Striatal Activation Predicts Differential Therapeutic Responses to Methylphenidate and Atomoxetine

Abstract

Objective:

Methylphenidate has prominent effects in the dopamine-rich striatum that are absent for the selective norepinephrine transporter inhibitor atomoxetine. This study tested whether baseline striatal activation would predict differential response to the two medications in youth with attention-deficit/hyperactivity disorder (ADHD).

Method:

A total of 36 youth with ADHD performed a Go/No-Go test during functional magnetic resonance imaging at baseline and were treated with methylphenidate and atomoxetine using a randomized cross-over design. Whole-brain task-related activation was regressed on clinical response.

Results:

Task-related activation in right caudate nucleus was predicted by an interaction of clinical responses to methylphenidate and atomoxetine (F_1,30 = 17.00; p < .001). Elevated caudate activation was associated with robust improvement for methylphenidate and little improvement for atomoxetine. The rate of robust response was higher for methylphenidate than for atomoxetine in youth with high (94.4% vs. 38.8%; p = .003; number needed to treat = 2, 95% CI = 1.31–3.73) but not low (33.3% vs. 50.0%; p = .375) caudate activation. Furthermore, response to atomoxetine predicted motor cortex activation (F_1,30 = 14.99; p < .001).

Conclusion:

Enhanced caudate activation for response inhibition may be a candidate biomarker of superior response to methylphenidate over atomoxetine in youth with ADHD, purportedly reflecting the dopaminergic effects of methylphenidate but not atomoxetine in the striatum, whereas motor cortex activation may predict response to atomoxetine. These data do not yet translate directly to the clinical setting, but the approach is potentially important for informing future research and illustrates that it may be possible to predict differential treatment response using a biomarker-driven approach.

Clinical trial registration information:

Stimulant Versus Nonstimulant Medication for Attention Deficit Hyperactivity Disorder in Children; https://clinicaltrials.gov/; NCT00183391.

Notable individual differences in response to stimulant and nonstimulant medications for attention-deficit/hyperactivity disorder (ADHD) highlight the importance of identifying predictors of differential response.¹ The majority of youth with ADHD respond to both psychostimulants, including methylphenidate, and nonstimulants, such as atomoxetine.² However, there is considerable variability in individual response to treatment; almost 40% of youth with ADHD derive little benefit from atomoxetine,³ and approximately one-third of youth respond preferentially to either methylphenidate or atomoxetine.⁴ This pattern of clinical response may reflect the partially overlapping and partially distinct pharmacological profiles of methylphenidate and atomoxetine: although both medications inhibit the norepinephrine transporter, only methylphenidate blocks the dopamine transporter.^5,6 This difference in affinity for the dopamine transporter could potentially represent the neuropharmacological basis for preferential response to methylphenidate and, if so, could ultimately be exploited to identify individuals who respond preferentially to methylphenidate over atomoxetine.

Differences in the affinity of methylphenidate and atomoxetine for the dopamine transporter manifest most clearly in dissociable effects on neural activity in the striatum. Single doses of methylphenidate but not atomoxetine increase striatal dopamine levels⁷ and task-related neural activity,^8,9 reflecting the expression of abundant dopamine transporters but sparse norepinephrine transporters in the striatum.^10,11 Clinical response to methylphenidate therapy in adults with ADHD was predicted by methylphenidate-elicited increases in striatal dopamine¹² and was positively related to striatal dopamine transporter availability,¹³ with greater transporter binding sites in responders than in nonresponders,^13,14 and reduction in binding sites during treatment correlating with symptomatic improvement.¹⁵ The corresponding lack of binding sites for atomoxetine to directly influence striatal function makes it unlikely that these same effects are implicated in the therapeutic mechanism of action.¹⁶ Measures of striatal function may therefore potentially serve as candidate biomarkers for individuals with ADHD who benefit preferentially from the dopaminergic effects of methylphenidate therapy, while conversely predicting nonresponse to the selective noradrenergic effects of atomoxetine.

It is important to consider what the potential role of biomarker predictors of treatment response in youth with ADHD might be. There would seem to be little clinical need for predictive biomarkers of response to methylphenidate treatment, since the medication is effective for most individuals with ADHD and efficacy can be rapidly determined. However, psychostimulants are not a viable option for a sizable minority of youth with ADHD for a variety of reasons (e.g., comorbidity, tolerability, preference). Developing a biomarker-driven approach to predict response and nonresponse to nonstimulants such as atomoxetine, particularly in relation to the methylphenidate response, would be extremely beneficial, given the lower overall response rate for atomoxetine and the extended duration of exposure needed to determine optimal response.³ Beyond clinical utility, the identification of neural markers of differential response would have considerable heuristic value with regard to determining the mechanisms of action of these frequently used medications for ADHD.

The present study used functional magnetic resonance imaging (fMRI) to test the value of baseline striatal activation for predicting differential response to methylphenidate versus atomoxetine in youth with ADHD. Youth underwent scanning once at baseline, while off medication for at least 2 weeks, while performing a Go/No-Go test of response inhibition off medication, and were treated with both medications in randomized order in a cross-over design. Our previously published study found that clinical improvement was differentially associated with reductions in activation for methylphenidate in several brain regions and corresponding gains in activation for atomoxetine in parallel groups of youth with ADHD that included 27 of the participants in the current study.¹⁷ Based on these findings and on the pharmacological profiles of the two medications, we hypothesized that elevated baseline activation in the striatum would predict superior theraputic response to methylphenidate over atomoxetine.

METHOD

Study Participants

A total of 36 youth, 7 to 17 years of age, with ADHD were recruited from a National Institutes of Health–funded comparator study of methylphenidate and atomoxetine treatment. All participants met DSM-IV criteria for ADHD, any subtype, on the Kiddie Schedule for Affective Disorders and Schizophrenia for School-Aged Children,¹⁸ and were rated at least 1.5 SD above age and gender norms on the ADHD Rating Scale–IV–Parent Version (ADHD-RS-IV).¹⁹ Participants had a mean ± SD Full Scale IQ of 98.1 ± 12.5 on the Wechsler Intelligence Scale for Children–Fourth Edition (WISC-IV) (Table 1).

TABLE 1.

Demographic and Clinical Characteristics

Characteristic	Whole Sample (N = 36)	Low Caudate Activation (n = 18)	High Caudate Activation (n = 18)	p
Age, y, mean (SD)	11.0 (2.4)	11.5 (2.9)	10.6 (1.9)	.245b
Male sex, n (%)	30 (83.3)	16 (88.9)	14 (77.8)	.371c
Full scale IQ, mean (SD)a	98.1 (12.5)	99.0 (9.5)	97.3 (14.7)	.377b
Hollingshead SES	32.5 (17.3)	32.0 (17.0)	32.9 (18.7)	.155b
DSM-IV ADHD subtype
Combined type, n (%)	22 (61.1)	10 (55.6)	12 (66.7)	.494c
Predominantly inattentive type, n (%)	14 (38.9)	8 (44.4)	6 (33.3)
Prior treatment for ADHD, n (%)	8 (22.2)	4 (22.2)	4 (22.2)	>.999c
Comorbid ODD, n (%)	17 (47.2)	10 (55.6)	9 (50.0)	.738c
ADHD-RS-IV
Inattentive score, mean (SD)	21.9 (4.2)	23.3 (2.9)	20.6 (4.9)	.074b
Hyperactive/impulsive score, mean (SD)	15.9 (7.8)	13.9 (9.0)	17.9 (5.8)	.120b
ADHD symptom total score, mean (SD)	37.8 (9.8)	37.2 (11.0)	38.5 (8.6)	.688b
Go/No-Go task performance
Commission errors, %, mean (SD)	26.2 (14.6)	25.3 (16.6)	27.1 (12.7)	.729b
Omission errors, %, mean (SD)	5.9 (7.1)	5.0 (5.3)	6.9 (8.6)	.423b
Reaction time, milliseconds, mean (SD)	497.3 (97.1)	491.0 (87.0)	503.7 (108.2)	.700b
Criterion, mean (SD)	−0.5 (0.3)	−0.5 (0.3)	−0.5 (0.2)	.901b

Note: ADHD = attention-deficit/hyperactivity disorder; ADHD-RS-IV= ADHD Rating Scale-IV; ODD = oppositional defiant disorder; SES = socioeconomic status.

^aFull scale IQ not available for 2 participants in the low caudate activation group and 1 in the high caudate activation group.

^bStudent t test for independent samples.

^cPearson χ² tests.

Exclusion criteria included poor response or tolerability to an adequate trial of either medication, substance abuse history or a positive urine screen, participation in a treatment study in the past 30 days, primary diagnosis of a mood or anxiety disorder requiring treatment, past or present diagnosis of a psychotic disorder, or any medical condition that could affect brain function. In all, 28 participants were medication-naive at the beginning of the study; 7 participants had a history of prior stimulant treatment, and 1 participant had previously been treated with atomoxetine. The protocol required a minimum of 2 weeks off medication before beginning assessment.

This study was approved by the institutional review board of the Icahn School of Medicine at Mount Sinai. Written informed consent and verbal assent certified by an unaffiliated witness were obtained from all parents and participants, respectively. Consent was additionally obtained from 16 youth who did not complete the procedures, 4 for excessive motion or anxiety during the baseline scan, 5 who were never randomized to treatment (i.e., screen fails or withdrawal of consent), and 7 because they dropped out of the study before completing both treatments.

Study Design

Participants underwent scanning with fMRI at baseline, following a minimum of 2 weeks off medication, and were treated with osmotically released (OROS) methylphenidate (McNeil-PPC Inc.) and atomoxetine (Eli Lilly and Company) in a double-blind, cross-over design. The cross-over design of the clinical trial included two baseline periods, one prior to each medication. A total of 30 youth (83%) underwent scanning during the first baseline period and 6 youth (17%) at the second baseline. All participants were medication free for 2 weeks before scanning.

Medication was administered in randomized order in two treatment blocks, each lasting approximately 8 weeks, separated by a 2-week placebo washout. Medication was titrated to a standard of optimal response and tolerability using stepped sequential dose escalating procedures, with an absolute dose schedule for methylphenidate and a weight-adjusted schedule for atomoxetine. Methylphenidate was initiated at 18 mg/d and titrated upward in 18-mg/d increments to a maximum daily dose of 72 mg. Atomoxetine was started at a daily dose of 0.5 mg/kg and was titrated to 1.8 mg/kg (dose levels of 0.5, 1.0, 1.4, and 1.8 mg/kg), with a maximum total daily dose of 120 mg.

End-of-treatment assessments were conducted after at least 2 weeks of stable response at the optimal dose. Clinical response was measured with the ADHD-RS-IV, and magnitude of improvement was calculated as the percent change in ADHD-RS-IV total score (i.e., difference of baseline and posttreatment scores divided by the baseline score, and multiplied by 100). Response to treatment was categorized as poor (<30% decrease in ADHD-RS-IV from baseline), moderate (30%–50% decrease), or robust (>50% reduction from baseline). The mean ± SD daily dose at the end-of-treatment assessment was 54.5 ± 17.5 mg for methylphenidate and 1.23 ± 0.47 mg/kg for atomoxetine. The mean ± SD length of treatment was 53.9 ± 17.6 days for methylphenidate and 53.4 ± 25.3 days for atomoxetine (t₃₅ = 0.10, p = .92).

Go/No-Go Task

Participants performed an established Go/No-Go task during fMRI.^17,20 The task measured the ability to inhibit responses to rare nontargets (No-Go trials) in the context of responding to frequent targets (Go trials), and consisted of six runs that each lasted 4 minutes. Each run began with 10 seconds of fixation and contained 57 trials, with 43 (75%) Go trials and 14 (25%) No-Go trials. Stimuli were presented for 500 milliseconds with an interstimulus interval of 3,500 milliseconds. Promotional images from the Spiderman movie were used as stimuli. Participants were instructed to respond as quickly and accurately as possible with the right hand using a fiber-optic button system. The percentage of commission errors on No-Go trials served as the measure of response inhibition; the percentage of omission errors and reaction time were calculated for Go trials. In addition, the signal detection variable criterion (c) was calculated from the hit (Go trial) and false alarm (No-Go trial) rates to provide a pooled measure of performance on the task. Criterion was a measure of response bias, with negative values indicating a bias to respond (as opposed to not responding).²¹

Image Acquisition

Participants underwent scanning on the same 3.0-Tesla Siemens Allegra (Siemens Medical Systems) head-dedicated MRI scanner. Six series of 120 functional T2*-weighted images depicting the blood oxygenation level–dependent (BOLD) signal were acquired in the axial plane using gradient-echo echoplanar imaging (repetition time = 2,000 milliseconds; echo time = 40 milliseconds; flip angle = 90° field of view = 210 mm; 64 × 64 matrix; slice thickness = 3 mm; interslice gap = 1 mm; 28 slices). A high-resolution T2-weighted anatomical volume of the brain was acquired at the same 28 axial slice locations with a turbo spin-echo pulse sequence (slice thickness = 4 mm; in-plane resolution = 0.41 mm²).

Image Preprocessing and First-Level Analysis

Functional images were preprocessed and analyzed using SPM8 software (Wellcome Trust Center for Neuroimaging, London, UK). Six whole-brain BOLD time series, each representing a single run of the Go/No-Go task, were acquired for each participant. The BOLD time series were motion-corrected for each participant individually, and time series that included a volume with more than one voxel (4 mm) of motion were discarded. In all, 20 (9%) of the 216 time series acquired for the whole sample were discarded for excessive motion. Participants each contributed a mean ± SD of 5.5 ± 0.6 time series to the analyses. The remaining BOLD time series were corrected for the staggered acquisition of slices during echo-planar imaging, co-registered to the high-resolution T2-weighted image, spatially normalized to the standard 2-mm Montreal Neurological Institute (MNI) template, and smoothed with an 8-mm Gaussian kernel. Event-related analyses were conducted with subject-specific general linear models that fit β weights to regressors for the four trial events (correct No-Go, correct Go, incorrect No-Go, incorrect Go) in each run, as well as six motion parameters of no interest,²² convolved with the default SPM hemodynamic response function.²³ The neural effect of response inhibition was modeled by applying appropriate linear contrasts to parameter estimates for correct No-Go events minus correct Go events, yielding a contrast map for each participant.

Group Analysis

The contrast maps of all participants were entered into second-level random-effects general linear models conducted using SPM8 software. An initial one-sample t test was conducted to confirm prior reports of frontostriatal activation for response inhibition.^17,20 Whole-brain activation associated with symptomatic improvement for methylphenidate versus atomoxetine was tested using a multiple linear regression analysis that included three regressors for the following: (1) ADHD-RS-IV percent change score for methylphenidate, (2) ADHD-RS-IV percent change score for atomoxetine, and (3) an interaction term (i.e., product), which identified activation related to differential response to the two treatments. Both change score regressors were centered on zero. Treatment order and the signal detection variable criterion were entered as covariates in the regression. Criterion was included to account for the influence of behavioral performance on the relation of activation to treatment response. The results of the group analyses were thresholded at p < .005 with a cluster size of k > 100 voxels to protect against false-positive results.²⁴ A Monte Carlo simulation (procedure described by Slotnick and Schacter²⁵) that accounted for image resolution and smoothing parameters established that a cluster extent of 100 contiguous resampled voxels (2 mm³) corrected for multiple voxel comparisons at p < .01.

Post hoc analyses were conducted to further characterize activation associated with differential response to methylphenidate and atomoxetine. We extracted β values from clusters identified in the regression analysis using masks from the “aal.002” atlas²⁶ to restrict data extraction to the region of interest. Youth with ADHD were classified into high and low activation groups based on a median split of the extracted beta values. Follow-up tests examined the rate of robust responses (dichotomously defined) to methylphenidate versus atomoxetine in the whole sample and separately for the high- and low-activation groups using McNemar χ² tests and the number needed to treat (NNT).²⁷

RESULTS

Clinical Response and Behavioral Performance

The ADHD-RS-IV change scores for methylphenidate and atomoxetine are plotted in Figure S1 (available online). Response to treatment varied from poor to robust for both medications; the ADHD-RS-IV change score ranged from −20.0% to 97.1% for methylphenidate and from −14.5% to 96.9% for atomoxetine. Mean ADHD-RS-IV change score was 59.5 ± 29.0% for methylphenidate and 39.3 ± 33.0% for atomoxetine (paired t₃₅ = 3.27, p = .002). A total of 23 youth (63.9%) showed robust clinical responses to methylphenidate, whereas 15 youth (41.7%) responded robustly to atomoxetine (χ² = 3.55, p = .096; NNT = 6, 95% CI = −31.93 to 2.38). The preponderance of preferential responders to methylphenidate in the sample is evident in Figure S1 (available online). Performance measures on the Go/No-Go test are provided in Table 1. Neither ADHD-RS-IV change score correlated with the percentage of omission errors or the percentage of commission errors (all p > .05), but the change score for atomoxetine correlated significantly with reaction time on Go trials (r = 0.39, p = .02). The number of BOLD time series discarded for excessive motion was not related to the percent change scores for atomoxetine (r = 0.10, p = .56) or methylphenidate (r = −0.23, p = .18).

Whole-Brain Activation for Response Inhibition

Analysis of the entire sample of youth with ADHD revealed a robust pattern of frontostriatal activation for response inhibition (correct No-Go events minus correct Go events) that included bilateral caudate nucleus (Figure 1). Cluster coordinates and statistics are provided in Table S1, available online. There were no differences in task-related activation between youth with and without prior treatment at the time of the fMRI scan or between youth who underwent scanning during the first and second baseline periods.

FIGURE 1.

Baseline activation for response inhibition (correct No-Go minus correct Go events) in youth with attention-deficit/hyperactivity disorder (ADHD). Note: The figure was thresholded at cluster p value corrected <.005.

Whole-Brain Activation Related to Treatment Response

The multiple regression analysis revealed a significant main effect of the ADHD-RS-IV change score for atomoxetine on activation in right motor cortex (MNI coordinates: x = 60, y = −6, z = 36; cluster size = 104 voxels; F_{1, 30} = 14.99; p < .001; Figure 2A). The magnitude of motor cortex activation was positively related to symptomatic improvement for atomoxetine, such that greater activation predicted better clinical response to atomoxetine (Figure 2B). There were no significant main effects of the ADHD-RS-IV change score for methylphenidate on task-related activation in any brain regions (Figure S2, available online).

FIGURE 2.

(A) Symptomatic improvement for atomoxetine predicted baseline activation for response inhibition (correct No-Go minus correct Go events) in right motor cortex in youth with attention-deficit/hyperactivity disorder (ADHD); (B) Scatterplot of the association between right motor cortex activation (in b values) and percent change on the ADHD Rating Scale–IV–Parent Version total symptom score for atomoxetine.

The interaction term in the regression analysis identified a cluster of activation for response inhibition in right caudate nucleus (MNI coordinates: x = 8, y = 10, z = 0; cluster size = 157 voxels; F_{1, 30} = 17.00; p < .001; Figure 3A), reflecting an interaction of the ADHD-RS-IV change scores for methylphenidate and atomoxetine on the magnitude of activation in this cluster. As shown in Figure 3B, elevated right caudate nucleus activation was predicted by a combination of robust symptomatic improvement for methylphenidate and little to no improvement for atomoxetine. Stated another way, elevated activation in the right caudate nucleus distinguished youth with ADHD who responded robustly to methylphenidate and poorly to atomoxetine.

FIGURE 3.

(A) Differential symptomatic improvement for methylphenidate and atomoxetine predicted baseline activation in right caudate nucleus for response inhibition (correct No-Go minus correct Go events) in youth with attention-deficit/hyperactivity disorder (ADHD); (B) Heat map of the association between right caudate nucleus activation (in β values) and percent change on the ADHD Rating Scale–IV–Parent Version total symptom scores for methylphenidate and atomoxetine. Note: The map illustrates the association of elevated caudate activation with the combination of robust improvement for methylphenidate and little to no improvement for atomoxetine. The heat map was generated from scatter data using SigmaPlot software version 12.5 (Systat Software, Inc., San Jose, CA) with inverse distance interpolation.

Post hoc tests of β values extracted from the right caudate nucleus cluster were conducted to further characterize the interaction effect in the regression analysis. A median split of the β values for right caudate nucleus activation was used to divide the sample into high-activation (n = 18) and low-activation (n = 18) groups that did not differ on clinical or demographic variables (Table 1). Of note, 17 (94.4%) of 18 youth in the high–caudate-activation group showed robust clinical responses to methylphenidate, whereas only 7 (38.8%) demonstrated robust responses to atomoxetine (χ² = 9.31, p = .003; NNT = 2, 95% CI = 1.31–3.73). Among the 18 youth in the low–caudate-activation group, only 6 (33.3%) showed robust responses to methylphenidate and 9 (50.0%) demonstrated robust responses to atomoxetine (χ² = 1.80, p = .375; NNT = −6, 95% CI = −6.63 to 2.07).

DISCUSSION

The results of this study confirm the hypothesized association of elevated caudate nucleus activation for response inhibition with superior response to methylphenidate over atomoxetine. Furthermore, the magnitude of motor cortex activation at baseline was found to predict clinical response to atomoxetine. To our knowledge, these are the first findings to identify possible biomarker predictors of differential response to the stimulant methylphenidate over the nonstimulant atomoxetine, as well as overall response to atomoxetine in youth with ADHD. These candidate biomarkers may potentially be of interest to clinicians, given both the high level of nonresponse to atomoxetine and the extended time required to achieve optimal response to this medication.³ The findings have considerable promise for directing future research on personalized treatment of ADHD. However, they are not yet sufficiently developed to be applied to clinical practice.

Caudate nucleus activation for response inhibition may be a candidate biomarker of differential response to methylphenidate and atomoxetine in youth with ADHD. Specifically, elevated activation in the right caudate nucleus predicted robust response to methylphenidate and little to no response to atomoxetine. The magnitude and potential clinical value of enhanced caudate nucleus activation for predicting superior response to methylphenidate over atomoxetine are best illustrated by the NNT measures. In the current study, it was necessary to treat six youth with ADHD in the overall sample to detect one individual who responded robustly to methylphenidate compared to atomoxetine; this is similar to and even slightly better than the NNT of 9 previously reported in a large, parallel-group comparator study of these two medications.⁴ However, consideration of the baseline magnitude of caudate nucleus activation for response inhibition greatly improved this NNT measure. One of every two youth who presented with elevated right caudate activation showed robust improvement in ADHD symptoms for methylphenidate compared to atomoxetine. The greatly improved NNT associated with high caudate activation presumably reflects the filtering of youth with low caudate nucleus activation, who showed a much lower rate of robust response to methylphenidate and no preference for either treatment. The identification of a biomarker that predicts preferential response to a stimulant over a nonstimulant medication, in association with a specific neural function (i.e., response inhibition) in a particular brain region (i.e., caudate nucleus) in one-third of youth with ADHD, is a potentially important step in the development of targeted approaches to treatment.

The present findings are partially consistent with previous studies in adults with ADHD that reported a link between clinical response to methylphenidate and baseline levels of striatal dopamine¹² and dopamine transporters.^13–15 However, in the current study, elevated right caudate nucleus activation predicted response to methylphenidate, but only when considered in combination with response to atomoxetine. Methylphenidate response alone was not related to caudate nucleus activation, possibly because the high rate of robust response to methylphenidate in our sample (23 of 36 youth; 63.9%) may have presented insufficient variability to detect associations with brain activation outside the context of response to atomoxetine. In addition, the primary focus on predictors of differential response, including the use of a cross-over design to directly compare two Food and Drug Administration–approved medications for ADHD in the absence of a placebo condition, and a task-based neuroimaging measure (i.e., BOLD signal change for response inhibition) may have also contributed to the failure to find a striatal marker specific to methylphenidate response.

The delineation of neurobiologically valid markers of differential response to stimulant and nonstimulant medications has heuristic value for understanding the therapeutic mechanisms of these different classes of medications. It is noteworthy that the robust clinical response to the nonselective catecholamine reuptake inhibitor methylphenidate and the poor response to the selective norepinephrine reuptake inhibitor atomoxetine were together associated with elevated activation, specifically in the striatum, the site of the densest dopamine transporter concentrations in the brain,¹⁰ but little norepinephrine transporter expression.¹¹ Thus, the difference in affinity for dopamine transporters and, by extension, the effects on striatal function between the two medications, may represent the neuropharmacological basis for preferential response to methylphenidate.¹⁶ The additional therapeutic action of methylphenidate in the striatum could also account for the greater response to methylphenidate than to atomoxetine found both in this study and in larger clinical trials.^4,28

The identification of elevated caudate nucleus activation as a potential biomarker for differential response is consistent with a large body of literature implicating striatal dopaminergic mechanisms in the therapeutic actions of methylphenidate, but not atomoxetine, in individuals with ADHD.¹⁶ Single doses of methylphenidate increase extracellular dopamine levels in the striatum,⁷ which acutely enhances local activation,^8,9 purportedly via D1 and D2 receptor mechanisms that have opposing excitatory and inhibitory effects on activation in cortical and subcortical regions.²⁹ However, short-term daily methylphenidate treatment has been linked to the downregulation of elevated striatal dopamine transporter levels in adults with ADHD,¹⁵ whereas 1 year of daily treatment has been reported to both enhance striatal dopamine transporter levels³⁰ and to downregulate dopamine release and postsynaptic receptor levels.¹² The present results suggest that these adaptations in striatal dopamine function, which would dampen activation in the striatum and downstream regions,²⁹ may be particularly relevant to the therapeutic effects of methylphenidate for symptoms of ADHD related to caudate nucleus hyperactivation. However, changes in striatal activation with treatment can only be inferred from the baseline scans obtained in the current study. Our previously published, parallel-group fMRI study, in which many (but not all) of the same participants underwent scanning before and after treatment, found no evidence that striatal mechanisms contributed to clinical improvement for either methylphenidate or atomoxetine.¹⁷ Furthermore, the positive clinical responses to both medications in at least some youth with lower caudate activation point to the involvement of therapeutic mechanisms beyond the striatum (e.g., the inferior frontal gyrus).

The current findings may have the greatest potential clinical impact in relation to the need to identify the ~40% of youth with ADHD who derive little benefit from atomoxetine treatment.³ The identification of atomoxetine as a relatively poor treatment for symptoms of ADHD in the context of caudate nucleus hyperactivation raises the question of whether youth with elevated caudate activation may comprise a substantial proportion of atomoxetine nonresponders, and argue against treating these youth with atomoxetine as a first option. Moreover, the finding of a positive response to atomoxetine related to higher baseline motor cortex activation is partially consistent with the improvement-related inhibition of motor cortex activation found for both atomoxetine and methylphenidate in our parallel-group study.¹⁷ Together, these findings implicate motor cortex inhibition in the therapeutic actions of atomoxetine, at least within the context of a response inhibition task, and suggest that the capacity for reduction in motor cortex activation may be an improvement-limiting factor for atomoxetine therapy. The potential of this and other biomarker studies³¹ to explain and to predict the largely bimodal response profile of atomoxetine warrants further investigation of this motor cortex mechanism.³ Using the two biomarkers (i.e., high right caudate and bilateral motor cortex activation) together might represent a fruitful approach, but it is beyond the scope and power of the current study.

The findings of this study should be considered in light of several potential limitations. First, the identification of striatal predictors of response to methylphenidate may have benefited from a larger sample size. The relatively small sample size may have presented insufficient variability in methylphenidate response to detect associations with striatal activation outside the context of the atomoxetine response. Second, carryover effects of treatment on baseline activation cannot be completely ruled out for the six youth who underwent scanning during the second baseline period. However, all participants were medication free for at least 2 weeks before scanning, and there were no differences in baseline activation between youth scanned during the first and second baseline periods. Moreover, excluding the data from the six youth scanned during the second baseline period did not meaningfully change the results of the study. Finally, the use of a whole-brain approach to analyze task-related activation in this study might seem counterintuitive, given the specific a priori hypotheses regarding striatal activation as a predictor of differential response. Yet, this approach offered the opportunity to test the specificity of the findings to the striatum, which would not have been possible using a region-of-interest approach.

In summary, the present study identifies baseline right caudate nucleus activation for response inhibition as a predictor of differential response to the stimulant methylphenidate and the nonstimulant atomoxetine in youth with ADHD, and motor cortex activation as a predictor of overall response to atomoxetine. Our findings indicate that regional localization of functional neural anomalies linked to ADHD, in this case, in the right caudate nucleus and motor cortex, and the availability of pharmacological targets for medications (e.g., catecholamine transporters) in these regions, may be key determinants of differential treatment response in youth with ADHD. Although the data presented here do not translate directly to the clinical setting, the approach is potentially important, and provides a foundation for future investigations.

Supplementary Material

Supplemental material cited in this article is available online.