Cerebral Cortex, Vol. 11, No. 7, 648-665,
July 2001
© 2001 Oxford University Press
Real and Illusory Contour Processing in Area V1 of the Primate: a Cortical Balancing Act
Section of Neurobiology, Yale University School of Medicine, 333 Cedar Street, New Haven, CT 06520, USA
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
It is known that neurons in area V2 (the second visual area) can signal the orientation of illusory contours in the primate. Whether area V1 (primary visual cortex) can signal illusory contour orientation is more controversial. While some electrophysiology studies have ruled out illusory signaling in V1, other reports suggest that V1 shows some illusory-specific response. Here, using optical imaging and single unit electrophysiology, we report that primate V1 does show an orientation-specific response to the ‘abutting line grating’ illusory contour. However, this response does not signal an illusory contour in the conventional sense. Rather, we find that illusory contour stimulation leads to an activation map that, after appropriate subtraction of real line signal, is inversely related to the real orientation map. The illusory contour orientation is thus negatively signaled or de-emphasized in V1. This ‘activation reversal’ is robust, is not due merely to presence of line ends, is not dependent on inducer orientation, and is not due to precise position of line end stimulation of V1 cells. These data suggest a resolution for previous apparently contradictory experimental findings. We propose that the de-emphasis of illusory contour orientation in V1 may be an important signal of contour identity and may, together with illusory signal from V2, provide a unique signature for illusory contour representation.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
Visual contours abound in natural scenes. Some visual contours are clearly defined by luminance contrast (e.g. Fig. 1a
, river bank, lower circle). Other contours (e.g. Fig. 1a, canyon wall detail) are defined in a less direct manner, often inferred by local visual cues, e.g. texture (Julesz, 1984; Leventhal et al., 1998), motion (Marcar et al., 1995) and occlusion (Baumann et al., 1997). How does the visual system encode these inferred or ‘higher order’ contours?
|
Here, we have addressed this question by studying the cortical processing of one type of higher order contour, the ‘abutting line grating’ illusory contour (Fig. 1b
) (Kanisza, 1974; Soriano et al., 1996). In this contour type, displaced line gratings induce the perception of orientation even in the absence of luminance contrast across the contour. Previous studies have drawn different conclusions regarding how visual cortex processes this type of illusory contour. Electrophysiological studies in the primate have shown that orientation-selective cells in area V1 (primary visual cortex) are well activated by real (luminance defined) contours (Hubel and Wiesel, 1968). In area V2 (the second visual area), however, cells are activated by both real and illusory contours of the same orientation (von der Heydt and Peterhans, 1989). The possible existence of such ‘illusory contour’ cells in V1 is more controversial. In the primate, electrophysiological studies have concluded that illusory contour cells are virtually absent in V1 (Peterhans and von der Heydt, 1989; von der Heydt and Peterhans, 1989). Grosof and co-workers suggested that primate V1 cells can respond to illusory contours defined by displaced grating stimuli (although their ‘illusory’ contour stimuli also comprised real luminance contrast edges) (Grosof et al., 1993). In the cat, Sheth et al. (Sheth et al., 1996) reported an illusory contour response in area 17 (the cat primary visual area), thus concluding that illusory contour processing indeed commences in the first rather than second visual area (Sheth et al., 1996). Lower resolution functional imaging studies of humans suggest, too, that V1 may play some role in illusory contour signaling (Hirsch et al., 1995; Mendola et al., 1999; Seghier et al., 2000). Whether these different conclusions are due to species differences or to differences in experimental method needs further examination. Regardless of whether primary visual cortex encodes illusory contours or not, the encoding of real and illusory orientation by single V2 neurons raises significant questions. Since illusory contour cells in V2 respond to both real and illusory contours of the same orientation, their signal can be ambiguous. How, then, are real and illusory contours sufficiently differentiated by the visual system? One possible way in which real and illusory contours may be distinguished by V1 and V2 is by some integration of their respective responses.
In this paper, using optical imaging and electrophysiological methods, we investigate the processing of the abutting line grating illusory contour by V1 and V2 in the anesthetized primate. In particular, we examine whether V1 shows evidence of any illusory-specific activations and, if so, what role these activations play in illusory contour processing. We find that V1 does demonstrate orientation-selective response to illusory contours but, surprisingly, one that is complementary to that shown by V2. We propose that this signal in V1, together with the illusory signal in V2, serves to distinguish the real versus illusory nature of visual contours.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
Surgical Preparation
Experiments were performed under protocols approved by Yale Animal Care and Use Committee. Five adult cynomologus macaque monkeys and one adult rhesus macaque monkey were administered ketamine (10 mg/kg i.m.) and atropine sulfate (0.05 mg/kg) and prepared for surgery. Following intubation and catheterization for intravenous drug administration, animals were anesthetized with thiopental sodium (Abbott Laboratories, North Chicago, IL; induction 10 mg/kg, maintenance 1–2 mg/kg per h), paralyzed with vercuronium bromide (Organon, West Orange, NJ; induction 0.1 mg/kg, maintenance 100 µg/kg per h), and artificially respirated. Anesthetic depth was assessed continuously via implanted wire EEG electrodes, end-tidal CO2, oximetry and heart rate monitoring, and by regular testing for response to toe pinch. Eyelids were retracted with specula. Pupils were dilated with atropine and eyes were focused with customized primate contact lenses (Danker Laboratories Inc., Sarasota, FL) onto a computer screen (Barco Calibrator PCD-321, Belgium) at 145 cm distance. Eyes were aligned by converging the receptive fields (RFs) of a V1 binocular cell with a Risley prism over one eye. Under aseptic surgical conditions, a craniotomy (in most instances 10 x 6 mm, 5–15 mm lateral to midline, 14–20 mm rostral to occipital cranial ridge) and durotomy were performed to expose cortex posterior to the lunate sulcus. Such exposures, from which all our recordings were obtained, gave us access to cortical areas representing eccentricities of 2–5° below the horizontal meridian and along the vertical meridian.
Optical Imaging
An optical chamber was adhered to the skull, filled with sterile silicone oil and sealed with a glass window. Images were acquired using an Imager 2000 system (Optical Imaging Inc., Germantown, NY) with 630 nm illumination. Image data was binned to yield response map dimensions of 324 x 240 pixels. Each stimulus condition was presented in randomized order for 3 s with a 10–15 s interstimulus interval (Bonhoeffer and Grinvald, 1996). For each stimulus condition, we collected 15 consecutive 200 ms image frames after stimulus onset and these were stored for subsequent analysis. Signal-to-noise ratio was enhanced by trial averaging (40–100 trials per stimulus condition), and by synchronization of acquisition with heart rate and respiration. Animals were positioned on a floating bench (Newport, Irvine, CA) to minimize motion artifacts. For ocular dominance maps, electromechanical shutters (Uniblitz, Rochester, NY) were placed in front of the eyes for monocular stimulation.
Visual Stimuli
Illusory contour stimuli were created using a custom-made C-language program and were presented binocularly to the animal. These achromatic illusory contour gratings were composed of short acute (45°, e.g. Fig. 2a) or short obtuse (135°) lines (bright lines against a dark background, 1 pixel wide, 0.03° width) spaced 0.25° apart, a spacing which has been shown to be effective for illusory contour cells in V2 [(von der Heydt and Peterhans, 1989); at 2–5° eccentricity V2 receptive fields typically have receptive field sizes of 1–2° and V1 0.2–0.5° (Hubel and Wiesel, 1974; Dow et al., 1981; Gattass et al., 1981; Roe and Ts'o, 1995)]. These inducing elements were aligned, with a column spacing of 1.25 cycles/degree, to produce a percept of either horizontal (Fig. 2a, left) or vertical (Fig. 2a, right) illusory contours. To minimize response to real inducer orientation, these rows of aligned inducers were together drifted back and forth [0.8°/s, drift range two cycles, three second presentation time, screen dimensions 13° (w) x 10° (h)] in the direction along the orientation of inducing lines, producing the percept of illusory contour motion orthogonal to the illusory contour orientation (see Fig. 2a). Thus, since element size, orientation, spacing and motion were identical for both illusory horizontal and illusory vertical stimuli, the only difference between these two conditions was the arrangement of the inducing lines, i.e. the orientation of the illusory contour. Responses to illusory contour gratings were compared with responses to identically spaced (1.25 cycles/degree) and drifting real line gratings [see Fig. 2b; 1 pixel width, 0.8°/s, drift range two cycles, 3 s presentation time, screen dimensions 13° (w) x 10° (h)] presented binocularly at four primary orientations (horizontal, acute, vertical and obtuse). Luminance values were measured using a calibrated photometer (Minolta Chromameter CS-100, Ramsey, NJ) and were constant across stimuli (background luminance 0.1 cd/m2, line luminance 40.0 cd/m2, global stimulus luminance 8.0 cd/m2).
|
Electrophysiology
Subsequent to imaging, the glass window and silicone oil were removed and the cortex was stabilized with agar. Glass-coated tungsten electrodes (Ainsworth, Northampton, UK) were inserted into superficial layers of V1 cortex. Response characteristics and RFs of single units were determined using a hand-held visual projection lamp. Units were selected for quantitative study only if they exhibited clear orientation selectivity as determined from an audio monitor. Single units were isolated and spike activity was collected (Spike2, Cambridge Electronic Design Ltd, Cambridge, UK) in response to sequences of oriented real and illusory small-field stimuli (see Figs 7a and 9). As for full field gratings, different illusory stimuli differed only in their illusory contour orientation and were composed of identical inducing lines with 0.25° line spacing. For each real or illusory orientation condition, a single real or illusory contour was swept back and forth (0.8°/s) across a static 2° aperture centered on the receptive field center. In the illusory stimulus conditions, there was no motion of the real lines, only coherent sweeping of the line end positions, producing a percept of a moving oriented illusory contour. Spontaneous activity levels were collected during blank screen presentation. Modulation indices were calculated for real (modr) and illusory (modi) contour stimulation at preferred and non-preferred orientations (see Fig. 7c):
 |
 |
 |
|
|
Image Analysis
Optical imaging maps shown in this paper are either ‘difference’ maps (responses to one stimulus condition subtracted from responses to another stimulus condition) or ‘single condition’ maps (responses show specific activations associated with one stimulus condition only). Single condition maps provide the most reliable indication of stimulus-specific activations, but this may occur at the expense of signal-to-noise ratio. Difference maps yield better overall signal quality, but activations common to both stimulus conditions are eliminated. In our image analysis design, we have taken advantage of this latter limitation. Single condition illusory contour responses contain intrinsically significant real orientation signal component (due to the orientation of real inducing elements). Thus, location of illusory specific responses can only be revealed by canceling out common real contributions via a difference map derivation (e.g. horizontal illusory minus vertical illusory) (Sheth et al., 1996).
Difference Maps
Difference maps were obtained for pairs of stimulus conditions by subtracting summed frames acquired within 3 s of stimulus onset. (Intrinsic cortical responses have relatively slow time-courses peaking 2–4 s after stimulus onset.) These maps (pixel values 0–255) indicate the relative preference of each location in the image for one (darker pixels) or other (lighter pixels) of a pair of conditions (e.g. horizontal or vertical). Gray values indicate equal preference for either stimulus.
Single Condition Maps
Single condition responses were obtained by first summing frames acquired within 3 s of stimulus onset. Single condition maps were subsequently derived relative to a reference ‘blank’ condition. For illusory contour condition maps, and for real oriented stimulus condition maps in some control experiments, we used a blank screen (null orientation stimulus, luminance 8.0 cd/m2) to generate the reference blank response. For all other real oriented stimulus condition maps we used a ‘cocktail blank’ reference, constructed by summing the responses to all four cardinal orientations (Bartfeld and Grinvald, 1992; Bonhoeffer and Grinvald, 1993). Such cocktail blanks are ‘activated’ blanks that serve to average out biological artifacts such as those due to blood vessels (Bartfeld and Grinvald, 1992). In contrast to difference maps, single condition maps indicate the response magnitude at each location in the image for a particular stimulus condition. Thus, darker pixels indicate strongest responses and lighter pixels indicate weakest responses.
First Frame Subtraction
In all quantitative image processing and analyses in this paper, the method of ‘first frame subtraction’ (Bonhoeffer and Grinvald, 1996) was used to remove blood vessel artifact. Since blood vessel artifacts tend to persist throughout the 3 s period of imaging, subtraction of the first frame (in our case 200 ms frame) from each of the subsequent 200 ms frames can reduce these image contributions. (Since the intrinsic signal has a slow onset, there is virtually no stimulus-specific response in the first few hundred milliseconds.) While this method is effective in removing blood vessel artifact, this image enhancement occurs at the expense of overall signal-to-noise ratio (e.g. increased ‘shot’ noise in image). Thus, the appearance of images can be qualitatively different depending on method of image calculation [e.g. compare Fig. 4a (without first-frame subtraction) with Fig. 5b (with first-frame subtraction)]. In this paper, we chose to use first-frame subtraction for all our automated quantitative analyses to reduce the likelihood of vessel artifacts confounding the determination of map correlations.
|
|
R/R). Lower panel, spatially filtered real map. (c) Difference map for illusory stimulation. Horizontal and vertical illusory responses were subtracted to reveal illusory components (top panel). Darkest areas indicate strongest horizontal illusory response and lightest areas indicate strongest vertical illusory response (scale, right, reflectance change –Thresholding
To compare the locations of imaged illusory domains with real orientation domains, we generated thresholded maps (80th percentile, pixel histogram distribution) of each of the real single condition (horizontal, acute, vertical and obtuse) orientation maps. Illusory contour difference maps were thresholded at the 80th percentile and above for horizontal illusory domains, and at the 20th percentile and below for vertical illusory domains. We have used these cut-off levels as a rule of thumb because they are in qualitative agreement with visual inspection of the image. The conclusions drawn from our statistical comparisons of image maps (see spatial correlation methods, below) are not affected by the precise cut-off. Prior to thresholding, images were spatially filtered using a 9 x 9 pixel moving window low pass filter.
Elimination of Small Activation Domains
We presumed that if an activated group of pixels was too small, then it was less likely to be a true orientation domain and more likely to be due to noise. Thus, although infrequent, thresholded pixel clusters (defined as groups of adjacent pixels) containing less than 10 pixels (i.e. less than 50 µm breadth) were excluded from analysis.
Test of Spatial Correlation
To test for the presence of spatial correlation between a pair of thresholded maps (e.g. mapA and mapB), we used a non-parametric statistical method (Cole, 1949; Sorenson, 1976). This method tests for the degree of spatial correlation and does not address the issue of relative signal magnitude. The presence or absence of suprathreshold activation for every pixel in each map pair was tabulated in 2 x 2 contingency tables. For each comparison, we calculated a 
2 statistic (Cole, 1949) to test for statistical significance:
 | (1) |
2 statistic (d.f. = 1) exceeded the 0.001 significance level. Where spatial correlation was deemed significant, coefficients of spatial correlation, c, were calculated (–1.0 < c < +1.0) (Cole, 1949):
 | (2) |
 |
 |
Coefficients within the range 0.2 < c < 1.0 indicated a significant degree of domain overlap, and coefficients within the range –1.0 < c < –0.2 indicated a significant degree of domain segregation. Coefficients within the range –0.2 < c < 0.2 were deemed neither significantly overlapped nor segregated (Cole, 1949). We considered that coefficients could vary with changes in threshold criteria. We therefore tried different threshold levels (90th/10th and 70th/30th percentile, cf. 80th/20th percentile criteria) in some cases. We determined that although the precise coefficient value can vary with different cut-offs, the sign of the coefficient value remains robust.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionNotesReferences |
|---|
We used conventional intrinsic optical imaging methods to map the spatial distribution of cortical activity in area V1 of five anesthetized adult macaque monkeys. For comparison, we mapped area V2 activity in two of these five cases, and in one additional adult macaque monkey. In each case, a portion of areas V1 and V2 posterior to the lunate sulcus was exposed, and the V1/V2 border location was determined via ocular dominance column mapping (e.g. Figs 3c, 6a
). We obtained and then compared cortical orientation maps in response to presentation of drifting real and illusory contour stimuli (Fig. 2).
|
|
To investigate the spatial relationships between real and illusory maps, we first derived high resolution spatial maps for real orientation preference (Fig. 3
). Single condition maps were obtained for each of the four cardinal orientations (real lines were 1 pixel wide full screen length oriented at 0, 45, 90 or 135°) (Fig. 3a). Difference maps (Fig. 3b) were obtained by subtracting orthogonal conditions. In some cases, orientation maps were supported by electrophysiological sampling (Fig. 3d). The spatial locations of orientation domains were then determined by low-pass filtering and thresholding (Fig. 3a, colored outlines to right of each single condition map).
Illusory contour orientation maps (e.g. Figs 4b and 5c) were then obtained by subtracting image responses to orthogonal illusory stimuli composed of identical inducers. Such subtraction removes signal common to both stimulation conditions — that is, the contribution by real oblique contour activation (Sheth et al., 1996). Any remaining differential signal can then be attributed to differential illusory contour activation.
We have organized the paper in the following manner. Our observations regarding real and illusory responses in V1 and V2 are presented in Figures 4–7. Control data are presented in Figures 8–10. For control conditions, we used randomly positioned oblique short line stimuli with the same motion as illusory contour stimuli (no illusory contour percept) (see Fig. 10), and also blank screen stimuli with a global luminance equivalent to the real and illusory contour stimuli. Our interpretation of this differential signal as due to illusory contour activation was further controlled by examining dependence of this signal on inducer orientation and position (see Figs 8 and 9). Additional data that serve to aid in the interpretation of our observations (single condition maps) are presented in Figures 11–12. Our summary and speculations regarding these observations are presented in Figure 13.
|
|
|
|
|
Reversed Alignment of Illusory and Real Domains in Area V1
Previous studies had suggested that primate V1 cells exhibit little or no response to illusory contours (Peterhans and von der Heydt, 1989; von der Heydt and Peterhans, 1989). We were thus surprised to find that optical images from area V1 in response to illusory contour stimuli revealed orientation-dependent response domains. Figure 4 shows a typical V1 cortex case where illusory response domains are evident. These illusory response domains (Fig. 4b) were comparable in size and spacing to real line orientation domains (Fig. 4a) in V1, although overall differential signal amplitudes were reduced relative to real response maps (imaging activation typically 20–50% in magnitude). We were further intrigued when we compared real and illusory response maps and did not observe overlap between real and illusory domain maps of the same orientation preference. Rather, in V1, horizontal real domains tended to overlie vertical illusory domains and vertical real domains tended to overlie horizontal illusory domains (compare lower panels of Fig. 4a and b). Orange circles demarcate strongest (darkest) horizontal real domains in Figure 4a, but these same domain locations tend to coincide with strongest (lightest) vertical illusory domains in Figure 4b. Thus, when the illusory activation map is compared with the real activation map in V1, we find an apparent inversion in response, i.e. an ‘activation reversal’ map.
We then quantified these impressions. We uniformly applied low-pass and threshold criteria to images in order to delineate the locations and extents of activation domains, and then examined statistically the spatial overlap (i.e. spatial correlation) of these illusory contour and real line domains. As mentioned above, the illusory mapping signal is relatively small and thus susceptible to noise. To increase our confidence in our analysis, we therefore designed an analysis method that was conservative in two ways. First, we confined the test of correlation to only the strongest horizontal/vertical (either real or illusory) signal domains by thresholding the maps. Secondly, we used a non-parametric statistical analysis that (unlike parametric methods) does not require the assumption of image pixel normal distribution. With this approach, our statistic indicates primarily the presence or absence of spatial overlap of domains, rather than pixel-for-pixel correlation across unthresholded signal range. Indeed, because of the constraints associated with this approach, the extent of spatial correlation across conditions may sometimes be underestimated.
Figure 5 shows an example of how we quantitatively compared our V1 orientation maps. Strongest domain locations were obtained first by low-pass filtering and then thresholding each real response map (Figs 5a and b). Illusory response maps (Fig. 5c, top) were similarly low-pass filtered (Fig. 5c, middle) and thresholded (Fig. 5c, bottom). We then overlaid these strongest domain responses in one color-coded map (Fig. 5d) The presence of significant spatial correlation between these strongest domains in the illusory and real maps was tested using a 2 statistic (see Material and Methods). For correlations that are significant, this method yields a numerical correlation index (c) that ranges from –1 (completely non-overlapped) to 0 (neither clearly overlapping or non-overlapping) to +1 (completely overlapping). Although, with this approach, the significance levels and the correlation indices can change with different low-pass and threshold parameters, we find that both the sign and approximate magnitude of the correlation index are persistent and robust over multiple low-pass and threshold parameters.
We applied these analysis procedures to image data from five cases of V1. Analysis results are summarized in Table 1. Two of the cases are shown in detail in Figure 5d and e. Consistent with our previous qualitative observations, horizontal illusory domains (red outline) tend to overlie vertical real domains (blue), and vertical illusory domains (black outline) tend to overlie horizontal real domains (orange). In all five cases of V1 studied, statistically significant spatial correlations (P < 0.001, 2 test) were found between horizontal illusory and vertical real maps, and between vertical illusory and horizontal real maps (Table 1, area V1, orthogonal orientation data, positive indices). Between co-oriented real and illusory domains, significant inverse correlations (P < 0.001, 2 test) were found (Table 1, area V1, matching orientation data, negative indices) in all five cases.
|
To test whether illusory activation maps may be related to the orientation of the inducing lines, we compared the distribution of illusory domains with real acute orientation domains (the inducing line orientation). Of the 10 such comparisons made in V1 (Table 1
, 10 comparisons in last two columns), there were no instances of significant positive spatial correlation between illusory contour domains and real oblique orientation domains. Thus it is unlikely that illusory component maps were an artifactual consequence of insufficient ‘nulling’ of contributions from real inducing lines. In addition, maps obtained from subtraction of different randomly positioned oblique line segments did not produce significant differential response, demonstrating that the illusory contour signal did not result merely from the presence of line ends (see also Fig. 10). Neither were these signals due to differential motion since line segments were moved along the inducing real line orientation in both illusory conditions. We therefore attribute these activations as responses to illusory contours. These data thus suggest that the activation pattern in V1 during illusory contour stimulation is reversed relative to that during real line stimulation, a pattern we term ‘activation reversal’. We will examine in a later part of this paper whether this reversal constitutes an absolute inversion in response or a relative inversion in response.
Alignment of Real and Illusory Orientation Domains in Area V2
For comparison, we also examined responses in area V2. In the macaque monkey, area V2 is located anterior to V1 on the lip and in the depths of the lunate sulcus. The portion of V2 available for imaging varies between 0 and 2 mm in antero-posterior extent. V2 cortex was sufficiently exposed for imaging in only three of six cases studied. We found a pattern of response in area V2 (Fig. 6) that was distinct from that of area V1. The orientation images obtained in response to real line stimulation in V2 were similar to those obtained previously with domain sizes of ~500 µm (Fig. 6b), larger in size than those in V1 (Ts'o et al., 1990; Roe and Ts'o, 1995; Roe and Ts'o 1997). When we imaged V2 during illusory contour stimulation (Fig. 6c), we also observed a clustering of cortical activation in difference maps. Although the magnitude of illusory response signal was smaller in general, these orientation-dependent response domains were comparable in size and spacing to real line orientation domains in V2 (Fig. 6b). Consistent with and further supporting previous electrophysiological findings in V2, we find that areas of dense activation often showed alignments between real and illusory contour domains with the same orientation preference (compare Fig. 6b with Fig. 6c, lower panels). These instances of domain alignment were accompanied by some spatial differences between real and illusory maps in V2 (e.g. Fig 6c, gray areas outside circled zones), perhaps related to the orientation of inducing real line components (Ramsden et al., 1999b).
As done for V1 analyses, we quantified the spatial overlap in V2 by calculating the statistical relationship between thresholded real and illusory maps. Of the three cases of V2 cortex studied, all exhibited a significant positive spatial correlation (P<0.001, 2 test) between real and illusory domain maps of same orientation preference (Table 1, area V2, matching orientations data, positive indices). In the case shown in Figure 6 (Table 1, case F), the correlation index between horizontal real and horizontal illusory maps is +0.34 and between vertical real and vertical illusory maps is +0.27. As shown in Table 1, correlation indices in V2 ranged from +0.26 to +0.39, confirming significant overlap of real and illusory domains of the same orientation preference. Note that correlation indices in V2 tend to be somewhat lower than in V1 (see text below and Table 1). This may reflect a more complex organization for contour processing in V2 (Ramsden et al., 1999b). An orientation-matched correlation between real and illusory domains in V2 is consistent with (although not necessarily predicted by) previous electrophysiological findings showing that more than one-third of oriented V2 cells can have similar orientation tuning preference for oriented real line and higher order contour stimuli (Peterhans and von der Heydt, 1989; von der Heydt and Peterhans, 1989). Overall correlations between real and illusory orientation maps have been reported in area 18 of the anesthetized cat, although specific domain alignments were not compared (Sheth et al., 1996). This is the first report of such domain correlations in primate V2.
In sum (Fig. 6d), our optical imaging data demonstrate that differential illusory responses are evident in primate V2 and V1. These responses show characteristic and distinct alignments with real orientation domains. We find that horizontal illusory stimuli produce strongest activation in the horizontal real domains of V2. In contrast, the same horizontal illusory stimulus produces weakest activation in the horizontal real domains of V1.
Activation Reversal in V1 is Confirmed by Single Unit Electrophysiology
Given the surprising results we obtained in V1, we chose to further examine the responses of single V1 neurons to real and illusory contour stimulation using electrophysiological methods. As with our imaging recordings, we presented sequences of preferred and non-preferred (i.e. orthogonal) oriented drifting real lines or illusory contours (see Materials and Methods). Only cells with clearly oriented responses (n = 25) were considered for study. Figure 7 illustrates responses from a V1 cell with a 135° orientation preference (Fig. 7b). This cell exhibited a much stronger response to a 135° oriented line than a 45° oriented line (Fig. 7a, middle section). However, when alternating sequences of 45°/135° illusory contours were presented (Fig. 7a, left section), we obtained the opposite response pattern. The cell exhibited a greater response to a 45° illusory contour than a 135° illusory contour. Thus, when real line orientations were presented to illusory responsive cells (Fig. 7a, middle section) the strongest responses occurred at the preferred orientation. When illusory contour orientations were presented, we observed the weakest responses at the preferred orientation: i.e. an ‘activation reversal’ response pattern. These modulations were compared with spontaneous firing (Fig. 7a, right section; solid line = mean, dotted lines = ±1 SD). Indeed, when stimulated by illusory contours at the preferred real line orientation, responses could in some cases be less than mean spontaneous firing (Fig. 7a, left section). These activation patterns were observed using both small and full screen stimuli.
To quantify these observed modulatory effects, we calculated real (modr) and illusory (modi) modulation indices (Fig. 7c) for each cell when stimulated by preferred and non-preferred orientation sequences. Modulation indices range from –1 (strong activation reversal) to 0 (no modulation) to +1 (strong in-phase activation). In addition, we determined modulation indices for spontaneous activity (similarly calculated for alternating epochs, see Fig. 7a, right section) of all cells sampled, and used the mean index ± 2 SD (Fig. 7e, dotted lines) as a significance criterion. Three possible distributions are depicted in Figure 7d for real versus illusory modulation indices: uncorrelated activation (left), in-phase activation (middle) and activation reversal (right). As shown in Figure 7e, our data support the latter. Of 25 cells, 12 (48%) had illusory modulation indices exceeding those of spontaneous conditions (Fig. 7e, dotted line, mean spontaneous modulation – 2 SD). Modulation indices were negative for all of these 12 cells, demonstrating the presence of activation reversal at the level of single units. Considered as a population, these data also support the imaging results. The mean illusory modulation index was significantly below zero (i.e. the mean spontaneous index) (P = 0.01, paired t-test). The overall distribution of illusory modulation indices (mean –0.20, median –0.25, SD 0.23, range –0.68 to +0.19) tended towards negative values. Such a skewed distribution is consistent with a population response demonstrated by our imaging maps showing an activation reversal response in V1 (Fig. 7d, right).
Control Experiments
Area V1 map reversals occurred consistently across five different animal preparations during illusory processing. This suggests that they do not arise due to some chance consequence of a specific stimulus configuration selectively influencing a specific portion of cortical retinotopy. Nevertheless, we considered the possibility that the apparent illusory contour response map in V1 arose not from the higher order contour per se but rather from configurational aspects of the inducer elements themselves. For example, perhaps there was some specific interaction between the real and illusory orientation signaling that might lead to activation patterns dependent on a specific inducer angle (i.e. an inducer orientation dependence). Alternatively, due to the small size of some V1 RFs, perhaps there is an imbalance in the number of lines entering the RF during vertical versus horizontal illusory stimulation (i.e. an inducer position dependence). To investigate these possibilities we performed further control experiments.
Control Study 1: Inducer Orientation Independence
To investigate whether these illusory maps might be dependent on inducer element orientation, in three experiments we compared illusory orientation images obtained with inducing elements of different orientations (acute and obtuse). These experiments gave similar results. Figure 8 illustrates such a comparison (same case as shown in Fig. 3). Single condition orientation reference maps were first derived in response to horizontal, acute, vertical and obtuse oriented lines (cf. Fig. 3a). In Figure 8a we show illusory orientation maps obtained from the same cortex using either acute inducers (left) or obtuse inducers (right). The illusory horizontal and vertical activation zones are circled in red and black, respectively, for the acute induction condition (left); and in pink and brown, respectively, for the obtuse induction condition (right). Each illusory activation pattern bears the predicted reversed relationship with the real orientation map. When acute inducers are used, the horizontal illusory map (red outline) overlaps with the vertical real domains (blue from Fig. 3a) (Fig. 8b, top left panel). When obtuse inducers are used, the horizontal illusory domains (pink outline) also overlap with the vertical real domains (blue) (Fig. 8b, top right panel). When horizontal illusory maps obtained with acute or obtuse induction are compared directly, there is a high degree of overlap (Fig. 8b, bottom left panel, red and pink are overlapped). Figure 8c illustrates a similar relationship between the vertical illusory (black outline, acute inducers; brown outline, obtuse inducers) and horizontal real domains (orange). Thus, regions most strongly activated by real vertical are most weakly activated by illusory vertical; those most strongly activated by real horizontal are most weakly activated by illusory horizontal. These qualitative observations are also supported by spatial correlation indices (Table 2). Positive indices are obtained between orthogonal real and illusory maps, for both acute and obtuse inducers. Negative indices are obtained between matching real and illusory maps, for both acute and obtuse inducers. Thus, when inducer orientation is varied, we see no evidence of consistent change or shift of the map away from or towards the inducer orientation domains in V1. These results illustrate that the activation reversal pattern observed in V1 is predicted by the illusory contour orientation and is obtained independent of the inducer orientation.
|
Control Study 2: Inducer Position Independence
The interpretation of V1 responses may be complicated by the small receptive field size of V1 neurons. V2 responsiveness is more easily related to the alignment of line ends because the stimuli are designed so that each receptive field [typically 0.5–2° in size (Roe and Ts'o, 1995)] is stimulated by at least two or three sets of line ends (Peterhans and von der Heydt, 1989). At the eccentricity at which we record, the sizes of V1 cell classical receptive fields can be as small as 0.25° or less. Thus, it could be argued that some of the response modulation may be due to small differences in the precise geometry of line end positions. To examine this issue, we have studied how small shifts in the position of the illusory contour stimulus affects the response of V1 neurons.
Figure 9a illustrates a V1 cell (RF size ~0.2°) located in an obtuse orientation domain that is selective for 135° orientation. Figure 9b schematizes stimulation of this cell by acute (left panel) and obtuse (right panel) illusory contour stimuli. With this stimulus geometry, the receptive field is stimulated by two to four line ends of the illusory contour stimulus. To assess the effect of different positions of these line ends on the response of this V1 cell, we shifted the illusory contour stimulus (in four pixel offsets) such that the center of the stimulus was positioned at one of nine locations relative to the RF (Fig. 9c). Although we do not have the ability to know precisely what part of the receptive field each line end is entering, we can be sure that the cell is experiencing a slightly different stimulus geometry with each of the nine stimulus positions. Furthermore, our choice of offset and range of positions ensures that a full ‘cycle’ of inducer spacing has been sampled.
Using this approach, we repeated the experiment described in Figure 7, using alternating 45° and 135° real lines and illusory contours (Fig. 9d). The centers of both real and illusory line stimuli were similarly shifted in 4 pixel offsets. With the center of the stimulus at position 1 (Fig. 9d, top row, left), this V1 cell responds robustly to the 135° real line stimulus (unshaded epoch) and is relatively quiescent during 45° real line (shaded epoch) stimulation. Consistent with previous examples, this response pattern is reversed for illusory line stimuli (top row, right), such that a better response is obtained with 45° illusory line stimulus (shaded epoch) than with 135° illusory line stimulus (unshaded epoch). When the stimulus is moved to a different position (e.g. position 2) the response to 135° real line is slightly reduced (row 2, left), perhaps due to imprecise centering of the receptive field or to substructure within the receptive field. However, the response to illusory contours still shows a reversal at this position (row 2, right). Indeed, such a reversal is present in the remaining seven of the nine stimulus positions (lines 3–9). Differences in response magnitude are observed at different stimulus positions, both in terms of absolute response magnitude and relative preferred/non-preferred response ratio. Thus, there may be some effect of the precise stimulus geometry on the cell's response. However, in all positions the cell's orientation-selective response reverses with illusory contour stimulation. Although only one cell was mapped in such detail, these results strengthen our findings by showing that while the magnitude of response may be modulated by details of stimulation geometry, the pattern of activation reversal is not. That is, the activation reversal pattern observed in V1 is predicted by the illusory contour orientation and is obtained regardless of relative inducer position. These observations further support the activation reversals obtained with imaging.
Control Study 3: Random Line Control
Although illusory difference maps are ideal for revealing preference for one illusory stimulus over another (relative change), they are less useful for revealing the changes of activation associated with a given illusory stimulus condition (absolute change). For example, do strong illusory difference signals result from an increase in vertical domain activation, a decrease in horizontal domain activation, or a combination of both? To address this issue, we devised a difference map paradigm involving only one rather than two illusory contour stimuli. This paradigm involves subtraction of a random line stimulus from an illusory contour stimulus. At the top right of Figure 10a, we illustrate a random line stimulus where constituent real lines share the same size and orientation as the horizontal illusory stimulus and thus has identical overall line density and luminance. These random lines are also moved with identical motion along their axis of orientation. The random line stimulus differs from the illusory line stimulus only in its lack of line end alignment, and therefore lack of illusory contour percept. Thus, akin to subtracting blank from horizontal real, subtracting randomly positioned lines from horizontal illusory (Fig. 10a, right) eliminates the real component of the signal (due to orientation and motion of real line elements), leaving illusory-specific signal. In effect, this produces a difference map that discriminates signal changes associated with only one illusory contour condition.
Two random line experiments were conducted, both with similar results. One example of such a subtraction is shown in Figure 10a. Subtraction of random line response from horizontal illusory response evoked a differential map (right) that is qualitatively similar to that associated with the subtraction of vertical illusory responses from horizontal illusory response (left). Although not identical (see below), spatial overlap of these activation regions is high (compare alignment of red outlines). In both maps, dark regions (outlined in red) are associated with preferred horizontal illusory activation.
We next compared these random line subtracted maps with real orientation maps. Figure 10b illustrates that, consistent with the observations shown in Figures 3–5