Cerebral Cortex Advance Access originally published online on August 18, 2004
Cerebral Cortex 2005 15(4):460-478; doi:10.1093/cercor/bhh148
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Cerebral Cortex V 15 N 4 © Oxford University Press 2004; all rights reserved
Functional Architecture of Retinotopy in Visual Association Cortex of Behaving Monkey
1 Center for Molecular and Behavioral Neuroscience, Rutgers University, Newark, NJ, USA, 2 Present address: Institute of Physiology, Medical School of Pecs, Hungary
Address correspondence to Ralph M. Siegel, Center for Molecular and Behavioral Neuroscience, Rutgers University, 197 University Avenue, Newark, NJ 07102, USA. Email: axon{at}cortex.rutgers.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
While the receptive field properties of single neurons in the inferior parietal cortex have been quantitatively described from numerous electrical measurements, the visual topography of area 7a and the adjacent dorsal prelunate area (DP) remains unknown. This lacuna may be a technical byproduct of the difficulty of reconstructing tens to hundreds of penetrations, or may be the result of varying functional retinotopic architectures. Intrinsic optical imaging, performed in behaving monkey for extended periods of time, was used to evaluate retinotopy simultaneously at multiple positions across the cortical surface. As electrical recordings through an implanted artificial dura are difficult, the measurement and quantification of retinotopy with long-term recordings was validated by imaging early visual cortex (areas V1 and V2). Retinotopic topography was found in each of the three other areas studied within a single day's experiment. However, the ventral portion of DP (DPv) had a retinotopic topography that varied from day to day, while the more dorsal aspects (DPd) exhibited consistent retinotopy. This suggests that the dorsal prelunate gyrus may consist of more than one visual area. The retinotopy of area 7a also varied from day to day. Possible mechanisms for this variability across days are discussed as well as its impact upon our understanding of the representation of extrapersonal space in the inferior parietal cortex.
Key Words: extrastriate visual cortex • optical imaging • parietal cortex • spatial perception • visual fields • visual pathways
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The inferior parietal lobule of the macaque monkey is comprised of a multitude of visual areas, each of which has neurons with large receptive fields (Steinmetz et al., 1987
; Ben Hamed et al., 2001). Various authors have suggested that the inferior parietal lobule plays an important role in the transformation of a retinotopic to an egocentric coordinate system or in the planning of motor movements (Andersen et al., 1990; Stein, 1992; Siegel and Read, 1997b; Previc, 1998; Snyder et al., 1998). For example, retinotopic signals are combined with eye position signals to generate a novel intermediate representation. Studying and comparing the precise visual topography in those areas can clarify the nature of transformation from a retinotopic to an egocentric representation.
Electrophysiological and optical imaging studies have shown that the dorsal prelunate area (DP) and area 7a of the macaque inferior parietal lobule have ‘gain fields’, i.e. show differential activation with varying angles of gaze (Andersen et al., 1985, 1990; Read and Siegel, 1997; Siegel et al., 2003). Further electrophysiological measurements have been used to demonstrate that the retinotopic location of stimuli is encoded in area 7a neurons (Andersen et al., 1985, 1990; Motter et al., 1987; Read and Siegel, 1997; Merchant et al., 2001). However, none of these studies have demonstrated a topographic organization of retinotopy across the cortical surface, either due to the technical inadequacy of long-term electrical recordings, or due to the intrinsic variability in the distribution of receptive fields. Thus, the retinotopic organization in the upper layers across the two-dimensional surface of the inferior parietal lobule was investigated using intrinsic optical imaging in the behaving monkey for extended periods of time. Intrinsic optical imaging has the advantage that it allows repeated measurements of cortical activity over tens of square millimeters. The mechanism underlying intrinsic imaging is mainly an initial increase in deoxyhemoglobin within the first 3 s of stimulus onset, which leads to decreased reflectance of light at wavelengths in the 500–600 nm range (Malonek et al., 1997).
Optic flow visual stimulation was utilized as it is a powerful cue for spatial location, and area 7a neurons are selective to these stimuli (Read and Siegel, 1997; Siegel and Read, 1997a). Analogous to gain field optical studies (Siegel et al., 2003), it was hypothesized that the position of the stimulus within the visual field should modulate the spatial distribution of the optical signal if the area under study has an orderly retinotopic organization. Areas that possess little or no retinotopy should yield noisy optical maps without systematic variation due to retinal stimulus location.
Single-unit recordings are often used to verify the results of optical recordings (Grinvald et al., 1986; Arieli et al., 1995; Maldonado et al., 1997; Ramsden et al., 2001; Landisman and Ts'o, 2002). Others have shown examples of electrode penetrations through artificial dura (Arieli and Grinvald, 2002; Arieli et al., 2002). While we have confirmed gain fields using a few electrical recordings (Siegel et al., 2003), it is our experience that introducing a low-impedance electrode through the artificial dura leaves a small hole and can damage the underlying pia and cortex due to the presence of a ‘neomembrane’ (Arieli et al., 2002; Chen et al., 2002; Siegel et al., 2003). Avoiding electrode penetration is one factor that has allowed us to optically record up to four years' data in one animal, enabling repeated measurements of cortical functions. Thus, we needed to see if it was possible to study visual topography optically without using microelectrodes. Areas V1 and V2 were included, as their visual topography is well known. In those areas, the retinotopic organization was confirmed in behaving monkeys with stimuli consisting of elongated vertical stripes near and on the vertical meridian. Appropriate retinotopic activation patterns were also observed with the optic flow stimuli in areas V1 and V2. The study of the inferior parietal lobule (dorsal prelunate and 7a) showed a coarser visual topography across the cortical surface that varied from day to day.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The intrinsic reflectance signal of the dorsal prelunate cortex (DP) and area 7a was studied in two male rhesus monkeys (Macaca mulatta, M1R and M2L). Additional recordings were made over areas V1 and V2 posterior to the lunate sulcus. Those areas were exposed within the optical chambers of both animals and served as a control for our methods. The two animals are the same as described in an earlier study (Siegel et al., 2003
). All procedures were carried out in accordance with the Rutgers University Animal Institutional Review Board and the NIH Guidelines for the Care and Use of Laboratory Animals. Surgical Procedures
After the initial training on the fixation task, a head post was implanted. Standard surgical procedures were performed under sterile conditions as described previously (Siegel et al., 2003). The head post was machined from a single stainless steel block with a 50.8 x 25.4 x 6.35 mm mounting surface to provide the exceptional rigidity needed for optical studies (Fig. 1B). The head post was embedded in Palacos orthopedic cement (Smith & Nephew, Richards, Memphis, TN), which in turn was secured to the skull with up to 24 titanium cranio-maxillofacial screws (Synthes, Paoli, PA). In subsequent surgery, a stainless steel recording chamber (20 mm inner diameter) was placed over the inferior parietal lobule based on magnetic resonance images. Within the chamber, a craniotomy and durotomy were performed to expose the cortical surface. An artificial dura was inserted into the craniotomy according to published methods (Shtoyerman et al., 2000; Arieli et al., 2002; Siegel et al., 2003). With this artificial dura covering and protecting the cortex, we were able to collect intrinsic images over a period of >3 years in M1R. In M2L, however, the growth of a neomembrane over the cortex precluded imaging after 
8 months.
|
'' screws to a steel plate bolted to a CX-95A carrier attached to a horizontal Newport X-95 beam that provides excellent rigidity. Middle panel: a front view of the assembly on the floating air table including the primate chair in the lower portion of the image. The CCD camera and lenses are in the upper left portion of the image. The monkey's head is attached to the horizontal X-95 rail in the front. Right panel: a closer view of the camera and lens system, which is secured on a X-26 steel rail that can be slid up and down. This sliding rail in turn is attached firmly to AX-95 devices, whose angle can also be adjusted. Two Nikon 50 mm f1.2 lenses are illustrated. (C) Anatomical location of optical chambers for both monkeys (left, M2L; right, M1R) superimposed on structural MRIs. Within one chamber, two or three exposures were selected: one posterior centered over the lunate sulcus (LS), which included areas V1/V2 and DPv (lower insets with angioarchitectonic maps at 540 nm); and one anterior at the tip of the superior temporal sulcus (STS), which includes areas DP and 7a (upper insets with angioarchitectonic maps at 540 nm). In M1R, a third camera placement slightly further posterior was used to image V1/V2. Scale bar, 2 mm.Visual Stimuli and Behavioral Task
The monkeys were trained on a detection task that required fixation on a central target. An infrared eye camera (ISCAN, Cambridge, MA) monitored the eye position at 30 Hz; fixation outside the target of >1° terminated the trial. A small fixation target appeared in the center of the screen, and the animal continued the trial by pulling a lever attached to the primate chair within a reaction time of 150–800 ms. The monkey maintained fixation on the central target throughout the trial. Two seconds after the trial began, an optic flow stimulus (expansion) was displayed in one of nine locations (3 x 3 grid) centered around the fixation target (Fig. 1A). Two eccentricities (10° and 20°) and stimulus sizes (10° and 20° diameter) for the optic flow fields were used. There were 128 dots per display for the 20° stimulus size and 32 dots per display for the 10° stimulus. The dots all moved radially away from the center of the display at a constant velocity of 6°/s. The monkey had to detect a change in the structure of the stimulus (from structured expansion flow to unstructured expansion), and release the lever within 150–800 ms while maintaining fixation on the central fixation target. The time at which the motion change occurred varied randomly between 3000 and 5000 ms after stimulus onset. A correct response was rewarded with a drop of juice. For most of the experiments, visual topography was mapped using the optic flow stimulus.
In addition, a simpler stimulus was used to verify our ability to obtain fine scale topography in areas V1 and V2 in M1R. Elongated thin bars were presented within the lower contralateral (left) quadrant to map the representation of the vertical meridian. The bar stimulus (1° wide, 12° long) was presented at one of four eccentricities (0°, i.e. centered on the lower vertical meridian, and –1°, –2°, –3° off to the left visual field parallel to the lower vertical meridian; see Fig. 4B). The bar consisted of 128 moving dots that moved at a constant velocity of 6°/s either up or down. Dots in all stimuli had a staggered point life of 532 ms (Siegel and Read, 1997a). While the dots themselves moved, the bar itself was stationary. In this task, the monkey had to maintain fixation and respond to the dimming of the fixation square (0.5° x 0.5°) by releasing the lever. There was no change in the trajectory of the dots. The 10% dimming occurred at random times between 4000 and 5000 ms after fixation onset. In this experiment, a blank condition was included which consisted of the fixation dot alone on the blank screen. Thus, in the elongated thin bar stimulus task, the monkey responded only to the fixation point and his attention was always at the center of fixation.
|
A blank condition was not used in the optic flow task for the following reasons. In studies of area 7a, evidence was found that the attentional state is important (Siegel and Read, 1997a
; Read and Siegel, 1997). To ensure maximal visual responses from the attentional system, and to keep the attentional load consistent, the monkey always attended to the optic flow stimuli. This decision was initially made for the Siegel et al. (2003) optical imaging study and is continued here. If a blank condition were to be used, this would entail displaying only a fixation point to which the animal would direct his attention. A comparison between the blank and the experimental optic flow condition could be made; however, the comparison would not be valid because the effect of the stimulus condition and the attentional state would be confounded. Intrinsic Optical Imaging
For the optical imaging recordings, the monkey chair and the monkey's head were attached to a floating air table (Newport, Irvine, CA) via the stainless steel head post, various clamps and an assembly of X95 rails (Newport, Irvine, CA) (Fig. 1B). During data collection the table was floated and the monkey chair did not contact the ground. To stabilize the cortex and dampen pulsations, the chamber was filled with 0.9% sterile saline and hydraulically sealed with a silicon washer and glass window. Images were acquired using the Imager 2001 System (Optical Imaging Co., Rehovot, Israel) with 605 nm illumination, focusing 500 µm below surface vessels. A modified tandem lens system (Ratzlaff and Grinvald, 1991; Siegel et al., 2003) was used to magnify the imaged region. In this lens system, an inverted image in the focal plane of the Nikon 50 mm f1.2 objective lens was imaged by a Nikkor 60 mm macro lens. Alternatively, the Nikkor lens was replaced by an additional Nikon 50 mm f1.2 lens spaced from the camera to act as a macro lens (Fig. 1B). This system permitted a working distance of up to 80 mm and variable magnification. The Imager 2001 system collected 756 x 480 pixel images, which were binned on-line by 2 for the analysis; no other filters were applied to the data. For each stimulus condition, 37 consecutive frames at 7 Hz were collected. Signal-to-noise ratio was improved by trial averaging. Typically, one experiment consisted of 1–3 imaging runs per day, which resulted in 30–90 trials per stimulus position in the optic flow experiment. Approximately 60–100 trials were collected per stimulus position in the elongated thin bar experiment. Image data were collected for all trials, but trials with incorrect responses or artifacts (e.g. excessive motion) were later rejected for analysis (Siegel et al., 2003). Experiments with poor performance or high variability in a plot of the mean to the standard deviation of the optical signal (Siegel et al., 2003) were excluded.
As it was not possible to place the camera and head in identical positions for each experiment, the imaged regions varied slightly from day to day. However, by using the blood vessels as landmarks, the maps for each day were aligned via rotation and translation. This was achieved by superimposing the blood vessel maps in Photoshop (Adobe Systems, San Jose, CA). On the first map of the series of experiments the major blood vessels were outlined and a mask of the vessels was created. The subsequent blood vessel maps were then aligned with that mask using translation, rotation and magnification where appropriate. The parameters from this alignment were then applied to the optical maps.
Chamber Placements
Monkey M1R had a chamber over the inferior parietal lobule of the right hemisphere (Fig. 1C, right). This monkey's chamber implant and artificial dura proved to be of exceptional stability, and the tissue growth over the cortex was minimal, which allowed us to repeatedly image this animal over a period of >3 years using a series of tasks, some of which are reported elsewhere (Siegel et al., 2003). Monkey M2L had a chamber over the left hemisphere in a slightly more posterior location than M1R (Fig. 1C, left). In this animal, a neomembrane grew rapidly and covered the cortex within several months. While this neomembrane was almost opaque to green (540 nm) illumination, it was sufficiently transparent to red (605 nm) light for 8 months (Fig. 7A) to allow intrinsic imaging and alignment of the larger blood vessels. In both monkeys, areas 7a and DP were exposed within the chamber, as well as early visual areas posterior to the lunate sulcus (LS). Two sections within the optical chamber were targeted, one more anterior location including area 7a and more dorsal portions of the dorsal prelunate (DPd) on either side of the superior temporal sulcus (STS), and a more posterior location on either side of the LS, including V1 and V2 and ventral portions of DP (DPv). (The nomenclature for the DP subregions was chosen to be descriptive; the correspondence with actual visual areas is discussed where appropriate.) Areas V1 and V2 will be referred to as ‘V1/V2’.
|
Analysis of Optical Imaging Data
As the optical imaging signal develops >2000 ms after stimulus onset, the analysis uses the reflectance signal within the last 2000–3000 ms of the stimulation period (Grinvald et al., 1986; Shtoyerman et al., 2000; Siegel et al., 2003). Baseline normalized single condition maps were always computed to show the individual activation to one stimulus condition. The second half (i.e. 1000 ms) of visual fixation (see Fig. 1A) preceding the onset of visual stimulation was used to normalize the maps. In the elongated thin bar experiment, blank subtraction was also performed. The blank subtraction was computed as the difference in two baseline normalized maps, i.e. the blank condition map (fixation point only) was subtracted from the stimulus condition map (elongated thin bar plus fixation point).
In the optic flow experiments using nine different stimuli, two regression models were used in order to determine the location and shape of the region of the visual field that would activate each pixel. Both models were chosen from the class of general linear models, as these cover the range of receptive field shapes reported in the dorsal stream, and whose parameters can be estimated using closed forms (Read and Siegel, 1997; Anderson and Siegel, 1999). Such an approach was chosen to match the underlying neural retinotopic response as closely as possible by regarding each pixel as a ‘neuron’. Two models were used here. The first was a purely linear model for both the horizontal and vertical position of the stimulus. Such receptive fields have been reported in area 7a (Read and Siegel, 1997). The linear regression was performed independently for each of the 378 x 240 pixels with the equation
 | (1) |
x(I,J) and y(I,J) are the slopes of the regression for the horizontal and vertical coordinates (x and y), ß(I,J) is the intercept, and 
i(I,J) the error value. To demonstrate the dependence of the intrinsic signal on the x- and y-position of the stimulus, the angle (in 360° space) and the amplitude of the x- and y-vectors were computed from the rectangular coordinate system of (x, y). As the optical signal is inversely related to neural activity (Grinvald et al., 1986; Frostig et al., 1990; Shtoyerman et al., 2000), all parameters were multiplied by –1. The angles were color coded in the functional maps. Thus, each color corresponds to an angle (e.g. 0° or 360°, cyan; 180°, red) within the visual field for which the maximal response would be obtained. To quantify the visual field representation within the maps and to compare them between experiments, regions of interest were selected on the optical maps. Although the imaged region varied between experiments, the same region of interest could be easily located on each map based on the characteristic angioarchitecture. Typically, two locations per cortical area (50 x 50 pixels, equivalent to 1–1.3 mm) were selected, typically one on the medial and one on the lateral part within the chamber. Large blood vessels were avoided, as they may be modulated by the visual signal (Siegel et al., 2002). Thus, for each region of interest, the average x- and y-slopes from the linear regression analysis were computed, and the corresponding angle and amplitude of the vector calculated. The resulting angle measurements were analyzed with the circular statistics program Oriana (Kovach Computing Services, Anglesey, UK).
While linear receptive fields have been reported in the parietal cortex, peaked receptive fields were also observed (Read and Siegel, 1997). A linear regression might be thought inappropriate for V1/V2 to model the retinotopic response. Thus, a second model from the class of general linear models was selected for modeling the retinotopic response. A quadratic function (i.e. second-order linear model)
 | (2) |
xx(I,J) and yy(I,J) as the quadratic coefficients. Its shape can be either a ‘peak’ (or ‘valley’). The width and location of the peak can be evaluated from the regression coefficients. While it is an implicit assumption that visual neurons would encode the position of a stimulus mostly as a peaked function, i.e. a decline in neural activity as the ‘optimal’ stimulus moves away from the receptive field center (Hubel and Wiesel, 1968; Schiller et al., 1976), more complex receptive fields have been reported in the dorsal stream (Motter et al., 1987; Pigarev et al., 2001, 2002). Conveniently the quadratic fit accommodates these more complex shapes of retinotopic activation. The signs of the quadratic coefficients determine the shape of the quadratic function (negative, peak; positive, valley). Various combinations of positive and negative quadratic x- and y-coefficients yield various shapes (peak, valley and saddles). Figure 2 illustrates the various shapes that can be derived from this combination of x- and y-coefficients.
|
In order to display these multi-dimensional data, masks were created to reveal the dominant sign of the quadratic coefficient. For each x- and y-coordinate, it was first determined whether the quadratic coefficient was negative or positive. This allowed the generation of two types of mask (positive, ‘valley’ or negative, ‘peak’) for each data set. For example, an x-peak mask displays only those pixels that have a negative sign of the x-quadratic coefficient, and masks pixels with a positive sign.
The width of the quadratic function is an estimate of the tuning width for stimulus position (Anderson and Siegel, 1999). A large absolute value of the quadratic coefficient corresponds to a narrow peak or valley, whereas a small coefficient indicates a broader tuning. The width of the modeled activated region of cortex was calculated from the equation 
X50 is the shift in position along the horizontal axis from the quadratic center that results in a 50% change in firing rate from the peak. Similar calculations were made for the vertical coordinate y. Thus, for the selected regions of interest one can also calculate the average width of the activated region along the x- and y-coordinates.
The third aspect of the quadratic analysis is the location of the function's peak within the two-dimensional x- and y-space. The combination of the linear and quadratic coefficients provides the location of the centers of peaks (or valleys) for each parabola using the equation –x/2xx for the horizontal coordinate, with a similar equation for the vertical axis (Anderson and Siegel, 1999). With this formula, the location of the modeled center of activation can be computed within the visual field. Again, these rectangular coordinates are transformed into polar coordinates to create the angle maps. For each region of interest the x- and y-coordinates of the centers are used to locate the preferred visual field location.
Comparison of the First- and Second-order General Linear Models
Both the linear and the quadratic fits were performed for every pixel. This raises the question about which was the better fit. One might simply use the variance accounted for (i.e. the normalized sum square error) on a pixel-by-pixel basis. However, this criterion is flawed, as increasing the number of parameters necessarily increases the variance accounted for (i.e. the R2). Therefore, the addition of parameters needs to be penalized. To compare the two regression models directly, the Akaike Information Criterion (AIC) was used; this balances the number of parameters and residual variance (Akaike, 1974; Siegel and Birks, 1988). For each pixel, the AIC value was computed and the results converted into a binary map where white pixels indicate that the linear model was better, and black pixels indicate that the quadratic model was the better fit.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Behavioral Performance
For the main experiments with the optic flow stimuli, 34 experiments were performed for M1R (21 over V1/V2 and DPv, 13 over DP and 7a); 12 experiments were performed for M2L (7 over V1/V2 and DPv, 5 over DP and 7a). For both monkeys, the percentage correct performance was always >90% for all positions (M1R and M2L, mean 96%). The reaction times for the different retinal locations of the stimuli were analyzed using a stepwise regression analysis (Fig. 3). In M1R, 31 of 34 experiments showed a significant dependence on the position of the stimulus, whereas in M2L, all 12 experiments showed a significant effect of position. The coefficients of the regression were analyzed, and for both monkeys it was found that the fastest reaction times occurred when the stimulus was over the fovea, as illustrated with an experiment for M2L (Fig. 3A,B). Each monkey's dependence of reaction time on position was similar, with M1R showing slightly weaker effects as evident in the smaller regression coefficients (Fig. 3C). Thus the monkey's psychophysical performance was reliable and consistent over time within the ranges specified by the defined limits.
|
Retinotopic Mapping of Areas V1 and V2 with Elongated Bar Stimuli
To establish that retinotopic maps could be obtained under the current experimental conditions in the fixating animal, an experiment was performed in M1R using elongated thin bar stimuli defined by moving dots (Fig. 4). It would be expected that a narrow bar in the visual field should result in a band of activation across the cortical surface. With the stimulus centered on the lower vertical meridian, activation was observed 1–2 mm behind the vessel over the lunate sulcus (Fig. 4A), indeed in an elongated narrow region (Fig. 4C1). This activation pattern is consistent with the known representation of the V1/V2 border along the lower vertical meridian. When the bar was placed –1° off to the contralateral field, the activation shifted posteriorly (Fig. 4C2). There was also a slight tilt from the 0° activation, probably because the stimuli were not presented within a polar arrangement (as typically used to exploit the conformal map of cortex: Tootell et al., 1988), but rather shifted parallel to the 0° bar. In the subsequent map of the –2° bar stimulus (Fig. 4C3), the band of activation had moved further posterior so that it was barely visible within the chamber.
Overall, the retinotopic activation from the thin bar stimuli was consistent within and between experiments. This is illustrated in Figure 4D, where the one-dimensional profiles of the percentage change in reflectance signal are shown for two bar locations (0°, D1; –1°, D2) for three experiments. For the one-dimensional profiles, a section across the cortex was chosen orthogonal to the band of activation for the 0° bar (Fig. 4A,C). Between experiments (i.e. days), the pattern of activation was very stable with respect to location and amplitude of the signal for both eccentricities. This is evident in the high degree of overlap between the profiles from different days. Consistent results were also obtained within a day's experiment as demonstrated by splitting the data set in half (not shown).
From the width of the band of activation, an approximation of the cortical magnification factor was calculated. With the stimulus (1° width) at 0° (vertical meridian) the band of activated cortex measured 2.1 mm (at half height). This is consistent with the known cortical magnification factor that ranges between 3.0 mm/deg and 1.7 mm/deg visual angle at eccentricities between 5° and 10°, respectively (see Discussion). The activation from the stimulus centered on the vertical meridian indicated that the V1/V2 border was located 1–2 mm behind the lunate sulcus in M1R.
Mapping of Areas V1/V2 and the Ventral Portion of the Dorsal Prelunate Sulcus (DPv) with Expansion Optic Flow
Recordings from V1/V2 were used to explore the response to large field optic flow stimuli. The responses to these less traditional stimuli were also used to further validate the recording methods. Simultaneously, recordings were collected from the ventral portion of the prelunate sulcus (DPv) to determine the presence and nature of its retinal topography to expanding dot stimuli.
Single Condition Maps
Single condition maps (baseline normalized) for each of the nine stimulus positions illustrate the primary finding of a variation of the reflected light with the retinotopic position (Fig. 5A). The angioarchitectonics collected at 605 nm illumination from M2L mainly reveal the large blood vessel over the LS with areas V1/V2 posterior and DPv anterior of the vessel (Fig. 7A, see also the inset in Fig. 1C, M2L). The nine single condition maps in Figure 5A demonstrate that in areas V1/V2 relative darkening of the cortex occurs when the stimulus appears in the center (0°,0°) or just below (0°,–20°), as can be seen in the region of interest (black square). The amount of cortex in V1/V2 that is activated is substantial and is not punctate. This suggests that the point image of the optic flow stimuli covers a substantial portion of the exposed V1/V2 cortex. DPv, on the other hand, was mostly activated by stimuli in the upper right corner (20°,20°), as can be seen in the region of interest (white square). Again large portions of cortex were activated.
|
|
Time Course
For a selected region of interest (black squares in Fig. 5A), the time course of the optical signal over V1/V2 was obtained for the nine different stimulus locations (Fig. 5B). All time courses began similarly (shaded area between time –1000 and 0 ms) with a slight increase in the reflectance signal, followed by a decline. This represents a dependence of the reflected light from the initial events of the trial (onset of fixation point, initial saccade to fixation point, etc.). As the maximum visually evoked response peaks at 2000 ms after stimulus onset and either decays or remains constant to 3000 ms after stimulus onset, the intrinsic signal was extracted at this time period (i.e. signal change compared to baseline). Thus, the greatest attenuation in reflected light is found for the center position (0°,0°), followed by the position just below (0°,–20°). A graph of reflected light as a function of retinotopic position indeed shows the minimal response at (0°,0°), as can be seen in the mean reflectance (Fig. 6A). The same style plot (Fig. 6D) was created for a region of interest in DPv (small white squares in Fig. 5A). Here the interpolated maximum was located in the upper contralateral quadrant at (20°,20°).
The baseline-normalized single-condition maps and the region-of-interest graphs therefore show that the optical response depends on the retinotopic position of the optic flow stimulation. Regression models were used to quantify the tuning pixel by pixel for the entire maps and across all experiments.
First-order General Linear Model (V1/V2)
A simple linear regression model (equation 1) was first used to quantify these effects (Fig. 6B). This figure shows the region of interest over V1/V2 (black squares, Fig. 5A). The linear fit for the region of interest captures one aspect of the data, i.e. the strength of the lower field response. Note that the regression surface was multiplied by –1 to correct for the inverse relationship between neuronal firing and optical signal. For the entire maps, the process of the linear fit was performed on a pixel-by-pixel basis rather than on single regions of interest and parameter maps were generated. Therefore, each pixel has a regression surface associated with it.
The intercept parameter map ‘ß(I,J)’ (Fig. 7B) is the modeled change of the reflected light from the cortex when the stimulus is in the center position (xi = yi = 0° in equation 1). In areas V1/V2, the map of the horizontal x-coefficient (x) does not show a strong modulation (Fig. 7C), whereas the y-coefficient (y) map is dark (negative values) (Fig. 7D). (Note that the brightness of the pixels in the coefficient images indicates the percentage change in signal per degree of visual angle, i.e. %/deg, not the signal amplitude.) The resulting color-coded angle map (Fig. 7F) shows that the V1/V2 cortex is mostly active when the stimulus is in the lower contralateral quadrant (blue-green), in particular along the lower vertical meridian (bright green-yellow), which demarcates the border between area V1 and V2.
First-order General Linear Model (DPv)
A regression of the region of interest in Figure 5A (white square) is used to demonstrate the dependence of its signal upon retinotopic position. The resulting regression is a plane with the strongest signal in the upper visual field. A pixel-by-pixel analysis provides the quantification of the optical signal across the cortex (Fig. 7B–F). The activation over DPv is mainly related to stimuli in the upper contralateral quadrant and horizontal meridian (blue colors; Fig. 7F). The amplitude map (Fig. 7E) illustrates that the major blood vessels generate a very strong signal (white pixels), while the strength of the retinotopic signal across the cortex is reasonably uniform. The strong vessel signal could be responsible for the ‘rainbow’-like color change near the LS.
Second-order Quadratic General Linear Model (V1/V2)
A linear modulation of stimulus position within the visual field is obviously a poor model for V1/V2. As described in the Materials and Methods, parameters for a second-order general linear model were computed (equation 2). Figure 6C illustrates the quadratic surface for the region of interest (black square in Fig. 5A) in V1/V2. It is a peaked function with its center just below the fovea. In Figure 8, the quadratic regression was computed for the data of Figure 5 on a pixel-by-pixel basis. The first-order coefficients (Fig. 8A,C) are similar to the ones obtained with the linear model (Fig. 7C,D). However, the second-order coefficients provide a quadratic dependency. The quadratic coefficients are predominantly negative over V1/V2, which indicates peaked functions for this area (Figs 2A, 8B,D; see Fig. 6C). The peak centers of the retinotopic activation were computed on a pixel-by-pixel basis for the entire map (see Materials and Methods). For example, the linear x-coefficients (x) are mostly positive (Fig. 8A) and the quadratic x-coefficients (xx) mostly negative (Fig. 8B), and this yields positive values for the x-centers. The linear and quadratic y-coefficients (y and yy) are predominantly negative (Fig. 8C,D), which yields negative values for the y-centers. As a result, most x- and y-centers for V1/V2 in M2L are located within the lower contralateral (right) quadrant or close to the lower vertical meridian as shown in the color-coded angle map (Fig. 8E). This is consistent with the known visual topography and with the results from the study with the elongated thin bar stimulus.
|
Second-order Quadratic General Linear Model (DPv)
While a linear model is perhaps not the best for V1/V2, its appropriateness for DPv is unknown. Figure 6F shows the quadratic fit for the region of interest (white square, Fig. 5A), which is a saddle shape. The pixel-by-pixel analysis indeed reveals a mix of positive (yy) and negative (xx) coefficients (Fig. 8B,D), which correspond to saddle points (Figs 2C,D and 6F). The DPv centers are also located in the contralateral field closer to the horizontal meridian. However, the modeled retinotopic activation for DPv may be quite complex with a combination of positive and negative quadratic coefficients. Before cataloging the retinotopic activation of the DPv pixels, it was necessary to compare the linear and quadratic models.
Comparison of First- and Second-order Models for Both Imaged Regions
To determine whether the linear or quadratic model was best for each pixel, the AIC was computed. By comparing the resulting scores for each pixel using a binary map, the linear fit proved to be the better model for DPv, while the map over V1/V2 was best modeled by the quadratic function (Fig. 9A). (The ‘better model’ is defined as having a majority of the pixels for that region selected by the AIC.) Areas V1/V2 were best modeled as a quadratic function, and DPv as a linear function for 10 of 17 experiments in M1R, and for 6 of 7 experiments in M2L.
|
|
To establish which sign of quadratic coefficient was dominant for a particular area, the angle map (Fig. 8E) was overlaid with masks for positive and negative quadratic coefficients (‘valleys’ and ‘peaks’; Fig. 10). Most of the V1/V2 retinotopic activation was best modeled with ‘peaked functions’ (Fig. 10A,C). This pattern was very consistent for all experiments and both monkeys. In each of the M2L region of interest measurements (total 14; 7 experiments, 2 regions of interest per area, x and y), the quadratic terms were negative. In M1R, V1/V2 also had mostly peaked functions, 27 for x, 24 for y (total 32; 16 experiments, 2 regions of interest per area). (Although there are saddle points in the DP regions, these are not the appropriate model for the data, as the linear regression was selected by the AIC.)
|
Summary of Retinotopic Organization in V1/V2 and DPv
To quantitatively assess the retinotopic organization of the optical signal across experiments, regions of interest were selected medially and laterally within one area. Within selected regions of interest, the width of the peaked function for V1/V2 was calculated to provide an estimate of the modeled width of retinotopic activation (see Materials and Methods). In M1R, with the 20° stimuli the average width was 24° ± 2.75° (mean ± standard error), n = 14. For the smaller 10° stimuli, which were also presented closer to the center, the average width was 38.6° ± 6.1°, n = 18. The difference between the two stimulus sets was not significant (t = –1.97, df = 30, P > 0.05). In M2L, the average width of retinotopic activation was 15.8° ± 10.1°, n = 12 for the 20° stimulus, and 11.3° ± 9.9°, n = 2 for the 10° stimulus (t = 0.58, df = 12, P > 0.05).
Similar to the single example for M2L (Figs 5–7
), all the regions of interest selected in areas V1/V2 consistently showed preference for stimuli in the lower contralateral visual field and along the lower vertical meridian using the second-order model. This was also the case for monkey M1R (Fig. 12A). This preference for a particular angle was confirmed using the Raleigh test for uniformity in both animals (Zar, 1984). The Raleigh test is a circular statistic that tests whether there are significant deviations from a uniform distribution. The null hypothesis of a uniform distribution of directions was rejected (P < 0.05). This finding confirms that the retinotopic activation was consistent across experimental sessions (days) and matches the results using the elongated bar stimuli. Hence the values of mean direction and the confidence interval provide a meaningful summary of the tuning for the region of interest (Fig. 12A).
|
The DPv data were also tested for uniformity and were found to be uniform for both animals (P > 0.05). This indicates variability in the measurements across days with our stimulus paradigm as evidenced by the high circular standard deviations (Fig. 12B).
Mapping of the Dorsal Portion of the Dorsal Prelunate (DPd) and Area 7a with Expansion Optic Flow
In both animals, the draining vein, which lies between the dorsal apex and the dorsal-most portion of the STS, was used to divide DPd and area 7a (Fig. 1C, insets; Fig. 11A). The regression parameters were estimated for both the first- and second-order models. The two models were then compared using the AIC. For DPd and area 7a, the retinotopic activation was always best modeled by a linear function (Fig. 9B). A typical example of such an experiment is presented in Figure 11 for M1R. The image of the angioarchitecture under green light (Fig. 11A) shows a clear view of the cortex without any signs of neomembrane growth. The intercept (ß) and amplitude maps (Fig. 11B,E) indicate that strong signals originate from the blood vessels. The optical signal amplitude over cortex free of large blood vessels appears relatively even. The map of the horizontal coefficient (x, Fig. 11C) shows modulation from lateral (light, positive values) to medial (dark, negative values) in area 7a, whereas the vertical coefficient (y, Fig. 11D) map appears rather uniform. The DPd cortex also shows modulation along the x-axis with mostly dark pixels (negative values). This is summarized in the colored angle map (Fig. 11F), where the purple and dark blue colors suggest an emphasis on upper contralateral activation both in DPd and 7a.
Raleigh's test for uniformity was applied for the DPd and 7a angle data for the regions of interest for both animals. DPd was significantly non-uniform in both animals (M1R, P < 0.01; M2L, P < 0.05; Fig. 12C), whereas area 7a in both animals showed uniform distributions (Fig. 12D), as suggested by the large standard deviations. Thus, the regions of interest in DPd had repeatable retinotopic representations while nearby area 7a did not. The retinotopic representation of DPd was different for the two hemispheres in the two animals, with one predominantly in the upper visual field (M1R) and one in the lower visual field (M2L).
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Intrinsic optical imaging was utilized for mapping retinotopic organization in striate and extrastriate parietal areas in behaving monkeys. The validity of the stimulus and recording method was established by imaging early visual cortex with its known retinotopy. Two stimulus types (elongated thin bars and larger circular expansion optic flow) were used in V1/V2, and results from both agreed with the known visual topography. Areas of the inferior parietal lobule were then tested for retinotopic organization with optic flow stimuli.
Selection of Model for Quantitative Mapping of Retinotopic Activation
Regression models were used to reduce and summarize the large amounts of optical imaging data. Two general linear models were used to model the dependence of the optical signal upon the stimuli; each pixel was modeled independently. This approach permitted a direct numerical comparison of a first-order linear model and a second-order quadratic model. An alternative choice to the second-order quadratic is an exponential Gaussian model, which is commonly used to represent receptive fields in single-unit data (Ben Hamed et al., 2001). Like the second-order quadratic, the receptive field width and center can be computed from the Gaussian model coefficients. However, the determination of the coefficients of the Gaussian model must be performed by numerical gradient descent methods or annealing to minimize an objective function. This is further complicated by the necessity to compute a DC offset (i.e. the signal at the extreme edges of the receptive field.) While various powerful minimization methods are currently available, there is no guarantee that the global minimum, and hence the best fit to the data, will be obtained. Therefore the general linear model, with its closed form solutions, was selected. Further, the general linear model had the advantage of direct comparison with previously published results from single-unit recordings (Andersen et al., 1990; Read and Siegel, 1997).
Time Course of Signal During the Fixation Period
In the present study, there was a variation in the optical signal that occurred after the initial fixation onset. This type of change was seen in each of the areas recorded reported here as well as in Siegel et al. (2003). In V1 of anesthetized animals, such variation is not reported (Shtoyerman et al., 2000). In one published figure of a time course for V1 of behaving monkey (Grinvald et al., 1991), the fixation event is not shown, as the monkey was passively viewing the display. In the other figure (Fig. 4 of Shtoyerman et al., 2000), the baseline is flat during the initial fixation period. That Shtoyerman et al. (2000) do not report the modulation during the initial fixation period may be because the time course of their data is normalized by the time course of a blank condition that might also vary in time. As the time course in the current study is not normalized by a blank condition, the modulation of signal during the initial fixation period is thus revealed. It could be due to a number of events that occur during the start of a trial. The fixation point is illuminated, the monkey makes a saccade and there are shifts in attention. The ‘baseline normalization analysis’ described in the Materials and Methods was thus used to ensure the stability of the data over the duration of the trial and allowed us to avoid using a normalization procedure with a ‘blank’ condition.
Validation of Method in V1/V2
As electrophysiological recordings compromise our optical imaging, two different measurements were made in areas V1 and V2 to establish the validity of the optical imaging method and the choice of stimuli. First, areas V1 and V2 of M1R were mapped with elongated bar stimuli placed along the lower vertical meridian at increasing eccentricities up to 3° away from the meridian. Many researchers have used intrinsic optical imaging to reveal retinotopic organization in visual cortex in various anesthetized species (Mc Loughlin and Blasdel, 1998; Macknik and Haglund, 1999; Blasdel and Campbell, 2001; Bosking et al., 2002; Lyon et al., 2002; Schiessl and McLoughlin, 2003). To our knowledge there is only one imaging study, using voltage-sensitive dyes, that has examined retinotopy in the behaving monkey (Slovin et al., 2002). Our experiments demonstrated that primary visual cortex is sensitive to slight spatial displacements of a line or band stimulus, and that these topographically organized responses can be reliably visualized by optical imaging. This was confirmed with intrinsic imaging in the behaving monkey with the current elongated thin bar experiment that showed consistent activation along the V1/V2 border when the stimulus was centered on the lower vertical meridian. The results from this experiment are also consistent with previous electrical and 2-deoyxglucose studies (Gattass et al., 1981; Dow et al., 1985; Tootell et al., 1988) as well as with the cited optical studies of retinotopy in macaque monkey V1 and V2. To the best of our knowledge, the present study is the first intrinsic optical recording of retinotopy in behaving monkey V1 confirming the anesthetized optical studies and the voltage-sensitive dye imaging study (Slovin et al., 2002).
The present study did not observe a mirror-image representation of the –1° or –2° bar in V2. There are a number of reasons for this. First, V2 in the monkey may start too close to the lunate sulcus to be imaged. Further, V2 cortex curves down into the lunate sulcus so that the imaged cortex becomes out of focus or unobservable. This can be seen on the blood vessel map that becomes out of focus anterior to the V1/V2 border. It is also possible that the stimuli consisting of small moving dots were not ideal for activating area V2. The superficial layers of V1 contain many neurons (up to 30%) that prefer small stimuli (‘end-stopped cells’), whereas in V2 no such differences were reported between laminae (Heider et al., 2000). Further, optical imaging studies of V1 and V2 (Roe and Ts'o, 1995; Xiao and Felleman, 2004) suggest a highly complex mapping across V2.
Reliable and reproducible retinotopic activation was obtained with the elongated thin bar stimuli, consistent with the known retinotopy and cortical magnification factors. The representation of the elongated bar stimulus across the cortical surface was fine enough to establish that millimeter scale features could be recorded with our system in behaving monkey. The activation pattern obtained with the bar stimuli confirmed that our optical imaging approach can be successfully utilized to measure retinotopic activation in the behaving monkey, and thus serves as a basis for subsequent studies using optic flow in visual association cortex. The reliability of these V1 results also suggests that the variability of activation that we describe in other visual areas does not arise from inadequacies in the optical imaging method.
A second approach to validate our methods was to map V1/V2 with optic flow stimuli. Optic flow was selected for a number of reasons. First, there is a dearth of studies examining the response of neurons in these early visual areas to optic flow. Second, optic flow stimuli are extensively used for studying extrastriate cortex. Third, it is not known a priori whether large field optic flow stimuli can be used to extract retinotopic organization in V1/V2 or elsewhere.
In the current optical imaging study, much of the imaged cortex was modulated for each stimulus location, indicating that the expansion optic flow stimuli had a large cortical point image. Small millimeter and submillimeter point images, as typically seen (Macknik and Haglund, 1999; Blasdel and Campbell, 2001; Bosking et al., 2002), were not observed with the 10° or 20° optic flow stimuli. The inability to record fine scale changes across the cortex appears not to be a technical limitation, as we can observe finer scaled structures in V1 with the elongated thin bar stimuli. The activation of a substantial portion of the imaged cortex by the optic flow stimuli appears to be a substantial and repeatable finding. Even though much of the cortex was activated by any one of the nine stimuli, the differences in activation between each stimulus, as evident by inspection of the single condition maps or by the regression methods, could be extracted and analyzed to yield a retinotopically based representation of the visual field on a pixel-by-pixel basis. It was clear that the central and lower contralateral stimuli best activated the V1/V2 cortex. This is consistent with previous multi-unit mapping (Gattass et al., 1981) and imaging studies (Blasdel and Campbell, 2001; Brewer et al., 2002; Fize et al., 2003) using more classical stimuli, as well as with the results using the elongated thin bar stimuli described herein.