The convergence of eye-specific thalamic inputs to visual cortical neurons forms the basis of binocular vision. Inputs from the same eye that signal light increment (On) and decrement (Off) are spatially segregated into subregions, giving rise to cortical receptive fields (RFs) that are selective for stimulus orientation. Here we map RFs of binocular neurons in the mouse primary visual cortex using spike-triggered average. We find that subregions of the same sign (On–On and Off–Off) preferentially overlap between the 2 monocular RFs, leading to binocularly matched orientation tuning. We further demonstrate that such subregion correspondence and the consequent matching of RF orientation are disrupted in mice reared in darkness during development. Surprisingly, despite the lack of all postnatal visual experience, a substantial degree of subregion correspondence still remains. In addition, dark-reared mice show normal monocular RF structures and binocular overlap. These results thus reveal the specific roles of experience-dependent and -independent processes in binocular convergence and refinement of On and Off inputs onto single cortical neurons.
Optimal functioning of the nervous system requires precise and selective connections between neurons, which are established through elaborate developmental processes. Such accurate wiring is particularly important for higher sensory and associative centers that receive converging inputs from multiple parallel channels. In the primary visual cortex (V1), individual simple cells receive 2 streams of thalamic inputs that separately signal light increment (On) and decrement (Off) (Hubel and Wiesel 1959). Importantly, the precise spatial layout of the On and Off subregions in the receptive fields (RFs) renders the simple cells selective for stimulus orientation (Hubel and Wiesel 1962; Ferster and Miller 2000). Simple cells also receive another set of On and Off inputs, those from the other eye, making them binocularly responsive (Hubel and Wiesel 1959, 1962). Individual V1 neurons are tuned to similar orientations through the 2 eyes (Nelson et al. 1977; Bridge and Cumming 2001; Wang, Sarnaik and Cang 2010), a feature presumably required for normal binocular perception. While numerous studies have investigated orientation selectivity development (Hubel and Wiesel 1963; Chapman and Stryker 1993; Chapman and Godecke 2000; White et al. 2001; Rochefort et al. 2011), how the 2 sets of On/Off inputs converge and give rise to binocularly matched orientation preference is largely unknown.
A theoretic model of the development of monocular RFs (Miller 1994) also explains their binocular development (Erwin and Miller 1998). In this model, if the same-sign inputs from the 2 eyes (On–On, Off–Off) are temporally correlated at the same retinotopic locations, the 2 monocular RFs of individual neurons would develop subregion correspondence, that is, spatial overlap between same-sign subregions. Alternatively, subregion anti-correspondence would develop if the correlation were between inputs of opposite signs (On–Off). Thus, whether subregion correspondence or anti-correspondence emerges depends on the process that drives input correlations, but either of these arrangements can give rise to binocularly matched orientation preferences. Experimental evidence for subregion correspondence has been indirect, being inferred from the near-zero disparity tuning of simple cells in cats (Anzai et al. 1999). However, a substantial fraction of cells in V1 are not tuned to disparity (∼60% in anesthetized cats [Ferster 1981]; and ∼50% in awake monkeys [Prince et al. 2002]) and yet they have binocularly matched orientation preference. Therefore, the full scope and degree of spatial relationship between the 2 monocular RFs remain to be investigated. Even more importantly, what developmental factors contribute the establishment of this binocular relationship is still completely unknown.
Mice have emerged as a useful model in studying visual cortical function (Ohki et al. 2005; Niell and Stryker 2008; Jia et al. 2010; Liu et al. 2010; Smith and Hausser 2010; Van den Bergh et al. 2010; Bonin et al. 2011; Ko et al. 2011; Atallah et al. 2012; Vreysen et al. 2012; Wilson et al. 2012) and development (Smith and Trachtenberg 2007; Kuhlman et al. 2011; Rochefort et al. 2011; Li, Lu et al. 2012; Li, Ma et al. 2012). We have recently discovered in mice that the binocular matching of orientation preferences requires normal vision during development (Wang, Sarnaik and Cang 2010), suggesting that visual experience may be a signal that correlates inputs from the 2 eyes. As visually induced activities are likely correlated binocularly between inputs of the same signs, subregion correspondence should develop. Alternatively, if other processes drive binocular convergence, both subregion correspondence and anti-correspondence would be possible. The aim of this study, therefore, is to examine the binocular relationship of cortical RFs in mice and whether it requires visual experience to establish.
Using 2 separate methods to map cortical RFs, we find that mouse V1 neurons overwhelmingly display subregion correspondence, which is significantly degraded in mice deprived of visual experience during development. These results support the aforementioned model and provide an RF basis, that is, overlap of same-sign subregions, for the experience-dependent matching of orientation preference. Surprisingly, substantial subregion correspondence remains in the visually deprived mice, revealing a novel role of experience-independent processes in driving binocular convergence. Also, monocular RF structures and orientation tuning are normal in the visually deprived mice. These results elucidate the exquisite specificity with which experience-dependent and -independent processes coordinate to assemble thalamocortical circuits during development.
Materials and Methods
Wild-type C57BL/6 mice of both sexes, aged between postnatal day (P) 31–36, were used in our experiments. Normal-reared mice (NR, n = 66) were kept in a 12 h light:12 h dark cycle. Dark-reared animals were kept in constant, complete darkness either from postnatal day 11 (DR11, n = 81 mice) or from birth (DR0, n = 24 mice) with their dams until the day of recording at P31–36. For the mice dark-reared from birth, pregnant females were put in the dark 2–7 days before estimated delivery. All experimental protocols were approved by Northwestern University Institutional Animal Use and Care Committee.
In vivo extracellular, single-unit recordings were made in anesthetized mice, following our published procedures (Wang et al. 2009; Wang, Sarnaik and Cang 2010; Wang, Sarnaik, Rangarajan et al. 2010). Briefly, animals were anesthetized with urethane (1.2–1.3 g/kg, i.p.) and supplemented with the sedative chlorprothixene (10 mg/kg, i.m.). Atropine (0.3 mg/kg) and dexamethasone (2.0 mg/kg) were administered subcutaneously to minimize mucus secretion and edema, respectively. The level of anesthesia was confirmed by the lack of toe pinch reflex and additional urethane (0.2–0.3 g/kg) was administered as needed. Temperature was monitored with a rectal probe and maintained at 37°C through a feedback heater control module (FHC, Bowdoinham, ME, USA). Silicone oil was applied on both eyes to prevent drying. A tracheotomy was performed to deliver oxygen during the recordings in animals weighing over 20 g. Most animals weighed lesser than this and could be maintained in a stable condition by oxygen delivery through the nose. A small craniotomy of ∼ 2 mm2 was made over the binocular zone of the primary visual cortex, 3.0–3.2 mm lateral from the midline suture, and 0.5–1 mm anterior from the lambda suture in the left hemisphere, to expose the brain for recording with 5–10 MΩ tungsten microelectrodes (FHC, Bowdoinham). In a few experiments, intrinsic imaging (Kalatsky and Stryker 2003) was performed to confirm the location of the binocular zone (Wang, Sarnaik and Cang 2010).
Electrical signals, both spikes (filtered between 0.5 and 7 kHz and sampled at 25 kHz) and field potentials (filtered between 10 and 300 Hz and sampled at 800 Hz), were acquired using a System 3 workstation (Tucker Davis Technologies, FL, USA). Spike waveforms were further sorted offline into single units using OpenSorter (Tucker Davis Technologies). Electrode penetrations were made perpendicular to the pial surface and majority of the recorded cells (∼70%) were between 350 and 550 μm in depth, corresponding to the thalamo-recipient layer IV (Niell and Stryker 2008; Liu, Li et al. 2009). In a few experiments, LFP recordings at different depths of regular spacing were made to estimate the cortical layers at each depth, using current-source density analysis (Freeman and Nicholson 1975; Niell and Stryker 2008). This confirmed that our depth measurements were largely accurate.
In a few separate experiments (n = 4 mice), we simultaneously mapped RFs and monitored eye positions under the same experimental conditions as the binocular recordings. The pupil position was acquired with a high-speed CCD camera (UNIQ Vision, Inc., 400 Hz) and illuminated by an infrared LED ring. Images were captured and processed with a Labview-based eye-tracking software (Sakatani and Isa 2004) and analyzed with custom MATLAB code.
The animals were euthanized at the end of recordings by an overdose of pentobarbital solution (150 mg/kg, Euthasol from Virbac).
Visual stimuli were generated using customized routines written in MATLAB (Mathworks, MA, USA) (Niell and Stryker 2008) using the Psychophysics Toolbox extensions (Brainard 1997; Pelli 1997). They were displayed on a flat panel CRT video monitor (40 × 30 cm, 60 Hz refresh rate, mean luminance ∼ 35 cd/m2) placed 25 cm away, in front of the animal. The viewing distance was chosen following previous physiological studies of the mouse visual cortex (Dräger 1975; Mangini and Pearlman 1980; Gordon and Stryker 1996; Niell and Stryker 2008) and supported by the fact that mouse eyes have a large depth of focus due to their small size (Green et al. 1980; Lin et al. 2004; de la Cera et al. 2006). The stimulus was presented separately through the 2 eyes to map monocular RFs of binocular V1 neurons. In experiments using spike-triggered analysis, Gaussian noise movies were generated in the frequency domain and filtered to contain only a subset of frequencies relevant to mouse V1 neurons (Niell and Stryker 2008) to increase the signal-to-noise ratio of the responses (Ringach et al. 1997) (low-pass, spatial frequency cut-off at 0.08 cpd, temporal frequency at 4 Hz). These dense noise stimuli were slightly more effective in eliciting responses from mouse V1 cells than those used before (Niell and Stryker 2008) and did not require post hoc corrections of their response frequency spectra. The movie was generated at 60 × 60 pixels and displayed at 600 × 600 pixels on the monitor, covering ∼60° × 60° in visual space. It contained 9,000 frames and was presented at 30 Hz (5 min in duration). The movie was multiplied with a sinusoidally varying contrast of 0.1 Hz to prevent response adaptation and presented 3–4 times in order to evoke sufficient number of spikes. The visual stimuli were displayed on a gray scale of 255 levels, and gamma-corrected in nearly all experiments. In a small number of early experiments lacking this (n = 29 cells), a post hoc correction was performed.
In some experiments, we mapped the RFs with sparse noise stimuli consisting of single, bright squares (4°) flashed at one of many locations on a square grid (Liu, Li et al. 2009; Wang, Sarnaik, Rangarajan et al. 2010). The grid was usually 11 × 11 or 15 × 15 in dimensions, and in a few cases 9 × 9. The squares were flashed On for 500 ms, Off for 500 ms between flashes, and were displayed on a gray background (luminance = 10 cd/m2). They were repeated for at least 5 times at each location and presented in a pseudorandom sequence. The stimulus set included a blank condition, in which no flash was displayed, in order to determine the spontaneous firing rate.
In all cells, we also measured their orientation tuning by presenting drifting sinusoidal gratings (0–360°, 12 linear steps; spatial frequency 0.01–0.08 cpd, 4 logarithmic steps; temporal frequency 2 Hz; contrast 100%) (Niell and Stryker 2008; Wang et al. 2009; Wang, Sarnaik and Cang 2010). Each condition was presented for 1.5 s, with a 0.5 s inter-stimulus interval, in a pseudorandom order, for 4–6 trials. The response to a particular stimulus condition was obtained by averaging the number of spikes over the 1.5 s stimulus duration, across all trials and subtracting the spontaneous rate, which was the mean response to a randomly interleaved blank condition.
Spatio-temporal RFs of V1 neurons were obtained from their responses to the noise movies using spike-triggered average (STA) (Jones and Palmer 1987; Ringach et al. 1997), by calculating the mean of the movie frames preceding each spike. This was done for 9 consecutive frames before each spike, and the frame at which the maximum amplitude of STA occurred was selected as the RF for all further analyses. This was usually the second (66 ms latency) or third (99 ms) frame before the spike. Pearson's correlation coefficients were calculated for pairs of unfiltered spatial STAs. For all further analysis of RF structure the spatial STAs were first thresholded. We used 3 different values of cut-offs to ensure that our conclusions did not depend on the threshold. For all the results presented in the main text, we determined the thresholds by analyzing the 4 corners (10 × 10 pixels each) of the STA (60 × 60 pixels). These regions clearly did not contain any part of the RFs and their STA values were used to calculate the mean and standard deviation of the noise. We first thresholded the data with an objective cut-off (mean ± 2 × standard deviation of the noise). All pixels that had values within the threshold were set to zero. Since this is a stochastic threshold, some points may exceed the threshold by chance. In such cases we manually removed the points that were scattered away from the obvious RF. In no case were individual subregions or their precise areas selected manually. Every STA was visually inspected multiple times to ensure that no responsive regions were discarded due to thresholding. In addition, we used thresholds of mean ± 3 × standard deviation of the noise and 30% of the positive (On) and negative (Off) peak STAs, and all the major conclusions stayed the same.
After thresholding, all connected pixels of the same sign were assigned as 1 subregion. For each subregion, we calculated its overlap indices with all subregions through the other eye. The overlap index (OI) quantifies how much of a given subregion (or a whole RF) overlaps with another one, that is, the ratio of the overlapped pixels over the total pixels of the subregion in question. The maximum of all the OIs of each subregion was selected and assigned a positive sign if the maximum overlap happened between the subregions of the same sign (On–On or Off–Off), or a negative sign if opposite signs overlapped. This value was referred to as the subregion OI (sOI) and used to quantify the degree of subregion correspondence (range: −1 to 1). For each cell, we also calculated a correspondence index, which was defined as the median of all the sOIs of the subregions in this cell's 2 monocular RFs. Similarly, we quantified the spatial overlap between the contralateral and ipsilateral RFs of individual neurons by calculating the OI between the 2 RFs without considering On/Off signs.
The orientation of each RF was determined by first performing a 2-dimensional Fourier transform of the STA (post-thresholding) using the built-in MATLAB function fft2 and then calculating the angle of its weighted circular mean. The preferred orientation in response to drifting gratings was calculated as the circular mean weighted by the responses (spontaneous subtracted) to all the 12 directions (Niell and Stryker 2008). The correlation between the RF orientation and the preferred orientation was preserved even without thresholding the RFs (data not shown). Cells were classified into simple and complex based on the discrete Fourier transform of their responses to drifting gratings at the preferred direction and spatial frequency (Niell and Stryker 2008). Cells that showed a temporal modulation at the stimulus frequency (an F1/F0 ratio ≥1) through both eyes were classified as simple and the rest, which showed an F1/F0 ratio <1 through at least 1 eye were considered complex.
The responses to flashing spots at each location were analyzed by counting spikes within a time window of 250 ms (from 50 to 300 ms after flash onset or offset) in each trial. We separated the spikes after flash onset and offset in order to obtain the On and Off subregions. The spontaneous firing was analyzed in the blank stimulus condition and the mean + 2 × standard deviation of the spontaneous rate was calculated as the threshold. The cell was considered responsive, at a given grid location, if there were more spikes than the threshold in at least 50% of the trials. The mean spike rate during the 250 ms time window at each responsive location, after subtraction of the mean spontaneous rate, was used for all further analyses. From these thresholded data, continuous responsive locations were selected manually to obtain individual subregions. The overall RF overlap and subregion correspondence were analyzed by the same method as for the STA data.
We generated randomized distributions from our datasets of sOI in order to test the significance of the observed subregion correspondence. Specifically, individual subregions of an RF were randomly assigned On or Off with equal probability, followed by calculation of sOI, as described above. The obtained random sOI distributions were then compared with the observed. In order to test the significance of subregion overlap as a function of RF orientation matching, we rotated the ipsilateral RFs by 36 different angles (0° to 360° at 10° steps) relative to the contralateral RF, followed by calculation of sOI in each case. The obtained sOI values and their corresponding RF orientation differences were analyzed in the same way as the observed data and compared with it to test statistical significance.
All statistical tests were evaluated at α = 5% probability of false positives. Two-sided statistical tests were performed. In figures, *P < 0.05, **P < 0.01, and ***P < 0.001. All values, except where mentioned and in Figs 5B, 6F, and 7F, are expressed as mean ± standard error of mean. In Figs 5B, 6F, and 7F, values are expressed as median ± standard error of the median, which was calculated using a smooth bootstrapping method.
Nonparametric tests that do not require any assumptions about the distribution of the data were used in all cases. Comparison of distributions was done using the two-sample Kolmogorov–Smirnov test (K–S test) and comparisons between means or medians of datasets were done using two-sample Mann–Whitney U test or Wilcoxon signed-rank test (for paired data). Comparison of a dataset with either a standard uniform distribution or a randomized distribution was done using one-sample K–S test. Statistical analyses and graphing were done in MATLAB and Prism (GraphPad Software, Inc.).
Binocular Matching of RF Orientations in Mouse Cortical Neurons
We determined the RF structure of individual V1 neurons separately for each eye (Fig. 1A), in mice after the critical period of visual cortical development (postnatal day 31–36). We mapped the RFs using spike-triggered average (STA) with dense noise stimuli (Jones and Palmer 1987). In addition, we recorded each cell's responses to drifting sinusoidal gratings in order to determine the relationship between RF structure and orientation tuning, and their degree of binocular matching.
The STA method was effective in mapping RFs in cells with simple-like characteristics, such as temporal modulation in response to drifting gratings (Fig. 1B, n = 59/64 cells had an F1/F0 ratio of ≥1 through both eyes). Most of these cells were located in the thalamo-recipient layer IV (Fig. 1C, n = 45/64 cells at cortical depths of 350–550 μm). Monocular RF geometry was similar between the 2 eyes and consisted of similar numbers of alternating On and Off subregions (Fig. 1D and E, subregions/RF: On, 1.30 ± 0.05, Off, 1.40 ± 0.06, P = 0.10, Wilcoxon signed-rank test). The subregion sizes were also similar between the 2 eyes (contra, 123.4 ± 5.8 deg2, ipsi, 110.7 ± 5 deg2, P = 0.3), but slightly more subregions were observed for the contralateral RFs (Fig. 1D and E, subregions/RF: contra, 1.44 ± 0.06, ipsi, 1.27 ± 0.06, P = 0.03, Wilcoxon signed-rank test).
We next examined whether the orientations of the 2 monocular RFs are matched binocularly for individual cortical neurons. Each RF's orientation was determined as the angle along which the RF subregions were oriented (see Materials and Methods). The RF orientation was indeed binocularly matched, with its interocular difference distributed near zero (Fig. 1F; median of their absolute values ± standard error = 15.5 ± 2.1°). As expected from the feedforward model proposed by Hubel and Wiesel (1962) and shown by previous studies for contralateral RFs in mice (Niell and Stryker 2008; Liu et al. 2011), the RF orientation of individual neurons correlated with their orientation preference through both contralateral and ipsilateral eyes (Fig. 1G, P < 0.001 for both, compared with a uniform distribution if there were no correlation). Consequently, the similar RF orientations through the 2 eyes provide an RF basis for the binocularly matched orientation preference that was revealed in our previous study using grating stimuli (Wang, Sarnaik and Cang 2010).
Subregion Correspondence between Contralateral and Ipsilateral RFs
The matching of RF orientations could be a consequence of spatial overlap of either same-sign or opposite-sign subregions between the 2 monocular RFs. We thus determined which of these 2 binocular relationships exist in these neurons. First, we examined whether the contralateral and ipsilateral RFs overlap spatially without considering the sign of component subregions, using an overall RF OI (ranging from 0 to 1, see Materials and Methods for details). RFs through the 2 eyes were largely overlapped (Fig. 2A and B; median = 0.51 ± 0.02), thus allowing analysis of subregion correspondence. We then, for each subregion in an RF, calculated an sOI, which quantifies the degree of overlap between a given subregion and its most-overlapped subregion through the other eye. The sOI magnitude ranged between 0 (no overlapping subregion) and 1 (complete subregion overlap) and were assigned positive, if the overlap was between subregion pairs of the same signs (On–On or Off–Off), or negative, if between opposite signs (On–Off or Off–On). Interestingly, we observed a 3:1 prevalence of positive values of sOI (Fig. 2C; n = 218/347 subregions, 63%) over negative ones (n = 66/347 subregions, 19%), indicating a preference for subregion correspondence over anti-correspondence. There was also a substantial fraction of zeros in the distribution (n = 63/347 subregions, 18%), due to the presence of individual subregions that did not overlap with any subregion in the other eye.
We then analyzed subregion correspondence for individual neurons by calculating a correspondence index, determined for each neuron as the median of overlap indices of all subregions in both RFs. Using this metric, we can classify cells as exhibiting subregion correspondence, if majority of their subregions have positive sOI (correspondence index > 0) or anti-correspondence if majority have negative sOI (correspondence index < 0). Importantly, the vast majority of cells (n = 45/64, ∼70%) showed a positive correspondence index (Fig. 2D), whereas only a few (n = 12/64, 19%) showed negative values (nearly a 4:1 ratio). Moreover, the distribution of correspondence index was significantly skewed toward positive values when compared with a random distribution, with equal likelihood of correspondence and anti-correspondence (Fig. 2D, P < 0.001). Finally, because neither of the metrics used above took into account the strength of the responses, we also calculated a cross-correlation coefficient using the raw, unfiltered RFs. The positively skewed distribution (Fig. 2E, positives = 53/64, ∼83%) confirms similar sign and magnitude of the 2 RFs, that is, subregion correspondence.
Furthermore, we examined the relationship between subregion correspondence and binocular matching of RF orientation. For this, we binned the cells into 3 groups based on the difference in their RF orientations through the 2 eyes: matched (difference within 30°), intermediate (30 − 60°), and unmatched (60 − 90°). In the cells that had binocularly matched orientations (n = 51/64), the ratio of positive-to-negative correspondence index was nearly 5:1 (Fig. 2F), much greater than the 1:1 ratio seen in cells with unmatched orientations (n = 5/64; Fig. 2F). Together, our results indicate that the same-sign RF subregions through the 2 eyes show spatial overlap in most cells, likely giving rise to the binocularly matched orientation preferences in the cortex.
By also monitoring eye positions in a few experiments, we examined whether the observed subregion correspondence could be influenced by potential eye movements during the recording sessions. We did not observe any rapid eye movement in the 12 sessions (n = 13 cells, 4 mice), and any change in eye position was due to small and slow drifts. The change in eye positions (Fig. 2G, 1.09 ± 0.2 deg/session horizontally, 0.91 ± 0.28 deg/session vertically), over typical recording sessions (∼1 h), was very small compared with the typical RF (subregion area = ∼100 deg2). Importantly, the small changes were not accumulating, thus no systematic errors could have added up during the course of recording. Indeed, repeated stimulation of the same eye yielded RFs that were highly reproducible within recording durations (Fig. 2G) and showed high correlation between the raw (unfiltered) RFs (Fig. 2H, media n = 0.93). These analyses thus indicate that the small, slow change in eye position, if at all existent, does not affect the mapped RF structures appreciably.
Visual Deprivation Degrades Subregion Correspondence and Binocular Matching of RF Orientations
Visual experience activates the same spatial locations, simultaneously through the 2 eyes, providing coordinated binocular inputs that could in principle lead to subregion correspondence. Therefore, we studied the effect of visual deprivation on subregion correspondence and the consequent matching of RF orientations. We reared mice in complete darkness from P11 (DR11), shortly before eye opening in mice, until the time of recording at P31–36 (Fig. 3A). As shown previously (White et al. 2001; Wang, Sarnaik and Cang 2010; Rochefort et al. 2011), dark-rearing during this period of development had no effect on orientation selectivity development in mice (data not shown). Consistently, the RFs in these mice displayed alternating and oriented On and Off subregions (Fig. 3B), similar to those in the NR mice. Monocular RF geometry was largely similar between the DR11 (n = 74) and NR (n = 64) RFs, in terms of number of subregions (Fig. 3C–D, DR11: On, 1.30 ± 0.05/RF, Off, 1.40 ± 0.06/RF, P = 0.9; NR: On, 1.30 ± 0.05/RF, Off, 1.40 ± 0.06/RF, P = 0.7) and contralateral RF area (Fig. 3E, DR11: 372.7 ± 15.0 deg2; NR: 354.7 ± 15.2 deg2, P = 0.57), though the ipsilateral RFs were slightly larger in the DR11 mice (Fig. 3E, DR11: 331.0 ± 15.7 deg2; NR: 282.0 ± 13.2 deg2, P = 0.03). Importantly, dark rearing had no effect on the correlation between RF orientation and orientation preference to drifting gratings through either eye (Fig. 3F, P < 0.001 for both eyes, compared with a uniform distribution).
Despite the lack of visual experience, the spatial overlap between contralateral and ipsilateral RFs in the DR11 mice was similar to that in the NR mice (Fig. 4A, P = 0.38), allowing us to analyze subregion correspondence. Notably, visual deprivation caused a deficit in subregion correspondence (Fig. 4B), with fewer same-sign overlaps in the DR11 mice (positive sOIs: n = 222/394, ∼56%; compared with n = 218/347, ∼63% in NR). This degradation of subregion correspondence in DR11 mice was especially pronounced at high values of sOI (Fig. 4B,C), shifting the distribution leftward (Fig. 4C; P = 0.02 for the positive sOIs, only non-zero values plotted). Interestingly, although more negative sOIs were seen in the DR11 mice (104 of 394, ∼26% compared with 66 of 347, ∼19% in NR), their distributions were similar in the 2 groups (Fig. 4C; P = 0.55). Note that the decrease in sOI with dark-rearing is a robust phenomenon seen even without separation into positive and negative sOI and at different thresholds for filtering RFs (Fig. 4D; P = 0.007, mean ± 2SD, P = 0.04, mean ± 3SD, P = 0.002, 30% of peak). Importantly, this decrease was also evident without any filtering in the decreased correlation coefficients between the 2 RFs (Fig. 4E,F; P = 0.02).
Surprisingly, substantial subregion correspondence still remained in the visually deprived mice. There were still about twice as many subregions that had positive sOIs (∼56%) than negative ones in the DR11 mice (∼26%, Fig. 4D). This result was also evident from cell-based analysis using the correspondence index (Fig. 4G,H). Visual deprivation significantly decreased the degree of correspondence and shifted the distribution leftwards (NR vs. DR11, P = 0.02), but the distribution in the DR11 mice was clearly different from the random distribution (Fig. 4G,H; dotted line, P < 0.001).
To understand the effect of the degraded, yet non-random, subregion correspondence on RF orientations in the visually deprived mice, we analyzed sOI values as a function of RF orientation matching (Fig. 5A). In the NR mice, as expected from the analysis using correspondence index (Fig. 2F), the sOI values were overwhelmingly positive for the cells with binocularly matched RF orientation (Fig. 5B, median sOI = 0.36 ± 0.04, n = 347 subregions, P < 0.001 compared with random), whereas positive and negative values were equally possible for the unmatched cells (sOI = 0 ± 0.16, P = 0.04 compared with the matched group).
Remarkably, visual deprivation from P11 did not change sOI values in any of the 3 groups (Fig. 5B, n = 394 subregions). The cells with matched RF orientations in the DR11 mice still exhibited similar subregion correspondence as NR mice (Fig. 5B, sOI = 0.33 ± 0.06, P = 0.18 compared with NR; P < 0.001 vs. random), accounting for the nonrandom distribution of correspondence indices in these mice (Fig. 4H,I). Importantly, the degraded subregion correspondence in the DR11 mice (Fig. 4) was reflected in the much fewer number of binocularly matched cells in the DR11 mice (49% in DR11, vs. 80% in NR, Fig. 5B). Consequently, the RF orientations through the 2 eyes were greatly mismatched in the DR11 mice (Fig. 5C; P < 0.001 compared with NR mice), and the binocular matching of orientation tuning in these mice was also disrupted (Fig. 5D, P = 0.03), consistent with our previous report (Wang, Sarnaik and Cang 2010). Notably, however, the persistent subregion correspondence in these mice was also reflected in a slightly larger fraction of cells that had matched orientations than expected if RF orientations through the 2 eyes were completely independent (Fig. 5B, 49% in DR11, vs. 33% in random, P = 0.002).
Light-Independent Component of Subregion Correspondence
The apparent subregion correspondence in mice dark-reared from P11 could be due to light exposure before the onset of visual deprivation. This possibility was supported by a recent report that mouse visual cortical neurons responded to light flashes through closed eyelids at P10 (Rochefort et al. 2011). Although visual inputs before eye-opening are more diffuse, they are not completely without orientation information or spatial contrast and are capable of driving patterned vision through naturally closed eyelids (Krug et al. 2001). In addition, visual inputs before eye-opening could affect binocular development by modulating spontaneous retinal activity (Renna et al. 2011) with the potential to synchronize retinal waves in the 2 eyes and consequently influence binocular development such as subregion correspondence. To test this possibility, we initiated dark rearing before birth by placing pregnant mice in darkness 2–7 days before delivery and performed the same RF mapping experiments at P31–36 (Fig. 6A). The RFs in these mice had similar structures as those in mice reared with normal visual experience and in mice dark-reared from P11 (Fig. 6B; subregions/RF: On, 1.20 ± 0.06; Off, 1.40 ± 0.07; RF area: contra, 355.6 ± 19.6 deg2, and ipsi, 314.3 ± 17.7 deg2. All P > 0.05 compared with NR and DR11). The correlation between the RF orientation and tuned orientation in these mice was also normal (P < 0.001 for both eyes, compared with a uniform distribution), ruling out any role of visual experience in the development of monocular RFs in mice.
We next analyzed the binocular relationship of RF structure and subregion correspondence in the mice dark-reared from birth (DR0, n = 47 cells and 243 subregions) and compared with those dark-reared from P11 (DR11). First, there was no difference in overall RF overlap between the 2 groups of mice (Fig. 6C, P = 0.82). Second, there was no further decrease in subregion correspondence in the DR0 animals, either in the distribution (Fig. 6D, P = 0.92 for the positive arm and P = 0.63 for the negative arm) or in the percentage of positive sOIs (56% in DR11 and 56% in DR0; Fig. 6E). Even the degree of subregion correspondence in the cells with binocularly matched RF orientations was similar in the 2 dark-reared groups (Fig. 6F, sOI median = 0.35 ± 0.05 in DR0 and 0.33 ± 0.06 in DR11, P = 0.49), near that of the NR mice (median = 0.36 ± 0.04). Consistent with the lack of further degradation of subregion correspondence with earlier visual deprivation, the binocular matching of RF orientations in the DR0 mice was similarly disrupted as in the DR11 mice (Fig. 6G, P = 0.51, compared with DR11). In other words, the substantial subregion correspondence and RF orientation matching seen in the DR11 mice (Figs 4 and 5) was not due to light exposure before P11. Together, these results indicate that a light-independent mechanism, in addition to visual experience after P11, drives the binocular matching of RF orientations by aligning same-sign subregions in the 2 RFs.
Subregion Correspondence in RFs Mapped with Flashed Spots
To confirm and extend the above results that were obtained using STA in response to dense noise stimuli, we mapped RFs in a separate set of experiments using stimuli consisting of single, small bright squares flashed at different locations on a large grid (Liu, Li et al. 2009; Wang, Sarnaik, Rangarajan et al. 2010, also see Materials and Methods for details). The RFs were determined using the peri-stimulus histograms of the flash-evoked responses at different locations (Fig. 7A) and the responses to the onset and offset of the flash were separated to obtain the On and Off subregions, respectively. Compared with the STA, the spot stimuli were able to map the overlapped On–Off subregions within individual RFs (e.g., cell 11 in Fig. 7B), thus characterizing both simple and complex cells (36% complex cells, compared with 8% with STA, n = 75 cells). However, the flashed spots were less effective in mapping Off subregions than On subregions (24.3% Off and 75.7% On, n = 214 subregions), compared with the almost equal numbers of On and Off subregions mapped using noise stimuli (52% Off and 48% On, n = 347 subregions). As a result, most RFs did not display Off subregions in response to the flashed spots (subregions/RF, On = 1.08 ± 0.03, Off = 0.35 ± 0.04, e.g., cell 29 in Fig. 7B), consistent with a previous report of On-dominant RFs in mouse V1 using similar stimuli (Liu, Li et al. 2009). Furthermore, Off subregions were much smaller than On subregions (Off, 102.2 ± 13.8 deg2; On, 196.3 ± 14.7 deg2, P < 0.001), and the total RF areas (contra, 266.5 ± 23.1 deg2; and ipsi, 183.0 ± 20.3 deg2, P = 0.001) were smaller than those mapped with noise stimuli (contra, 354.7 ± 15.2 deg2, ipsi, 282.0 ± 13.2 deg2; P < 0.001 for both). Overall, single flashed spots seemed less effective in evoking responses from mouse V1 cells, but nevertheless they offered the opportunity to analyze RFs with overlap between On and Off subregions.
We also used the same spot stimuli to map RFs in mice dark-reared from P11 (DR11) to confirm the results of RF overlap and subregion correspondence obtained by the STA. First, dark rearing did not significantly alter the overall overlap between the contralateral and ipsilateral RFs (Fig. 7C; P = 0.08, NR, n = 75, DR11, n = 89 cells), just as observed with the STA. Furthermore, we quantified subregion correspondence using the sOI and found that it was degraded by visual deprivation, as seen from decrease in high positive values, that is, the leftward shift of the positive sOI distribution (Fig. 7D,E; NR vs. DR11, positive sOIs, P = 0.04, negative sOIs, P = 0.99). Importantly, in the cells that were binocularly matched in their preferred orientations, similar levels of subregion correspondence was observed in both normal and dark-reared mice (Fig. 7F; P = 0.67), thus confirming all the STA results (Figs 4 and 5).
In summary, we have demonstrated that both visual experience-dependent and independent mechanisms align same-sign subregions. Given the fact that RF structures determine orientation preference in simple cells (Hubel and Wiesel 1962; Dräger 1975; Movshon et al. 1978; Reid and Alonso 1995; Niell and Stryker 2008), our results suggest correspondence of subregions as a means of matching RF orientations and hence orientation preference during development (Wang, Sarnaik and Cang 2010). The binocular relationships demonstrated here provide valuable insights into the developmental mechanisms of binocular visual circuits.
Binocular convergence at the visual cortex serves important functions, such as depth perception, increasing acuity, fusing the views from the 2 eyes to eliminate blind spots, and increasing the field of view. Here we show that the converging thalamic inputs representing the 2 eyes are in register spatially and the same-sign subregions of the 2 monocular RFs are aligned, leading to binocularly matched orientation tuning. Given that neighboring cortical neurons display considerable scatters in their retinotopic locations (Bonin et al. 2011) and lack orientation columns in rodents (Ohki et al. 2005), these levels of precision in binocular convergence are remarkable and must require elaborate developmental processes to establish. We show that visual experience can improve the overlap of same-sign subregions of binocular RFs and give rise to similar preferences for orientations between the 2 eyes. On the other hand, some experience-independent processes can establish normal monocular RF structures as well as gross binocular overlap and, surprisingly, a substantial degree of subregion correspondence. Together, these results delineate the specific roles of experience-dependent and -independent components in the complex sequence of binocular development.
Subregion Correspondence and Visual Experience
How can subregion correspondence and the consequent matching of RF orientations be mediated by visual experience? An obvious candidate is Hebbian plasticity driven by the correlated inputs from the 2 eyes. Indeed, mathematical models based on such correlations in inputs from the lateral geniculate nucleus (LGN) have been proposed to explain the development of orientation selectivity, ocular dominance, and binocular matching in cats (Miller 1994; Erwin and Miller 1998). In these models, binocular matching of RF orientations could arise from different patterns of correlation between eye-specific LGN inputs. Correlated activity between the same-sign inputs that represent identical retinotopic locations would lead to subregion correspondence, whereas correlated activity between opposite-sign inputs would lead to anti-correspondence (Erwin and Miller 1998). Importantly, both scenarios would result in binocularly matched orientation preference. Experimentally, recordings in the developing ferret dLGN indeed revealed same-sign correlations (Ohshiro and Weliky 2006). Although the exact shape of the correlation was different from what was originally proposed, it could still produce appropriate cortical RFs in computer simulations under certain constraints (Ohshiro and Weliky 2006). Consistent with these predictions, our results indicate that subregion correspondence is the predominant form of organization and requires visual experience. With visual experience, corresponding regions in the 2 retinas perceive same contrast changes simultaneously, resulting in correlation between same-sign inputs to the cortex. Our results therefore provide the first direct experimental support for the correlation-based models.
Although our results rule out the possibility that correspondence and anti-correspondence were equally likely, the observed distribution of subregion overlap indices (Fig. 2) is wide compared with what would be predicted by the mathematical model of cat V1 (100% correspondence, i.e., all sOIs = 1). Because our study is the first to systematically quantify subregion correspondence in any species, it is unclear whether the overlap distribution is narrower in cats and monkeys. On the basis of the available data of phase and positional disparity distribution in cats (Anzai et al. 1999), a similarly wide distribution of subregion overlap may indeed be the case. If so, a revision of the model would be needed to be more biologically realistic.
Experience-Independent Binocular Development
In the absence of visual experience, segregation of On and Off subregions within monocular RFs and gross overlap between the 2 RFs is completely unaffected. These could be established through Hebbian processes driven by spontaneous activity. Waves of action potentials propagate across ganglion cells in the retina before vision onset (Galli and Maffei 1988; Meister et al. 1991; Wong et al. 1993). Structured spontaneous activities are also seen in the dLGN (Weliky and Katz 1999) and visual cortex (Chiu and Weliky 2001; Hanganu et al. 2006), and their patterns are modulated by the retinal waves (Hanganu et al. 2006). According to the correlation-based model described above (Miller 1994), On–Off subregion segregation within a monocular RF could arise from anti-correlation between On and Off inputs of the same eye. Such anti-correlation is obvious during normal visual experience, but could also exist in spontaneous activity, as shown in the asynchronous firing of On and Off RGCs during retinal waves in mice (Kerschensteiner and Wong 2008). Our results of normal monocular RF structure in visually deprived mice thus highlight the potential role of structured spontaneous activity in the development of orientation selectivity. At the same time, however, it is still surprising that visual deprivation did not affect monocular cortical RFs despite its known impact on retinal development (Tian and Copenhagen 2001, 2003; Giovannelli et al. 2008; Di Marco et al. 2009; Burnat et al. 2012). In particular, many of the bistratified On–Off RGCs in mice are converted into monostratified On or Off cells after eye-opening and this process is disrupted by visual deprivation (Tian and Copenhagen 2001, 2003; Liu et al. 2007; Liu, Robinson et al. 2009). Given that most dLGN cells are of single On/Off sign (Grubb and Thompson 2003; Marshel et al. 2012), the potential impact of increased On–Off RGCs on cortical development might be limited.
It is even less intuitive how even the partial degree of subregion correspondence could be attained in the absence of obvious common inputs to the 2 eyes (i.e., visual experience). In order for the spontaneous activity to drive subregion correspondence, the same-sign thalamocortical inputs that represent similar retinotopic locations through the 2 eyes must be correlated. Amazingly, certain degrees of binocular correlation may indeed exist in mice even at the level of retinal waves (Ackman et al. 2012), possibly via direct retino-retinal connections (Muller and Hollander 1988; Tóth and Strznicky 1989; Thanos 1999; Ackman et al. 2012) or retinopetal modulations (Gastinger et al. 2006). Such binocular correlation in spontaneous retinal activity may be further enhanced at the LGN by cortical feedback projections (Weliky and Katz 1999), thus leading to subregion correspondence. In kittens, binocularly matched orientation maps appear before visual experience (Crair et al. 1998) and can develop in the absence of coordinated binocular activity (Godecke and Bonhoeffer 1996). Thus the experience-independent mechanisms discussed above may be shared between different species. Importantly, at the level of individual cells, the experience-independent processes alone cannot establish normal levels of subregion correspondence, as we have shown in the DR mice. During normal development, subregion correspondence is increased with visual experience, which leads to further alignment of RF orientations and binocular matching of orientation tuning.
We have also found that the retinotopic positions of the 2 monocular RFs are binocularly aligned independent of visual experience. This is consistent with the fact that cortical retinotopy reaches adult levels before eye opening (Cang, Renterìa et al. 2005). Graded expression of molecular cues, such as ephrin-As and their receptor EphAs, guide retinotopic map formation throughout central visual areas (Huberman et al. 2008), including thalamocortical projections (Cang, Kaneko et al. 2005). Since these guidance cues appear to be expressed in similar gradients across eye-specific layers in the LGN (Cang, Kaneko et al. 2005), they could align the maps binocularly. The thalamocortical maps established by guidance cues are further refined by activity-dependent processes driven by retinal waves (Cang, Renterìa et al. 2005). Consequently, if retinal waves are indeed correlated between the 2 eyes, the binocular alignment of retinotopic positions could be further refined by the structured spontaneous activity.
Binocular Relationship of Monocular RFs and Disparity Tuning
The 2 RFs of most cortical neurons only overlap partially in both normal and visually deprived mice. Such spatial offset between the positions of the 2 monocular RFs are also observed in cats and monkeys, and it makes those cells respond maximally when the object falls at noncorresponding points in the 2 retinas, thus resulting in selectivity for disparity (Hubel and Wiesel 1970b; Nelson et al. 1977; Poggio and Fischer 1977; Ferster 1981). In addition to positional difference, differences in subregion arrangement within the RFs, the phase difference, can give rise to disparity tuning in cats (DeAngelis et al. 1991; Ohzawa et al. 1996). Both position and phase differences can co-exist in the 2 RFs (Anzai et al. 1997, 1999), and existing disparity tuning data in cats suggest that they obey subregion correspondence (Erwin and Miller 1999). Our study shows that such a spatial organization is also observed in mice. Furthermore, we have revealed, for the first time, the developmental factors that give rise to these binocular relationships. Whether such binocular organization also gives rise to disparity tuning and further to depth perception in mice is an interesting question that merits future investigations.
Implications for Neural Development
Our understanding of the development of binocular cortical circuits is largely derived from studies of ocular dominance plasticity, where monocular deprivation during a critical period weakens the inputs of the deprived eye and strengthens those of the open eye. Ocular dominance plasticity is widely studied as an experimental paradigm to understand the formation of thalamocortical circuits (Hubel and Wiesel 1970a; Hubel et al. 1977; Crair et al. 1998) and to probe sensitive periods of cortical plasticity (Hubel and Wiesel 1970a; Hensch 2005). Our current study shifts the focus from this competition-based plasticity induced by visual manipulations to correlation-based refinement of binocular convergence in normal development. In fact, simple competition between thalamocortical inputs of the 2 eyes cannot explain how the ipsilateral inputs that mature later are able to compete with the already strong contralateral inputs to a binocular cell (Crair et al. 1998). Instead, correlated activity between the 2 inputs could not only strengthen the weak inputs onto particular binocular neurons via Hebbian plasticity, but also drive the matching of RF structure and orientation tuning, as we observe here.
In conclusion, our experiments indicate that experience-independent mechanisms are sufficient in arranging inputs from the same eye to establish normal monocular RFs. Their binocular relationship requires visual experience to reach normal level of precision, but can be achieved, to certain extent, with experience-independent mechanisms, making these circuits more resilient to dysfunction. Such interplay between experience-independent and -dependent mechanisms may also underlie the development of higher brain functions, where convergence of multiple streams of information is needed, such as multisensory integration, language, and social interactions.
This work was supported by US National Institutes of Health (NIH) grants (EY020950 and EY018621), a Sloan Research Fellowship, a Klingenstein Fellowship Award in Neurosciences, and a Brain Research Foundation Seed Grant to J.C.
We thank Dr. Cris Niell for help with data analysis, Dr. Lupeng Wang for help with eye tracking experiments and members of the Cang laboratory for helpful discussions and comments. Conflict of Interest: None declared.