Novel motor skills are learned through repetitive practice and, once acquired, persist long after training stops 1,2 . Earlier studies have shown that such learning induces an increase in the efficacy of synapses in the primary motor cortex, the persistence of which is associated with retention of the task [3][4][5] . However, how motor learning affects neuronal circuitry at the level of individual synapses and how long-lasting memory is structurally encoded in the intact brain remain unknown. Here we show that synaptic connections in the living mouse brain rapidly respond to motor-skill learning and permanently rewire. Training in a forelimb reaching task leads to rapid (within an hour) formation of postsynaptic dendritic spines on the output pyramidal neurons in the contralateral motor cortex. Although selective elimination of spines that existed before training gradually returns the overall spine density back to the original level, the new spines induced during learning are preferentially stabilized during subsequent training and endure long after training stops. Furthermore, we show that different motor skills are encoded by different sets of synapses. Practice of novel, but not previously learned, tasks further promotes dendritic spine formation in adulthood. Our findings reveal that rapid, but long-lasting, synaptic reorganization is closely associated with motor learning. The data also suggest that stabilized neuronal connections are the foundation of durable motor memory.Fine motor movements require accurate muscle synergies that rely on coordinated recruitment of intracortical synapses onto corticospinal neurons 6,7 . Obtaining new motor skills has been shown to strengthen the horizontal cortical connections in the primary motor cortex 4,5 . In this study, we taught mice a single-seed reaching task (Supplementary Movie 1). The majority of 1-month-old mice that underwent training gradually increased their reaching success rates during the initial 4 days, and then levelled off (n = 42, Fig. 1a, b). There were a few mice (n = 5) that engaged in extensive reaching, but continually failed to grasp the seeds. These mice normally gave up reaching after 4-8 days (Fig. 1b) investigate the process of learning-induced synaptic remodelling in the intact motor cortex, we repeatedly imaged the same apical dendrites of layer V pyramidal neurons marked by the transgenic expression of yellow fluorescent protein (YFP-H line) in various cortical regions during and after motor learning, using transcranial two-photon microscopy 8 ( Supplementary Fig. 1). Dendritic spines are the postsynaptic sites of most excitatory synapses in the brain and changes in spine morphology and dynamism serve as good indicators of synaptic plasticity 9,10 . Spines that were formed and eliminated were identified by comparing images from two time points, and then normalized to the initial images. Imaged regions were guided by stereotaxic measurements, ensuring the imaged neurons resided in the primary motor cortex. In several experiments,...

Abstract. The alignment of serial-section electron microscopy (ssEM) images is critical for efforts in neuroscience that seek to reconstruct neuronal circuits. However, each ssEM plane contains densely packed structures that vary from one section to the next, which makes matching features across images a challenge. Advances in deep learning has resulted in unprecedented performance in similar computer vision problems, but to our knowledge, they have not been successfully applied to ssEM image co-registration. In this paper, we introduce a novel deep network model that combines a spatial transformer for image deformation and a convolutional autoencoder for unsupervised feature learning for robust ssEM image alignment. This results in improved accuracy and robustness while requiring substantially less user intervention than conventional methods. We evaluate our method by comparing registration quality across several datasets.