BRAIN RESEARCH 1104 (2006) available at

Transcription

1 available at Research Report Comparative analysis of event-related potentials during Go/NoGo and CPT: Decomposition of electrophysiological markers of response inhibition and sustained attention Elif Kirmizi-Alsan a, Zubeyir Bayraktaroglu a, Hakan Gurvit b, Yasemin H. Keskin a, Murat Emre b, Tamer Demiralp a, a Department of Physiology, Istanbul Faculty of Medicine, Istanbul University, Istanbul Tip Fakültesi, Fizyoloji Anabilim Dali, Çapa-Istanbul, Turkey b Department of Neurology, Istanbul Medical Faculty, Istanbul University, Turkey ARTICLE INFO ABSTRACT Article history: Accepted 2 March 2006 Available online 7 July 2006 Keywords: Event-related potential Go/NoGo Continuous performance task Time-frequency analysis Sustained attention Response inhibition Neuropsychological tests target specific cognitive functions; however, numerous cognitive subcomponents are involved in each test. The aim of this study was to decompose the components of two frontal executive function tests, Go/NoGo (GNG) and cued continuous performance task (CPT), by analyzing event-related potentials (ERPs) of 24 subjects both in time and time-frequency domains. In the time domain, P1, N1, P2, N2 and P3 peak amplitudes and latencies and mean amplitudes of 100 ms time windows of the post-p3 time period were measured. For GNG, the N1 amplitude and for both GNG and CPT N2 amplitudes were significantly higher in the NoGo condition compared with the Go condition. P3 had a central maximum in the NoGo conditions of both paradigms in contrast to a parietal maximum in the Go conditions. All peaks except P1 and mean amplitudes of the post-p3 period were more positive in CPT compared to those of GNG. N1, N2 and P3 latencies were longer for the NoGo condition than the Go condition in the CPT. In time-frequency analyses, the NoGo condition evoked higher theta coefficients than the Go condition, whereas the CPT and GNG paradigms differed mainly in the delta band. These results suggest that theta component reflects response inhibition in both GNG and CPT, whereas delta component reflects the more demanding sustained attention requirement of the CPT. The latency prolongation observed with the NoGo condition of the CPT paradigm was thought to be due to perseverance/inhibition conflict enhanced by the primer stimuli in CPT Elsevier B.V. All rights reserved. 1. Introduction Go/NoGo (GNG) and continuous performance task (CPT) are two neuropsychological tests that are designed to measure complex attentional functions such as response inhibition and sustained attention, which are thought to be mediated by the prefrontal cortex (Weintraub, 2000). Sustained attention is defined as the ability to maintain an efficient level of responding on a demanding task over a period of time (Ward, 2004). CPT is used to measure sustained attention but is also sensitive to response inhibition and has been used for the assessment of numerous clinical entities such as attention Corresponding author. address: demiralp@istanbul.edu.tr (T. Demiralp) /$ see front matter 2006 Elsevier B.V. All rights reserved. doi: /j.brainres

2 115 deficit disorder, schizophrenia and depression (Ballard, 1996; Gonzalez-Garrido et al., 2001; van Leeuwen et al., 1998; Zillessen et al., 2001), whereas GNG paradigm is a purer test of response inhibition (Weintraub, 2000) and has been especially used for disorders such as neurological conditions involving orbito-frontal cortex and depression (Kaiser et al., 2003; Leimkuhler and Mesulam, 1985). The GNG paradigm consists of a series of two different stimuli, where generally the probabilities of the Go and NoGo stimuli are equal. The simple CPT paradigm has a similar design, where however a relatively small number of Go stimuli are presented within a large set of varying distractor stimuli that correspond to the NoGo condition. The complex cued version of the CPT further includes a primer stimulus, which is followed either by the Go stimulus that has to be responded to as fast as possible or any distractor corresponding to the NoGo condition. Any other stimulus that does not follow the primer is a meaningless distractor. The presence of a large set of interspersed distractors in the cued CPT paradigm makes it more complicated for the subject to focus his/her attention on the Go and NoGo stimuli, hence builds a higher demand of sustained attention compared with the GNG paradigm (Weintraub, 2000). A higher level of vigilance is also needed because the accurate performance depends on both the accurate representation of the primer and the maintenance of this information during the delay period (Dias et al., 2003). Additionally, although the Go and NoGo stimuli have equal probabilities in both the GNG and CPT paradigms, we assume that the presentation of the primer stimulus generates a bias for a coming Go stimulus and a preparation for a fast motor response. Such bias produced by the primer of the cued CPT paradigm should create a difficulty for the subject to resist to respond and to enhance the trend for perseverance to the motor task. Electrophysiological correlates of the abovementioned cognitive domains have been analyzed using event-related potential (ERP) studies based on various experimental paradigms. One of the most studied ERP measures, the P3, is typically obtained with an oddball paradigm as a positive going potential that peaks approximately 300 ms after stimulus onset with a parietal maximum, and is commonly thought to reflect the amount of attentional resources directed to the task-relevant stimulus (Donchin and Coles, 1988; Polich, 1999; Polich and Kok, 1995). Pfefferbaum et al. (1985) found a P3 with a maximum at Pz on Go condition and a P3 with a centro-parietal maximum for NoGo condition with a semantic GNG paradigm. While the Go P3 has the same characteristics as the oddball-p3, the NoGo P3 has a different topography, which shows that it corresponds to a distinct neuronal process. Additionally, visual GNG tasks are known to produce an enhanced fronto-central N2 wave (Falkenstein et al., 1999). N2 is a negative ERP wave with a fronto-central distribution that peaks around 250 ms after stimulus presentation and is enhanced when there is a tendency to make a prepotent but incorrect response, which is the case for both GNG or CPT tasks (Kopp et al., 1996; Nieuwenhuis et al., 2004). In most ERP studies, the CPT task instead of a simple GNG task has been employed for the testing of Go and NoGo responses, although CPT by design is significantly more demanding on cognitive domains such as resistance to perseverance and sustained attention. The aim of this study is to determine the specific electrophysiological subcomponents corresponding to response inhibition, resistance to perseverance and sustained attention through a comparative analysis of ERPs elicited by GNG and CPT paradigms on the same subject group, under same conditions and with similar stimulus characteristics. ERPs were compared in two separate stages, where in the first stage the Go vs. NoGo responses of both paradigms were compared to obtain the signal features reflecting the response inhibition, and in a second stage, the compatible conditions of both paradigms were compared to obtain the components reflecting the sustained attention and the resistance to perseverance. The time-frequency analysis based on Wavelet Transform (WT) has been shown to be a powerful tool for the decomposition of ERPs into functional components (Ademoglu, 1995; Ademoglu et al., 1998; Basar et al., 1999, 2001; Demiralp and Ademoglu, 2001; Demiralp et al., 1999a,b, 2001a,b; Devrim et al., 1999; Samar et al., 1995; Yordanova et al., 2000). ERPs consist not only of sequential but also parallel activations of different neuronal groups. Hence, temporally overlapping activations are necessarily involved in the ERP generation (Basar, 1980). Therefore, the analysis of the ERPs in the time domain using peak amplitudes and latencies that reflect superimposed activities of different subsystems is inefficient for detecting the parallel activations. However, different frequency characteristics of the mechanisms operating in parallel may help in the decomposition of the ERPs into functional components. Therefore, the analysis of the timevarying frequency content of the ERPs by using Wavelet Transform has revealed significant additional information that is not available in the time domain (Ademoglu, 1995; Ademoglu et al., 1997, 1998; Basar et al., 1999, 2001; Demiralp Fig. 1 Grand averages of Go and NoGo conditions of GNG and CPT paradigms on Cz. The basic wave pattern of P1, N1, P2, N2 and P3 peaks was visible in both GNG and CPT. There was a general positive shift in CPT amplitudes, which was seen as less negative N1 and N2 amplitudes, more positive P2 and P3 amplitudes and a continuous positive shift at the post-p3 period compared to the GNG recordings.

3 116 BRAIN RESEARCH 1104 (2006) and Ademoglu, 2001; Demiralp et al., 1999a,b, 2001a,b; Devrim et al., 1999; Samar et al., 1995; Yordanova et al., 2000). For the sake of completeness and comparability with earlier studies carried out in the time domain, we will report the GNG and CPT-ERP results both in the time and time-frequency domains, and try to relate the findings in both representations with each other, where the time domain peaks and troughs are assumed to reflect superimposition of parallel processes that are more purely reflected in time-frequency coefficients. 2. Results Mean reaction times for GNG and CPT were ± ms and ± ms, respectively, and the difference between reaction times was not significant. The mean number of omission errors was 0.42 ± 1.64 and the mean number of commission errors was 0.83 ± 1.05 in GNG, whereas these were 0.83 ± 1.31 and 0.79 ± 1.47 in the CPT, respectively. Because Go and NoGo stimuli numbers were different for GNG and CPT paradigms (50 and 40, respectively), omission and commission error ratios were obtained by calculating number of omissions/total number of Go's and number of commissions/total number of NoGo's for each subject to enable a statistical comparison. The mean value of the omission error ratios was ± for GNG and ± for CPT and the mean value of the commission error ratios was ± for GNG and ± for CPT. There was no significant difference in either the commission or the omission error ratios between the two paradigms Analyses in the time domain The basic wave pattern of P1, N1, P2, N2 and P3 peaks was visible for GNG and CPT paradigms. The ERPs of Cz in Fig. 1 show the P2, N2 and P3 peaks, whereas the N1 and P1 peaks that are better observed in posterior locations can be seen in superposed plots of all channels in Fig. 6. In order to analyze the effect of the response inhibition in time domain, the amplitudes and latencies of these wave components in the Go condition of each paradigm were compared with those in the NoGo condition, and then in a second step the compatible conditions of GNG and CPT paradigms have been compared to investigate the effects of sustained attention and resistance to perseverance. Additionally, as seen in Fig. 1, the CPT showed a general positive shift, which resulted in less negative N1 and N2 amplitudes, more positive P2 and P3 amplitudes and a continuous positive shift at the post-p3 period compared to the GNG recording. Therefore, the GNG vs. CPT comparison was extended to include the post-p3 period. For this purpose, mean amplitudes of 100 ms time windows between 450 ms and 950 ms of the compatible conditions were compared between paradigms. The statistical results on time domain variables are shown in Tables 1 3. Table 1 The results of repeated measures ANOVA performed on the amplitudes and latencies of the N1 and P2 potentials A Go vs. NoGo ERPs Factors df N1 amplitude N1 latency P2 amplitude P2 latency GNG Condition 1, AP 2,46 LAT 2, Condition AP 2,46 Condition LAT 2,46 CPT Condition 1, AP 2,46 LAT 2, Condition AP 2, Condition LAT 2,46 B GNG vs. CPT ERPs Factors df N1 amplitude N1 latency P2 amplitude P2 latency Go Paradigm 1, AP 2,46 LAT 2, Paradigm AP 2, Paradigm LAT 2,46 NoGo Paradigm 1, AP 2,46 LAT 2, Paradigm AP 2,46 Paradigm LAT 2,46 (A) The results of the Go vs. NoGo comparisons for the GNG and CPT paradigms. (B) The results of GNG vs. CPT comparisons for the Go and NoGo conditions. P < P < P <

4 117 Table 2 The results of repeated measures ANOVA performed on the amplitudes and latencies of the N2 and P3 potentials A Go vs. NoGo ERPs Factors df N2 amplitude N2 latency P3 amplitude P3 latency GNG Condition 1, AP 3, LAT 2, Condition AP 3, Condition LAT 2, CPT Condition 1, AP 3, LAT 2, Condition AP 3, Condition LAT 2, B GNG vs. CPT ERPs Factors df N2 amplitude N2 latency P3 amplitude P3 latency Go Paradigm 1, AP 3, LAT 2, Paradigm AP 3, Paradigm LAT 2, NoGo Paradigm 1, AP 3, LAT 2, Paradigm AP 3, Paradigm LAT 2, (A) The results of the Go vs. NoGo comparisons for the GNG and CPT paradigms. (B) The results of GNG vs. CPT comparisons for the Go and NoGo conditions. P < P < P < Go vs. NoGo conditions There was no significant difference in either the amplitudes or the latencies of the P1 wave between the Go and NoGo conditions. Go trials in both paradigms differed from the NoGo trials at N1, N2 and P3 peaks (Fig. 1). The results of the Condition AP LAT-ANOVA design on the amplitudes and latencies of these wave components are shown in Tables 1A and 2A. In the GNG paradigm, the amplitudes of the N1 and N2 wave were significantly higher for the NoGo trials compared to the Go trials (Condition: F(1,23) = 4.94, P < 0.05 [ 1.88 μv vs μv] and F(1,23) = 7.07, P < 0.05 [ 0.25 μv vs μv], respectively) (Fig. 2A), whereas no significant overall amplitude difference was observed between the Go and NoGo P3 amplitudes. However, the NoGo P3 showed a central maximum in comparison to the parietal maximum Table 3 The results of repeated measures ANOVA performed on the mean amplitudes of post-p3 period between the GNG and CPT paradigms for the Go and NoGo conditions GNG vs. CPT Mean amplitudes of post-p3 time windows Factors df (ms) (ms) (ms) (ms) (ms) Go Paradigm 1, AP 3, LAT 2, Paradigm AP 3, Paradigm LAT 2,46 NoGo Paradigm 1, AP 3, LAT 2, Paradigm AP 3,69 Paradigm LAT 2,46 Bonferonni correction is applied on the results. P < P <

5 118 BRAIN RESEARCH 1104 (2006) of the Go P3 (Condition AP: F(3,69) = 28.09, P < 0.001) (Fig. 2A). For the CPT paradigm, the statistical comparisons between the Go and the NoGo conditions were similar to those of the GNG paradigm in terms of the N2 and P3 amplitudes. The amplitudes of the N2 wave were significantly higher for the NoGo trials compared to the Go trials (Condition: F(1,23) = 7.15, P < 0.05 [3.05 μv vs μv]) (Fig. 2A). Although no significant overall amplitude difference was observed between the Go and NoGo P3 amplitudes, there was a significant anteriorization of the NoGo P3 to a central maximum as in the GNG paradigm (Condition AP: F(3,69) = 45.36, P < 0.001) (Fig. 2A). None of the ERP waves showed any latency differences between the Go and NoGo conditions of the GNG paradigm, whereas N1, N2, and P3 latencies of the NoGo condition in the CPT paradigm were longer than those of the Go condition (Condition: F(1,23) = 7.21, P < 0.05 [102.0 ms vs ms]; F (1,23) = 8.18, P < 0.01 [240.2 ms vs ms]; F(1,23) = 43.38, P < [375.6 ms vs ms], respectively) (Fig. 1) GNG vs. CPT paradigms There was no significant difference in either the amplitudes or the latencies of the P1 wave between the two paradigms. The GNG paradigm differed from the CPT paradigm at all other peaks and at the post-p3 period. The results of the Paradigm AP LAT-ANOVA design on the amplitudes and latencies of these wave components are shown in Tables 1B and 2B. First of all, independent from the paradigm, typical N2 and P3 topographies were observed for both the Go and NoGo conditions. N2 wave had the typical fronto-central maximum for both Go and NoGo conditions (AP: F(3,69) = 7.81, P < 0.01 and F(3,69) = 10.4, P < 0.001, respectively) of both paradigms, whereas the P3 amplitudes showed a parietal maximum in the Go condition and a central maximum in the NoGo condition as expected (AP: F(3,69) = 11.52, P < and AP: F(3,69) = 17.23; P < 0.001, respectively) (Fig. 2A). For the CPT paradigm, N1 amplitude in the NoGo condition (Paradigm: F(1,23) = 9.93, P < 0.01 [ 0.22 μv vs μv]) and the N2 amplitudes in both Go and NoGo conditions (Paradigm: F (1,23) = 30.43, P < [1.47 μv vs μv] and F(1,23) = 31.44, P < [3.05 μv vs μv]) were smaller, whereas P2 and P3 amplitudes in both Go (Paradigm: F(1,23) = 5.69, P < 0.05 [7.35 μv vs μv] and F(1,23) = 61.82, P < [14.39 μv vs μv]) and NoGo conditions (Paradigm: F(1,23) = 17.49, P < [7.04 μv vs μv] and F(1,23) = 70.62, P < [15.46 μv vs μv]) were larger. The decrease of the N2 amplitude in the CPT was more pronounced in the fronto-central regions (Paradigm AP: F(3,69) = 18.24, P < 0.001) for the NoGo trials (Fig. 2A). The increase in the P3 amplitudes in the CPT were more pronounced in the parietal region (Paradigm AP: F (3,69) = 8.39, P < 0.01) with the Go condition, whereas it was significantly greater over the central areas with the NoGo condition (Paradigm AP: F(3,69) = 20.22, P < 0.001) (Fig. 2A). The latencies of the N1, P2, N2 and P3 waves of the CPT-ERPs were significantly longer than those of the GNG for the NoGo condition (Paradigm: F(1,23) = 7.51, P < 0.05 [102.0 ms vs ms]; F(1,23) = 4.87, P < 0.05 [157.0 ms vs ms]; F (1,23) = 6.61, P < 0.05 [240.2 ms vs ms]; and F(1,23) = 19.79, P < [375.6 ms vs ms], respectively). In the Go condition, there was no significant latency difference between Fig. 2 Scalp topographies of the (A) N1, P2, N2, P3 waves and (B) means of the 100 ms time windows in the post-p3 period between 450 and 950 ms. N2 wave had the typical fronto-central maximum for both Go and NoGo conditions of both paradigms with a significant increase in amplitude in the NoGo condition. P3 amplitudes showed a parietal maximum in the Go condition and a central maximum in the NoGo condition of both paradigms as expected. The N1 and N2 waves showed significant decreases and P2 and P3 significant increases in the CPT paradigm compared with the GNG paradigm. Mean amplitudes of the first four time windows between 450 and 850 ms of the Go condition were overall more positive for the CPT paradigm compared to GNG, where the difference was most prominent in the fronto-central leads in the ms period. For the NoGo condition, mean amplitudes of the first two ( ms) and the last ( ms) time windows were higher for the CPT compared to the GNG paradigm.

6 119 the CPT and GNG, except the fronto-centrally shorter latency of the P3 wave in the CPT paradigm (Paradigm AP: F(3,69) = 4.93, P < 0.05). For the post-p3 period, mean amplitudes of 100 ms time windows between 450 and 950 ms were tested for significant differences between the compatible conditions of both paradigms with a Paradigm AP LAT-ANOVA design. The first four time windows (i.e., , , and ms) of the Go condition were overall more positive for the CPT paradigm compared to GNG (Paradigm: F(1,23) = 22.74, P < [4.9 μv vs. 2.4 μv]; F(1,23) = 24.56, P < [4.46 μvvs.1.46μv]; F(1,23) = 35.04, P < [4.43 μvvs.1.34μv]; F(1,23) = 14.49, P < [3.64 μvvs μv], respectively). While no topography effect was observed for the first two time windows, the difference was most prominent in the fronto-central leads in the ms period (Paradigm AP: F(3,69) = 5.67, P <0.01;F(3,69) = 5.62, P < 0.01, respectively) (Fig. 2B). For the NoGo condition, mean amplitudes of the first two ( and ms) and the last ( ms)timewindowswerehigherfortheCPTcomparedtothe GNG paradigm (Paradigm: F(1,23) = 48.16, P < [5.68 μv vs μv]; F(1,23) = 28.72, P < [4.21 μvvs.1.64μv]; F(1,23) = 9.95, P <0.01[4.23μVvs.2.65μV], respectively) (Figs. 1 and 2B). In summary, for both the Go and the NoGo conditions of the CPT paradigm, P2 and P3 amplitudes were more positive and N1 and N2 amplitudes were less negative, and mean amplitudes of all the time windows throughout the post-p3 period were more positive than those of the GNG paradigm. These phenomena seemed to be the result of a superimposed positive slow wave in the CPT starting around N1 and remaining throughout the rest of the epoch Analyses in the time-frequency plane In the present study, the significant differences within the delta and theta bands were dissociating the main subprocesses investigated. The rectified WT coefficients (Fig. 3) were used for statistical evaluation of the frequency bands, and the reconstructed time courses of the theta and delta bands were depicted in Fig. 4 for a simple visualization of the results. The Bonferonni-corrected statistical results for the theta and delta coefficients are shown in Tables 4 and 5. Similar to the time domain results, first the differences between the Go vs. NoGo conditions of each paradigm were tested with a Condition AP LAT-ANOVA, which was followed by the test of differences between the compatible conditions of both paradigms with a Paradigm AP LAT-ANOVA Go vs. NoGo Theta 1 coefficient, which reflects the theta response as early as the first 167 ms, was significantly higher in the NoGo trials than in the Go trials in both GNG and CPT paradigms (Condition: F(1,23) = 16.44, P < [16 vs. 12.5] and F Fig. 3 Grand averaged time-frequency surfaces and ERPs recorded from Cz for Go and NoGo conditions of GNG and CPT paradigms and the difference-surfaces and difference-waveforms for the paradigm and the condition comparisons. The time-frequency surfaces are generated by rectifying the WT coefficients to represent the signal energy within that time-frequency range. The lower row displays the CPT-GNG differences for NoGo and Go conditions, whereas the NoGo Go differences for the CPT and GNG paradigms are shown on the right column. It can be observed that the CPT-GNG difference is mainly a long lasting delta difference, whereas the NoGo Go difference contains significant theta coefficients.

7 120 BRAIN RESEARCH 1104 (2006) be observed in the GNG paradigm within the delta range. However, the antero-posterior distribution of the D3 coefficient differed between the Go and NoGo conditions of both paradigms (Condition AP: F(3,69) = 14.32, P < and F (3,69) = 26.7, P < 0.001, respectively) (Fig. 5B). The topography of the D3 coefficient was parallel to that of P3, that is, the positivity was more anterior in the NoGo condition than in the Go condition for both the GNG and the CPT. Fig. 4 The interpolated waveforms for the theta and delta bands of Go and NoGo conditions in GNG and CPT paradigms on Cz. The NoGo condition of both the GNG and CPT paradigms induced larger theta oscillations compared with the Go condition, whereas the main difference in the delta range occurred between the compatible conditions of the GNG and CPT paradigms. (1,23) = 8.91, P < 0.01 [17.08 vs ], respectively) (Fig. 3). This theta component had a fronto-central distribution in both GNG and CPT paradigms (Condition AP: F(3,69) = 6.55, P < 0.01 and F (3,69) = 8.17, P < 0.01, respectively) (Fig. 5A). The temporal position of Theta 1 (0 167 ms) roughly corresponds to P1 N1 complex suggesting that it contributes to P1 and/or N1 wave. The time domain analyses did not provide any differentiation between the Go and the NoGo conditions in P1, whereas there was a significant difference in N1 only in the GNG paradigm. However, we obtained a consistent Go vs. NoGo effect for both GNG and CPT paradigms with the T1 coefficient. Theta 3 coefficient also differentiated the Go from the NoGo trials of the CPT, where it was higher in the NoGo trials (Condition: F(1,23) = 10.80, P < 0.01 [16.49 vs. 9.39]) (Fig. 3). This effect also had a highly significant fronto-central distribution as for Theta 1 (Condition AP: F(3,69) = 11.06, P < 0.001) (Fig. 5A). Although the T3 of the GNG paradigm did not show a significant overall difference between the Go and NoGo conditions, in the NoGo condition, its power was significantly higher over the fronto-central areas than in the Go condition (Condition AP: F(3,69) = 7.90, P < 0.01) (Fig. 5A). The D1 did not show any difference between the Go and NoGo conditions in both GNG and CPT paradigms. In the CPT paradigm, the D2, D3 and D5 coefficients showed a significant difference between the Go and the NoGo conditions (Condition: F(1,23) = 56.81, P < [45.67 vs ]; F(1,23) = 25.48, P < [42.99 vs ]; F(1,23) = 8.86, P < 0.01 [33.53 vs ], respectively), whereas no significant overall difference could GNG vs. CPT The T3 and T5 coefficients of the NoGo condition were higher for the CPT compared with those of the GNG (Paradigm: F (1,23) = 12.80, P < 0.01 [16.49 vs ] and F(1,23) = 11.56, P < 0.01 [5.99 vs. 3.63], respectively). The paradigm change induced highly significant differences in the delta frequency range. For the Go condition, the CPT paradigm produced significantly higher powers than the GNG paradigm in the D2 to D6 coefficients corresponding roughly to the ms time window (Paradigm: F(1,23) = 57.77, P < [78.83 vs ]; F(1,23) = 12.24, P < 0.01 [42.99 vs ]; F (1,23) = 18.89, P < [34.59 vs. 17.7]; F(1,23) = 79.76, P <0.001 [33.53 vs ]; F(1,23) = 14.06, P < [23.76 vs ], respectively), while in the NoGo condition, the D3 and D6 coefficients of the CPT paradigm were significantly higher than those of the GNG paradigm (Paradigm: F(1,23) = 64.08, P <0.001 [73.98 vs ] and F(1,23) = 34.00, P < [28.23 vs ], respectively). The D2 increase in the CPT paradigm was maximal over centro-parietal leads in both Go and NoGo conditions (Paradigm AP: F(3,69) = 19.42, P <0.001andF(3,69) = 16.05, P <0.001)(Fig. 5B). Although a significant paradigm AP effect is observed for D3 in the NoGo condition in terms of a higher increase in the CPT in parieto-occipital leads (Paradigm AP: F (3,69) = 11.07, P < 0.001), this significance seems to result from the strong main paradigm effect. The D4 and D5 increase in the Go condition (Paradigm AP: F(3,69) = 10.36, P < and F (3,69) = 13.17, P < 0.001, respectively) as the D5 and D6 increase in the NoGo condition of the CPT was maximal over the frontocentral leads (Paradigm AP: F(3,69) = 8.46, P < 0.01 and F (3,69) = 8.69, P <0.01,respectively)(Fig. 5B). In short, the delta power increase in the post-p3 period of the CPT was localized to the fronto-central leads. 3. Discussion Although in the ERP literature CPT and GNG are considered as similar paradigms, and CPT is often used as a GNG paradigm, some important differences are established by the neuropsychological literature between these two paradigms. CPT is implemented as a neuropsychological test for the assessment of sustained attention, perseverance and response inhibition, whereas the GNG is considered more or less a purer response inhibition task (Weintraub, 2000). The aim of the present study was to explore the electrophysiological correlates of these differences between the CPT and the GNG tasks Behavioral results An important point of the study design was to avoid any learning effect and interference between the GNG and CPT

8 121 Table 4 The results of repeated measures ANOVA performed on six theta coefficients A Go vs. NoGo Theta coefficients Factors df T1 T2 T3 T4 T5 T6 GNG Condition 1, AP 3, LAT 2, Condition AP 3, Condition LAT 2,46 CPT Condition 1, AP 3, LAT 2,46 Condition AP 3, Condition LAT 2,46 B GNG vs. CPT Theta coefficients Factors df T1 T2 T3 T4 T5 T6 Go Paradigm 1,23 AP 3, LAT 2, Paradigm AP 3, Paradigm LAT 2, NoGo Paradigm 1, AP 3, LAT 2, Paradigm AP 3,69 Paradigm LAT 2,46 (A) The results of the Go vs. NoGo comparisons for the GNG and CPT paradigms. (B) The results of GNG vs. CPT comparisons for the Go and NoGo conditions. Bonferonni correction is applied on the results. P < P < tests carried out by the same subjects. The use of the Go stimulus of the CPT paradigm as the NoGo stimulus in the GNG paradigm was successful to avoid any learning effect that might be observed if the Go stimuli of both paradigms would be the same. The fact that there were no statistically significant differences in the reaction times or error rates between the GNG and CPT shows that the performance in GNG was not influenced by CPT. On contrary direction, this design could have the possibility to introduce a higher demand on inhibitory processes in GNG paradigm, since the previously responded stimuli (letter Z which was a target in CPT) would have to be inhibited. However, had this been the case, then one would expect a higher commission error rate in the NoGo condition of the GNG paradigm, which was not the case. Therefore, it is verified that the study design avoided any interference that could be introduced by the consecutive application of both paradigms in the same subjects Time domain results The anteriorization of the P3 and more pronounced frontal N2 in the NoGo condition compared to the Go condition observed in both paradigms were in line with the literature (Falkenstein et al., 1999; Fallgatter and Strik, 1999; Fallgatter et al., 1999; Strik et al., 1998) and were regarded as the correlates of response inhibition, which was the common feature of both paradigms (Roberts et al., 1994). In the GNG paradigm in addition to the N2 increase and the P3 anteriorization, a significant N1 increase was observed in the NoGo-ERPs compared with the Go-ERPs, which however was not replicated in the CPT probably due to the masking effect of the positive shift in both Go and NoGo responses of the CPT. Several studies have shown the importance of the ms interval after stimulus onset for the NoGo decision (Gemba and Sasaki, 1990; Hoshiyama et al., 1996, 1997; Sasaki et al., 1989; Schluter et al., 1998). De Jong et al. (1990) observed a modulation of the N1 wave in the ERPs elicited by the auditory stop signals, when successfully and unsuccessfully inhibited responses were compared. Filipovic et al. (2000) also reported an early negativity corresponding to the N1 wave that might be a more specific electrophysiological reflection of the NoGo decision proper compared with the effects observed in later peaks such as N2 and P3. The N1 effect observed in the GNG paradigm is compatible with these earlier results. The original results of this study are the highly significant differences between the CPT- and GNG-ERPs. Firstly, the latencies of N1, N2 and P3 waves of the CPT paradigm showed a significant prolongation in the NoGo condition compared with the Go condition, whereas no such difference was observed between the Go and NoGo conditions of the GNG paradigm. Additionally, the NoGo latencies of the CPT were significantly longer than those of the GNG paradigm, whereas the P3 latency in the Go condition was shorter in frontocentral area compared with the Go condition of the GNG. The primer A preceding a distractor letter in the NoGo condition of the CPT could be causing a perseverance to motor response

9 122 BRAIN RESEARCH 1104 (2006) Table 5 The results of repeated measures ANOVA performed on six delta coefficients A Go vs. NoGo Delta coefficients Factors df D1 D2 D3 D4 D5 D6 GNG Condition 1,23 AP 3, LAT 2, Condition AP 3, Condition LAT 2, CPT Condition 1, AP 3, LAT 2, Condition AP 3, Condition LAT 2, B GNG vs. CPT Delta coefficients Factors df D1 D2 D3 D4 D5 D6 Go Paradigm 1, AP 3, * LAT 2, Paradigm AP 3, Paradigm LAT 2,46 NoGo Paradigm 1, AP 3, LAT 2, Paradigm AP 3, Paradigm LAT 2, (A) The results of the Go vs. NoGo comparisons for the GNG and CPT paradigms. (B) The results of GNG vs. CPT comparisons for the Go and NoGo conditions. Bonferonni correction is applied on the results. P < P < and accordingly shorten the latencies in the Go-ERPs but prolong the NoGo-ERP latencies, when the response has to be inhibited. In this case, a larger CNV would be expected following the primer in CPT (Dias et al., 2003), which can be observed in Fig. 6. The prolongation of the P3 latencies observed also in earlier studies with the NoGo condition of the CPT (Fallgatter and Strik, 1999; Fallgatter et al., 1999) were assigned to the more demanding task requirements of the inhibition condition (NoGo) as compared to the execution condition (Go). However, in the present study, the cued CPT but not the simple GNG paradigm produced a prolongation of P3 latencies in the NoGo condition. Therefore, it might be suggested that the latency effect seen with CPT is actually an electrophysiological trace of a conflict between perseverance and inhibition. This state of being locked to the prepotent response induced by the primer could be augmented by the clinical conditions that are prone to impulsivity, inability to resist inappropriate response tendencies, increased perseveration, impaired perseverance, such as orbito-frontal damage, and ADHD (Leimkuhler and Mesulam, 1985; Weintraub, 2000). That this effect cannot be observed in the behavioral responses in terms of a higher commission error rate in CPT but only in ERP latencies, shows the importance of using electrophysiological measurements for the evaluation of such pathologies. Secondly, the amplitudes of the N1 and N2 waves were smaller and those of the P2 and P3 waves were higher in the CPT paradigm. These findings were accompanied by the higher positivity in the post-p3 period of the CPT, which suggested that these observations could be the result of a superimposed positive slow wave. This positive difference between the CPT and GNG paradigms displays a continuous effect that is distributed in time, whereas the electrophysiological differences between the Go and NoGo conditions are limited to temporary changes in N2 and P3 amplitudes. Had this been confirmed, then the slow wave could have been claimed to be the trace of the sustained attention component, which we assume to be more dominant in the CPT paradigm compared with the GNG. From this perspective, it was possible to define the start of the superimposed slow wave observed as a shift from baseline on the grand averaged ERP waves as early as 100 ms and continuing throughout the rest of the epoch. This effect could be explained by a baseline difference between both paradigms due to the higher CNV observed after the primer of the CPT. However, as seen in Fig. 6, even if the baseline is chosen as the prestimulus period of the primer, the responses to the CPT targets do not return to the baseline before 1 s, which is not the case for the GNG-ERPs. Although the early positive shift after the target stimulus could be explained by a CNV resolution, the later post-p3 positivity of the CPT-ERPs cannot to be explained by such mechanism Decomposition of ERPs in the time-frequency domain ERPs are considered to reflect both serial and parallel processes running during the cognitive tasks. There have been studies attempting to discriminate these parallel processes by time-frequency analysis of ERPs (Ademoglu, 1995;

10 123 Ademoglu et al., 1997, 1998; Basar et al., 1999, 2001; Demiralp and Ademoglu, 2001; Demiralp et al., 1999a,b, 2001a,b; Samar et al., 1995; Yordanova et al., 2000). Considering that the observation of a positive shift of the peak amplitudes in the time domain might be the result of the superimposition of a slow positive shift reflecting such a parallel process, it should have been possible to isolate this slow wave by the timefrequency analysis. Based on this assumption, wavelet decomposition was applied, the results of which have shown Fig. 6 Grand averages of the period between the primer and the subsequent target stimulus of CPT (upper plot), and the period between any NoGo stimulus that is followed by a Go stimulus in the GNG paradigm. ERPs obtained from all 30 channels are superimposed. A clear negative shift (CNV) can be seen in the CPT but not in the GNG paradigm. that the theta band ( Hz) power in the early poststimulus period (first 167 ms) differentiates the Go and the NoGo conditions of both GNG and CPT paradigms, hence purely reflects the response inhibition effect. In the time domain, in this early time window, an increased N1 amplitude was observed only in the NoGo-ERPs of the GNG paradigm. However, the early theta effect between the Go and NoGo responses corresponding to the same time window as N1 is much more evident and consistent in both GNG and CPT paradigms. Filipovic et al. (2000) observed a significant negative difference in the N1 latency range by subtracting the Go responses from the NoGo responses in the time domain and suggested that this early negativity is a more specific electrophysiological reflection of the NoGo decision. The early theta difference consistently observed in the present study Fig. 5 Scalp topographies of the absolute values of six theta (A) and six delta band (B) coefficients in Go and NoGo conditions of GNG and CPT paradigms. T1 and T3 show a clear condition effect with higher fronto-central midline amplitudes in the NoGo condition of both paradigms, whereas T3 and T5 show also a paradigm effect with significantly higher amplitudes in CPT. The main effect for the D2 D6 coefficients is the paradigm effect in terms of significantly higher amplitudes in CPT. D3, however, additionally shows a significant condition effect, because the topography of D3 in Go and NoGo conditions changes in-line with the P3 topography in time domain. The Go-D3 has a parietal maximum, whereas the NoGo-D3 has a central maximum.

11 124 BRAIN RESEARCH 1104 (2006) seems to isolate this early element of the NoGo decision from other ERP components efficiently. The midline fronto-central distribution of this theta coefficient (Fig. 5A) also fits well with the functional anatomy of response inhibition. A similar effect with a similar topography was also observed for the theta coefficient reflecting the ms time window (T3) (Fig. 5A). This pattern in the timefrequency transformed data suggests that the early theta effect in the first 167 ms is part of an oscillatory activity that lasts until N2-P3 time window and gets superimposed on a strong delta response within this latency range. This oscillatory pattern can be clearly observed in the reconstructed theta waveforms in Fig. 4, where at least two cycles of theta oscillations after the stimulation time-point show a significant amplitude increase in the NoGo condition compared with the Go condition of both paradigms. An additional finding is that this later theta component differs also significantly between the NoGo-ERPs of both paradigms in terms of a higher theta activity in the CPT. This effect might be the result of the higher demand for inhibition in the CPT paradigm due to the higher motor preparation through the primer stimulus as discussed above. The delta response within the ms poststimulus time window showed a highly significant difference between the paradigms (Fig. 3). The CPT produced significantly higher delta response compared with the GNG (Fig. 4). This paradigm effect showed a fronto-central distribution for the late delta in the post-p3 period (Fig. 5B). Because the sustained attention is the cognitive domain, in which the GNG and CPT paradigms differ the most, and the main signal difference between the ERPs of these two paradigms is the addition of a long positive delta-range component in the CPT potentials, we suggest that this delta component reflects the sustained attention. In the earlier part of the delta response, a condition effect was also observed between the Go- and NoGo-ERPs of both the GNG and CPT paradigms. The D3 coefficient that represents the delta response within the ms showed topographies corresponding to the Go and NoGo P3s in the time domain (Fig. 5B). D3 had a parietal maximum for the Go-ERP, whereas it was centrally distributed for the NoGo condition leading to a significant condition effect for D3 in addition to the paradigm effect mentioned above. This shows that the main constituents of the Go P3 and NoGo P3 are also within the delta range. It has been observed in earlier studies that the strongest constituent of the P3-family potentials is a delta oscillation with the same topography of the P3 in the time domain, which is superimposed by different processes within theta and other higher frequency bands with specific topographies according to the specific tasks applied (Basar-Eroglu et al., 1992; Demiralp and Ademoglu, 2001; Demiralp et al., 1999a, 2001a,b; Devrim et al., 1999; Klimesch et al., 2000). It can be considered that the lower probability of the Go stimuli in the CPT paradigm might also contribute to the difference between the ERPs of the GNG and CPT paradigms in terms of a larger P3 amplitude due to oddball effect in the CPT in addition to the difference in sustained attention. To avoid this confounding factor, the GNG paradigm could be designed to include only 10% Go and 90% NoGo stimuli. However, such a design would introduce three major disadvantages to the problem of isolating electrophysiological signs of inhibition and sustained attention: First, it would add a significant sustained attention component to the GNG paradigm, because the detection of the rare Go stimuli would increase the effort to maintain the focus of the attention during longer periods of time. Second, the NoGo conditions of both paradigms would not be comparable in terms of their relative probabilities (10% in CPT vs. 90% in GNG). And third, the Go vs. NoGo comparison within the GNG paradigm for the isolation of the inhibition component would include the confounding effect of different stimulus probabilities (10% vs. 90%). Therefore, we chose equal relative probabilities for Go and NoGo conditions in both paradigms. Additionally, as mentioned above, the oddball effect has been shown to mainly affect a single delta coefficient within the P3 latency and with a P3 like topography (Demiralp et al., 1999a), which seems to be the D3 coefficient in the present study that showed a significant parietal vs. central distribution in the Go and NoGo conditions, respectively. However, the delta effect that differentiated both the Go and NoGo conditions of the CPT paradigm from those of the GNG paradigm was distributed within a latency range between 100 and 950 ms and showed a frontal topography especially in the post-p3 period. Therefore, it is unlikely that the long-lasting delta increase in the CPT stems from the oddball effect. The importance of the time-frequency analysis in the decomposition of the functional components of ERPs is that, in contrast to frequency domain analysis with no temporal resolution, it can dissociate these two processes the long lasting delta effect that differs between the CPT and GNG paradigms and the delta effect in the P3 latency range that differs between the Go and NoGo conditions of each paradigm. The time-frequency analysis with its temporal resolution can dissociate these two delta processes using their different temporal characteristics (Demiralp and Ademoglu, 2001). This point is also important for the interpretation of oscillatory ERP components, because it shows that oscillations cannot be attributed to absolute functions without considering their temporal and topographic characteristics. Oscillations within the same frequency range but with different temporal dynamics and topographies may correspond to the activities of distinct neuronal processes. 4. Conclusion The results show that by comparing the ERPs evoked by neuropsychological tasks, and decomposing the ERP signals into time-frequency components, the signal parameters that more specifically correlate with parallel cognitive subfunctions can be illuminated. The present study specifically demonstrated that purification of delta component of CPT- ERPs could have a higher sensitivity for sustained attention compartment; whereas, response inhibition was shown to be evaluated better by the theta components of both CPT- and GNG-ERPs, and perseverance/inhibition conflict can be determined by focusing on the difference between the Go and the NoGo latencies of the CPT-ERPs.

12 Experimental procedures 5.1. Subjects Twenty-four healthy right-handed volunteers (13 males and 11 females) were recruited as subjects with a mean age of 25.8 ± 5.6 and a mean education of 17.8 ± 3.3 years Experimental paradigms A visual GNG paradigm, where the stimuli were a subset of the CPT stimuli, was used to achieve maximum similarity between the paradigm characteristics. The stimuli for the GNG paradigm consisted of 100 stimuli, A and Z appearing in a random order with equal probabilities where A was the target stimulus. The CPT paradigm consisted of 400 stimuli, 10 distractors (B, C, D, E, F, G, H, J, K, L), 1 primer A, 1 target Z appearing with the following probabilities: 20% primers, 10% Go stimuli (any Z after an A ), 10% NoGo stimuli (any distractor letter after an A ) and 60% distractors. By this way, the probability ratios of the Go vs. NoGo conditions have been kept equal in both paradigms. Each stimulus was presented on the center of a 15 inch computer screen as white letters of cm over a gray background from a distance of 80 cm with a visual angle of 3 15 both horizontally and vertically for 200 ms, and the interstimulus interval (ISI) was 1500 ms. Subjects sat in a dimly lit electrically shielded room with the computer mouse under their right hand. Each subject performed the CPT first. The subjects were told to respond to A s in the GNG paradigm and any Z after an A in the CPT paradigm with the left mouse button with an equal emphasis on speed and accuracy. Two types of errors were possible in each condition: an error of omission when the subject fails to respond to a target letter, and an error of commission when the subject fails to inhibit to respond to the distractor letter. The CPT task was different from the GNG, as the letter Z became the target only if the primer letter A preceded it. The rationale behind the choice of the stimuli was to choose physically similar stimuli in both paradigms, while avoiding the problem of the learning effect that was expected to enhance performance in GNG as a result of performing CPT prior to GNG. To overcome this problem, the Go stimulus of CPT was used as the NoGo stimulus in GNG ERP recordings A software designed in our laboratory for the 32 channel digital EEG amplifier of the La Mont Medical Inc. was used for the recording and analysis of the ERPs. Stimulus presentation is carried out by a Matlab program developed in our lab using the Psychophysics Toolbox extensions (Brainard, 1997; Pelli, 1997). EEG was amplified with a band pass of Hz from 30 scalp electrodes, Oz, O1, O2, Pz, P3, P4, P7, P8, Cz, C3, C4, T7, T8, Fz, F3, F4, FCz, FC3, FC4, CPz, CP3, CP4, FT7, FT8, F7, F8, TP7, TP8, FP1, FP2, and sampled at 200 Hz. Linked earlobes were used as the reference. The electro-oculogram (EOG) was recorded bipolarly between the electrodes on the external canthus of the right eye and on the nasion to identify the ocular artifacts introduced by both vertical and horizontal eye movements. After building the ERP epochs of 1500 ms duration between 500 and 1000 ms, trials with EEG or EOG amplitudes exceeding ±90 μv were rejected automatically as artifact. ERPs were averaged for the Go and NoGo conditions of both the GNG and CPT paradigms and the mean amplitude of the 200 ms period before the onset of the stimulus was subtracted from the averaged ERPs for baseline correction Time domain analyses The P1, N1, P2, N2 and P3 peaks were identified as the most negative and positive deflections within the following time windows that were determined using the grand-average potentials: P1 70 to 140 ms, N1 70 to 140 ms, P2 140 to 250 ms, N2 140 to 300 ms and P3 260 to 450 ms. The post-p3 period was evaluated through mean amplitudes of 5 time windows ( ms, ms, ms, ms, ms). The P1 was measured only in occipital channels, whereas N1 and P2 were measured in frontal, central and parietal channels. All other variables were measured in frontal, central, parietal and occipital channels Time-frequency analyses For the time-frequency analysis, Quadratic B spline wavelet transform has been applied as shown earlier by Ademoglu (1995), Ademoglu et al. (1997, 1998), Demiralp et al. (1999a,b) and Demiralp and Ademoglu (2001). The key points of the wavelet analysis are summarized below: Wavelet transform The wavelet transform performs a signal analysis in both time and frequency using families of functions h a,b : h a,b(t) ¼jaj 1 2 h ð t b a Þ a;bar;ap0; ð1þ that are generated from a single function h(t) by the operation of dilations (a) and translations (b). The continuous wavelet transform (CWT x ) of a signal x(t)at scale a and position b can be generally defined as: * + CWT x (b;a) ¼ x(t);jaj 1 2 h* ð t b Þ a Z l ¼jaj 1 2 xt ðþh*ð t b Þdt; ð2þ a l where * represents the complex conjugation and x, h represents the inner product. Eq. (2) is interpreted as a multiresolution decomposition of the signal into a set of frequency channels, which in case of a = 2 j (j = integer) would have one and the same bandwidth in a logarithmic scale. This property is known as constant-q or constant relative bandwidth frequency analysis by octave band filters (Samar et al., 1995). This allows observing higher frequencies in shorter and lower frequencies in longer time windows that are optimal for both time and frequency. ERPs have been demonstrated to have spectra displaying constant-q behavior (Basar et al., 1975; Samar et al., 1995; de Weerd and Kap, 1981).

13 126 BRAIN RESEARCH 1104 (2006) Linear time shifting of the wavelet basis function relative to the ERP is performed for each scale (frequency band), and wavelet coefficients are computed and rectified to measure the energy present in the ERP at the corresponding point in time at that scale. This permits the time localization of components in each frequency band and guarantees that the captured components are separated in time and frequency. A quadratic B-spline wavelet basis function in the form of multiresolution scheme was used (Ademoglu, 1995; Ademoglu et al., 1997, 1998). A fast WT algorithm in a standard pyramidal filter scheme is used to decompose ERPs (Ademoglu, 1995; Ademoglu et al., 1998; Demiralp and Ademoglu, 2001). The application of a five-octave wavelet transform to the data sampled at 200 Hz yields five sets of coefficients in the Hz (high gamma), Hz (gamma), Hz (beta), Hz (alpha) and (theta) frequency ranges, and a residue in the Hz (delta) frequency band. Fig. 7 schematically illustrates this approach such that each octave contains the number of coefficients necessary to provide the relevant time resolution for that frequency range with variation in color employed to illustrate the amplitudes of the coefficients at a specific frequency and time. The interpolation of the coefficients by quadratic spline functions is used to reconstruct the analyzed signal and to represent the full time course of each frequency band (Fig. 7). In the present study, 6 delta and 6 theta coefficients were obtained for the 1000 ms poststimulus period. The window size for delta and theta resolution levels was 167 ms. The precise time windows represented by each delta and theta coefficient were as follows: coefficient 1 (0 167 ms), coefficient 2 ( ms), coefficient 3 ( ms), coefficient 4 ( ms), coefficient 5 ( ms), coefficient 6 ( ms). These time-frequency regions are designated by a letter representing the frequency band (i.e., D for delta, T for theta) and the coefficient number representing the time window (e.g., the theta component in the ms latency range will be noted as T3). The statistical evaluation has been carried out on the rectified WT coefficients to compare the differences of the powers in specific time-frequency regions (Fig. 3). For a simpler visualization of the time courses of the investigated frequency ranges, the reconstructed waveforms have been depicted in Fig Statistical analyses The amplitudes and latencies of the P1, N1, P2, N2 and P3 peaks, mean amplitudes of the post-p3 time windows and the wavelet coefficients were compared between conditions (Go vs. NoGo) and paradigms (GNG vs. CPT) using separate repeated measures ANOVA designs. For the first series of analyses, the Go vs. NoGo potentials were compared separately for the GNG and CPT paradigms with the following factors: Experimental Condition (Condition [2 levels]: Go vs. NoGo) Antero-posterior distribution (AP [4 levels]: frontal, central, parietal, occipital) Lateral distribution (LAT [3 levels]: left, midline, right). In the second line of analyses, Go and NoGo potentials of the GNG were compared with corresponding potentials of the CPT paradigm separately: Experimental Paradigm (Paradigm [2 levels]: GNG vs. CPT) Antero- Fig. 7 Schematic diagram of the Discrete Wavelet Transform (DWT). The signal in the time domain (upper trace) is transformed onto the time-frequency plane using a discrete WT (lower left panel). The interpolation of the coefficients by quadratic spline functions can be used to reconstruct the analyzed signal and to represent the full time course of each frequency band (lower right panel). The scalp topographies of some wavelet coefficients are shown to demonstrate that the coefficients within the same or nearby time windows may have distinct scalp distributions suggesting different neuronal processes running in parallel.