Motion adaptation reveals that the motion vector is represented in multiple coordinate frames

Abstract

Abstract

Abstract:

Abstract  Accurately perceiving the velocity of an object during smooth pursuit is a complex challenge: although the object is moving in the world, it is almost still on the retina. Yet we can perceive the veridical motion of a visual stimulus in such conditions, suggesting a nonretinal representation of the motion vector. To explore this issue, we studied the frames of representation of the motion vector by evoking the well known motion aftereffect during smooth-pursuit eye movements (SPEM). In the retinotopic configuration, due to an accompanying smooth pursuit, a stationary adapting random-dot stimulus was actually moving on the retina. Motion adaptation could therefore only result from motion in retinal coordinates. In contrast, in the spatiotopic configuration, the adapting stimulus moved on the screen but was practically stationary on the retina due to a matched SPEM. Hence, adaptation here would suggest a representation of the motion vector in spatiotopic coordinates. We found that exposure to spatiotopic motion led to significant adaptation. Moreover, the degree of adaptation in that condition was greater than the adaptation induced by viewing a random-dot stimulus that moved only on the retina. Finally, pursuit of the same target, without a random-dot array background, yielded no adaptation. Thus, in our experimental conditions, adaptation is not induced by the SPEM per se. Our results suggest that motion computation is likely to occur in parallel in two distinct representations: a low-level, retinal-motion dependent mechanism and a high-level representation, in which the veridical motion is computed through integration of information from other sources.

Introduction

In everyday situations we perceive objects and interact with them. Consider the case in which you wish to swat a mosquito buzzing around. If it is stationary you would fixate on it, in the hope of producing a final strike, but when it is moving, one typically follows it using smooth-pursuit eye movements (SPEM) to maintain it within the high-resolution fovea. Unfortunately, tracking a moving object generates a problem: when performing precise SPEM, although the object is moving in the world, it remains practically still on the retina. Other sources of information must be used to perceive the veridical motion of an object in such conditions. 

Different studies combining the motion aftereffect (MAE) with SPEM, have tried to identify the various cues used for motion computation. The general consensus is that the MAE can arise from mechanisms at multiple stages of motion processing, leading to a similar behavioral manifestation (Anstis, Verstraten, & Mather, 1998; Swanston, 1994). The early work of Anstis and Gregory (1965) emphasized the importance of retinal background motion in MAE induction. However, others were unable to replicate these results and the question of whether motion on the retina (without an anchor to other objects or background) can cause MAE remains controversial (Swanston, 1994). A major element generating robust MAE is the relative motion of one visual object with respect to another (object or background) on the retina (Mack et al., 1987; Swanston, 1994; Swanston & Wade, 1992; Wade & Swanston, 1996). In a well designed experiment (yet without measuring eye movements), Swanston and Wade (1992) used vertical gratings that were either physically moving or moving on the retina due to SPEM. These gratings were positioned above and below a retinally stationary central grating. The MAE was determined by the relative motion of the central grating with respect to the flanking gratings. Other extraretinal oculomotor elements that can be incorporated at different stages in the motion processing pathway can also induce a MAE (Chaudhuri, 1990a, 1991; Davies & Freeman, 2011; Freeman, 2007; Freeman & Sumnall, 2005; Freeman, Sumnall, & Snowden, 2003; Haarmeier, Bunjes, Lindner, Berret, & Thier, 2001). The integration of such nonvisual signals with retinal motion information may result in the representation of veridical motion, such that it allows discriminating between movement caused by gaze shifts (e.g., smooth pursuit) and movement in the real world. Chaudhuri (1990a, 1991) argued that the MAE can be generated directly by the inhibition of SPEM. He noticed that after a period of prolonged unidirectional smooth pursuit, if the tracking target is extinguished during the postadaptive period, the eyes continue to drift in the tracking direction (pursuit afternystagmus). Chaudhuri suggested that the visual system, in an effort to maintain fixation upon the target, produces a motor signal in the opposite direction in order to offset the residual afternystagmus. The perceptual registration of this efferent signal produces the MAE. Others demonstrated that an efference copy of the eye movements can induce a MAE which can be combined with the retinally induced MAE at subcortical or cortical levels (Davies & Freeman, 2011; Freeman, 2007; Freeman & Sumnall, 2005; Freeman et al., 2003; Haarmeier et al., 2001). 

A somewhat different way to assess veridical motion during SPEM is to use a representation of motion that is based on the change in the object's position with time in a different coordinate frame than the original, retinal one. Such a coordinate frame can be relative to the head, the body, an object, or even in world coordinates (see Anstis et al., 1998; Ilg, 2008; Swanston, 1994; Wade & Swanston, 1996, for a review). For simplicity, all representations of the motion vector beyond a strictly retinotopic one are termed spatiotopic motion representations. Note that we are referring here to the encoding of the motion vector (i.e., assessment of the veridical velocity) rather than the representation of the location of the motion in the visual field (as in the case of constant motion inside an aperture). Encoding the motion vector in spatiotopic coordinates may be especially useful for the extraction of motion direction and speed during SPEM. Under such conditions the retinotopic representation conveys no motion information about the pursued object, as it is practically immobile on the retina. A spatiotopic representation, in contrast, would convey precise information about the object's motion, undisturbed by any gaze shifts. To our knowledge, well controlled, direct evidence for the spatiotopic representation of the motion vector in adaptation is still lacking (Swanston, 1994). 

We set here to test, in a well controlled paradigm, if motion during SPEM is also represented in a spatiotopic fashion, apart from the known retinotopic representation. We used the MAE, generating adaptation to random-dot motion that was either present only in retinal coordinates (Retina condition) or only in spatiotopic coordinates (Screen condition). Specifically, in the Retina condition, subjects pursued a single moving dot while the background dots were stationary, thereby generating motion of the dots on the retina. Conversely, in the Screen condition, the subjects pursued a moving dot that moved at the same speed and direction as the adapting random-dot stimulus, such that the retinal motion was effectively zero. Our hypothesis was that if a spatiotopic representation exists, one should observe adaptation effects in the Screen condition. Alternatively, if adaptation is based solely on retinal motion, adaptation should only be observed in the Retina condition. We find adaptation effects in both cases, suggesting that both levels of representation of motion coexist. In addition, we show that in our specific conditions, pursuit alone (in the absence of random-dot background) does not generate a significant MAE and thus the observed retinotopic and spatiotopic MAEs cannot be solely explained by efference copy induced adaptation. 

Materials and methods

Subjects

Seventeen subjects (ages 20–33 years, 9 females) participated in the experiment. Three subjects were discarded due to noisy behavioral performance (see data analysis) and another due to inaccurate SPEM (see eye tracking), leaving a total of n = 13. Eleven of them were naïve and two were experienced subjects (authors T. S. M. and A. M). Subjects had normal visual acuity by self-report. They all gave written informed consent. Experimental procedures were approved by the ethics committee of the Hebrew University of Jerusalem. 

Stimuli and experimental settings

Stimuli were created in MATLAB (MATLAB version 7.7.0.471 Natick, Massachusetts: The MathWorks Inc. 2008) and Psychophysics Toolbox (Brainard, 1997) and were shown on a 19-inch CRT monitor (Graphics Series G90fB, View Sonic, Los Angeles, USA, 1280 × 1024, 75 Hz). Viewing distance was 52 cm and was enforced using a chin and forehead rest. All experiments were conducted in the dark and the edges of the screen were barely visible (white dots' luminance was 60 cd/m² and the black background's luminance was <10⁻³ cd/m²). 

Stimuli were arrays of large-field random white dots on a black background (dots' size: 0.2° diameter, 100% coherence, array size 41.8 × 33.5°, dots' density: 0.36 dot/°², dots' lifetime: uniformly distributed with a 6-s maximum; Figure 1a, b), excluding a horizontal strip in the middle of the screen (sized 41.8 × 4.9°) where no dots were presented. 

Figure 1

 

Experimental design and analysis. (a) Illustration of the stimuli used in the No Adaptation (NA) condition and the trial's temporal sequence. Subjects judged whether a kinetic random-dot array (the test stimulus) moved to the right or to the left. (b) Illustration of the stimuli used in the Screen and Retina (S&R) condition and the trial's temporal sequence. First, while fixating, subjects saw a moving dot array (the adapting stimulus) for 40 s. Then, subjects judged whether the dots moved to the right or to the left in a 200-ms moving dot array (the test stimulus) while they were intermittently exposed to the adapting stimulus for 5 s (top-up). This temporal sequence applies to all other experimental conditions. Due to fixation, the moving dots in this configuration moved both on the screen and on the retina. (c) Results of one subject in the NA (light grey circles) and S&R (black circles) conditions, showing an adaptation bias toward the unadapted direction. Data were fitted with a psychometric function and the Point of Subjective Equality (PSE) was calculated. Adaptation bias was measured as the shift in PSE (ΔPSE) between NA and S&R conditions.

Experimental design

Subjects participated in three experimental adaptation conditions (Screen and Retina, Screen, and Retina; see Figure 2a), one baseline condition (No Adaptation [NA]; Figure 1a), and one control condition (Pursuit Only [PO], Figure 2a). Seven subjects (Regular group) performed the experiment with the adapting and top-up (a shorter adapting stimulus reinforcing adaptation) stimuli moving to the right as described below. The other 6 subjects (Mirror group) performed the same four conditions with the adapting and top-up stimuli moving from right to left. As expected, a repeated-measures ANOVA with adaptation condition and adaptation direction as factors revealed no effect of adaptation direction (F_(1,11) = 0.25, p = 0.63) and no interaction between adaptation condition and adaptation direction (F_(2,22) = 0.99, p = 0.39). We therefore collapsed the data across groups and refer to the data as if the adapting and top-up stimuli moved to the right. 

Figure 2

 

Adaptation effects: spatiotopic motion causes larger adaptation than retinotopic motion. (a) An illustration of the adaptation stimuli used in the Screen and Retina (S&R) condition, the Screen condition, the Retina condition, and the Pursuit Only (PO) condition. Note that in the S&R condition retinotopic motion (indicated by Retina) and spatiotopic motion (indicated by Screen) are equal and no eye movements (indicated by Eye) are made. In the Screen condition only spatiotopic motion and eye movements are present, and in the Retina condition only retinotopic motion and eye movements are present. In the PO condition only pursuit eye movements are made, without any visual motion in the background. (b) Left panel: ΔPSE results across subjects (black symbols) and the median ΔPSE (grey rectangle) in the S&R condition, showing a significant adaptation bias. The red line indicates no adaptation. Middle panel: A scatter plot showing adaptation effects in the Screen condition and in the Retina condition (black symbols represent individual adaptation effects, grey rectangle represents the median adaptation effect). The Screen condition adaptation effect is significantly stronger. The green line and the blue line indicate zero adaptation effect in the Screen and in the Retina conditions respectively. The grey line specifies the 45° line on which the two effects equate. Right panel: ΔPSE results across subjects (black symbols) and the median ΔPSE (grey rectangle) in the Pursuit Only (PO) condition, showing no significant adaptation. The orange line indicates no adaptation. (c) Left panel: Exemplary horizontal eye movement traces of one subject in a single adaptation trial, under the three conditions (S&R - red line, Screen - green line, Retina - blue line, target position - dashed black line). Eye movements were very accurate in both Screen condition and in the Retina condition. Right panel: Mean pursuit gain and SD for each subject in the Screen, Retina, and PO conditions (dotted line indicates perfect pursuit gain = 1).

The subjects went through different adaptation conditions on separate days to minimize long-term adaptation effects between different conditions. Each daily experimental adaptation condition was preceded by a short NA session to assure that the baseline level of performance remained constant. The order of the experimental conditions was randomized between subjects. Experiments were conducted in a self-paced manner without any perceptual performance feedback. Gaze position at the flashed fixation point at the beginning of each trial was used as a trigger to display the stimulus. 

No adaptation condition

The temporal sequence of the NA session was as follows: each trial started with a central red fixation point (0.3° diameter; Figure 1a). Immediately following a successful fixation, the central dot disappeared and a 200-ms movie of a white random-dot array (test stimulus) appeared. The dots coherently moved to the right or to the left at one of five different speeds (range: 0–1 °/s; Figures 1a, c). Next, the dot array disappeared and the subject reported the dots' motion direction by a key press. There were 125 trials, divided into five blocks, such that each of the five speeds was repeated 25 times. A drift correction procedure was performed before each block. The temporal order of the blocks was randomized across subjects. 

Adaptation conditions

In the Screen and Retina (S&R) adaptation condition, a central fixation point first appeared. Successful fixation triggered a 40-s movie of a white random-dot array (adapting stimulus) moving coherently to the right at a speed of 8.3°/s (Figure 1b). Note that since the subjects maintained fixation on a static central point, the motion of the adapting stimulus on the screen and upon the retina were identical. After this initial adaptation phase, the testing began: a central dot was presented and its successful fixation triggered a 5-s top-up movie with identical parameters to the adapting stimulus. Following this, the central dot disappeared and a test stimulus (one of six possible speeds, 0–5°/s; Figures 1b, c) was shown for 200 ms. After the disappearance of the test stimulus, subjects indicated the dots' motion direction (as in the NA condition). There were 150 test trials, each preceded by a top-up sequence. Each of the six test speeds was repeated 25 times in a random order. The 150 trials were divided into five blocks, with drift correction procedures performed after the initial adaptation and every block. 

The Screen adaptation condition began with a fixation point located at the leftmost part of the screen. A successful fixation triggered the adapting stimulus (for 40 s) as in the S&R condition. Simultaneously with its appearance, the red fixation point started moving at the same speed and direction as all the other random dots in the adapting stimulus. After 5 s the point reached the right end of the screen, disappeared (triggering the disappearance of the adapting stimulus), and reappeared at its starting location. Its refixation triggered the reappearance of the adapting stimulus once again. The subjects were required to pursue the red target accurately and make a saccade to the starting location, resulting in a chainsaw SPEM, repeated eight times. Therefore, while the adapting stimulus was moving on the screen, it was (to the level of pursuit precision) static on the retina. After this initial adaptation phase, the testing began using the same sequence of events as in the S&R adaptation condition. The test speed was between 0–3°/s. 

The Retina adaptation condition was identical to the Screen adaptation condition in its pursuit dynamics, except the eye movement was in the opposite direction. The difference between the two conditions was that the adapting stimulus and top-up stimuli were static random-dot arrays rather the moving ones. Thus, in the Retina condition, the adapting stimulus was moving on the retina in the absence of veridical motion on the screen. 

The Pursuit Only (PO) adaptation condition was identical to the Retina adaptation condition in its pursuit dynamics. However, during adaptation and top-up periods, only the pursuit target was apparent on the screen. A delay of 2.5 s was added between the presentation of the 200-ms test stimulus and the top-up stimulus of the next trial to guarantee complete elimination of any dot traces in the background during the following pursuit. 

Eye tracking

A video-based infrared desk-mounted eye tracker (Eye Link1000, SR Research, Ontario, Canada) with a sampling rate of 500 Hz was used for recording eye movements. The manufacturer's software was used for stimuli presentation, calibration, validation, drift correction, and determination of periods of fixation. The eye-position data was used to automatically monitor online performance of the task. Throughout the experiment, failure to maintain fixation or execution of unwarranted saccades during the pursuit automatically aborted the trial and the trial was recycled. 

Eye-movement data were analyzed using the manufacturer's software and Matlab R2007b. Blinks and saccades were filtered out and data were smoothed using a running average with a 10-ms window. Mean pursuit velocity for a period from 300 ms to 5000 ms after target onset in each adaptation and top-up trial was computed, and the pursuit-gain ( $\frac{\text{mean pursuit velocity across trials}}{\text{target velocity}}$) was used to measure the accuracy of the pursuit. One subject was excluded (post-hoc) from further analysis due to an extremely low pursuit-gain (< 0.6). 

Data analysis

All data were analyzed using SPSS for Windows (Rel. 16.0.1. 2007, SPSS Inc., Chicago, IL), Microsoft Excel (2007), and Matlab software. A psychometric function was fitted to the behavioral data (frequency of left and right responses for each test speed) using the following formula:  

 

(Figure 1c). 

The goodness of fit of the data to the assigned psychometric curve was determined by the model's explained variance (R²). The logistic regression was calculated using the logit function: $\text{logit}\left({\text{p}}_{\text{i}}\right)=\text{ln}(\frac{{\text{p}}_{\text{i}}}{1-{\text{p}}_{\text{i}}})$. We only analyzed the data from 14 subjects whose performance was well matched by the psychophysical function (R² > 0.9; explaining more than 90% of the variance in the baseline [NA] condition). 

The change in the Point of Subjective Equality (ΔPSE) was computed to describe the adaptation-induced changes in motion perception. The PSE was individually determined for the baseline (NA) condition and for the adaptation condition performed just after it. This was done by finding the psychometric curve's speed in which 50% of the responses were rightward. The PSE in the NA condition was then subtracted from the PSE in adaptation condition, to obtain our measure of true adaptation level (ΔPSE measure; Figure 1c). 

Results

Subjects performed a motion discrimination task, judging if a random-dot array moved to the right or to the left, with and without prior motion adaptation (Figure 1a, b). Adaptation was induced in three possible configurations (Figure 2a): in the S&R condition, subjects observed a kinetic dot array on the background while fixating on the center of the screen. Thus, adaptation could result from retinal motion-sensitive mechanisms, as well as screen motion-sensitive ones. In the Screen condition, subjects pursued a target moving in tandem with the kinetic dot array in the background. Adaptation, therefore, could only arise from mechanisms sensitive to the veridical motion, since there was minimal retinal motion (to the extent that the pursuit was slightly inaccurate). In the Retina condition, only retinal-motion adaptation was induced by having the subjects pursue a target moving to the left while a static dot array was presented in the background. This resulted in a motion vector in the same direction and speed as in the other two conditions. 

Eye movements

Eye movements were monitored online throughout the experiment to ensure that the acquired data comprised of very accurate fixation and SPEM. In the S&R condition, subjects held fixation with an average accuracy of 0.56° in the horizontal axis and 0.65° in the vertical axis. Pursuit gain, used to quantify SPEM accuracy, was high in both pursuit conditions: Screen pursuit gain mean = 0.98 (SD = 0.04) and Retina pursuit gain mean = 0.86 (SD = 0.07, Figure 2c). Exemplary eye movement traces of one subject and the mean pursuit gain and SD for each subject in the three pursuit conditions are shown in Figure 2c. 

Motion adaptation

We find, as has been previously shown in numerous experiments (Anstis & Gregory, 1965; Krekelberg, van Wezel, & Albright, 2006; Sekuler, Watamaniuk, & Blake, 2002), that in the standard condition (termed here S&R), motion adaptation leads to a biased perception toward the nonadapted direction. The median effect was ΔPSE = 1.54°/s (Range = 2.42°/s; Wilcoxon Signed Ranks test: Z = −3.18, p = 0.001; see Figure 2b). Smaller, yet significant adaptation effects were seen in both the Retina condition (median ΔPSE = 0.18°/s, Range = 1.00°/s; Wilcoxon Signed Ranks test: Z = −3.11, p = 0.002) and the Screen condition (median ΔPSE = 0.46°/s, Range = 1.55°/s; Wilcoxon Signed Ranks test: Z = −3.18, p = 0.001). Importantly, adaptation was significantly larger in the Screen condition in which there was motion on the screen but no net retinal motion than in the converse Retina condition (Wilcoxon signed ranks test: Z = −2.13, p = 0.03, see Figure 2b). 

The adaptation effect in the S&R condition was not the linear sum of the adaptation effects in the Screen condition and the Retina condition (Related Samples Wilcoxon Signed Rank test: Z = 2.621, p = 0.009). Also, no significant correlations were found between the individual adaptation effects in the S&R condition and the Screen condition (r₍₁₃₎ = 0.02, p = 0.95), the Retina condition (r₍₁₃₎ = 0.37, p = 0.21), or their linear sum (r₍₁₃₎ = 0.19, p = 0.54). 

Although the mean pursuit gain in the Screen condition was 0.98, representing a close to perfect pursuit, one might suggest that the MAE in the Screen condition could have resulted simply from the residual retinal motion generated by the small deviations of the eye movements from the pursuit target. If that were the case, one would expect that subjects whose pursuit speed gain in the Screen condition was low would show greater adaptation since this would lead to larger retinal-slip speeds. This should yield a negative correlation between the pursuit gain and the MAE size in the Screen condition. However, the correlation (across subjects) between the pursuit gain in the Screen condition and Screen MAE was virtually null (r₍₁₃₎ = 0.046, p = 0.88, see Figure 3a). Furthermore, three of the subjects had a pursuit gain above one during the Screen condition. Thus, on average, their retinal slip was in the opposite direction to that of the pursuit. If indeed the size of the Screen adaptation effect was influenced by the retinal slip in this condition, a smaller MAE effect should be observed in these subjects. However, the effect size in these subjects did not differ significantly from that of subjects whose Screen pursuit gain was smaller than one (independent samples Mann-Whitney U test: U = 14.00, p = 0.87). We conclude that the MAE in the Screen condition was probably not related to the small retinal slip motion that is always present to some extent during smooth pursuit. 

Figure 3

 

The size of Screen and Retina MAEs is not related to the pursuit gain in those conditions. (a) A scatter plot of the pursuit gain and the MAE in the Screen condition (black symbols represent individual scores) showing nonsignificant negligible linear correlation (black line). (b) A scatter plot of the pursuit gain and the MAE in the Retina condition (black symbols represent individual scores) showing nonsignificant negligible linear correlation (black line). (c) A scatter plot of the difference in the adaptation effect in the two conditions (ΔPSE Screen - ΔPSE Retina) and the difference between the pursuit gain in the two conditions (Pursuit Gain_Screen − Pursuit Gain_Retina; black symbols represent individual scores), showing nonsignificant negligible linear correlation (black line).

A key finding here is the larger Screen MAE in comparison with the Retina MAE. A possible explanation for this difference may be the smaller pursuit gain in the Retina condition (0.86 vs. 0.98 in the Screen condition), resulting in a slower retinal speed than the nominal speed and possibly leading to weaker adaptation in the Retina condition. However, the difference in the adaptation effect in the two conditions (ΔPSE_Screen - ΔPSE_Retina), and the difference between the pursuit gain in the two conditions (Pursuit Gain_Screen − Pursuit Gain_Retina) were not significantly correlated across subjects (r₍₁₃₎ = −0.21, p = 0.49, see Figure 3c). This suggests that the larger MAE in the Screen condition than in the Retina condition did not result from the small difference in the pursuit gain between the two conditions. 

The MAE could possibly result from a pursuit oculomotor signal per se, without any important role of the visual random-dot stimulus in the background (Chaudhuri, 1990a, 1991; Davies & Freeman, 2011; Freeman, 2007; Freeman & Sumnall, 2005). To rule this out, we conducted another control experiment, the Pursuit Only (PO) adaptation condition, which was performed by eight of the original subjects. As before, subjects performed a motion discrimination task with and without prior-motion adaptation. However, in the PO experiment during the adaptation phase, subjects tracked a sole pursuit target in the absence of any additional visual stimuli in the background. No significant adaptation was observed under these conditions (median ΔPSE = 0.14, Range = 4.24; Wilcoxon signed ranks test: Z = −0.56, p = 0.58; see Figure 2b). The same subgroup of subjects showed a significant adaptation in the Retina condition (median ΔPSE = 1, range = 5.59: Wilcoxon signed ranks test: Z = −2.52, p = 0.01) in which the adaptation effect across all subjects was the smallest, suggesting that the lack of significance in the PO experiment was not due to mere lack of statistical power. These results show that a smooth pursuit of a sole dot in the dark (in the absence of a visual stimulus in the background) does not induce a clear MAE when using our test (random-dot array) stimulus, although it will generate an MAE for a dot presented in the fovea (Chaudhuri, 1990a, 1991; Freeman & Sumnall, 2005). 

We conclude that during smooth pursuit, motion adaptation is more sensitive to the veridical movement in screen coordinates than to the retinal aspects of motion. 

Discussion

We report here that significant motion adaptation occurs when viewing a dot array, which moves in spatiotopic coordinates (e.g., on the screen) but is immobile in retinotopic coordinates (due to concurrent smooth-pursuit eye movement, SPEM). This spatiotopic adaptation effect is significantly greater than the adaptation induced by viewing a stationary dot array that moves only on the retina. 

Based on these results we can speculate that some form of representation of the motion vector (explicit or implicit) exists in spatiotopic coordinates. Note that a spatiotopic representation of the motion vector is distinctly different from the issue of the reference frame of the place where movement occurs (e.g., within a specific aperture). Spatiotopic encoding of the motion vector endows an advantage for the extraction of motion parameters during SPEM, a condition in which the retinotopic representation of the pursued object motion is unsuitable as the object is almost immobile on the retina. 

Second, we show that the visual system is more sensitive to exposure to an object moving only in spatiotopic coordinates than to motion solely in retinotopic coordinates. Possibly, the representation of motion in spatiotopic coordinates is more closely related to our perceptual experience especially when the retinal and veridical motion vectors differ, as in the case of object-motion extraction during SPEM. 

Our data reveal no linear relationship between the adaptation effect in the S&R condition and the adaptation effect in the Screen or Retina conditions. The S&R MAE is neither the linear sum of the MAEs in the two conditions, nor is it correlated with either of them. We argue that this is to be expected due to the qualitative difference between these conditions, namely the lack of SPEM in the S&R condition, unlike the two other experimental conditions. The underlying brain mechanisms supporting motion perception during fixation and pursuit are at least partially different (see Lisberger, 2010 for a review; Luebke & Robinson, 1988). Therefore, direct comparison of the MAE size in the S&R condition and the two other conditions is somewhat problematic. 

We also report here that merely tracking a pursuit target, in the absence of any additional visual stimuli in the background, does not lead to any significant motion adaptation. Hence we demonstrate that, in our experimental setup, the efference copy of the SPEM (i.e., the extraretinal signal) cannot cause adaptation per se and cannot explain the observed spatiotopic and retinotopic MAEs. 

According to the subjects' reports, the dot array's motion was perceived as it actually occurred: in the Screen condition, it was seen as moving, while in the Retina condition it was perceived as static (similar to Mack et al., 1987). Thus, although the visual system was adapted by the retinotopic motion stimulus, it did not produce a conscious percept of motion during the presentation of the adapting stimulus. 

Taken together, these data suggest that two different adaptation mechanisms may operate here: a mechanism that induces adaptation based on the retinal motion signals, which is likely to be based on motion processing at lower-level regions of the visual hierarchy (where the conscious motion percept is not yet formed) and a mechanism which induces adaptation based on spatiotopic motion signals, which is probably based on motion processing higher in the visual system hierarchy where a percept of the veridical motion is already formed. 

Recent studies in monkeys indicate that, indeed, a nonretinal representation of the motion vector may be present in the visual system. Typically, whereas neurons in cortical area middle temporal (MT) are selective to the motion vector parameters (direction and speed) in retinal coordinates, neurons further downstream in the motion processing pathway (e.g., middle superior temporal (MST) area) encode motion velocity in spatiotopic coordinates (Ilg, 2008; Ilg, Schumann, & Thier, 2004; Inaba, Miura, & Kawano, 2011; Inaba, Shinomoto, Yamane, Takemura, & Kawano, 2007). For instance, Inaba et al. (2011, 2007) recorded the activity of MT and MST neurons in monkeys that were pursuing a moving target with a random-dot array moving at various speeds in the background. They found that MST neurons were tuned to the dots' veridical motion while MT neurons encoded their retinal motion. In MST, the activity of a subgroup of neurons called visual tracking (VT) neurons (which receive inputs regarding the retinal image motion, eye, and head movements) reflects the target trajectory in world-centered coordinates (see Ilg, 2008 for a review). It seems likely that the source of the retinal input is the earlier retinal representation of object motion in MT (possibly using the visual-only neurons in MST as interfacing elements) (Ilg et al., 2004). These signals are integrated with eye-movement related signals (which are not a direct efference-copy signal but perhaps originate from the frontal eye fields) and with head-movement related signals from the semicircular channels of the inner ear (Ilg, 2008). Therefore, the retinotopic mechanism of adaptation may possibly be located in MT or earlier (e.g., V3A) while the spatiotopic higher-level one is likely to be present in MST. 

The existence of a spatiotopic mechanism tuned to head-centered motion was shown by Champion & Freeman (2010). They demonstrated that in a velocity judgment task with SPEM, subjects used mixed strategies to decide which of three stimuli was the odd one. On some trials, subjects relied on the stimulus' relative retinal motion and on the pursuit target's motion separately, yet on other trials they relied on the explicit head-centered velocity of the stimulus. We show here that a spatiotopic motion mechanism (which might also be head-based) can explain a different behavioral phenomenon: the MAE during SPEM. 

The fact that the retinotopic adaptation we report here is smaller than the spatiotopic adaptation might be partially explained by the suppression of visual responses to reafferent motion (Chukoskie & Movshon, 2009). Reafferent motion is the retinal motion of veridically-static objects induced by pursuit eye movements. It usually does not lead to a percept of motion, similar to the motion in our Retina condition. Chukoskie and Movshon (2009) recorded the responses of single neurons in monkeys' MT and MST during pursuit. They found that cells in MST (and to a lesser degree in MT) were often suppressed when retinal motion was induced by the pursuit movement, possibly reflecting the suppression of such spurious motion. This suppression may contribute to the smaller retinotopic MAE (compared to the spatiotopic MAE) in our experiment. 

In a pioneering early study, Anstis and Gregory (1965) measured the MAE induced by random-dot array motion under configurations like our S&R, Screen, and Retina conditions. Unlike here, they found no Screen MAE and concluded that the MAE is determined by retinal motion and is independent of SPEM. Uncontrolled retinal motion due to the lack of eye tracking, the use of static test stimuli, the very slow pursuit speed, or the performance of the experiments in full light might all explain these contradicting results. Contrary to Anstis and Gregory's results, Davies and Freeman (2011) did find a MAE in a configuration resembling our Screen condition. Like them, Mack et al. (1987) could not replicate Anstis and Gregory's results and showed that the MAE size depended on the induced motion of visual edges that appear to move due to SPEM. However, our findings are unlikely to be explained by visual edges effects because all experiments were performed in the dark, making edges barely visible. Moreover, viewing conditions (e.g., retinal edges) were identical in the Retina and in the Screen conditions yet the MAEs in these two conditions differ significantly. Similarly, as shown in the results section, the slight differences in the pursuit gain between the two conditions are unlikely to explain the greater MAE in the Screen condition. Nevertheless, in the Screen condition retinal motion of the stationary dots relative to the moving edges cannot be completely ruled out as a confound (however unlikely it is due to the edges' minimal contrast). In order to empirically determine if its contribution is significant, one can conduct the same experiment inside a Ganzfeld drum where there are no edges and therefore no relative motion. 

The difference in the adaptation effect size between the Screen and the Retina conditions might potentially be explained by differential attention load. Although both conditions require pursuit, Heinen, Jin, & Watamaniuk (2011) have shown that when a large random-dot array moves in synchrony with a foveal pursuit target (like in our Screen condition), attention can be disengaged from the fovea without diminishing pursuit accuracy. In such a case, two different pools of attention are available for pursuit: a foveal one and a peripheral one. Pursuit can be performed using the peripheral pool, freeing the foveal pool for other attention demanding tasks. In contrast, when only the foveal target is present, attention must be focused on the fovea to accurately pursue it. In this light, one may argue that in our Retina condition, subjects focused their attention on the pursuit target as the background dots did not move along with it, while in the Screen condition subjects could spread their attention to the peripheral random-dot array which moved along with the pursuit target. Thus, the spotlight of attention could potentially have been larger in the Screen condition than during the Retina condition, giving more weight to the background dots and consequently, generating a greater MAE in the Screen condition than in the Retina condition. Indeed, the MAE is modulated by attention: performing an alphanumeric discrimination task in the fovea while a texture moves at the background reduces the background-induced MAE (Chaudhuri, 1990b). However, the modulation of the MAE depends on the contrast of the adapting pattern: as adaptation contrast increases, so does the strength of the aftereffect, but there is a compressive nonlinearity in the aftereffect such that when the adaptation contrast exceeds about 2–3% contrast, the aftereffect strength remains nearly constant (See Burr & Thompson, 2011 for a review). So, in high contrast adapting patterns as the random-dot array we have used (close to 100%), no attention modulation of the MAE should be seen, even if there were differences in the allocation of attention between the Screen and the Retina conditions. Moreover, it is unlikely that the attentional window was wider because of the rigorous online exclusion criteria of the tracking task, requiring subjects to stay focused on the target and the lack of any attention demanding parallel task. 

A different perspective from ours was taken by studies asking whether the MAE could result from the efference copy of the oculomotor signal on its own (Chaudhuri, 1990a, 1991; Davies & Freeman, 2011; Freeman, 2007; Freeman & Sumnall, 2005; Freeman et al., 2003). All these studies have suggested that the degree of MAE generated during smooth pursuit is linked to the repetitive SPEM (pursuit followed by a saccade back to the starting location, followed by pursuit, etc.). Chaudhuri (1990a, 1991) suggested a postadaptation oculomotor mechanism in which after the repetitive SPEM stop, they give rise to residual nystagmus eye movements that must be suppressed for correct fixation. This suppression can generate the MAE. Another possible oculomotor source of MAE was suggested to occur already at the time of adaptation, where oculomotor signals directly cause the adaptation of cortical visual areas (Freeman & Sumnall, 2005). Both Chaudhuri (1990a, 1991) and Freeman & Sumnall (2005) asked subjects to track a small pursuit target with repetitive SPEM without any visible background and measured the direction of MAE using a static central fixation point. Contrary to our findings in the control PO condition, both studies report a MAE under these conditions. The spatial specificity of the MAEs reported in these studies and here may provide a clue to their source: the MAE induced by visual inputs is mostly restricted to the specific portion of the retina exposed to motion (Knapen, Rolfs, & Cavanagh, 2009; Swanston, 1994). Conversely, the oculomotor MAE is not retinotopic (Davies & Freeman, 2011). In Chaudhuri's (1990a, 1991) and Freeman and Sumnall's (2005) studies, the central-dot test stimulus appeared exactly at the same location as the adapting stimulus. Therefore, the test stimulus location was centered on the exact retinal area that was adapted by the visual pursuit target. Hence, the MAE could have resulted also from a visual effect (as in our Screen condition) and not only from an oculomotor (extraretinal) one. In contrast, in our PO experiment the test stimulus was a large-field dot array that did not overlap with the adapting stimulus' location (in the fovea). Due to the spatial specificity of the visual MAE we could test the oculomotor MAE separately. It is important to note first that there are other differences between our experimental setting and the ones used in Chaudhuri's (1990a, 1991) and Freeman & Sumnall's (2005) studies that might contribute to these conflicting results. Among these are the use of static vs. dynamic test stimuli, the testing procedure, the different size and location of the retinal region exposed to adaptation, etc. Second, in a different experiment in Freeman & Sumnall's (2005) study, a MAE without overlap between adaptation and test stimuli was observed. However, this was done using an adapting parafoveal dot array which elicited look nystagmus eye movements (rather than a single central dot and SPEM, as in our case). All in all, the lack of a significant MAE in the PO condition suggests that in our experimental setup SPEM per se cannot cause adaptation and therefore cannot explain our spatiotopic and retinotopic MAEs (though it is important to note that the oculomotor-related signal is crucial for the formation of the spatiotopic representation of motion). 

Our results show that motion computation can utilize spatiotopic and retinotopic coordinates independently. Future research is needed to investigate the exact nature and location of the underlying mechanisms responsible for the two computation types. In addition, the relevant reference frame of the reported spatiotopic encoding of the motion vector (head, body, screen, or world-based) is yet to be discovered. 

Taken together, our data demonstrate that the motion vector is represented in spatiotopic coordinates, in parallel with a more low-level mechanism encoding motion in retinotopic coordinates. Furthermore, the visual system is more prone to adaptation that results from motion in spatiotopic than in retinotopic coordinates. Indeed, the representation of motion in spatiotopic coordinates is more closely related to our perceptual experience, especially when the retinal and veridical motion vectors differ. 

Acknowledgments

Tal Seidel Malkinson's work was supported by the German-Israeli Foundation for Scientific Research and Development (GIF) grant number 1108-79.1/2010. 

We would like to thank Drs. Concetta Morrone for kindly providing us with the MATLAB software generating the random-dot stimuli, Shaul Hochstein for suggesting the Pursuit Only control experiment, and Tanya Orlov for her helpful comments on the manuscript. 

Commercial relationships: none. 

Corresponding author: Tal Seidel Malkinson. 

Email: [e-mail]. 

Address: Department of Neurobiology and Department of Psychology, Hebrew University, Jerusalem, Israel.