Motor learning selectively strengthens cortical and striatal synapses of motor engram neurons

Abstract

Learning and consolidation of new motor skills require plasticity in the motor cortex and striatum, two key motor regions of the brain. However, how neurons undergo synaptic changes and become recruited during motor learning to form a memory engram remains unknown. Here, we train mice on a motor learning task and use a genetic approach to identify and manipulate behavior-relevant neurons selectively in the primary motor cortex (M1). We find that the degree of M1 engram neuron reactivation correlates with motor performance. We further demonstrate that learning-induced dendritic spine reorganization specifically occurs in these M1 engram neurons. In addition, we find that motor learning leads to an increase in the strength of M1 engram neuron outputs onto striatal spiny projection neurons (SPNs) and that these synapses form clusters along SPN dendrites. These results identify a highly specific synaptic plasticity during the formation of long-lasting motor memory traces in the corticostriatal circuit.

Keywords: corticostriatal circuit; dendritic spines; memory engram; motor learning; synaptic clustering; synaptic plasticity; two-photon imaging.

Figure 1:. Reactivation of motor cortex engram neurons correlates with behavioral performance on the forelimb reaching task.

(A) Experimental timeline and schematic drawing of the reaching task. (B) Average behavioral performance of mice trained on the forelimb reaching task, including all mice used in this study. n = 17 Early TRAP mice, 28 Late TRAP Learner mice, 13 Late TRAP Non-Learner mice. Error bars, SEM. (C) Representative image of TRAP-tdTomato (magenta) fluorescence and c-fos immunostaining (green) in the primary motor cortex (M1). Left: an overview of entire brain hemisphere, upper right: enlarged view of M1, lower right: enlarged view of the white square in M1. Green arrows denote cells with only c-fos expression, magenta arrows denote cells with only TRAP labeling, white arrows denote cells with c-fos expression and TRAP labeling. (D) Schematic drawing showing TRAPing of activated neurons with 4-OHT in c-fos TRAP-Ai9 mice (magenta) and immunostaining for c-fos proteins (green). Reactivated neurons show TRAP/c-fos double-positive signals. (E) Number of TRAP-tdTomato labeled cells in M1 (analyzed area: ~1.13 mm²). Bars denote mean and circles individual mice. Error bars, SEM. Control: 72.90 ± 9.93, n = 19 mice; Early TRAP: 86.34 ± 14.81, n = 11 mice; Late TRAP Non-Learner: 95.70 ± 17.72, n = 10 mice; Late TRAP Learner: 139.6 ± 25.24, n = 11 mice. p = 0.887, Control vs. Early TRAP; p = 0.647, Control vs. Non-Learner; p = 0.011, Control vs. Learner, One-way ANOVA with Tukey’s multiple comparison test. (F) Number of c-fos-expressing cells in M1 (analyzed area: ~1.13 mm²). Bars denote mean and circles individual mice. Error bars, SEM. Control: 429.5 ± 67.32, n = 19 mice; Early TRAP: 424.3 ± 42.92, n = 11 mice; Late TRAP Non-Learner: 340.0 ± 58.05, n = 10 mice; Late TRAP Learner: 505.7 ± 120.0, n = 11 mice. p = 0.606, One-way ANOVA. (G) Fraction of TRAP-labeled cells with c-fos- and TRAP-double labeling in M1. Bars denote mean and circles individual mice. Error bars, SEM. Control: 21.39% ± 3.26%, n = 19 mice; Early TRAP: 20.65% ± 1.92%, n = 11 mice; Late TRAP Non-Learner: 21.59% ± 2.86%, n = 10 mice; Late TRAP Learner: 34.99% ± 4.54%, n = 11 mice. p = 0.998, Control vs. Early TRAP; p > 0.999, Control vs. Non-Learner; p = 0.015, Control vs. Learner, One-way ANOVA with Tukey’s multiple comparison test. (H) Correlation between fraction of c-fos- and TRAP-double labeled cells in M1 and reaching performance (success rate) for individual Early TRAP or Late TRAP mice. Line represents linear regression. Early TRAP mice: n = 11, r² = 0.072, p = 0.426; Late TRAP mice: n = 21, r² = 0.375, p = 0.003, Pearson correlation. (I) Correlation between fraction of c-fos- and TRAP-double labeled cells in M1 and reaching performance (success rate) for Learner and Non-Learner Late TRAP mice. Line represents linear regression. Non-Learner mice: n = 10, r² = 0.014, p = 0.745; Learner mice: n = 11, r² = 0.567, p = 0.008, Pearson correlation. *p < 0.05; ns, non-significant. See also Figure S1.

Figure 2:. Dendritic spine plasticity during motor learning is specific to engram neurons in motor cortex.

(A) Experimental timeline and schematic drawing of in vivo two-photon imaging experiment. (B) Representative image showing the post-hoc identification of dendrites from TRAP neurons. Arrow heads denote dendrites from non-TRAP neurons (Thy1-YFP signal only, green), arrows denote dendrites from TRAP neurons (overlap of Thy1-YFP signal (green) and TRAP-tdTomato signal (magenta), white). (C) Representative images of dendrites from TRAP (top) and non-TRAP (bottom) neurons repeatedly imaged throughout training. Arrow heads denote spines formed during first two days of training that did not persist, arrows denote spines formed during first two days of training and persisted until end of training. (D) Average spine density normalized to baseline spine density in trained (red) and control (gray) mice over the course of motor learning. Control: n = 5 mice with 22 dendritic segments and 1019 spines; Trained: 4 mice with 20 dendritic segments and 866 spines. p = 0.261, 2-way repeated-measures ANOVA. (E) Average normalized spine density of TRAP dendrites in trained (red) and control (gray) mice. Control TRAP: n = 5 mice with 11 dendritic segments and 470 spines; Trained TRAP: n = 4 mice with 10 dendritic segments and 436 spines. p = 0.040, 2-way repeated-measures ANOVA. (F) Average normalized spine density of TRAP dendrites (red solid line) and non-TRAP dendrites (light red dashed line) in trained mice. TRAP: n = 4 mice with 10 dendritic segments and 436 spines; Non-TRAP: 4 mice with 10 dendritic segments and 430 spines. p = 0.044, 2-way repeated-measures ANOVA. (G) Average normalized spine density of TRAP dendrites (gray solid line) and non-TRAP dendrites (light gray dashed line) in control mice. TRAP: n = 5 mice with 11 dendritic segments and 470 spines; Non-TRAP: 5 mice with 11 dendritic segments and 549 spines. p = 0.521, 2-way repeated-measures ANOVA (H) Survival of learning-induced spines (formed on first two days of training) on TRAP dendrites (red solid line) and non-TRAP dendrites (light red dashed line) in trained mice. TRAP: n = 4 mice with 36 newly formed spines; Non-TRAP: 5 mice with 27 newly formed spines. p = 0.008, 2-way repeated-measures ANOVA. Difference persisted at day 15. TRAP: 43.0% ± 3.5%; Non-TRAP: 13.4% ± 3.6%. p = 0.001, paired t-test. (I) Survival of spines formed on first two days on TRAP dendrites (gray solid line) and non-TRAP dendrites (light gray dashed line) in control mice. TRAP: n = 5 mice with 25 newly formed spines, Non-TRAP: 5 mice with 29 newly formed spines. p = 0.766, 2-way repeated-measures ANOVA. No difference at day 15. TRAP: 19.2% ± 16.0%; Non-TRAP: 21.5% ± 8.8%. p = 0.909, paired t-test. *p < 0.05; **p < 0.01; ns, non-significant. See also Figure S2.

Figure 3:. Motor learning strengthens behavior-relevant M1 projections to striatal SPNs

(A) Schematic drawing of M1 TRAP neurons and their projection to striatal SPNs. (B) Fraction of SPNs receiving direct projection from TRAPed M1 neurons. Higher proportion of cells received projections from M1 engram neurons in late TRAP mice compared to early TRAP and control mice. Control: 66.67%, n = 24 neurons from 6 mice; Early TRAP: 81.25%, n = 32 neurons from 11 mice; Late TRAP: 97.06%, n = 34 neurons from 12 mice. Control vs. Early: p = 0.232; Control vs. Late TRAP: p = 0.003; Early vs. Late TRAP: p = 0.009, Fisher’s exact test. (C) Representative traces of optogenetic induced EPSCs (oEPSCs) in striatal SPNs. (D) oEPSCs amplitude in striatal SPNs. oEPSC amplitude in late TRAP mice is higher than in either control or early TRAP mice. Control: 113.27 ± 28.88 pA, n = 24 neurons from 6 mice; Early TRAP: 78.66 ± 18.58 pA, n = 32 neurons from11 mice; Late TRAP: 221.80 ± 33.22 pA, n = 34 neurons from 12 mice. Control vs. Early TRAP: p > 0.999; Control vs. Late TRAP: p = 0.008; Early vs. Late TRAP: p = 0.0003, Kruskal-Wallis test with Dunn’s multiple comparison. (E) Top: Schematic drawing of functional connectome analysis using combined 2-photon Ca²⁺ imaging with optogenetic stimulation. Axonal projections from TRAPed M1 neurons are shown in red and dendrites from striatal SPNs are shown in black. Bottom: Schematic drawing of NMDAR-mediated Ca²⁺ response at a dendritic spine receiving direct input from a TRAP axon terminal during optogenetic stimulation. (F) Top: Representative images of a segment of dendrite during three repeated optogenetic stimulation pulses. Yellow arrowheads indicate Ca²⁺ responses on active spines. Scale bar, 1 μm. Bottom: Fluorescence response from Ca²⁺ imaging using Fluo-5F of the two identified active spines above. Blue bars indicate timing of optogenetic stimulation. (G) Density of active spines on striatal SPNs. SPNs of late TRAP mice receive more projection from M1. Early TRAP: 0.012 ± 0.003 μm⁻¹, n = 14 dendrites from 9 cells and 6 mice; Late TRAP: 0.11 ± 0.02 μm⁻¹, n = 19 dendrites from 9 cells and 6 mice. p < 0.0001, Mann-Whitney test. (H) All spine density of striatal SPNs. No difference between early and late TRAP mice. Early TRAP: 0.60 ± 0.04 μm⁻¹, n = 14 dendrites from 9 cells and 6 mice; Late TRAP: 0.56 ± 0.03 μm⁻¹, n = 19 dendrites from 9 cells and 6 mice. p = 0.418, Mann-Whitney test. (I) Ca²⁺ response probability of active spines on SPNs during repeated optogenetic stimulation of TRAPed M1 inputs. Ca²⁺ probability is significantly higher in late TRAP mice than in early TRAP mice. Early TRAP: 0.44 ± 0.07, n = 19 spines from 9 cells and 6 mice; Late TRAP: 0.76 ± 0.02, n = 260 spines from 9 cells and 6 mice, p < 0.0001, Mann-Whitney test. (J) Distribution of Ca²⁺ response probabilities. Distribution in late TRAP mice is shifted toward the higher end. Early vs. Late TRAP: p = 0.003, Kolmogorov-Smirnov test. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, non-significant. See also Figure S3.

Figure 4:. Behavior-relevant inputs to SPNs form clusters during motor learning

(A) Spatial distribution of active spines along SPN dendrites in early TRAP mice. Upper: representative dendrite image with active spines labeled (circles). Lower: 5 representative dendrospinograms, identifying the spatial location of somas (black circles), active spines (colored vertical lines), and non-active spines (black vertical lines). (B) Spatial distribution of active spines along SPN dendrites in late TRAP mice. (C) Spatial distribution of active spines along SPN dendrites in Thy1-ChR2 mice. (D) Cumulative distribution of nearest neighbor distance (NND) for late TRAP (left) and Thy1-ChR2 mice (right). Red and green lines denote the NND distribution of active spines and black lines denote the distribution from Monte Carlo simulated “control spines” with 95% confidence interval (dashed). (E) Z-score of the NND distribution for late TRAP (red) and Thy1 group (green). Red and green shaded area marks the area under the curve that exceeds the 95% confidence interval of simulated “control spines”, representing the stretch of dendrite with clustered active spines. (F) Nearest neighbor index (NNI) in late TRAP and Thy1 mice. The NNI of late TRAP mice is significantly smaller than that of Thy1-ChR2 mice. Late TRAP: 0.60 ± 0.05, n = 19 dendrites from 9 cells and 6 mice; Thy1-ChR2: 0.71 ± 0.09, n = 13 dendrites from 9 cells and 6 mice. p = 0.001, Mann-Whitney test. (G) Ca²⁺ response probability of synapses inside and outside of clusters in late TRAP mice. Ca2+ probability is higher in spines inside clusters than outside. Left: comparison of NND and spine Ca2+ probability, right: comparison of average Ca²⁺ probability inside (NND < 10 μm) and outside (NND > 10 μm) of clusters. NND < 10: 0.74 ± 0.02, n=245 spines; NND > 10: 0.49 ± 0.08, n = 15 spines. p = 0.0015, Mann-Whitney test. (H) Ca²⁺ response probability of synapses inside and outside of clusters in late Thy1-ChR2 mice. No difference between spines inside and outside of clusters. NND < 10: 0.61 ± 0.024, n=122 spines; NND > 10: 0.52 ± 0.06, n = 22 spines. p = 0.105, Mann-Whitney test. (I) Schematic model showing increased numbers of active spines (green) that receive input from M1 engram neurons (magenta) and formation of clusters of active spines on striatal SPN dendrites after motor learning. *p < 0.05; **p < 0.01; ns, non-significant. See also Figure S4.