Remodelling of corticostriatal axonal boutons during motor learning

Abstract

Motor skill learning induces long-lasting synaptic plasticity at dendritic spines^1-4 and at the outputs of motor cortical neurons to the striatum^5,6. However, little is known about corticostriatal axon activity and structural plasticity during learning in the adult brain. Here, using longitudinal in vivo two-photon imaging, we tracked thousands of corticostriatal axonal boutons in the dorsolateral striatum of awake mice. We found that learning a new motor skill dynamically regulated these boutons. The activities of motor corticostriatal axonal boutons exhibited selectivity for rewarded movements (RM) and unrewarded movements (UM). Notably, boutons on the same axonal branches showed diverse responses during behaviour. Motor learning significantly increased the proportion of RM boutons and reduced the heterogeneity of bouton activities. Moreover, motor learning induced profound structural dynamism in boutons. By combining structural and functional imaging, we saw that newly formed axonal boutons were more likely to exhibit selectivity for RM and were stabilized during motor learning, whereas UM boutons were selectively eliminated. These findings reveal a novel form of plasticity in corticostriatal axons and show that motor learning drives dynamic bouton reorganization to support motor skill acquisition and execution.

Fig. 1. Longitudinal two-photon Ca²⁺ imaging of corticostriatal axonal boutons during motor learning.

a, Schematic of lever-pushing task in which mice received water rewards following a cue. Example shows two rewarded (RM) and one unrewarded (UM) movement during the ITI. b–d, Behavioural improvements over training (n = 17 mice): increased success rate (b), decreased reaction time (c) and reduced movements during the ITI (d). Grey lines represent individual mice and the black line shows the group average. e, Representative RM trajectories on day 1 and day 11 from one mouse. Grey lines represent single trials, black line shows the average and the red dotted line dashed line indicates movement onset. f, Movement trajectories became more consistent across trials during training (r = 0.44, P = 1.05 × 10⁻⁹, Pearson’s correlation). CC, cross-correlation. g, Schematic of viral injection in M1 and imaging in DLS. 2P microscope, two-photon microscope. h, Example GCaMP6s-labelled corticostriatal axons on day 1 and 11. Scale bars, 20 μm. i, Task-related activity traces from boutons on day 3. Red lines indicate movement, black bars show lever pushes, blue represents cue and red represents reward. j, Top, individual (grey) and average (black) RM and UM trajectories. Bottom, averaged activity of 426 boutons aligned to RM or UM onset; boutons sorted by peak activity time. k, Increased trial-to-trial bouton activity correlation during RM in late versus early learning (**P = 0.007, Wilcoxon rank sum test, n = 13 mice). l, Trial-to-trial activity correlation plotted against movement similarity shows stronger coupling over learning (repeated measures two-way ANOVA, Bonferroni post hoc correction, P < 0.001 at multiple bins). m,n, Top, averaged bouton images from example axons 1 (m) and 2 (n). Bottom: ΔF/F₀ traces from three boutons with bouton-specific Ca²⁺ events, including detected events (filled dots) and the corresponding absences (open dots). Error bars indicate s.e.m. Source Data

Fig. 2. Reward- and movement-related activity of M1 corticostriatal boutons.

a, Example peri-movement activity of three boutons during RM (top) and UM (bottom) trials. Left, RM-selective bouton. Middle, RM–UM both. Right, UM-selective bouton. Resp, responsive. b, PCA embedding of all boutons (n = 3,744 RM, n = 4,211 UM, 8 mice). RM-only (red) and UM-only (blue) boutons are distinct. c, Top, lever movement trajectory. Second row, activity of 57 boutons (15 UM and 42 RM) from one mouse. Each row represents one bouton. Third row, PC1 (orange) and PC2 (blue) of bouton population activity. Bottom, behavioural annotations: cue, movement (move) and reward. d, 3D PCA trajectories of neural activity for RM and UM trials from one representative session. e, Trajectory selectivity index for RM versus UM trials at early and late learning stages (P < 0.05, Wilcoxon rank sum test, n = 8 mice). Shaded areas represent s.e.m. f, Change in bouton reward selectivity during motor learning. RM: P = 0.003, UM: P = 0.0011, both: P = 0.96, un-resp: P = 0.19; Wilcoxon rank sum test, n = 8 mice. Un-resp, unresponsive. g,h, Example of a bouton gaining RM selectivity (g) and the fate of early UM boutons over learning (h). i,j, Example trials with dissimilar (i) and similar (j) movement trajectories and corresponding bouton ensemble activity. k, Ensemble activity difference negatively correlates with movement similarity in late stage (r = −0.46, P = 1.89 × 10⁻⁵), but not early (r = −0.04, P = 0.73; Pearson’s correlation, n = 13 mice) learning. l, Ensemble differences between trials with most similar or least similar movement trajectories (early: P = 0.88, late: P = 0.0035, two-sided Wilcoxon rank sum test, n = 13 mice). *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant. Error bars represent s.e.m. Source Data

Fig. 3. Heterogeneous activity of boutons on the same corticostriatal or thalamostriatal axon.

a, Top, GCaMP6s image showing a single axon with clear axon and bouton morphology. Bottom, representative Ca²⁺ traces of two boutons on the same axon during RM and UM trials. Red and blue lines represent RM and UM initiation, respectively, red arrowheads show Ca²⁺ transients and stars indicate bouton-specific (heterogeneous) events. b, Fraction of unified Ca²⁺ transients at early and late stages of motor learning. P = 0.001, two-sided Wilcoxon rank sum test, n = 8 mice. c, Relative fraction of RM-related same and unique peaks at early and late stages (same: P = 3.1 × 10⁻⁴, unique: P = 3.1 × 10⁻⁴, Wilcoxon rank sum test, n = 8 mice). d, GCaMP6s image of corticostriatal axon and bouton structures (top), and boutons selective to RM or UM. e,f, Trial-averaged Ca²⁺ activity from two boutons shown in d during RM (top) and UM (bottom) trials on day 14. Both RM- and UM-selective boutons were found on the same axon. Grey represents Ca²⁺ transients in individual trials (ΔF/F₀) and black shows average of all trials in one day (day 14). g, Axon heterogeneity at M1 early (left), M1 late (middle) and PF late (right) stages (P = 0.028, Wilcoxon rank sum test, n = 8 mice for M1, n = 3 mice for PF). h, Schematic of virus injection in PF and imaging in DLS. i, Representative image of GCaMP6s-labelled thalamostriatal axons in DLS on day 16. Scale bar, 5 μm. j, Top, RM and UM lever individual (grey) and average (black) trajectories. Bottom, corresponding averaged Ca²⁺ activity from 2,254 boutons. k, GCaMP6s image of thalamostriatal axons (top) and identified boutons responsive to RM or UM trials (bottom). *P < 0.05, **P < 0.01, ***P < 0.001. Error bars represent s.e.m. Source Data

Fig. 4. Structural plasticity of corticostriatal and thalamostriatal axonal boutons.

a,b, Repeated imaging shows bouton formation (arrowhead) and elimination (arrow) in control (a) and trained (b) mice. c, Corticostriatal bouton density increased significantly in trained mice versus controls at multiple timepoints (P < 0.05; n = 143–146 axons from 8–9 mice). Wilcoxon rank sum test, control: n = 146 axons, 9 mice; trained: n = 143 axons, 8 mice. d,e, Trained mice exhibited increased bouton formation on day 4 and elimination on day 6. Formation, day 4: P = 0.01; elimination, day 6: P = 0.027, Wilcoxon rank sum test; control: n = 9 mice; trained: n = 8 mice. f,g, Similar imaging of thalamostriatal axons in control (f) and trained (g) mice showed bouton turnover. h, Thalamostriatal bouton density remained unchanged across 7 days (P > 0.05, Wilcoxon rank sum test, control: n = 46 axons, 3 mice; trained: n = 59 axons, 4 mice. i,j, No significant differences in bouton formation (i) or elimination (j) in thalamostriatal axons between groups. P > 0.05 for days 1–7, Wilcoxon rank sum test; control: n = 3; trained: n = 4 mice. k, New boutons formed on day 4 and survived. Day 8: P = 0.027, day 10: P = 0.0037, two-sided Wilcoxon rank sum test; control: n = 9 mice; trained: n = 8 mice). l–n, Bouton density at earlier stages correlated with later density (l) (day 10 versus day 4 (m): r = 0.51, P = 7.46 × 10⁻¹¹; versus day 8 (n): r = 0.73, P = 9.23 × 10⁻²⁵; Pearson’s correlation, n = 143 axons). o, GCaMP6s images of averaged GCaMP6s signal from day 1 and 9 reveal bouton formation and elimination. p,q, RM boutons formed at a higher rate than UM boutons at the late stage (P = 0.0019), with no difference in elimination rates. r,s, Bouton density increased in early RM-responsive axons (r; P = 5.8 × 10⁻⁴) and decreased in early UM-responsive axons (s; P = 0.0047, Wilcoxon rank sum test, n = 8 mice). t, Model for bouton turnover on axons with motor learning. Bouton density increases in axons that become RM-responsive during learning and decreases in those that become UM-responsive. *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant. Error bars represent s.e.m. Source Data

Extended Data Fig. 1. Corticotriatal axonal activities during RM, reward delay and reward omission trials.

a, Averaged activity pattern of same population of boutons during RM, delay reward and reward omission trials (423 boutons from one example mouse). b, Fraction of delay reward modulated boutons (n = 4 mice). c, Mean activity of delay reward modulated boutons during delay reward and reward omission trials (P = 0.029, two-sided Wilcoxon rank sum test, n = 4 mice). *p < 0.05. Error bars represent SEM. Source Data

Extended Data Fig. 2. Corticostriatal axonal bouton responses to cue and punishment.

a, Averaged activity pattern of same population of boutons during RM, cue and punishment (457 boutons from one example mouse). b, Fraction of cue modulated boutons (n = 4 mice). c, Fraction of punishment modulated boutons (n = 4 mice). Error bars represent SEM. Source Data

Extended Data Fig. 3. Averaged activity pattern of RM only, UM only and RM UM both boutons.

a, Averaged activity pattern of RM only boutons during RM trials (left) and UM trials (right), sorted according to the activity peak time in RM trials (one example mouse). b, Averaged activity pattern of UM only boutons during RM trials (left) and UM trials (right), sorted according to the activity peak time in UM trials (one example mouse). c, Averaged activity pattern of RM UM both boutons during RM trials (left) and UM trials (right), sorted according to the activity peak time in RM trials (one example mouse). d, Cumulative distribution of activity peak time of RM only (red), UM only (blue) and RM UM both (yellow) boutons (n = 8 mice). e, Fraction of boutons with activity peaks during the lever movement period (peak at −0.4 s to 0.5 s, n = 8 mice, RM vs UM, P = 0.0047, UM vs Both, P = 0.0019, two-sided Wilcoxon rank sum test). f, Fraction of boutons with activity peaks after the lever movement (peak at >0.5 s, n = 8 mice, RM vs UM, P = 0.0047, UM vs Both, P = 0.0019, two-sided Wilcoxon rank sum test). **p < 0.01. Error bars represent SEM. Note that more RM only and RM UM both boutons activated during lever pushing period, while more UM only boutons activated after lever pushing. Source Data

Extended Data Fig. 4. Dynamic changes of bouton selectivity during motor learning.

a-d, Fate of classified RM only (a), UM only (b), RM-UM both (c) and Un-responsive (d) boutons identified by their responses during early stages of training (n = 8 mice). e-h, Origin of classified RM only (e), UM only (f), RM-UM both (g) and Un-responsive (h) boutons identified by their responses during late stages of training (n = 8 mice). i, Fraction of boutons for 16 types defined by their dynamic change (n = 8 mice). Error bars represent SEM. j, Fraction of stable and switching boutons. Stable boutons represent the boutons that were classified as the same types at the early and late stages, while switching boutons represent boutons that were classified as different types at the early and late stages (P = 1.6 × 10⁻⁴, two-sided Wilcoxon rank sum test, n = 8 mice). ***P < 0.001, error bars represent SEM. Source Data

Extended Data Fig. 5. Stability of bouton representation during motor learning.

a, RM UM representation on two different days at late stage of learning (P = 0.8, P = 1, P = 0.8 respectively for RM, UM and RM-UM both, two-sided Wilcoxon rank sum test, n = 6 mice). b-e, Reward representation for 2 mice with good behavior performance at the early stage of learning. b, Success rate of example mouse 1 across training days. c, Change of RM UM representation during motor learning for mouse 1. d, Success rate of example mouse 2 across training days. e, Change of RM UM representation during motor learning for mouse 2. Source Data

Extended Data Fig. 6. Heterogeneity of UM-related bouton activities located on the same axon.

a, Top: example of average GCaMP6s image showing a single axon with clear axon and bouton morphology. Bottom, representative Ca²⁺ traces of two distinct boutons (red arrow) located on the same axon. Red vertical line: initiation of RM; blue vertical line, initiation of UM; red arrowhead, detected Ca²⁺ transients; stars, heterogeneous local Ca²⁺ transients. b, Relative fraction of UM related same peaks and unique peaks at early and late stage (P(Same)=0.0019, P(Unique)=0.0019, two-sided Wilcoxon rank sum test, n = 8 mice). c, Mean amplitude of RM and UM related peaks at early stage of learning (P = 0.33, two-sided Wilcoxon rank sum test, n = 8 mice). d, Mean amplitude of RM and UM related peaks at late stage of learning (P = 0.33, two-sided Wilcoxon rank sum test, n = 8 mice). **P < 0.01, NS, not significant. Error bars represent SEM. Source Data

Extended Data Fig. 7. Correlation between axon bouton and shaft activity.

a, Top: example of average GCaMP6s image showing a single axon with clear axon and bouton morphology. The red circles represent boutons, and the green areas represent axon shaft. Bottom, representative Ca²⁺ traces of two distinct boutons (1 and 2) located on the same axon, and the Ca²⁺ trace of the axonal shaft. b, Mean correlation between axon shaft and boutons located on the same axon at early and late stage, note that learning increased the correlation along the same axon (P = 0.015, two-sided Wilcoxon rank sum test, n = 8 mice). c, Fraction of unique peaks existed in boutons, but not on axon shaft at early and late stage, learning decreased the heterogeneity between boutons and axon shaft after learning (P = 0.038, two-sided Wilcoxon rank sum test, n = 8 mice). d, The correlation for smaller and larger peaks between bouton and axon shaft at early and late stages of learning (P (Early)=0.038, P(Late)=0.003, two-sided Wilcoxon rank sum test, n = 8 mice). * p < 0.05, ** p < 0.01. Error bars represent SEM. Source Data

Extended Data Fig. 8. Analysis of axon heterogeneity using GCaMP8f activity.

a, Schematic diagram showing the sites of virus injection (M1) and imaging (dorsolateral striatum, DLS). b, Top: example of averaged GCaMP8f image showing a single axon with clear axon and bouton morphology. Bottom, representative Ca²⁺ traces of three distinct boutons (red arrows) located on the same axon. Red vertical line: initiation of RM; blue vertical line, initiation of UM; red arrowhead, detected Ca²⁺ transients; stars, heterogeneous local Ca²⁺ transients. c, fractions of unified Ca²⁺ transients at late stages of motor learning (n = 3 mice). d, Axon heterogeneity at late stage of learning (n = 3 mice). Error bars represent SEM. Source Data

Extended Data Fig. 9. Fraction of boutons Ca²⁺ events absent from axonal shaft across different detection thresholds.

a, Fraction of unique peaks in boutons (compared to axon shaft) calculated using different peak detection threshold (from 0.5 SD to 3 SD, two-sided Wilcoxon rank sum test, n = 8 mice). b, Fraction of unique peaks in boutons plotted against different peak detection thresholds. Note the heterogeneity was not affected by the threshold both at early and late stage, but the fraction of unique peaks at late stage were significantly lower than that at early stage (Early stage, P = 1, one-way ANOVA, n = 8 mice; Late stage, P = 0.96, one-way ANOVA, P = 0.03, two-way ANOVA, between early and late stage, n = 8 mice). *p < 0.05. Error bars represent SEM. Source Data

Extended Data Fig. 10. Same peak fraction across different detection thresholds.

a, Fraction of same peaks between boutons within the same axon calculated using different peak detection threshold (from 0.5 SD to 3 SD, two-sided Wilcoxon rank sum test, n = 8 mice). b, Fraction of same peaks between boutons within the same axon plotted against different peak detection thresholds (from 0.5 SD to 3 SD, P = 0.006, between early and late stage, P = 1.9 × 10⁻⁷, between different thresholds, two-way ANOVA, n = 8 mice). **P < 0.01. Error bars represent SEM. Source Data

Extended Data Fig. 11. Axon heterogeneity of RM and UM axons.

Axon heterogeneity of RM and UM axons at early and late stage (P(RM) = 0.0059, P(UM) = 0.44, two-sided Wilcoxon rank sum test, n = 8 mice). ** p < 0.01, NS, not significant, error bars represent SEM. Source Data

Extended Data Fig. 12. Prediction of axon origin using pairwise correlation in axon activity.

a, Distribution of pairwise axon correlation (n = 8 mice, 7988 axon pairs). The red dashed line indicates the boundary between two distribution clusters (with axon correlation of 0.7, as the putative cutoff between axon pairs from the same neurons and pairs from different neurons). b, Axon heterogeneity plotted against pairwise axon correlation. c, Axon heterogeneity in axons originating from putative different and the same neurons (P = 0.5, two-sided Wilcoxon rank sum test, n = 8 mice). NS, not significant. Error bars represent SEM. Source Data

Extended Data Fig. 13. Spatial distribution of newly formed and eliminated boutons.

a, Spatial distribution of newly-formed and eliminated boutons along M1 axons throughout training. Black vertical lines indicate boutons that persisted throughout the imaging sessions, red circles indicate newly formed boutons, and blue circles indicate eliminated boutons. b, Cumulative distribution of nearest neighbor distance for newly formed boutons in the control (black) and training (red) group (P = 2.47 × 10⁻⁶, two-sided Kolmogorov–Smirnov test). c, Average nearest neighbor distance of newly formed boutons in control (black) and training group (red) mice (P = 0.01, two-sided Wilcoxon rank sum test, control: n = 9 mice; training: n = 8 mice). d, Average nearest neighbor distance of eliminated boutons in control (black) and trained (red) mice (P = 0.17, two-sided Wilcoxon rank sum test, control: n = 9 mice; training: n = 8 mice). e, Cumulative distribution of nearest neighbor distance for newly formed boutons in shuffled (black) and training (red) group. f, Average nearest neighbor distance of newly formed boutons in shuffled (black) and training group (red) mice (P = 0.0014, two-sided paired t-Test, n = 8 mice). *p < 0.05, **p < 0.01, NS, not significant. Error bars represent SEM. Source Data

Extended Data Fig. 14. Changes of bouton densities across days.

a-d, The normalized bouton density plotted against the density on adjacent imaging days. e-h, Bouton density on day 10 plotted against its density on day 2 (e), day 4 (f), day 6 (g), and day 8 (h). The bouton densities across days were positively correlated (P = 2.29 × 10⁻¹⁰ (a), P = 3.64 × 10⁻¹² (b), P = 5.28 × 10⁻¹⁴ (c), P = 9.23 × 10⁻²⁵ (d), P = 0.0082 (e), P = 7.46 × 10⁻¹¹ (f), P = 3.72 × 10⁻¹⁵ (g), P = 9.23 × 10⁻²⁵ (h), two-sided Pearson’s correlation, n = 143 axonal segments). Source Data

Extended Data Fig. 15. Example traces showing movement trajectories and calcium activity of an active (a) and inactive (b) bouton.

Gray traces represent single trial movements or calcium activity, while black traces represent the mean. Source Data