Spatial clustering of orientation preference in primary visual cortex of the large rodent agouti

Summary

All rodents investigated so far possess orientation-selective neurons in the primary visual cortex (V1) but – in contrast to carnivores and primates – no evidence of periodic maps with pinwheel-like structures. Theoretical studies debating whether phylogeny or universal principles determine development of pinwheels point to V1 size as a critical constraint. Thus, we set out to study maps of agouti, a big diurnal rodent with a V1 size comparable to cats'. In electrophysiology, we detected interspersed orientation and direction-selective neurons with a bias for horizontal contours, corroborated by homogeneous activation in optical imaging. Compatible with spatial clustering at short distance, nearby neurons tended to exhibit similar orientation preference. Our results argue against V1 size as a key parameter in determining the presence of periodic orientation maps. They are consistent with a phylogenetic influence on the map layout and development, potentially reflecting distinct retinal traits or interspecies differences in cortical circuitry.

Graphical abstract

Highlights

• •Agouti V1 neurons are among the highest orientation- and direction-selective neurons in rodents
• •They respond best to low spatial frequencies and with a bias for horizontal orientations
• •There is no evidence of systematic periodic maps of orientation columns for agouti
• •Neurons along the vertical cortical axis tend to have similar orientation preferences

Biological Sciences; Neuroscience; Developmental Neuroscience; Cellular Neuroscience; Sensory Neuroscience

Introduction

Many early sensory cortical areas are orderly arranged in functional structures, where neurons sharing similar response properties lie close to each other. A well-described example is the organization of orientation-selective neurons into domains or columns of iso-preference in the primary visual cortex. Like receptive field position, orientation preference is preserved along the axis perpendicular to the cortical surface but changes systematically around singularities in the axis parallel to it, forming pinwheel-like structures. Therefore, the same preferred orientation repeats in area-specific intervals forming regular orientation preference maps (OPMs) that have been found in V1 and V2 of all carnivores (e.g. Bonhoeffer and Grinvald, 1991; Hubel and Wiesel, 1962; 1963; Chapman et al., 1996), primates (e.g. Essen and Zeki, 1978; Hubel and Wiesel, 1968; Ts'o et al., 1990; Blasdel and Salama, 1986), and their close relatives (see Kaschube et al., 2010) studied so far. Although all investigated rodent species presented orientation-selective neurons, no such maps have been observed in this order (rats: Ohki et al., 2005, gray squirrel: Van Hooser et al., 2005, mice: Bonin et al., 2011). This lack of classical columns gave rise to the terms “interspersed” or “salt-and-pepper” organization implying a random spatial arrangement of orientation-selective neurons in these animals. Noteworthy, reports in lab mice (Ringach et al., 2016; Kondo et al., 2016) and grasshopper mouse (Scholl et al., 2016) indicate that the arrangement is not entirely random. Instead, these authors describe functional modules of roughly 30 μm diameter which run perpendicular to the cortical surface and contain neurons tuned to similar orientations, which they thus termed “mini-columns”. Along this line, several studies, mostly theoretical, have proposed that the formation of OPMs might not be solely dependent on phylogeny but might be due to self-organization (Kaschube et al., 2010), constrained by factors such as brain size (Keil et al., 2012; Meng et al., 2012), number, and density of neurons (Weigand et al., 2017). The main argument behind these ideas is that, within bigger and more neuron dense cortical areas, an interspersed organization imposes higher wiring costs than periodic orientation columns. Since neurons that respond to similar preferences tend to interconnect (i.e. Löwel and Singer, 1992; see also Hebb, 1949), shorter axons with closer pathways and faster signal transmission between those neurons might be achieved through a more periodic spatial arrangement of orientation selectivity when assuming a dynamical map formation (Wolf and Geisel, 1998; Koulakov and Chklovskii, 2001). According to this argument, for bigger brains, it would be more efficient to cluster their neurons by functional similarity in modules as observed in OPMs (for review see Chklovskii and Koulakov, 2004).

Not surprisingly, given their availability, all rodents studied so far have been of smaller brain size and of nocturnal (mice, rats) or crepuscular habits (gray squirrels). The theoretical considerations above predict that rodents with a larger V1 size such as agouti and capybara might present OPMs (Weigand et al., 2017). Thus, we set out to study the functional responses and anatomical layout of orientation-selective neurons in the red-rumped agouti Dasyprocta leporina. Agoutis are diurnal rodents native to South and Central America with body and brain size comparable to that of cats. Their V1 surface extends for over a centimeter lateral from the lateral sulcus and even longer in the antero-posterior axis and spans around 320–340 mm² (Dias et al., 2014; their Figure 2; personal communication M. Garcia). Although the agouti's retinotopic layout has been studied before (Picanço-Diniz et al., 1991), no orientation selective visual responses of this species were ever recorded. In the present study, we obtained maps of retinotopy and orientation-selective responses from intrinsic signal imaging. Subsequently, we quantified receptive field size in detail and neuronal response tuning to several stimulation parameters such as contour orientation, direction of movement, and spatial and temporal frequency (TF) from in vivo parallel electrophysiological recordings in the anesthetized agoutis' visual cortex. We used vertical and horizontal grids of multi-electrodes to determine the topographic layout of orientation selectivity. The same experiments were performed in anesthetized cats (Brodmann areas 17 and 18) in order to directly compare agouti neuron selectivity and layout with the well-established cat model.

Figure 2.

Orientation selectivity in primary visual cortex of agouti and cat

(A and B) Examples of isolated units for agouti (A) and for cat area 18 (B) stimulated at 0.08 cpd. Dashed lines indicate double Gaussian fits to the mean firing rates. Light gray lines indicate pre-stimulus firing rate. Error bars, standard deviation. Note that here agouti and cat units prefer the horizontal direction of movement and have thus vertical orientation preference.

(C) Distributions of orientation selectivity indices (OSIs) in both species. Agouti selectivity indices (n = 349) are significantly lower than indices for cat area 18 (n = 97, Mann-Whitney U, p < 0.0001) and cat area 17 (n = 22, Mann-Whitney U, p < 0.0001).

(D) Distributions of direction selectivity indices (DSIs) in both species. For (C and D), cat distributions of OSI and DSI are divided into Brodmann areas 17 (gray) and 18 (black). Only indices of neurons that passed both the spike-dependent threshold and a static threshold of 0.1 are depicted. Error bars are SEM (standard error of the mean).

Results

Guided by Dias et al. (2014), we identified the anatomical location of V1 in agoutis and performed extracellular electrophysiological recordings from single units and optical imaging of intrinsic signals. V1 identity was histologically confirmed postmortem. In order to compare our results in the novel rodent species to a well-established non-rodent model with similar V1 size, we also performed the same experiments in cats.

Receptive field size

Agoutis possess a visual streak (Figure 1A, Picanço-Diniz et al., 1991) and laterally positioned eyes. Therefore, we mapped and evaluated the area (see Transparent Methods) of the aggregate classical receptive fields (aCRF) based on multi-unit activity as a function of eccentricity by stimulating monocularly in the range between 5 and 118°. In cats, we stimulated binocularly after aligning the eyes using an optic prism. Although we did not record central binocular responses in the agouti, we still obtained a considerable overlap in the eccentricities of the aCRF recorded in both species. Agoutis can have small receptive fields at a wide range of eccentricities (Figure 1B).

Figure 1.

Receptive field size in agoutis and cats

(A) Sketch of the electrophysiologically sampled area in agouti V1.

(B) aCRF mapping examples from agouti V1 at three different eccentricities (E) along the visual streak (elevation less than 5°) and two fields from cat area 18. Scale bars represent 2°.

(C) Receptive field size as a function of elevation. Data from 5 cats and 11 agoutis: 125 multi-units for cat A18, 82 for cat A17, and 401 for agouti V1.

(D) Receptive field size as a function of eccentricity. More lateral part of the visual streak in agouti (light violet shade in A, 30–120 deg). Agouti aCRFs do not increase at large eccentricities (≥90 deg).

(E) Receptive field size as a function of eccentricity. Zoom into the central visual field <25 deg (dark violet shade in A).

(F) Cumulative distributions of aCRF size for agouti V1, cat area 17, and 18. Note that agouti V1 neurons tend to have aCRFs of similar size as cat area 17.

SL, lateral sulcus; HM, horizontal meridian; VM, vertical meridian. Cortical coordinates: M, medial; L, lateral; A anterior; P, posterior; adapted from Picanço-Diniz et al. (1991).

All the units recorded had aCRFs close to the horizontal meridian of the visual field (±15° in elevation) (Figures 1C–1E). We quantified the area of multi-units with well-defined aCRFs (n = 401 in agouti, n = 125 in cat A18, n = 82 in cat A17). Our measurements for agouti aCRF size are in the 1–22 deg² range. Cat A18 aCRF sizes are in the 1–32 deg² range, and A17 aCRF sizes are in the 0.5–22 deg² range. Agouti aCRF sizes are much more similar to cat A17 than cat A18 sizes (mean and SD values: cat A17 = 7.2 ± 4.3; cat A18 = 11.5 ± 6.5; agouti V1 = 6.2 ± 3.7; Figure 1F).

Cat area 18 exhibits bigger CRFs (2–32 deg² range reported in (Hubel and Wiesel, 1965)) and selectivity to lower spatial frequencies (SFs) than area 17 (Movshon et al., 1978). Characteristically, for cat area 18, it can be observed that the cat's aCRF size increases rapidly with elevation and eccentricity (Figures 1C and 1E) while the agoutis' aCRFs remain small at large eccentricities of up to 120° (Figure 1F).

Orientation and spatial frequency selectivity

Previous studies (Picanço-Diniz et al., 1991; Dias et al., 2014) recorded retinotopic visual responses in agouti V1, but orientation selectivity in this species has not been reported yet. Thus, we used drifting sine wave gratings to characterize functional visual responses. The gratings consisted of twelve directions of movement, three TFs (2, 4, and 8 Hz), and seven SFs (0.04, 0.08, 0.16, 0.32, 0.64, 1.28, and 2.56 cpd). Single units that passed a sign test with a 95% confidence interval (p < 0.05) between their pre-stimulus and evoked firing rates were considered visually driven and included in the analysis. Agouti neurons fired more vigorously at 0.08 and 0.16 cpd (see Figure 3A) and ceased to respond reliably to higher SFs than 0.64 cpd at all TFs tested. They further responded much less selectively to 8Hz < 4 Hz < 2 Hz stimulation. For stimulation with 8 Hz, we encountered almost no visually driven cells and with 4 Hz much less than with 2 Hz, which met our selectivity criterion (see below). In addition, the few selective responses obtained at 4 or 8 Hz did not behave qualitatively different than at 2 Hz for different SFs. Thus, in the following comparative and spatial analysis, we focused on data obtained at 2 Hz and at SFs <0.64 cpd.

Figure 3.

Agouti V1 neurons are selective for spatial frequency

(A) Population mean firing rate of all orientation-selective agouti single units evoked at different spatial frequencies (black line) and during the pre-stimulus period (gray line).

(B) Counts of neurons that evoked maximum mean firing rate at that SF.

(C) Spatial frequency tuning curve example. Mean maximum firing rate obtained with an optimal grating at each SF. Error bars, standard deviation.

(D) Polar plot of the example agouti single unit (upper) and of a cat area 17 single unit (lower) at five different SFs.

(E) Orientation selectivity index of the two neurons of D at different SFs. Note that the agouti neuron fires most at 0.08 and exhibits highest orientation selectivity at 0.16 cpd. For comparison, the cat area 17 neuron remains selective at high SFs.

(F) Spatial frequency selectivity of the mean orientation selectivity indices for agouti, cat A17, and cat A18 in the same scale.

(G) Same as (F) zooming in on the agouti curve. Note that agoutis show small orientation selectivity indices, with optimal OSI tuning at 0.32 cpd in the visual area investigated. Only indices of neurons that passed both the OSI spike-dependent threshold and a static threshold of 0.1 were included (n = 349). Error bars are SEM (standard error of the mean), except for (C)

Some examples of response profiles of single units to several orientations and directions are depicted in Figures 2A and 2B (for peri-stimulus time histograms (PSTHs) see Figure S3). We computed, based on each profile, an orientation selectivity index and a direction selectivity index (OSI and DSI, respectively,(Wunderle et al., 2013) (Conde-Ocazionez et al., 2018), and see Methods), where an index of 1 indicates a neuron that responds to only one stimulus orientation or direction, and an index of 0 indicates equal responsiveness across all orientations or directions.

For the quantitative comparison of cell populations between both species, the best OSI or DSI obtained per single unit for a given SF was considered (Figures 2C and 2D). This was done in order to count each single unit's OSI and DSI once and at its optimal stimulation. In addition, only single units fulfilling the criteria of (i) a positive sign test, (ii) a significant OSI, i.e., above the spike-dependent OSI threshold (Figure S2), (iii) with an OSI greater than a static threshold of minimum 0.1 were considered (see Wunderle et al., 2013; Peiker et al., 2013 for similar selection criteria).

Overall, agouti selectivity indices in this sample are significantly lower than those of cats with a median OSI for agouti V1 at 0.2 (n = 349), for cat A18 at 0.43 (n = 97), and for cat A17 at 0.43 (n = 22) (Mann-Whitney U, p < 0.0001 for both comparisons between agouti V1 and cat A18 or cat A17).

Though the difference between the species is smaller for the DSI than the OSI distributions, agouti neurons are notably less direction selective than those of cat area 18 (Figure 2D). For DSIs selected according to the same criteria (crossing the spike-dependent DSI threshold and a static DSI threshold of minimum 0.1), the median DSI for agouti V1 in the sample is 0.17 (n = 261), for cat A18 is 0.25 (n = 91), and for cat area 17 is 0.21 (n = 18) (Mann-Whitney U, p = 0.04 for the comparison between agouti V1 and cat A18).

Although the spike-dependent threshold corrects for high indices, which are rate related, we also checked whether agouti indices might be lower because of lower absolute firing rates and thus lower signal-to-noise ratio caused by anesthesia. To this end, we compared median OSI indices between rate-matched samples of the two species (OSI >0.1, 10 sps < evoked rate <20 sps). The result supports the conclusion that the difference in selectivity between the two species is not related to firing rate (median OSI agouti V1, 0.16, n = 91, versus median OSI cat area 18, 0.36, n = 21; Mann-Whitney U p < 0.0001).

In addition to the OSI, we further calculated the classical orthogonal modulation depth index (OMDI) of orientation selectivity (see Methods, e.g. Niell and Stryker, 2008; Mazurek et al., 2014). Accordingly, we calculate 0.3 for agouti V1 neurons, 0.63 for cat A18, and 0.49 for cat A17 (see Bachatene et al., 2016, for similar values). Having excluded extreme indices of 1 because of 0 firing in the opposite directions, median OMDIs are still higher than the spike-rate-dependent OSIs but highly significantly correlated with them (Wilcoxon signed rank, p < 0.0001). So, both species differences and also the dependence of orientation selectivity on SF (data not shown) remained similar. In addition, the angle of the grating direction, which evoked the optimal response, differed only 11.4 ± 9.4° from the preferred angle obtained by vector summation.

Although the population of agouti neurons responded over a broad range of SFs, they fired more vigorously at lower SFs (Figure 3A) with most of the neurons exhibiting highest firing rates at 0.08 cpd (Figures 3B and 3C). Orientation-selective firing could be observed up to 0.64 cpd (Figure 3D). Since SF and orientation tuning have been observed not to be independent response characteristics in cats (e.g. Jones et al., 1987), we investigated whether the degree of orientation selectivity of agouti neuron responses varied as a function of SF.

Although OSI values of agouti single units crossing the threshold are much lower and less modulated with SF than those of cat neurons (Figures 3F and 3G), the mean OSI of the those responses presents a SF tuning profile with a peak for 0.32 cpd (Figure 3G). The cat SF profiles are congruent with the cat literature (i.e. Ribot et al., 2013). For example, as expected, 0.64 cpd stimulation still produced highly selective responses for cat area 17 (Figure E, F) but not for cat area 18 (Figure 3F).

In addition to the OSI and DSI values, we computed the half-width at half-height (HWHH) of the orientation and direction tuning curves of agouti neurons.

This parameter quantifies the sharpness of the orientation or direction tuning curves, as third selectivity indicator, by fitting the normalized firing rate responses to Gaussian curves (Figure 4A, also see Methods). In this analysis, we only considered neurons with an orientation (for orientation HWHH) or direction (for direction HWHH) selectivity index higher than 0.2.

Figure 4.

Sharpness of orientation and direction tuning curves in agoutis and cats

(A–C) (A) Examples of Gaussian fits of orientation tuning curves elicited at 0.32 cpd for one agouti and one cat single unit. Mean population half-width at half-height (HWHH) of orientation (B) and direction (C) selectivity tuning curves. Firing rates for each stimulation orientation/direction were normalized for every neuron with a selectivity index >0.2. A single peak Gaussian function was fitted to each of the normalized firing rates for orientation/direction, for each spatial frequency. Note that tuning curves of cat area 18 neurons have much lower HWHH values than agouti, especially for direction selectivity. Data from 2 cats and 9 agoutis. Error bars: SEM. Stars indicate significant differences between agoutis and cats (Wilcoxon rank-sum test; p < 0.001; n = agouti units, cat units)

In comparison to cats, we found for orientation (Figure 4B) and direction tuning curves (Figure 4C), larger HWHH values for agouti units than for cat units, reflecting broader tuning curves (as depicted in the examples in Figure 4A).

The discrepancy between HWHH values of the two species is particularly high for direction selectivity. As expected for area 18 (Orban et al., 1981), these neurons are well tuned for direction of motion, and thus, the direction tuning curves are very narrow.

We further observed that agouti orientation HWHH is in accordance with what was reported for the gray squirrel, the only other large visual rodent studied so far (<70 deg, Van Hooser et al., 2005). Noteworthy, in the cited study, a more rigid criterion (OSI >0.5) was applied to sample the orientation-selective units from which the HWHH was calculated. With such a high threshold, Van Hooser et al. (2005), report a median HWHH of 28 deg. Therefore, we repeated the HWHH in our data for cells with an OSI >0.5 and obtained an agouti median HWHH for orientation tuning lower than 23 deg for all of the five SFs studied (similar results for cats, data not shown). Although not conclusive, agouti neurons seem to have a sharper orientation tuning than the gray squirrel neurons.

Spatial layout of orientation selective neurons

In order to analyze the spatial layout of neurons with similar orientation preference, we analyzed their differences in orientation preference as a function of cortical distance. Preferred angle differences may range from 0 deg to 90 deg.

We compared this measure for the same type of recordings in agoutis and cats. For a salt-and-pepper-like layout, the angle difference between pairs of neurons should be independent of the cortical distance. In contrast, for periodically repeating orientation columns that compose orientation preference maps as in cats, the angle difference between pairs of neurons should increase smoothly with distance in all directions parallel to the cortical surface until targeting the proximal domain of orthogonal orientation preference. Similarly, this difference should be minimal when comparing adjacent cells, as within a cortical column or at the same electrode.

It should be noted that all cortical distances shown in Figure 5 are distances between the electrodes at which the two neurons being compared were recorded. Therefore, the distances declared here are not exact neuronal distances but a close estimate. Whenever a comparison is declared to be of 0 μm distance, it means that the two neurons were recorded from the same electrode. In Figures 2A and 2B, we illustrate that clearly different waveforms from the same multi-unit with similar orientation tuning could be isolated in both species.

Figure 5.

Spatial layout: tuning difference between cell pairs as a function of distance

Mean pairwise angle difference (A) and pairwise Pearson correlation of firing rates to stimulation angle (B) as a function of horizontal cortical distance for cat A18 and agouti V1. For each cell, only the orientation preference at the spatial frequency, which elicited the highest OSI, was considered. Gray dotted lines represent the calculation for the shuffled data. Asterisks depict significant differences between the correlations obtained for recorded and shuffled data. Significance criterion: Mann-Whitney p < 0.001 (Bonferroni correction for 10 multiple comparisons) for the pairwise comparison of the data points and the shuffled data. Data from 2 cats and 5 agoutis. For raw p values, see Tables S1 and S2, for number of data points, see Table S3. Note that agouti neuron similarity is only different than chance for neurons that lie close to each other. Error bars: SEM.

Figure 5A displays the angle differences as a function of horizontal cortical distance for pairwise cell comparisons, considering the orientation preference obtained with the SF that elicited the highest OSI. For comparisons of distances equal or greater than 250 μm, agouti orientation preference differences do not deviate from the estimate for a random spatial layout (dotted gray line, shuffled data). In contrast, the cat's angle difference profile shows the expected smooth increase of angle differences across cortical distance with a characteristic valley around 900 μm roughly corresponding to the distance between iso-orientation columns. This indicates a structure in cats' OPMs which agoutis lack. However, the result that agouti angle differences deviate significantly from the shuffled data at 0 μm is compatible with the interpretation that neurons recorded from the same electrode do have similar orientation preference, implying a local structure (see also Figure 6A).

Figure 6.

Orientation preference distribution across vertical and horizontal dimensions

(A) Polar plots of example single units recorded from a vertical double shank probe at 0.08 cpd. Waveforms from separated units are color coded. Note that orientation preference in agoutis is relatively stable along the vertical axis. Representatively, units on both shanks prefer orientations varying around the horizontal axis.

(B) Angle differences are higher across (gray filled circle, n = 121) than along shanks (empty circle, n = 145; Mann-Whitney-U, p < 0.005) and between units separated from the same site (black filled circle, n = 24; Mann-Whitney-U, p < 0.0001).

Pairwise differences (C) Pairwise comparisons of orientation preference across all electrodes of each device. Median and interquartile ranges are depicted by circles and lines, respectively. Empty and filled circles depict pairwise comparisons between single units from Neuronexus probes or electrode arrays, respectively. Probe electrode sample from a vertical cylinder (A, orthogonal to cortical surface). Array sample from horizontal planes (parallel to cortical surface). Angle difference (top) and tuning similarity (bottom). Both measures indicate a greater similarity between neurons arranged vertically. ∗ depicts p < 0.0001 (Mann-Whitney U test, for raw p values see Tables S4 and S5). Probe electrode data from 3 agoutis. Array electrode data from 5 agoutis.

(D) Percentage of agouti single units with different orientation preferences above selectivity threshold categorized in 12 groups of ±7.5 deg at 0.16 cpd and 2 Hz (n = 85 neurons). Strikingly, in the overall sample, the horizontal orientation preference predominates.

In Figures 3 and 4, we had observed that orientation selectivity indices of agoutis are much lower and tuning curves are more flat than those of cats. This opens the question whether the calculated preferred orientation is actually able to accurately reflect the overall functional layout in those animals.

Therefore, we aimed for a measure, which takes not only the difference between “best angles” into consideration but also the shape of each cell's tuning curve. To accomplish this aim, we calculated the orientation tuning similarity, which is the Pearson correlation between the orientation tuning curves (i.e. the firing rates to each stimulus orientation; see Methods) of the two neurons to be compared (Figure 5B, see also plots separated by SF in Figure S4)

For both cat and agouti, the Pearson correlation reaches its positive peak at the smallest comparable distance. For agoutis, the maximum correlation obtained is only 0.5, drops to the shuffled data mean already at the next available distance point, and stays constant across the remaining distances. For cats, the correlation starts at much higher values (0.7), reaches negative correlation values, and also exhibits the characteristic reversion around 900 μm, already discussed above for the “angle difference” measure. Here, similar to both measures of angle difference and tuning similarity, we observe a maximum negative difference in neuronal similarity at around 700 μm. This is to be expected as the distance between cross-orientation columns in cat area 17 is approximately 500 μm and a little bit larger in area 18 (e.g. Löwel et al., 1987).

From these plots, we draw the conclusion that agouti neurons situated very close to each other (i.e. were recorded within the same electrode) have a higher feature similarity than expected by chance, but this does not hold for neurons recorded at different recording sites. Noteworthy, the agouti's tuning correlation also never reaches negative values, suggesting that clearly segregated domains of orthogonal orientation preference at such regular intervals as known for cat orientation preference maps, for example, are rather unlikely.

Clustering of iso-orientation cells in agouti V1 along the vertical axis and orientation bias

Inspired by the above results and the “mini-column” findings in the rodent visual cortex in the recent years (Ringach et al., 2016; Kondo et al., 2016; Maruoka et al., 2017), we investigated if the orientation preference similarity of nearby single units holds along the axis perpendicular to the cortical surface. To this end, we recorded with vertical probe electrodes with 16 recording sites of 100-μm inter-site distance inserted perpendicularly to the cortical surface (Figure 6A). We observed that the orientation preference of single units along the shank tended to be more similar than between two shanks spaced 500 μm apart (Mann-Whitney-U test, along versus across shanks, n = 222, p < 0.0001 Figure 6B).

Thus, we also statistically compared angle differences and tuning similarity of all single unit pairs either along the probe shank dimensions (Figure 6C, black empty circles, perpendicular to the cortical surface) or across the array dimension (gray full circles, parallel to the cortical surface). For both types of electrodes and each SF analyzed, the angle difference (Figure 6C, top) and tuning similarity (Figure 6B, bottom) were calculated between all recording sites of a probe or an array where an orientation selective unit response was recorded.

For the majority of tested SFs, angle differences were significantly smaller along the vertical than the horizontal dimension indicating greater similarity of orientation preferences of different recording sites in depth. In accordance, the similarity index was also significantly greater for neurons along the vertical dimension. It seems that neurons of similar orientation preference are grouped together on a short-range scale in a columnar-like manner but that orientation preference is not organized periodically across the horizontal plane.

In further support of this observation, the same result is obtained when analyzing the spatial layout among multi units (Figure S5).

Strikingly, in the example of Figure 6A, close to horizontal orientation preferences (slightly left oblique for shank 1 and right oblique for shank 2), seem to dominate for units both along and across the two shanks. Although units responding to vertical contours also occurred (see Figure 2A), the example recording matches an overall bias in the sample of agouti units preferring stimulation with horizontal contours (Figure 6D). When separating units according to their orientation preference at 0.16 cpd in 12 categories of equal size (±7.5 deg), the majority of units express horizontal preference.

Variation of preferred orientation across spatial frequency

In the previous section, we established the similarity of orientation preferences of neurons (inter-neuronal comparison) by comparing their preferred angle at the SF, which elicited the highest OSI. Next, we investigated the stability of the preferred angle of selective neurons across SFs of stimulation (intra-neuronal comparison). It turned out that agouti – in contrast to cat – orientation preference varies considerably across SFs (Figure 7A). Thus, we analyzed the stability of angle preference for every neuron that showed an orientation-selective response above the threshold to at least two SFs. We used two metrics of preferred orientation dispersion, namely, the angle range (Figure 7B, top), which is the largest difference in the preferred angles of all orientation-selective responses for a given neuron, and the circular variance of the preferred orientations (see Methods) of all orientation-selective responses of that neuron (Figure 7B, bottom). Agouti orientation-selective V1 neurons exhibit a larger variability in their preferred angles across different SFs than neurons in cat areas 17 and 18.

Figure 7.

Neuronal (in)stability of preferred angles across different spatial frequencies

(A) Orientation preference and selectivity index at different SFs for three example units from agouti (red, blue, gray squares) and one from cat area 17 (green circles). Note that agouti orientation preference varies much more than the cat's.

(B) Overall cumulative distribution of angle range (top) and circular variance (bottom) for all cells.

(C) Mean of the pairwise within-cell comparisons. For each pair of orientation selective responses (to different spatial frequencies) crossing the selectivity threshold, the difference of preferred angle was computed. The octaves denote the difference in spatial frequencies of the responses being compared (e.g. comparisons between 0.04-0.08 cpd and 0.32–0.64 cpd are both one octave apart).

(D) Zoom into the agouti curve shown in (C). Note that orientation difference increases with SF difference. Error bars: SEM.

We then analyzed whether a systematic dependence on SF was present within the observed variability. For this, we computed intra-neuron pairwise angle differences between the orientation-selective responses and classified them according to the distance in SF octaves (Figures 7C and 7D). Here, angle differences from all single units entered which met the selection criteria for at least two SFs of the groups 0.04, 0.08, 0.16, 0.32, and 0.64 cpd (312 agouti, 22 cat A17, and 83 cat A18 single units).

The data for agouti show a displacement toward larger values than cat, as expected from the higher variability discussed above, and also an increase in angle difference as the difference in octaves increases. Cat data, on the other hand, show lower angle difference values than those of agouti for every octave comparison.

Only area 18 (and not 17) appears to show a systematic variation of angle difference with octave difference, although orientation-selective responses of A18 to 0.64 cpd are, as expected from the area 18 SF cutoff, scarce and therefore that data point might not be representative. Pattadkal et al. (2018) have shown that mouse V1 neurons present a shift in the preferred angle with SF. We also found that in our agouti data but not for cat area 17.

Orientation preference maps in agouti

In order to confirm the lack of large-scale periodic orientation columns, we obtained maps of intrinsic signals in three agoutis. Intrinsic signal imaging is known to have a spatial resolution of at least 100 μm (Grinvald et al., 1986). We imaged the field of view exposed lateral to the lateral sulcus during monocular stimulation using two monitor positions of different eccentricity (0–40 deg and 40–80 deg). The checker bar protocol gave rise to specific retinotopic activations in the intrinsic signal maps predicted from Picanço-Diniz et al. (1991, their Figure 1B). According to the movement of the horizontal checker bar from upper to lower visual field, the activation shifted gradually from lateral posterior to medial anterior (Figure 8A). When moving the vertical checker bar in the visual field from medial to lateral, the activation in the single condition maps moved gradually from lateral anterior to medial posterior (Figure 8B). As expected for the enlarged representation of the visual streak, the maps evoked by the central position of the bar (+10 deg until −5 deg) covered together the largest part of the region of interest (yellow-orange-red in the color map). Similar maps were obtained from the other two animals. The landmarks obtained are in concordance with the descriptions of retinotopic maps (Picanço-Diniz et al., 1991; Dias et al., 2014).

Figure 8.

Intrinsic signal maps of retinotopy

Left, upper image: color overlay on the vessel image of position specific activation with bars of 5 deg width spanning the whole monitor in horizontal (A, 6 positions) or vertical (B, 8 positions, indicated by checker sketches on top) orientation. Lower image: sketch depicting iso-elevation (A) or iso-azimuth lines. Right: single condition maps.

(A) The activation starts with 15 deg in the upper visual field activating the lateral-posterior part of the craniotomy migrating to the anterior part. The visual streak extends from red to yellow.

(B) A vertical bar positioned at the visual field's midline evokes activity at anterior-lateral part moving to medial. HM, horizontal meridian; VM, vertical meridian. Cortical coordinates: M, medial; L, lateral; A, anterior; P, posterior.

Once having successfully obtained intrinsic signals from checker bars, we continued stimulation with gratings of four different orientations (eight directions of motion) and a SF adequate for agouti (0.16 cpd at 2 Hz) on the ideal monitor position.

To avoid any artificial periodicity in the single condition maps by filtering, image subtraction, or cocktail blank normalization, we used only first-frame correction as a preprocessing step. This evoked rather homogeneous activations in the region of interest (ROI) visually stimulated (stimROI). In addition, not all stimuli were equally efficient in evoking an intrinsic signal. Interestingly, the best activation seemed to be obtained with horizontal gratings. Accordingly, yellow was the dominating color in the compound angle map (Figures 9A and 9B). This finding repeated in all three animals and in both regions of interest along the visual streak representation we obtained per hemisphere/animal (Figure 9C, n = 6). For quantification of the coverage of preferences, orientation preference per pixel after vector summation was categorized in four groups of equal size because four different orientations were used for visual stimulation. Confirming the orientation bias observed in the electrophysiological data of single units (Figure 6C), it turned out that indeed the horizontal and/or the right oblique orientation preference dominate in the vector map (Figure 9C).

Figure 9.

Orientation maps obtained by grating stimulation in agouti primary visual cortex

(A) Unfiltered averaged single condition maps after first frame correction.

(B) Left, polar (saturation of the color codes for vector strength) map according to the color bar below; blue outline, stimROI; red outline, shiftROI. right: vessel image of the recorded area. Scale bar, 1 mm; cortical coordinates as in Figure 8.

(C) Relative count of pixels preferring one of the four angle categories as indicated below in two ROIs, a medial and a more temporal one. Note that preferences for either horizontal or right oblique (recording the left hemisphere) dominate in all animals and ROIs.

(D) Mean spatial correlation coefficients between all frames within the stimulated ROI (blue), an ROI shifted in an area not stimulated by that monitor position (red) and on a rubber brain illuminated with red light (green) for C50. Error bars: SEM.

Since the maps did not present any obvious periodicity or complementarity as known from maps in carnivores and primates, we tested for response reliability to the same stimulus orientation at different moments of the recording session. To this end, we spatially correlated the raw activation values of the first frame corrected pixels in the visually stimulated region of interest (stimROI) of the summed frames per recording block and condition with each other. Correlation values obtained with the same grating at different moments of time were significantly higher than with frames obtained at orthogonal orientations. This was not the case for pixels in a region of interest outside the visually stimulated area (shiftROI) close by (Figure 9D, Mann-Whitney-U test, difference between group 0 vs 90: n = 2182, stimROI, p = 0.0006; shiftROI, p = 0.6). The same result was obtained for the two other animals C36 and C38 (Figure S6).

Since the intrinsic signal is very small, it might reflect overall differences in illumination of the high-contrast whole-field grating on the monitor being reflected into the recording chamber, especially when there is no clear visually evoked signal amplitude. Therefore, we compared correlation values to those obtained from a control recording on a homogeneous rubber surface being exposed to the same illumination and stimulation conditions. As expected for a homogeneous surface, correlation values were higher than in the “real” experiment but homogeneously distributed over all angle differences, thus excluding any systematic influence of that or any other systematic variable (Mann-Whitney-U test, difference between group 0 vs 90: lumROI, n = 2210, p = 0.58).

Discussion

We characterized orientation- and direction-selective neurons and their spatial layout in the primary visual cortex of agoutis, while recording from multi-electrodes in horizontal or vertical arrangements and obtaining intrinsic signals during stimulation with sinusoidal gratings. In order to relativize our results to a mammal with high orientation selectivity, similar V1 size, and periodic OPMs, we performed the same experiments in cats.

Agouti aCRFs were of small to moderate size throughout the horizontal streak representation. Although neurons exhibited clear orientation and direction preference, selectivity was only half as high as for cats. Neurons responded best at lower SFs (0.08–0.32 cpd), up to 0.64 cpd, with a bias for horizontal contour preference.

Although preferred orientation seems to be similar along the axis perpendicular to the cortical surface, no systematic periodicity was found parallel to it – in contrast to cat area 17 and 18. Optical imaging of intrinsic signals supported these electrophysiological findings, as we observed homogeneous V1 activation by stimulation with oriented gratings.

Receptive fields and spatial acuity

According to our results, agouti aCRFs (1–22 deg²) can be as small as cat area 17 and 18 fields in the central visual field representation (our own comparison data, and Hubel and Wiesel, 1962). As expected for a rodent, their size does not increase much with eccentricity in the vicinity of the horizontal meridian (±15° elevation), which includes the representation of the agouti's horizontal streak. Previous studies on agouti (Picanço-Diniz et al., 1991) report slightly larger RF sizes of 8–33 deg² close to the midline until up to 71 deg² in the temporal periphery. The difference might be partially due to our focus on the region close to the horizontal meridian, thus undersampling high/low elevations and far periphery. Further, the automatic algorithm used here delineates aCRFs from the top 30% of the PSTH responses, in contrast to hand mapping taking into account the last spike detected.

Agouti aCRFs are also smaller than those of rats (7–130 deg², Girman et al., 1999), much smaller than those of mice (50–700 deg², Niell and Stryker, 2008), and appear in the low range of the rabbit's receptive field size (1–60 deg², Murphy and Berman, 1979), which – as agouti – is more diurnal than rats and mice (Jilge, 1991).

Thus, our data suggest that agoutis exhibit high visual acuity along their visual streak. Behavioral and visual evoked potential reports indicate an SF cutoff at about 0.6 cpd for mice (Porciatti et al., 1999; Prusky and Douglas, 2004), which is very similar to the cutoff for single unit spiking responses in agoutis. However, multi-unit activity in mice indicates much lower optimal SFs (0.02 cpd, Niell and Stryker, 2008), whereas agouti V1 neurons responded most selectively to 0.32 cpd, about half the cat's optimal SF in V1 (our data; Movshon et al., 1978), and similar to ferrets (Baker et al., 1998).

Orientation and direction selectivity

With the vector summation method, we find highly selective cells in both agouti and cats, but our median values are lower than those reported previously for mice (Niell and Stryker, 2008; Tan et al., 2011) and cats (Gardner et al., 1999; Carandini and Ferster, 2000; Scholl et al., 2013, 2016). When calculating the orthogonal modulation depth index (OMDI), the overall indices rise to the values reported in the previous literature, and, importantly, are highly correlated to the values we used for our further analysis. Moreover, having performed the same experiments in both species allows us to compare agouti and cat selectivity internally, independently of absolute selectivity values. Noteworthy, agoutis tend to have lower firing rates, which might account for reduced OSI values. Thus, we implemented a rate-dependent selectivity index and also compared OSI distribution for rate-matched samples. Still, the remaining data revealed that orientation selectivity in agouti is lower than that in the cat primary visual cortex. Lower selectivity could thus be a consequence of different circuits generating orientation and direction selectivity in the different orders.

Tan et al. (2011) as well as Bopp et al. (2014) found less evidence for lateral inhibition in mouse V1, as opposed to cats. In cortical layouts with modular periodic maps such as primates and carnivores, lateral inhibition might be a crucial mechanism contributing to their high orientation selectivity.

Animals with modular pinwheel-like structures exhibit considerable selectivity in their excitatory long-range connectivity linking preferentially neurons responding to similar orientations and along a similar axis (Schmidt et al., 1997a) (Schmidt et al., 1997b). In contrast, orientation-selective neurons of mice receive inputs that are distributed over a broad range of preferred orientations (Jia et al., 2010); Iacaruso et al., 2017; Lee et al., 2019). Pattadkal et al. (2018) demonstrate that orientation selectivity in rodents lacking orderly maps can emerge from random intracortical connectivity but predict that orientation preference shifts considerably in dependence of SF. In support of the hypothesis that this feature could be common to rodents – and distinct from cats – our data show a clear shift in orientation preference that increases with the difference in stimulated SF also in the big rodent agouti.

Noteworthy, while in carnivores, primates, and rodents, feature selectivity seems to arise from the alignment of geniculo-cortical axons firstly in the cortex (e.g. (Reid and Alonso, 1995; Chapman et al., 1991, (Lien and Scanziani, 2013, 2018; Scholl et al. (2013)) suggest that cortical orientation selectivity in rodents may also be directly adopted from orientation-selective neurons in the lateral geniculate nucleus.

Functional layout

Pinwheel-like OPMs are present in all primates and carnivores studied so far. In contrast, rodents and lagomorphs (Van Hooser, 2007) show an interspersed organization of orientation selective neurons in their visual cortices. Although this suggested phylogeny as a main reason for the observed differences, theoretical studies proposed that a small V1 size could be an alternative reason for the lack of OPMs (Meng et al., 2012) and that it may put constrains on map structure (Keil et al., 2012). The agouti has the largest rodent V1 studied so far. Accordingly, Weigand et al. (2017), from simulations of a transition from interspersed to periodic OPMs and its dependence on the number of interconnected neurons, suggested that agoutis (and capybara) “likely possess OPMs”.

Contrary to this theoretical prediction, our results do not support the presence of periodic orientation preference maps in agouti V1. Although we cannot rule out the possibility of orientation maps with a periodicity smaller than 250 um, due to our electrode spacing, such periodicity (with its implied pinwheel density) is unexpected for a visual cortex the size of the agouti's (Kaschube et al., 2010). Interestingly, in agreement with an absence of modularity, the agouti visual cortex also lacks CO blobs (Dias et al., 2014).

Arguing against the hypothesis of a uniform mechanism constrained by brain size is also a recent report of pinwheels and OPMs in mouse lemur V1, the smallest primate studied so far (Ho et al., 2020).

To be further noted, agoutis – as other rodents – have been reported to possess lower neuronal densities than species with OPMs (review in Weigand et al., 2017; personal communication, M. Garcia), which opens the possibility that although absolute size does not matter, density of neurons might. Lower densities of striate neurons could go along with lower density of thalamo-cortical afferents and maybe absent clustering of ON/OFF responses, which is held responsible for spatially ordering cortical neurons in carnivores and primates (Kremkow et al., 2016; Kremkow and Alonso, 2018).

Yet, not all neural densities of the investigated species are firmly known, and theoretical considerations speak against neuronal density as a major factor (Ibbotson and Jung, 2020). More characteristics of rodents are laterally positioned eyes and a low central-peripheral density ratio of retinal ganglion cells due to the absence of a fovea. Thus, Ibbotson and Jung (2020) discuss that the central-peripheral ratio is more closely associated than neuronal density with the expression of pinwheel-like structures predicting an interspersed layout for big rodents such as agoutis.

Alternative ideas – independent of brain size or neuron density – posit rodent-specific connectivity influencing the excitation-inhibition balance (Ohki and Reid, 2007; Hansel and van Vresswjik, 2012; Sadeh and Rotter, 2015) or the extent of astrocyte arbors of restricting the size of hypercolumns (Philips et al., 2017) as responsible for the lack of columns in rodents. Given the low selectivity and the lack of large-scale periodicity of orientation preference in agoutis, our results are consistent with these models.

Nevertheless, it is important to note that interspersed does not necessarily equal random organization. Recent studies have indicated that a “mini-columnar” structure of iso-oriented cells is present in mice (Ringach et al., 2016; Scholl et al., 2016; Kondo et al., 2016; Maruoka et al., 2017). Our results are consistent with this organizational structure since neurons recorded from the same site or along the axis perpendicular to the cortical surface tend to be more similar than neurons recorded parallel to it. However, mouse “mini-columns” have a radius of less than 50 μm, measured via 2-photon imaging at a high resolution (Ringach et al., 2016; Kondo et al., 2016). Such a periodicity would escape the resolution of intrinsic signal imaging and also electrophysiological recordings from 1 МΩ electrodes that sample from a radius of about 50–100 μm around their tip (Henze et al., 2000), rendering it impossible to assert if the local structure we see in agoutis reflects the same feature. Moreover, the similarity of orientation preference along the vertical axis could be partially explained by the pronounced orientation anisotropy for horizontal preference in our data. On the other hand, an evenly distributed orientation bias would be expected to shift differences toward smaller orientation preference differences also in the horizontal cortical axis. Since these differences are increasing more rapidly with distance in the horizontal than in the vertical axis, our data essentially favor a “mini-columnar” organization.

It would be important to include the orientation anisotropy into models of the visual cortical layout because it seems to be a feature common to species with a visual streak and low central-periphery ratio (rats: Girman et al., 1999; mouse: Dräger 1975; Salinas et al., 2017; hamster: Tiao and Blakemore 1976 and rabbits: Bousfield 1977; Murphy and Berman, 1979) and could be inherited from a bias in the retino-geniculo-cortical input.

In conclusion, our data fill an important gap of knowledge about cortical evolution as they support that phylogenetic trait – including a specific retinal layout – is more predictive for formation of periodic orientation maps than brain size. Future studies should elucidate the agoutis' functional layout on the cellular level and its local circuits producing feature selectivity.

Limitations of the study

It should be noted that the smallest available distance between simultaneously recorded channels in the horizontal axis was 250 μm and in the vertical axis 100 μm. This limits the resolution of angle differences in space and thus our conclusion pointing toward units preferring similar orientation preference clustering together. However, the observation that single units isolated from the same multi-unit (0 μm angle difference) frequently had more similar orientation preference supports that our data are in agreement with a not entirely random organization, i.e., with the previously observed mini-columns in rodents.

Resource Availability

Lead Contact

Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, Kerstin Schmidt (kschmidt@neuro.ufrn.br).

Material Availability

This study did not generate new unique reagents. The original datasets from the agouti species have not been deposited in a public repository yet.

Data and Code Availability

Data and codes that support the findings of this study are available from the authors upon reasonable request.

Methods

All methods can be found in the accompanying Transparent Methods supplemental file.

Published: January 22, 2021