Local and global predictors of synapse elimination during motor learning.

During learning, synaptic connections between excitatory neurons in the brain display considerable dynamism, with new connections being added and old connections eliminated. Synapse elimination offers an opportunity to understand the features of synapses that the brain deems dispensable. However, with limited observations of synaptic activity and plasticity in vivo, the features of synapses subjected to elimination remain poorly understood. Here, we examined the functional basis of synapse elimination in the apical dendrites of L2/3 neurons in the primary motor cortex throughout motor learning. We found no evidence that synapse elimination is facilitated by a lack of activity or other local forms of plasticity. Instead, eliminated synapses display asynchronous activity with nearby synapses, suggesting that functional synaptic clustering is a critical component of synapse survival. In addition, eliminated synapses show delayed activity timing with respect to postsynaptic output. Thus, synaptic inputs that fail to be co-active with their neighboring synapses or are mistimed with neuronal output are targeted for elimination.

Divergent Learning-Related Transcriptional States of Cortical Glutamatergic Neurons.

Experience-dependent gene expression reshapes neural circuits, permitting the learning of knowledge and skills. Most learning involves repetitive experiences during which neurons undergo multiple stages of functional and structural plasticity. Currently, the diversity of transcriptional responses underlying dynamic plasticity during repetition-based learning is poorly understood. To close this gap, we analyzed single-nucleus transcriptomes of L2/3 glutamatergic neurons of the primary motor cortex after 3 d motor skill training or home cage control in water-restricted male mice. "Train" and "control" neurons could be discriminated with high accuracy based on expression patterns of many genes, indicating that recent experience leaves a widespread transcriptional signature across L2/3 neurons. These discriminating genes exhibited divergent modes of coregulation, differentiating neurons into discrete clusters of transcriptional states. Several states showed gene expressions associated with activity-dependent plasticity. Some of these states were also prominent in the previously published reference, suggesting that they represent both spontaneous and task-related plasticity events. Markedly, however, two states were unique to our dataset. The first state, further enriched by motor training, showed gene expression suggestive of late-stage plasticity with repeated activation, which is suitable for expected emergent neuronal ensembles that stably retain motor learning. The second state, equally found in both train and control mice, showed elevated levels of metabolic pathways and norepinephrine sensitivity, suggesting a response to common experiences specific to our experimental conditions, such as water restriction or circadian rhythm. Together, we uncovered divergent transcriptional responses across L2/3 neurons, each potentially linked with distinct features of repetition-based motor learning such as plasticity, memory, and motivation.

Meta-reinforcement learning via orbitofrontal cortex.

The meta-reinforcement learning (meta-RL) framework, which involves RL over multiple timescales, has been successful in training deep RL models that generalize to new environments. It has been hypothesized that the prefrontal cortex may mediate meta-RL in the brain, but the evidence is scarce. Here we show that the orbitofrontal cortex (OFC) mediates meta-RL. We trained mice and deep RL models on a probabilistic reversal learning task across sessions during which they improved their trial-by-trial RL policy through meta-learning. Ca2+/calmodulin-dependent protein kinase II-dependent synaptic plasticity in OFC was necessary for this meta-learning but not for the within-session trial-by-trial RL in experts. After meta-learning, OFC activity robustly encoded value signals, and OFC inactivation impaired the RL behaviors. Longitudinal tracking of OFC activity revealed that meta-learning gradually shapes population value coding to guide the ongoing behavioral policy. Our results indicate that two distinct RL algorithms with distinct neural mechanisms and timescales coexist in OFC to support adaptive decision-making.

Learning binds new inputs into functional synaptic clusters via spinogenesis.

Learning induces the formation of new excitatory synapses in the form of dendritic spines, but their functional properties remain unknown. Here, using longitudinal in vivo two-photon imaging and correlated electron microscopy of dendritic spines in the motor cortex of mice during motor learning, we describe a framework for the formation, survival and resulting function of new, learning-related spines. Specifically, our data indicate that the formation of new spines during learning is guided by the potentiation of functionally clustered preexisting spines exhibiting task-related activity during earlier sessions of learning. We present evidence that this clustered potentiation induces the local outgrowth of multiple filopodia from the nearby dendrite, locally sampling the adjacent neuropil for potential axonal partners, likely via targeting preexisting presynaptic boutons. Successful connections are then selected for survival based on co-activity with nearby task-related spines, ensuring that the new spine preserves functional clustering. The resulting locally coherent activity of new spines signals the learned movement. Furthermore, we found that a majority of new spines synapse with axons previously unrepresented in these dendritic domains. Thus, learning involves the binding of new information streams into functional synaptic clusters to subserve learned behaviors.

The NMDA receptor subunit GluN3A regulates synaptic activity-induced and myocyte enhancer factor 2C (MEF2C)-dependent transcription.

N-Methyl-d-aspartate type glutamate receptors (NMDARs) are key mediators of synaptic activity-regulated gene transcription in neurons, both during development and in the adult brain. Developmental differences in the glutamate receptor ionotropic NMDA 2 (GluN2) subunit composition of NMDARs determines whether they activate the transcription factor cAMP-responsive element-binding protein 1 (CREB). However, whether the developmentally regulated GluN3A subunit also modulates NMDAR-induced transcription is unknown. Here, using an array of techniques, including quantitative real-time PCR, immunostaining, reporter gene assays, RNA-Seq, and two-photon glutamate uncaging with calcium imaging, we show that knocking down GluN3A in rat hippocampal neurons promotes the inducible transcription of a subset of NMDAR-sensitive genes. We found that this enhancement is mediated by the accumulation of phosphorylated p38 mitogen-activated protein kinase in the nucleus, which drives the activation of the transcription factor myocyte enhancer factor 2C (MEF2C) and promotes the transcription of a subset of synaptic activity-induced genes, including brain-derived neurotrophic factor (Bdnf) and activity-regulated cytoskeleton-associated protein (Arc). Our evidence that GluN3A regulates MEF2C-dependent transcription reveals a novel mechanism by which NMDAR subunit composition confers specificity to the program of synaptic activity-regulated gene transcription in developing neurons.

Reciprocal Activation within a Kinase-Effector Complex Underlying Persistence of Structural LTP.

Long-term synaptic plasticity requires a mechanism that converts short Ca2+ pulses into persistent biochemical signaling to maintain changes in the synaptic structure and function. Here, we present a novel mechanism of a positive feedback loop, formed by a reciprocally activating kinase-effector complex (RAKEC) in dendritic spines, enabling the persistence and confinement of a molecular memory. We found that stimulation of a single spine causes the rapid formation of a RAKEC consisting of CaMKII and Tiam1, a Rac-GEF. This interaction is mediated by a pseudo-autoinhibitory domain on Tiam1, which is homologous to the CaMKII autoinhibitory domain itself. Therefore, Tiam1 binding results in constitutive CaMKII activation, which in turn persistently phosphorylates Tiam1. Phosphorylated Tiam1 promotes stable actin-polymerization through Rac1, thereby maintaining the structure of the spine during LTP. The RAKEC can store biochemical information in small subcellular compartments, thus potentially serving as a general mechanism for prolonged and compartmentalized signaling.

GIT1 regulates synaptic structural plasticity underlying learning.

The signaling scaffold protein GIT1 is expressed widely throughout the brain, but its function in vivo remains elusive. Mice lacking GIT1 have been proposed as a model for attention deficit-hyperactivity disorder, due to alterations in basal locomotor activity as well as paradoxical locomotor suppression by the psychostimulant amphetamine. Since we had previously shown that GIT1-knockout mice have normal locomotor activity, here we examined GIT1-deficient mice for ADHD-like behavior in more detail, and find neither hyperactivity nor amphetamine-induced locomotor suppression. Instead, GIT1-deficient mice exhibit profound learning and memory defects and reduced synaptic structural plasticity, consistent with an intellectual disability phenotype. We conclude that loss of GIT1 alone is insufficient to drive a robust ADHD phenotype in distinct strains of mice. In contrast, multiple learning and memory defects have been observed here and in other studies using distinct GIT1-knockout lines, consistent with a predominant intellectual disability phenotype related to altered synaptic structural plasticity.

Regulation of Rho GTPase proteins during spine structural plasticity for the control of local dendritic plasticity.

While it is generally appreciated that learning involves the structural rearrangement of neuronal circuits, the underlying orchestration of molecular events that drives these changes is not as well understood. Recent studies on the spatiotemporal organization of synaptic signaling events have provided new insights into the biochemical underpinnings of various expressions of structural neuronal plasticity, as well as the functional consequences that emerge because of the particular behavior of the molecules involved. In particular, activity patterns of and interplay among a class of morphogenic signaling proteins, the Rho GTPases, and their downstream signals, are found to be critical for linking neuronal activity with various forms of neuronal plasticity. We review recent findings on this topic and discuss their physiological implications.

Reorganization of corticospinal output during motor learning.

Motor learning is accompanied by widespread changes within the motor cortex, but it is unknown whether these changes are ultimately funneled through a stable corticospinal output channel or whether the corticospinal output itself is plastic. We investigated the consistency of the relationship between corticospinal neuron activity and movement through in vivo two-photon calcium imaging in mice learning a lever-press task. Corticospinal neurons exhibited heterogeneous correlations with movement, with the majority of movement-modulated neurons decreasing activity during movement. Individual cells changed their activity across days, which led to changed associations between corticospinal activity and movement. Unlike previous observations in layer 2/3, activity accompanying learned movements did not become more consistent with learning; instead, the activity of dissimilar movements became more decorrelated. These results indicate that the relationship between corticospinal activity and movement is dynamic and that the types of activity and plasticity are different from and possibly complementary to those in layer 2/3.

Circuit Mechanisms of Sensorimotor Learning.

The relationship between the brain and the environment is flexible, forming the foundation for our ability to learn. Here we review the current state of our understanding of the modifications in the sensorimotor pathway related to sensorimotor learning. We divide the process into three hierarchical levels with distinct goals: (1) sensory perceptual learning, (2) sensorimotor associative learning, and (3) motor skill learning. Perceptual learning optimizes the representations of important sensory stimuli. Associative learning and the initial phase of motor skill learning are ensured by feedback-based mechanisms that permit trial-and-error learning. The later phase of motor skill learning may primarily involve feedback-independent mechanisms operating under the classic Hebbian rule. With these changes under distinct constraints and mechanisms, sensorimotor learning establishes dedicated circuitry for the reproduction of stereotyped neural activity patterns and behavior.

Rho GTPase complementation underlies BDNF-dependent homo- and heterosynaptic plasticity.

The Rho GTPase proteins Rac1, RhoA and Cdc42 have a central role in regulating the actin cytoskeleton in dendritic spines, thereby exerting control over the structural and functional plasticity of spines and, ultimately, learning and memory. Although previous work has shown that precise spatiotemporal coordination of these GTPases is crucial for some forms of cell morphogenesis, the nature of such coordination during structural spine plasticity is unclear. Here we describe a three-molecule model of structural long-term potentiation (sLTP) of murine dendritic spines, implicating the localized, coincident activation of Rac1, RhoA and Cdc42 as a causal signal of sLTP. This model posits that complete tripartite signal overlap in spines confers sLTP, but that partial overlap primes spines for structural plasticity. By monitoring the spatiotemporal activation patterns of these GTPases during sLTP, we find that such spatiotemporal signal complementation simultaneously explains three integral features of plasticity: the facilitation of plasticity by brain-derived neurotrophic factor (BDNF), the postsynaptic source of which activates Cdc42 and Rac1, but not RhoA; heterosynaptic facilitation of sLTP, which is conveyed by diffusive Rac1 and RhoA activity; and input specificity, which is afforded by spine-restricted Cdc42 activity. Thus, we present a form of biochemical computation in dendrites involving the controlled complementation of three molecules that simultaneously ensures signal specificity and primes the system for plasticity.

Autocrine BDNF-TrkB signalling within a single dendritic spine.

Brain-derived neurotrophic factor (BDNF) and its receptor TrkB are crucial for many forms of neuronal plasticity, including structural long-term potentiation (sLTP), which is a correlate of an animal's learning. However, it is unknown whether BDNF release and TrkB activation occur during sLTP, and if so, when and where. Here, using a fluorescence resonance energy transfer-based sensor for TrkB and two-photon fluorescence lifetime imaging microscopy, we monitor TrkB activity in single dendritic spines of CA1 pyramidal neurons in cultured murine hippocampal slices. In response to sLTP induction, we find fast (onset < 1 min) and sustained (>20 min) activation of TrkB in the stimulated spine that depends on NMDAR (N-methyl-d-aspartate receptor) and CaMKII signalling and on postsynaptically synthesized BDNF. We confirm the presence of postsynaptic BDNF using electron microscopy to localize endogenous BDNF to dendrites and spines of hippocampal CA1 pyramidal neurons. Consistent with these findings, we also show rapid, glutamate-uncaging-evoked, time-locked BDNF release from single dendritic spines using BDNF fused to superecliptic pHluorin. We demonstrate that this postsynaptic BDNF-TrkB signalling pathway is necessary for both structural and functional LTP. Together, these findings reveal a spine-autonomous, autocrine signalling mechanism involving NMDAR-CaMKII-dependent BDNF release from stimulated dendritic spines and subsequent TrkB activation on these same spines that is crucial for structural and functional plasticity.