Learning binds new inputs into functional synaptic clusters via spinogenesis

Abstract

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.

Extended Data Figure 1.. Experimental setup, learning metrics, and image quality assurance

(a,b) Task performance improves over days of training. (a) The percentage of trials resulting in reward significantly increases over learning (p = 2e-31; Pearson’s correlation coefficient). Data points correspond to the mean fraction of successful trials ± SEM. **** p < 0.001. n = 53 mice. (b) The reaction time (time from cue onset to movement onset; black) as well as the time from cue to reward (red) and movement onset to reward significantly decrease over learning (p = 8e-35 for cue-to-reward; p = 7e-50 for cue-to-movement; p = 6e-4 for movement onset to reward, Pearson’s correlation coefficient). Data points correspond to mean time ± SEM. **** p < 0.001. n = 53 mice. (c) Schematic of imaging schedule. In each animal, three dendritic fields were imaged on each of the first three training days (Imaging Session 1, IS1), after which each field was imaged in 5-day intervals for two additional imaging sessions (IS2 and IS3). This yielded three imaging sessions for each field: early, middle, and late sessions. (d) Spine density does not significantly change over learning (1-way ANOVA, main effect of session: F = 0.94, d.f. = 2, p = 0.39). Individual dendrites are plotted as colored dots; black line corresponds to the median spine density ± bootstrapped 95% CI.n = 76 dendrites / 23 mice. (e) Spine event frequency is stable over learning for MRS (p = 0.46, rank-sum test), and decreases for nonMRSs (p = 1e-5, rank-sum test). Spine event frequencies are higher for MRSs in late sessions (early: p = 0.17, late: p = 7e-7). Bars represent median ± bootstrapped 95% CI. n = 898 early MRSs, 820 late MRSs, 869 early nonMRSs, 788 late nonMRS. Multiple comparisons were corrected using the FDR method. (f) Example lever position trace, showing movement periods in green shadings. (g) Example of frame-wise 2d correlation with the corresponding calculated reference image (i.e., an iteratively aligned and average-projected version of the imaging field) associated with the behavioral window shown in (d). Prior to motion correction (red line), frame correlations with the reference image show with a decrease during movements. After motion correction, mean frame correlations are higher regardless of movements (blue line). (h) Example x-y pixel shifts used for motion correction of each frame in the behavioral window shown in (f,g), showing that extra correction during movements successfully compensated for movement artifacts. (i) Average projection images from the ~4s behavioral window indicated in (g). Prior tomotion correction, images are blurry and individual structures are difficult to resolve because of misalignment across frames (top). After motion correction, images are sharp, and individual spines are visible (bottom). (j) Motion correction generates stable frames over movement. When aligned to movement onset, frame correlations with the reference image (as in (e)) decrease sharply during movements in the raw, pre-correction images (red line), but are stable during movements for the post-correction frames (blue line). Curves represent mean change in reference imaged correlation. Shaded portions correspond to SEM.

Extended Data Figure 2.. Workflow of in vivo imaging followed by correlated ex vivo electron microscopy

(a) Workflow of correlated light and electron microscopy (CLEM) to identify the in vivo- imaged dendrites for EM. From top left moving clockwise:in vivo images of dendrites imaged during an experiment are acquired a final time for reference of the most updated structural information. Vascular maps are then generated by 2p in vivo imaging of TRITC-Dextran (injected retro-orbitally on final day of in vivo imaging), capturing an ~200𝜇m z-stack in the area surrounding the dendrite of interest for detailed structural landmarks (panel 2 and white square in panel 3) and a low-zoom (usually 1x / ~1000𝜇m) to provide the larger structural context with respect to the vasculature. Arrows indicate the vasculature used for aligning the subsequent images. After the animal is perfused with fixative, thin sections are cut, and bright-field imaging is used to find the vasculature and approximate the location of the dendrite (white square in panel 4). Confocal images are then acquired using the slice containing the target vasculature so as to ensure the location of the target dendrite (white square in panel 5). The slice of interest is then processed for EM, thereby permanently rendering the tissue photo-inaccessible. The tissue is then exposed to iterative steps of X-Ray tomography and trimming (guided by software-based non-rigid alignment) so as to isolate the dendrite within a small volume amenable to EM. Shown is an alignment of the confocal stack with the x-ray tomogram. Colored arrowheads point to the dendrites of interest identified in subsequent images. Serial-section scanning-blockface EM is then performed on the final tissue block, and the confocal images are aligned to the output using the accumulated fiducial structures over previous imaging modalities. The final result (panel 8) is an aligned stack of confocal fluorescence and EM data, allowing a fluorescence overlay onto the EM stack. Using this alignment, the target dendrite and constituent spines are then reconstructed using the in vivo 2p image as reference for the domains of interest. These experiments were repeated independently in n = 4 mice with comparable results. (b) After reconstruction of the target dendrite(s) and spines (left), the synaptically connected axonal partner of each spine was traced to the full extent of the imaging volume or to the furthest point of confident identification (right). In this volume, 59 axons were successfully reconstructed to various extents. (c) Results of all CLEM experiments from the current study. Left, in vivo 2-photon average projections, aligned EM data, and the corresponding reconstruction of the target dendrites are shown. Reconstruction images are angled so as to accentuate the correspondence between 2p and EM. Middle, zoomed-in visualizations of the reconstructions of target dendrites prior to reconstruction of synaptically connected axons. The number of axons reconstructed in each volume were: 137 (top), 74 (middle), and 60 (bottom), for a total of 271 axons. Note that the axonal partners of spines near the ends of dendrites were often not reconstructed due to their proximity to the edge of the sample volume.

Extended Data Figure 3.. Characterization of movement-related signals at single dendritic spines

(a) Schematic of MRS definition. The dot product of the ΔF/F₀ trace with binarized movement traces defines the movement score, x, for a given spine. Movements are then shuffled in time 1000x (without breaking individual movements), and the corresponding shuffled scores calculated. The real and shuffled scores are then ranked together. Those spines whose movement score is above the 97.5^th percentile are considered MRS. (b) Example traces of movement onset-aligned activity for 12 individual MRSs (green, top) and 12 new spines (blue, bottom). Movement onset is indicated by black vertical dashed line. Activity onset timing is indicated with red triangles. Mean ± SEM. (c) Explanation of event detection based on previously published^, methods. Left, full ΔF/F₀ trace for an example spine. For each trace, noise is estimated based on the negative portion of the trace, reflected about the x-axis. Values above 1x the standard deviation of this trace are considered above the “baseline” threshold (light blue), and values above 2x the standard deviation are considered above the “active” threshold (red). Green rectangle indicates portion of the trace inspected in right. Upper right, zoomed-in section of the trace for easier visualization. Tan region indicates example region used to demonstrate end and start times of active periods. End times (red asterisk) are defined as when a smoothed version of the trace falls below the active threshold, and start times (green asterisk) are defined by moving backwards in time from each end point to the first point where the raw trace falls below the baseline threshold. All frames meeting these criteria are assigned a value of 1, and all other frames a value of 0. In this way, events are considered continuous unless they fall below the baseline threshold. The resulting binarized trace (bottom right) indicates when spines are active and inactive. (d) Relationship of spine active periods to movements. Using the binarized activity described in (c), the relationship of spines’ activity can be shown as either the fraction of movements during which a given spine is active (left) or the fraction of activity for a given spine that occurs during movements (right). Spines are subdivided into MRSs and nonMRS, as described in (a). Bars correspond to the medians ± 95% CI. MRSs are active during a higher fraction of movements than nonMRSs in both early (p = 8e-52; rank-sum test) and late (p = 2e-70; rank-sum test) sessions, and while MRSs are active during a similar fraction of movements across sessions (p = 0.95; rank-sum test), nonMRSs show a significant decrease (p = 5e-7; rank-sum test). Similarly, a higher fraction of MRS activity occurs during movements than nonMRSs in both early (p = 1e-72; rank-sum test) and late (p = 4e-93; rank-sum test) sessions, though both MRSs (p = 9e-101; rank-sum test) and nonMRS (p = 4e-105; rank-sum test) show an increased fraction of activity occurring during movements by late sessions. All comparisons were two-sided. Multiple comparisons were corrected using the FDR method. **** p < 0.001. (e) The co-activity rates of simulated spine pairs (Methods) correlate better with the geometric mean of the constituent spine pairs’ event frequencies than the arithmetic mean, demonstrating that the geometric mean is a more suitable normalization factor for co-activity rates. Overall statistics for arithmetic mean: r = 0.91, p <0.0001; geometric mean: r = 0.95, p < 0.0001. The r-value was higher for the fit between geometric mean and co-activity rates in 1000/1000 shuffles.

Extended Data Figure 4.. Spine identification and CLEM-based corroboration

(a) Example of ROIs. Top, in vivo image under consideration. Bottom, elliptical ROIs manually drawn in the initial analysis of this dendrite. Magenta ellipses correspond to ROIs that were successfully located in EM, while blue ellipses show those that were not confidently identified. Two example regions are indicated for closer inspection in (b). (b) Zoomed-in regions shown in (a) for closer inspection of ROI drawing. Top row: deconvolved in vivo images of the two regions shown in (a). Deconvolved images were used throughout this study to aid in spine identification. Second row: corresponding EM reconstructions showing the same spines imaged in vivo. Third row: reconstructions of the associated regions of dendrite, but with post-synaptic density (PSD) reconstructions shown in red. Fourth row: example EM micrographs demonstrating PSDs on a selection of spines (numbers correspond to spines in above image). Scale bars in the fourth row (green) correspond to 0.5μm. (c) Summary of the fraction of ROIs identified in vivo that were successfully located in EM. The vast majority of structures drawn as spines in vivo were also found in EM. Of those structures that were not located, most were due to technical failures in EM acquisition (1 of the 16 of such failures was due to debris from a previous slice obfuscating the region of the target dendrite, and 15 were due to slices that were skipped, likely due to uneven cutting). We also identified one highly localized dendritic region that appeared damaged/blebbed despite a healthy appearance in vivo, preventing the assessment of two spines observed in vivo. The four remaining ROIs were not located due to unknown reasons, suggesting either mislabeling or elimination of these spines between the final in vivo imaging session and EM processing (see a potential example of this in the blue-encircled spine in (b). (d) Summary of the fraction of all spines identified in EM (within the dendrites captured in vivo) that were also labeled as ROIs in vivo. The majority of spines that were not identified in vivo were co-axial with either another (usually larger, see (e)) spine (red slice), or the dendritic branch (purple slice). A small subset of spines located in EM were obscured by another fluorescent structure (e.g., a labeled axon) in vivo (blue slice). The remaining spines were not identified in vivo for unknown reasons, which might include the rapid formation of spines between the final imaging session and EM processing, or most of these spines being below the detectable size for in vivo imaging (see (e)). (e) Summary histograms of the volumes (as calculated in EM) of spines that were also captured in vivo (green) and spines that were not identified in vivo (gray). The vast majority of spines that were identified in EM but not labeled in vivo were small in size; 67% of “missed” spines were within the 3^rd percentile of the “captured” population. The medians of the “missed” and “captured” groups are indicated with black and green arrows, respectively.

Extended Data Figure 5.. Analysis of no-task controls

(a) Schematic of “no-task” condition compared to the typical “learning” condition. Unlike in the learning condition, the no-task condition administers water rewards at the end of each cue period, regardless of the movement of the lever. Mice were head-fixed in the experimental apparatus only during the 9 imaging sessions. (b) Heat map of lever press correlations across sessions of the “no-task” experiment. Black portions correspond to days during which mice were not placed in the experimental apparatus. n = 14 mice. (c) Lever movement correlations, both within and across adjacent sessions, did not increase over time in the no-task controls. Significance of the relationship was evaluated with Pearson’s correlation coefficient, with p-values shown in figure text. Data points correspond to mean ± SEM correlation values for each session. n = 14 mice. (d) Activity event frequency of different spine types for the no-task condition. In both early and late sessions, MRSs and nonMRSs show comparable event frequencies (early: p = 0.07, late: p = 0.4, rank-sum test). MRSs and nonMRS slightly decrease (MRSs: p = 0.0035, nonMRSs: p = 9e-8) their activity event frequencies over the experiment. Multiple comparisons corrected using the FDR method. When pooled with the learning-group data, only late MRSs are significantly higher in the learning group than in the no-task control after correcting for multiple comparisons (p = 0.006, rank-sum test). Conversely, nonMRS showed higher event frequencies in both early and late sessions for the no-task group (early: p = 2e-4, late: p 0.002, rank-sum test). n = 433 early MRSs, 1270 early nonMRSs, 506 late MRSs, 918 late nonMRSs. Bars represent median ± 95% CI. Individual data points (spines) shown as black dots. (e) MRS density was lower in the no-task group (0.21/μm, 95% CI = [0.17, 0.25]) than the learning group ((0.34/μm, 95% CI = [0.29, 0.38]) (p = 4e-4, rank-sum test). Bars represent median values ± 95% CI. Circles correspond to individual dendrites, color-coded by animal. Note that all imaged dendrites are shown, including those that show no new spine formation. n = 40 no-task group dendrites, 76 learning-group dendrites. (f) The new spine density on imaged dendrites was lower in the no-task group (0.02/μm, 95% CI = [0.010 0.027]) than the learning group (0.039/μm, 95% CI = [0.022, 0.054]) (p = 0.04, rank-sum test). Bars represent median values. Circles correspond to individual dendrites, color-coded by animal. Note that all imaged dendrites are shown, including those that show no new spine formation. n = 40 no-task control dendrites, 76 learning-group dendrites. (g) A higher fraction of new spines was transient in the control group vs. learning group (p = 0.02, Chi-squared test). (h) Functional clustering was intact in the no-task condition, but co-activity was generally less pronounced. Much like the learning condition, MRSs showed higher overall co-activity levels than nonMRSs (3-way ANOVA [MRS status × distance × session], main effect of MRS status: F = 976.58, d.f. = 1, p = 2e-211; post-hoc using least-significant difference: p < 0.0001 for MRS vs. nonMRS across all distance bins when collapsing across sessions or when preserving session identity). We observed a main effect of distance (F = 109.38, d.f. = 5, p = 5e-211), and the first distance bin was higher than all other bins (p = 1e-30, post-hoc using LSD), indicating that functional clustering is present. By comparing to the learning-group data (Fig 1j), we find that overall co-activity values were higher in the learning group (4-way ANOVA [MRS status × distance × session × condition], main effect of condition, p = 6e-54; post-hoc using LSD). We also observe a significant MRS status × condition interaction (F = 106.27, d.f. = 1, p = 7e-25), with post-hoc analysis revealing that both MRSs (p = 2e-57) and nonMRS (p = 1e=5) show higher co-activity rates in the learning condition than their no-task counterparts. n = 6376 early MRS pairs, 49894 early nonMRS pairs, 8088 late MRS pairs, 47274 late nonMRS pairs. Y-axis scale was set to match that of the learning group (Fig. 1j). Mean ± SEM. (i) Neither MRS density (r = 0.03, p = 0.64) nor nonMRS density (r = −0.10, p = 0.16, Pearson’s correlation coefficient) changed as a function of distance to the nearest new spine in the no-task controls. Mean ± SEM. (j) New spines formed during the no-task control are weakly functionally clustered with MRSs. Like in the learning condition, new spines formed in the no-task condition show slightly higher overall co-activity rates with MRSs vs. nonMRSs (2-way ANOVA [MRS status × distance], main effect of MRS status: F = 4.2, d.f. = 1, p = 0.04) and closer spine pairs generally had higher co-activity (main effect of distance: F = 2.3, d.f. = 5, p = 0.04). No significant interaction term was observed. When comparing to the learning condition data (Fig. 4c), we observed a significant main effect of MRS status and distance (3-way ANOVA [MRS status × distance × condition], MRS status: F = 22.85, d.f. = 1, p = 2e-6; distance: F = 14.87, d.f. = 5, p = 2e-14), but not condition (i.e. no-task vs. learning). However, we found a significant interaction between distance and condition (F = 5.66, d.f. = 5, p = 3e-5), suggesting that the spatial relationship of co-activity rates differs across the two conditions. Indeed, post-hoc comparisons show that only the learning condition shows significant functional clustering, such that the first distance bin is greater than all other distance bins (p <0.0001 for learning condition at distance = 0–5um vs. all other bins; p > 0.05 for no-task condition at distance = 0–5um vs. all other bins). n = 789 new spine-MRS pairs, 2029 new spine-nonMRS pairs. Y-axis scale was set to match that of the learning group (Fig. 3c). Mean ± SEM. (k) Like in the learning condition, new spine-MRS pairs show activity that is enriched during movements, more so than new spine activity alone (p = 7e-5; sign-rank test) or MRS activity alone (p = 0.037; sign-rank test). Multiple comparisons were corrected for using FDR. n = 21 fields showing new spine-MRS co-activity / 25 total fields. Bars represent median ± 95% CI. (l) Movements containing new spine-MRS co-activity are not more correlated with the learned movement pattern (LMP) than movements lacking such activity (p = 0.43, sign-rank test). Data points correspond to the median correlation of movements occurring with (blue) or without (red) co-activity of new spine-MRS pairs in a given field. n = 18 fields showing new spine-MRS co-activity during at least 3 movements / 21 fields showing new spine-MRS co-activity / 25 total fields. (m) The functional separation of sustained new spine-MRS pairs from other pair types is conspicuously absent. New spine-MRS pairs, irrespective of survival, showed slightly higher overall co-activity rates (3-way ANOVA [MRS status × distance × survival], main effect of MRS: p = 0.001), and spine pairs were generally functionally clustered (main effect of distance, p = 0.02), but no interaction terms were significant. Notably, we observed a main effect of new spine survival (p = 0.01), but post-hoc inspection revealed that transient new spine-MRS co-activity was actually slightly higher than the other groups (p < 0.02 for all comparisons). Mean ± SEM.

Extended Data Figure 6.. Additional spatial analyses of dendritic structural and functional features with respect to new spines

(a) Overall spine density as a function of distance from new spines. While there is a trend towards higher spine densities closer to new spines, the effect is not significant (Pearson’s correlation coefficient; r = −0.10, p = 0.13). Data points correspond to the mean spine density ± SEM of all dendrites displaying new spine formation (n = 50 dendrites across 21 mice). (b) A large fraction of new spine-MRS co-activity occurs during movements regardless of the distance between the spines (ANOVA with post-hoc test using the least-significant difference; main effect of distance p = 0.11). Data points correspond to the mean ± SEM of all new spine-MRS pairs (n = 1658 pairs across 50 dendrites from 21 mice). (c) Movements coincident with new spine-MRS coactivity have similarly high correlations with the learned movement pattern regardless of the distance between the spines (ANOVA with post-hoc test using the least-significant difference; main effect of distance: p = 0.85). Data points correspond to the mean (± SEM) correlation of movements coincident with co-activity of a given new spine-MRS pair (n = 1658 pairs / 50 dendrites / 21 mice). (d) The fraction of movements coincident with co-activity of a particular new spine-MRS pair shows a significant decrease with increasing distance between the spines, illustrating that a higher number of movements is encoded by closer spine pairs (ANOVA with post-hoc test using the least-significant different; main effect of distance: p = 0.9e-05; p <0.02 for post-hoc comparison of first distance bin vs. all other distance bins). Mean ± SEM. (n = 1658 pairs / 50 dendrites / 21 mice).

Extended Data Figure 7.. Supporting evidence of in vivo spine volume estimates using iGluSnFR

(a) Effect of different enlargement threshold cutoff values (from 1.1, light green, to 2, magenta) on the relationship between the probability of MRS enlargement and distance to the nearest new spine. Data points correspond to the mean probability of spine enlargement (± SEM) for MRSs in a given distance bin for each new spine imaged. n = 118 new spines, 697 MRSs present on 50 new-spine-containing dendrites across 21 mice. (b) Summary of Pearson’s statistics for data shown in (a). The relationship is significant for threshold values up to 1.5-fold change in spine volume. Data points correspond to either the calculated Pearson’s correlation coefficient (black, left axis), or the corresponding p-value (red, right axis) calculated from statistical tests on the data groups shown in (a). (c,d) Relationship between estimated spine volume and spine event frequency in early (c) and late (d) sessions, revealing a lack of positive correlation, suggesting that our methods using iGluSnFR fluorescence do not overestimate the volume of highly active spines. Linear fit of data shown in red. Significance of the relationship was determined from Pearson’s correlation coefficient, with r and p-values shown in figure text. **** p < 0.001, ns: not significant. (e) Evaluation of the effect of removing active periods from image projections prior to calculation of spine volume. In 11 of the 23 “learning group” animals (769 spines), spine volume was re-calculated after removal of all frames in which a given spine was considered “active”. “Activity-removed” spine volume estimates correlate strongly and significantly with spine volume estimates from full-length time series projections. Unity line shown in dashed blue. Linear fit of data shown in red. Significance of the relationship was determined from Pearson’s correlation coefficient, with r and p-values shown in figure text. **** p < 0.001 (f) Example in vivo, EM, and reconstruction images illustrating the presence of spine apparati, a signature of mature and potentiated spines, in spines that showed enlargement in vivo during learning. Green arrows indicate spines of interest. Each spine of interest in the in vivo images is outlined for spine size comparisons (red = early; green = late), and the early and late outlines are overlaid in the bottom left corner of each late session. Red arrows in the EM images indicate spine apparati. Reconstructed spine apparati are shown as yellow structures within spines (magenta).

Extended Data Figure 8.. Cases of axon-sharing between spines

Two additional example cases of axon sharing between spines on imaged dendrites. The left-most column shows reconstructions of the dendrites and its spines along with the axon being shared between two spines. Each synapse occurring with the dendrite of interest is circled, and corresponding EM images illustrating synaptic contact are shown on the right (further indicated with red arrows). The top example shows a double connection between two pre-existing spines immediately after a dendritic branch point. The bottom example shows a double connection formed by a new spine and a nearby pre-existing spine.

Extended Data Figure 9.. Limiting analysis to rewarded movement-related spines (rMRSs) yields reproduces main findings

(a) Pre-existing rMRSs show strong functional clustering. Data points correspond to mean ± SEM. n = 443 early rMRSs / 2052 condendritic rMRS-rMRS pairs; 1472 early nonrMRS-nonrMRS pairs; 571 late rMRSs / 3278 late nonrMRS-nonrMRS pairs. (b) rMRS density nearby (≤10μm) new spines is higher than expected by chance, as estimated by randomizing the location of new spines. Statistical significance was determined by a one-sided comparison of the median of each shuffle across all new spines to the real data median; the resulting fraction of shuffles that are consistent with the null hypothesis (i.e. that there is not a higher nearby MRS density than expected by chance) corresponds to the p value; p<0.001. Inset, median nearby rMRS density of all new spines compared to the median chance estimate. Error bars correspond to bootstrapped 95% CI. n = 118 new spines across 50 dendrites. (c) rMRS density decreases as a function of distance from the nearest new spines, whereas nonrMRS density does not. MRS density: Spearman’s r = −0.16, p = 0.01; nonMRS density: Spearman’s r = 0.03, p = 0.67). Data points correspond to mean ± SEM. n = 118 new spines across 50 dendrites. (d) rMRS enlargement probability decreases as a function of distance from the nearest new spine. Spearman’s r = −0.11, p = 0.048). Data points correspond to mean ± SEM. n = 118 new spines; 443 rMRSs present on 50 new spine-containing dendrites. (e) New spines are functionally clustered with rMRSs in a manner similar to MRSs. Data points correspond to mean ± SEM. n = 1091 new spine-rMRS pairs; 2774 new spine-nonrMRS pairs / 118 new spines across 50 dendrites / 21 mice. (f) The fraction of activity of new spine-rMRS pairs occurring during movements trends higher than either new spine activity alone (p = 1e-6) or rMRS activity alone (p = 0.1; sign-rank rest correcting for multiple comparisons using the FDR method). n = 34 fields. Bars correspond to the median fraction of cases for each group present in each of the 34 fields imaged. Error bars represent bootstrapped 95% CI. (g) Rewarded movements associated with new spine-rMRS co-activity are not more correlated with the learned movement pattern (LMP) than rewarded movements lacking such activity (p = 0.14). Thus, it appears that preferential occurrence of co-activity during rewarded movements lead to the similarity of movements to LMP shown in Fig. 4g. (h) The survival of new spines can be predicted by co-activity with nearby rMRSs. New spine-rMRS pairs showed higher overall co-activity rates than other spine pairs (3-way ANOVA [rMRS status × distance × survival], main effect of rMRS status: p = 3e-11), and co-activity values were significantly impacted by distance (main effect of distance, p = 0.0006). We observed a main effect of new spine survival (p = 0.007), and trends for interaction terms distance × survival (p = 0.055) and rMRS × distance × survival (p = 0.058). Given the trend in the three-way interaction term, we performed post-hoc comparisons of the four groups across each distance bin. Sustained new spine-rMRS co-activity at close distances (within 5μm) was significantly higher in 17 of the 18 possible comparisons against the other three groups (p values range from 7e-12 to 0.02 for the 17 significant comparisons, p = 0.47 for the final comparison), and was higher than values within the same group at longer distances (>15, p <0.008). Conversely, transient new spines show co-activity that is not higher at closer distances (transient new spine-MRS pairs’ co-activity within 5μm is not different than any other distance bin (p values range from 0.27 to 0.73); transient new spine-nonMRS pairs’ show co-activity that is significantly higher (p = 0.02) than one other bin and significantly lower (p = 0.046) than another bin), suggesting that no significant functional clustering is present between transient new spines and other spines. Further, neither transient spine group showed a significant correlation with distance (transient new spines-MRS co-activity vs. distance, r = 0.06, p = 0.39; transient new spines-nonMRS co-activity vs. distance: r = −0.07, p = 0.14), while sustained new spine co-activity is significantly negatively correlated with distance (sustained new spine-MRS co-activity vs. distance: r = −0.18, p = 0.0001; sustained new spine-nonMRS co-activity vs. distance: r = −0.14, p = 6e-5). Thus, surviving new spines’ co-activity with rMRSs is higher and more spatially selective than transient new spines. n = 28 transient new spines, 230 transient new spine-rMRS pairs, 669 transient new spine-nonrMRS pairs; 57 sustained middle-session new spines, 530 sustained new spine-rMRS pairs, 1259 sustained new spine-nonrMRS pairs. Mean ± SEM.

Extended Data Figure 10:. Main findings are robust against changes in the threshold for MRS definition

Shifting the threshold for defining MRSs to being greater than either the 95^th percentile of shuffles or the 99.5^th percentile of shuffles has negligible effects on the main findings. (a) Functional clustering of pre-existing spines is similar when using both 95^th and 99.5^th percentiles to define MRSs. 95^th percentile cutoff: MRSs are still more co-active than nonMRS (3-way ANOVA, main effect of MRS status: F = 1092.04, d.f. = 1, p = 2e-233), and spines show distance-dependent co-activity rates (main effect of distance: F = 132.07, d.f. = 6, p = 3e-138), such that closer pairs are more co-active (Pearson’s correlation coefficient, r values range from 0.14–0.22 and p < 0.0001 for all 8 separate groups shown). More conservative cutoff results in higher co-activity rates of MRS pairs. n = 687 early MRSs / 4318 early MRS-MRS pairs / 1080 early nonMRSs / 13621 early nonMRS-nonMRS pairs / 685 late MRSs / 4687 late MRS-MRS; 1230 late nonMRS / 12318 late nonMRS-nonMRS pairs at the 99.5^th percentile; n = 954 early MRSs / 7719 early MRS-MRS pairs / 813 early nonMRSs / 8534 early nonMRS-nonMRS pairs / 868 late MRSs / 7313 late MRS-MRS pairs / 788 late nonMRSs / 8967 late nonMRS-nonMRS pairs. (b) The higher density of nearby MRSs is robust to the different MRS thresholds. Main plots show cumulative probability distributions of the nearby MRS density for real new spines (light blue) and chance, estimated by shuffling new spine locations 1000 times for each new spine (250 individual shuffles shown in gray, median of all shuffles shown in black). Statistical significance was determined by comparing the median of each shuffle across all new spines to the real data median; the resulting fraction of shuffles that are consistent with the null hypothesis (i.e. that there is not a higher nearby MRS density than expected by chance) corresponds to the p value. 95^th percentile MRS density: p = 0.003. 99.5^th percentile MRS density: p = 0.012. n = 118 new spines across 50 dendrites. (c) The distance-dependent decrease in MRS density is similar across different thresholds. 95^th percentile: MRSs: Spearman’s rank coefficient, r = −0.12, p = 0.045, nonMRS: r = 0.07, p = 0.212. MRSs show a significantly different slope than nonMRS (p < 0.001). 99.5^th percentile: MRSs: r = −0.11, p = 0.052. n = 1390 new spine-MRS pairs / 2475 new spine-nonMRS pairs at 99.5^th percentile; 1731 new spine-MRS pairs / 2131 new spine-nonMRS pairs at 95^th percentile. (d) Volume increase probability with respect to new spine location is similar across different MRS thresholds. 95^th percentile cutoff: r = −0.07, p = 0.062 (Pearson’s correlation coefficient. 99.5^th percentile cutoff: r = −0.-8, p = 0.06 (Pearson’s correlation coefficient). n = 1390 new spine-MRS pairs / 2475 new spine-nonMRS pairs at 99.5^th percentile; 1734 new spine-MRS pairs / 2131 new spine-nonMRS pairs at 95^th percentile. (e) New spine functional clustering is nearly identical across different thresholds. 95^th percentile MRSs: New spine-MRS pairs are more co-active than new spine-nonMRS pairs (2-way ANOVA, main effect of MRS status: F = 18.71, d.f. = 1, p = 2e-5). New spine co-activity generally shows a distance-dependent change in co-activity rates (main effect of distance: F = 20.97, d.f. = 6, p = 1e-20), an effect that did not depend on MRS status, (MRS status by distance interaction: F = 1.5, d.f. = 5, p = 0.19). Both groups showed higher co-activity at the 0–5μm bin than all other bins (post-hoc using LSD: p < 0.05 for all). n = 1390 new spine-MRS pairs / 2475 new spine-nonMRS pairs at 99.5^th percentile; 1734 new spine-MRS pairs / 2131 new spine-nonMRS pairs at 95^th percentile. (f) The functional separation of transient and sustained new spines with respect to their co-activity with nearby spines is comparable for both thresholds. 95^th percentile cutoff: as in other conditions, new spine-MRS pairs are more co-active than new spine-nonMRS pairs (3-way ANOVA, main effect of MRS status: F = 24.91, d.f. = 1, p = 7e-7), and pairs are differentially co-active at closer distances (main effect of distance: F = 5.77, d.f. = 5, p = 3e-5). Only surviving new spines show a significant negative correlation with distance (surviving new spine-MRS vs. distance, r = −0.16, p = 2e-5 [Spearman’s: r = −0.13, p < 0.001]; surviving new spine-nonMRS vs distance, r = −0.13, p = 0.003 [Spearman’s r =0– 0.13, p = 0.003] ; transient new spine-MRS co-activity vs distance, r = −0.07, p = 0.2 [Spearman’s r = −0.09, p = 0.13] , transient new spine-nonMRS co-activity vs. distance: r = −0.04, p = 0.54 [Spearman’s r = −0.06, p = 0.38], Pearson’s correlation coefficient). n = 339 transient new spine-MRS pairs / 560 transient new spine-nonMRS pairs / 618 sustained new spine-MRS pairs / 1171 sustained new spine-nonMRS pairs at 99.5^th percentile; 451 transient new spine-MRS pairs / 448 transient new spine-nonMRS pairs / 833 sustained new spine-MRS pairs / 956 sustained new spine-nonMRS pairs at 95^th percentile. (g) The fraction of activity occurring during movements is highly similar for different MRS thresholds. The fraction of new spine-MRS co-activity events occurring during movements is higher than the fraction of new spine-only (95^th: p = 1e-6; 99.5^th: p = 7e-5) and MRS-only (95^th: p = 0.003; 99.5^th: p = 2e-4, rank-sum test) events. n = 34 fields. Median ± 95% CI. (h) The correlation of movements coinciding with new spine-MRS co-activity with the learned movement pattern (LMP) is robust to MRS threshold. Both thresholds yield MRSs that, when co-active with new spines, signal movements more similar to the LMP than movements lacking such activity (95^th: p = 0.002, 99.5^th: p = 0.0009, sign-rank test). n = 34 fields. Individual data points correspond to the median correlation of all movements concurrent with co-activity (blue) or lacking such activity (red) of each imaged field. Overall median of each group shown in color-coded line.

Figure 1:. Characterization of movement-related signals in dendritic spines in M1 during motor learning

(a) Schematic of experimental setup. (b) Task structure. (c) Lever movement traces during rewarded trials in sessions 1 and 14 for one mouse. Grey: ten individual trials; black: average of all trials. (d) Correlogram of lever trajectory correlation within and across sessions (n = 53 mice). Each box corresponds to the median pairwise correlation coefficients of rewarded movement trajectories over a 3 s window (as shown in c), averaged across animals. (e) The mean lever trajectory correlation increases both within (black line; center diagonal from d; p = 0.00005, Pearson’s correlation) and across (gray line; +1 diagonal from d; p = 0.0002, two-sided Pearson’s correlation coefficient) learning sessions. n = 53 mice. Data points correspond to mean ± SEM. (f) Example average-intensity projection of a single-plane in vivo imaging time series of an iGluSnFR-expressing apical dendrite of a L2/3 excitatory neuron in M1, along with associated fluorescence traces of a subset of spines (demarcated with blue outline). Numbers correspond to associated spine traces. Spines classified as MRSs are labeled in green. Portions of each trace classified as “active” are demarcated with overlaid green (for MRSs) or blue (for other spines) lines. (g) Comparison of movement-onset-aligned average fluorescence traces of MRSs (green) and nonMRSs (red). Spine numbers correspond to the spines in f. (h) Movement onset-aligned, z-scored activity from all imaged MRSs pooled across all sessions, sorted by peak timing. n = 2554 MRSs. (i) Onset timing histogram of MRS activity with respect to movement onset. Gray shaded region corresponds to the period after movement onset (t = 0). Median onset value (-280ms) is plotted as a vertical dashed red line. n = 2554 MRSs. (j) Pairs of MRSs show higher co-activity rates than non-MRS pairs in both early (1–3) and late (11–13) sessions (2-way ANOVA (F = 418.3, d.f. = 3;) with post-hoc test using the least-significant difference (LSD): early MRS vs early nonMRS: p = 5e-114; late MRS vs. late nonMRS: p = 5e-157). Closer pairs show higher co-activity rates in all groups (2-way ANOVA, main effect of distance: F = 78.1, d.f = 6, p = 1e-96; group × distance interaction: F = 2.8, d.f. = 18, p = 6e-5; first distance bin vs all other bins: p<0.0001 for all groups), and co-activity rates significantly negatively correlate with distance (early MRS pairs: r = −0.15, p = 1e-25; early nonMRS pairs: r = −0.09, p = 1e-8; late MRS pairs: r = −0.12, p = 4e-12; late nonMRS pairs: r = −0.18, p = 6e-30). Co-activity rates were normalized in each spine pair by the geometric mean of the activity rates of the two spines in the pair. n = 894 early MRSs / 6990 early MRS-MRS pairs; 873 early nonMRSs / 9434 early nonMRS-nonMRS pairs; 818 late MRSs / 6556 late MRS-MRS pairs; 1097 late nonMRS-nonMRS pairs. Cases of only one MRS being present on an imaged dendrite (6 total MRSs / dendrites) were necessarily excluded from co-activity analysis of MRSs. Mean ± SEM.

Fig. 2.. iGluSnFR fluorescence signals are dependent on glutamate sensitivity

Comparison of iGluSnFR signals using excitation at 925 nm, which was used for the experiments described in this study, with signals using excitation at 810 nm, which is near the isosbestic point of iGluSnFR such that iGluSnFR signals are independent of glutamate concentration. Separate cohorts of mice were used for 810 nm and 925 nm imaging. (a) Comparison of the fraction of time spines were considered “active” while imaging iGluSNFR with excitation at 925nm (red) vs. 810nm (grey). Left, histograms of the fraction of time active (i.e. total time above threshold divided by the total imaging time) for all spines imaged across all animals in both groups. Right, bar graph summary of the fraction of time active for each animal. p<0.001, two-sided rank-sum test. Median ± bootstrap 95% confidence intervals. n = 5 mice / 17 fields / 33 dendrites / 484 spines imaged at 810nm; n = 23 mice / 45 fields / 76 dendrites / 1915 spines for 925nm. Individual data points show average for each mouse. (b) Comparisons of co-activity rates based on excitation wavelength. Left, histograms of all co-activity rates across all spine pairs within the same imaging field imaged at the corresponding wavelength. n = 5 mice / 17 fields / 33 dendrites / 484 spines / 4909 condendritic spine pairs imaged at 810nm; n = 23 mice / 45 fields / 76 dendrites / 1915 spines / 94060 condendritic spine pairs imaged at 925nm. Right, bar graph summary of the co-activity rates of each animal imaged at the corresponding wavelength. p = 4e-4, two-sided rank-sum test. Median ± bootstrap 95% confidence intervals. Individual data points show median for each mouse. n = 5 mice imaged at 810nm; n = 23 mice imaged at 925nm.

Figure 3:. New spines form near enlarged, movement-related spines in the motor cortex during motor learning

(a) In vivo images (average intensity projections of time series) of an iGluSnFR-expressing dendrite from early (top) and late (bottom) learning sessions, with MRSs (green) and nonMRSs (red) labeled. Arrowhead indicates a new spine. Asterisks indicate spines used for spine volume examples in (e). (b) New spines form in regions of higher MRS density. Left, cumulative probability plot of the density of MRSs within 10μm of new spines (light blue) in the learning condition compared to shuffled values found by randomizing new spine locations 10,000 times (black) from the same data. Right, analogous plot for the no-task control condition. Individual shuffles plotted in grey. Insets, the median density of nearby MRSs is significantly higher for new spines than for shuffled data (p = 0.003), an effect that was similar but did not reach statistical significance in no-task controls (p = 0.06; p-value calculated as the fraction of shuffles following the null hypothesis via a one-sided query of whether the median of each shuffle was greater than the median of the real data). Median ± bootstrapped 95% confidence intervals. n = 118 new spines (learning group); n = 51 new spines (no-task control group). (c) MRS density, but not nonMRS density, decreases as a function of distance from new spines. (2-way ANOVA [MRS label × distance], significant interaction of MRS label with distance (F = 2.27, d.f. = 5, p = 0.046), showing that the effect of MRS status on the density measurement depends on distance from the new spine. Correspondingly, MRS density is significantly negatively correlated with distance (Pearson’s correlation coefficient, r = −0.15, p = 0.009), while nonMRS show no such correlation (r = 0.10, p = 0.09)). Mean ± SEM. n = 118 new spines; 697 MRSs present on 50 new spine-containing dendrites. All statistical tests two-sided. (d) Left, in vivo images of an example MRS that showed 58% volume enlargement over learning (displayed spine indicated with * in a). Spine outline from early (green) and late (red) sessions (defined automatically by intensity values above background, Methods) are shown in each panel, and overlaid in the middle panel for comparison. Right, 3d surface plots of the spine showing clear increase in intensity across the spine head. (e) New spines form nearby enlarged MRSs. Left, MRSs near sites of new spine formation show a larger mean spine volume change (left; r = −0.09, p = 0.04, Pearson’s correlation coefficient) and are more likely to show enlargement (volume >1.1× early session volume, right) than MRSs in sites further from new spines (p = 5e-5; Pearson’s correlation coefficient of enlargement probability vs. distance). n = 118 new spines; 697 MRSs present on 50 new spine-containing dendrites. Right, in the no-task control condition, neither the fold-change in MRS volume (Pearson’s correlation coefficient, r = −0.015, p = 0.79) nor the probability of MRS enlargement (Pearson’s correlation coefficient, r = 0.02, p = 0.68) change as a function of distance to new spines. n = 51 new spines; 339 MRSs on new spine-containing dendrites. Mean ± SEM. All statistical tests two-sided. (f) Pre-existing spines that show enlargement over learning tend to house spine apparati. Left-most panels: early and late in vivo images showing dendrite used in CLEM. Second-to-left panel, EM reconstruction of the dendrite shown at left, with Greek letters indicating fiducial spines for alignment. Two outlined regions indicate spines under examination. Zoomed-in reconstruction images for both of these spines are provided at right, and include reconstructed spine apparati (yellow) contained within, as well as a schematized sectioning plane from which EM micrographs are drawn. Beside each zoomed-in window is a corresponding EM micrograph showing the spines of interest (purple outline) and highlighted spine apparati (yellow outline). Right-most panels show early and late in vivo images showing spine enlargement, with early (red) and late (green) spine outlines provided for comparison. Overlaid outlines for the spines are shown to the right for clarity. (g) Pie charts showing the fraction of spines housing a spine apparatus. The spines that enlarged during learning in vivo (left) are more likely to house a spine apparatus than non-enlarged spines (right) (Pearson’s chi-squared test of independence; p = 2e-7). (h) Pie chart showing the fraction of new spines that house a spine apparatus. This fraction (0.17) is significantly smaller (p = 5e-8; Pearson’s chi-squared test of independence) than the overall fraction of pre-existing spines from (g) (70/123 = 0.57).

Figure 4:. Functional clusters of new spines and movement-related spines preferentially encode learned movements

(a) Top, images of a new spine from peri-movement periods, averaged across movements. Vertical dashed line indicates movement onset. Bottom, average movement onset-aligned activity of the new spine shown above. Mean ± SEM. (b) Left, movement onset-aligned, z-scored activity of all imaged new spines, sorted by timing of peak activity. Right, histogram of onset timing of new spine activity with respect to movement onset. Gray shaded region indicates time periods after movement onset (t = 0). Median onset timing (-232 ms) is indicated with a vertical red dashed line. n = 118 new spines. (c) New spines show higher co-activity rates with MRSs than with non-MRSs (2-way ANOVA [MRS label × distance], main MRS label effect: F = 31.12, d.f. = 1, p = 3e-8). All pairs showed co-activity that depended on inter-spine distance (main distance effect: F = 20.4, d.f. = 5, p = 5e-20), and each group shows a significant decrease over distance, wherein the 0–5μm bin is significantly higher than other distance bins (post-hoc using LSD: new spine-MRS: p <0.001 vs. all other bins; new spine-nonMRS: p <0.005 vs all other bins). Further, co-activity rates are significantly negatively correlated with distance (Spearman’s rank coefficient, new spine-MRS vs distance, r = −0.16, p = 1e-7; new spine-nonMRS: r = −0.20, p = 5e-14). n = 1658 new spine-MRS pairs; 2207 new spine-nonMRS pairs. Mean ± SEM. (d) Example traces illustrating a new spine only activity event, a MRS only activity event, and a co-activity event. Shown are both the ΔF/F₀ trace (upper) as well as binarized “event” traces (lower) indicating active and inactive periods for each spine. Colored regions on the ΔF/F₀ trace also indicate active periods for each spine. Gray blocks demarcate the periods defined by each of the three activity types. (e) New spine-MRS co-activity occurs preferentially during movements. The fraction of new spine-MRS co-activity events occurring during movements is higher than the fraction of new spine-only (p = 4e-7) and MRS-only events (p = 3e-4; two-sided sign-rank test correcting for multiple comparisons using the FDR method). n = 34 fields. Median ± bootstrapped 95% confidence intervals. (f) Example lever traces of individual movements that coincide with new spine-MRS co-activity (left; blue lines) compared to movements that lack new spine-MRS co-activity (right; red lines). The “learned movement pattern” (the average of all rewarded movements over the late (11–14) learning sessions) is overlaid over each group (black line). (g) Movements with new spine-MRS co-activity (light blue) are more similar to the learned movement pattern than movements lacking such activity (red) (p = 0.003; two-sided igned-rank test). Each data point corresponds to the median correlation value of all new spine-MRS pairs within a given field (n = 34 fields). The medians of these values indicated as horizontal lines of the corresponding color.

Figure 5:. Correlated light- and electron microscopy (CLEM) reveals patterns of microstructures surrounding functional clusters

(a) Example CLEM images showing filopodial clustering around new spines. Left, early (session 2) and late (session 12) in vivo images (average projections of time series) showing new spine formation (blue arrowheads) over learning. Middle, magnified late-session in vivo image with labeled fiducial structures (Greek letters) for alignment with EM image. Right, EM reconstruction of the same dendrite, with new spines (cyan structures, blue arrowheads) and fiduciary structures (Greek letters) labeled. Filopodia (red structures) not visible in vivo are revealed by EM to cluster around new spines. (b) Example filopodium (red) identified in EM showing lack of synaptic contact (absence of postsynaptic density and aggregated presynaptic vesicles) over the full length of the structure. Inset in the upper right shows the fully reconstructed filopodium (arrow). (c) Left, histogram of the distances between filopodia and their closest new spine demonstrating that filopodia (red) cluster around new spines. True distances were compared to chance, calculated by shuffling the filopodia locations 10,000 times (gray). Inset, bar graph summary of data. Median ± bootstrapped 95% confidence intervals. (n = 36 filopodia/ 6 dendrites/ 5 fields/ 4 animals; p = 0.0003). P-value calculated as the fraction of shuffles following the null hypothesis by a one-sided comparison of whether the median of each shuffle was ≥ the median of the real data. *** p < 0.001. Right, cumulative probability distributions of the distance between filopodia and their nearest new spine compared to shuffled data. Distributions of shuffles separated by individual dendrites shown in gray shades, and the overall shuffled distribution in black. (d) CLEM identification of new spine synapsing with MSBs. Left, example in vivo images of a dendrite showing new spine formation by the final session. Background manually removed for clarity. Middle, EM reconstruction of the region surrounding the new spine, with new spine highlighted (cyan), along with zoomed-in image of the reconstructed region (dotted-line box shows region of interest; other spines removed in zoomed-in image for clarity), with the axon associated with the new spine and the allodendritic spine added. Right, EM image showing the new spine synapsing with a multi-synaptic bouton (MSB), with both synapses on the bouton indicated with red arrows. (e) Summary of the prevalence of MSBs on different spine types. Left, new spines (light blue) show a higher fraction of MSBs than pre-existing spines (red) (chi-square test, p = 1e-7). Right, MSBs are more common on recently formed new spines (those that formed between the middle and late sessions; magenta) than older new spines (those that formed between the early and middle sessions; dark blue) (Pearson’s chi-squared test of independence, p = 0.04). (f) Left, example images (four sections) of the group of a new spine, a MSB, and an allodendritic spine from d, showing the presence of a spine apparatus (yellow outlined region) in the allodendritic spine. Inset above shows the volume reconstruction, with a partially transparent allodendritic spine showcasing the internal spine apparatus. Right, new spines connected to a MSB are less likely to house a spine apparatus than the paired allodendritic spines, suggesting that the allodendritic spines are more mature than the new spines (p = 0.005; Pearson’s chi-squared test of independence). (g) Example in vivo images over early, middle, and late sessions of learning showing “transient” new spine formation; i.e. the new spine appears in the middle session and is eliminated by the late session (open arrowheads indicate pre-formation and post-elimination; filled arrowhead indicates new spine). (h) New spine survival is related to the local co-activity structure with task-related spines. Both sustained and transient new spines show significantly higher co-activity rates with MRSs than non-MRSs on middle learning sessions (3-way ANOVA [MRS label × distance × survival]: main MRS label main effect: F = 30.86, d.f. = 1, p = 3e-8; survival of new spine main effect: F = 7.75, d.f. = 1, p = 0.005; sustained new spine – MRS vs. sustained new spine – nonMRS co-activity: p = 2e-8; transient new spine –MRS vs. transient new spine – nonMRS co-activity: p = 0.004, post-hoc test using the least-significant difference). Sustained new spine-MRS co-activity rates are significantly higher than both transient new spine groups (vs. transient new spine-MRS pairs: p = 0.009; vs. transient new spine-nonMRS: p = 8e-9). A main effect of distance was observed (main distance effect: F = 5.46, d.f. = 5, p = 1e-4), and the MRS label × distance × survival interaction was significant (F = 2.38, d.f. = 5, p = 0.04), suggesting that the distance-dependent relationship of MRS status with co-activity depends on the survival of the new spine, and justifying the statistical separation of the data at these three levels. Individual inspection of the data grouped in this way (i.e. one group corresponding to each line) reveals that only sustained new spines showed a significant negative correlation of co-activity rates with distance (sustained new spine-MRS: r = −0.16, p = 8e-5; sustained new spine-nonMRS: r = −0.14, p = 5e-4; transient new spine-MRS: r = 0.02, p = 0.74; transient new spine-nonMRS: r = −0.08, p = 0.17). n = 28 transient new spines, 418 transient new spine-MRS pairs, 481 transient new spine-nonMRS pairs; 57 sustained middle-session new spines, 765 sustained new spine-MRS pairs, 1024 sustained new spine-nonMRS pairs. Mean ± SEM.

Figure 6:. Most new spines represent novel connections on the dendritic segment

(a) Example of axon-sharing between pre-existing spines. Left-most panel, in vivo image (average projection of time series) of a dendritic segment found to have a pair of pre-existing spines that share an axon (red circles). Fiducial structures for EM alignment are denoted by Greek letters. Second-to-left panel, corresponding EM reconstructions of the same dendritic segment. Fiducial structures marked with Greek letters. Reconstruction on the right shows a zoom-in portion of the dendrite (demarcated by dashed box at left), with reconstruction of axon (yellow) shared by the two pre-existing spines (red circles) added. A zoomed-in image (demarcated by grey dashed-line boxes) are provided at right. Zoomed-in versions are rotated for clarity. Within each zoomed-in image, the approximate EM slicing plane (red planes) are indicated. Right, EM micrographs of each spine showing the presence of a synapse (red arrows) with the axon of interest. (b,c) Example reconstructions of new spines and their associated axons demonstrating the lack of axon-sharing by new spines. Left, in vivo images showing new spine formation between early and late sessions (empty arrowheads for pre-formation new spine location, filled arrowheads for new spines). Middle, as in a, EM reconstructions of the corresponding dendritic regions and EM images illustrating synaptic connections with the axon of interest. Right, zoomed-out reconstructions of the dendritic and axonal segments. The approximate EM sectioning plane (red) is indicated. The axon in b runs parallel to the dendrite, and does not contact nearby spines on the targeted dendrite (zoomed-in window provided for clarity). The axon in c makes contact with another spine, but does not form a synapse, and the remainder of the axon runs nearly orthogonal to the imaged dendrite and does not ramify within the captured volume. (d) Most new spines do not share axons with nearby spines within the EM volume (Pearson’s chi-squared test of independence; p = 4e-6, sharing vs. non-sharing). (e,f) Most new spine-MRS pairs do not show co-activity rates consistent with axon sharing. e) Left, pixel-wise correlation map of the dendritic segment shown in a, relative to the central pixel in “pre-existing spine 1” (top circle). High correlation values outside the basis spine (pre-existing spine 1) are only present in “pre-existing spine 2”, the spine found to share an axon with pre-existing spine 1, as in a. Middle, example activity event traces illustrating the highly coherent activity of the confirmed axon-sharing pair of spines shown at left. Right, example activity event traces of a new spine-MRS pair from the same dendrite. f) Histograms of normalized co-activity rates for all new spine-MRS pairs (green; n = 1658 pairs) and all pre-existing spine pairs that are confirmed in EM to share axons (blue; n = 3 pairs). New spine-MRS pairs show a lower median co-activity rate (0.12) vs confirmed axon-sharing pairs (0.32; p = 0.009, two-sided rank-sum test).

Figure 7:. Proposed model of locally facilitated spinogenesis during motor learning

(a) During early stages of learning, task-related inputs (presynaptic axons 1–3; task-related activity indicated by green haze) to pre-existing spines (MRSs) on M1 L2/3 apical dendrites support task performance. (b) A subset of these task-related synapses undergo synaptic potentiation (red fill in indicated spine), likely leading to local facilitation in the dendrite. The locally facilitated region becomes a hot-spot for nascent filopodia (dashed outlines), which sample the surrounding neuropil for pre-existing axonal boutons. (c) Some of the filopodia make contact with a pre-existing bouton, forming new dendritic spines (blue), and resulting in MSBs shared with allodendritic spines (spines from dendrites 2 and 3). Thereafter, new spines that show synchronized activity (green haze within spines) with surrounding task-related spines are selectively maintained, while those with less synchronous activity (red haze) are eliminated. This process selects new spines that contribute to functional clustering of task-related information. (d) The surviving new spines participate in a new or modified synaptic cluster wherein correlated, task-related inputs to new spines and nearby spines (green haze) robustly drive the postsynaptic activity and the execution of the learned movement. The elimination of new spines receiving locally asynchronous activity (c) prevents the inclusion of functionally different inputs into the dendritic repertoire. Dendrite 2 was excluded from the drawing to reflect the unknown fate of multi-synaptic boutons (i.e. whether they split into distinct boutons or the other spine retracts).