Adaptation to Binocular Anticorrelation Results in Increased Neural Excitability Reuben Rideaux, Elizabeth Michael, and Andrew E. Welchman Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/32/1/100/1861419/jocn_a_01471.pdf by guest on 05 May 2021 Abstract ■ Throughout the brain, information from individual sources Receptive fields of disparity-tuned simple cells in macaque V1. converges onto higher order neurons. For example, information Neuron, 38,103–114, 2003; Cumming, B. G., & Parker, A. J. from the two eyes first converges in binocular neurons in area V1. Responses of primary visual cortical neurons to binocular dispar- Downloaded from http://direct.mit.edu/jocn/article-pdf/32/1/100/1931154/jocn_a_01471.pdf by guest on 26 September 2021 Some neurons are tuned to similarities between sources of infor- ity without depth perception. Nature, 389, 280–283, 1997], their mation, which makes intuitive sense in a system striving to match function has remained opaque. To determine how neural mech- multiple sensory signals to a single external cause—that is, estab- anisms tuned to dissimilarities support perception, here we use lish causal inference. However, there are also neurons that are electroencephalography to measure human observers’ steady- tuned to dissimilar information. In particular, some binocular neu- state visually evoked potentials in response to change in depth rons respond maximally to a dark feature in one eye and a light after prolonged viewing of anticorrelated and correlated feature in the other. Despite compelling neurophysiological and random-dot stereograms (RDS). We find that adaptation to an- behavioral evidence supporting the existence of these neurons ticorrelated RDS results in larger steady-state visually evoked [Katyal, S., Vergeer, M., He, S., He, B., & Engel, S. A. Conflict- potentials, whereas adaptation to correlated RDS has no effect. sensitive neurons gate interocular suppression in human visual These results are consistent with recent theoretical work sug- cortex. Scientific Reports, 8, 1239, 2018; Kingdom, F. A. A., gesting “what not” neurons play a suppressive role in supporting Jennings, B. J., & Georgeson, M. A. Adaptation to interocular stereopsis [Goncalves, N. R., & Welchman, A. E. “What not” difference. Journal of Vision, 18, 9, 2018; Janssen, P., Vogels, R., detectors help the brain see in depth. Current Biology, 27, Liu, Y., & Orban, G. A. At least at the level of inferior temporal 1403–1412, 2017]; that is, selective adaptation of neurons tuned cortex, the stereo correspondence problem is solved. Neuron, 37, to binocular mismatches reduces suppression resulting in 693–701, 2003; Tsao, D. Y., Conway, B. R., & Livingstone, M. S. increased neural excitability. ■ INTRODUCTION maximizes the image similarity between the two eyes It remains an important challenge in neuroscience to un- (Fleet, Wagner, & Heeger, 1996; Ohzawa, DeAngelis, & derstand how the brain combines a pair of 2-D retinal im- Freeman, 1990). This makes intuitive sense; however, some ages to support 3-D perception. Classically, this problem disparity-selective neurons in V1 appear poorly optimized for has been framed as one of matching features between such a computation in that they respond maximally to differ- the two eyes, that is, solving the “stereoscopic correspon- ent images presented on the two retinae (Read & Cumming, dence problem,” so that the depth of objects can be tri- 2007; Cumming & Parker, 1997). Moreover, binocular neu- angulated (Julesz & Chang, 1976; Marr & Poggio, 1976). rons can show tuning to images that are difficult to imagine This problem is nontrivial, as the number of “false matches” being produced in the real world. A prime example of this is (i.e., correspondences between features that do not origi- the test of neural function with anticorrelated RDSs (aRDSs) nate from the same object) rapidly increases with the in which the polarity of image features is reversed between number of to-be-matched elements. the two eyes. Unlike correlated RDSs (cRDSs), viewing Random-dot stereograms (RDSs) are frequently used to in- aRDS does not support reliable depth perception; neverthe- vestigate binocular vision because of their ability to divorce less, some disparity-selective neurons in V1 respond strongly information about 2-D form from differences between the to these stimuli. Despite empirical evidence supporting the two eyes. These stimuli are composed of many self-similar existence of these neurons in macaques (Janssen, Vogels, features, potentially posing a severe challenge to establishing Liu, & Orban, 2003; Tsao, Conway, & Livingstone, 2003; binocular correspondence. The classic framework for under- Cumming & Parker, 1997) and humans (Katyal, Vergeer, standing stereopsis is to find correspondence by considering He, He, & Engel, 2018; Kingdom, Jennings, & Georgeson, a range of potential disparities and selecting the offset that 2018), their functional role remains opaque. Recent theoretical work suggested a potential explana- tion for neurons tuned to mismatched binocular fea- University of Cambridge tures. In their binocular likelihood model of stereopsis, © 2019 Massachusetts Institute of Technology. Published under a Journal of Cognitive Neuroscience 32:1, pp. 100–110 Creative Commons Attribution 4.0 International (CC BY 4.0) license. https://doi.org/10.1162/jocn_a_01471 Goncalves and Welchman (2017) suggested a simple de- University of Cambridge Ethics Committee; all observers coding rule for binocular neurons: Information about provided written informed consent. depth can be read out from a population of binocular The sample sizes used in the experiments were se- neurons where the decoding scheme is based on the lected based on previous studies using similar tech- cross-correlation between the encoding receptive fields. niques to study stereopsis (Cottereau, McKee, Ales, & Under this scheme, the activity of a binocular neuron Norcia, 2012; Cottereau, McKee, & Norcia, 2012). can lead to increased excitation for a particular depth interpretation or drive suppression of a specific depth Apparatus and Stimuli estimate. By reading out a population of binocular neu- rons, it is possible to derive a likelihood estimate of the Stimuli were generated in MATLAB (The MathWorks, Inc.) Downloaded from http://mitprc.silverchair.com/jocn/article-pdf/32/1/100/1861419/jocn_a_01471.pdf by guest on 05 May 2021 depth of the scene. This provides a plausible explanation using Psychophysics Toolbox and Eyelink Toolbox exten- for why neurons should respond to binocular correspon- sions (Cornelissen, Peters,&Palmer,2002;Brainard, dences that do not relate to a single physical object in the 1997; Pelli, 1997; see psychtoolbox.org/). Binocular pre- environment. In particular, the “what not” responses of sentation was achieved using a pair of Samsung 2233RZ binocular neurons can be used to drive suppression of LCD monitors (120 Hz, 1680 × 1050) viewed through Downloaded from http://direct.mit.edu/jocn/article-pdf/32/1/100/1931154/jocn_a_01471.pdf by guest on 26 September 2021 unlikely interpretations of the scene. Despite this theoret- mirrors in a Wheatstone stereoscope configuration. The ical promise, there is little empirical evidence for the role viewing distance was 50 cm, and participant head position of “what not” responses in the human visual system. was stabilized using an eye mask, headrest, and chin rest. The idea that binocular mismatches are used to drive Eye movement was recorded binocularly at 1 kHz using an suppression in visual cortex yields a distinct prediction EyeLink 1000 (SR Research Ltd.). concerning the balance of excitation and inhibition fol- Adaptation stimuli consisted of RDS (12° × 12°) on a lowing a period of adaption. In particular, adapting the mid-gray background surrounded by a static grid of black responses of units that drive suppression should lead and white squares intended to facilitate stable vergence. to less inhibition, thereby increasing the net excitation Dots in the stereogram followed a black or white of the cortex. To investigate the role of “what not” re- Gaussian luminance profile, subtending 0.07° at half sponses within the visual cortex, here we use electro- maximum. There were 108 dots/deg2, resulting in ∼38% encephalography to measure human observers’ brain coverage of the background. In the center of the stereo- activity during and after prolonged viewing of aRDS. gram, four wedges were equally distributed around a cir- Specifically, we measure steady-state visually evoked po- cular aperture (1.2°), each subtending 10° in the radial tentials (SSVEP) in response to cRDS and aRDS, following direction and 70° in polar angle, with a 20° gap between adaptation to either aRDS or cRDS. We find that, follow- wedges (Figure 1A). The wedge formation was used to ing adaptation to aRDS, SSVEP amplitude in response to perceptually accentuate the near/far regions from the cRDS increases relative to a preadaptation baseline. surrounding zero disparity surface. Dots constituting These results are consistent with the idea that “what the wedges were offset by 10 arcmin between the left not” responses play a suppressive role in supporting and right eyes, and the remaining dots had zero offset. stereopsis; that is, selective adaptation of “what not” re- This disparity was large enough to clearly distinguish sponses reduce suppression, resulting in increased neu- the near/far surface from the surrounding region while ral excitability. still being comfortable to stereoscopically