Refinement of efficient encodings of movement in the dorsolateral striatum throughout learning

Abstract

The dorsal striatum plays a critical role in action selection, movement, and sensorimotor learning. While action-specific striatal ensembles have been described, the mechanisms underlying their formation and evolution during motor learning remain poorly understood. Here, we employed longitudinal two-photon Ca²⁺ imaging of dorsal striatal neurons in head-fixed mice as they learned to self-initiate locomotion. We found that both direct- and indirect-pathway spiny projection neurons (dSPNs and iSPNs, respectively) exhibited robust activation during early locomotor bouts, with activity gradually diminishing across sessions. For dSPNs, action onset and offset ensembles progressively emerged from an initially broad population of nonspecific neurons. In contrast, iSPN ensembles originated from neurons responsive to opposing actions before refining into onset- or offset-specific populations. These findings demonstrate that striatal ensemble activity becomes more selective over time, with a reduction in nonspecific neuronal activation and an increase in the efficiency of striatal encoding for learned motor actions.

Keywords: CP: Neuroscience; DLS; direct and indirect pathways; longitudinal 2P imaging; motor learning; striatal ensembles.

Figure 1.. Striatal activation decreases throughout learning

(A) AAV delivery of GCaMP6s into the DLS of D1-Cre or A2a-Cre mice followed by gradient refractive index (GRIN) lens implantation. (B) Two-photon (2P) imaging during head-restrained, self-generated locomotion. Bottom: DLS longitudinal imaging; scale bar: 100 μm. (C) Average velocity (n = 14 mice; p = 0.0068). (D) Average time spent running (n = 14 mice; p < 0.0001). (E and F) Percentage of imaged neurons positively modulated during locomotion for dSPNs (E; n = 6 mice; p = 0.440) and iSPNs (F; n = 6 mice; p = 0.130). (G and H) SPN event rate as a function of locomotion speed for dSPNs (G; n = 6 mice; p = 0.979) and iSPNs (H; n = 6 mice; p = 0.003). (I and J) Left: dSPN summary. Right: iSPN summary. Top: average speed. Middle: average standardized dF/F of SPNs. Bottom: average Ca²⁺ event rate (normalized). All analyses are centered around motion onsets and offsets. (K–N) Across-day percentages of dSPNs active during onsets (K, n = 6 mice; p = 0.004), offsets (L, n = 6 mice; p = 0.038), iSPNs active during onsets (M, n = 6 mice; p = 0.008), and offsets (N, n = 6 mice; p = 0.008). Statistical significance was assessed by repeated measures (RM) one-way ANOVA with multiple comparisons (C–F and K–N) and two-way RM ANOVA with multiple comparisons (G and H).

Figure 2.. Specific decrease in the proportion of movement-nonspecific SPNs with learning

(A and B) Average standardized dF/F of dSPNs (A) and iSPNs (B) SPNs are categorized as active only during motion onsets (onset only, green), offsets (offset only, red), active during both onsets and offsets (both, yellow), or not active during either (neither, blue). (C and D) Representative images of SPN activation types during sessions 1 (left) and 8 (right). Scale bars: 100 μm. (E) Percentage of dSPNs across days categorized as onset only (green; n = 6 mice; p = 0.189), offset only (red; n = 6 mice; p = 0.814), both (yellow; n = 6 mice; p = 0.027), and neither (blue; n = 6 mice; p = 0.018). (F) Percentage of iSPNs across days categorized as onset only (green; n = 6 mice; p = 0.525), offset only (red; n = 6 mice; p = 0.660), both (yellow; n = 6; p = 0.029), and neither (blue; n = 6 mice; p = 0.003). (G) Average dSPN activation types for early vs. late periods (onset only p = 0.102; offset only p = 0.710; both p = 0.018; neither p = 0.084). (H) Average iSPN activation types for early vs. late periods (onset only p = 0.176; offset only p = 0.468; both p = 0.064; neither p = 0.063). Statistical significance was assessed by RM one-way ANOVA with multiple comparisons (E and F) and two-sided paired t tests (G and H).

Figure 3.. Refinement of movement-specific SPN ensembles throughout learning

(A and B) Percentage of late-stage (days 7 + 8) onset (A) and offset (B) dSPN ensembles that were significantly active during early stages (days 1 + 2). Left: observed activation overlap (obs), right: average overlap of shuffled activation vectors (see STAR Methods) (n = 6 animals; dSPN onset: p = 0.017; dSPN offset: p = 0.003). (C and D) Percentage of late-stage (days 7 + 8) onset (C) and offset (D) iSPN ensembles that were significantly active during early stages (days 1 + 2). Left: observed activation overlap (obs), right: average overlap of shuffled activation vectors (see STAR Methods) (n = 6 animals; iSPN onset: p = 0.047; iSPN offset: p = 0.582). (E and F) Early vs. late activation bias averaged for each animal (across all dSPNs) for onset-specific (E; n = 6 animals; p = 0.004) and offset-specific (F; n = 6 animals; p = 0.010) dSPN ensembles. (G and H) Early vs. late activation bias averaged for each animal (across all iSPNs) for onset-specific (G; n = 6 animals; p = 0.019) and offset-specific (H; n = 6 animals; p = 0.011) iSPN ensembles. Colors depict action bias (more positive values → onset-biased → green, negative values → offset-biased → red, closer to zero → more nonspecific → yellow, E–H). (I) Within-day similarity indices for onset-specific dSPN ensembles (green; n = 6 mice/8 days/459 onsets; p = 0.045) and offset-specific dSPN ensembles (red; n = 6 mice/8 days/480 offsets; p = 8.837e− 04). (J) Across-day similarity indices for onset-specific dSPN ensembles (green; 6 mice/8 days/209 onsets; p = 0.045) and offset-specific dSPN ensembles (red; n = 6 mice/8 days/209 offsets; p = 0.013). (K) Within-day similarity indices for onset-specific iSPN ensembles (green; n = 6 mice/8 days/546 onsets; p = 0.637) and offset-specific iSPN ensembles (red; n = 6 mice/8 days/556 offsets; p = 0.062). (L) Across-day similarity indices for onset-specific iSPN ensembles (green; 6 mice/8 days/210 onsets; p = 0.159) and offset-specific iSPN ensembles (red; n = 6 mice/8 days/194 offsets; p = 0.042). (I–L) Right: boxplots depicting summary of early (days 1 + 2) vs. late (days 7 + 8) average (per animal) similarity indices (I, p = 0.016; J, p = 0.078; K, p = 0.016; and L, p = 0.047). (M) Comparison of early similarity indices between SPN types (pooling “within” and “across” days, onset + offset ensembles from I–L) (dSPN early n = 24, iSPN early n = 24; p < 0.0001). (N) Representative principal-component analysis (PCA) + t-distributed stochastic neighbor embedding (tSNE) dimensionality reduction of an animal’s dSPN population activity across onsets (green) and offsets (red), with early (days 1 + 2) bouts in lighter colors and later (days 7 + 8) bouts in darker colors. (O) Separability score (onsets vs. offsets) of all D1-Cre animals comparing the early- to late-stage dSPN population separability of actions (n = 6 mice; p = 0.016). (P) Representative PCA + tSNE dimensionality reduction of an animal’s iSPN population activity across onsets (green) and offsets (red), with early (days 1 + 2) bouts in lighter colors and later (days 7 + 8) bouts in darker colors. (Q) Separability score (onsets vs. offsets) of all A2a-Cre animals comparing the early- to late-stage iSPN population separability of actions (n = 6 mice; p = 0.886). Statistical significance was assessed by two-sided paired t tests (A–H, I–L, right, O, and Q), linear mixed-effects modeling (I–L, left), and two-sided unpaired t tests (M).

Figure 4.. Striatal movement-specific activity becomes more efficient throughout learning

(A and B) Average performance of action-specific linear regression models fitting dSPN activity to velocity, trained on subsets of onsets (A; n = 6 mice; p = 0.486) or offsets (B; n = 6 mice; p = 0.087). (C and D) Average performance of action-specific linear regression models fitting iSPN activity to velocity, trained on subsets of onsets (C; n = 6 mice; p = 0.503) or offsets (D; n = 6 mice; p = 0.814). (E and F) Model performance of action-specific (onset-specific, E; offset-specific, F) linear regression models trained on subsets of dSPNs of varying sizes (see STAR Methods; onsets: n = 6 mice, p = 0.579; offsets: p = 0.059). (G and H) Model performance of action-specific (onset-specific, G; offset-specific, H) linear regression models trained on subsets of iSPNs of varying sizes (see STAR Methods; onsets: n = 6 mice, p = 0.253; offsets: p = 0.059). (I and J) Average efficiency score (see STAR Methods) plotted for dSPN encodings of onsets (I; n = 6 mice; p = 0.015) or offsets (J; n = 6 mice; p = 0.040). (K and L) Average efficiency score (see STAR Methods) plotted for iSPN encodings of onsets (K; n = 6 mice; p = 0.076) or offsets (L; n = 6 mice; p = 0.358). (I–L) Right: comparison of early- (days 1 + 2) vs. late- (days 7 + 8) stage efficiency scores (I, p = 0.006; J, p = 0.034; K, p = 0.042; and L, p = 0.045). Statistical significance was assessed by RM one-way ANOVA with multiple comparisons (A–D), Friedman test (I–L, left), two-way RM ANOVA with multiple comparisons (E–H), and two-sided paired t tests (I–L, right).