 |
||||
|
||||
|
||||
|
||||
The Journal of Neuroscience, August 20, 2003, 23(20):7690-7701
Previous Article | Next Article
Time Course and Time-Distance Relationships for Surround Suppression in Macaque V1 Neurons
Wyeth Bair,1,2 James R. Cavanaugh,2 andJ. Anthony Movshon1,2 1Howard Hughes Medical Institute and 2Center for Neural Science, New York University, New York, New York 10003
Abstract
Top
Abstract
Introduction
Iso-orientation surround suppression is a powerful form of visual contextual modulation in which a stimulus of the preferred orientation of a neuron placed outside the classical receptive field (CRF) of the neuron suppresses the response to stimuli within the CRF. This suppression is most often attributed to orientation-tuned signals that propagate laterally across the cortex, activating local inhibition. By studying the temporal properties of surround suppression, we have uncovered characteristics that challenge standard notions of surround suppression. We found that the latency of suppression depended on its strength. Across cells, strong suppression arrived on average 30 msec earlier than weak suppression, and suppression sometimes arrived faster than the excitatory CRF response. We compared the relative latency of CRF response onset and offset with the relative latency of suppression onset and offset. Response onset was delayed relative to response offset in the CRF but not in the surround. This is not the expected result if neurons targeted by suppression are like those that generate it. We examined the time course of suppression as a function of distance of the surround stimulus from the CRF and found that suppression was predominantly sustained for nearby stimuli and predominantly transient for distant stimuli. By comparing the latency of suppression for nearby and distant stimuli, we found that orientation-tuned suppression could effectively propagate across 6 - 8 mm of cortex at1 m/sec. This is considerably faster than expected for horizontal cortical connections previously implicated in surround suppression. We offer refinements to circuits for surround suppression that account for these results and describe how feedback from cells with large CRFs can account for the rapid propagation of suppression within V1.
Key words: macaque monkey; primary visual cortex; surround suppression; contextual modulation; response latency; propagation velocity; cortical feedback
Introduction
Receptive fields of neurons in the visual cortex can be decomposed into a classical receptive field (CRF), in which stimuli directly elicit a neuronal response, and an area beyond the CRF, in which stimuli elicit no response of their own but can profoundly modulate a CRF-driven response (for review, see Allman et al., 1985; Gilbert, 1992
Here we study the time course of orientation-tuned surround suppression (Blakemore and Tobin, 1972
To further resolve mechanisms of surround suppression, we considered the implications of a simple circuit (Fig. 1) for the timing of suppression. Here, an orientation-tuned target neuron (triangle at left) is suppressed by an inhibitory neuron (black circle) that is driven by similarly tuned neurons displaced laterally in the cortex (triangles at right). We examined two aspects of timing for this circuit. First, we compared the timing of response onsets and offsets for CRF stimuli with that for surround stimuli. We recently reported that response onset latency was consistently longer than offset latency (in >95% of cells and by 10-20 msec on average) for cortical neurons driven by dynamic CRF stimuli (Bair et al., 2002
View larger version (22K):
[in this window]
[in a new window]
Figure 1. Putative circuit for iso-orientation surround suppression in V1. The target neuron (left triangle) receives inhibition from a nearby neuron (large filled circle) that is driven by excitatory neurons displaced laterally in cortex (3 triangles at right) that have orientation tuning similar to the target cell. Preferred stimuli confined to the CRFs for three excitatory neurons are indicated by the circular patches of sinusoidal grating shown on the tilted plane that represents a two-dimensional visual field. Arrows emanating from the stimuli indicate localized feedforward inputs to the excitatory neurons. We find that this circuit is insufficient to account for some temporal features of surround suppression.
Materials and Methods
Electrophysiology. We recorded extracellularly from primary visual cortex of anesthetized, paralyzed macaque monkeys (two Macaca nemestrina and eight M. fascicularis). Detailed methods for this type of recording are available in an article by Levitt et al. (1994Tungsten-in-glass microelectrodes (Merrill and Ainsworth, 1972
At the end of experiments, animals were perfused with 4% paraformaldehyde in saline. Cortical sections of 40 µm were mounted on slides and stained for Nissl substance with cresyl violet. We reconstructed electrode tracks by locating the electrolytic lesions and by visualizing tissue damage from the passage of the electrode. We determined the laminar location of each cell on the basis of visual inspection of landmarks in the stained sections, as described by Cavanaugh et al. (2002
Visual stimuli. Visual stimuli were generated by custom software on a Cambridge Research Systems (Kent, UK) 2/2 board and presented on a standard cathode ray tube (CRT) at 100 Hz vertical refresh with mean luminance of 33 cd/m 2. The CRT was placed farther from the monkey's eye for smaller RFs and closer for larger RFs (range, 80-180 cm). For each cell whose action potential waveform was well isolated from the noise, we systematically optimized the orientation, spatial period, and temporal frequency of a sine wave grating at 100% contrast that drifted for 2-4 sec behind a circular aperture chosen to approximate the CRF. We then used the smallest patch of optimal grating that provided a reliable response to find the center of the RF. Next, we presented stimuli in an interleaved manner consisting of either a single circular patch of variable diameter or a single annular patch of variable inner diameter For annuli, the outer diameter was set to the maximum extent of our screen, which ranged from 10 to 20° of visual angle, depending on the screen placement. We defined the CRF region to be the smallest circular patch that produced at least 95% of the maximum response (for a detailed description, see Cavanaugh et al., 2002
Our dynamic center-surround stimulus consisted of two patches of drifting grating that were presented in the CRF and surround apertures defined above. The stimulus consisted of a random sequence of states (Fig. 2C) in which each grating drifted in either the preferred direction or in an orthogonal direction for one full cycle of sinusoidal modulation. At the end of each state, each grating was randomly and independently assigned to have either the preferred or orthogonal direction for the next state. Thus, at each state boundary (i.e., after each full cycle of drift), each grating either drifted seamlessly or made a 90° change in orientation and direction. Opportunities for state transitions occurred simultaneously for the CRF and surround. The two possible directions of motion for the CRF and surround created four possible states and 16 state transitions, which occurred at random. The state transitions of interest for this study are labeled above the icons in Figure 2C. When describing states and state transitions, we will refer to orientation and direction as "preferred" when they are the same as those preferred by the CRF, even when referring to stimuli that are located in the surround. Likewise, "orthogonal" always means at 90° to preferred. Thus, a stimulus in the surround that has the orientation preferred by the CRF is called preferred even though it is a suppressive stimulus.
View larger version (74K):
[in this window]
[in a new window]
Figure 2. Tuning curves for orientation and size were used to configure our dynamic center-surround stimulus. A, The mean firing rate of an example neuron (3 trials, 4 sec/trial) is plotted against the drift direction (in degrees, relative to preferred) of a sinusoidal grating. Motion was always orthogonal to the orientation of the grating; therefore, we refer to this as an orientation tuning curve. Error bars indicate SEM. The stimulus of preferred orientation is shown by the icon above the peak at zero. The spontaneous rate for this neuron was zero. B, A size tuning curve (filled circles) shows mean response versus stimulus diameter for the same neuron. The optimal size (small icon) was smaller than that in A, which accounts for larger responses here. The annulus tuning curve (open circles) shows the response to gratings presented in an annular window versus the inner diameter of the annulus. The outer diameter was set to fill our screen. The CRF region (small icon) was the disk that optimized the size tuning curve. The surround region (large icon) was the annulus with the smallest inner diameter that did not elicit a response above the spontaneous firing rate. The arrow marks the inner diameter for the surround region. Icons here and in A are drawn to scale. C, The icons depict a succession of six states of the dynamic center-surround stimulus (see Materials and Methods). The gratings in the CRF (center disk) and surround (annulus) drifted in either the preferred direction (shown as vertical) or in an orthogonal direction (horizontal). For clarity, the surround annulus outer diameter is shown here at 50% of the size determined in B. Below the icons, orientation is plotted versus time for the CRF (black line) and surround (gray line). The state transitions of interest here are labeled above the bent arrows. This sequence was chosen for convenient demonstration; the actual sequence of transitions was random.
The dynamic stimulus was presented in trials lasting 30 sec and was interleaved with a center-alone and a surround-alone control in which the contrast of the grating in the other aperture was set to zero. There were 4 sec of mean gray between trials. If the surround-alone stimulus yielded a substantial response to the transition from orthogonal to preferred, the inner diameter of the annulus was increased to eliminate the response. The temporal frequency of the grating was sometimes increased from the optimum (if the optimum was <4 Hz and if doing so did not cause the firing rate to drop to <90% of optimal) because this increased the number of times that each state transition was tested. We used temporal frequencies ranging from 3.1 to 25 Hz, but most cells were tested at either 6.25 or 12.5 Hz. The spatial phase of the surround grating was set to match that of the center grating. In two cells, we varied the phase of the surround grating and found that it did not affect our results. This is consistent with reports that surround suppression is primarily phase-insensitive (Nelson and Frost, 1978
To test the dependence of surround suppression on the distance of the surround stimulus from the CRF, we designed a second, simpler experiment that allowed us to focus on the timing of the onset of suppression. In this experiment, the optimal drifting grating for the CRF appeared on a mean gray background for 1 sec, and a surround stimulus appeared 400 msec later and lasted for 300 msec. Six stimuli were interleaved in a block-wise random manner with 2 sec of mean gray between stimuli. Five of the six stimuli had surrounds, which extended from the outer border of our screen to an inner diameter that varied across stimuli, and the sixth stimulus had no surround. This allowed us to study the time course of suppression for stimuli at various distances from the CRF with reference to the time course of the response to the CRF stimulus when no surround stimulus was imposed.
Data analysis. The response to a particular transition of the dynamic stimulus was computed by averaging together segments of spike trains (1 msec resolution) shortly before and after each occurrence of the transition. The spike trains were aligned to the time of the state transition before averaging, and the resulting average is referred to as a peristimulus time histogram (PSTH). PSTHs were filtered in time with a Gaussian of an SD of 2 msec and plotted in units of spikes per second. Responses to the stimuli in which a surround was flashed on were computed in a similar manner, with the onset of the stimulus taken as the temporal reference.
We computed the latency of response to a change in the visual stimulus (e.g., a state transition or the appearance of a surround stimulus) by first subtracting the reference response (the PSTH in response to no change in the stimulus) from the response PSTH for the change. We then identified the maximum of the absolute value of this response difference, and we searched backward in time from the maximum to the point at which the difference was equal to 5% of the maximum. The time of this point was defined as the response latency. Thus, latencies were the times of 5% rise to the peak response difference. This is the same method that we used previously (Bair et al., 2002
We compared our method with one that chooses the latency to be the point at which the response difference reaches a criterion statistical significance. Because the latter requires the response to reach a fixed level, the latency estimate will grow as the change in response is scaled down. This is undesirable because later we show that, on average, the strength of suppression is approximately scaled down for far surround stimuli (see Results). We tested both latency methods on PSTHs for simulated Poisson spike trains in which the firing rate changed at a fixed time from an initial level to a final level in 20 msec (which was typical for our data). As expected, the 5% latency method was substantially less dependent on the final rate than was the statistical criterion method for both rate increases and decreases. For changes in the rate approximating the smallest in our data, even the 5% latency method began to show an increased bias (up to 4 msec longer than the true latency). However, such small rate changes occurred only for the most distant surround stimuli, for which the bias operates in a conservative manner with respect to our conclusions (see Discussion).
We computed the suppression strength, 1 - rs/rp, for responses to the dynamic center-surround stimulus from the firing rate (rp) during the preferred stimulus (preferred CRF, orthogonal surround) and the rate during suppression (rs; preferred CRF, preferred surround; recall that, in the surround, the orientation preferred by the CRF is typically the most suppressive). Firing rates were computed in a time window equal to one period (drift cycle) of the stimulus. This window was centered in an epoch twice as long, which corresponded to two consecutive periods of the pertinent stimulus. This ensured that the response was measured in an epoch that began after the signals arrived in the cortex (i.e., it accounted for the visual latency).
The spontaneous firing rate was computed in 500 msec epochs preceding the onset of each trial of the dynamic center-surround stimulus.
We computed a cortical magnification factor (CMF) using the same equation that we used previously (Cavanaugh et al., 2002
where E is the receptive field eccentricity in degrees (Van Essen et al., 1984
(1)
Results
Temporal characteristics of surround suppression
By definition, stimuli outside the CRF do not elicit responses; therefore, we used stimuli in the CRF to evoke responses that the surround could modulate. We chose drifting (rather than static) sinusoidal gratings because they maximized the sustained portion of the CRF-evoked response. This worked well for complex cells, which have relatively unmodulated responses to drifting gratings. Simple cells, however, responded with strong modulation at the fundamental temporal frequency, obscuring the temporal profile of surround suppression and preventing the measurement of response latencies. Therefore, we present results for complex cells only. There is strong evidence that surround suppression operates similarly for simple and complex cells (Dreher, 1972We studied surround suppression in 92 orientation-tuned complex cells. A typical orientation tuning curve is shown for an example neuron in Figure 2A, where vertically and horizontally oriented icons depict the preferred and orthogonal stimuli, respectively. After quantitatively optimizing the orientation, direction, and spatial and temporal frequencies of a circular patch of drifting grating, we varied the diameter of the optimal patch to compile a size tuning curve. The example neuron had an optimal diameter of
Within the CRF and surround regions, we presented a random, dynamic stimulus that consisted of a grating drifting at either the preferred orientation or the orientation orthogonal to preferred. We will refer to these orientations as preferred and orthogonal regardless of whether they occurred in the CRF or surround. Figure 2C shows a sample stimulus sequence for six periods of the dynamic center-surround stimulus. The stimulus icons show the CRF and surround orientation in each period, which lasted for the time that it took the gratings to drift one spatial cycle (typically 160 msec; range, 80-320 msec). At the end of the period, each grating was randomly and independently assigned either the preferred or orthogonal orientation for the next period. Thus, at each period boundary, a grating either drifted seamlessly or made an abrupt 90° orientation shift. For example, at the first transition in Figure 2C, the CRF stimulus changed from orthogonal to preferred (i.e., horizontal to vertical), whereas the surround stimulus remained orthogonal. This transition, labeled onset, usually provoked a strong CRF-driven excitatory response. At the next transition, suppression, the surround changed to the preferred orientation. This configuration usually maximized surround suppression. The suppression is released at the next transition (release), when the surround changes back to orthogonal. At the fourth transition (offset), the CRF stimulus changed from preferred to orthogonal. The final transition shown in Figure 2 involves a simultaneous change in the CRF and surround stimuli that sets the excitation of the preferred CRF stimulus against the suppression of the surround. During the actual stimulus, these five state transitions occurred at random and among the 16 possibilities (see Materials and Methods).
Our first goal was to assess whether the timing of response onsets and offsets driven by the CRF and surround were consistent with the circuit in Figure 1. To do this, we focus on the responses to the five stimulus transitions just described. Figure 3 shows the responses for the example cell, which was typical of the population. The onset of the preferred CRF stimulus (Fig. 3A, change between left, right icons) caused a sudden increase in mean firing rate (black line) 52 msec after the stimulus transition (t = 0). The response curve was computed by averaging together the spike trains surrounding all occurrences (here, 119) of this stimulus transition. Also shown is the reference curve (gray line), which is the average response for two periods of the stimulus depicted by the left icon (i.e., orthogonal orientation is maintained in CRF and surround). We defined the response latency to be the time at which the response diverged from the reference curve by a criterion amount (see Materials and Methods), which is indicated by the distance between the pair of dots beneath the arrow (see Fig. 3A legend). For the opposite stimulus transition, i.e., a change from preferred to orthogonal in the CRF, a decrease in the rate ensued (Fig. 3B, black line). The reference curve (gray line) shows the response to the maintained preferred stimulus. For clarity, the change in response (response minus reference) is plotted in the gray inset. The offset latency (29 msec) was
View larger version (36K):
[in this window]
[in a new window]
Figure 3. The dynamic center-surround stimulus elicited rapid increases and decreases in firing rate via the CRF and the surround. In each plot, the black trace shows the mean firing rate (calibration in A) versus time for the stimulus transition indicated by the pair of icons. The icons straddle t = 0, the time of the stimulus transition. A-D, Reference curves (gray lines) show the response to two consecutive periods of the stimulus indicated by the left icon. All curves are averages over 100-120 repeats (see Materials and Methods). A, The response increased when the CRF stimulus changed from orthogonal (shown as horizontal in icons) to preferred (vertical) but the surround orientation stayed orthogonal (nonsuppressive). The reference response,
The time course for surround suppression is shown in Figure 3C, where only the surround stimulus changed. The latency of suppression (indicated by the arrow) was larger than both the onset and offset latencies for the CRF. However, the suppression was nearly absolute, and its onset was as abrupt as the rate changes elicited via the CRF. The release of suppression (Fig. 3D) was not as abrupt, but it occurred with a latency similar to that of suppression. The data in Figure 3, A and C, imply that the suppressive signal from the surround took longer to affect the spike rate than did the CRF excitation. A compelling demonstration of this is provided by the response to the simultaneous change of the CRF and surround from the orthogonal to the preferred orientation (Fig. 3E). The response to this simultaneous transition (thick black line) followed the response for CRF onset alone (gray dashed line, replotted from Fig. 3A) for
For each orientation-tuned complex cell that had orientationtuned suppression, we measured the latencies for the stimulus transitions described in Figure 3A-D. Their distributions are plotted in Figure 4. The histograms for the CRF (Fig. 4A,B) show that the mean offset latency was shorter than the mean onset latency (arrows show means). When offset latency was plotted against onset latency for each cell, nearly all of the points fell below the diagonal (Fig. 4E), indicating that stimulus offset consistently affected neuronal output more rapidly than stimulus onset. This temporal asymmetry, or onset delay, is characteristic of signals driven by changes between preferred and antipreferred stimuli in the CRF (Bair et al., 2002
View larger version (43K):
[in this window]
[in a new window]
Figure 4. Comparison of response timing for CRF and surround signals across cells. A-D, The distribution of response latencies (see Materials and Methods) for CRF transitions (gray histograms) and surround transitions (open histograms) are plotted for 92 cells (arrows show means). On average, CRF offset (A) was earliest (35 msec; SD, 10 msec). CRF onset (B) occurred on average 17 msec later (52 msec; SD, 13 msec). Surround modulation occurred the latest on average [61 and 60 msec for suppression (C) and release (D), respectively; SD, 17 msec]. E, For each neuron, offset time is plotted against onset time. Almost all points fall below the unity diagonal, indicating that almost all cells responded to the offset faster than to the onset of the preferred stimulus in the CRF. Offset latency also was significantly less variable than onset latency (SD, 9.7 and 13 msec; F test, p = 0.009). F, For the surround, suppression time is plotted against release time. These values were on average not different across cells (paired t test, p = 0.87; n = 85) and were significantly correlated (r = 0.70; p < 10-6).
The second prominent feature of the latency histograms is that suppression is delayed on average relative to CRF onset (Fig. 4, compare B, C). The average delay, 9 msec (SD, 15 msec), was shorter than the 15-20 msec range reported by previous studies (Knierim and Van Essen, 1992
Plotting suppression latency versus CRF onset latency (Fig. 5A) shows that suppression occurred later than CRF onset for most cells, but that for some cells (points below the diagonal), suppression began before the CRF response. Examples of responses for two such cells (Fig. 5A, circled points) are shown in Figure 5B, where the divergence of the solid black and gray lines shows the time of onset (same format as Fig. 3A), and the divergence of the dashed black and gray lines shows the time of suppression (similar to Fig. 3C). In both examples, suppression starts
View larger version (41K):
[in this window]
[in a new window]
Figure 5. Latency of suppression relative to CRF onset. A, Suppression latency is plotted against CRF onset latency for 87 cells. In most cells, suppression arrived later than CRF excitation, but in some (points below diagonal line), suppression arrived sooner. B, For two example cells (A, C, circled points) for which suppression came before onset, response curves are shown for suppression (suppr.; dashed lines) and onset (solid lines). Solid lines show response (black) and reference (gray) curves for the transition like that of Fig. 3A. Dashed lines show the response (black) and reference (gray) curves for the transition like that of Fig. 3C. Open circles mark 5% latencies, as in A and C. C, For each cell, suppression delay (suppression latency minus CRF onset latency) is plotted against suppression strength (1 indicates complete suppression; 0, no suppression). There is a significant negative correlation between suppression delay and strength (r = -0.48; p < 10-5; n = 87). Strong suppression often occurred as early as CRF onset, whereas weak suppression was delayed by
Figure 5A also reveals a positive correlation between suppression and onset latencies (r = 0.44; p < 0.0001; n = 87), indicating that, on average, cells responding early to CRF stimulation also received suppression early. If CRF and surround signals operated through primarily independent mechanisms and if suppression entailed extensive pooling, one might expect the timing of these signals to be uncorrelated. Perhaps the observed correlation is caused by some delaying operation that acts downstream from the point where CRF and surround signals are combined. This is possible if the two signals are combined early in a subset of cells whose outputs incur delays along their paths to drive other cells downstream within V1.
An unexpected relationship was revealed when suppression delay (suppression latency minus CRF onset latency) was plotted against suppression strength (Fig. 5C). Suppression delay was inversely correlated with suppression strength across cells (r = -0.48; p < 0.0001; n = 87). The delay was on average
However, an important possibility to consider is that the negative correlation between suppression strength and suppression latency could be caused by variations in our choice of the inner diameter for the surround stimulus. For surround stimuli placed farther from the CRF, we expect suppression to be weaker (Jones, 1970
Time course as a function of distance
An important yet untested prediction of the circuit in Figure 1, in which surround suppression arises via horizontal propagation of signals from cells with similar tuning in nearby columns of visual cortex, is that the distance traveled should be related to the time of arrival of suppression. To test for such a relationship, we measured the time course of suppression for surround stimuli at various distances from the CRF in 31 cells. We presented surround stimuli for 300 msec during an ongoing optimized CRF stimulus. We used brief stimuli to maximize the number of repeated trials and thereby increase the accuracy of our timing estimates. Figure 6A, right panel, depicts the stimulus at an instant when the surround was present. The surround grating extended from the outer edge of the video display to an inner circular border that was concentric with the circular CRF stimulus in the center of the display. The diagram shows the far surround stimulus (that most remote from the CRF), whereas the two dashed circles mark the inner borders for a mid surround (at an intermediate distance) and the near surround (that closest to the CRF).
View larger version (62K):
[in this window]
[in a new window]
Figure 6. The time course of suppression is shown for surround stimuli that lie at several distances from the CRF for two example neurons. A, The stimulus display (right) shows the CRF stimulus, which was a central patch of grating optimized for the CRF, and the far surround stimulus, which was a grating extending from the edges of the display to a circular inner border. The dashed circles mark inner borders for two other surround stimuli: the mid and near surrounds. The CRF stimulus appeared for 1 sec starting at t = 0, and the surround stimulus appeared for 300 msec starting at t = 400 msec (timing bar at bottom). The preferred orientation and spatial frequency were as shown. The mean firing rates (n = 20 trials, convolved with Gaussian; SD, 4 msec) are shown for the near, mid, and far surrounds (thick to thin lines). The response to the CRF stimulus alone is shown by the dotted line. The near surround strongly suppressed the response to the CRF (thickest black line; arrows indicate onset of suppression). The suppression caused by the mid surround initially had a time course similar to that of the far surround but became weaker after
Responses to four stimulus conditions are shown in Figure 6A for one example neuron. The dashed line shows the response to the CRF stimulus presented alone, and the three solid lines, from thick to thin, show the responses to near, mid, and far surround stimuli, respectively. The thick line shows that the firing rate of this cell was strongly suppressed during a period of
For each complex cell, we computed the change in suppression onset time,
t, associated with the change from the near to far surround. In particular,
View larger version (30K):
[in this window]
[in a new window]
Figure 7. Dependence of latency and time course of surround suppression on surround distance and its implications for propagation velocity. A, The x-axis plots the latency for far suppression minus that for near suppression,
If horizontal axonal projections mediate surround suppression, we would expect our data to reflect their conduction speed. Several studies suggest that propagation speed in such thin, unmyelinated fibers is
It might be argued that we overestimated propagation speed because some cells driving the surround suppression had RFs that overlapped the inner edge of the surround stimulus and hence were closer to the center of the CRF. However, such cells probably did not contribute to the short latency of suppression. Rossi et al. (2001
The most striking change that we observed for more distant surrounds was neither the change in latency nor the expected weakening of suppression. Rather, the examples in Figure 6 show that responses to far surround stimuli were dominated by a transient component that was less evident for near surrounds. Most cells (22 of 31) showed clear signs that the sustained component of suppression weakened more than the initial transient component for far surrounds, but several others (4 of 31) had strictly sustained suppression that decreased evenly in time for surround stimuli lying farther from the CRF. To quantify this trend across the database, we computed for each cell the average suppression as a function of time for the response to surrounds at each of five successive distances. All resulting curves for a given cell were shifted by the latency of near suppression onset for that cell, making t = 0 the time of near suppression onset. The time-shifted curves for near suppression were averaged across all cells to produce the near surround average in Figure 7C (solid line). This plot shows that suppression peaked at 85% at t = 41 msec (filled circle). The conventional, steady-state method of estimating suppression from the size tuning curve yielded a significantly smaller value of 58% (SEM, 5%, computed as the ratio of response for the optimal size to the response asymptote for large stimuli; Cavanaugh et al., 2002
It is worth emphasizing that curves for each cell in Figure 7C were aligned to its time of near-suppression onset, yet the population averages all show rapidly increasing suppression starting at time 0 (or a few milliseconds thereafter for the far surround curve). This reinforces the observation made above that, for a substantial fraction of cells, there was very little change in timing of the onset of suppression with distance from the CRF.
Finally, an important aspect of our study is that we tested only the part of the surround that did not significantly overlap the CRF. For each cell, we attempted to move the inner diameter of our surround stimulus closer to the CRF but found that the surround stimulus then elicited a CRF response that obscured the time course of suppression. On average, our surround stimuli were separated from the CRF stimulus by 28% of the CRF diameter Therefore, we cannot say how the timing of suppression depends on distance within the very near surround. However, we used high-contrast stimuli, which minimize the apparent size of the CRF (Kapadia et al., 1999
Discussion
We will first briefly discuss two basic novel findings about the time course of surround suppression: that, on average, more strongly suppressed cells were suppressed sooner and that suppression had a transient component that was prominent for surround stimuli lying farther from the CRF. We will then discuss the answers to the two questions that motivated this study. First, is the asymmetry in the timing of CRF response onset and offset reflected in the timing of surround suppression? Second, how does the timing of suppression depend on the distance of the surround stimulus from the CRF? The answers were not the ones anticipated for the simple circuit of Figure 1. We will argue that the lack of onset-offset asymmetry for surround suppression may be accounted for in terms of the dynamics of signals at the inhibitory relay, whereas the effectively rapid propagation of signals from distant surround stimuli constrains the circuitry carrying the excitatory responses that drive the inhibition.Delay of signals from the surround relative to those from the CRF
The delay of surround suppression with respect to CRF excitation has been reported previously to range from 15 to 60 msec (Knierim and Van Essen, 1992Onset and transience of surround suppression
Iso-orientation surround suppression achieved its maximum strength on average within 50 msec from its onset (Fig. 7C). Thus, despite its modulatory nature, surround suppression can act as suddenly as the direct-driving CRF signals. In addition to its abrupt onset, suppression often had a strong transient component that was revealed as surround stimuli were placed farther from the CRF. We had expected suppression from distant surrounds to be weaker (Jones, 1970There is no onset delay for surround suppression
We previously reported that neurons in the visual system have an onset-offset asymmetry in their CRF-driven responses (Bair et al., 2002Alternatively, it is possible that the onset delay exists in the signal that drives suppression but is canceled by the inhibitory relay. How this could happen is depicted by the four traces in Figure 8A. Trace F plots the level of activity in the CRF-driven feedforward inputs to the cells (S) that supply the surround signal. The S cells are activated (high state, trace S) by F after a delay (light gray band) that corresponds to their CRF onset delay (17 msec on average). Trace I shows that the response of the inhibitory relay follows its excitatory input from S with little delay. This is supported by reports that inhibitory neurons are activated very rapidly and can retain the temporal structure of their inputs (Simons, 1978
View larger version (47K):
[in this window]
[in a new window]
Figure 8. Refinements to the model shown in Figure 1 to account for the timing of the surround response. A, This hypothetical timing diagram can account for the presence of the onset delay for the CRF and its absence for the surround. The circuit (at the left) consists of a population of surround neurons (S), which inhibit a target neuron (T) via an inhibitory relay (I). The target and surround neurons have the same orientation preference. The four thick traces show the feedforward input (F) and responses versus time for three elements in the circuit. Active and inactive states are shown by high and low values, respectively. The CRF input to the target neuron (T) is assumed to be active (preferred stimulus) at all times (not plotted). The feedforward input driving the surround neurons is meant to indicate the timing of signals as they arrive in cortex; i.e., the subcortical delay is not represented. When the input is high, the surround stimulus is in the preferred configuration. The response of the surround neurons (S) has a delayed onset (light gray band), but response offset has little delay. Thus, the top traces (F, S) show the onset-offset asymmetry for neurons responding to their CRF stimulus (i.e., the bar marked offset is shorter than that marked onset). The response of the inhibitory relay (I) follows rapidly the signal from S. The response of the target neuron (T) is initially high (its CRF stimulus is preferred; data not shown) and is then suppressed by the inhibitory signal. When inhibition from I turns off, the response of the target neuron should recover rapidly (dashed line). However, if the recovery of the response of T were delayed (light gray band), then the latency of its release from suppression (release bar) would be equal to its latency for suppression onset (suppression bar), consistent with our observations. B, Cells (triangles) in V1 (bottom gray box) project via fast axons (arrows) to a higher visual area (top gray box) where neurons have larger CRFs created by convergent input. If cells from the higher area project back (dotted line) directly or indirectly to local inhibitory neurons in V1 (black circle), then suppression from far regions of the surround could arrive on target cells in V1 with little additional delay compared with suppression from the near surround. This could explain the data points in Fig. 7A that fall along the vertical line for zero delay.
Thus, we hypothesize that the persistence of GABAA inhibition may be responsible for both the onset delay observed for the CRF and the lack of a similar delay for surround suppression. Such an asymmetry is consistent with the notion that inhibition has veto power over excitation.
Suppression delay versus distance
We measured the change in latency of suppression as the surround stimulus was moved farther from the CRF. For many cells, suppression was somewhat delayed for more distant surrounds, but for a substantial fraction of cells, there was very little delay, even for surround stimuli that were far from the CRF and evoked relatively weak suppression. The apparent propagation speed of suppression was >1 m/sec for 40% of cells, whereas only 1 of 31 cells yielded a speed <0.2 m/sec. This effectively rapid propagation occurred in cells recorded in layers 2 and 3, layer 4, and deep layers. The most widely cited circuit for iso-orientation surround suppression involves long-range horizontal projections (Allman et al., 1985Can faster propagation within V1 explain our data? Higher conduction speeds have been found using electric shock. In cat cortical slices, Hirsch and Gilbert (1991
Alternatively, rapid effective propagation of suppression could result from feedback from visual areas with large RFs, such as V2 or MT (Allman et al., 1985
Lamme, 1995
