Contrast sensitivity revealed by spontaneous eyeblinks: Evidence for a common mechanism of oculomotor inhibition

Abstract

Spontaneous eyeblinks are known to serve important physiological functions, and recent evidence shows that they are also linked to cognitive processes. It is yet unclear whether this link reflects a crude rate modulation or, alternatively, an automatic and precise process, tightly linked to the low-level properties of sensory stimuli. We have recently reported (Y. S. Bonneh, Adini, & Polat, 2015) that, for microsaccades, the onset and release from inhibition in response to transient stimuli depend systematically on the low-level stimulus parameters. Here we reanalyzed our previous data for both microsaccades and eyeblinks for observers with sufficient blinking (>10% of trials, 18 of 23 observers tested) who watched and silently counted sequences of Gabor patches at 1 Hz with varied contrast and spatial frequency. We found that spontaneous eyeblinks, although less frequent, were similar to microsaccades in their modulation pattern in response to transient stimuli, demonstrating inhibition and rebound, which were dependent on the contrast and spatial frequency of the stimuli. The average blink response time, measured as the latency of the first blink following its release from inhibition, was longer for lower contrast and higher spatial frequency. Importantly, it was highly correlated with a similar measure for microsaccades as well as with psychophysical measures of contrast sensitivity. These results suggest that both eyeblinks and microsaccades are linked to the same inhibitory mechanism that presumably turns off oculomotor events while processing previous events and generates a rebound effect upon its release. The onset of both eyeblinks and microsaccades may thus reflect the time course of this mechanism and the associated cognitive process.

Introduction

Our eyes are never at rest, exhibiting a rich repertoire of ocular dynamics of saccades, microsaccades, pursuit, drift, tremor, pupil dilation, and eyeblinks. Eyeblinks, which occur spontaneously at a low rate (e.g., 15–20 per min; R. K. McIntire, Macy, Seo, Nelson, & Kolbe, 2014), are less frequent than saccades (a rate of approximately three saccades per second; Otero-Millan, Macknik, Langston, & Martinez-Conde, 2013) but several times more frequent than are needed for moistening the eyes (Doane, 1980), suggesting an additional role related to visual processing. Although this suggested role is largely unknown, there is accumulating evidence that links eyeblinks to cognitive processes and states in terms of both the average rate and specific timing (Bacher & Smotherman, 2004). The evidence can be divided into three categories: (a) blink rate modulation by the sustained cognitive state, such as the level of vigilance; (b) suppression of blinking during important visual information; and (c) blink rate modulation time-locked to specific events, not necessarily visual. The effect of a sustained cognitive state can be demonstrated, for example, by the finding that reduced vigilance during a prolonged time on a task increased the average eyeblink rate and duration (L. K. McIntire, McKinley, Goodyear, & McIntire, 2014), presumably because of reduced inhibitory strength. Similarly, the blink rate increased over time for a continuous performance task in which observers were expected to suppress their blinking for an anticipated target presentation—more so in people with attention-deficit/hyperactivity disorder (Fried et al., 2014, figure 11d, e). On the other hand, the blinking rate decreased when cognitive demand in a visuomotor and memory task was increased (Veltman & Gaillard, 1998). 

More relevant to the current study is the evidence for event-related modulation of blinking. A number of studies have demonstrated that people tend to inhibit eyeblinks during exposure to important visual information, presumably to minimize the loss of such information (Fried et al., 2014; Fukuda, 1994; Ichikawa & Ohira, 2004; Nakano, Yamamoto, Kitajo, Takahashi, & Kitazawa, 2009; Shultz, Klin, & Jones, 2011). Nakano et al. (2009) found that the timing of blink generation is synchronized between individuals at the attentional breakpoints of video stories. Notably, Shultz et al. (2011) found that while viewing video scenes, toddlers dynamically adjusted the timing of their blink inhibition with respect to salient onscreen events. However, event-related modulation of blinking goes beyond avoiding the loss of visual information. For example, the timing of blink generation was synchronized between individuals at the attentional breakpoints of speech (Nakano & Kitazawa, 2010). Moreover, eyeblinks were inhibited prior to stimulus onset in a discrimination task for purely auditory stimuli (Fukuda, 1994) and tended to occur immediately before or soon after a vocal response in a Stroop test (Oh, Han, Peterson, & Jeong, 2012). These and other findings led Nakano, Kato, Morito, Itoi, and Kitazawa (2013) to suggest that eyeblinks are not just optimally inhibited; they are also actively involved in the release of attention. They provided fMRI evidence for blink-triggered brain activation patterns, which are consistent with this hypothesis (Nakano et al., 2013). 

Saccades, large or small, like eyeblinks, are oculomotor behaviors that generate transient visual stimulation on the retina at discrete points in time. Microsaccades are small saccades that occur during fixation, and like spontaneous eyeblinks, they appear stochastically. They are known to be inhibited in response to sensory transients with a time course that depends on the stimulus parameters and attention (Engbert, Mergenthaler, Sinn, & Pikovsky, 2011). This phenomenon, known as “microsaccade inhibition,” has been widely used in recent years to study attention and cognitions (for a review, see Rolfs, 2009). 

We have recently reported (Y. S. Bonneh et al., 2015) that, for microsaccades, the onset and release from inhibition in response to transient stimuli depend systematically on the low-level stimulus parameters. In these experiments, our goal was to obtain involuntary measures that do not require any behavioral response or explicit perceptual decisions and thus could be used in the future with noncommunicating individuals. The general methodology used was therefore based on passive viewing of a regular (every second) slide show of visual onsets having different properties (contrast and spatial frequency) mixed in a random order. The results showed that the microsaccade response time (RT), as measured by the latency of the first microsaccade relative to the stimulus onset following its release from inhibition was sensitive to the contrast and spatial frequency of the stimulus and could be used to extract a contrast response function without the observers' response. Moreover, the contrast detection thresholds, measured behaviorally for different spatial frequencies, were highly and positively correlated with the microsaccade RT measured at high contrast. 

In the current study, we reanalyzed our data for eyeblinks. Initially, we did not consider eyeblinks to be of interest for these data because they were infrequent and their temporal precision was unclear. However, when properly computing the blink onset times, we found that, although less frequent, eyeblinks were very similar to microsaccades regarding their temporal entrainment and response pattern as a function of the low-level properties of the stimuli of contrast and the spatial frequency. 

Methods

The methods are identical to those used in our previous paper on microsaccades (Y. S. Bonneh et al., 2015) except for details regarding the detection of blinks. We will repeat these methods here for convenience. 

Subjects

Overall, 18 observers (ages 25–50) with normal or corrected-to-normal vision participated in the experiments: 14 in the contrast experiment, 13 in the spatial frequency experiment, and seven in the contrast detection experiment. Additional observers, including the first author (n = 7, ∼30%), were fully tested but were excluded at the data analysis stage due to a low blink rate (<10% of trials, see below). All participants signed an informed consent form as required and approved by the Helsinki Committee. 

Apparatus

Stimuli were displayed on a 22-in. CRT monitor using an in-house-developed platform for psychophysical and eye-tracking experiments (PSY) developed by the first author, running on a Windows PC. The video format was true color (RGB) at a 100-Hz refresh rate with 1024 × 768 pixel resolution occupying a 33.4° × 25.4° area. The background luminance was 40 cd/m², and luminance values were gamma-corrected. For presenting very low-contrast stimuli when testing the detection threshold, luminance dithering (2 × 2 pixels) was applied. The sitting distance was 0.6 m, and all experiments were administered in dim light. Eye movements were recorded monocularly with an Eyelink 1000 infrared system (SR Research, Ontario, Canada) with a sampling rate of 500 Hz. Movements of the head were limited by a chin and forehead rest. Recording was performed from the right eye although viewing was binocular. A standard 9-point calibration was performed before each session although the absolute position of the eyes was never used and was not important in this study. 

Stimuli and procedures

In three eye-tracking experiments, the observers passively viewed a “slide show” of repeated visual onsets while they silently counted the displayed items and reported the total number at the end of the run. In these experiments, vertical Gabor patches were flashed for 100 ms at fixation at an exact 1-Hz repetition rate. A static small (0.12° in diameter) fixation point was constantly presented. Each run consisted of 100 presentations lasting for 100 s, and each observer was tested in multiple runs as specified for each of the following experiments. In the contrast experiment, the Gabor patches had a spatial frequency of 3 c/° and a Gaussian envelope of σ = 4 cycles with varied contrast of 0.8%–50% in factor two jumps presented in random order. Among the 14 participants, 10 were tested in five runs each, and four were tested in 10–20 runs each. In the spatial frequency experiment, the Gabor patches had a fixed contrast of 25% with varied spatial frequency from a predefined list of 0.2, 1, 2, 4, 6, and 8 c/° with a fixed envelope of σ = 2.7°, presented in random order. Among the 13 participants, five were tested in five runs each, four in 10 runs each, and four observers in ∼20 runs each. 

In an additional contrast detection threshold experiment, the observers were tested for determining the psychophysical contrast detection threshold as a function of spatial frequency, also known as the contrast sensitivity function. The detection thresholds were measured in a subset of the observers that had sufficient (>10% of trials) blinking (n = 7) for the same stimuli as in the spatial frequency experiment, using a standard temporal two-alternative forced choice staircase procedure (one-up, three-down) applied to target contrast as was previously done (e.g., Y. Bonneh & Sagi, 1998); this is described in more detail in our previous study (Y. S. Bonneh et al., 2015). Participants were tested on three to four runs each. 

Data analysis

Blink detection

Eyeblinks were detected with a simple algorithm tuned to the way the blinks were represented by the eye tracker used. Blink periods were first defined as zero pupil size, producing approximate events of eye-close (transition to zero) and eye-open (change from zero). In order to improve the accuracy of these approximations, the vertical trace was further analyzed in a local window of 100 ms prior to an approximate onset and 150 ms after an approximate offset. This was done because blinks were typically preceded by vertical movement of the eyes (known as Bell's phenomenon; see Costela et al., 2014, for measures of this effect). The blink onset was then defined as the time of the first value of the vertical trace larger than the median of the local window. Similarly, the blink offset was defined as the time of the last value smaller than the median of the local window. 

Microsaccade detection

Microsaccades were detected for comparison with eyeblinks using the algorithm introduced by Engbert and Kliegl (2003) as in our previous study (Y. S. Bonneh et al., 2010). All microsaccades pooled together per experiment were verified to obey the “common sequence” pattern (as in Y. S. Bonneh et al., 2010). See our previous paper for more details. 

Epoch extraction

Epochs were extracted and time-locked to the stimulus onset with one epoch per stimulus presentation in a time range of 0.5 s before until 1 s after stimulus onset. An exception to this is the 4-s nonoverlapping epochs extracted for the entrainment demonstration (Figure 1c). In analyzing microsaccades, periods of missing data, such as during eyeblinks, were locally discarded from further analysis with an additional margin of 0.05 s without discarding the whole epoch. 

Figure 1

 

Basic properties of spontaneous blinking. Data were extracted from the two experiments in a and b and from the frequency experiment only in c. (a) Blink duration distribution across subjects and the two experiments. (b) Blink rate averages for the 27 different data sets, collected from 18 different observers. (c) Entrainment of blinks compared with microsaccades by repeated stimuli (denoted by a gray bar) at a fixed 1-Hz rate. Rate modulation functions expressed as percentage from average were first averaged for each observer across long (4-s) nonoverlapping epochs (all frequencies <8 c/°, n = ∼350 per observer) and then averaged across observers (n = 13) with error bars showing 1 SE across observers. Note the similarity between microsaccades and eyeblinks, especially at the rising period, and the similarity between the four independent repetitions presented here to illustrate entrainment.

Microsaccade and blink rate analysis

The rate modulation function for both eyeblinks and microsaccades was calculated by convolving a raw rate estimate of one microsaccade/blink per sample duration at the time of onset with a Gaussian window having a sigma of 50 ms as in our previous studies (Y. S. Bonneh et al., 2015; Y. S. Bonneh et al., 2010). We also tested the causal kernel previously used by Widmann, Engbert, and Schroger (2014), which produced similar results, but preferred to keep the previous method for consistency. The rates were averaged across epochs within observer and then across observers to compute the event-related modulation of microsaccades and blinks with equal contribution from each observer. In some cases, we computed rate modulation functions as the percentage change in rate from average, computed per epoch (dividing the rate modulation by its average, subtracting one and multiplying by 100), averaged across epochs and then across observers. 

Microsaccade and blink RT

Microsaccade RT and blink RT were computed per epoch as the latency of the first microsaccade/blink after stimulus onset in different time windows. These windows were selected to separately explore early (around 0–200 ms) and late (above 200 ms) processes as indicated by the rate modulation functions and were slightly fine-tuned to minimize the average standard error in each experiment (see below for the statistical tests that take this tuning into account). Epochs with no microsaccades/blinks in the specified window were excluded from the average. An additional microsaccade/blink “hit rate” measure was computed as the percentage of epochs with at least one microsaccade/blink in the specified time interval. In computing error bars for the RT values averaged across subjects, we applied the Cousineau and Morey method (Morey, 2008), which is appropriate for within-subject effects, allowing for a better “rule of eye” for significance. In this method, data were first averaged and normalized (demeaned) for each subject, then averaged across subjects with error bars reflecting 1 SE of these means, then adjusted by adding the grand average across all conditions (e.g., contrast levels) and subjects. The results of this procedure is identical to normal average except for the error bars, which are usually smaller and make a better representation of the within-subject effects. We then applied the correction factor for the error bars (square root of n/[n − 1]; n is the number of conditions). For n = 7 or 8 (number of conditions in our experiments), this implies a very small correction (7%–8%) for the error bars. 

Assessment of the statistical significance

We used nonparametric permutation tests (Efron & Tibshirani, 1998) to test the dependence of the blink RT and blink hit rate on the contrast and spatial frequency. For each test, we randomly permuted (1,000 permutations) the labels of the observations (i.e., the contrast or spatial frequency of each epoch) and recalculated group average blink RT and hit functions in exactly the same way as the standard computation (averaged and demeaned per observer and then averaged across observers, e.g., as in Figure 2b). We then quantified the effect by the correlation coefficient R between blink RT and Log(contrast) for the contrast experiment and between blink RT and spatial frequency of 2 c/° and above for the spatial frequency experiment. We then computed the p value as the fraction of permutations in which the original correlation was exceeded by the correlation of the permuted data (in absolute values). In order to account for the multiple intervals and interval optimization in computing blink RT and hit rate, we repeated the permutation test described above for all possible valid intervals in 50-ms steps. For the main “late” effect of release from inhibition, the interval had to start in the range 150–400 ms to have duration in the range 300–800 ms and end no later than 800 ms, which produced 34 possible intervals. For the “early” effect of the onset of inhibition, the end point was limited to 400 ms, and the starting point was in the range 0–100 ms, thus consisting of 16 possible intervals. We computed both the p values for the optimal interval as well the p values extracted from the combined effect of all possible intervals. We report the latter but note that these were similar. 

Figure 2

 

The effect of contrast on the spontaneous blinking pattern. Stimuli consisted of vertical Gabor patches (3 c/°) with varied contrast, briefly flashed (100 ms, denoted by a gray bar in a), presented at fixation at 1 Hz in passive viewing. (a) Blink rate modulation functions. The upper panel shows two raster plots of blink onsets for samples of 300 epochs for low (1.6%, in red) and high (50%, in blue) contrast patches, one line per epoch and one dot per blink. The lower panel shows the corresponding rate modulation functions for different contrast levels (those corresponding to the upper panel are in red, blue), time-locked to the stimulus onset (time 0). The data were first averaged across epochs within observer and then across observers (n = 14). Error bars (50% and 1.6% only for clarity) denote 1 SE of the mean across observers, down-sampled for clarity. (b–d) The effect of contrast on blink inhibition and release time and strength. (b) Blink RT, computed as the average onset time of the first blink in the specified time window (250–750 ms), corresponding to the release from inhibition. The ms-RT computed in the same way is shown for comparison. (c) Blink hit rate (percentage) in the same time window. (d) Blink hit rate in an early time window (50–250 ms), corresponding to the onset of inhibition. Contrast is plotted on the x-axis in log units, ranging from 0.8% to 50%. In b through d, values were first averaged and normalized within observers, then averaged across observers and readjusted by adding the grand average. Error bars denote 1 SE across observers. The results of a linear regression (for contrast ≥1.6%) are also shown (|R| > ∼0.9 in all cases), with p values computed using a nonparametric permutation test (see Methods). As shown, with increased contrast, blink inhibition is released earlier (b), it has a stronger rebound effect (c), and is stronger immediately after stimulus onsets (d).

Results

The results are summarized below. We first describe the basic blink properties and blink entrainment, then the results for the contrast and spatial frequency experiments, and finally, we compare eyeblinks, microsaccades, and psychophysical thresholds. 

Blink properties

Figure 1 summarizes the basic blink properties and entrainment. As shown in Figure 1a, blink duration, collapsed across all observers and experiments pooled together (totaling 6,800 blinks), appeared normally distributed with a peak around 270 ms. Interestingly, there were some very short blinks (<50 ms), which could possibly be an artifact of the eye-tracker system, but they had no impact on the results and were not discarded from the analyses. Figure 1b shows that the average blink rate varied dramatically across observers, mainly in the range of approximately five to 25 blinks/min with two extreme cases (two and 33 blinks/min). This is equivalent to one blink in 10%–50% of the trials (presented at 1 Hz) on average. An additional five observers (not shown) were discarded from the analysis because of low blink rates (<10% of trials) with one exception of a low blink rate not discarded (see below). Figure 1c shows the effect of blink entrainment by the repeated stimuli (1 Hz) in the spatial frequency experiment collapsed across all frequencies below 8 c/° within observer and then averaged across observers. The data from the contrast experiment were not included here due to their larger temporal variability. Rate modulation functions, presented as the percentage change in rate from average (see Methods), were averaged in long (4-s) nonoverlapping epochs, thus demonstrating four repetitions computed from independent measures. Figure 1c shows that both eyeblinks and microsaccades were entrained by the periodic stimuli with inhibition that starts long before the onset of the stimulus, thus reflecting anticipation. Although the eyeblinks and microsaccades showed a similar time course, the actual microsaccade peak rates were approximately seven times larger and did not reach zero rates whereas the blink rates approached zero before and immediately after stimulus onset. 

Contrast

The results for the contrast experiment are plotted in Figure 2. Figure 2a plots the blink rate modulation functions for different target contrast levels: each first averaged across epochs within observer and then averaged across observers (n = 14). The upper panel shows examples of the data that underlie the rate modulation functions in terms of a raster plot with a dot for every blink. All rate modulation functions share the same shape: (a) early inhibition that starts prior to stimulus onset, possibly due to the temporal regularity (1 Hz), which allows for precise anticipation and does not depend on the stimulus; (b) stimulus-dependent inhibition that reaches the minimum (maximal rate inhibition) at 150–200 ms after onset; (c) release from inhibition that strongly depends on contrast, both in terms of the magnitude of the release peak and, to a lesser extent, its latency. Note that higher contrast results in faster onset and stronger rate inhibition as well as a faster release of inhibition and a stronger magnitude of that release. This is manifested by the peaks of the rate modulation functions, which were in the range 300–500 ms after stimulus onset, depending on contrast. We used this analysis for comparison with other studies that used rate modulation functions and to motivate the second analysis that followed, which we found more accurate and reliable. We therefore did not apply statistics to the rate modulation functions, but the error bars indicate the strength of the measured effects (only for the difference between the highest and lowest contrast levels where error bars are shown). 

The results of the second type of analysis are shown in Figure 2b through d. Two measures were computed: the blink RT (Figure 2b) and the blink hit rate (Figure 2c, d), i.e., the percentage of epochs including one or more blinks computed in two time windows: early (50–250, Figure 2d) and late (250–750, Figure 2b, c) (see Methods). Data were averaged across observers and 1 SE error bars computed with normalization (demeaned within observer, averaged, and then grand average added; Morey, 2008, see Methods), The time windows for the analysis were selected to capture the onset and release from inhibition observed in the blink rate modulation functions (Figure 2a) with slight fine-tuning to minimize the average standard error across all contrast levels for all observers together (see Methods). We noted that optimizing these time windows per observer could further reduce variability. 

The results for the blink RT (Figure 2b), reflecting the release from inhibition at a late time window of 250–750 ms, show a monotonic decrease as a function of contrast expressed in log units, which starts from 1.6% and could be approximated by a linear regression with R = −0.92 and p = 0.004 computed using a nonparametric permutation test across all possible time windows (see Methods). This computation was applied to all correlations reported below. Note that only trials with blinks at the specified time window were taken into account in computing blink RT. The percentage of these trials is shown in Figure 2c as the hit percentage (blink hits) at that window, which ranged from ∼8%–16% and increased as a function of log contrast with a linear regression fit of R = 0.96 and p = 0.001 (permutation test). Figure 2d plots the blink hit percentage in the early time window (50–250 ms), showing a monotonic decrease from 2.5% to 1% as a function of log contrast, which could be approximated by a linear regression with R = −0.89 and p = 0.007 (permutation test). The results for the blink RT in the early window (not shown) did not show a consistent dependency on contrast, possibly because of the sparse data of a very low blink rate. Figure 2b also plots the microsaccade RT (ms-RT), computed for the same data. As shown, the ms-RT was similar to the blink RT, but it was slightly faster for some contrast levels (by 20–30 ms). We noted that the ms-RT shown here is based on data that are only partially identical to our previous study (Y. S. Bonneh et al., 2015) (two participants omitted because of blink rate too low; <10% of trials, see Methods; two new added). 

Spatial frequency

The results for the spatial frequency experiment are plotted in Figure 3 with analysis similar to the contrast experiment above. Figure 3a plots the blink rate modulation functions for different target frequencies, each first averaged across epochs within observer and then averaged across observers (n = 13). As shown, the rate peaks decreased and (to a lesser extent) the latency increased with higher spatial frequency. Note that the blinking was almost completely inhibited prior to stimulus onset. The results for the blink RT (Figure 3b), reflecting the release from inhibition at a late time window of 250–600 ms, show a roughly linear increase above 2 c/° with increasing frequency (from ∼380 ms to ∼430 ms) and a slight increase also for low frequencies below 2 c/°. The linear increase from and above 2 c/° was highly significant with R = 0.97 and p = 0.0004 in the permutation test (see Methods). We noted that the time window end point, which we used for spatial frequency was shortened by 150 ms (600 ms instead of 750 ms) to properly focus on the release from inhibition, which was faster for frequency compared to contrast (compare the rate modulation functions in Figures 2a and 3a). This choice was tested for significance by the nonparametric significance test that explored all possible intervals (see Methods). 

Figure 3

 

The effect of spatial frequency on the spontaneous blinking pattern. Stimuli consisted of vertical Gabor patches at high contrast (25%) with varied spatial frequency but with fixed envelopes, briefly flashed (100 ms), presented at fixation at 1 Hz, in passive viewing. Examples of low and high spatial frequency stimuli are shown in a with the shaded bar denoting the stimulus presentation duration. (a) Blink rate modulation functions for different spatial frequencies, time-locked to the stimulus onset (time 0), averaged across epochs within observer and then across observers (n = 13). Error bars (1 and 8 c/° only for clarity) denote 1 SE of the mean across observers, down-sampled for clarity. (b) Blink RT, computed as the average onset time of the first blink in the specified time window (250–600 ms). Values were first averaged and normalized within observers then averaged across observers and readjusted by the grand average. Error bars denote 1 SE across observers. The ms-RT, computed similarly, is shown in cyan for comparison. Note the gradual increase in blink and microsaccade RTs from 2 to 8 c/°, and below 2 c/°. (c) Blink hit rate around the release peaks (250–600 ms), which appears to mirror the blink RT curve. No such effect was found for microsaccades. For both blink RT (b) and hit rate (c), the values were linearly correlated with the spatial frequency for 2 c/° and above, with p values shown for the nonparametric permutation test. Measures for the initial period of blink inhibition (not shown) did not result in consistent effects.

Figure 3c plots the percentage of epochs with one or more blinks in the same window of 250–600 ms in which the blink RT was computed. The values were low, ranging from 8% to 16% of the trials. However, the effect of spatial frequency clearly mirrors the effect of latency (blink RT) with a peak at 2 c/°, a roughly linear decrease above 2 c/°, and a small decrease for low spatial frequencies below 2 c/°. A nonparametric permutation test (see Methods) revealed that the linear fit from and above 2 c/° for the blink hits was highly significant: R = −0.94, p = 0.003. In Figure 3b, we also plotted the ms-RT, computed for the same data. As shown, the ms-RT was similar to the blink RT, but it was slightly faster (by 10–20 ms). We noted that the ms-RT shown here is based on data that are only partially identical to our previous study (Y. S. Bonneh et al., 2015) (five participants were omitted due to their low blink percentage). 

Blinks compared to microsaccades

Because, according to our data, both the blinks and microsaccades exhibited a similar event-related response, we made a detailed comparison that is summarized in Figure 4. Figure 4e plots the event-related rate modulation of microsaccades and blinks for three spatial frequencies (1, 4, and 8 c/°). As shown, the microsaccade rate peaks are approximately five times larger than those for blinks with roughly similar peak latency and a similar decrease in magnitude for a higher spatial frequency (e.g., 8 c/° in red). One observed difference is that the blink rate peaks for the higher spatial frequencies (4 and 8) are “blurred” or flat with unclear latency. This could be due to the smaller number of trials with blinks at high frequencies (8% at 8 c/°, Figure 3c) or, alternatively, due to a larger latency variability for the blinks. 

Figure 4

 

Blinks compared to microsaccades. (a–d) Correlation plots between blink and microsaccade RTs for the contrast (a–b) and spatial frequency (c–d) experiments with correlation computed across observer averages (b, d) and observer per condition (a ,c). (e) We superimposed the rate modulation functions of microsaccades and blinks for three spatial frequencies (1, 4, and 8 c/°; data from all observers pooled together). Note that the microsaccade rates are by far larger (more than four times larger). (f) A scatter plot of correlation coefficients computed for each observer between the blink RT and contrast (red) and spatial frequency (blue, only frequencies ≥2 c/°). The plot demonstrates the variability and precision of the measures across observers, assuming a high linear correlation, positive for frequency and negative for contrast. Note the three complete “outlier” observer measures marked in bold; all from the contrast experiment, one for ms-RT, one for blink RT, and one for both.

Figure 4a and b and c and d plot the correlation between blink RT and ms-RT for the contrast and spatial frequency experiments, respectively. The left column (Figure 4a and c) shows scatter plots with each point corresponding to the average RT for one observer and for one contrast (Figure 4a) or spatial frequency (Figure 4c). A linear regression analysis (“robust fit”) revealed a significant correlation of R = 0.28 (p = 0.01 for significant correlation, two outliers excluded) and R = 0.49 (p = 0.0002) for the contrast and spatial frequency data, respectively. A much higher correlation was found for the averages of these data across observers as shown for both contrast (R = 0.78 [0.99 without one outlier], Figure 4b) and spatial frequency (R = 0.98, Figure 4d). 

We further plotted the correlation coefficients (R) measured in the two experiments for different observers in a scatter plot that compares blink RT and ms-RT data with a point per observer (Figure 4f). This plot allows one to assess the relative quality and accuracy of the blink RT and ms-RT measures for contrast and spatial frequency, assuming that a higher correlation reflects more accurate measures. The plot shows comparable but somewhat lower correlations for the blink RT (positive for frequency and negative for contrast) compared with microsaccades with few outliers. These differences are likely to be due to the smaller frequency of blinks. 

Finally, we used the spatial frequency experiment data set to investigate the relationship between microsaccades and eyeblinks within epochs, following stimulus onset (0–1000 ms time range). Among all epochs from all observers pooled together (n = ∼20,000), ∼75% had only microsaccades, ∼10% had only eyeblinks, ∼15% had none, and ∼10% had both. We calculated the rate modulation functions for microsaccades from epochs that contained no eyeblinks (75%) and for eyeblinks from epochs that did not contain microsaccades (about half of the epochs with eyeblinks). In both cases, we found a similar rate modulation function to that obtained with the whole data set (similar to Figure 4e). In examining a shorter interval of 0–600 ms, which is more relevant to the observed rate modulation, we found that only 2% of the epochs had both microsaccades and eyeblinks in this interval, and excluding these epochs did not alter the effect of spatial frequency on the rate modulation function as well as RTs for both eyeblinks and microsaccades. In general, the two oculomotor behaviors of microsaccades and eyeblinks appear to be mutually exclusive and independent. 

Blink RT versus behavioral measures of contrast sensitivity

We compared the blink RT as a function of spatial frequency to the psychophysical contrast detection threshold (see Methods) for a subset of the observers (n = 7) as we previously did for microsaccades. The results are summarized in Figure 5. In Figure 5a, we compare the psychophysical thresholds averaged across observers (in log units) with the blink RT by expressing both in z values (demeaned and divided by standard deviation per observer, then averaged across observers). As shown, the two plots are roughly similar except for the low spatial frequency of 0.2 c/° (which we excluded from the z-score calculation of the other dots to allow optimal scaling of the results). Comparing Figure 5a to Figure 3b, which is the same plot for the full set of observers (n = 13) revealed that the deviant measure at 0.2 c/° is likely to be a measurement outlier. Moreover, other than 0.2 c/°, the results are also similar to our previous results with microsaccades although microsaccades showed a better and almost perfect fit to the psychophysical data (Y. S. Bonneh et al., 2015). A more detailed comparison is provided in Figure 5b, showing a scatter plot of these z values with one point per observer and frequency, excluding the low spatial frequency (0.2 c/°), showing a relatively high correlation (R = 0.73). This correlation implies that faster blink RT corresponds to a better detection threshold. 

Figure 5

 

Blink RT compared with the psychophysical contrast detection threshold in the spatial frequency experiment. The detection threshold was measured in a standard two-alternative, forced choice staircase paradigm for a subset of the observers, which also showed a significant rate of blinking (n = 7). Values were first averaged and normalized within observers, then averaged across observers, and finally readjusted by adding the grand average. Error bars denote 1 SE across observers. (a) The blink RT expressed in z values (subtracting mean, dividing by standard deviation, excluding 0.2 c/°), superimposed on a z-value plot of the threshold data from the same observers. Note the similarity between the psychophysical threshold and the blink RT measured for high-contrast stimuli in passive viewing except for 0.2 c/°. (b) A correlation plot of the individual observers' z-value data from a, excluding the deviant point of 0.2 c/°, comparing detection threshold and blink RT. Each point represents one observer at one spatial frequency. Note the highly significant correlation (R = 0.73).

Discussion

Eyeblinks, like microsaccades, are oculomotor behaviors that generate transient visual stimulation on the retina in discrete points in time. We have recently reported (Y. S. Bonneh et al., 2015) that for microsaccades the onset and release from inhibition in response to transient stimuli depend systematically on the low-level stimulus parameters, such as contrast and spatial frequency, and on contrast sensitivity. In the current study, we found that exactly the same happens for eyeblinks. We showed this by reanalyzing our data for eyeblinks from the two main experiments in which observers watched a slide show of Gabor patches at 1 Hz with varied contrast and spatial frequency. In the first experiment, we found that the eyeblink inhibition and the rebound effect on its release were stronger and that the latency of its release (blink RT) was shorter with increased contrast, showing an inversely linear relation to Log(contrast) (Figure 2). In the second experiment, we found that the blink RT increased as a function of spatial frequency for 2 c/° and above (Figure 3). Importantly, it was highly correlated with the microsaccade RT (Figure 4) as well as with the detection threshold measured psychophysically for different spatial frequencies (Figure 5). The results suggest a generalized and common oculomotor inhibition mechanism as discussed below. 

The basic properties of eyeblinks

The eyeblink data, which we collected under conditions of repetitive (1 Hz) stimulation, were generally similar in properties to previous eyeblink studies (e.g., R. K. McIntire et al., 2014) in terms of blink duration (an average duration of 250–300 ms) and blink rate (∼15–25 blinks/min) as shown in Figure 1. Importantly, the blink rate largely varied across individual observers and was low (<10% of trials with eyeblinks) for five observers, including the first author, which were discarded from the data analysis. A large variability in blink rates was also reported in previous studies (e.g., Nakano, Kuriyama, Himichi, & Nomura, 2015). Despite this large variability, the computed properties, including blink duration as well as blink RT changes as a function of stimulus contrast and spatial frequency, were similar and produced reliable group averages as shown in Figures 2 and 3. This suggests that, even for observers with very low spontaneous blink rates, one might be able to obtain reliable blink RT measures by simply collecting more data. Note, however, that only truly spontaneous blinks produce reliable results because attempts to intentionally increase the blink rate by the first author failed to produce meaningful results. 

Eyeblink entrainment is another basic property found in the current study with a very regular rate modulation, similar but not identical to the microsaccade rate entrainment (Figure 1c). This entrainment implies precise temporal anticipation. The rate modulation entrainment of eyeblinks was similar to that of microsaccades (Figure 1c) with some notable differences including peaks that were seven times smaller (see Results). 

Alternative measures for blink and microsaccade inhibition

In our previous study (Y. S. Bonneh et al., 2015), we investigated different measures for microsaccade inhibition by computing both rate modulation functions as well as ms-RT measures in early and late temporal windows, corresponding to the onset and release from inhibition, respectively. Microsaccade rate modulation functions have been widely investigated, focusing on the release from the inhibition part (e.g., Rolfs, Kliegl, & Engbert, 2008; Widmann et al., 2014). We show this type of analysis for eyeblinks in Figures 2a and 3a. Following our previous microsaccade study, we also extracted quantitative measures based on averaging the exact latency of eyeblink onsets (blink RT) in specific temporal windows. We found consistent results for the release from inhibition (Figures 2b and 3b). However, unlike for microsaccades, we could not extract a measure for the onset of eyeblink inhibition because of the very low rate of eyeblinks around the stimulus presentation in both the contrast and spatial frequency experiments. Nevertheless, we show that this stimulus-specific inhibition indeed exists for contrast by analyzing the rate in the early inhibitory temporal window (50–250 ms, Figure 2d). We show that a measure of early eyeblink inhibition, although based on very low rates (∼1% of trials), is significantly modulated by contrast with stronger inhibition attained with higher contrast (Figure 2d). However, no such consistent effect was found for spatial frequency (data not shown) probably because of the low blink rate. An additional discussion of the alternative measures, including choosing a temporal window that applies to eyeblinks as well as to microsaccades appears in our previous paper (Y. S. Bonneh et al., 2015). 

Blink RT versus ms-RT and psychophysical measures of contrast sensitivity

In the current study, we followed our previous microsaccade study (Y. S. Bonneh et al., 2015) and compared blink RT measures for high-contrast Gabor patches with different spatial frequencies to the corresponding detection thresholds in the same observers although with a slightly smaller sample (n = 7, two excluded for low blink rates). There was one deviant measure of blink RT at the lowest spatial frequency (0.2 c/°), which was inconsistent with the contrast detection threshold, the ms-RT, and the blink RT for the whole group (Figure 5a). We have no explanation for this discrepancy other than the existence of a measurement outlier. Note that our behavioral measures of contrast sensitivity, with highest sensitivity at 2 c/°, are compatible with previous studies on Gabor patches (Peli, Arend, Young, & Goldstein, 1993). When we excluded the lowest spatial frequency deviant point, we obtained a high correlation (R = 0.73) between blink RT and detection thresholds expressed in log units (Figure 5b), but ms-RT had a higher correlation (R = 0.87, not shown). 

In comparing blink RT with ms-RT (Figure 4), we found that the detailed correlations (per observer and contrast or frequency) were significant but relatively low whereas correlating the averages across observers produced high correlations (|R|> = ∼0.9). Overall, ms-RT appears to be more precise than blink RT does because it produced higher correlations with frequency and contrast within observers (Figure 4f), but this most likely reflects the approximately seven times higher rate of microsaccades as compared with eyeblinks. If this is the case, then this difference could potentially be overcome by collecting more data. Another notable difference between the blink and microsaccade RT measures is that ms-RTs were generally faster by 10–30 ms (Figures 2b and 3b) but with some exceptions. It is not clear whether this reflects an inaccuracy in estimating the blink onset, and this requires further investigation. An additional discussion of the comparison between the oculomotor and behavioral measures, which refers to microsaccades but also applies to eyeblinks, appears in our previous paper (Y. S. Bonneh et al., 2015). 

The effect of spatial frequency: A functional interpretation

Our results of slower microsaccade and blink RT for higher spatial frequency are consistent with the magno-parvocellular distinction, according to which low spatial frequencies are processed more quickly by the magnocellular pathway as compared with high spatial frequencies that are processed more slowly by the parvocellular pathway (Merigan & Maunsell, 1993). Interestingly, the difference between the oculomotor RT for high and low spatial frequency is consistent with a detailed object recognition model based on predictive top-down computation (Bar, 2004). Accordingly, fast processing of the low spatial frequency information is used to provide initial predictions about the most likely interpretation of the input. These predictions are then back-propagated to resolve ambiguities and expedite the processing of the slowly arriving high spatial frequency information. The measured time lag between the high and low frequency responses of this object recognition system was found to be 50–80 ms for a difference factor of ∼5 in spatial frequency (Bar et al., 2006), which is roughly consistent with our results (Figure 3b shows a speed difference of ∼50 ms between 2 and 8 c/°). Note that our measures refer to physically defined spatial frequencies whereas the object recognition studies often refer to the relative frequencies expressed in cycles per picture (e.g., Bar et al., 2006). Interestingly, our finding of a linear relationship between spatial frequency and blink RT (for 2 c/° and above) suggests that regardless of the absolute frequency, the relative frequency will determine the latency advantage of the lower frequency and could thus be used to constrain the high spatial frequency information that arrives later. Finally, it is interesting to consider the expected blink and microsaccade RTs in response to natural objects that contain a wide range of spatial frequencies. One possibility is that the oculomotor inhibition will last until the input is completely resolved, including the interactive integration of low and high spatial frequency information. Alternatively, the oculomotor RT is determined by the fastest channel and thus should correspond to its response to the lower spatial frequencies in the image. 

Comparison to previous eyeblink studies

The finding that, by just analyzing the onset times of spontaneous eyeblinks during passive viewing of a stream of events, one can extract a relative measure of contrast response and contrast sensitivity is surprising and novel. Previous studies did not compute stimulus-triggered eyeblink rate modulation functions and latency measures as a function of the quantitative measures of stimulus strength, but there are few studies that have analyzed event-related responses. Fukuda (1994) computed eyeblink rate modulation in response to auditory tones and visual patches and found poststimulus positive peaks that were higher when observers had to discriminate (e.g., silently count the number of high tones). Shultz et al. (2011) found that eyeblink inhibition was time-locked to salient social and physical events in video clips with saliency denoted by a subjective rating. The current findings suggest that eyeblink analysis and measures could be more precise than previously reported and could be used to produce measures of cognitive and sensory processing. 

A general oculomotor inhibition mechanism

The current findings on eyeblinks, combined with the extensive literature on microsaccade inhibition and our previous study, strongly suggest the existence of a general oculomotor inhibition mechanism, the effect of which is as follows: In response to sensory or cognitive events, eye-movement rates first decrease, then increase above baseline before returning to baseline. This stereotypical time course (decrease–increase–baseline) is modulated by the properties of the stimulus as well as by attention and expectation (for a review on microsaccades, see Rolfs, 2009). This pattern is observed for microsaccades, for eyeblinks (the current study), and possibly for drift as well (Y. Bonneh, Fried, Arieli, & Polat, 2014; see also Ahissar, Arieli, Fried, & Bonneh, 2014). The important evidence related to eyeblinks consists of the current findings of stimulus-specific inhibition (Figure 2d) and release from inhibition (Figure 2b and 3b) with additional preliminary evidence for temporally unpredictable stimulus-triggered inhibition (unpublished data; it requires further investigation). 

The suggested common oculomotor inhibition mechanism also implies common underlying physiological mechanisms. These mechanisms were discussed in detail in relation to microsaccades in our previous study (Y. S. Bonneh et al., 2015) and are suggested to be applicable to eyeblinks. In short, the model suggested by Engbert (2012) provides a good framework for interpreting the current eyeblink results as well as the previous microsaccade results, assuming a central role for the superior colliculus and an integrative effect of low-level visual properties that affect feed-forward collicular responses and top-down cortical signals. A functional interpretation of the general oculomotor inhibition mechanism is discussed next. 

A functional interpretation of blink inhibition

One common interpretation of the phenomena of blink inhibition, or the stimulus-driven modulation of blink rate, is the notion of avoiding the loss of valuable visual information (Shultz et al., 2011). Accordingly, humans have the capacity for controlling the timing of blinks in order to minimize the chance of losing critical information while viewing a stream of visual events (Nakano et al., 2009). This, however, could not be the full explanation for two reasons. First, eyeblinks are also inhibited for nonvisual stimuli; eyeblink generation was synchronized between individuals at the attentional breakpoints of speech (Nakano & Kitazawa, 2010) and was inhibited prior to stimulus onset in a purely auditory discrimination task (Fukuda, 1994). Second, eyeblink inhibition appears to resemble microsaccade inhibition in showing a stereotypical inhibit–release–baseline response triggered and modulated by the stimulus even when temporally unpredictable (Fukuda, 1994), suggesting a general pattern of oculomotor inhibition discussed above. 

The alternative interpretation is that sensory stimuli or anticipated sensory stimuli trigger oculomotor inhibition to prevent the loss of possibly relevant visual information associated with the event until the processing of the event is terminated. Upon the termination of processing, the inhibition is removed, and a transient, self-generated “reset signal” is issued via a single eyeblink and/or microsaccade, marking the termination of a cognitive process or the disengagement of attention (Nakano et al., 2013). In the current study, the latency of this “reset signal,” i.e., the blink RT, depended, like ms-RT, on the contrast and spatial frequency of the stimuli, reflecting the processing time of these stimuli (Y. S. Bonneh et al., 2015). This interpretation is consistent with the suggestion and supporting evidence of Nakano et al. (2013) that eyeblinks are actively involved in the release of attention. It is also consistent with the findings that eyeblinks tend to cause perceptual switching in bistable perception phenomena, thus demonstrating the “reset” capacity of eyeblinks. Given the above interpretation, one could consider these “reset signals” as reflecting milestones in cognitive processes or “cognitive punctuation marks.” 

Blink and microsaccade RTs as implicit measures of low-level visual properties

The current results suggest that both blink and microsaccade RTs could be used to measure low-level visual properties implicitly without the observer's response and perhaps even without attentional engagement. However, only spontaneous eyeblinks are appropriate, and their typical rate is typically approximately seven times lower than that of microsaccades. Moreover, some observers show very low rates of spontaneous blinking. This makes eyeblinks less practical in terms of measuring visual properties in noncommunicating individuals, but the eye-tracking technology needed for blink detection is much more affordable. 

Acknowledgments

We thank Oren Kadosh for his contribution to the experimental data collection and Dr. Moshe Fried for helpful discussions. The study was supported by the Israel Science Foundation (ISF, 188/10, UP). 

Commercial relationships: none. 

Corresponding author: Yoram S. Bonneh.