Research Article

Temporal Specificity in the Cortical Plasticity of Visual Space Representation

See allHide authors and affiliations

Science  14 Jun 2002:
Vol. 296, Issue 5575, pp. 1999-2003
DOI: 10.1126/science.1070521


The circuitry and function of mammalian visual cortex are shaped by patterns of visual stimuli, a plasticity likely mediated by synaptic modifications. In the adult cat, asynchronous visual stimuli in two adjacent retinal regions controlled the relative spike timing of two groups of cortical neurons with high precision. This asynchronous pairing induced rapid modifications of intracortical connections and shifts in receptive fields. These changes depended on the temporal order and interval between visual stimuli in a manner consistent with spike timing–dependent synaptic plasticity. Parallel to the cortical modifications found in the cat, such asynchronous visual stimuli also induced shifts in human spatial perception.

Persistent synaptic modification can be induced by repetitive pairing of pre- and postsynaptic spikes, and the direction and magnitude of the modification depend on the relative spike timing (1–6). Presynaptic spiking within 10s of milliseconds before postsynaptic spiking induces synaptic potentiation, whereas the reverse order results in synaptic depression. Spike timing–dependent plasticity has been widely observed at excitatory synapses in vitro, and its functional significance has been discussed in a variety of neuronal circuits (7–10). However, whether and how this mechanism operates in vivo in response to sensory stimuli is only beginning to be examined experimentally (11–13). In the primary visual cortex, timing of visual stimuli can directly affect timing of neuronal spiking and thus, may play a role in activity-dependent cortical plasticity.

A simple working hypothesis of how relative timing of visual stimuli can affect cortical representation of visual space is illustrated inFig. 1. Asynchronous stimuli in retinal regions A and B may evoke asynchronous spiking in two groups of cortical neurons, a and b (Fig. 1, A and B). According to spike timing–dependent synaptic plasticity, AB stimulation may strengthenab connections and weakenba connections (Fig. 1C, left). Because intracortical connections contribute significantly to shaping the cortical receptive fields (14), the synaptic modifications induced by AB stimuli may cause the receptive fields of both groups to shift toward A(15) (Fig. 1D, left). In a typical population-decoding scheme, the perceived stimulus location is determined by the spatial profile of population neuronal response (16). The leftward shift (toward A) of receptive fields should cause a rightward shift in the response profile, hence a rightward shift in perception (Fig. 1E, left). By symmetry, BAstimuli that activate group b before a should cause the opposite changes at all levels (Fig. 1, right column).

Figure 1

A hypothetical scheme in which stimulus timing affects cortical modification. (A) Intracortical excitatory connections between two groups of neurons, a andb, whose receptive fields fall in A andB, respectively. Blue, group a; red, groupb. Arrows indicate directions of axonal projections. Only a subset of connections between a and b are shown; connections within each group are omitted. (B) Relative spike timing of groups a and b in response toAB and BA stimuli. (C) Strength of connections between a andb (represented by line thickness) after conditioning, obtained from a simulation with 500 pairs ofAB (left) or BA(right) conditioning (17). (D) Receptive fields after conditioning. Arrows indicate directions of shifts. (E) Spatial profiles of the population responses evoked by a test stimulus (vertical line) at A/B border before (solid curve) and after (dashed) conditioning.

Asynchronous Visual Conditioning

Receptive fields of primary visual cortical neurons recorded in anesthetized adult cats were mapped with strips of drifting gratings (Fig. 2A) (17). During visual conditioning, random spatial patterns were flashed asynchronously in two adjacent retinal regions (A andB, with receptive field of the recorded neuron inB) (Fig. 2B) to manipulate the relative spike timing of two groups of cortical neurons (a and b). In order for these stimuli to be effective in cortical modification, the precision of spike timing in the responses must be commensurate with the temporal requirement of spike timing–dependent synaptic plasticity on the order of 10s of milliseconds. We measured the spike timing precision by comparing the responses of each neuron toAB and BA stimuli. Spiking of a group b neuron during ABstimuli (A/B interval: 8.3 ms) lagged behind its spiking during BA stimuli by ∼8 ms, indicating time-locking of the spikes to the flashed stimuli (Fig. 2C). The precision of spike timing is also reflected in the cross-correlation between the responses to ABand BA stimuli (Fig. 2D), which exhibited a distinct peak at ∼8 ms, corresponding to theA/B interval. Such correspondence was also found in other experiments in which the A/B intervals ranged from 0 to 66.7 ms. The mean width at half height of the peak, which reflects spike timing precision, was 26 ± 15 ms (SD) for the 145 neurons examined (Fig. 2E). With such precision, A/Binterval in the conditioning stimuli can effectively determine the relative spike timing between neuronal groups a andb.

Figure 2

Spike timing of cortical neurons in response to asynchronous conditioning stimuli. (A) Receptive field of a neuron (upper panel) and the mapping stimuli (lower panel). Error bars, ±SEM. Solid lines in lower panel delineate regions A andB; dotted lines indicate strip positions. (B) Two types of conditioning stimuli, AB andBA. The spatial patterns were either random checkerboards (8 × 8 pixels/region, in experiments shown in Fig. 4) or random bars at preferred orientation of the cell (16 bars/region, in experiments shown in Fig. 5) at 100% contrast. Size: 3.3 × 1.8° to 12.4 × 11.5°/region. In each conditioning pairA and B were each stimulated for 1 frame (8.3 ms); A/B interval (−8 to 8 frames) and resting period (11 to 12 frames) were blank screens (<1 cd m−2). Dotted lines delineating A andB were not part of the stimuli. Average cortical firing rate during conditioning: 2.9 ± 1.5 spikes/s (SD, n = 123). (C) Responses of the neuron in (A) evoked by the stimuli in (B). (Upper) Spiking in response toAB (red) and BA(blue) conditioning (600 consecutive pairs each) at an 8.3-ms interval. (Lower) Post–stimulus time histograms of the responses. For this cell, the duration of the response to ABstimuli is slightly longer than that to BAstimuli. However, for a population of cells analyzed (n= 90), we found no significant difference in duration between the responses to AB andBA stimuli (P > 0.40,t test). (D) Cross-correlation between spike trains evoked by AB andBA stimuli. (E) Distribution of spike timing precision of cortical neurons, measured by the width at half height of the peak in the cross-correlogram. Arrow indicates the mean.

Excitatory Intracortical Connections

To monitor intracortical synaptic modification, we applied cross-correlation analysis of simultaneously recorded cortical cell pairs (18–23). First, we analyzed the relationship between synaptic modifications and changes in intracortical cross-correlation using the model circuit depicted inFig. 1A (17). As a result of spike timing–dependent synaptic plasticity, repetitiveAB conditioning potentiatedab connections and depressedba connections, whereasBA conditioning induced the opposite changes. Accompanying these synaptic modifications, two types of changes were observed in the cross-correlation between the spike trains of model neurons. First, for intragroup cross-correlation [bi , bj ], conditioning induced changes in correlation strength (measured by the area under the peak, Fig. 3A), reflecting modifications of the common inputs from group a neurons tobi and bj (see Fig. 3A legend). Second, for intergroup cross-correlation [ai , bj ], conditioning induced shifts in correlation asymmetry (24) (Fig. 3B), reflecting modifications of both ab andba connections (see legends).

Figure 3

Relationship between intracortical synaptic modifications and changes in cross-correlation as revealed by simulation. (A) Shuffle-subtracted cross-correlograms (black) between a pair of group b neurons (bi and bj ) during control (middle), after AB (left) and afterBA (right) conditioning. Blue lines, correlation due to monosynaptic common inputs from group aneurons to bi and bj (Fig. 1, A and C), obtained by eliminating all connections in the circuit except those from each group a neuron tobi or bj . (B) Shuffle-subtracted cross-correlograms between an intergroup pair,ai and bj . Red lines, correlation due to mono- and di-synapticbj ai connections, obtained by eliminating all connections except those originating from bj and those projecting toai . Blue lines, correlation due to mono- and di-synaptic ai bj connections.

In the first set of conditioning experiments, we measured changes in correlation strength between intragroup cell pairs (defined as groupb). Shuffle-subtracted, flank-normalized cross-correlograms (17) were computed from the responses to mapping stimuli (Fig. 2A) before and after conditioning (Fig. 2B). In both examples shown in Fig. 4A, 1200 pairs ofAB conditioning at 8.3-ms interval (∼2 min) induced a marked increase in correlation strength (dashed line), indicating potentiation of ab connections (Fig. 3A, left), whereas the same amount of BAconditioning induced a decrease in correlation strength (Fig. 4A, dotted line), indicating depression of the connections (Fig. 3A, right). For the 138 cell pairs examined, 1200 pairs of conditioning at 8.3-ms interval (AB) induced an increase of 29.8 ± 4.2% (SEM, P < 10−5,t test), and conditioning at −8.3-ms interval (BA) induced an 18.2 ± 3.1% reduction (P < 10−5). We further measured the effect of conditioning at A/B intervals ranging from –66.7 to 66.7 ms and found significant changes only within ±50 ms (Fig. 4B). To examine the persistence of the change induced by 1200 pairs of AB conditioning (at the 8.3-ms interval) we measured the cross-correlation continuously in the absence of further conditioning. The change in correlation strength was found to persist for up to 8 min (Fig. 4C).

Figure 4

Conditioning-induced changes in intracortical cross-correlation. (A) Two examples of changes in intragroup correlation strength. Shuffle-subtracted, flank-normalized cross-correlograms (smoothed with 1.5-ms boxcar convolution) are shown during control (solid) and afterAB (dashed) and BAconditioning (dotted). The effects were significant for bothAB (P < 10−5 for both examples) and BA (P < 0.05 and P < 10−5 for first and second examples, respectively) conditioning, as assessed by nonparametric bootstrap (42). (Insets) Receptive fields of the pair; dashed line, A/B border. The gray curve was shifted up for clarity. Error bars, ±SEM. (B) Changes in correlation strength as a function of A/Binterval. Error bars, ±SEM (n = 67 to 138). Asterisks indicate data points significantly different from 0 (*P< 0.05; **P < 0.01; ***P < 0.005;t test). (C) Persistence of the effect (n = 89). Time 0 is defined as the end of conditioning. (D) Two examples of changes in intergroup correlation asymmetry. Cross-correlograms are shown afterAB (dashed) andBA conditioning (dotted). The difference in asymmetry after the two types of conditioning was significant (P < 5 × 10−6 and P< 0.05 for the first and second examples, respectively). Control cross-correlogram was omitted for clarity. (E) Shifts in correlation asymmetry as a function of A/Binterval (n = 28 to 58). The effects ofAB and BAconditioning were combined as 1/2 × [asymmetry afterAB (dashed) − asymmetry afterBA (dotted)], and the left side of the curve (dashed) was plotted as the exact opposite of the right. (F) Persistence of the shift (n = 34).

In a parallel set of conditioning experiments, we measured shifts in the asymmetry of cross-correlation between group a and group b neurons. In both examples shown in Fig. 4D, the shuffle-subtracted cross-correlograms measured afterAB (dashed) and BA(dotted) conditioning at an 8.3-ms interval were laterally displaced from each other. The rightward shift of the peak afterAB conditioning suggests strengthening ofab connections or weakening ofba connections (Fig. 3B, left), whereas the leftward shift after BA conditioning suggests the opposite changes (Fig. 3B, right). Because the effect ofBA conditioning on correlation [ai , bj ] is equivalent to the effect of AB conditioning on correlation [bj , ai ] and because for each pair the designation of a andb is arbitrary, the measured effects ofAB and BAconditioning were combined (see Fig. 4E legend). For the 58 cell pairs examined, 1200 pairs of conditioning stimuli at an 8.3-ms interval induced a 4.9 ± 1.5% (SEM, P < 0.01,t test) shift in correlation asymmetry in the predicted direction. We further measured the shift as a function ofA/B interval from 0 to 66.7 ms (Fig. 4E) and found the effect restricted to a window of ∼50 ms. Thus, shifts in the intergroup correlation asymmetry also depend on theA/B interval in a manner consistent with spike timing–dependent synaptic modification. In addition, we examined the persistence of the shift induced by 1200 pairs of conditioning stimuli (at an 8.3-ms interval) and found the effect to last for several minutes (Fig. 4F). This is similar to the persistence of conditioning-induced change in intragroup correlation strength (Fig. 4C), consistent with the idea that both types of changes in cross-correlation reflect the same set of intracortical synaptic modifications.

Cortical Receptive Fields

Conditioning-induced intracortical synaptic modifications may lead to shifts in receptive fields (Fig. 1D). We thus compared the receptive fields of each neuron mapped before and after conditioning (17). For both group b neurons shown inFig. 5A, 800 pairs (∼1.5 min) ofAB conditioning at an 8.3-ms interval induced a shift of the receptive fields toward A, whereasBA conditioning induced a shift away fromA. To quantify this effect, we fitted each receptive field with a Gaussian function and computed the difference between the peak positions of the Gaussian fits before and after conditioning (normalized by the width at half height). For the 154 groupb neurons examined, 800 pairs of ABconditioning (8.3-ms interval) induced a 1.86 ± 0.53% (SEM,P < 0.001, t test) shift towardA, whereas 800 pairs of BAconditioning (−8.3 ms) induced a 1.69 ± 0.53% (P < 0.002) shift away from A. We further measured the effect of conditioning as a function ofA/B interval from −33.3 to 33.3 ms and found significant shifts only at 8.3 and –8.3 ms (Fig. 5B), indicating temporal specificity. Thus, the direction and magnitude of shift in cortical receptive field depend on the order and interval betweenA and B, consistent with the prediction shown inFig. 1D. In addition to the positions, we also compared the receptive field sizes (widths of the Gaussian fits) before and after conditioning. Conditioning induced an overall reduction of receptive field size, but the effect exhibited no systematic dependence onA/B interval (Fig. 5C).

Figure 5

Conditioning-induced shifts in cortical receptive fields. (A) Results from two example group bneurons. Receptive fields are shown during control (solid) and afterAB (dashed) and BA(dotted) conditioning. Arrow indicates peak position of the Gaussian fit. (B) Normalized receptive field shift as a function of A/B interval (n = 82 to 154). Positive shift: toward A. Error bars, ±SEM. Asterisks indicate data points significantly different from 0 (***P < 0.005; t test). (C) Change in receptive field size as a function ofA/B interval, measured from the same cells as in (B). The effects were not significantly different betweenAB and BAconditioning at each interval (P > 0.15,t test).

Spatial Perception

To test whether the asynchronous conditioning can induce shift in perceptual localization (Fig. 1E), we also performed psychophysical experiments on human subjects. The conditioning stimuli were similar to those used in the physiological experiments, and a three-bar bisection test (25) was performed at the A/Bborder (Fig. 6A) to measure the perceived position of the middle bar before and after conditioning (17). For all four subjects tested, we found that 400 pairs (∼50 s) of AB conditioning induced a shift in the perceived location of the middle bar towardB, whereas BA conditioning induced a shift toward A, consistent with the prediction shown inFig. 1E. Furthermore, a significant effect was observed only atA/B intervals within ±20 ms (Fig. 6B), satisfying the temporal specificity required by the model. To further characterize the conditioning-induced perceptual shift, we measured the time course for both the induction and decay of the effect. At intervals of 8.3 ms (AB) and –8.3 ms (BA), significant shift was induced after only 100 to 200 pairs of conditioning stimuli (Fig. 6C), indicating rapid onset of the effect. After 400 pairs of conditioning stimuli, the effect showed a slight but noticeable decay over a 2-min period (Fig. 6D), suggesting that the persistence of the perceptual effect is also on the order of minutes, comparable to that observed for the intracortical synaptic modifications (Fig. 4, C and F).

Figure 6

Conditioning-induced shifts in human perceptual localization. (A) Test block with three-bar bisection task. Dotted lines delineating A and B were not part of the stimuli. (B) Shift in perceptual localization as a function of A/B interval for four subjects (400 conditioning pairs at each interval). Positive Δthreshold: towardA (shift of percept toward B). Error bars, ±SEM (n = 5 to 17). Asterisks indicate data points significantly different from 0 (*P < 0.05; **P < 0.01; ***P < 0.005;t test). YF and YD, authors; HY, experienced observer; QT, naı̈ve subject. (C) Dependence of perceptual shift on the amount of conditioning at ±8.3-ms intervals. Data points represent mean shifts (±SEM) of YF, YD, and HY. The data of QT was not included because no significant effect was observed at ±8.3 ms. (D) Persistence of the shift induced by 400 conditioning pairs. Data points represent mean shifts of YD and YF (persistence was not measured for HY and QT).


Asynchronous visual stimuli in different retinal regions induced rapid changes in cortical representation of visual space, which was likely mediated by spike timing–dependent modification of intracortical connections. Previous studies have shown that visual stimuli can induce various changes in adult cortical neurons. For example, in contrast adaptation, a few seconds of visual stimulation in the receptive field can induce a significant reduction in the response amplitude (26), along with changes in receptive field size and position (27). These effects may be due to reduction in excitability of the cortical neurons (28, 29) or homosynaptic short-term depression of the thalamic inputs (30). Synchronous stimulation of the receptive field center and part of the unresponsive surround can induce preferential expansion of the receptive field toward the costimulated surround region (31), which is likely mediated by Hebbian modification of intracortical connections. Visual conditioning with an artificial scotoma can also induce a rapid expansion of cortical receptive field inside the scotoma (32) and a shift in perceptual localization near the edge (25). The underlying mechanism is thought to be the strengthening of excitatory intracortical connections (22), although an alternative explanation based on contrast adaptation has also been proposed (33). In our study although mechanisms related to contrast adaptation may have caused the conditioning-induced reduction in receptive field size (Fig. 5C), they cannot account for the shift in receptive field position because the effect depends critically on the order and interval between stimuli in Aand B (Fig. 5B). Instead, spike timing–dependent modification of intracortical connections (1, 3,5, 6) provides the most natural explanation. This form of synaptic plasticity is also believed to underlie conditioning-induced changes in cortical orientation tuning in both developing (13) and adult (12) animals.

In vitro, activity-dependent synaptic modification can last for hours (34, 35). The ongoing spiking activity in vivo, however, may continuously modulate synaptic connections and obliterate the effects of conditioning over several minutes (Fig. 4, C and F). In addition, a reduction in modulatory inputs (e.g., those from the basal forebrain) caused by anesthesia may further limit the extent and persistence of conditioning-induced cortical modifications (36). Persistence on the order of several minutes to hours has been observed for other adult cortical modifications in vivo, including shifts in cortical receptive fields (31) and orientation tuning (12) induced by visual conditioning and modifications of intracortical connections induced by auditory conditioning (20). Note that the persistence of synaptic modification in the auditory cortex appears to depend on the duration of conditioning (20). Although ∼2 min of conditioning induced a short-term modification in the visual cortex in our study, longer conditioning may cause more lasting changes.

The simple model depicted in Fig. 1 considers only spike timing–dependent modification of excitatory intracortical connections. Changes in inhibitory connections were not analyzed because they are difficult to measure with cross-correlation analysis (19). In the mammalian central nervous system, activity-induced modification of inhibitory synapses appears to be insensitive to the order of pre- and postsynaptic spiking (37), thus it may not contribute directly to the asymmetric stimulus-timing dependence of the cortical modifications observed here. Another set of excitatory connections not considered in the model are “feedforward,” thalamic inputs. In principle, the asynchronous conditioning stimuli may also modify the thalamic inputs, and such modifications may cause changes in the strength of correlation between intragroup cortical cell pairs (Fig. 4, A to C) and shifts in cortical receptive fields (Fig. 5). However, modification of the thalamic inputs alone cannot account for the observed shifts in intergroup correlation asymmetry (Fig. 4, D to F), as shown by simulation studies (38). Thus, the cortical modifications observed in this study are likely to result at least partly from spike timing–dependent plasticity of excitatory intracortical connections. In particular, horizontal connections in layer 2/3 of the primary visual cortex play an important role in experience-dependent cortical reorganization during development (39), and spike timing–dependent modification of these connections is a robust phenomenon in vitro (6).

Spike timing on the order of 10s of milliseconds plays an important role in both coding of visual information (40,41) and modification of neuronal circuitry (1–6). We found that spiking of the cortical neurons was time-locked to the flashed visual stimuli with a precision of ∼30 ms (Fig. 2, C to E), allowing timing of visual stimuli on the order of 10s of milliseconds to be represented by spike timing in the cortical circuit. The marked dependence of cortical modifications on stimulus timing, observed at the levels of intracortical synaptic connectivity, receptive field properties, and human visual perception, attests to the critical role of stimulus timing in activity-dependent plasticity of adult visual cortex.

  • * These authors contributed equally to the work.

  • To whom correspondence should be addressed. E-mail: ydan{at}


View Abstract

Stay Connected to Science

Navigate This Article