Orientation Selectivity in Pinwheel Centers in Cat Striate Cortex

See allHide authors and affiliations

Science  06 Jun 1997:
Vol. 276, Issue 5318, pp. 1551-1555
DOI: 10.1126/science.276.5318.1551


In primary visual cortex of higher mammals neurons are grouped according to their orientation preference, forming “pinwheels” around “orientation centers.” Although the general structure of orientation maps is largely resolved, the microscopic arrangement of neuronal response properties in the orientation centers has remained elusive. The tetrode technique, enabling multiple single-unit recordings, in combination with intrinsic signal imaging was used to reveal the fine-grain structure of orientation maps in these locations. The results show that orientation centers represent locations where orientation columns converge containing normal, sharply tuned neurons of different orientation preference lying in close proximity.

In recent years, optical imaging has enabled the investigation of neuronal response properties over large areas of the visual cortex in vivo (1-3). These experiments have revealed that orientation selectivity is not organized in parallel bands but in iso-orientation domains that are arranged radially in a pinwheel-like fashion (4). Optical imaging studies have shown that the magnitude of the orientation signal in the centers of these pinwheels is low (1, 3, 4), suggesting that the population of neurons in these locations might mainly consist of unoriented cells. However, because of their relatively low spatial resolution, imaging studies cannot reliably determine the physiological characteristics of individual neurons in these regions. We have previously reported that in some locations of cat striate cortex, adjacent cells display large differences in orientation preference (5). Because this is an alternative explanation for the low magnitude of the optical orientation signal, we conjectured that these regions may correspond to the pinwheel centers in the orientation preference map.

In five halothane-anesthetized adult cats, we used optical imaging based on intrinsic signals to record the orientation preference maps of visual areas 17 and 18 (6). The animals were stimulated with drifting square wave gratings of different orientations. The image of the visual cortical surface obtained in one experiment along with the corresponding “angle” and “polar” maps is shown in Fig.1 (7). After obtaining these maps, we used tetrodes, which enable simultaneous and separable recording of small numbers of neighbouring neurons (8, 9), to record from pinwheel centers or iso-orientation domains while presenting the same drifting square wave gratings used in the imaging session. Subsequently, we extracted and sorted the individual spike trains from the multi-unit recording, computed peri-stimulus time histograms (PSTHs) and tuning curves for all cells (9, 10), and compared the response properties for neurons within and outside of orientation centers.

Figure 1

Orientation map obtained by intrinsic signal imaging. (A) Vascular pattern of the cortical surface. This area contains portions of visual areas 17 and 18. The circles in the panels show the locations of the tetrode penetrations. These penetrations were aimed at orientation centers or at iso-orientation domains [see (B)]. The two asterisks indicate the locations of the recordings shown in Fig. 2. Scale bar, 1 mm. (B) Color-coded orientation preference (“angle”) map for the cortical region shown in (A). The responses to four stimulus orientations were summed vectorially, and the preferred orientation for every point is color-coded as shown on the right. This angle map reveals the typical organization of orientation preference maps into pinwheel centers (where different orientation preferences converge; for example, left asterisk) and iso-orientation domains (where neurons of similar orientation preference are grouped together; for example, right asterisk). (C) Polar map, combining the color code for preferred orientation with a brightness code representing the strength of orientation tuning [for details see (3, 4, 16)]. Dark regions represent areas of weak tuning, whereas bright areas represent strong orientation preference. Note that dark areas are prevalent in pinwheel centers (for instance, middle circle lower row). L, lateral; M, medial; P, posterior; and A, anterior.

In 24 penetrations in pinwheel centers, we recorded at 77 sites and extracted a total of 345 single neurons (Table 1). As a control, we performed 20 penetrations in iso-orientation domains (11) yielding 81 sites and a total of 348 neurons. By statistically comparing the firing rate before and during stimulus presentation, we found 8% of the cells in pinwheel centers and 7% of the cells in iso-orientation domains to be unresponsive to the presented stimuli. Fourteen percent of the cells in pinwheel centers and 17% in iso-orientation domains were not tuned (12). For the remaining tuned cells, tuning bandwidths and firing rates were determined and were indistinguishable between orientation centers and iso-orientation domains (Table 1). This finding demonstrates that neurons in pinwheel centers are as selective for stimulus orientation as those in other locations of the cortex and that pinwheel centers do not contain larger populations of unselective cells, as might be inferred from the reduced brightness of the polar maps characteristic for these locations.

Table 1

Statistics for all the data and for a subset sampled from upper cortical layers (above 700 μm) grouped by recording locations in pinwheel centers and iso-orientation domains. The Kolmogorov-Smirnov (KS) test was used to determine if the two samples were derived from the same population with respect to tuning bandwidth and firing rate. All values are well above 0.05, indicating that even this sensitive test cannot detect differences between the two samples. The t test (two-tailed; TT) was used to prove that the two samples significantly (P < 0.01) differ in orientation scatter and orientation range. Because here the scatter of the different parameters within the populations is the relevant entity, standard deviations (which describe this scatter) are used rather than standard errors of the mean (which describe the precision of the measurement of the population’s mean).

View this table:

One implication of this finding is that pinwheel centers would be expected to show greater local variance in orientation preference than iso-orientation domains. Using circular statistics, we defined the measures of orientation scatter and orientation range (13) and found, as expected, that both measures were significantly larger for sites recorded in pinwheel centers than for sites recorded in iso-orientation domains (Table 1). This result is further illustrated by Fig. 2, which displays the response properties of a group of cells recorded from a pinwheel center (panels A, B, and C) and another group in an iso-orientation domain (panels D, E, and F). These plots show that the neurons recorded from the pinwheel center are indeed tuned to widely different orientations (orientation range 70° in Fig. 2, A to C), whereas the cells recorded in the iso-orientation domain are tuned to similar orientations (and directions). This is most clearly visible in panels B and E of Fig. 2 where Gaussian fits of the orientation tuning curves for each of the cells are shown (10).

Figure 2

Examples of response properties of single units recorded at the center of a pinwheel (A to C) and in the middle of an iso-orientation domain (D to F). (A) and (D) depict data from three neurons. For each neuron 50 superimposed waveforms recorded from the four wires of the tetrode are shown separately. Note that the pattern of spike amplitudes across the four channels varies from cell to cell. (B) and (E) show tuning curves [Gaussian fit (10)] from recordings in a pinwheel center and an iso-orientation domain. Numbers (1, 2, and 3) and colors label tuning curves from cells whose spike-waveforms (A and D) and PSTHs (C and F) are shown in the other panels. Unlabeled, gray curves represent neurons from the same recording site that were simultaneously recorded but not displayed further. (C) and (F) are peristimulus time histograms for the same three neurons shown in (A) and (D). The cells were stimulated by monocular presentation of an oriented drifting square wave grating. Each row corresponds to the cell’s response to a different stimulus orientation ranging from 0° to 360° in 22.5° steps. In this example, neurons from the pinwheel center (A to C) exhibit wide variance in orientation preferences (orientation scatter: 22.6°, orientation range: 70°), whereas cells from the iso-orientation domain (D to F) show small variance (orientation scatter, 12.3°; orientation range, 29°).

Occasionally, we found sites within iso-orientation domains that displayed an unexpectedly large orientation scatter (14). These data seem to be at odds with the optical imaging data where iso-orientation domains are represented as regions with a homogeneous population of neurons with similar orientation tuning. However, optical imaging based on intrinsic signals mainly records neuronal activity from cells lying less than 600 to 800 μm below the cortical surface (15, 16). Therefore, to compare directly the optical and electrophysiological recordings, we determined the orientation scatter as a function of the recording depth (Fig. 3). In the case of the pinwheel centers there is no trend of the orientation scatter to change with increasing depth. In contrast, in iso-orientation domains a linear regression shows a significant (P < 0.05) positive slope, indicating that in this case orientation scatter continuously increases with depth. Consequently, if only data obtained from sites located in the upper 700 μm are considered (17), the difference in orientation scatter between pinwheel centers and iso-orientation domains becomes much more pronounced than if the comparison is made for the complete data set (Table 1). Similarly, another related measure clearly reveals a difference between pinwheel centers and iso-orientation domains (Table 1): especially in the upper 700 μm of the cortex, the percentage of sites in which the orientation range exceeds 60° is much larger in pinwheel centers (23%) than in iso-orientation domains (3%).

Figure 3

Orientation scatter in pinwheel centers (A) and iso-orientation domains (B) as a function of depth. The orientation scatter at each site is plotted against cortical depth. The gray scale of the squares (see bottom of the figure) codes for the number of neurons that contributed to the data point. For the calculation of the regression line, the contribution of each data point was weighted according to the number of neurons sampled.

Our observation that orientation scatter increases with cortical depth in iso-orientation domains could have at least two explanations: First, it could reflect a true increase in the heterogeneity of orientation tuning in deep layers of cortex. This is supported by some earlier single-unit studies which also reported that orientation scatter increases with cortical depth (18). Alternatively, the strong curvature of the imaged cortical region on the lateral gyrus might account—at least in part—for the larger orientation scatter in deeper layers. Assuming a strictly columnar arrangement, it is likely that columnar width would decrease with increasing cortical depth, thereby placing cells of different orientation preference in closer proximity. The present data do not allow us to distinguish between these two possible explanations.

For the recordings targeted at pinwheel centers, it could be argued that, despite our effort to hit these locations, we were slightly off-target and missed a small population of untuned cells located precisely in the pinwheel centers. One way to address this concern is to score only those sites as successful pinwheel penetrations in which we found an orientation range larger than 60°, because this should only occur in the immediate vicinity of pinwheel centers. Interestingly, when only these data are considered, there are even fewer untuned cells (4%; 3/72) than in the whole data set for pinwheel penetrations (14%; 48/345). This further reinforces our conclusion that neurons in pinwheel centers are not less tuned than those located in iso-orientation domains.

Taken together the results of this study demonstrate that neurons in or near pinwheel centers exhibit the same proportion of unresponsive and orientation-tuned cells, have similar bandwidths and firing rate distributions, and are thus as selective for stimulus orientation as neurons in iso-orientation domains. Thus, the regions of reduced brightness in the polar maps, characteristic for pinwheel centers, do not reflect a lack of neuronal orientation selectivity, but rather result from the summation of cellular responses with greater variance in orientation preference.

These results differ from predictions made by some models of orientation preference that assume orientation centers as regions of decreased orientation selectivity [see, for example, (19)]. Our data suggest that pinwheel centers, with respect to orientation preference, do not represent functionally distinct compartments within striate cortex: Their orientation-tuning properties appear to be indistinguishable from those of iso-orientation domains. This has important implications for cortical organization. It means that the anatomical connectivity of pinwheel centers requires a remarkable degree of specificity. If orientation selectivity arises by the alignment of thalamic afferents (20), the thalamocortical projections into pinwheel centers will require a much higher degree of precision than similar projections into iso-orientation domains. By the same token, if cortico-cortical connections projecting to pinwheel centers link regions of similar orientation preference (21), they would require substantially more accuracy than those projecting to iso-orientation domains. It is remarkable that developmental mechanisms seem to be able to provide this degree of topographic precision for setting up the neural network of the visual cortex.

  • * These authors contributed equally to the work.

  • To whom correspondence should be addressed at The Center for Neuroscience, University of California, Davis, CA 95616, USA. E-mail: pedro{at}


View Abstract

Navigate This Article