Psychophysiology, 46 (2009), 1–10. Wiley Periodicals, Inc. Printed in the USA. Copyright r 2009 Society for Psychophysiological Research DOI: 10.1111/j.1469-8986.2009.00847.x

Contextual modulation of oculomotor control reflected in N2 and saccade reaction time distributions

JASPER G. WIJNEN AND K. RICHARD RIDDERINKHOF Amsterdam Center for the Study of Adaptive Control in Brain and Behavior, Department of Psychology, University of Amsterdam, Amsterdam, the Netherlands

Abstract Avoiding reflexive saccades triggered by salient yet task-irrelevant stimuli requires the engagement of control processes that inhibit attention toward irrelevant objects and prevent reflex-like oculomotor action. In the current study participants made saccades to visual targets to the left and right of fixation as directed by target appearance. A distractor could either be presented in the same (congruent trials) or the opposite hemifield (incongruent trials) as the target. Trial context was manipulated, creating risky (mostly incongruent blocks), safe (mostly congruent blocks), or neutral conditions. Electroencephalogram was measured to examine if any contextual modulations would be reflected in the N2 peak frequently associated with performance monitoring. The results are discussed within the framework of dual mechanisms of cognitive control and suggest that a high-risk context facilitates avoidance of capture, lessening the need for reactive control. Descriptors: Attentional capture, Inhibitory control, Saccades, EEG, Distributional analysis

decade and has been termed attentional capture (for reviews, see Ruz & Lupian˜ez, 2002; Theeuwes & Godijn, 2001). Automatic movement of the eyes toward such sudden onset distractors has been observed in many experiments (e.g., Kramer, Hahn, Irwin, & Theeuwes, 2000; Theeuwes, Kramer, Hahn, & Irwin, 1998) and has been termed oculomotor capture. In such experiments, subjects are confronted by salient distractor stimuli and asked to move their eyes to a target stimulus that is presented simultaneously with the distractor. In many cases the subjects’ eyes move initially to distractor stimuli before moving on to the target stimulus, even though instructions designate a direct movement to the target. The context in which distracting stimuli appear may modulate oculomotor capture. For example, in a context where congruent trials (in which target and distractor coincide spatially within the same hemifield) are most prevalent, subjects might be inclined to let themselves be guided to some extent by the salient distractor. On the other hand, when distractor and target appear more frequently in opposite hemifields (incongruent trials), subjects are better off inhibiting activation of the response elicited by the distractor and/or enhance activation triggered by the target. Such strategic adjustments of attentional focus have been reported previously in studies using related tasks. Gratton, Coles, and Donchin (1992) used an Eriksen flanker task where a centrally presented stimulus is flanked by distractor stimuli indicating either the same or the opposite response, and subjects are instructed to react only to the central stimulus. Slower reaction times (RTs) and more errors are typically observed when flanker stimuli are incongruent with the target stimulus. This effect was reduced, however, when the probability of a given trial being

This study aims to identify electrophysiological correlates of adaptive oculomotor control and to combine this data with relatively new distribution-analytical techniques designed to highlight inhibitory effects of cognitive control in saccadic reaction-time data. Primarily we are interested in whether these control mechanisms are modulated by the context within which control was invoked. The control of saccadic eye movements comes into play when exogenous (i.e., stimulus-driven) influences compel us to make an eye movement toward the stimulus while set goals require us to look in a different direction. The control mechanisms that guide our behavior through such conflicts and that allow us to produce an appropriate ‘‘correct’’ response may be affected by the degree to which we expect distracting stimuli to appear and/or the degree to which we are likely to commit errors. Here we explore the effects of such contextual modulation on brain activity measures related to cognitive control as well as on the dynamics of selective inhibition of distracting stimuli as revealed by distributional analysis. Saccade programming directed at relevant goals may be disrupted by the appearance of bright colors, sudden flashes, or fastmoving objects. The ability of these types of stimuli to seize our attention automatically and orient it to the eliciting event has been the object of a considerable amount of research in the last

The present study was supported by grants from the Netherlands Organisation for Scientific Research (NWO) to the second author. Address reprint requests to: Jasper G. Wijnen, Amsterdam Center for the Study of Adaptive Control in Brain and Behavior, Department of Psychology, Roetersstraat 15, 1018 WB Amsterdam, the Netherlands. E-mail: [email protected] 1

2 incompatible was increased. Likewise, Ridderinkhof (2002) varied the frequency of congruent to incongruent trials using the Simon task, where subjects have to ignore location-based information and should instead base their response on stimulus color. When congruent trials (in which location corresponded spatially to the response indicated by the color) were more frequent, subjects responded faster in congruent compared to incongruent trials, whereas the reverse was true when incongruent trials were more frequent. This pattern of reversed congruity effects in contexts associated with a high frequency of incongruity was also found by Logan and Zbrodoff (1979) using a Strooplike task. A number of explanations have been offered for these findings. Logan and Zbrodoff (1979) suggested that subjects adopt a strategy that allows them to be guided more by distracting stimuli when such distractors provide valid cues for the to-be-issued response. Gratton et al. (1992) distinguished an early and rapid parallel mode of processing from a subsequent and slower focused mode. When levels of distraction are high, subjects shift to a slower and more focused mode of stimulus processing. In their view, a top-down process determines the choice between the two modes of processing, which, in turn, may be affected by the context of the stimulus (i.e., expectancies with respect to the amount of congruent and incongruent noise) and the expected utility of one strategy over another. In a similar vein, Ridderinkhof (2002) interpreted the reversal of the Simon effect as the result of a strategic adjustment of the task set (the set of guidelines that determines how perception is turned into action). Subjects may be inclined to act impulsively or with caution depending on the expected gains and risks. This degree of cautiousness may be reflected in a selective suppression of response activation triggered by distracting stimulus features. De Pisapia and Braver (2006) put forward a neurocomputational network incorporating some of these strategic processes by invoking proactive as well as reactive control mechanisms. When incongruent distractors occur frequently in an experimental block, a sustained proactive control is generated, facilitating active maintenance of task-set information and able to bias information processing toward a more exclusive focus on task-relevant stimulus features. A second reactive control mechanism suppresses task-irrelevant information starting as soon as conflict is detected. The model provided an excellent fit for empirical reaction time data gathered using a Stroop color word task. In addition, activation in the model’s conflict unit corresponded closely to activation in medial frontal cortex as measured by brain imaging methods (fMRI). Maintaining high levels of cognitive control may not be the most appropriate thing to do under all circumstances, first because maintaining such control is effortful and second because acting more impulsively is beneficial to speed and may be acceptable when the risk of errors is low. Convergent evidence points to a number of specific regions in the brain that play a role in evaluating such risks. Posterior areas of medial frontal cortex (MFC), in particular the rostral cingulate zone (RCZ), are implicated in monitoring the risk of errors or more generally the risk of losing reward (e.g., Brown & Braver, 2005; Magno, Foxe, Molholm, Robertson, & Garavan, 2006; for a review, see Ridderinkhof, Ullsperger, Crone, & Nieuwenhuis, 2004). These structures are thought to signal the need for increased control to other regions in the brain (e.g., lateral prefrontal cortex) that can implement appropriate behavioral adjustments. More dorsal parts of posterior MFC (in particular the supplementary motor area [SMA] and the adjacent pre-SMA) have been associated

Wijnen and Ridderinkhof with goal-directed action selection (Nachev, Rees, Parton, Kennard, & Husain, 2005; Rushworth, Buckley, Behrens, Walton, & Bannerman, 2007). One psychophysiological measure that has been reliably associated with control adjustments and goal-directed action selection is an event-related brain potential (ERP) labeled the N2. The N2 component appears as a negative frontocentral shift occurring between 250 and 350 ms after stimulus presentation. Enlarged N2 amplitudes in conflict tasks have been interpreted to reflect RCZ activity related to monitoring the preliminary activation of erroneous responses (Yeung, Botvinick, & Cohen, 2004) or pre-SMA activity related to selecting the appropriate action in the face of competing ones (Burle, van den Wildenberg, & Ridderinkhof, 2009). In oculomotor capture tasks, the N2 may similarly reflect the activity of these dorsal MFC areas in goal-directed action selection and/or in monitoring the preliminary activation of erroneous saccades. On the behavioral side, a number of measures exist that can provide useful information on the strength of cognitive control. Strong control during conflict tasks (such as the Eriksen Flanker, the Stroop, and the oculomotor capture tasks) may reduce overall effects of congruence on accuracy and RT. Such a reduction of the congruence effect is found when salient task-irrelevant elements appear further apart in time from target stimuli (e.g., Eimer & Schlaghecken, 1998), indicating that selective suppression of task-irrelevant elements may build up and become more efficient over time. This buildup of efficiency is also evident when subjects take longer to respond, resulting in reduced interference effects for slower responses (Burle, Possamaı¨ , Vidal, Bonnet, & Hasbroucq, 2002). In general, spatial conflict tasks show large interference effects in the faster portions of the RT distribution, which progressively decrease with longer response times (Ridderinkhof, van den Wildenberg, Wijnen, & Burle, 2004). This progressive decrease is sensitive to factors that modulate the level of control, such as strategic factors (Ridderinkhof, 2002), intraindividual variability (Burle et al., 2002), or interindividual variability (Ridderinkhof, Scheres, Oosterlaan, & Sergeant, 2005; Wylie, Ridderinkhof, Eckerle, & Manning, 2007).

The Present Study In the oculomotor capture task (Theeuwes et al., 1998), subjects are confronted with six figure-eight shapes contained in gray circles, presented equi-spaced on an imaginary circle around a central fixation point. After 1 s all but one of the circles change color and all figure-eight premasks change into letters. The character in the circle that does not change color changes into a C or a reversed C. Subjects carry the instruction to quickly make a saccade to the circle that retained its color and to react subsequently with a button press to the orientation of the C. On some of the trials an irrelevant additional circle abruptly appears simultaneously with the color change. When such an object is presented, in many instances the subjects’ eyes first move into the direction of the abrupt onset before moving on to the target circle. To make the paradigm more suitable for the distribution-analysis method described earlier, we developed a modified version that includes congruent and incongruent conditions (as opposed to the distractor and no distractor trials used in the original), thus allowing us to record congruence effects (Wijnen & Ridderinkhof, 2007). Participants viewed four circles and were asked to respond to the one that did not change color while being simultaneously distracted by a displacement of one of the other circles. They were

3

Contextual modulation of oculomotor control instructed to indicate the side of the screen that contained the target by making a saccade toward that side. Incongruent trials were defined as trials where the target and the distractor appeared in different hemifields. Distractors appeared on average several hundreds of milliseconds before the target, eliciting a strong exogenous pull of the eyes toward their location. On incongruent trials, such distractors are generally considered to initially activate the incorrect response, leading either to a fast error or selective suppression of the response toward the distractor. The initial activation is seen most clearly in conditional accuracy functions (plotting accuracy against RT), whereas the subsequent selective suppression is expressed most clearly in delta plots derived from cumulative density functions. In a context where the risk of erroneous saccades as driven by incongruent distractors is more prevalent, proactive control is expected to increase. This strategic adaptation to the context will enable participants to process conflicting stimuli more efficiently and will make ignoring irrelevant features easier. Because the role of the distractor is reduced, effects of congruence on overall RT and accuracy are expected to be weaker as well. An important focus in the current study is on the specific mechanism by which such performance-enhancing strategy shifts are implemented on a neurophysiological level. Increased proactive control might result in decreased capture of overt attention by the distractor, in increased inhibition of response activation triggered by the distractor, or both. If increased proactive control in ‘‘risky’’ contexts results in reduced oculomotor capture, then this effect on initial distractor-driven saccade activation should be manifest in improved accuracy for fast responses, as expressed in conditional accuracy functions (CAFs) as well as in a smaller congruence effect on the N2 (reflecting facilitated action selection and less need for top-down control). If increased proactive control results in stronger inhibitory oculomotor response control, then this should be expressed in RT distributions by showing a greater reduction of interference effects for slower responses. Additionally, congruence effects on N2 amplitudes should then be larger for the risky context condition, reflecting stronger cognitive control responses. Behavioral evidence consistent with the latter hypothesis was observed previously in similar context effects using a (manual) Simon task (Ridderinkhof, 2002). Note that although a distinction between these two hypotheses cannot be made by looking at overall RTs and accuracy in isolation, the additional information provided by the analyses of RT distributions will allow us to make a strong case for one hypothesis over the other while convergent evidence can be obtained from the physiological data. The present study will test these hypotheses for the case of oculomotor control. Oculomotor capture tends to invoke selective response suppression more strongly than versions of conflict tasks with manual responses (Wijnen & Ridderinkhof, 2007). The present study will add to our understanding of adaptive control in conflict tasks in two ways: first, by examining the extent to which contextual modulations of oculomotor capture involve distractor-driven saccade activation and/or subsequent selective suppression of such erroneous saccades and, second, by examining how activity in the dorsal MFC (as expressed in the N2) is modulated as a function of the risk of erroneous oculomotor activation. To that end, the probability of distractor incongruity was manipulated parametrically in a conflict task previously shown to be suitable for examining oculomotor capture and selective saccade suppression (Wijnen & Ridderinkhof, 2007).

Method Participants Twelve undergraduate students (9 female; mean age: 20.33 years, SD: 2.27, minimum: 18, maximum: 25) from the University of Amsterdam took part in the experiment and obtained course credits for their participation. All participants reported to be healthy and to have normal or corrected-to-normal vision. Subjects were tested individually in a quiet and dimly lit university chamber. Stimuli Participants sat in a comfortable chair at a distance of 55 cm in front of a computer screen on which stimuli where shown against a black background. A small white cross (subtending a visual angle of approximately 0.51  0.51) served as a fixation point during the experiment. Each trial started with the presentation of the fixation point only. After 400 ms four green circles (diameter 1.61) were added to the display. These were situated in each of the four corners of the display with a vertical distance of 2.51 of visual angle and a horizontal distance of 10.81 of visual angle from fixation. The four circles were presented for a duration of 400 ms and for another 200 ms without the presence of the fixation cross. At this time, one of the circles changed position, moving vertically to a spot at the same height as the fixation point. After a jittered stimulus onset asynchrony (SOA; 20, 80, 140, 200, 260, 320, or 380 ms), three of the four circles changed from a green to an equiluminant blue color (as shown in Figure 1). The circle that changed position was never the one that retained its original color. This configuration remained on the screen for 1000 ms, after which the next trial started. On each trial participants were asked to respond with an eye movement to the circle that did not change color (sustained color-singleton target). The circle displacement was irrelevant for the execution of the task (offset–onset distractor). Trials in which the target and the distractor were presented within the same hemifield are designated congruent trials; in all other cases the trial is labeled incongruent (congruence manipulation). Procedure and Design The experiment involved one session lasting approximately 2 h. After participants received one practice block of 224 trials, electrodes were attached to the them (as described below) and

Figure 1. A possible incongruent experimental trial. The circle positioned top-left moves to the center-left position, forming the distractor. After a variable interval, three of the circles change color. The circle positioned top-right is the only one not to change color and forms the target.

4 they received another block of practice. After that, participants entered the experimental phase, which consisted of nine blocks of 224 trials. In one third of the blocks, 75% of trials were congruent; in another third of the blocks, 75% of trials were incongruent. In these blocks the participants were informed of the context condition with the message that, ‘‘In this block the moving circle will frequently, but not always, appear in the same/ opposite half of the screen as the unique circle.’’ For the rest of the blocks and for the two practice blocks the ratio of congruent to incongruent trials was 1:1. Participants were also informed of this ratio (except in practice blocks). In all blocks it was stressed to the participant to withhold their response until the circles on the screen changed color, but to react as rapidly as possible when the target appeared. The sequence of block types was counterbalanced over subjects with the constraint that all blocks of one context type were always consecutive. Apparatus Recordings of the electroencephalogram (EEG) were made from 30 leads using an Easycap with sintered Ag-AgCl electrodes referenced to the left earlobe. Vertical electrooculogram (VEOG) was recorded bipolarly from electrodes placed above and below the left eye, and horizontal EOG was recorded bipolarly from electrodes placed on the outher canthi of the two eyes. A ground electrode was placed at AFz (10/20 system). Electrode impedances were kept below 10 kO. EEG was recorded using Neuroscan amplifiers and Scan 4.2 software with a sample rate of 500 Hz with low-pass filter at 30 Hz and time constant of 1 s. Analysis of the Oculomotor Data Saccade RTs were determined with HEOG using a procedure adopted from Eimer and Schlaghecken (2001). The HEOG was epoched off-line into intervals starting 100 ms before distractor onset and ending 1400 ms after distractor onset. The 100-ms interval prior to distractor onset served as baseline. Saccade onset was determined for individual trials in the following way. Each HEOG epoch was scanned within the 1400-ms interval following distractor onset for the first uninterrupted positive-going or negative-going deflection, consisting of a series of at least 15 consecutive data points (equivalent to 30 ms) and exceeding an amplitude of  30 mV. For saccade accuracy purposes, trials were scored as an error trial if the first eye movement that satisfied the above criteria moved away from the target according to HEOG. When the criteria were met, the latency of the first data point of that time series was defined as saccade onset. A reliable saccade onset time could be scored for 99.4% of trials using these criteria. In the analysis of saccadic reaction times (SRTs), distractortarget SOA was subtracted from the SRTs. When the resulting SRT was faster than 100 ms or a saccade was detected before the presentation of the target, then the saccade was considered a fast guess and the trial dropped from analysis (21.1%). Outliers were removed from the data using a recursive procedure adopted from McCormick (1997), which involved temporarily removing the fastest and slowest RTs from each condition for each participant. After removal the mean and standard deviation for the remaining data was calculated. If either of the two removed data points fell outside an interval bounded by 4 SD from the mean, that data point was removed permanently. If the data point fell within the interval it was returned to the data set. This procedure continued until no more data points were removed permanently. By using this procedure 0.20% of the data were removed as outliers.

Wijnen and Ridderinkhof To capture the dynamics of the congruence effect, the SRTs for each subject were rank-ordered for each of the context and congruence conditions. Only correct trials were used for this purpose. The SRTs were then divided into five equal-size speed bins, thus comprising vincentized cumulative density functions (CDFs; Vincent, 1912). Delta plots were made by computing the congruence effect for each bin and plotting this against the average SRTcollapsed across congruence conditions for that speed bin and context condition. Conditional accuracy functions were computed to investigate the occurrence of fast guesses and incidence of oculomotor capture in the different conditions. Again, for each of the conditions five speed bins were created but now including errors and fast guesses (with the exception of saccades made within 100 ms after distractor presentation [3.54% of all trials]). For each speed bin, accuracy and mean SRTwere determined. Outliers were removed as before (0.12%). Analyses of variance (ANOVAs) with repeated measures were employed to assess the experimental effects on SRT and saccade accuracy. The within-subject variables in both ANOVAs were congruence, context, and speed bin. For illustrational purposes a second set of analyses was done on these same data, with accuracy scores reversed for incongruent trials. Rather than showing the percentage of correct saccades per experimental condition and speed quantile, as in CAFs, the percentage of saccades to the distractor is shown, creating in effect a conditional capture function (CCF). As the evidence for a particular target location is still meager for reactions that fall within the first quantiles, CCFs for both congruent and incongruent trials are expected to start at about the same percentage level. Because on congruent trials the distractor location is also the target location, the percentage of eye movements toward the distractor should quickly reach an asymptote near 100% for this condition. For incongruent trials CCFs are instead expected to develop toward an asymptote near 0%. Furthermore, CCFs for incongruent trials are expected to reach this asymptote quicker when the block contains mostly incongruent trials. Analysis of EEG Ocular artifacts were removed from the data by splitting the signal into independent components using the independent component analysis procedure as implemented in Brain Vision Analyzer. Components corresponding to blinks and horizontal eye movements were identified and removed from the raw signal. To remove the large slow-frequency P3 component from the raw data, a bandpass filter (3.5–20 Hz; cf. Donkers & van Boxtel, 2004) was applied off-line. ERPs were created by extracting single trial distractor-locked epochs starting from 200 ms predistractor to 1400 ms postdistractor. The 200-ms interval preceding the onset of the distractor was used as a baseline. After we set the baseline, the data were segmented again, this time locked to the target ( 200 to 1000 ms). Epochs containing muscle and other artifacts were excluded from further analysis. Segments were removed when the minimum and maximum amplitude within the segment differed more than 100 mV, when amplitudes exceeded 100 mV or 100 mV, or when two consecutive samples differed by more than 50 mV. After that, averaged waveforms for congruent and incongruent trials were generated, separately for each of the three context conditions and for each subject. Subject averages were scanned for the occurrence of an N2 defined as the most negative peak in a 160–380-ms time window after the target stimulus. A repeated measures ANOVA was executed with congruence and context as within-subject fac-

Contextual modulation of oculomotor control

5

tors and N2 amplitude at six midline electrodes (Fz, FCz, Cz, CPz, Pz, and Oz) as dependent variables. Results All reported p values were corrected using the Greenhouse– Geisser adjustment of the degrees of freedom when necessary. Overall RT and Accuracy SRTs on congruent trials were significantly shorter compared to SRTs on incongruent trials, F(1,11) 5 14.69, p 5 .003. In addition, more errors were made on incongruent trials compared to congruent ones, F(1,11) 5 21.97, p 5 .001). Most important, context modulated the effects of congruence on both RTs and accuracy, with the largest effects of congruence appearing in the 75% congruent blocks (RT: F[2,22] 5 10.25, p 5 .001; Accuracy: F[2,22] 5 14.75, po.001; see Figure 2). The main effect of context (RT: F[2,22] 5 2.20, p 5 .134; accuracy: F[2,22] 5 7.07, p 5 .004) is a direct derivate of the interaction between context and congruence. Because the main effect of context pools across congruent and incongruent conditions that differ in trial counts, interpretation of this main effect is not straightforward. Fast Guesses Subjects were more likely to guess fast in blocks containing mostly congruent trials and least likely in mostly incongruent blocks (see Figure 3). A fast guess was defined as a reaction given prior to 100 ms after target presentation. The mean number of fast guesses differed significantly between the context conditions, F(2,22) 5 8.52, p 5 .002. Distributional Analysis Delta plots show a sharp reduction of the congruence effect with increasing RTand even a reversal of the congruence effect at the slowest RTs for the experimental blocks with mostly incongruent trials (see Figure 4). A two-way interaction effect between speed bin and congruence showed this reduction to be significant, F(4,44) 5 46.92, po.001. Importantly, this interaction was not modulated by context, F(8,88) 5 0.25, p 5 .826. CAFs (Figure 5a) revealed a large number of fast errors, especially on incongruent trials. A two-way interaction between speed bin and congruence confirmed the difference between congruent and incongruent conditions, F(4,44) 5 27.80, po.001. Speed bin and context did not interact, F(8,88) 5 2.14, p 5 .094. Fast errors on incongruent trials were particularly prevalent in experimental blocks where most trials were congruent. In contrast, fast errors on congruent trials were most likely to be seen in mostly incongruent blocks. These effects of context dissipated with longer SRTs (Speed bin  Congruence  Context: F[8,88] 5 6.44, p 5 .001). Reversing the accuracy scores for incongruent trials created a CCF that plots the percentage of eye movements toward the distractor against mean SRT (Figure 5b). This illustrated that 77.23% of the fastest saccades (those in the first quintiles) initially went to the distractor. No effects of congruence, F(1,11) 5 1.24, p 5 .29, were found on the first quintile, nor was there an interaction between congruence and context, F(2,22) 5 0.83, p 5 .68, corresponding to the lack of such effects in fast guesses. As in the fast guesses, context did have a reliable effect on the direction of reactions in the first quantile, showing less capture by the distractor on mostly incongruent blocks, F(2,22) 5 14.87, po.001. An interaction between context and

Figure 2. Mean saccade reaction times (top), error rates, error rates in the first CAF quantile and N2 amplitudes at FCz (bottom) in congruent (light gray) and incongruent (dark gray) trials for the three context conditions (MI: mostly incongruent, EQ: 50% congruent, MC: mostly congruent). Error bars represent the standard error of the mean.

quantile replicates what was found in the CAF analyses: Context effects dissipate as both asymptotes are reached, F(8,88) 5 6.12, po.001. Because incongruent trial scores are reversed, this finding no longer shows up as a three-way interaction between context, congruence, and quantile, F(8,88) 5 1.06. EEG Analysis The midline ERP waveforms are shown in Figure 6 as a function of congruence and context.1 The analyses of N2 amplitude revealed a pattern reminiscent of our earlier results on the SRT 1 A lateralized ERP component known as the N2pc might be useful in determining the extent to which target and distractor captured attention (cf. Hickey, McDonald, & Theeuwes, 2006). However, the current experimental design did not lend itself to compare N2pc effects across experimental conditions in a meaningful way, because both target and distractor are present on every trial and neither was ever presented on the vertical meridian.

6

Figure 3. Incidence of fast guesses defined as a saccade executed earlier than 100 ms posttarget. The bars represent the percentage of fast guesses in the mostly incongruent (MI, left) condition, the context condition where the ratio of congruent to incongruent trials was equal (EQ, middle), and the mostly congruent condition (MC, right). Error bars represent the standard error of the mean.

and accuracy data. N2 amplitudes were larger on those trials where accuracy was low (and the risk of making errors high; see Figure 2). Congruence did not have a significant effect on N2 amplitude except on the Oz electrode (Fz: F[1,11] 5 3.06, p 5 .108; FCz: F[1,11] 5 2.07, p 5 .179; Cz: F[1,11] 5 2.56, p 5 .139; CPz: F[1,11] 5 1.97, p 5 .188; Pz: F[1,11] 5 2.50, p 5 .142; Oz: F[1,11] 5 7.83, p 5 .017), where incongruent trials tended to evoke larger N2s. Most important, this effect of congruence was modulated by context at all electrode sites (Fz: F[2,22] 5 8.61, p 5 .009; FCz: F[2,22] 5 8.57, p 5 .007; Cz: F[2,22] 5 13.59, p 5 .001; CPz: F[2,22] 5 17.18, po.001; Pz: F[2,22] 5 15.27, po.001; Oz: F[2,22] 5 6.26, p 5 .014; see Figure 6). The N2 effect (the increase in N2 amplitude in incongruent compared to congruent conditions) was most pronounced for experimental blocks where only 25% of trials were incongruent. Neither congruence nor the modulation of congruence by context interacted with electrode position (Congruence  Position: F[5,55] 5 0.82, p 5 .94; Congruence  Context  Position: F[10,110] 5 1.73, p 5 .18; see Figure 7). Note that Greenhouse–Geisser adjustments for the degrees of freedom were found to be necessary on some but not all electrodes (as indicated by Mauchley’s test of sphericity).

Discussion The present study investigated how oculomotor capture might be affected by the context in which distracting and target objects appear. The dynamics of capture by salient distractors and the successful inhibition of such capture can be examined by inspecting cumulative density functions and delta plots, which can provide information on the timing at which selective response inhibition processes may start to override oculomotor capture. Alternatively, the effects of capture may be visible in conditional accuracy functions and conditional capture functions, which can zoom in on early control processes by comparing the incidence of capture in the fastest saccades versus the slower ones. Stimulus contexts, such as the likelihood of incongruent distractors or of making an error, may influence processes leading to oculomotor capture and affect the strength or speed with which inhibition of distractors is invoked. To examine such contextual modulations, we presented participants with visual stimulus ar-

Wijnen and Ridderinkhof

Figure 4. Delta plots displaying the congruence effect size (incongruent minus congruent; y-axis) plotted against mean reaction time across congruence conditions (x-axis) for five speed quantiles. Experimental blocks in which 25% of trials was congruent (MI) are displayed with white diamonds. Black squares are used for blocks where 50% of trials was congruent (EQ). Mostly congruent blocks are displayed with white triangles (MC).

rays and asked them to ignore salient distractors in these arrays while making an eye movement to a target singleton. Trials could be either congruent when target and distractor appeared in the same hemisphere or incongruent if target and distractor appeared in opposite hemispheres. Stimulus context was manipulated by varying the ratio of congruent to incongruent trials between blocks. EEG was used to record brain activity during the task. ERPs elicited by the target were extracted using standard signal averaging procedures. We expected a component of the ERP thought to be sensitive to MFC-mediated control processes, the N2, to be modulated by (reactive and/or proactive) oculomotor response control adjustments caused by the trial context. Effects of context clearly impacted overall SRT and accuracy data. Subjects reacted more quickly and were more accurate on congruent trials, but these differences almost vanished in blocks where incongruent trials were most prevalent and were enhanced in mostly congruent blocks. Conversely, costs were incurred for rare congruent and incongruent trials, where targets appeared at unexpected locations. In terms of De Pisapia and Braver’s (2006) dual control network, these effects could be explained by assuming that proactive control is stepped up in the risky mostly incongruent blocks. This may lead to biasing of information processes to focus more exclusively on stimuli that are relevant to the task or to a strengthening of reactive control in this risky context, serving to inhibit activation that may result in a saccade to the distractor. Evidence Pointing to Stronger Proactive Control Leading to Avoidance of Capture When the distractor location frequently indicates the subsequent target location, as in our mostly congruent blocks, the oculomotor system may strategically use this information to prepare for a saccade in the direction of the first salient object entering the visual field. In such relatively safe contexts, effortful proactive control is kept relatively low. Stimulus salience rather than goalrelated factors are dominant in determining which action gets expressed in behavior and, consequently, the distractor becomes a powerful force for triggering saccades. CAFs and CCFs indicate that, for the safe mostly congruent context, fast saccades went to the distractor location by an overwhelming margin of almost 85% regardless of the congruence condition. Significantly

Contextual modulation of oculomotor control

7

Figure 5. Conditional accuracy functions (CAF) and conditional capture functions (CCF) displaying the percentage of errors and the percentage of eye movements toward the distractor (y-axis) plotted against mean reaction time (x-axis) for five speed quantiles. Incongruent trials are displayed with gray lines. Black lines and gray markers are used for congruent trials. Separate CAFs and CCFs are shown for mostly incongruent blocks (MI, diamonds, thin lines), 50% congruent blocks (EQ, squares, lines with intermediate thickness), and mostly congruent blocks (MC, triangles, thick lines).

fewer saccades went immediately to the distractor in the riskier context conditions. Fast guesses were also more prevalent in mostly congruent blocks, suggesting that response thresholds may be lowered with weak proactive control. Evidence Pointing to Stronger Proactive Control Leading to More Efficient Reactive Control In all of the context conditions the appearance of the distractor may have triggered inhibition of its location in order to prevent an eye movement to it. This type of control can be considered reactive, as control develops from the moment a conflict (in this case either between the salient distractor and the instruction to react only to the target or between distractor and target after both have been presented on an incongruent trial) is detected. Earlier studies indicated that, when subjects take a longer time to respond, the effect of the distractor on oculomotor response time becomes smaller, indicating a suppression of task-irrelevant elements that builds up over time (Wijnen & Ridderinkhof, 2007). This effect was replicated in this study. The buildup of selective suppression yielded a reduction (and in some cases even a reversal) of the congruence effect for slower reaction times. A reversal of congruence effects is difficult to explain in terms of passive dissipation of distractor-triggered activation over time; instead, such cases point to an active process that selectively inhibits response activation caused by the distractor, thus yielding relative facilitation of the correct response. Most important, SRT distributional data did not yield any indication that the buildup of selective suppression was differentially affected by the context of the trial, either in strength, latency, or buildup speed. Delta plots for each of the three context conditions were similar in slope and were only shifted along the y-axis. In the risky context conditions, the effects of congruence on RT started lower for the fastest saccades, indicating that, at the time of execution of these earliest saccades, interference effects have already been partially prevented or resolved. Context Modulations of the N2 Effect Reflect the Need for Ad Hoc Reactive Control Adjustments ERP results reveal effects of congruence on the N2 peak (starting approximately 230 ms after target onset) that are modulated by the context of the trial. An enlarged N2 peak for incongruent

trials is reliably found in the Eriksen Flanker task (Bartholow et al., 2005; Heil, Osman, Wiegelmann, Rolke, & Hennighausen, 2000; Kopp, Rist, & Mattler, 1996; Yeung et al., 2004), which also calls for the selective inhibition of a response triggered by salient but irrelevant stimulus features. The findings in these studies and the present study are consistent with the idea that the N2 reflects control processes (such as monitoring for activation of incorrect responses or goal-directed selection of appropriate actions in the face of competing ones) called upon to avoid errors, as triggered by distracting stimuli that evoke responses that are incompatible with the goal-directed response. Such a control adjustment is reactive in the sense that it is only invoked after the presentation of conflicting stimuli. The congruence effect on the N2 was modulated by trial context, which, as suggested by our distributional analyses, leads to avoidance of capture rather than stronger suppression of distractor-triggered activation. The pattern of these modulations showed that the N2 effect of congruence is reduced in riskier contexts and enhanced in safer contexts. These results can be explained by assuming that proactive control, guided by the knowledge of the relative frequency of congruent and incongruent trials, prepares the oculomotor system for the most likely type of trial. Thus, subject preparedness to deal with incongruent trials is high in risky, mostly incongruent blocks, and, because the analysis of RT distributions indicates that such preparation enables subjects to avoid paying attention to the distractor (rather than leading to a stronger suppression of distractor- triggered activation), the need for forceful reactive control adjustments is not quite as high as in the other two context conditions, leading to a smaller effect of congruence on the N2 amplitude. In safe, mostly congruent blocks, a congruent trial is expected, and bottom-up saliency is permitted to take a more active role in governing response selection. When a rare and unexpected incongruent trial is then presented, reactive control needs to step in swiftly to prevent an error. The amplified N2 in this interpretation represents (signaling of erroneous activation calling for) enhanced demands on goal-directed action selection. It is worth examining a number of alternative possible explanations for the interaction between congruence and context on N2 amplitude. A case could be made that the N2 represents a signal

8

Wijnen and Ridderinkhof

Figure 6. Grand average waveforms from six midline leads time locked to target onset. Negative voltage is plotted upward. Thin lines are used for congruent trials and thick lines are used for incongruent trials. The left column displays data from blocks in which most trials were incongruent (MI), the middle column is used for blocks where congruent and incongruent trials were equally prevalent (EQ), whereas the right-most column is used for blocks where most trials were congruent (MC).

that communicates the need for increased control on incongruent trials but is also affected by the well-established finding of enlarged N2 amplitudes for deviant infrequent ‘‘oddball’’ stimuli (see, e.g., the excellent review on the N2 by Folstein & van Petten, 2008). Although such a reading cannot be ruled out, it must be noted that it is only the combination of the locations of the target and the distractor that vary in probability between the context conditions, whereas studies where an N2 enlargement in reaction to stimulus infrequency is found usually employ simpler stimuli (i.e., a single letter or geometric figure is used for both standard and deviant stimuli). For instance, a study by Bartholow et al. (2005) using a

manual Eriksen flanker task found that the N2 was not sensitive to variations in the probability of congruency relations, suggesting that it is not the infrequency of the congruency relation per se that causes N2 enlargements akin to those found in oddball studies. Another explanation for the findings in this study might be found by adapting the hypothesis that the N2 reflects a mismatch or violation of internally generated expectations (Folstein & van Petten, 2008). Clearly, once the location of the distractor is revealed, subjects are likely to develop expectations as to the location of the target in both the mostly congruent and mostly incongruent blocks. Such an interpretation, however, would not

9

Contextual modulation of oculomotor control

that unexpected detection of response conflict and sudden need for recruitment of neural structures responsible for selective suppression is reflected in the larger N2 amplitudes for rare incongruent trials, these same structures, once engaged, do not ‘‘work harder’’ to inhibit erroneous responses. The absence of a significant effect of context on the delta-plot slopes indicates that buildup of selective suppression progresses equally fast for mostly congruent and mostly incongruent blocks, although the latter situation has a better starting position (i.e., there is less capture to begin with). It remains possible that a more automatic, exogenously driven and rapidly operating form of reactive attentional inhibition (that can be dissociated from a slower and more voluntary form of response inhibition; Eimer & Schlaghecken, 2003) is affected by context but is not expressed in the RT distribution; our data do not speak to this possibility.

Figure 7. N2 effects of congruence as a function of trial context and electrode position. Gray bars represent mostly incongruent (MI) blocks, darker gray bars are used for blocks where 50% of trials was congruent (EQ), and black bars represent mostly congruent blocks (MC).

necessarily invalidate an explanation in terms of response conflict. Indeed, certainly in the case of saccadic tasks expectations, for an important object to appear in the visual field might be neurally represented in the form of a buildup of activity for that location in structures responsible for oculomotor execution (i.e., the superior colliculi). Thus, once a target appears in a different location a response conflict immediately ensues. Effects of congruence on the N2 did not show a frontocentral maximum typical of conflict tasks. Although more centrally focused N2 effects of compatibility are not unprecedented (see Bartholow et al., 2005), the lack of an anterior focus might also be attributable to task characteristics unique to the present study or the fact that subjects responded with saccades rather than manually. Furthermore the removal of the P3 component, although helpful in highlighting the N2 component, might complicate comparisons across studies with respect to the source location of the N2 (not the strong point of the ERP method in any case). As the delta plots indicate, although safe, mostly congruent contexts seem to lead to strong MFC-mediated control reactions (on rare incongruent trials), this is not accompanied by stronger selective oculomotor suppression. Although it may be the case

Context Affects Proactive Control, Which, in Turn, Modulates the Need for Reactive Control In sum, in the present study the behavioral effects of trial context primarily seemed to affect early oculomotor capture and the magnitude of interference effects on overall SRT, but not the strength of selective suppression of incorrectly captured responses. In mostly incongruent contexts, proactive control was used so as to set up the system to protect against oculomotor capture, as evidenced by the behavioral as well as ERP data. In mostly congruent contexts, in contrast, proactive control yielded a less conservative regime, allowing the system to rely more on distractors. As a consequence, the incidental and unexpected incongruent distractor yielded excessive capture (witness the CAF and CCF data) and increased demands on selection of the appropriate saccade (witness the contextual modulation of the N2 effect). The increased level of incorrect capture was apparently resolved by goal-directed action selection processes (although we cannot exclude the possibility that automatic inhibition during earlier phases of attentional processing played a role). We conclude that the riskier contexts created by manipulating the ratio of incongruent to congruent trials mainly affects control processes that are proactive and, as such, are in place before the appearance of both distractor and target stimuli. Because proactive control prevents capture, this, in turn, affects the need for and preparedness of reactive control systems. In contrast, this study did not reveal effects of this manipulation on the speed and strength with which reactive control processes selectively inhibit activation triggered by irrelevant stimuli.

REFERENCES Bartholow, B. D., Pearson, M. E., Dickter, C. L., Sher, K. J., Fabiani, M., & Gratton, G. (2005). Strategic control and medial frontal negativity: Beyond errors and response conflict. Psychophysiology, 42, 33–42. Brown, J. W., & Braver, T. S. (2005). Learned predictions of error likelihood in the anterior cingulate cortex. Science, 307, 1118–1121. Burle, B., Possamaı¨ , C. A., Vidal, F., Bonnet, M., & Hasbroucq, T. (2002). Executive control in the Simon effect: An electromyographic and distributional analysis. Psychological Research, 66, 324–336. Burle, B., van den Wildenberg, W. P. M., & Ridderinkhof, K. R. (2009). Evidence for selective response activation and inhibition in conflict tasks: An electromyographic and electroencephalographic study. Manuscript in preparation. De Pisapia, N., & Braver, T. S. (2006). A model of dual control mechanisms through anterior cingulate and prefrontal cortex interactions. Neurocomputing, 69, 1322–1326.

Donkers, F. C., & van Boxtel, G. J. (2004). The N2 in go/no-go tasks reflects conflict monitoring not response inhibition. Brain & Cognition, 56, 165–176. Eimer, M., & Schlaghecken, F. (1998). Effects of masked stimuli on motor activation: Behavioral and electrophysiological evidence. Journal of Experimental Psychology: Human Perception and Performance, 24, 1737–1747. Eimer, M., & Schlaghecken, F. (2001). Response facilitation and inhibition in manual, vocal, and oculomotor performance: Evidence for a modality-unspecific mechanism. Journal of Motor Behavior, 33, 16–26. Eimer, M., & Schlaghecken, F. (2003). Response facilitation and inhibition in subliminal priming. Biological Psychology, 64, 7– 26. Folstein, J. R., & van Petten, C. (2008). Influence of cognitive control and mismatch on the N2 component of the ERP: A review. Psychophysiology, 45, 142–170.

10 Gratton, G., Coles, M. G. H., & Donchin, E. (1992). Optimizing the use of information: Strategic control of activation of responses. Journal of Experimental Psychology: General, 121, 480–506. Heil, M., Osman, A., Wiegelmann, J., Rolke, B., & Hennighausen, E. (2000). N200 in the Eriksen-task: Inhibitory executive process? Journal of Psychophysiology, 14, 218–225. Hickey, C., McDonald, J. J., & Theeuwes, J. (2006). Electrophysiological evidence of the capture of visual attention. Journal of Cognitive Neuroscience, 18, 604–613. Kopp, B., Rist, F., & Mattler, U. (1996). N200 in the flanker task as a neurobehavioral tool for investigating executive control. Psychophysiology, 33, 282–294. Kramer, A. F., Hahn, S., Irwin, D. E., & Theeuwes, J. (2000). Age differences in the control of looking behavior: Do you know where your eyes have been? Psychological Science, 11, 210–217. Logan, G. D., & Zbrodoff, N. J. (1979). When it helps to be misled: Facilitative effects of increasing the frequency of conflicting stimuli in a Stroop-like task. Memory and Cognition, 7, 166–174. Magno, E., Foxe, J. J., Molholm, S., Robertson, I. H., & Garavan, H. (2006). The anterior cingulate and error avoidance. The Journal of Neuroscience, 26, 4769–4773. McCormick, P. A. (1997). Orienting attention without awareness. Journal of Experimental Psychology: Human Perception and Performance, 23, 168–180. Nachev, P., Rees, G., Parton, A., Kennard, C., & Husain, M. (2005). Volition and conflict in human medial frontal cortex. Current Biology, 15, 122–128. Ridderinkhof, K. R. (2002). Micro- and macro-adjustments of task set: Activation and suppression in conflict tasks. Psychological Research, 66, 312–323. Ridderinkhof, K. R., Scheres, A., Oosterlaan, J., & Sergeant, J. (2005). Distribution-analytical techniques in the study of ADHD: Delta plot analyses reveal deficits in response suppression that are eliminated by methylphenidate treatment. Journal of Abnormal Psychology, 114, 197–215.

Wijnen and Ridderinkhof Ridderinkhof, K. R., Ullsperger, M., Crone, E. A., & Nieuwenhuis, S. (2004). The role of the medial frontal cortex in cognitive control. Science, 306, 443–447. Ridderinkhof, K. R., van den Wildenberg, W. P. M., Wijnen, J. G., & Burle, B. (2004). Response inhibition in conflict tasks is revealed in delta plots. In M. Posner (Ed.), Cognitive neuroscience of attention. New York: Guilford Press. Rushworth, M. F. S., Buckley, M. J., Behrens, T. E. J., Walton, M. E., & Bannerman, D. M. (2007). Functional organization of the medial frontal cortex. Current Opinion in Neurobiology, 17, 220–227. Ruz, M., & Lupia´n˜ez, J. (2002). A review of attentional capture: On its automaticity and sensititvity to endogenous control. Psicolo´gica, 23, 283–309. Theeuwes, J., & Godijn, R. (2001). Attentional and oculomotor capture. In C. L. Folk & B. S. Gibson (Eds.), Attraction, distraction, and action: Multiple perspectives on attentional capture (pp. 121–150). Amsterdam: Elsevier. Theeuwes, J., Kramer, A. F., Hahn, S., & Irwin, D. E. (1998). Our eyes do not always go where we want them to go: Capture of the eyes by new objects. Psychological Science, 9, 379–385. Vincent, S. B. (1912). The function of the vibrissae in the behavior of the white rat. Behavioral Monographs, 1, 1–82. Wijnen, J. G., & Ridderinkhof, K. R. (2007). Response inhibition in motor and oculomotor conflict tasks: Different mechanisms, different dynamics? Brain & Cognition, 63, 260–270. Wylie, S. A., Ridderinkhof, K. R., Eckerle, M. K., & Manning, C. A. (2007). Inefficient response inhibition in individuals with mild cognitive impairment. Neuropsychologia, 45, 1408–1419. Yeung, N., Botvinick, M. M., & Cohen, J. D. (2004). The neural basis of error detection: Conflict monitoring and the error-related negativity. Psychological Review, 111, 931–959.

(Received October 25, 2007; Accepted December 9, 2008)