Electrophysiological Predictors of Single-Trial Response ...scott/pdf/Burwell_SPR2013poster...Figure...
1
Figure 3. Cm-cluster ICs (A) and centroid (B) Electrophysiological Predictors of Single-Trial Response Accuracy Located in Superior Frontal Cortex Scott J. Burwell1, Scott Makeig2, William G. Iacono1, & Stephen M. Malone1 University of Minnesota - Twin Cities1, University of California - San Diego2 Background Superior medial frontal (e.g., pre-supplementary motor area or pre-SMA) neural circuitry is activated in both response-selection and response-inhibition tasks and is thought to guide the appropriateness of one’s behavior (Mostofsky & Simmonds, 2008) Excitation of medial frontal cortical neurons prior to the onset of a voluntary motor response (rather than after) on a given trial has been posited as a mechanism for resolving response-conflict and choosing appropriate behavior (Isoda & Hikosaka, 2007). To date, the vast majority of research surrounding action-monitoring (the self-awareness and guiding of one’s performance) has focused on “reactive” brain responses that follow an already-executed erroneous motor response such as the error-related negativity (ERN or Ne) and positivity (Pe). Increasing interest is being given to processes that precede responses (”proactive”) during cognitive challenge and such processes have implications in the etiology of behaviorally disinhibited disorders such as attention-deficit hyperactivity disorder (ADHD) and substance dependence (Mostofsky & Simmonds, 2008; Isoda & Hikosaka, 2007). EEG is well suited to distinguish between proactive (response-preceding) and reactive (response-following) brain processes because of its temporal precision. Moreover, using advanced blind source separation techniques such as independent components analysis can aid in identifying the time-courses and spatial locations of the neural generators of EEG. The aim of the current study was to identify sources that contribute to the execution of task-appropriate (correct performance) and task-inappropriate (errors of commission) button-presses during a cognitive flanker task. Moreover, we sought to understand if such processes were related to individual differences in ADHD symptom counts. Results & Discussion References Akalin Acar, Z. & Makeig, S. (2010). Neuroelectromagnetic forward head modeling toolbox. J Neurosci Methods, 190, 258 - 270. Delorme, A., Palmer, J., Onton, J., Oostenveld, R. & Makeig, S. (2012). Independent EEG sources are dipolar. PLoS ONE, 7(2), e30135. Delorme, A. & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics. J Neurosci Methods, 134, 9 - 21. Isoda, M. & Hikosaka, O. (2007). Switching from automatic to controlled action by monkey medial frontal cortex. Nat Neurosci, 10(2), 240 - 248. McLoughlin, G., Palmer, J.A., Rijsdijk, F. & Makeig, S. (in press). Genetic overlap between evoked frontocentral theta-band phase variability, reaction time variability, and attention-deficity/hyperactivity disorder symptoms in a twin study. Biological Psychiatry. Mostofsky, S.H. & Simmonds, D.J. (2008). Response inhibition and response selection: two sides of the same coin. J Cognitive Neurosci, 20(5), 751 - 761. Method Results Amplitude on all three of the Cm measures (PNe, Ne, and Pe) was predictive of trial-level accuracy on the flanker task (Table 1). Specifically, enhancement of the response-preceding PNe was associated with about 30% increase in the odds of appropriate responding for a given trial. Conversely, enhancements of the response-following Ne and Pe were indicative of decreased odds for correct performance. This tendencey for incorrectly-performed motor responses to occur before the peak Cm amplitude and correctly-performed motor responses to occur after peak Cm amplitude is borne out in Figure 4. When entered into regression as predictors of ADHD symptoms, only the response-preceding PNe activation from the Cm cluster was associated with ADHD; for both hyperactive and inattentive symptom types, decreased PNe was indicative of more symptoms. This finding is illustrated in Figure 5. Discussion These results support the notion that neurocognitive processes leading up to an individual’s behavior are important for the appropriateness of that behavior. The fact that symptom counts for ADHD were inversely related to the strength of the response-preceding PNe component suggests that adolescents with ADHD may have deficiencies in neurocircuitry critical for appropriate behavioral constraint and response selection. Without sufficient arousal of brain regions that undergird the production of context-appropriate behavior, the ability for one to adhere to social and occupational norms may be compromised. The association between ADHD-prone individuals and weaker response-preceding Cm activation shown here is in line with recent research illustrating that inter-trial Cm theta activations and motor response times are more variable in those with the disorder (McLoughlin et al., in press). Such results may have implications for how childhood disruptive disorders such as ADHD are assessed and treated. Considerations & future directions Because of similarities in spatial location and the low spatial resolution of EEG, it is possible that the Cm cluster of ICs reported here has similar neuroanatomical origin as frontal-midline neurocircuitry commonly investigated as it pertains to the reactive detection of one’s errors (Ne or ERN, generated by anterior cingulate cortex). However, such circuitry typically lies more anterior than the location of the Cm cluster studied here (proximal to pre-SMA) and is rarely discussed as it pertains to proactive mechanisms. By incorporating more subject-specific BEM head models, we hope to better distinguish between closely situated neural networks that contribute to proactive and reactive action monitoring processes. Alternatives to the clustering technique used here as well as causal modeling approaches for accuracy prediction analyses are being pursued. Acknowledgements We would like to thank staff and participants at the Minnesota Center for Twin and Family Research for making this research possible. This project was funded by AA 017314 and the SPR Research Fellowship Travel Award. Contact Information S. Burwell: [email protected] Attn: MCTFR N218 Elliott Hall (Psychology Department) 75 East River Road Minneapolis, MN 55455-0344 Sample: adolescent monozygotic twins (N = 48 pairs, age range 14 - 16) drawn from the Adbrain study at the Minnesota Center for Twin and Family Research. EEG task/data: Sixty-one scalp electrodes positioned at standard 10/20 locations recorded EEG continuously throughout completion of an Eriksen flanker task. On each trial, subjects view a five-letter array (e.g., SSSSS, HHHHH, SSHSS, HHSHH) and are instructed to respond with either a left or right button-press to indicate the target (center) stimulus (see Figure 1). EEGs were decomposed into independent components (ICs) using an adaptive mixture algorithm (Delorme et al., 2012) to recover putative source generators. Equivalent current dipoles were fit to realistic boundary element (BEM) head models derived from anatomical MRIs. At the time of dipole-fitting, subject-specific forward BEM models for 23% of the current sample had already been computed using a recently-developed MATLAB software (see Figure 2; Akalin Acar & Makeig, 2010); therefore, only this subset of the presented source-localized data is based on subject-specific MRIs. The remaining subjects’ dipoles were fit using the standard head model from the Montreal Neurological Institute. Only ICs that had less than 15% residual variance between dipole projection and IC scalp topography were considered for the current study. Locatized dipoles were input to a k-means clustering algorithm to identify ICs across the entire sample that may have been generated in comparable regions of cortex. A prominent central-midline cluster (out of several others, hereinafter referred to as “Cm”) was the focus of the present study. Dipoles as part of the Cm cluster are plotted in Figure 3A and the cluster centroid in 3B. Trial-level IC-activations were epoched -2 to 2 seconds relative to the onset of the button-press. For the sake of computational efficiency, only trials containing an errorneous button-press (N), the preceding correct button-press (N-1), or the succeeding correct button-press (N+1) were analyzed. ICs for the Cm-cluster were back-projected to electrode Cz and trial-level voltage data were normalized to the average power within a pre-response baseline (-700 to -200 milliseconds). Trial measures were derived with a peak-picking algorithm as the most positive peak in the window preceding the response (PNe or pre-Ne; -200 to 0 ms), peri-response (Ne; -100 to 100 ms), and post-response (Pe; 0 to 200 ms). Regressions were carried out using generalized estimating equations (GEEs), modeling within-family correlations as independent. Logit regressions were used to understand whether changes in the three Cm-measures (PNe, Ne, Pe) affected performance (0 = error, 1 = correct; controlling for reaction time). Odds ratios were computed by exponentiating the product between the standard deviation of the predictor and the raw regression coefficient. To investigate whether activations by the Cm-cluster had any predictive value of ADHD, we regressed (GEE Poisson regression) DSM-IV-TR ADHD hyperactive and inattentive symptoms collected from subjects based on each twin’s mean score for PNe, Ne, and Pe, seperately. 1.2 0.8 0.4 0 -0.4 -0.8 -1.2 Grand-Average ERPs -1000 -800 -600 -400 -200 0 200 400 600 time [ms] (button-press) Correct Incorrect amplitude Figure 2. Scalp, skull, cerebrospinal fluid (CSF), and brain mesh layers for boundary element head models for a pair of monozygotic twins. Scalp Skull CSF Brain HHHHH SSSSS HHSHH (left-button “H”) (right-button “S”) (right-button “S”) trials Figure 1. Eriksen flanker task B A ERPIMAGE: trials sorted by stimulus-onset 5000 4000 3000 2000 1000 trials ERPIMAGE: split by accuracy 5000 4000 3000 2000 1000 -1000 -800 -600 -400 -200 0 200 400 trials PNe Ne Pe 15.78 27.44 -24.77 (2.61) (3.34) (3.70) 1.28* 1.57* 0.66* Predictor B (S.E.) Odds Ratio Logistic regression coefficients predicting trial-level accuracy with response-preceding positive (PNe; -200 to 0 ms) and response-following negative (Ne; -100 to 100 ms) and positive (Pe; 0 to 200 ms) predictors, adjusted for stimulus-onset timing. *p < .001 Table 1. Trial-level accuracy predicted by Cm-activations preceding and following the button press. Figure 4 (right). Grand-average ERPs and ERPIMAGEs for correctly- and incorrectly-executed trials, locked to response-onset. Solid vertical lines on the left and right of the top panel indicate mean stimulus-onset time and mean accuracy, respectively. The dotted vertical lines along the right edge of ERPIMAGE plots signify trials where performance was substantially worse (left) or better (right) than would be expected (p < .01, per binomial test). -1000 -800 -600 -400 -200 0 200 400 600 time [ms] 0.2 0 0.6 -0.2 0.4 amplitude ADHD Control Figure 5. Attention-deficit hyperactivity disorder (ADHD) and control groups designated by lower- and upper-quartile symptom counts. On trials that were executed correctly, weaker pre-response positivity (i.e., PNe in Table 1) was indicative of ADHD hyperactive (odds ratio = 0.68, p = .003) and inattentive (odds ratio = -0.72, p = .013). This reduction in pre-response activation associated with ADHD symptoms was not evident on incorrect trials nor were other (Ne and Pe) components reduced on correct trials (p’s > .2). * Cm-cluster pre-response activation on correctly-performed trials is weakened in adolescents with high ADHD symptoms Motor response-preceding frontal cortical activations predict trial accuracy, deficient in ADHD time [ms] Correct Incorrect
Transcript of Electrophysiological Predictors of Single-Trial Response ...scott/pdf/Burwell_SPR2013poster...Figure...
-
Figure 3. Cm-cluster ICs (A) and centroid (B)
Electrophysiological Predictors of Single-Trial Response AccuracyLocated in Superior Frontal Cortex
Scott J. Burwell1, Scott Makeig2, William G. Iacono1, & Stephen M. Malone1University of Minnesota - Twin Cities1, University of California - San Diego2
BackgroundSuperior medial frontal (e.g., pre-supplementary motor area or pre-SMA) neural
circuitry is activated in both response-selection and response-inhibition tasks and is thought to guide the appropriateness of one’s behavior (Mostofsky & Simmonds, 2008)
Excitation of medial frontal cortical neurons prior to the onset of a voluntary motor response (rather than after) on a given trial has been posited as a mechanism for resolving response-con�ict and choosing appropriate behavior (Isoda & Hikosaka, 2007).
To date, the vast majority of research surrounding action-monitoring (the self-awareness and guiding of one’s performance) has focused on “reactive” brain responses that follow an already-executed erroneous motor response such as the error-related negativity (ERN or Ne) and positivity (Pe).
Increasing interest is being given to processes that precede responses (”proactive”) during cognitive challenge and such processes have implications in the etiology of behaviorally disinhibited disorders such as attention-de�cit hyperactivity disorder (ADHD) and substance dependence (Mostofsky & Simmonds, 2008; Isoda & Hikosaka, 2007).
EEG is well suited to distinguish between proactive (response-preceding) and reactive (response-following) brain processes because of its temporal precision. Moreover, using advanced blind source separation techniques such as independent components analysis can aid in identifying the time-courses and spatial locations of the neural generators of EEG.
The aim of the current study was to identify sources that contribute to the execution of task-appropriate (correct performance) and task-inappropriate (errors of commission) button-presses during a cognitive �anker task. Moreover, we sought to understand if such processes were related to individual di�erences in ADHD symptom counts.
Results & Discussion
ReferencesAkalin Acar, Z. & Makeig, S. (2010). Neuroelectromagnetic forward head modeling toolbox. J
Neurosci Methods, 190, 258 - 270.
Delorme, A., Palmer, J., Onton, J., Oostenveld, R. & Makeig, S. (2012). Independent EEG sources are dipolar. PLoS ONE, 7(2), e30135.
Delorme, A. & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics. J Neurosci Methods, 134, 9 - 21.
Isoda, M. & Hikosaka, O. (2007). Switching from automatic to controlled action by monkey medial frontal cortex. Nat Neurosci, 10(2), 240 - 248.
McLoughlin, G., Palmer, J.A., Rijsdijk, F. & Makeig, S. (in press). Genetic overlap between evoked frontocentral theta-band phase variability, reaction time variability, and attention-de�city/hyperactivity disorder symptoms in a twin study. Biological Psychiatry.
Mostofsky, S.H. & Simmonds, D.J. (2008). Response inhibition and response selection: two sides of the same coin. J Cognitive Neurosci, 20(5), 751 - 761.
Method
Results
Amplitude on all three of the Cm measures (PNe, Ne, and Pe) was predictive of trial-level accuracy on the �anker task (Table 1). Speci�cally, enhancement of the response-preceding PNe was associated with about 30% increase in the odds of appropriate responding for a given trial. Conversely, enhancements of the response-following Ne and Pe were indicative of decreased odds for correct performance. This tendencey for incorrectly-performed motor responses to occur before the peak Cm amplitude and correctly-performed motor responses to occur after peak Cm amplitude is borne out in Figure 4.
When entered into regression as predictors of ADHD symptoms, only the response-preceding PNe activation from the Cm cluster was associated with ADHD; for both hyperactive and inattentive symptom types, decreased PNe was indicative of more symptoms. This �nding is illustrated in Figure 5.
Discussion
These results support the notion that neurocognitive processes leading up to an individual’s behavior are important for the appropriateness of that behavior.
The fact that symptom counts for ADHD were inversely related to the strength of the response-preceding PNe component suggests that adolescents with ADHD may have de�ciencies in neurocircuitry critical for appropriate behavioral constraint and response selection. Without su�cient arousal of brain regions that undergird the production of context-appropriate behavior, the ability for one to adhere to social and occupational norms may be compromised. The association between ADHD-prone individuals and weaker response-preceding Cm activation shown here is in line with recent research illustrating that inter-trial Cm theta activations and motor response times are more variable in those with the disorder (McLoughlin et al., in press). Such results may have implications for how childhood disruptive disorders such as ADHD are assessed and treated.
Considerations & future directions
Because of similarities in spatial location and the low spatial resolution of EEG, it is possible that the Cm cluster of ICs reported here has similar neuroanatomical origin as frontal-midline neurocircuitry commonly investigated as it pertains to the reactive detection of one’s errors (Ne or ERN, generated by anterior cingulate cortex). However, such circuitry typically lies more anterior than the location of the Cm cluster studied here (proximal to pre-SMA) and is rarely discussed as it pertains to proactive mechanisms. By incorporating more subject-speci�c BEM head models, we hope to better distinguish between closely situated neural networks that contribute to proactive and reactive action monitoring processes.
Alternatives to the clustering technique used here as well as causal modeling approaches for accuracy prediction analyses are being pursued.
Acknowledgements
We would like to thank sta� and participants at the Minnesota Center for Twin and Family Research for making this research possible.
This project was funded by AA 017314 and the SPR Research Fellowship Travel Award.
Contact Information
S. Burwell: [email protected]
Attn: MCTFRN218 Elliott Hall (Psychology Department)75 East River RoadMinneapolis, MN 55455-0344
Sample: adolescent monozygotic twins (N = 48 pairs, age range 14 - 16) drawn from the Adbrain study at the Minnesota Center for Twin and Family Research.
EEG task/data: Sixty-one scalp electrodes positioned at standard 10/20 locations recorded EEG continuously throughout completion of an Eriksen �anker task. On each trial, subjects view a �ve-letter array (e.g., SSSSS, HHHHH, SSHSS, HHSHH) and are instructed to respond with either a left or right button-press to indicate the target (center) stimulus (see Figure 1).
EEGs were decomposed into independent components (ICs) using an adaptive mixture algorithm (Delorme et al., 2012) to recover putative source generators.
Equivalent current dipoles were �t to realistic boundary element (BEM) head models derived from anatomical MRIs. At the time of dipole-�tting, subject-speci�c forward BEM models for 23% of the current sample had already been computed using a recently-developed MATLAB software (see Figure 2; Akalin Acar & Makeig, 2010); therefore, only this subset of the presented source-localized data is based on subject-speci�c MRIs. The remaining subjects’ dipoles were �t using the standard head model from the Montreal Neurological Institute.
Only ICs that had less than 15% residual variance between dipole projection and IC scalp topography were considered for the current study. Locatized dipoles were input to a k-means clustering algorithm to identify ICs across the entire sample that may have been generated in comparable regions of cortex. A prominent central-midline cluster (out of several others, hereinafter referred to as “Cm”) was the focus of the present study. Dipoles as part of the Cm cluster are plotted in Figure 3A and the cluster centroid in 3B.
Trial-level IC-activations were epoched -2 to 2 seconds relative to the onset of the button-press. For the sake of computational e�ciency, only trials containing an errorneous button-press (N), the preceding correct button-press (N-1), or the succeeding correct button-press (N+1) were analyzed.
ICs for the Cm-cluster were back-projected to electrode Cz and trial-level voltage data were normalized to the average power within a pre-response baseline (-700 to -200 milliseconds).
Trial measures were derived with a peak-picking algorithm as the most positive peak in the window preceding the response (PNe or pre-Ne; -200 to 0 ms), peri-response (Ne; -100 to 100 ms), and post-response (Pe; 0 to 200 ms).
Regressions were carried out using generalized estimating equations (GEEs), modeling within-family correlations as independent. Logit regressions were used to understand whether changes in the three Cm-measures (PNe, Ne, Pe) a�ected performance (0 = error, 1 = correct; controlling for reaction time). Odds ratios were computed by exponentiating the product between the standard deviation of the predictor and the raw regression coe�cient.
To investigate whether activations by the Cm-cluster had any predictive value of ADHD, we regressed (GEE Poisson regression) DSM-IV-TR ADHD hyperactive and inattentive symptoms collected from subjects based on each twin’s mean score for PNe, Ne, and Pe, seperately.
1.2
0.8
0.4
0
-0.4
-0.8
-1.2
Grand-Average ERPs
-1000 -800 -600 -400 -200 0 200 400 600
time [ms]
(button-press)CorrectIncorrect
amp
litu
de
Figure 2. Scalp, skull, cerebrospinal �uid (CSF), and brain mesh layers for boundary element head models for a pair of monozygotic twins.
Scalp Skull CSF Brain
HHHHH
SSSSS
HHSHH
(left-button “H”)
(right-button “S”)
(right-button “S”)
trials
Figure 1. Eriksen �anker task
B
A
ERPIMAGE: trials sorted by stimulus-onset
5000
4000
3000
2000
1000
tria
ls
ERPIMAGE: split by accuracy
5000
4000
3000
2000
1000
-1000 -800 -600 -400 -200 0 200 400
tria
ls
PNeNePe
15.7827.44
-24.77
(2.61)(3.34)(3.70)
1.28*1.57*0.66*
Predictor B (S.E.) Odds Ratio
Logistic regression coe�cients predicting trial-level accuracy with response-preceding positive (PNe; -200 to 0 ms) and response-following negative (Ne; -100 to 100 ms) and positive (Pe; 0 to 200 ms) predictors, adjusted for stimulus-onset timing. *p < .001
Table 1. Trial-level accuracy predicted by Cm-activations preceding and following the button press.
Figure 4 (right). Grand-average ERPs and ERPIMAGEs for correctly- and incorrectly-executed trials, locked to response-onset. Solid vertical lines on the left and right of the top panel indicate mean stimulus-onset time and mean accuracy, respectively. The dotted vertical lines along the right edge of ERPIMAGE plots signify trials where performance was substantially worse (left) or better (right)than would be expected (p < .01, per binomial test).
-1000 -800 -600 -400 -200 0 200 400 600
time [ms]
0.2
0
0.6
-0.2
0.4
amp
litu
de
ADHDControl
Figure 5. Attention-de�cit hyperactivity disorder (ADHD) and control groups designated by lower- and upper-quartile symptom counts. On trials that were executed correctly, weaker pre-response positivity (i.e., PNe in Table 1) was indicative of ADHD hyperactive (odds ratio = 0.68, p = .003) and inattentive (odds ratio = -0.72, p = .013). This reduction in pre-response activation associated with ADHD symptoms was not evident on incorrect trials nor were other (Ne and Pe) components reduced on correct trials (p’s > .2).
*
Cm-cluster pre-response activation on correctly-performed trials is weakened in
adolescents with high ADHD symptoms
Motor response-preceding frontal cortical activations predict trial accuracy, de�cient in ADHD
time [ms]
Correct
Incorrect
