|Home | About | Journals | Submit | Contact Us | Français|
Novel motor skills are learned through repetitive practice and, once acquired, persist long after training stops1,2. 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 task3–5. 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 neurons6,7. Obtaining new motor skills has been shown to strengthen the horizontal cortical connections in the primary motor cortex4,5. 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). To 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 microscopy8 (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 plasticity9,10. 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, intracortical microstimulation was performed at the end of repetitive imaging to confirm that images were taken from the functionally responding motor cortex (Fig. 1c, Supplementary Notes and Supplementary Fig. 2).
Unexpectedly, we found that motor learning led to rapid formation of dendritic spines (spinogenesis) in the motor cortex contralateral to the reaching forelimb. One-month-old mice that finished 30 reaches with more than 10 successes in the first day of training were imaged within 1 h of the training session and showed 10.6 ± 1.1% new spines which were not in the images acquired the day before training. This spine formation was more than double that found in age-matched controls, which were handled similarly and imaged over the same period of time, but not trained (Fig. 1d–f, 4.7 ± 0.6% in general controls, P < 0.001). In contrast, spine elimination measured in the same images was not significantly altered by motor learning during single training sessions (Fig. 1f, P > 0.9). In addition, mice that went through shaping but not training (shaping controls) or mice that were trained to reach for a seed too far away to grasp (activity controls) did not show an increase in spine formation rates (Fig. 1f, P > 0.1 with general control, P < 0.001 with trained mice; see Methods for all control conditions). This suggests that refinement of fine motor movements, rather than other training-related experiences or unskilled motor activity, drives robust spine formation. Furthermore, the percentage of spines formed immediately after the first training session is linearly correlated with the number of successful reaches during the training session, revealing a direct link between learning and spine formation (Fig. 1g, r2 = 0.77).
Perfection of a motor skill often requires persistent practice over time. To examine how prolonged learning affects spine dynamics, we trained and imaged mice over different periods of time (that is, from 2 to 16 days). We found that training for 2 days and longer resulted in significant increases, not only in spine formation, but also in spine elimination (Fig. 2a, b, P < 0.005 at all time points). Although delayed, this increase in spine elimination ultimately resulted in the total spine density in the trained animals returning to control levels by day 16 (Fig. 2c). As a control, we measured spine formation and elimination over a 4-day training period in the ipsilateral (to the trained limb) primary motor cortex and the contralateral posterior sensory cortex, and found no significant increase in spine formation or elimination in either case (Fig. 2a, b, d, e, P > 0.2). In addition, mice that failed to learn also failed to show an increase in either spine formation or elimination in the contralateral motor cortex (Figs 1b and 2a, b, f, P > 0.6). Therefore, the observed changes in spine dynamics are region- and learning-specific, indicating that motor learning causes synaptic reorganization in the corresponding motor cortex.
The enhanced spine loss after rapid spinogenesis reflects a rewiring of the neuronal circuitry in response to learning, rather than a simple addition of new spines. To examine how learning reorganizes synaptic connections, we imaged the same mice three times, classified imaged spines into new and pre-existing spines based on their appearance in the initial two images, and then quantified their survival percentages in the third images (Fig. 3a). Our data show that new spines are less stable than pre-existing spines in general (Fig. 3b, c). Specifically, in control mice, 43.8 ± 3.1%, 25.8 ± 5.3% and 19.2 ± 4.6% of the spines that formed between days 0 and 4 remained by days 6, 8 and 16, respectively. During the same period of time, 96.7 ± 0.5%, 94.9 ± 1.1% and 92.8 ± 1.9% of the pre-existing spines remained (Fig. 3d, P < 0.001 compared to new spines). These results suggest that new spines are initially unstable and undergo a prolonged selection process before being converted into stable synapses. In addition, we found that new spines were significantly more stable in trained mice, with 64.1 ± 2.2%, 55.3 ± 4.1% and 51.0 ± 4.8% of the spines that formed during the initial 4-day training remaining by days 6, 8 and 16, respectively (Fig. 3d, P < 0.001 compared to new spines in control mice). In contrast, pre-existing spines in trained mice were significantly less stable than control mice over the same time periods (Fig. 3d, P < 0.05). More importantly, when the fate of the new spines formed during initial learning (day 0–4) was examined months later (day 120), we found that 42.3 ± 2.9% of new spines persisted in the mice trained for 16 days during adolescence, whereas only 13.5 ± 1.7% of new spines remained in the control mice (Fig. 3d, P < 0.001). In addition, we found that spine formation and stabilization were associated with behavioural improvement. More new spines were formed daily during the learning acquisition phase (days 1–4) than during the learning maintenance phase (days 5–16); the new spines that were formed during learning acquisition, but not during maintenance, were preferably stabilized with continuous training (Supplementary Notes and Supplementary Fig. 3). Taken together, these data indicate that motor learning selectively stabilizes learning-induced new spines and destabilizes pre-existing spines. The prolonged persistence of learning-induced synapses provides a potential cellular mechanism for the consolidation of lasting, presumably permanent, motor memories.
Dendrites in the mammalian brain contain not only spines but also filopodia. Filopodia are long, thin protrusions without bulbous heads, and make up ~10% of the total dendritic protrusions in the motor cortex of 1-month-old mice. Previous studies suggest that filopodia are precursors of dendritic spines11,12. We found that filopodia were very dynamic in the mouse motor cortex in vivo. Most of them turned over within 1 day in control mice (79.3 ± 12.8% formation and 87.6 ± 5.9% elimination), and motor learning had no effect on filopodial formation and elimination (91.0 ± 15.3% formation and 86.5 ± 8.8% elimination, P > 0.2). Among the filopodia observed in the initial images, few of them became spines over the following day in control mice (6.3%). However, this filopodium-to-spine transition was enhanced by motor skill learning (13.1%). Furthermore, 25% of new spines formed from filopodia on training day 1 persisted after another 4 days of training, indicating a contribution of filopodia to the rewired neuronal circuitry. Furthermore, when filopodia and spines were pooled together for analysis, there was a ~10% increase in the dynamics of both control and training categories. Thus, the conclusion of motor learning on total protrusions was consistent with the spine analysis alone (Supplementary Fig. 4).
One of the important characteristics of motor learning is that, once the skill is well learned, its further maintenance does not require constant practice. To test whether lasting motor memories might be contained within structurally stable neural circuits, we trained young mice for 8–16 days to acquire the reaching skill, housed them in control cage conditions for 4 months, and retrained them on the same task in adulthood. We found that these pre-trained mice maintained skilful performance with high success rates even on the first day of reintroducing the reaching task (Fig. 4a). Imaging of these pre-trained adult mice showed that spine formation and elimination during retraining were similar to those of naive adults without training (Fig. 4b, e, g, P > 0.1 for 4 and 8 days). In contrast, naive adults learning the reaching task for the first time had a learning curve similar to adolescent mice, and showed significantly higher spine formation and elimination than control adults (Fig. 4a–c, g, 4 and 8 days, P < 0.01 with control for both formation and elimination). Next, we asked if learning a novel motor skill continued to drive synaptic reorganization in the pre-trained brain. To do this, we trained mice that had been pre-trained on the reaching task with a new motor task—the capellini handling task, which also requires fine forelimb motor skills (see Methods).We found that pre-trained mice, similar to naive adults, had enhanced spine formation and elimination during the training of this novel skill task (Fig. 4d, f, g, P < 0.001 compared to control adults). Despite high spine dynamics induced by novel skill learning, most spines that were formed during adolescent learning of the reaching task and maintained in adults persisted after training with the capellini handling task (95.6 ± 7.7%), suggesting that already stabilized synapses are not perturbed by novel learning in adults. These results indicate that synaptic structural coding outlasts the early learning experience and persists in adulthood to support later maintenance of motor skills. The fact that novel learning experiences continue to drive synaptic reorganization without affecting the stability of synapses formed during previous learning further suggests that different motor behaviours are stored using different sets of synapses in the brain.
Our study investigated the process of synapse reorganization in the living brain during natural learning, distinguishing it from several studies where changes were triggered by non-physiological sensory manipulation13–18. Although rapid synapse formation has been observed during long-term potentiation in vitro19,20, we show, for the first time, that synapse formation in the neocortex begins immediately as animals learn a new task in the living brain (within 1 h of training initiation). Such high spine formation does not occur with motor activity alone or later practice of the established skill. The rapidity of the response contradicts the general assumption that significant synaptic structural remodelling in motor cortex takes days to occur, following more subtle cellular activity and changes in synaptic efficacy4,21,22. One recent study on brain slices shows that glutamate-sensitive currents expressed in newly formed spines are indistinguishable from mature spines of comparable volumes23, further suggesting that the new spines formed during learning are probably active. Furthermore, the persistence of new spines over months provides a long-lasting structural basis for the enhanced synaptic strength that is retained even when the task performance is discontinued.
Many previous studies have used fixed tissue preparation to investigate changes in synapse number and dendritic complexity after motor skill learning24–28. Our in vivo imaging of superficial dendrites from layer V pyramidal neurons revealed that postsynaptic dendritic spine addition was rapid, but eventually counteracted by the loss of pre-existing spines, resulting in a time-dependent spine density change during motor learning. Although the synaptogenesis observed in our study is compatible with earlier results, its temporal relationship with behavioural improvement and the contribution of synapse elimination in circuitry reorganization in other brain layers and regions during motor learning require further investigation. This eventual balancing of synapse number could be a homeostatic mechanism by which the output layer V neurons integrate converging inputs into superficial cortical layers to govern precise fine motor control.
Young (1 month old) and adult (> 4 months old) mice expressing YFP in a small subset of cortical neurons (YFP-H line29) were used in all the experiments. Young mice were trained on the single-seed reaching task for up to 16 days and displayed a stereotypical learning curve (Fig. 1b). Naive adult mice and mice that had been previously trained with the single-seed reaching task in adolescence were trained with either the same reaching task or a novel capellini handling task for up to 8 days (see Methods). Apical dendrites of layer V pyramidal neurons, 10–100 μm below the cortical surface, were repeatedly imaged in mice under ketamine–xylazine anaesthesia with two-photon laser scanning microscopy. Spine dynamics in the motor cortex and other regions were followed over various intervals. Imaged regions were initially guided by stereotaxic measurements. In 14 mice, intracortical microstimulation (see Methods) was performed at the end of repetitive imaging to determine the location of acquired images relative to the functional forelimb motor map (Supplementary Fig. 2). In total, 32,079 spines from 209 mice were tracked over 2–4 imaging sessions, with 121 mice imaged twice, 79 mice three times and 9 mice imaged four times. Spine formation and elimination rates in each mouse were determined by comparing images of the same dendrites acquired at two time points; all changes were expressed relative to the total number of spines seen in the initial images. The number of spines analysed and the percentage of spine elimination and formation under various experimental conditions are summarized in Supplementary Table 1. To quantify spine size, calibrated spine head diameters were measured over time30 (Supplementary Notes). All data are presented as mean ± s.d., unless otherwise stated. P-values were calculated using the Student's t-test. A non-parametric Mann–Whitney U-test was used to confirm all conclusions.
We thank D. States, W. Thompson, L. Hinck, D. Feldheim, J. Ding, X. Li, A. Lin and C. Cirelli for critical comments on this manuscript; A. Sitko for her pilot studies of skilled reaching in mice, and D. Adkins, J. Kleim and N. Thomas for their assistance with intracortical microstimulation procedures. This work was supported by grants from the Ellison Medical Foundation, the DANA Foundation, and the National Institute on Aging to Y.Z.
Author Contributions T.X. and X.Y. contributed equally to this work. Both of them performed in vivo imaging, analysed the data, made figures and participated in the discussion. A.J.P., W.F.T. and J.A.Z. trained all the mice used in the experiments. K.T. and T.J. developed behavioural methods, performed the intracortical microstimulation experiments, and provided comments for the manuscript. Y.Z. initiated the project, did data analysis and wrote the manuscript.