Exp Brain Res (2010) 203:453–463 DOI 10.1007/s00221-010-2250-2

RESEARCH ARTICLE

Monitoring antisaccades: inter-individual differences in cognitive control and the influence of COMT and DRD4 genotype variations Emmanouil Kattoulas • Ioannis Evdokimidis • Nicholas C. Stefanis • Dimitrios Avramopoulos Costas N. Stefanis • Nikolaos Smyrnis

•

Received: 13 November 2009 / Accepted: 8 April 2010 / Published online: 24 April 2010 Ó Springer-Verlag 2010

Abstract Conscious monitoring of behavior is an essential control function for adaptation and learning. Antisaccade performance was investigated in a large sample of young healthy men in two tasks, one that required conscious error monitoring and one that did not. Conscious error monitoring did not lead to changes in error rate between the two tasks, while other antisaccade parameters were significantly modulated. Application of signal detection theory showed a large inter-individual variability in error detection sensitivity: the majority of individuals were unable to monitor antisaccade errors (chance error detection group), while a minority that successfully monitored their errors (non-chance error detection group) were worse in antisaccade performance in both tasks. These results were explained by the hypothesis of two modes of antisaccade processing favored by each one of the two groups:

a mode of conscious cortical cognitive control leading to error monitoring, worse performance and no post-error adaptation and a mode of non-conscious subcortical control leading to chance error monitoring, post-error slowing and better performance of the antisaccade task. This hypothesis was corroborated by the results of the genotype analysis. Error-monitoring sensitivity in the non-chance error detection group was modulated by COMT genotype variations that in turn did not have an effect on error rate. On the other hand, DRD4 genotype variations were related to differences in antisaccade error rate while not affecting error-monitoring sensitivity. Keywords Saccade  Inhibition  Error detection  Post-error adaptation  Gene  Single nucleotide polymorphism

Introduction E. Kattoulas  I. Evdokimidis  N. Smyrnis Cognition and Action Group, Neurology Department, School of Medicine, Eginitio Hospital, National and Kapodistrian University of Athens, Athens, Greece N. C. Stefanis  N. Smyrnis (&) 1st Psychiatry Department, School of Medicine, Eginitio Hospital, National and Kapodistrian University of Athens, 74 V. Sofias Ave., 11528 Athens, Greece e-mail: [email protected] E. Kattoulas  N. C. Stefanis  C. N. Stefanis University Mental Health Research Institute, National and Kapodistrian University of Athens, Athens, Greece D. Avramopoulos School of Medicine, Department of Psychiatry and Behavioral Sciences, Johns Hopkins University, Baltimore, MD, USA

Conscious monitoring of errors in the performance of cognitive tasks is an essential function for optimal performance adaptation and learning. Many influential models of cognitive control incorporate an error-monitoring system that is believed to rely on the activity of prefrontal and cingulate cortical areas (Carter et al. 1998; Miller and Cohen 2001; Botvinick et al. 2001, 2004; Ridderinkhof et al. 2004; Velanova et al. 2008) as well as dopaminergic pathways involving the basal ganglia and the mesolimbic dopamine nuclei (Holroyd and Coles 2002). Not all tasks that require cognitive control are accompanied by conscious monitoring, as can be inferred by error recognition. An example of this exception is the antisaccade task, which involves the inhibition of a visually triggered saccade in order to perform a voluntary saccade in the

123

454

opposite direction (Hallett 1978; Fischer and Weber 1992). Mokler and Fischer (1999) showed that less than halve of the errors in task performance (error saccade before the appropriate antisaccade) were consciously detected by normal subjects. Previous studies on whether error awareness is linked to post-error behavioral adjustments yielded contradictory results, some reporting post-error adjustments only after detected errors (Nieuwenhuis et al. 2001; Klein et al. 2007), while others demonstrated this effect after both aware and unaware errors (Rabbitt 2002; Endrass et al. 2005). Event-related potential studies in humans showed that both recognized and unrecognized antisaccade errors produced an error-related negativity (Nieuwenhuis et al. 2001; Endrass et al. 2007) thought to reflect error processing in the anterior cingulate cortex (Van veen and Carter 2002). In a recent fMRI study, both recognized and unrecognized errors produced activation of the rostral cingulate cortex bilaterally (Klein et al. 2007). These studies then showed that for this task a large proportion of errors committed activate errormonitoring areas but still remain undetected and do not lead to performance adjustments. In all the aforementioned studies of antisaccade error detection, subjects were pooled together and trials were divided into detected and undetected errors or correct responses. This type of analysis though fails to address the important question of whether each individual can consciously detect antisaccade errors. The present study assessed performance in an antisaccade task with error monitoring and a simple antisaccade task where error monitoring was not a task requirement, in a sample of 609 apparently healthy young men. Using signal detection theory (MacMillan and Creelman 2005), we investigated individual differences in error detection sensitivity as well as whether the extra requirement of error detection leads to differences in antisaccade task performance. In the same population of young men, the catechol-omethyltransferase (COMT) and dopamine D4 receptor (DRD4) gene variations were measured. The val158 met single nucleotide polymorphism (SNP) of COMT gene is known to result in differences in dopamine regulation especially in the prefrontal cortex (Chen et al. 2004) and differences in performance of executive function tasks (Egan et al. 2001) and monitoring tasks (Fossella et al. 2002; Blasi et al. 2005). The variations in SNP-521 of DRD4 gene are known to result in differential transcription of the DRD4 receptor (Okuyama et al. 1999). This polymorphism was correlated with error-related negativity in an error-monitoring task (Kra¨mer et al. 2007). Both COMT and DRD4 genes are dopamine activity– related genes and linked to either attention control or error monitoring, so we investigated their effects in task performance and error detection sensitivity in the antisaccade task at population level.

123

Exp Brain Res (2010) 203:453–463

Materials and methods Participants The data for this analysis are derived from a large epidemiological study (ASPIS, Athens Study of Psychosis proneness and Incidence of Schizophrenia) in which a sample of 2,130 young Greek men 18–24 years of age, selected from the conscripts of the Greek Air Force, performed a battery of eye movement, including 90 trials of the simple antisaccade task and cognitive tasks (Evdokimidis et al. 2002; Smyrnis et al. 2002, 2003). A subgroup of 690 individuals within this population performed 40 trials of the monitoring antisaccade task. Each individual was informed about the experimental procedures and study goals and gave written consent to participate in the study. The study protocol was approved by the ethics committee of the University Mental Health Research Institute. DNA extraction and COMT and DRD4 typing A mouthwash for DNA extraction was obtained from all 2,130 subjects. DNA was extracted from mouthwash as follows: 15 ml of sterile saline solution was supplied, and the subjects were instructed to perform a rigorous mouthwash for approximately 20 s. The sample was centrifuged at 1,440 g for 15 min at 4°C. The pellet was resuspended in 2 ml lysis buffer (2% SDS, 0.1 M NaCl, 0.05 M trisHCl (pH 8.0) 1 mM EDTA) and incubated at 37°C for 16 h with 2 mg of proteinase K. Proteins were precipitated and removed with the addition of 1 ml of 6 M NaCl and centrifugation at 2,160 g for 20 min at 4°C. DNA was precipitated from the supernatant with ethanol (Avramopoulos et al. 2002). All genotyping was performed blind to phenotype measures by K-Biosciences (Herts, UK) (http://www.kbioscience.co.uk/) using a competitive allele-specific PCR system (CASP). Overall genotyping error rate has been estimated to be B0.3% (based on the repeatability of the genotype calls in the same samples). Procedure The apparatus and procedure used for administering the eye movement tasks in the ASPIS study were described in detail in our previous reports (Evdokimidis et al. 2002; Smyrnis et al. 2003). Eye movements were recorded from the right eye using the IRIS SCALAR infrared device. Stimulus presentation and recording of the responses was accomplished with a program written in Turbo Pascal 7.0 for DOS. A 12-bit A/D converter was used for data acquisition (Advantech PC-Lab Card 818L). Eye

Exp Brain Res (2010) 203:453–463

movement data were sampled at 600 Hz and stored in each PC for off-line data processing. Each trial of the simple antisaccade task started with the appearance of a central fixation target (white cross 0.5 9 0.5 degrees). After a variable period of 1–2 s, the central target was extinguished and an identical target appeared randomly at one of 9 target amplitudes from the center (2–10 degrees at 1 degree intervals) and two directions (left or right). The subjects were instructed to make an eye movement as quickly as possible to the opposite direction from that of the peripheral target and to hold that position until the peripheral target disappeared and the central fixation target reappeared 1.5 s later. In the monitoring antisaccade task, an inter-trial interval of 6 s was added, during which the experimenter asked the subject whether he performed a correct antisaccade in the previous trial or he made an error. The correct response was defined as a single saccadic eye movement (antisaccade) in the opposite direction of the peripheral target. An error was defined as the case in which the eyes made one or more saccades in the direction of the peripheral target (error pro target saccade) even if these were then followed by a correction saccade in the opposite direction, which was true in the vast majority of cases. Data processing An interactive PC program (created using the TestPointÒ CEC) was used for detection and measurement of saccades from the eye movement record. The details of the preprocessing of the eye movement records in the antisaccade task are given in our previous reports (Evdokimidis et al. 2002; Smyrnis et al. 2003). Figure 1a presents the horizontal position (solid line) and instantaneous speed record (dotted line) for a correct antisaccade trial. The instantaneous speed record was used to define the onset and the end of the antisaccade using the criterion that a number of 5 consecutive speed values (8 ms duration at 600 Hz) were above a predefined noise level. The noise level was determined by taking the root mean square of the signal in 15 windows of 20 values each, covering the first 500 ms of the 1–2 s period of central fixation and then taking the median value of these 15 values. The end of the saccade was the return of the instantaneous speed to the noise level. The onset of the antisaccade was used to compute the latency of the correct antisaccade (LA), and the end of the antisaccade was used to compute its amplitude (AA) in the position record (Fig. 1a). Figure 1b presents the horizontal position (solid line) and instantaneous speed record (dotted line) of a trial in which an error prosaccade occurred and then a correction antisaccade followed. The onset and the end of the erroneous prosaccade were defined in the instantaneous speed record using the same criteria as

455

a AA

490deg/sec

7 deg LA AE

AC

b

200 msec LE

LC

Target onset Fig. 1 Position and instantaneous speed record of a correct antisaccade trial (a) and an error prosaccade with correction antisaccade trial (b). LA: correct antisaccade latency AA: correct antisaccade amplitude LE: erroneous prosaccade latency AE: erroneous prosaccade amplitude LC: corrective antisaccade latency AC: corrective antisaccade amplitude

for the correct antisaccades. These time points were used to compute the latency of the error prosaccade (LE) and its amplitude (AE) in the position record (Fig. 1b). Then, the program defined the onset and the end of all subsequent movements in the direction of the target until the onset of the movement in the opposite direction of the target (correction antisaccade, Fig. 1b). The time from the end of the last movement toward the direction of the target until the onset of the correction antisaccade was computed and is referred to as the latency to correction (LC). The end of the correction antisaccade was used to compute the amplitude of the correction antisaccade (AC). Subject data analysis In this first part of the analysis, we derived performance indices for each subject in the simple and the monitoring antisaccade task. In the simple antisaccade task, we used the first 40 trials that were presented out of the total of 90 in order to have the same number of trials as for the monitoring antisaccade task. We excluded all trials with artefacts (blinks, etc.) in the period of the execution of the first saccade or any type of eye movement in the period of 100 ms before the appearance of the peripheral stimulus. We also excluded all trials in which the first saccade (whether correct antisaccade or erroneous prosaccade) had a response latency that was less than 80 ms (predictive saccades) and all trials where the first saccade was less than 1 deg in amplitude. In order for each subject to be included in the analysis, we used the criterion that he performed at least 20 valid trials in the simple antisaccade task (50% of total trials measured) and 20 valid trails in the monitoring

123

456

Exp Brain Res (2010) 203:453–463

antisaccade task. Using these exclusion criteria, we retained 609 subjects (88.3% of the total number of subjects) for which all subsequent analyses were performed. The percentage of trials in which an erroneous prosaccade occurred was computed for each subject in each task. This percentage was normalized via an arcsine transformation for percentages (Snedecor and Cochran 1980), and the transformed values were used for all statistical analyses. The median latency for correct antisaccades, the median latency for error prosaccades and the median latency to correction were also computed for each subject and each task. Finally, the median amplitude for correct antisaccades, error prosaccades and correction antisaccades were computed for each subject and each task. The differences between these values between the two tasks were studied using paired t-tests. The responses of each subject in the monitoring antisaccade task were grouped in four categories depending on the eye movement recognition. Thus, a correct antisaccade could either be reported as such (N1, correct recognition) or as error prosaccade (N2, false alarm). An error prosaccade could be falsely reported as correct antisaccade (N3, miss) or detected and reported as error (N4, hit). The percentage of correct error detections (hits) was defined as: PðhÞ ¼ ðN4=ðN3 þ N4ÞÞ  100

ð1Þ

The percentage of erroneous detections of correct antisaccades as errors (false alarms) was defined as: PðfaÞ ¼ ðN2=ðN1 þ N2ÞÞ  100

ð2Þ

By transforming these two percentages in z-scores, a dprime score for error detection in the antisaccade task was computed as follows (MacMillan and Creelman 2005): d-prime ¼ ZðPðhÞÞ  ZðPðfaÞÞ

ð3Þ

In cases where the percentage of hits or the percentage of false alarms was 0 or 100%, we used the formulas ‘ 9 N and 1-‘ 9 N, respectively, to compute the zscores. N was the number of valid trials in the task (MacMillan and Creelman 2005). The d-prime score is a measure of detection sensitivity that is free of response bias (MacMillan and Creelman 2005). This means that an individual with a large error percentage in the antisaccade task will not by necessity have a lower d-prime compared to an individual with a lower error percentage in the task because d-prime depends equally on detected errors and false alarms. A d-prime score that is greater than zero indicates that error detection sensitivity for this subject is above chance level, while a d-prime score at or below zero indicates that error detection sensitivity is at chance level. Using the d-prime score of each subject, we

123

divided our population of 609 subjects into a group of individuals with non-chance error detection (d-prime [ 0) and a group of subjects with chance error detection (d-prime B 0). Single trial data analysis For the analyses on single trial data, we used the pooled single trials from all subjects in the monitoring antisaccade task that were divided in two groups: (1) the group of trials from all subjects with non-chance error detection (d-prime [ 0) and (2) the group of trials from all subjects with chance error detection (d-prime B 0). We then used these two groups to perform the following analyses: Average pooled sensitivity measurement The measurement of d-prime with only a few trials per subject is prone to errors due to large random variation (MacMillan and Creelman 2005). Pooling data from many subjects gives a more reliable estimate of the true d-prime of the population. Especially in the case where the number of trials for each subject is very small, as was the case here, the d-prime of the pooled data is a better estimate than the mean of the d-primes for each subject (MacMillan and Creelman 2005). We pooled all responses in the monitoring antisaccade task (N1, N2, N3, N4) for all subjects with chance error detection (d-prime B 0) and all subjects with d-prime [ 0. We then computed the pooled p(h) and p(fa) to derive a d-prime score for the pooled data for each one of the two groups using formula (3). Then, we used the following formula to compute the variance for each one of these pooled d-prime scores (MacMillan and Creelman 2005): h i d-prime var ¼ ½PðhÞ  ð1  PðhÞÞ= Nerrors  ðUðPðhÞÞÞ2 þ ½PðfaÞ  ð1  PðfaÞÞ=½Ncorrect  ðUðPðfaÞÞÞ2 

ð4Þ

In formula (4), Nerrors is the total number of errors in the task for each group (equals the number of hits and misses) and Ncorrect is the total number of correct antisaccades in each group (equals the number of correct recognitions and false alarms). The function U for each proportion of hits (P(h)) and false alarms (P(fa)) is given by the formula: 2 p UðPÞ ¼ 1= ð2  pÞ  e1=2ðPÞ ð5Þ Using formula (4), the variance of the pooled d-prime was computed, and then standard error of d-prime was derived by:

Exp Brain Res (2010) 203:453–463

d-prime se = d-prime var1=2

457

ð6Þ

The 95% confidence intervals (CI) for the d-prime were derived by: 95% CI ¼ d-prime  1:96  ðd-prime seÞ

ð7Þ

Saccade parameters, error recognition and post-error slowing Trials for correctly identified error prosaccades (hits) and unidentified error prosaccades (misses) were pooled within each group of chance and non-chance error detection. Then, we compared the mean latency of error prosaccades, the mean amplitude of error prosaccades, the mean latency of correction antisaccades and the mean amplitude of correction antisaccades between hits and misses separately within each group of chance and non-chance error detection using t-tests. For the estimation of post-error slowing, latencies of correct antisaccades following an error prosaccade were compared to the latencies of correct antisaccades following a correct trial. We compared latencies for all subjects polled together, as well as separately for each group of chance and non-chance error detection. The analysis was performed for both the simple and the monitoring antisaccade task.

between the monitoring and simple antisaccade task. There was no significant difference in error rate between the two tasks. The latency for correct antisaccades, error prosaccades and correction antisaccades after the occurrence of an error were higher in the monitoring antisaccade task compared to the simple antisaccade task (Table 1). The amplitude of all eye movements was smaller in the monitoring antisaccade task compared to the simple antisaccade task (Table 1). Error detection in the monitoring antisaccade task The majority of erroneous prosaccades were followed by a corrective antisaccade (81.4 vs. 18.6% not followed by a corrective antisaccade). Still, not all of the erroneous prosaccades were detected. Error detection performance for all subjects is presented in Table 2. The majority of individuals were at chance level in their ability to detect errors, having a d-prime score of zero or less than zero. A d-prime could not be computed for 53 individuals who had perfect performance in the task (groups 1 and 2 in Table 2). The difference in the proportions of individuals with chance (d-prime B 0) and non-chance (d-prime [ 0) error detection in the population was highly significant (P-two sided \ 10-4). For the group of non-chance error detection, the pooled percentage of hits (see ‘‘Methods’’) was 35%, and

Relation of COMT and DRD4 genotype to error performance and error detection sensitivity From the 556 individuals that had a valid d-prime score, we had COMT genotype data for 518 and DRD4 genotype data for 457. The effect of COMT genotype on error rate and in the monitoring antisaccade task was investigated for all subjects using a regression model with val load as the independent factor (values of: 0 for met/met, 1 for met/val and 2 for val/val). The same analysis was used to study the effect of COMT genotype on d-prime. This analysis was performed only in the group with non-chance error detection, since trying to evaluate the d-prime variation in chance error detection group is meaningless. The effect of DRD4 genotype variations were also studied on error rate and d-prime using the same regression model. In this case, the independent predictor was the T load (values of: 0 for C:C, 1 T:C and 2 T:T).

Results Comparison of monitoring and simple antisaccade tasks Table 1 presents the results of the t-tests comparing antisaccade error rate, eye movement latency and amplitude

Table 1 Comparison of performance in the simple and monitoring antisaccade tasks Saccade parameter

Mean (SD) Mean (SD) Paired t (P) df Simple Monitoring

Error rate (%)

20.5 (17.6) 19.5 (17.5) 1.9 (0.06)

Antisaccade Lat. (ms)

257 (40)

306 (66)

23 (\10-5) 607

Error Prosac. Lat. (ms)

199 (48)

219 (76)

5.5 (\10-5) 525

Correction Lat. (ms)

135 (66)

150 (76)

4 (\10-4)

Antisaccade Amp. (deg) 8 (2.8)

7.1 (2.9)

8.3 (\10-5) 607

Error Prosac. Amp. (deg)

5 (2.2)

4.4 (1.9)

5.5 (\10-5) 525

Correction Amp. (deg)

11.3 (4.6)

9.5 (4.1)

8.1 (\10-5) 497

608

501

Table 2 Categorization of subjects based on their success in error detection (d-prime) Number 1. Performed no errors and reported no errors (d-prime not defined)

20 (3.3%)

2. Performed no errors and reported errors (d-prime not defined)

33 (5.4%)

3. d-prime B 0 chance error detection

370 (60.8%)

4. d-prime [ 0 non-chance error detection

186 (30.5%)

Total

609

123

458

Exp Brain Res (2010) 203:453–463

Table 3 Comparison of chance and non-chance error detection groups Mean (SD) d-prime B 0

Mean (SD) d-prime [ 0

t (P)

df

Age (years)

20.9 (1.8)

21.1 (1.8)

1.3 (0.2)

554

Education (years)

12.7 (1.9)

12.6 (2)

0.8 (0.4)

554

Error rate (%)

18.6 (15.9)

26.7 (18.4)

5.6 (\10-5)

554

Antisaccade Lat. (ms)

305 (59)

310 (69)

0.9 (0.4)

554

Error Prosac. Lat. (ms)

223 (78)

219 (80)

0.5 (0.6)

554

Correction Lat. (ms)

153 (72)

159 (86)

0.7 (0.5)

536

Antisaccade Amp. (deg)

7.1 (2.8)

7.1 (3)

0.1 (0.9)

554

Error Prosac. Amp. (deg)

4.3 (1.9)

4.5 (1.8)

1 (0.3)

554

Correction Amp. (deg)

9.5 (4.1)

9.1 (4.1)

1 (0.3)

536

Error rate in simple task

19.5 (16.1)

26.5 (19.1)

4.5 (\10-5)

554

Group differences in antisaccade task parameters In this analysis, differences between the chance and nonchance error detection groups were investigated. Table 3 presents the comparison of the demographic features and monitoring antisaccade task parameters between the two groups. Individuals with chance error detection made fewer errors in the monitoring antisaccade task compared to individuals with non-chance error detection. These same individuals also performed better in the simple antisaccade task (Table 3). For each subject, the percentage of errors did not differ between the monitoring and simple antisaccade task (for the non-chance error detection group: paired t-test t = 0.4, P = 0.7, df = 185, for the chance error detection group: paired t-test t = 0.76, P = 0.4, df = 369). Figure 2 presents the relationship between antisaccade error rate and error detection sensitivity (d-prime) in the monitoring antisaccade task.

123

1.0

0.8

percentage of errors

the pooled percentage of false alarms was 20% resulting in a pooled d-prime score of 0.46 (95% confidence intervals 0.41–0.5, see ‘‘Methods’’) confirming that this group in average had better than chance error detection sensitivity. For the group of chance error detection, the pooled percentage of hits was 9.5%, and the pooled percentage of false alarms was 14.2% resulting in a pooled d-prime score of -0.24 (95% confidence intervals -0.27 to -0.2), confirming that this group in average had chance error detection sensitivity. Thus, the reason for the lower d-prime scores in the chance error detection group seems to be the lower detection of erroneous movements. This group has also a lower percentage of false alarms, a result compatible with the possibility that these subjects report movements as correct by chance. In the remaining sections of the results, we analyzed data separately for the two groups of individuals with chance and non-chance error detection performance.

0.6

0.4

0.2

0.0 -3

-2

-1

0

1

2

3

4

5

dprime

Fig. 2 Scatter plot of antisaccade performance (percentage of errors) versus error detection performance (d-prime) for all subjects in the monitoring antisaccade task

The two groups of individuals with chance and nonchance error detection did not differ in age or years of formal education (Table 3). Group differences in the relation of saccade parameters to response type In this analysis, we tried to identify which characteristics of the antisaccade error lead to its recognition by the subject. In order to do this, we investigated whether identified antisaccade errors (hits) differed in the characteristics of error prosaccade or correction antisaccade (latency and/or amplitude) from unidentified errors (misses). It was further hypothesized that if such differences would be present, they should be present only in the non-chance error detection group. In Table 4, the pooled means of error prosaccade and correction antisaccade parameters for hits and misses in the monitoring antisaccade task are compared separately for

Exp Brain Res (2010) 203:453–463

459

Table 4 Saccade parameter differences between hits and misses for non-chance (A) and chance (B) error detection groups Saccade parameter

Mean for hits (SD)

Mean for misses (SD)

t(P)

df

222 (109)

219 (91)

0.6 (0.6)

A Error Prosac. Lat. (ms)

-3

1,469

Correction Lat. (ms)

179 (121)

158 (104)

3.3 (10 )

1,253

Error Prosac. Amp. (deg)

5.2 (3.1)

4.6 (2.8)

3.4 (\10-3)

1,469

Correction Amp. (deg)

10.2 (6.4)

10.3 (5.5)

0.2 (0.8)

1,196

Error Prosac. Lat. (ms)

212 (99)

218 (85)

0.8 (0.4)

2,030

Correction Lat. (ms)

164 (116)

150 (108)

1.6 (0.1)

1,745

Error Prosac. Amp. (deg)

4.8 (3.1)

4.7 (3)

0.4 (0.7)

2,030

Correction Amp. (deg)

10 (5.2)

10.3 (5.7)

0.6 (0.6)

1,679

B

the non-chance error detection group (Table 4A) and the chance error detection group (Table 4B) (see ‘‘Methods’’). Table 4A shows that the latency of the correction antisaccade and the amplitude of the error prosaccade were significantly larger for hits than for misses in the nonchance error detection group. No significant differences were observed for any of the parameters in the group of individuals with chance error detection (Table 4B). In conclusion, this analysis showed that detected errors had larger error amplitudes and longer latencies to the correction antisaccade compared to undetected errors, and this effect was only present in the non-chance error detection group as expected. Group differences in post-error slowing In this analysis, post-error adaptation was investigated. In order to do this, we compared the antisaccade latency of correct movements following an error prosaccade to the antisaccade latency of correct movements following a correct antisaccade. For the monitoring antisaccade task, no significant differences were observed when all subjects were polled together (mean latency for antisaccades following an error prosaccade = 314.6 ± 107.6, mean latency for antisaccades following a correct antisaccade = 313.6 ± 104.2, t = 0.39, P [ 0.05, df = 12,803). The same was true when the analysis was performed separately for the group of chance error detection (mean latency for antisaccades following an error prosaccade = 313.9 ± 106.8, mean latency for antisaccades following a correct antisaccade = 311.4 ± 102, t = 0.74, P [ 0.05, df = 7,846) and non-chance error detection (mean latency for antisaccades following an error prosaccade = 315.9 ± 109.2, mean latency for antisaccades following a correct antisaccade = 320.8 ± 109.1, t = -1.09, P [ 0.05, df = 3,426). No significant differences were observed when only movements that followed a

correctly identified error prosaccade or correct antisaccade were included. The same analysis was performed for the simple antisaccade task. When movements from all subjects were pooled together antisaccades following an error prosaccade showed a significantly larger latency compared to antisaccades following a correct antisaccade (mean latency for antisaccades following an error prosaccade = 267.9 ± 81.1, mean latency for antisaccades following a correct antisaccade = 263.1 ± 72.4, t = -2.88, P \ 0.01, df = 14,324). When a separate comparison for the two groups was performed, a significant difference was observed for the chance error detection group (mean latency for antisaccades following an error prosaccade = 267 ± 85.1, mean latency for antisaccades following a correct antisaccade = 262.4 ± 71.8, t = -2.12, P \ 0.05, df = 8,791) but not for the non-chance error detection (mean latency for antisaccades following an error = 269.4 ± 76.3, mean latency for antisaccades following a correct antisaccade = 265 ± 73.4, t = -1.55, P [ 0.05, df = 4,043). Group differences in the relation of COMT and DRD4 genotype to error rate and error detection sensitivity In this final analysis, we investigated whether the COMT and DRD4 genotype variations were related to the antisaccade error rate and error detection sensitivity in the monitoring antisaccade task (d-prime score). The frequencies of subjects for each genotype (the three COMT genotype groups and the three DRD4 genotype groups) both in the total sample and the two groups (chance and non-chance error detection group) were not deviant from those expected for genotypes in Hardy–Weinberg equilibrium (Table 5). Also, the frequencies for each genotype in the two groups of chance and non-chance detection did not differ from each other both for COMT (X2 = 0.003 P = 0.98) and DRD4 gene (X2 = 4.06 P = 0.13).

123

460

Exp Brain Res (2010) 203:453–463

Table 5 COMT and DRD4 genotype frequencies in the sample met/met

met/val

val/val

HWE (P)

All subjects (518)

116

269

133

0.83

d-prime B 0 (345)

80

177

88

0.91

d-prime [ 0 (177)

36

92

45

0.62

T:T

T:C

C:C

HWE (P)

All subjects (457)

120

203

134

0.48

d-prime B 0 (303)

47

70

37

0.43

d-prime [ 0 (154)

73

133

97

0.25

A: COMT

B: DRD4

HWE(P) Hardy–Weinberg equilibrium X2 test P value

Figure 3a shows the effect of the COMT genotype variations on error rate in the monitoring antisaccade task for all subjects. The regression linear model for the val load showed no significant effect of COMT on error rate (val load regression: r = 0.04, F1,564 = 0.99, P = 0.32). Figure 3b shows the effect of the COMT genotype variations on d-prime. This analysis was performed only for the nonchance error detection group (see ‘‘Methods’’). There was a marginally significant decrease in d-prime with increasing val load (r = 0.15, F1,171 = 3.77, P = 0.05). Further excluding 15 individuals with the lowest d-prime scores (lower than 0.1) resulted in a significant effect of val load on d-prime (r = 0.18, F1,156 = 5.16, P = 0.02).

Discussion The first question that this study tried to answer was whether error monitoring would result in a difference in performance in the antisaccade task. It was found that error rate did not differ between the two tasks (simple vs. monitoring) for each subject, suggesting that specifically instructing subjects to monitor their performance does not affect task performance. Other parameters though differed significantly between the two tasks (increased latency and decreased amplitude of antisaccades as well as error prosaccades and correction antisaccades for the monitoring

c

25

Error rate (%)

Error rate (%)

a

Figure 3c shows the effect of the DRD4 genotype variations on error rate for all subjects. The regression model showed a significant decrease in error rate with increasing T load (r = 0.1, F1,495 = 5.49, P = 0.02). Figure 3d shows the effect of the DRD4 genotype variations on d-prime for the group of non-chance error detection. The effect was not significant (regression: r = 0.007, F1,152 = 0.007, P = 0.93). In conclusion, this analysis showed that error-monitoring sensitivity (d-prime) was modulated by COMT but not DRD4 genotype variations. Increasing val load led to decreasing sensitivity in the non-chance error detection group. In contrast, antisaccade error rate was modulated by DRD4 but not COMT genotype variations. Error rate decreased with increasing T load.

20

15

10 met/val

15

10

val/val

1.5

d 1.5

1.0

1.0

d-prime

b

0.5

0.0 met/met

met/val

val/val

Fig. 3 a Graphic representation of the mean error rate (and standard error of the mean) for the three COMT genotype groups (presented in the X-axis) b. Graphic representation of the mean d-prime score (and standard error of the mean) for the group of subjects with non-chance error detection for the three COMT genotype groups (presented in the X-axis) c. Graphic representation of the mean error rate (and standard

123

20

d-prime

met/met

25

C:C

T:C

T:C

C:C

T:T

0.5

0.0 T:T

error of the mean) for the three DRD4 genotype groups (presented in the X-axis) d. Graphic representation of the mean d-prime score (and standard error of the mean) for the group of subjects with non-chance error detection for the three DRD4 genotype groups (presented in the X-axis)

Exp Brain Res (2010) 203:453–463

antisaccade task). One possible explanation for these differences could be the longer inter-trial interval used in the monitoring antisaccade task, necessary for the subjects to report performance. This large interval was not present in the simple antisaccade task. In a previous study (Smyrnis et al. 2002), we investigated the effect of inter-trial interval on antisaccade latency, and we observed a decrease in latency with increasing interval, opposite to the present study’s effect. Still, we did not study the effects of very large inter-trial intervals as the one used in the monitoring antisaccade task. Thus, different task rhythms could lead to different responses. Another possible explanation could be the dual task condition in the monitoring antisaccade task (perform an antisaccade and report the outcome) compared to the simple antisaccade task. Still, one would expect to observe differences in task performance if this was indeed a dual task paradigm. The second question this study tried to answer was how successful were the individual subjects in error detection. Mokler and Fischer (1999) have already shown that only 40% of antisaccade objective errors could be detected consciously, suggesting that conscious error monitoring in this task is far from being accurate. This study investigated the error detection sensitivity of each individual using signal detection theory. The results broaden the original observation of Mokler and Fischer (1999), showing that the majority of individuals in this large population of apparently healthy young men were at chance level in their ability to consciously detect antisaccade errors in performance, while only a minority of about 30% of individuals were above chance level. Thus, we divided the population in two groups of individuals: the chance error detection and the non-chance error detection group. We argued that maybe instructing subjects to monitor performance would have an effect on task performance only for subjects that successfully monitor their performance namely the nonchance error detection group. These differences might dissolve by including the majority of individuals with chance error detection. The results were again counterintuitive but clear. There was no difference in error rate between the two antisaccade tasks in the group with nonchance error detection sensitivity, indicating that even for this specific group of non-chance error detection specifically instructing subjects to monitor their performance does not interfere with antisaccade performance. We further investigated differences between the two groups of subjects. We observed that the non-chance error detection group produced more objective errors in both antisaccade tasks than the chance error detection group. It should be noted that the increased number of antisaccade errors in the non-chance error detection group is not a trivial finding, since an increase in error rate does not necessarily mean that the error detection sensitivity will

461

increase as well. Error detection sensitivity as measured by d-prime depends equally on detection of true errors and avoidance of false alarms. Actually, using the pooled data analysis for the two populations, it was shown that the chance error detection group had a lower pooled percentage of both correctly identified errors and false alarms compared to the non-chance error detection group. Thus, we could hypothesize that subjects that cannot monitor their performance use the strategy to report most movements as correct. For the non-chance error detection group, the detected errors had significantly larger error amplitudes and longer latencies for the corrective antisaccade confirming previous results (Mokler and Fischer 1999; Endrass et al. 2007) compared to the non-detected errors. This difference was absent in the group of chance error monitoring as would be expected if these individuals reported errors at chance. Finally, the analysis of post-error adaptation confirmed yet another difference between the two groups’ individuals with chance and non-chance error monitoring. Delay in antisaccade reaction time (post-error slowing) was observed after the commission of an error in the simple antisaccade task, as would be expected (Rabbitt 1966). This was not observed in the monitoring antisaccade task, probably due to the longer latencies for all movements observed in the task or to the large inter-trial interval for error reporting in this task. Interestingly, it was shown that post-error adaptation was observed only in the chance error detection group. This finding further strengthens our hypothesis for the separate groups of individuals that perform the antisaccade task in a different mode of performance. Summarizing, the first part of the study demonstrated that for each individual subject the instruction to actively monitor the performance does not influence task performance. We suggest two different hypotheses that could explain this finding. According to the first hypothesis, the two procedures, namely the antisaccade task and task monitoring, are totally independent processes, and the presence of the latter does not influence the performance in the former. This could be the case if task monitoring is, for example, an offline procedure, minimizing cognitive demands during task performance. According to the second hypothesis, the two procedures are distinct but strongly linked together, meaning that task monitoring might be present even when it is not a task prerequisite. In the second part of this study, we identified two distinct groups of subjects: the first group includes the majority of subjects that report performance errors at chance, whereas the second group includes a minority of subjects that identifies and reports errors according to their metric characteristics. The second group displays worse task performance namely more antisaccade errors, for both the simple and the monitoring antisaccade task than the chance error detection

123

462

group. Also, this group shows no post-error adaptation in the simple antisaccade task, while on the other hand, the first group of subjects that displays better task performance but cannot consciously detect errors displays post-error adaptation in the simple antisaccade task. The difference in task performance between the two groups makes more plausible the second hypothesis namely that task performance and error monitoring are two procedures that are distinct but strongly linked together. Subjects either monitor their performance in the antisaccade task even if they are not explicitly instructed to do so and are effective in error detection at the expense of task performance or (and this includes the majority of subject in our analysis) they are unable to detect errors, even if error recognition is a task requirement but perform better in the antisaccade task. For the first group of subjects, we could hypothesize that part of their attentional resources is allocated to successfully detect errors in performance, while for the majority of subjects their inability to consciously monitor performance could result to the focusing of attentional resources to successful task performance (inhibiting error prosaccades). The finding that post-error slowing is observed only for the group of subjects that do not consciously detect errors in performance comes in accordance with previous studies stating that post-error adaptation was observed even in those error trials that were not consciously registered and signaled (Rabbitt 2002; Endrass et al. 2005) and is not influenced by drugs that have an effect on error detection (Riba et al. 2005). A recent study using event-related potentials and fMRI (Marco-Pallare´s et al. 2008) suggested that brain areas associated with error detection can be dissociated from those involved in post-error adaptive actions. Adding to those findings, our study also favors the hypothesis that some subjects use error detection procedures at the expense of performance and in that case no post-error adaptation procedures are involved. The majority of subjects do not involve error detection mechanisms in task execution and, in that case, post-error adaptation is observed and task performance is better. The antisaccade task that was used in our paradigm requires the subject to inhibit a reflexive saccade toward a novel peripheral target and instead generate a voluntary movement to the contralateral visual field (Hallett 1978). An error in the task is thought to reflect an inability to inhibit prepotent responses to the target (Massen 2004). In order to avoid this error, the level of activity of fixation neurons in the superior colliculus is increased (a form of a neural ‘‘brake’’) (Schall et al. 2002; Munoz and Everling 2004). The inhibition signal to the superior colliculus comes from the basal ganglia and the dorsolateral prefrontal cortex (Pierrot-Deseilligny et al. 2004). In accordance with our previous hypothesis, we could speculate that for the group of subjects that cannot effectively

123

Exp Brain Res (2010) 203:453–463

monitor their performance, the major source of inhibitory signals is subcortical, that is from the basal ganglia. On the other hand, for the minority of subjects that successfully monitor performance, we could hypothesize that DLPFC is the major origin of inhibitory signals, a cortical area that, as such, is more accessible to awareness. Finally, we studied the effect of COMT and DRD4 genotype variations on error detection sensitivity and objective error performance (error rate). Error detection sensitivity decreased with increasing val load in the nonchance error detection group. This is in agreement with previous works (Egan et al. 2001; Blasi et al. 2005) linking COMT variations to prefrontal activity which, according to our previous hypothesis, is strongly linked to antisaccade performance in this group of subjects. In contrast to this observation, increasing DRD4 T load resulted in a decrease in error rate, while error detection sensitivity was not affected by DRD4 genotype variations. Studies with D4 receptor knockout mice suggest that D4 receptor expression rate leads to compensatory changes in dopamine production in the basal ganglia (Rubinstein et al. 1997). Additionally, previous studies have shown the presence of D4 receptors not only in prefrontal areas but also in the striatum (for a review Oak et al. 2000). Finally, Kra¨mer et al. (2007) observed that the DRD4 T load had a specific effect on error-related negativity. These authors showed that T:T homozygotes had a larger error-related negativity compared to C:C homozygotes and explained their result as an indication that this polymorphism had a specific effect on error processing by the mesencephalic dopamine pathways. So, the specific effect of DRD4 variation on error rate observed in our study can be explained through the relation of D4 receptor expression not only with prefrontal circuits but also with basal ganglia activity. In accordance with our hypothesis, basal ganglia activity is linked to the inhibitory processes of the antisaccade task for the majority of individuals displaying optimal antisaccade performance. In conclusion, this study leads to the hypothesis of two distinct modes of cognitive control in the antisaccade task. As inferred by inter-individual differences in task performance and error recognition in a large population of healthy young male, subjects are able to use one of the two distinct modes. These modes of cognitive control were found to be selectively modulated by genotype variations of the COMT and DRD4 genes. Future research is necessary to replicate the original findings and elucidate the exact neurophysiological background of each mode of cognitive control. These results could also lead to testable hypotheses concerning the behavior of patients suffering from CNS disorders such as schizophrenia and Parkinson’s disease. It is known that these patients present with deficits in objective antisaccade performance (Everling and Fischer 1998). It would be interesting to investigate error

Exp Brain Res (2010) 203:453–463

monitoring sensitivity in these patients and its relation to COMT and DRD4 genotype variations. Acknowledgments This work was supported by the grant ‘‘EKBAN 97’’ to Professor C.N. Stefanis from the General Secretariat of Research and Technology of the Greek Ministry of Development. Conflict of interest statement nical support for this project.

‘‘Intrasoft Co’’ provided the tech-

References Avramopoulos D, Stefanis NC, Hantoumi I, Smyrnis N, Evdokimidis I et al (2002) Higher scores of self reported schizotypy in healthy young males carrying the COMT high activity allele. Mol Psychiatry 7:706–711 Blasi G, Mattay VS, Bertolino A, Elvevag B, Callicott JH et al (2005) Effect of catechol-O-methyltransferase val158met genotype on attentional control. J Neurosci 25:5038–5045 Botvinick MM, Braver TS, Barch DM, Carter CS, Cohen JD (2001) Conflict monitoring and cognitive control. Psychol Rev 108:624–652 Botvinick MM, Cohen JD, Carter CS (2004) Conflict monitoring and anterior cingulate cortex: an update. Trends Cogn Sci 12:539– 546 Carter CS, Braver TS, Barch DM, Botvinick MM, Noll D et al (1998) Anterior cingulate cortex, error detection and the online monitoring of performance. Science 280:747–749 Chen J, Lipska BK, Halim N, Ma QD, Matsumoto M et al (2004) Functional analysis of genetic variation in catechol-O-methyltransferase (COMT): effects on mRNA, protein, and enzyme activity in postmortem human brain. Amer J Hum Genet 75:807– 821 Egan MF, Goldberg TE, Kolachana BS, Callicott JH, Mazzanti CM et al (2001) Effect of COMT Val108/158 Met genotype on frontal lobe function and risk for schizophrenia. Proc Natl Acad Sci USA 98:6917–6922 Endrass T, Franke C, Kathmann N (2005) Error awareness in a saccade countermanding task. J Psychophysiol 19:275–280 Endrass T, Reuter B, Kathmann N (2007) ERP correlates of conscious error cognition: aware and unaware errors in an antisaccade task. Eur J Neurosci 26:1714–1720 Evdokimidis I, Smyrnis N, Constantinidis TS, Stefanis NC, Avramopoulos D et al (2002) The antisaccade task in a sample of 2,006 young men. I. Normal population characteristics. Exp Brain Res 147:45–52 Everling S, Fischer B (1998) The antisaccade: a review of basic research and clinical studies. Neuropsychologia 36:885–899 Fischer B, Weber H (1992) Characteristics of ‘‘anti’’ saccades in man. Exp Brain Res 89:415–424 Fossella J, Sommer T, Fan J, Wu Y, Swanson JM et al (2002) Assessing the molecular genetics of attention networks. BMC Neurosci 3:14 Hallett PE (1978) Primary and secondary saccades to goals defined by instruction. Exp Brain Res 18:1279–1296 Holroyd CB, Coles MGH (2002) The neural basis of human error processing: reinforcement learning, dopamine and the errorrelated negativity. Psychol Rev 109:679–699 Klein TA, Endrass T, Kathmann N, Neumann J, von Cramon DY et al (2007) Neural correlates of error awareness. Neuroimage 34:1774–1781

463 Kra¨mer UM, Cunillera T, Ca`mara E, Marco-Pallare´s J, Cucurell D et al (2007) The impact of catechol-O-methyltransferase and dopamine D4 receptor genotypes on neurophysiological markers of performance monitoring. J Neurosci 27:14190–14198 Macmillan NA, Creelman CD (2005) Detection theory: a user’s guide, 2nd edn. Lawrence Erlbaum Associates, Mahwah Marco-Pallare´s J, Camara E, Mu¨nte TF, Rodrı´guez-Fornells A (2008) Neural mechanisms underlying adaptive actions after slips. J Cogn Neurosci 20:1595–1610 Massen C (2004) Parallel programming of exogenous and endogenous components in the antisaccade task. Q J Exp Psychol A 57:475–498 Miller EK, Cohen JD (2001) An integrative theory of prefrontal cortex function. Annu Rev Neurosci 24:167–202 Mokler A, Fischer B (1999) The recognition and correction of involuntary prosaccades in an antisaccade task. Exp Brain Res 125:511–516 Munoz DP, Everling S (2004) Look away: the anti-saccade task and the voluntary control of eye movement. Nature 5:218–228 Nieuwenhuis S, Ridderinkhof KR, Blom J, Band GPH, Kok A (2001) Error-related brain potentials are differentially related to awareness of response errors: evidence from an antisaccade task. Psychophysiology 38:752–760 Oak JN, Oldenhof J, Van Tol HHM (2000) The dopamine D4 receptor: one decade of research. Eur J Pharmacol 405:303–327 Okuyama Y, Ishiguro H, Toru M, Arinami T (1999) A genetic polymorphism in the promoter region of DRD4 associated with expression and schizophrenia. Biochem Bioph Res Co 258:292–295 Pierrot-Deseilligny C, Milea D, Muri R (2004) Eye movement control by the cerebral cortex. Curr Opin Neurol 17:17–25 Rabbitt PM (1966) Errors and error-correction in choice-response tasks. J Exp Psychol 71:264–272 Rabbitt PM (2002) Consciousness is slower than you think. Q J Exp Psychol A 55:1081–1092 Riba J, Rodriguez-Fornells A, Monte A, Munte TF, Barbanoj MJ (2005) Noradrenergic stimulation enhances human action monitoring. J Neurosci 25:4370–4374 Ridderinkhof KR, Van den Wildenberg WPM, Segalowitz SJ, Carter CS (2004) Neurocognitive mechanisms of cognitive control: the role of prefrontal cortex in action selection, response, inhibition, performance monitoring and reward-based learning. Brain Cogn 56:129–140 Rubinstein M, Philips TJ, Bunjow JR, Falzone TL, Driewczapolski G et al (1997) Mice lacking dopamine D4 receptors are supersensitive to ethanol, cocaine and methamphetamine. Cell 90:991– 1001 Schall JD, Stuphorn V, Brown JW (2002) Monitoring and control of action by the frontal lobes. Neuron 36:309–322 Smyrnis N, Evdokimidis I, Stefanis NC, Constantinidis TS, Avramopoulos D et al (2002) The antisaccade task in a sample of 2, 006 young males. II. Effects of task parameters. Exp Brain Res 147:53–63 Smyrnis N, Evdokimidis I, Stefanis NC, Avramopoulos D, Constantinidis TS et al (2003) Antisaccade performance of 1,273 men: effects of schizotypy, anxiety, and depression. J Abnorm Psychol 112:403–414 Snedecor GW, Cochran WG (1980) Statistical methods, 7th edn. The Iowa State University Press, Ames Van Veen V, Carter CS (2002) The anterior cingulate as a conflict monitor: fMRI and ERP studies. Physiol Behav 77:477–482 Velanova K, Wheeler ME, Luna B (2008) Maturational changes in anterior cingulate and frontoparietal recruitment support the development of error processing and inhibitory control. Cereb Cortex 18:2505–2522

123