- Email: [email protected]
Spatial frequency characteristics of nearby neurons in cats’ visual cortex
Neuroscience Letters 418 (2007) 242–247
Spatial frequency characteristics of nearby neurons in cats’ visual cortex St´ephane Molotchnikoff a,∗ , Pierre-Camille Gillet a , Svetlana Shumikhina a,b , Marilyn Bouchard a a
D´epartement de Sciences Biologiques, Universit´e de Montr´eal, C.P. 6128, succ. Centre-ville, H3C 3J7, Montr´eal, PQ, Canada b Institute of Higher Nervous Activity and Neurophysiology, Russian Academy of Sciences, Moscow 117485, Russia Received 9 October 2006; received in revised form 1 March 2007; accepted 16 March 2007
Abstract Various methods have allowed mapping of responses to several stimulus features on the cortical surface, particularly edge orientation and motion direction. The cortical mapping of spatial frequencies (SF), which is the basic property that leads to perception of spatial details of visual objects, is still controversial. We recorded simultaneously extracellular action potentials from neighboring cells in superficial layers of the area 17–18 border region of anesthetized cats. Responses of nearby cells to sine-wave gratings of varying SF were analyzed. Spatial frequency tuning curves were cross-correlated to establish the degree of similarity between the curves and optimal SFs were compared for each pair of neurons. The investigation showed that only about a half of nearby neurons exhibited close optimal SFs and similar tuning curves. The results suggest that SF channels do not show a clear clustering within a small pool of neurons. Such organization may contribute to the perception of spatial details at all orientations and motion directions. © 2007 Elsevier Ireland Ltd. All rights reserved. Keywords: Vision; Spatial frequency tuning; Clustering; Visual cortex; Cat
Investigations of evoked responses recorded from cortical single cells established that neurons are grouped within functional and anatomical domains. For instance, Hubel and Wiesel [14] reported many decades ago, the systematic columnar grouping of neurons with similar orientation preference and receptive field locations. Similarly, cells driven through the same eye are grouped in ocular dominance columns. These clusters have been well documented with various techniques including optical imaging and electrophysiological recordings [2,7,18]. Less can be said of the cortical representation of spatial frequency (SF) [24]. In recent years, several models have been hypothesized. It has been proposed that orientation columns are organized along high- or low SFs [4,3,16,22]. Others [9] suggested that SF preferences are continuously distributed with an absence of clear clustering, while Maffei and Fiorentini [19] described a laminar organization for SF. Alternatively, Issa et al. [17], suggested that pinwheel centers of the orientation layout alternate as loci for columns of high and low SFs. Others [8,27] depicted a cylinder
∗
Corresponding author. Tel.: +1 514 343 6616; fax: +1 514 343 2293. E-mail address: [email protected] (S. Molotchnikoff).
0304-3940/$ – see front matter © 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.neulet.2007.03.043
like model with an additional scheme that CO blobs are sites of low SF preference. On the other hand, Bressloff and Cowan [5] conjectured a spherical model instead in which orientations are distributed along azimuthal axes while low- and high SFs are located in polar regions. More recently, Sirovich and Uglesich [24], using optical imaging methods, described an absence of any grouping along SF preferences because every pixel may be interpreted as an admixture of low- and high-pass SF cellular populations. Finally, Tolhurst and Thompson [26] reported that neurons within 100 were likely to have similar optimal SFs, but even cells recorded close together sometimes differed markedly in their preferred SF. Hence, the conclusions are not easy to draw. The above brief survey strongly indicates that the cortical functional organization of SF preferences is still unsettled and merits further investigation. Electrophysiological recordings provide better resolution than imaging techniques. Thus, we decided to perform simultaneous recordings from nearby neurons in visual cortex. With a single electrode, we recorded multiunit neuronal activity from a relatively small pool of cells responding to the optimal for multiunit activity orientation. This procedure allows testing of adjacent cells subjected to identical
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
conditions and it offers several significant advantages. First, cells are situated close to each other, and second, nearby neurons are tested simultaneously so differences in SF tuning properties cannot be ascribed to fluctuations in vegetative physiology that may arise across different recording locations and different cats. Two to four (and rarely five) nearby neurons could be sorted out and for each unit we examined the SF tuning curves. Then tuning preferences between neighboring cells were correlated to uncover whether neurons optimally sensitive to a narrow band of SFs were grouped together. Both simple and complex cell responses were analyzed. We show that only about a half of nearby neurons show similar optimal SFs. This is partly in disagreement with results of DeAngelis et al. [7] who undertook a similar attempt to study SFs of simultaneously recorded neurons (by the same electrode) in cat visual cortex and found a higher degree of SF clustering. However, only responses of simple cells were analyzed in their study and only optimal SFs of nearby neurons were compared but not the entire SF tuning curves. Furthermore, simple cells were recorded in cortical layers 3, 4 and 6 but not in the superficial layers as in our experiments. Hence, our study provides additional details on SF clustering in visual cortex. Adult cats (2.5–3.2 kg) were used in the investigation. All experimental protocols followed the regulations of the Canadian Council on Animal Care (CCAC) and the corresponding US National Institutes of Health regulations and were approved by the Institutional Animal Care and Use Committee at the Universit´e de Montr´eal. Animals premedicated with Atravet (acepromazine maleate, 1 mg/kg, i.m.) and atropine sulfate (0.04 mg/kg, i.m.) were initially anesthetized with ketamine hydrochloride (25 mg/kg, i.m.). Xylocaine (lidocaine hydrochloride, 2%) was used as a local anesthetic. Cats were prepared for electrophysiological recordings in a conventional fashion. Animals were paralysed with gallamine triethiodide (flaxedil, initial dose 40 and 10 mg/kg during the experiment, i.v.) and anesthesia was maintained with a mixture of gases (N2 O/O2 – 70/30 supplemented with 0.5% Isoflurane) for the duration of the experiment. Flaxedil was delivered to the animals continuously in the mixture of 5% dextrose in lactated Ringer’s solution. Plano contact lenses with artificial pupils (5 mm diameter) were placed on the cat’s eyes to prevent the cornea from drying. The pupils were dilated with atropine sulfate (1%) and the nictitating membranes were contracted with phenylephrine hydrochloride (2.5%). A heating pad was used to maintain the body temperature at 37.5 ◦ C. Electroencephalogram, electrocardiogram and expired CO2 were monitored throughout the experiment to ensure an adequate level of anesthesia. The end tidal CO2 partial pressure was kept constant between 28 and 30 mmHg. The antibacterial agent tribrissen (24%, 30 mg/kg per day, s.c.) and the antibiotic duplocillin (0.1 ml/kg, i.m.) were administered to the animals. Multiunit activity in the visual cortex (area 17 (P = 0.0–5.0 mm; L = 0.5–2.0 mm [21]) superficial layers, 300–600 m deep (up to ∼1000 m in few cases), was recorded by tungsten microelectrodes enclosed in stainless steel tubing (impedance 10 M at 1000 Hz). The boundaries of area 17 were recognized according to the stereotaxic atlas
243
of Reinoso-Suarez [21] (Fig. 1D). After the microelectrodes were inserted, the cortex was covered by warm agar (3–4% in saline) and wax. The neuronal action potentials were amplified and sent to a computer for voltage discrimination and recording with 0.05 ms resolution for on-line and off-line analyses (Experimenter’s WorkBench in earlier experiments and SciWorks in later experiments, DataWave Technologies). Multiunit recordings from one electrode usually included 2–4 well isolated single units which were thresholded, that is, isolated from the noise. After clearly detectable activity (single spikes within multiunit activity should exhibit an amplitude well above the neuronal noise) was obtained on the microelectrode, the compound receptive field (cRF) of the group of cells was determined using a hand-held projector with a narrow slit of light projected on a translucent screen placed 57 cm from the cat’s eyes. The compound receptive field corresponded to the sum of classical receptive fields of individual cells recorded by the same electrode. Qualitative properties such as dimensions, orientation and directional selectivity, etc. For quantitative tests, visual stimuli were generated by commercial software (Vision Works for neurophysiology program of Vision Research Graphics Inc., Durham, USA) and displayed on a cathode ray screen (Mitsubishi Electronics, effective display area of 380 mm × 285 mm, with a refresh rate of 120 Hz) centered on the cRF and synchronized with the data acquisition processes. Drifting, sinusoidally modulated grating patches covering the cRF (5.0 cd/m2 , 80% contrast) had a square aperture surrounded by a dark background. In all tests, optimal orientation and direction, with temporal frequency set at 2 Hz, were presented in blocks of interleaved trials and shown to the dominant eye. During these runs, peristimulus time histograms (PSTHs) were accumulated. Each stimulus was presented for 4096 ms, 25 times. Spatial frequency tuning curves were obtained by presenting drifting gratings of different SF (0.15–2 cyc/deg) in a pseudorandom sequence in steps of 0.5 octaves (with some additional values of 0.25 octaves). Spontaneous activity was recorded in the absence of visual stimulation for the same period of time. Individual units were sorted out from within multiunit activity by a spike separation method using commercial software. Spike sorting is based on the established assumption that action potentials from different cells have different amplitudes and temporal characteristics and that these characteristics are stable during a single trial recording and across trials. Because spike separation is performed off-line, attention was focused on data acquisition. Tests were made during control recordings to insure that a time window of on-line unit extraction was sufficient to reproduce fully spike waveforms off-line. During the recordings, the action potentials were detected by their voltage threshold crossing and the unit extraction was centered on the peak of action potentials. Usually, 2–3 ms of digitized voltages with a peak pre-time of 0.5–0.7 ms were sufficient to reproduce the shape of action potentials. The spike sorting procedure was performed automatically by the software using eight parameters such as amplitude (height) and width of peaks and valleys of the action potentials, spike areas and ratio of peaks. These features (eight parameters) formed clusters and the Z-score was obtained to esti-
244
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
Fig. 1. Tuning curves to spatial frequency (SF) of two neighboring cells recorded from two experiments (Sites 1 and 2). First site: panels A and B showing band-pass and low-pass curves, respectively. Optimal SF = 0.21 in (A) and 0.15 in (B), cycles per degree (cyc/deg). Goodness-of-fit R2 = 0.93 and 0.96, respectively, for curve fit. Action potential waveforms are shown in panel (C). Z-score = 4.4. (D) Location of microelectrodes on the cortical surface (26 sites in the area 17–18 border region). Second site: panels (E and F) showing two band-pass curves. Optimal SFs are 0.35 and 0.8 cyc/deg. Goodness-of-fit R2 = 0.9 and 0.88, respectively. y-axis: number of spikes per second, ±S.E.M. Spike waveforms, clusters and auto-correlograms (note absence of spikes during the absolute refractory period) of sorted neurons are shown in (G). Z-score = 6.5. Amplitudes of action potentials indicated in V. Time scale: 1.8 ms in (C) and 2.2 ms in (G). Note stability of clusters in spontaneous activity and in responses of sorted cells. The third green cluster corresponds to a third neuron which was of a low-pass type and had a preferred SF of 0.15 cyc/deg.
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
mate the statistical significance of spike separation (Z-score had to be superior to 2.5). Elliptical cluster boundaries were used. Color coded discriminated spikes were individually visualized to ensure the waveform stability of spikes belonging to a particular cluster. Superposition and averaging of different waveforms in the chosen time window allowed confirming spike separation (Fig. 1C and G). Corrections could be made by manually adjusting cluster boundaries. As an additional control, a raster plot of activity with color coded isolated spikes and histograms of auto- and cross-correlation analyses between isolated spikes were checked for possible errors of spike separation [11,12,23] (Fig. 1G). Excluded action potentials were globally considered as noise. The procedure was similar to one described by Fee et al. [10]. Regularly, up to four neurons could be reliably separated from the activity recorded by the same electrode, and the isolated units differed in shape and amplitude. We used both Autocut and Autosort programs. The latter allowed some additional testing using the principal component analysis. Spatial frequency tuning curves of individual neurons were classified as low-pass (LP), band-pass (BP) and high-pass (HP). The individual tuning functions were fitted by Log Gaussian functions Y = Baseline X/Center ∧ + Amplitude × exp(−0.5 × ln 2 Width where Baseline is the level of spontaneous activity, Amplitude is the height of tuning (maximum response), X is a spatial frequency, Center is the cell optimal SF and Width is the width of tuning at half height. The optimal SFs and bandwidths of individual cells were derived from the equation. To estimate the similarity of SF tuning curves, the Pearson correlation coefficient between two curves (at stimulus conditions) was computed. Linear regression analysis was performed to obtain the probability values. Each recorded neuron was classified as simple or complex cell on the basis of their response modulation (AC/DC ratio) [25]. Cells for which AC/DC ratio exceeded 1.0 were considered to be simple while cells with AC/DC ratio less than 1.0 were classified as complex. Spike sorting yielded 82 cells from 26 recorded neuronal pools. This comprised 35 simple and 47 complex cells. For every neuron, we measured the SF tuning curve with the average response magnitude. Except for SF, the other various gratings parameters were fixed. Hence, we determined the evoked firing rate in relationship to the SF. Fig. 1 illustrates tuning curves of two units recorded simultaneously from two recording sites. The spike waveforms are displayed in each case in panels C and G. The cluster analysis indicated a highly significant separation, Z score = 4.4 for cells A, B; and Z = 6.5 for cells E, F. For cells A and B (first site) the respective optimal SFs were 0.21 and 0.15 cyc/deg. More interestingly, these two cells of this pool showed quite different tuning curves. Indeed, cell A was a band-pass cell (bandwidth at half height 0.22 cyc/deg) while the companion neuron was a low-pass cell (bandwidth at half height 0.6 cyc/deg); its firing rate gradually diminished as the
245
stimulating SF increased. The second example from a different experiment illustrates two cells (E and F) that presented bandpass tuning curves. However, the respective optimal SFs were very different, 0.35 and 0.8 cyc/deg. That is a difference of 1.25 octaves. In addition, the bandwidths were also distinct. Cell F was very narrowly tuned around its preferred SF (bandwidth at half height = 0.12 cyc/deg), the bandwidth at half height of the nearby unit (E) was equal to 0.6 cyc/deg. It should be noted that three cells were sorted out at this site (all three clusters are shown in Fig. 1G). The third neuron was low-pass and had an optimal SF of 0.15 cyc/deg (SF tuning curve not shown). Thus neighboring cortical neurons have different SF properties even when the units are located nearby. Fig. 2A shows the preferred or optimal SF distribution. The distribution of optimal SFs spans a range from 0.13 to 2.0 cycles per degree with a mean value of 0.31 ± 0.03 (mean ± S.E.M.). This range is comparable to what has been reported originally for area 17 in cats [20], where no units were reported to respond to SF lower than 0.13 cyc/deg, nor higher than 2.0 cyc/deg. The distribution of Fig. 2A is biased towards low SFs perhaps reflecting the fact that some of our recording sites were located closely to the area 17/18 border and may include units located within area 18 where neurons prefer lower SFs [20]. No significant difference was observed between SF preferences of simple and complex cells (0.30 ± 0.06 and 0.32 ± 0.04, respectively) though complex cells seemed to prefer higher SFs than simple cells as suggested with previous studies [1]. Because no significant difference was found between simple and complex cells we will not describe further results according cell types. Panel (B) of Fig. 2 illustrates the distribution of curve types. The units exhibiting a band-pass property constitute about two thirds (58.6%) of the population while low-pass cells made up about 40% of the group. High-pass cells were very infrequent (one cell). In order to examine whether cells recorded by the same electrode tip belonged to one single functional group we correlated pair wise optimal SFs of all cells sorted out from the same site. The scatter plot shown in Fig. 2C shows that only about a half (43) of pairs were positioned on the equality line indicating the same optimal SF for both cells. Even in the range 0.2–0.8 cyc/deg a significant portion of neighboring cells had dissimilar SF preferences especially those of band-pass type (Fig. 2D). It is worth noting that only 7 pairs out of 43 have identical optimal SFs. Fig. 2E supplements the previous analysis; it shows the distribution of the correlation coefficient (r, Pearson coefficient). For this computation, we correlated the data points of the entire tuning curves between neighboring neurons recorded simultaneously at one site. High correlation values (∼0.7 and higher, see significant values on the graph) indicate that both cells presented very analogous curves; conversely, a weak or even negative correlation points toward a lack of similarity between both curves. This histogram shows that less than half of pairs of neighboring cells have comparable curves (44.9%, r ≥ 0.7). Half of cell pairs (49%) had coefficient values ranging from −0.4 to +0.4. The distribution of Fig. 2E was compared with a distribution of correlation coefficients for pairs of neurons belonging to different sites (non-neighboring
246
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
Fig. 2. Population analyses. (A) Distribution of optimal SFs in cycles per degree (cyc/deg). Optimal SFs are derived from curve fit and rounded. (B) SF tuning curve type distribution with the proportion of each curve type indicated above the bars, LP: low-pass, BP: band-pass, HP: high-pass. (C) The optimal SF of one cell is plotted against the optimal SF of the companion cell. Symbols (+) and (䊉) indicate non-significant (ns) and significant (sign., p < 0.05) correlation between the full tuning curves for each unit, respectively. (D) Only band-pass cells are plotted (43 out of 98 pairwise combinations). Note that since some cells have the same optimal SFs some points are superimposed. (E, F) Distributions of the Pearson correlation coefficient (r) between tuning curves of nearby neurons (E, N = 98) and non-neighboring cells (F, N = 126).
cells, shuffling by one experiment was performed) (Fig. 2F). Importantly, these two distributions were not significantly different (p = 0.29, one-tailed Mann–Whitney non-parametric test), thus indicating the high degree of similarity between SF preferences of pairs of nearby cells and non-neighboring neurons. The results further confirm that only about a half of nearby neurons show similar SF preferences and that there is an absence of a clear clustering of SFs in the superficial layers of the area 17–18 border region in cat visual cortex. In addition to demonstrating that nearby cortical cells, tested simultaneously, exhibit very different optimal SFs, the present study revealed several new findings. First, tuning curves of
neighbouring units display various profiles. Second, even nearby neurons whose tuning curves are of band-pass type exhibited different preferred SFs. Thus, results of the present investigation suggest that the majority of neighboring cells exhibit different optimal SFs and dissimilar tuning curves. Only about less than half (45%) of cells exhibited a high pair-wise correlation suggestive of a grouping of neurons with similar tuning curves, hence only these cells have a close optimal SF. Globally, however, our results seem to indicate an absence of a clear clustering or, at best, a very sparse clustering of SFs in the superficial layers of the area 17–18 border region in cat visual cortex. This conclusion appears to be in agreement with Issa et al. [17] maps which
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
show that SF preferences at the extremes of the cat’s SF spectrum are separated by no more than 0.75 mm. This conclusion was suspected by Tolhurst and Thompson [26], although with a quite different approach; they reported optimal SFs can differ markedly from neuron to neuron. It should be noted however that since 75% of units in our sample had preferences between 0.15 and 0.3 cyc/deg, the clustering of the shuffled data was already quite tight. This could influence a comparison of SF tuning curves on a single electrode to tuning curves from separate recordings (shuffled control) showing no statistical difference in the distributions. Computations performed on extracellular recordings made in the hippocampus suggest that spike amplitude of neurons superior to 60 V may be recorded within a radius of 50 [6,13]. Though the spike amplitudes are different in the cortex and hippocampus some assumptions may be reasonable. The average spike amplitude of the present study was 30 ± 3 V. Because the amplitude of extracellular spikes decreases rapidly with distance, the distance between the cell body of the contributing to the multiunit activity neuron and the recording tip is at most 100–150 from the recorded loci. That is roughly 2–3 orientation columns, which spans about 20–30◦ of axis of orientation [2,15]. Although, we did not measure orientation tuning curves, the optimal orientation of the grating for multiunit activity was determined by finding which grating axis evoked the largest firing. Thus, it is most likely that sorted cells whose optimal orientations were too different from that of the multiunit responded too weakly to be analyzed since they were not optimally stimulated. For that reason, it is logical to assume that we studied cells sharing similar or close optimal orientation. Therefore, they belong to the same functional domain. Furthermore, it has been demonstrated that orientation preference is unrelated to the optimal SF [14,24,26] at one particular site. Such a lack of grouping along similar SF preferences allows very large combinations of orientations and SFs to be overlaid. Indeed the presence of cells with several optimal SFs, sometimes several octaves apart, within a relatively small pool of cells that share close orientation preferences means that a large SF spectrum is represented in a narrow band of orientations and vice versa. Such an organization may be the neural basis that underlies the successful capture of multiple levels of spatial detail at all orientations at each location in the visual field. Acknowledgements The research was supported by NSERC (Canada) grants to S. Molotchnikoff. We thank Dr. S. Itaya and unknown reviewers for helpful comments. References [1] C. Bardy, J.Y. Huang, C. Wang, T. FitzGibbon, B. Dreher, ‘Simplification’ of responses of complex cells in cat striate cortex: suppressive surrounds and ‘feedback’ inactivation, J. Physiol. (Lond.) 574 (2006) 731–750. [2] A. Basole, L.E. White, D. Fitzpatrick, Mapping multiple features in the population response of visual cortex, Nature 423 (2003) 986–990.
247
[3] T. Bonhoeffer, A. Grinvald, Optical imaging based on intrinsic signals, in: A.W. Toga, J.C. Mazziotta (Eds.), Brain Mapping: The Methods, Academic Press, New York, 1996, pp. 55–97. [4] T. Bonhoeffer, D.S. Kim, D. Malonek, D. Shoham, A. Grinvald, Optical imaging of the layout of functional domains in area 17 and across the area 17/18 border in cat visual cortex, Eur. J. Neurosci. 7 (1995) 1973–1988. [5] P.C. Bressloff, J.D. Cowan, A spherical model for orientation and spatialfrequency tuning in a cortical hypercolumn, Philos. Trans. R. Soc. Lond. B Biol. Sci. 358 (2003) 1643–1667. [6] G. Buzs`aki, Large-scale recording of neuronal ensembles, Nat. Neurosci. 7 (2004) 446–451. [7] G.C. DeAngelis, G.M. Ghose, I. Ohzawa, R.D. Freeman, Functional microorganization of primary visual cortex: receptive field analysis of nearby neurons, J. Neurosci. 19 (1999) 4046–4064. [8] R.L. De Valois, K.K. De Valois, Spatial Vision, Oxford University Press, New York, 1988, 381 pp. [9] R.M. Everson, A.K. Prashanth, M. Gabbay, B.W. Knight, L. Sirovich, E. Kaplan, Representation of spatial frequency and orientation in the visual cortex, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8334–8338. [10] M.S. Fee, P.P. Mitra, D. Kleinfeld, Automatic sorting of multiple unit neuronal signals in the presence of anisotropic and non-Gaussian variability, J. Neurosci. Methods 69 (1996) 175–199. [11] G.L. Gerstein, Cross-correlation measures of unresolved multi-neuron recordings, J. Neurosci. Methods 100 (2000) 41–51. [12] L. Hazan, M. Zugaro, G. Buzs´aki, Klusters, NeuroScope, ND Manager: A free software suite for neurophysiological data processing and visualization, J. Neurosci. Methods 155 (2006) 207–216. [13] D.A. Henze, Z. Borhegyi, J. Csicsvari, A. Mamiya, K.D. Harris, G. Buzs`aki, Intracellular features predicted by extracellular recordings in the hippocampus in vivo, J. Neurophysiol. 84 (2000) 390–400. [14] D.H. Hubel, T.N. Wiesel, Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex, J. Physiol. (Lond.) 160 (1962) 106–154. [15] D.H. Hubel, T.N. Wiesel, Uniformity of monkey striate cortex: a parallel relationship between field size, scatter, and magnification factor, J. Comp. Neurol. 158 (1974) 295–305. [16] M. H¨ubener, D. Shoham, A. Grinvald, T. Bonhoeffer, Spatial relationships among three columnar systems in cat area 17, J. Neurosci. 17 (1997) 9270–9284. [17] N.P. Issa, C. Trepel, M.P. Stryker, Spatial frequency maps in cat visual cortex, J. Neurosci. 20 (2000) 8504–8514. [18] S. LeVay, T.N. Nelson, Columnar organization of the visual cortex, in: A.G. Leventhal (Ed.), The Neural Basis of Visual Function, CRC, Boston, 1991, pp. 266–315. [19] L. Maffei, A. Fiorentini, Spatial frequency rows in the striate visual cortex, Vision Res. 17 (1977) 257–264. [20] J.A. Movshon, I.D. Thompson, D.J. Tolhurst, Spatial and temporal contrast sensitivity of neurones in areas 17 and 18 of the cat’s visual cortex, J. Physiol. (Lond.) 283 (1978) 101–120. [21] F. Reinoso-Suarez, Topographischer Hirnatlas der Katze, Herausgegeben von E. Merck A.G., Darmstadt (1961). [22] D. Shoham, M. Hubener, S. Schulze, A. Grinvald, T. Bonhoeffer, Spatiotemporal frequency domains and their relation to cytochrome oxidase staining in cat visual cortex, Nature 385 (1997) 529–533. [23] S. Shumikhina, J. Guay, F. Duret, S. Molotchnikoff, Contextual modulation of synchronization to random dots in the cat visual cortex, Exp. Brain Res. 158 (2004) 223–232. [24] L. Sirovich, R. Uglesich, The organization of orientation and spatial frequency in primary visual cortex, Proc. Natl. Acad. Sci. U.S.A. 101 (2004) 16941–16946. [25] B.C. Skottun, R.L. De Valois, D.H. Grosof, J.A. Movshon, D.G. Albrecht, A.B. Bonds, Classifying simple and complex cells on the basis of response modulation, Vision Res. 31 (1991) 1079–1086. [26] D.J. Tolhurst, I.D. Thompson, Organization of neurones preferring similar spatial frequencies in cat striate cortex, Exp. Brain Res. 48 (1982) 217–227. [27] M.A. Webster, R.L. De Valois, Relationship between spatial-frequency and orientation tuning of striate-cortex cells, J. Opt. Soc. Am. A 2 (1985) 1124–1132.
Spatial frequency characteristics of nearby neurons in cats’ visual cortex St´ephane Molotchnikoff a,∗ , Pierre-Camille Gillet a , Svetlana Shumikhina a,b , Marilyn Bouchard a a
D´epartement de Sciences Biologiques, Universit´e de Montr´eal, C.P. 6128, succ. Centre-ville, H3C 3J7, Montr´eal, PQ, Canada b Institute of Higher Nervous Activity and Neurophysiology, Russian Academy of Sciences, Moscow 117485, Russia Received 9 October 2006; received in revised form 1 March 2007; accepted 16 March 2007
Abstract Various methods have allowed mapping of responses to several stimulus features on the cortical surface, particularly edge orientation and motion direction. The cortical mapping of spatial frequencies (SF), which is the basic property that leads to perception of spatial details of visual objects, is still controversial. We recorded simultaneously extracellular action potentials from neighboring cells in superficial layers of the area 17–18 border region of anesthetized cats. Responses of nearby cells to sine-wave gratings of varying SF were analyzed. Spatial frequency tuning curves were cross-correlated to establish the degree of similarity between the curves and optimal SFs were compared for each pair of neurons. The investigation showed that only about a half of nearby neurons exhibited close optimal SFs and similar tuning curves. The results suggest that SF channels do not show a clear clustering within a small pool of neurons. Such organization may contribute to the perception of spatial details at all orientations and motion directions. © 2007 Elsevier Ireland Ltd. All rights reserved. Keywords: Vision; Spatial frequency tuning; Clustering; Visual cortex; Cat
Investigations of evoked responses recorded from cortical single cells established that neurons are grouped within functional and anatomical domains. For instance, Hubel and Wiesel [14] reported many decades ago, the systematic columnar grouping of neurons with similar orientation preference and receptive field locations. Similarly, cells driven through the same eye are grouped in ocular dominance columns. These clusters have been well documented with various techniques including optical imaging and electrophysiological recordings [2,7,18]. Less can be said of the cortical representation of spatial frequency (SF) [24]. In recent years, several models have been hypothesized. It has been proposed that orientation columns are organized along high- or low SFs [4,3,16,22]. Others [9] suggested that SF preferences are continuously distributed with an absence of clear clustering, while Maffei and Fiorentini [19] described a laminar organization for SF. Alternatively, Issa et al. [17], suggested that pinwheel centers of the orientation layout alternate as loci for columns of high and low SFs. Others [8,27] depicted a cylinder
∗
Corresponding author. Tel.: +1 514 343 6616; fax: +1 514 343 2293. E-mail address: [email protected] (S. Molotchnikoff).
0304-3940/$ – see front matter © 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.neulet.2007.03.043
like model with an additional scheme that CO blobs are sites of low SF preference. On the other hand, Bressloff and Cowan [5] conjectured a spherical model instead in which orientations are distributed along azimuthal axes while low- and high SFs are located in polar regions. More recently, Sirovich and Uglesich [24], using optical imaging methods, described an absence of any grouping along SF preferences because every pixel may be interpreted as an admixture of low- and high-pass SF cellular populations. Finally, Tolhurst and Thompson [26] reported that neurons within 100 were likely to have similar optimal SFs, but even cells recorded close together sometimes differed markedly in their preferred SF. Hence, the conclusions are not easy to draw. The above brief survey strongly indicates that the cortical functional organization of SF preferences is still unsettled and merits further investigation. Electrophysiological recordings provide better resolution than imaging techniques. Thus, we decided to perform simultaneous recordings from nearby neurons in visual cortex. With a single electrode, we recorded multiunit neuronal activity from a relatively small pool of cells responding to the optimal for multiunit activity orientation. This procedure allows testing of adjacent cells subjected to identical
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
conditions and it offers several significant advantages. First, cells are situated close to each other, and second, nearby neurons are tested simultaneously so differences in SF tuning properties cannot be ascribed to fluctuations in vegetative physiology that may arise across different recording locations and different cats. Two to four (and rarely five) nearby neurons could be sorted out and for each unit we examined the SF tuning curves. Then tuning preferences between neighboring cells were correlated to uncover whether neurons optimally sensitive to a narrow band of SFs were grouped together. Both simple and complex cell responses were analyzed. We show that only about a half of nearby neurons show similar optimal SFs. This is partly in disagreement with results of DeAngelis et al. [7] who undertook a similar attempt to study SFs of simultaneously recorded neurons (by the same electrode) in cat visual cortex and found a higher degree of SF clustering. However, only responses of simple cells were analyzed in their study and only optimal SFs of nearby neurons were compared but not the entire SF tuning curves. Furthermore, simple cells were recorded in cortical layers 3, 4 and 6 but not in the superficial layers as in our experiments. Hence, our study provides additional details on SF clustering in visual cortex. Adult cats (2.5–3.2 kg) were used in the investigation. All experimental protocols followed the regulations of the Canadian Council on Animal Care (CCAC) and the corresponding US National Institutes of Health regulations and were approved by the Institutional Animal Care and Use Committee at the Universit´e de Montr´eal. Animals premedicated with Atravet (acepromazine maleate, 1 mg/kg, i.m.) and atropine sulfate (0.04 mg/kg, i.m.) were initially anesthetized with ketamine hydrochloride (25 mg/kg, i.m.). Xylocaine (lidocaine hydrochloride, 2%) was used as a local anesthetic. Cats were prepared for electrophysiological recordings in a conventional fashion. Animals were paralysed with gallamine triethiodide (flaxedil, initial dose 40 and 10 mg/kg during the experiment, i.v.) and anesthesia was maintained with a mixture of gases (N2 O/O2 – 70/30 supplemented with 0.5% Isoflurane) for the duration of the experiment. Flaxedil was delivered to the animals continuously in the mixture of 5% dextrose in lactated Ringer’s solution. Plano contact lenses with artificial pupils (5 mm diameter) were placed on the cat’s eyes to prevent the cornea from drying. The pupils were dilated with atropine sulfate (1%) and the nictitating membranes were contracted with phenylephrine hydrochloride (2.5%). A heating pad was used to maintain the body temperature at 37.5 ◦ C. Electroencephalogram, electrocardiogram and expired CO2 were monitored throughout the experiment to ensure an adequate level of anesthesia. The end tidal CO2 partial pressure was kept constant between 28 and 30 mmHg. The antibacterial agent tribrissen (24%, 30 mg/kg per day, s.c.) and the antibiotic duplocillin (0.1 ml/kg, i.m.) were administered to the animals. Multiunit activity in the visual cortex (area 17 (P = 0.0–5.0 mm; L = 0.5–2.0 mm [21]) superficial layers, 300–600 m deep (up to ∼1000 m in few cases), was recorded by tungsten microelectrodes enclosed in stainless steel tubing (impedance 10 M at 1000 Hz). The boundaries of area 17 were recognized according to the stereotaxic atlas
243
of Reinoso-Suarez [21] (Fig. 1D). After the microelectrodes were inserted, the cortex was covered by warm agar (3–4% in saline) and wax. The neuronal action potentials were amplified and sent to a computer for voltage discrimination and recording with 0.05 ms resolution for on-line and off-line analyses (Experimenter’s WorkBench in earlier experiments and SciWorks in later experiments, DataWave Technologies). Multiunit recordings from one electrode usually included 2–4 well isolated single units which were thresholded, that is, isolated from the noise. After clearly detectable activity (single spikes within multiunit activity should exhibit an amplitude well above the neuronal noise) was obtained on the microelectrode, the compound receptive field (cRF) of the group of cells was determined using a hand-held projector with a narrow slit of light projected on a translucent screen placed 57 cm from the cat’s eyes. The compound receptive field corresponded to the sum of classical receptive fields of individual cells recorded by the same electrode. Qualitative properties such as dimensions, orientation and directional selectivity, etc. For quantitative tests, visual stimuli were generated by commercial software (Vision Works for neurophysiology program of Vision Research Graphics Inc., Durham, USA) and displayed on a cathode ray screen (Mitsubishi Electronics, effective display area of 380 mm × 285 mm, with a refresh rate of 120 Hz) centered on the cRF and synchronized with the data acquisition processes. Drifting, sinusoidally modulated grating patches covering the cRF (5.0 cd/m2 , 80% contrast) had a square aperture surrounded by a dark background. In all tests, optimal orientation and direction, with temporal frequency set at 2 Hz, were presented in blocks of interleaved trials and shown to the dominant eye. During these runs, peristimulus time histograms (PSTHs) were accumulated. Each stimulus was presented for 4096 ms, 25 times. Spatial frequency tuning curves were obtained by presenting drifting gratings of different SF (0.15–2 cyc/deg) in a pseudorandom sequence in steps of 0.5 octaves (with some additional values of 0.25 octaves). Spontaneous activity was recorded in the absence of visual stimulation for the same period of time. Individual units were sorted out from within multiunit activity by a spike separation method using commercial software. Spike sorting is based on the established assumption that action potentials from different cells have different amplitudes and temporal characteristics and that these characteristics are stable during a single trial recording and across trials. Because spike separation is performed off-line, attention was focused on data acquisition. Tests were made during control recordings to insure that a time window of on-line unit extraction was sufficient to reproduce fully spike waveforms off-line. During the recordings, the action potentials were detected by their voltage threshold crossing and the unit extraction was centered on the peak of action potentials. Usually, 2–3 ms of digitized voltages with a peak pre-time of 0.5–0.7 ms were sufficient to reproduce the shape of action potentials. The spike sorting procedure was performed automatically by the software using eight parameters such as amplitude (height) and width of peaks and valleys of the action potentials, spike areas and ratio of peaks. These features (eight parameters) formed clusters and the Z-score was obtained to esti-
244
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
Fig. 1. Tuning curves to spatial frequency (SF) of two neighboring cells recorded from two experiments (Sites 1 and 2). First site: panels A and B showing band-pass and low-pass curves, respectively. Optimal SF = 0.21 in (A) and 0.15 in (B), cycles per degree (cyc/deg). Goodness-of-fit R2 = 0.93 and 0.96, respectively, for curve fit. Action potential waveforms are shown in panel (C). Z-score = 4.4. (D) Location of microelectrodes on the cortical surface (26 sites in the area 17–18 border region). Second site: panels (E and F) showing two band-pass curves. Optimal SFs are 0.35 and 0.8 cyc/deg. Goodness-of-fit R2 = 0.9 and 0.88, respectively. y-axis: number of spikes per second, ±S.E.M. Spike waveforms, clusters and auto-correlograms (note absence of spikes during the absolute refractory period) of sorted neurons are shown in (G). Z-score = 6.5. Amplitudes of action potentials indicated in V. Time scale: 1.8 ms in (C) and 2.2 ms in (G). Note stability of clusters in spontaneous activity and in responses of sorted cells. The third green cluster corresponds to a third neuron which was of a low-pass type and had a preferred SF of 0.15 cyc/deg.
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
mate the statistical significance of spike separation (Z-score had to be superior to 2.5). Elliptical cluster boundaries were used. Color coded discriminated spikes were individually visualized to ensure the waveform stability of spikes belonging to a particular cluster. Superposition and averaging of different waveforms in the chosen time window allowed confirming spike separation (Fig. 1C and G). Corrections could be made by manually adjusting cluster boundaries. As an additional control, a raster plot of activity with color coded isolated spikes and histograms of auto- and cross-correlation analyses between isolated spikes were checked for possible errors of spike separation [11,12,23] (Fig. 1G). Excluded action potentials were globally considered as noise. The procedure was similar to one described by Fee et al. [10]. Regularly, up to four neurons could be reliably separated from the activity recorded by the same electrode, and the isolated units differed in shape and amplitude. We used both Autocut and Autosort programs. The latter allowed some additional testing using the principal component analysis. Spatial frequency tuning curves of individual neurons were classified as low-pass (LP), band-pass (BP) and high-pass (HP). The individual tuning functions were fitted by Log Gaussian functions Y = Baseline X/Center ∧ + Amplitude × exp(−0.5 × ln 2 Width where Baseline is the level of spontaneous activity, Amplitude is the height of tuning (maximum response), X is a spatial frequency, Center is the cell optimal SF and Width is the width of tuning at half height. The optimal SFs and bandwidths of individual cells were derived from the equation. To estimate the similarity of SF tuning curves, the Pearson correlation coefficient between two curves (at stimulus conditions) was computed. Linear regression analysis was performed to obtain the probability values. Each recorded neuron was classified as simple or complex cell on the basis of their response modulation (AC/DC ratio) [25]. Cells for which AC/DC ratio exceeded 1.0 were considered to be simple while cells with AC/DC ratio less than 1.0 were classified as complex. Spike sorting yielded 82 cells from 26 recorded neuronal pools. This comprised 35 simple and 47 complex cells. For every neuron, we measured the SF tuning curve with the average response magnitude. Except for SF, the other various gratings parameters were fixed. Hence, we determined the evoked firing rate in relationship to the SF. Fig. 1 illustrates tuning curves of two units recorded simultaneously from two recording sites. The spike waveforms are displayed in each case in panels C and G. The cluster analysis indicated a highly significant separation, Z score = 4.4 for cells A, B; and Z = 6.5 for cells E, F. For cells A and B (first site) the respective optimal SFs were 0.21 and 0.15 cyc/deg. More interestingly, these two cells of this pool showed quite different tuning curves. Indeed, cell A was a band-pass cell (bandwidth at half height 0.22 cyc/deg) while the companion neuron was a low-pass cell (bandwidth at half height 0.6 cyc/deg); its firing rate gradually diminished as the
245
stimulating SF increased. The second example from a different experiment illustrates two cells (E and F) that presented bandpass tuning curves. However, the respective optimal SFs were very different, 0.35 and 0.8 cyc/deg. That is a difference of 1.25 octaves. In addition, the bandwidths were also distinct. Cell F was very narrowly tuned around its preferred SF (bandwidth at half height = 0.12 cyc/deg), the bandwidth at half height of the nearby unit (E) was equal to 0.6 cyc/deg. It should be noted that three cells were sorted out at this site (all three clusters are shown in Fig. 1G). The third neuron was low-pass and had an optimal SF of 0.15 cyc/deg (SF tuning curve not shown). Thus neighboring cortical neurons have different SF properties even when the units are located nearby. Fig. 2A shows the preferred or optimal SF distribution. The distribution of optimal SFs spans a range from 0.13 to 2.0 cycles per degree with a mean value of 0.31 ± 0.03 (mean ± S.E.M.). This range is comparable to what has been reported originally for area 17 in cats [20], where no units were reported to respond to SF lower than 0.13 cyc/deg, nor higher than 2.0 cyc/deg. The distribution of Fig. 2A is biased towards low SFs perhaps reflecting the fact that some of our recording sites were located closely to the area 17/18 border and may include units located within area 18 where neurons prefer lower SFs [20]. No significant difference was observed between SF preferences of simple and complex cells (0.30 ± 0.06 and 0.32 ± 0.04, respectively) though complex cells seemed to prefer higher SFs than simple cells as suggested with previous studies [1]. Because no significant difference was found between simple and complex cells we will not describe further results according cell types. Panel (B) of Fig. 2 illustrates the distribution of curve types. The units exhibiting a band-pass property constitute about two thirds (58.6%) of the population while low-pass cells made up about 40% of the group. High-pass cells were very infrequent (one cell). In order to examine whether cells recorded by the same electrode tip belonged to one single functional group we correlated pair wise optimal SFs of all cells sorted out from the same site. The scatter plot shown in Fig. 2C shows that only about a half (43) of pairs were positioned on the equality line indicating the same optimal SF for both cells. Even in the range 0.2–0.8 cyc/deg a significant portion of neighboring cells had dissimilar SF preferences especially those of band-pass type (Fig. 2D). It is worth noting that only 7 pairs out of 43 have identical optimal SFs. Fig. 2E supplements the previous analysis; it shows the distribution of the correlation coefficient (r, Pearson coefficient). For this computation, we correlated the data points of the entire tuning curves between neighboring neurons recorded simultaneously at one site. High correlation values (∼0.7 and higher, see significant values on the graph) indicate that both cells presented very analogous curves; conversely, a weak or even negative correlation points toward a lack of similarity between both curves. This histogram shows that less than half of pairs of neighboring cells have comparable curves (44.9%, r ≥ 0.7). Half of cell pairs (49%) had coefficient values ranging from −0.4 to +0.4. The distribution of Fig. 2E was compared with a distribution of correlation coefficients for pairs of neurons belonging to different sites (non-neighboring
246
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
Fig. 2. Population analyses. (A) Distribution of optimal SFs in cycles per degree (cyc/deg). Optimal SFs are derived from curve fit and rounded. (B) SF tuning curve type distribution with the proportion of each curve type indicated above the bars, LP: low-pass, BP: band-pass, HP: high-pass. (C) The optimal SF of one cell is plotted against the optimal SF of the companion cell. Symbols (+) and (䊉) indicate non-significant (ns) and significant (sign., p < 0.05) correlation between the full tuning curves for each unit, respectively. (D) Only band-pass cells are plotted (43 out of 98 pairwise combinations). Note that since some cells have the same optimal SFs some points are superimposed. (E, F) Distributions of the Pearson correlation coefficient (r) between tuning curves of nearby neurons (E, N = 98) and non-neighboring cells (F, N = 126).
cells, shuffling by one experiment was performed) (Fig. 2F). Importantly, these two distributions were not significantly different (p = 0.29, one-tailed Mann–Whitney non-parametric test), thus indicating the high degree of similarity between SF preferences of pairs of nearby cells and non-neighboring neurons. The results further confirm that only about a half of nearby neurons show similar SF preferences and that there is an absence of a clear clustering of SFs in the superficial layers of the area 17–18 border region in cat visual cortex. In addition to demonstrating that nearby cortical cells, tested simultaneously, exhibit very different optimal SFs, the present study revealed several new findings. First, tuning curves of
neighbouring units display various profiles. Second, even nearby neurons whose tuning curves are of band-pass type exhibited different preferred SFs. Thus, results of the present investigation suggest that the majority of neighboring cells exhibit different optimal SFs and dissimilar tuning curves. Only about less than half (45%) of cells exhibited a high pair-wise correlation suggestive of a grouping of neurons with similar tuning curves, hence only these cells have a close optimal SF. Globally, however, our results seem to indicate an absence of a clear clustering or, at best, a very sparse clustering of SFs in the superficial layers of the area 17–18 border region in cat visual cortex. This conclusion appears to be in agreement with Issa et al. [17] maps which
S. Molotchnikoff et al. / Neuroscience Letters 418 (2007) 242–247
show that SF preferences at the extremes of the cat’s SF spectrum are separated by no more than 0.75 mm. This conclusion was suspected by Tolhurst and Thompson [26], although with a quite different approach; they reported optimal SFs can differ markedly from neuron to neuron. It should be noted however that since 75% of units in our sample had preferences between 0.15 and 0.3 cyc/deg, the clustering of the shuffled data was already quite tight. This could influence a comparison of SF tuning curves on a single electrode to tuning curves from separate recordings (shuffled control) showing no statistical difference in the distributions. Computations performed on extracellular recordings made in the hippocampus suggest that spike amplitude of neurons superior to 60 V may be recorded within a radius of 50 [6,13]. Though the spike amplitudes are different in the cortex and hippocampus some assumptions may be reasonable. The average spike amplitude of the present study was 30 ± 3 V. Because the amplitude of extracellular spikes decreases rapidly with distance, the distance between the cell body of the contributing to the multiunit activity neuron and the recording tip is at most 100–150 from the recorded loci. That is roughly 2–3 orientation columns, which spans about 20–30◦ of axis of orientation [2,15]. Although, we did not measure orientation tuning curves, the optimal orientation of the grating for multiunit activity was determined by finding which grating axis evoked the largest firing. Thus, it is most likely that sorted cells whose optimal orientations were too different from that of the multiunit responded too weakly to be analyzed since they were not optimally stimulated. For that reason, it is logical to assume that we studied cells sharing similar or close optimal orientation. Therefore, they belong to the same functional domain. Furthermore, it has been demonstrated that orientation preference is unrelated to the optimal SF [14,24,26] at one particular site. Such a lack of grouping along similar SF preferences allows very large combinations of orientations and SFs to be overlaid. Indeed the presence of cells with several optimal SFs, sometimes several octaves apart, within a relatively small pool of cells that share close orientation preferences means that a large SF spectrum is represented in a narrow band of orientations and vice versa. Such an organization may be the neural basis that underlies the successful capture of multiple levels of spatial detail at all orientations at each location in the visual field. Acknowledgements The research was supported by NSERC (Canada) grants to S. Molotchnikoff. We thank Dr. S. Itaya and unknown reviewers for helpful comments. References [1] C. Bardy, J.Y. Huang, C. Wang, T. FitzGibbon, B. Dreher, ‘Simplification’ of responses of complex cells in cat striate cortex: suppressive surrounds and ‘feedback’ inactivation, J. Physiol. (Lond.) 574 (2006) 731–750. [2] A. Basole, L.E. White, D. Fitzpatrick, Mapping multiple features in the population response of visual cortex, Nature 423 (2003) 986–990.
247
[3] T. Bonhoeffer, A. Grinvald, Optical imaging based on intrinsic signals, in: A.W. Toga, J.C. Mazziotta (Eds.), Brain Mapping: The Methods, Academic Press, New York, 1996, pp. 55–97. [4] T. Bonhoeffer, D.S. Kim, D. Malonek, D. Shoham, A. Grinvald, Optical imaging of the layout of functional domains in area 17 and across the area 17/18 border in cat visual cortex, Eur. J. Neurosci. 7 (1995) 1973–1988. [5] P.C. Bressloff, J.D. Cowan, A spherical model for orientation and spatialfrequency tuning in a cortical hypercolumn, Philos. Trans. R. Soc. Lond. B Biol. Sci. 358 (2003) 1643–1667. [6] G. Buzs`aki, Large-scale recording of neuronal ensembles, Nat. Neurosci. 7 (2004) 446–451. [7] G.C. DeAngelis, G.M. Ghose, I. Ohzawa, R.D. Freeman, Functional microorganization of primary visual cortex: receptive field analysis of nearby neurons, J. Neurosci. 19 (1999) 4046–4064. [8] R.L. De Valois, K.K. De Valois, Spatial Vision, Oxford University Press, New York, 1988, 381 pp. [9] R.M. Everson, A.K. Prashanth, M. Gabbay, B.W. Knight, L. Sirovich, E. Kaplan, Representation of spatial frequency and orientation in the visual cortex, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8334–8338. [10] M.S. Fee, P.P. Mitra, D. Kleinfeld, Automatic sorting of multiple unit neuronal signals in the presence of anisotropic and non-Gaussian variability, J. Neurosci. Methods 69 (1996) 175–199. [11] G.L. Gerstein, Cross-correlation measures of unresolved multi-neuron recordings, J. Neurosci. Methods 100 (2000) 41–51. [12] L. Hazan, M. Zugaro, G. Buzs´aki, Klusters, NeuroScope, ND Manager: A free software suite for neurophysiological data processing and visualization, J. Neurosci. Methods 155 (2006) 207–216. [13] D.A. Henze, Z. Borhegyi, J. Csicsvari, A. Mamiya, K.D. Harris, G. Buzs`aki, Intracellular features predicted by extracellular recordings in the hippocampus in vivo, J. Neurophysiol. 84 (2000) 390–400. [14] D.H. Hubel, T.N. Wiesel, Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex, J. Physiol. (Lond.) 160 (1962) 106–154. [15] D.H. Hubel, T.N. Wiesel, Uniformity of monkey striate cortex: a parallel relationship between field size, scatter, and magnification factor, J. Comp. Neurol. 158 (1974) 295–305. [16] M. H¨ubener, D. Shoham, A. Grinvald, T. Bonhoeffer, Spatial relationships among three columnar systems in cat area 17, J. Neurosci. 17 (1997) 9270–9284. [17] N.P. Issa, C. Trepel, M.P. Stryker, Spatial frequency maps in cat visual cortex, J. Neurosci. 20 (2000) 8504–8514. [18] S. LeVay, T.N. Nelson, Columnar organization of the visual cortex, in: A.G. Leventhal (Ed.), The Neural Basis of Visual Function, CRC, Boston, 1991, pp. 266–315. [19] L. Maffei, A. Fiorentini, Spatial frequency rows in the striate visual cortex, Vision Res. 17 (1977) 257–264. [20] J.A. Movshon, I.D. Thompson, D.J. Tolhurst, Spatial and temporal contrast sensitivity of neurones in areas 17 and 18 of the cat’s visual cortex, J. Physiol. (Lond.) 283 (1978) 101–120. [21] F. Reinoso-Suarez, Topographischer Hirnatlas der Katze, Herausgegeben von E. Merck A.G., Darmstadt (1961). [22] D. Shoham, M. Hubener, S. Schulze, A. Grinvald, T. Bonhoeffer, Spatiotemporal frequency domains and their relation to cytochrome oxidase staining in cat visual cortex, Nature 385 (1997) 529–533. [23] S. Shumikhina, J. Guay, F. Duret, S. Molotchnikoff, Contextual modulation of synchronization to random dots in the cat visual cortex, Exp. Brain Res. 158 (2004) 223–232. [24] L. Sirovich, R. Uglesich, The organization of orientation and spatial frequency in primary visual cortex, Proc. Natl. Acad. Sci. U.S.A. 101 (2004) 16941–16946. [25] B.C. Skottun, R.L. De Valois, D.H. Grosof, J.A. Movshon, D.G. Albrecht, A.B. Bonds, Classifying simple and complex cells on the basis of response modulation, Vision Res. 31 (1991) 1079–1086. [26] D.J. Tolhurst, I.D. Thompson, Organization of neurones preferring similar spatial frequencies in cat striate cortex, Exp. Brain Res. 48 (1982) 217–227. [27] M.A. Webster, R.L. De Valois, Relationship between spatial-frequency and orientation tuning of striate-cortex cells, J. Opt. Soc. Am. A 2 (1985) 1124–1132.