A deep-learning strategy to identify cell types across species from high-density extracellular recordings.

High-density probes allow electrophysiological recordings from many neurons simultaneously across entire brain circuits but don't reveal cell type. Here, we develop a strategy to identify cell types from extracellular recordings in awake animals, revealing the computational roles of neurons with distinct functional, molecular, and anatomical properties. We combine optogenetic activation and pharmacology using the cerebellum as a testbed to generate a curated ground-truth library of electrophysiological properties for Purkinje cells, molecular layer interneurons, Golgi cells, and mossy fibers. We train a semi-supervised deep-learning classifier that predicts cell types with greater than 95% accuracy based on waveform, discharge statistics, and layer of the recorded neuron. The classifier's predictions agree with expert classification on recordings using different probes, in different laboratories, from functionally distinct cerebellar regions, and across animal species. Our classifier extends the power of modern dynamical systems analyses by revealing the unique contributions of simultaneously-recorded cell types during behavior.

Emergence of syntax and word prediction in an artificial neural circuit of the cerebellum.

The cerebellum, interconnected with the cerebral neocortex, plays a vital role in human-characteristic cognition such as language processing, however, knowledge about the underlying circuit computation of the cerebellum remains very limited. To gain a better understanding of the computation underlying cerebellar language processing, we developed a biologically constrained cerebellar artificial neural network (cANN) model, which implements the recently identified cerebello-cerebellar recurrent pathway. We found that while cANN acquires prediction of future words, another function of syntactic recognition emerges in the middle layer of the prediction circuit. The recurrent pathway of the cANN was essential for the two language functions, whereas cANN variants with further biological constraints preserved these functions. Considering the uniform structure of cerebellar circuitry across all functional domains, the single-circuit computation, which is the common basis of the two language functions, can be generalized to fundamental cerebellar functions of prediction and grammar-like rule extraction from sequences, that underpin a wide range of cerebellar motor and cognitive functions. This is a pioneering study to understand the circuit computation of human-characteristic cognition using biologically-constrained ANNs.

P-sort: an open-source software for cerebellar neurophysiology.

Analysis of electrophysiological data from Purkinje cells (P-cells) of the cerebellum presents unique challenges to spike sorting. Complex spikes have waveforms that vary significantly from one event to the next, raising the problem of misidentification. Even when complex spikes are detected correctly, the simple spikes may belong to a different P-cell, raising the danger of misattribution. To address these identification and attribution problems, we wrote an open-source, semiautomated software called P-sort, and then tested it by analyzing data from P-cells recorded in three species: marmosets, macaques, and mice. Like other sorting software, P-sort relies on nonlinear dimensionality reduction to cluster spikes. However, it also uses the statistical relationship between simple and complex spikes to merge disparate clusters and split a single cluster. In comparison with expert manual curation, occasionally P-sort identified significantly more complex spikes, as well as prevented misattribution of clusters. Three existing automatic sorters performed less well, particularly for identification of complex spikes. To improve the development of analysis tools for the cerebellum, we provide labeled data for 313 recording sessions, as well as statistical characteristics of waveforms and firing patterns of P-cells in three species.NEW & NOTEWORTHY Algorithms that perform spike sorting depend on waveforms to cluster spikes. However, a cerebellar Purkinje-cell produces two types of spikes; simple and complex spikes. A complex spike coincides with the suppression of generating simple spikes. Here, we recorded neurophysiological data from three species and developed a spike analysis software named P-sort that relies on this statistical property to improve both the detection and the attribution of simple and complex spikes in the cerebellum.

Deleting Mecp2 from the cerebellum rather than its neuronal subtypes causes a delay in motor learning in mice.

Rett syndrome is a devastating childhood neurological disorder caused by mutations in MECP2. Of the many symptoms, motor deterioration is a significant problem for patients. In mice, deleting Mecp2 from the cortex or basal ganglia causes motor dysfunction, hypoactivity, and tremor, which are abnormalities observed in patients. Little is known about the function of Mecp2 in the cerebellum, a brain region critical for motor function. Here we show that deleting Mecp2 from the cerebellum, but not from its neuronal subtypes, causes a delay in motor learning that is overcome by additional training. We observed irregular firing rates of Purkinje cells and altered heterochromatin architecture within the cerebellum of knockout mice. These findings demonstrate that the motor deficits present in Rett syndrome arise, in part, from cerebellar dysfunction. For Rett syndrome and other neurodevelopmental disorders, our results highlight the importance of understanding which brain regions contribute to disease phenotypes.

Roles of the Cerebellum in Motor Preparation and Prediction of Timing.

The cerebellum is thought to have a variety of functions because it developed with the evolution of the cerebrum and connects with different areas in the frontoparietal cortices. Like neurons in the cerebral cortex, those in the cerebellum also exhibit strong activity during planning in addition to the execution of movements. However, their specific roles remain elusive. In this article, we review recent findings focusing on preparatory activities found in the primate deep cerebellar nuclei during tasks requiring deliberate motor control and temporal prediction. Neurons in the cerebellum are active during anti-saccade preparation and their inactivation impairs proactive inhibitory control for saccades. Experiments using a self-timing task show that there are mechanisms for tracking elapsed time and regulating trial-by-trial variation in timing, and that the cerebellum is involved in the latter. When predicting the timing of periodic events, the cerebellum provides more accurate temporal information than the striatum. During a recently developed synchronized eye movement task, cerebellar nuclear neurons exhibited periodic preparatory activity for predictive synchronization. In all cases, the cerebellum generated preparatory activity lasting for several hundred milliseconds. These signals may regulate neuronal activity in the cerebral cortex that adjusts movement timing and predicts the timing of rhythmic events.

50 Years Since the Marr, Ito, and Albus Models of the Cerebellum.

Fifty years have passed since David Marr, Masao Ito, and James Albus proposed seminal models of cerebellar functions. These models share the essential concept that parallel-fiber-Purkinje-cell synapses undergo plastic changes, guided by climbing-fiber activities during sensorimotor learning. However, they differ in several important respects, including holistic versus complementary roles of the cerebellum, pattern recognition versus control as computational objectives, potentiation versus depression of synaptic plasticity, teaching signals versus error signals transmitted by climbing-fibers, sparse expansion coding by granule cells, and cerebellar internal models. In this review, we evaluate different features of the three models based on recent computational and experimental studies. While acknowledging that the three models have greatly advanced our understanding of cerebellar control mechanisms in eye movements and classical conditioning, we propose a new direction for computational frameworks of the cerebellum, that is, hierarchical reinforcement learning with multiple internal models.

A cerebello-olivary signal for negative prediction error is sufficient to cause extinction of associative motor learning.

The brain generates negative prediction error (NPE) signals to trigger extinction, a type of inhibitory learning that is responsible for suppressing learned behaviors when they are no longer useful. Neurons encoding NPE have been reported in multiple brain regions. Here, we use an optogenetic approach to demonstrate that GABAergic cerebello-olivary neurons can generate a powerful NPE signal, capable of causing extinction of conditioned motor responses on its own.

New Cerebello-Cortical Pathway Involved in Higher-Order Oculomotor Control.

The cerebellum and the basal ganglia play an important role in the control of voluntary eye movement associated with complex behavior, but little is known about how cerebellar projections project to cortical eye movement areas. Here we used retrograde transneuronal transport of rabies virus to identify neurons in the cerebellar nuclei that project via the thalamus to supplementary eye field (SEF) of the frontal cortex of macaques. After rabies injections into the SEF, many neurons in the restricted region, the ventral aspects of the dentate nucleus (DN), the caudal pole of the DN, and the posterior interpositus nucleus (PIN) were labeled disynaptically via the thalamus, whereas no neuron labeling was found in the anterior interpositus nucleus (AIN). The distribution of the labeled neurons was dorsoventrally different from that of DN and PIN neurons labeled from the motor cortex. In the basal ganglia, a large number of labeled neurons were confined to the dorsomedial portion of the internal segment of the globus pallidus (GPi) as more neurons were labeled in the inner portion of the GPi (GPii) than in the outer portion of the GPi (GPio). This is the first evidence of a projection between cerebellum/basal ganglia and the SEF that could enable the cerebellum to modulate the cognitive control of voluntary eye movement.

Entrained neuronal activity to periodic visual stimuli in the primate striatum compared with the cerebellum.

Rhythmic events recruit neuronal activity in the basal ganglia and cerebellum, but their roles remain elusive. In monkeys attempting to detect a single omission of isochronous visual stimulus, we found that neurons in the caudate nucleus showed increased activity for each stimulus in sequence, while those in the cerebellar dentate nucleus showed decreased activity. Firing modulation in the majority of caudate neurons and all cerebellar neurons was proportional to the stimulus interval, but a quarter of caudate neurons displayed a clear duration tuning. Furthermore, the time course of population activity in the cerebellum well predicted stimulus timing, whereas that in the caudate reflected stochastic variation of response latency. Electrical stimulation to the respective recording sites confirmed a causal role in the detection of stimulus omission. These results suggest that striatal neurons might represent periodic response preparation while cerebellar nuclear neurons may play a role in temporal prediction of periodic events.

Different contributions of preparatory activity in the basal ganglia and cerebellum for self-timing.

The ability to flexibly adjust movement timing is important for everyday life. Although the basal ganglia and cerebellum have been implicated in monitoring of supra- and sub-second intervals, respectively, the underlying neuronal mechanism remains unclear. Here, we show that in monkeys trained to generate a self-initiated saccade at instructed timing following a visual cue, neurons in the caudate nucleus kept track of passage of time throughout the delay period, while those in the cerebellar dentate nucleus were recruited only during the last part of the delay period. Conversely, neuronal correlates of trial-by-trial variation of self-timing emerged earlier in the cerebellum than the striatum. Local inactivation of respective recording sites confirmed the difference in their relative contributions to supra- and sub-second intervals. These results suggest that the basal ganglia may measure elapsed time relative to the intended interval, while the cerebellum might be responsible for the fine adjustment of self-timing.

Single-Unit Extracellular Recording from the Cerebellum During Eyeblink Conditioning in Head-Fixed Mice.

This chapter presents a method for performing in vivo single-unit extracellular recordings and optogenetics during an associative, cerebellum-dependent learning task in head-fixed mice. The method uses a cylindrical treadmill system that reduces stress in the mice by allowing them to walk freely, yet it provides enough stability to maintain single-unit isolation of neurons for tens of minutes to hours. Using this system, we have investigated sensorimotor coding in the cerebellum while mice perform learned skilled movements.

Cerebellar Roles in Self-Timing for Sub- and Supra-Second Intervals.

Previous studies suggest that the cerebellum and basal ganglia are involved in sub-second and supra-second timing, respectively. To test this hypothesis at the cellular level, we examined the activity of single neurons in the cerebellar dentate nucleus in monkeys performing the oculomotor version of the self-timing task. Animals were trained to report the passage of time of 400, 600, 1200, or 2400 ms following a visual cue by making self-initiated memory-guided saccades. We found a sizeable preparatory neuronal activity before self-timed saccades across delay intervals, while the time course of activity correlated with the trial-by-trial variation of saccade latency in different ways depending on the length of the delay intervals. For the shorter delay intervals, the ramping up of neuronal firing rate started just after the visual cue and the rate of rise of neuronal activity correlated with saccade timing. In contrast, for the longest delay (2400 ms), the preparatory activity started late during the delay period, and its onset time correlated with self-timed saccade latency. Because electrical microstimulation applied to the recording sites during saccade preparation advanced self-timed but not reactive saccades, regardless of their directions, the signals in the cerebellum may have a causal role in self-timing. We suggest that the cerebellum may regulate timing in both sub-second and supra-second ranges, although its relative contribution might be greater for sub-second than for supra-second time intervals.SIGNIFICANCE STATEMENT How we decide the timing of self-initiated movement is a fundamental question. According to the prevailing hypothesis, the cerebellum plays a role in monitoring sub-second timing, whereas the basal ganglia are important for supra-second timing. To verify this, we explored neuronal signals in the monkey cerebellum while animals reported the passage of time in the range 400-2400 ms by making eye movements. Contrary to our expectations, we found that neurons in the cerebellar dentate nucleus exhibited a similar preparatory activity for both sub-second and supra-second intervals, and that electrical simulation advanced self-timed saccades in both conditions. We suggest that the cerebellum plays a causal role in the fine adjustment of self-timing in a larger time range than previously thought.

Facilitation of temporal prediction by electrical stimulation to the primate cerebellar nuclei.

The cerebellum is known to be involved in temporal information processing. However, the underlying neuronal mechanisms remain unclear. In our previous study, monkeys were trained to make a saccade in response to a single omission of periodically presented visual stimuli. To detect stimulus omission, animals had to predict the timing of each next stimulus. During this task, neurons in the cerebellar dentate nucleus exhibited a transient decrement of activity followed by a gradual increase in firing rate that peaked around the time of the next stimulus (Ohmae et al., 2013). In the present study, to address how these two components of neuronal activity contributed to omission detection, we applied electrical microstimulation to the recording site at different timing during the task. We found that electrical stimulation just before the stimulus omission shortened the latencies of both contraversive and ipsiversive saccades. Because the changes in saccade latency non-linearly depended on the timing of stimulation in each inter-stimulus interval, and electrical stimulation just before the early stimulus in the sequence failed to evoke saccades, the neuronal activity in the dentate nucleus might regulate temporal prediction rather than facilitating saccade execution. Our results support the hypothesis that the firing modulation in each inter-stimulus interval in the dentate nucleus represents neuronal code for the temporal prediction of next stimulus.

Chromatin remodeling inactivates activity genes and regulates neural coding.

Activity-dependent transcription influences neuronal connectivity, but the roles and mechanisms of inactivation of activity-dependent genes have remained poorly understood. Genome-wide analyses in the mouse cerebellum revealed that the nucleosome remodeling and deacetylase (NuRD) complex deposits the histone variant H2A.z at promoters of activity-dependent genes, thereby triggering their inactivation. Purification of translating messenger RNAs from synchronously developing granule neurons (Sync-TRAP) showed that conditional knockout of the core NuRD subunit Chd4 impairs inactivation of activity-dependent genes when neurons undergo dendrite pruning. Chd4 knockout or expression of NuRD-regulated activity genes impairs dendrite pruning. Imaging of behaving mice revealed hyperresponsivity of granule neurons to sensorimotor stimuli upon Chd4 knockout. Our findings define an epigenetic mechanism that inactivates activity-dependent transcription and regulates dendrite patterning and sensorimotor encoding in the brain.

Two different mechanisms for the detection of stimulus omission.

Although we can detect slight changes in musical rhythm, the underlying neural mechanism remains elusive. Here we show that two distinct mechanisms are automatically selected depending on the speed of the rhythm. When human subjects detected a single omission of isochronous repetitive auditory stimuli, reaction time strongly depended on the stimulus onset asynchrony (SOA) for shorter SOAs (<250 ms), but was almost constant for longer SOAs. For shorter SOAs, subjects were unable to detect stimulus omission when either monaural stimuli or those in different frequencies were randomly presented. In contrast, for longer SOAs, reaction time increased when different tempos were presented simultaneously to different ears. These results suggest that depending on the speed of rhythms, the brain may use either temporal grouping of discrete sounds or temporal prediction of upcoming stimuli to detect the absence of a regular stimulus. Because we also found a similar relationship between reaction time and SOA for both visual and tactile stimuli, dual detection strategies could be generalized to other sensory modalities.

Climbing fibers encode a temporal-difference prediction error during cerebellar learning in mice.

Climbing fiber inputs to Purkinje cells are thought to be involved in generating the instructive signals that drive cerebellar learning. To investigate how these instructive signals are encoded, we recorded the activity of individual climbing fibers during cerebellum-dependent eyeblink conditioning in mice. We found that climbing fibers signaled both the unexpected delivery and the unexpected omission of the periocular airpuff that served as the instructive signal for eyeblink conditioning. In addition, we observed that climbing fibers activated by periocular airpuffs also responded to stimuli from other sensory modalities if those stimuli were novel or if they predicted that the periocular airpuff was about to be presented. This pattern of climbing fiber activity is markedly similar to the responses of dopamine neurons during reinforcement learning, which have been shown to encode a particular type of instructive signal known as a temporal difference prediction error.

Decoding the timing and target locations of saccadic eye movements from neuronal activity in macaque oculomotor areas.

OBJECTIVE: The control of movement timing has been a significant challenge for brain-machine interfaces (BMIs). As a first step toward developing a timing-based BMI, we aimed to decode movement timing and target locations in a visually guided saccadic eye movement task using the activity of neurons in the primate frontal eye field (FEF) and supplementary eye field (SEF). APPROACH: For this purpose, we developed a template-matching method that could recruit a variety of neurons in these areas. MAIN RESULTS: As a result, we were able to achieve a favorable estimation of saccade onset: for example, data from 20 randomly sampled FEF neurons or 40 SEF neurons achieved a median estimation error of ∼10 ms with an interquartile range less than 50 ms (± ∼25 ms). In the best case, seven simultaneously recorded SEF neurons using a multi-electrode array achieved a comparable accuracy (10 ± 30 ms). The method was significantly better than a heuristic method that used only a group of movement cells with sharp discharges at the onset of saccades. The estimation of target location was less accurate but still favorable, especially when we estimated target location at a timing of 200 ms after the onset of saccade: the method was able to discriminate 16 targets with an accuracy of 90%, which differed not only in their directions (eight directions) but also in amplitude (10/20°) when we used data from 61 randomly sampled FEF neurons. SIGNIFICANCE: The results show that the timing, amplitude and direction of saccades can be decoded from neuronal activity in the FEF and SEF and further suggest that timing-based BMIs can be developed by decoding timing information using the template-matching method.

Temporally specific sensory signals for the detection of stimulus omission in the primate deep cerebellar nuclei.

The cerebellum is implicated in sensory prediction in the subsecond range. To explore how neurons in the cerebellum encode temporal information for the prediction of sensory events, we trained monkeys to make a saccade in response to either a single omission or deviation of isochronous repetitive stimuli. We found that neurons in the cerebellar dentate nucleus exhibited a gradual elevation of the baseline firing rate as the repetition progressed. Most neurons showed a transient suppression for each stimulus, and this firing modulation also increased gradually, opposed to the sensory adaptation. The magnitude of the enhanced sensory response positively correlated with interstimulus interval. Furthermore, when stimuli appeared unexpectedly earlier than the regular timing, the neuronal modulation became smaller, suggesting that the sensory response depended on the time elapsed since the previous stimulus. The enhancement of neuronal modulation was context dependent and was reduced or even absent when monkeys were unmotivated to detect stimulus omission. A significant negative correlation between neuronal activity at stimulus omission and saccade latency suggested that the timing of each stimulus was predicted by the amount of recovery from the transient response. Because inactivation of the recording sites delayed the detection of stimulus omission but only slightly altered the detection of stimulus deviation, these signals might be necessary for the prediction of stimulus timing but may not be involved only in the generation of saccades. Our results demonstrate a novel mechanism for temporal prediction of upcoming stimuli that accompanies the time-dependent modification of sensory gain in the cerebellum.

[Neural representation of time].

Temporal information is essential for perception and behavior. Although the neural substrates for temporal processing have been elucidated in many different conditions, how individual neurons in each network represent time remains largely unknown. Here we review previous models of time representation in the brain, and propose that these models can be classified into four different groups based on two viewpoints. The first viewpoint is that temporal information is either prospective or retrospective. For example, the online control of movement timing requires prospective or predictive information, whereas the duration discrimination of previously presented stimuli depends on retrospective temporal information. The other viewpoint is whether neuronal coding is based on modulation of the firing rate in each neuron (rate coding) or the occurrence of synchronous activity across multiple neurons (temporal coding). The accumulator model and state-dependence model both represent time by modulating the rate of neuronal firing depending on the elapsed time, thereby providing the prospective and retrospective information, respectively. In contrast, temporal coding is used by the coincidence detection and entrainment/synchronization models acquired through learning. This classification might be helpful for comprehensive understanding of the neuronal mechanisms of temporal processing, each of which is implemented by the intrinsic property of each sensory system and/or by a dedicated network specialized for timing. We also propose a model incorporating serial stages of temporal processing to reproduce a fixed time interval, and suggest that future physiological and pharmacological experiments might prove our hypothesis.