Hybrid dedicated and distributed coding in PMd/M1 provides separation and interaction of bilateral arm signals.

Pronounced activity is observed in both hemispheres of the motor cortex during preparation and execution of unimanual movements. The organizational principles of bi-hemispheric signals and the functions they serve throughout motor planning remain unclear. Using an instructed-delay reaching task in monkeys, we identified two components in population responses spanning PMd and M1. A "dedicated" component, which segregated activity at the level of individual units, emerged in PMd during preparation. It was most prominent following movement when M1 became strongly engaged, and principally involved the contralateral hemisphere. In contrast to recent reports, these dedicated signals solely accounted for divergence of arm-specific neural subspaces. The other "distributed" component mixed signals for each arm within units, and the subspace containing it did not discriminate between arms at any stage. The statistics of the population response suggest two functional aspects of the cortical network: one that spans both hemispheres for supporting preparatory and ongoing processes, and another that is predominantly housed in the contralateral hemisphere and specifies unilateral output.

Corticostriatal plasticity is necessary for learning intentional neuroprosthetic skills.

The ability to learn new skills and perfect them with practice applies not only to physical skills but also to abstract skills, like motor planning or neuroprosthetic actions. Although plasticity in corticostriatal circuits has been implicated in learning physical skills, it remains unclear if similar circuits or processes are required for abstract skill learning. Here we use a novel behavioural task in rodents to investigate the role of corticostriatal plasticity in abstract skill learning. Rodents learned to control the pitch of an auditory cursor to reach one of two targets by modulating activity in primary motor cortex irrespective of physical movement. Degradation of the relation between action and outcome, as well as sensory-specific devaluation and omission tests, demonstrate that these learned neuroprosthetic actions are intentional and goal-directed, rather than habitual. Striatal neurons change their activity with learning, with more neurons modulating their activity in relation to target-reaching as learning progresses. Concomitantly, strong relations between the activity of neurons in motor cortex and the striatum emerge. Specific deletion of striatal NMDA receptors impairs the development of this corticostriatal plasticity, and disrupts the ability to learn neuroprosthetic skills. These results suggest that corticostriatal plasticity is necessary for abstract skill learning, and that neuroprosthetic movements capitalize on the neural circuitry involved in natural motor learning.

Robust Brain-Machine Interface Design Using Optimal Feedback Control Modeling and Adaptive Point Process Filtering.

Much progress has been made in brain-machine interfaces (BMI) using decoders such as Kalman filters and finding their parameters with closed-loop decoder adaptation (CLDA). However, current decoders do not model the spikes directly, and hence may limit the processing time-scale of BMI control and adaptation. Moreover, while specialized CLDA techniques for intention estimation and assisted training exist, a unified and systematic CLDA framework that generalizes across different setups is lacking. Here we develop a novel closed-loop BMI training architecture that allows for processing, control, and adaptation using spike events, enables robust control and extends to various tasks. Moreover, we develop a unified control-theoretic CLDA framework within which intention estimation, assisted training, and adaptation are performed. The architecture incorporates an infinite-horizon optimal feedback-control (OFC) model of the brain's behavior in closed-loop BMI control, and a point process model of spikes. The OFC model infers the user's motor intention during CLDA-a process termed intention estimation. OFC is also used to design an autonomous and dynamic assisted training technique. The point process model allows for neural processing, control and decoder adaptation with every spike event and at a faster time-scale than current decoders; it also enables dynamic spike-event-based parameter adaptation unlike current CLDA methods that use batch-based adaptation on much slower adaptation time-scales. We conducted closed-loop experiments in a non-human primate over tens of days to dissociate the effects of these novel CLDA components. The OFC intention estimation improved BMI performance compared with current intention estimation techniques. OFC assisted training allowed the subject to consistently achieve proficient control. Spike-event-based adaptation resulted in faster and more consistent performance convergence compared with batch-based methods, and was robust to parameter initialization. Finally, the architecture extended control to tasks beyond those used for CLDA training. These results have significant implications towards the development of clinically-viable neuroprosthetics.

Representation of Muscle Synergies in the Primate Brain.

Evidence suggests that the CNS uses motor primitives to simplify movement control, but whether it actually stores primitives instead of computing solutions on the fly to satisfy task demands is a controversial and still-unanswered possibility. Also in contention is whether these primitives take the form of time-invariant muscle coactivations ("spatial" synergies) or time-varying muscle commands ("spatiotemporal" synergies). Here, we examined forelimb muscle patterns and motor cortical spiking data in rhesus macaques (Macaca mulatta) handling objects of variable shape and size. From these data, we extracted both spatiotemporal and spatial synergies using non-negative decomposition. Each spatiotemporal synergy represents a sequence of muscular or neural activations that appeared to recur frequently during the animals' behavior. Key features of the spatiotemporal synergies (including their dimensionality, timing, and amplitude modulation) were independently observed in the muscular and neural data. In addition, both at the muscular and neural levels, these spatiotemporal synergies could be readily reconstructed as sequential activations of spatial synergies (a subset of those extracted independently from the task data), suggestive of a hierarchical relationship between the two levels of synergies. The possibility that motor cortex may execute even complex skill using spatiotemporal synergies has novel implications for the design of neuroprosthetic devices, which could gain computational efficiency by adopting the discrete and low-dimensional control that these primitives imply. SIGNIFICANCE STATEMENT: We studied the motor cortical and forearm muscular activity of rhesus macaques (Macaca mulatta) as they reached, grasped, and carried objects of varied shape and size. We applied non-negative matrix factorization separately to the cortical and muscular data to reduce their dimensionality to a smaller set of time-varying "spatiotemporal" synergies. Each synergy represents a sequence of cortical or muscular activity that recurred frequently during the animals' behavior. Salient features of the synergies (including their dimensionality, timing, and amplitude modulation) were observed at both the cortical and muscular levels. The possibility that the brain may execute even complex behaviors using spatiotemporal synergies has implications for neuroprosthetic algorithm design, which could become more computationally efficient by adopting the discrete and low-dimensional control that they afford.

Advances in neuroprosthetic learning and control.

Significant progress has occurred in the field of brain-machine interfaces (BMI) since the first demonstrations with rodents, monkeys, and humans controlling different prosthetic devices directly with neural activity. This technology holds great potential to aid large numbers of people with neurological disorders. However, despite this initial enthusiasm and the plethora of available robotic technologies, existing neural interfaces cannot as yet master the control of prosthetic, paralyzed, or otherwise disabled limbs. Here I briefly discuss recent advances from our laboratory into the neural basis of BMIs that should lead to better prosthetic control and clinically viable solutions, as well as new insights into the neurobiology of action.

Continuous closed-loop decoder adaptation with a recursive maximum likelihood algorithm allows for rapid performance acquisition in brain-machine interfaces.

Closed-loop decoder adaptation (CLDA) is an emerging paradigm for both improving and maintaining online performance in brain-machine interfaces (BMIs). The time required for initial decoder training and any subsequent decoder recalibrations could be potentially reduced by performing continuous adaptation, in which decoder parameters are updated at every time step during these procedures, rather than waiting to update the decoder at periodic intervals in a more batch-based process. Here, we present recursive maximum likelihood (RML), a CLDA algorithm that performs continuous adaptation of a Kalman filter decoder's parameters. We demonstrate that RML possesses a variety of useful properties and practical algorithmic advantages. First, we show how RML leverages the accuracy of updates based on a batch of data while still adapting parameters on every time step. Second, we illustrate how the RML algorithm is parameterized by a single, intuitive half-life parameter that can be used to adjust the rate of adaptation in real time. Third, we show how even when the number of neural features is very large, RML's memory-efficient recursive update rules can be reformulated to also be computationally fast so that continuous adaptation is still feasible. To test the algorithm in closed-loop experiments, we trained three macaque monkeys to perform a center-out reaching task by using either spiking activity or local field potentials to control a 2D computer cursor. RML achieved higher levels of performance more rapidly in comparison to a previous CLDA algorithm that adapts parameters on a more intermediate timescale. Overall, our results indicate that RML is an effective CLDA algorithm for achieving rapid performance acquisition using continuous adaptation.

Dynamic changes of rodent somatosensory barrel cortex are correlated with learning a novel conditioned stimulus.

The rodent somatosensory barrel cortex (S1bf) has proved a valuable model for studying neural plasticity in vivo. It has been observed that sensory deprivation or conditioning reorganizes sensory-driven activity within S1bf. These observations suggest a role for S1bf in somatosensory learning. This study evaluated the hypothesis that the response properties of extracellularly recorded neurons in S1bf would change as subjects learned to respond to stimulation of S1bf. Intracortical microstimulation (ICMS) of S1bf was used as a means for bypassing feedforward drive from the sensory periphery, midbrain, and thalamus while exciting local cortical networks. To separate the learning of this conditioned stimulus-conditioned response (CS-CR) from other elements of the task, we employed a cross-modal transfer schedule. Long-Evans rats were initially trained to respond to an auditory stimulus. All subjects were then implanted in both S1bfs with chronic microwire arrays for recording neural activity and delivering ICMS. Next, this association was transferred to ICMS of one hemisphere's S1bf. S1bf responded to ICMS with a brief increase in firing rate followed by a longer reduction in activity. We observed that the duration of reduced activity elicited by ICMS increased as the subjects began to respond correctly more often than expected by chance, and the magnitude of the initial positive response increased as they consolidated this CS-CR. Subsequent ICMS of the opposite S1bf revealed that this CS-CR did not generalize across hemispheres. These results suggest that a mechanism involving a single hemisphere's S1bf tunes cortical responses in concert with changes in rodent behavior during somatosensory learning.

Design and analysis of closed-loop decoder adaptation algorithms for brain-machine interfaces.

Closed-loop decoder adaptation (CLDA) is an emerging paradigm for achieving rapid performance improvements in online brain-machine interface (BMI) operation. Designing an effective CLDA algorithm requires making multiple important decisions, including choosing the timescale of adaptation, selecting which decoder parameters to adapt, crafting the corresponding update rules, and designing CLDA parameters. These design choices, combined with the specific settings of CLDA parameters, will directly affect the algorithm's ability to make decoder parameters converge to values that optimize performance. In this article, we present a general framework for the design and analysis of CLDA algorithms and support our results with experimental data of two monkeys performing a BMI task. First, we analyze and compare existing CLDA algorithms to highlight the importance of four critical design elements: the adaptation timescale, selective parameter adaptation, smooth decoder updates, and intuitive CLDA parameters. Second, we introduce mathematical convergence analysis using measures such as mean-squared error and KL divergence as a useful paradigm for evaluating the convergence properties of a prototype CLDA algorithm before experimental testing. By applying these measures to an existing CLDA algorithm, we demonstrate that our convergence analysis is an effective analytical tool that can ultimately inform and improve the design of CLDA algorithms.

Task-dependent changes in cross-level coupling between single neurons and oscillatory activity in multiscale networks.

Understanding the principles governing the dynamic coordination of functional brain networks remains an important unmet goal within neuroscience. How do distributed ensembles of neurons transiently coordinate their activity across a variety of spatial and temporal scales? While a complete mechanistic account of this process remains elusive, evidence suggests that neuronal oscillations may play a key role in this process, with different rhythms influencing both local computation and long-range communication. To investigate this question, we recorded multiple single unit and local field potential (LFP) activity from microelectrode arrays implanted bilaterally in macaque motor areas. Monkeys performed a delayed center-out reach task either manually using their natural arm (Manual Control, MC) or under direct neural control through a brain-machine interface (Brain Control, BC). In accord with prior work, we found that the spiking activity of individual neurons is coupled to multiple aspects of the ongoing motor beta rhythm (10-45 Hz) during both MC and BC, with neurons exhibiting a diversity of coupling preferences. However, here we show that for identified single neurons, this beta-to-rate mapping can change in a reversible and task-dependent way. For example, as beta power increases, a given neuron may increase spiking during MC but decrease spiking during BC, or exhibit a reversible shift in the preferred phase of firing. The within-task stability of coupling, combined with the reversible cross-task changes in coupling, suggest that task-dependent changes in the beta-to-rate mapping play a role in the transient functional reorganization of neural ensembles. We characterize the range of task-dependent changes in the mapping from beta amplitude, phase, and inter-hemispheric phase differences to the spike rates of an ensemble of simultaneously-recorded neurons, and discuss the potential implications that dynamic remapping from oscillatory activity to spike rate and timing may hold for models of computation and communication in distributed functional brain networks.

Oscillatory phase coupling coordinates anatomically dispersed functional cell assemblies.

Hebb proposed that neuronal cell assemblies are critical for effective perception, cognition, and action. However, evidence for brain mechanisms that coordinate multiple coactive assemblies remains lacking. Neuronal oscillations have been suggested as one possible mechanism for cell assembly coordination. Prior studies have shown that spike timing depends upon local field potential (LFP) phase proximal to the cell body, but few studies have examined the dependence of spiking on distal LFP phases in other brain areas far from the neuron or the influence of LFP-LFP phase coupling between distal areas on spiking. We investigated these interactions by recording LFPs and single-unit activity using multiple microelectrode arrays in several brain areas and then used a unique probabilistic multivariate phase distribution to model the dependence of spike timing on the full pattern of proximal LFP phases, distal LFP phases, and LFP-LFP phase coupling between electrodes. Here we show that spiking activity in single neurons and neuronal ensembles depends on dynamic patterns of oscillatory phase coupling between multiple brain areas, in addition to the effects of proximal LFP phase. Neurons that prefer similar patterns of phase coupling exhibit similar changes in spike rates, whereas neurons with different preferences show divergent responses, providing a basic mechanism to bind different neurons together into coordinated cell assemblies. Surprisingly, phase-coupling-based rate correlations are independent of interneuron distance. Phase-coupling preferences correlate with behavior and neural function and remain stable over multiple days. These findings suggest that neuronal oscillations enable selective and dynamic control of distributed functional cell assemblies.

Distinct neural representations during a brain-machine interface and manual reaching task in motor cortex, prefrontal cortex, and striatum.

Although brain-machine interfaces (BMIs) are directly controlled by the modulation of a select local population of neurons, distributed networks consisting of cortical and subcortical areas have been implicated in learning and maintaining control. Previous work in rodent BMI has demonstrated the involvement of the striatum in BMI learning. However, the prefrontal cortex has been largely ignored when studying motor BMI control despite its role in action planning, action selection, and learning abstract tasks. Here, we compare local field potentials simultaneously recorded from the primary motor cortex (M1), dorsolateral prefrontal cortex (DLPFC), and the caudate nucleus of the striatum (Cd) while nonhuman primates perform a two-dimensional, self-initiated, center-out task under BMI control and manual control. Our results demonstrate the presence of distinct neural representations for BMI and manual control in M1, DLPFC, and Cd. We find that neural activity from DLPFC and M1 best distinguish between control types at the go cue and target acquisition, respectively. We also found effective connectivity from DLPFC→M1 throughout trials across both control types and Cd→M1 during BMI control. These results suggest distributed network activity between M1, DLPFC, and Cd during BMI control that is similar yet distinct from manual control.

Invariant neural dynamics drive commands to control different movements.

It has been proposed that the nervous system has the capacity to generate a wide variety of movements because it reuses some invariant code. Previous work has identified that dynamics of neural population activity are similar during different movements, where dynamics refer to how the instantaneous spatial pattern of population activity changes in time. Here, we test whether invariant dynamics of neural populations are actually used to issue the commands that direct movement. Using a brain-machine interface (BMI) that transforms rhesus macaques' motor-cortex activity into commands for a neuroprosthetic cursor, we discovered that the same command is issued with different neural-activity patterns in different movements. However, these different patterns were predictable, as we found that the transitions between activity patterns are governed by the same dynamics across movements. These invariant dynamics are low dimensional, and critically, they align with the BMI, so that they predict the specific component of neural activity that actually issues the next command. We introduce a model of optimal feedback control (OFC) that shows that invariant dynamics can help transform movement feedback into commands, reducing the input that the neural population needs to control movement. Altogether our results demonstrate that invariant dynamics drive commands to control a variety of movements and show how feedback can be integrated with invariant dynamics to issue generalizable commands.

Distinct neural representations during a brain-machine interface and manual reaching task in motor cortex, prefrontal cortex, and striatum.

Although brain-machine interfaces (BMIs) are directly controlled by the modulation of a select local population of neurons, distributed networks consisting of cortical and subcortical areas have been implicated in learning and maintaining control. Previous work in rodents has demonstrated the involvement of the striatum in BMI learning. However, the prefrontal cortex has been largely ignored when studying motor BMI control despite its role in action planning, action selection, and learning abstract tasks. Here, we compare local field potentials simultaneously recorded from primary motor cortex (M1), dorsolateral prefrontal cortex (DLPFC), and the caudate nucleus of the striatum (Cd) while nonhuman primates perform a two-dimensional, self-initiated, center-out task under BMI control and manual control. Our results demonstrate the presence of distinct neural representations for BMI and manual control in M1, DLPFC, and Cd. We find that neural activity from DLPFC and M1 best distinguishes control types at the go cue and target acquisition, respectively, while M1 best predicts target-direction at both task events. We also find effective connectivity from DLPFC → M1 throughout both control types and Cd → M1 during BMI control. These results suggest distributed network activity between M1, DLPFC, and Cd during BMI control that is similar yet distinct from manual control.

Detecting event-related changes of multivariate phase coupling in dynamic brain networks.

Oscillatory phase coupling within large-scale brain networks is a topic of increasing interest within systems, cognitive, and theoretical neuroscience. Evidence shows that brain rhythms play a role in controlling neuronal excitability and response modulation (Haider B, McCormick D. Neuron 62: 171-189, 2009) and regulate the efficacy of communication between cortical regions (Fries P. Trends Cogn Sci 9: 474-480, 2005) and distinct spatiotemporal scales (Canolty RT, Knight RT. Trends Cogn Sci 14: 506-515, 2010). In this view, anatomically connected brain areas form the scaffolding upon which neuronal oscillations rapidly create and dissolve transient functional networks (Lakatos P, Karmos G, Mehta A, Ulbert I, Schroeder C. Science 320: 110-113, 2008). Importantly, testing these hypotheses requires methods designed to accurately reflect dynamic changes in multivariate phase coupling within brain networks. Unfortunately, phase coupling between neurophysiological signals is commonly investigated using suboptimal techniques. Here we describe how a recently developed probabilistic model, phase coupling estimation (PCE; Cadieu C, Koepsell K Neural Comput 44: 3107-3126, 2010), can be used to investigate changes in multivariate phase coupling, and we detail the advantages of this model over the commonly employed phase-locking value (PLV; Lachaux JP, Rodriguez E, Martinerie J, Varela F. Human Brain Map 8: 194-208, 1999). We show that the N-dimensional PCE is a natural generalization of the inherently bivariate PLV. Using simulations, we show that PCE accurately captures both direct and indirect (network mediated) coupling between network elements in situations where PLV produces erroneous results. We present empirical results on recordings from humans and nonhuman primates and show that the PCE-estimated coupling values are different from those using the bivariate PLV. Critically on these empirical recordings, PCE output tends to be sparser than the PLVs, indicating fewer significant interactions and perhaps a more parsimonious description of the data. Finally, the physical interpretation of PCE parameters is straightforward: the PCE parameters correspond to interaction terms in a network of coupled oscillators. Forward modeling of a network of coupled oscillators with parameters estimated by PCE generates synthetic data with statistical characteristics identical to empirical signals. Given these advantages over the PLV, PCE is a useful tool for investigating multivariate phase coupling in distributed brain networks.

Redundant information encoding in primary motor cortex during natural and prosthetic motor control.

Redundant encoding of information facilitates reliable distributed information processing. To explore this hypothesis in the motor system, we applied concepts from information theory to quantify the redundancy of movement-related information encoded in the macaque primary motor cortex (M1) during natural and neuroprosthetic control. Two macaque monkeys were trained to perform a delay center-out reaching task controlling a computer cursor under natural arm movement (manual control, 'MC'), and using a brain-machine interface (BMI) via volitional control of neural ensemble activity (brain control, 'BC'). During MC, we found neurons in contralateral M1 to contain higher and more redundant information about target direction than ipsilateral M1 neurons, consistent with the laterality of movement control. During BC, we found that the M1 neurons directly incorporated into the BMI ('direct' neurons) contained the highest and most redundant target information compared to neurons that were not incorporated into the BMI ('indirect' neurons). This effect was even more significant when comparing to M1 neurons of the opposite hemisphere. Interestingly, when we retrained the BMI to use ipsilateral M1 activity, we found that these neurons were more redundant and contained higher information than contralateral M1 neurons, even though ensembles from this hemisphere were previously less redundant during natural arm movement. These results indicate that ensembles most associated to movement contain highest redundancy and information encoding, which suggests a role for redundancy in proficient natural and prosthetic motor control.

Emergence of a stable cortical map for neuroprosthetic control.

Cortical control of neuroprosthetic devices is known to require neuronal adaptations. It remains unclear whether a stable cortical representation for prosthetic function can be stored and recalled in a manner that mimics our natural recall of motor skills. Especially in light of the mixed evidence for a stationary neuron-behavior relationship in cortical motor areas, understanding this relationship during long-term neuroprosthetic control can elucidate principles of neural plasticity as well as improve prosthetic function. Here, we paired stable recordings from ensembles of primary motor cortex neurons in macaque monkeys with a constant decoder that transforms neural activity to prosthetic movements. Proficient control was closely linked to the emergence of a surprisingly stable pattern of ensemble activity, indicating that the motor cortex can consolidate a neural representation for prosthetic control in the presence of a constant decoder. The importance of such a cortical map was evident in that small perturbations to either the size of the neural ensemble or to the decoder could reversibly disrupt function. Moreover, once a cortical map became consolidated, a second map could be learned and stored. Thus, long-term use of a neuroprosthetic device is associated with the formation of a cortical map for prosthetic function that is stable across time, readily recalled, resistant to interference, and resembles a putative memory engram.

Selective modulation of cortical population dynamics during neuroprosthetic skill learning.

Brain-machine interfaces (BMIs) provide a framework for studying how cortical population dynamics evolve over learning in a task in which the mapping between neural activity and behavior is precisely defined. Learning to control a BMI is associated with the emergence of coordinated neural dynamics in populations of neurons whose activity serves as direct input to the BMI decoder (direct subpopulation). While previous work shows differential modification of firing rate modulation in this population relative to a population whose activity was not directly input to the BMI decoder (indirect subpopulation), little is known about how learning-related changes in cortical population dynamics within these groups compare.To investigate this, we monitored both direct and indirect subpopulations as two macaque monkeys learned to control a BMI. We found that while the combined population increased coordinated neural dynamics, this increase in coordination was primarily driven by changes in the direct subpopulation. These findings suggest that motor cortex refines cortical dynamics by increasing neural variance throughout the entire population during learning, with a more pronounced coordination of firing activity in subpopulations that are causally linked to behavior.

Diverse operant control of different motor cortex populations during learning.

During motor learning,1 as well as during neuroprosthetic learning,2-4 animals learn to control motor cortex activity in order to generate behavior. Two different populations of motor cortex neurons, intra-telencephalic (IT) and pyramidal tract (PT) neurons, convey the resulting cortical signals within and outside the telencephalon. Although a large amount of evidence demonstrates contrasting functional organization among both populations,5,6 it is unclear whether the brain can equally learn to control the activity of either class of motor cortex neurons. To answer this question, we used a calcium-imaging-based brain-machine interface (CaBMI)3 and trained different groups of mice to modulate the activity of either IT or PT neurons in order to receive a reward. We found that the animals learned to control PT neuron activity faster and better than IT neuron activity. Moreover, our findings show that the advantage of PT neurons is the result of characteristics inherent to this population as well as their local circuitry and cortical depth location. Taken together, our results suggest that the motor cortex is more efficient at controlling the activity of pyramidal tract neurons, which are embedded deep in the cortex, and relaying motor commands outside the telencephalon.

Skilled independent control of individual motor units via a non-invasive neuromuscular-machine interface.

Objective.Brain-machine interfaces (BMIs) have the potential to augment human functions and restore independence in people with disabilities, yet a compromise between non-invasiveness and performance limits their relevance.Approach.Here, we hypothesized that a non-invasive neuromuscular-machine interface providing real-time neurofeedback of individual motor units within a muscle could enable independent motor unit control to an extent suitable for high-performance BMI applications.Main results.Over 6 days of training, eight participants progressively learned to skillfully and independently control three biceps brachii motor units to complete a 2D center-out task. We show that neurofeedback enabled motor unit activity that largely violated recruitment constraints observed during ramp-and-hold isometric contractions thought to limit individual motor unit controllability. Finally, participants demonstrated the suitability of individual motor units for powering general applications through a spelling task.Significance.These results illustrate the flexibility of the sensorimotor system and highlight individual motor units as a promising source of control for BMI applications.