Modulation of ERP components by task instructions in a cued go/no-go task

Transcription

1 Psychophysiology, 53 (2016), Wiley Periodicals, Inc. Printed in the USA. VC 2015 The Authors. Psychophysiology published by Wiley Periodicals, Inc. on behalf of Society for Psychophysiological Research DOI: /psyp Modulation of ERP components by task instructions in a cued go/no-go task IDA EMILIA AASEN a,b AND JAN FERENC BRUNNER b,c,d a Department of Psychology, Norwegian University of Science and Technology (NTNU), Trondheim, Norway b Department of Neuropsychology, Helgeland Hospital, Mosjøen, Norway c Department of Neuroscience, NTNU, Trondheim, Norway d Department of Physical Medicine and Rehabilitation, St. Olavs Hospital, Trondheim University Hospital, Norway Abstract The present study investigated how components of ERPs are modulated when participants optimize speed versus accuracy in a cued go/no-go task. Using a crossover design, 35 participants received instructions to complete the task prioritizing response speed in half of the task, and accurate responding in the other half of the task. Analysis was performed on the contingent negative variation (CNV), P3go, and P3no-go and the corresponding independent components (IC), as identified by group independent component analysis. After speed instructions, the IC CNV late, P3go anterior, P3no-go early, and P3no-go late all had larger amplitudes than after accuracy instructions. Furthermore, both the IC P3go posterior and IC P3go anterior had shorter latencies after speed than after accuracy instructions. The results demonstrate that components derived from the CNV and P3 components are facilitated when participants optimize response speed. These findings indicate that these ERP components reflect executive processes enabling adjustment of behavior to changing demands. Descriptors: ERP, ICA, Go/no-go, Executive control, Optimization, P3, CNV Executive function is a collective term for the neural processes that enable effective performance and adaptation to novel, conflicting, or complex situations (Lezak, Howieson, Bigler, & Tranel, 2012). The cued go/no-go task is an experimental paradigm that requires activation of executive functions to achieve fast and accurate performance. In addition to the effectiveness of specific executive processes, however, it has been suggested that specific mechanisms that facilitate other processes are of key importance for optimal performance (Forstmann et al., 2008; Stuss et al., 2005). In the cued go/no-go task, several executive and facilitating processes can be investigated. Just like the activation of any other cognitive process, the effect of facilitation is reflected in altered neural activity. The present study investigates such alterations when instructing participants to optimize performance speed or accuracy in a cued go/no-go task while keeping other task aspects unchanged. That way, the executive processes needed to perform The authors would like to thank John Polich for helpful comments on previous versions of this manuscript. The Regional Committees for Medical and Health Research Ethics in Norway approved the present study. Both authors contributed equally to this work, and the order of authorship is arbitrary. Address correspondence to: Ida Emilia Aasen, Department of Psychology, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway. ida.aasen@ntnu.no This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoNoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. 171 the task are the same in both conditions, while the relative facilitation of the different processes is affected. The neural correlates of optimization of speed or accuracy can be investigated by computing ERPs from EEG recorded during task performance. Previous studies have found that such manipulations can modulate both the latencies and amplitudes of several ERP components (e.g., Band, Ridderinkhof, & van der Molen, 2003; Boehm, van Maanen, Forstmann, & van Rijn, 2014; Christensen, Ivkovich, & Drake, 2001; Kutas, McCarthy, & Donchin, 1977; Pfefferbaum, Ford, Johnson, Wenegrat, & Kopell, 1983). A methodological challenge in ERP research, however, is the fact that ERPs represent the sum of activity from various locations occurring at overlapping points in time and representing different cognitive processes (Kappenman & Luck, 2012). A common approach to this challenge is to manipulate a task in a manner changing the overlap between different subcomponents (e.g., Hohnsbein, Falkenstein, & Hoormann, 1995; Verleger, Baur, Metzner, & Smigasiewicz, 2014; Verleger, Metzner, Ouyang, Smigasiewicz, & Zhou, 2014). These methods, however, necessitate knowledge about how to affect the latencies of the underlying components differentially, and do not enable separation of components that always follow each other closely in latency. Such manipulations can also not rule out that other unknown components affect the components of interest. Another approach to solving the problem of component overlap is the use of blind source separation methods, such as principal or independent component analysis (ICA; Makeig & Onton, 2012; Spencer, Dien, & Donchin, 2001). ICA is a method that uses the

2 172 I.E. Aasen and J.F. Brunner input data to learn how to separate the mixed signals recorded from multiple scalp locations into several temporally independent and spatially fixed information sources. ICA defines each information source by a spatial filter. When these filters are applied to data collected from many scalp locations, each independent component (IC) filter focuses on one source of information in the data, canceling out the contributions of all but one of the source signals contributing to the multichannel data (Makeig & Onton, 2012). In the present study, three well-studied ERP components, the contingent negative variation (CNV), P3go, and P3no-go were decomposed using group ICA to enable studying how the underlying components are modulated when individuals are instructed to optimize speed versus accuracy in the cued go/no-go task. In the following, the traditional ERP components will be referred to as raw ERP components, whereas the ICA-derived components will be referred to as ICs. The CNV Introduction of a cue activates preparatory processes that significantly reduce reaction time (RT; Niemi & N a at anen, 1981; Sanders & Wertheim, 1973). The CNV is a slow negative potential elicited in the time interval between the cue and the imperative stimulus. This potential consists of multiple components reflecting different preparatory processes (Brunia, van Boxtel, & B ocker, 2012; Rohrbaugh et al., 1997). CNVs presenting with the largest negativities early in the interstimulus interval have been linked to encoding of relevant stimulus characteristics (Bender et al., 2012; Ruchkin et al., 1997), as well as maintenance and rehearsal processes related to the cue information (Rohrbaugh et al., 1997). The late CNV presents largest negativity immediately before the imperative stimulus. When interstimulus intervals are short, the early and late CNV cannot be identified as separate components in the raw ERP. These components can, however, be separated by using ICA, even in paradigms with 1-s interstimulus intervals (Brunner et al., 2015; Jervis et al., 2007). The late CNV has been suggested to reflect activation of the contingency templates, or schema, for when and how to respond to the upcoming target stimulus (Brunia et al., 2012; Strack, Kaufmann, Kehrer, Brandt, & Sturmer, 2013). We refer to this schema as the task set. Unlike the early CNV, this component has been found to be larger under instructions for speed than for accuracy (Loveless & Sanford, 1974). Because it is modulated in response to time pressure, this potential has been suggested to be facilitated by effortful allocation of attention (Falkenstein, Hoormann, Hohnsbein, & Kleinsorge, 2003), or energization (Brunner et al., 2015). We prefer the concept of energization because it is specified as a facilitation process that is under voluntary control, and required for initiating and maintaining responses (Stuss & Alexander, 2007). This process is different from other, more automatic facilitation processes, such as arousal, emotional valence, or motivation, which also intensify processing, but do this effortlessly. In addition to facilitating the task set proactively, Brunner et al. (2015) suggested that energization is also required for processes occurring after the imperative stimulus. The P3 Family The most studied group of ERP components elicited after an imperative stimulus is the P3 family. These components are all regarded as reflecting cognitive processes, although the exact processes reflected in the different P3 components are a matter of much debate. What is better known are the effects that different manipulations have on these components. P3 amplitude is larger when participants devote more attention or effort to a task, but is smaller when tasks become more perceptually or cognitively demanding (Johnson, 1986; Luck, 2005). It is therefore important to differentiate between the cognitive load on the specific P3 process (the amount of information that must be handled by the P3 process) and the facilitation (i.e., the intensity) of this process when investigating P3 components experimentally. Often, these effects become mixed. For instance, making a task more difficult can simultaneously increase both cognitive load and facilitation. In the present study, cognitive load is therefore kept constant, while facilitation processes are affected by instructing participants to optimize speed or accuracy. In the cued go/no-go paradigm, two P3 components can be identified following the imperative stimulus: the P3go and the P3no-go. These components differ both in latency and topography, suggesting that they reflect different cognitive processes (Falkenstein, Hoormann, & Hohnsbein, 1999; Pfefferbaum, Ford, Weller, & Kopell, 1985). The processes reflected by different P3 components may therefore not be affected in the same way by a speed accuracy manipulation. Prediction of the effects of the speed accuracy manipulation on specific P3 components therefore depends on an understanding of which processes the different components reflect. The P3go: Functional significance and subcomponents. The parietal P3go has been suggested to be similar to the target P3, or P3b, from the oddball paradigm (Barry & Rushby, 2006). The latency of this component has often been said to reflect the time it takes to classify an incoming stimulus as a target (see Polich, 2007; Verleger, 1988). According to the context-updating theory (Donchin, 1981; Donchin & Coles, 1988; Polich, 2003, 2007), the P3b is suggested to index an attention-driven comparison process that evaluates a task-related, and subjectively unexpected, stimulus in working memory. This theory has received some criticism, and Verleger (1988) argued that the P3b is better described as being evoked by all target events rather than unexpected events only. There has been some debate on whether this ERP component is only related to stimulus evaluation, or whether it is also related to the specific response to be made. Using a simple RT paradigm, as well as two- and four-choice RT tasks, Falkenstein and colleagues (Falkenstein, Hohnsbein, & Hoormann, 1993, 1994; Hohnsbein et al., 1995) identified two partially overlapping P3 components the stimulus-related P-SR, and the choice-related P-CR. Whereas the P-SR had a central distribution and was not affected by RT or complexity of the choice, the P-CR had a parietal distribution and was clearly delayed with increasing RT and choice complexity. They interpreted this finding as indicating that the P-SR is related to stimulus classification, whereas the P-CR is related to response selection. Comparing stimulus- and response-locked components, however, Verleger, Jaskowski, and Wascher (2005) found that P3b amplitudes calculated as stimulus- and response-locked components were equally large. The amplitude of the centrally located component (P3a), however, was clearly larger for stimulus- than response-locked components. They therefore argued that the P3b represents a link between perceptual classification and response selection, rather than being related to perceptual processes or response selection alone. They concluded that the P3b might reflect the process of monitoring whether the stimulus is classified correctly to the appropriate response. Recently, Verleger, Baur et al.

3 Modulation of ERP components by task instructions 173 (2014) found further support for the view that the P3b represents a reactivation of a learned stimulus-response (S-R) link. They demonstrated that the P3b amplitude dramatically decreased when the S-R link was made less straightforward. Other processes in the go condition. Evaluative processes are frequently suggested to be active after errors and in conflict situations. The go condition does not involve the possibility of commission errors. This condition does, however, involve the possibility of slow responses errors of speed (Luu, Flaisch, & Tucker, 2000). The cause of errors of speed is inattention in the time interval just prior to and during the presentation of the target stimulus. Optimization of speed depends on detecting this type of inattention to enable maintaining or adjusting task-set activation as needed in the following trials (i.e., a process of self-monitoring). The existence of such a monitoring process that intercedes to bias attention back to the task-related schema when performance decays due to inattention has previously been suggested by Manly and colleagues (Manly, Robertson, Galloway, & Hawkins, 1999). Possibly, the two overlapping P3 components that have been observed in parietal and central areas may differentially reflect reactivation of the S-R link and self-monitoring. Speed accuracy effects on the P3go. Several studies have demonstrated that target-p3 latency is shorter under high as compared to low time pressure (Kutas et al., 1977; Pfefferbaum et al., 1983; Ramautar, Kok, & Ridderinkhof, 2004), which is in line with the suggestion that the P3b reflects an S-R link. Some studies have also found increased target-p3 amplitudes in conditions where response speed is increased (Carrillo-de-la-Pe~na & Cadaveira, 2000; Falkenstein et al., 1993; Friedman, Vaughan, & Erlenmeyer- Kimling, 1978; Roth, Ford, & Kopell, 1978). These amplitude changes could be the result of increased overlap in P3 subcomponents due to latency changes, or reflect real amplitude changes in one or several subcomponents. The use of ICA to separate P3go subcomponents when studying the effects of speed and accuracy instructions can contribute to answering these questions, and give indications about the functional significance of the different components. The P3no-go. The frontocentral P3no-go from cued go/no-go tasks is evoked when a no-go imperative stimulus unexpectedly follows a go cue. The exact process reflected in this ERP component is still debated. The P3no-go has been associated with inhibitory (Fallgatter & Strik, 1999; Gajewski & Falkenstein, 2013; Kok, Ramautar, De Ruiter, Band, & Ridderinkhof, 2004; Smith, Johnstone, & Barry, 2006; Wessel & Aron, 2014) and evaluative processes (Bruin, Wijers, & van Staveren, 2001; Schmajuk, Liotti, Busse, & Woldorff, 2006; Sehlmeyer et al., 2010). A recent review of studies on the P3no-go, however, concluded that the latter hypothesis has received more empirical support than the former (Huster, Enriquez-Geppert, Lavallee, Falkenstein, & Herrmann, 2013). In that review, it was suggested that the P3no-go could reflect either the evaluation of the outcome of inhibition, or an evaluation of the inhibitory process itself. Like the CNV and P3go, it has also been suggested that the P3no-go is comprised of overlapping components reflecting different processes. Brunner et al. (2013) noted that the two subcomponents that they identified seemed to reflect consecutive processes, with one of the components always appearing shortly after the other (IC P3no-go early and IC P3no-go late ). They suggested that these components might independently reflect the proposed inhibitory and evaluative functions of the P3no-go. In a follow-up study, however, the inhibitory hypothesis of the IC P3no-go early was not supported, whereas the monitoring hypothesis of the IC P3no-go late was supported (Brunner et al., 2015). Rather than being related to inhibition, it was suggested that the IC P3no-go early might reflect implementation of an alternative, subdominant task set. Others have also argued that processes evoked by the no-go condition may be better understood within a broader framework of response selection and execution processes, rather than inhibition only (Jasinska, 2013; Rae, Hughes, Weaver, Anderson, & Rowe, 2014). Speed accuracy effects on the P3no-go. The effects of speed and accuracy instructions on the P3no-go have not been previously studied, although several studies have found larger P3no-go amplitudes in fast as compared to slow responders (Dimoska, Johnstone, & Barry, 2006; Smith et al., 2006). An interesting study by Benikos and Johnstone (2009) also showed that the P3no-go amplitude is larger under fast as compared to slow event rates in healthy controls, indicating that faster responding may modulate the P3no-go amplitude intraindividually. Brunner et al. (2015) demonstrated that the IC P3no-go early correlated significantly with neuropsychological indexes of energization, including reaction time. This finding could indicate that the IC P3no-go early may be the subcomponent of the P3no-go that is modulated by response speed. The IC P3no-go late was not related to the energization indices, but was highly related to accuracy (negatively correlated to errors) and therefore suggested to be the subcomponent of the P3no-go reflecting a monitoring process. This component was suggested to be modulated by emotional factors. Although monitoring is often considered to be a cognitive process without any explicit emotional content, conflict situations and errors have been associated with emotional responses and activity in brain regions associated with negative affect (Ichikawa et al., 2011; Shackman et al., 2011). There is also some evidence that there is a relationship between the P3no-go amplitude and anxiety (Sehlmeyer et al., 2010). Study Aims and Hypotheses This study investigates whether ICs from the cued go/no-go task are differentially modulated when participants are instructed to optimize either speed or accuracy. A schematic model presenting the proposed processes involved in the different conditions of the task is presented in Figure 1. In line with the findings by Falkenstein et al. (2003) and Loveless and Sanford (1974), the raw CNV component was expected to be larger in the speed than in the accuracy condition. The amplitude IC CNV late, but not the IC CNV early, was hypothesized to be facilitated by the speed instructions. The processes proposed to be reflected in the CNV ICs are sketched on the left side of Figure 1. The effects on the raw P3go were expected to reflect previous findings, which have demonstrated latency effects on parietal, but not central, parts of the target P3 when RT is affected (Falkenstein et al., 1993; Verleger et al., 2005). Decomposing the P3go should result in one IC corresponding to the parietally maximal P3b, which was expected to have shorter latency in the speed as compared to the accuracy condition. The amplitude of the P3b component was not expected to be affected by the manipulation, as correct target-response mapping should be equally important to obtain both fast and accurate responses. It was also hypothesized that a subcomponent of the P3go, reflecting self-monitoring of attention, would show larger

4 174 I.E. Aasen and J.F. Brunner Figure 1. Schematic illustration of processes hypothesized to be involved in the cued go/no-go task. The CNV is thought to reflect at least two processes: encoding of cue information and attention to the dominant task set. In the go condition, the P3go is thought to reflect reactivation of the target-response link (A 5 go) and evaluation of whether the preparatory attention to the go task set was optimal. In the no-go condition, the P3no-go is thought to reflect implementation of the subdominant task set, and subsequent evaluation of the level of control over this implementation process. Several of these processes are hypothesized to be modulated by instructions to optimize speed or accuracy. The late CNV and the early P3no-go are suggested to be facilitated by an effortful, energetic process (energization). The monitoring process in the no-go condition is suggested to be modulated by emotional processes. The speed instruction is also hypothesized to result in quicker reactivation of the target-response link and more resources invested in evaluating preparatory attention. amplitude in the speed as compared to the accuracy condition. This is because the use of information from the current trial to finetune energization of the late CNV in the following trials would be particularly important in the speed condition. If a P3go IC reflects such a self-monitoring process, the amplitude of this subcomponent should also correlate with the IC CNV late amplitude. The proposed processes reflected in the P3go ICs are sketched on the right side, lower panel of Figure 1. The raw P3no-go amplitude was hypothesized to be larger in the speed than in the accuracy condition in line with studies finding larger P3no-go amplitudes in fast as compared to slow responders (Dimoska et al., 2006; Smith et al., 2006). The amplitude of the centrally maximal IC P3no-go early was hypothesized to be facilitated more in the speed than the accuracy condition, reflecting increased energization to implement a subdominant task set. If our working hypothesis is wrong and this component reflects response inhibition, however, the component should have shorter latency than the mean RT in both conditions and have a shorter latency in the speed than in the accuracy condition. The type of monitoring reflected in the amplitude of the frontocentral IC P3no-go late could be related either to the focus on evaluating whether the (non)response was correct, or to the evaluation of how close one was to committing a premature response in the given trial. Whereas the first alternative should result in larger amplitudes when focusing on accuracy than speed, the second alternative should result in larger amplitudes under high response speed. As previous studies of the P3no-go have only indicated increases in amplitude related to increased response speed, only the latter alternative is sketched in Figure 1. Participants Method A total of 26 female and 9 male university students (mean age years, SD 5 2.6) was recruited to participate in the experiment through presentations during lectures that the participants were attending. Written informed consent was obtained from all participants prior to their participation in the experiment. Criteria for exclusion were a history of neurological or psychiatric disorder, or previous experience with the cued go/no-go task. No participants were excluded from the study. The spatial filters for the independent components were obtained through application of ICA to ERPs recorded from the 35 participants, as well as 193 healthy adult volunteers (109 females, mean age years, SD 5 4.2, previously reported in Brunner et al., 2015). These volunteers completed the same cued go/no-go task that was used in the experiment, but under instructions to balance speed and accuracy (mean RT ms, mean number of commission errors 5 1.1). The volunteers were recruited among acquaintances of researchers and staff, family, and friends of braininjured patients, and a healthy control group from a study on prematurely born children. The resulting spatial filters were then applied to the ERPs of the 35 students recruited for the experiment. EEG Recording EEG was recorded using a 21-channel Mitsar ( EEG system, with a 19-channel tin electrode cap with electrodes Fz, Cz, Pz, Fp1/2, F3/4, F7/8, T3/4, T5/6, C3/4, P3/ 4, and O1/2, fitted according to the International system, and the cap electrodes were referenced to earlobe electrodes. The ground electrode was placed on the forehead (between the Fp1/2 and Fz electrodes). Impedance was kept below 5 kx. The bandpass was Hz, and the sampling rate was 250 Hz. It has previously been argued that slow wave components such as the CNV can be reliably extracted, even when such low frequency cutoffs are used (Padilla, Wood, Hale, & Knight, 2006). ERPs were computed offline in the common average montage. Trials with omission and commission errors were excluded from averaging. Behavioral Task Participants sat approximately 1.5 m from a computer screen while performing the cued go/no-go task, which consisted of 400 pairs of images, with a new pair presented every 3 s (the task used by Brunner et al., 2015). An illustration of the task and the stimulus categories can be seen in Figure 2. The task contained three categories of stimuli animals (A), plants (P), and humans (H), presented

5 Modulation of ERP components by task instructions 175 Figure 2. Illustration of the cued go/no-go task. Left: Participants were instructed to respond to all animal-animal (A-A) pairs (top) and withhold responses to the other three stimulus combinations. The cue and imperative stimulus in A-A and plant-plant (P-P) pairs were always identical. Human stimuli in P-H pairs were combined with a sound. Right: The size of the stimuli relative to the computer screen. in four equally probable combinations: A-A, P-P, A-P, or P-H. The first and second stimulus in A-A and P-P pairs were always identical. There were 20 different images in each category to reduce habituation due to repetition of stimuli. Also, to maintain alertness, the H stimuli were paired with different 70 db sounds consisting of different frequencies (500, 1,000, 1,500, 2,000, and 2,500 Hz). Each stimulus was presented for 100 ms, and the interstimulus interval between the first and second image in a pair was 1,000 ms. The images were selected from children s textbooks and were of approximately equal luminance and size. All images were presented against a white background, and a blank screen occupied the interstimulus intervals. The trials were grouped into four blocks of 100 trials. Each block contained a new set of five images of each category. The image pairs were presented pseudorandomly so that each block contained an equal number of trials of each type. Participants were instructed to press the left mouse button for every A-A (go) trial as quickly as possible and to withhold the response on every A-P (no-go) trial. All participants responded using the index finger on the right hand. No response was required for the P-P or P-H trials. All trials starting with A required subjects to continue attending to the second stimulus in the pair. Verbal speed accuracy manipulations have the possible disadvantage that the effect of these instructions wane over time in favor of the individuals automatic response tendencies (Heitz, 2014). To minimize this effect, participants had a short break every 50 trials, during which the task instructions were repeated. Mean RT, SD of the RT, and the number of omission and commission errors were calculated separately for each participant. A response was regarded as correct if it occurred 200 to 1,000 ms after the second stimulus in go trials. Instructions Under both conditions, participants were instructed to respond to the target stimuli as quickly and accurately as possible. In one condition, they were told to optimize speed and, in the other, to optimize accuracy. The instructions were as follows: Accuracy: I want you to respond as quickly and accurately as you can to the target stimuli, but most of all, it is very important that you do not commit a single mistake! Speed: I want you to respond as quickly and accurately as possible to the target stimuli. In this condition, however, I want you to really push yourself to be as fast as possible, even to the point where you might commit one or two premature responses. The most important thing is that you respond very quickly to all targets. Thus, although the weighting of speed and accuracy was manipulated, the instructions in both conditions were designed to keep participants within the limits of maintaining the conflicting demands of fast and accurate responses. The order in which participants received their instructions was randomized: 18 participants were instructed to focus on speed in the first half of the task, and 17 participants were instructed to focus on accuracy during the first half. After 200 trials, the instructions were reversed. Correction of EEG Artifacts Eyeblink artifacts were corrected by applying ICA to the raw EEG, zeroing the activation curves of ICs corresponding to eyeblinks (see Jung et al., 2000; Vigario, 1997). Epochs of EEG with excessive absolute amplitude in unfiltered EEG (> 100 lv), epochs of excessive faster frequency activity (> 35 lv infrequencybandsof Hz), or excessive slower frequency activity (> 50 lv infrequency bands of 0 1 Hz) were automatically marked and excluded from further analysis. After artifact rejection, the mean number of artifact- and error-free trials used to compute ERPs was 95.5 (SD 5 5.8; range ) for the accuracy cue condition (all trials starting with an animal, used to compute the CNV), and 90.7 (SD 5 6.0; range ) for the speed cue condition. In the go condition, the ERPs were computed for a mean of 47.7 trials (SD ; range ) in the accuracy condition, and 47.0 trials (SD ; range ) in the speed condition. In the no-go condition, the mean number of trials for computing the ERPs was 47.8 (SD 5 3.1; range ) for the accuracy condition, and 43.7 (SD 5 3.3; range ) for the speed condition.

6 176 I.E. Aasen and J.F. Brunner Decomposing ERPs Into Independent Components ICA can be applied to continuous EEG, single trial ERPs, or ERPs averaged across trials. By applying ICA to averaged ERPs from many individuals (group ICA; Kropotov & Ponomarev, 2009; Liu et al., 2009; Olbrich, Maes, Valerius, Langosch, & Feige, 2005), one can obtain ICs that are less affected by oscillatory activity and noise than ICA applied to single trial ERPs or EEG. Another advantage of group ICA is that, by applying the same spatial filters to all participants ERPs, the possibility of comparing the latency and amplitude of an IC peak across individuals is maintained. This method, however, assumes that any particular component has approximately the same topography in all individuals. This is an assumption that ignores differences in individual cortical configuration that may project equivalent potentials to different scalp areas in different subjects (Makeig & Onton, 2012). In the present study, ICA was conducted using the infomax algorithm (Makeig, Jung, Bell, Ghahremani, & Sejnowski, 1997) on a collection of 19-channel ERPs from 228 individuals. To obtain a reliable decomposition, ICA depends on a large number of training points to learn the weights in the unmixing matrix and, in general, larger amounts of clean input data result in more reliable decompositions. Onton and Makeig (2006) suggest that the number of training points should, as a rule of thumb, be larger than the number of electrodes squared (N 2 ) times a factor, k, whichagain depends on the number of electrodes, the algorithm used, data quality, and frequency range. For data collected from a large number of channels (> 64), Makeig and Onton (2012) suggest that k could be set at 20. The same heuristic has, however, not been investigated in data from fewer channels. Because of this, it was expected that a number of training points falling well above but below should be sufficient to obtain a reliable decomposition. In the present study, a sufficient number of training points was achieved by using ERPs from a large number of subjects (n 5 228) as input for the ICA decomposition. ICA was conducted separately for three different conditions. In the cue condition, ERPs were selected from the time interval of 0 1,100 ms after onset of the first stimulus (S1). In go and no-go trials, ERPs were selected from the time interval of ms after onset of the second stimulus (S2). ICA was performed on the ERP Scalp Location 3 Time series matrix. The output of the ICA is a number of temporally independent and spatially fixed components. The number of components equals the number of channels (19 in this case). The raw data for the group ICA, the ERPs, are expected to consist of a mixture of a few large source components plus many smaller source components. The ICA is expected to separate large source components into different components well, whereas the remaining components consist of mixtures of smaller source components and noise (Makeig et al., 1999). The ICAs for the go and no-go conditions were therefore calculated separately, to maximize the probability of finding all the relevant P3go and P3no-go components. Taking into account the 19 electrodes, the sampling rate of 250 Hz, and the time interval of 0.7 s, the matrix for go and no-go conditions consisted of 19 rows and columns, giving 39,900 points in time for computing the unmixing matrices for these conditions. ICA for the relevant (A) and irrelevant (P) cue condition was calculated conjointly to enable selecting components that differentiated between relevant and irrelevant cues. For ICA application in the cue condition, a time interval of 1.1 s was used, resulting in a matrix consisting of 19 rows and columns, giving 125,400 points in time for computing the unmixing matrix. Further details of this method are described in Kropotov and Ponomarev (2009). Measurement of Latency and Amplitude of ERP Components The amplitudes of the raw ERP components were measured in the three midline electrodes (Fz, Cz, Pz), with the exception of the P3 in the go condition, where no clear component could be observed at Fz. As argued by Luck (2005), only electrodes where the component of interest is present should be included in the analyses to avoid excessive noise. As the ICs are characterized by fixed topographies and time courses, the amplitude of the back-projected ICs were only measured at the electrode where the signal was maximal (Cz for IC CNV late,icp3go anterior,icp3no-go early, and IC P3no-go late, and Pz for IC CNV early and IC P3go posterior ). For the P3 components, amplitude was measured at the maximum local peak in the defined time interval (Luck, 2005). This time interval was selected for the P3go and P3no-go components separately based on the onset and offset of the grand mean potentials from the entire group of 35 participants across the speed and accuracy conditions in the midline electrodes. Being slow fluctuations without clear peaks, the CNV components were measured as the mean amplitude in a time interval. The amplitude of the raw CNV and IC CNV late were measured as the mean in the 100 ms before the presentation of S2, while the amplitude of IC CNV early was measured as the mean in the time interval of 600 to 800 ms after S1. This time window was chosen based on the group s mean IC CNV early, which had largest negativity in this time interval. Baseline was adjusted to the mean voltage 100 ms before stimulus presentation relative to S1 for the CNV components and relative to S2 for the P3go and P3no-go components. To ensure that effects on the amplitudes of post-s2 components were not simply the result of differences in the pre-s2 baseline period (CNV), the amplitudes of the P3go and P3no-go components were calculated relative to both pre-s2 and pre-s1 baselines. Analyses of the P3 components relative to a pre-s1 baseline are included in the Appendix. The latencies of the P3go and P3no-go components were measured using a relative criterion version of the fractional area (FA) approach. The method was chosen because it has been proven to be more accurate than peak latency measures (Hansen & Hillyard, 1980; Kiesel, Miller, Jolicoeur, & Brisson, 2008; Luck, 2005), and has been shown to result in very high test-retest reliability when measuring the latency of the P3no-go (Brunner et al., 2013). The onset of the time window for measuring FA latency for each individual is found by identifying the largest amplitude within a defined time window, going back to the lowest point within this time window, and then finding the point in time where a prespecified percentage of this difference is reached (50% for the P3no-go and 70% for the P3go). The offset is defined as the point in time within the time window where the amplitude returns to, or comes closest to the onset point again. The latency is then defined as the median of the FA. The total time window for calculation of the FA was limited according to the observed onset and offset of the grand mean ERP component in the midline electrodes. The time interval was thus limited to ms for the P3go components. For the P3no-go components, the FA time window was limited to ms. Statistics All data were analyzed using IBM SPSS Statistics 21.0 for Mac. Tests of significant differences among parameters in the speed and accuracy conditions and among the three midline electrodes were calculated using repeated measures analyses of variances (ANOVAs) with electrode (Fz, Cz, Pz) and condition (speed vs.

7 Modulation of ERP components by task instructions 177 Figure 3. Grand mean raw ERPs and independent components (ICs) for the group of healthy subjects (n 5 228) in the visual cued go/no-go task. a: Schematic illustration of the task with time (ms) and amplitude (lv) scale for the ERPs. b: ERPs at Fz, Cz, and Pz electrodes for go (A-A, green), no-go (A-P, red), and irrelevant (P-P, black) conditions. Arrows indicate points in time and conditions under which the topographic maps are made. c: Maps of the raw CNV, go, and no-go components taken at their peak amplitudes as indicated by arrows in (b). d: Topographies (left) and time courses (right) of the two CNV ICs back-projected to Fz, Cz, and Pz. e: Topographies (left) and time courses (right) of the two P3go ICs backprojected to Fz, Cz, and Pz. f: Topographies (left) and time courses (right) of the two P3no-go ICs back-projected to Fz, Cz, and Pz. g: Illustration of the timing and overlap of the raw CNV and ICs, P3go and ICs, and P3no-go and ICs at the Cz and Pz electrodes. accuracy) as within-subject factors. To correct for the nonsphericity of the electrode factor, degrees of freedom were corrected using the Greenhouse-Geisser epsilon adjustment (Jennings & Wood, 1976). Only corrected probability values are reported. Significant main effects of electrode position and significant Condition 3 Electrode interactions were then explored using Scheffe s procedure. Planned comparisons between ERP parameters and between ERP and behavioral parameters were conducted using Pearson s or Spearman s correlations as appropriate for parametric and nonparametric variables. Alpha was set at.05. Results Independent Components of the CNV The grand mean ERPs from the group of 228 subjects and a topographic map of the raw CNV are shown in Figure 3b. As can be observed in Figure 3c, the late CNV had a centroparietal distribution. Applying ICA to the ERPs in the interstimulus interval resulted in three ICs with negative slow fluctuations specific to the cue condition, thus qualifying as CNV components. Of these components, one was much weaker in relative power than the two others, and was therefore discarded from further analysis. The two CNV ICs and topographic maps are shown in Figure 3d. One component, referred to as IC CNV early, had a parietal distribution, with a positive peak at 400 ms followed by a negative peak at 700 ms. The second component, IC CNV late, had a centroparietal distribution, with the largest negativity preceding the second stimulus. Independent Components of the P3Go The grand mean raw P3go component from the group of 228 subjects had a centroparietal distribution at the mean peak latency of 325 ms at Pz, which can be seen in Figure 3b,c. Applying ICA to the collection of individual ERPs from the go condition resulted in two ICs with positivity coinciding in latency with observed peaks

8 178 I.E. Aasen and J.F. Brunner Table 1. Descriptives, Behavioral Parameters Accuracy Speed n Mean (SD) Mean (SD) F 2 g p RT (42.5) 270 (22.6) 86.77***.72 SD RT (21.0) 50.8 (18.4) 8.28**.20 Commissions (.55) 4.56 (2.18) ***.79 Omissions (.77).44 (.96) Note. Means and SDs of the behavioral parameters of the 35 participants in the speed and accuracy conditions. RT 5 reaction time; g p 2 5 partial eta squared. **p <.01. ***p <.001. in the raw P3go component. Figure 3e shows the two components and their topographies. One, referred to as IC P3go posterior, had a parietal distribution and a mean peak latency of 325 ms, matching the description of P3b. The second component, IC P3go anterior,had a centroparietal distribution and a mean peak latency of 330 ms. Independent Components of the P3No-Go The grand mean raw P3no-go and topographic map from the group of 228 subjects is shown in Figure 3b,c. The P3no-go had a central distribution at the mean peak latency of 340 ms at Cz. Applying ICA to the collection of individual ERPs for the no-go condition resulted in two components contributing to the central positive P3no-go component. These components and their topographies are shown in Figure 3f. One of these, referred to as IC P3no-go early, had a central distribution and a mean peak latency of 320 ms. The second component, referred to as IC P3no-go late, had a frontocentral distribution and a mean peak latency of 380 ms. Effects of the Speed Versus Accuracy Instructions on the Behavioral Data There was a significant effect of the instructions on the behavioral data, with the speed condition resulting in shorter RTs, F(1,34) , p <.001, smaller SDs of the RT, F(1,34) , p <.01, and more commission errors, F(1,34) , p <.001. The descriptive data for the behavioral parameters in the two conditions are presented in Table 1. Table 2. Descriptives, Raw ERP Parameters Amplitude (lv) Latency (ms) Accuracy Speed Accuracy Speed n Mean (SD) Mean (SD) Mean (SD) Mean (SD) CNV 35 Fz.45 (1.13).63 (1.33) Cz (1.25) (1.25) Pz (1.08) (1.10) P3go 35 Cz 7.82 (3.12) 8.63 (2.93) (25.9) (28.8) Pz 9.05 (2.72) 9.66 (2.60) (20.4) (21.5) P3no-go 35 Fz 6.57 (3.39) 7.68 (3.60) (26.3) (23.1) Cz (3.85) (3.48) (23.3) (21.2) Pz 7.79 (2.69) 8.31 (2.93) (53.0) (58.5) Note. Means and SDs of the ERP amplitudes and latencies in the speed and accuracy conditions for the 35 participants. Effects of the Speed Versus Accuracy Instructions on ERP Amplitudes Table 2 shows the descriptive data for the raw ERP components in the speed and accuracy conditions. Table 3 shows the descriptive data for the independent components. CNV amplitudes in the speed and accuracy conditions. The 2 3 3(Condition3 Electrode) repeated measures ANOVA on the raw CNV amplitudes revealed a significant main effect of electrode position, F(2,68) , p <.001, as well as a significant interaction between the instructional manipulation and electrodes, F(2,34) , p <.01. Summary of the ANOVA results for the raw ERP component amplitudes can be seen in Table 4. Post hoc tests using Scheffe s procedure revealed that the amplitude was significantly larger at the Cz (M , SD ) compared to the Fz (M 5.54, SD ) electrode (p <.001), and larger at the Pz (M , SD ) compared to the Fz electrode (p <.001). No significant difference was found between the amplitude at the Cz compared to the Pz electrode. Furthermore, pairwise comparisons of the CNV amplitude using Scheffe s procedure in each electrode in the speed versus accuracy conditions revealed a significant effect of the instruction at the Pz electrode (p <.01), with significantly larger amplitudes in the speed as compared to the accuracy condition. The effect of the instructional manipulation was not significant at Fz or Cz. Figure 4 demonstrates the differences in amplitudes for the raw CNV and independent components in the two conditions. The amplitude of the IC CNV late, F(1,34) , p <.01, was significantly larger in the speed than in the accuracy condition. There was no significant effect of the instructional manipulation on the amplitude of the IC CNV early. Table 5 shows the summary of the ANOVA results for the ICs. P3go amplitudes in the speed and accuracy conditions. The repeated measures ANOVA showed a significant main effect of the instructions, F(1,34) , p <.01, on the raw P3go with significantly larger amplitudes in the speed as compared to the accuracy condition. 1 There was also a main effect of electrode position, F(1,34) , p <.05, with larger amplitudes at the Pz as compared to the Cz electrode. There was no significant interaction between instruction and electrode position. There was also a significant effect of the instructions on the IC P3go anterior, F(1,32) , p <.01, with larger amplitudes in the speed than in the accuracy condition, but not on the IC P3go posterior amplitude. Figure 5b demonstrates the effect of the instructions on the P3go components. As the IC CNV late was hypothesized to reflect preparation for the dominant task set (A-A 5 go), and the IC P3go anterior was expected to reflect evaluation of this preparatory attention, it was investigated whether the amplitudes of these components correlated with each other. The IC CNV late correlated significantly with the IC P3go anterior amplitude in both the accuracy (r 52.42, p <.05) and the speed condition (r 52.41, p <.05), with larger IC CNV late amplitudes being related to larger IC P3go anterior amplitudes. P3no-go amplitudes in the speed and accuracy conditions. Figure 5c shows the effect of the instructions on the raw P3no-go 1. This effect was not significant when the ANOVA was conducted on amplitudes measured relative to a pre-s1 baseline. See Appendix for details.

9 Modulation of ERP components by task instructions 179 Table 3. Descriptives, Independent Components Amplitude (lv) Latency (ms) Accuracy Speed Accuracy Speed n Mean (SD) Mean (SD) Mean (SD) Mean (SD) IC CNV early (Pz) (.97) (1.24) IC CNV late (Cz) (.96) (1.12) IC P3go anterior (Cz) (2.68) 7.33 (3.65) (27.5) (24.6) IC P3go posterior (Pz) (2.73) 6.78 (3.08) (20.9) (23.6) IC P3no-go early (Cz) (3.85) 9.56 (3.96) (25.4) (22.9) IC P3no-go late (Cz) (3.00) 9.35 (3.26) (22.9) (21.2) Note. Means and SDs of the independent component (IC) ERP amplitudes and latencies in the speed and accuracy conditions for the 35 participants. and ICs. The repeated measures ANOVA revealed significant main effects of the instructions, F(1,34) , p <.001, and electrode position, F(2,68) , p <.001, as well as a significant Electrode 3 Condition interaction, F(2,68) , p <.05. Post hoc analyses using Scheffe s procedure revealed that the amplitude was significantly larger at the Cz (M , SD ) than at the Fz (M , SD ) electrode (p <.001), and larger at the Cz than at the Pz (M , SD ) electrode (p <.001), but not significantly different at the Fz compared to the Pz electrode. Pairwise comparisons of the raw P3no-go in each electrode in the speed versus accuracy conditions using Scheffe s procedure revealed a significant effect of the instruction at the Fz (p <.05) and Cz (p <.001) electrodes, with significantly larger amplitudes in the speed as compared to the accuracy condition. The effect of the instruction was not significant at Pz. Significant effects of the instruction on the IC P3no-go early, F(1,34) , p <.05, and IC P3no-go late, F(1,34) , p <.001, amplitudes were also found, with significantly larger amplitudes in the speed than in the accuracy condition. Effects of the Speed Versus Accuracy Instructions on ERP Latencies Table 4. Summary of Repeated Measures ANOVA (Condition 3 Electrode) Results for the Raw ERP Parameters in the Speed and Accuracy Conditions Amplitude df F 2 g p Latency df F g p 2 CNV (2 3 3) Condition 1, Electrode 2, ***.72 C 3 E 2, **.15 P3go (2 3 2) Condition 1, **.25 1, Electrode 1, *.12 1, C 3 E 1, , P3no-go (2 3 3) Condition 1, ***.31 1, Electrode 2, ***.72 2, C 3 E 2, *.11 2, Note. F values and effect sizes (partial eta squared, g p 2 ) for the effects of condition (C, accuracy vs. speed) and electrode position (E) on amplitude and latency variables. The degrees of freedom (df) are adjusted using a Greenhouse-Geisser correction where the assumptions of sphericity are not met. Only uncorrected df are reported. *p <.05. **p <.01. ***p <.001. Table 2 shows the means and SDs of the latencies of the raw ERP components and ICs in the speed and accuracy conditions. Figure 5 shows the grand mean P3go and P3no-go components in the midline electrodes, as well as the P3go and P3no-go ICs. The repeated measures ANOVA revealed no significant results of instruction, electrode position, or interaction effects on the latency of the raw P3go component. The instructional manipulation did have a significant effect on the IC P3go anterior, F(1,32) , p <.001, and IC P3go posterior, F(1,34) , p <.001, latencies, however. For both ICs, the latencies were significantly later in the accuracy as compared to the speed condition, which can be observed in Figure 5b. As shown in the scatter plots in Figure 6, the latency of the IC P3go posterior correlated significantly with RT in the accuracy (r 5.53, p <.001) condition. The correlation was also significant in the speed condition (r 5.44, p <.01). Based on the outlier labeling rule (Hoaglin & Iglewicz, 1987), however, one subject had an RT above the upper bound and was defined as an outlier. When excluding this subject, the correlation was no longer significant (r 5.27, p 5.12). In spite of the IC P3go anterior latency being shorter after speed than accuracy instructions, the latency of this component was not significantly related to RT in either condition. Regarding the raw P3no-go, the repeated measures ANOVA revealed no significant effects of instruction or electrode position, neither were there any significant effects of the instruction on the latencies of the IC P3no-go early or IC P3no-go late. Discussion The goal of this study was to investigate the modulating effects of instructions to optimize speed or accuracy on CNV and P3 ERP parameters elicited in the cued go/no-go task. The behavioral data confirmed that the manipulation was successful, as all participants had shorter RTs and higher rates of commission errors in the speed as compared to the accuracy condition. The optimization of speed compared to accuracy led to significantly larger amplitudes of the raw late CNV, as well as the P3go and P3no-go. No significant changes in the latencies of the P3go or P3no-go were found. The existence of overlapping components when measuring raw ERP components, however, makes it difficult to draw certain conclusions about how the instructions affect the underlying processes. Indeed, the investigation of the ICs derived from these ERP components gave a more differentiated picture. The CNV Independent Components The IC CNV late had larger amplitude in the speed as compared to the accuracy condition, whereas the IC CNV early amplitude remained unchanged across conditions. This is in line with our predictions and the results of previous findings (Loveless & Sanford, 1974), indicating that the proactive processes necessary for