Spatiotemporal dynamics of visual attention during saccade preparation: Independence and coupling between attention and movement planning

Abstract

During the preparation of a saccadic eye movement, a visual stimulus is more efficiently processed when it is spatially coincident with the saccadic target as compared to when the visual and the saccadic targets are displayed at different locations. We studied the coupling between visual selective attention and saccadic preparation by measuring orientation acuity of human subjects at different locations relative to the saccadic target and at different delays relative to the saccade cue onset. First, we generalized previous results (E. Castet, S. Jeanjean, A. Montagnini, D. Laugier, & G. S. Masson, 2006) revealing that a dramatic perceptual advantage at the saccadic target emerges dynamically within the first 150–200 ms from saccade cue onset. Second, by varying the validity of the spatial cue for the discrimination task, we encouraged subjects to modulate the spatial distribution of attentional resources independently from the automatic deployment to saccadic target. We found that an independent component of attention can be voluntarily deployed away from the saccadic target. The relative weight of the automatic versus the independent component of attention increases across time during saccadic preparation.

Introduction

Psychophysical experiments, using a dual-task paradigm, have convincingly shown that a stimulus presented in the period preceding saccade execution is best processed when its location coincides with the saccadic target (Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler, Anderson, Dosher, & Blaser, 1995). A local improvement of perceptual performance is commonly interpreted as the footprint of selective attention. Therefore, it is now widely established that there exists a tight coupling between the preparation of a saccadic eye movement and a shift of covert selective attention (Findlay & Gilchrist, 2003). The preparation of other types of goal directed movements has also revealed a coupling with a shift of visual attention (Baldauf, Wolf, & Deubel, 2006). 

Several studies have addressed the question whether the coupling between saccadic programming and attentional shift is mandatory. The most frequent answer to this question is a positive one: It has actually proven very hard to perform attention-demanding tasks at locations dissociated from the saccadic target immediately before the initiation of an eye movement (Deubel & Schneider, 2003). 

However, Kowler et al. (1995) found that although programming a saccade requires a spatial shift of attention to the saccadic target, some attention can be diverted from the saccadic goal with relatively little cost for the latency of the eye movements. This fact was illustrated through an attentional operating characteristic (AOC) curve (by analogy with the ‘receiver operating characteristics’ curve; Green & Swets, 1966) representing the trade-off between saccadic promptness (or inverse latency) and perceptual accuracy in a letter discrimination task. Three different points along the AOC curve were obtained by means of a verbal instruction to the subject encouraging either to concentrate on the discrimination task while sacrificing saccadic promptness if needed, or to perform saccades with the shortest latency as possible while disregarding performance in the perceptual task, or to adopt an intermediate strategy between the two previous ones. 

One possible reason why the existence and the magnitude of an independent component of attentional resources during saccadic preparation are still debated is that early studies lacked a fine temporal sampling of the attentional effect during the short period immediately preceding the movement. 

Indeed, very few studies have attempted to address the dynamical evolution of the presaccadic attentional deployment (Castet, Jeanjean, Montagnini, Laugier, & Masson, 2006; Deubel & Schneider, 2003; Doré-Mazars, Pouget, & Beauvillain, 2004; Shepherd, Findlay, & Hockey, 1986). We have recently shown that orientation discrimination acuity improves gradually and systematically at the saccadic target location during saccade preparation (Castet et al., 2006). In addition, we found that visual performance, although globally impaired with respect to the saccade goal, improved across time even at locations away from the saccadic target, but still adjacent to it (same eccentricity, ±45°). Unfortunately, our previous data could not precisely discriminate between two hypotheses: (1) Saccade-triggered attention is focused on the movement goal, but an independent component of voluntary attention—of increasing strength across time—can be deployed in parallel either to the same or to a different location; (2) selective attention is locked uniquely to the saccadic target (with increasing strength across time) but its focus is extended over a broad area (see for example Intriligator & Cavanagh, 2001), including locations at ±45°. 

The aim of the present research work is to shed further light on the spatiotemporal dynamics of the perceptual resources during saccadic preparation. Importantly, different from our previous study, we test visual performance during saccadic preparation also at very distant locations from the movement target in order to avoid the abovementioned confound concerning the size of the attentional focus. More generally, we propose here to estimate the independent component of attention by parametrically varying the need of attentional resources at different locations relative to the saccadic goal and measuring performance in a demanding visual task. To do this, we use a dual task, in which subjects have to execute an accurate saccade toward a centrally cued location (primary task), while they also have to discriminate the orientation of a Gabor patch presented at different locations and delays before saccade initiation (secondary task). The secondary task is designed as a Posner-like experiment (Posner, 1980) in order to induce observers to voluntary modulate the spatial deployment of their attention according to a probability schedule for the location of the perceptual test. The probability schedule (or validity condition, in the Posner's terminology) could either favor the saccadic target location (synergistic condition), the location opposite to it (conflict condition), or, finally, both locations equally (neutral condition). The measure of orientation discrimination ability has been previously used to assess the spatiotemporal effects of attentional cueing (Pestilli & Carrasco, 2005; Solomon, 2004). 

Shepherd et al. (1986) had used a similar paradigm to investigate the presaccadic attentional shift induced with an endogenous cue. Differently from us, they used detection reaction time as a measure of selective attention and they concluded that the planning of an eye movement implies a mandatory and an absolute capture of attention to the saccadic target. Attentional resources could be made free from the saccadic target only well after the end of the saccade. 

In contrast, we show here that perceptual performance is significantly modulated by the probability schedule, indicating that subjects are capable, under certain conditions, of dissociating a part of voluntary attention from the saccadic target location. In particular, we present experimental evidence that early after cue onset perceptual resources can (almost) optimally be diverted from saccadic target location. Later, around 150–200 ms after cue onset, and therefore very close to saccade initiation, the attentional capture at saccadic target becomes stronger. 

Methods

Experimental design

Experiments 1 and 2 were designed as a dual task, where the execution of a saccade constitutes the primary task, whereas an orientation discrimination judgment constitutes the secondary task. The control Experiment 1b does not involve the saccade but its structure is otherwise very similar to the other ones. The perceptual task (orientation discrimination) was designed as a Posner-like paradigm (Posner, 1980) where the target location cue is only probabilistically defined; that is, it indicates the spatial location where the discrimination target will be presented with a given probability (or validity in Posner's terminology) denoted by p. Subjects were informed at the beginning of the session about the probability p that the discrimination patch would appear at the saccadic target. For each probability condition, within an experimental block we measured performance both at saccade target and away from it. 

The sequence of stimuli and the experimental task are shown in Figure 1. 

Figure 1

 

(a) Temporal sequence of stimuli. Observers fixate the center of a circular array of eight disks. The fixation point is then replaced by a saccadic cue. The offset of the cue is the go signal to execute the saccade. In this example, an oblique saccade has to be made to the upward–left disk. After a variable delay, all disks are replaced by vertical Gabor patches except for one of them that is slightly tilted. The tilted patch can appear at the location of the saccadic target (valid trials) or opposite to it (invalid trials), as indicated by the dashed red circles (invisible in the experiment). At the beginning of each experimental block, observers were informed about the probability condition; that is, the percentage of trials in which the tilted patch was going to be presented at the saccadic target. (b) Possible spatial configurations for the saccadic and the perceptual target locations: The orientation of the saccadic cue is along one of the diagonal directions and it determines the two possible positions of the perceptual target.

Observers fixated a dot displayed at the center of a circular array of eight disks (eccentricity = 6°, diameter = 2°). The background was gray (∼20 cd/m ²). The disks and the post‐stimulus mask consisted of white noise dots (0.1 × 0.1°) whose contrast randomly ranged from 0% to ±50% relative to the mean luminance of the display. In Experiment 2, the maximum pixel noise contrast could be set at one of three possible values, ±25%, ±50%, or ±75%, in order to adjust the difficulty of the task at the individual level and avoid ceiling effect or chance-level performance in the discrimination task. 

After a random delay (between 375 and 750 ms), the fixation point was replaced by an arrow indicating the randomly selected direction of the required oblique saccade, which could be upward–left, downward–left, upward–right, or downward right (±45°, ±135°; see Figure 1b). The duration of the arrow was 6.25 ms (one frame). This arrow acted not only as a cue for the saccadic target location, but also as a probabilistically defined cue for the discrimination target. For instance, in a block corresponding to the probability condition p = 75%, an arrow pointing upward–right would instruct (1) a saccade to the upward–right circle and, at the same time, (2) that with 75% probability the discrimination target will be presented at the upward–right circle location (valid trials), whereas only with 25% probability will it be presented opposite to it (invalid trials), namely, at the downward–left circle location. 

Subjects were instructed to perform a saccade “as soon as the central arrow disappeared.” After a variable delay (cue to target onset asynchrony, CTOA: 6 or 150 ms) from the arrow cue offset, all disks were replaced with Gabor patches (2 cycle/deg gratings in cosine phase; seven of them were vertical, one tilted) for a very short duration (either 25 or 50 ms, i.e., four or eight frames) adjusted to the subject's performance during a training session. The tilted patch (discrimination target, or DT) could be presented in one of two relative locations with respect to the cued saccadic target (ST): same (relative angle = 0°) or opposite (180° apart) to it (see Figures 1a and 1b). The relative location of the tilted patch with respect to the saccadic target was randomly selected on each trial within the probability schedule of the current block. 

A mask consisting of static noise dots covering the whole screen immediately followed the Gabor patches for 37.5 ms (six frames) in order to minimize visible persistence. After this, the initial array of eight noise disks was displayed again. 

After the execution of the saccade, observers reported the orientation of the tilt with respect to vertical by pressing one of two buttons. The button pressing automatically initiated the next trial. 

Subjects were explicitly instructed not to sacrifice saccadic accuracy or latency in favor of a better performance in the perceptual task. In particular, online auditory feedback was given if the saccade was not accurate (and therefore rejected), namely, if its amplitude or direction departed from the required values by more than 20% in Experiment 1 and 50% in Experiment 2. Only accurate saccades were further analyzed. For example, if in Experiment 1 the central arrow pointed toward the 6° eccentric disc located at 45° in the first quadrant, the corresponding saccade was only accepted and included in the analysis if it had an amplitude within 6° ± 1.2° and a direction angle within 45° ± 9°. A verbal encouragement was given by the experimenter, if needed, to reduce saccadic latency as much as possible in order to enforce the execution of a true dual task and discourage a strategic postponing of saccade initiation at the advantage of perceptual performance. Conversely, trials in which the saccadic latency was too short, that is, when the eye movement started before the offset of the Gabor patches, were excluded from analysis a posteriori, but no feedback was given to the subject in this case. Exceedingly long latency trials (>1500 ms) were also excluded from analysis, but they represented a negligible fraction of occurrences (<1%). 

Three probability conditions were tested in Experiment 1: p = 75%, p = 50%, and p = 25%, whereas only conditions p = 75% and p = 25% were used in Experiment 2. 

Three subjects were tested for Experiment 1 (two authors of the paper and one naive, average age 32), whereas 10 naive subjects (average age 26) were tested for Experiment 2. 

Experiment 1b, covert attention

The three subjects tested in Experiment 1 were also tested with a second control experiment, involving a single attentional task. In practice, the same experimental sequence of Figure 1 was used and the only difference was that subjects were required to maintain fixation on the central fixation spot instead of performing the saccade. Stimulus arrangement, presentation durations, CTOA values, and everything else were the same as in Experiment 1. 

In a typical Posner-like design, three probability conditions (75%, 50%, and 25%) were used for subject AM and EC, whereas subject AS was tested only with p = 75% and p = 50%. 

Stimuli and eye movement recordings

Eye movements were recorded with an Eyelink II infrared video-based eyetracker (500 Hz; resolution <0.1°). Online check of saccade parameters was performed at each trial and auditory feedback about inaccurate saccades was immediately given. Data about saccades as well as orientation judgment were stored and analyzed offline with MATLAB (The MathWorks, Inc.). 

Stimuli were displayed on a 21-in. color monitor (GDM-F520, Sony, Japan) driven by a display controller (Cambridge Research System Visage 2/3F, Cambridge, UK) with a 160-Hz frame rate (frame duration = 6.25 ms). At a viewing distance of 80 cm, the average separation between adjacent pixels subtended 0.034° of visual angle. The screen subtended 27° × 20°. A lookup table in the software was used to linearize the intensity response of the screen phosphors at an 8-bit luminance resolution. 

Data analysis

Thresholds and percent correct discrimination. An adaptive staircase method was used in Experiment 1 to estimate orientation discrimination threshold for three subjects. In Experiment 2, the test Gabor stimulus had always a relative tilt of ±30° and the percent of correct orientation judgments was assessed for each of 10 naive subjects. In Experiment 2, the presentation duration of the Gabor patches and the contrast of the pixel noise mask were adjusted, during the initial training session, for each individual participant, in order to obtain a performance between chance-level and perfect performance. 

The adaptive staircase method used in Experiment 1 required a much larger number of experimental trials and a more extended training phase (over several daily sessions). In practice, each subject was trained with all experimental conditions (the three probability conditions) until her or his staircase performance curves consistently converged to similar values in different experimental blocks. Each threshold was typically obtained by collecting between 250 and 750 accurate trials after the training phase. Between 10% and 20% (across subjects and different conditions) of the trials were rejected and excluded from analysis because they did not meet the required accuracy criteria. 

For Experiment 2, we wanted to limit the experimental sessions to a maximum of three 1-hr daily sessions, and to do so, we only allowed 1.5 hr of training phase at most. We collected typically between 70 and 210 accurate trials to evaluate each data point, whereas the fraction of inaccurate rejected trials was in this case more variable and ranged between 10% and 30%. 

In Experiment 1, the data points, consisting of the measurements collected over all blocks, were fitted with a linear logistic model of the psychometric function (Collett, 2003): 

$\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(p_i\right)=\log (p_i/(1-p_i\left)\right)={\beta }_0+{\beta }_1{x}_i,$

(1)

where p_i is the fraction of times the test stimulus was judged as tilted clockwise with respect to the vertical and x_i is the actual orientation of the test. The estimated parameters β₀ and β₁ were used to calculate the 0.75 and 0.5 thresholds: 

$\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}{\mathrm{d}}_{0.75}=(\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}(0.75)-{\beta }_0)/{\beta }_1,$

(2)

 

$\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}{\mathrm{d}}_{0.5}=-{\beta }_0/{\beta }_1.$

(3)

 

The orientation discrimination threshold was defined as the difference between the 0.75 and 0.5 thresholds: 

$\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{c}\mathrm{r}\mathrm{i}\mathrm{m}\mathrm{i}\mathrm{n}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{d}=g({\beta }_0,{\beta }_1)=\mathrm{l}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(0.75\right)/{\beta }_1.$

(4)

 

The standard error of thresholds was estimated with a standard method of error propagation for the approximate variance of a bivariate function (Collett, 2003, p. 108). The 95% confidence interval was evaluated by multiplying the standard error by a factor 1.96. 

For each psychometric function, we measured the deviance statistic D to assess whether the fitted function displayed a lack of fit (Collett, 2003, p. 65). Values of D can be compared with tables of the χ² distributions with (n–m) degrees of freedom, where n is the number of orientations tested and m is the number of unknown parameters included in the logistic model (here, m = 2). If the observed value of the D statistic exceeds the (100α)% point of the χ² distribution, where α is sufficiently small, then the lack of fit is declared significant at the (100α)% level. All thresholds reported (except one) rely on fitted functions for which no lack of fit was observed with α = .05. 

For Experiment 2, we simply evaluated the fraction of correct responses for each observer in the different conditions. Group average performance was compared between conditions. 

Results

Experiment 1, orientation discrimination thresholds

Orientation discrimination thresholds are shown in Figure 2 for three subjects, three probability conditions ( p = 75%, p = 50%, and p = 25%), and two CTOA values. The blue curve corresponds to threshold estimates for the trials in which the test Gabor patch was presented at the saccadic target locations whereas in red are displayed the thresholds for trials where the tilted Gabor patch was presented opposite to the saccadic target. 

Figure 2

 

Results of Experiment 1. Orientation discrimination thresholds are plotted as a function of the delay between saccadic cue offset and Gabor patches' onset for the three observers (rows). The three columns correspond to the three probability conditions ( p = 75%, 50%, and 25%). Threshold values at (opposite to) the saccadic target location are plotted in blue (red). Error bars represent the 95% confidence interval.

Error bars represent 95% confidence intervals, thus nonoverlapping error bars between two data points indicate a highly significant statistical difference between those data. 

Note that the well-established presaccadic attentional benefit at the saccadic target would correspond in terms of our threshold measure to a lower threshold value at saccadic target than opposite to it. If one concentrates on the two first subjects (highly trained subjects, authors of the paper), it is clear that this benefit is generally present but it is strongly modulated by the probability condition and the CTOA value. In particular, for subject AM and EC, the presaccadic attentional benefit is canceled in the p = 50% condition (middle column) for the shortest CTOA, namely, when the test Gabor patch is presented immediately (only one frame) after cue onset. In the p = 25% condition (right column), which leads subjects to voluntarily deploy attention away from the saccadic goal, the benefit is even reversed, that is, a perceptual advantage is observed opposite to the saccadic target, for the shortest CTOA and it becomes nonsignificant for the longest CTOA (150 ms after cue onset). 

A strong reduction of discrimination threshold was consistently observed at the saccadic target between the shortest and the longest CTOA. Such improvement of visual performance, represented by a threshold reduction of a factor 2 or 3, is highly significant for subject EC and AM, for whom the 95% confidence error bars never overlap between the first and second CTOA. This finding replicates a previous result of our group (Castet et al. 2006) obtained in a condition of absolute certainty about the location of the test perceptual stimulus (for a comparison between present and previous results, see Discussion). Importantly, the strong improvement at ST was observed also in the conflict condition (p = 25%), suggesting that the progressive summoning of resources at saccadic target between 6 and 150 ms is tightly linked to the oculomotor planning. In contrast, no consistent perceptual improvement across time was observed at the location opposite to ST, not even in the p = 25% condition (with the exception of subject EC), which globally favors that location. 

We believe that the perceptual improvement at saccadic target across time during the movement preparation provides an insight about the characteristic dynamics of the saccade-related modulation of attentional resources. 

Subject AS reveals a different pattern of results. Similar to the other subjects, the orientation discrimination threshold is consistently reduced, at the ST location, between the shortest and the longest CTOA (highly significant difference for condition p = 75% and p = 50%). However, the relative performance measured at and opposite to saccadic target location was practically unaffected by the probability condition, and a strong perceptual benefit for perception at the saccadic goal was permanently observed. This pattern of results is suggestive of an automatic and mandatory attentional capture (Deubel & Schneider, 2003) by the saccadic preparation. The attentional capture would be effective very early during saccadic preparation and it would further increase across time. 

We introduce now a quantity that we call spatial specificity and that provides a compact measure of the relative perceptual advantage at different locations, namely, at the saccadic target location versus opposite to it. The spatial specificity is defined as  

$\mathrm{S}\mathrm{p}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}\mathrm{s}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}=\log \left[\right(\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{e}\mathrm{S}\mathrm{T})/(\mathrm{t}\mathrm{h}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{h}\mathrm{o}\mathrm{l}\mathrm{d}\mathrm{a}\mathrm{t}\mathrm{S}\mathrm{T}\left)\right].$

(5)

 

Note that a value around zero for the spatial specificity corresponds to an equal performance in the discrimination task. In turn, we assume that this condition is achieved through an even distribution of perceptual resources. Positive values of the spatial specificity correspond instead to a perceptual advantage measured at the saccadic target and negative values correspond to a perceptual advantage opposite to the saccadic target. 

Spatial specificity is plotted in Figure 3 for all three subjects and probability conditions and the values obtained for the short and the long CTOA can be directly compared in each panel. As expected, for all subjects, the spatial specificity undergoes an increase (a shift toward more positive values) between CTOA = 6 and 150 ms, in line with the finding that the selective advantage at saccadic location builds up across time during the first 150–200 ms of saccadic preparation (Castet et al., 2006). 

Figure 3

 

Spatial specificity. The individual orientation discrimination threshold was measured opposite to the target location as well as at the saccadic target. The logarithmic ratio of these two values (spatial specificity) is plotted as a function of the probability condition for the three subjects. The spatial specificity for the shortest CTOA (cyan curve) can be compared with the spatial specificity obtained for the longest one (blue curve). Note that the spatial specificity is zero when perceptual performance is equal at and opposite to the saccadic target; a value bigger than 0 corresponds to a perceptual advantage at the target location and a negative value reveals an advantage away from the saccadic target.

A repeated measures two-way nonparametric Friedman analysis of variance (see Conover, 1999) revealed that both the probability condition and the CTOA affected spatial specificity significantly (df = 2, χ² = 11.51, and p < .005 for the p condition effect; df =1, χ² = 7, and p < 0.01 for the CTOA effect). However, it is clear from inspection of Figure 3 that a strong, monotonic modulation of spatial specificity (responsible for the significance of the ANOVA test) is induced by the manipulation of the probability of the perceptual target for two subjects (EC and AM) but not for the third one. 

In particular, subjects EC and AM show an interesting pattern of results for the shortest CTOA: In this condition, the spatial specificity measure crosses the zero point (uniform distribution of perceptual resources) for the neutral condition p = 50% and attains symmetric positive and negative value for the high- and low-probability condition, respectively, in agreement with an optimal distribution of attentional resources depending on the probability condition. 

Saccadic latency and check of task duality

Saccadic latency averaged for the three subjects of Experiment 1 to 222, 212, and 201 ms, respectively, these values being well in the typical range of regular, visually guided saccades. In addition, the individual variability of the saccadic population was relatively small ( SD between 25 and 35 ms for all subjects). The saccadic latency histograms are shown in Figure 4a for the three subjects and all experimental conditions. Latency distributions do not appear to be affected by the current manipulation of probability (a more detailed analysis is reported below). 

Figure 4

 

Saccadic latency and check of task duality. (a) Individual saccadic latency histograms for all conditions of Experiment 1. (b) Individual mean saccadic latency difference between the longest and the shortest CTOA. Blue (red) points represent the latency across trials where the discrimination stimulus was presented at (opposite to) the saccadic target. Error bars represent the 95% confidence interval. A negative latency difference, as observed here, reveals that saccadic latency was longer for the shortest CTOA, at odds with the hypothesis of a strategic delay for saccadic preparation.

To control whether the temporal proximity between saccadic onset and the time of presentation of the perceptual target played a major role in the observed results on orientation acuity, we performed a post hoc analysis on trials grouped depending on their saccadic latency. We divided trials for the CTOA = 150-ms condition (for which the proximity to saccade onset is expected to be more relevant) into halves, depending on having a saccadic latency below or above the median latency. We did not find any significant difference, in any subject or probability condition, in the orientation discrimination performance across latency groups. 

A possible confound in the interpretation of the present results could be grounded on whether subjects have the capacity to perform a dual task, as required to do. Let us assume that observers are capable to strategically procrastinate saccade preparation and execution until most information about the test perceptual stimulus is acquired (serial strategy). In this situation, consistent attentional resources could be still available after cue presentation and before the true preparation of the eye movement. However, if a serial, instead of parallel, execution of the two tasks (orientation discrimination and saccade preparation) is actually applied, we would predict that saccadic latencies would be longer for the longest CTOA than for the shortest one. To rule out this possibility, we computed the average latency difference between long and short CTOAs for the three subjects and the three probability conditions. Data were collapsed across relative spatial location (at versus opposite to saccade target) after checking that similar results were obtained for these two cases. Results are presented in Figure 4b (error bars correspond to 95% confidence intervals), and they clearly suggest that for none of the subjects the “procrastination” strategy was applied. In contrast, saccadic latency proved to be in general mildly longer for the short than for the long CTOA (very significant difference for subject EC), indicating that subjects were truly performing a dual task and they were not implementing a serial strategy. 

Experiment 1b, covert attention task

The strong modulatory effect of the probability condition of perceptual performance during saccadic programming observed for the two trained subjects suggests that part of the presaccadic shift of attention toward the saccadic target can be voluntarily counteracted rather than being an absolute, mandatory phenomenon (an ability which is perhaps only present in highly trained observers). 

Thus, in Experiment 1, even during saccadic preparation, an independent component of attention could be deployed at different locations across space depending on spatial cues and on the reliability attributed to them but largely independently of the saccadic preparation. 

To compare such independent component of attention during saccadic preparation with the capacity to orient attention in a simple cueing task, we tested the same three subjects of Experiment 1 with a control experiment ( Experiment 1b), involving no shift of the gaze. The visual stimuli, the probability conditions, and the test delays used were the same as in Experiment 1, and so was the perceptual judgment required about the relative orientation of the Gabor test. In practice, our control experiment replicates in its design the typical Posner experiment (Posner, 1980), with valid trials where the arrow cue indicates the actual location of the Gabor test and invalid trials where the latter is located opposite to the cued position. 

To illustrate in a compact way the distribution of perceptual resources in space, we used the same variable defined by Equation 5 in Experiment 1, the spatial specificity. For the present control task ( Experiment 1b), the spatial specificity is defined as the relative advantage (expressed as log threshold ratio) in orientation discrimination at the location cued by the arrow versus the location opposite to it. This definition allows us to directly compare spatial specificity for the present task and the dual task ( Experiment 1). Recall that in Experiment 1 the spatial specificity was defined as the relative perceptual advantage at the saccadic target location (coincident with the arrow-cued location) versus opposite to it. 

In Figure 5, we compare the spatial specificity evaluated for Experiment 1 (“dual” in the figure legend) and Experiment 1b (“covert” task), at the two different CTOAs. A similar monotonic dependence of spatial specificity upon the probability condition is apparent for the two tasks for subject EC and AM, whereas spatial specificity is rather flat and always significantly above zero for subject AS Interestingly, for all subjects, spatial specificity is very similar for the dual and the covert attention task for the shortest CTOA (error bars do largely overlap). This fact suggests again that, early after cue onset, the distribution of voluntary attention is not strongly affected by the impending execution of the cued saccade. Finally, the finding that subject AS does not show any effect of the probability condition in the control task, that is, when the preparation of a saccade is not explicitly required, suggests a possible explanation for her peculiar behavior in Experiment 1: She can perhaps not avoid a strong effect of automatic attentional orienting (Tipples, 2002) induced by the simple presence of the arrow cue. 

Figure 5

 

Experiment 1b, simple covert attention task. Spatial specificity for the dual and the simple covert attention task: The log-threshold ratio for orientation discrimination opposite to the cued location versus at the cued location is compared for the results of Experiment 1b (orange, without saccades, covert task) and Experiment 1 (green, with saccades, dual task). The spatial specificity is shown for the three observers as a function of the probability condition. The upper (lower) panels report the spatial specificity values for the shortest (longest) CTOA.

Experiment 2, perceptual performance for a large pool of naive subjects

The absence of any modulation of voluntary attention on the perceptual performance of one subject (completely naive to the object of this study) raised the issue whether the otherwise clear pattern of results revealed for the two trained subjects (and authors of the paper) could ever be observed in naive nontrained subjects. More generally, we took in consideration the possibility that the manipulation of the independent component of attention during saccadic preparation could result in highly variable results across different subjects. It is already established that demanding attentional tasks exhibit large variability at the population level (Fan, McCandliss, Sommer, Raz, & Posner, 2002). 

To address this issue, we decided to repeat Experiment 1 in a more compact version (three 1-hr daily sessions as compared to 2 months of daily experimental sessions in Experiment 1) and a larger pool of subjects. We tested 10 naive subjects with the dual task illustrated in Figure 1. The perceptual test was a Gabor patch tilted at ±30°. We only administered two probability conditions ( p = 75% and p = 25%), and we evaluated perceptual performance in terms of percent of correct responses. 

The mean saccadic latency ranged across subjects between 190 and 330 ms with a group average of 258 ± 17 ms. Again, we performed an offline analysis of the individual latency data to check that subjects were not strategically postponing the saccade in order to optimize perceptual performance. On the basis of this analysis, we decided to exclude one subject from the following analysis because she had significantly longer saccadic latencies for trials with the longest CTOA as compared to the shortest, and this was true in all probability conditions and relative locations of the target. Interestingly, her mean latency difference (long CTOA-short CTOA) reached almost 100 ms, suggesting that she was literally postponing saccade initiation for about the difference in CTOA (∼144 ms). 

Figures 6a and 6b show the group average ( N = 9) percent correct in the two probability conditions and for different locations of the discrimination target relative to the saccadic target. The perceptual advantage at saccadic target is apparent in the p = 75% conditions, but such an advantage is strongly reduced in the p = 25%, and it is even (nonsignificantly) reversed for the shortest CTOA (red point is slightly above blue point). In addition, the percent correct performance does dramatically improve across time at the saccadic target location. In contrast, and similar to Experiment 1, group-averaged visual performance appears to be constant across time at the location opposite to ST. A nonparametric two-way (Relative position × CTOA) repeated measures Friedman analysis of variance revealed a significant main effect of the relative position of the test Gabor with respect to saccade target ( df = 1; repetitions = 9, χ ² = 23.13, p < 10 ⁻⁵) for the condition p = 75%. This effect was nonsignificant ( df =1; repetitions = 9, χ ² = 2.8, p = .09) for p = 25%. The main effect of CTOA was nonsignificant for p = 25% ( df = 1; χ ² = 1.98, p = .16) and close to significance criterion for p = 75% ( df = 1; χ ² = 3.4, p = .06). With this nonparametric test, it is not possible to evaluate the significance of interaction between the two factors. Therefore, we performed a post hoc Wilcoxon signed rank test (paired measures, comparison of median) to test the hypothesis that perceptual performance is different across different CTOAs when the perceptual and the saccadic targets are at the same location. We found that for both probability conditions the effect of CTOA is significant ( p < .03). 

Figure 6

 

Experiment 2, results. The group average ( N = 9) of the percent of correct orientation discrimination judgments is shown for the two probability conditions ( p = 75% and p = 25%, panels a and b). Perceptual performance evaluated at the saccadic target is plotted in blue, whereas performance opposite to it is in red. Error bars represent the standard error of the mean. (c) The mean difference of performance of discrimination at the saccadic target versus opposite to it provides a measure of the spatial specificity. The dependence of the spatial specificity on the probability condition is plotted for the shortest (cyan) and the longest CTOA (blue).

This pattern of results substantially replicates the findings obtained for subject EC and AM in terms of orientation discrimination threshold, indicating that the main findings of Experiment 1 are robust across a large pool of untrained subjects. 

Similarly to the analysis performed for Experiment 1, we evaluated the spatial specificity of the attentional resources during the preparation of the required saccade. Given that this time we only had percent correct data, instead of threshold, we used a different, more appropriate definition of spatial specificity. Now the spatial specificity is defined as the difference of percent correct evaluated at the saccadic target location versus opposite to it. Again, a null spatial specificity corresponds to an even repartition of the attentional resources in space, whereas a positive (negative) value indicates a selective perceptual advantage (disadvantage) at the saccadic target location as compared to the opposite location. 

Figure 6c shows the group average spatial specificity as a function of the probability condition and for the two different delays after cue onset. Similarly to what we observed for Experiment 1, the spatial specificity is higher for the p = 75% than for the p = 25% condition; in addition, it undergoes a positive shift across time, indicating that the presaccadic perceptual advantage develops gradually and becomes more relevant at CTOA = 150 ms. A nonparametric two-way ( p condition × CTOA) repeated measures Friedman analysis of variance confirmed that both main factors were highly significant (nine repetitions; p condition: df = 1, χ ² = 10,54, p < .002; CTOA: df = 1, χ ² = 5.05, p<.03). To have an indication of possible interaction effects, we performed a standard repeated measures two-way ANOVA ( p condition × CTOA), which confirmed significance of the main factors but not of the interaction term. Once again, the probability condition affects dramatically the spatial specificity indicating that the presaccadic shift of attention to the saccadic target can be successfully opposed when a perceptual task has to be performed away from that location. 

To perform a direct comparison and to validate the structural similarity between Experiments 1 and 2, we tested the two highly trained subjects of Experiment 1 (EC and AM, authors of the paper) with the new task ( Experiment 2): Results are shown in Figure 7. Note that the effects across probability and CTOA turned out to be similar, for the two experienced subjects, both in terms of perceptual thresholds in Experiment 1 and in terms of percent of correct judgments in Experiment 2. At the qualitative level, the main pattern of results obtained for EC and AM is consistent with the observations for the large pool of naive subjects in Experiment 2 (compare Figure 7 with Figures 6a and 6b). Finally, for these two subjects, the inversion effect between the two probability conditions at the short CTOA is particularly strong, suggesting that a long training can facilitate the early deployment of the independent component of attention (see also Kristjansson, 2006) away from the saccadic target. 

Figure 7

 

Experiment 2, trained subjects. The percent of correct orientation judgments as a function of CTOA is shown for the two highly trained subjects (E. C. and A. M., on the upper and the lower panels, respectively) who had performed Experiment 1. Results corresponding to the p = 75% ( p = 25%) condition are shown in the left (right) column. Perceptual performance evaluated at the saccadic target is plotted in blue, whereas performance opposite to it is in red.

Trade-off between attention and saccadic latency

Kowler et al. (1995) reported that a part of attentional resources can be diverted from the saccadic target location during saccadic preparation. They analyzed the cost that the allocation of this independent component imposed to the saccadic latency and they quantified the trade-off between the accuracy of perceptual performance away from the saccadic target and the saccadic reaction time in an AOC curve. With our experiments, we have also demonstrated that attentional resources can be partly diverted (especially early in time) from the saccadic goal, allowing a relative perceptual improvement at a location away from the saccadic target. However, it is important to check whether such a diversion of attentional resources implied a sensitive cost in terms of saccadic latency. 

A trade-off between attentional deployment away from the saccadic target and saccadic promptness should result in a globally longer latency for saccades in the p = 25% than that in the p = 75% conditions. The p = 25% condition is actually the most difficult one, in which attention has to “resist” the natural tendency to shift toward the saccadic goal. We tested this hypothesis for the three subjects of Experiment 1 plus the nine subjects whose data were analyzed in Experiment 2. A one-tailed paired t test of the difference of mean latency for the two probability conditions turned out to be nonsignificant ( p > .05). However, by inspecting individual values of the mean latency difference (condition p = 25% − condition p = 75%), we observed that for one of the naive subjects (C.P.) tested in Experiment 2, such value was well above zero (about 95 ms) and it was clearly an outlier (>group mean+2std) with respect to the group distribution. This finding suggests that some individuals, but not all of them, have to pay a significant price in terms of delayed reaction times in order to improve performance away from the saccadic target. 

The main effects revealed by the analysis of Experiments 1 and 2, however, could not be explained by a pure trade-off between latency and visual performance. When subject C.P. was excluded from analysis, the main effects displayed in Figure 6 remain significant. Indeed, we repeated all the nonparametric statistical tests for the analysis of the effect of the crucial variables in the experiment after exclusion of subject C.P. We did not find any relevant change in the results, with the exception that the main effect of CTOA on the spatial specificity was not significant, although close to the classically used criterion (two-way Friedman, eight repetitions, main effect of p condition: df = 1, χ ² = 8.39, p = < .005; main effect of CTOA: df = 1, χ ² = 3.45, p = .06). 

In addition, another observation argues against a big impact of the saccadic latency cost: For the three subjects in Experiment 1, the mean latency difference across probability conditions averaged to 0.9, 1.8, and 2.6 ms, respectively, a difference which could hardly justify the strong performance difference between conditions. 

Absolute spatial distribution of perceptual resources

Experiments requiring a high degree of focused attention have shown that there exists an inhomogeneity of perceptual performance across the horizontal meridian (He, Cavanagh, & Intriligator, 1996; Previc, 1990): Performance is typically better in the lower than that in the upper visual field. 

We tested the preference for the lower hemifield in our orientation acuity data of Experiments 1 and 2. A repeated measures two-way Friedman's analysis of variance (Absolute position × Relative position to the ST) of the percent correct data of Experiment 2 revealed a significant main effect of the absolute position ( df = 1, χ ² = 7.98, p < .005). Post hoc analysis revealed that the perceptual performance was always better in the lower hemifield, both in the case in which the discrimination and the saccade target were at the same location and when they were at opposite locations and for both probability conditions ( p = 75% and p = 25%). Indeed, a repeated measures two-way Friedman's analysis of variance of the performance difference (lower hemifield minus upper hemifield) failed to show any significant effect of either probability condition ( p > .9) or relative position of the discrimination target with respect to the saccadic target ( p > .5). 

For the threshold data of Experiment 1, we obtained the same pattern of results: The threshold ratio (discrimination in the upper hemifield/lower hemifield) was always greater than 1 (indicating a robust preference for the lower hemifield) and no significant effect of the probability condition ( p > .5) or of the relative position with respect to the saccadic target ( p > .9) was observed. 

This finding, which is in line with previous experimental work in the attentional literature, suggests that the preference for the lower hemifield is a robust phenomenon in vision. However, the lack of a strong dependence of the lower hemifield preference on the specific probability condition seems to support the idea that this inhomogeneity is related to a general visual advantage in the lower hemifield (Carrasco, Penpeci-Talgar, & Cameron, 2001) rather than to an attention-specific effect (He et al., 1996). In any case, the lack of an interaction effect between ST location and the preference for the lower hemifield seems to indicate that the absolute structure of the visual performance fields is not affected by the planning of a saccade either upward or downward. 

Finally, no systematic inhomogeneity of the perceptual performance across the vertical meridian (i.e., a preference for either the right or the left hemifield) was detected. 

Discussion

We have studied the spatiotemporal distribution of attention during the preparation of a cued saccadic eye movement. To do so, we have used a dual-task design in which a central cue provides simultaneously exact information about the target location for the instructed saccade (primary task) and probabilistic information about the location of a perceptual target (secondary task). The secondary task is designed as a blocked Posner-like experiment, where, at the beginning of an experimental block, subjects are instructed about the probability condition; that is, the probability that the perceptual test (a tilted Gabor patch) will be presented at the location indicated by the arrow, coincident with the saccadic target. When the probability is high ( p = 75%), the optimal position for the focus of voluntary selective attention is coincident with the position of the saccadic target. In contrast, in the neutral ( p = 50%) and in the low probability condition ( p = 25%), an optimal distribution of selective attention should favor a relative dissociation between the loci of movement planning and selective visual processing. 

If attention is obligatorily locked on the saccadic goal, we would expect to find a null effect of the manipulation of probability on the perceptual performance and we would also expect to replicate the well-known perceptual advantage at saccadic target. If, on the contrary, two parallel and (at least partly) independent processes, involving attentional deployment to different locations, can be executed, namely, the saccade preparation and the voluntary selection of the highly relevant processing location, then we would expect a significant modulation of the perceptual performance. 

We actually found evidence of a strong effect of the probability condition on the ability to discriminate the relative orientation of a Gabor test stimulus briefly presented either at the saccadic target location or opposite to it. This suggests that it is possible to divert an independent component of attention from the saccadic goal. Importantly, we could exclude that (1) the withdrawal of attention from the saccadic goal was explained by an increase of saccadic latency or a loss of accuracy; (2) the capacity to deploy attention to a distinct location than the saccadic target is limited to a restricted number of highly trained subjects (see Experiment 2). The modulatory effect of the probability condition was significant for both CTOAs tested. However, for the longest one (∼160 ms after cue onset), the presaccadic attentional capture was apparent in all conditions for almost all subjects, meaning that the perceptual test is anyway best processed at the saccadic goal. 

With this study, we were able to replicate the dramatic progressive improvement of visual performance at the saccadic goal that we had previously observed in a simpler experimental paradigm. Such an improvement corresponds to a reduction of orientation discrimination thresholds by a factor 2 or 3 (as observed in Experiment 1), and most importantly, it remains highly significant when tested on a large pool of naive subjects ( Experiment 2). Different from our previous study, we did not detect any improvement of visual performance across time far from the saccadic target. We believe that this robust dynamic modulation of performance at the saccadic target is the true signature of the mandatory presaccadic shift of attention. 

Comparison with previous behavioral studies

Shepherd et al. (1986) used a similar dual-task paradigm to study the dynamic relationship between eye movements and spatial attention. They measured detection reaction times for a probe presented at different locations and delays from cue onset, while they modulated across blocks the probabilistic validity of the cue indicating the location of the perceptual stimulus. In contrast with our results, Shepherd et al. concluded that selective attention was obligatorily locked to the saccadic target starting from the moment the cue is presented. Although the reason for the lack of agreement between those results and ours is not completely clear, one possible explanation is in the different measure of perceptual performance used: detection reaction times in the Shepherd et al. paper and orientation acuity performance in our case. Reaction time measures have often led to contradictory results in the domain of selective attention and in particular concerning the coupling between oculomotor preparation and attentional shift (Klein, 1980; Remington, 1980). In addition, by a closer look to their data, it is clear that, when voluntary attention is deployed away from the saccadic target, the presaccadic facilitation at saccadic goal is quantitatively reduced (although still present). Therefore, it is legitimate to think that reaction times are not a sensitive enough measure to reveal the fine time course of the presaccadic modulation of visual perception (see also Deubel & Schneider, 1996). 

Later, Deubel and Schneider (1996, 2003) have addressed the issue whether the coupling between saccadic control and visual attention should be considered as obligatory. They tested it with a dual task (primary task: delayed saccade; secondary task: letter discrimination at different locations), which shared some aspects with the experiment presented here. Their conclusion was that indeed the coupling is obligatory and that a strong presaccadic perceptual advantage was steadily present before the movement initiation, at any delay from cue onset (Deubel & Schneider, 2003). 

One major confound possibly at the origin of the discrepancy between these studies and our current results is that Deubel and Schneider (1996, 2003) have used delayed saccade tasks, in which the spatial cue for the perceptual target location was presented a long time (1500 ms) before the saccadic cue and the latter was presented well before the go-signal for the eye movement. We believe that a crucial point in our experimental design was the simultaneous presentation of the perceptual and oculomotor spatial cue and the requirement to process both tasks in parallel. This fact enabled us to measure the progressive buildup of the perceptual advantage at saccade target. 

Kowler et al. (1995) performed a pioneering experiment leading to a quantification of the independent component of attention in terms of an AOC curve. However, in that case the modulation of the part of attention diverted from saccadic target was obtained through a simple verbal encouragement to subjects (“privilege the perceptual discrimination and sacrifice saccadic latency” or the other way around). The advantage of manipulating cue validity is to allow a parametric control of the postulated independent component of attention. We were thus able to reveal a dramatic effect of this voluntary deployment of attention which did not imply any cost for saccadic latency. 

More recently, Doré-Mazars et al. (2004) measured the capability to detect a letter change at different locations within a letter string and at different delays from the string onset (coincident with the go-signal) during the phase of preparation of a voluntary saccade toward the letter string. They found that overall the letter change was best detected when it occurred close to the saccade landing point. However, when an unexpected group of letters was presented at the beginning of the letter string, it acted as a distractor and turned out to divert perceptual resources from the saccadic target location, especially if it occurred early during saccade preparation. It is important to point out that the study by Doré-Mazars et al. demonstrated that a component of attention could actually be diverted from the saccadic target location by means of a manipulation of an exogenous cue (i.e., the abrupt onset of a salient letter group). Although in the present work we used a radically different experimental design and in particular, we manipulated the distribution of attention by means of an endogenous cue associated to a given probability condition, we found a similar time course of attentional deployment during saccade preparation. 

Time course of covert endogenous attention: faster than we thought?

Several psychophysical studies involving exogenous and endogenous orienting of attention have been reviewed recently (Carlson, Hogendoom, & Verstraten, 2006). The results were summarized in the following way: “Generalizing across orientation [of attention] studies, peripheral cues seem to involve shift times of around 75–175 ms, with centrally cued shifts taking between 200 and 300 ms.” 

These estimates refer usually to when attention is fully deployed, that is, after reaching a saturation level for the attentional benefits. Some authors have shown, indeed, that a gradual buildup of attention can be measured almost instantaneously after the presentation of an exogenous or endogenous cue (Cheal & Lyon, 1991; Nakayama & Mackeben, 1989). It is interesting to point out, however, that if the above duration estimates for attentional shifts were taken at face value, no benefit due to an endogenous shift of attention could in principle take place during the typical delay corresponding to the preparation of a saccade. 

In contrast, both in our single covert attention task and in our dual tasks, we found a significant effect of the probability condition in the shaping of attentional resources already for the shortest CTOA. In this condition, the perceptual probe was presented only 6 ms after cue offset and it remained visible during either 25 or 50 ms. A powerful visual mask was then superimposed on the stimulus, limiting the visual persistence phenomenon. The attentional modulation induced by the probability condition proved to be strong and robust both for trained and for naive subjects. This indicates that measurable effects of endogenous precueing on visual perception can be detected very early after the onset of the cue and well before the effects of the saccade-induced attentional shift. 

In a recent review, Kristjansson (2006) has reported a number of experimental paradigms which reveal the human ability to rapidly learn particular associations between cue and target useful to perform attentional tasks. The repetition of the same cue–target association across a few trials in an experimental sequence can indeed result in a dramatic improvement of performance. It is interesting to point out that our conflict condition (p = 25%) corresponds to an unusual cue–target association (attention has preferentially to be shifted away from the arrow end) and it actually took several trials to adapt to this condition. Nevertheless, after a while, subjects were able to optimally exploit this association and rapidly deployed attention to the location opposite to the arrow end and to the saccadic target. 

Neuronal correlates of the dynamic deployment of the attentional focus

Bisley and Goldberg (2003, 2006) suggested that neuronal activity in the lateral intraparietal area (LIP) represents a saliency map where the highest peak of activation encodes the current focus of attention and predicts the relative perceptual performance. In a dual task involving delayed saccades and orientation discrimination of a Landolt ring, both behavioral and electrophysiological data indicated that attention was generally pinned to the saccadic target, before movement initiation. However, the onset of a visual distractor away from the saccadic target induced a temporary shift of attention to that location, proving that attention is not strictly “locked” to saccadic target. Interestingly, the time course of the relative activation in LIP corresponding to different locations in space matched the relative performance in discrimination. The authors observe that the activity related to the processing of the saccade target in LIP does not seem to reflect saccadic planning per se but rather the relative salience of the target location. 

When considering the spatial aspects of our results in view of these electrophysiological studies, we could speculate that the parietal cortex, and LIP in particular, has an important role in the control of the effects we observed in this study. For instance, the relevance of the location opposite to saccade target in the low-probability condition ( p = 25%) could result in a corresponding peak of activation early after cue onset. Later, the increasing behavioral salience of the saccadic target would take over by producing a relative maximum of activity across LIP population. 

Concerning the temporal aspects, it is hard to make a comparison between the effect we observed and the work by Bisley and Goldberg (2003, 2006). Most electrophysiological recordings in LIP have been done while using delayed saccade paradigms. This experimental design favors a serial strategy, with the execution of the saccade following, possibly after a long delay, the attentional task. On the contrary, a parallel strategy was strongly encouraged in our paradigm, by the constraint to simultaneously program a saccade for immediate execution and perform a demanding visual task. 

Another cortical area that plays a crucial role in the control of eye movements and of attentional shifts is the frontal eye fields area. Thompson, Bichot, and Schall (1997) as well as other single cell studies from the same group (Murthy, Thompson, & Schall, 2001) have investigated the neuronal representation of overt and covert shifts of attention in the FEF. They have found evidence that there is a possible dissociation between FEF activity related to attentional shifts and to pure motor programming. However, in most of these studies, the behavioral performance was tested using a combination of visual and saccadic tasks to be executed following a complex serial, or mixed serial-parallel schedule. For instance Murthy et al. (2001) trained monkeys to perform a covert visual search aiming to find an oddball target among distractors. A saccade to the target had to be planned and executed in most trials, but in a minority of trials the target swapped position with a distractor and in this case the saccade had to reach the distractor location. Therefore, in this and similar paradigms, it is the experimental design itself that imposes an artificial temporal dissociation between the deployment of voluntary attention and the saccadic program. 

A recent different electrophysiological approach (microstimulation in FEF and recordings in lower visual areas: Awh, Armstrong, & Moore, 2006; Moore, Armstrong, & Fallah, 2003) indicated that the presaccadic attentional effects at the target location result from a descending signal from the frontal oculomotor structures. Yet a further distinct hypothesis is put forward by Super, van der Togt, Spekreijse, and Lamme (2004) to explain the observed presaccadic spatially specific enhancement of neuronal activity in V1, which could be related to the attentional effects. They present evidence to suggest that an efferent signal is sent from the early visual cortex to the oculomotor cortex. 

Conclusion

In this study, we have confirmed and generalized our previous result concerning the dynamic improvement of orientation acuity at the saccadic target location within the first 150–200 ms during the preparation of an eye movement (Castet et al. 2006). Such improvement is dramatic (especially when measured in terms of discrimination threshold) and it proved to be robust across a large population of naive subjects. It has to be underlined that the previously observed result was replicated here with a larger number of alternative positions for the saccadic and the perceptual target, as well as by using the diagonal axes instead of the cardinal directions. 

We have also clarified a previous finding which could lead to two alternative interpretations. In the previous paper, a dynamic increase of orientation acuity was observed away from—but relatively close to—the saccadic target (Castet et al., 2006). It was not clear whether this increase was due to the progressive build-up of the independent voluntary attention or rather to saccade-induced effects operating over a large area around the target. In the current work, we do not find any significant perceptual improvement across time when the discrimination target was presented at a location far from the saccadic target, although subjects were encouraged to deploy attention to that location by the probability condition (p = 25%). This fact seems to support the explanation of our previous findings based on an extended focus of the saccade-induced dynamic attentional effect. 

The present paper reports another important finding indicating that the link between selective attentional resources and saccadic preparation is not in an all-or-none mode, nor is it constant across time. We showed that during the period between saccadic cue onset and movement initiation, different probability conditions induce a parametric modulation of the spatial distribution of attentional resources, which is independent of the modulations induced by the saccadic program. In particular, a mandatory attentional deployment at the saccadic target location was observed only for the longest CTOA. 

It is commonly assumed that the presaccadic shift of attention toward the movement goal serves the ecological purpose to optimally select the target, especially when the latter is embedded in a cluttered environment (Liversedge & Findlay, 2000). Our results concerning the temporal dynamics of the spatial specificity of attention is somewhat at odds with this interpretation. In keeping with the study by Doré-Mazars et al. (2004), our findings suggest that the need for focused attention onto the saccadic target becomes really strong only relatively late after the cue onset. If an attentional shift was necessary to select the target for the impending movement, it should ideally occur as soon as possible, well before the motor program is completed. This issue deserves further investigations. 

Acknowledgments

AM was founded by a European Individual Fellowship Marie Curie. 

Commercial relationships: none. 

Corresponding author: Anna Montagnini. 

Email: [e-mail]. 

Address: 31 Chemin Joseph Aiguier, 13402 Marseille Cedex 20, France.