1. Neuroscience

Visual mode switching learned through repeated adaptation to color

  1. Yanjun Li  Is a corresponding author
  2. Katherine EM Tregillus
  3. Qiongsha Luo
  4. Stephen A Engel  Is a corresponding author
  1. Department of Psychology, University of Minnesota, United States
https://doi.org/10.7554/eLife.61179
Share this article
Cite this article
  1. Yanjun Li
  2. Katherine EM Tregillus
  3. Qiongsha Luo
  4. Stephen A Engel
(2020)
Visual mode switching learned through repeated adaptation to color
eLife 9:e61179.
https://doi.org/10.7554/eLife.61179
  2. Download BibTeX
  3. Download .RIS

Abstract

When the environment changes, vision adapts to maintain accurate perception. For repeatedly encountered environments, learning to adjust more rapidly would be beneficial, but past work remains inconclusive. We tested if the visual system can learn such visual mode switching for a strongly color-tinted environment, where adaptation causes the dominant hue to fade over time. Eleven observers wore bright red glasses for five 1-hr periods per day, for 5 days. Color adaptation was measured by asking observers to identify ‘unique yellow’, appearing neither reddish nor greenish. As expected, the world appeared less and less reddish during the 1-hr periods of glasses wear. Critically, across days the world also appeared significantly less reddish immediately upon donning the glasses. These results indicate that the visual system learned to rapidly adjust to the reddish environment, switching modes to stabilize color vision. Mode switching likely provides a general strategy to optimize perceptual processes.

Introduction

When the visual system encounters different environments – for example a change in overall brightness, focus, or color – sensory processing also changes, in order to maintain accuracy and efficiency. Some of the processes producing such adjustments, called visual adaptation, unfold gradually (Clifford et al., 2007; Kohn, 2007; Wark et al., 2007; Webster, 2015). For example, putting on sunglasses can alter the color of an apple, making it difficult to determine if it is ripe, but as our visual system adapts, the apple’s apparent color gradually returns to normal. For common environmental changes, it would be beneficial if the visual system could remember past adaptation, and rapidly switch to the appropriate state (Engel et al., 2016; Yehezkel et al., 2010). Such visual mode switching would aid the many functions that adaptation serves, including improving the detection or discrimination of objects and their properties (Dragoi et al., 2002; Krekelberg et al., 2006; McDermott et al., 2010; Müller et al., 1999; Wissig et al., 2013) and making neural codes more efficient (Seriès et al., 2009; Sharpee et al., 2006; Wainwright, 1999).

Empirical evidence for learning to switch visual modes is sparse and inconclusive, however. A few studies have found preliminary support for learning effects on visual adaptation (Engel et al., 2016; Yehezkel et al., 2010), but others have found little to no effect of experience (Tregillus et al., 2016; Vinas et al., 2012). Notably, previous work has not measured the consequences of moving in and out of an environment multiple times per day over many days, and none has tested for changes in the time course of adaptation with experience. Thus, it remains unclear whether people can learn to rapidly switch visual modes with experience.

Here, we used color adaptation to test for such learning: Observers wore a pair of tinted glasses, which made the world appear very reddish (the spectral transmission of the glasses as well as the monitor gamut with and without the glasses are shown in Figure 1). Color adaptation in such situations is relatively well-understood, and one of its main effects is that the dominant color of the environment fades over time (e.g. Belmore and Shevell, 2008; de La Hire, 1694; Eisner and Enoch, 1982; Neitz et al., 2002; von Kries, 1902), restoring the world to its prior, ‘normal’ appearance.

Figure 1
Download asset Open asset
Glasses' characteristics and experimental procedures.

(A) The red glasses used in this study and their transmission spectrum. The glasses filter out most of the energy at short wavelengths and maintain most of the energy at long wavelengths. (B) Monitor gamut through (solid line) and without (dashed line) the glasses plotted in CIE color space. The glasses compress the gamut and shift it toward red chromaticity. For example, the greenest light produced by the monitor (black dot) falls in an orange part of color space through the red glasses (gray dot). (C) Experimental procedures. The upper panel indicates the times when the observers wore the glasses within 1 day. Two test sessions were conducted, during the first and last 1 hr of wearing the glasses. The lower panel illustrates the test procedure in each session. Orange bars indicate the time of test: 5 min before putting on the glasses, immediately after putting on the glasses, then following 10, 25, 40, and 55 min of wearing the glasses. Observers then removed the glasses and were tested immediately, and 10, 20, and 30 min later. (D) Test display. Observers adjusted the color of a square centered on a background image of a naturalistic office environment, presented on a monitor in a fully lit room. The fixed image of the office and skyline was presented on the test display to give observers context information when making the adjustments. A black square of 5.7° separated the 0.5° square test patch from the background image. The test patch was presented for 200 ms at 1.5 s intervals, and the observer’s goal was to set it to appear unique yellow. Observers viewed the test display through a 3-foot felt-lined tunnel, on a calibrated monitor, in the fully lit lab room.

Observers in the present experiment donned and removed the glasses multiple times a day for 5 consecutive days. We hypothesized that color adaptation would speed up and/or get a head start over days, such that observers would experience a much smaller perceptual change in the color of the world when they put on the red glasses, providing evidence that they had learned to switch modes. Because it may involve mechanisms beyond classical adaptation, we will use the term ‘rapid adjustment’ to refer to this possible empirical evidence for mode switching -- that as soon as observers put the glasses on, their effects were less prominent. Different potential mechanisms behind the adjustment will be considered in the Discussion.

Observers wore the red glasses for five 1-hr periods, each separated by 1 hr without glasses (Figure 1). To track adaptation, we asked observers to make unique yellow settings, identifying the wavelength of light that appears neither reddish nor greenish (Jameson and Hurvich, 1955). Unique yellow is a commonly used measure in color perception, in part because observers are highly consistent in their judgments (e.g. Brainard et al., 2000; Jameson and Hurvich, 1955; Neitz et al., 2002).

On each day, observers were tested in two sessions, once in the morning and once in the afternoon, for 5 days in a row. In each session, they performed: 1 test before putting on the glasses; 5 tests with the glasses on; and 4 tests after removing the glasses. During each test, observers made unique yellow settings for five 1-min blocks. Within each block, observers set as many matches as they could. Each datapoint in Figures 2–5 represents the average settings across a 5-min test. The tests were all conducted in a fully lit lab room in order to provide information about the visual environment present. In a follow-up, conducted about 1 month after the main experiment, observers participated in one additional and identical testing session.

Figure 2 with 2 supplements see all
Download asset Open asset
Results of the main experiment and the follow-up session.

Mean unique yellow settings represented in hue angle are plotted as a function of time for 5 days and the follow-up test. The black dots are baseline settings, made at the beginning of each test session with glasses off. The white background indicates morning sessions, and the light gray background indicates afternoon. The red dots plot settings with glasses on and the green dots are settings after removing the glasses. Successive symbols are plotted for each 5-min test (see Figure 1C). The gray bars represent standard errors of the mean, computed across participants (N = 11).

Figure 2—source data 1

Figure 2 data.

https://cdn.elifesciences.org/articles/61179/elife-61179-fig2-data1-v1.xlsx

Results

The world appeared very reddish when observers first put on the glasses, and the redness faded over time as vision adapted. Figure 2 plots mean unique yellow settings (quantified as hue angle, see Materials and methods) as a function of time, averaging across 11 observers, for the 5 days. The relatively small number (around 220) on the very first test with glasses on (red dots) indicates that observers’ unique yellow was physically relatively green, which was required to cancel the redness produced by the glasses. The upward slope of each session’s 5 settings shows that observers added less green to unique yellow over time, adapting to the red environment during the 1 hr of wearing the glasses, with the world looking less and less red. This pattern can be seen both in the morning (with white background in Figure 2) and the afternoon (with light gray background) session on all 5 days.

Adjusting to the glasses became faster and stronger

Across days, observers learned to rapidly adjust to the red glasses. That is, when they first put the glasses on, the world appeared less and less reddish. This is visible in the graph by the rising trend of the first unique yellow setting in each session across days. A linear trend analysis (Figure 3 red dots) showed that this increase was reliable (yt=4.06t−76.7+et, 95% CI [2.91, 5.21], t = 6.87, p <0.0001). A number of different mechanisms could account for this empirical observation (see Discussion) but the changes were not due to lingering overall adaptation across days, as baseline measurements made before putting the glasses on showed a very different trend (see below).

Figure 3 with 1 supplement see all
Download asset Open asset
Rapid adjustment, total adaptation, and color aftereffect across 5 days.

Red dots show rapid adjustments, computed as mean settings from the first 5-min test of each session with the glasses on. Total adaptation effects, denoted by the pink dots, are mean settings from the test taken after 1hr of wearing the glasses. Green dots are mean settings of the first 5-min test after removing the glasses. Data have been corrected for possible baseline shifts by subtracting the baseline value for each morning session, taken immediately before putting the glasses on. The black dashed lines are linear fits to the rapid adjustment, total adaptation, and the aftereffect. Both rapid adjustment and total adaptation effect grew significantly over days, and there was a trend for aftereffects to decrease across day.

Figure 3—source data 1

Figure 3 data.

https://cdn.elifesciences.org/articles/61179/elife-61179-fig3-data1-v1.xlsx

How rapidly did this effect arise? Each datapoint in Figures 2 and 3 represents mean unique yellow settings averaged across the five 1-min blocks that comprised each test. To better judge the timing of effects, we repeated our analysis using observers’ averaged settings within only the first 1-min block. We also repeated the analysis using observers’ very first unique yellow setting in the first block. In both cases, unique yellow after donning the glasses again shifted significantly across days (t = 6.72, p<0.001 for the first block; t = 4.11, p<0.01 for the first setting), suggesting observers adjusted to the red glasses relatively quickly (Figure 3—figure supplement 1; Figure 2—figure supplement 2 shows the complete time course of our results as a function of 1-min blocks. We did not have priori expectations about the subtle trends from block to block, and so leave their examination to future work).

The amount of gradual adaptation to the red glasses during the 1 hr of testing, on the other hand, did not change across days. To estimate this quantity, we calculated the slope of the unique yellow settings within each 1 hr session. The grand average slope was 13.30° of hue angle toward red per hour, and there were no significant changes in slopes across test sessions (ANOVA, F9,100 = 1.06, p=0.40). Given the increasing rapid and constant gradual effects, it is not surprising that total adaptation, the sum of the rapid and the gradual effects, quantified by the last setting with glasses on in each session, also increased across days (yt=3.55t−60.6+et, 95% CI [2.14, 4.96], t = 3.68, p < 0.01, Figure 3, pink dots).

Learned mode switching was long-lasting

About 1 month (36 ± 7 days) after the main experiment, observers returned for a follow-up test. (Figure 2, right). Rapid adjustment to the glasses remained strong; the first test of unique yellow settings was redder than the settings from the first day of the main experiment (t = −4.83, p<0.001). However, the effect was somewhat diminished, as the follow-up settings were greener than those made on day 5 of the main experiment (t = 3.28, p<0.01). About 66% of the change across the 5 days was maintained in the follow-up test.

A trend for color aftereffect to change across days

When observers removed the red glasses, they experienced a classical color aftereffect (von Helmholtz, 1924; Krauskopf and Gegenfurtner, 1992; van Lier et al., 2009), and reported the world looked slightly greenish, thus they added red to cancel out this aftereffect when making their unique yellow settings (Figure 2, green dots). There was a trend for the immediate aftereffect to become less strong across days, evident in the analyses of the first 5-min test, the first 1-min block, and the first individual match setting (all p<0.1 and p>0.05; Figure 3 green dots show the means of the first 5-min tests). We tracked the further decay of the aftereffect for half an hour after removing the glasses, as observers’ settings shifted back toward baseline. The decay followed a roughly exponential shape, as previously reported for color aftereffects (Fairchild and Lennie, 1992; Fairchild and Reniff, 1995; Wright and Parsons, 1934). The decay constant, as measured by an exponential fit, did not change over days (F9,96 = 0.01, p=1).

Baseline unique yellow became slightly greener across days

Baseline values of unique yellow on each day were measured as the mean setting from the first 5-min test of the morning session, made before putting the glasses on; these settings were preceded by many hours (averaging approximately 15) without glasses wear, and were made without the glasses on. We observed a small but significant shift in baseline unique yellow settings over time, visible in Figure 2 (black dots) as the hue angle of baseline shifting toward green (yt=−0.94t+298.3+et, 95% CI [−1.30, −0.58], t = −3.33, p < 0.01). This is surprising because adapting to the red glasses makes redness more neutral over time, thus resulting in redder unique yellow (see Discussion).

To make sure our main finding of greater rapid adaptation did not depend upon this shift in baseline, we corrected its effect by subtracting the baseline setting in the morning test session on each day from all settings within the day. These baseline-corrected results showed a very similar overall pattern across days as the uncorrected data, although some effects became slightly larger (Figure 2—figure supplement 1).

Color constancy increased across days

Color constancy, an important benefit of adaptation, is the extent to which objects appear the same color despite changes in viewing conditions (e.g. Brainard and Radonjić, 2014; Foster, 2011; Witzel and Gegenfurtner, 2018). Such stability against transient features of the environment allows color appearance to provide reliable information about object identity and state (e.g. the ripeness of an apple).

One definition of perfect color constancy is when the same physical entity, a surface or light source, is perceived as identical under different viewing conditions. In experiments on monitors, where experimenters only have direct access to pixel intensities, perceived surfaces are usually estimated using modeling of likely lights and surfaces. However, the use of colored glasses in our study affords us a more direct approach.

Specifically, if observers in our experiment had perfect color constancy, then the same physical pixels on the monitor, regardless of whether they were seen as surfaces or light sources (our experiment was ambiguous in this regard), should appear unique yellow both with and without the glasses, despite the glasses’ dramatic effect on the spectrum of light reaching the eye. If these conditions hold then the only difference between the two unique yellow settings would be the difference in viewing conditions: That is, the same physical world (monitor pixels) would be perceived identically (i.e. unique yellow) across the two situations, a reasonable definition of perfect color constancy.

To estimate the amount of constancy, we characterized the physical color reaching the eye using the relative gain of the long-wavelength (L) and medium-wavelength (M) photoreceptors. This measure assumes that unique yellow settings correspond to a balancing point between the L and M cone responses, where a scale factor (gain) may be applied to responses of one of the cone classes: L = k*M. Effects of adaptation, or other plasticity, on unique yellow can be quantified by solving for k, which is equal to L/M (Neitz et al., 2002).

Figure 4 plots our results using this metric and shows that color constancy improved across days. The black dots are baseline unique yellow settings before putting on the glasses; as expected, they fell around 1, where the gain of the L and M cones was equal. The red dashed line at the top of the plot reflects perfect color constancy with glasses on, calculated by assuming that the physical color corresponding to unique yellow did not change from baseline on the first day. This identical spectrum of light would of course result in very different cone absorptions with the glasses on than off, because of the glasses’ effect on the light reaching the photoreceptors. On the other hand, if observers completely lacked color constancy, unique yellow settings with glasses on would simply remain at baseline values.

Figure 4
Download asset Open asset
Results plotted as relative gain of cones, L/M.

The red symbols show the relative gain of L and M cones (k = L/M, see text) for settings with glasses on, corrected for the red glasses transmittance. The black dots are baseline settings taken at the beginning of each test session with glasses off. If the observers showed complete absence of color constancy, the unique yellow settings with glasses on should have been at the same level as this baseline. The red dashed line above corresponds to the baseline unique yellow corrected for the red glasses’ transmittance. If observers had perfect color constancy, their settings would produce identical physical colors on the monitor with and without glasses, and so should fall here when glasses were worn.

Figure 4—source data 1

Figure 4 data.

https://cdn.elifesciences.org/articles/61179/elife-61179-fig4-data1-v1.xlsx

Across days, observers’ unique yellow settings (red dots) steadily rose toward the perfect color constancy line, indicating that color constancy improved. The very first time they put on the glasses, observers showed about 68% of perfect constancy, as calculated by the ratio between (1) the Euclidean distance between baseline and the first unique yellow setting with glasses on and (2) the distance between baseline and perfect constancy. This pre-existing constancy was presumably due to the rapid adaptation that produces the color constancy we experience in most situations (e.g. Rinner and Gegenfurtner, 2000; Smithson and Zaidi, 2004; Webster and Mollon, 1995). The amount of constancy grew significantly as observers learned to immediately adjust to the red glasses (t = 4.60, p<0.001), and exceeded 80% on the 5th day.

Individual differences in learning

Figure 5 plots changes in adaptation for individual observers. Some observers showed a large increase in the amount of rapid adjustment over 5 days (sample single observer shown in upper panel in Figure 5A, gray circle in Figure 5B), while others demonstrated a flatter pattern (lower panel in Figure 5A, black circle in Figure 5B). To test if the individual differences were statistically reliable, we computed the Pearson correlation between the changes in rapid adjustment from the first day to the fifth day, and the changes from the first day to the follow-up test. This correlation was significant (r = 0.81, p=0.003, Figure 5B), indicating that observers who had a larger learning effect over 5 days also retained larger amounts a month later, a form of test-retest reliability. Thus, individuals appear to differ in their ability to learn to rapidly switch visual modes.

Figure 5
Download asset Open asset
Individual differences in learning to adjust rapidly.

(A) Complete time courses for twoobservers. One observer (upper panel) showed a gradual increase of rapid adjustment during the 5 days. This observer also retained the stronger rapid adjustment in the follow-up test. Another observer (lower panel), showed a flatter pattern across days and little effect of learning in the follow-up test. (B) Test-retest reliability of individual differences. The change in rapid adjustment to the glasses (relative to the 1st day) measured on the 5th day significantly correlated with the change measured in follow-up test, across observers. This indicates observers differed in their ability to learn to rapidly switch visual modes. Red dots represent observers and the dashed line is the least-square fit. The light gray and black circles denote the individuals plotted in the upper and lower portion of panel A, respectively.

Figure 5—source data 1

Figure 5 data.

https://cdn.elifesciences.org/articles/61179/elife-61179-fig5-data1-v1.xlsx

Discussion

Through experience, observers learned to rapidly adjust to the red glasses, with the world appearing less and less reddish as soon as they put them on. In general, such rapid adjustment allows us to compensate for changes in the visual environment (e.g. Dragoi et al., 2002; Krekelberg et al., 2006; McDermott et al., 2010; Müller et al., 1999; Wissig et al., 2013), while also improving neural coding efficiency (e.g. Seriès et al., 2009; Sharpee et al., 2006; Wainwright, 1999).

In situations where different visual environments alternate frequently, like wearing and removing glasses, the visual system repeatedly readjusts itself. Our results suggest that observers can learn to make the adjustments more efficiently over time, to the point where they can adjust almost immediately upon entering the new environment. Such visual mode switching should enable people to better handle the demands of the complex and changing visual world.

Relation to prior work

It is well accepted that color adaptation has a ‘fast’ and a ‘slow’ mechanism and involves both receptoral and postreceptoral visual processes (e.g. Augenstein and Pugh, 1977; Fairchild and Reniff, 1995; Rinner and Gegenfurtner, 2000). One plausible interpretation of our results depends on these well-studied mechanisms; it is possible that through practice a fast adaptation mechanism became able to produce stronger and more rapid effects. In the motor-learning literature, this possibility has been termed ‘meta-learning’ because it affects parameters that govern the rate of adaptation, itself a kind of learning (e.g. Zarahn et al., 2008). Other alternative mechanisms are possible, however, including storage, and retrieval of adapted states (e.g. Lee and Schweighofer, 2009). Future work will explore these and other possibilities (see also below).

Past work examining visual mode switching has produced mixed results. For example, observers who adapted to cylindrical lenses, creating a sort of astigmatism, showed fast re-adaptation in a second testing session (Yehezkel et al., 2010). However, clinically astigmatic observers showed little change in adaptation during 6 months following their initial prescription of corrective lenses (Vinas et al., 2012). Conflicting results also appeared in color perception, where in one study adapting to yellow filters produced little change in adaptation across 5 days (Tregillus et al., 2016), while another report showed that long-term habitual wearers of red and green lenses can adapt more rapidly than naive observers to the color changes the lenses produce (Engel et al., 2016). Variability in observer populations and experimental procedures may account for these mixed findings. A final bit of evidence for mode switching comes from a different paradigm, in which learning of a visual discrimination task was specific to the visual system’s adaptive state, as manipulated by inducing a motion aftereffect (McGovern et al., 2012).

Our paradigm differed from past work in that observers adapted to very strong perceptual changes multiple times a day, and we tracked the detailed time course of adaptation in a test setting with rich cues to context (see below). Together, these factors likely produced larger changes and more reliable measurements of adaptation than observed previously. Testing whether factors such as the frequency of environmental change have an influence on the learning effect that we observed here is an important direction for future research.

Past work on long-term adaptation to colored environments, for example wearing red glasses or living under red lights continuously for part of the day, has found that adaptation grows stronger over days (Belmore and Shevell, 2008; Belmore and Shevell, 2011; Eisner and Enoch, 1982; Hill and Stevenson, 1976; Kohler, 1963; Neitz et al., 2002). However, these studies did not measure the time course of adaptation, or if observers could learn to rapidly switch between the different viewing conditions.

These past results were also highly variable, both within and between studies (Belmore and Shevell, 2008; Belmore and Shevell, 2011; Eisner and Enoch, 1982; Eskew and Richters, 2008; Hill and Stevenson, 1976; Kohler, 1963; Neitz et al., 2002; Tregillus et al., 2016), similar to the inconsistency in prior results on mode switching. One reason for this variability may be that observers were tested with little context present. For example, most tests were made in a completely darkened room, presenting only a single small test patch, making it difficult for the visual system to determine viewing conditions, and hence the appropriate adaptive state. The test setting in our experiment provided many cues that the visual system could use to tell which environment was present, that is whether the red glasses were on or off. These context cues may be necessary for mode switching to occur, although precisely which cues are important for which environments remains to be determined.

Other results from present work

Unexpectedly, we found that the baseline unique yellow setting, made immediately prior to the introduction of the red glasses each morning, shifted toward physically more greenish across days. The shift was in the opposite direction from the color that the glasses produced and from the shift of the adaptation effect within 1 hr. A similar trend in baseline settings was also found in two previous studies (Engel et al., 2016; Tregillus et al., 2016). While we can only speculate as to the cause of this pattern, it could be due to the aftereffect following the glasses’ removal. At that point, observers' judgments indicated that the world looked greenish to them, consistent with classical color aftereffects (von Helmholtz, 1924; Krauskopf and Gegenfurtner, 1992; van Lier et al., 2009). Adaptation across days to this greenish tint could have produced a shift in unique yellow toward green when not wearing the glasses. Long-term adaptation to aftereffects appears to be possible in other domains (Murch and Hirsch, 1972; Sheth and Shimojo, 2008).

The strengthened rapid adaptation we observed substantially improved observers’ color constancy, that is the stability of perceived color despite the changes in viewing conditions (e.g. Brainard and Radonjić, 2014; Foster, 2011; Witzel and Gegenfurtner, 2018). Rapid adaptation, and even faster processes including 'simultaneous' local contrast, are likely major mechanisms that serve this constancy, (e.g. Rinner and Gegenfurtner, 2000; Smithson and Zaidi, 2004). A current debate in the field is whether constancy is improved for familiar, natural illuminant changes, which our visual systems may have encountered most often (Rüttiger et al., 1999; Delahunt and Brainard, 2004; Pearce et al., 2014; Radonjić and Brainard, 2016; Weiss et al., 2017). Our results suggest that training with repeated exposure can improve color constancy, at least for a very strong and unfamiliar illumination change. More generally, observers show some amount of color constancy, and a variety of other perceptual constancies, in most natural settings, without any training. The extent to which these forms of visual mode switching are inborn, determined during development, or learned as an adult remains under investigation (e.g. Jameson and Hurvich, 1989; Sugita, 2004; Yang et al., 2015).

Relatedly, the aftereffect measured immediately upon removing the red glasses shifted toward the baseline across days, implying a faster readjustment to familiar, natural conditions over time. However, this trend was relatively small, of only modest statistical reliability, and could be specific to switches from the unnatural red-glasses conditions. The small size of the effect, if real, could be because observers have already partly learned to rapidly adjust to the natural environment, which remains controversial, as mentioned above.

Mechanisms producing more rapid adjustment

Neurally, adaptation to changes in the dominant color has effects on several sites within the retina (Boynton and Whitten, 1970; Lee et al., 1999; Rieke and Rudd, 2009) as well as cortical stages of color processing (e.g. Engel and Furmanski, 2001; Rinner and Gegenfurtner, 2000). One hint toward the neural locus of change in our experiment is that behavioral changes across days were not observed in adaptation within the hour of glasses wearing. This independence from classical adaptation, which partly arises early in the visual system, suggests that mode switching may arise relatively late in processing (Rinner and Gegenfurtner, 2000). Identifying more precisely the extent to which learning can affect these different stages of processing could be profitably addressed in the future.

Computationally, one can view adaptation as the result of an inference process, in which the visual system must determine whether the visual environment has changed (Grzywacz and de Juan, 2003; Kording et al., 2007; Wark et al., 2009). Through exposure to the alternating colored and uncolored environment, observers in our experiment may have learned: (1) that the red environment was more likely (i.e. it had higher prior probability); (2) to more efficiently extract evidence of the red environment (giving it a higher likelihood); (3) that the red environment was likely to persist for a long time (making it costly to not adapt); (4) to speed inference by remembering, rather than re-inferring, the past adaptive state for the red environment. All these possibilities could produce stronger immediate adjustment, and they are not mutually exclusive. Future work could determine which factors are responsible for the changes in rapid adjustment across days.

Individual differences

What are the sources of individual differences in the ability to learn to rapidly switch between the two states? Past work has shown that observers may display very different amounts of experimentally measured color constancy, depending upon whether they were asked to make judgments of surface reflectance or of reflected light (Arend and Reeves, 1986; Arend and Goldstein, 1987; Radonjić and Brainard, 2016). In a given task, observers could potentially use either of these strategies. We gave specific instructions in order to limit the impact of strategy selection (see Materials and methods); however, it is still possible that some observers could be ‘thinking’ more or less in making their unique yellow judgments, which could be one source of the individual differences we found here. Compliance in wearing the glasses could also theoretically account for differences, but we closely monitored compliance, and failures were very few. Future work can examine whether individual differences in other aspects of color perception, or vision more generally, can account for individual differences in mode-switching.

In sum, our results demonstrate that the visual system can learn to rapidly adjust to an experienced environment. This mode switching lessens the perceptual changes produced by changing viewing conditions, which could aid a number of perceptual tasks, for example recognition of objects or materials, discrimination between similar objects or materials, as well as improved communication with other observers. Mode switching is not limited to color vision. Similar rapid re-adaptation has been reported in audition (Hofman et al., 1998) and sensorimotor paradigms, in which observers adapt to prisms that rotate or displace their visual field (e.g. Redding et al., 2005), or force fields that disturb their motor outcomes (e.g. Wolpert and Flanagan, 2016). Visual mode switching also resembles context-dependent learning that arises in conditioning and other memory paradigms. Mode switching may be a general solution to the problem of maintaining consistent behavior in a changing world.

Materials and methods

Observers

Request a detailed protocol

Observers included author YL and 11 members (21 to 37 years of age) of the University of Minnesota community. All had normal color vision, as assessed by the Ishihara Color Blindness Test, and normal or corrected-to-normal (using contact lenses) visual acuity. None had worn red glasses for extended periods of time prior to this study. One of the observers recruited reported that she changed her criterion for unique yellow during the study, and her data showed very large variance in baseline across days. Her data were excluded from further analysis. Experimental procedures were approved by the University of Minnesota Institutional Review Board. All observers provided written, informed consent before the start of the study.

Apparatus

Visual stimuli were presented on a NEC MultiSync FP2141 cathode ray tube monitor, with screen resolution of 1024*768 pixels, and a refresh rate of 85 Hz. The monitor was calibrated using a Photo Research PR655 spectroradiometer, with gun outputs linearized through look-up tables. All visual stimuli were delivered in Matlab using the psychophysical toolbox (Brainard, 1997). Viewing distance was maintained at 50 cm with a chinrest.

Glasses

Request a detailed protocol

Observers wore a commercial pair of bright red glasses made by SomniLight (Shawnee, KS). Black baffling was added on the top of the frame to prevent light from bypassing the glasses from above. The glasses filter out most of the light at short wavelengths and let pass most of the light at long wavelengths. We measured the glasses transmittance by placing the glasses in front of the spectroradiometer and recording sunlight. The spectral transmission of the glasses (Figure 1A) shows that the transmittance is above 90% at wavelengths over 620 nm, and less than 10% at wavelengths below 550 nm.

To characterize the effect of the glasses on our testing display, we measured the gamut of the monitor with and without the glasses. Figure 1B demonstrates that the gamut of the monitor seen through the glasses becomes compressed and shifts toward red chromaticity.

Procedure

Request a detailed protocol

In the main experiment, observers wore the glasses for five 1-hr periods per day, for 5 consecutive days. On each day, observers came to the lab in the morning and wore the red glasses for 1 hr, while participating in a testing session. Then, they left the lab and attended to their routine everyday activities, experiencing a variety of illumination conditions. They were asked to put on the glasses again 1 hour after they took off the glasses in the lab. During the day, they wore the glasses for three 1-hr periods, each separated by 1 hr without glasses. At the end of the fourth 1-hr period without glasses, they came back to the lab for a second testing session, identical to that in the morning. Figure 1C, upper panel, illustrates the procedure of the experiment. In a follow-up test session conducted about 1 month after the main experiment, observers came back and performed one additional and identical testing session.

Observers completed all tests in a fully lit room (with no window), with the aim of measuring perceptual experience in a context like their natural environment while adapting to the glasses. The screen was viewed through a 3-foot felt-lined ‘tunnel’, so that ambient light reaching our test display was not a significant factor. Observers sat in front of the ‘tunnel’ with their heads positioned on a chinrest located at its entrance.

During the test sessions, observers adjusted the color of a 0.5° square centered on a background image of a naturalistic environment (an office scene). The mean luminance of the background office image was 20 candela/m2. A black square of 5.7° separated the test patch from the background image (Figure 1D). The goal was to set the small square to unique yellow. We gave instructions "Your task is to adjust the small patch to yellow, which contains no red nor green in it, based on the light reaching your eye. Try not to think about what the color of the patch on the screen should be" to observers for both tests with and without the red glasses.

The small patch was presented for 200 ms at 1.5 s intervals. To make adjustments observers pressed the left and down arrow buttons to reduce redness in the patch, right and up arrow buttons to reduce greenness in the patch, and then pressed the space bar when they had set the patch to appear neither reddish nor greenish. The left and right arrow buttons were for coarse adjustments, and the up and down arrow buttons were for finer adjustments. Observers had 20 s at the most to make one single adjustment so that they did not get stuck in making one single setting and did not adapt to the test patch.

Stimuli were created using a modified version of the MacLeod-Boynton color space (MacLeod and Boynton, 1979), scaled and shifted so that the origin corresponds to a nominal white point of Illuminant C and so that sensitivity is roughly equated along the two axes (Webster et al., 2000).

We began by computing cone responses from the stimulus spectrum using the Smith and Pokorny, 1975 cone fundamentals scaled so that the sum of L cone and M cone responses equaled 1 and the S cone responses divided by this sum also equaled 1. We then computed initial coordinates in the MacLeod-Boynton color space as rmb=(L−M)/(L+M) and bmb=S/(L+M). Finally, we scaled and shifted these coordinates:

LM=(rmb−.6568)×2168
S=(bmb−.01825)×6210

where LM is the scaled red-green coordinate, and S is the scaled S-cone coordinate, 0.6568 and 0.01825 are the MacLeod-Boynton coordinates of Illuminant C, and 2168 and 6210 are constants that scale the LM and S axes so that a value of 1 is roughly equal to detection threshold (Webster and Mollon, 1995).

All settings fell along the nominally iso-luminant plane (defined by the LM and S axes, with luminance set to 51 candela/m2) when not wearing the glasses in order to reduce brightness effects on the judgments. The photopic luminosity function we used to define nominal isoluminance was the CIE Photopic V(λ) modified by Judd, 1951.

In performing the unique yellow task, observers moved the stimulus along a circle in this plane. Thus, results are shown in ‘Hue Angle,’ where luminance and contrast (i.e. distance from the origin in the plane) were held constant. The stimuli were not adjusted for the glasses, and thus were likely not held at strictly constant luminance or contrast for judgments made while the glasses were on. The radius of the hue circle used was 80, which is a chromatic contrast of roughly 80 times detection threshold (see above) and was kept constant during the adjustment procedure.

Observers could adjust the angle of the stimulus with coarser or finer steps of 5 or 1 degree of hue angle per button press, respectively. Button presses had no effect once observers reached a green endpoint at 200° in hue angle and a red endpoint at 360° of hue angle. At the beginning of each trial, the hue angle of the stimuli was set randomly from 290 ± 45°. We tracked observers’ responses and stored each step of their adjustments. Examination of these data confirmed that they were not using the red or green endpoint as an anchor for their settings (e.g. always moving to the endpoint and then moving a fixed number of steps back).

At the beginning of each test session, observers performed five 1-min blocks of this task with natural vision. Then, they put the glasses on and immediately did 5 blocks of the task again. During each block, observers made as many matches as they could and between blocks, there was a break of a few seconds. Observers were also tested after 10, 25, 40, and 55 min of wearing the glasses. Between tests observers took a short walk and/or watched videos of their choice, or texted, on a computer or their phone.

After 1 hr, observers removed the glasses and were immediately tested again. Further tests were performed 10, 20, and 30 min after removing the glasses. The full test procedure is illustrated in the lower panel of Figure 1C.

Data analysis

Request a detailed protocol

Initial analyses averaged hue angle across tests and observers, and plotted them as a function of test time and day. In order to compare unique yellow settings with and without the glasses, we also characterized the results in terms of relative gain of the cone photoreceptors (Neitz et al., 2002). The analysis assumes that unique yellow settings correspond to a balancing point between the L and M cone responses, where a scale factor (gain) is applied to responses of one of the cone classes: L = k*M. Effects of adaptation can be quantified by solving for k using estimates of the cone responses to the stimulus for each unique yellow setting.

We computed relative gains of cones as follows: First, we calculated the spectra of the unique yellow settings by multiplying the RGB values of the observers’ settings by the gun spectra of the monitor and summing the outputs of the three guns. For the settings made with the glasses on, we further multiplied the monitor spectra by the transmission spectrum of the glasses. The spectra of the settings were then multiplied by the cone fundamentals to compute cone absorptions, using Stockman and Sharpe, 2000 fundamentals, with peaks scaled to 1. Lastly, the absorptions were converted into relative gain by the ratio of L/M (which solves for k in the equation above). This same quantity was computed for settings made both with and without the glasses.

Data availability

Data and code for figure reconstruction have been deposited in OSF. Dataset name is VisualModeSwitchingDataset. https://osf.io/eru9g/, https://doi.org/10.17605/OSF.IO/ERU9G.

The following data sets were generated

References

  9. Book
    1. Brainard DH
    2. Radonjić A
    (2014) Color constancy
    In: Chalupa L. M, Werner J. S, editors. The New Visual Neurosciences. MIT Press. pp. 545–556.
    https://doi.org/10.1007/s00417-014-2661-z
    1. Hill AR
    2. Stevenson RW
    (1976)
    Long-term adaptation to ophthalmic tinted lenses
    Modern Problems in Ophthalmology 17:264–272.
  26. Conference
    1. Judd DB
    (1951)
    Report of U.S. secretariat committee on colorimetry and artificial daylight
    Proceedings of the Twelfth Session of the CIE.
    1. Kohler I
    (1963)
    The formation and transformation of the perceptual world
    Psychological Issues 3:1–173.
    1. Radonjić A
    2. Brainard DH
    (2016) The nature of instructional effects in color constancy
    Journal of Experimental Psychology: Human Perception and Performance 42:847–865.
    https://doi.org/10.1037/xhp0000184
  57. Book
    1. von Kries J
    (1902)
    Theoretische Studien Über Die Umstimmung Des Sehorgans
    Albrecht-Ludwigs-Universitä.

Decision letter

  1. Joshua I Gold
    Senior Editor; University of Pennsylvania, United States
  2. Marisa Carrasco
    Reviewing Editor; New York University, United States
  3. David Brainard
    Reviewer
  4. Larry Maloney
    Reviewer; NYU, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

To maintain accurate perception, vision has to adapt in a changing environment. This paper provides convincing behavioral evidence that the visual system can learn to rapidly adjust to an experienced environment. Such learning may help stabilize vision perception and optimize perceptual processes. For instance, it may support enhanced color constancy across spectral environments an individual human encounters with some regularity, and aid many perceptual tasks, for example recognition of objects or materials.

Decision letter after peer review:

Thank you for submitting your article "Visual mode switching learned through experience" for consideration by eLife. Your article has been reviewed by Joshua Gold as the Senior Editor, a Reviewing Editor, and three reviewers. The following individuals involved in review of your submission have agreed to reveal their identity: David Brainard (Reviewer #1); Larry Maloney (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, we are asking editors to accept without delay manuscripts, like yours, that they judge can stand as eLife papers without additional data, even if they feel that they would make the manuscript stronger. Thus the revisions requested below only address clarity and presentation.

Summary:

It is very well established that chromatic adaptation occurs on short time scales (milliseconds to minutes) and reasonably well established that it occurs on longer time scales (hours to weeks). The question asked here is whether the visual system can learn, over a relatively long time scale (days), to accelerate/enhance its short-term adaptation (seconds to tens of minutes). Such learning might support enhanced color constancy across spectral environments an individual encounters with some regularity. The question is theoretically interesting and not much studied experimentally. The experiments appear to have been carefully executed and analyzed. The authors determine that there are two adaptation processes (rapid and gradual) and that the rapid process learns to anticipate. This "learning to adapt" effect allows individuals to adjust more rapidly to a change in visual experience when they have experienced a similar change previously. The effect was long lasting, still present in diminished form a month after completion of the main part of the experiment. They also found evidence for increased color constancy across days.

The experiments document the effect but do not much constrain the nature or site of the underlying mechanisms. In this sense, this study opens the door to future computational and experimental studies that attempt to dissect and explain the phenomenon.

Essential revisions:

1) The key phenomenon to explain is the decreasing onset of the rapid effect after change in illuminant condition. The idea that there are two stages of adaptation one of which can be driven by predictions of the external environment is exciting. The authors advance a probabilistic model of change detection but no details are given. There is a literature on two-stage models of adaptation (e.g., Pugh and Augenstein, 1977) that should be mentioned.

2) The authors portray their main finding sometimes as "increases in the amount of adaptation the visual system produces immediately upon putting on the glasses" or as learning "to shift rapidly to a partially adapted state". Thy do not seem to be equivalent mechanisms. The first refers to neural circuits adapting more rapidly/extensively after experience with the environment. The second suggests a learning effect-where the short timescale (e.g. seconds) adaptation is the same but the learning allows circuits a head start. It would be helpful to lay out these scenarios clearly in the text and use a consistent characterization of the main result.

3) It is interesting that the after-effect does not change with experience. Although subjects adjust more quickly to the glasses (e.g. they've learned the relevant adaptation state, Figure 2 red points), they haven't learned to undo the effect of removing the glasses (Figure 2, green points). The authors should discuss what they think about these findings.

4) We commend the authors for presenting a summary of the individual differences seen in the data, which are large. The origin of these differences is not clear, and together with mixed reports in the few studies of this sort that are reported in the literature mean some caution is required in interpretation of results. The possibility of some subjects "thinking about what they see" and responding on that basis comes to mind. This is a thorny issue that plagues most studies of constancy and is not reason to hold up publication of the present work, but it should be discussed.

5) How sure are you that subjects complied with glasses wearing instructions throughout the day? Could compliance relate to individual differences?

6) Color constancy: It is not clear what the logic is and how the color constancy calculations were done.

To make a perfect constancy prediction, one starts with some analysis of what surface reflectance corresponded to unique yellow in the glasses off condition, based on some assumption about ambient illumination and some constraints on surface reflectance functions. Then one asks what settings would correspond to that surface reflectance function under the changed illumination (here induced by the glasses). A calculation like this may have been done, but it is not provided.

Why is the proposed metric a measure of constancy? The rationale for the metric used (and claims made) is a single reference. Please clarify and help the readers understand better how the index captures constancy.

7) There are clearly a number of different ways to represent these data. Hue angle in the stimulus is fine and direct, and the L/(L+M) used in Figure 4 is also fine. Another alternative is the relative gain of the L and M cones at unique yellow, on the assumption that unique yellow represents L-kM = 0 for some k in each state of adaptation. That then gives k = L/M. Would this gain oriented expression of the data lead to more insight than the L/(L+M) version? It may be worth taking a look as it refers the data back to a hypothesized mechanistic state of the visual system. Would viewing either this or the L/(L+M) representation on an expanded within session time scale, separately for each session, reveal more about the dynamics, especially if (as suggested below in the context of Figure S2) the individual settings made once per minute in the within-session blocks were shown explicitly.

8) One aspect of the design that is likely important is that observers alternated between the two environments several times per day. It may be that frequency of environmental change is an important factor. Another factor is that the environment they were in was moderately complex, an office.

From Figure 1 it looks like the ambient environment for the experiment varied from day to day with changes in outdoor lighting coming in through the windows. Was there any characterization of this? Are the authors at all concerned that this variation may have affected results? Can they give any guidance to future researchers who would like to try to replicate their experiments as closely as possible, in terms of room size, relative amount occupied by windows, what the indoor lights were, ambient illumination level relative to display, etc.

9) Please say a little more about conversion to MB space and displayed stimuli.

a) How were peaks cone fundamentals scaled relative to each other when computing LMS, for subsequent computation of L/L+M and S/L+M. It seems that those two quantities are the Lmb and Smb passed into the computation of LM and S. Not sure S is the best choice of notation for the latter.

b) What photopic luminosity function was used to define nominal isoluminance. Given use of Stockman-Sharpe fundamentals one might infer the new CIE standard that is a weighted sum of those, or you might have used CIE 1931, or Judd-Vos, or.…

b) Please give the actual radius of the hue circle used in the adjustments, as well as the hue angle spacing for coarse and fine adjustments so that it would be possible for someone to produce your stimuli.

c) Was a full hue circle used, or were there endpoints? If end points, how confident are you that subjects didn't use those as a reference and count steps from there, or less explicitly anchor their adjustments to an estimated midpoint of the range provided? Learning of such strategies could masquerade as learning to adapt.

d) What is the luminance of the test patch that is being adjusted? Subsection “Apparatus” says the background luminance was 41.85 cd/m2, but later and in the picture, this is described/shown as black, which is surprising unless the ambient in the room was very high luminance.

10) Figure S2, top panel interpretation. The pattern of results is a little hard to interpret. We'd expect the least adaptation for the first setting, so in general these should be lower on the y-axis than the corresponding points in Figure 2. That does not appear to be the case in many instances. See the first group of glasses on settings, for example. Any comment? Are the first settings just really noisy? It might be clearer if each individual setting were plotted, rather than just providing the comparison of the first to the mean.

https://doi.org/10.7554/eLife.61179.sa1

Author response

Essential revisions:

1) The key phenomenon to explain is the decreasing onset of the rapid effect after change in illuminant condition. The idea that there are two stages of adaptation one of which can be driven by predictions of the external environment is exciting. The authors advance a probabilistic model of change detection but no details are given. There is a literature on two-stage models of adaptation (e.g., Pugh and Augenstein, 1977) that should be mentioned.

We agree that the manuscript failed to emphasize the past literature on two-stage models of adaptation, and our revision has corrected this oversight. We also have added some speculative discussion on mechanistic models. Specifically, we now mention the work cited above in our discussion, along with Fairchild and Reniff, (1995) and Rinner and Gegenfurtner, (2000), both of whom also propose “fast” and “slow” adaptation mechanisms.

Subsection “Relation to prior work”:

“It is well accepted that color adaptation has a ‘fast’ and a ‘slow’ mechanism and involves both receptoral and postreceptoral visual processes (e.g. Augenstein and Pugh, 1977; Fairchild and Reniff, 1995; Rinner and Gegenfurtner, 2000). One plausible interpretation of our results depends on these well-studied mechanisms; it is possible that through practice a fast adaptation mechanism became able to produce stronger and more rapid effects. In the motor-learning literature this possibility has been termed ‘meta-learning’ because it affects parameters that govern the rate of adaptation, itself a kind of learning (Zarahn et al., 2008). Other alternative mechanisms are possible, however, including storage, and retrieval of adapted states (Lee and Schweighofer, 2009). Future work will explore these and other possibilities (see also below).”

2) The authors portray their main finding sometimes as "increases in the amount of adaptation the visual system produces immediately upon putting on the glasses" or as learning "to shift rapidly to a partially adapted state". Thy do not seem to be equivalent mechanisms. The first refers to neural circuits adapting more rapidly/extensively after experience with the environment. The second suggests a learning effect-where the short timescale (e.g. seconds) adaptation is the same but the learning allows circuits a head start. It would be helpful to lay out these scenarios clearly in the text and use a consistent characterization of the main result.

The manuscript was unfortunately unclear on this very important point, and we have revised it accordingly. In the revised manuscript we now use the term "rapid adjustment" in the Introduction and Results. We have added text to the Introduction stating that this language is meant to only refer to the empirical observation that as soon as the observers put the glasses on, they were more adjusted to them and that a number of different mechanisms, such as the ones referred to above, could account for this. We have added to the Discussion that either or both of the above interpretations are consistent with our present data.

Introduction:

“We hypothesized that color adaptation would speed up and/or get a head start over days, such that observers would experience a much smaller perceptual change in the color of the world when they put on the red glasses, providing evidence that they had learned to switch modes. Because it may involve mechanisms beyond classical adaptation, we will use the term “rapid adjustment” to refer to this possible empirical evidence for mode switching -- that as soon as observers put the glasses on, their effects were less prominent. Different potential mechanisms behind the shift will be considered in the Discussion.”

Subsection “Relation to prior work”:

“One plausible interpretation of our results depends on these well-studied mechanisms; it is possible that through practice a fast adaptation mechanism became able to produce stronger and more rapid effects. In the motor-learning literature this possibility has been termed ‘meta-learning’ because it affects parameters that govern the rate of adaptation, itself a kind of learning (Zarahn et al., 2008). Other alternative mechanisms are possible, however, including storage, and retrieval of adapted states (Lee and Schweighofer, 2009). Future work will explore these and other possibilities (see also below).”

3) It is interesting that the aftereffect does not change with experience. Although subjects adjust more quickly to the glasses (e.g. they've learned the relevant adaptation state, Figure 2 red points), they haven't learned to undo the effect of removing the glasses (Figure 2, green points). The authors should discuss what they think about these findings.

This too is an important point, and we now discuss it more completely in the paper. We conducted more detailed analyses of the aftereffect, measured immediately upon removing the glasses, and found a trend for it to become less strong over days. As for our rapid adaptation to the glasses, we examined changes across days in the first 5-minute test, the first one-minute block, and the first individual match setting (see reply to comment, below). This change of the after-effect was not reliable at p < 0.05 in any of these analyses, but was p < 0.1 in all three. We now address this trend in the discussion as well, where we suggest it provides further evidence for improved mode-switching with experience. The reduced learning effect across days could be because observers already have partly learned to switch rapidly to natural viewing conditions through daily experience, or even evolution.

Subsection “Other Results from Present Work”:

“Relatedly, the after-effect measured immediately upon removing the red glasses shifted toward the baseline across days, implying a faster readjustment to familiar, natural conditions over time. However, this trend was relatively small, of only modest statistical reliability, and could be specific to switches from the unnatural red-glasses conditions. The small effect, if real, could be because observers have already partly learned to rapidly adjust to the natural environment, which remains an unresolved debate, as mentioned above.”

4) We commend the authors for presenting a summary of the individual differences seen in the data, which are large. The origin of these differences is not clear, and together with mixed reports in the few studies of this sort that are reported in the literature mean some caution is required in interpretation of results. The possibility of some subjects "thinking about what they see" and responding on that basis comes to mind. This is a thorny issue that plagues most studies of constancy and is not reason to hold up publication of the present work, but it should be discussed.

This is another important point that we have added to our Discussion. Past work has shown that observers may display different amounts of experimentally measured color constancy, depending upon whether they were asked to make judgments of surface reflectance or of reflected light (Arend and Reeves, 1986; Arend and Goldstein, 1987; Radonjić and Brainard, 2016). Observers could potentially use either of these strategies, depending upon what they were thinking during a task. In order to eliminate the impact of strategy selection, we gave instructions “Your task is to adjust the small patch to yellow, which contains no red nor green in it, based on the light reaching your eye. Try not to think about what the color of the patch should be” for both tests with and without the red glasses. However, it’s still possible that some observers who were not complying with the instructions could be "thinking" more or less in making their unique yellow judgements. We now discuss this possibility in the manuscript, as one source of the big individual differences we found here.

Subsection “Individual differences”:

“What are the sources of individual differences in the ability to learn to rapidly switch between the two states? Past work has shown that observers may display very different amounts of experimentally measured color constancy, depending upon whether they were asked to make judgments of surface reflectance or of reflected light (Arend and Reeves, 1986; Arend and Goldstein, 1987; Radonjić and Brainard, 2016). In a given task, observers could potentially use either of these strategies. We gave specific instructions in order to limit the impact of strategy selection (see Materials and methods), however, it is still possible that some observers could be "thinking" more or less in making their unique yellow judgements. This could be one source of the individual differences we found here.”

5) How sure are you that subjects complied with glasses wearing instructions throughout the day? Could compliance relate to individual differences?

We believe that compliance was good and apologize for leaving discussion of it out of the original manuscript. We sent text reminders to the observers on the first and second day about wearing the glasses during the day. In addition, every day when participants came in for the afternoon session, an RA asked them in person if they did as they were instructed in terms of glasses wearing. All reported wearing the glasses for 5 hours each day. Two participants reported having to delay wearing the glasses for ~30 minutes, one time on one day, and their testing sessions were accordingly delayed by 30 minutes to allow for full "glasses-on" time. It is possible that observers could be not reporting some lack of compliance, but we believe that this is not likely to be very frequent. We now discuss compliance in the manuscript.

Subsection “Individual differences”:

“Compliance in wearing the glasses could also theoretically account for them, but we closely monitored compliance, and failures were very few. Future work can examine whether individual differences in other aspects of color perception, or vision more generally, can account for individual differences in mode-switching.”

6) Color constancy: It is not clear what the logic is and how the color constancy calculations were done.

To make a perfect constancy prediction, one starts with some analysis of what surface reflectance corresponded to unique yellow in the glasses off condition, based on some assumption about ambient illumination and some constraints on surface reflectance functions. Then one asks what settings would correspond to that surface reflectance function under the changed illumination (here induced by the glasses). A calculation like this may have been done, but it is not provided.

Why is the proposed metric a measure of constancy? The rationale for the metric used (and claims made) is a single reference. Please clarify and help the readers understand better how the index captures constancy.

This suggestion is an excellent approach for studies conducted with lighting and/or surfaces simulated using a display device. Our understanding of the suggestion is that a perfect constancy prediction can be made by characterizing the observers' unique yellow settings in terms of an underlying physical stimulus, in this case the surface reflectance. The prediction of perfect constancy is that unique yellow should correspond to the same surface reflectance with and without the red glasses. Because of the use of display technology, and perhaps the observer's task of adjusting color rather than simulated reflectance, an experiment may not have direct access to the underlying surfaces and illuminants, and so whether or not there is a physical match (i.e. in reflectances) must be estimated using the modeling approach the reviewer outlines.

Sometimes, however, one does have more direct access to the underlying physics. An experiment could, for example, ask observers to choose unique yellow from a large set of physical (not simulated) Munsell papers under different physical illuminants. In this case, the perfect match prediction would be that the exact same paper (with the same reflectance) is selected under the different illuminants. In this case no modeling calculations are required for a good perfect constancy prediction.

We believe that the use of colored filters affords us this latter approach. Specifically, the perfect constancy prediction, i.e. whether the physics of a match, viewed with the red glasses on, matches the physics of the match, viewed with the red glasses off, in our case comes down to whether the physical content of the viewed scene was identical in the two viewing conditions. We do not have access to the observers' inferred reflectance functions, and indeed it is possible that they perceived our test as a light source rather than a surface, but we assume that if the light leaving the monitor was set identically with and without glasses, then a perfect match was made, because the identical physical situation in the world was perceived identically. That is, the only difference between the two unique yellow settings was that the observer was wearing the red glasses in one case – both the physical world and the perception it created (i.e. unique yellow) was constant in the two settings. To us, this seems a reasonable definition of constancy, and we apologize that it was not better explained in the manuscript. We now do so more completely.

Now, the reviewer also correctly points out that there is an additional question of what units to use to calculate how close the observer is to the perfect constancy prediction. Of course, many different units are possible, as long as they are corrected for the effects of the red glasses, and these could capture either how close the emitted light is to the perfect constancy prediction or how close the inferred surface reflectance is to that prediction. To avoid replotting our results too many times, we now choose to use the units that were suggested in comment 7) for this purpose, and have added citations supporting their use.

Subsection “Baseline unique yellow became slightly greener across days”:

“One definition of perfect color constancy is when the same physical entity, a surface or light source, is perceived as identical under different viewing conditions. In experiments on monitors, where experimenters only have direct access to pixel intensities, perceived surfaces are usually estimated using modeling of likely lights and surfaces. However, the use of colored glasses in our study affords us a more direct approach.

Specifically, if observers in our experiment had perfect color constancy, then the same physical pixels on the monitor, regardless of whether they were seen as surfaces or light sources (our experiment was ambiguous in this regard), should appear unique yellow both with and without the glasses, despite the glasses’ dramatic effect on the spectrum of light reaching the eye. If these conditions hold, then the only difference between the two unique yellow settings would be the difference in viewing conditions: That is, the same physical world (monitor pixels) would be perceived identically (i.e. unique yellow) across the two situations, a reasonable definition of perfect color constancy.

To estimate the amount of constancy, we characterized the physical color reaching the eye using the relative gain of the long-wavelength (L) and medium-wavelength (M) photoreceptors. This measure assumes that unique yellow settings correspond to a balancing point between the L and M cone responses, where a scale factor (gain) may be applied to responses of one of the cone classes: L = k*M. Effects of adaptation, or other plasticity, on unique yellow can be quantified by solving for k, which is equal to L/M (Neitz et al., 2002).”

7) There are clearly a number of different ways to represent these data. Hue angle in the stimulus is fine and direct, and the L/(L+M) used in Figure 4 is also fine. Another alternative is the relative gain of the L and M cones at unique yellow, on the assumption that unique yellow represents L-kM = 0 for some k in each state of adaptation. That then gives k = L/M. Would this gain oriented expression of the data lead to more insight than the L/(L+M) version? It may be worth taking a look as it refers the data back to a hypothesized mechanistic state of the visual system.

This is a great suggestion: We agree that the relative gain of the L and M cones at unique yellow is better at characterizing the mechanism that the weighting factors of cones are being adjusted. This has been previously used by Neitz, (2002) to reveal the plastic neural mechanism that compensates for the large variation in the ratio of L to M cones in the retina across populations. We have now replotted our Figure 4 in units of L/M. We then quantify the amount of color constancy using this unit, addressing comment 6). This is calculated by the ratio between (1) the Euclidean distance between baseline and the first unique yellow setting with glasses on and (2) the distance between baseline and the complete adaptation/perfect color constancy prediction (see above).

Would viewing either this or the L/(L+M) representation on an expanded within session time scale, separately for each session, reveal more about the dynamics, especially if (as suggested below in the context of Figure S2) the individual settings made once per minute in the within-session blocks were shown explicitly.

We agree that it is also important to look at the finer dynamics of the adaptation effects. We now make clearer in the paper that each point in our original plot represents the average settings across a 5-minute test, and that each test is in turn comprised of 5 one-minute blocks. Within each block, observers set multiple matches, and we can use those to examine the time course even more finely, which we do in response to comment 10) below.

Figure 2—figure supplement 2 plots the unique yellow settings, represented in hue angle, for each 1-minute block. To keep the plot of manageable size, we averaged the morning and afternoon session within a day, yielding 250 blocks in total. This averaging did not affect overall trends in the data. In the figure, one dot represents one block and 5 connected dots show the 5 blocks of each test. The upper section of the figure represents settings made with glasses off. The black dots are baseline settings before putting on the red glasses. The lower section of the figure shows settings made with glasses on. For reasons described in comment 10) below, the means in this figure were calculated while excluding the first match from tests 2-5, though again that did not affect the overall pattern of our results.

We include this in the supplementary material of the paper (Figure 2—figure supplement 2), replacing parts of our previous supplemental figures that also showed details of timing. Several factors complicate the interpretation of the plot: one cannot simply "connect all the dots" to see a fine-scale representation of the time course of adaptation. First, there are varying amounts of time between tests, from 5 to 15 minutes (see Figure 1C for complete testing schedule). In addition, between tests observers were viewing uncontrolled stimuli, though while keeping the glasses on or off as appropriate, that could differ substantially in color statistics from our test display (see also response to comment 10)). Finally, there was certainly adaptation to the testing display itself within a test. Accounting for these factors to make inferences about the fine-scale time course is difficult, and we plan to take up this challenge in a future publication, since the fine timing is generally an orthogonal point to our hypothesis about mode-switching. There are nevertheless interesting trends in the data, and we plan to share our complete data set for interested readers to examine. Note further that these factors do not invalidate our analyses that use just the first 1-minute block (or the first match) following donning or removal of the glasses, to characterize effects across testing sessions.

Subsection “Adjusting to the glasses became faster and stronger”:

“Figure 2—figure supplement 2 shows the complete time course of our results as a function of one-minute blocks. We did not have priori expectations about the subtle trends from block to block, and so leave their examination to future work.”

8) One aspect of the design that is likely important is that observers alternated between the two environments several times per day. It may be that frequency of environmental change is an important factor. Another factor is that the environment they were in was moderately complex, an office.

From Figure 1 it looks like the ambient environment for the experiment varied from day to day with changes in outdoor lighting coming in through the windows. Was there any characterization of this? Are the authors at all concerned that this variation may have affected results? Can they give any guidance to future researchers who would like to try to replicate their experiments as closely as possible, in terms of room size, relative amount occupied by windows, what the indoor lights were, ambient illumination level relative to display, etc.

It is an important point that the frequency of environmental change might have an influence on the learning effect we observed here; we are currently conducting a follow-up experiments where subjects wore the same pairs of glasses as in the present study for 5 hours continuously per day for 5 days, producing less alternation between the two environments within each day. Increased adaptation to the glasses was also found in the 4 subjects that we have tested so far. Further analyses will be done after we collect more data, but collection has been slow due to the pandemic, and so we prefer to publish this result in a future paper. For now, we have added this as a future direction to the Discussion.

“Testing whether factors such as the frequency of environmental change have an influence on the learning effect that we observed here is an important direction for future research.”

We sincerely apologize for a misunderstanding due to our unclear description of Figure 1. Figure 1D depicts the image observers viewed on the computer screen during testing, when they were making the unique yellow adjustments. It does not depict our testing environment itself. The fixed image of the office (not a rendered scene, did not change over time) was presented on the test display to give observers context information when making the adjustments; specifically, to make it obvious to the visual system that they were wearing glasses. This might not be the case if testing was conducted while viewing a small patch on a black screen.

Our experiment was conducted in a room underground with no windows, and the ambient lighting was from stable overhead room lights. The screen was viewed through a 3-foot felt lined 'tunnel', so that ambient light reaching our test display was not a significant factor. Subjects sat in front of the tunnel with their heads located at the entrance of the tunnel. We now describe our testing room and display more clearly in the manuscript. Finally, we now also make clear that between testing sessions observers moved around freely in their everyday lives, experiencing a variety of illumination conditions.

Figure 1D legend:

“The fixed image of the office and skyline was presented on the test display to give observers context information when making the adjustments.

Observers viewed the test display through a 3-foot felt-lined tunnel, on a calibrated monitor, in a fully lit lab room.”

Subsection “Procedure”:

“Observers completed all tests in a fully lit room (with no window), with the aim of measuring perceptual experience in a context like their natural environment while adapting to the glasses. The screen was viewed through a 3-foot felt-lined ‘tunnel’, so that ambient light reaching our test display was not a significant factor. Observers sat in front of the ‘tunnel’ with their heads positioned on a chinrest located at its entrance.”

9) Please say a little more about conversion to MB space and displayed stimuli.

a) How were peaks cone fundamentals scaled relative to each other when computing LMS, for subsequent computation of L/L+M and S/L+M. It seems that those two quantities are the Lmb and Smb passed into the computation of LM and S. Not sure S is the best choice of notation for the latter.

b) What photopic luminosity function was used to define nominal isoluminance. Given use of Stockman-Sharpe fundamentals one might infer the new CIE standard that is a weighted sum of those, or you might have used CIE 1931, or Judd-Vos, or.…

We now provide a more detailed description of the conversion to MB space and the displayed stimuli. We have added the following information to our Materials and methods section.

Subsection “Procedure”:

“Stimuli were created using a modified version of the MacLeod-Boynton color space (MacLeod and Boynton, 1979), scaled and shifted so that the origin corresponds to a nominal white point of Illuminant C and so that sensitivity is roughly equated along the two axes (Webster et al., 2000).

We began by computing cone responses from the stimulus spectrum using the Smith and Pokorny, (1975) cone fundamentals scaled so that the sum of L cone and M cone responses equaled 1 and the S cone responses divided by this sum also equaled 1. We then computed initial coordinates in the MacLeod-Boynton color space as rmb = (L-M)/(L+M) and bmb = S/(L+M). Finally, we scaled and shifted these coordinates:

LM = (rmb -.6568) x 2168

S = (bmb –.01825) x 6210

Where LM is the scaled red-green coordinate, and S is the scaled S-cone coordinate, 0.6568 and 0.01825 are the MacLeod-Boynton coordinates of Illuminant C, and 2168 and 6210 are constants that scale the LM and S axes so that a value of 1 is roughly equal to detection threshold (Webster and Mollon, 1995).

All settings fell along the nominally iso-luminant plane (defined by the LM and S axes, with luminance set to 51 candela/m2) when not wearing the glasses in order to reduce brightness effects on the judgements. The photopic luminosity function we used to define nominal isoluminance was the CIE Photopic V(λ) modified by Judd (1951).”

To calculate L/M in the revised manuscript (and L/(L+M) in the original) we used the Stockman and Sharpe, (2000) cone fundamentals, with the peaks of cone fundamentals scaled to 1.

b) Please give the actual radius of the hue circle used in the adjustments, as well as the hue angle spacing for coarse and fine adjustments so that it would be possible for someone to produce your stimuli.

The radius of the hue circle used in this study was 80, which is the chromatic contrast. This contrast is at a nominal threshold and was kept constant during the adjustment procedure. The coarser and finer steps of adjustment were +/- 5 and +/- 1 degree respectively per button press. We now report these parameters in our Materials and methods section.

Subsection “Procedure”:

“The radius of the hue circle used was 80, which is a chromatic contrast of roughly 80 times detection threshold (see above) and was kept constant during the adjustment procedure.

Observers could adjust the angle of the stimulus with coarser or finer steps of 5 or 1 degree of hue angle respectively per button press.”

c) Was a full hue circle used, or were there endpoints? If end points, how confident are you that subjects didn't use those as a reference and count steps from there, or less explicitly anchor their adjustments to an estimated midpoint of the range provided? Learning of such strategies could masquerade as learning to adapt.

This also is an important point that we now add in the revised manuscript. We did not use the full hue circle, and we had a green endpoint at 200 degrees in hue angle and a red endpoint at 360 degrees of hue angle. At the beginning of each trial, the hue angle of the stimuli was set randomly from 290±45 degrees. We tracked observers’ responses and stored each step of their adjustments. Only a few observers in some blocks at the beginning of the experiment hit the green or red limit, making them highly unlikely to be used as an anchor point.

Subsection “Procedure”:

“Button presses had no effect once observers reached a green endpoint at 200 degrees in hue angle and a red endpoint at 360 degrees of hue angle. At the beginning of each trial, the hue angle of the stimuli was set randomly from 290±45 degrees. We tracked observers’ responses and stored each step of their adjustments. Examination of these data confirmed that they were not using the red or green endpoint as an anchor for their settings (e.g. always moving to the endpoint and then moving a fixed number of steps back).”

d) What is the luminance of the test patch that is being adjusted? Subsection “Apparatus” says the background luminance was 41.85 cd/m2, but later and in the picture, this is described/shown as black, which is surprising unless the ambient in the room was very high luminance.

We apologize for not being clear on the description of our test display and the misunderstanding caused by Figure 1D. As noted above, Figure 1D shows what was presented on the monitor to the observers when they were making the unique yellow settings. The office scene is not the room that observers were tested in; it is the background image displayed on the monitor and that was for giving observers context while they were making adjustments. The black square was 5.7 degrees of visual angle, and it separates the 0.5-degree test patch from the office image (the background). The background luminance was 20 cd/m2, and the luminance of the test patch was 51 cd/m2. We now have added this information to the paper.

Subsection “Procedure”:

“The mean luminance of the background office image was 20 candela/m2.”

“All settings fell along the nominally iso-luminant plane (defined by the LM and S axes, with luminance set to 51 candela/m2) when not wearing the glasses in order to reduce brightness effects on the judgments.”

10) Figure S2, top panel interpretation. The pattern of results is a little hard to interpret. We'd expect the least adaptation for the first setting, so in general these should be lower on the y-axis than the corresponding points in Figure 2. That does not appear to be the case in many instances. See the first group of glasses on settings, for example. Any comment? Are the first settings just really noisy? It might be clearer if each individual setting were plotted, rather than just providing the comparison of the first to the mean.

We thank the reviewer for the observation and agree that the first unique yellow setting made with glasses on in each test would be expected to be greener/lower on the y-axis than the subsequent matches. So, we examined the first match in the first block of each test (to define terms: each five-minute test was comprised of 5 one-minute blocks; each morning or afternoon session had 5 tests with glasses on, separated by 5-10 minutes according to the schedule shown in manuscript Figure 1C). We found that with glasses on, the first match in the first block in all but the first test in each session was unexpectedly larger in hue angle (less green) than the following matches in that block. But the first match in the first block of the first test immediately after putting on the glasses did not differ from other matches.

This pattern is shown in Author response image 1 which plots the first eight individual matches in each of the 5 blocks averaged across days. The linked red dots are the first matches in each block linked across tests, and the rest of the colored lines are the subsequent 2-8 matches in each block. The first matches in the first block of tests 2-5 made with glasses on were much higher on the y-axis than other matches (i.e. red lines in tests 2-5 for block 1 are outliers).

This pattern has a simple explanation: tests 2-5 were conducted after the 5 -10 minutes break, during which observers watched videos or texted on their phones or laptop computers. The first test, however, immediately followed a 'baseline' test with glasses off, where observers were viewing our test display. According to most observers, the screen of the phones or the computers they viewed during the break looked redder than other objects in the environment when they wore the glasses, probably because they had higher luminance than other indoor objects. This was particularly true for white backgrounds while texting. So, after viewing their phone/laptop screen for a period of time, observers likely became more adapted to reddishness than after viewing our test display. Then when observers transitioned from their phone/computer to the test display, they were in a more adapted state, so they set the first unique yellow match to be redder than other matches, which were made after looking at the test display for a while. This effect was not present on the first match in the first block of the first test upon putting on the glasses because there was no transition between different displays before the first test (which was conducted immediately after a baseline test).

Because these 4 points (first match of the first block with glasses on in test 2-5) were apparently taken under different conditions, we removed them when computing the block averages in Figure 2—figure supplement 2. Because our main results figures averaged across all blocks in a test, they were not affected by inclusion of these 4 points to a noticeable extent, and so we have left them unchanged (though we made sure all statistical trends were maintained if the outliers were removed). Note also that, figures that plot only the first test with glasses on (i.e. the scatter plots in Figure 3, and Figure 3—figure supplement 1) are unaffected by these outliers as well, because of the immediate transition from the baseline test. However, our previous plot of the entire time course using the first match was, as the reviewer noted, distorted, as we now know, by these outlier points, and so we have removed it from the manuscript.

Author response image 1
Download asset Open asset
https://doi.org/10.7554/eLife.61179.sa2

Article and author information

Author details

  1. Yanjun Li

    Department of Psychology, University of Minnesota, Minneapolis, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Methodology, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    li000611@umn.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0210-2305

  2. Katherine EM Tregillus

    Department of Psychology, University of Minnesota, Minneapolis, United States
    Contribution
    Conceptualization, Data curation, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-9543-6012

  3. Qiongsha Luo

    Department of Psychology, University of Minnesota, Minneapolis, United States
    Contribution
    Data curation
    Competing interests
    No competing interests declared

  4. Stephen A Engel

    Department of Psychology, University of Minnesota, Minneapolis, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - review and editing
    For correspondence
    engel@umn.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5241-6433

Funding

National Science Foundation (NSF-BCS 1558308)

  • Stephen A Engel

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This study was funded by NSF-BCS 1558308. We thank Victoria Papke and the many other research assistants who helped with this study. Michael A Webster and Rhea Eskew provided valuable discussion, and we thank the three reviewers.

Ethics

Human subjects: Experimental procedures were approved by the University of Minnesota Institutional Review Board (STUDY00002354). All subjects provided written, informed consent before the start of the study.

Senior Editor

  1. Joshua I Gold, University of Pennsylvania, United States

Reviewing Editor

  1. Marisa Carrasco, New York University, United States

Reviewers

  1. David Brainard
  2. Larry Maloney, NYU, United States

Publication history

  1. Received: July 17, 2020
  2. Accepted: December 5, 2020
  3. Version of Record published: December 15, 2020 (version 1)

Copyright

© 2020, Li et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 869
    Page views
  • 94
    Downloads
  • 4
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Yanjun Li
  2. Katherine EM Tregillus
  3. Qiongsha Luo
  4. Stephen A Engel
(2020)
Visual mode switching learned through repeated adaptation to color
eLife 9:e61179.
https://doi.org/10.7554/eLife.61179

Categories and tags

Research organism

Further reading

    1. Neuroscience

    Cholinergic and noradrenergic axonal activity contains a behavioral-state signal that is coordinated across the dorsal cortex

    Lindsay Collins, John Francis ... David A McCormick
    Research Article Updated

    Fluctuations in brain and behavioral state are supported by broadly projecting neuromodulatory systems. In this study, we use mesoscale two-photon calcium imaging to examine spontaneous activity of cholinergic and noradrenergic axons in awake mice in order to determine the interaction between arousal/movement state transitions and neuromodulatory activity across the dorsal cortex at distances separated by up to 4 mm. We confirm that GCaMP6s activity within axonal projections of both basal forebrain cholinergic and locus coeruleus noradrenergic neurons track arousal, indexed as pupil diameter, and changes in behavioral engagement, as reflected by bouts of whisker movement and/or locomotion. The broad coordination in activity between even distant axonal segments indicates that both of these systems can communicate, in part, through a global signal, especially in relation to changes in behavioral state. In addition to this broadly coordinated activity, we also find evidence that a subpopulation of both cholinergic and noradrenergic axons may exhibit heterogeneity in activity that appears to be independent of our measures of behavioral state. By monitoring the activity of cholinergic interneurons in the cortex, we found that a subpopulation of these cells also exhibit state-dependent (arousal/movement) activity. These results demonstrate that cholinergic and noradrenergic systems provide a prominent and broadly synchronized signal related to behavioral state, and therefore may contribute to state-dependent cortical activity and excitability.

    1. Neuroscience

    Early language exposure affects neural mechanisms of semantic representations

    Xiaosha Wang, Bijun Wang, Yanchao Bi
    Research Article Updated

    One signature of the human brain is its ability to derive knowledge from language inputs, in addition to nonlinguistic sensory channels such as vision and touch. How does human language experience modulate the mechanism by which semantic knowledge is stored in the human brain? We investigated this question using a unique human model with varying amounts and qualities of early language exposure: early deaf adults who were born to hearing parents and had reduced early exposure and delayed acquisition of any natural human language (speech or sign), with early deaf adults who acquired sign language from birth as the control group that matches on nonlinguistic sensory experiences. Neural responses in a semantic judgment task with 90 written words that were familiar to both groups were measured using fMRI. The deaf group with reduced early language exposure, compared with the deaf control group, showed reduced semantic sensitivity, in both multivariate pattern (semantic structure encoding) and univariate (abstractness effect) analyses, in the left dorsal anterior temporal lobe (dATL). These results provide positive, causal evidence that language experience drives the neural semantic representation in the dATL, highlighting the roles of language in forming human neural semantic structures beyond nonverbal sensory experiences.

    1. Neuroscience

    Two conserved vocal central pattern generators broadly tuned for fast and slow rates generate species-specific vocalizations in Xenopus clawed frogs

    Ayako Yamaguchi, Manon Peltier
    Research Article Updated

    Across phyla, males often produce species-specific vocalizations to attract females. Although understanding the neural mechanisms underlying behavior has been challenging in vertebrates, we previously identified two anatomically distinct central pattern generators (CPGs) that drive the fast and slow clicks of male Xenopus laevis, using an ex vivo preparation that produces fictive vocalizations. Here, we extended this approach to four additional species, X. amieti, X. cliivi, X. petersii, and X. tropicalis, by developing ex vivo brain preparation from which fictive vocalizations are elicited in response to a chemical or electrical stimulus. We found that even though the courtship calls are species-specific, the CPGs used to generate clicks are conserved across species. The fast CPGs, which critically rely on reciprocal connections between the parabrachial nucleus and the nucleus ambiguus, are conserved among fast-click species, and slow CPGs are shared among slow-click species. In addition, our results suggest that testosterone plays a role in organizing fast CPGs in fast-click species, but not in slow-click species. Moreover, fast CPGs are not inherited by all species but monopolized by fast-click species. The results suggest that species-specific calls of the genus Xenopus have evolved by utilizing conserved slow and/or fast CPGs inherited by each species.