The intrinsic functional architecture of early cortical areas in highly visual mammals is characterized by the presence of domains and pinwheels, with orientation preference of the inputs to these regions being more and less selective, respectively. We exploited this organizational feature to investigate mechanisms supporting extraclassical surround suppression, a process thought to be critical for figure ground segregation and form vision. Combining intrinsic signal optical imaging and single-unit recording in V1 of anesthetized cats, we show for the first time that the orientation tuning of the suppressive surround is sharper for domain than for pinwheel neurons. This difference depends on high center gain and is more pronounced in superficial cortex. In addition, when we remove the near component of the surround stimulus, the strength of suppression induced by the iso-oriented surround is significantly reduced for domain neurons but is unchanged for orthogonal oriented surrounds. This leads to broader orientation tuning of suppression that renders domain cells indistinguishable from pinwheel cells. Because the limited receptive field of the near surround can be accounted for by the lateral spread of long-range connections in V1, our findings suggest that intrinsic V1 circuits play a key role in the orientation tuning of extraclassical surround suppression.
In species with highly developed visual systems, such as cat and monkey, neurons in early cortical areas are organized into functionally homogeneous domains based on similar orientation preference (Albus 1979; Grinvald et al. 1986; Born and Tootell 1992; Malonek et al. 1994; Lyon et al. 2002; Villeneuve et al. 2009). Such organization is thought to promote more efficient wiring between distant neurons preferring like-orientations, which optimizes the generation of more complex, dynamic, and expansive receptive fields (Mitchison and Crick 1982; Gilbert and Wiesel 1989; Malach et al. 1993; Toth et al. 1996; Bosking et al. 1997; Somers et al. 2002; Changizi and He 2005; Shmuel et al. 2005). In addition to domains, pinwheel centers exist where multiple domains with different orientation preferences converge (Bonhoeffer and Grinvald 1991; Ohki et al. 2006) and exhibit high heterogeneity of orientation preference within the local environment (Mariño et al. 2005; Nauhaus et al. 2008). Synaptic inputs to pinwheel neurons are less selective for orientation than inputs to neurons located in the more homogenous domain regions (Schummers et al. 2002; Mariño et al. 2005; Stimberg et al. 2009). A substantial portion of these inputs can arise through local and long-range cortical circuits (Vidyasagar et al. 1996; Bringuier et al. 1999; Somers et al. 2002; Mariño et al. 2005; Girardin and Martin 2009) and as such likely influence the differences in orientation tuning between pinwheel and domain neurons (Dragoi et al. 2001; Nauhaus et al. 2008).
A long-standing, unresolved issue in V1 is whether intrinsic circuits play a critical role in the modulation of the classical receptive field (CRF) center by the extraclassical surround (Gilbert and Wiesel 1989; Toth et al. 1996; Bullier et al. 2001; Angelucci et al. 2002, Cavanaugh et al. 2002a; Levitt and Lund 2002; Stettler et al. 2002; Series et al. 2003; Chisum and Fitzpatrick 2004; Schwabe et al. 2006). By definition, the extraclassical surround alone does not elicit a response from V1 cells (Allman et al. 1985) but instead typically acts by suppressing responses recorded from the CRF/center (Sceniak et al. 1999; Walker et al. 2000; Liu et al. 2011), especially when center gain is high (Hirsch and Gilbert 1991; Weliky et al. 1995). Furthermore, the extraclassical surround has an orientation tuning independent of the center (DeAngelis et al. 1994; Sillito et al. 1995; Sengpiel et al. 1997; Cavanaugh et al. 2002b) that is thought to play a critical role in figure ground segregation and provide the building blocks of form perception (Knierim and van Essen 1992; Sillito et al. 1995; Series et al. 2003).
Here we take 2 approaches to determine whether intrinsic V1 architecture plays a role in extraclassical surround suppression. First, we relate the orientation tuning of the suppressive surround to a cell's position in the orientation map, domain versus pinwheel. Second, we measure the impact of the “near” component of the surround on the orientation tuning of suppression. In doing so, we find that orientation tuning of surround suppression is significantly sharper for neurons located in domains, and this difference is most pronounced in the superficial layers. Moreover, the orientation tuning of suppression depends primarily on the inclusion of the near surround component, which likely corresponds to the receptive field range covered by long-range lateral connections in V1 (Salin et al. 1992; Angelucci et al. 2002; Levitt and Lund 2002; Cantone et al. 2005; Schwabe et al. 2006). Thus, our data not only suggest that a cell's position in the orientation preference map has a significant impact on the orientation tuning of extraclassical surround suppression but also that intrinsic V1 circuits play a key role.
Materials and Methods
The experiments were performed in 10 adult cats of either sex. All procedures were approved by the Institutional Animal Care and Use Committee of the University of California, Irvine. Animals (2.4–5.0 kg) were initially anesthetized with a mixture of ketamine (21 mg/kg, intramuscular [IM]) and xylazine (3 mg/kg, IM). A tracheotomy was performed to allow artificial respiration, and anesthesia was maintained with isofluorane (0.2–1.0%) in a mixture of nitrous oxide and oxygen (67/33%). To ensure proper level of anesthesia throughout the duration of the experiment, the electrocardiograph, electroencephalography, and lung pressure were monitored. The end-tidal CO2 was maintained between 4.0% and 4.5%.
Pupils were dilated and accommodation paralyzed with atropine sulfate (1%). The nictitating membranes were retracted with 2.5% phenylephrine hydrochloride and corneas protected with zero-power, air-permeable contact lenses that were cleaned daily. To prevent eye movements, neuromuscular blockade was induced with a bolus of vecuronium bromide (0.6 mg in 1 ml, IV) and maintained throughout the experiment with 0.2 mg/kg/h mixed with dexamethasone (0.5 mg/kg/h, IV) in a solution of 5% dextrose and lactated Ringer's.
A craniotomy was made above the dorsal surface of area 17 (V1), and a recording chamber (20 mm diameter) was fixed to the skull with dental cement. The dura was removed and the chamber filled with silicon oil and sealed with glass for optical imaging. At the end of each experiment, the animals were euthanized with a lethal dose of sodium pentobarbital (40 mg/kg) and perfused through the heart with 0.9% saline, followed by a 4% solution of paraformaldehyde in phosphate buffer.
Following perfusion, the brains were removed and stored in a solution of 30% sucrose in phosphate buffer for a minimum of 3 days. Brain sections were cut parasagitally at 50 μm and observed untreated under fluorescent and bright-field light using a Zeiss Axioplan2 microscope. Digital images of electrode tracks stained with the fluorescent dye, 1, I′-dioctadecyl-3.3,3′,3′-tetramethyl- indocarbocyanine perchlorate (Dil, see Single-Unit Recording), were captured using a low-light–sensitive video camera (Cooke Sensicam QE) and Neurolucida software.
Visual Display Apparatus and Optics
Corrective lenses, if necessary, were used in the optical imaging and single-unit recording experiments to focus the eyes on a tangent screen 37 cm away. Artificial pupils (3 mm in diameter) were placed in front of the eyes. The location of the optic disc and the area centralis for each eye were plotted daily using a fiber-optic projector (Pettigrew et al. 1979).
The visual stimuli were displayed on a gamma calibrated 21 inch View Sonic Graphics Series G225f CRT screen with a mean monitor luminance of 50 cd/m2 set to a spatial resolution of 640 × 480 pixels and with a frame refresh rate of 100 Hz in a noninterlaced mode. Stimuli were generated using a G5 Mac with an ATI Radeon 9200 graphics card running EXPO software (courtesy of Dr Peter Lennie).
Optical Imaging and Orientation Preference Maps
CCD Camera and Optics
Optical images were captured with a 12-bit video camera (Adimec-1000M/D, Netherlands; >60 dB signal-to-noise ratio) consisting of a 1004 × 1004 array of pixels equipped with a 50/50 mm tandem-lens combination (Ratzlaff and Grinvald 1991) positioned over the exposed cortex such that its optical axis was perpendicular to the cortical surface. In order to uniformly illuminate the cortex for imaging, 2 adjustable light guides powered by a 100 W tungsten halogen light source driven by a DC power supply (Kepco Inc., Flushing, NY) were passed through a filter and pointed at the cortical surface. Initially, a reference map of the blood vessel pattern at the surface of the cortex (see Fig. 1A) was obtained by using green light at 546 nm, and a region of interest was demarcated. The camera was then focused ∼400 μm below the surface of the cortex and data collected using red light at 605 ± 10 nm. The range of power output of the Kepco power supply was kept between 8–9 A and 9.9–10.0 V.
Visual Stimuli and Data Acquisition
To obtain orientation preference maps (e.g., Bonhoeffer and Grinvald 1991; Mariño et al. 2005), the intrinsic signals were recorded in response to binocularly viewed full screen, high-contrast (100%) square-wave gratings (0.2–0.3 cycles/degree), in 1 of 8 orientations (22.5 × 180°), and drifting at a temporal frequency of 2 Hz. To prevent transitory orientation nonspecific responses, each trial started with a 500 ms static grating, followed immediately by 4 s drifting in one direction and then 4 s drifting in the opposite direction, during which cortical images were captured (over 6 frames without binning) by the Optical Imager 3001 (Optical Imaging Inc., Durham, NC). Each block of trials consisted of the 8 oriented stimuli and a single blank stimulus (uniformly gray screen set to the mean background luminance) presented in pseudorandom order. In addition, to allow for relaxation of activity-dependent vascular changes and prevent neuronal habituation during the data acquisition period, each trial was separated by an 8 s interstimulus interval (ISI), during which time a uniformly gray screen of mean background luminance was presented. Each orientation and blank trial was repeated 40 to 60 times to generate each orientation map.
Single-condition maps for each orientation (see Fig. 1B–E) were derived by first integrating each trial block over its 6 frames and then averaging across the repetitions and dividing by the average of the blank trials (as we have done previously; Mariño et al. 2005). Colored orientation preference maps (i.e., Fig. 1F) were generated through pixel-by-pixel vector summation of the 8 single conditions (Blasdel and Salama 1986; Bonhoeffer and Grinvald 1991). Masks were created for each orientation map to exclude pixels where the map could not be determined due to bone, major blood vessels, or the convolutions of the cortex.
Stability of Orientation Maps
Special care was taken in choosing highly accurate and reproducible maps. Map locations were considered reproducible where there was no displacement of pinwheel centers between 2 maps generated from subsets of 4 alternate orientations (for more description, see supplementary material in Mariño et al. 2005). The first consisted of orientations 0°/180°, 45°/225°, 90°/270°, and 135°/315°. The second set consisted of orientations 22.5°/202.5°, 67.5°/247.5°, 112.5°/292.5°, and 157.5°/337.5°. Color maps from these 2 subsets were compared with each other and to the color map generated from all 8 orientations. Pinwheel positions (and surrounding domains) that were congruent between these 3 maps were deemed acceptable for single-unit recordings.
Single-cell extracellular recordings were made with commercial epoxy-insulated tungsten microelectrodes (5–7 MΩ, FHC, Bowdoin, ME). Electrodes were aligned perpendicular to the cortical surface to target either domains or pinwheels based on the relationship of the surface blood vessel pattern to the orientation map (see Fig. 1A–F). To eliminate the effect of parallax during positioning, we guided the electrode onto the target via triangulation through several vantage points. Once the electrode was inserted, the chamber surrounding the craniotomy was filled with 1.5% agar solution in saline and sealed with physiological wax (melting point 45 °C) in order to reduce brain pulsation. Electrodes were advanced through the cortex by a computer-controlled micropositioner (Motion Controller SMC-100, Newport) fixed to a stereotaxic arm (KOPF Instruments). Action potentials of single neurons were amplified using an Xcell-3 four-channel amplifier (FHC), with the amplified signals also broadcast over a loudspeaker for subjective analysis. Single neurons were carefully isolated by manipulating the electrode in steps as small as 1 μm until amplitudes of spontaneously evoked spikes were well above the background neural activity. Expo's data acquisition and signal-processing capabilities were used for online window discrimination of individual visually evoked action potentials.
Cells were recorded at all cortical depths. See Table 1 for numbers of domain and pinwheel cells recorded at superficial (<600 μm), intermediate (600–1200 μm), and deep (>1200 μm) levels, cortical depths which generally correspond to supragranular, granular, and infragranular layers, respectively, of cat area 17. The angle and overall cortical depth of electrode penetrations were assessed following the procedure of DiCarlo et al. (1996), in which electrodes were dipped (5–10 times) in DiI (50 mg/ml) prior to penetrating the brain and serial sections were observed following histological processing. This allowed for confirmation of the perpendicularity of our penetrations and provided overall measures of the depth of penetration. As shown in the examples in Figure 1G,H, the angles of our electrode penetrations were very near perpendicular to the cortical surface.
|Cortical depth (μm)||HC CRF||HC C-S||LC CRF||LC C-S||ΔNear C-S|
|Total number of cells||58||48||41||43||47||32||33||33||24||26|
|Cortical depth (μm)||HC CRF||HC C-S||LC CRF||LC C-S||ΔNear C-S|
|Total number of cells||58||48||41||43||47||32||33||33||24||26|
Note: C-S, center-surround; HC, high contrast; LC, low contrast; Dom, domain; PW, pinwheel.
The zero harmonic (mean rate) and first harmonic components of the accumulated response were computed for each stimulus. Simple and complex cell designations were determined by the ratio of the first harmonic and mean of the response to a drifting grating stimulus (Skottun et al. 1991). The great majority of our cells were complex (86 of 106).
Identification of the CRF
All visual stimuli were drifting achromatic sinusoidal gratings (unless stated otherwise) enveloped by sharp edged circular apertures, presented on a gray background of the same mean luminance as the stimuli. Each stimulus had a 2–6 s duration and was repeated 2–4 times with a 6 s ISI, during which the animal viewed the background screen. A blank stimulus (background screen) was used to determine the spontaneous activity.
Initially, the approximate orientation and location of each receptive field center were found by using a circular square-wave grating patch (4° in diameter). Each cell was stimulated through the dominant eye with the nondominant eye occluded. To determine the exact position of the receptive field, the grating patch was systematically moved under mouse control until maximal activity was detected audibly. Then, the orientation of the grating was varied to determine the orientation that evoked the greatest response. Next, the orientation was kept constant and the diameter of the patch decreased to as small as 1° to pinpoint the receptive field's center, similar to procedures used by others (Walker et al. 2000; Sceniak et al. 2006). We further confirmed that the stimulus was positioned on the center of the receptive field from the cell's response to the grating summation field test (aperture tuning, see below). If the magnitude of response did not monotonically increase to the initial expansion of the stimuli diameters (0.5°, 1°, 2°, and 3°), then we considered the stimulus to be offset in relation to the receptive field center and the position of the receptive field was reassessed. Finally, for many of the cells to further ensure that the position and size of the center stimulus were not misaligned and also that the extraclassical surround stimuli (see below) did not evoke a response on their own, an annular grating with an outer diameter of 30° was presented at multiple inner diameters (0.1°, 0.5°, 1°, 2°, 3°, 4°, 5°, 6°, 7°, 8°, 9°, 10°, 11°, 12°, 18°, 20°, and 23°) at the preferred orientation (Sengpiel et al. 1997; Walker et al. 2000).
Once the center of each cells’ receptive field was identified, the CRF properties were tested in detail in the following sequence: First, the orientation tuning of the neuron was remeasured with a 100% contrast grating at a fixed diameter of 10°, drifting at 4 Hz and with orientations sampled every 22.5° (16 steps). Next, the grating with the preferred orientation was used to find the optimal spatial and temporal frequencies. Then we found the CRF size (grating summation field; Sceniak et al. 1999; Cavanaugh et al. 2002a), which we define here as the aperture size resulting in the greatest mean response. This was done by presenting a set of 100% contrast, orientation, and spatiotemporal optimized gratings in apertures varying in diameter (0.2°–30°, in 18 steps) and in pseudorandom order (Sengpiel et al. 1997; Walker et al. 2000). Next, the contrast profile (10–100%, in 10 steps) of the cell was measured at the optimal parameters. Finally, if response saturated before 100% contrast, then the aperture tuning protocol was run once more at the maximum non-saturating contrast.
In both monkey and cat, the optimal receptive field size of a V1 cell will often increase when stimulus contrast is decreased, causing the CRF at low contrast to be larger than the CRF at high contrast (Sceniak et al. 1999; Tailby et al. 2007). Therefore, the CRF aperture-tuning protocol was repeated for each cell at the contrast level corresponding to 30% of the saturating response (low contrast). Our low contrast had a range of 8–50% and a mode of 30%.
Spatial eccentricities of all receptive fields in our experiment were constrained within 2° above and 16° below the area centralis.
Extraclassical Surround Stimuli
High-Contrast Center/High-Contrast Surround
Once the optimal parameters of each cell's center (CRF) component were established, we proceeded to measure the orientation tuning of the suppressive extraclassical surround (also referred to as the surround). Both the inner and outer diameters of the surround stimuli were determined individually for each cell from the aperture tuning test described above (i.e., the grating summation field). Specifically, the inner and outer diameters of the surround stimuli were set as the diameters that evoked the optimal response and maximum suppression, respectively, for the cell. In addition, for most cells, the inner surround diameter was confirmed by the annulus test.
To measure the orientation tuning of the cell's suppressive surround, the optimal orientation, aperture, and spatiotemporal frequencies for the center stimulus were kept constant. Directly abutting the outer circumference of the center stimulus was a surround grating of the same phase, and spatial and temporal frequencies. Whereas the center stimulus was maintained at the preferred orientation/direction throughout, the surround stimulus was shown with 1 of 16 possible directions (in 22.5° increments). Both the center and surround were shown at high contrast.
Low-Contrast Center/High-Contrast Surround
To examine the effect of contrast on the tuning of the surround, the above procedure was repeated with 2 modifications: 1) the center was presented at the optimal low-contrast center size as determined by the aperture tuning protocol (see above) and 2) the inner diameter of the high-contrast surround was adjusted to account for the expanded size of the low-contrast center.
High-Contrast Center/High-Contrast Surround: Removal of Near Surround
To test the impact of the near surround on the orientation tuning of extraclassical surround suppression, we removed the near component of the surround and compared responses with full surround conditions. Parameters were the same as the High-Contrast Center/High-Contrast Surround condition above, with the exception that the near surround was replaced with the mean luminance background. The near surround was determined as the difference between the CRF size defined at high and low contrasts. This definition of the near surround is based on the overlap of receptive field sizes in monkey V1 and the receptive field representation covered by long-range lateral connections (Angelucci et al. 2002; Cavanaugh et al. 2002a; Angelucci and Sainsbury 2006). Similar estimates can be made for carnivore V1 based on comparisons of the long-range intrinsic connectivity and receptive field sizes in cat (Salin et al. 1992; Li CY and Li W 1994) and ferret (Cantone et al. 2005). We refer to the remaining surround as the “far” component based on the description of Angelucci et al. (2002). While more recent descriptions of the far surround also include a larger facilitative component (Ichida et al. 2007; Schwabe et al. 2010), it is important to make clear that we only examine the smaller suppressive component of the surround. We consider the impact of this on our far surround results in the Discussion.
We quantified the homogeneity of orientation preference of the local environment for each recording site on the orientation map as in Mariño et al. (2005) and Nauhaus et al. (2008). For the orientation preference maps, the homogeneity index (HI) for a cortical location x is defined as
Center Selectivity Index
This index has a value of 0 for a data set falling uniformly on a circle (not selective for orientation) and a value of 1 for a data set with responses to only a single direction. For cells with orientation selectivity but no direction selectivity, we modified equation (2) by substituting 2θn for θn and halving the resulting angle. Neurons that were directionally selective tended to have high values of the direction selectivity index, while nondirectional but orientation selective neurons tended to have high values of the orientation selectivity index. Similar to previous studies (i.e., Cavanaugh et al. 2002b), we always computed both selectivity indices for each neuron and chose the larger value as the CSI.
Iso-orientation Suppression Depth
Surround Selectivity Index
To estimate the surround selectivity of response suppression, based on Cavanaugh et al. (2002b) and Naito et al. (2007), equation (2) was modified by converting the effect of the surround stimulus into a mean vector given by
Unless otherwise stated, the Mann–Whitney U test was used to assess the significance of differences between responses of cells located in domains and pinwheel centers. The Wilcoxon matched-pairs signed-rank test was used for pairwise comparisons of CSIs, surround selectivity indices (SSIs), and ISDs measured at high and low contrasts and for surround conditions with and without the near component. The Spearman test was used for assessing the correlation between variables. Mean values given in the text are accompanied by standard deviations. In addition, Monte Carlo permutation tests were done to confirm the significance of the effects of removing the near surround on the tuning and depth of suppression of the suppressive surround.
To examine the impact of the V1 orientation map on the extraclassical surround, we recorded from 117 isolated neurons that were located in either domains or pinwheels (see orientation map examples in Figs 1–3). Ninety-one percent of the neurons (106/117) showed suppression by one or more surround orientation. The remaining 11 neurons showed no suppression by any surround and are not included for further analysis.
Center Orientation Selectivity of Pinwheel and Domain Neurons
Prior to examining the extraclassical surround, we first addressed the relationship between the orientation selectivity of the CRF of V1 neurons and their positions in the orientation preference map. We did this in order to provide a point of comparison for the orientation selectivity of the suppressive extraclassical surround (see next section). In addition, we were motivated by discrepancies in previous work showing that orientation selectivity for pinwheel and domain neurons is not significantly different (Maldonado et al. 1997; Schummers et al. 2002; Mariño et al. 2005) but that orientation tuning widths are (Nauhaus et al. 2008). Furthermore, previous studies measured orientation selectivity or tuning widths with large-field stimuli, extending well beyond the cells’ CRFs; yet, extraclassical surround stimuli can significantly sharpen orientation tuning through center-surround interactions (Chen et al. 2005; Xing et al. 2005; Okamoto et al. 2009). Therefore, it has yet to be determined whether the orientation selectivity of the center alone significantly differs between domain and pinwheel neurons.
For each neuron located within a domain (n = 58) or pinwheel center (n = 48), we carefully determined the location of the receptive field center and then established the optimal visual stimulus parameters. At high stimulus contrast, the size of the CRF (center) was first determined by an aperture tuning protocol as the smallest aperture eliciting the greatest response. Next, drifting gratings matching the center were presented at 8 orientations, 2 directions each, to cover 360°. A CSI for each neuron was calculated (eqs 2 and 3), with an index score of 1 indicating either high orientation or direction selectivity and 0 indicating equal selectivity to all directions (Cavanaugh et al. 2002b; Naito et al. 2007). The top row of Figure 2 shows response profiles of 4 domain cells (Fig. 2A–D), and the lower row (Fig. 2E–H) shows the profiles for 4 pinwheel cells. All domain cells show moderate to high orientation or direction selectivity, while some pinwheel cells exhibit relatively high orientation selectivity (Fig. 2E) and others very low selectivity (Fig. 2G,H).
Figure 2J shows the CSI distribution for the pinwheel and the domain populations. The 2 populations overlap in CSI values ranging between 0.3 and 0.77; however, the CSI distribution for domain cells is skewed toward higher values (high selectivity), whereas the pinwheel distribution is skewed toward lower. On average, the orientation selectivity difference between domain (0.59 ± 0.21) and pinwheel (0.42 ± 0.19) neurons is highly significant (P < 0.0001). We also examined whether CSI for pinwheel and domain cells varied by electrode depth relative to the cortical surface. As shown in Table 1, the numbers of pinwheel and domain cells were distributed fairly evenly across different electrode depths, which we categorized into 3 regions: superficial (<600 μm), intermediate (600–1200 μm), and deep (>1200 μm), although a slightly higher percentage of the neurons were located superficially. Overall, we find that domain cells are more orientation selective than pinwheel cells in each of the 3 depths (data not shown), although due to higher variability in the intermediate (P = 0.08) and deep (P = 0.23) zones the difference is only significant for superficial cells (P < 0.05).
The observed dependence of CSI on a cell's position in a domain or pinwheel is influenced, at least in part by the varying degree of orientation preference homogeneity of the local pool of neurons (<250 μm; Mariño et al. 2005; Nauhaus et al. 2008; Stimberg et al. 2009). We measured the degree of local homogeneity at each recording site using a HI (eq. 1) and compared it with the cell's CSI value (Fig. 2J). Similar to previous reports (Mariño et al. 2005; Nauhaus et al. 2008), the HIs of our domain locations were consistently high, ranging from 0.73 to 0.95 (mean: 0.86 ± 0.07), and the HIs of our pinwheel centers were relatively low, ranging from 0.10 to 0.38 (mean: 0.17 ± 0.08). As already seen in the histogram distribution, the CSI values for both pinwheel and domain neurons are spread over a wide range. However, there is a positive and significant correlation between the position of the cell within the orientation map (HI) and its CSI (r2 = 0.33, P = 0.001). In conclusion, for the population as a whole, the position of the cell within the V1 orientation map has a significant influence on the orientation selectivity of the center.
The Suppressive Surround and Its Relation to the Orientation Map
Having established that orientation selectivity of the center significantly differs for neurons located in domains and pinwheels, we next asked whether the orientation map also has an effect on the orientation selectivity and the suppression strength of the extraclassical surround. The suppressive surround properties were recorded for 84 of the 106 neurons described above. While holding the center stimulus at the cell's preferred orientation, we varied the orientation of the extraclassical surround. In total, we presented 16 surround directions tilted at 22.5° increments. In this fashion, rather than manipulating the impact of the local pool of neurons related to the CRF, we were likely isolating the contribution of long-range horizontal projections within V1 (Ts'o et al. 1986; Gilbert and Wiesel 1989; Somers et al. 2002; Stettler et al. 2002; Series et al. 2003; Chisum and Fitzpatrick 2004) and/or feedback from higher visual areas (Bullier et al. 2001; Angelucci et al. 2002; Cavanaugh et al. 2002a; Cantone et al. 2005; Bardy et al. 2009).
Figure 3 shows the relationship between the orientation selectivity of the center and the suppressive surround for 2 domain (Fig. 3A,B) and 2 pinwheel (Fig. 3C,D) neurons. In Figure 3A, the solid trace shows the orientation selectivity of a domain cell's center response, with the optimal response at 90° denoted by the star. The dotted trace shows the surround selectivity profile for the same cell. Here the cell's response is most suppressed when the orientation of the surround stimulus is aligned with the preferred orientation of the center (iso-orientation suppression). The strength of the surround suppression rapidly attenuates on both flanks as the orientation of the surround deviates from the center orientation, becoming ineffective at 45° offset and beyond. For this cell, surround suppression is also direction selective, as no suppression is apparent at the opposite direction (270°). Figure 3B shows strong iso-orientation suppression for a second domain neuron, with a peak orientation at 270°. This cell is also strongly suppressed by nearby nonpreferred orientations. In addition, unlike the previous example, this cell also shows strong suppression for the opposite surround direction (90°). Nevertheless, for both domain neurons, the strength of iso-orientation suppression (star) is clearly greater than the suppression by the orthogonal surround orientations, denoted by the inverted “T” (⊥).
Figure 3C shows an example for an orientation selective pinwheel cell. In this case, the surround has a suppressive effect at all orientations resulting in a flat tuning surround profile. Figure 3D shows another example of a pinwheel cell with peak center selectivity at 337.5°. Here the surround also exerts broad suppression but with slightly greater suppressive depth at iso-orientation.
We consistently observed that the surround strength of iso- relative to ortho-orientation suppression was stronger in domain than in pinwheel neurons (see examples in Fig. 3). To quantify this observation, we devised an index for ISD. The ISD directly compares suppression strength resulting from iso- and ortho-oriented surrounds (eq. 4). A value of 1 indicates that the surround has no suppressive effect at ortho-orientations and exerts complete suppression at iso-orientation; 0 indicates that the suppressive effect of the surround is equal in magnitude for both iso- and ortho-oriented surrounds. Negative values indicate that the surround is more suppressive at ortho- compared with iso-orientation. For example, for the 2 domain neurons (Fig. 3A,B), the ISD values are 0.76 and 0.77 meaning iso-orientation suppression is over 75% greater than ortho-orientation suppression. In contrast, values for the 2 pinwheel neurons (Fig. 3C,D) are substantially lower (−0.12 and 0.18), with the ISD for the pinwheel cell in Figure 3C being negative because the average magnitude of ortho-orientation suppression (⊥) is 12% greater than iso-orientation suppression (star).
The relationship between the ISD and orientation homogeneity of the local environment is shown in Figure 4A. The data show that there is a significant positive correlation between these variables (r2 = 0.32, P = 0.001). The histogram in Figure 4B shows the ISD distribution for pinwheel and domain cells. Both show a wide range of ISD values, with pinwheel neurons skewed toward lower values indicating a broader surround tuning. Overall, the average ISD for domain neurons is double the strength of the pinwheel cells (ΔISD = 0.17; domain: 0.33 ± 0.21 vs. pinwheel: 0.16 ± 0.19; P < 0.001). Furthermore, when examined by electrode depth (Fig. 4C), there are consistently higher ISD values for domain cells compared with pinwheels, with the difference being the greatest for cells encountered above 600 μm (ΔISD = 0.23; domain: 0.41 ± 0.18 vs. pinwheel: 0.18 ± 0.19; P < 0.001). The difference in ISD for cells at intermediate depths is slightly less (ΔISD = 0.20) and also significant (P = 0.05).
The difference in the depth of suppression between domain and pinwheel neurons is clearly visible in the average tuning curves shown in Figure 5A. For cells encountered in the first 600 μm of the electrode descent, the depth of surround suppression at iso-orientation is similar for domain and pinwheel cells (60%; normalized response of 0.4) but for pinwheel neurons suppression at ortho-orientation is on average over 3 times stronger than domain neurons (47% for pinwheels vs. 15% for domains; P < 0.005). For cells encountered in the intermediate (600–1200 μm) and deep (>1200 μm) electrode penetrations, the tuning profiles of the domain and pinwheel populations start to converge (Fig. 5B,C). Overall, the ISD values of these 3 population pairs show that the relative iso-orientation suppression strength progressively weakens with electrode descent for domain, but not pinwheel neurons.
In addition to the ISD, we also calculated a SSI for each domain and pinwheel neuron (eq. 5; Cavanaugh et al. 2002b; Naito et al. 2007). The SSI measures the orientation (or direction) selectivity of surround suppression; an index value of 1 indicates that suppression occurs following only one surround orientation, whereas 0 means all orientations suppress equally. As for the ISD, SSI values for our example domain cells (0.39 and 0.35; Fig. 3A,B) were both higher than the example pinwheel cells (0.04 and 0.26; Fig. 3C,D). Because SSI and CSI are both measures of circular variance, this allows for direct comparison between the selectivity of the center and surround. As shown in Figure 6A, for the majority of neurons (63 of 84), the center has a higher orientation selectively (CSI) than the surround (SSI). Of the remaining cells, some had CSI and SSI values that were the same (n = 13; touching the diagonal), whereas others had higher SSIs (n = 8). Overall, we find no correlation between the 2 indices (r2 = 0.04, P = 0.74), suggesting that separate mechanisms mediate the orientation selectivity of the center and surround.
White and black arrowheads on the ordinate in Figure 6A show the average SSI values for the pinwheel (0.19 ± 0.13) and domain (0.26 ± 0.19) populations, respectively. Although the domain population has a modestly higher SSI mean, this difference is not significant (P = 0.13). However, as shown in Figure 6B, there is a modest positive correlation between SSI and HI (r2 = 0.21, P = 0.05). Furthermore, when examined by electrode depth (Fig. 6C), the SSI difference between the superficial neurons is also significant (ΔSSI = 0.14; domain: 0.34 ± 0.18 vs. pinwheel: 0.20 ± 0.16; P < 0.05). This SSI difference is also apparent in the average suppression tuning profiles shown in Figure 5A. For both domain and pinwheel neurons, suppression by the iso-oriented surround is nearly identical, whereas for domain neurons, nonpreferred surround orientations provide noticeably weaker suppression and are therefore more selective. Or, put another way, pinwheel neurons are suppressed more strongly by a broader range of surround orientations and therefore are less selective.
In summary, we find that the overall orientation selectivity of the extraclassical surround (SSI) and, to an even greater extent, the relative depth of suppression by an iso-oriented surround (ISD) depend on position in the orientation map. That is, compared with neurons located in pinwheels, domain neurons show higher orientation selectivity of the suppressive surround and greater relative suppression by iso-oriented surrounds.
Near and Far Surround Effects Relative to the Orientation Map Position
Several recent reports have shown that the full extraclassical suppressive surround can be divided into “near” and “far” subcomponents (Angelucci et al. 2002; Bair et al. 2003). Thus, we next examined whether the orientation tuning of the suppressive surround can be attributed more specifically to either component. Figure 7A illustrates the aperture tuning profile of a V1 neuron to high- and low-contrast gratings. Reduction of stimulus contrast results in a reduced response and an expansion of the optimal aperture size (Sceniak et al. 1999; Cavanaugh et al. 2002a; Tailby et al. 2007). The near surround is quantified as the region corresponding to the increase in area of the receptive field when the stimulus contrast is reduced from high to low, as described for V1 cells in monkey (Angelucci et al. 2002; Levitt and Lund 2002; Schwabe et al. 2006) and ferret (Cantone et al. 2005); Similar estimates can also be made for cat (Tusa et al. 1978; Salin et al. 1992; Li CY and Li W 1994; Tailby et al. 2007). In other words, the near surround is suppressive at high center contrast and facilitative at low. The remainder of the extraclassical surround beyond the near component can be considered as the far surround (Angelucci et al. 2002; Bair et al. 2003), but only the suppressive component of the far surround, as the larger facilitative component of the far surround (Ichida et al. 2007; Shushruth et al. 2009) was not determined for our experiments (see Discussion). Figure 7A also shows the relationship between the dimensions of the physical stimulus under low- and high-contrast conditions, and the relatively small and large extents of the near and far surrounds, respectively, are indicated.
Low-Contrast Center Orientation Selectivity of Pinwheel and Domain Neurons
As a result of lowering center contrast, part of what was operationally defined (see Materials and Methods) as the extraclassical surround is now incorporated within the center. This center summation at low contrast invites the question of what effect it has on the center orientation selectivity. Thus, we first determined what impact lowering the stimulus contrast has on the CSI of V1 cells and its relation to the orientation map. As shown in Figure 7B, a reduction in the center contrast maintains the average CSI difference between the domain (n = 44) and the pinwheel (n = 36) populations (0.59 ± 0.20 vs. 0.39 ± 0.19; P < 0.0001).
Figure 7C shows the scatter plot for the pinwheel and domain CSI values under low and high center contrast conditions. There is not a significant shift in CSI as a function of contrast for the whole population (P > 0.05). Likewise, for pinwheel neurons, the mean CSI shows no significant change (high contrast: 0.39 ± 0.17; low contrast: 0.41 ± 0.20; P = 0.62). For the domain neurons, however, there is a significant increase in CSI when contrast is lowered (high contrast: 0.54 ± 0.20; low contrast: 0.60 ± 0.19; P < 0.05).
Low-Contrast Center/High-Contrast Surround
We next asked what impact lowering center contrast has on the orientation tuning of surround suppression (ISD and SSI) for pinwheel and domain neurons. In Figure 8, the high- and low-contrast values for each cell were determined from the contrast profile shown in the lower right of each panel. For high contrast (black diamonds), we chose the maximum response prior to saturation or supersaturation (80% for Fig. 8A and 100% for Fig. 8B–D). For low contrast (gray diamonds), we chose the corresponding contrast that evoked 30% of maximum response. In the large left hand panels, the horizontal black and gray lines show the peak response of the center under high and low contrast, respectively. It is worth noting that the difference between the center responses at high and low contrast may not match a 30% difference. This is because contrast values were chosen based on the contrast tuning profile using the optimal stimulus aperture for high contrast (i.e., the high-contrast center size). Once the low-contrast value was chosen, the center size was redetermined at low contrast (see Materials and Methods). At low contrast, the center typically expands and therefore firing rate will be higher than that elicited by the smaller aperture size used in the contrast tuning profile (e.g., compare firing rates at 7° and 11° along the low-contrast aperture tuning profile in Fig. 7A).
Figure 8A,B shows the surround tuning profiles of 2 domain neurons. What is immediately apparent is a flattening of the suppression tuning curves under low center contrast (gray trace). At low center contrast, the ortho-oriented surround stimuli (⊥) are now as effective in suppressing the cells’ center response as the iso-oriented surrounds (star), unlike for high center contrast (black trace). As such, ISD values at low center contrast are substantially reduced; high- (black value) to low- (gray value) contrast ISDs drop from 0.43 to −0.04 in Figure 8A and 0.51 to 0.22 in Figure 8B.
Figure 8C,D shows the surround profiles of 2 pinwheel cells. For the first pinwheel cell (Fig. 8C), surround suppression is fairly well tuned at high center contrast, with an ISD of 0.56. At low center contrast, however, surround suppression is equal in strength for most surround orientations, yielding a much lower ISD (−0.07). For the second pinwheel cell (Fig. 8D), surround suppression at low contrast is also less orientation tuned compared with the high-contrast condition. The ISD values are again substantially lower at low center (−1.88) compared with high (0.11) center contrast.
The observed flattening of the surround suppression tuning under low center contrast is reflected in the normalized tuning curves for both domain (n = 33) and pinwheel (n = 33) cell populations (Fig. 9A) and is primarily due to increases in suppression by the nonpreferred orientations. That is, compared with the high center contrast condition (Fig. 9B), iso-orientation suppression under low center contrast (Fig. 9A) does not change significantly in strength for domain or pinwheel cells (P > 0.4; see comparisons to the left of each set of bar graphs in Fig. 9C). On the other hand, ortho-orientation suppression strength significantly increases for both domain (P = 0.007) and pinwheel (P = 0.009) neurons when center contrast is reduced (Fig. 9C). These results suggest that broad orientation tuning of suppression under low-contrast center conditions is invariant to position in the orientation map.
High-Contrast Center/High-Contrast Surround: Removal of Near Surround
Because of the expansion of the receptive field at low contrast, by definition, the near surround has been incorporated into the center. By default, removal of the near surround leaves only the far component of the suppressive surround (Fig. 7A; see Angelucci et al. 2002; cf. Ichida et al. 2007). One interpretation of the broad suppression observed under the low-contrast center/high-contrast surround stimulus combination (Fig. 9A) is that the far surround lacks orientation tuning. However, it is also possible that lowering the center gain by reducing contrast contributes to the observed broadening of surround suppression so that weak suppression from nonoptimal surround orientations are now just as effective in suppressing the center response (Cavanaugh et al. 2002b). To address the effects of removing the near surround without the confound of reduced center gain, we measured orientation tuning using center-surround stimuli with the near component of the surround removed for a subset of 46 neurons that showed low-contrast summation of the CRF. In these trials, however, the center contrast was kept high and constrained to the smaller, high-contrast defined center size (Fig. 10A). With this configuration, there is a gap between the outer edge of the CRF and the inner edge of the surround. To be clear, this gap represents the removal of the near surround, and only the far surround remains (see Discussion for issues related to defining the full extent of the far surround). In the example shown in Figure 10A, the near surround gap is 2°. Supplementary Figure 1 shows the size of the near surround in relation to the size of the full suppressive surround for our sample.
Figure 10B–E shows 4 cell examples where the tuning of the surround with the near component removed (Δnear) is compared with the tuning of the full suppressive surround. For the 2 domain cells shown (Fig. 10B,C), removal of the near surround results in a decrease in iso-orientation suppression (stars), whereas ortho-orientation suppression strength (⊥) is either only slightly reduced (Fig. 10B) or unchanged (Fig. 10C). As a result, full to Δnear ISD values for the 2 domain cells decrease substantially (from 0.76 to 0.32 in Fig. 10B; from 0.51 to 0.06 in Fig. 10C). For the first pinwheel cell (Fig. 10D), the removal of the near surround causes some broadening in the tuning of suppression, resulting in a slight drop in ISD (full ISD = 0.56, Δnear ISD = 0.35). The pinwheel neuron in Figure 10E lacks surround orientation tuning when stimulated with the full surround (full ISD: 0.07). For this cell, masking of the near surround causes an attenuation of suppression strength across almost all orientations including iso-orientation, and as such, the ISD remains relatively low (Δnear ISD: 0.17).
The scatter plot in Figure 11A compares the ISDs for each domain (n = 24) and pinwheel (n = 26) neuron with (abscissa) and without (ordinate) the presence of the near surround. For domain cells, the exclusion of the near surround causes a significant 40% reduction in mean ISD (0.37–0.21, P < 0.01; black arrows in Fig. 11A). As illustrated in the average suppression tuning curves in Figure 11C and the left pair of bar graphs in Figure 11B, this drop in ISD for domain cells is due to a highly significant reduction in iso-orientation suppression (0.67–0.48 in Fig. 11B; P < 0.01). The robustness of this effect was confirmed by running Monte Carlo permutations (Supplementary Fig. 2A). What is important to note is that there is no significant change in ortho-orientation suppression (P = 0.67; Fig. 11B; also see Fig. 11C), and this was also confirmed using Monte Carlo simulations (Supplementary Fig. 2B). Therefore, changes in ISD cannot be due to a general ceiling effect. On the other hand, when the near surround is removed for cells located at pinwheels (Fig. 11D), the reduction in iso-orientation suppression (0°) is much smaller and not significant (only 16%, P = 0.08; Fig. 11B; Supplementary Fig. 2C). Moreover, suppression at all nonpreferred orientations is also slightly reduced for the pinwheel cells (Fig. 11D); therefore, the mean ISD does not change (0.17–0.14, P = 0.13; Fig. 11A; also see Fig. 11D and Supplementary Fig. 2D).
Based on these results, we conclude that the near surround contributes to orientation tuning of the surround through increasing iso-orientation suppression, but not by adding to suppression strength at nonpreferred orientations. Furthermore, the tuning impact of the near surround is significantly more pronounced for neurons located in domains, especially more superficially in cortex, and therefore depends on a cells’ position in the orientation map.
We find that orientation tuning of the extraclassical suppressive surround depends on the cell's location in the V1 orientation preference map. Compared with cells located in pinwheels, the orientation selectivity of the surround (SSI) and depth of suppression by an iso-oriented surround (ISD) are greater for cells located in domains. In addition, we find that the influence of the orientation map on suppression depends primarily on the smaller, near component of the surround. Because the near surround is thought to be mediated primarily through long-range lateral connections within V1, our results suggest a key role of intrinsic connectivity in the orientation tuning of surround suppression.
Pinwheel Targeting and Electrode Deviation
In our study, 4 steps were taken in order to minimize geometrical errors: First, we used maps that were highly stable and we could confirm the position of the pinwheels by comparing subsets of the orientation trials against each other (see Materials and Methods); second, to minimize parallax, we used triangulation by targeting the recording electrode on the pinwheel landmark from several vantage points; third, in some cases, the electrodes were coated with DiI so that the perpendicularity of the electrode with respect to the horizontal contour of the cortical surface and underlying layers could be confirmed through histology; fourth, neurons were grouped by electrode depth into superficial (<600 μm), intermediate (600–1200 μm), and deep (>1200 μm) regions in order to restrict the possible contamination from effects of electrode deviation encountered deeper in the penetration. We chose these ranges because they also correspond to the depths of the supragranular, granular, and infragranular layers, respectively. Indeed, we find that differences in surround suppression between domain and pinwheel neurons are greatest for superficial cells; whereas the domain/pinwheel differences in surround tuning between cells located in intermediate and deeper regions become smaller and less significant.
We do not solely attribute domain/pinwheel cell convergence in the surround orientation tuning in deeper neurons primarily to electrode deviation, as there are genuine physiological differences in surround suppression between superficial and deeper layer cells. For example, it has been reported that superficial cells show greater suppression strength (Jones et al. 2000; Akasaki et al. 2002; Okamoto et al. 2009) and higher extraclassical surround orientation selectivity (Cavanaugh et al. 2002b) compared with deeper lamina, though neither of these results were related to the V1 orientation map. In addition, anatomical differences related to orientation preference of long-range lateral connections may also exist. In superficial layers of V1 in cat and other species such as ferret, tree shrew, and monkey, long-range connections occur as refined patches predominantly between like-orientation domains (i.e., Rockland et al. 1982; Gilbert and Wiesel 1989; Gilbert 1992; Malach et al. 1993; Weliky et al. 1995; Bosking et al. 1997; Kisvarday et al. 1997; Stettler et al. 2002). Conversely, patchy long-range intrinsic connections in the infragranular layers in V1 of tree shrew (Lyon et al. 1998) and several species of nonhuman primate (Lyon and Kaas 2001, 2002a, 2002b, 2002c) appear less refined and may therefore be less orientation specific with respect to the orientation map. As discussed in the next section, the degree of orientation specificity of these connections may play a role in the suppressive surround.
In addition to physiological and anatomical evidence that would suggest better surround tuning in superficial layers, our electrode track reconstructions confirm that the angle of penetration allowed the electrode tip to stay within the effective range of a pinwheel, even at cortical depths of 2 mm. Nauhaus and colleagues (2008) have shown that a deviation of 115μm from a pinwheel center occludes the orientation tuning differences between neurons located at pinwheels and cells located at other sites in the orientation map. This translates into error tolerances of 3.3°, 5.5°, and 11° of deviation from the norm for electrode descents of 2000, 1200, and 600 μm, respectively. Furthermore, electrode deviations from the center of domains are subject to double the angle tolerances compared with pinwheel locations (ibid). Based on DiI reconstructions, our electrode penetrations had angle deviations <3.3° (Fig. 1G,H).
In conclusion, we are confident that geometrical errors in landmark targeting and electrode deviation do not significantly skew our results, especially for neurons located in the superficial depths. However, due to lack of DiI staining for all electrode penetrations, we cannot completely rule out some possible contamination through deviation in the deeper layers.
Relating Tuned Surround Suppression to Intrinsic Organization of V1
As alluded to above, a possible mediating mechanism for the orientation tuning of extraclassical surround suppression is long-range lateral connectivity within V1 (Ts'o et al. 1986; Bolz and Gilbert 1989; Hirsch and Gilbert 1991; Malach et al. 1993; Weliky et al. 1995; Somers et al. 2002; Chisum and Fitzpatrick 2004). These inputs arise from distances greater than 250 μm and therefore originate from neurons with visual field representations that are at least partially outside the target cell's CRF (Gilbert 1992; Brown et al. 2003). Importantly, these inputs can provide suppression through either direct inhibition from long-range projecting basket cells (Kisvarday et al. 1997) or indirect inhibition from horizontal pyramidal cells that synapse onto local inhibitory neurons (Ahmed et al. 1994; Anderson et al. 1994).
However, a number of studies have emphasized additional sources for the extraclassical suppressive surround: feedforward inputs relayed from the lateral geniculate nucleus (LGN) and feedback connections from higher visual areas. Feedforward mechanisms can account for suppression that occurs too quickly to rely on cortical processes (Webb et al. 2005; Xing et al. 2005; Alitto and Usrey 2008; Liu et al. 2011), whereas feedback from higher cortical areas can account for extraclassical receptive field sizes that are spatially too big to be covered by monosynaptic long-range lateral connections within V1 (Salin et al. 1992; Bullier et al. 2001; Angelucci et al. 2002; Cavanaugh et al. 2002a; Levitt and Lund 2002; Cantone et al. 2005; Schwabe et al. 2006; cf. Liu et al. 2011). As proposed for surround suppression mediated through long-range lateral connections, suppression in V1 through feedback can also be accounted for by synapses onto local inhibitory neurons (Anderson and Martin 2009; Schwabe et al. 2010).
Roles of Long-Range Lateral and Feedback Connections
In comparison to feedback connections, our results are more supportive for long-range lateral connections playing a role in the orientation tuning of extraclassical surround suppression in 2 ways. First, we show that the orientation selectivity (and relative tuning strength) of the suppressive surround is significantly better for domain neurons than for pinwheel cells (Figs 4–6). One explanation for this is that inputs corresponding to the extraclassical receptive field, such as long-range lateral inputs or feedback from higher order areas, arise predominantly from neurons preferring the iso-orientation. As noted above, these inputs, if excitatory, could provide suppression through contacts onto local inhibitory neurons (Ahmed et al. 1994; Anderson et al. 1994; Anderson and Martin 2009; Schwabe et al. 2010). Extensive work in cat (Gilbert and Wiesel 1989; Kisvarday et al. 1997), ferret (Weliky et al. 1995), tree shrew (Rockland et al. 1982; Bosking et al. 1997), and monkey (Malach et al. 1993; Stettler et al. 2002) has shown that domains preferentially receive inputs from like-oriented domains within V1. However, limited work in monkey relating the orientation preference of feedback from V2 to V1 has yielded conflicting results (Stettler et al. 2002; Shmuel et al. 2005). Therefore, whether or not feedback is orientation tuned remains controversial. Moreover, the orientation preference of feedback to V1 from higher order areas beyond V2 in any species has not been reported. Therefore, based on existing evidence, intrinsic long-range connections in V1 is a good candidate for mediating orientation tuned suppression.
Second, we show that masking of the near surround significantly broadens orientation tuning of the suppressive surround for domain neurons. At first glance, a key factor arguing against a major contribution from long-range lateral connections is the limited region of the extraclassical receptive field that these inputs represent, which at least in monkey V1 is considered to correspond only to the relatively small region of the near surround (Angelucci et al. 2002; Levitt and Lund 2002). In cat (Salin et al. 1992; Li CY and Li W 1994) and ferret (Cantone et al. 2005), similarly small estimates of the receptive field extent covered by long-range lateral connections can also be made. Despite the relatively small size of the near surround (see Supplementary Figure 1), we show that it nevertheless has a large and significant effect on the orientation tuning of the surround suppression for domain neurons (Fig. 11C). Importantly, for these cells, when the near surround is masked, orientation tuning of suppression significantly broadens. This is due to a significant reduction in suppression by the iso-orientation but no significant change in orthogonal orientation (Fig. 11B). This finding therefore links the presence of the near surround specifically to iso-orientation suppression and the remaining suppressive surround, which we refer to as the “far” surround, to suppression at both iso- and ortho-orientations.
In this way, the suppression from the near surround of domain cells is more sharply tuned to orientation than the far suppressive surround. This supports the finding that preferential connectivity between like-oriented domains occurs within V1 but that feedback connectivity can be more broadly tuned with respect to orientation domains (Stettler et al. 2002). However, as noted above, others have found feedback from V2 to be tuned with respect to the orientation map (Shmuel et al. 2005). In addition, studies have shown that inactivation of higher visual area feedback in cat can reduce the orientation selectivity of surround suppression in V1 (Bardy et al. 2009). Therefore, more investigation is needed to determine how well orientation tuned feedback connections really are. In addition, our far suppressive surround results should be interpreted with the understanding that other recent investigations have estimated far surround size using inward expanding surround annuli, which can provide larger estimates and may also include facilitative modulation of the center response (Ichida et al. 2007; Shushruth et al. 2009; Schwabe et al. 2010). We used grating summation tests and chose the largest outer diameter that evoked the maximum suppression; thus, our estimates of the far surround did not include the facilitative component that can occur more distally (see, e.g., the reduction in suppression for the largest aperture sizes in Fig. 10A). While the orientation selectivity of the far surround determined through inwardly expanding annuli has not been reported, it is possible that they could yield different orientation tuning of surround suppression than we report here.
Finally, in contrast to domains, for cells at pinwheel centers, the masking of the near surround results in a general reduction in suppression for all surround orientations with no significant change in the already broad orientation tuning (Fig. 11D). This suggests that the near surround of pinwheel cells is not as well tuned to orientation and is consistent with evidence that long-range inputs arise from a broader array of orientation preferences (Yousef et al. 2001; Buzas et al. 2006).
Role of Feedforward Circuits
It is clear that feedforward inputs from LGN neurons in cat and monkey also exhibit extraclassical surround suppression (Solomon et al. 2002; Nolt et al. 2004). Some forms of surround suppression in the LGN can occur too rapidly to be a product of feedback from cortex (Alitto and Usrey 2008) and can even occur when V1 activity is silenced (Sceniak et al. 2006). Furthermore, broadly tuned suppression in V1 occurs very early in time, suggesting it is mediated through the faster propagating feedforward afferents (Xing et al. 2005). However, Xing et al. (2005) have also shown that orientation tuned suppression arises about 15 ms later, which is consistent with the slower propagation speed of unmyelinated long-range horizontal projections originating from 1.5 to 3 mm away (0.1–0.2 mm/ms; Grinvald et al. 1994; Bringuier et al. 1999). In addition, examination of receptive field spatial summation of cat V1 neurons suggests that earlier arising suppression is likely to result from feedforward afferents and is weaker than later suppression likely mediated by cortical circuits (Liu et al. 2011). Moreover, at least in monkey V1, the extent of the extraclassical surround thought to be covered by most feedforward afferents is substantially smaller than the range covered by long-range lateral connections (Angelucci et al. 2002; Angelucci and Sainsbury 2006). Taken together, this recent literature suggests a more limited role for feedforward afferents in the orientation tuning of near surround mediated suppression.
Surround Mechanisms Involved during Low-Contrast Center Conditions
Thus far, we have discussed our results for high-contrast center-surround stimuli. Under these conditions, we show that when the near surround is removed, the orientation tuning of the suppressive surround is substantially and significantly decreased for domain neurons, but not for pinwheel neurons. In addition, a moderate, but more broadly tuned, suppression remains due to the presence of the far surround. In this report, we also find that a high-contrast extraclassical surround leads to broad suppression (nearly flat) of responses to low-contrast center stimuli for both domain and pinwheel cells (Fig. 9). This suggests that surround suppression of a V1 cell is invariant to position in the orientation map when center gain is low.
For our population, the reduction in the orientation selectivity of the suppressive surround was due to a significant increase in ortho-suppression, with no significant change in iso-suppression (Fig. 9C). Though not involving orientation maps, similar results have been shown for monkey V1 where the strength of orthogonal surround suppression increases substantially, while iso-orientation suppression increases as well, but to a lesser degree (Levitt and Lund 1997; Cavanaugh et al. 2002b). The most likely explanation for this effect is that a reduction in center gain, due to lowering center contrast, allows the more weakly suppressive orthogonal surrounds to exert greater suppression on the center response (ibid). However, models by others have suggested that low center gain should lead to weaker surround suppression (Schwabe et al. 2006), and recent experiments show that suppression by the far surround is actually stronger when the center contrast is high, not when it is low (Schwabe et al. 2010). In our data, we can also compare suppression strength of the far surround when center contrast is high (Fig. 11B) to suppression strength when center contrast is low (Fig. 9C). For domain cells, far surround iso-suppression strength now slightly increases from 0.49 to 0.59 when center contrast decreases, though this effect is not significant (P = 0.12). While this result is consistent with the findings of Levitt and Lund (1997) and Cavanaugh et al. (2002b), it still conflicts with Schwabe et al. (2010). A likely reason for this discrepancy is the different criteria used to select the low contrast. In the latter study, low contrasts were selected that elicited responses 2 standard deviations above spontaneous levels. We chose low-contrast values that evoked 30% of maximum response, and Cavanaugh et al. (2002b) chose low contrasts that evoked at least 20% of maximum, both of which could have led to relatively higher low contrasts compared with those used by Schwabe et al. (2010). Based on their model (Schwabe et al. 2006, 2010), these different contrast levels would then explain the discrepancy. Moreover, the model predicts that for mid-level low contrasts, slight misalignments of the center receptive fields can alter the gain and have a dramatic affect on suppression strength, and this could be another factor in the discrepancy.
Interestingly, Webb et al. (2005) suggest that broader orientation tuning of surround suppression under low-contrast center conditions is dependent on feedforward mechanisms. This conclusion was based on observations that surround suppression was not susceptible to contrast adaptation and showed little interocular transfer. Feedforward-mediated extraclassical suppression is also supported by other evidence showing that untuned suppression does not require cortical activity (Sceniak et al. 2006), it arrives relatively early (Xing et al. 2005; Alitto and Usrey 2008), and it may even be relayed from retinal ganglion cells (Solomon et al. 2006; Alitto and Usrey 2008).
Influence of Orientation Map on Center Selectivity
While the main focus of our results is on extraclassical surround suppression, we are also the first to report that orientation selectivity of the CRF (CSI) differs significantly between pinwheel and domain neurons, with domain neurons showing higher selectivity (Fig. 2I,J). This is not entirely unexpected since a recent report found that orientation tuning widths can be much broader for neurons located closer to pinwheel centers (Nauhaus et al. 2008). Nevertheless, previous work had found no significant pinwheel/domain differences in orientation selectivity (Maldonado et al. 1997; Schummers et al. 2002, 2007; Mariño et al. 2005). One reason for this discrepancy may be due to the comparatively small sample size of cells in Schummers et al. (2002) and Mariño et al. (2005) because of the demanding nature of the whole-cell recording techniques they employed. Even in our large data set, we find a high degree of overlap in the selectivity of domain and pinwheel cells. Therefore, a smaller sample size may have been less likely to yield a significant difference.
An additional important difference between our study and earlier studies, including studies involving higher cell numbers (i.e., Maldonado et al. 1997; Schummers et al. 2007), is the difference in stimulus size employed. We restricted our stimuli to the carefully defined CRF, whereas previous studies used stimuli that filled the entire display field. From previous work comparing orientation tuning widths, it is now established that stimulus size significantly affects orientation tuning of V1 cells in cat (Chen et al. 2005; Okamoto et al. 2009) and monkey (Xing et al. 2005). This difference in stimulus size could therefore have skewed earlier results as both pinwheel and domain cells would have been more sharply tuned.
Impact of Contrast on Center Selectivity
It is well documented that the CRF is not a passive filter, and stimulus contrast plays a critical role in determining its size (Sceniak et al. 1999; Angelucci et al. 2002; Levitt and Lund 2002; Angelucci and Sainsbury 2006; Tailby et al. 2007). In agreement with these previous studies done in monkey and cat V1, for our cells, the reduction of contrast generally resulted in an expansion of the center receptive field size. On average, we observed a 36% increase in the receptive field size when stimulus contrast was reduced to a contrast that evoked 30% of optimal response. Furthermore, compared with high contrast, under low contrast, the CSI values for the pooled population did not systematically change (Fig. 7C).
At first glance, our results appear to contradict the findings of others showing an inverse relationship between the circular variance of the orientation response and contrast in ferret and monkey V1 (Alitto and Usrey 2004; Johnson et al. 2008). One possible reason for this discrepancy, however, is that our stimuli were confined to the CRF (defined at each contrast), whereas the previous studies used full field stimuli (ibid). As already noted above, inclusion of the extraclassical suppressive surround can significantly sharpen orientation tuning widths (Chen et al. 2005; Xing et al. 2005; Okamoto et al. 2009), and this could also have had an effect on the selectivity.
The physiological mechanism of contrast-dependent receptive field spatial expansion still remains to be conclusively resolved as it has been observed in the retina (Nolt et al. 2004), LGN (Solomon et al. 2002; Nolt et al. 2004; cf. Sceniak et al. 2006), and in the visual cortex (Sceniak et al. 1999; Angelucci et al. 2002; Cavanaugh et al. 2002a; Tailby et al. 2007). Nonetheless, it has been proposed that cortical networks are mainly responsible for the expansion of the receptive field at low contrasts since the effect seems to be considerably larger in cortex (Sceniak et al. 1999) than in the LGN (Solomon et al. 2002; Nolt et al. 2004; Sceniak et al. 2006), and pharmacological inactivation of V1 has been shown to cancel the effect in LGN afferents (Sceniak et al. 2006). As discussed earlier, long-range lateral connections within V1 are thought to provide the anatomical basis for this phenomenon (Angelucci et al. 2002; Angelucci and Sainsbury 2006).
Significance of Modular Organization
We conclude that neural properties related to stimulus orientation, such as the center response and near surround suppression, are affected by modular organization of orientation preference (orientation maps) within V1. Because these maps are only present in a select group of highly visual mammals that include carnivores, tree shrews, and primates, but exclude rodents (Lyon 2007; Van Hooser 2007), our results provide insights into functions that manifest in more complex cortical organization. It would be interesting to determine whether even more complex interactions with orientation processing occur in the monkey, particularly in relation to other forms of modular organization such as cytochrome oxidase blobs (Livingstone and Hubel 1984) and other functional characteristics such as color (Johnson et al. 2008; Lu and Roe 2008). In addition, a detailed comparison of center and surround properties of V1 neurons in highly visual species without orientation maps, such as squirrels (Van Hooser et al. 2005), would also be an important next step to help us better understand the evolutionary advantages of the intrinsic architecture underlying orientation preference.
Whitehall Foundation (#2008-05-33, #2009-12-44) to D.C.L.
We are grateful to Nelson Espinoza and Roger Geertsema for assistance on initial experiments; to Daniel Shima, Jorge Mariño, and S. Sushruth for technical advice; and Xiangmin Xu, Marcel Stimberg, Ian Nauhaus, and Emily Grossman for helpful comments on the manuscript. Conflict of Interest: None declared.