and evolutionary older part of the body dynamics. Here, we address the question of how SC dynamics affects motor learning within the cerebellum; in particular, does the SC facilitate cerebellar motor learning or constitute a biological constraint? We provide an exploratory framework by integrating biologically plausible cerebellar and SC computational models in a musculoskeletal upper limb control loop. The cerebellar model, equipped with the main form of cerebellar plasticity, provides motor adaptation; whilst the SC model implements stretch reflex and reciprocal inhibition between antagonist muscles. The resulting spino-cerebellar model is tested performing a set of upper limb motor tasks, including external perturbation studies. A cerebellar model, lacking the implemented SC model and directly controlling the simulated muscles, was also tested in the same. The performances of the spino-cerebellar and cerebellar models were then compared, thus allowing directly addressing the SC influence on cerebellar motor adaptation and learning, and on handling external motor perturbations. Performance was assessed in both joint and muscle space, and compared with kinematic and EMG recordings from healthy participants. The differences in cerebellar synaptic adaptation between both models were also studied. We conclude that the SC facilitates cerebellar motor learning; when the SC circuits are in the loop, faster convergence in motor learning is achieved with simpler cerebellar synaptic weight distributions. The SC is also found to improve robustness against external perturbations, by better reproducing and modulating muscle cocontraction patterns.

Pathophysiology of Dyt1-Tor1a dystonia in mice is mediated by spinal neural circuit dysfunction

Dystonia, a neurological disorder defined by abnormal postures and disorganized movements, is considered to be a neural circuit disorder with dysfunction arising within and between multiple brain regions. Given that spinal neural circuits constitute the final pathway for motor control, we sought to determine their contribution to this movement disorder. Focusing on the most common inherited form of dystonia in humans, DYT1-TOR1A, we generated a conditional knockout of the torsin family 1 member A (Tor1a) gene in the mouse spinal cord and dorsal root ganglia (DRG). We found that these mice recapitulated the phenotype of the human condition, developing early-onset generalized torsional dystonia. Motor signs emerged early in the mouse hindlimbs before spreading caudo-rostrally to affect the pelvis, trunk, and forelimbs throughout postnatal maturation. Physiologically, these mice bore the hallmark features of dystonia, including spontaneous contractions at rest and excessive and disorganized contractions, including cocontractions of antagonist muscle groups, during voluntary movements. Spontaneous activity, disorganized motor output, and impaired monosynaptic reflexes, all signs of human dystonia, were recorded from isolated mouse spinal cords from these conditional knockout mice. All components of the monosynaptic reflex arc were affected, including motor neurons. Given that confining the Tor1a conditional knockout to DRG did not lead to early-onset dystonia, we conclude that the pathophysiological substrate of this mouse model of dystonia lies in spinal neural circuits. Together, these data provide new insights into our current understanding of dystonia pathophysiology.

Upper and lower motor neuron neurophysiology and motor control

This chapter considers the principles that underlie neurophysiological studies of upper motor neuron or lower motor neuron lesions, based on an understanding of the normal structure and function of the motor system. Human motor neurophysiology consists of an evaluation of the active components of the motor system that are relevant to volitional movements. Relatively primitive motor skills include locomotion, much dependent on the spinal cord central pattern generator, reaching, involving proximal and distal muscles activation, and grasping. Humans are well prepared to perform complex movements like writing. The role of motor cortex is critical for the motor activity, very dependent on the continuous sensory feedback, and this is essential for adapting the force and speed control, which contributes to motor learning. Most corticospinal neurons in the brain project to brainstem and spinal cord, many with polysynaptic inhibitory rather than excitatory connections. The monosynaptic connections observed in humans and primates constitute a specialized pathway implicated in fractional finger movements. Spinal cord has a complex physiology, and local reflexes and sensory feedback are essential to control adapted muscular contraction during movement. The cerebellum has a major role in motor coordination, but also consistent roles in sensory activities, speech, and language, in motor and spatial memory, and in psychological activity. The motor unit is the final effector of the motor drive. The complex interplay between the lower motor neuron, its axon, motor end-plates, and muscle fibers allows a relevant plasticity in the movement output.

ERR2 and ERR3 promote the development of gamma motor neuron functional properties required for proprioceptive movement control

The ability of terrestrial vertebrates to effectively move on land is integrally linked to the diversification of motor neurons into types that generate muscle force (alpha motor neurons) and types that modulate muscle proprioception, a task that in mammals is chiefly mediated by gamma motor neurons. The diversification of motor neurons into alpha and gamma types and their respective contributions to movement control have been firmly established in the past 7 decades, while recent studies identified gene expression signatures linked to both motor neuron types. However, the mechanisms that promote the specification of gamma motor neurons and/or their unique properties remained unaddressed. Here, we found that upon selective loss of the orphan nuclear receptors ERR2 and ERR3 (also known as ERRβ, ERRγ or NR3B2, NR3B3, respectively) in motor neurons in mice, morphologically distinguishable gamma motor neurons are generated but do not acquire characteristic functional properties necessary for regulating muscle proprioception, thus disrupting gait and precision movements. Complementary gain-of-function experiments in chick suggest that ERR2 and ERR3 could operate via transcriptional activation of neural activity modulators to promote a gamma motor neuron biophysical signature of low firing thresholds and high firing rates. Our work identifies a mechanism specifying gamma motor neuron functional properties essential for the regulation of proprioceptive movement control.

Chemogenetic modulation of sensory afferents induces locomotor changes and plasticity after spinal cord injury

Neuromodulatory therapies for spinal cord injury (SCI) such as electrical epidural stimulation (EES) are increasingly effective at improving patient outcomes. These improvements are thought to be due, at least in part, to plasticity in neuronal circuits. Precisely which circuits are influenced and which afferent classes are most effective in stimulating change remain important open questions. Genetic tools, such as Designer Receptors Exclusively Activated by Designer Drugs (DREADDs), support targeted and reversible neuromodulation as well as histological characterization of manipulated neurons. We therefore transduced and activated lumbar large diameter peripheral afferents with excitatory (hM3Dq) DREADDs, in a manner analogous to EES, in a rat hemisection model, to begin to trace plasticity and observe concomitant locomotor changes. Chronic DREADDs activation, coupled with thrice weekly treadmill training, was observed to increase afferent fluorescent labeling within motor pools and Clarke's column when compared to control animals. This plasticity may underlie kinematic differences that we observed across stages of recovery, including an increased and less variable hindquarters height in DREADDs animals, shorter step durations, a more flexed ankle joint early in recovery, a less variable ankle joint angle in swing phase, but a more variable hip joint angle. Withdrawal of DREADDs agonist, clozapine-N-oxide (CNO) left these kinematic differences largely unaffected; suggesting that DREADDs activation is not necessary for them later in recovery. However, we observed an intermittent “buckling” phenomenon in DREADDs animals without CNO activation, that did not occur with CNO re-administration. Future studies could use more refined genetic targeted of specific afferent classes, and utilize muscle recordings to find where afferent modulation is most influential in altering motor output.

Synaptic Projections of Motoneurons Within the Spinal Cord

Motoneurons have long been considered as the final common pathway of the nervous system, transmitting the neural impulses that are transduced into action.While many studies have focussed on the inputs that motoneurons receive from local circuits within the spinal cord and from other parts of the CNS, relatively few have investigated the targets of local axonal projections from motoneurons themselves, with the notable exception of those contacting Renshaw cells or other motoneurons.Recent research has not only characterised the detailed features of the excitatory connections between motoneurons and Renshaw cells but has also established that Renshaw cells are not the only target of motoneurons axons within the spinal cord. Motoneurons also form synaptic contacts with other motoneurons as well as with a subset of ventrally located V3 interneurons. These findings indicate that motoneurons cannot be simply viewed as the last relay station delivering the command drive to muscles, but perform an active role in the generation and modulation of motor patterns.

Epidural Electrical Stimulation: A Review of Plasticity Mechanisms That Are Hypothesized to Underlie Enhanced Recovery From Spinal Cord Injury With Stimulation

Spinal cord injury (SCI) often results in life-long sensorimotor impairment. Spontaneous recovery from SCI is limited, as supraspinal fibers cannot spontaneously regenerate to form functional networks below the level of injury. Despite this, animal models and humans exhibit many motor behaviors indicative of recovery when electrical stimulation is applied epidurally to the dorsal aspect of the lumbar spinal cord. In 1976, epidural stimulation was introduced to alleviate spasticity in Multiple Sclerosis. Since then, epidural electrical stimulation (EES) has been demonstrated to improve voluntary mobility across the knee and/or ankle in several SCI patients, highlighting its utility in enhancing motor activation. The mechanisms that EES induces to drive these improvements in sensorimotor function remain largely unknown. In this review, we discuss several sensorimotor plasticity mechanisms that we hypothesize may enable epidural stimulation to promote recovery, including changes in local lumbar circuitry, propriospinal interneurons, and the internal model. Finally, we discuss genetic tools for afferent modulation as an emerging method to facilitate the search for the mechanisms of action.

The recovery of standing and locomotion after spinal cord injury does not require task-specific training

After complete spinal cord injury, mammals, including mice, rats and cats, recover hindlimb locomotion with treadmill training. The premise is that sensory cues consistent with locomotion reorganize spinal sensorimotor circuits. Here, we show that hindlimb standing and locomotion recover after spinal transection in cats without task-specific training. Spinal-transected cats recovered full weight bearing standing and locomotion after five weeks of rhythmic manual stimulation of triceps surae muscles (non-specific training) and without any intervention. Moreover, cats modulated locomotor speed and performed split-belt locomotion six weeks after spinal transection, functions that were not trained or tested in the weeks prior. This indicates that spinal networks controlling standing and locomotion and their interactions with sensory feedback from the limbs remain largely intact after complete spinal cord injury. We conclude that standing and locomotor recovery is due to the return of neuronal excitability within spinal sensorimotor circuits that do not require task-specific activity-dependent plasticity.

Pioneers in CNS inhibition: 2. Charles Sherrington and John Eccles on inhibition in spinal and supraspinal structures

This article reviews the contributions of the English neurophysiologist, Charles Scott Sherrington [1857-1952], and his Australian PhD trainee and collaborator, John Carew Eccles [1903-1997], to the concept of central inhibition in the spinal cord and brain. Both were awarded Nobel Prizes; Sherrington in 1932 for "discoveries regarding the function of neurons," and Eccles in 1963 for "discoveries concerning the ionic mechanisms involved in excitation and inhibition in central portions of the nerve cell membrane." Both spoke about central inhibition at their Nobel Prize Award Ceremonies. The subsequent publications of their talks were entitled "Inhibition as a coordinative factor" and "The ionic mechanism of postsynaptic inhibition", respectively. Sherrington's work on central inhibition spanned 41 years (1893-1934), and for Eccles 49 years (1928-1977). Sherrington first studied central inhibition by observing hind limb muscle responses to electrical (peripheral nerve) and mechanical (muscle) stimulation. He used muscle length and force measurements until the early 1900s and electromyography in the late 1920s. Eccles used these techniques while working with Sherrington, but later employed extracellular microelectrode recording in the spinal cord followed in 1951 by intracellular recording from spinal motoneurons. This considerably advanced our understanding of central inhibition. Sherrington's health was poor during his retirement years but he nonetheless made a small number of largely humanities contributions up to 1951, one year before his death at the age of 94. In contrast, Eccles retained his health and vigor until 3 years before his death and published prolifically on many subjects during his 22 years of official retirement. His last neuroscience article appeared in 1994 when he was 91. Despite poor health he continued thinking about his life-long interest, the mind-brain problem, and was attempting to complete his autobiography in the last years of his life.

Proprioception: Bottom-up directive for motor recovery after spinal cord injury

Proprioceptive feedback provides movement-matched sensory information essential for motor control and recovery after spinal cord injury. While it is understood that the fundamental contribution of proprioceptive feedback circuits in locomotor recovery is to activate the local spinal cord interneurons and motor neurons in a context-dependent manner, the precise mechanisms by which proprioception enables motor recovery after a spinal cord injury remain elusive. Furthermore, how proprioception contributes to motor learning mechanisms intrinsic to spinal cord networks and gives rise to motor recovery is currently unknown. This review discusses the existence of motor learning mechanisms intrinsic to spinal cord circuits and circuit-level insights on how proprioception might contribute to spinal cord plasticity, adaptability, and learning, in addition to the logic in which proprioception helps to establish an internal motor command to execute motor output using spared circuits after a spinal cord injury.

V2a interneuron diversity tailors spinal circuit organization to control the vigor of locomotor movements

Locomotion is a complex motor task generated by spinal circuits driving motoneurons in a precise sequence to control the timing and vigor of movements, but the underlying circuit logic remains to be understood. Here we reveal, in adult zebrafish, how the diversity and selective distribution of two V2a interneuron types within the locomotor network transform commands into an appropriate, task-dependent circuit organization. Bursting-type V2a interneurons with unidirectional axons predominantly target distal dendrites of slow motoneurons to provide potent, non-linear excitation involving NMDA-dependent potentiation. A second type, non-bursting V2a interneurons with bidirectional axons, predominantly target somata of fast motoneurons, providing weaker, non-potentiating excitation. Together, this ensures the rapid, first-order recruitment of the slow circuit, while reserving the fast circuit for highly salient stimuli involving synchronous inputs. Our results thus identify how interneuron diversity is captured and transformed into a parsimonious task-specific circuit design controlling the vigor of locomotion.

Brain pathology is related to total daily physical activity in older adults

OBJECTIVE

To test the hypothesis that brain pathology is associated with total daily physical activity proximate to death in older adults.

METHODS

We studied brain autopsies from 428 decedents of the Rush Memory and Aging Project. The quantity of all physical activity was measured continuously for up to 10 days with actigraphy (Actical; Philips Healthcare, Bend, OR). Multiple regression analyses controlling for age and sex were used to examine the relation of brain indexes to total daily physical activity and other clinical covariates proximate to death.

RESULTS

Average total daily activity was 1.53 × 105 counts/d (SD 1.14 × 105 counts/d), and mean age at death was 90.6 (SD 6.12) years. Nigral neuronal loss (estimate -0.232, standard error [SE] = 0.070, p = 0.001) and macroinfarcts (estimate -0.266, SE 0.112, p = 0.017) were independently associated with total daily physical activity proximate to death, accounting for an additional 2.4% of the variance of total daily activity. Other postmortem indexes (Alzheimer disease, Lewy bodies, TAR DNA-binding protein 43, hippocampal sclerosis, microinfarcts, atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy) were not associated with total daily activity. In 295 cases (70%), we derived a measure of white matter tissue integrity from postmortem brain MRI. This metric accounted for an additional 5.8% of the variance of total daily physical activity when controlling for age, sex, nigral neuronal loss, and macroinfarcts.

CONCLUSION

Macroinfarcts, nigral neuronal loss, and white matter pathologies are related to total daily physical activity in older adults, but further studies are needed to explain its pathologic basis more fully.

Evidence for startle as a measurable behavioral indicator of motor learning

The ability of the classic startle reflex to evoke voluntarily prepared movement involuntarily has captured the attention of neuroscientists for its wide-ranging functional utility and potential uses in patient populations. To date, there is only one documented task resistant to the startReact phenomenon-index finger abduction. Previous reports have suggested the lack of startReact is due to different neural mechanisms driving individuated finger movement and more proximal joint control (e.g. elbow, wrist movement). However, an alternative hypothesis exists. Though not particularly difficult to execute, isolated index finger abduction is rarely performed during activities of daily living and is not a natural correlate to common individuated finger tasks. We propose that startReact can be evoked during individuated finger movements but only during tasks that are highly trained or familiar. The objective of this study was to determine the impact of a 2-week training regimen on the ability to elicit startReact. We found evidence in support of our hypothesis that following training, individuated movements of the hands (specifically index finger abduction) become susceptible to startReact. This is significant not only because it indicates that individuated finger movements are in fact amenable to startReact, but also that startle has differential response characteristics in novel tasks compared to highly trained tasks suggesting that startle is a measurable behavioral indicator of motor learning.

Non-functionalized soft alginate hydrogel promotes locomotor recovery after spinal cord injury in a rat hemimyelonectomy model

BACKGROUND

Spinal cord injury (SCI) and the consecutive devastating neurological sequelae have an enormous individual and economic impact. Implantation of functionalized hydrogels is a promising approach, because they can serve as a matrix for the regenerating tissue, carry and release bioactive molecules and various cell types. We already demonstrated that non-functionalized soft alginate hydrogel supported axonal outgrowth and protected neurons against oxidative stress in vitro. Here, we investigated the effects of such soft alginate hydrogels on locomotor recovery in small and large spinal cord lesions.

METHOD

Hemimyelonectomy of 2 mm or 4 mm length was performed in rats and soft alginate hydrogel was implanted. Functional recovery of the hindlimbs was assessed in the open field [Batto Beattie Bresnahan (BBB) score] and using swimming test [Louisville Swim score (LSS)] for 140 days post injury (DPI). Reference histology was performed.

RESULTS

Rats that received an alginate implant into 2 mm spinal cord lesions demonstrated significantly improved locomotor recovery compared to controls detectable already at 10 DPI. At 140 DPI, they reached higher LSS and BBB scores in swimming and open field tests, respectively. However, this beneficial effect of alginate was lacking in animals with larger (4 mm) lesions. Histological examination suggested that fibrous scarring in the spinal cord was reduced after alginate implantation in comparison to controls.

CONCLUSIONS

Implantation of soft alginate hydrogel in small spinal cord lesions improved functional recovery. Possible underlying mechanisms include the mechanical stabilization of the wound, reduction of secondary damage and inhibition of fibrous scarring.

Escape from homeostasis: spinal microcircuits and progression of amyotrophic lateral sclerosis

In amyotrophic lateral sclerosis (ALS), loss of motoneuron function leads to weakness and, ultimately, respiratory failure and death. Regardless of the initial pathogenic factors, motoneuron loss follows a specific pattern: the largest α-motoneurons die before smaller α-motoneurons, and γ-motoneurons are spared. In this article, we examine how homeostatic responses to this orderly progression could lead to local microcircuit dysfunction that in turn propagates motoneuron dysfunction and death. We first review motoneuron diversity and the principle of α-γ coactivation and then discuss two specific spinal motoneuron microcircuits: those involving proprioceptive afferents and those involving Renshaw cells. Next, we propose that the overall homeostatic response of the nervous system is aimed at maintaining force output. Thus motoneuron degeneration would lead to an increase in inputs to motoneurons, and, because of the pattern of neuronal degeneration, would result in an imbalance in local microcircuit activity that would overwhelm initial homeostatic responses. We suggest that this activity would ultimately lead to excitotoxicity of motoneurons, which would hasten the progression of disease. Finally, we propose that should this be the case, new therapies targeted toward microcircuit dysfunction could slow the course of ALS.

The level of performance stabilization influences motor adaptation on an isometric force control task

This study compared the effects of two levels of performance stabilization on the adaptation to unpredictable perturbations in an isometric control force task with the goal of controlling 40% of the maximum force. The experiment consisted of pre-exposure and exposure phases. In the pre-exposure two levels of performance stabilization were manipulated: a stabilization group (SG) performed three trials in a row while maintaining 40% of the maximum force for three seconds and an absolute error less than or equal to 5% (i.e., the criteria of performance), and a superstabilization group (SSG) performed six blocks of the same criteria. During the exposure phase, the task was the same as that in the pre-exposure phase; however, it was inserted 9 trials of perturbations when the task goal changed to 60% of the maximum force. We measured the %RMSE, RMS from the biceps and triceps brachii and co-contraction. In the pre-exposure phase, both groups showed similar performance and muscle activity. When exposed to the perturbations, SSG performance more quickly returned to the previous level of accuracy, showed lower muscle activation and demonstrated a greater muscle co-contraction than did SG. The results give support to the adaptive process model on motor learning.

Motor learning characterization in people with autism spectrum disorder: A systematic review

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder primarily characterized by deficits in social interaction, communication and implicit skill learning.

Lack of adaptation during prolonged split-belt locomotion in the intact and spinal cat

KEY POINTS

During split-belt locomotion in humans where one leg steps faster than the other, the symmetry of step lengths and double support periods of the slow and fast legs is gradually restored. When returning to tied-belt locomotion, there is an after-effect, with a reversal in the asymmetry observed in the early split-belt period, indicating that the new pattern was stored within the central nervous system. In this study, we investigated if intact and spinal-transected cats show a similar pattern of adaptation to split-belt locomotion by measuring kinematic variables and electromyography before, during and after 10 min of split-belt locomotion. The results show that cats do not adapt to prolonged split-belt locomotion. Our results suggest an important physiological difference in how cats and humans respond to prolonged asymmetric locomotion.

ABSTRACT

In humans, gait adapts to prolonged walking on a split-belt treadmill, where one leg steps faster than the other, by gradually restoring the symmetry of interlimb kinematic variables, such as double support periods and step lengths, and by reducing muscle activity (EMG, electromyography). The adaptation is also characterized by reversing the asymmetry of interlimb variables observed during the early split-belt period when returning to tied-belt locomotion, termed an after-effect. To determine if cats adapt to prolonged split-belt locomotion and to assess if spinal locomotor circuits participate in the adaptation, we measured interlimb variables and EMG in intact and spinal-transected cats before, during and after 10 min of split-belt locomotion. In spinal cats, only the hindlimbs performed stepping with the forelimbs stationary. In intact and spinal cats, step lengths and double support periods were, on average, symmetric, during tied-belt locomotion. They became asymmetric during split-belt locomotion and remained asymmetric throughout the split-belt period. Upon returning to tied-belt locomotion, symmetry was immediately restored. In intact cats, the mean EMG amplitude of hindlimb extensors increased during split-belt locomotion and remained increased throughout the split-belt period, whereas in spinal cats, EMG amplitude did not change. Therefore, the results indicate that the locomotor pattern of cats does not adapt to prolonged split-belt locomotion, suggesting an important physiological difference in the control of locomotion between cats and humans. We propose that restoring left-right symmetry is not required to maintain balance during prolonged asymmetric locomotion in the cat, a quadruped, as opposed to human bipedal locomotion.

Spinal control of motor outputs by intrinsic and externally induced electric field potentials

Despite numerous studies on spinal neuronal systems, several issues regarding their role in motor behavior remain unresolved. One of these issues is how electric fields associated with the activity of spinal neurons influence the operation of spinal neuronal networks and how effects of these field potentials are combined with other means of modulating neuronal activity. Another closely related issue is how external electric field potentials affect spinal neurons and how they can be used for therapeutic purposes such as pain relief or recovery of motor functions by transspinal direct current stimulation. Nevertheless, progress in our understanding of the spinal effects of electric fields and their mechanisms has been made over the last years, and the aim of the present review is to summarize the recent findings in this field.

Perception of Gait Asymmetry During Split-Belt Walking

Optimization of gait rehabilitation using split-belt treadmills critically depends on our understanding of the roles of somatosensory perception and sensorimotor recalibration in perceiving gait asymmetry and adapting to split-belt walking. Recent evidence justifies the hypothesis that perception of gait asymmetry is based mainly on detection of temporal mismatches between afferent inputs at the spinal level.

Spinal microcircuits comprising dI3 interneurons are necessary for motor functional recovery following spinal cord transection

The spinal cord has the capacity to coordinate motor activities such as locomotion. Following spinal transection, functional activity can be regained, to a degree, following motor training. To identify microcircuits involved in this recovery, we studied a population of mouse spinal interneurons known to receive direct afferent inputs and project to intermediate and ventral regions of the spinal cord. We demonstrate that while dI3 interneurons are not necessary for normal locomotor activity, locomotor circuits rhythmically inhibit them and dI3 interneurons can activate these circuits. Removing dI3 interneurons from spinal microcircuits by eliminating their synaptic transmission left locomotion more or less unchanged, but abolished functional recovery, indicating that dI3 interneurons are a necessary cellular substrate for motor system plasticity following transection. We suggest that dI3 interneurons compare inputs from locomotor circuits with sensory afferent inputs to compute sensory prediction errors that then modify locomotor circuits to effect motor recovery.

Toward an Integration of Deep Learning and Neuroscience

Neuroscience has focused on the detailed implementation of computation, studying neural codes, dynamics and circuits. In machine learning, however, artificial neural networks tend to eschew precisely designed codes, dynamics or circuits in favor of brute force optimization of a cost function, often using simple and relatively uniform initial architectures. Two recent developments have emerged within machine learning that create an opportunity to connect these seemingly divergent perspectives. First, structured architectures are used, including dedicated systems for attention, recursion and various forms of short- and long-term memory storage. Second, cost functions and training procedures have become more complex and are varied across layers and over time. Here we think about the brain in terms of these ideas. We hypothesize that (1) the brain optimizes cost functions, (2) the cost functions are diverse and differ across brain locations and over development, and (3) optimization operates within a pre-structured architecture matched to the computational problems posed by behavior. In support of these hypotheses, we argue that a range of implementations of credit assignment through multiple layers of neurons are compatible with our current knowledge of neural circuitry, and that the brain's specialized systems can be interpreted as enabling efficient optimization for specific problem classes. Such a heterogeneously optimized system, enabled by a series of interacting cost functions, serves to make learning data-efficient and precisely targeted to the needs of the organism. We suggest directions by which neuroscience could seek to refine and test these hypotheses.

Genetically identified spinal interneurons integrating tactile afferents for motor control

Our movements are shaped by our perception of the world as communicated by our senses. Perception of sensory information has been largely attributed to cortical activity. However, a prior level of sensory processing occurs in the spinal cord. Indeed, sensory inputs directly project to many spinal circuits, some of which communicate with motor circuits within the spinal cord. Therefore, the processing of sensory information for the purpose of ensuring proper movements is distributed between spinal and supraspinal circuits. The mechanisms underlying the integration of sensory information for motor control at the level of the spinal cord have yet to be fully described. Recent research has led to the characterization of spinal neuron populations that share common molecular identities. Identification of molecular markers that define specific populations of spinal neurons is a prerequisite to the application of genetic techniques devised to both delineate the function of these spinal neurons and their connectivity. This strategy has been used in the study of spinal neurons that receive tactile inputs from sensory neurons innervating the skin. As a result, the circuits that include these spinal neurons have been revealed to play important roles in specific aspects of motor function. We describe these genetically identified spinal neurons that integrate tactile information and the contribution of these studies to our understanding of how tactile information shapes motor output. Furthermore, we describe future opportunities that these circuits present for shedding light on the neural mechanisms of tactile processing.