Background
Attention-deficit hyperactivity disorder (ADHD), characterized by impaired attention and impulsivity/behavioral control, is a common neurodevelopmental disorder persisting across the lifespan [
1]. Among treatment options, methylphenidate is one of the most common, efficacious, and tolerable psychostimulants as pharmacotherapy for ADHD [
2]. Methylphenidate is effective in reducing core symptoms and associated behavioral problems of ADHD, as well as improving academic performance, quality of life, and neuropsychological functions [
3,
4]. However, around 30% of individuals with ADHD exhibit poor responses to methylphenidate [
5]. Studies indicate that individuals with ADHD having poor intelligence quotient, higher disease severity, and a family history of a psychiatric disorder [
6] show a poor response to methylphenidate. Further, higher anxiety levels [
7], as well as co-occurring personality, substance use [
8], alongside anxiety disorders [
9] are related to suboptimal methylphenidate responses in individuals with ADHD.
Investigating an individual’s neurobiological variations may provide better explanations and have translational potentials to help identify those poor responders before the start of methylphenidate prescription. This endeavor is of clinical significance to reduce the suffering of unnecessary drug-related side-effects, the delay from receiving more effective treatment, and the discouragement of the patients and their families. Methylphenidate binds to the dopamine transporter and norepinephrine transporter and blocks their reuptake, thereby increasing the extracellular levels of these neurotransmitters [
10]. Earlier pharmacogenetic studies have indicated that certain polymorphisms in norepinephrine [
11] or serotonin transporter genes [
12], as well as dopamine receptor genes [
13] may be associated with responses to methylphenidate. A positron emission tomography study on healthy adults suggests that inter-individual variability in the amount of dopamine released by neurons associates with the degree to which dopamine at synaptic levels increases the following blockade of dopamine transporters by methylphenidate [
14]. In individuals with ADHD, methylphenidate also has been shown to increase striatal dopamine availability, which in turn may further affect the corticostriatal systems subserving ADHD symptoms and behaviors related to executive dysfunctions [
15]. However, brain phenotypes of individuals with ADHD who likely respond to methylphenidate remain elusive [
16].
Among the scarce published reports [
16], structural MRI studies yielded mixed findings that individuals with ADHD who are poor responders to methylphenidate appear to have thinner medial frontal lobe [
17], smaller corpora callosa white matter (WM) volumes [
18], smaller inferior posterior cerebellar volumes, greater caudate volumes and asymmetry [
19], and smaller caudate and accumbens volumes concentrations [
20]. Most of these studies consisted of limited sample sizes (N of ADHD < 30 in total) [
18‐
20] and did not report or address issues of in-scanner motion. Studies have confirmed that head motion in the MRI scanner would introduce inaccuracy when estimating gray matter (GM) volume and thickness [
21,
22]. In addition, most of the studies included participants with major psychiatric comorbidity [
17] or with a prior methylphenidate exposure [
17]. Co-occurring mental health issues per se have been shown to significantly affect treatment responses to methylphenidate [
15]. A meta-analysis of ADHD-associated brain structural alterations revealed that studies with a higher percentage of psychostimulant-treated participants tend to be associated with fewer differences in the striatum (specifically smaller volumes in ADHD) [
23]. This suggests that stimulant exposure would affect brain structures in ADHD. Interestingly, the findings from this meta-analysis also indirectly converge to show that structural correlates of responses to psychostimulant might involve the striatum. In sum, despite the inconsistency in directions [
19,
20] and methodological caveats, these studies suggest that poor responders are essentially characterized by altered striatal structures, among other mixed findings [
16]. Further, functional image studies show frontostriatal connectivity measured by resting-state functional MRI [
24], as well as striatal regional cerebral blood flow and its dopamine transporter binding estimated by single-photon emission computed tomography [
25] are related to treatment responses in individuals with ADHD.
In addition to the striatal regions and associated corticostriatal circuitries, the default-mode network (DMN) maybe another brain circuitry involved in methylphenidate effects and responses. The DMN comprises the posterior cingulate/precuneus, medial prefrontal cortex, and lateral inferior parietal cortex [
26]. It exhibits reduced activity when task-positive networks (mainly the fronto-parieto-striatal circuitries) activate in response to external tasks and is most active when people are engaged in internally-oriented and stimulus-independent cognition [
27]. ADHD is characterized by altered DMN connectivity [
28] and increased DMN-task-positive network connections [
28,
29]. Through increasing dopamine and norepinephrine, methylphenidate has been consistently reported to suppress DMN activities in patients with ADHD while engaging in cognitive tasks (i.e., reducing the DMN-task-positive network connection) [
30‐
33]. Reduction in DMN activity or connectivity with task-positive networks also is associated with methylphenidate-related improvements in core symptoms of individuals with ADHD [
34]. The medial prefrontal cortex, which is thinner in poor responders at baseline, as shown in the aforementioned naturalistic study [
17], partially corresponds to the one hub region of the DMN. Despite its convergingly essential role in the pathophysiology of ADHD and methylphenidate effects on brain function, the DMN has never been specifically investigated whether its structure and function are related to treatment response to methylphenidate in ADHD.
In this context, this study aimed to characterize the baseline brain structural correlates that distinguished good and poor responders to methylphenidate in medication-naïve patients with ADHD, who did not have major psychiatric comorbidities. Based on the mass-univariate analysis of voxel-based morphometry (VBM), we first employed a combination of both unbiased exploratory whole-brain and hypothesis-driven approaches targeting striatal and DMN regions. Further, we leveraged a multivariate pattern classification method that takes into account interactions between regions, and is capable of making predictions for individual subjects based on brain imaging patterns [
35]. This machine learning approach can complement group-level inferences from the preceding mass-univariate analysis [
36]. We hypothesized that the ADHD-poor responder group, relative to the good responder group, would have smaller baseline striatal volumes [
20,
23]. Taking a typical negative connection relationship between the DMN and frontostriatal task-positive networks [
27], as well as a notion that brain structural covariance results from functional connectivity [
37], we further hypothesized larger regional GM volumes in the regions within the DMN in the poor responder group at baseline. These brain patterns, among other regional structural information, would provide multivariate indicators to predict an individual’s medication response.
Discussion
Using a distinct design in a sample free from confounds of psychiatric comorbidities and medication exposure, we found that participants with ADHD with poor methylphenidate responses clinically had smaller regional volumes in the left putamen and larger precuneus volumes at baseline, compared with the good responders. Multivariate pattern recognition also identified that volumetric information among these two regions alongside the left frontoparietal regions occipital lobes and posterior/inferior cerebellum could differentiate between good and poor responders. This regional structural information, especially striatal volume, if replicated, might serve as a potential biomarker for methylphenidate responses in ADHD.
The finding of greater left putamen volumes in good responders was largely consistent with the prior study showing smaller striatal volumes concentration in individuals with ADHD-poor responders [
20]. Several PET studies have demonstrated that methylphenidate could increase striatal dopamine availability [
70,
71], and this mechanism is related to the binding to the dopamine transporter [
72]. Patients with ADHD with higher striatal dopamine transporter availability show a better response to methylphenidate treatment [
15]. Further, striatal dopamine receptor (D2) availability is positively associated with methylphenidate response [
73]. In parallel, larger GM volumes are believed to be related to an increased density of neurons and more synapses in the local brain region [
74]. The local neural density and the synapses also represent the reciprocal connections within the local brain region and clustering [
75]. Furthermore, earlier studies suggest a direct positive correlation between striatal GM volume and D2 receptors [
76,
77]. Taken together, the relationship between larger putamen volumes and a good methylphenidate response may be explained by that higher striatal volumes could indicate higher D2 receptor availability locally, which leads to an increase in the methylphenidate efficiency.
To the best of our knowledge, this study is the first work demonstrating the poor responder group, relative to the good responder group, had higher GM volumes of the precuneus, a hub of the DMN [
78]. The result echoes earlier evidence that methylphenidate influences DMN activities [
30‐
33] in patients with ADHD. This finding of larger precuneus echoes a similar finding from an earlier PET work. Namely, Tomasi et al. [
79] reported that dopamine modulates attention in part by regulating neuronal activity in the posterior parietal cortex, including the precuneus. Higher striatal dopamine transporter levels, which result in enhanced clearance of dopamine and weaker dopamine signals, are associated with lower deactivation in the DMN during an attention-requiring task [
79]. This suggests that the DMN deactivation, which should be normally remarkable to facilitate optimal performances during external cognition [
80], and could be enhanced by methylphenidate in individuals with ADHD, is mediated by striatal dopamine levels [
79]. Taken together, we speculate that lower putamen volumes, as shown in the poor responders, may have lower D2 and dopamine transporter availability, which results in less dopamine increased by methylphenidate. Given negative functional connectivity between precuneus and putamen [
81], as well as corticostriatal projections between the precuneus and putamen [
82], this lower level of striatal dopamine increase may lead to less dopamine available being used in the precuneus, contributing to less deactivation of the DMN as subserved by methylphenidate. The concomitant larger precuneus volumes in the poor responders might represent a compensatory process for the preceding mechanism. Combining the above mechanisms and our finding of a negative correlation between the volumes of precuneus and putamen, the pattern might be alternatively explained by the notion that structural covariance reflects brain functional connectivity and is resulted from direct structural connections through trophic effects [
37]. Specifically, the putamen-associated network [
83] has an anti-correlated functional relationship with the DMN/precuneus [
27,
81]. Future studies need to replicate the current findings and investigate such speculations.
Although the striatum and DMN were specifically targeted revealed in the univariate VBM analysis in this study, we note that methylphenidate-associated brain structural and functional changes also involve other brain systems and areas, which might be associated with treatment responses as well. For example, methylphenidate has effects on modulating the attention network [
84,
85] and normalizes activation of the dorsolateral inferior prefrontal cortex to improve attention [
86]. In addition, methylphenidate also could regulate brain activity in premotor cortices [
87], which may be associated with its beneficial effects on hyperactivity symptoms. Whether these other putative brain systems/regions are associated with responses to methylphenidate treatment warrants further investigation.
Our finding of the regional brain volume difference between different drug responders by using voxel-based morphometry was also supported by using a machine learning approach, which selected the precuneus and left putamen, endorsing the aforementioned mass-univariate findings. In addition to these two regions, the SVM classification identified that regions of the discriminative pattern most predictive of treatment responses were in the left frontoparietal regions, mid and posterior cingulum gyrus, occipital lobes, as well as posterior cerebellum. There is functional connectivity between the striatum and the posterior cingulate, middle/inferior frontal gyrus [
83], and structural connections between the striatum and cerebellum as well as the frontal gyrus, respectively [
88]. A human PET study demonstrated that the frontal area and cingulate gyrus are the regions of dopaminergic projection [
89]. In macaque monkeys, axons with dopamine transporter are presented in cerebellum lobules III and IX [
90]. Norepinephrine transporters distribute not only in high-density regions such as the thalamus and locus coeruleus but also in the low-density regions in the frontal, parietal, and occipital cortex [
91]. In these cerebral cortices, norepinephrine transporters are the major transporters for dopamine and norepinephrine reuptake [
92]. Altogether, these brain regions selected by multivariate pattern classification contained transporters involving in dopamine and norepinephrine reuptake, which is implicated in methylphenidate mechanisms. Moreover, striatum has direct structural and functional connections with most of these regions [
83,
88], which might synergistically contribute to mechanisms underlying responses to methylphenidate with striatum. The finding of the classification accuracy of over 85% based on brain structure measures are promising and, if replicated, suggest that it may be possible in the future to use machine learning-based pattern recognition analyses to aid in the classification of medical response before the application of methylphenidate for patients with ADHD.
Several limitations must be considered while interpreting the results. First, the drug response was evaluated through a retrospective chart review, which consisted of patients’ current progress and detailed medication profile that were sufficient for rating the CGI-I. But there were no details about further cognitive function or the life quality profile. Of note, this study also was limited by a lack of placebo-controlled design. However, this study is a follow-up analysis based on the cohort originally for the cross-sectional study purpose, and the two child psychiatrists who rated the CGI-I were blinded to the participants and their attending psychiatrists. This approach may account for some extents of this caveat. We acknowledge that a prospective longitudinal design may be a more robust approach, e.g., the MTA study [
93], to answer such research questions. Second, despite the present “pure” phenotype without confounding effects from psychotropic agents and co-occurring major psychiatric problems, readers need to notice the caveat of generalizability of our results based on such a sample recruited from one medical center in Taiwan. Third, we excluded participants who were lost to follow-up within 1 month of starting using methylphenidate. These patients with ADHD may be more likely to have poor clinical outcomes. However, the percentage of good responders herein was 80%, approximately the ratio of responses to methylphenidate reported before [
5], indicating that the current sample was representative of the general ADHD population. Fourth, despite the fact that cerebral morphometric alterations may be different between ADHD subtypes [
94], we did not undertake the subgroup analysis based on the subtype, given the limited sample size of the poor responder group. But there was no difference in ratios of subtypes between the good and poor responder groups (Table
1). Future relevant larger studies could benefit from ADHD subtyping analyses. Lastly, the study participants had a relatively wide age range. The mega-analysis using the cross-sectional ENIGMA dataset showed that ADHD had smaller putamen volume in participants with broader age ranges [
95]. The altered putamen volume in ADHD was unaltered with age development based on another large NeuroIMAGE sample [
96]. To balance the statistical power and difficulty recruiting such a medication-naïve and comorbidity-free sample, we still employed the current sampling approach. The linear and square terms of age was controlled in every model to minimize the confounding effect. However, we acknowledge that some age-related effects may not be excluded statistically. Future studies of a similar kind will need to take developmental issues into account.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.