HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) All Versions of this Article: 94/5/3538    most recent 00106.2005v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via ISI Web of Science (4) Citing Articles via Google Scholar Google Scholar Articles by Heimel, J. A. Articles by Nelson, S. B. Search for Related Content PubMed PubMed Citation Articles by Heimel, J. A. Articles by Nelson, S. B.

Laminar Organization of Response Properties in Primary Visual Cortex of the Gray Squirrel (Sciurus carolinensis)

Department of Biology, Brandeis University, Waltham, Massachusetts

Submitted 28 January 2005; accepted in final form 29 June 2005

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
The gray squirrel (Sciurus carolinensis) is a diurnal highly visual rodent with a cone-rich retina. To determine which features of visual cortex are common to highly visual mammals and which are restricted to non-rodent species, we studied the laminar organization of response properties in primary visual area V1 of isoflurane-anesthetized squirrels using extra-cellular single-unit recording and sinusoidal grating stimuli. Of the responsive cells, 75% were tuned for orientation. Only 10% were directionally selective, almost all in layer 6, a layer receiving direct input from the dorsal lateral geniculate nucleus (LGN). Cone opponency was widespread but almost absent from layer 6. Median optimal spatial frequency tuning was 0.21 cycles/°. Median optimal temporal frequency a high 5.3 Hz. Layer 4 had the highest percentage of simple cells and shortest latency (26 ms). Layers 2/3 had the lowest spontaneous activity and highest temporal frequency tuning. Layer 5 had the broadest spatial frequency tuning and most spontaneous activity. At the layer 4/5 border were sustained cells with high cone opponency. Simple cells, determined by modulation to drifting sinusoidal gratings, responded with shorter latencies, were more selective for orientation and direction, and were tuned to lower spatial frequencies. A comparison with other mammals shows that although the laminar organization of orientation selectivity is variable, the cortical input layers contain more linear cells in most mammals. Nocturnal mammals appear to have more orientation-selective neurons in V1 than diurnal mammals of similar size.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
A major feature of sensory cortex is the presence of columns of neurons with similar response properties. Most neurons in a vertical column of primary visual cortex (V1) of mammals as diverse as carnivores, primates, and tree shrews respond best to stimuli presented at a narrow range of orientations. These columns are organized horizontally to form smooth, repeating functional maps of orientation preference across the cortical surface. Rodents, however, appear to lack such maps. Recently, we showed that even a large and highly visual rodent, the gray squirrel (Sciurus carolinensis), lacks this horizontal organization (Van Hooser et al. 2005). Another major feature of V1 is its laminar division into horizontal layers containing neurons that vary in their functional response properties. This is particularly evident in animals that have been traditionally used for vision research such as the cat and the monkey. In murid rodents, anatomical lamination is less prominent, and there is an absence of pronounced functional laminar organization. For instance, in mouse and rat, direction selectivity is almost uniform across layers (Girman et al. 1999; Metin et al. 1988) unlike cats and primates in which specific sublayers have high direction selectivity (DeBruyn et al. 1993; Gilbert 1977; Hawken et al. 1988; Orban et al. 1986; Snodderly and Gur 1995). Another example is that although many studies find high proportions of simple cells in layer 4 (and 6) of area 17 in the cat (Bullier and Henry 1979; Henry et al. 1979; Hubel and Wiesel 1962; LeVay et al. 1987) and macaque (Hubel and Wiesel 1968; Kagan et al. 2002; Livingstone and Hubel 1984; Ringach et al. 2002; Schiller et al. 1976a), so far only one study has found a similar trend in a rodent species (Parnavelas et al. 1983). These differences in laminar specificity might be due to differences in animal size or to the degree of specialization for vision. On the other hand, it may be the case that all rodents lack a strong laminar organization in V1. If so, this would indicate a more general lack of functional organization in rodent visual cortex and would suggest that mechanisms for vertical and horizontal organization are related or interdependent.

To answer this question, we studied the vertical organization of a highly visual rodent, the gray squirrel. This animal has large eyes and relatively large and well-laminated visual brain structures (Hall et al. 1971; Van Hooser et al. 2003). Its visual acuity [2.8–3.9 cycles/° (cpd)] (Jacobs et al. 1982) is better than a tree shrew's (1.2–2.4 cpd) (Petry et al. 1984), and its visual cortex is comparable in size to a ferret's (Hall et al. 1971; Law et al. 1988). Orientation selectivity is very common among V1 neurons of squirrels (Polkoshnikov and Supin 1988; Van Hooser et al. 2005).

Unlike nocturnal rodents, gray squirrels have good color vision. The gray squirrel retina contains rods and cones in a 2:3 ratio (Anderson and Fisher 1976; Cohen 1964; West and Dowling 1975). It contains two classes of cones, a short wavelength S-cone with a peak sensitivity of 444 nm and a medium wavelength M-cone with a peak sensitivity of 543 nm and a rod photoreceptor with a peak sensitivity at 502 nm (Blakeslee et al. 1988). Gray squirrels also have a much higher percentage of color-selective optic nerve fibers than cats have (6 vs. <1%) (Blakeslee et al. 1988; Pearlman and Daw 1970). This provides an extra opportunity for a comparative study of dichromatic color vision outside the primate order.

Apart from laminar variations (see e.g., Table 2), there are also differences in the overall degree of selectivity for different stimulus features. Perhaps the most obvious are the large differences across mammals in the spatial acuity of neurons in V1. This most likely reflects differences in eye size and retinal ganglion cell density. Other differences, such as the varying level of orientation selectivity and temporal frequency tuning across species are not easily explained by differences in size (see Table 2 for numbers and references). A pattern may emerge if we would have knowledge of the response properties of a diurnal rodent.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
Surgical preparation and maintenance

Adult gray squirrels (Sciurus carolinensis) weighing 300–700 g were prepared for single-unit recording using methods similar to those of Van Hooser et al. (2003). Animals were initially anesthetized with a mixture of ketamine and acepromazine maleate (90 mg/ml ketamine, 0.91 mg/ml acepromazine maleate, 0.5 ml/kg initial dose im). A tracheotomy was performed for artificial ventilation and administration of isoflurane anesthesia after surgery (0.5–1.5% isoflurane in 50/50 oxygen/nitrous oxide). End tidal CO₂ was measured and kept at 4% by adjusting the ventilation rate within 35–90 strokes/min. The animal was mounted in a custom stereotaxic frame. Rectal temperature was recorded and maintained at 38°C. Electroencepholography (EEG) and heart rate were monitored continuously to assess the level of anesthesia. If spindle activity disappeared from the EEG or if the heart rate changed in response to a toe pinch, the isoflurane concentration was increased. If the EEG showed long periods of low activity, the level was decreased. Typically 1% isoflurane was adequate to maintain this level of anesthesia. Preparations were generally stable for >30 h. To minimize eye movements, a paralytic agent (10 mg/ml gallamine triethoiodide, 0.5 ml/h iv) was infused. Eyelids were held open with loose sutures. Pupils were dilated with 1% atropine sulfate. Plano contact lenses were inserted to prevent drying. In earlier experiments (Van Hooser et al. 2003), we found that it was not necessary to adjust the focus of the eyes. All procedures were approved by the animal care and use committee at Brandeis University.

Recording

The primary visual cortex in the gray squirrel is a large area surrounding the posterior medial pole of the cortex (Hall et al. 1971). Part of it is folded beneath the brain surface, but all of the central vision representation is easily accessible. The anterior-lateral border of V1 represents the vertical meridian. We made a craniotomy (3 x 3 mm²) centered on average at 6.5 mm posterior and 3.5 mm lateral from bregma. In early experiments, when recording close the vertical meridian, we mapped the progression of receptive field positions at several locations to ascertain that we were in V1 and not in neighboring V2, which has a mirrored retinotopy. Under our anesthesia regime, we learned that we could easily determine whether we were in V1 or V2 because the amount of multiunit response to visual stimuli in and around layer 4 is much lower in V2 (Hall et al. 1971). In the first half of the experiments, the dura was resected, and later the dura was left intact. Warm artificial cerebral spinal fluid was used to keep the brain moist, and 3% agar in lactated Ringer solution was used to protect the brain and minimize pulsations. We used high-impedance micro-electrodes (10 M, FHC, Bowdoinham, ME) and a custom-developed multiple window discriminator to isolate single units. We recorded data on 21 penetrations that were roughly perpendicular to the cortical surface. We searched for well-isolated units, so our sample may be biased toward larger cells. After recording from one cell, the electrode was advanced 40 µm, and a subsequent cell was included if the electrode had advanced 100 µm or if the cell showed different response properties from the previous cell. Histological reconstructions showed that penetrations were generally within 15^o of perpendicular to the cortical surface (Van Hooser et al. 2005).

Reconstruction of recording sites

At the end of each penetration, three electrolytic lesions (9 µA, 3 s, tip negative) were made along the electrode track to recover the recording positions. The electrodes were coated in DiI (Molecular Probes, Eugene OR) so that the tracks could later be found (DiCarlo et al. 1996; Snodderly and Gur 1995).

The animals were given an overdose of ketamine/acepromazine or sodium thiopental and perfused transcardially with 0.1 M phosphate-buffered solution (PBS) followed by 4% paraformaldehyde. Coronal sections (50 µm) were prepared on a cryostat. Sections were Nissl-stained, as previously described (Van Hooser et al. 2005). Cortical layers 2/3–6 could easily be identified, as shown in Fig. 1. Sublayers 3c and 6b (Kaas et al. 1972b; Moore 2001) could often be distinguished and occasionally a sub-lamination of layer 4 was apparent but not consistently enough to include this in our laminar analysis. Cytochrome oxidase staining did not reveal any further sublamination or tangential organization (Moore 2001). We used the lesion sites to calibrate the electrode manipulator readings to laminar positions in the cortex using a technique similar to (Hawken et al. 1988). Recording sites were assigned a layer and a relative depth within that layer. By studying lesions and dye labeling from the same penetration and laminar borders on several neighboring sections, we estimate that the error in the reconstruction depth is 70 µm. For seven penetrations, we could not reconstruct the relative depths for the recording sites with certainty and left these cells out of the laminar analysis.

View larger version (149K): [in this window] [in a new window]   FIG. 1. Reconstructing an electrode track. Nissl-stained coronal section of primary visual cortex (V1) with DiI-coated microelectrode track and electrolytic lesion. Layers 2/3–6, and white matter are distinguishable.

Visual stimulation was provided by an Apple Macintosh PowerMac G4 and a gamma-corrected Samsung SyncMaster 900SL CRT monitor running custom-developed stimulation software (Van Hooser et al. 2003) using Matlab (The MathWorks, Natick, MA) and the Psychophysics Toolbox (Brainard 1997; Pelli 1997). Stimuli were shown at a distance of 57 cm from the eyes with a refresh rate of 120 Hz.

Color stimulation

We measured the spectra of the red, green, and blue phosphors (denoted R, G, and B) on our monitor from 400 to 700 nm in steps of 4 nm using a PR-650 spectral radiometer (Photo Research, Chatsworth, CA) kindly lent to us by R. Clay Reid of Harvard University. Using activity spectra S and M for the blue and green cones, respectively, estimated from Blakeslee et al. (1988) and correcting these values for preretinal absorbence (Yolton et al. 1974), we computed a matrix relating RGB color values (r, g, b) to relative cone activation (S_act; M_act) = (R · S, G · S, B · S; R · M, G · M, B · M) · (r; g; b) = (0.0008, 0.00057, 0.0299; 0.0135, 0.0742, 0.0191) · (r; g; b). Using this equation we computed an "equiluminant" pair of colors, blue (r,g,b) = (0, 0, 1) and green (1, 0.42, 0). The blue color mainly activates the S cone, whereas the green color mainly activates the M cone. The total activation of the two cone types, (M_act+S_act), is very similar for the two colors. The luminance contrast of this stimulus (computed using this total activation) is only 2%.

We also used cone-isolating stimuli (Estevez and Spekreijse 1982) to assay the cone inputs to a subset of the recorded cells. We developed one pair of colors for each cone: m₊ and m_– for the M cone, and s₊ and s_– for the S cone. The color pairs were chosen so that a transition from one color to the other would dramatically change the drive to the cone of interest while minimally influencing the drive to the other cone. Thus switching from s_– to s₊ dramatically increases the drive to S cones, but M cones respond with equal intensity to both s_– and s₊. We found color pairs that meet the cone isolating criteria and provide strong contrast: m₊ = (1, 1, 0), m_– = (0, 0, 0.24), s₊ = (0, 0, 0.92), s_– = (0.98, 0.11, 0). The m and s pairs correspond to 90 and –10% M-cone contrast and –5 and 90% S-cone contrast. The luminance contrast between the pairs (using dL/mean L) is 96% for m_– and m₊ and 95% for s_– and s₊. The pairs do not provide the maximum possible contrasts because they were originally designed using less accurate measurements of the CRT phosphor spectra.

It is difficult to assess the validity of our computations psychophysically. We have simulated a behavioral experiment by Blakeslee et al. (1988) in which gray squirrels were required to identify a colored light among three choices: two achromatic lights (4,800 K temperature) and a monochromatic light that varied in wavelength from 470 to 510 nm. The investigators observed a clear neutral range between 495 and 505 nm where the gray squirrels could not distinguish the colored light from achromatic light. We computed the ratio of green to blue cone activation for a 4,800 K source using the cone spectra estimated above and correcting for retinal absorbance, and then computed the same ratio for stimulation with monochromatic light. The achromatic and monochromatic activation ratios were equal for a 502-nm monochromatic stimulus, which is in excellent agreement with the neutral range in the preceding text, suggesting our choices for cone-isolating stimuli are appropriate. In addition, in this study and in pilot experiments in the LGN we encountered some cells that responded exclusively to either the S- or M-cone isolating stimuli, consistent with the idea that these stimuli are truly cone isolating.

Experimental protocol

Receptive fields were initially mapped by hand on a tangent screen. Stimuli were presented monocularly to the dominant eye. We did not score ocular dominance or binocularity, as receptive fields in the two eyes did not always overlap, possibly due to the paralysis of the animal. Coarse orientation tuning curves were measured using large (16 x 16°) sinusoidal gratings drifting in 12 equally spaced different directions. These gratings had a spatial frequency of 0.2 cpd and drifted with a temporal frequency of 4 Hz. Most cells responded vigorously to at least one direction. As with all our tests, all stimulus conditions were pseudorandomly interleaved and shown five times. Between all of our stimuli a gray background (luminance: 45 cd/m²) was shown for 3 s. Except for the contrast tuning test, contrast was 80% for all tests.

We determined the optimal direction (maximum modulation or firing rate, see following text) and used this for all subsequent stimuli. For two-thirds of our experiments, a small patch of this grating (5 x 5° square rotated to match optimal orientation) was shown at nine overlapping positions around the hand-mapped center to determine the receptive field center more accurately. Centered at this position we showed drifting sinusoidal gratings of nine different spatial frequencies between 0.015 and 1.6 cpd in the original larger window of 16 x 16°. Further tests used gratings with the optimal spatial frequency. Temporal frequency was varied between 0.5 and 32 Hz, and the optimal temporal frequency was used for the remaining tests. Contrast tuning was measured using drifting gratings at seven contrasts levels between 2 and 100%. To determine latency and spatial linearity, we flashed static gratings with 12 different spatial phases.

For a subset of the cells, color sensitivity was assessed by measuring spatial frequency tuning again using an equiluminant colored grating drifting at 4 Hz. For some cells, we measured cone inputs and cone opponency by showing a series of linear combinations of cone-isolating colors, with m₊ overlapping with s_– and m_– with s₊. In the early spatial phase of the sinusoidal grating, we used the combination m₊ + (1 – )s_–, and in the late phase, we used m_– + (1 – )s₊. Thus for = 0(1) the stimulus is S(M)-cone isolating.

The total characterization procedure took a little less than an hour per cell. Not all cells could be kept well isolated long enough to run all tests. The results of all tests run are included in the analysis.

Data analysis and statistics

Average firing rates (DC or F₀ component) and response modulation at the drifting frequency (F₁ component) were computed. The parameter value that gave the largest F₀ or F₁ component overall was considered optimal and was used in subsequent tests. For the analysis of the data and fitting of tuning curves, we did not use the F₁ modulation but used the average firing rate for all cells. This allowed us to compare the tuning of cells with different F₁/F₀ ratios.

For several parameters, we defined low and high cut-offs as the stimuli where the firing rate reached half the maximum firing rate. The spontaneous rate was not subtracted unless otherwise stated. For each cell we computed the orientation selectivity index (OSI) as follows,

, where _k is the drift direction in radii and F₀( _k) is mean firing rate in response to a stimulus with that direction. The OSI and the related circular variance (1 – OSI) are often used as a measure of orientation selectivity in V1, see e.g., (Worgotter and Eysel 1987). Von Misen functions p₀ + p₁exp{[cos(2 – _pref)] – 1}/p₃ provide the best fit to orientation tuning curves (Swindale 1998). From Von Misen fits, we computed the tuning width as half width at half the height (HWHH) of the maximum F₀ response (Henry et al. 1974). A directionality index (DI) was computed as (R_optimal – R_oppposite)/R_optimal, subtracting the spontaneous firing rate from all responses. Spatial and temporal frequency responses were fit with difference of Gaussians. For the temporal frequency tuning, a constraint was added that >150 Hz the cells would respond at the spontaneous rate. The fits were not very sensitive to the point at which this constraint is enforced.

Few, if any, of the response properties were normally distributed. To compare scalar properties between different groups of cells, we used a nonparametric Kruskal-Wallis rank test, unless otherwise stated in the text. To calculate the significance of differences in categorical properties, we applied a ² test. For computation of the significance of bimodality, we used a Matlab implementation of Hartigan's DIP test (Hartigan and Hartigan 1985) by F. Mechler (available at http://manuelita.psych.ucla.edu/dario/neurodata.htm). In bar graphs, significance is shown by asterisks, *P < 0.05, **P < 0.01, and ***P < 0.001.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
We studied the functional response properties of 221 neurons in the primary visual cortex of 16 gray squirrels using sinusoidal gratings. Of these cells, 27 did not respond to visual stimulation or had peak firing rates <3 Hz, and we excluded these from our analysis. Our sample of cells was well distributed across cortical layers. We could reconstruct the locations of 41, 32, 27, and 28 cells in layers 2/3, 4, 5, and 6, respectively. We did not record any cells in layer 1. The same cells were used for one additional test in which the orientation preference and tuning width dependence on contrast was measured. The results of this test and the absence of a smooth horizontal organization of orientation preference were reported earlier (Van Hooser et al. 2005).

Gray squirrel V1 is divided into a lateral binocular zone that mediates the central 30^o of vision in each hemisphere and a medial monocular zone (Hall et al.1971; Kaas et al. 1972a). Receptive fields of cells in our study all fell between –30 and 10° of the horizontal meridian and were <60° lateral of the vertical meridian, and three-quarters (151/194) of these cells were within the binocular zone. Of these, only 8% (12/151) responded preferentially to the ipsilateral eye.

Relative modulation

The relative modulation F₁/F₀, i.e., the ratio of response modulated at the drift frequency to the average firing rate, is a measure of the linearity of input summation of a neuron. Researchers in other mammals have reported a bimodal distribution of F₁/F₀ with a valley near 1 (Schiller et al. 1976c; Skottun et al. 1991). We did not find such a bimodal distribution if all cells are considered (Fig. 2C). A bimodality is apparent if we only consider the more oriented cells, i.e., all cells with an orientation tuning half-width at the half height of the maximum response (HWHH, see Orientation tuning) less than the population median, (Fig. 2C, dark bars), but this was not significant (P = 0.09). A bimodality does not necessarily imply the existence of two distinct populations but may reflect a nonlinearity in the relative modulation measure itself that could make a unimodal distribution of cell parameters appear bimodal (Mata and Ringach 2005; Mechler and Ringach 2002). However, because the relative modulation correlates to several other response properties, such as the preferred spatial frequency and the contrast linearity (see following text), it is convenient for presentation purposes, to take the F₁/F₀ = 1 as a boundary between two classes. We did not measure subfields and cannot apply the original definition of simple and complex cells based on subfield segregation (Hubel and Wiesel 1962), but the 146 cells with smaller F₁/F₀ values, we call "complex," and the 48 cells with ratios above this value we call "simple."

View larger version (32K): [in this window] [in a new window]   FIG. 2. Modulation and linearity. A: running median and quartiles of F1/F0 for drifting gratings over cortical depth. Moving window corresponds to 250 µm. Dots are individual cells. B: median and quartiles of F1/F0 ratios for different layers. Layers 4 and 6 show slightly higher F1/F0 ratios (P = 0.01), i.e., cells are more linear in their spatial summation. C: distribution of F1/F0 ratios appears bimodal for oriented cells (dark bars) but not if unoriented cells are included (light bars). D: spatial F1/F0 for static gratings is lower in layers 2/3 (P = 0.008) and higher in layer 4 and 6, but only the latter is significant by itself (P = 0.05). E: histogram of F1/F0 static. Dark bars, oriented cells; light bars, unoriented cells.

The response to flashed stationary gratings at different spatial phases is another indication of the spatial linearity of a cell. The spatial linearity is defined as the ratio of the spatial F₁ component and spatial F₀ component of the response. The measure is very similar to the F₁/F₀ value computed for drifting gratings, and the two measures are highly correlated (r = 0.6, P < 10^–11). Its distribution is very skewed and only shows a hint of bimodality with a notch around F₁/F₀ = 0.75. The laminar distribution, shown in Fig. 2D, is similar to that of the temporal modulation ratio of B, but the differences are more pronounced. Layers 2/3 and 5 are on average more nonlinear than layers 4 and 6.

When making the distinction between simple and complex cells based on relative modulation due to moving stimuli in primates, one finds much higher percentages of simple cells than when using static stimuli and subfield segregation for the classification (Bullier and Henry 1980; Kagan et al. 2002). There is no such difference between the two measures within our dataset: a quarter (26%) of the cells are simple based on the criterion that (temporal) F₁/F₀ > 1 for drifting gratings, whereas 27% have (spatial) F₁/F₀ > 0.75 for static gratings.

Orientation tuning

The degree of orientation tuning varied enormously, from cells that were completely indifferent to orientation, e.g., the example shown in Fig. 3 A, top, to cells that responded very selectively to certain orientations, e.g., Fig. 3A, bottom. The median, first and third quartiles of the OSI are 0.29, 0.13, and 0.54 as shown in the cumulative histogram in Fig. 3B by the dashed and dotted vertical lines.

View larger version (29K): [in this window] [in a new window]   FIG. 3. Orientation and direction selectivity. A: examples of orientation tuning of an unselective cell (top) with orientation selectivity index (OSI) = 0.04 and of a highly orientation-selective cell (bottom) with OSI = 0.61 and half width at half height (HWHH) of 24°. Error bars, SE; solid lines, Von Misen fits; dotted horizontal lines, spontaneous rates. B: cumulative histogram of OSI. Black line, all cells; gray line, simple cells; dashed line, median; dotted line, quartiles. Simple cells have higher OSI. C: cumulative histogram of HWHH. Simple cells have a smaller HWHH. A HWHH of 90° means that the response did not drop below half its maximum. D: little dependence of orientation selectivity on cortical depth. E: cumulative histogram of DI. Directionality is higher in simple cells (P = 0.001). F: median and quartiles of directionality index (DI) in different layers. Direction selectivity is higher in layer 6 (P = 0.0001).

We found little difference in the distribution of orientation selectivity across the different layers of the cortex, as shown in Fig. 3D. An analysis of the horizontal organization across the cortical surface of orientation preferences of cells in this dataset has been published previously (Van Hooser et al. 2005).

Direction tuning

The median DI is a low 0.23, Fig. 3E. Using the criterion of DI >2/3, only 10% of the cells are directionally selective. The directional selectivity is significantly higher in layer 6 than in other layers as illustrated in Fig. 3F (P = 0.0001). Simple cells are more directional than complex cells (median: 0.34 vs. 0.18, P = 0.001). The given numbers and statistics are very similar if the spontaneous rate is not subtracted (not shown).

Spatial frequency tuning

Spatial frequency tuning with drifting gratings in the optimal direction was recorded in 192 neurons and fit with a difference of Gaussians. The median optimal spatial frequency is 0.21 cpd with complex cells preferring slightly higher frequencies than simple cells (P = 0.006; Fig. 4B, 2 middle curves). This difference is more pronounced for the high cut-off spatial frequency shown by the right two curves in Fig. 4B and Table 1 (P = 0.0002). The high- and low-frequency cut-off frequencies (frequencies with responses of half the maximum response) were estimated using the fit. If a cell's response did not drop below half the maximum response for the lowest frequency that we used in our tests (0.015 cpd), the cell was classified as low-pass. The majority (82%) of the cells show band-pass spatial frequency tuning like the example shown in Fig. 4A, top. The median bandwidth log₂(high cut-off frequency/low cut-off frequency) of the band-pass cells is 2.1 octaves. The full distribution of bandwidths is shown in Fig. 4C.

View larger version (32K): [in this window] [in a new window]   FIG. 4. Spatial frequency. A: examples of band-pass (top) and low-pass (bottom) spatial frequency tuning. Curves are fits with a difference of 2 Gaussians. Dashed straight lines, low cut-off, optimal, and high cut-off frequencies; dotted horizontal line, the spontaneous rate. B: cumulative histogram of spatial frequency tuning. Black lines, all cells; gray lines, simple cells. Left-most curves, low cut-offs; middle 2 curves, optimal frequencies; right-most curves, high cut-off frequencies. C: histogram of bandwidth log2(high cut-off frequency/low cut-off). White bars, complex cells; gray bars, simple cells. D: cells in layer 5 have a tendency to be low-pass (P = 0.12).

The highest preferred spatial frequencies are found centrally within the visual field. More peripherally the preferred spatial frequencies drop significantly. For the 144 cells within the central 30° of the visual field, the median preferred spatial frequency is 0.23 cpd, whereas for the 48 cells in our sample outside that region, the median is 0.17 cpd (P < 10^–6).

The spatial frequency bandwidth is positively correlated with the orientation tuning width (r = 0.26, P < 10^–3) and the temporal frequency bandwidth (r = 0.25, P = 0.002). This means that in general, cells do not sacrifice selectivity in one of these features to obtain higher selectivity in other features. An illustration of this is that the preferred spatial frequency is even negatively correlated with the orientation tuning width (r = –0.32, P < 10^–5).

Temporal frequency tuning

Using drifting gratings, we measured the cells' temporal frequency tuning from 0.5–32 Hz. About a third of the cells fired significantly above the spontaneous rate at 32 Hz, but as our monitor had a refresh rate of 120 Hz, we could not test cells at much higher frequencies. A fairly typical example of the tuning of a band-pass neuron is shown in Fig. 5A. Median optimal temporal frequency is 5.3 Hz. Simple and complex cells do not differ from each other in their temporal frequency tuning, see Fig. 5B. Median high temporal frequency cut-off is 17 Hz. At the lowest temporal frequency (0.5 Hz), 10% of the cells still respond above half of the maximum response. These cells are classified as low-pass. The other cells have a median temporal bandwidth of 3.0 octaves, Fig. 5C.

View larger version (34K): [in this window] [in a new window]   FIG. 5. Temporal frequency. A: example temporal frequency tuning curve. Error bars show SE. The curve is a fit with a difference of Gaussians. Dashed verticals lines (from left to right), low cut-off, optimal and high cut-off frequency. B: cumulative histogram of temporal frequency tuning. Black lines, all cells; gray lines, simple cells; left-most curves, low cut-offs; middle curves, optimal frequencies; right-most curves, high cut-off frequencies. C: bandwidth log2(high cut-off frequency/low cut-off); white bars, complex cells; gray bars, simple cells. D: optimal frequency is higher in layers 2/3 (P = 0.004). E: low cut-off frequency is also higher in layers 2/3 (P = 0.0002)

Cells in layers 2/3 have a higher optimal temporal frequency (P = 0.004), Fig. 5D, and a higher low cut-off frequency (P = 0.0002), E, than cells in other layers. There are no significant laminar differences in the high temporal cut-off frequency. Optimal velocity is significantly higher in layers 2/3 (median: 31°/s, P = 0.006) and lower in layer 4 (median: 19°/s, P = 0.04).

The temporal frequency tuning of cells responding to the central 30° of vision (median: 6.2 Hz) and those responding to the periphery (median 5.3 Hz) is not significantly different. Preferred velocities are significantly higher in the periphery than in central vision because cells mediating central vision have higher preferred spatial frequencies than those in the periphery.

Responses to static stimuli

Latency. After determining a cell's optimal parameters for drifting gratings, we used these parameters to show stationary gratings. The spatial phase with the strongest response within 1 s was used to calculate the cell's latency, defined as the time at which the firing rate reached half its maximum value. The latency is strongly laminar dependent, as can be seen in Fig. 6, A and B. The median latency of the whole population is 34 ms, with first and third quartiles of 25 and 52 ms, shown in Fig. 6C. Cells in layer 4 have a significantly shorter latency (P = 0.001) than those in other layers, with a layer median of 26 ms. Layers 2/3 have significantly longer latencies with a median of 48 ms (P = 0.0003).

View larger version (26K): [in this window] [in a new window]   FIG. 6. Latency. A: depth profile of latency shows strong laminar dependence. B: median and quartiles of latency across layers. Layer 4 latency is shorter and layers 2/3 latency is longer than the average population. C: cumulative histogram. Simple cells (gray curve) tend to have shorter latencies than complex cells (black curve).

Response duration

Figure 7A shows a few examples of neuronal responses to the onset of stationary gratings. Some neurons have a very transient response consisting of a single peak with a latency between 18 and 80 ms. Many other cells continue to fire afterward at a lower rate but above spontaneous levels, and some of these cells show a pronounced second wider peak at a latency 100 ms. We did not see an obvious correlation between the state of the animal and response duration as we often recorded sustained cells after encountering transient cells and vice versa.

View larger version (24K): [in this window] [in a new window]   FIG. 7. Response to stationary stimuli. A: examples of peri-stimulus histograms of responses after onset of stationary grating at time 0. Numbers give the transience index (see text). B: cumulative histogram of transience values. No difference exists between transience of complex and simple cells. C: transience vs. cortical depth. Layer 4/5 border contains very sustained cells.

Spontaneous and maximum rates

Spontaneous firing rates are between 0.01 and 20 Hz, with an overall median rate of 0.46 Hz (see Fig. 8A). Complex cells have significantly higher firing rates than simple cells with a median 0.57 versus 0.22 Hz (P = 0.006). There is a clear laminar dependence. Cells in layer 5 have significantly higher spontaneous rates than the general population, (median: 1.7 Hz, P < 10^–4), whereas layers 2/3 have a much lower firing rate (median: 0.19 Hz, P < 10^–5). The median maximum average firing rate in response to five cycles of an optimal drifting grating of 100% contrast is 19 Hz. There are no significant laminar differences in the maximum average firing rate. Spontaneous and maximum rates are correlated (r = 0.4, P < 10^–6).

View larger version (25K): [in this window] [in a new window]   FIG. 8. Spontaneous rate. A: cumulative histogram shows a range of 0.01–20 Hz rates and a lower rate for simple cells (gray line). B: layers 2/3 have a very low spontaneous rate, while layer 5 has a high rate. C: depth profile. Cells in supragranular layers have low spontaneous rates.

The three examples in Fig. 9A illustrate the wide range of contrast response curves observed. All curves could be well fit by the nonlinear Naka-Rushton equation (Albrecht and Hamilton 1982; Naka and Rushton 1966) R = R_s + gCⁿ/(sⁿ + Cⁿ), where R_s is the spontaneous rate and C is the contrast. The fitting parameters are: the contrast at mid-saturation s, the gain g, and the exponent n. For most cells in the LGN a good fit can be achieved with n = 1 (see e.g., Van Hooser et al. 2003), but many cells in our V1 sample are more nonlinear and require fits with higher exponents. From the fits, we could calculate the relative maximum gain (RMG). This is the maximum slope of the graphs in Fig. 9A if the maximum rate minus the spontaneous rate is normalized to 1. The RMG is a measure for the nonlinearity in response to contrast and is much less dependent on the fitting procedure than the exponent n or the gain g. A completely linear contrast response curve would correspond with RMG = 0.01%^–1 = 1. Any nonlinear curve will have a higher RMG. The median RMG is 2.8. We did not find any simple cells with a very nonlinear contrast response curve as can be seen from the simple cell RMG distribution in Fig. 9B. The general difference between the linearity of complex and simple cells is also clear in their population medians of 2.8 and 2.4 (P = 0.004).

View larger version (22K): [in this window] [in a new window]   FIG. 9. Contrast tuning. A: examples of contrast dependence, from very linear (top) to very nonlinear (bottom). Solid lines, Naka-Rushton fits; dashed lines, the C50 value, the contrast at half the peak response. The value in the bottom right corner is the relative maximum gain (maximum slope of graph with normalized response). B: cumulative histogram of relative maximum gain. Complex cells generally have higher maximum relative gains than simple cells (median: 2.8 vs. 2.4, P = 0.004). C: cumulative histogram of contrast at half-peak response (C50). Complex cells tend to have lower C50 values (median: 34 vs. 41%, P = 0.02).

Color

We ran two tests specifically designed to study color sensitivity. We calculated an S-cone isolating color pair s₊ and s_– and an M-cone isolating pair m₊ and m_–. The color s₊ excites the S cone much more than s_– does, whereas both equally activate the M cone. The pair m₊ and m_– differentially activate the M cone. At the optimal spatial frequency, we showed a family of drifting grating stimuli that linearly interpolate between the S-cone isolating grating and the M-cone isolating grating (Chatterjee and Callaway 2002; Diller et al. 2004; Johnson et al. 2004). The color of the up-phase was m₊ + (1 – )s_- and the down-phase was m_– + (1 – )s₊. Thus for = 0, this constitutes an S-cone isolating stimulus; for = 1, an M-cone isolating stimulus at the opposite phase; and intervening values of are combinations of the two. We also measured the spatial frequency tuning using a pair of equiluminant colors that gave maximally different responses in each cone type while the sum of the two cone-type activity was identical.

To improve our intuition for how spatial profiles of cone input influence the responses to these stimuli, we constructed linear model cells, shown in Fig. 10A, and computed their responses. The graphs in the left column show spatial profiles W_S(x) of S- (solid line) and W_M(x) of M-cone (dashed line) input along the direction of movement of the stimulating grating. The middle graphs show model spatial frequency tuning for luminance contrast gratings in black and the spatial frequency tuning for equiluminant color gratings in gray. The right column shows responses for the mixtures of cone isolating stimuli. These pictures illustrate a few properties of the stimuli. If the profiles of S- and M-cone input are identical up to a constant positive factor, i.e., [W_S(x)/[W_S(x) + W_M(x)] = c everywhere and 0 c 1, there will be a mixture _min of the two cone-isolating stimuli that gives no response, such as in the top figure. For an ideal pair of cone-isolating stimuli that equally modulate the S and M cones, this point would be equal to the relative S-cone weight, i.e., _min = c, for a linear neuron. If a neuron receives only M-cone input, _min = 0, whereas only S-cone input corresponds to _min = 1. When S- and M-cone inputs are almost identical, i.e., c 1/2, the response to an equiluminant grating vanishes. In general, if the S- and M-cone inputs are not proportional to each other, a cell responds for all . One exception to this rule is when both S- and M-cone inputs are on-like or both are off-like. Even if the spatial profiles are not proportional to each other, there will be a mixture where there is no response, see Fig. 10A, second row. In such a case, the spatial frequency tuning will be low-pass for both black-and-white and colored stimuli. In recognition of these properties, we name nonopponent those cells that have a point in the color mixture stimulus where they do not fire above spontaneous rate (by 3 SD) and thus lack cone opponency. Cells that respond above spontaneous rate for mixtures of all , we name cone-opponent. Cells are called band-pass if they are spatial band-pass filters for both the contrast luminance and the equiluminant gratings and low-pass otherwise.

View larger version (22K): [in this window] [in a new window]   FIG. 10. Color. A: examples of linear cell model responses. Left: spatial profiles in the drift direction of M-cone (dashed) and S-cone (solid) input. Middle: spatial frequency tuning for luminance (black) and equiluminant (gray) gratings. Right: response for mixtures of phase-shifted S- and M-cone isolating stimuli from S-cone isolating at = 0 to M-cone isolating at = 1. B: examples of measured responses. Left: spatial frequency tuning for luminance (black) and equiluminant (gray) gratings. Right: response for mixtures of cone isolating stimuli. C: histogram of the cone balance giving the least response. Few cells respond least to M-cone isolating stimuli. Colors correspond to legend of E. D: the optimal spatial frequency is most often lower for equiluminant gratings than for luminance gratings. E: total and laminar percentages of color groups. Cone opponency is almost absent in layer 6, very distinct from the other layers (P = 0.0001). F: percentage per class of cells with HWHH <90°. Cells with band-pass spatial frequency tuning are nearly all orientation-selective, whereas only around half of the cells in other classes are oriented (P < 0.0001). There is no difference between cone-opponent and nonopponent cells (P = 0.27). G: percentage per class of cells with DI > 0.5. Direction selectivity is nearly absent in cone-opponent cells. There is no difference between band-pass cells and low-pass cells (P = 0.4).

A histogram of the cone balance _min is plotted in Fig. 10C. The colors correspond to the colors in E. Most of the cells have a low _min, which means they respond preferentially to M-cone input, like the top three examples shown in Fig. 10B. The distribution of _min is not significantly different for cone-opponent versus nonopponent cells (comparing the dark bars with the lighter bars). The distribution of _min is wide but skewed toward low values. The average value of _min is 0.21 for cone-opponent cells and 0.22 for nonopponent cells for which it is an approximation to the relative S-cone weight. The majority of cells (68%) show a significant response for all , even at _min, and are classified as cone-opponent. A fifth (26/122) of the cells do not fire (3 SDs) above spontaneous rate for the S-cone isolating stimulus, whereas a 10th (12/122) of the neurons do not respond to the M-cone isolating grating.

Figure 10D shows that the preferred spatial frequency is significantly lower for equiluminant gratings than for luminance contrast gratings (0.14 vs. 0.21 cpd, P < 10^–8). The percentage of low-pass cells is 29% when measured with equiluminant gratings, whereas it is only 19% measured with a luminance contrast gratings. This suggests that the spatial antagonism between excitation and suppression is not as well matched per cone type as it is for the summed cone inputs.

The laminar distribution of cell classes, shown in Fig. 10E, reveals a striking absence of cone-opponent cells in layer 6 (significantly different from other layers, P = 0.0002). The layer 4/5 border has relatively many cone-opponent cells (not shown).

Cone opponency and orientation tuning are not correlated. Figure 10F shows that the percentage of oriented cells (HWHH < 90°) in the two cone-opponent categories is similar to that in the nonopponent categories (P = 0.3). The plot, however, does show that nearly all of the spatial band-pass cells are tuned for orientation while only around half of the low-pass cells are (P < 0.0001).

The situation is very different for cone opponency and direction selectivity. Cone-opponent cells have a much lower direction selectivity than nonopponent cells (median: 0.16 vs. 0.38, P < 10^–4). Figure 10G shows that virtually none of the opponent cells have a direction bias (DI > 0.5), which is very different from the nonopponent cells (P < 10^–4). There is no difference in direction selectivity of band- and low-pass cells (P = 0.4).

Cone-opponent cells have a higher high cut-off temporal frequency (median: 21 vs. 15 Hz, P = 0.01) and a much shorter latency (median: 31 vs. 48 ms, P = 0.0001). Only 16% of the nonopponent cells have latencies <34 ms, the overall population median. The cone-opponent cells are also much more sustained (median TI: 0.81 vs. 0.99, P < 0.001) and have a lower C₅₀ (median: 31 vs. 44%, P = 10^–5). The groups of cone-opponent and nonopponent cells were not significantly different in other properties such as F₁/F₀ ratio, optimal spatial and optimal temporal frequencies.

To compare with Johnson et al. (2001) and Shapley and Hawken (2002), we computed the color sensitivity, defined as the ratio of the maximum firing rate for the preferred equiluminant grating over the maximum rate for the luminance contrast stimulus. The median color sensitivity is 0.71. Three-quarters of the cells (116/154) are in the color-luminance category of Johnson et al. (2001) with color sensitivities between 0.5 and 2. Only one cell in our sample responds to the equiluminant grating at more than twice the rate of the luminance contrast grating. Of course, we may have missed cells that do not respond at all to luminance contrast gratings because we used black-and-white gratings when searching for cells. Color sensitivity is not different in the different cell classes nor is it significantly correlated to direction selectivity or orientation tuning, optimal spatial frequency tuning, temporal frequency tuning, or latency.

Apart from the laminar differences, we did not find a clustering of cone opponency or color sensitivity for nearby neurons on a single penetration. The cone opponency was not significantly different on 9 of 10 penetrations for which we studied cone opponency. The 10th penetration had a lower color opponency but a Kruskal-Wallis P value of 0.04 only. Color sensitivity varied even less between penetrations. Thus we did not find evidence for dramatic clustering of color processing, although we cannot exclude the possibility of a more subtle organization.

Multivariate analysis

Except for the F₁/F₀ ratio, none of the cells' individual properties show nontrivial bimodality or clustering seen by eye or by the Hartigan's dip test (Hartigan and Hartigan 1985). To see if our data suggest the existence of discrete groups of cell types, we looked at the clustering of cells in multi-dimensional space. We took various subsets of the measured properties and projected these to their principal component axes. The Hartigan's dip test did not report a significant bimodality along any of these projections. This certainly does not exclude the existence of cell classes, but we have not found an indication that there are. There is not one axis that explains most of the variance, suggesting that most properties vary mostly independently. Principal component analysis reflected the correlations that we have already mentioned, such as that between sharper orientation tuning and higher F₁/F₀ ratio but did not unveil any additional underlying structure.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION GRANTS ACKNOWLEDGMENTS REFERENCES

 
We found all the major classes of cell types reported in the V1 of other mammals. Unoriented cells, simple and complex oriented cells, both with and without end-stopping are present in gray squirrel V1. We first compare the individual response properties in the gray squirrel to those in other species before discussing their laminar organization.

Orientation tuning

Orientation-selective neurons in the gray squirrel have tuning widths that are very similar to those observed in other mammals. If we consider only the oriented cells (with HWHH < 90°), the median tuning width (HWHH) of gray squirrel V1 neurons is 25°. This is remarkably similar to both carnivores and primates, 19–25° in cat (Felis domesticus) (Gilbert 1977; Henry et al. 1974), 21° (mean) in ferret (Mustela putorius furo) (Alitto and Usrey 2004), 29° in mink (Mustela vison) (LeVay et al. 1987), 27° in baboon (Papio ursinus) (Kennedy et al. 1985), 27° in owl monkey (Aotus trivirgatus) (O'Keefe et al. 1998) and a median width of 24° at 71% of the maximum response in macaque (Macaca fascicularis and M. mulatta) (Gur et al. 2004; Ringach et al. 2002).

A quarter of the cells are unoriented and have HWHH of 90°. This is much more than in the cat, where unoriented cells are very rare (Baumgartner et al. 1965; Gilbert 1977; Hubel and Wiesel 1962; 1%, Murphy and Berman 1979; 5%, Kato et al. 1978), and a little more than most reports in primate, such as macaque (16% M. mulatta, Schiller et al. 1976a; 14% M. fascicularis, De Valois et al. 1982b) and owl monkey (10%, O'Keefe et al. 1998), although one study in macaque reports a similar number (31% M. nemestrina) (Bullier and Henry 1980). It is similar to the slightly smaller tree shrew (Tupaia glis), where also a quarter of the cells is not oriented (Humphrey and Norton 1980). In comparison with other rodents, the gray squirrel number falls in the middle. The mouse (Mus musculus) has many more unoriented neurons in V1, 55% (Drager 1975), 46% (Metin et al. 1988), 49% (Mangini and Pearlman 1980). So has the golden hamster (Mesocricetus auratus), 60% (Tiao and Blakemore 1976). In the related red squirrel (S. vulgaris), 44% (Polkoshnikov and Supin 1988) of cells sampled in V1 and 17% of cells sampled in the binocular region (Polkoshnikov and Revishchin 1998) were classified as nonoriented. Both tree squirrel species have more unoriented cells than the latest estimates in the rat (Rattus norvegicus), 16% (Parnavelas et al. 1981) and <7% (Girman et al. 1999). Older studies in the rat differ and find as many as 70% unoriented cells (Shaw et al. 1975; Wiesenfeld and Kornel 1975). Girman et al. (1999) suggest that their use of a different anesthetic, paralyzing the animal, and a smaller craniotomy make their higher estimate of orientation selectivity in the rat more reliable. Table 2 shows orientation tuning across several mammalian species.

The fraction of oriented cells across mammals shows two trends. Our interpretation of the literature is that nocturnal and crepuscular species, which have rod-dominated retina's or like the owl monkey only have a single cone-type (Jacobs et al. 1993), have more oriented cells than diurnal species of similar size. It is difficult to answer the question whether diurnal animals trade orientation-selective neurons or direction-selective neurons for color-selective neurons. We do see, however, that cone-opponent neurons are just as likely to be strongly orientation-selective as nonopponent neurons. There is also no significant negative correlation between the orientation index and color sensitivity. Another trend is that the smallest animals, mouse (adult weight: 25 g) and hamster (adult weight: 90–150 g), have the most unoriented cells and the largest animals, cat and primate have the fewest. An exception to this is the rat (adult weight: 350 g), which is smaller than gray squirrel (adult weight: 600 g) and has more oriented cells, but the rat is nocturnal and may have more oriented cells than a similarly sized diurnal animal.

Simple/complex

Oriented simple and complex cells (Hubel and Wiesel 1962) have been reported in several mammals. We did not measure subfield segregation but made a simple/complex classification based on the relative modulation. This classification is correlated to the original classification based on a neuron's subfield segregation (Mata and Ringach 2005; Movshon et al. 1978a; Schiller et al. 1976c; Skottun et al. 1991; but see Kagan et al. 2002). Simple cells, i.e., oriented cells with segregated ON and OFF subfields (Hubel and Wiesel 1962) showing relatively linear spatial summation, tend to have a high F₁/F₀ (Kagan et al. 2002). Complex cells are nonlinear in their spatial summation and tend to have a low F₁/F₀. Many complex cells in the macaque, however, have a high relative modulation (Kagan et al. 2002), and a study in the cat showed that many simple cells have a low modulation ratio (Movshon et al. 1978a). We called the three-quarters of the cells in our sample that had F₁/F₀ < 1 "complex" and the remaining quarter "simple." Our data do not allow us to tell whether simple and complex cells are truly separate cell classes or different ends of a continuous population as suggested recently (Chance et al. 1999; Mata and Ringach 2005; Mechler and Ringach 2002). The simple/complex distinction based on F₁/F₀ ratio certainly correlates with several other response properties. The population of simple-like cells with F₁/F₀ > 1 responds faster, is more sharply tuned for orientation, has higher direction selectivity, lower spontaneous rate, and lower maximum contrast gain, and responds to lower spatial frequencies. The increased modulation as well as selectivity for orientation and direction, and the lower spontaneous rate of simple cells, could be the effect of a higher spike threshold in these cells compared with complex cells (Priebe et al. 2004).

In gray squirrel, simple cells are more orientation-selective than complex cells. This is also true in cat (Gizzi et al. 1990; Murphy and Berman 1979), ferret (Alitto and Usrey 2004), tree shrew (Kaufmann and Somjen 1979), and macaque (Gur et al. 2004; Schiller et al. 1976b; Ringach et al. 2002). An exception is the wallaby marsupial (Ibbotson and Mark 2003) in which a larger fraction of the population of complex cells was sharply tuned than the simple cell population. In rodents, the situation is unclear, as relatively few of the oriented cells are complex in mouse (Mangini and Pearlman 1980) and rat (Girman et al. 1999). The lower spontaneous rate of simple cells is a universal finding, reported in mouse (Drager 1975), rat (Parnavelas et al. 1981), cat (Pribram et al. 1981), macaque (Schiller et al. 1976a), and wallaby (Ibbotson and Mark 2003).

The question of whether simple and complex cells prefer different spatial frequencies has been studied extensively in cat. Maffei and Fiorentini (1973) reported that complex cells prefer lower spatial frequencies. Other authors do not find any significant difference in preferred spatial frequencies (Ikeda and Wright ^</