All animal care and use procedures were approved by the Animal Experiment Committee of the University of Tokyo. Adult C57BL/6 J mice (Japan SLC, both sexes, 3–6 months old) were housed under controlled conditions (23°C, 55% humidity) on a 12-h light/12-h dark cycle (lights on at 8:00 and off at 20:00). Food and water were available ad libitum, with fewer than five animals per cage. Grin2a (NR2A) homozygous knockout mice were backcrossed to a C57BL/6 J background for over 11 generations using breeding pairs originally provided by Dr. M. Mishina (University of Tokyo) (Fagiolini et al., 2003).

Under 1–1.5% isoflurane anesthesia (25–30 g body weight at surgery), stainless steel screws (1.1 mm diameter) were implanted over the right somatosensory cortex (1.5 mm lateral to the midline, 1.0 mm posterior to bregma) for EEG recordings. A reference electrode was implanted in the cerebellum (midline, 1.5 mm posterior to lambda). For EMG recordings, a stainless steel wire (100 μm) was inserted into the cervical trapezius muscle. For monopolar LFP recordings, insulated stainless steel wires (200 μm diameter) were stereotaxically implanted to target the anterior cingulate cortex (0.6, 0.3, 1.2), V1 binocular zone (medial: −3.4, 2.6, 0.4; lateral: −3.4, 3.0, 0.4), V2 higher-order cortical visual area lateromedial (V2LM: −3.4, 3.6, 0.4) (Wang and Burkhalter, 2007), and the corresponding regions in the left hemisphere. LFP electrodes were implanted 400 μm below the cortical surface in the binocular zone of V1, primarily targeting layer 4 and the upper portion of layer 5. These layers are known to produce the largest VEP responses in the SRP experiments (Frenkel et al., 2006). Electrodes were secured with petroleum jelly and dental acrylic. An aluminum frame (CF-10; Narishige, Tokyo, Japan) was attached to dental acrylic for head fixation. Local anesthesia (lidocaine) and antibiotics (ampicillin) were administered during surgery, and buprenorphine (0.05 mg/kg) was subcutaneously administrated for postoperative analgesia. After a one-week recovery period, animals were gradually habituated to the recording chamber and head-fixation setup until they remained calm for at least 30 min.

Flash stimulation was delivered using two fiber-coupled white LEDs (1.3 mm diameter) with optical fibers directed at each eye to cover the binocular zone (Figure 1A). The timing and duration of LED flashes were controlled by a TTL pulse generator (OPTG8, Doric Lenses, Quebec, Canada) and a constant-current stimulator (SEG-3104, Nihon Kohden, Tokyo, Japan). Light intensity was maintained at 100 cd/m² throughout the experiment. Visual stimulation began after an 8 min dark adaptation, with alternating 1 Hz flashes (10 msec duration) to each eye. Each eye received 100 flashes per session, repeated three times (300 flashes total) with 300 s intervals between repetitions. Sessions were conducted daily for 4–10 days (mean = 6.8 days). EEG, EMG, LFP, and LED timing data were recorded simultaneously and animal behavior was monitored using an infrared camera. Variations in the protocol are described in the respective sections.

Extracellular spike activity was recorded using bipolar electrodes made of twisted stainless wires (100 μm), implanted bilaterally in the binocular zone of V1 (−3.4, ±3.0, 0.4). A reference electrode with minimal neuronal activity was selected. Signals were band-pass filtered (0.15–8 kHz) and digitized at 40 kHz (MAP system, Plexon, Dallas, TX) whenever the signal exceeded a preset threshold of 4 standard deviations above the baseline. LFPs were recorded simultaneously with MUA. Electrode positions were marked post-experiment by electrolytic lesioning (50 μA for 10 s) and confirmed histologically. MUA data were processed using Offline Sorter (Plexon), and LFP and MUA analyses, including spike-triggered averages, were performed using NeuroExplorer (Nex Technologies, Madison, WI).

For chronic recordings, tethered mice with implanted electrodes were connected to a 16-channel commutator (Plexon) and allowed to move freely within a shielded cage. LFPs were recorded (filtered 0.7–170 Hz, sampled at 1 kHz) using the MAP system and analyzed offline. Visual stimuli (10 msec duration, 1 Hz, 3 × 100 flashes with 400 s intervals) were presented over 4 days using an LED array positioned above the cage (30–60 lx at floor level). Animals were kept awake by gentle tapping on the cage during visual stimulation in the dark. Sleep recordings were performed for 7–10 days to assess the relationship between visual plasticity and sleep. Sleep stages were scored manually based on EEG and EMG signals.

The NMDA receptor antagonist MK801 (FUJIFILM Wako Pure Chemical Corporation, Osaka, Japan) was dissolved in saline (0.1 mg/mL) and administered intraperitoneally (0.5 mg/kg) 15 min before daily recording sessions for 6 consecutive days. For ACC inactivation, the GABA_A receptor agonist muscimol (Merck, Darmstadt, Germany) was dissolved in saline (1 or 10 mM) and applied unilaterally (0.3 μL) to the ACC via a microsyringe 30 min before recording under light isoflurane anesthesia. The injection cannula remained in place for 30 s post-injection. A guide cannula (C313G, 7 mm length, P1 Technologies, Roanoke, VA) and an internal dummy cannula (C313DC, 8 mm length) were implanted along with the EEG/EMG and LFP recording setups. The injection cannula tip (C313I, 360 μm diameter, 8 mm length) was extruded 1 mm from the guide cannula end, targeting the ACC (0.6, 0.3, 1.2) ipsilateral to the EEG recording site.

SVM training and classification were performed with minor modifications to previously described methods (Makino et al., 2019). Positive and negative peak amplitudes were extracted from each VEP event by detecting the maximum and minimum potentials within a specified time window. Additionally, the total area under the VEP waveform was calculated by summing all areas above and below the baseline between 0 and 500 msec after visual stimulation. One or more of these indices, derived from one or more electrodes, were used for SVM training and classification. Randomly selected 75% of VEP events were used as the training dataset, while the remaining 25% were classified as left or right eye stimulation events using the fitcsvm and predict functions in MATLAB (MathWorks, Natick, MA) with a linear kernel. This process was repeated 1,000 times with different datasets, and mean classification accuracy was calculated for each mouse.

Data are presented as mean ± SEM. Statistical analyses were performed using Prism 9 (GraphPad Software, La Jolla, CA). Two-sided unpaired Student’s t-tests were used for comparisons between control and experimental groups, unless otherwise stated. For data with significantly different variances, Mann–Whitney U tests or Welch’s corrected t-tests were applied. Statistical significance was set at p < 0.05.

We investigated visual responses in the primary visual cortex (V1) of awake head-fixed adult mice (3–6 months old) during brief light stimulation targeting the binocular zone. Optical fibers connected to a white LED were securely positioned near the eyes and remained fixed throughout the experiments (Figure 1A). Neurons in the binocular zone of V1 receive inputs from both eyes, allowing quantification of responses to ipsilateral and contralateral stimuli (Figure 1B). Each eye was stimulated sequentially with brief flashes (10 msec) at a low frequency (0.5 Hz) with a 1 s interval between the eyes. Local field potential (LFP) recordings in V1 showed responses to visual stimulation from each eye (Figure 1C). Stimulus-triggered averaged visual responses, visually evoked potentials (VEPs), revealed distinct characteristic components: an early negative peak (N1, ~30 msec latency), a first positive peak (P1, ~100 msec), a second negative peak (N2, ~200 msec), a second positive peak (P2, ~300 msec), and a third negative peak (N3, ~400 msec), lasting approximately 600 msec (Figure 1D). N1 likely reflects input from the lateral geniculate nucleus, while later components may involve intracortical processing (Schroeder et al., 1998). Using this setup, we performed repeated light stimulation experiments for at least 4 days, with 3 sessions per day (Figure 1E).

Repeated light stimulation led to a gradual increase in V1 LFP amplitude, while somatosensory EEG activity remained unchanged (Figure 2A). Progression from the initial to the final stimulation revealed that P1 and N2 peaks became more prominent, with P2 and N3 peaks emerging later, while N1 remained stable (Figure 2B). Comparison of VEPs between day 1 and day 6 demonstrated a pronounced increase in N2 amplitude (Figures 2C,D). This response gain was generally greater for contralateral inputs than for ipsilateral inputs (Figure 2E). Combined data from both eyes showed significant increases in N2, P2, and N3 amplitudes by day 5, while N1 remained unchanged (Figure 2F). The time course of N2 amplitude exhibited steady growth, reaching a plateau around day 4 (Figures 2G,H).

In addition to gradual potentiation, visual responses increased rapidly within seconds of stimulation onset (Figure 3A). Initially weak responses strengthened, eventually resembling a typical VEP waveforms. Quantitative analysis of response strength, comparing early (first 0–10) with later (190–200) stimulations, confirmed rapid potentiation (Figures 3B,C). The rapid potentiation of VEPs was maintained even after the termination of the visual stimulation session and persisted when the next session began on day 1 and day 2 (Supplementary Figure S1). We refer to the cumulative process observed over several days (Figure 2) and the short-term enhancement (Figure 3) collectively as flash-evoked visual plasticity (FVP). Similar FVP was observed in young mice (P30), around the critical period for ocular dominance plasticity (Supplementary Figure S2).

The observed rapid potentiation may reflect changes in brain states, such as increased alertness or attention, rather than purely visual plasticity. We examined changes in the brain state during visual stimulation. Specifically, we quantified beta activity (20–30 Hz) and gamma activity (30–100 Hz) during the first session of visual stimulation by comparing the pre-stimulation, stimulation, and post-stimulation phases. EEG beta activity in the somatosensory cortex is commonly considered a marker of alertness or wakefulness, whereas gamma activity of LFP in the anterior cingulate cortex (ACC) is often associated with attention. Our results indicate that while beta and gamma activities in the V1L region increased markedly during stimulation, the increases in the EEG (somatosensory cortex) and ACC were more gradual and transient, peaking shortly after the onset of visual stimulation (Supplementary Figure S3). These findings suggest that changes in alertness or attention may contribute to the rapid process but are unlikely to fully account for it.

To determine whether FVP varies across cortical regions, we examined V1 (medial and lateral binocular zones), V2LM, and ACC, which are involved in attention, anticipatory activity and memory processing (Makino et al., 2019; Sidorov et al., 2020), using simultaneous LFP recordings (Figure 4A). While the V1 and V2LM regions displayed robust FVP, the ACC did not show any changes after repeated visual stimulation (Figure 4B). The increase in N2 amplitude was significant in the visual areas, but not in the ACC, paralleling the lack of somatosensory EEG changes (Figure 4C). Given the long latency of the N2 peak (~200 msec), cortico-cortical or cortico-subcortical interactions may contribute to the FVP. However, muscimol-induced inhibition of ACC activity after establishing the FVP did not affect V1 responses (Supplementary Figures S4A,B), suggesting that the FVP is localized within the visual cortex.

Because neurons in V1’s binocular zone receive inputs from both eyes, we investigated pathway-specific FVP. We first stimulated one eye for 4–5 days, and then stimulated the other eye for 4–5 days in the same animals (Figures 5A,B). Both stimulations significantly increased the N2 amplitude (~150% increase in each session), comparable to alternating eye stimulation (Figure 5C). This finding suggests pathway-selective plasticity and a locus of initial plasticity in the V1.

LFP, multiunit activity (MUA) recordings from the same electrode showed a close association between VEP negative peaks and neuronal activity (Figures 6A,B). The MUA peaks M1 and M2 were aligned with the VEP peaks N1 and N2, respectively. Neuronal activity significantly increased with VEP amplitudes, indicating that changes in VEPs closely corresponded to local cortical activity (Figure 6C).

Changes in brain responses would reflect the reorganized sensory information processing induced by sensory experience. To assess changes in sensory processing, we used a support vector machine (SVM) classifier to decode visual inputs (right or left eye) from VEP data (Figures 7A,B). Seventy-five percent of the visual response data from head-fixed mice were used to train the SVM classifier, while the remaining 25% of the data were reserved for testing the classifier’s performance. SVM analysis based on VEP peaks (P1, P2, N1, N2, N3) and the area under the curve revealed high classification accuracy, especially with visual cortex data compared with non-visual EEG or ACC data (Figure 7C). The classifier achieved over 95% accuracy with optimized time windows, with higher accuracy observed in responses from the late phase (day 4) compared with the initial phase (day 1) VEPs, suggesting enhanced sensory discrimination through the FVP (Figures 7D,E). No significant changes in VEP N1 amplitude were detected before and after FVP (Figures 2D–F). However, SVM analysis revealed a significant increase in discriminative power following FVP (Figure 7D). This is likely due to the SVM leveraging simultaneously recorded data from multiple visual regions, surpassing the simple averaging of the N1 amplitude (Figures 2D–F).

NMDA receptors play a crucial role in plasticity. We used Grin2a (a gene encoding NR2A) homozygous knockout (NR2A KO) mice (Sakimura et al., 1995; Fagiolini et al., 2003) to test the involvement of NMDA receptors in the FVP. NR2A protein, a subunit of NMDA receptors, develops postnatally and is incorporated into NMDA receptors (Quinlan et al., 1999). Lack of the NR2A subunit decreases NMDA receptor currents in the visual cortex (Miyamoto et al., 2003), resulting in the loss of plasticity of orientation selectivity (Fagiolini et al., 2003) and sensory experience-dependent slow-wave activity during non-REM sleep in developing mice (Miyamoto et al., 2003). We found that these mice lacked FVP, implicating NMDA receptor function in this plasticity process (Figures 8A,B), as well as other types of visual receptive field properties and learning behaviors (Frenkel et al., 2006). Heterozygous NR2A knockout mice showed intermediate FVP compared with wild-type and homozygous knockout mice (Supplementary Figures S5A,B). Additionally, pharmacological inhibition of NMDA receptors in wild-type mice using MK801, an open-channel NMDA receptor blocker, before stimulation abolished FVP, confirming its dependence on NMDA receptor signaling (Figures 8C,D). Rapid potentiation was also impaired in NR2A KO mice (Figure 8E).

We tested FVP in freely moving mice by delivering LED stimulation from above in a recording chamber (Figure 9A). VEP recordings showed that the visual response amplitudes (P1 and N2 peaks) gradually increased with repeated stimulation, similar to head-fixed mice (Figures 9B–D). N2 peak amplitude significantly increased by day 4 compared with that on day 1 (Figure 9E).

Occasional spontaneous 3–6 Hz oscillations have been observed in the visual cortex during wakefulness, displaying spike-and-wave patterns that resemble evoked responses (Senzai et al., 2019) and spontaneous activities (Miyamoto et al., 2019) (Figure 10A). These spontaneous oscillations became more evident after prolonged stimulation (day 4 vs. day 1) (Figure 10B). We also examined the effects of FVP on the activity during non-REM (NREM) sleep. Following multiple days (4–10 days, average 6.8 days) of stimulation (300 pulses at 1 Hz every 3 h), delta (0.5–4 Hz) rhythms in V1 increased significantly during NREM sleep, while somatosensory EEG activity remained unchanged (Figure 10C). Normalized V1 delta activity to EEG to reduce potential changes associated with global brain activity confirmed this increase (Figure 10D), whereas theta activity (6–10 Hz) during REM sleep or gamma activity during waking (40–60 Hz) remained stable (Figure 10E). This suggests that FVP selectively enhances delta activity during NREM sleep in V1.