Introduction

Attentional control is the ability to select relevant visual information in the presence of other irrelevant stimuli [1, 2]. This selection is also referred to as top-down or task-driven attention, and can be contrasted to bottom-up attention, which automatically selects stimuli based on their unique properties [3–5]. Top-down selection is typically investigated by having humans and non-human animals attend to one out of several stimuli, either by instruction or manipulating reward probabilities [6–8].

A separate line of research has focused on visual attention in the context of motor actions. Eye movements [9–15] as well as hand movements [16–19] are preceded by shifts of attention to their movement targets. These premotor attention shifts occur automatically when we explore or interact with our environment–without any instructions or reward manipulations [20]. Attentional control and premotor attention frequently share a common goal: Behaviorally relevant or rewarded objects are typically also movement targets [21, 22]. However, the relationship between premotor attention and attentional control has not yet been defined.

Several frameworks can be used to describe the relationship between attentional control and premotor attention. The key approach to study attentional control is the visual search paradigm, which requires top-down or voluntary attention to detect the task relevant information [2]. Unfortunately, visual search studies frequently require gaze to be maintained stable and do not measure eye movements. They typically assume that gaze, under free-viewing conditions, would be directed to the attended location [23–25]. Yet, although it is implied in these studies that attentional control drives premotor attention, this relationship is not directly investigated. Based on the visual search paradigm, a more formal framework has been proposed, referred to as salience maps or priority maps [26–29]. Within these maps, bottom-up and top-down signals are thought to be integrated in a winner-take-all process. Subsequently, the highest activity peak on the map determines the attentional focus, to which eye movements can be potentially directed [26]. This framework again links attentional control and premotor attention without explicitly testing their relationship.

While visual-search-based theories assume that eye movements follow the attentional focus, the premotor theory of attention proposes the opposite, namely that visual attention is a product of the motor system. In order to shift attention covertly (i.e., without moving the eyes), a motor program still has to be prepared, yet not necessarily executed [30, 31]. While this theory can explain attention shifts to movement targets in the absence of any instruction or reward manipulation, a tight coupling does not prove that visual attention in fact arises from motor preparation. Instead, the reverse might be true, and successful movements depend on preceding attentional selection to specify the motor coordinates for an upcoming movement [20, 32]. Based on this assumption, the attentional landscapes framework was proposed [20], which explicitly deals with multiple attentional foci, as they can occur during simultaneous eye-hand movements [33–35]. While cognitive control and premotor attention are closely linked in both frameworks, the implied direction of this relationship is opposite.

The above discussed frameworks vary markedly in their assumptions. They either do not specify the relationship between attentional control and premotor attention directly (visual search & priority maps), or agree on a tight link but assume unilateral influences in opposite directions (premotor theory of attention & attentional landscapes). Since these contrasting assumptions have not yet been addressed, it is still an open question how reflexive information selection in the context of motor actions interacts with adaptive, controlled attentional selection.

We investigated these two components of attentional selection using a classical dissociation approach in which we pitted premotor attention shifts against the capability to maintain voluntary spatial attention. This approach can reveal competition or prioritization between premotor and voluntary attention. Our participants were required to attend to a given location (voluntary attention) while simultaneously preparing an eye movement (premotor attention to eye) and/or hand movement (premotor attention to hand) to another location. If all three tasks interact equally with each other, this would indicate dual-task costs, whereas distinct interaction patterns can differentially support or refute the above discussed frameworks. We used local visual discrimination performance as a proxy of visuospatial attention during premotor and voluntary selection, and systematically biased participants’ deployment of voluntary attention by informing them about which location was most likely to contain the discrimination signal– a briefly presented oriented noise patch. Our data of Experiment 1 revealed that any type of attentional selection, voluntarily as well as premotor, was associated with improved discrimination performance at the target location. Furthermore, we observed no indication of attentional competition between the eye and the hand motor target. In striking contrast, voluntary attentional selection suffered with every motor action being planned, revealing that eye and hand movement preparation abolishes attentional control. This was even the case when conditions for attentional control were optimized: In Experiment 2, the critical discrimination signal predictably occurred at a fixed location, facilitating voluntary attentional selection by reducing both location uncertainty and memory load. Our results demonstrate that the intrinsic effect of premotor attention is so strong that even under conditions benefitting top-down attentional control, the visual system selectively prioritizes attention shifts to movement.

Materials and methods

Experimental design

Experiment 1 comprised seven tasks (randomized block design): Attention-only, Eye-only, Hand-only, Eye-Hand, Attention-Eye, Attention-Hand, Attention-Eye-Hand. Compared to earlier studies [33–35] this design allows us to investigate all possible combinations of spatial attention related to eye movement preparation, hand movement preparation, and voluntary attention. Fig 1A depicts the sequence for the Attention-Eye-Hand task (see S1 Video for a trial demonstration): Participants initially fixated a central fixation target (FT) comprising a black (~0 cd/m2) and white (~98 cd/m2) “bull’s eye” (radius 0.5°) on a gray background (~60 cd/m2). Their right index finger remained on a gray oval (0.6° x 0.65°, ~22 cd/m2) slightly below the eye fixation. At the beginning of each block, four equally spaced locations were marked by gray circles (radius 1.7°) at an eccentricity of 8° from fixation. Note that the distance between the lower two circles was large enough (~11.3°) that participants’ hand did not occlude the objects. One of the four locations (randomly selected and constant throughout a block) was framed in dark gray (~24 cd/m2), indicating the attention target (AT), i.e. the location that participants should aim to attend to endogenously. Note that no such attention target was indicated in the Eye-only, the Hand-only, and the Eye-Hand task. Once stable eye and finger fixation was detected within a 2.5° radius virtual circle centered on the fixation targets for 200 ms, four streams of 1/f spatial noise patches (radius 1.7°) appeared at the marked locations. Each noise stream consisted of randomly generated 1/f noise patches windowed by a symmetrical raised cosine (radius 1.7°, sigma 0.57), refreshing at 60 Hz (Fig 1B). After 400–800 ms, two arrow cues appeared nearby the FT, indicating the eye and the hand movement targets (MT1 & MT2). The movement targets were selected randomly for each trial and could coincide with the attention target as well as with each other. The onset of the arrow cues was the go-signal for both movements, which had to be executed as fast and precise as possible. Participants reached towards either of the two potential movement targets while simultaneously making a saccade towards the other–at free choice. We asked them to stay fixated at the respective motor target until the end of the trial. In the Attention-only task no cues occurred, and in the single movement tasks (Eye-only, Hand-only, Attention-Eye, and Attention-Hand) only one arrow occurred–and only one movement was executed while the other effector remained at the fixation target. Note that in all single and combined movement conditions (unlike in the Attention-only condition) participants had to decipher the orientation cue(s) presented at fixation, which, next to movement preparation, may have further affected their ability to maintain voluntary attention at the instructed location. Even though we have designed the arrow cues to be easily distinguishable (high contrast, sufficient size), we cannot rule out an impact of this additional requirement on peripheral discrimination performance. 100–150 ms after cue onset (within the movement latency), one of the 1/f noise streams was briefly replaced by an orientation-filtered noise stimulus, showing a 40° clockwise or counterclockwise orientation. Participants were informed that this test signal would appear at the attention target location in 75% of trials (in tasks without an attention target, the test was equally likely to appear at any of the four locations). After 50 ms the test was masked by the reappearance of non-oriented 1/f noise for another 700 ms (Fig 1C provides an overview of stimulus timing). Afterwards, the screen turned blank and participants indicated via button press in a non-speeded manner whether they had perceived the orientation to be tilted clockwise or counterclockwise. They received auditory feedback for incorrect responses. Since task instructions are a crucial factor in determining the balance between voluntary and premotor attention [5], we encouraged participants to treat each task with equal importance by asking them to “take [their] time to give the best [clockwise / counterclockwise] answer” as well as to “move [their] eyes (and/or finger) as fast and as precise as possible”.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Methods.

(a) Example trial sequence of the ATT-EYE-HAND task (Experiment 1). Throughout the block, the endogenous attention target (AT) was marked by a dark gray circle. Participants maintained central eye and finger fixation until two black arrow cues occurred that marked two of four noise patches as movement targets. Participants reached towards one and simultaneously saccaded towards the other motor target. Before movement onset, one of the noise streams showed a clockwise or counterclockwise orientation signal. After the movements and a masking period, participants indicated their discrimination judgment via button press. (b) Noise streams used as discrimination stimuli. Each of the four noise streams consisted of a succession of randomly generated 1/f noise patches. The test stream comprised a 50 ms sequence of orientation filtered 1/f noise patches showing a clockwise or counterclockwise tilt. (c) Stimulus timing. Fixation (FT) and attention target (AT) remained on the screen throughout the trial. 400 to 800 ms after the onset of four noise pre-masks (M), the motor cues were presented. 100 ms after cue onset, one of the noise streams contained the orientation test signal, which was masked after 50 ms.

https://doi.org/10.1371/journal.pone.0262567.g001

A threshold task preceded the experiment to ensure a consistent level of discrimination difficulty across participants. The threshold task visually matched the main experiment but no arrow cues were presented and participants were instructed to maintain eye and finger fixation. Furthermore, they were informed at which of the 4 locations the test would be presented in 100% of trials. We used a procedure of constant stimuli and randomly selected the orientation filter strength alpha (corresponding to the visibility of the orientation tilt) out of six linear steps of filter widths. By fitting cumulative Gaussian functions to the discrimination performance via maximum likelihood estimation, we determined the filter width corresponding to 75% correct discrimination performance for each participant and used this value for the main experiment.

Participants performed 66 experimental blocks (2 Attention-only, 3 Eye-only, 2 Hand-only, 8 Eye-Hand, 11 Attention-Eye, 8 Attention-Hand, and 32 Attention-Eye-Hand blocks) of at least 66 trials each, resulting in a total of 4,356 trials per participant. We controlled online for violations of eye and finger fixation (outside 2.5° from FT before the cue onset), too short (<170 ms) or too long (>700 ms) movement latencies, and incorrect eye or hand movements (not landing within 2.5° from motor target center). Erroneous trials were repeated in random order at the end of each block. Overall, 567 ± 117 (mean ± SEM) trials per participant were repeated due to eye movement errors, 441 ± 73 due to finger movement errors.

Task, stimuli, and timing of Experiment 2 were equivalent to Experiment 1, except that we facilitated voluntary attention deployment by presenting only one stream of 1/f noise, to which participants were endogenously attending throughout. The location of this noise stream (attention target; AT) again was indicated at the beginning of each block (either the upper right or the upper left location, randomly selected). As in the previous experiment, depending on the pre-block instruction, participants had to either exclusively attend to the noise stream (Attention-only), or attend to the noise stream and perform eye- (Attention-Eye), hand- (Attention-Hand), or simultaneous eye-hand-movements (Attention-Eye-Hand) to randomly selected motor target(s) indicated by centrally presented arrow cue(s). Unlike in Experiment 1, attention and movement targets never coincided. Furthermore, for each trial we randomly selected the orientation filter strength out of eight linear steps of filter widths (i.e. visibility level; alpha 5 to 75) and fitted cumulative Gaussian functions to the obtained group average discrimination performance via maximum likelihood estimation.

After an initial training (one block of 30 trials for each movement condition), participants performed 13 experimental blocks (3 Attention-only, 3 Attention-Eye, 3 Attention-Hand, and 4 Attention-Eye-Hand blocks) of at least 80 trials each, resulting in a total of 1,130 trials per participant. We controlled online for violations of eye and finger fixation (outside 2.5° from the FT before the cue onset), too short (<170 ms) or too long (>700 ms) movement latencies, and incorrect eye or hand movements (not landing within 2.5° from motor target center). Erroneous trials were repeated in random order at the end of each block. Overall, 145 ± 67 trials per participant were repeated due to eye movement errors, 130 ± 29 due to finger movement errors.

Results

In Experiment 1 participants were instructed to perform different combinations of three possible tasks: endogenously attending to a specific location (Attention), executing an eye movement (Eye), and executing a hand movement (Hand) to a centrally cued target. Concurrently, they performed a two-alternative forced-choice discrimination task based on oriented pink noise patches [40]. Orientation discrimination performance at the endogenously attended location, at the motor target location(s), and at neutral locations (i.e., movement-irrelevant, non-target control locations) served as a proxy for visuospatial attention during motor target and endogenous perceptual selection. Altogether, the experiment comprised seven tasks: Attention-only, Eye-only, Hand-only, Eye-Hand, Attention-Eye, Attention-Hand, Attention-Eye-Hand.

We first evaluated the pattern of attentional selection during each of the above conditions (Fig 2A). In the Attention-only task, we biased discrimination signal probability to guide voluntary attention: the discrimination signal was most likely to appear at the to be attended location (75% probability). Performance at the attention target was better than at the non-targets (p = 0.001, d = 1.869), indicating that participants deployed voluntary attention to the most probable discrimination signal location [3]. In the Eye-only and the Hand-only task, performance at the eye target (p = 0.001, d = 1.986) and the hand target (p = 0.004, d = 1.626) was similarly enhanced relative to the movement-irrelevant locations, which were equally likely to contain the discrimination signal. This demonstrates that attention shifted to the movement targets, independent of discrimination signal probability.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Voluntary and premotor attentional selection.

Discrimination performance at the different test locations of the single target tasks (a), double target tasks (b), and triple target task (c) of Experiment 1. Black lines within each whisker plot indicate the group average. Colored bars depict the 95% confidence interval, dots represent individual subject data, and dashed lines mark chance performance. *p < 0.05, significant difference between a location and the neutral location of the respective task. (d) Discrimination performance at the eye target (left), hand target (middle), and attention target (right) as a function of the experimental task. *p < 0.05, all significant difference between tasks are indicated by horizontal lines. Other conventions as in (a).

https://doi.org/10.1371/journal.pone.0262567.g002

Next, we investigated interactions between these three sources of attentional selection (Fig 2B). In the Eye-Hand task, participants simultaneously performed eye and hand movements to either shared or separate targets. When the two movements were made to separate target locations, we observed improved performance both at the eye (p = 0.003, d = 2.527) and the hand target (p = 0.007, d = 1.389) compared to the movement-irrelevant locations, and the attentional benefit at the two movement targets did not differ (p = 0.219, d = 0.531). When participants made simultaneous eye-hand movements to a shared target, performance at that location was also significantly improved (p = 0.005, d = 1.543), and comparable to performance when eye and hand movements were directed to separate locations (compared to the eye target: p = 0.431, d = 0.216; compared to the hand target: p = 0.515, d = 0.240). Importantly, relative to single effector movements (i.e. Eye-only and Hand-only), combined effector movements to separate locations did neither significantly reduce discrimination performance at the eye target (p = 0.819, d = 0.094) nor at the hand target (p = 0.366, d = 0.225). In summary, this demonstrates that during simultaneous eye-hand movements, attention is deployed to both movement targets in parallel without any observable cost, which is in line with previous studies [33–35; but see 41].

To investigate how voluntary attentional control interacts with motor planning, we asked participants to attend at one location while preparing an eye or hand movement to another (Attention-Eye task, Attention-Hand task). These two tasks create a conflict: while the discrimination signal was most likely to appear at the voluntary attention target, the movement target was more likely to be at a different, non-predictable location. In the Attention-Eye task, when eye movement target location and voluntarily attended location coincided, this mutual target, as expected, received a discrimination benefit (p = 0.001, d = 3.260). When voluntary attention and eye movement were directed to separate locations, we observed enhanced performance at the eye target (p = 0.001, d = 2.214) and a small performance benefit at the attention target (p = 0.015, d = 0.900; note that p_corrected = 0.060). Moreover, performance at the attention target was worse than at the eye target (p = 0.001, d = 1.501). We observed similar results for voluntary attention during hand movement preparation. In the Attention-Hand task, performance at the attention target (p = 0.007, d = 1.373), the hand target (p = 0.001, d = 3.430), and the shared hand-attention target (p = 0.001, d = 3.295) was significantly enhanced. Again, the attentional benefit at the attention target was smaller than at the hand target (p = 0.001, d = 1.935). To summarize, in contrast to the Eye-Hand task, in which attention was equally distributed to both movement targets, attention was clearly biased towards the motor target in the Attention-Eye and the Attention-Hand tasks.

We put further stress on attentional control by asking participants to simultaneously attend to a location while preparing both an eye and a hand movement (Fig 2C, Attention-Eye-Hand). As before, we observed a clear attentional benefit at the eye target (p = 0.001, d = 2.444), the hand target (p = 0.002, d = 1.719), and the combined eye-hand target (p = 0.001, d = 1.745). However, even though the discrimination signal was most likely to appear at the voluntary attention target, participants were not able to maintain voluntary attention there–as it did not differ from neutral performance (p = 0.534, d = 0.208).

A direct comparison of performance across the different motor tasks showed that this decrease in performance was limited to voluntary attention and did not apply to movement targets (Fig 2D). Performance at the eye target was consistently enhanced whether only an eye movement was prepared, or the eye movement was accompanied by either a hand movement (p = 0.819, d = 0.094), voluntary attentional selection (p = 0.893, d = 0.036), or both (p = 0.645, d = 0.201). Likewise, attention was consistently deployed to the hand target, independently of whether the hand movement was accompanied by an eye movement (p = 0.366, d = 0.225), voluntary attentional selection (p = 0.180, d = 0.595), or both (p = 0.150, d = 0.447). In other words, performance at the movement targets in the combined eye-hand movement task was statistically indistinguishable from the respective performance in the single (eye only or hand only) tasks, demonstrating that the attentional selection of one motor target did not affect the selection of the other. In direct contrast, voluntary attentional control was hampered by motor programming: performance at the attention target was reduced whenever a single eye movement (p = 0.001, d = 1.340) or single hand movement (p = 0.003, d = 1.157) were planned. Importantly, performance decreased even further when both an eye and a hand movement simultaneously were directed away from the attended location (compared to single eye movement: p = 0.001, d = 0.898; compared to single hand movement: p = 0.009, d = 0.977). Thus, while attentional control was already affected by single movements, it was practically annihilated during simultaneous eye and hand movement preparation.

We observed that voluntary attention was reduced when participants made an eye or hand movement. Conversely, however, voluntary attention did not affect perceptual performance at the movement targets. We next investigated whether voluntary attention interfered with eye or hand movement preparation in any other way, for example by decreasing movement accuracy or prolonging movement latencies.

We first compared eye and hand landing positions across the different motor tasks (Fig 3A). Generally, when two movement targets were cued (Eye-Hand and Attention-Eye-Hand task) participants tended to select the upper locations as eye targets and the lower locations as hand targets. However, neither eye nor hand movement precision–measured as the average distance of the movement endpoint from motor target center – differed significantly between the respective single movement tasks (Eye-only / Hand-only) and the multiple target tasks (Eye movement precision: Eye-only vs. Eye-Hand: p = 0.259, d = 0.326, vs. Attention-Eye: p = 0.959, d = 0.017, vs. Attention-Eye-Hand: p = 0.400, d = 0.251; Hand movement precision: Hand-only vs. Eye-Hand: p = 0.056, d = 0.635, vs. Attention-Hand: p = 0.740, d = 0.076, vs. Attention-Eye-Hand: p = 0.037). Thus, neither the requirement to program a second movement nor to deploy voluntary attention affected eye and hand movement precision. In contrast, we observed interactions between eye and hand movement control with respect to movement latencies. Compared to the Eye-only task (Fig 3B; left), eye movement onsets were significantly delayed in tasks in which also a hand movement had to be prepared (Eye-only vs. Eye-Hand: p = 0.001, d = 1.647, vs. Attention-Eye-Hand: p = 0.001, d = 1.607), which is in line with earlier work [33]. Having to attend voluntarily, however, did not slow down eye movement execution (Eye-only vs. Attention-Eye: p = 0.131, d = 0.226). Likewise, hand movement latencies (Fig 3B; right) were slightly prolonged by simultaneous eye movement preparation (Hand-only vs. Eye-Hand: p = 0.014, d = 0.417, vs. Attention-Eye-Hand: p = 0.012, d = 0.426), but again not by voluntary attention (Hand-only vs. Attention-Hand: p = 0.275, d = 0.044). Note that the effect of hand movement on eye movement execution was considerably more pronounced than vice versa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Movement latencies and precision.

(a) Normalized eye (top row) and hand (bottom row) landing frequency maps averaged across participants in the respective tasks. White values depict the average distance between movement endpoint and target center as well as the 95% confidence interval. (b) Relative frequency of eye (left) and hand (right) latencies in the respective tasks. (c) Eye and hand movement latencies (left) and landing errors (right) across all tasks split as to whether the movement was made to an endogenously attended location (AE, AH) or not (E, H). Error bars denote the 95% confidence interval.

https://doi.org/10.1371/journal.pone.0262567.g003

To investigate the influence of voluntary attentional control on movement execution, we evaluated movement latencies (Fig 3C; left) and landing errors (defined as the distance between movement endpoint and target center; Fig 3C; right) depending on whether the movement was made to the voluntarily attended location or not. Neither for eye nor for hand movements we observed a significant difference in latencies (Eye-only vs. Attention-Eye: p = 0.164, d = 0.085; Hand-only vs. Attention-Hand: p = 0.118, d = 0.072) or landing errors (Eye-only vs. Attention-Eye: p = 0.517, d = 0.132; Hand-only vs. Attention-Hand: p = 0.865, d = 0.037), demonstrating that attentional control affected neither movement onset nor precision.

Our results showed that preparing eye or hand movements interferes with voluntary attention. In an attempt to reduce the interference of motor preparation on attentional control, we optimized conditions to favor voluntary attention deployment in Experiment 2. In this experiment, we presented only one noise stream that always contained the discrimination signal, which removes any potential uncertainty as to where to attend or respond (Fig 4A). Participants either attended to that location (Attention-only), or attended and made eye (Attention-Eye), hand (Attention-Hand), or simultaneous eye-hand movements (Attention-Eye-Hand) away from this location. In this experiment, we varied the width of the orientation filter used to create the discrimination signal (the smaller the width, the clearer the orientation) and assessed perceptual performance by measuring psychometric thresholds–an alternative approach to quantify attention [5].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Costs of action-selection for voluntary attention.

(a) Stimuli configuration in Experiment 2. One noise stream was presented at a fixed location, marked by a circle. Participants maintained central eye and finger fixation until one (Attention-Eye or Attention-Hand) or two (Attention-Eye-Hand) black arrow cues occurred, indicating the motor target(s). Participants reached and / or saccaded towards the motor target(s) upon cue onset. In the Attention task, no cue occurred and participants maintained fixation. After 100–200 ms the noise stream displayed a brief orientation, which participants discriminated via button press at the end of the trial. (b) Group-averaged psychometric functions (discrimination performance vs. orientation filter width) for each task. Error bars denote the between subject SEM. (c) Discrimination performance at the voluntarily attended location for the different tasks. Black lines within each whisker plot indicate the group average. Colored bars depict the 95% confidence interval, dots represent individual subject data. Significant differences between tasks are indicated by horizontal lines (*p < 0.05).

https://doi.org/10.1371/journal.pone.0262567.g004

In line with our previous findings, motor preparation consistently reduced participants’ ability to voluntary attend (Fig 4B). This was evident in the psychometric functions’ slopes and thresholds. In the Attention-only task, the slope (m = 0.59 [0.37, 1.58]) was steeper than in the Attention-Eye task (m = 0.37 [0.26, 0.65]) and the Attention-Hand task (m = 0.32 [0.17, 0.57]). In the hardest version of the task (Attention-Eye-Hand), the detrimental effect of motor-preparation on attentional control was even more pronounced (m = 0.29 [0.07, 1.38]). When estimating thresholds at a fixed discrimination performance level of 75% we observed a matching pattern: the highest threshold in the Attention-only task (α = 47.21 [41.89, 53.51]), followed by the two single movement tasks (Attention-Hand: α = 42.09 [31.11, 49.12], Attention-Hand: α = 40.16 [31.40, 47.22]). Correspondingly, performance in the Attention-Eye-Hand task yielded the lowest threshold (Attention-Eye-Hand: α = 20.50 [0.98, 36.44]).

To ease the comparison to Experiment 1, Fig 4C depicts the average discrimination accuracy at the voluntarily attended location for each task. As before, voluntary attentional control was consistently hampered by motor programming. Performance at the attention target was reduced when a single eye movement (p = 0.001, d = 1.156) or single hand movement (p = 0.026, d = 1.175; note that p_corrected = 0.052) were planned, and decreased even further when both an eye and a hand movement simultaneously were directed away from the attended location (compared to single eye movement: p = 0.001, d = 1.456; compared to single hand movement: p = 0.001, d = 1.330). Thus, even though we provided optimal conditions for voluntary attentional control by decreasing task difficulty and memory load, motor preparation nonetheless markedly impaired voluntary attentional selection.

Discussion

We studied the relationship between voluntary attentional selection and attention shifts before goal-directed motor actions and observed robust perceptual benefits–a marker for attention deployment–under the typically investigated single-task conditions: at eye movement targets [9–15], hand movement targets [16–19], as well as endogenously attended locations [1, 3, 42]. Under multiple-task conditions, when participants selected two or even three locations, we observed a dissociation between the investigated sources of attention. There was no attentional competition between eye and hand movement targets, yet voluntary attentional selection suffered significantly from every simultaneous action plan. This reveals that the visual system selectively prioritizes premotor shifts of attention over voluntary attentional control and demonstrates that voluntary attentional selection is extremely fragile.

One might have expected that, in order to maintain attentional control, discrimination performance at the motor targets would drop. Instead, although task instructions and stimulus probabilities particularly facilitated voluntary attentional orienting, it was consistently disrupted by motor actions–even when no stimuli were presented at the motor goals, arguing against the possibility that participants voluntarily attended the movement targets. Further, it has been shown that reflexive, non-instructed eye movements likewise interfere with voluntary attentional control [4, 43–45]. Combined, the circumstances under which eye or hand movements do not impact voluntary attentional control have yet to be defined.

Earlier studies observing that eye movements compete with voluntary attention or vice versa [12, 46] cannot rule out that attentional performance was reduced due to a secondary task–as is typically observed in multi-tasking studies (dual- as compared to single-task condition; see also [47]). Using different levels of task difficulty (one, two, or three attention targets), we show that voluntary attention was increasingly impaired, whereas performance at eye and hand movement targets did not suffer. This dissociation–a cost for endogenous attention yet no interference for eye and hand premotor attention–rules out that increased task or attentional selection difficulty caused the loss of attentional control and indicates that earlier results [12, 46] are likewise not explained by multi-tasking costs. Further, even when there was only a single location to be voluntarily attended (Experiment 2), attentional control decreased whenever an action was planned away from this location. This shows that neither reduced stimulus location uncertainty nor the attention target being the only salient stimulus presented could prevent the observed loss of attentional control. Future studies should assess the time course of this disruptive effect of motor programming on voluntary attention, which, based on previous results on premotor attention [e.g., 33, 34, 46–48], likely builds up throughout movement preparation.

Different aspects of our results are not compatible with other major frameworks referring to the relationship between attentional control and premotor attention. First, our present and earlier observations of multiple, simultaneous attentional peaks [33–35, 49, 50] are incompatible with priority map models assuming a strict winner-take-all attentional selection, in which performance benefits should occur only at the highest peak [26–29]. Second, our finding that eye and hand movements draw attention away from the voluntary attended location is incompatible with the view that motor actions merely follow the current focus of voluntary attention, as is assumed in visual search frameworks [2, 26, 51, 52]. Third, the premotor theory of attention [30, 31] assumes that any shift of attention is equivalent to a saccade plan, which offers two testable predictions. First, when the saccade target matches the endogenously attended location, saccadic latencies should be shorter. Our data show that they are not. Second, when the saccade is directed away from the endogenously attended location, two “saccade plans” are technically required. It has been shown that the preparation of two movements (e.g., saccades) results in diminished allocation of attention to the second motor goal [53]. This competition implies that, according to the premotor theory, saccade target and endogenously attended location should compete with each other, i.e., attention allocation to both locations should decrease. Our data show that this also was not the case. In summary, our findings cannot be explained by priority map, visual search, or the premotor theory of attention frameworks without modifying their core assumptions.

However, our findings are in line with the proposal of an attentional landscape–a map representing the attentional distribution across space [20]. This framework allows for simultaneous attention deployment to several action-relevant locations, observed as multiple "attentional peaks". These peaks can vary in magnitude to reflect different degrees of attentional allocation. Our results verify that attention can be allocated to multiple locations at a time. Multiple stimuli are typically assumed to compete for attentional selection [54, 55]. We found that motor goals are prioritized in this attentional selection. Further in line with this view, we observed that the peak performance associated with endogenous attentional selection decreases gradually with every motor action added.

Some of our behavioral effects can be linked to neurophysiological studies. First, activity within visual areas is consistently modulated by visual attention [56, 57], resulting in corresponding perceptual changes [58–60]. Several studies on voluntary attention (control), have observed this modulation throughout the hierarchy of the visual system, ranging from occipital [61–63] over parietal [64, 65] to frontal cortex [66–69]. Second, eye movement preparation modulates neuronal activity in visual cortical areas in a manner indistinguishable from voluntary attention [70]. The neural sources for this modulation are assumed to be fronto-parietal feedback connections converging onto earlier visual areas [57, 71–73]. This has led to multiple proposals that oculomotor areas could serve as an attention source or map [28, 29, 74]. Third, our behavioral observation of multiple attentional peaks can be related to simultaneous and distinct activity peaks observed in human and monkey neurophysiology studies [75–79].

The neurophysiological basis for other key aspects of our findings however is still lacking. First, there is no evidence of the source of premotor attention signals to visual cortex before hand movements. These feedback-signals could originate from reach-related or oculomotor areas–which has yet to be investigated. It is unknown whether neuronal activity associated with premotor attention before eye and hand movements occurs simultaneously in the same area, or in separate areas. Given how consistently attention is shifted before reaching [16, 17, 19] and grasping movements [18, 80], understanding the underlying neural circuitry is crucial to comprehend the mechanisms that govern attentional selection in real-life situations. Second, interactions between endogenous and premotor attention likewise have not yet been explained neurophysiologically. Oculomotor areas are thought to be the common source for presaccadic and covert voluntary attention [58, 71, 72]. It is therefore unclear why both eye movement and hand movement planning do compete with voluntary attention, but no competition is observed between multiple movement targets [33–35, 47, 81, 82]. Third, we do not know whether neuronal modulations associated with eye, hand, and voluntary spatial attention originate in a common area (e.g., frontal, parietal, visual or subcortical areas), suggesting a common attentional map, or arise from different areas. In the latter case, attention to multiple targets could activate separate areas without integrating activity between them. As evidence in favor of a common attentional map has mainly been collected under experimental conditions requiring eye and hand fixation, those conclusions may be biased. It is equally possible that separate, effector-specific maps show attentional modulation during eye and hand movement target selection. Such separate maps could explain the absence of premotor attentional competition between different effectors.

In everyday life, we continuously explore and interact with our environment. Our findings reveal that whenever our eye or hand movement goals do not match our attentional control settings, attentional control cannot be maintained. Thus, attentional control is likely to fail as frequently as we move. We typically avoid this failure by aligning our attentional control and movement goals. While the classical understanding of attention underscores covert attentional orienting in the absence of motor actions, such situations of immobility are rare, if not artificial. Actions are typically considered the consequence of attentional control. Our data however show that actions take precedence over attentional control. We therefore propose to refocus from considering action as the strict consequence of voluntary attentional control to viewing action as the main determinant of successful or failed visual selection.

Supporting information

Acknowledgments

The authors are grateful to Andreas Oschwald and the members of the Deubel lab for helpful comments and discussions in the beer garden.