Saccadic suppression

Another idea to emerge early in the twentieth century was that visual sensitivity is actively reduced during saccades. Holt (1903) concluded that saccades “condition a momentary visual central anaesthesia,” a loss of sensitivity. However, evidence for suppression by saccades is contradictory. Many researchers (Krauskopf et al., 1966; Latour, 1962; Zuber and Stark, 1966) have reported weak threshold elevation for detecting spots of light flashed briefly during saccades (two- to threefold), and Krauskopf et al. (1966) found no threshold elevation at all. By contrast, Bridgeman et al. (1975) reported a strong reduction in sensitivity for detecting displacement during saccades when the displacement occurred at about the same time as the start of a saccade.

Dodge (1900) and Woodworth (1906) concluded that there was no requirement for a central change in visual functions, arguing that image motion during saccades was too rapid to be seen and caused what Campbell and Wurtz (1978) later termed a grayout. However, measurements of contrast sensitivity (during normal vision) show that this idea cannot be generally true. Although stimuli of high spatial frequency become unresolvable at saccadic speeds, stimuli of low spatial frequency can be seen only in motion but become considerably more visible at saccadic speeds (Burr and Ross, 1982). Thus, during saccades, the normally invisible low spatial frequencies that predominate in natural scenes (Field, 1987) should become abruptly salient, posing a potential problem for vision.

It seems that at least part of the reason motion is not seen during saccades is that the low spatial frequencies that would normally be so conspicuous are suppressed during saccades (Burr et al., 1982, 1994; Volkmann et al., 1978). An example of this frequency-dependent suppression is shown in Figure 93.1A–C, which compares contrast sensitivity for detecting briefly flashed horizontal gratings during saccades (filled symbols) with that for fixation (open symbols). Sensitivity is very similar at the higher spatial frequencies, but at low spatial frequencies sensitivity during saccades is reduced sharply, reaching a tenfold reduction of sensitivity at 0.02 c/deg; these are the very frequencies that would otherwise be visible and highly conspicuous during saccades. The selectivity for spatial frequency might explain some of the conflicting data from earlier studies. Loss of sensitivity should depend on the spatial frequency content of the experimental stimuli, typically high (e.g., small spots of light) in the luminance threshold studies (Krauskopf et al., 1966; Latour, 1962; Zuber and Stark, 1966) but low (large targets) in displacement studies (Bridgeman et al., 1975).

Figure 93.1..  

A–C, Contrast sensitivity for detecting a horizontal grating briefly displayed either at the beginning of a large (30 degrees) horizontal saccade (filled circles) or during free viewing (open circles). Measurements were made for 2500 trolands, 1 troland, and 0.014 troland. Sensitivity is greater during normal than saccadic viewing at low spatial frequencies, but the two curves converge at high frequencies; the point of convergence decreases steadily with retinal illuminance. Note that at the higher spatial frequencies of the lowest illuminance (C), contrast sensitivity was actually higher during saccades than during normal vision (see also Fig. 93.2). Measurements were not made above 3 c/deg, as the large saccade would cause smearing of the grating. D–F, Contrast sensitivity for continuously displaced horizontal gratings, either stationary (filled squares) or drifting at 10 Hz (open squares). There is an advantage for drifting gratings at low spatial frequencies. The spatial frequency at which the motion curve peels away from the stationary curve changes with illuminance and corresponds very closely to the point at which the sensitivity during saccades (above) diverge from normal sensitivity, strongly suggesting that motion mechanisms are selectively suppressed during saccades. (Data reproduced with permission from Burr et al., 1982.)

Burr et al. (1982) also reported qualitative changes in motion perception during saccades. Observers viewed at close distance a high-contrast scene back-projected through a deflectable mirror. Displacing the scene abruptly at saccadic speeds and amplitudes caused a strong sensation of motion that instantly commanded attention. However, if the displacements of the scene were the result of a saccade, the motion was sensed, but lacked the salience and the alarming sensation that usually accompany fast motion in normal viewing: subjects observed that the image had been displaced, but they did not report feeling startled. This qualitative impression, together with Bridgeman et al. (1975) demonstrations of large desensitization to displacement during saccades, suggested that the frequency selectivity of the suppression may reflect desensitization of motion mechanisms.

A good deal of evidence points to motion desensitization during saccades. However, it should be pointed out that it is difficult to test motion directly during saccades, for several important technical reasons. One is that, by definition, motion requires moderately long stimulus durations, necessarily exceeding the duration of maximal saccadic suppression. Another is that the movement of the eyes will introduce image motion, which is difficult to take into account accurately in calculating the real retinal velocity of external motion. However, it is possible to infer the action of motion mechanisms using brief stimuli that contain a wide range of temporal frequencies and hence will excite motion mechanisms (tuned to all directions) as well as mechanisms that respond best to stationary stimuli.

An attempt to do this is shown in Figure 93.1. As mentioned earlier, sensitivity during saccades becomes progressively compromised at low spatial frequencies, while at higher spatial frequencies, above 3 c/deg, there is virtually no suppression at all. Figure 93.1D–F shows steady-state sensitivity measurements for stationary and drifting (10 Hz) gratings measured under comparable conditions. As had been well documented previously (Burr and Ross, 1982), gratings in motion are more conspicuous at low spatial frequencies. Importantly, the spatial frequency at which sensitivity for drifting gratings begins to exceed that of stationary gratings is very similar to the spatial frequency at which saccadic suppression begins (illustrated by the vertical dashed lines). As brief stimuli comprise a wide range of temporal frequencies, the similarity in frequency ranges suggests that it is the motion mechanism that is selectively impaired during saccades. To be certain that the correspondence in spatial frequency was not merely a coincidence, the measurements were repeated at two lower levels of illumination. At all three levels of illuminance, both sets of curves—sensitivity to brief displays in normal and saccadic viewing and sensitivity to continuously drifting or stationary stimuli—peeled apart at about the same spatial frequency, consistent with the suggestion that the mechanisms suppressed during saccades are those responsible for the higher sensitivity to motion in normal viewing.

Despite the problems in examining motion perception directly during saccades, there have been several attempts to do so. For example, thresholds for detecting an abrupt change in the speed of a drifting grating are far poorer during saccades than during normal vision (Burr et al., 1982). The discrimination of motion of random-dot patterns is also severely impaired during saccades (Ilg and Hoffmann, 1993; Shiori and Cavanagh, 1989). And discrimination of two-frame motion sequences is severely impaired when one frame is presented near saccadic onset (Burr et al., 1999).

What may be the mechanism whereby motion sensitivity is reduced? Recent anatomical and physiological advances have shown that vision, at least in the early stages of visual analysis, is processed through two largely independent streams: the magno- and parvocellular systems (see Chapter 30). Although these two systems are not completely separate, parvocellular function can be probed by using equiluminant stimuli, which are modulated in color but not in luminance. The magno system there may respond spuriously to some equiluminant stimuli, but it is known to be incapable of color discriminations (Merigan, 1989). Figure 93.2 shows forced-choice discriminations of either the color of equiluminant red-green stimuli or the luminance of equichromatic yellow-black stimuli (both of very low spatial frequency) as a function of time after saccadic onset. Luminance discrimination (filled circles) was severely impaired just after saccadic onset, by one log unit, steadily improving to normal levels over a 200 msec period. Chromatic sensitivity (filled squares), on the other hand, was not at all impaired at around the time of the saccade, and actually improved over the period following the saccade, by approximately a factor of 2. This is an example of saccadic enhancement of contrast sensitivity, similar to that previously observed by Burr et al. (1982) for relatively high frequencies of luminance modulation (see Figure 93.1C) that presumably stimulate the same P-pathways.

Figure 93.2..  

Contrast sensitivity for discriminating the color or the luminance of a broad horizontal bar briefly presented at a given time after the onset of a saccade (abscissa) for two subjects. Bars were modulated either in color (equiluminant red-green: square symbols) or in luminance (yellow-black: circle symbols). Sensitivity is expressed as the inverse of rms cone contrast. The dotted line shows the chromatic sensitivity in free viewing, the dashed line the luminance sensitivity. Note that not only is there no desensitisation for chromatic discrimination just after the saccade, there is actually a marked increase in sensitivity for a period of up to 200 msec after the saccade. (Data reproduced with permission from Burr et al., 1994.)

These results, showing that equiluminant stimuli (irrespective of the spatial frequency) are not suppressed during saccades and can actually be enhanced, imply that saccadic suppression is specific to the magnocellular pathway. The parvocellular pathway, essential for chromatic discrimination, is left unimpaired. Using a different technique, Uchikawa and Sato (1995) arrived at a similar conclusion. They measured incremental spectral sensitivity for detecting monochromatic discs displayed against a white background during normal viewing and saccades. They showed that during saccades, the spectral sensitivity curve showed a marked decrease at ∼570 nm (known as Sloan's notch), a clear signature of the spectrally opposed mechanisms of the parvocellular system. In normal viewing, this decrease was absent (for brief stimuli), suggestive of magnocellular function. Their results are replotted in Figure 93.3, together with representative measurements of responses of P and M retinal ganglion cells of macaque monkey. The psychophysical detection thresholds during normal viewing follow closely the responses of M cells, while during saccades they are more like those of P cells.

Figure 93.3..  

Sensitivity (in relative units) for detecting a monochromatic bar briefly presented on a white background in normal viewing (open circles) or at the onset of a 6 degree saccade. In normal viewing, the curve has a broad peak around 550 nm and closely follows the sensitivity of a sample of retinal M cells in the macaque monkey (taken from Zrenner, 1983). During saccades, however, the form of the curves changes dramatically to reveal Sloan's notch, a sharp dip in sensitivity at around 570 nm. The dashed lines show the average response of a population of P cells. While not following the human sensitivity data exactly, they show the same characteristic dip for middle wavelengths. This is very strong evidence for suppression of magnocellular activity during saccades. (Adapted with permission from Uchikawa and Sato, 1995, and Zrenner, 1983.)

A fundamental question provoked by these studies was whether saccadic suppression results from a central signal, such as the corollary discharge proposed by Sperry (1950) and Von Holst and Mittelstädt (1954), or whether the visual motion caused by the eye movement itself masks vision during saccades (Campbell and Wurtz, 1978; Castet et al., 2001; MacKay, 1970, 1973). There is good evidence that image motion of the kind caused by saccades can mask brief stimuli (Campbell and Wurtz, 1978; Derrington, 1984; MacKay, 1973), but is this the only, or indeed the principal, mechanism at work? Diamond et al. (2000) simulated visual saccades by optically deflecting the display at suitable speed, amplitude, and acceleration, and measured contrast sensitivity to briefly displayed gratings. When the target gratings were displayed on an otherwise blank screen, simulated saccades had little effect on thresholds compared with real saccades. However, when a high-contrast random pattern was added to the display to provide a strong spurious visual motion signal, the simulated saccade produced a suppression that was comparable in magnitude to and lasted longer than that produced by the real saccade. This result suggests that visual masking can be important for vision at around the time of saccades but that it is not the only mechanism. There must also be a signal of nonvisual origin that accompanies each real saccade to decrease sensitivity to low-frequency, luminance-modulated stimuli. Other evidence for a nonvisual suppression signal is that visual phosphenes generated by applying weak electrical signals to the eye in darkness are suppressed during saccades by a comparable amount to real light images (Riggs et al., 1974).

The similarity between the time course of saccadic suppression and visual masking observed by Diamond et al. (2000) could indicate that these two phenomena have a common site of action. This seems reasonable, given the evidence that saccadic suppression occurs early, preceding the site of contrast masking (Burr et al., 1994) and motion analysis (Burr et al., 1999). One interesting possibility is that both saccadic suppression and masking act on contrast gain mechanisms of cortical and/or geniculate cells. This idea predicts that saccades should not only decrease sensitivity, as shown in Figures 93.1 and 93.2, but should also cause the system to respond more rapidly. This is what occurs. During saccades, the temporal impulse response function becomes more rapid (Burr and Morrone, 1996), as many models of contrast gain (e.g., Shapley and Victor, 1981) would predict.

The fact that the impulse response accelerates rather than decelerates during saccades suggests that although strongly attenuated, the magno system remains active during saccades (a parvo-dominated response should be slower). This is consistent with experiments demonstrating that under certain conditions, saccades in the direction of a rapidly moving, high-contrast grating can improve direction discrimination of that stimulus (Castet and Masson, 2000; Garcia-Perez and Peli, 2001). While these demonstrations clearly do not refute the existence of a centrally driven suppression mechanism, they do highlight the important point that centrally driven saccadic suppression only attenuates motion detection, not eliminating it completely, leaving an important role for other mechanisms, such as masking, in natural (usually visually rich) viewing conditions (Campbell and Wurtz, 1978; Derrington, 1984; Diamond et al., 2000; MacKay, 1973).

The studies discussed above all refer to moderate to large saccades and might not be applicable to the microsaccades that are normally made, together with slow drifts, around the fixation point. Most evidence suggests that small saccades cause little or no threshold elevation (Krauskopf et al., 1966; Sperling, 1990), indicating that the effects of the image tremor may be controlled by other means. Murakami and Cavanagh (1998, 2001) recently proposed that the retinal motion generated by microsaccades is eliminated by subtracting a baseline speed, estimated from the minimal retinal jitter, from the velocity signals of local-motion detectors. Evidence favoring this model is derived from the observation that if a region of the retina is adapted to jittering motion and a static pattern is subsequently inspected, the unadapted (but not the adapted) region appears to jitter. Murakami and Cavanagh claim that the reduction in motion sensitivity caused by adaptation reduces the estimate of the baseline jitter so that the motion caused by eye jitter becomes superthreshold in the unadapted region and hence visible. This idea is particularly interesting in the context of older theories that suggested that stabilization is achieved by subtraction of extraretinal signals. In such cases, there is subtraction of a speed scalar (not a spatial displacement vector) from velocity estimates that have been extracted by specialized motion detectors.

In conclusion, human psychophysical data clearly suggest an extraretinal suppression of early visual activity during saccades. There is also much neurophysiological evidence pointing to the underlying neural mechanisms of suppression, but there is no clear consensus on this: some studies show clear evidence of suppression, but others do not. Still others point to more complicated effects, such as an inversion of directional selectivity of MT cells during saccades (Thiele et al., 2002). For a recent review of some of this literature, readers are referred to Ross et al. (2001).