Visual experience can cause functional modification in the adult visual cortex; however, how cortical receptive fields (RFs) are dynamically modified by natural scene stimulation remains unclear. Here, using in vivo patch-clamp recordings from neurons in the rat primary visual cortex (V1), we showed that minutes of conditioning with natural movies could increase the similarity between cortical RF structure and the subset of movie images that depolarized the cell. This effect lasted for a few minutes in the absence of further movie stimulation. Manipulating the statistics of the movies by temporal shuffling or spatial whitening showed that the spatiotemporal correlation of the movie was important in inducing the RF modification. Furthermore, the movie-induced RF modification required the activation of N-methyl-d-aspartate receptors. Such rapid RF modification may play an important role in the dynamic coding of natural scenes.
Sensory experience plays an important role in shaping mammalian cortical circuits during both development and adulthood (Buonomano and Merzenich 1998; Gilbert 1998; Karmarkar and Dan 2006; Feldman 2009). For example, during the critical period, monocular deprivation causes a shift in ocular dominance in V1 (Wiesel and Hubel 1963) and a restructuring of geniculocortical arbors (Antonini and Stryker 1993), and early experience with moving visual stimuli has a significant impact on the development of cortical direction selectivity in visually naive animals (Li et al. 2008). During adulthood, the visual circuits can be modified by various manipulations, including perceptual learning (Schoups et al. 2001; Frenkel et al. 2006) and visual deprivation by retinal lesion (Kaas et al. 1990; Gilbert and Wiesel 1992). Adult cortical receptive field (RF) can also be modified at a time scale on the order of minutes by repetitive visual stimulation (Dragoi et al. 2000; Yao and Dan 2001; Fu et al. 2002) or by changing the statistics of visual stimuli (David et al. 2004; Sharpee et al. 2006).
Many forms of activity-dependent RF plasticity are likely mediated by synaptic modifications. For example, synchronous visual stimulation and iontophoretic cortical activation can change the orientation tuning and ocular dominance of V1 neurons (Frégnac et al. 1992; Shulz and Frégnac 1992). Synchronous stimulation of the RF and a region outside the RF can induce RF expansion toward the co-stimulated surround (Eysel et al. 1998). These effects are consistent with a Hebbian synaptic learning rule (Hebb 1949), according to which synaptic connection is strengthened when pre- and postsynaptic neurons are active simultaneously. Previous studies also revealed stimulus-timing-dependent changes in the cortical RF position and orientation tuning (Schuett et al. 2001; Yao and Dan 2001; Fu et al. 2002), which may be mediated by spike timing-dependent plasticity of synaptic connections (Bi and Poo 1998; Caporale and Dan 2008).
Although the aforementioned studies using artificial stimuli provided insight into the synaptic mechanism of experience-dependent RF modification, the effect of more natural sensory stimulation on adult cortical response remains poorly understood. A recent study in cat V1 showed that repeated stimulation with movies of natural scenes induced rapid improvement in response reliability and caused reverberation of the movie-evoked responses in spontaneous activity (Yao et al. 2007), but it remains unclear how RF is dynamically modified after movie stimulation and whether N-methyl-d-aspartate (NMDA) receptor-dependent synaptic plasticity is involved. Furthermore, it is not clear whether the movie-induced response modification (Yao et al. 2007) is due to the spatial or temporal correlations present in the movie. On the other hand, previous studies have shown that RFs measured with natural stimuli differed from those measured with artificial stimuli (David et al. 2004; Felsen et al. 2005; Sharpee et al. 2006), yet it is unclear whether natural stimuli cause RF changes that can be observed after the termination of natural stimulation or the RF difference is due to different mapping stimuli.
Using in vivo patch-clamp recordings from neurons in the rat primary visual cortex, we measured synaptic RFs before and after repetitive conditioning with movies of natural scenes. By changing the statistical properties of the conditioning stimuli, we were able to address whether RF modification depends on the spatiotemporal correlation of the stimulus. We showed that repetitive stimulation with natural movies caused enhancement and reduction in the amplitudes of visually evoked responses at different RF locations. After movie stimulation, the spatial structure of RF became more similar to the subset of movie images that depolarized the cell. This effect depended on the spatiotemporal correlation of the movie and required the activation of NMDA receptors. Furthermore, during movie conditioning, the spike rate to the preferred images of neurons selectively increased. Our results demonstrate that movie stimulation can cause rapid RF modification, which may enhance the stimulus selectivity of the neuron.
Materials and Methods
Surgery and Preparation
The use and care of animals complied with the guidelines of the Animal Advisory Committee at the Shanghai Institutes for Biological Sciences, and all the experiments were approved by the Animal Care and Use Committee at the Institute of Neuroscience, Chinese Academy of Sciences (Shanghai, China). Adult Long-Evans rats (200–350 g) were anesthetized with an intraperitoneal injection of urethane (1.3 g/kg) or a mixture of fentanyl (0.3 mg/kg) and medetomidine (0.3 mg/kg). Additional doses were administered when needed. The animal was head fixed into a stereotaxic apparatus, with a stream of O2 passing over the nose. Body temperature was maintained at 38°C with a heating blanket (FHC Inc.). A craniotomy (diameter 0.5 mm) was made above the right V1 (6.5–7.0 mm posterior to the bregma and 3.0–4.0 mm lateral to the midline), and the dura was removed. The left eye was fixed with a metal ring to prevent eye movement and occasionally irrigated with a sterile saline solution.
Patch recordings were made with a MultiClamp 700B amplifier (Molecular Devices). Electrodes were pulled with a micropipette puller P-97 (Sutter Instrument) from unfilamented borosilicate glass (outer diameter: 1.5 mm; inner diameter: 0.8 mm, Sutter Instrument) to tips of 2 μm. The pipettes (4–7MΩ) were filled with internal solution containing (in mM) K-gluconate 136.5, KCl 17.5, NaCl 9, MgCl2 1, 4-(2-hydroxyethyl)-1-iperazineethanesulfonic acid 10, and ethylene glycol tetraacetic acid 0.2 (pH 7.3 and mOsm 290). Amphotericin B (0.375 mg/mL, EMD Biosciences) was added before the experiment. The electrodes were inserted perpendicular to the cortical surface and advanced with a motorized micromanipulator (MX7600, Siskiyou). Visual responses were recorded in the current-clamp mode. The mean resting potential of the cells was −62.7 ± 8.0 mV (mean ± standard deviation [SD]).
For a subset of cells (n = 24 in Fig. 3E, n = 15 in Fig. 5B–D, n = 48 in Fig. 5E, n = 26 in Fig. 6A, n = 48 in Fig. 7B, and n = 20 in Fig. 7C), single-unit extracellular recordings were made with single tungsten microelectrodes (5MΩ, A-M systems) or a 32-channel silicon probe (A4 × 8-5 mm-200-200-177, impedance 1.1–2.3 MΩ, NeuroNexus Technologies). The neural responses were amplified and filtered using an A-M systems Model 1800 amplifier or a Cerebus 128-channel system (Blackrock microsystems). Unit isolation was based on cluster analysis of waveforms. The data were sampled at 10 kHz by a Digidata 1440A digitizer (Molecular Devices) or at 30 kHz by the Cerebus 128-channel system. All the data were analyzed with the custom software in Matlab.
Visual stimuli were presented with a 7″ LCD monitor (Xenarc Technologies Corp.; refresh rate 60 Hz; mean luminance 97.3 cd/m2) placed 35 mm from the left (contralateral) eye. All the stimuli were updated every 2 frames, resulting in an effective frame rate of 30 Hz. The RF was mapped using sparse-noise stimuli, in which a white (or black) square (8.9o–11.2o) was flashed on a gray background at each of the 8 × 8 positions in a pseudorandom sequence. During movie conditioning, the RF was repetitively stimulated with a movie of time-varying natural scenes (van Hateren and Ruderman 1998). Each movie was 24.7 s long, containing 730 consecutive images (32 × 32 pixels, root mean square (RMS) contrast of 0.29) followed by 12 blank images. A total of 8 natural movies were used in this study. The mean luminance for each image in the conditioning stimuli was the same. Both the sparse-noise stimuli and the conditioning stimuli subtended 71.1° × 71.1°–90° × 90°.
To examine whether the spatiotemporal statistics of the conditioning stimuli influence RF modification, we also used 3 additional types of stimuli: Temporally shuffled movie, spatially whitened movie, and low-spatial-resolution white noise. To generate the shuffled movie, the images in the natural movie were temporally shuffled to remove temporal correlations. To generate the whitened movie, for each image in the natural movie, we randomized the spatial power spectrum by drawing the amplitude at each frequency randomly from a uniform distribution on [0 1]. In total, 6 shuffled movies and 6 whitened movies were used. For white noise conditioning, the movie images were replaced with 730 frames of spatiotemporal white noise (8 × 8 pixels, RMS contrast of 0.29). When we measured the responses of the same cell to the 4 types of conditioning stimuli (Fig. 5A–D), different stimuli were randomly interleaved and each stimulus was repeated 15 times.
RF Measurement, Conditioning, and Data Analysis
For each cell, we first mapped the RF using 50 or 75 repeats of sparse-noise stimuli. We then presented 5 blocks of conditioning and mapping stimuli. In each block, 5 repeats of movie stimuli (or other types of conditioning stimuli) were followed by 10 or 15 repeats of RF mapping stimuli. After each block of conditioning, 2 s of blank frames were presented to measure the spontaneous activity. The responses to the mapping stimuli in all the 5 blocks were averaged to obtain the RF after conditioning. The subthreshold (or spike) responses to sparse-noise stimuli were binned at 33.3 ms, and the spatiotemporal RFs (STRF) were computed using reverse correlation (Jones and Palmer 1987). To determine whether the response at a stimulus position was significant, we computed a baseline activity using the mean response within 0.5–1.5 s after stimulus onset at each position, and defined a threshold as 5 times SD above the baseline. For each stimulus position, the subthreshold (or spike) response was considered to be significant if the peak amplitude was above the threshold. For those positions at which the visual responses were significant, we compared the amplitudes of peak responses before and after the conditioning by computing the percent change in amplitude. For each cell, we computed the variance of the RF at each time delay and defined a signal-to-noise ratio (SNR) as the ratio of the maximum variance to the mean variance for RFs at delays 0.5–1.5 s after stimulus onset (Malone et al. 2007). For most cells, the SNRs for the ON and OFF regions were imbalanced, and the ON and OFF regions of the RF were largely overlapped (see Supplementary Material and Supplementary Fig. S1). To analyze the effect of conditioning, we picked the dominant RF region for each cell (Supplementary Material). Among the cells for which we measured RFs and analyzed the RF spatial profiles, only those with SNR >1.5 were included (patch recordings, n = 133; extracellular recordings, n = 48). The STRF was normalized to have unit variance and the spatial RF map at the delay with peak variance was used for further analysis.
To examine the relationship between RF modification and movie responses, we grouped the movie images into 3 subsets, S+, So, and S– images, by the following method. For the responses averaged over the last 2 blocks of conditioning (binned at the stimulus frame rate, 33.3 ms), we computed the mean and SD for the time bins above the mean potential, and defined an upper threshold as 1 SD above the mean (Fig. 3A, red dashed line). A lower threshold for the time bins below the mean potential was defined in a similar manner (Fig. 3A, blue dashed line). We defined those images that caused responses higher than the upper threshold as S+ images, and those that caused responses lower than the lower threshold as S− images. In addition to using 1 SD to define the upper and lower threshold, we also tried 0.75 SD and 1.25 SD in the analysis of RF modification (Supplementary Fig. S4). The third group of images, So, was defined as those images corresponding to the time bins within 0.25 SD of the mean potential (Fig. 3A, black dashed lines). In choosing the images for each group, we assumed that the response latency was 100 ms (corresponds to 3 frames). For the spatial RF measured before and after (RFbefore and RFafter) movie stimulation, we computed the correlation coefficient (CC) between the RF and each image in the 3 groups. For each cell, we further computed the difference between the mean CC measured using RFafter and RFbefore (defined as ΔCC), for each group of images.
To estimate the persistence of movie-induced RF modification, we mapped RF by 50 repeats of sparse noise every 3 min after the last block of conditioning and mapping, and computed the CC between the RF and each of the S+ images. Because stable patch recordings could be sustained only for a limited time and the spike RF also showed similar modification as the synaptic RF (Fig. 3D), for this experiment we performed extracellular recordings (n = 24, 22, and 20 for the 3 time points in Fig. 3E, respectively) as well as patch recordings (n = 43, 16, and 13 for the 3 time points in Fig. 3E, respectively). For extracellular recordings, we included only those cells in which both the SNR of RF >1.5 and the movie-evoked firing rate >1.5 spikes/s.
To define S+ images for extracellular responses, we binned the movie-evoked peristimulus time histogram (PSTH) at the stimulus frame rate and set a threshold of 0.25 of the highest amplitude of the PSTH. Those images that caused extracellular responses higher than the threshold were defined as S+ images. The reason to choose 0.25 for the threshold was that, for the subset of cells recorded intracellularly in which the movie-evoked firing rate >1.5 spikes/s (n = 15), the median firing rate for those bins corresponding to S+ images was 0.25 of the maximum rate.
To analyze the spiking responses during the 5 blocks of movie conditioning, we set a threshold of 0.25 of the highest amplitude of the movie-evoked PSTH, and defined event (non-event) as those time bins that were above (below) the threshold.
To compare extracellular responses of the same cell to the 4 types of conditioning stimuli (Fig. 5A–D), the spike trains were binned at the stimulus frame rate (33.3 ms). To quantify the reliability of responses to repetitive conditioning, we computed the average between-trial CC for responses to the same stimulus. To quantify the degree of neural selectivity for the responses to each type of visual stimulus (Fig. 5D), we computed lifetime response sparseness S: S = 1 − a, where a denotes the activity fraction.Haider et al. 2010). As a nonparametric measure of neuronal selectivity, sparseness makes no assumption about the form of firing rate distribution, the tuning properties of the neuron, and the number of stimuli in the set (Rolls and Tovee 1995). For highly selective neurons (e.g., those respond to a single frame in the movie), S will approach 1. For nonselective neurons (e.g., those respond to all movie frames), S will approach 0.
Cortical Application of APV
To block the NMDA receptors, a standing pool of 1 mM d-2-amino-5-phosphonovaleric acid (APV) (Tocris Bioscience) was applied to the surface of primary visual cortex. We performed extracellular recordings to estimate the effect of APV application on spontaneous firing (n = 26, Fig. 6A) and movie-evoked responses (n = 20, Fig. 7C), and patch recordings (n = 13, Fig. 6E) to measure RF before and after movie conditioning in the presence of APV. For patch recordings, the spontaneous membrane potential responses of most cells showed visible hyperpolarization within 1 min of APV application. RF mappings and movie conditioning were performed after 5 min of APV application. To compensate for the hyperpolarization effect, a small amount of positive current (0–50pA, depending on the degree of hyperpolarization and the input resistance of the cell) was injected to depolarize the cell to the potential before APV application. Only those cells whose membrane potential showed >2 mV hyperpolarization after APV application were included in the analysis.
Movie-Stimulation-Induced Synaptic Modification in Receptive Field
RFs of V1 neurons were measured under current clamp with sparse noise (Fig. 1A,B) before and after repetitive visual stimulation with a movie of natural scene (see Materials and Methods). In many neurons, the movie-evoked responses showed sparse episodes of depolarization, which were reliable over trials (Fig. 1B). When the RF was measured again after the movie conditioning, we found that the peak response amplitudes were enhanced in some positions but reduced in other positions (P < 0.01, Wilcoxon rank-sum test, for all 4 positions marked with * in Fig. 1B). Over the population, movie stimulation induced significant changes (P < 0.05, Wilcoxon rank-sum test) in response amplitude for 1.5 ± 0.2 positions (or 26% ± 4% positions; mean ± standard error of the mean [SEM]) in the RF (Fig. 2A,B). Among all the positions at which the responses were significant (see Materials and Methods), both reduction and enhancement in response amplitude were observed (Fig. 2C), indicating that movie stimulation caused disparate circuit modifications. To analyze the movie-induced RF change, we extracted the spatial profile of the RF at the time delay of maximum response (see Materials and Methods), and computed the CC between the spatial RF measured before and after movie stimulation (Fig. 2D). As a control, we also estimated the degree of spontaneous RF change after blank screen stimulation (same duration as the movie). The CC measured with movie conditioning was significantly smaller than that measured with blank control (P < 0.0002, Wilcoxon rank-sum test, Fig. 2D), indicating that the movie-induced RF change cannot be accounted for by the spontaneous fluctuations in the responses or the eye movements of the animal.
Increased Similarity Between Receptive Field and Movie Images
Previous studies showed that experience-dependent plasticity depends on the pattern of sensory inputs and the corresponding neural activity (Karmarkar and Dan 2006). Since the RF measured by sparse-noise stimuli could account for a small fraction of the movie-evoked responses (see Supplementary Material and Supplementary Fig. S2), we examined the relationship between RF modification and the movie responses. As seen from Figure 1B, the neuron was markedly depolarized in response to some images of the conditioning movie. According to the Hebb's rule, we reasoned that the synaptic inputs associated with depolarization may be strengthened, and RF measured after the conditioning may be more similar to those images that caused depolarization. To test this conjecture, we identified 3 groups of movie images, S+, So, and S− images, based on the movie-evoked membrane potential response (see Materials and Methods, Fig. 3A). Over the population, the responses evoked by S+ and So images were 8.6 ± 0.5 and 2.3 ± 0.3 mV (mean ± SEM) above the mean spontaneous response, respectively, whereas the response evoked by S− images was 2.3 ± 0.3 mV (mean ± SEM) below the spontaneous level (Supplementary Fig. S3). For each group of movie images, we measured the similarity of the image to RFbefore (or RFafter) by CC. Figure 3B,C shows the probability distribution of CCs for 2 example cells, computed between RF (upper, RFbefore; lower, RFafter) and each image in the 3 groups (red, S+ images; black, So images; blue, S− images). For both cells, the mean CC for S+ images was higher for RFafter (Fig. 3B, CC = 0.1; Fig. 3C, CC = 0.38) than for RFbefore (Fig. 3B, CC = 0.05; Fig. 3C, CC = 0.34), indicating an increase in similarity between S+ images and RF structure after the movie conditioning. For each cell, we further computed a ΔCC, the difference between the mean CC measured using RFafter and RFbefore, for each group of images. Over the population (n = 43), we found that ΔCC was significantly larger than zero for S+ images (P< 0.0005, Wilcoxon signed-rank test) and So images (P < 0.02, Wilcoxon signed-rank test), but not for S− images (P = 0.15, Wilcoxon signed-rank test, Fig. 3D left panel). This indicates that the spatial profile of synaptic RF became more similar to S+ and So images after movie stimulation. To reduce the arbitrariness in choosing the upper (lower) threshold used to define S+ (S−) images (see Materials and Methods), we repeated the analysis of Figure 3D using another 2 different thresholds and found that ΔCC remained significant for S+, but not for S− images (Supplementary Fig. S4). For cells in which both the synaptic RF and the spike RF were measured, when we performed the analysis using the spike RF, we found that ΔCC was significantly larger than zero only for the S+ images (P < 0.05, Wilcoxon signed-rank test, n = 29, Fig. 3D, right panel). This result is consistent with the effect that spike threshold serves to enhance feature selectivity of cortical neurons (Priebe and Ferster 2008). To examine the persistence of RF modification, we continued to map RF in the absence of further movie conditioning (see Materials and Methods). We found that the effect of movie conditioning persisted for several minutes (Fig. 3E). The rapid decay of RF modification may be due to the spontaneous activity or spiking activity evoked by the mapping stimuli, which can reverse activity-induced synaptic modifications in vivo (Zhou et al. 2003).
Dependence of RF Modification on the Spatiotemporal Correlation of Conditioning Stimulus
Natural scenes exhibit significant correlations in space and time (Field 1987; Dong and Atick 1995). To examine whether the RF modification depends on the spatiotemporal characteristics of the conditioning stimuli, we also tested the effect using 3 additional types of stimuli: Temporally shuffled movie, in which the images in the natural movie were temporally shuffled; spatially whitened movie, in which the pixels in each image were spatially uncorrelated; and low-spatial-resolution white noise (see Materials and Methods). For white noise stimuli, we used a spatial resolution lower than that of natural movie so that the stimuli could effectively drive the neurons (Supplementary Fig. S5). When we measured the effect of the 3 types of conditioning stimuli on synaptic RF, we found that none of them induced significant change in CC (Fig. 4A–C), indicating that both the spatial and the temporal correlations of the conditioning stimulus are important in inducing the RF modification.
Why do natural movie conditionings induce RF modification more efficiently than the other types of visual stimuli? To address this question, we compared the response properties to natural movies with those to other stimuli. The mean membrane potential to natural movies was not significantly different from that to each of the 3 types of stimuli (Fig. 4D, P > 0.4 for all the 3 comparisons, Wilcoxon rank-sum test). The reliability of membrane potential response (see Materials and Methods) to whitened movies, however, was significantly lower than that to natural movies (P < 0.01, Wilcoxon rank-sum test), whereas the reliability to shuffled movies (or white noise) was not significantly different from that to natural movies (Fig. 4E, P > 0.2 for shuffled movies, P > 0.3 for white noise, Wilcoxon rank-sum test). To further examine the response difference, we made extracellular recordings to measure the responses of each cell to the 4 types of stimuli (see Materials and Methods). As shown in the example in Figure 5A, both the firing rate and the temporal structure of this neuron's responses appeared different to the 4 stimuli. Over the population (n = 15), we found that the mean firing rate to whitened movies was significantly lower than that to natural movies (P < 0.00005, Wilcoxon signed-rank test), whereas the mean rate to shuffled movies (white noise) was higher than (similar to) that to natural movies (Fig. 5B). The reliability of response to whitened movies was significantly lower than that to natural movies (P < 0.0002, Wilcoxon signed-rank test), whereas the reliability to shuffled movies (white noise) was similar to that to natural movies (Fig. 5C). This indicates that whitened movies did not drive the neurons as efficiently as natural movies. We next determined what other aspect of the responses may exhibit difference between natural movie and shuffled movie (or white noise) stimulation. Visual inspection of the example cell in Figure 5A shows that the spikes appeared in fewer frames under the stimulation of natural movie compared with shuffled movie or white noise, suggesting a difference in neuronal selectivity for different stimuli. To quantify the degree of selectivity, we used the measure of lifetime sparseness (see Materials and Methods). Sparseness approaches 1 for highly selective neurons and approaches 0 for nonselective neurons. Over the population, the response sparseness was significantly higher for natural movies than for shuffled movies (P < 0.0002, Wilcoxon signed-rank test) or white noise (P < 0.0002, Wilcoxon signed-rank test, Fig. 5D). Although the response sparseness for whitened movies was similar to that for natural movies (P > 0.5, Wilcoxon signed-rank test, Fig. 5D), this was likely due to the lower firing rate to whitened movies. When we further examined the relationship between the response sparseness and RF modification, we found that ΔCC for S+ images was correlated with the sparseness of movie-evoked response (r = 0.37, P < 0.01, Fig. 5E). Thus, the effectiveness of natural movies at inducing RF modification may be partially accounted for by the higher degree of response sparseness.
Dependence of RF Modification on NMDA Receptors
Many forms of activity-dependent synaptic modification require activation of NMDA receptors (Bear 1996; Malenka and Nicoll 1999; Dan and Poo 2004). To test whether the mechanism underlying movie-induced RF changes involves NMDA receptor-dependent synaptic plasticity in the cortex, we locally applied NMDA receptor antagonist APV to the visual cortex (see Materials and Methods). Previous in vivo studies showed that APV application could cause significant reduction in both the spontaneous and visually evoked activity (Fox et al. 1989; Miller et al. 1989). In our study, we found that the spontaneous firing rate decreased by ∼50% a few minutes after APV (1 mM) application and the effect lasted at least 30 min (Fig. 6A). Patch recordings showed that the spontaneous membrane potential was hyperpolarized by −4.4 ± 2.2 mV (mean ± SD) after the application of APV (Fig. 6B). To compensate for the effect of hyperpolarization, we injected a small amount of depolarizing current during RF mappings and natural movie conditioning (see Materials and Methods), which ensured that the movie stimuli could evoke robust depolarization and spiking responses in the presence of APV (Fig. 6C). While the movie-evoked responses were reliable and sparse, the probability distributions of CCs were similar between RFbefore and RFafter (Fig. 6D). Across the population, ΔCC was not significantly different from zero for any of the 3 groups of movie images (P > 0.3, Wilcoxon signed-rank test, n = 13, Fig. 6E). Thus, the movie-induced RF modifications depend on the activation of NMDA receptors.
Selective Increase in Spike Response to Movie Images
Since the RF became more similar to the S+ images, we wondered whether repetitive stimulation with natural movies could enhance the image selectivity of the neuron. We thus examined the firing rate of the neurons over the 5 blocks of movie conditioning (see Materials and Methods). Because S+ images that caused depolarization often evoked high firing rates, we divided the time bins of the movie-evoked PSTH into event bins (Fig. 7A, red) and non-event bins (Fig. 7A, blue) (see Materials and Methods). Presumably, the event spikes were evoked by the S+ images that the neuron preferred and the non-event spikes by the non-S+ images. Over the population (n = 48), we found a marked increase in the number of event spikes over the 5 blocks of movie stimulation, but no significant change in the spikes of non-events (Fig. 7B). Thus, consistent with the increased similarity between the spike RF and the S+ images (Fig. 3D, right), repetitive movie stimulation induced a preferential increase in the responses to the preferred images, resulting in enhanced stimulus selectivity of the neurons. When we applied APV to the visual cortex, neither event spikes nor non-event spikes showed significant change over time (n = 20, Fig. 7C), further suggesting that the movie-induced modifications require the action of NMDA receptors.
We have shown that repetitive movie stimulation caused an increase in similarity between the RF structure and the subset of movie images that depolarized the cell. This RF change is consistent with the Hebbian synaptic learning rule (Hebb 1949), which can account for several forms of cortical RF modification induced by visual stimulation (Eysel et al. 1998; Eyding et al. 2002; Fu et al. 2002). Whereas several previous studies of RF plasticity have controlled the precise timing of spikes or manipulated the postsynaptic activity in response to sensory stimulation (Frégnac et al. 1992; Shulz and Frégnac 1992; Cruikshank and Weinberger 1996; Yao and Dan 2001; Fu et al. 2002; Yao et al. 2004; Meliza and Dan 2006; Dahmen et al. 2008), the RF modification in the current study was induced under a more natural stimulation condition. The movie-induced RF modification may contribute to the experience-dependent enhancement in stimulus selectivity of cortical neurons.
To examine whether the Hebb's rule can account for the RF modification, we constructed a Hebbian neural network model to perform further analysis (Supplementary Material). The model consisted of a single postsynaptic neuron driven by 32 × 32 presynaptic neurons (Supplementary Fig. S6A), and synaptic modifications were implemented on the basis of the correlation between pre- and post-synaptic spiking activity (Dayan and Abbott 2001). We simulated a total of 43 postsynaptic neurons, and set each RFbefore to be those measured from the cells in Figure 3D. We found that the model could predict the RF changes observed in the experiment (Supplementary Fig. S6C–E), supporting the notion that RF modification is mediated by Hebbian synaptic plasticity.
We found that RF modification depended on the activation of NMDA receptors, further suggesting activity-dependent synaptic modification as the underlying synaptic mechanism. Because the APV was locally applied to the cortical surface, the synapses modified by movie stimulation were likely to be the intracortical connections. Intracortical connections have been shown to contribute significantly to shaping cortical RFs (Gilbert 1998; Fu et al. 2002), and various forms of plasticity have been demonstrated at synapses of intracortical connections in visual cortical slices (Hirsch and Gilbert 1993; Sjostrom et al. 2001; Froemke and Dan 2002). Previous studies showed that stimulus-timing-dependent orientation plasticity can be transferred interocularly (Schuett et al. 2001; Yao et al. 2004), suggesting that these effects are mainly mediated by intracortical connections. It is of interest for future study to further investigate whether the sites of movie-induced synaptic modifications occur mainly at intracortical connections or at the feedforward thalamic connections.
The movie-induced change in RF spatial profile resulted from a combination of enhancement and reduction in the amplitudes of local synaptic responses at different RF locations. A similar effect has been reported in previous studies of RF plasticity in the rat visual cortex (Meliza and Dan 2006) and tadpole tectum (Vislay-Meltzer et al. 2006), cortical frequency tuning in the auditory system (Dahmen et al. 2008), and the whisker-evoked responses in the barrel cortex (Jacob et al. 2007). The effects observed in these studies may reflect the spread of long-term potentiation or long-term depression to neighboring synapses (Engert and Bonhoeffer 1997; Fitzsimonds et al. 1997; Tao et al. 2001) as well as the overlapping populations of presynaptic neurons activated by the sensory stimulus. In our study, the heterogeneous changes in the amplitude of local synaptic responses may be due to the complex spatiotemporal pattern of movie images. Assuming that these images activated converging synaptic inputs to the postsynaptic neuron, the excitatory inputs that were associated with the depolarization and spiking of the postsynaptic neuron may be potentiated, and those that were not correlated with the postsynaptic depolarization may be depressed.
We showed that manipulating the statistics of natural scenes movies by temporal shuffling (spatial whitening) or using white noise as conditioning stimuli could diminish the RF modification. Responses to whitened movies showed lower firing rates and lower reliability compared with those to natural movie stimulation. Thus, the failure of whitened movies to induce RF changes may simply be due to its low efficacy in driving the cortical neurons. While the analysis of firing rate and response reliability demonstrated that shuffled movies (or white noise) could drive the cells as well as natural movies, the sparseness of spiking response to natural movies was significantly higher than that to shuffled movies (or white noise). Furthermore, the sparseness of movie-evoked response was significantly correlated with the degree of RF modification, suggesting that response sparseness may play an important role in inducing RF modification. Our result also suggests that the sparse cortical response evoked by natural stimulation (Vinje and Gallant 2000; Haider et al. 2010) may contribute to the mechanism underlying the adaptation of the visual system to efficient coding of natural scenes (Attneave 1954; Barlow 1961; Simoncelli and Olshausen 2001).
This work was supported by 973 program (2011CBA00405) and the Hundred Talents Program of the Chinese Academy of Sciences.
We thank Y. Dan for discussions and critical comments on the manuscript. We also thank H. Zhong and W. Xu for their technical assistance. Conflict of Interest: None declared.