|Home | About | Journals | Submit | Contact Us | Français|
Surround suppression (SS) refers to a reduction in the effective stimulus contrast in one visual location produced by a stimulus presented in an adjacent location. This type of suppression is tuned for orientation and spatial frequency and is thought to be a cortical process. In this paper we used psychophysical measurements to determine whether S-cone-driven signals are affected by surround suppression and, if so, whether S-cone and achromatic signals interact at spatially-remote locations. Our results revealed three important aspects of surround suppression.
Firstly, we show that S-cone probes are suppressed by simultaneous S-cone contrast surrounds and that this suppression has the characteristics of a cortical mechanism. Secondly, we show that when probes and surrounds are presented simultaneously, there are no suppressive interactions between S-cone and luminance stimuli. Finally, we demonstrate that this apparent independence is an artifact of signal timing: When the S-cone components of the stimuli precede the luminance components by approximately 40ms, we find a significant interaction between the two pathways. The amplitude of this interaction depends critically upon the relative onset times of the two components. These results indicate that some component of surround suppression depends on neural computations that occur after the S- and Luminance pathways are combined in striate cortex. In addition, the strong dependence of the magnitude of surround suppression on temporal ordering suggests that much of the effect is driven by transient signals.
Many visual targets can be weakened or rendered invisible by a more intense stimulus (often called a ‘mask’) in close spatial and temporal proximity. This effect has long been a useful tool for studying the early visual system. By varying spatial, chromatic and temporal properties of the mask relative to the target, we can learn about how neurons responding to different features interact with each other. In recent years, psychophysicists studying cortical mechanisms have used this technique to demonstrate that contrast-driven neural interactions are tuned to orientation, spatial frequency, direction of motion and a variety of other dimensions.
The most general finding of these studies is that as the contrast of the masking stimulus increases, target (probe) contrast must also increase to produce the same criterion detectability. It was thought initially that this psychophysical non-linearity reflected the saturation of the neural firing rate observed in single cells: A larger incremental change in contrast would be needed to produce the same incremental change in firing rate, and thus more contrast was needed in the probe target for the observer to detect a change. However, subsequent neurophysiological studies demonstrated that there were more factors at work. Although the firing rate of pre-cortical neurons stimulated by very small patches do exhibit a simple saturating contrast response function (Bonin, Mante et al. 2005), the main determinant of neural activity in cortex is the local population response which adjusts individual firing rates in order to cope with increases in stimulus magnitude, a process known as contrast normalization (Carandini, Heeger et al. 1997). This type of normalization shifts the response function of a single neuron to a higher range; the shift prevents saturation and maintains sensitivity to small changes within the range. Normalization also produces the characteristic contrast increment function observed in human psychophysical data. In 1994, Foley provided psychophysical evidence that contrast normalization accounted for many of the observations on human contrast sensitivity (Foley 1994).
Most psychophysical studies of contrast masking have superimposed the mask on the probe target. A few, however, have examined the effects of high contrast annular surrounds on contrast detection and contrast appearance of a central probe (Werner 1935; Cannon and Fullenkamp 1988; Snowden and Hammett 1998; Xing and Heeger 2000; Zenger-Landolt and Koch 2001). Petrov and colleagues (Petrov, Carandini et al. 2005; Petrov and McKee 2006) showed that these two types of local and long-range masking, which they called overlay masking and surround suppression respectively, had quite different properties. Surround suppression is more tightly tuned for orientation and spatial frequency than overlay suppression, it has a different dependence on contrast, and it appears to be weak or absent in the fovea but highly significant in the periphery. It operates across distances in visual space that scale with eccentricity but not the spatial frequency of the probe stimulus – suggesting a mechanism that operates across a fixed distance on cortex. At large eccentricities surround suppression can operate over distances that are surprisingly large; at an eccentricity of eleven degrees, a surround separated from a target by four degrees can still elevate target thresholds significantly and metacontrast effects have been reported over even larger ranges (Stoper 1978). The spatial extent and strong spatial tuning of surround suppression also reinforces the idea that it is cortical in origin since receptive fields in the LGN and retina are relatively small and weakly-tuned for properties such as orientation and spatial frequency. Correlates of this remotely-induced change in effective contrast have been measured extensively with single-unit electrophysiology (Blakemore and Tobin 1972; Levitt and Lund 1997; Sceniak, Ringach et al. 1999; Walker, Ohzawa et al. 1999; Bair, Cavanaugh et al. 2003), EEG (Haynes, Roth et al. 2003), MEG (Ohtani, Okamura et al. 2002) and fMRI (Williams, Singh et al. 2003; Zenger-Landolt and Heeger 2003).
Until now, surround suppression has been studied almost exclusively using achromatic luminance stimuli. Although the early human visual system contains three chromoluminance channels generated by combining cone signals in different ways, the effects of surround suppression on contrast detection thresholds have not been measured in the two opponent color systems. While many researchers have demonstrated local interactions between superimposed luminance and chromatic signals (e.g Drum 1983, Stockman and MacLeod 1993, Chen, Foley and Brainard 1993), reports of suppressive effects due to remote, chromatic stimuli are much rarer.
Experiments concerning remote suppressive effects in the S-cone and luminance pathways are particularly interesting because these pathways are relatively independent prior to visual cortex (De Valois, Abramov et al. 1967; Young 1986; Hendry and Reid 2000). Although this separation is not absolute and some interactions between chromatic and luminance channels have been demonstrated to date (Drum 1983; Lee and C. F. Stromeyer 1989; Stockman, MacLeod et al. 1993; Chen, Foley et al. 2000) they have been measured almost exclusively in spatially overlapping stimuli and often with uniform fields. It is therefore difficult to assign them exclusively to cortical or pre-cortical mechanisms. Even in the rare cases when interactions have been measured between spatially-separated S-cone signals and luminance contrast signals (Singer and D'Zmura 1994) these measurements have concerned perceptual appearance rather than detection thresholds which are likely to depend on different neural mechanisms (Cannon 1985; Snowden and Hammett 1998).
The experiments described in this paper ask two questions: Firstly, do signals initiated in the S-cones undergo surround suppression in a qualitatively similar manner to pure luminance signals. Secondly, if they do, are S-cone and luminance surround suppression mechanisms independent or can signals in one channel interact with those in another? The answers to these questions are important because they help to constrain the neural mechanisms that underlie long-range contrast normalization in cortex.
The test stimuli consisted of phase-randomized Gabor ‘probe’ patches presented either in isolation or with an annular surround. The relative phases of the probe and annulus regions were randomized from trial to trial since surround suppression is phase insensitive (Petrov and McKee 2006). All experiments were two-alternative forced choice (2AFC) contrast detection threshold measurements. Our experiments were spatial 2AFC with stimuli presented to the left and right of fixation and the subject asked to indicate which location contained the probe Gabor patch. Our measurements were all made at an eccentricity of five degrees and the spatial frequency of the Gabor patch was 1.5 cycles per degree (cpd).
For any block of trials, the contrast and orientation of the surround was constant. An adaptive staircase procedure, ‘QUEST’ (Watson and Pelli 1983) varied the contrast of the probe to determine the contrast that yielded 79% correct responses. Each staircase terminated after 300 trials and the current estimate of the mean and standard deviation of the probe threshold was recorded. Each measurement was repeated at least three times per subject. Where shown, threshold data from individual runs were combined across subjects using an average weighted by the variance of each individual measure (Bevington 1969).
Our stimuli differ in several important ways from those often used to study surround effects in other psychophysical studies.
Our measurements were made with stimuli presented in the periphery, where simultaneous surround suppression has been shown to be maximal (Petrov, Carandini et al. 2005).
To reduce uncertainty about probe location, thin grey rings were present around all potential probe locations throughout the experiments so that they were unambiguous. This is particularly important when probes are presented without the surrounding annuli since the increase in spatial uncertainty caused by the absence of high-contrast flanking regions can raise the detection threshold of this types of probe artificially (Petrov, Verghese et al. 2006).
Stimuli were presented for short intervals (120 msec) to minimize the possibility of subjects making eye movements towards the target locations. The rapid presentation time meant that there was insufficient time to execute a saccade from fixation to either of the target locations before the targets disappeared.
A gap of at least one stimulus grating wavelength (1λ - approximately 0.67° of visual angle) was maintained between the probe and annulus locations to minimize the effects of border segmentation effects or overlay masking on the detection thresholds. In addition, the relative phase of the probe and surround regions was randomized to ensure that any potentially phase-sensitive border effects due to contrast induction have, on average, zero effect.
Finally, we performed contrast detection threshold measurements which are likely to recruit a small, well-tuned neuronal population rather than contrast appearance measurements which are likely to involve population averages of all neurons with RFs overlying the probe region (Cannon and Fullenkamp 1991; Xing and Heeger 2001).
Our stimuli and surrounds were defined by contrast modulations in one of two directions in cone-contrast space: One direction was designed to stimulate only the S-cones, leaving the quantal catch in the L- and M- cones unaffected, thereby generating a strong and independent signal in the opponent S-(L+M) pathway. The other direction was designed to stimulate all cone classes equally, generating a signal in the achromatic luminance pathway.
Generation of truly S-cone isolating stimuli is a demanding task. Approximate S-cone isolation can be achieved computationally by using silent-substitution techniques (Estevez and Spekreijse 1982) based on photometric calibration of individual monitor spectra (Brainard 1989) and published measurements of human cone photoabsorption spectra (Stockman, MacLeod et al. 1993). However, the effects of optical aberrations and macular pigment density vary across subjects as well as across the retina (Cottaris 2003). For this reason, we required each subject to perform a calibration setting to determine their S-cone isolating cone contrast settings for each spatial frequency. In this procedure, we presented nominally S-cone pathway-isolating sine-wave gratings of the same spatial frequency as the probe stimuli, flickering at 8Hz. We then allowed the subject to adjust the amount of luminance and opponent red/green contrast until the flicker was minimized. These calibration settings were repeated ten times and the mean cone contrast vector of each set of ten measurements was taken to be the S-cone isolating stimulus. These minimum flicker settings have been shown to be effective at finding S-cone isolation and give almost identical results to other procedures such as finding points of maximum transient tritanopia or minimally-distinct border settings (Cavanagh, MacLeod et al. 1987; Smithson 2005).
All stimuli were generated using the Psychophysics toolbox on a PowerPC Mac running OS9 with a 10-bit graphics card (ATI Radeon 9000). They were presented on a Sony Multiscan200 monitor running at 100Hz with a resolution of 1024×768 pixels. Monitor calibration was performed at 1nm resolution using a fiber-optic photospectrometer (USB2000, OceanOptics, FL). Subjects sat at a distance of 650mm from the screen which subtended a total angle of 26 by 20° of visual angle. The background of the screen was maintained at a mean gray luminance of 34cd/m2 and all stimuli were calibrated in units of cone contrast relative to this background.
Six observers (4 women) all with normal or corrected to normal vision participated in the experiments. All observers were experienced psychophysical observers and three were naïve to the aims of the experiments. Experiments were performed in accordance with the Smith-Kettlewell Eye Research Institute institutional review board protocols. Not all observers participated in all experiments but all experiments were performed by at least three observers and data from all participants are shown for each experiment.
We first tested whether probes defined by S-cone contrast were suppressed by simultaneously-presented surround contrast. These data are shown in Figure 2. In all subjects, and for all surround contrasts, contrast detection thresholds for S-cone-defined probes increased in the presence of co-oriented S-cone surrounds and the degree of suppression increased with the contrast of the surround. On average, surrounds with a high level of S-cone contrast (approximately 50%) almost doubled the contrast required to detect the central probe region at threshold. These suppression factors are broadly consistent with those measured for luminance stimuli in our own lab, although the contrast dependence is far more linear over the range we measured – most likely because we are unable to generate contrasts high enough to produce the saturation commonly seen with luminance surround suppression (see Petrov 2005 (Petrov, Carandini et al. 2005)).
For luminance-defined stimuli, surrounds suppress most effectively when they match the probe region in orientation and spatial frequency. This tuning is relatively strong: collinear luminance surrounds can raise luminance probe detection thresholds by factors of three or four while orthogonal surrounds have no detectable effect. By comparison, recent measurements of the orientation tuning of overlay masking (where the mask and probe are superimposed) show far weaker tuning (Foley and Chen 1997; Holmes and Meese 2004; Petrov, Carandini et al. 2005). The strong orientation tuning that we measure for luminance surrounds is therefore likely to be a signature of cortical processing and is one way of differentiating surround suppression from weakly-tuned subcortical normalization processes such as overlay masking.
Because the surrounds we used in these experiments were separated from the probe regions, the suppression must have been due to mechanisms that were able to integrate over two thirds of a degree of visual space. This, in turn, makes it unlikely that we are measuring a pre-cortical mechanism since even generous estimates of S-cone-driven receptive field sizes at this eccentricity are less than a degree (Wandell 1995; Calkins 2001).
However, it is conceivable that some of the suppressive effects measured in Experiment 1 were due to broadly-tuned overlay masking mechanisms operating at the border of the probe region. We therefore measured the orientation and spatial frequency tuning of the suppressive mechanism in order to distinguish between the effects of overlay masking and surround suppression.
Firstly, we measured the orientation tuning of the suppressive effects found with the original stimulus configuration (gap of 1λ). These data are shown in Figure 3a. They show a small but significant effect of orientation with collinear surrounds generating more suppression than orthogonal surrounds. These data are well-fit by a Gaussian function with an offset of 1.3 and a full width at half maximum (FWHM) of 57° (σ=34°). Secondly, we found that S-cone surround suppression was also tuned for spatial frequency (Figure 3b): Probes were maximally suppressed by surrounds of a matching spatial frequency. Because of the upper limits on the spatial frequency imposed by the limited S-cone resolution at 5° eccentricity, it was not possible to measure a broad spatial frequency tuning curve for these data.
S-cone surround suppression is tuned for both orientation and spatial frequency, and low-spatial-frequency surrounds had little effect on high-spatial-frequency probes. However, we did measure significant suppression from orthogonal surrounds and from surrounds with spatial frequencies that were several times greater than the probe's. This untuned component could either reflect a contribution from pre-cortical normalization mechanisms with large suppressive fields or it could originate from cortical neurons with weak orientation and/or spatial frequency tuning. We ran two experiments to distinguish between these two possibilities. Firstly we measured the effect of unambiguous S-cone overlay masking in order to compare it to the effect from spatially distant masks. Secondly, we measured the spatial extent of the spatially-distant suppressive mechanism in order to distinguish it from overlay masking which is a local effect (Petrov, Carandini et al. 2005).
Pre-cortical contrast normalization mechanisms have a well-defined effect on contrast detection thresholds. Although pre-cortical normalization processes are thought to be almost entirely suppressive, they can either increase or decrease contrast detection threshold depending on the total amount of contrast in the probe region. The shape of the threshold-versus-contrast (TVC) ‘dipper function’ that describes this effect is well understood in the contrast normalization literature and can be derived from an analysis of the underlying neural response-versus-contrast (RVC) functions of the neural population (Foley 1994). To a good approximation, the detection thresholds are inversely proportional to the slope of the contrast response function and in psychophysics it is common to estimate the RVC curve from the TVC data.
The shape of the TVC curve for S-cone stimuli has been measured in a previous study and it was found that it could be fitted by the same functions used for luminance data (Chen, Foley et al. 2000). We repeated these measurements for a subset of the overlay contrast range and found very similar results. Our data are plotted in Figure 4. Note that we could only measure thresholds for a reduced contrast range because the probe and mask stimuli were now overlaid and their sum could not exceed the monitor gamut (Chen et al extended their display range by combining images from two monitors optically). Nevertheless, the overlay masking data can be compared directly to the surround suppression data over a large part of the range.
The most striking aspect of these data is that pure S-cone overlay masking reduced contrast detection thresholds over the entire contrast range that we could measure. The dipper for the S-cone TVC function is remarkably broad relative to that of the achromatic TVC curve although it is likely that S-cone masks do generate suppression at higher contrast levels than we were able to generate with our display system. These curves are essentially identical to those measured by Chen et al over the same contrast range. By comparison, S-cone contrast in the surround always increased detection thresholds. These data suggest very strongly that the suppressive effects that we measure for annular surrounds remote from the probe region are not due to pre-cortical contrast normalization processes.
How remote must the surround region be from the probe before it ceases to suppress it? For luminance probes presented at six degrees eccentricity, statistically significant threshold elevations are generated by surrounds at distances of up to three degrees (Petrov and McKee 2006). This distance is one of the most compelling reasons for believing that surround suppression is a cortical phenomenon. The spatial resolution of the S-cone system is at almost a log-unit lower than that of the luminance system (Green 1972; Hess, Mullen et al. 1989; Calkins 2001); therefore, the extent of surround suppression for the S-cone system might be substantially larger than the extent in the luminance system.
To examine this issue, we first measured the effect of suppressive surrounds presented at increasing distances from the probe region. These data are shown in Figure 5 together with data from achromatic surround suppression stimuli for comparison. These data show that S-cone surrounds exert a significant suppressive effect at distances of up to 2.5° from the probe region. At this distance, we measure no significant effect of achromatic surrounds on achromatic probe regions but our conditions were somewhat different from those used by Petrov et al (Petrov and McKee 2006) and may not have been optimal for the luminance pathway.
In general, we find that the absolute level of surround suppression for a given stimulus depends on several factors including spatial frequency and stimulus duration. It is possible to measure far larger suppressive effects for the luminance system when we use stimuli that have a higher spatial frequency and shorter duration but it is difficult to make measurements with S-cone isolating stimuli with the same spatiotemporal characteristics because the detection thresholds for these stimuli are outside the gamut of our display system.
So far we have measured the effects of simultaneous surround suppression in configurations where the probe contrast and surround had the same chromaticity. If the suppressive signal is computed from a mixed population of cells that respond to both luminance and S-cone contrast, we might expect that luminance contrast surrounds would suppress S-cone contrast probes. Other researchers have shown that long-range masking may occur between stimuli of different chromatic types (Foster 1979; Reeves and Bearse 1988) although the spatial and chromatic channels in these experiments were not perfectly isolated. We therefore asked whether high-contrast luminance annuli act to suppress S-cone probe regions in a manner similar to that found for the within-chromatic-class stimuli above.
Figure 6 shows the effects of placing achromatic annuli of differing contrasts around S-cone probe regions in the periphery (top row). The S-cone annulus data is repeated from Figure 2 for comparison.
As shown in Figure 2, co-linear annular S-cone gratings increase detection thresholds for S-cone probes by as much as a factor of two. The effect is approximately linear with contrast over the range of contrasts that we can generate on our display system.
In comparison, simultaneously-presented achromatic contrast in the surround region has only a very weak effect on S-cone contrast detection thresholds. The absence of suppressive effects on S-cone probes due to achromatic surrounds is particularly striking when their contrast is expressed in multiples of contrast detection thresholds. The contrast of the highest contrast luminance annulus was approximately 50 times the luminance probe detection threshold (not shown). In comparison, luminance surrounds of as little as three times the detection threshold (approximately 3% RMS cone contrast) have significant effects on the detection thresholds of luminance probes and S-cone surrounds at three times threshold can double the contrast detection threshold of S-cone defined probes.
Are S-cone and luminance components truly independent in these surround-suppression stimuli? Very few neurons in primary visual cortex are exclusively tuned to a single chromatic direction (Johnson, Hawken et al. 2001) and single-unit measurements in V1 indicate that achromatic surrounds have significant effects on the responses of neurons that respond to S-cone stimuli (Solomon, Peirce et al. 2004). Even though the chromatic tuning of centers and surrounds is more consistent in area V2, we would still expect to measure some effect of luminance surrounds on S-cone centers, given the fact that our 50% contrast luminance surrounds must be driving the majority of V1 neurons to saturation.
One possibility is that the low-contrast S-cone and high-contrast luminance signals arrive at a critical site at different times. Our stimuli are extremely brief in comparison with those used by Solomon et al who averaged spike rates recorded over two seconds and so relative timing effects may be more important. It is well-established that although S-cone signals are not particularly sluggish when measured in the LGN (Calkins 2001), they develop a temporal lag relative to luminance signals in V1 (Cottaris and De Valois 1998) and this lag can be measured psychophysically in stimuli where luminance and S-cone signals are overlaid (Lee and C. F. Stromeyer 1989; Stockman, MacLeod et al. 1993). Perhaps compensating for this lag leads to more powerful interactions between spatially remote stimuli?
In our final experiment, we measured contrast detection thresholds for brief (50ms) S-cone stimuli in the presence of luminance contrast surrounds that were displayed with a relative temporal offset (stimulus onset asynchrony, SOA). The SOA was varied from −60 to 60ms in steps of 20ms with a negative SOA indicating that the probe appeared before the surround.
The results are shown in Figure 7. Data points to the right of zero are thresholds for S-cone probes in the presence of luminance surrounds when the surround was presented after the probe. Points to the left are thresholds when the surround preceded the probe. Although there is only a slight increase in threshold due to high-contrast surrounds presented simultaneously with the probe region, there is a clear increase in threshold when the surround is presented roughly 40ms after the probe region.
There are clearly strong suppressive interactions between spatially-separated luminance and S-cone signals in the human visual system but they depend critically on the relative timing of the different chromatic components of the stimulus.
We have shown here that detection thresholds for equiluminant S-cone gratings are elevated by simultaneous S-cone contrast outside the classical receptive field. This stimulus suppression appears to be qualitatively similar to that experienced by low-contrast achromatic luminance gratings in the presence of simultaneous higher-contrast achromatic luminance surrounds and it is distinct from pre-cortical overlay masking. Moreover, there are significant interactions between chromatic and achromatic mechanisms with luminance contrast surrounds suppressing the apparent contrast of S-cone probe stimuli when the onset of the surround is delayed with respect to probe onset.
These data suggest that S-cone and luminance surround suppression engage similar cortical mechanisms. As luminance-driven surround suppression effects are measured routinely in primate primary visual cortex (e.g. (Bair, Cavanaugh et al. 2003; Solomon, Peirce et al. 2004)), it is natural to ask if this area might also be a site of the S-cone surround suppression that we measured. Solomon et al have studied the effects of chromatic and achromatic suppressive surrounds in areas V1 and V2 using single-unit electrophysiology (Solomon, Peirce et al. 2004). Several of their findings are relevant to the results presented here: Firstly, they find that luminance contrast surrounds suppress both luminance and color/luminance CRFs in V1 and V2. Secondly, they show that chromatic surrounds generate weak or little suppression of chromatic CRFs in V1 but more suppression of both chromatic and achromatic CRFs in V2. Finally, in common with other groups (Johnson, Hawken et al. 2001), they find that V1 neurons with strong chromatically-tuned CRFs have almost no orientation tuning.
Are our data consistent with the response properties of V1 neurons? At first glance, it would seem that they are not because Solomon et al find little evidence for suppression of any V1 neurons with strong chromatic tuning. Instead, the properties we measure for S-cone surround suppression more strongly resemble those found in area V2 where Solomon et al found significant suppression of chromatically-tuned units.
However, it is possible that psychophysical detection of chromatic stimuli is mediated by cells that are not exclusively tuned to equiluminant directions in color space. For example, Solomon et al (along with others e.g. (Lennie, Krauskopf et al. 1990; Johnson, Hawken et al. 2001)) measure many V1 neurons that respond to both chromatic and achromatic contrast. Although they respond to isoluminant color, these units have good spatial frequency and orientation tuning and are likely to respond well to the oriented bandpass stimuli used in our experiments.
Therefore, our data do not rule out an early site for S-cone surround suppression as measured by contrast detection tasks. Our S-cone isolating stimuli will drive color/luminance cells in both the probe and surround regions. According to Solomon et al, color/luminance cells in the probe regions would therefore be subject to surround suppression from the color/luminance cells in the surround. The same neurons would be driven by pure luminance contrast and S-cone isolating contrast and so we would expect to see an interaction between these two types of stimuli.
In particular, the fact that we measure significant orientation tuning in the surround suppression driven by S-cone isolating stimuli suggests that these effects may not depend on the most chromatically-sensitive neurons in V1, since these are essentially untuned for orientation. The neurons most likely to detect the probe at contrast threshold are those that respond to color and luminance (and are therefore subject to suppression by luminance surrounds). This type of chromoluminance detection mechanism has been described and modeled extensively by Chen et al (Chen, Foley et al. 2000; Chen, Foley et al. 2000). The role of the spatially untuned, chromatically selective neurons may be to code stimulus chromaticity rather than spatial structure (Johnson, Hawken et al. 2004; Solomon and Lennie 2005). According to this line of reasoning, pure appearance measurements would find less orientation tuning and less interaction between luminance surrounds and chromatic probes as these judgments may depend on the outputs of neurons that are part of a more specialized color pathway that shows chromatically-specific gain control both for spatially overlapping (Solomon and Lennie 2005) and remote (Solomon, Peirce et al. 2004) regions of contrast. In fact this is what Singer and D'Zmura demonstrated for superthreshold chromatic induction effects (Singer and D'Zmura 1994). They showed that although chromatic contrast presented in an annulus can induce appearance changes in a spatially remote center region, these changes are not orientation tuned and the inductive effects are rather small.
The sensitivity to the relative timing of the stimulus onsets is also of interest. A recent paper by Ishikawa et al (Ishikawa, Shimegi et al. 2006) demonstrates clearly that surround suppression is closely linked to the phenomenon of metacontrast masking (Alpern 1952). Surround suppression generally refers to effects measured with simultaneous central probe and surround presentation, but this is a special case in a continuum of effects measured over a wide range of stimulus onset, or more likely, stimulus offset asynchronies (SOAs) (Macknik and Livingstone 1998). Breitmeyer (Breitmeyer and Ganz 1976; Breitmeyer and Ogmen 2000; Ogmen, Breitmeyer et al. 2006) and others Ishikawa (Ishikawa, Shimegi et al. 2006) have proposed that surround suppression for luminance stimuli is mediated by two mechanisms that may correspond loosely to the magno- and parvo-cellular (or ‘sustained and transient’) pathways. One mechanism has relatively weak spatial frequency and orientation tuning and high contrast sensitivity. The other mechanism is strongly-tuned to orientation and spatial frequency, but has low contrast sensitivity. Suppressive effects at short SOAs reflect the inhibitory input of the mechanism with weak contrast sensitivity while effects at longer SOAs are the result of the mechanism with greater contrast sensitivity. The reasoning is that while the central probe regions are detected by parvocellular pathway neurons with good spatial frequency resolution, the surround signals carried by the magnocellular pathway dominate suppression. These ‘transient’ signals arrive in cortex first and must therefore be lagged more in order to have them interact with the probe regions. Although the functional differences between the achromatic response properties of parvocellular and magnocellular pathways may be slightly less distinct than this logic supposes (Levitt, Schumer et al. 2001), the underlying observation that “…any stimulus gives rise to a family of impulse responses varying in latency, strength, and persistence, which depend on the spatial-frequency composition of the stimulus” (Breitmeyer and Ganz 1976) is surely sound. The fact that we measure such clear differences in suppression latency with chromatic centers and achromatic surrounds may be a result of the relatively strong functional separation of the magnocellular and koniocellular pathways.
The observation that luminance surrounds are most effective in suppressing S-cone probes when they are lagged by approximately 40ms is particularly intriguing in the light of work by Cottaris and De Valois (Cottaris and De Valois 1998), (De Valois, Cottaris et al. 2000) indicating that chromatic and luminance signals in V1 are lagged relative to each other by time intervals of the order of 10-30ms with magnocellular and achromatic signals arriving earlier than parvocellular and chromatic signals (particularly chromatic signals with S-cone inputs). The lags we measure are also consistent with reaction time differences for S-cone and opponent L/M stimuli demonstrated by Smithson and Mollon (Smithson and Mollon 2004) although we note that the contrasts of our probe and surrounds regions are very different, both in terms of absolute cone contrast and multiples of detection threshold.
The existence of this lag also indicates that the interaction we see is not due to contamination of the S-cone stimuli by luminance transients or residual S-cone contributions to the magnocellular pathway (Stockman, MacLeod et al. 1993). Pure luminance stimuli demonstrate profound surround suppression at zero SOA despite significant contrast difference between the probe and surround regions. If detection of our nominally S-cone isolating probe regions was supported by signals leaking into pre-cortical luminance pathways, we would expect to see equally-strong suppressive effects for simultaneous probe and surround presentations. Dynamic interactions between spatially-overlapping S-cone and luminance signals have been reported before both for uniform (Stockman, MacLeod et al. 1993) and patterned (Lee and C. F. Stromeyer 1989) fields and both groups report a temporal lag between the luminance and S-cone signals. Stockman and MacLeod (Stockman, MacLeod et al. 1993) demonstrated temporally-lagged suppressive interactions between flickering, uniform S-cone and Luminance stimuli with the positive S-cone inputs suppressing luminance inputs at a temporal lag of approximately 17ms. In a motion discrimination task, Lee and Stromeyer (Lee and C. F. Stromeyer 1989) found that S-cone signals interacted with the luminance pathway with a positive sign at a lag of around 40 ms under weak adapting conditions such as those used in this study. In both cases, these effects were attributed to post-receptoral but pre-cortical mechanisms, partly because of the high temporal frequency of the S-cone modulation used. However, it is possible to measure EEG responses in primary visual cortex that track stimulus modulation frequencies well above the flicker fusion threshold (Van Der Tweel 1964). In our own EEG work, we regularly measure robust responses to S-cone flicker above 20Hz and several researchers have demonstrated that both high frequency chromatic and luminance modulations generate activation and adaptation in visual cortex even when they are not perceptible (Shady, MacLeod et al. 2004; Jiang, Zhou et al. 2007). It is possible, therefore, that some of the interactions at short temporal phase offsets measured by Stockman and MacLeod were of a cortical rather than a retinal origin. The temporal lag that they measured also matches that measured by De-Valois et al which appears to be added to S-cone signals after they have entered primary visual cortex. In addition, as we describe above, there are good reasons for differentiating between local ‘overlay masking’ effects that operate on the scale of a retinal or LGN receptive field and cortical surround suppression which operates over much longer distances. The effects measured by Stockman and MacLeod and Lee and Stromeyer may belong to the former category, while the effects described in this paper may belong to the latter.
If the temporal lags in the interactions between luminance surrounds and chromatic probes have direct physiological correlates, they may provide a powerful method for differentiating between the influences of different visual pathways early in the visual system. It is likely, for example, that interactions between isoluminant (L−M)-cone opponent chromatic stimuli and S-cone isolating stimuli will be largely free of a rapid magnocellular component (or else such a component could be eliminated by the presence of constant luminance noise (Smithson and Mollon 2004)). It would be instructive to measure the temporal dependencies of these types of stimuli and compare them with the data presented here.
We have shown that surrounding regions composed of either S-cone and luminance contrast raise the contrast detection thresholds of S-cone probes. Because of their distance and tuning properties and because S-cone surround suppression is qualitatively similar to conventional luminance surround suppression these effects are likely to be cortical in origin. The effect of luminance surrounds on S-cone probes is only manifest when the two stimulus components are presented asynchronously. The degree of asynchrony is similar to that measured for S-cone stimuli in V1 electrophysiologically and it is consistent with the suggestion of De Valois and Cottaris that signals from the S-cone isolating stimuli undergo an additional processing stage immediately after they enter V1. The site of action of cortical surround suppression is therefore likely to be downstream of this site.
Supported by NIH/NEI grants: EY017071,EY018157 and by NSF grant 820101001
I would like to thank Suzanne McKee for valuable discussions and suggestions during the preparation of this manuscript.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.