Assessing the kaleidoscope of monocular deprivation effects

Abstract

Short-term monocular deprivation (∼150 min) temporarily shifts sensory eye balance in favor of the deprived eye (Lunghi, Burr, & Morrone, 2011; Zhou, Clavagnier, & Hess, 2013), opposite to classic deprivation studies (Hubel & Wiesel, 1970). Various types of deprivation—light-tight, diffuser lenses, image degradation—have been tested, and it seemed that a deprivation of contrast was necessary, and sufficient, for these shifts. This could be accommodated in a feedforward model of binocular combination (Meese, Georgeson, & Baker, 2006; Sperling & Ding, 2010), in which the shift reflects a (persistent) reweighting induced by an interocular gain control mechanism tasked with maintaining binocular balance (Zhou, Clavagnier, et al., 2013). Here, we used a novel “kaleidoscopic” monocular deprivation that, although it rendered images fractionated and uninformative, preserved gross luminance, color, spatial frequency, motion, and contrast information, effectively sneaking the image degradation past early, feedforward mechanisms, targeting higher levels. Kaleidoscopic deprivation produced effects indistinguishable from traditional light-tight patching. This rules out contrast imbalance as the sole factor driving these shifts in sensory eye balance. In addition, since the suppression of the kaleidoscopic image likely requires feedback from higher-level processes capable of determining the behavioral relevance of an eye's information (Foley & Miyanshi, 1969; Jiang, Costello, & He, 2007; Kovács, Papathomas, Yang, & Fehér, 1996; Wolf & Hochstein, 2011), feedforward-only models may need to be elaborated.

Introduction

Ocular dominance plasticity following short-term monocular deprivation in adult humans

Lunghi et al. (2011) first observed that short-term (150-min) monocular deprivation with a translucent patch affected the dynamics of binocular rivalry, resulting in the previously deprived eye prevailing twice as often as the nondeprived eye. Zhou, Clavagnier, et al. (2013) extended this finding by showing that the deprived eye has greater influence in dichoptic phase combination, global motion coherence (GMC), and contrast matching tasks. As well, they observed comparable effects with both light-tight and diffuser-lens deprivation (a ground glass that eliminates contrast but preserves overall illumination). In an effort to evaluate the determinants of this new form of short-term plasticity, Zhou, Reynaud, and Hess (2014) studied the consequence of other types of monocular deprivation such as band-pass filtering, parametric reductions in contrast, and contrast-preserving phase scrambling. They concluded that deprivation in overall monocular contrast was not necessary to induce short-term deprivation effects, because the removal of just high-frequency contrast information was sufficient (i.e., low-pass filtering triggered deprivation effects, while high-pass did not). In addition, because phase scrambling did not induce deprivation effects, it seemed that a reduction in the amplitude of the high-frequency components was not just sufficient but also necessary to induce shifts in eye balance. Taken together, it seemed that feedforward, contrast energy–based models of deprivation sufficed to explain these effects. Under this idea, deprivation triggers interocular contrast gain control (Moradi & Heeger, 2009; Shapley & Enroth-Cugell, 1984; Sperling & Ding, 2010), whereby gain is increased in the deprived eye and decreased in the nondeprived eye (Zhou, Clavagnier, et al., 2013) in an assumedly compensatory, homeostatic attempt (Fox & Stryker, 2017; Turrigiano & Nelson, 2000) to restore interocular balance. Upon resumption of normal binocular vision, observed deprivation effects, then, reflect the persistence of this reweighting, giving the deprived eye greater inhibitory influence over the contralateral eye and thereby greater representation in binocular/dichoptic tasks and percepts, for up to 30 min (Zhou, Clavagnier, et al., 2013). 

One challenge to this story, though, was Zhou et al.'s (2014) failure to find any effect of deprivation with a 40% contrast reduction of the deprived eye. A contrast gain control system, purportedly responsible for maintaining balance between the eyes under normal viewing conditions, failing to respond to such a large reduction in the contrast of one eye would be of limited utility. Further complicating this picture were recent findings of Bai, Dong, He, and Bao (2017). They were able to find deprivation effects with a pink-noise deprivation (in which the power spectrum is preserved and phase information is replaced with that of white noise), even though this manipulation nominally preserves high spatial frequency power. However, this effect was found only on the dynamics of binocular rivalry, extending the results of Lunghi et al. (2011), but did not influence dichoptic phase combination, consistent with the phase-scrambling results of Zhou et al. (2014).¹ A summary of these results is provided in Table 1. 

Table 1

 

Overview of previous work on short-term monocular deprivation effects. Notes: Rows show the deprivation method (ordered by the root mean square [RMS] contrast of the deprived eye), and columns show the psychophysical test used to assess sensory eye balance, pre- and postdeprivation. Cells contain the relevant citation(s), including the present work. Cells with bold text indicate negative results, that is, no effect of deprivation. Empty cells indicate a heretofore unstudied combination of deprivation and test.

 

Novel kaleidoscopic deprivation

Given these challenges to the feedforward, contrast-deprivation triggered account of short-term monocular deprivation, we hypothesized that feedback from higher-level areas, tuned to image utility, may play a role. To test this, we used a monocular, kaleidoscopic deprivation that obliterated useful ecological information in the eye's image while preserving gross low-level visual properties such as luminance, color, spatial frequency, motion, and contrast (root mean square [RMS] contrast was >90% that of the original image). Figure 1 shows the view of a scene as seen through the kaleidoscopic lens. In our procedure, while wearing the kaleidoscopic lens on the deprived eye, the nondeprived eye was left open (Figure 1c), and participants were asked to engage in normal, active vision outside the laboratory for the 150-min deprivation period. This means that although the conflict (lack of synchronicity; Zhou et al., 2014) between the images may be apparent to early visual areas, the determination of which eye is degraded is indeterminate. However, the degradation of the kaleidoscopic image, and consequently the trigger to alter interocular balance, should be clear to any higher area sensitive to image utility for active vision. Thus, in the conflict between the two eyes' images during deprivation, the eye that allows for useful interaction with the environment (i.e., the open eye) is prioritized, suppressing the less-useful one (the kaleidoscopic patched). In the discussion, we speculate that it is this relative suppression, not the particulars of the image degradation per se, that triggers the shift in sensory eye balance (unfortunately, none of the deprivation studies we reviewed addressed the effects of interocular competition during the deprivation period as a potential trigger to deprivation). 

Figure 1

 

Illustration of the effect of kaleidoscopic deprivation (a) on a normal scene taken from the environment around the laboratory (b). Note that contrast is not significantly affected, and gross luminance, color, motion, and spatial frequency information are preserved. Panel (c) shows an observer wearing the spectacles fitted with the kaleidoscopic lens. Observers engaged in normal, active vision during the 150-min deprivation period.

Methods

Participants

Seven participants between 20 and 35 years of age with normal or corrected-to-normal vision participated in the study. To participate, observers had to have visual acuity ≥20/25, stereoacuity ≥30′ of arc, and normal color vision.² In addition, participants performed two pretest sessions of the dichoptic GMC task and the binocular rivalry task as practice and screening. Participants with GMC thresholds more than 1.5 SD above the group mean or 2 SD above their own thresholds from the two pretests were excluded from the study (4 potential participants were thus excluded based on the former criteria, and 1 participant was excluded based on the later). Lastly, participants performed a sighting test³ to determine their dominant (“preferred”) eye. Participants were naive to the purpose of this study. All observers gave informed consent prior to participation. The study was approved by the Institutional Review Board of University of Massachusetts Boston. 

Apparatus

Ocular balance tasks were programmed using custom scripts in MATLAB (version R2015a; MathWorks, Natick, MA), using Psychophysics Toolbox functions (Brainard, 1997; Kleiner, Brainard, Pelli, Ingling, & Murray, 2007; Pelli, 1997). GMC and rivalry stimuli were presented on a calibrated three-dimensional (3D) ASUS monitor with a resolution of 1,080 × 1,024. Dichoptic stimuli were presented using NVIDIA 3D Vision 2 LCD shutter goggles synchronized to the monitor at 120 Hz with interleaved frames presented to each eye at 60 Hz. Unless otherwise specified, observers were seated 57 cm from the display in a quiet, dark room. Monocular quick contrast sensitivity function (qCSF) measurements were made on a PF790 CRT monitor (resolution: 1,024 × 768 at 75 Hz) that had been linearized and fitted with a VideoSwitcher (X. Li & Lu, 2012) to expand gray-scale bit depth from 8 to 16 bits. The kaleidoscopic lens (“Future Eyes Kaleidoscope Glasses and Crystal Necklace Monocles,” 2017) was fitted into one of the openings in a pair of spectacles, with the other opening left empty (Figure 1c). 

Tests of sensory eye balance

Dichoptic GMC task

Global motion processing has been known to involve two processing stages: an initial local motion detection stage that is contrast dependent and a later motion integration stage that is contrast invariant (Hess, Hutchinson, Ledgeway, & Mansouri, 2007). Evaluating dichoptic GMC thresholds offers insights into the effect of deprivation on dichoptic interactions. Global motion stimuli were composed of a field of dots, undergoing translational movement, presented within a visible rectangular aperture to aid binocular fusion. The aperture subtended 8° of visual angle from a viewing distance of 57 cm. Dots had a Weber contrast of 64% presented on a uniform, gray background (with an average luminance of ∼12.5 cd/m², as measured through the shutter goggles). The dot field consisted of 100 dots (each subtending 0.1°), some of which were designated as “signal” dots and had coherent, infinite lifetime translational motion. The remaining dots were designated as “noise” dots and each had a consistent, but random angle, motion vector. All dots drifted at 6°/s and were redrawn at a random location within the test window if they reached the aperture border. Dots could be presented dichoptically, with signal dots in one eye and noise in the other, or monocularly, with both signal and noise dots in one eye and empty background in the other (Figure 2a and 2b, respectively). GMC thresholds were measured using a single-interval, direction discrimination task controlled by an adaptive PSI procedure (Prins, 2012). On each trial, observers were asked to discriminate the global direction of the dot field (left or right), and their response was recorded by a key press. Threshold values, then, reflect the proportion of signal dots that are required for at-threshold direction discrimination for a particular observer and condition. The two dichoptic and the two monocular conditions were mixed within blocks. There were 180 trials for each condition within one session. Each block took approximately 10 min to complete. 

Figure 2

 

Schematic of global motion coherence (GMC) tests. Sessions were blocked by condition (dichoptic or monocular). Each block had 180 trials and used an adaptive procedure to determine coherence thresholds (i.e., the percentage of signal dots required to achieve 75% correct performance). Panel (a) shows an example (not drawn to scale) of a dichoptic GMC test, here with signal dots presented to the left eye and noise to the right eye (images were interleaved temporally via LCD shutter goggles). Which eye received signal or noise was randomized, trial to trial. Panel (b) shows an example of a monocular GMC test, with signal+noise presented to one eye (in this example, the left) and the other eye (here, the right) shown the empty background. In both dichopic and monocular tests, the binocular percept was identical, a field of randomly moving dots with a small subset field of dots moving, near threshold, to the left or right (randomized from trial to trial). The observer was instructed to indicate the direction of this motion with a keypress.

Binocular rivalry task

Interocular rivalry is a well-established probe of neural competition (Tong, Meng, & Blake, 2006). In our tests, two orthogonal sine wave gratings with a spatial frequency of five cycles per degree (cpd), orientation of ±45,° and Michelson contrast of 30% were dichoptically presented within an alignment ring to facilitate binocular fusion (Klink, Brascamp, Blake, & van Wezel, 2010). Stimuli subtended 1.5° × 1.5° of visual angle and were presented centrally on a uniform, gray background (with an average luminance of ∼12.5 cd/m², as measured through the shutter goggles) for 120 s. Observers continuously reported their perceptual experience by pressing either the left or right arrow keys to indicate an exclusive percept of the left or right tilted monocular stimulus, respectively. Observers were asked to hold the corresponding key down to indicate the duration of the unitary percepts. Both keys were released when mixed percepts (plaid fusion; patchwork) were seen (see Figure 5a). 

Monocular contrast sensitivity functions

Because short-term deprivation effects last for only about 20 to 30 min postpatching (Lunghi, Emir, Morrone, & Bridge, 2015; Zhou, Clavagnier, et al., 2013; Zhou et al., 2014) our estimate of the contrast sensitivity function had to be brief. We used the qCSF method, a procedure for rapid estimation of spatial CSFs that combines Bayesian adaptive inference with a trial-to-trial information gain strategy (Lesmes, Lu, Baek, & Albright, 2010). The stimuli we used were adapted from Lesmes et al. (2010). Gabor stimuli (Gaussian-windowed sinusoidal gratings) appeared with spatial frequencies ranging from 0.25 to 36 cpd and were rendered on a 400- × 400-pixel grid, subtending 5.6° × 5.6° of visual angle, from a viewing distance of 128 cm. The Gabor patches were either oriented 45° to the left or right, and observers were asked to perform an orientation discrimination task. Gratings were presented for 130 ms, and observers judged the orientation of the grating as tilted right or left by pressing the corresponding arrow keys. A block consisted of 200 trials. The CSF can be captured with four parameters: peak contrast sensitivity (“gain”), bandwidth (full width at half height), truncation (low-frequency plateau), and peak frequency (the spatial frequency at which there was maximum sensitivity, capturing the left-right position of the CSF). As a broad sensitivity metric, however, one can instead estimate the area under the log CSF (AULCSF), which, all else being equal, changes when gain, bandwidth, and/or truncation changes (Lesmes et al., 2010). We chose this, along with the gain, as our main measure of pre- versus postdeprivation effects on contrast sensitivity. Then, to capture (potential) left-right shifts in the CSF, we looked for pre- versus postdeprivation changes in peak frequency. 

Experimental procedure

After prestudy screening (see above), all participants were scheduled for four visits. On each visit, observers were assigned a type of eye patch (light-tight or kaleidoscopic) and the eye to be patched (preferred or nonpreferred). The order was randomized across participants. The main experiment had three phases: (1) predeprivation measures of sensory eye balance (GMC thresholds and binocular rivalry tests), (2) 150 min of deprivation (during which the observer was asked to engage in everyday activities around the lab and building), and (3) postdeprivation measures of sensory eye balance (identical to the set of predeprivation measures). In a second phase of testing, participants were run to assess contrast sensitivity functions of the deprived and nondeprived eyes, monocularly. Those observers were scheduled for two visits, for light-tight patch and kaleidoscopic deprivation, on his or her preferred eye. An overview of the schedule is provided in Table 2. 

Table 2

 

Schedule of conditions.

 

Results

Dichoptic GMC

Here, results reflect the proportion of signal dots required for threshold motion direction discrimination. Predeprivation, we determined dichoptic GMC thresholds both when signal dots were presented in the to-be-deprived eye (and noise in the nondeprived eye) and also when signal dots were presented to the nondeprived eye (and noise to the deprived eye; Figure 2a). Postdeprivation, we repeated these measurements. This allowed us to note changes by taking the difference between thresholds in an eye, predeprivation minus postdeprivation. Presented this way, one can determine whether there are sensitivity changes in one, or both, eyes in the native units of percent coherence. Postdeprivation, we expected lower thresholds when signal was presented to the (assumedly strengthened) deprived eye, as compared with predeprivation, yielding a positive pre-post difference. We expected the opposite pattern when signal was presented to the (assumedly weakened) nondeprived eye, yielding a negative pre-post difference. We performed a repeated-measures analysis of variance (ANOVA) with (pre-post) threshold differences as the dependent variable. There were three, two-level factors (relevant to the eye receiving signal during GMC threshold measurements): deprivation type (whether that eye had undergone light-tight or kaleidoscopic deprivation), eye preference (whether that eye was the preferred or nonpreferred eye), and deprivation state (whether that eye was the deprived eye). As expected, we found a significant main effect of deprivation state, F(1, 6) = 25.568, p = 0.002; ηp² = 0.81, but no effect of deprivation type, F(1, 6) = 1.591, p = 0.254, nor eye preference, F(1, 6) = 0.001, p = 0.977. In short, there was an increase in sensitivity (“strengthening”) in the deprived eye and a concomitant decrease in sensitivity (“weakening”) in the nondeprived eye, and the magnitude of this effect was indistinguishable for light-tight and kaleidoscopic deprivation (Figure 3a,b). 

Figure 3

 

Change in dichoptic global motion coherence (GMC) thresholds as a result of light-tight and kaleidoscopic monocular deprivation. Panels (a) and (b) show changes in thresholds (pre- minus postdeprivation) as a result of light-tight and kaleidoscopic deprivation, respectively (collapsed over eye preference). In both panels, red, filled symbols show threshold changes when signal dots are presented to the deprived eye (and noise to the nondeprived), and open symbols show threshold changes when signal dots are presented to the nondeprived eye (and noise dots to the deprived). Values are shown for each observer (s1–s7) as well as an overall average. Panel (c) shows dichoptic GMC threshold ratios, calculated by taking the ratio of thresholds when the signal is presented to the deprived eye versus the nondeprived eye, pre- and postdeprivation (results are collapsed across eye preference and observer). Values close to 1 thus indicate balance between the eyes, whereas lower values indicate a strengthening of the deprived eye (and/or a weakening of the nondeprived eye). Panel (d) shows a normalization of these ratios, dividing the postdeprivation ratio by the predeprivation ratio, and is included to facilitate comparison with previous studies (normalized coherence ratios <1 indicate an increase in sensitivity in the deprived eye and/or a decrease in sensitivity in the nondeprived). Error bars indicate SEM.

To facilitate comparison with previous studies, we also calculated threshold ratios by taking the ratio of dichoptic GMC thresholds of the two eyes, predeprivation (i.e., signal in to-be-deprived eye/signal in to-be-nondeprived eye) and postdeprivation (signal in deprived eye/signal in nondeprived eye). This way, deprivation state is collapsed, but deprivation period (i.e., pre- vs. postdeprivation) becomes a factor. (And, because eye preference was not a significant factor in the earlier analysis, this factor was collapsed.) We then performed a 2 × 2 repeated-measures ANOVA on these ratios, with factors deprivation type and deprivation period. As expected, the analysis showed a significant main effect of deprivation period, F(1, 6) = 59.919, p = 0.0002; ηp² = 0.909, but no significant effect of deprivation type, F(1, 6) = 0.001, p = 0.975. Prior to deprivation, threshold ratios were close to 1, indicating relative balance between the eyes. Postdeprivation, however, the coherence threshold ratios fell significantly, to 0.64 (SD = 0.19) and 0.58 (SD = 0.14) for light-tight and kaleidoscopic deprivation, respectively, consistent with the strengthening of the deprived eye and weakening of the nondeprived eye (Figure 3c). We then normalized these coherence threshold ratios, simply dividing the postdeprivation ratio by the respective predeprivation ratio (Figure 3d). The resulting normalized values of 0.66 (SD = 0.29) and 0.53 (SD = 0.09) for light-tight and kaleidoscopic deprivation, respectively, were comparable to previously reported normalized ratios of approximately 0.5 from Zhou, Clavagnier, et al. (2013).⁴ 

Monocular GMC

We measured monocular GMC thresholds (both signal and noise dots presented to the same eye during the test, with the other eye left unstimulated) pre- and postdeprivation, in the deprived and the nondeprived eye (Figure 2b). It has been established that at suprathreshold contrast levels, overall stimulus contrast does not have an effect on monocular GMC thresholds; performance is dependent only on coherence for contrasts higher than approximately 10% (Mansouri et al., 2008). Thus, even if deprivation manifests as a change in contrast gain in the deprived eye (and/or the nondeprived eye), there should be no change in monocular GMC thresholds. Just as with dichoptic GMC thresholds, we computed the difference between pre- and postdeprivation thresholds as our dependent variable (Figure 4). A 3 × 2 repeated-measures ANOVA with eye preference (preferred or nonpreferred), deprivation type (light-tight or kaleidoscopic), and deprivation state (deprived or nondeprived) as factors showed no effects of deprivation state, F(1, 6) = 3.397, p = 0.115; deprivation type, F(1, 6) = 0.992, p = 0.358; or eye preference, F(1, 6) = 0.009, p = 0.926. In short, as expected, deprivation had no measurable effect on monocular GMC thresholds. 

Figure 4

 

Monocular global motion coherence (GMC) thresholds following light-tight and kaleidoscopic deprivation. Panels (a) and (b) show differences in thresholds (pre- minus postdeprivation) as a result of light-tight and kaleidoscopic deprivation, respectively (collapsed over eye preference). In both panels, red, filled symbols show differences when stimuli (both signal and noise dots) were presented to the deprived eye (with the nondeprived eye being left unstimulated), and open symbols show threshold changes when stimuli were presented to the nondeprived eye. Values are shown for each observer (s1–s7) as well as an overall mean. Error bars indicate SEM.

Figure 5

 

Schematic and results of binocular rivalry tests. Panel (a) shows an example of a binocular rivalry test, here with the left-tilted windowed sinusoid presented to the left eye and the right-tilted to the right eye. A schematic is shown to represent a potential time course of binocular percepts. The observer was instructed to hold down a key to indicate which grating (left- or right-tilted) prevailed, that is, was the exclusive percept at any given time (the absence of a keypress indicated that there was a mixed percept) for a period of 120 s. The prevalence duration for an eye is the sum of these periods. Panel (b) shows rivalry prevalence duration ratios following light-tight and kaleidoscopic monocular deprivation (collapsed over eye preference and averaged over observers). Ratios were computed by taking the total duration of an eye's prevalence (i.e., the sum of all phases of dominance), postdeprivation, divided by the duration of its prevalence, predeprivation. Values, then, greater than 1.0 indicate an increase in prevalence postdeprivation of the indicated eye (deprived or nondeprived) and values less than 1.0, a decrease. Error bars indicate SEM. Individual subject data are presented in Supplementary Figure S1.

Binocular rivalry

To estimate the extent of each eye's dominance during rivalry, pre- versus postdeprivation, we determined the total accumulated time⁵ that each eye's image prevailed during the first 30 s of each trial, a period matching previous work, and most likely to reveal a deprivation effect (Kim, Kim, & Blake, 2017; Lunghi et al., 2011). To facilitate comparison with previous work, collapsing over eye preference, we calculated prevalence ratios, dividing the postdeprivation prevalence duration by the predeprivation duration, for the deprived, and the nondeprived, eye. This ratio was calculated for each type of deprivation. A two-way repeated-measures ANOVA on the ratios showed no effect of deprivation type, F(1, 6) = 0.051, p = 0.83, but a significant effect of deprivation state, F(1, 6) = 8.88, p = 0.025, ηp² = 0.597, indicating that the prevalence of the deprived eye was increased (to a similar extent for either deprivation type) as a result of deprivation. The average prevalence ratios of 1.21 (SD = 0.45, light-tight) and 1.19 (SD = 0.49, kaleidoscopic) for the deprived eye (Figure 5b, filled, red symbols) and 0.87 (SD = 0.37; light-tight) and 0.96 (SD = 0.16; kaleidoscopic) for the nondeprived eye (Figure 5b, open, blue symbols) are comparable to Lunghi et al.'s (2011) values of 1.27 for the deprived eye and 0.9 for the nondeprived eye (following light-tight deprivation).⁶ The rates of mixed percepts with similar tests have been found to be low (Lunghi et al., 2011) to moderate (∼20%; Bai et al., 2017). Our rates were higher, at approximately 50%. (We do not have a satisfactory answer for why mixed percept rates might vary so widely, especially given the similarities between the test stimuli used across these studies. It is true, though, that judgments of mixed versus exclusive percepts are subjective and therefore subject to criterion changes based on instruction and/or response method.) Mixed percept rates did not show any differences post- versus predeprivation nor between light-tight and kaleidoscopic patching (rates were 52% [CI ±5.3%] and 48% [9.3%] for light-tight patching, pre- and postdeprivation, respectively, and 52% [7.9%] and 51% [6%] for kaleidoscopic patching, pre- and postdeprivation, respectively). 

Monocular contrast sensitivity

We measured the CSF (Lesmes et al., 2010), for each eye, pre- and postdeprivation, for each participant. We then calculated the ratio of post to pre values for three parameters: peak contrast sensitivity, peak spatial frequency, and AULCSF. These ratios were examined in a 2 × 2 repeated-measures ANOVA with deprivation type (light-tight and kaleidoscopic) and deprivation state (deprived or nondeprived) as independent variables. For peak contrast sensitivity, there was a significant main effect of deprivation state, F(1, 5) = 16.672, p = 0.01, ηp² = 0.769, but no effect of deprivation type, F(1, 5) = 0.208, p = 0.667, and no interaction, F(1, 5) = 0.125, p = 0.738 (Figure 6). For AULCSF, there was a significant main effect of deprivation state, F(1, 5) = 10.422, p = 0.023, ηp² = 0.68, but no main effect of deprivation type, F(1, 5) = 0.379, p = 0.565, and no interaction, F(1, 5) = 0.004, p = 0.951. Taken together, these results indicate that light-tight and kaleidoscopic deprivation results in a relative increase of contrast sensitivity in the deprived eye. For peak spatial frequency, there was no significant effect of deprivation state, F(1, 5) = 3.493, p = 0.121, or deprivation type, F(1, 5) = 0.038, p = 0.853, nor an interaction, F(1, 5) = 0.05, p = 0.832. 

Figure 6

 

Parameters from monocular contrast sensitivity functions (CSF). CSFs were determined pre- and postdeprivation, after light-tight and kaleidoscopic deprivation, for the deprived eye (red symbols) and nondeprived eye (blue symbols). Panel (a) shows the post/pre ratios for peak contrast sensitivity, averaged over observers. Values greater than 1 indicate a strengthening of an eye (i.e., greater sensitivity, postdeprivation), whereas values lower than 1 indicate a weakening. Panel (b) shows the post/pre ratios for area under the log contrast sensitivity function (AULCSF), averaged over observers. Again, values greater than 1 indicate a strengthening of an eye (i.e., greater sensitivity, postdeprivation), whereas values lower than 1 indicate a weakening. Panel (c) shows the post/pre ratios for peak frequency, averaged over observers. Here, values greater than 1 indicate a shift of spatial frequency tuning toward higher frequencies, whereas values less than 1 indicate a shift toward lower frequencies. Error bars indicate SEM. Individual subject data on AULCSF and peak contrast sensitivity are presented in Supplementary Figures S2 and S3.

Discussion

We showed that 150 min of monocular deprivation using a kaleidoscopic patch, which fractionated an eye's image but preserved gross luminance, color, spatial frequency, motion, and contrast information, biased sensory eye balance in favor of the deprived eye to the same extent as a conventional light-tight patch. This effect was observed in dichoptic GMC tests, rivalry tests, and measures of monocular CSFs.⁷ Consistent with previous results (Zhou, Clavagnier, et al., 2013), we observed deprivation effects in both the deprived eye and in the open, nondeprived eye, supporting the interocular, push-pull nature of the effect. 

The present results are challenging to account for with current feedforward models of monocular deprivation effects. There, weight is shifted toward the eye with the degraded image, as determined by low-level visual mechanisms triggered by the attenuation of contrast in one eye as induced, for instance, by light-tight or diffuser glass patching (Lunghi et al., 2011; Zhou, Clavagnier, et al., 2013), particularly in high spatial frequencies (Zhou et al., 2014). However, kaleidoscopic deprivation sneaks its degradation past these low-level mechanisms; there is no image-quality basis on which to disfavor the kaleidoscopic eye. However, there is an image-utility basis. Here, we speculate that higher-level visual areas, sensitive to the relative usefulness of each eye's image during active vision, alter the duty cycle of rivalry (Jiang et al., 2007; Kovács et al., 1996; Tong et al., 2006) during deprivation, actively suppressing the uninformative kaleidoscopic eye. This suppression, then, mimics the suppression induced by the light-tight patch and similarly induces a shift in interocular gain control weights. As with traditional patching, then, a lingering of these weights temporarily biases sensory eye balance when binocular vision is restored. This explanation finds support in recent models of rivalry in which perceptual competition is modulated by top-down (attentional) influences, which invoke contrast gain control mechanisms to drive the visibility of an eye's image and are subject to adaptation (Kang & Blake, 2010; H.-H. Li, Rankin, Rinzel, Carrasco, & Heeger, 2017; Platonov & Goossens, 2013). 

Can changes in contrast gain account for deprivation effects?

Because the sensory balance shifts we observed were induced by a kaleidoscopic deprivation that did not affect contrast, we know that the trigger to a shift in sensory eye balance need not be based on deprivation of contrast. However, it may still be that the effect of deprivation manifests as a change in interocular contrast gain. Given that, one can ask whether the deprivation-induced changes in interocular contrast that have been reported are sufficient to produce observed changes in, for instance, GMC thresholds. Put another way, can dichoptic GMC threshold shifts be explained by a contrast increase in the deprived eye and/or a relative drop in contrast in the nondeprived eye? 

Two main tests have been used to measure interocular contrast gain changes, per se, as a result of deprivation. The first is to look for changes in (at-threshold) monocular contrast sensitivity. Although Lunghi et al. (2011) did not report any changes in contrast sensitivity (as measured by detection thresholds) after 150 min of light-tight patching, Zhou, Clavagnier, et al. (2013) did, for both light-tight and diffuser deprivation, as evidenced by an increase in contrast sensitivity (for a 0.3 c/d grating) in the deprived eye and a concomitant decrease in sensitivity in the nondeprived eye. This is consistent with our CSF measurements, discussed above, which showed an increase in contrast sensitivity (decrease in contrast thresholds) in the deprived eye and a concomitant decrease in sensitivity in the nondeprived eye. Taken together, this helps establish that deprivation does affect contrast processing per se and moreover that it is fundamentally an interocular effect, because sensitivity changes are be observed even in the nondeprived eye. 

The second method is to measure changes in (suprathreshold) interocular perceived contrast. This measure is even more relevant to deprivation studies, we would argue, since the tests for deprivation effects such as rivalry, binocular phase combination (Ding & Sperling, 2006), and GMC thresholds employ suprathreshold stimuli. Lunghi et al. (2011) reported a postdeprivation imbalance in perceived contrast, resulting in a nondeprived/deprived contrast ratio of 0.74⁸ (i.e., apparent contrast in the nondeprived eye was 0.74 that of the deprived eye). This is consistent with the postdeprivation biases in interocular rivalry they observed, but they deemed it quantitatively insufficient to fully account for the observed deprivation effect. Zhou, Clavagnier, et al. (2013), using a binocular contrast matching task, reported an interocular perceived contrast ratio of approximately 0.85.⁹ Again, this was consistent with the deprivation-induced changes in interocular GMC thresholds they observed (and indeed Zhou, Clavagnier, et al., 2013; Zhou, Thompson & Hess, 2013, explicitly attributed their deprivation effects to changes in interocular gain) but also was apparently insufficient to account for the entire imbalance: In a separate experiment, a contrast imbalance of approximately 0.70¹⁰ had to be used in their interocular GMC stimuli to “cancel out” the bias induced by deprivation. What kind of contrast imbalance, then, would be necessary to account for our GMC results? 

To answer this, we conducted a separate test to determine how dichoptic GMC thresholds (in general, no deprivation involved) were affected by the contrast ratio of the dichoptic motion stimuli (i.e., the relative contrast of the signal and noise dots in each eye). We presented fixed-contrast dots to one eye and parametrically reduced the contrast of the dots in the other. We used six different interocular contrast ratios ranging from 1 (where the contrast to each eye was equal and set to a Weber contrast of 64%) to 0.5 (where the contrast to one eye was half the other eye, i.e., Weber contrast of 32%) in 10 steps. Eight observers each performed six blocks of 360 trials (with dichoptic signal/noise presentation to each eye counterbalanced and randomized, trial by trial), with the interocular contrast ratio fixed within each block. All blocks were run in a single session, and each block took about 10 min to complete. As expected, when signal dots were presented with lower contrast, dichoptic GMC thresholds increased, that is, the task got more difficult (Figure 7a, upper curve), and when noise dots were presented with lower contrast, thresholds were reduced (Figure 7a, lower curve). Using those values, we can then determine the function that relates interocular contrast ratios to GMC coherence threshold ratios (thresholds when signal is presented to the reduced contrast eye divided by thresholds when the noise is presented to the reduced contrast eye; Figure 7b). Looking back to our main experiment, the normalized GMC threshold ratios we observed (collapsed across eye deprived and observer) were 0.66 and 0.53 for light-tight and kaleidoscopic deprivation, respectively (Figure 3d). According to the function here, this result could be driven by an imbalance in contrast of approximately 0.80 to 0.70 (Figure 7b, dotted arrows). Notably, this imbalance is comparable to the values of 0.74 (Lunghi et al., 2011) and 0.85 (Zhou, Clavagnier, et al., 2013) observed by the previous work discussed above. In short, this indicates that the deprivation effects we observed on GMC thresholds (induced by either light-tight or kaleidoscopic patching) are quantitatively well-accounted for by a shift in interocular contrast. We argue then that, whatever the trigger, and whatever mechanisms may respond to deprivation, the effect manifests as a shift in interocular contrast (Spiegel, Baldwin, & Hess, 2017). 

Figure 7

 

The relationship between dichoptic GMC thresholds and interocular contrast ratios. Panel (a) shows dichoptic GMC thresholds as the interocular contrast imbalance between the eyes is increased. Panel (b) shows the dichoptic GMC threshold ratio (where the threshold when noise is presented to the lower contrast eye [and signal to the full-contrast eye] is divided by the threshold when signal is presented to the lower contrast eye [and noise to the full-contrast eye]) as a function of interocular contrast ratio. Dotted arrows indicate the contrast imbalance necessary to achieve the GMC threshold ratios that were observed in the dichoptic tests from our main experiment. Error bars indicate SEM.

This conclusion dovetails nicely with recent neurophysiology, which has suggested a direct relationship between short-term, monocular deprivation and contrast gain changes, as well as helping to localize where the changes in interocular balance are instantiated. For instance, cortical steady-state visually evoked potential responses associated with a (diffuser lens) deprived eye showed increased contrast reactivity, postdeprivation (Zhou, Baker, Simard, Saint-Amour, & Hess, 2015). Similarly, visually evoked potential measures from primary visual cortex confirmed an increase in the (diffuser lens) deprived eye (in the early, C1 component, associated with retinotopic areas) and a negatively correlated decrease in the nondeprived eye, a further confirmation of the push-pull nature of the effect (Lunghi, Berchicci, Morrone, & Di Russo, 2015). Further, given that deprivation effects find longest-term expression (up to 180 min, postdeprivation) on rivalry duty cycles for chromatic, isoluminant test stimuli, especially implicates the P pathway in primary visual cortex (Lunghi et al., 2013). Finally, it has been shown in animal models that contrast gain control is GABA mediated in primary visual cortex (Katzner, Busse, & Carandini, 2011). In adult humans, resting GABA concentration in V1, as measured by (ultra-high-field 7T) magnetic resonance imaging, dropped after short-term monocular deprivation (again, with a diffuser lens), with the amount of the drop being related to the strength of the behavioral deprivation effect (as measured by binocular rivalry; Lunghi, Emir, et al., 2015). Put together, the psychophysical data and the neurophysiology, then, paint a consistent story: short-term, monocular deprivation triggers plasticity in the primary visual cortex, facilitating a shift in interocular contrast gain, whereby gain is increased in the deprived eye and reduced in the nondeprived eye, a homeostatic attempt to restore interocular balance. 

Suppression during deprivation may reconcile previous results

Our current results with short-term, light-tight deprivation are in agreement with previous studies: Sensory eye balance shifts in favor of the deprived eye (Lunghi et al., 2011; Zhou, Clavagnier, et al., 2013). Although sufficient to induce such effects, light-tight patching is not necessary; other types of monocular deprivation, such as diffuser lens (Zhou, Clavagnier, et al., 2013), large (80%) global contrast reductions (Zhou et al., 2014), and low-pass filtering (Zhou et al., 2014), are also effective. On the other hand, though, other methods of monocular image degradation, such as phase scrambling and pink noise (Bai et al., 2017; Zhou et al., 2014), failed to induce deprivation. Taken together, it seemed that a monocular reduction in high spatial frequency contrast was both necessary and sufficient to induce shifts in sensory eye balance. This dovetailed nicely with contrast-based models of the induced deprivation effects themselves, in which a contrast imbalance during deprivation led to (sustained) changes in the weights of interocular gain control (Zhou, Clavagnier, et al., 2013), that action of an assumedly homeostatic mechanism attempting to restore interocular balance prior to binocular combination. However, our present results with kaleidoscopic patch and to some extent those of Bai et al. (2017) are inconsistent with this picture, because short-term deprivation effects were found with deprivations that did not significantly reduce high-frequency contrast (for example, pink noise, or the kaleidoscopic patching of the present study). Further complicating the picture was Zhou et al.'s (2014) failure to find any deprivation effect with a deprivation condition that used a sizeable (40%) reduction in monocular contrast. How can all these results be reconciled? We speculate here that deprivation effects do not depend on the contrast in the eyes' images per se but are instead determined by how likely (and how often) a particular deprivation causes that eye's image to be suppressed during the deprivation period. Most straightforwardly, it is the suppression of the degraded image, not the image degradation itself, that triggers a reweighting of interocular contrast gain. 

The present results challenge conventional models of ocular dominance plasticity and binocular combination that appeal only to feedforward processes (Chadnova, Reynaud, Clavagnier, & Hess, 2017; Meese et al., 2006). Ding and Sperling (2006) raised the possibility of complex higher-order visual processes (such as attention) influencing ocular dominance. Indeed, in the rivalry literature, for instance, meaningful information (Kovács et al., 1996) and natural images (Baker & Graf, 2009) or images with faces (Jiang et al., 2007), recognizable figures (Yu & Blake, 1992), or those targeted by attention (Dieter, Brascamp, Tadin, & Blake, 2016; Lack, 1978; H.-H. Li et al., 2017; Meng & Tong, 2004) have been shown to dominate in rivalry (for a review, see Wolf & Hochstein, 2011). There is growing evidence that attention can modulate eye-specific information (Zhang, Jiang, & He, 2012; Zhaoping, 2008), and this idea has been confirmed recently in the domain of binocular rivalry (H.-H. Li et al., 2017). Connecting these mechanisms further, H.-H. Li et al. (2017) attributed (and modeled) rivalry as an interplay of attention and interocular gain control (Baker, Meese, & Summers, 2007; Moradi & Heeger, 2009; Nichols & Wilson, 2009; Wilke, Logothetis, & Leopold, 2003). That means that the manipulation used to create the deprived eye's image will matter, because these will influence whether, and how often, the deprived eye is suppressed (Brascamp, Klink, & Levelt, 2015). Also, the details of the procedure of the deprivation period will matter, because this can also influence the relative dominance of the eyes' images. For instance, it has been shown that attention can influence rivalry, boosting the representation of the attended image (Dieter et al., 2016; Lack, 1978; H.-H. Li et al., 2017; Meng & Tong, 2004), and that the usefulness of an image toward visually guided behavior may also influence interocular suppression (Foley & Miyanshi, 1969). 

This framing gives another perspective on the studies shown in Table 1. With light-tight or diffuser patch deprivation, then, the prediction is clear: Total suppression of the deprived eye during the deprivation period will trigger a maximal compensatory interocular gain reweighting, in favor of the deprived eye, and thereby a maximal deprivation effect when binocular vision is restored. As well, the expectation is the same for the mean color and 80% contrast reduction conditions (where, similarly to the diffuser condition, deprivation approaches a ganzfeld). With the other manipulations, we can make only informed speculation on the nature of the interocular competition during deprivation, as none of the deprivation studies discussed here reported it, and most work on binocular rivalry has not been done with full-field, temporally dynamic, natural image manipulations (Blake & Wilson, 2011). That said, it is reasonable to expect the horizontal-pass and low-pass manipulations to again be similar to the diffuser condition; the deprived eye should be largely suppressed (Chai, Papathomas, Zhuang, & Alais, 2009; Fahle, 1982; Meese & Hess, 2004), yielding substantial deprivation effects. At the other extreme are the high-pass and 40% contrast reduction manipulations. These are most likely to support fusion (indeed, fusion under such conditions is an assumption that underlies binocular phase combination tasks). This lack of suppression, then, would not yield any deprivation effect. 

In between these two extremes lie the various phase scrambling conditions. These should yield considerable binocular rivalry (with a nominal bias toward the nondeprived eye, due to image statistics that identify it as a “natural” image versus its scrambled counterpart; Baker & Graf, 2009). Here, though, the details of the deprivation procedure should matter the most. When participants are asked to passively view a movie through one eye during deprivation (and a scrambled version of the movie in the other deprived eye, as in Zhou et al., 2014), near-balanced rivalry will likely be the norm (Wolf & Hochstein, 2011), and the “open” eye will not dominate sufficiently to trigger reweighting, helping to explain why there is no net deprivation effect under these passive viewing conditions (Zhou et al., 2014). However, if participants could free-view their environment during deprivation, as in Bai et al. (2017), or, better yet, were actively encouraged to go about their daily business outside the lab, as in the present study, then the nondeprived eye would likely dominate the deprived eye more thoroughly throughout the deprivation period (indeed, when debriefed, naive observers reported, and two expert observers confirmed, that during active vision, the kaleidoscopic image was nearly always suppressed, especially in central vision¹¹), resulting in the significant deprivation effects found here. Overall, these expectations qualitatively fit the pattern of results as shown in Table 1, whereas expectations based on the deprived eye's relative RMS contrast do not. 

In the present study, we demonstrated that robust shifts in sensory eye balance can be induced even with a monocular deprivation that does not affect low-level visual information. Although the presence of the monocular kaleidoscopic patch means there will be conflict between the eyes' images during deprivation, there are no feedforward triggers to interocular contrast gain control; there is a conflict, but which eye is to blame? The degradation induced by the kaleidoscopic deprivation would be noted only by a higher-level system sensitive to which visual stream was of greater utility for active vision (Foley & Miyanshi, 1969). One reason our study may have yielded stronger effects whereas previous work with non–contrast-based deprivation was inconsistent (showing no effect [Zhou et al., 2014] or mixed effects [Bai et al., 2017]) is that our observers explicitly engaged in active vision during deprivation. Successful interaction with the visual scene necessitates consistently prioritizing information received through the open, nondeprived eye and suppressing the kaleidoscopic, deprived eye. This, then, mimics the suppression during light-tight (or diffuser) patching and helps explain the similar effect sizes. A direct, parametric test of these ideas must be left for future work, but, suggestively, it has been found that “deprivation”—in the form of suppression of one eye from rivalry or continuous flash suppression—increased the dominance of the deprived eye during postdeprivation tests of rivalry (Kim et al., 2017). 

Acknowledgments

This work was supported by the Doctoral dissertation grant, an internal grant program at the University of Massachusetts Boston, awarded to the authors. 

Commercial relationships: none. 

Corresponding author: Mahalakshmi Ramamurthy. 

Email: . 

Address: Schepens Eye Research Institute, Harvard Medical School, Boston, MA, USA. 

Footnotes

¹  Bai et al. (2017) argued that there may be two separate mechanisms in play here—one responsible for ocular dominance for phase combination and one responsible for ocular dominance in rivalry—and these are selectively affected by deprivations of contrast and deprivations of phase information, respectively.

Footnotes

²  Visual acuity was measured using a Snellen acuity chart. Both monocular and binocular best corrected visual acuity was measured. Stereoacuity was measured using a random dot stereogram “Randot stereotest” (Stereo Optical Company). Color deficiencies were screened using a Waggoner evaluation kit.

Footnotes

³  Eye dominance can be determined through sighting test or by other sensory tests (Handa et al., 2004). There is still contention about how various tests correlate with each other (J. Li et al., 2010). We used a conventional measure of eye dominance: the sighting test. The procedure is similar to that used by Mansouri, Thompson, and Hess (2008), in which participants, with both eyes open, looked through a sighting hole at a distant target. Participants closed and opened each eye in succession to determine the eye that was aligned with the target (Rosenbach, 1903). This eye was defined as “preferred” for this study.

Footnotes

⁴  A normalized coherence threshold ratio of approximately 0.5 can be estimated from Zhou et al. Figure 2B right panel (2013), at the T4 time point (12 minutes post-deprivation, which is comparable to our testing duration). This value was similar for both light-tight and diffuser deprivation.

Footnotes

⁵  There is more than one way to measure rivalry dominance. Lunghi, Burr, and Morrone (2011) used the mean of the duration of the bouts of dominance, where here, similarly to Bai et al. (2017), we base the measure on the total duration. In general, the pattern of results should be the same. (One scenario in which they could differ, though, is if the bouts of dominance were to, say, decrease, but became more frequent. Here, the mean phase duration would indicate a decrease in dominance but accumulated time an increase; we think the latter is the more appropriate interpretation.)

Footnotes

⁶  As estimated from their Figure 1B at the 6-minute time point after deprivation, a point closest to when our rivalry tests occurred.

Footnotes

⁷  The only place where a deprivation effect was not found was where it was not expected: on monocular GMC thresholds. In these tests, both signal and noise dots are presented to the same eye (the other eye's image is left blank). It is known that once the contrast of dots is suprathreshold (here, dots were nominally at 64% Weber contrast), there is no effect of overall contrast on GMC thresholds; they are driven just by the coherence level (i.e., the signal-to-noise ratio; Mansouri et al., 2008). Thus, even if contrast gain were increased in the deprived eye and/or reduced in the nondeprived eye (which all our measures confirm was indeed the case), this would not affect monocular GMC thresholds. This condition was included as a planned, negative effect.

Footnotes

⁸  Lunghi et al. (2011) originally reported the ratio as deprived/nondeprived, so a value of 1.36 (i.e., the deprived eye had 1.36 the perceived contrast of the nondeprived). We have flipped the ratio here to facilitate comparison with the conventions in Zhou, Clavagnier, et al. (2013) and the present study.

Footnotes

⁹  As estimated from Zhou, Clavagnier, et al. (2013), figure 3.

Footnotes

¹⁰  As estimated from Zhou, Clavagnier, et al. (2013), figure 4. and the present study.

Footnotes

¹¹  In this context, it is worth revisiting the pattern of mixed percepts in our rivalry tests. Prolonged rivalry has been found to increase the incidence of mixed percepts (Klink et al., 2010). If there had been prolonged rivalry during deprivation (as opposed to consistent suppression of the kaleidoscopic eye's image), then the incidence of mixed percepts in our rivalry tests would be expected to increase, postdeprivation. However, we found no change.