Altered Neural Substrates of Cognitive Control in Childhood ADHD: Evidence From Functional Magnetic Resonance Imaging

Ten children (seven boys and three girls; mean age=8.8 years, SD=0.9) with combined-type ADHD and 10 age- and gender-matched comparison subjects (mean age=9.2 years, SD=1.3) participated for monetary compensation. The comparison subjects were a subset of a previously described comparison group (11). After complete description of the study to subjects, written informed consent from parents and assent from children were obtained. At the time of scanning, seven of the ADHD children were medication naive. In the other three ADHD children, medication had been withheld for 4 weeks (N=1) or 36 hours (N=2). Exclusion criteria were 1) full-scale IQ below 85 (estimated from performance on Vocabulary and Block Design subtests of the WAIS-III [ADHD subjects: mean=104.3, SD=16; comparison subjects: mean=128.4, SD=16.7]); 2) history of neurological or affective disorder; or 3) language disorder (subtest scores below 85 on the Wechsler Individual Achievement Test or Woodcock Mastery Scale). Inclusion criteria for the ADHD group were 1) DSM-IV criteria met as determined by the Diagnostic Interview Schedule for Children (19) (symptoms: mean=16.3, SD=1.7); 2) parent and teacher endorsements (i.e., ratings of “pretty much” or “very much”) on five out of nine items from the Swanson, Nolan, and Pelham (SNAP) Checklist; and 3) Child Behavior Checklist and Teacher Report Form (20) Attention Problem scores >60, a cutoff validated by Chen et al. (21) (ADHD: parent mean=71.3 [SD=7.1]; teacher mean=67.1 [SD=10.7]; comparison subjects: parent mean=50.2 [SD=0.48]). ADHD children with symptoms of conduct disorder (mean=0.78, SD=1.6) and oppositional defiant disorder (mean=4.9, SD=3.0) per the Diagnostic Interview Schedule for Children were not excluded. Among the 10 ADHD children, there was at least one symptom of conduct disorder reported in two children and at least one symptom of oppositional defiant disorder in seven children.

Stimuli were generated in Psyscope (22) and viewed via a magnet-compatible projector. Responses were recorded via right-hand held optical buttons. Head movement was minimized using a bite-bar formed with each subject’s dental impression.

Subjects performed two runs of a modified Eriksen flanker task (see reference 11 for details). Each trial consisted of a display comprising a central arrow and two flanking stimuli on either side. Subjects were instructed to press one of two buttons indicating the direction of the central arrow (left/right) while ignoring the flanking stimuli. On congruent and incongruent trials the flankers were arrows pointing in the same or different direction, respectively, relative to the central arrow. On neutral trials the flankers were diamonds that were not associated with a response. On no-go trials, the flanking stimuli were Xs that signaled that subjects should withhold their response.

Trials were presented in a rapid event-related design with a 3-second intertrial interval. On each trial, the display (800 msec) was followed by a blank screen (300 msec) and then a cross hair (1600 msec); intertrial interval was 300 msec. On fixation trials, the cross hair was presented for 2700 msec. The trial sequence followed a stochastic design in SPM 99, in which the probability of each trial type varied sinusoidally between 0 and 1 over a 30-second period. Each trial type had a probability function with a different phase; over time, all trial types occurred with equal probability. Across the two runs, there were 46–58 trials of each type (across subjects 51–52 trials per type) in addition to 44 fixation trials. The trial sequence was specified by one set of lists for half the subjects and another set for the remaining half; the order of the two runs was counterbalanced across subjects.

A 3-T MRI (GE 8.2.5, GE Medical Systems, Milwaukee) was used to acquire T₁-weighted flow compensated spin-echo anatomy images (TR=500 msec; minimum TE) in 16 contiguous 7-mm axial slices, parallel to the plane of the anterior commissure-posterior commissure. Functional acquisition included the same slices with a T₂*-sensitive gradient echo spiral pulse sequence with parameters of TR=1000 msec, TE=30 msec, field of view=24 cm, flip angle=60°, 64×64 matrix.

Using SPM 99 (Wellcome Department of Cognitive Neurology, London), images were corrected for differences in slice acquisition time and motion. All subjects displayed less than 1 mm of motion in x, y, and z directions throughout the course of each scan. Estimated motion parameters did not differ between ADHD children (mean=0.139 mm, SD=0.008) and comparison subjects (mean=0.159 mm, SD=0.005) (t=0.55, df=18, p=0.59). Images were normalized into a standard space (23) and interpolated to 2×2×2 mm cubic voxels. Normalized image volumes were spatially smoothed (6-mm full width at half-maximum Gaussian kernel) and temporally filtered (low-pass filter: 4-msec Gaussian; high-pass filter: calculated on the basis of trial frequency). fMRI responses were modeled by canonical hemodynamic response function and its temporal derivative for correct trials only. For each subject, activation maps were generated using linear contrasts identifying regions that were more active during incongruent relative to neutral trials and during no-go relative to neutral trials.

All analyses were generated by using a voxel-level height threshold of p<0.001 uncorrected for multiple comparisons and spatial extent threshold of 5 voxels; in the event that no activation reached threshold, successively liberal height thresholds, with a more stringent extent threshold, were employed to identify regions of subthreshold activation (e.g., uncorrected p<0.005 or p<0.01 with extent threshold of 10 voxels). First, to identify brain regions most consistently engaged during interference suppression (incongruent > neutral) and response inhibition (no go > neutral) within a group, individual activation maps for each contrast were averaged across subjects separately for the ADHD and comparison groups in a random effects model. Second, group differences were examined with group-by-condition repeated-measures analysis of variance (ANOVA) on signal intensity in functional regions of interest defined by the group average of the comparison group. Third, to identify regions in which the magnitude of activation was correlated with performance, simple regression analyses were performed separately for each group for 1) activation during interference suppression (incongruent > neutral) as a function of subjects’ ability to suppress interference (measured by the accuracy or reaction time difference between neutral and incongruent trials) and 2) activation during response inhibition (no go > neutral) as a function of subjects’ ability to withhold inappropriate responses (measured by the accuracy difference between no go and neutral trials). In the event that similar regions were identified across groups, regression analyses were also performed including all subjects.

Incorrect or absent responses on congruent, incongruent, and neutral trials and failure to withhold responses on no go trials were classified as errors. Subjects’ mean accuracy rates and response latencies were statistically analyzed separately in group (ADHD, comparison) by trial type (incongruent, neutral) repeated-measures ANOVAs including conditions that were relevant to the fMRI analysis. Overall accuracy was lower in ADHD than in comparison children (main effect of group: F=10.13, df=1, 18, p<0.01). Further, accuracy was lower for incongruent relative to neutral trials (main effect of trial type: F=9.44, df=1, 18, p<0.01). However, this result differed between ADHD and comparison children (group-by-trial type interaction: F=4.89, df=1, 18, p=0.04). Planned contrasts indicated that accuracy was significantly lower for incongruent relative to neutral trials in ADHD children (incongruent: mean=87.4%, SD=10.5; neutral: mean=92.3%, SD=6.7) (t=2.88, df=9, p=0.02) but not in the comparison subjects (incongruent: mean=98.1%, SD=2.1; neutral: mean=98.9%, SD=2.0) (t=1.07, df=9, p=0.31). Thus, comparison children were more successful at suppressing interference relative to ADHD children. Response latencies of correct trials did not differ between groups. Response latencies were longer during incongruent relative to neutral trials (main effect of trial type: F=23.62, df=1, 18, p<0.01), and this finding did not differ between groups, indicating that both groups were susceptible to interference. The mean magnitude of interference (incongruent relative to neutral trials) was 49.2 msec (SD=52.8) in ADHD children and 43.3 msec (SD=28.9) in comparison children. Thus, while failures of interference suppression were more frequent in ADHD children, temporal characteristics of successful interference suppression were similar in the two groups. For response inhibition, accuracy on no go trials was numerically higher for the comparison children (mean=92.2%, SD=4.0) than for ADHD children (mean=78.1%, SD=24.0). This effect was marginally significant (t=1.83, df=18, p=0.08) because of the large variability in performance within the ADHD group.

We examined whether regions identified by group averaging and the regression analyses correlated with ADHD symptom severity as indexed by scores on the attention problems subscale of the Child Behavior Checklist. Magnitude of activation in the regions of interest identified by group averaging in the left inferior frontal and right inferior parietal region for interference suppression and right precentral region for response inhibition did not correlate significantly with symptom severity. Similarly, none of the regions that correlated with successful interference suppression and response inhibition (identified by regression analyses) correlated significantly with symptom severity. These analyses suggest that variability in symptoms, as indexed by a common diagnostic instrument, are limited in sensitivity to cognitive control task performance and associated activation. However, our ADHD group was selected to be highly homogeneous (all combined subtype) and thus may have lacked the range of variability necessary to reveal correlational relationships.

In light of differences in average IQ between the ADHD and comparison groups, we examined whether a subset of the comparison children (N=5, mean=115) matched in IQ to the ADHD group showed significant activation in the regions of interest identified by group averaging (using two-tailed t tests). For interference suppression, magnitude of activation during incongruent relative to neutral trials was significantly greater in the left inferior frontal region (t=5.4, df=4, p=0.005) but marginally so in the right inferior parietal region (t=1.7, df=4, p=0.15). For response inhibition, magnitude of activation during no go relative to neutral trials was significantly greater in the right precentral region (t=3.6, df=4, p=0.02). Thus, despite a lower IQ, the subset of healthy children showed robust activation within frontal regions. It is unlikely, therefore, that reduced activation in these regions in the ADHD group was related to their lower IQ.

Despite differences in the magnitude and extent of activation, loci of frontal-temporal activation during interference suppression were similar between groups. Studies of adult interference control suggest that the lateral prefrontal cortex mediates selection from competing responses (26, 27). Superior interference suppression was associated with greater left inferior frontal engagement in both groups. Lateralization of the activation to the left hemisphere is consistent with verbal mediation of interference suppression. Indeed, fluid verbal ability was negatively correlated with magnitude of interference in healthy children (11). Verbal mediation is further suggested by activation of other language-related regions such as the left middle temporal cortex in both groups. The extent of frontal-temporal activation, however, was reduced in ADHD children, which may relate to reduced gray matter in those regions (28). Greater recruitment of the premotor cortex, a region known to be involved in response planning, was associated with better interference suppression in both groups, but premotor activation was more extensive in comparison children than in those with ADHD. Thus, similar frontal and temporal activation loci suggest that both groups engaged similar response selection and planning processes that contributed to interference suppression but that ADHD children engaged these areas to a lesser extent.

In contrast to group similarities in the location (but not magnitude) of activation in frontal and temporal cortices, the two groups differed in terms of parietal activation. Studies of adult interference control suggest that the parietal cortex represents visuospatial maintenance and attentional processes (26, 27). Right parietal activation was robust in the comparison children but weak in those with ADHD, suggesting reduced visuospatial attentional processing. Better interference suppression in ADHD children, however, was associated with recruitment of the medial bank of the left superior parietal sulcus. This region, extending posteriorly, has been associated with visuomotor attention (29). Greater recruitment of that region in ADHD children with better interference control may reflect a compensatory strategy that relied upon greater visuomotor attention in the face of reduced visuospatial attention.

The functional anatomy of response inhibition differed between the ADHD and comparison children. Response inhibition in the healthy children was related to premotor activation in the right hemisphere. ADHD children revealed frontal activation—only with lenient thresholds—in the right insula and supplementary motor cortex, regions that were not observed in the comparison children with lenient thresholds. Better response inhibition in ADHD children was associated with robust activation of the right posterior superior temporal cortex. This region has been associated with awareness of one’s own motor actions (30) and may reflect alternative, perhaps compensatory, processes in ADHD children. Those alternative processes were not evoked by the healthy children because temporal activation was not observed even at a lenient threshold. In other studies, regions that were more activated in ADHD than in comparison subjects were in the right inferior parietal cortex (16, 17). The loci of the putative compensatory activation across studies, right inferior parietal and superior temporal regions, correspond closely to regions exhibiting increased gray matter in ADHD relative to comparison children (28). The temporal rather than parietal locus of the posterior activation in the present study may reflect processes specific to current task demands. Our task was novel because it interleaved response inhibition (no go) with a variety of response selection trials (neutral, congruent, and incongruent). The variety of go conditions may have reduced the potency of the go response, yielding a marginal group difference in no go accuracy. Prior ADHD studies have reported significant group differences in performance and also more widespread frontal activation than that in our study (2, 16, 17). Those studies did not employ designs optimized to visualize selectively successful performance, as done in our study. Thus, design, task, and performance characteristics may account for differences across studies.