INTRODUCTION
Survivors of traumatic brain injury (TBI) exhibit severity-dependent impairments in a number of cognitive domains, including those that compromise the accuracy of action and performance monitoring. Impairments of this type belong to a broader constellation of deficits in cognitive control—the ability to orchestrate thought and action in accord with internal goals (Levine et al., 2002; Miller & Cohen, 2001). Cognitive control consists of evaluative and regulative component processes (Botvinick et al., 2001; Braver et al., 1999; Kerns et al., 2004; MacDonald et al., 2000). Evaluative component processes include monitoring for the presence of response conflict (simultaneously activated competing responses), monitoring performance for errors, and signaling the need to implement or adjust top-down control processes. These evaluative functions are critical for flexible adjustments of top-down control needed for adaptation to performance demands (e.g., correcting an error). The anterior cingulate cortex (ACC) plays a key role in the evaluative component of cognitive control (Gehring & Fencsik, 2001; Kerns et al., 2004; MacDonald et al., 2000; Miltner et al., 2003; van Veen & Carter, 2002a, 2002b). The regulative component, in contrast, is involved in the actual implementation of top-down control for task-relevant processes, allowing them to compete effectively against task-irrelevant processes. The dorsolateral prefrontal cortex (dlPFC) supports regulative control processes (Kerns et al., 2004; MacDonald et al., 2000; Perlstein et al., 2003). Thus, cognitive control has been conceptualized as a dynamic process implemented in a distributed cortical network that involves closely interacting but dissociable components: ACC-mediated evaluative processes indicate when control needs to be more strongly engaged by signaling the dlPFC, which, in turn, provides top-down support of task-appropriate behaviors in the implementation of this control.
Dysfunction of ACC-mediated evaluative control processes has direct implications for individuals who have suffered brain injuries. For example, animal research demonstrates ACC lesions alter the normal pattern of corrective behavior following errors, such that consecutive errors without appropriate correction are more common (Dias & Aggleton, 2000; Rushworth et al., 2003; Walton et al., 2003). Similar evidence comes from a human patient with a rare focal lesion of the rostral-to-middorsal ACC who was less likely than healthy controls to correct mistakes (Swick & Turken, 2002). In participants with TBI, where axonal shearing may be prominent in medial frontal regions, little research has examined the neural instantiation of evaluative control functions. One PET study suggested abnormalities of ACC glucose metabolism at rest in participants with TBI that correlated with subsequent neuropsychological performance (Fontaine et al., 1996), two fMRI studies using conflict-laden tasks (e.g., Stroop) found altered ACC activity in participants with TBI compared with controls (Scheibel et al., 2007; Soeda et al., 2005), and recent studies from our lab observed ACC dysfunction in participants with severe TBI during performance of a task requiring working memory (Perlstein et al., 2004), and diminished electrophysiological reflections of evaluative control, presumably mediated by the ACC (i.e., N450 and feedback-related negativity [FRN] components of the scalp-recorded event-related potential [ERP]) (Larson et al., 2007; Perlstein et al., 2006). A growing consensus from these studies is that ACC-mediated changes following TBI are the result of diffuse axonal damage that disturbs fronto-cortical and subcortical networks leading to subsequent evaluative control impairment (e.g., reduced post-error slowing or increased Stroop interference effects; see Larson et al., 2006a, and Seignourel et al., 2005).
The physiological and cognitive bases of evaluative control have been the subjects of many recent investigations. One putative reflection of the evaluative process of performance monitoring is an electrophysiological signature in the scalp recorded ERP known as the error-negativity (Ne) or error-related negativity (ERN; heretofore referred to as the Ne/ERN). The Ne/ERN is a fronto-medial maximal response-locked potential peaking within 100 ms after the commission of an error (Falkenstein et al., 1991). The precise cognitive mechanisms generating the Ne/ERN are under active debate (Holroyd & Coles, 2002; Yeung & Cohen, 2006), but have been attributed to detection of response conflict (Carter et al., 1998), detection of errors (i.e., mismatch between an intended and produced response; Falkenstein et al., 1991; Gehring et al., 1993), or an emotional context response to errors (Larson et al., 2006b; Vidal et al., 2000). ERP source localization and fMRI studies consistently implicate a region in the ACC as the primary neural generator of the Ne/ERN (van Veen & Carter, 2002a).
Although the Ne/ERN has received considerable attention in electrophysiological investigations of performance monitoring, researchers also examined a positive deflection occurring between 100 and 400 ms following the Ne/ERN known as the error positivity (Pe). The functional significance of the Pe remains controversial (Overbeek et al., 2005); however, the Pe can be differentiated from the P300 (Falkenstein et al., 2000) and may be associated with conscious error recognition, as it is diminished when subjects are unaware of on-line performance errors (Nieuwenhuis et al., 2001). Moreover, post-error RT slowing occurs only on trials that consist of an observable Pe (Mathalon et al., 2002), and Pe amplitude varies in relation to the degree of post-error slowing and autonomic nervous system activity (Hajcak et al., 2003a). Source localization studies of the Pe remain largely inconclusive, but suggest possible generators in either the caudal (Herrmann et al., 2004) or rostral portions of the ACC (Bush et al., 2000; see Overbeek et al., 2005 for review).
Performance/Error Monitoring After Brain Injury
One important aspect of evaluative control is the ability to monitor performance and implement strategic adjustments when current performance is inadequate. Stemmer et al., (2004) examined overt behavioral signs of error responses (e.g., exclamations, swearing, grimaces) during a flanker task and found that three of five stroke patients who experienced anterior communicating artery (ACA) aneurysms and subsequent lesions to the medial PFC, including the ACC, demonstrated poor ability to monitor performance. This study utilized ERPs to examine error-related neural activity; findings indicate decreased error-related neural activity in the patients with lesions to the medial PFC. Of note, some patients were aware of errors but did not produce a discernable Ne/ERN. Another study examined performance monitoring in everyday situations, such as wrapping a gift or packing a schoolbag. Participants with TBI showed poorer performance monitoring skills and corrected significantly fewer errors than control participants (Hart et al., 1998). O'Keeffe et al. (2004) utilized measures of electrodermal activity to examine autonomic responses to errors in participants with TBI. Participants with TBI detected significantly fewer errors on the task than demographically matched-control counterparts. In addition, electrodermal activity following errors was decreased in the participants with TBI relative to controls, with error detection rates and electrodermal activity being significantly correlated. Thus, participants with TBI exhibit impairments in the evaluative processes of conflict detection and resolution, leading to impaired processing of error-related conflict information and poor adjustments in performance (O'Keeffe et al., 2004). No studies to date, however, have examined the electrophysiological instantiations of performance monitoring deficits in participants with TBI. Yet, given the key role ACC-mediated conflict-detection/performance monitoring processes appear to play in signaling for the recruitment of regulative control mechanisms, it is important to characterize the functioning of such a process in survivors of TBI.
Current Study
The primary aims of the current study were, therefore, to extend previous findings of impaired performance monitoring in people with severe TBI and determine if electrophysiological indices of performance monitoring—the Ne/ERN and Pe components of the ERP, respectively—are attenuated in TBI. We predicted that patients with TBI would show smaller-amplitude electrophysiological activity (Ne/ERN; Pe) to error, relative to correct trials, compared to healthy controls.
METHODS
Participants
Participants with severe TBI were recruited from two Northern Florida trauma and rehabilitation hospitals; control participants were recruited via flyer and advertisement from the local community. Study enrollment initially included 21 participants with severe TBI and 21 healthy control participants. ERP data for one participant with TBI were lost because of equipment malfunction and one TBI participant performed the task incorrectly (i.e., responded to the word rather than the color for every trial); therefore, final analyses included 19 participants with severe TBI and 21 healthy control participants. All participants provided written informed consent according to procedures established by the University of Florida Health Science Center Institutional Review Board and were compensated for their participation.
TBI severity was determined from medical record review of lowest post-resuscitation Glasgow Coma Scale (GCS) score (Teasdale & Jennett, 1974), with severe TBI defined as a GCS score <9. Neurological indices including neuroradiological findings taken from acute computerized tomography (CT) scans, duration of loss of consciousness (LOC), and duration of post-traumatic amnesia (PTA) were also acquired from medical record review or, when LOC and PTA information were not available in medical records, from structured participant and significant other interview (King et al., 1997; McMillan et al., 1996). LOC and PTA data confirmed all patients met criteria for severe TBI as traditionally defined by LOC>6 hours and/or PTA>7 days (Bigler, 1990; Bond, 1986; Lezak et al., 2004).
Potential participants were excluded from the study if they had a history of psychotic or bipolar disorder, learning disability, alcohol or substance abuse, other acquired brain disorders (e.g., epilepsy, stroke), inpatient psychiatric treatment predating brain injury, clinically-significant depression or anxiety within two years prior to injury, current anti-epileptic medication use, or color-blindness as measured by the Ishihara pseudo-isochromatic color plates (Clark, 1924). Patients with language comprehension deficits or uncorrected visual impairments were also excluded.
Demographic characteristics and neuropsychological test summary data for control and TBI study participants are provided in Table 1. Injury characteristics (etiology, GCS, LOC, PTA, time since injury) and neuroradiological findings for the participants with TBI are presented in Table 2. Participants with TBI were at least four months post-injury, with the exception of one patient (two months post-injury) who was functioning well and desired to complete the study early before returning to employment responsibilities. No participants were engaged in legal action at the time of the study. Participant groups were comparable in age and education (see Table 1); groups did not significantly differ in gender distribution χ2(1) = 2.16, p > .14 (TBI: 15 male/4 female; Control: 12 male/9 female). Since previous studies demonstrate differences in Ne/ERN amplitude as a function of depressive or anxious symptoms (Hajcak et al., 2003b; Ruchsow et al., 2004; Ruchsow et al., 2006), TBI and control participants were administered the Beck Depression Inventory-2nd Edition (BDI-II; Beck, 1996) and the State-Trait Anxiety Inventory (STAI; Speilberger et al., 1983). Compared to controls, participants with TBI endorsed significantly more depressive symptoms; however, no individual scores met common clinical cut-offs for depression (BDI-II>21) and mean scores for both groups were within normal limits—not meeting criteria for minimal depression (BDI-II>13; see Beck, 1996). Participants with TBI also endorsed higher levels of state and trait anxiety symptoms.
Demographic and mean summary data for severe TBI and control groups
Injury characteristics and neuroradiological information for TBI patients (N = 19)
Assessment of Functioning
In an effort to characterize the cognitive functioning of participants with TBI, a battery of neuropsychological tests was administered to all participants. As presented in Table 1, and in-line with typical impairments seen after severe TBI, participants with severe TBI performed significantly worse than controls on tests broadly assessing attention (Digit Span Forward from the Wechsler Adult Intelligence Scale—3rd edition [WAIS-III]; Wechsler, 1997), processing speed (Trail Making Test Part A; Reitan, 1958), verbal fluency (Controlled Oral Word Association Test [COWAT] and Category Fluency; Benton & Hamsher, 1976), executive functioning (Trail Making Test Part B; Reitan, 1958/Digit Span Backward from the WAIS-III; Wechsler, 1997), and delayed verbal memory (Hopkins Verbal Learning Test–Revised [HVLT-R] long delay; Brandt & Benedict, 2001/Wechsler Memory Scale—Revised [WMS-R] Logical Memory II; Wechsler, 1987). In contrast, participants with TBI did not differ from controls on the initial encoding/immediate recall of verbal memory information (HVLT-R immediate recall/WMS-R Logical Memory I).
Similarly, the Frontal Systems Behavior Scale (FrSBe; Grace & Malloy, 2001) was used to assess self- and significant other-reported clinical symptomatology. The FrSBe is a 46-item behavior rating scale originally designed to measure behavioral change associated with frontal lobe injury. Each item is rated on a one to five point Likert-type scale, with one indicating “almost never” and five “almost always.” Thus, higher scores indicate more TBI-related symptoms. The FrSBe gathers information regarding behaviors from the patient (self-report) and a significant other and includes an overall composite score and three subscales that assess apathy, disinhibition, and executive function. Significant others who completed ratings in the current study were the primary caregivers of the participants with TBI and included 8 spouse/fiancée, 7 parents, 2 siblings, 1 grandparent, and 1 aunt. As can be seen in Table 1, participants with TBI endorsed higher levels of behavioral symptoms than control participants and endorsed increased levels of apathy and disinhibition. Of note, TBI participants did not endorse more difficulties with executive functioning than control counterparts.
Experimental Task
Participants performed a modified color-naming version of the single-trial Stroop task used by Kerns et al., (2004). In this task, participants are presented with one of three words (RED, GREEN, BLUE) printed in one of the same three colors. Congruent trials comprised words presented in their same color of ink (e.g., the word BLUE printed in blue ink); incongruent trials comprised color-words printed in a different color of ink (e.g., the word BLUE printed in red ink). Participants were instructed to respond as quickly and accurately as possible to the color of the word (while ignoring the word itself) with a button press to one of three color-coded response keys using the index, middle, and ring fingers of their right hand. Color-to-key mapping was practiced prior to task performance using 40 presentations of each color-key combination. Stroop trials were three seconds in duration and consisted of a Stroop color-word presented for 1.5 s followed by a 1.5 s-duration fixation cross to allow electrophysiological activity to return to baseline. Six blocks of 100 trials, for a total of 600 trials and approximately 30 minutes in EEG testing were presented. To increase the potency of the conflict stimulus, 70% of trials were congruent and 30% were incongruent.
Electrophysiological Data Recording, Reduction, and Measurement
EEG was recorded from 64 scalp sites using a geodesic sensor net and Electrical Geodesics, Inc., (EGI; Eugene, Oregon) amplifier system (20K gain, nominal bandpass = .10–100 Hz). Electrode placements enabled recording vertical and horizontal eye movements reflecting electro-oculographic (EOG) activity. EEG was initially referenced to Cz and digitized continuously at 250 Hz with a 16-bit analog-to-digital converter. A right posterior electrode approximately two inches behind the right mastoid served as common ground. Electrode impedance was maintained below 50 kΩ. EEG was segmented off-line and single trial epochs were rejected if voltages exceeded 100 μV, transitional (sample-to-sample) thresholds were greater than 100 μV, or eye-channel amplitudes were above 70 μV. EEG was digitally re-referenced to an average reference (Bertrand et al., 1985) in order to yield a reference-free representation of electrophysiological activity and to reconstruct the EEG at the Cz reference, then digitally low-pass filtered at 15 Hz.
Individual-subject response-locked averages were derived separately for correct and incorrect trials, collapsed across congruency to afford adequate signal-to-noise ratio, spanning 200ms prior to and 500 ms following response and baseline corrected using the 200 ms pre-response window. ERP trials containing errors of omission were excluded from averages. Electrode locations utilized were based on previous findings that the Ne/ERN and Pe are relatively focal over fronto-medial locations and centro-parietal areas, respectively (Falkenstein et al., 2000; Gehring et al., 1993), as well as the scalp-distribution maps of the present data. To ensure accurate characterization of Ne/ERN amplitude and prevent spurious findings as a result of potential group-wise latency differences, Ne/ERN amplitudes as well as the corresponding correct-trial amplitudes were extracted as the average of 15 ms pre- to 15 ms post-peak negative amplitude between 0 ms and 100 ms and averaged across four fronto-central electrode sites—4 (FCz), 65 (Cz), 5, and 55 (both sites anterior and slightly lateral to Cz—see Perlstein et al., 2006 for recording site figure). Given previous findings that the Pe is found at centroparietal electrode locations and is less punctate and more tonic than the Ne/ERN (Overbeek et al., 2005), Pe amplitude was measured as the averaged activity from 200 ms–400 ms of five centro-parietal electrode sites (65 [Cz], 18, 43, 30, and 34 [Pz]). Latency measurements for the Ne/ERN component were indexed as the peak negative-going amplitude.
Data Analysis
Median correct-trial RT (Ratcliff, 1993), arcsine transformed error rates (Neter et al., 1985), and ERP component amplitude and latency data were analyzed using separate repeated-measures analyses of variance (ANOVAs). The Huynh-Feldt epsilon adjustment was applied for ANOVAs with more than two levels of a within-subject factor and partial-eta2 (η2) reported as a measure of effect size. ANOVAs for RTs and error rates included the factors group (TBI, control) and congruency (congruent, incongruent), whereas ANOVAs for error-related ERP activity included the factors group and accuracy (correct, incorrect trial amplitudes). Tests of between-group simple effects were used to decompose interactions, whereas planned comparisons were used to examine the accuracy factor within each group. Cohen's-d effect sizes (Cohen, 1988) were calculated for condition-related effects.
RESULTS
Behavioral Data
Stroop Task behavioral performance
Overall RTs and error rates for the Stroop Task (Table 3) were not significantly correlated in control participants, r(20) = −0.17, p > .47, but were significantly positively correlated in participants with TBI, r(18) = .47, p < .04. Results suggest speed/accuracy trade-off did not influence control or TBI participants, as a positive correlation indicates longer RTs were associated with increased error rates, opposite the direction suggestive of a speed/accuracy trade-off.
Mean (±SD) error rates (percent) and reaction times (milliseconds) on the Stroop Task
Error rates
Control and TBI participant groups did not differ on total number of raw errors, t(38) = 1.54, p > .13, d = .49; participants with TBI averaged 48.79 ± 74.75 errors, while controls averaged 23.10 ± 16.53 errors. Similarly, groups did not significantly differ on the number of omission errors, t(38) = 1.13, p > .27, d = .36. Participants with TBI committed an average of 13.52 ± 25.21 omission errors, while control participants averaged 6.40 ± 13.38 omission errors.
A Group × Congruency ANOVA on arcsine-transformed error rates yielded only a significant main effect of congruency, F(1,38) = 83.68, p < .001, η2 = .69, reflecting significant error-rate interference, with both groups committing more errors to the incongruent than congruent condition, as revealed by planned contrasts [TBI: t(18) = 6.37, p < .001, d = .58; controls: t(20) = 6.57, p < .001, d = 1.34]. Neither the main effect of group, F(1,38) = 2.21, p > .15, η2 = .06, nor the Group × Congruency interaction, F(1,38) = 1.47, p > .23, η2 = .04, were significant.
Response Times
A Group × Congruency ANOVA on median RTs revealed the expected generalized slowing in participants with TBI, as reflected in a significant main effect of Group, F(1,38) = 11.75, p < .001, η2 = .24. Paralleling the error rate data, a main effect of congruency reflected the anticipated RT interference, F(1,38) = 73.19, p < .001, η2 = .66, with both groups showing longer RTs to the incongruent than congruent condition [TBI: t(18) = 4.97, p < .001, d = .97; controls: t(20) = 9.43, p < .001, d = 1.04]. The Group × Congruency interaction was not significant, F(1,38) = 0.84, p > .37, η2 = .02, indicating that the two groups showed equivalent levels of RT interference.
Event-Related Potential Data: Response-Related Activity
We first examined the number of trials retained for each condition to test for between-groups differences in signal-to-noise ratio (SNR). Controls and participants with TBI did not differ on number of trials retained for averaging in correct, t(38) = 0.55, p > .59, d = .17, or error conditions, t(38) = −1.24, p > .22, d = −.39. Response-locked correct-trial waveforms contained an average (±SD) of 401.1 ± 132.9 trials for participants with TBI and 422.5 ± 116.1 trials for controls, whereas response-locked error waveforms contained an average of 23.1 ± 33.96 trials for participants with TBI and 13.7 ± 7.9 trials for controls. Response-locked grand average ERP waveforms and spline-interpolated voltage maps for correct and error response-locked trials reflecting the fronto-medial Ne/ERN are shown in Fig. 1, whereas those for the centro-parietal Pe are shown in Fig. 2. Mean (±SD) component amplitude data are presented in Table 4.
Grand average ERP waveforms depicting response-locked correct- and error-related activity averaged across fronto-medial electrode locations for the Ne/ERN and top view of the spline-interpolated voltage distribution maps showing mean voltages for error-trial activity at 22 ms. Asterisk denotes the Ne/ERN.
Grand average ERP waveforms depicting response-locked correct- and error-related activity averaged across centro-parietal electrode locations for the Pe and top view of the spline-interpolated voltage distribution maps showing mean voltages for error-trial activity at 322ms for control and 286 ms for participants with TBI. Asterisk denotes the Pe component.
Mean (±SD) Ne/ERN and Pe component amplitude (μV) as a function of task condition
Ne/ERN. As anticipated, response-locked ERPs showed an early negative deflection that was larger in amplitude to error than correct trials in both groups, as confirmed by a Group × Accuracy ANOVA, which yielded a significant main effect of accuracy, F(1,38) = 41.09, p < .001, η2 = .52, and significant correct- versus error-trial planned contrasts in both the TBI, t(18) = 2.62, p < .02, d = .70, and control participants, t(20) = 6.12, p < .001, d = 1.65. More importantly, a significant Group × Accuracy interaction, F(1,38) = 12.99, p < .001, η2 = .26, reflected greater amplitude differences between correct- and error-related negativities in control than participants with TBI. Follow-up contrasts revealed that the error-trial Ne/ERN was significantly larger in the control relative to participants with TBI, t(38) = 2.65, p < .01, d = .83, whereas the ERP related to correct responses (correct-response negativity; CRN) did not differ between groups, t(38) = 1.69, p > .10, d = .54. Furthermore, the Group × Accuracy interaction was found in the absence of an overall main effect of group on Ne/ERN amplitude, F(1,38) = 2.58, p > .12, η2 = .06.
Pe
The Group × Accuracy ANOVA on the centro-parietal Pe revealed a significant main effect of accuracy, F(1,38) = 45.12, p < .001, η2 = .54, reflecting a more positive-going Pe on error than correct-trials in both the TBI, t(18) = 4.84, p < .001, d = 1.26, and control participants, t(20) = 4.91, p < .001, d = 1.22; however, the Group × Accuracy interaction was not significant, F(1,38) = 1.67, p > .20, η2 = .04. Subsequent tests of simple effects indicated the Pe significantly differed between groups on error, t(38) = 1.97, p = .05, d = .61, but not correct trials, t(38) = .60, p > .55, d = .18. These results, however, are difficult to interpret in the absence of a significant interaction effect and due to the relatively small sample size. Similar to the Ne/ERN, the main effect of group on Pe amplitude was not significant, F(1,38) = 2.52, p > .13, η2 = .06.
Peak latencies
A Group × Accuracy ANOVA yielded no significant main effects or interactions involving Ne/ERN latency (ps > .09).
DISCUSSION
Results of the current study were largely consistent with our primary prediction that participants with TBI would show a reduced-amplitude Ne/ERN on error relative to correct trials—reflective of an impaired neural mechanism of error processing or performance monitoring. Notably, participants with TBI did, on average, demonstrate a clearly discernable Ne/ERN, suggesting that, whereas error-related performance monitoring was impaired, it was not completely absent. Importantly, the finding that control and TBI groups did not significantly differ in correct-trial ERP amplitudes or the overall magnitude of response-related ERP amplitudes provides some evidence for specificity of the Ne/ERN reflection of performance monitoring deficits and does not reflect a more generalized decrement of response-locked ERP amplitudes in the TBI survivors.
We also observed a discernable positive-going deflection following the Ne/ERN in both TBI and control participant groups. This more phasic component, routinely referred to as the Pe, is believed to be involved in the conscious recognition and processing of errors—though the functional significance of the Pe remains under active debate (Overbeek et al., 2005). Current results indicate both TBI and control participants showed a significant difference in Pe amplitude between correct and error trials, but there was not a significant Group × Accuracy interaction. Thus, despite the significant difference between control and participants with TBI on subsequent contrasts for error, but not correct trials, no firm conclusions can currently be drawn about Pe amplitude in severe TBI.
While supporting our hypothesis that TBI survivors exhibit performance monitoring deficits, the ERP results do not directly address critical questions regarding the source or mechanism underlying the ERP manifestation of performance monitoring impairment. In addition to the potential limitations associated with interpretation of ERP findings in neurologically-impaired individuals (see later), reduced-amplitude ERP components may be because of a number of factors not directly reflective of an underlying impairment in the process of interest. For example, signal averaging for creation of ERPs is based on the assumption of across-trial time (and amplitude) invariance. However, it is conceivable that participants with TBI exhibit greater variability in the latency of peak response (i.e., “latency jitter”), violating this invariance assumption, and resulting in spuriously reduced-amplitude ERP components. Whereas methods exist for evaluating single-trial ERP data and adjusting for latency jitter (e.g., Möcks et al., 1988; Picton et al., 1995; Woody, 1967), their effectiveness is limited to large-amplitude ERP components (e.g., P300) and may not yield reliable or valid effects for the Ne/ERN or Pe components examined in the present research.
Nonetheless, results of this study suggest important implications for clinical application and future research. First, the pursuit of effective rehabilitation and compensatory strategies is dependent on accurate cognitive assessment and detailed understanding of the mechanisms underlying TBI-related impairment. The current study, reflecting the first published examination of an electrophysiological “marker” of performance monitoring in TBI, suggests a possible experimental method and electrophysiological marker for examining the behavioral and neurobiological changes in performance monitoring abilities in response to rehabilitation. Second, results suggest a continued need for emphasis on rehabilitation of performance monitoring deficits following TBI. Few empirically supported treatments currently exist that target such deficits, though investigators are currently working to validate potential treatments in this domain (Ownsworth et al., 2006). Finally, future studies should address the potential link between performance monitoring decrements and impairments in deficit awareness. For example, O'Keeffe et al., (2004) found attenuated electrodermal response to errors following TBI, as well as a relationship between performance monitoring abilities (error awareness), and amplitude of the error-related electrodermal response. Utilization of an electrophyiological marker, such as the Ne/ERN, in studies of performance monitoring and deficit awareness may provide much needed clarification on the neural mechanisms underlying impairments in deficit awareness.
Findings of the current study must be considered within the context of several potential limitations and alternative explanations, in addition to the ERP-related factors mentioned above. The task paradigm employed precluded our ability to unambiguously examine participants' facility to make reactive strategic adjustments following the commission of errors, such as post-error trial RT slowing, due to the probability distribution of congruent and incongruent trials (i.e., the preponderance of participant commission errors occurred in the incongruent condition and were followed by congruent trials due to the 70% congruent/30% incongruent proportion of trials employed; thus, post-error slowing was not evaluated because participants traditionally respond faster to congruent than incongruent trials). Post-error strategic adjustments, or the so-called “Rabbitt effect” (Rabbitt, 1966, 1968), have often been taken to reflect participants' top-down adjustments in performance strategy, perhaps reflecting the dynamic interplay between ACC-mediated evaluative and dlPFC-mediated regulative processes (Botvinick et al., 2004; Kerns et al., 2004). Nonetheless, our earlier behavioral study using a different task, demonstrated that participants with TBI were impaired in post-error RT slowing relative to healthy participants (Larson et al., 2006a). That is, relative to neurologically-normal comparison subjects, they showed smaller magnitude post-error slowing, suggestive of reduced post-error strategic adjustments in cognitive control.
Other important potential limitations with respect to comparing ERPs from neurologically-injured groups with neurologically-healthy comparisons should be noted. First, the current data indicate generalized slowing in participants with TBI that was not present for control participants. The potential variability in ERP component latency, as noted earlier, associated with variable response times may serve to spuriously reduce component amplitudes. Similarly, previous research indicates individuals who respond more quickly at the expense of accuracy show smaller Ne/ERN amplitude than those who attend more to accuracy than speed (Falkenstein et al., 2000; Gehring et al., 1993). Second, the possibility of alterations in cortical geometry, volume, and electrical conductivity in the TBI patient group may give rise to spurious amplitude reductions in the Ne/ERN. Specifically, the propagation or volume conduction of potentials to the scalp surface, and therefore their scalp distribution can be altered by the presence of injury-related factors. These can result in altered amplitude or scalp distribution of the ERP, challenging the assumption that using identical measurement electrode sites across the different groups will yield similar measurement sensitivity to the ERP components of interest. However, this explanation is unlikely in the present study given that the amplitude of CRN did not differ across groups. Finally, as a consequence of their impairment, participants with TBI committed more errors (although not statistically significant) on the Stroop task. Since the error-related ERP data were dependent on error rates, the TBI group contains nearly twice as many trials included in the ERP averages than controls, potentially making the error-related ERPs in the TBI group more reliable.
The absence of group-related differences in error rates may obscure the usefulness of the Ne/ERN as an electrophysiological reflection of impairment following severe TBI. However, previous studies conducted in our lab demonstrate Stroop tasks where the instruction context is held constant (i.e., all color-naming or all word-reading trials) are not sensitive to between-groups differences in error rates, whereas single-trial tasks that switch between color and word instruction contexts are sensitive to error rate differences (Seignourel et al., 2005; Perlstein et al., 2006). The absence of error rate differences may, therefore, be caused by the insensitivity of the current color-naming task or the relatively greater variability in error rates for participants with TBI than healthy comparison participants. Given these potential alternative explanations for the absence of between-groups error rate differences, the Ne/ERN may still serve a role as a marker of impaired performance monitoring following severe TBI. Future research will be required to better elucidate the potential functional consequences of this impairment.
It should also be noted that, despite the higher state and trait anxiety levels in the participants with TBI relative to controls, Ne/ERN amplitude was still reduced in the participants with TBI. Previous studies demonstrate that anxiety enhances Ne/ERN (Hajcak et al., 2003b) and higher anxiety in the TBI group would appear to work in opposition to our hypothesis of reduced error processing in participants with TBI. That is, despite the higher anxiety scores in participants with TBI, a Group × Accuracy interaction was, nonetheless, observed for the Ne/ERN. Depression, on the other hand, is associated with decreased Ne/ERN amplitude (Ruchsow et al., 2004). The finding of group differences in depression score may suggest that the level of depression in participants with TBI influenced findings of reduced ERP amplitude; however, depression scores for both groups were in the subclinical range and the lack of change in the pattern of significance for the Ne/ERN and Pe components when depression and anxiety scores were added as covariates indicates findings cannot be wholly accounted for by these variables.
Concluding Remarks
The present findings implicate an impaired performance monitoring mechanism in survivors of severe TBI. Thus, the present study fits well into a growing body of research indicating impaired performance monitoring following severe TBI and emphasizes the need for continued specific examination of this dysfunction and the development and validation of remediational or compensatory treatments.
ACKNOWLEDGMENTS
This original research has not been published elsewhere and is supported by a predoctoral National Institute of Health (NIH) Fellowship to MJL (F31 NS053335) and grants from the Evelyn F. McKnight Brain Research Grant Program, the Florida Brain and Spinal Cord Injury Research Trust Fund, and the NIH to WMP (K01 MH01857; R21 MH073076). The authors extend appreciation to Drew Nagle, Raechel Steckley, Cortney Mauer, Floris Singletary, and Ashley Carrol for their assistance in patient recruitment and data collection. Portions of this research were conducted in partial fulfillment of the Ph.D. dissertation of the first author.
