Thalamocortical boutons cluster by ON/OFF responses in mouse primary visual cortex

Abstract

In higher mammals, the thalamic afferents to primary visual cortex cluster according to their responses to increases (ON) or decreases (OFF) in luminance. This feature of thalamocortical wiring is thought to create columnar, ON/OFF domains in V1. We have recently shown that mice also have ON/OFF cortical domains, but the organization of their thalamic afferents remains unknown. Here we measured the visual responses of thalamocortical boutons with two-photon imaging and found that they also cluster in space according to ON/OFF responses. Moreover, fluctuations in the relative density of ON/OFF boutons mirror fluctuations in the relative density of ON/OFF receptive field positions on the visual field. These findings indicate a segregation of ON/OFF signals already present in the thalamic input. We propose that ON/OFF clustering may reflect the spatial distribution of ON/OFF responses in retinal ganglion cell mosaics.

NEW & NOTEWORTHY Neurons in primary visual cortex cluster into ON and OFF domains, which have been shown to be linked to the organization of receptive fields and cortical maps. Here we show that in the mouse such clustering is already present in the geniculate input, suggesting that the cortical architecture may be shaped by the representation of ON/OFF signals in the thalamus and the retina.

INTRODUCTION

In higher mammals, geniculate afferents cluster according to their ON/OFF responses (1–4). This feature of thalamocortical wiring is widely believed to induce columnar, ON/OFF cortical domains, where the responses of neurons are dominated by either the onset (ON) or the offset (OFF) of luminance within their receptive fields (2–6). We have recently found that the primary visual cortex in the mouse is also parceled into ON/OFF domains. If the origin of ON/OFF cortical domains is a consequence of the clustering of afferents by ON/OFF type, one would predict that ON/OFF thalamic inputs segregate in the mouse as well. Indeed, our main finding is that thalamocortical boutons in the mouse are spatially clustered according to their responses to light increments and decrements. Moreover, fluctuations in the dominance of ON/OFF responses in thalamic boutons mirror fluctuations in the density of ON/OFF receptive field centers across the visual field. We interpret this finding as indicating that biases in the representation of ON and OFF responses arise early in the visual system, possibly going back to the distribution of ON/OFF responses in the retinal ganglion cell mosaics themselves.

Understanding the organization and origin of ON/OFF domains has been attracting increased attention because their spatial layout is linked to the structure of receptive fields and cortical maps, including those for orientation, direction, and retinal disparity (2, 6–17). Simple cells with segregated ON/OFF subregions tend to be located between the centers of ON and OFF domains, with their receptive fields organized into ON/OFF subregions matching the retinotopic location of ON/OFF signals in nearby domains (6), in species both with and without orientation maps (18). Thus, the question of how ON/OFF domains are formed is of utmost relevance to our understanding of the cortical architecture. We consider the hypothesis that ON/OFF domains can be traced back to the statistics of the retinal ganglion cell mosaics in the retina. If true, this would mean that the organization of the cortical receptive fields is initially seeded by the spatial structure of signals from the retina and relayed by the geniculate (11, 12, 14, 19–22). We argue that the findings lend support for this hypothesis because ON/OFF domains correlate with fluctuations in the distribution of ON/OFF receptive fields across the visual field, a result that cannot be obtained if we assume a mechanism that simply sorts afferents at cortical sites according to ON/OFF types from a spatially uniform representation of ON/OFF responses at the input (17).

MATERIALS AND METHODS

Animals

All procedures were approved by UCLA’s Office of Animal Research Oversight (the Institutional Animal Care and Use Committee) and were in accord with guidelines set by the National Institutes of Health. A total of four C57BL/6J wild-type (WT) mice, male (1) and female (3), aged postnatal days (P)35–56, were used in the study. Data from six different cortical volumes were obtained.

Surgery

Two-photon imaging experiments were conducted through chronically implanted cranial windows over primary visual cortex. Carprofen was administered preoperatively (5 mg/kg, 0.2 mL after 1:100 dilution). Mice were anesthetized with isoflurane (4–5% induction; 1.5–2% surgery), and body temperature was maintained at 37.5°C with a feedback heating system. Eyes were coated with a thin layer of ophthalmic ointment to prevent desiccation. Anesthetized mice were mounted in a stereotaxic apparatus with blunt ear bars in the external auditory meatus to immobilize the head. A portion of the scalp overlying the two hemispheres of the cortex (∼8 mm × 6 mm) was removed to expose the underlying skull. The skull was dried and covered by a thin layer of Vetbond. After a drying period (15 min), the Vetbond provided a stable and solid surface to affix an aluminum bracket (a head holder) with dental acrylic. The bracket was then affixed to the skull, and the margins were sealed with Vetbond and dental acrylic to prevent infections. A craniotomy over monocular V1 on the left hemisphere was conducted with a high-speed dental drill. Special care was taken to ensure that the dura was not damaged during the process. A stock concentration of AAV1-CAG-GCaMP6s (Addgene RRID:SCR_002037: AAV1.CAG.GCaMP6s.WPRE.SV40; no. 100844-AAV1; titer: ∼2e13 GC/mL) was pressure injected with a Picospritzer III (Parker, Hollis, NH). We used a thin-walled glass pipette (Warner Instruments, no. 64-0800) pulled by Sutter P-1000 to create a sharp injection pipette (∼0.3- to 0.7-μm tip), and then the last 1–2 mm of the tip was broken to create a 6-μm tip for injection. The injection pipette was filled with the virus and positioned over the dorsal lateral geniculate nucleus (LGN), with coordinates 2.1 mm posterior from bregma and 2.3 mm lateral from the midline, with a micromanipulator. Then, the pipette was slowly lowered to a depth of 2,800 μm below the pial surface. Starting at a depth of 2,800 μm, 10 puffs were given at 15–20 psi with a duration of 10 ms, each puff separated by a 4-s interval and making injections every 10 μm moving up, with the last injection made at a depth of 2,600 μm. The total volume injected was ∼0.2 μL. A sterile 3-mm-diameter cover glass was placed directly on the exposed dura and sealed to the surrounding skull with Vetbond. The remainder of the exposed skull and the margins of the cover glass were sealed with dental acrylic. Mice were allowed to recover on a heating pad. Once alert, mice were transferred back to their home cage. Carprofen was administered postoperatively for 72 h. Mice were allowed to recover from surgery for 2 wk, after which we began monitoring the levels of expression of GCaMP6s in thalamic boutons in V1. If the expression levels were found to be adequate, we moved on to measure the visual responses of boutons and analyze their distribution in the cortex, as described below.

Two-Photon Imaging

At the beginning of each session, mice were briefly sedated and administered Texas red (ThermoFisher no. D3328, 0.1 mL sc, from a 2 mg/mL PBS stock). Mice were positioned on a running wheel and head-restrained under a resonant, two-photon microscope (Neurolabware, Los Angeles, CA) controlled by Scanbox acquisition software and electronics (Scanbox, Los Angeles, CA). After a waiting period of ∼20 min, we were able to image both geniculate boutons and the cortical vasculature on the green and red photomultiplier tube (PMT) channels, respectively (Fig. 1A). The light source was a Coherent Chameleon Ultra II laser (Coherent Inc, Santa Clara, CA). The excitation wavelength was set to 920 nm. The objective was an ×16 water immersion lens [Nikon; 0.8 numerical aperture (NA), 3-mm working distance (WD)]. The microscope frame rate was 15.6 Hz (512 lines with a resonant mirror at 8 kHz). The field of view was 516 μm × 308 μm in all instances. The objective was tilted to be approximately normal to the cortical surface. An electronically tuned lens (Optotune EL-10-30-C; Dietikon, Switzerland) was used to run independent sessions acquiring data from optical planes spaced 15 μm apart (to ensure disjoint sets of boutons in different sections) starting at a depth of ∼250 μm from the cortical surface (Fig. 1B). A total of six data sets from four mice were recorded, each with a different number of optical sections (see Table 1). Images were processed with a standard pipeline consisting of image registration (based on the Texas red signal), cell segmentation, and deconvolution with suite2p (https://suite2p.readthedocs.io/). For any one optical section, the location of the cells in the imaging plane was estimated as the center of mass of the corresponding region of interest (ROI) calculated by suite2p.

Figure 1.

Two-photon imaging from thalamic boutons in V1. A: The green photomultiplier tube (PMT) channel shows the expression of GCaMP6s in thalamic afferents and boutons. The red PMT channel shows a visualization of the vasculature by injections of dextran Texas red, which assisted in the registration of the images during our analyses. B: a typical experiment consisted in sampling a volume in V1 where adjacent optical sections were separated by 15 μm. C: sparse-noise stimulation was used to evoke response from thalamic boutons. D: reverse correlation of signals from each individual bouton with bright and dark stimuli led to ON and OFF kernels in each case. Shown are 2 examples of a bouton responding largely to bright stimuli (ON bouton) and one responding largely to dark stimuli (OFF bouton). The panels on right illustrate 2-dimensional Gaussian fits to the dominant kernels, which yield an estimate of the location of the bouton’s receptive field in visual space.

Table 1.

Summary of experiments

Data Set	Mouse ID	Sex	No. of planes	N on	N off	N on+off	Correlation	P Value
1	LGNP01	F	6	631	139	28	0.51	0.003
2	LGNP07	M	7	248	115	38	0.30	0.045
3	LGNP07	M	6	519	351	315	0.64	0.001
4	LGNP11	F	4	258	88	20	0.34	0.050
5	LGNP11	F	7	381	186	23	0.43	0.005
6	LGNP12	F	10	191	59	4	0.43	0.012

Each line summarizes each of the experiments conducted including the sex of each animal, the number of optical sections (planes) obtained, the total number of ON (N_on) and OFF (N_off) boutons that pass the data selection criteria, the number of boutons with both ON and OFF response (N_on+off), and the correlation and P value between the fluctuations of the spatial distribution of ON/OFF responses in boutons on the cortex and the fluctuations of the ON/OFF receptive fields centers on the visual field. Data sets 1, 3, and 5 correspond to the sample data shown in Figs. 2 and 3.

A camera synchronized to the frame rate of the microscope imaged the contralateral eye during data collection. These data were subsequently analyzed to determine the center and size of the pupil within the image plane. The distribution of eye movements was computed, yielding a mode and a standard deviation (SD). There were no obvious differences between the analyses performed on the entire data set or on data segments where the eye position was restricted to lie within 1 SD of the mode. Here, we report the analysis using the entire data set.

A summary of the data sets available is provided in Table 1. Across all the experiments, ON boutons were the most numerous, representing 68% of the population, whereas OFF boutons accounted for 26% of the population (Table 1). About 6% of boutons showed both ON and OFF kernels and were not considered any further in our analyses.

Visual Stimulation

A Samsung CHG90 monitor positioned 30 cm in front of the animal was used for visual stimulation. The screen was calibrated with a SpectraScan PR-655 spectro-radiometer (Jadak, Syracuse, NY), generating gamma corrections for the red, green, and blue components via a GeForce RTX 2080 Ti graphics card. Visual stimuli were generated by a Processing sketch written by our laboratory using OpenGL shaders (see http://processing.org). The screen was divided into an 18 × 8 grid, resulting in a tile size of 8° × 8°, thereby matching the size of a typical LGN center (23) (Fig. 1C). Each frame of the stimulus was generated by selecting the luminance of each tile randomly as either bright (10% chance), dark (10% chance), or gray (80% chance). The stimulus was flashed for 166 ms and appeared at a rate of 1 per second. The screen was uniform gray between stimuli. Transistor-transistor logic (TTL) pulses generated by the stimulus computer signaled the onset of stimulation. These pulses were time-stamped by the microscope with the frame and line number that was being scanned at that moment the signals occurred. Sessions lasted for 25 min, generating the response of cells in the population to 1,500 stimulus presentations.

Calculation of ON and OFF Kernels

For each bouton and tile in the stimulus, we calculated the average response of the bouton locked to the presentation of bright or dark stimulus over the 15 frames (1 s) following stimulus onset. The ON kernel at a delay of t frames after stimulus onset is represented as an image of equal size to the stimulus. The value of this image at tile location (i, j) corresponds to the average response following the presentation of a bright stimulus at that location t frames after stimulus onset. We denote this image by ON(t) and adopt a similar definition of the OFF kernel, OFF(t). For each time delay, we compute the norm of the kernel normalized by the norm at t = 0: S_on(t) = ON(t)/ON(0), and, similarly, we calculate S_off(t) = OFF(t)/OFF(0). These curves typically peak at delays of ∼5 frames (corresponding to ∼320 ms). We declared a bouton to have a significant ON kernel if its normalized norm attained a peak value larger than 5 and a two-dimensional Gaussian fit of the kernel at the peak delay time accounts for at least 50% of its variance. A similar definition applied to OFF kernels. As a result, a bouton could have no significant maps, either significant ON or OFF maps, or both (see Table 1). The two-dimensional Gaussians fit to the dominant ON/OFF kernels yield their center locations (y₁, y₂) on the visual field (Fig. 1D).

Canonical Correlation Analysis

Each bouton was assigned a coordinate in cortical space, (x₁, x₂, x₃) (Fig. 1B) and, for its dominant kernel, one in visual space, (y₁, y₂) (Fig. 1D). Canonical correlation analysis (MATLAB’s RRID:SCR_001622 cannoncorr() function) was done to find transformations $\widehat{x}=A\left(x-\overline{x}\right)$ and $\widehat{y}=B\left(y-\overline{y}\right)$ such that the covariance of $\widehat{x}$ and $\widehat{y}$ is diagonal and the correlations between matching canonical coordinates are maximized. The transformations are further constrained so that the variance of the canonical coordinates equals 1. In our case, the matrix A is n × 3, while the matrix B is n × 2, where n is the total number of boutons with at least one significant map. The goal of canonical correlation analysis is to represent cortical and visual space on a common two-dimensional latent space. The result is an optimal alignment of cortical and visual space under an affine transformation.

Density Estimation

Given a distribution of points in native or canonical space (either cortical or visual), we estimate the density distribution by $f\left(x\right)=\left(1/n\right){\displaystyle {\sum }_{i=1}^n{G}_{\sigma }\left(x\mathrm{-}{x}_i\right)}$, where G_σ(·) is a two-dimensional Gaussian kernel of width σ and {x_i} (i = 1,…, N) is the set of points under consideration (24). For canonical variables, we chose a width of σ = 0.25, following the rule of thumb bandwidth estimator 0.9 n^−1/5, with n ∼ 500, which is a typical size for our data (24). In native cortical space, we used σ = 48 μm. Estimates of $f_{\text{on}}^x$, $f_{\text{off}}^x$, $f_{\text{on}}^y$, and $f_{\text{off}}^y$ and their counterparts in canonical space were all obtained by this procedure.

To evaluate the likelihood that the observed fluctuations could arise by chance, we randomly shuffled the labels of ON and OFF cells in N = 1,000 simulated experiments. For each case, we calculated the distribution of fluctuations at each point in cortical and visual spaces, which enabled us to compute P = 0.001 level sets. Similarly, we computed the distribution of correlation coefficients between fluctuations of ON/OFF cells on the cortical surface and those of their center locations in the visual field, allowing us to calculate the statistical significance of the observed correlation in the original data.

Data and Code Availability

The data from all experimental sessions, including the raw kernels and estimated parameters, have been deposited at https://doi.org/10.6084/m9.figshare.21554547.v1. Sample code describing the structure of the database and the calculation of ON/OFF bouton distributions is provided along with the data in the same repository to assist in further analyses.

RESULTS

We used two-photon imaging in alert, head-fixed mice to characterize the visual responses of thalamic boutons in a volume of primary visual cortex. We expressed GCaMP6s in the LGN by injections of pAAV.CAG.GCaMP6s.WPRE.SV40 (Fig. 1A, left) and measured the responses of individual boutons with two-photon imaging. To assist in the registration of our images, we labeled the vasculature with Texas red (Fig. 1A, right). We used sparse-noise stimulation (Fig. 1C) to map the receptive fields of thalamocortical boutons within a cortical volume spanned by 6–10 optical sections 15 μm apart (Fig. 1B). A standard data analysis pipeline comprised of image registration (based on the red PMT channel), cell segmentation, signal extraction, and deconvolution steps yielded the estimated spiking of synaptic boutons. The centroid of the regions of interest (ROIs), along with the depth of the optical section, allowed us to assign each bouton a coordinate in cortical space: (x₁, x₂, x₃) (Fig. 1B). As the objective of the microscope was nearly normal to the cortex, (x₁, x₂) represents the projection of a bouton on the cortical surface and x₃ represents its depth.

We computed the ON and OFF receptive fields (or kernels) of each bouton in the population by correlating their responses with the locations of bright and dark patches in the stimuli (Fig. 1D). We defined ON/OFF kernels at the optimal delay for which their norms attained their maximum value. Boutons that had only a statistically significant ON kernel (68% of the population) were defined as ON (Fig. 1D); a similar definition was applied to OFF (26% of the population). A small fraction of boutons (∼6% of the population) had both ON and OFF kernels and were not considered further in the analysis.

Thalamocortical Boutons Cluster by ON/OFF Types

To test whether boutons cluster in cortical space according to their type, we computed the difference in the spatial density of ON and OFF boutons. Given a set of points $\left(x_1^i,x_2^i\right)$ corresponding to ON boutons (ignoring their depth) we estimated their density via a kernel estimate, $f_{\mathrm{on}}\left(x_1,x_2\right)=1/N_{\mathrm{on}}{\displaystyle {\sum }_{i}G\left(x_1-x_1^i,x_2-x_2^i\right)}$, where N_on is the number of ON boutons and G(·) is a Gaussian kernel (24) (see materials and methods). A similar density estimate can be obtained for OFF boutons, as $f_{\mathrm{off}}\left(x_1,x_2\right)=1/N_{\mathrm{off}}{\displaystyle {\sum }_{i}G\left(x_1-x_1^i,x_2-x_2^i\right)}$. The fluctuation in the density of ON and OFF boutons is given by the difference f_on (x₁, x₂) − f_off (x₁, x₂). We observed that f_on(x₁, x₂) and f_off(x₁, x₂) have nonuniform distributions that tend to peak at different locations (Fig. 2). As a result, the difference f_on(x₁, x₂) − f_off(x₁, x₂) had spatial structure with regions where the probability density of ON cells was higher than OFF cells (ON domains) and regions where the probability density of OFF cells was higher than ON cells (OFF domains). The statistical significance of these fluctuations was assessed by Monte Carlo simulations where the ON and OFF labels of the cells were randomly shuffled, allowing us to determine the locations where the deviations attained a significance at the 0.001 level (Fig. 2, right, blue and red level sets). These results are typical of our data sets, all of which exhibited regions with statistically significant ON and OFF clustering. We conclude that thalamocortical boutons in V1 are organized into ON/OFF domains. We emphasize that the segregation is far from perfect, but the analysis certainly demonstrates that boutons cluster by ON/OFF type beyond what might be expected by chance.

Figure 2.

Clustering of ON/OFF boutons. Each row corresponds to a separate experiment. Left: the distribution of ON boutons (projected onto the cortical surface by disregarding their depth). Center: the density of OFF boutons. Right: the fluctuations in density. Red and blue contours represent areas where the difference in density exceeds the magnitude of what would be expected by randomly shuffling the ON/OFF levels of the boutons at a significance level of P = 0.001.

Fluctuation of ON/OFF Boutons Correlates with Fluctuation of ON/OFF Receptive Field Centers

A possible explanation for the clustering of ON/OFF boutons is that there might already be an imbalance in the coverage of ON/OFF center receptive fields on the visual field, originating in the retinal ganglion cell mosaics, that is maintained because of retinotopic mapping in the thalamus and cortex. This hypothesis predicts a correlation between the fluctuations in the representation of ON/OFF signals by thalamic boutons and the fluctuations in the density of ON/OFF receptive field centers on the visual field (18).

To test this prediction, we bring cortical and visual fields into alignment using canonical correlation analysis (25), which generates two linear transformations of the data. Cortical space is mapped to ${\widehat{x}}_1=a_{11}(x_1-{\overline{x}}_1)+a_{12}(x_2-{\overline{x}}_2)+a_{13}(x_3-{\overline{x}}_3)$ and ${\widehat{x}}_2=a_{21}(x_1-{\overline{x}}_1)+a_{22}(x_2-{\overline{x}}_2)+a_{23}(x_3-{\overline{x}}_3)$, or in matrix form $\widehat{x}=\left(x-\overline{x}\right)A$. Similarly, the visual field is mapped by ${\widehat{y}}_1=b_{11}(y_1-{\overline{y}}_1)+b_{12}(y_2-{\overline{y}}_2)$ and ${\widehat{y}}_2=b_{21}(y_1-{\overline{y}}_1)+b_{22}(y_2-{\overline{y}}_2)$, or in matrix form $\widehat{y}=\left(y-\overline{y}\right)B$. The transformations maximize the correlations between the pairs $({\widehat{x}}_1,{\widehat{y}}_1)$ and $({\widehat{x}}_2,{\widehat{y}}_2)$, while ensuring the orthogonality of $({\widehat{x}}_1,{\widehat{y}}_2)$ and $({\widehat{x}}_2,{\widehat{y}}_1)$, and equalize the variance of all canonical variables to 1. The inclusion of cortical depth (x₃) allowed us to compensate for slight departures of the objective from the surface normal.

The outcome of canonical correlation analysis is a representation of each bouton in the population by its canonical coordinates in cortical space $({\widehat{x}}_1,{\widehat{x}}_2)$ and its canonical coordinates in visual space $\left({\widehat{y}}_1,{\widehat{y}}_2\right)$. Using this representation of the data, we can compute the fluctuation of the density of ON/OFF cells in the canonical cortical domain and the fluctuation in the density of ON/OFF receptive field centers in the canonical visual field and test whether they correlate with each other.

We used kernel density techniques to estimate the probability density of ON and OFF cells in canonical cortical space, denoted by $f_{\text{on}}^{\widehat{x}}$ and $f_{\text{off}}^{\widehat{x}}$, respectively (materials and methods). Similarly, we estimated the probability density of ON and OFF receptive field centers in canonical visual space, yielding $f_{\text{on}}^{\widehat{y}}$ and $f_{\text{off}}^{\widehat{y}}$. Fluctuations in the spatial distribution of ON and OFF cells in canonical cortical space were calculated as the difference $f_{\text{on}}^{\widehat{x}}-f_{\text{off}}^{\widehat{x}}$. Similarly, to measure fluctuations in the distribution of ON and OFF receptive field centers in canonical visual space, we calculated the difference $f_{\text{on}}^{\widehat{y}}-f_{\text{off}}^{\widehat{x}}$ (Fig. 3).

Figure 3.

Fluctuations in the density of ON/OFF boutons correlates with the density of ON/OFF-center receptive field centers on the visual field. Each panel corresponds to a separate experiment. Top, from left to right: the density of ON boutons in canonical cortical space, the density of OFF boutons, and their difference. Bottom, from left to right: the density of ON-center receptive fields in canonical visual field, the density of OFF-center receptive fields, and their difference. We see that in all cases the fluctuations correlate. Densities are normalized to their peak. Fluctuations are normalized to their maximum absolute value.

Replicating our observations in native cortical space, the distributions of $f_{\text{on}}^{\widehat{x}}$ and $f_{\text{off}}^{\widehat{x}}$ tend to be patchy, peaking in different locations, which results in the difference $f_{\text{on}}^{\widehat{x}}-f_{\text{off}}^{\widehat{x}}$ having statistically significant peaks and troughs (Fig. 3, top). We assessed the likelihood that the observed magnitudes in the fluctuations of $f_{\text{on}}^{\widehat{x}}-f_{\text{off}}^{\widehat{x}}$ could arise by chance by using Monte Carlo simulations where ON/OFF labels were randomly shuffled (materials and methods). Level sets were computed corresponding to the P = 0.001 significance level (Fig. 3, red and blue solid curves). As expected, we observe a clustering of ON/OFF boutons in the transformed canonical space as well (a linear transformation largely preserves the domains observed in cortical space).

Similarly, we can calculate the density of receptive field centers for ON ($f_{\text{on}}^{\widehat{y}}$ and OFF ($f_{\text{off}}^{\widehat{y}}$) boutons, as well as their fluctuations, $f_{\text{on}}^{\widehat{y}}-f_{\text{off}}^{\widehat{y}}$, in canonical visual space (Fig. 3, bottom). These data corroborate our prediction: fluctuations in $f_{\text{on}}^{\widehat{x}}-f_{\text{off}}^{\widehat{x}}$ are mirrored by fluctuations in the balance of ON/OFF receptive field centers on the canonical visual field, $f_{\text{on}}^{\widehat{y}}-f_{\text{off}}^{\widehat{y}}$. This is evidenced by the correlation between these functions (Fig. 3). Statistical significance was established by computing the likelihood that the observed level of correlation could arise by chance in controls that randomly shuffled ON/OFF labels (Fig. 3, P values; materials and methods).

DISCUSSION

We have recently shown that, like other mammals (1–4), mouse primary visual cortex is parceled into ON/OFF domains (18). Those data revealed that fluctuations in the distribution of ON/OFF neurons on the cortical surface, which define ON/OFF domains, were mirrored by fluctuations in the distribution of ON/OFF receptive field centers on the visual field. We suspected that the input to the cortex itself might be biased in its representation of ON/OFF signals. The present study was designed to investigate this question by measuring the spatial distribution of ON/OFF thalamic boutons in the cortex. Our findings show that 1) the distribution of ON/OFF thalamic boutons fluctuates in cortical space (Fig. 3, top, $f_{\text{on}}^{\widehat{x}}-f_{\text{off}}^{\widehat{x}}$), 2) the distribution of ON/OFF centers fluctuates across the visual field (Fig. 3, bottom, $f_{\text{on}}^{\widehat{y}}-f_{\text{off}}^{\widehat{y}}$), and 3) these two fluctuations correlate (comparison between Fig. 3, top and bottom.

The precise mechanism generating the clustering of afferents remains to be elucidated. One possibility is that ON/OFF signals segregate in the cortex via correlation-based, activity-dependent competition, even in a situation when the representation of ON/OFF signals at earlier stages in the visual system is uniform (17, 26). In this scenario, ON/OFF thalamic receptive fields covering the same part of the visual field end up projecting to different cortical targets (17). Our findings, however, are at odds with this picture, as it does not explain how fluctuations in the relative representation of ON/OFF signals in both cortical neurons (18) and boutons (present study) correlate with the distribution of ON/OFF receptive fields centers on the visual field (Fig. 3). Note that any spatial rearrangement of the location of thalamic boutons in cortical space does not affect the distribution of their receptive field centers on the visual field. Instead, we propose that the correlation between the ON/OFF representation in visual and cortical space can be explained by a simpler model in which a fluctuation in the representation of ON/OFF signals in the retina is “copied” into cortical space by the retinotopic map.

Indeed, a partial segregation of ON/OFF signals is already present at the scale of individual retinal ganglion cell receptive fields, as ON/OFF mosaics appear to “repel” each other (27). If the retinotopy conveys this pattern to the cortex, and the distance between adjacent afferents innervating V1 is comparable to the size of the thalamic arborizations (28), it is possible for individual afferents to dominate the center of the region they target and generate corresponding ON/OFF domains. This scenario is consistent with data from the cat, where receptive fields within a domain only change in location by 0.2 of the receptive field center size (6). Performing similar analyses in the mouse to investigate how the retinotopy and scatter of ON/OFF responses change within and across domains is an important step in understanding the spatial scales involved in the generation of the contrast polarity map and comparing between species.

Finally, a surprising feature of our data is that the proportion of ON boutons is larger than for OFF boutons (Table 1). This was unexpected because OFF responses dominate the cortex of mice (21, 29), cats, tree shrews, and primates (2, 30, 31). We can only speculate as to why this might be the case. One possibility is that the sparse-noise stimulus was more effective in driving ON inputs than OFF. The size of the checkers we used was 8° of visual angle, which could have biased the responses toward sustained ON inputs from the geniculate. We know, from a previous study, that the optimal spot diameter for sustained ON geniculate inputs is 11.6°, for sustained OFF 15.9°, and for transient OFF 14.0° (23). Thus, in our attempt to map the receptive fields with high spatial resolution and avoid surround suppression, we might have biased responses toward the ON pathway. To avoid this problem, future studies of ON/OFF domains may require the use of sparse stimuli with variable spot/checker sizes (23).

In conclusion, the findings demonstrate the clustering of afferents by ON/OFF responses in the mouse. The data are consistent with the notion that cortical receptive fields are shaped by the spatial distribution of ON and OFF inputs in the visual field, a property that is intrinsic to the retinal representation (11, 14, 19, 20). Of course, we have not yet conclusively demonstrated that the imbalances observed in the thalamic input correlate with ON/OFF domains in individual animals. We are currently testing whether this is the case by performing dual-color imaging where we reconstruct both the distribution of ON/OFF thalamic boutons and the ON/OFF responses of cortical neurons. The prediction is that fluctuations between ON/OFF signals in boutons and neurons will correlate in cortical space. Generally, data from such experiments will help us establish whether there is a correlation between the spatial distribution of ON/OFF boutons and the organization of cortical receptive fields in individual animals.

DATA AND CODE AVAILABILITY

The data from all experimental sessions, including the raw kernels and estimated parameters, have been deposited at https://doi.org/10.6084/m9.figshare.21554547.v1. Sample code describing the structure of the database and the calculation of ON/OFF bouton distributions is provided along with of the data in the same repository to assist in further analyses.

GRANTS

This work was supported by NIH Grant NS116471 (D.L.R.)

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

D.L.R. conceived and designed research; E.T. and D.L.R. performed experiments; D.L.R. analyzed data; D.L.R. interpreted results of experiments; D.L.R. prepared figures; D.L.R. drafted manuscript; E.T. and D.L.R. approved final version of manuscript.