Visual experience drives functional reorganization of cortical GABAergic circuits

Summary:

Sensory deprivation leads to extensive cortical plasticity, but the impact of enhanced sensory experience on the mature cortex remains poorly understood. Here, we examine how visually evoked activity in cortical circuits is shaped by repeated exposure to varied stimuli. The most prominent pattern of visually evoked activity in mouse primary visual cortex, beta oscillations (15–30Hz), arises from brief events of neural synchrony with a characteristic pattern of laminar propagation and relies on the activity of somatostatin-expressing (SST) GABAergic interneurons. We find that visually evoked beta activity is initially weak but robustly potentiated by repeated exposure to a diverse visual stimulus set, leading to enhanced recruitment of cortical neurons by these rhythmic network events. Cellular imaging further reveals that visual experience leads to increased visual responses in SST interneurons and suppressed responses in vasoactive intestinal peptide-expressing (VIP) GABAergic interneurons. In association with this rebalancing of inhibitory circuits, visual experience enhances visual selectivity in nearby pyramidal neurons. Visual experience thus selectively reorganizes adult dendrite-targeting inhibitory circuits, promoting network synchrony and enhancing sensory encoding by cortical excitatory projection neurons.

Introduction

Neocortical sensory circuits are crucial for detecting perceptual cues and statistical regularities in the surrounding environment. These circuits exhibit robust experience-driven plasticity that shapes their selectivity at all developmental stages. Visual deprivation during a juvenile critical period induces robust plasticity in the primary visual cortex (V1) that relies on GABAergic inhibition^1–14. In contrast, loss of sensory input in adulthood leads to cortical plasticity^15,16 that requires NMDA receptors (NMDARs) but may not rely on GABAergic circuits^15–17. Enhanced overall sensory experience may also contribute to plasticity in the adult cortex. Indeed, juvenile-like plasticity can be induced in the adult V1 cortex ^18,19 and mice raised in an enriched environment exhibit lifelong plasticity that is lost upon transfer to a standard environment^20–22. However, how activity in the adult cortex is shaped by visual experience as opposed to deprivation remains poorly understood.

V1 in rodents and other species exhibits distinct learning-related plasticity that can lead to either increased or decreased response magnitudes in excitatory and inhibitory neurons depending on context and reward contingencies^23–36. Repeated presentation of a single stimulus can evoke a stimulus-specific, NMDAR-dependent potentiation of visually evoked potential (VEP) responses in adult V1^37–39. This response potentiation relies on GABAergic inhibition and is associated with enhanced thalamocortical input to layer 4 and increased sensory-evoked firing^17,37–44. In contrast, other work has found that repeated presentation of single stimuli produces robust, NMDAR-dependent loss of response magnitude in excitatory and inhibitory neurons in V1 without changes in thalamocortical inputs^23,45. The precise relationship between visual experience and adult cortical plasticity thus remains unclear.

In many species, visual stimuli evoke robust rhythmic activity in V1, organizing spiking and enhancing encoding of relevant sensory information^46–58. Rhythmic synchronization typically arises across distant retinotopic cortical sites when they are stimulated by the same visual patterns^{47–49,51,57}, suggesting that the circuits underlying patterned activity are sensitive to the statistics of the sensory environment. Recent work using natural images found that visually evoked rhythms are linked to the predictability of visual features at one retinotopic site compared to the rest of the image⁵⁹. Rhythmic synchrony in V1 may be strengthened when stimuli match an internal representation of expected visual structure that is shaped by sensory experience^60–64. In mouse visual cortex, the most prominent visually evoked rhythm, beta activity (15–30Hz), relies on dendrite-targeting GABAergic interneurons that co-express the peptide somatostatin (SST-INs)^47,48 and is regulated by interneurons that co-express vasoactive intestinal peptide (VIP-INs)⁴⁹, suggesting that these inhibitory populations may potentially contribute to experience-dependent reorganization of mature cortical circuits.

Here, we examined the sensitivity of cortical circuits to repeated visual experience of single stimuli or varied stimulus sets by recording V1 local field potentials (LFP) and spiking activity longitudinally over several days. We find that visual experience causes a robust, selective potentiation of visually evoked cortical beta activity, characterized by increased amplitude of individual underlying network events synchronizing activity across cortical layers. Visual experience strengthens the entrainment of cortical neurons by beta events. This form of plasticity is induced by both drifting gratings and natural movies but requires exposure to a rich set of visual stimuli. Using longitudinal imaging of V1 neurons, we find that enriched visual experience transforms GABAergic activity by altering the balance of visually evoked activity in SST- and VIP-INs. In association, visual experience increases the visual selectivity of local pyramidal neurons (PNs). Together, these data suggest that varied visual experience reorganizes dendrite-targeting interneuron circuits in the adult visual cortex, enhancing network synchrony and visual encoding by excitatory neurons.

Results

Visual experience causes plasticity of evoked network events in V1

Previous work has identified both increased^17,37–44 and decreased^23,24,45 cortical sensory responses following repeated sensory experience, highlighting a potentially complex relationship between sensory input and plasticity in mature cortical circuits. To gain insight into how sensory experience modulates cortical activity, we first implanted visually naive adult mice with 16-channel silicon probes in V1 and performed longitudinal recordings of the LFP across cortical layers (Figure 1A, S1A). Implanted mice underwent a 7-day passive visual experience protocol (Fig. 1B) during which cortical activity was recorded throughout presentation of an extended stimulus set comprised of 64 drifting gratings varying in size, spatial frequency, and contrast (Figure 1A–B). Each session included 960 total stimulus presentations and 80 min of total visual exposure time.

Figure 1. Visual experience potentiates visually evoked beta activity.

(A) Experimental paradigm: LFP is recorded across V1 layers with chronically implanted electrode arrays in mice that are head-fixed on a running wheel while drifting gratings are presented on a screen. (B) Passive visual training schedule: over 7 days, visually naïve mice are exposed to gratings at 4 sizes (10, 20, 40, and 80 °), 4 contrasts (2, 8, 32, and 100%) and 4 spatial frequencies (0.01, 0.04, 0.16, and 0.64 cycles/°) resulting in a set of 64 stimuli (15 set repetitions; 960 trials per session). (C) Example raw and band-pass filtered (15–30Hz) LFP traces during a large, high-contrast stimulus (80°, 100%, 0.04 cycles/°) in a visually naive mouse (day 1; upper) and an experienced mouse (day 7; lower). (D) Power spectra of z-scored LFP evoked by a large, high-contrast grating stimulus (80°, 100%, 0.04 cycles/°) in visually naïve (yellow) and experienced (cyan) mice. Experience results in increased power in the beta band (15–30Hz) but not in the gamma band (30–80Hz; n = 7 mice). (E) Experience modulation index for visually evoked power in the beta and gamma bands across all 64 stimulus types (n = 7 mice). (F) Experience-induced change in visually evoked power (Day7 – Day1) for 100% contrast stimuli (16 stimulus types; n = 7 mice). (G) 7-day trajectory of baseline normalized beta power for 100% contrast stimuli (n = 7 mice). (H) Baseline normalized beta power for visually naïve mice (upper) and experienced mice (lower) across all 64 stimulus types (n = 7 mice). Data are presented as mean ± SEM; *: P<0.05, **: P<0.01,***: P<0.001. See Supplementary Table 1 for detailed statistical analysis. See also Figure S1.

Robust activity in the beta range (15–30Hz) is a key hallmark of visually evoked LFP activity in mouse V1^47–50 and may contribute to long-range synchronization of distant retinotopic areas^49,64. We observed modest evoked beta activity in response to highly salient stimuli in visually naïve mice. However, visual experience resulted in markedly enhanced visually evoked beta power (Figure 1C, 1D, S1A). To quantify the potentiation of activity across stimulus types, we developed an Experience Modulation Index (EMI, STAR Methods). EMI was positive across stimuli for power in the beta band but not for the adjacent gamma band (30–80Hz; Figure 1E, S1F–S1H), suggesting that visual experience selectively enhances beta activity. Potentiation of visually evoked beta activity was significant during periods of both quiescence and locomotion (Figure S1B), suggesting reorganization of underlying circuit interactions rather than enhancement of behavioral state-dependent dynamics. Power in the beta band increased across visual experience days (Figure 1F–1G), exhibiting an immediate increase and plateauing after the second day for stimuli that evoked strong initial beta responses (Figure S1G). In contrast, potentiation was more gradual for stimuli that initially evoked weak beta responses.

To further understand the impact of sensory experience on visually evoked activity in V1, we compared overall evoked LFP power, activity in the beta and gamma bands, and visual evoked potential (VEP) amplitudes for each stimulus type across the visual experience paradigm (Figure S1C–S1E). VEPs in V1 layer 4 (L4) arise largely from thalamocortical synaptic input³⁸ and increase in magnitude following repeated visual experience with a single stimulus^{17,37–43,65,66}. In good agreement with previous studies, we found that the L4 VEPs strengthened over the course of the passive experience protocol (Figure S1C, S1D). In accordance with previous work^47–49, we found that visually evoked beta activity (15–30Hz) in mouse V1 is maximal for stimuli with a spatial frequency of 0.04 cycles/° and increases with stimulus size and contrast (Figure 1G–1H). VEP amplitude was prominent at 0.04 and 0.16 cycles/° spatial frequencies, increased with size and contrast, and also exhibited potentiation across a wide range of stimuli. (Figure S1E). In contrast, we observed no change in visually evoked power in the gamma band for any stimulus (Figure S1F–S1H). Overall, these results suggest that visual experience enhances the network interactions underlying cortical beta activity.

Visual experience selectively potentiates the amplitude of beta events

We recently found that visually evoked beta activity emerges from specific cortical network events occurring rhythmically during visual stimulation⁶⁷. To examine how these events are regulated by visual experience, we employed a recently developed method, CBASS, for reliably detecting network events with high precision in both the frequency and temporal domains⁶⁷ (STAR Methods). CBASS examines the LFP within a specific frequency band (i.e., beta) and seeks a specific alignment of phase and amplitude across channels that is predictive of a particular state (i.e., visual stimulation). Consistent with our previous findings, the rate of beta events detected by CBASS increased during visual stimulation (Figure 2A, Figure S2) and power in the beta range increased during epochs with high beta event rate (Figure S2A). The average field around beta events had energy in the beta range (Figure S2D) and current source density (CSD) analysis revealed that they were associated with a propagation of activity from layer 4 to layer 2–3 followed by an activation of layers 5 and 6 (Figure 3B).

Figure 2. Enhanced beta activity reorganizes visually evoked spiking.

(A) LFP across channels (upper) and its short time Fourier transform (lower) around presentation of a large, high-contrast grating stimulus (80°, 100%, 0.04 cycles/°). CBASS ties visually evoked beta activity to specific network events (blue lines). Note that these events can also happen outside visual stimulation. (B) Averaged CSD around beta events in visually naïve mice (left) and experienced mice (right; n = 7 mice). (C) 7-day trajectory of beta event power (upper) and rate (lower) for 100% contrast stimuli at spatial frequencies of 0.01, 0.04, 0.16, and 0.64 cycles per degree (n = 7 mice). (D) MUA around beta events (blue) or a matched number of random events (gray) in visually naive (upper) and experienced mice (lower; n = 7 mice)(E) Pairwise Phase Consistency (PPC) spectra quantifying MUA-LFP phase-locking during beta events in visually naive (black) and experienced mice (blue; n = 7 mice). (F) 7-day trajectory of MUA-LFP PPC in the beta range (15–30 Hz) during beta events across days of visual exposure (n = 7 mice). Data are presented as mean ± SEM; *: P<0.05, **: P<0.01. See Supplementary Table 1 for detailed statistical analysis. See also Figure S2.

Figure 3. Experience-dependent response potentiation requires diverse stimuli.

(A) Baseline normalized beta band power evoked by 100% contrast stimuli in visually naïve mice (day 1; upper) and experienced mice (day 7; bottom) using 3 different exposure protocols having a matched number of trials (960 trials) between day 2 and 6: 1 Stim.: repetition of one stimulus (80°, 100%, 0.04 cycles/°; left), 16 Stim.: repetition of all full contrast stimuli (16 stimuli; middle) and 64 Stim.: repetition of the full 64 stimulus sets (right; n = 7 mice per protocol). The full stimulus set was displayed on day 1 and day 7 in the 1 Stimulus protocol. (B) Experience modulation index quantifying changes in baseline normalized beta power between day 1 and day 7 across all 100% contrast gratings for each protocol (n = 7 mice per protocol). (C) Same as (B) for beta event power. (D) Same as (B) for beta event rate. (E) Same as (B) for LFP-MUA PPC during beta events. Data are presented as mean ± SEM; *: P<0.05, **: P<0.01, ***: P<0.001. See Supplementary Table 1 for detailed statistical analysis. See also Figure S3

Visual experience resulted in a strong potentiation of beta event power (i.e. amplitude squared) across stimulus types (Figure 2B–2C) but a slight reduction in beta event rate (Figure 2C). Beta event entrainment of spiking, as measured by multiunit activity (MUA) and spike-LFP phase locking to beta frequencies, increased following visual experience (Figure 2D–2F, Methods). These effects were prominent during visual stimulation (Figure 2B–2F) and observed during quiescence and locomotion but not in the absence of a visual stimulus (Figure S2E–S2H). Activity in the nearby gamma range can likewise be linked to specific network events detected by CBASS⁶⁷. Gamma event rate increased during locomotion (Figure S2I–S2J) and LFP power increased in the gamma range during epochs with high gamma event rates (Figure S2I). Gamma events have faster dynamics, lower amplitude, and weaker activation of layer 6 than beta events (Figure S2K). In contrast to beta activity, neither the amplitude nor the rate of gamma events exhibited plasticity in response to the visual experience paradigm (Figure S2L). Similarly, visual experience did not alter entrainment of spiking by gamma events (Figure S2M–S2N). Visual experience thus selectively enhances the magnitude of individual visually evoked beta events and the degree to which they organize cortical spiking.

Experience-dependent plasticity requires exposure to diverse visual stimuli

To examine the conditions necessary to induce experience-dependent potentiation of visual responses, we exposed a cohort of visually naïve mice to repetition of a single highly salient grating stimulus (Figure S3A). The full stimulus set was presented on the first and last days of the protocol and only a single stimulus was presented on the intervening days. A second cohort of mice were exposed to a set of 16 full-contrast gratings (Figures S3B). Visually evoked beta activity in the 1-stimulus and 16-stimulus cohorts was then compared to that observed in the initial 64-stimulus cohort (Figure 3A). Neither the responses to the repeated single stimulus nor the full stimulus set on the first and last day exhibited plasticity of beta activity in the 1-stimulus cohort (Figure 3A, Figure S3A). In turn, we observed an intermediate degree of potentiation in response to the 16-stimulus protocol (Figure S3A, S3B). Potentiation of beta activity was strongest after exposure to the full 64-stimulus protocol (Figure 3B). Likewise, enhancement of beta event power, reduction in beta event rate, and MUA entrainment to beta events were robust following the 64-stimulus protocol, intermediate following the 16-stimulus protocol, and absent following the 1-stimulus protocol (Figure 3C–3E). Visual experience-dependent potentiation of evoked beta activity thus requires more than one repeated stimulus and increases with stimulus diversity.

Drifting grating stimuli evoke strong neural responses in V1 across species, but represent highly artificial stimulus conditions. We therefore recorded activity in a cohort of visually naïve mice repeatedly exposed to four natural movie clips at four sizes (Figure 4A–4C; see Methods). On the first and last days, mice were also presented with 16 full-contrast gratings. We found that beta activity exhibited plasticity in response to all natural movie clips, and this enhancement was most robust for the largest size stimuli (Figure 4D, 4E, S4A). In contrast, gamma activity in response to the movie clips remained unchanged (Figure S4B). Beta activity evoked by drifting gratings on the first and last day was also significantly potentiated across all stimuli (Figure 4E). Likewise, the power and entrainment of MUA by beta events increased and their rate decreased after visual exposure in response to both movie clips and grating stimuli (Figure 4F–4H). Experience-dependent potentiation of beta activity is thus generic to a broad range of visual statistics and generalizes to stimuli not shown in the repeated set.

Figure 4. Visually evoked beta activity is potentiated by natural images.

(A) Experimental paradigm: LFP is recorded across V1 layers with chronically implanted electrode arrays in head-fixed mice on a running wheel while natural movie clips are presented on a screen. (B) Passive visual training schedule: visually naïve mice are exposed to 4 movie 2 s clips at 4 sizes (10, 20 40 and 80 °) resulting in a set of 16 stimuli (days 1 and 8: 30 set repetitions, 480 trials; days 2 to 7: 60 set repetitions, 960 trials). On days 1 and 8, a set of 16 grating stimuli are also presented at 100% contrast (30 set repetitions; 480 trials). (C) Example frames for each movie clip. (D) 7-day trajectory of baseline normalized beta power evoked by two example movie clips (Movies 3 and 4). Exposure to natural movie clips potentiates beta power. (E) Experience modulation index quantifying changes in baseline normalized beta power between day 1 and day 8 across movie clips (green) and grating stimuli (blue; 7 mice). (F) Same as (E) for beta event power. (G) Same as (E) for beta event rate. (H) Same as (E) for LFP-MUA PPC during beta events. Data are presented as mean ± SEM; *: P<0.05, **: P<0.01, ***: P<0.001. See Supplementary Table 1 for detailed statistical analysis. See also Figure S4

Experience-dependent plasticity of GABAergic interneuron circuits

Previous work has found that visually evoked beta activity in V1 requires the activation of GABAergic interneurons co-expressing somatostatin (SST-INs)^47,48 and is negatively regulated by interneurons co-expressing vasoactive intestinal peptide (VIP-INs)⁴⁹. SST- and VIP-INs are mutually inhibitory^68–73, suggesting that expression of beta activity may be dynamically regulated by interactions between these two populations. We therefore examined how visual experience shapes the activity of SST- and VIP-INs in adult V1. We performed 2-photon Ca+ imaging to track visual responses of individual interneurons across days (Figure 5A, 5B), using the 16-stimulus protocol (Figure S3B; see Methods). Neurons were discarded if their receptive fields were misaligned with the visual stimuli (see Methods). Visual response amplitudes were significantly enhanced by visual experience in approximately half of the retained SST-INs (42/81 cells, 9 mice), whereas smaller proportions exhibited decreases (19/81 cells, 9 mice) or no change (20/81 cells, 9 mice) (Figure 5C–5F). The average visual response amplitude across the SST-IN population was robustly enhanced by visual experience (Figure 5G–H). These results remained consistent when analysis was restricted to periods of locomotion (Figure S5A–S5F). Furthermore, SST-INs whose responses were ultimately enhanced or suppressed exhibited distinct trajectories even on the initial days of visual experience and remained separated across days (Figures S5M–S5P), suggesting a rapid and robust differentiation between these subpopulations.

Figure 5. Visual experience induces opposing plasticity regimes in SST and VIP interneurons.

(A) Schematic representation of the experimental paradigm. Mice are head-fixed on a running wheel under a 2-photon microscope. Neural responses in V1 layer 2/3 are recorded via a cranial window, during presentation of a series of drifting gratings at 4 sizes (10, 20, 40, and 80 °) and 4 spatial frequencies (0.01, 0.04, 0.16, and 0.64 cycles/°). (B) Example SST-IN (upper) and its average response to a preferred stimulus (lower) over 8 days of visual experience. Blue shaded area indicates the visual stimulus. (C) Raster heatmap of population calcium activity. Z-scored calcium response averaged across all stimuli, represented by a colormap, around the visual stimulus indicated between dotted vertical lines, in SST-INs for each day of visual experience, sorted by the average response of the cells in naïve mice (first column) (n = 81 cells, from 9 mice). (D) Raster heatmap of the change in calcium response in each cell, calculated by subtracting first-day activity from last-day activity (first and last column in C), sorted by the magnitude of change across days (n = 81 cells, from 9 mice). (E) Z-scored response around visual stimulus averaged across all SST-INs in C, SEM shown in dark grey, visual stimulus window shown in light grey shade (n = 81 cells, from 9 mice). (F) Change in calcium response averaged across all cells in D (black, n = 81 cells). The subset of cells that were significantly up-regulated are denoted in red (n = 42 cells) and the subset that were significantly down-regulated are denoted in blue (n = 19 cells). (G) Average z-scored response across SST interneurons for each combination of size and spatial frequency over days of visual exposure (n = 81 cells). (H) Average z-scored response across visual stimuli for each SST interneurons between naive and visually experienced mice (n = 81 cells). (I) Same as C for VIP interneurons (n = 152 cells) (J) Same as D for VIP interneurons (n = 152 cells). (K) Same as E for VIP interneurons (n = 152 cells). (L) Same as F for VIP interneurons (red: upregulated, n = 8; blue: downregulated, n = 115; black: all, n = 152 cells). (M) Same as G for VIP interneurons (n = 152 cells). (N) Same as H for VIP interneurons (n = 152 cells). Data are presented as mean ± SEM; *: P<0.05, **: P<0.01, ***: P<0.001. See Supplementary Table 1 for detailed statistical analysis. See also Figure S5

In contrast to the SST-INs, the majority of VIP-INs exhibited reduced visual response amplitudes after visual experience (115/152 cells, 7 mice). In a small number of VIP-INs, responses were either enhanced (8/152 cells, 7 mice) or unchanged (29/152 cells, 7 mice) (Figure 5I–5L). Average visual response amplitudes across the VIP-IN population were decreased for most visual stimuli following the visual experience protocol (Figure 5M–N). These findings were largely unchanged by locomotion (Figure S5G–S5L). The small subset of VIP interneurons whose responses were enhanced following visual experience were largely those that were suppressed by visual stimulation on the initial days of visual experience (Figure S5Q–S5T), suggesting that visual experience causes VIP-INs to reduce their visual responses regardless of the direction of their initial response. In contrast, we found that parvalbumin-expressing interneuron (PV-IN) visual responses were largely unaffected by visual experience (Figure S6A–S6P), showing a small overall decrease in population visual response that was not replicated during locomotion. Varied visual experience thus has a small effect on soma-targeting inhibitory circuits. Overall, these results suggest that visual experience primarily rebalances the VIP-SST inhibitory circuit, suppressing VIP-INs and enhancing the visually evoked output of SST-INs.

Visual experience enhances visual selectivity in pyramidal neurons

SST-INs inhibit the dendrites of PNs, shaping input integration, Ca+ signaling, and synaptic plasticity^{69,71,74–76}. To examine how experience-dependent plasticity of SST-IN output affects excitatory neurons in the local circuit, we imaged the activity of layer 2/3 PNs. Visual experience resulted in a redistribution of responses in PNs (Figure 6A–6D), with subsets of PNs exhibiting suppression (148/313 cells, 9 mice), enhancement (106/313 cells, 9 mice), or no change (59/313 cells, 9 mice). However, the population average PN visual response amplitude remained unchanged (Figure 6E–6F). These results remained consistent when analysis was restricted to periods of locomotion (Figure S6Q–S6V). PNs whose responses were ultimately enhanced or suppressed began to diverge after the first day of visual experience (Figure S6W–S6Z) and exhibited distinct tuning for spatial frequency (Figure S6AA), suggesting that experience-dependent reorganization of the V1 circuit differentially affects PN populations with different visual selectivity.

Figure 6. Visual experience sharpens visual selectivity in pyramidal neurons.

(A) Raster images of average z-scored visual response around stimulus onset across stimuli in pyramidal neurons for each day of visual experience, sorted by magnitude of the cell’s initial response in naïve mice (n = 313 cells). (B) Raster image of the average response change around stimulus onset across stimulus type between naive and visually experienced mice in pyramidal neurons, sorted by change magnitude (n = 313 cells). (C) Average z-scored response around stimulus presentation across pyramidal neurons and stimuli over days of visual experience (n = 313 cells). (D) Average change in response between naive and visually experienced mice in upregulated (red, n = 106 cells) and downregulated subsets of pyramidal neurons (blue, n = 148 cells) and across all pyramidal neurons (n = 313 cells). (E) Average z-scored response across pyramidal neurons for each combination of size and spatial frequency over days of visual exposure (n = 313 cells). (F) Average z-scored response across visual stimuli for each pyramidal neuron between naive and visually experienced mice (n = 313 cells). (G) Average change in response between naive and visually experienced mice for each combination of size and spatial frequency for SST (n = 81 cells), VIP (n = 152 cells), and PV interneurons (n = 47 cells) and for pyramidal neurons (n = 313 cells). Stimulus combinations with significant changes are denoted by gray boxes. (H) Size tuning of two example pyramidal neurons in the naive (yellow) and visually experienced (cyan) states. (I) Average size tuning of pyramidal neurons in the naive (yellow) and visually experienced (cyan) states (n = 313 cells). (J) Histogram of the change in surround suppression index (SSI) for each pyramidal neuron between the first and last day of the visual experience protocol (n = 313 cells). (K) Schematic of changes induced in the V1 circuit by varied visual experience. In naive mice (left), SST interneurons are inhibited by visually responsive VIP interneurons, resulting in less SST-IN output to pyramidal neurons and decreased oscillatory synchrony. In visually experienced mice (right), decreased VIP interneuron visual responses and potentiated SST responses enhances the selectivity of pyramidal neuron visual responses and gives rise to stronger oscillatory synchrony. Data are presented as mean ± SEM; *: P<0.05, **: P<0.01, ***: P<0.001. See Supplementary Table 1 for detailed statistical analysis. See also Figure S6

Previous work found that SST-INs play a key role in mediating surround suppression in pyramidal neurons^77–79, suggesting that the experience-dependent enhancement of SST-IN output may shape PN selectivity for visual stimuli. Indeed, we observed an increase in PN response amplitudes for smaller stimuli (≤ 40 degrees) (Figure 6H–I) and an increase in PN surround suppression index (Figure 6J) following the visual experience protocol. Enhanced visual experience with varied stimuli may thus sharpen the selectivity of excitatory neurons for smaller visual stimuli. To examine whether rebalancing GABAergic inhibition also affects neural variability across stimulus repetitions, we assessed pairwise noise correlations between pyramidal neurons using a noise dependence measure (see Methods). We found that visual experience resulted in a significant decrease in noise dependence across days (Figure S6AC–S6AD). Together, our results suggest that visual experience organizes V1 circuit responses and optimizes encoding of stimuli by pyramidal neurons.

Discussion

Our results reveal a novel form of experience-dependent reorganization in the adult visual cortex that requires exposure to varied stimuli. We find that increased visual experience leads to a robust, selective enhancement of visually evoked beta activity. In response to varied visual stimuli over days, V1 exhibits increased amplitude of individual beta events in the cortical network and increased entrainment of spiking by those events. Underlying this augmentation of visually evoked patterned activity, we find that visual experience induces plasticity of visual responses in GABAergic interneurons, with SST-IN and VIP-INs exhibiting increased and decreased visually evoked activity, respectively. Finally, we find that visual experience increases the selectivity of nearby excitatory pyramidal neurons, leading to enhanced responses to small stimuli and increased surround suppression.

Visually evoked rhythmic synchrony in V1 emerges during the display of structured visual stimuli, particularly when visual features at one retinotopic site match that of the surround^{47,48,51,52,57,59,64}. Synchrony thus arises when stimuli conform to an internal representation of the expected structure of an image^60–63. However, it is unclear whether this representation is fixed early in development or remains sensitive to experience in maturity⁴². We find that visually evoked synchrony in the beta range in mouse V1 is plastic and strengthened by visual experience. Beta synchrony was induced both by drifting gratings and natural movie clips, suggesting that experience-dependent potentiation does not require a precise match between center and surround but rather a predictable relationship shaped by experience. Repeated exposure to a single static phase-reversing grating is reported to lead to modest potentiation of visually evoked synchrony in the alpha and beta bands in layer 4 V1⁴². However, our results suggest that beta plasticity is strongest upon experience of a varied stimulus set.

Previous work in mouse V1 found that visually evoked beta is reduced by the suppression of SST-INs^42,47,48 and reduced by activating VIP-INs⁴⁹. Experience-induced potentiation of beta could thus arise from either reduction in VIP-IN or enhancement of SST-IN activity. SST-INs are recruited by beta rhythmicity with a delay compared to excitatory and inhibitory fast spiking (FS) neurons^48,50. Increased SST-IN output may thus strengthen the refractory phase following synchronous discharge during beta events. In accordance with this hypothesis, our data indicate that experience-dependent plasticity increases the amplitude of beta events but reduces their rate. In addition, individual beta events more robustly entrained spiking following repeated visual experience, suggesting that beta potentiation reorganizes spike events and enhances spike synchrony in V1.

Repeated presentation of visual stimuli has been shown to result in a range of impacts on the amplitude of visually evoked cortical activity. Presentation of a single phase-reversed grating results in long-lasting potentiation of the amplitude of the layer 4 VEP that emerges across sessions, along with increased visually evoked firing of layer 4 excitatory neurons^{37–39,65,80,81}. This potentiation can be evoked by a range of stimuli but does not generalize to novel stimuli^37,38 and is associated with enhanced power in broadband alpha and beta frequencies (8–30Hz) and reduced gamma range activity^24,36,42. In contrast, repeated passive presentation of a single drifting grating stimulus results in decreased visually evoked responses in layer 4 and layer 2/3 excitatory neurons across days^23,24,45,65, suggesting that moving stimuli may lead to adaptation rather than potentiation of visual responses. Here, we found that repeated presentation of single and varied drifting grating stimuli and natural movies all resulted in robust potentiation of layer 4 VEPs. Because thalamocortical synaptic inputs make a substantial contribution to layer 4 VEPs, these results suggest that repeated presentation of a wide range of stimuli may potentiate the amplitude or synchrony of thalamocortical inputs. However, potentiation of visually evoked activity patterns within the cortex may instead depend on the number of stimulus repetitions or variation in stimulus features.

Variation in wakeful behavioral state, such as between quiescence and locomotion, regulates the amplitude of visual responses^82–85 and plays a role in experience-dependent plasticity^33–35. In addition, repeated stress can alter cortical sensory responses^86,87. Earlier studies incorporated up to two days of habituation to restraint prior to beginning repeated visual stimulation sessions^{17,37–42,65} and recorded cortical activity in mice restrained in a tube^{2,6,15,17,19,37–43,65,66,80,81}. In contrast, our protocol included extensive habituation and recordings in mice able to freely run on a wheel. Differences in either baseline behavioral state or locomotion may thus account for some reported differences across paradigms. However, we found that the response potentiation induced by repeated visual experience was robust to behavioral state, suggesting that this enhancement does not rely on arousal signals.

Cellular imaging revealed that in response to repeated exposure to varied stimuli, GABAergic interneuron circuits in V1 exhibited substantial changes in visual response amplitude. VIP- and SST-INs represent two distinct populations of GABAergic interneurons in the cortex, with distinct inputs, firing properties, and synaptic targets^68,88–93. VIP-INs primarily synapse on other GABAergic cells, including SST-INs, and make some synapses directly on PNs^{68,69,71,73,94}. In turn, SST-INs inhibit VIP-INs and innervate the dendrites of PNs, contributing to synaptic integration, calcium signaling and plasticity, and potentially gating inputs to apical dendritic segments^68,74–76. Both VIP- and SST-INs are sensitive to visual input, but whereas VIP-INs exhibit relatively small receptive fields⁹⁵, SST-INs are selective for larger stimuli^77–79,96. We found that a majority of V1 SST-INs exhibited progressive increases in visual response amplitude across visual presentation sessions. In contrast, almost all VIP-INs exhibited a decrease in visual responses over the same time period. In contrast, the soma-targeting PV-INs, which receive robust innervation from SST-INs^68,97 and may receive input from VIP-INs^72,98, exhibited only a small decrease in visual responses following visual experience. Together, these results suggest that visual experience selectively reshapes the dendrite-targeting VIP-SST circuit, leading to enhanced SST-IN output to local PN dendrites. In contrast, previous work found that repeated presentation of single stimuli leads to progressively decreased activation of layer 4 and layer 2/3 PV interneurons^{23,42,45,99,100}. Indeed, stimulus-specific response potentiation to phase-reversed grating stimuli may rely on activation of NMDA receptors in PV-INs¹⁷ and Nrst1 expressing cortico-thalamic neurons in layer 6^{40,41,101,102}.

Following repeated experience with varied stimuli, the responses of V1 PNs exhibited a selective increase in responses to small stimuli and increased surround suppression. Surround suppression in V1 has been suggested to emerge in part via inhibition from SST-INs, whose larger receptive fields may give rise to suppression of weaker PN responses to large diameter stimuli while permitting expression of strong PN responses to small diameter stimuli^77–79,96. Increased SST-IN output to PN dendrites following visual experience may thus lead to increased surround suppression of PNs and enhanced PN tuning for stimulus size. We further found that visual experience led to decreased pairwise noise correlations between local PNs, suggesting increased efficiency of visual encoding^103,104. Indeed, previous computational modeling results suggests that increased GABAergic inhibition may reduce noise correlations¹⁰⁵.

Taken together, our results further establish that experience leads to a sustained reshaping of the cortical response to visual stimuli. Exposure to varied stimuli leads to rebalancing of the evoked activity in GABAergic IN circuits, ultimately enhancing the visually evoked output from SST-INs. These circuit-level changes in inhibitory activation give rise to a robust strengthening of visually evoked beta rhythmicity and reorganization of the pattern of visually evoked spikes. In parallel, enhanced dendrite-targeting output from SST-INs is also associated with enhanced surround suppression and increased visual selectivity in local PNs. We find that this remodeling can be induced by a wide range of stimulus paradigms including drifting gratings and natural stimuli. These findings strongly suggest that the full expression of visual experience-dependent plasticity is contingent on specific aspects of the stimulus paradigm such as visual properties, stimulus repetition and behavioral state. Future studies should aim to further explore how the precise parameters of visual stimulation shape circuit interactions in the mature cortex and how different plasticity mechanisms engaged by sensory input may functionally interact to shape visual processing.

Methods

Experiments

Animals and viruses

All animal procedures were approved by the Institutional Animal Care and Use Committee of the Yale University School of Medicine (Protocol # 11317). Male and female of the following lines were used: C57Bl/6 mice, Sst-IRES-Cre mice (Jax 018973) crossed with Ai148 ((TIT2L-GC6f-ICL-tTA2)-D; Jax 030328), Vip-IRES-Cre mice (Jax 031628) crossed with Ai148, Pvalb-IRES-Cre mice (Jax 017320) crossed with Ai148, Thy1-GCaMP6s mice (Jax 024275), Emx1-cre mice (Jax 075934) injected with AAV9-Syn-FLEX-GCaMP6s (2 × 10¹² gc ml⁻¹; Addgene #100845) and Sst-IRES-Cre^+/− injected with AAV9-CaMKII2-GCaMP6s (2 × 10¹² gc ml⁻¹; Addgene #107790) and AAV9-CAG-Flex-ArchT (2 × 10¹² gc ml⁻¹; Addgene #209779). Mice were maintained on a 12-h light/dark cycle with ad libitum access to food and water. All experiments were conducted during the light phase when animals were between 3 and 6 months of age.

Surgery

Mice were anesthetized with isoflurane (1.5% in oxygen) and maintained at 37ºC for the duration of the surgery. Analgesia was provided with subcutaneous injections of Carprofen (5mg/kg). Lidocaine (1% in 0.9% NaCl) was injected under the scalp to provide topical analgesia. Eyes were protected from desiccation with ointment (Puralube). The scalp was resected, and the skull cleaned with Betadine. A surgical screw was implanted on the skull between the eyes and nuts were glued to the skull above the bregma suture, allowing the fixation of a headplate with bolts. Alternatively, the headplate were directly fixed onto the skull with C & B Metabond, (Butler Schein).

For chronic electrophysiology, two craniotomies were performed respectively above V1 on the left hemisphere (~0.15mm diameter; 3.75mm posterior, 2.5mm lateral from Bregma) and above the cerebellum (0.4mm diameter; ~6mm posterior from Bregma). An A16 probe with a CM16 connector (Neuronexus) was lowered into V1. Ground and reference wires were inserted above the cerebellum. Craniotomies were protected with Gelfoam (Pfizer), and all implants were affixed to the skull with C & B Metabond, (Butler Schein).

For calcium imaging, a 3 mm² square craniotomy was made over V1 on the left hemisphere. A custom-made glass window consisting of a 3 mm² square inner cover slip and 5 mm² round outer cover slip (both #1, Warner Instruments) glued with an ultraviolet-curing adhesive (Norland Products) was inserted into the craniotomy and secured to the skull with Cyanoacrylate glue (Loctite). A circular ring was glued to the titanium headpost and Metabond applied to cover all exposed skull.

For local virus injections, virus was first loaded into a glass micropipette. The tip of the pipette was lowered into the primary visual cortex through a craniotomy (3.75mm posterior, 2.5mm lateral from Bregma) at a depth of ~350 μm (QSI, Stoelting Co.).

Analgesics (5 mg/kg Carprofen) and anti-inflammatory steroids (2 mg/mL Dexamethasone) were given immediately after surgery and on the two following days and implanted mice were housed with food containing antibiotics (Sulfatrim, Butler Schein).

Chronic in-vivo extracellular electrophysiology

Mice were habituated to handling and head fixation for 5 days prior to recordings. Mice were head-fixed and freely running on a cylindrical wheel equipped with a magnetic angle sensor (Digikey) monitored wheel motion. The face (including the pupil and whiskers) was imaged with a miniature CMOS camera (Blackfly s-USB3, Flir) with a frame rate of 10 Hz.

For chronic electrophysiology, implants were connected to the recording apparatus (DigitalLynx system, Neuralynx). The most superficial contact point was used as a reference. Local Field Potentials, wheel motion, and timing signals for face movies and visual stimulus were acquired at a 40KHz sampling rate.

Chronic in-vivo two-photon cellular imaging

Imaging was performed using a resonant scanner-based two-photon microscope (MOM, Sutter Instruments) coupled to a Ti:Sapphire laser (MaiTai DeepSee, Spectra Physics) tuned to 920 nm for GCaMP6. Emitted light was collected using a 25 × 1.05 NA objective (Olympus). To prevent light contamination, the objective was enclosed in blackout material extending to the headpost. Images were acquired using ScanImage 4.2 at 30 Hz, 512 × 512 pixels at 150–250 μm depth from pia matter. Visual stimulation, wheel position, and Ca2+ imaging microscope resonant scanner frame ticks were digitized (5 kHz) and collected through a Power 1401 (CED) acquisition board using the Spike 2 software.

Visual stimulation hardware

Visual stimuli were generated using the Psychtoolbox Matlab extension and displayed on a 17” by 9.5” monitor situated 15 cm from the right eye. Screen display was linearized, and maximum luminance was adjusted to ~140 cd.sr/m². An iso-luminant grey background was displayed between visual stimuli.

Visual stimulation protocol

Visual exposure protocols lasted 6–8 days and used either vertical gratings drifting leftward at 1 or 2Hz or natural movies clips cropped from the movie Roman Holiday (1953). Stimuli were centered on the retinotopic location evoking the strongest visual response, measured with either MUA or visually evoked potential in LFP for electrophysiology, or fluorescence signal for calcium imaging. The following protocols were used:

−64 Stimuli:

Gratings were presented for 3s with a 2s inter-stimulus-interval (ITI) at all 64 combinations of 4 sizes (10, 20, 40, 80 ° diameter), contrasts (2, 8, 32 and 100%) and spatial frequencies (0.01, 0.04, 0.16 and 0.64 cycle/°). Each stimulus was repeated 15 times resulting in 960 total trials. This protocol was repeated 7 days.

−1 Stimulus:

A single highly salient grating (80°, 100%, 0.04 cycle/°), was presented for 3s with a 2s ITI for 960 trials between days 2 and 6. The 64-stimuli set was displayed on day 1 and 7 to probe for change across grating types.

−16 Stimuli:

Gratings were presented for 2s with a 5s ITI at full contrast at all 16 combinations of 4 sizes (10, 20, 40, 80 ° diameter) and spatial frequencies (0.01, 0.04, 0.16 and 0.64 cycle/°). Each stimulus was repeated 60 times resulting in 960 total trials. This protocol was repeated 7 days for electrophysiology recordings and 8 days for 2P recordings. This protocol was selected for 2P calcium recordings over the 64 stimuli since neuronal calcium responses were strongly modulated by behavioral states, thus requiring more repetitions for each stimulus.

-Natural Movies:

4 movie clips lasting 2s were presented with a 5s ITI, masked at 4 sizes (10, 20, 40, 80 ° diameter) resulting in 16 stimulus types. Between days 2 and 7 each stimulus was repeated 60 times resulting in 960 total trials. On days 1 and 7 each stimulus was repeated 30 times and 30 repetitions of the 16 Stimuli set were also displayed to test for changes across grating types, resulting in 960 total trials.

Data Analysis

Wheel and whisker motion

The first principal component of nose and whisker pad motion energy was computed from movie of the mouse face using FaceMap¹⁰⁶. Epochs of running and whisking were defined using a change point algorithm detecting local changes in the mean and variance of running speed and whisker pad motion. Briefly, the moving standard deviation of running speed or the first principal component of whisker motion were computed within a defined temporal window. The length t of this window determines the temporal resolution of the change-point analysis and was set to 4s for running speed and 500ms for facial motion. A first estimate of locomotion motion onset/offset times were then taken as the time when the moving standard deviations exceeded/ fell below 20% of its range above minimum. Estimates were refined in a window t around each onset/offset time by computing the time points corresponding to the maximum of the t-windowed moving forward/backward Z-score.

In-vivo electrophysiology

Data were analyzed in Matlab 2020b (Mathworks) using custom scripts. All time-series were down-sampled to 1KHz and aligned. Local field potential (LFP) recordings were high-pass filtered at 1Hz using a 2nd order Bessel filter.

Layer alignment of LFP and current source density (CSD) across recordings

To compute the average field potential around CBASS events across recordings, the LFP was linearly interpolated across channels to a common grid of laminar position. The CSD was derived as the second spatial derivative of the LFP across interpolated laminar positions. LFP channels were then mapped onto cortical layers using the CSD profile of visual responses. Two current sinks were identified and assigned respectively to layer 4 and layer 5b⁶⁷.

Visually Evoked Potential (VEP)

VEP is defined as the peak to trough amplitude of the LFP around the visual stimulus onset. Specifically, trough is restricted within 0–100 ms from the stimulus onset, while peak is restricted within 50–150 ms from the stimulus onset.

CBASS

CBASS (Clustering Band-limited Activity by State and Spectral features) ties a power increase in a defined frequency band (i.e., gamma (30–80Hz)) during a particular state (i.e., running) to the occurrence of defined events in the temporal domain (ref). A detailed description and implementations in matlab and python are available on (https://github.com/cardin-higley-lab/CBASS). Briefly, the multichannel LFP is filtered in the band of interest and candidate events are selected at the troughs of the filtered signal in a reference channel. In our case, the reference channel was chosen to be closest to layer 4. The spectrotemporal dynamics underlying each candidate event are parameterized using the real and imaginary part of the analytical representation (matlab function hilbert) of the filtered signal in each channel. Candidate events form a cloud in this parametric space where neighbors have similar spectro-temporal dynamics The event cloud is split randomly into $\mathit{n}$ partitions and a binomial test is performed in each partition to determine if events happen during the state of interest (i.e. running) at higher frequencies than overall. Partitioning is repeated $\mathit{N}$ time. A state enrichment score is calculated for events as the fraction of time they fell into an enriched partition. An optimization procedure is then applied to find the threshold yielding the most significant distance between events having a low and a high enrichment score in the feature space. Events above threshold are retained. Here we used n = 20 partitions and N = 1000. Different settings for these parameters have only a marginal influence on the result of the procedure.

The check the validity of the event partition, a state enrichment score is computed as described above on surrogate data having a spectrum and channel covariance matched with that of each recording. This surrogate data is constructed by decomposing the original LFP recording into principal components across channels (matlab function pca), randomizing the phase of their Fourier transform and remixing them. The fraction of candidate events above threshold on surrogate data indicates how likely pattern may be associated to the state of interest (i.e. locomotion) given the statistic of the signal.

Comparison of network activity within and outside CBASS event cycles

CBASS events are aligned to the trough of the band-pass filtered LFP in a reference channel. We defined each event’s boundaries as the peaks surrounding the event’s trough. Peak and troughs were determined as the 0 and π valued time points of the argument (matlab function abs) of the analytic representation (matlab function hilbert). Activity inside the event boundaries thus fell within a cycle centered on the trough. Epochs during and outside all CBASS event cycles were pooled separately and compared.

MUA distribution around CBASS events

MUA distribution around CBASS events was computed as follows. The lag separating each MUA spike from the nearest CBASS events was estimated. A histogram of lag values was then computed and normalized as follows. A baseline MUA rate $\mathit{p}$ was computed over samples. The variance of the rate over a window of $\mathit{n}$ samples was estimated assuming a binomial distribution as $\mathit{s}{\mathit{s}}_{\mathit{n}}^{\mathbf{2}}=\mathit{n}\mathit{p}(\mathbf{1}-\mathit{p})$. The normalized rate of events over a window of $\mathit{n}$ samples was then taken as ${\mathit{r}}_{\mathit{n}}=(\mathit{r}-\mathit{p})/\sqrt{{\mathit{s}}_{\mathit{n}}^{\mathbf{2}}}$ where $\mathit{r}$ is the event rate over samples and can be thought of as the number of standard deviations away from baseline. Thus, normalized histogram values represent standard deviation aways from the mean MUA rate under the assumption of an unmodulated binomial process.

Spectral analysis

The spectral power of a given time series was derived with Welch’s method. Each channel was divided into 500ms overlapping segments (75% overlap). Each segment was multiplied by a Hamming window and their Fourier transform was computed (matlab function fft). Power was derived as 10 times log10 of the squared magnitude of the Fourier Transform and expressed in dB. Power was averaged over segment and channels.

Band-limited LFP power was estimated by applying the Hilbert transform to zero-phase, band-pass filtered signals and computing instantaneous power as the squared magnitude of the analytic signal. Power was averaged across channels and stimulus epochs. Baseline power statistics were obtained from quiescent, no-stimulus periods, and stimulus evoked power was normalized by z-scoring relative to the baseline mean and standard deviation.

The spectral power of event-triggered averages was derived with a minimum bias multi-taper estimate. This differs from a classical multi-taper estimate in that Slepian tapers are replaced by a sinusoidal tapers sequence defined as:

$${\mathit{s}}_{\mathit{k}}=\sqrt{\mathbf{2}/\mathit{N}+\mathbf{1}}\mathbf{s}\mathbf{i}\mathbf{n}(\mathit{\pi }\mathit{n}\mathit{k}/\mathit{N}+\mathbf{1})$$

where$\mathit{N}$ is the number of samples in the triggered average, $\mathit{n}$ is the sample number and $\mathit{k}$ is the order of the taper. Sinusoidal tapers produce a spectral concentration almost comparable to that achieved with a Slepian sequence while markedly reducing local bias. The number of tapers was chosen to yield a bandwidth of 8Hz following the formula: $\mathit{K}=\mathit{r}\mathit{o}\mathit{u}\mathit{n}\mathit{d}\left(\right(\mathbf{4}\mathit{\pi }\mathit{N}\mathit{B}/\mathit{r})-\mathbf{1})$ where $\mathit{B}$ is the bandwidth and $\mathit{r}$ is the sample rate. Triggered averages were multiplied by each taper. Spectral power was then computed as described above and averaged over tapers.

For MUA phase locking quantification a spectro-temporal representations was first derived either for a set of frequencies using a wavelet transform (matlab function cwt) and a Morlet wavelet (matlab identifier cmor1–2) or across a full frequency band by computing the analytical representation of the filtered signal (matlab function Hilbert).

Spike phase locking was estimated using the Pairwise Phase Consistency defined as:

$$\widehat{PP{C}_f}=\sum _n^m\sum _m^N\frac{2\cdot \mathrm{c}\mathrm{o}\mathrm{s}\left({\theta }_n-{\theta }_m\right)}{N(N-1)}$$

where ${\mathit{\theta }}_{\mathit{k}}$ is the phase of the signal for frequency $\mathit{f}$ at the time of spike $\mathit{k}$ and $\mathit{N}$ is the total number of spikes. PPC provides an unbiased estimate of spike phase locking. However, estimate can be noisy if the spike number is inferior to 250. Thus, population estimates of PPC were derived by pooling MUA from all selected channels and the variance across mice was estimated by first calculating the PPC for individual mice.

Calcium signal pre-processing

Analysis of imaging data was performed either using ImageJ and custom routines in MATLAB (Mathworks) or with Suite2p¹⁰⁷. In the first case, motion artifacts and drifts in Ca2+ signals were corrected with the moco plug-in in ImageJ¹⁰⁸, and regions of interest (ROIs) were selected as previously described¹⁰⁹. Pixels in each ROI were averaged and the neuropil signal was subtracted^109–111. We observed no differences between the two pre-processing pipelines.

Calcium response to visual stimulation

Visual response was computed as the z-scored change in fluorescence during the 2 s visual stimulus (F) compared to the 1 s baseline before the stimulus (F₀), given by (F-mean(F₀₎)/std(F₀₎.

Visual responsiveness

The averaged z-score during the pre-stimulus 1-second period and 2-second stimulus period from all the trials were examined with a paired t-test to determine if each neuron is responsive to each stimulus on each session/day. Neurons that are responsive to a stimulus for more than a day-threshold were defined as responsive to that stimulus. Neurons responsive to at least one 20-degree or smaller size of stimulus, regardless of the spatial frequency, are defined as center-responsive. Only center-responsive neurons are included in the results.

Locomotion/Quiescent trials for cellular calcium recording

Wheel position was determined from the output of the linear angle detector. The circular wheel position variable was first transformed to the [−π, π] interval. The phases were then circularly unwrapped to get running distance as a linear variable, and locomotion speed was computed as the derivative of distance (cm/s). A change-point detection algorithm detected locomotion onset/offset times based on changes in standard deviation of speed. Locomotion onset or offset times were estimated from periods when the moving standard deviations, as determined in a 0.5s window, exceeded or fell below an empirical threshold of 0.1. Transition points were removed if a locomotion bout is under 1 second or a quiescent bout is under 2 seconds. Locomotion bouts were required to have average speed exceeding 0.5 cm/s. Only trials that completely fall into a locomotion bout or quiescent bout are categorized as a locomotion trial or quiescent trial respectively.

EMI

Experience Modulation Index is calculated using Z-scored Beta/Gamma power or calcium fluorescence with equation below:

$$\frac{Z_{Experienced}-Z_{Naive}}{Z_{Experienced}+Z_{Naive}}$$

Noise Correlation and Noise Dependence of Neuron Pairs

For each pair of neurons $i,j$, the stimulus-specific noise correlation coefficient ${\mu }_{i,j,s}$ was defined as the Pearson correlation coefficient between their trial-averaged z-scored responses for stimuli $s$:

$${\mu }_{i,j,s}=\frac{{\left({\mathit{x}}_{i,s}-\overline{{\mathit{x}}_{ı,s}}\right)}^{\mathrm{\top }}\left({\mathit{x}}_{j,s}-\overline{{\mathit{x}}_{Ј,s}}\right)}{\sqrt{Var\left({\mathit{x}}_{i,s}\right)Var\left({\mathit{x}}_{j,s}\right)}}$$

where ${\mathit{x}}_{\mathit{i},\mathit{s}}=\left\{{\mathit{x}}_{\mathit{i},\mathit{s},\mathbf{1}}\cdots ,{\mathit{x}}_{\mathit{i},\mathit{s},\mathit{T}}\right\}\in {\mathit{R}}^{\mathit{T}}$ refers to the average z -scored response of neuron $\mathit{i}$ to stimulus $\mathit{s}$ in each of the $\mathit{T}$ quiescent trails. The noise dependence ${\mathit{d}}_{\mathit{i},\mathit{j}}$ is the absolute value of the stimulus-specific noise correlation coefficient, averaged across all stimuli ($\mathrm{S}=16$), such that it only captures the magnitude of shared trial-to-trial variability, independent of the correlation sign:

$$d_{i,j}=\frac{1}{S}\sum _{s=1}^S\left|{\mu }_{i,j,s}\right|$$

Statistics

A detailed description of the statistics test in each figure panel and of statistical samples is provided in Extended Data Table 1. Unless otherwise noted, permutation tests were performed using mice as the statistical unit. Multiple comparisons were corrected using Benjamini & Hochberg procedure for false discovery rate.

Mixed-effects model

To assess changes in visual responses while accounting for the nested structure of the data (neurons recorded from the same mouse), we used linear mixed-effects (LME) models. First, to test whether the visual response change was significantly different from zero irrespective of stimulus identity, we fit an LME model with the change in response (deltaZ) as the dependent variable, an intercept as the fixed effect, and mouse identity included as a random intercept (deltaZ ~ 1 + (1 | MouseID)).

Second, to evaluate stimulus-specific response changes, we fit a separate LME model in which stimulus identity was included as a fixed effect without an intercept, allowing direct estimation of the mean response change for each stimulus, with mouse identity again included as a random intercept to account for repeated measurements within animals (deltaZ ~ −1 + Stimulus + (1 | MouseID)). Statistical significance of response changes was assessed using the estimated fixed-effect coefficients from the fitted models.