Resting-state functional and structural MRI data of ADHD patients (158 subjects) were obtained from the New York University Child Study Center for the ADHD-200 Global Competition and UCLA dataset (Bilder et al., 2016). The NYU dataset comprises 45 child ADHD patients and 73 adolescent ADHD patients, and the UCLA dataset is made up of 40 adult ADHD patients. Both datasets are made open to researchers online. For child and adolescent groups, psychiatric diagnoses were based on evaluations with the Schedule of Affective Disorders and Schizophrenia for Children—Present and Lifetime Version (KSADS-PL) administered to parents and children and the Conners’ Parent Rating Scale-Revised, Long version (CPRS-LV). Intelligence was evaluated with the Wechsler Abbreviated Scale of Intelligence (WASI). Inclusion in the ADHD group required a diagnosis of ADHD based on parent and child responses to the KSADS-PL as well as on a T-score greater than or equal to 65 on at least one ADHD related index of the CPRS-R: LV. Psychostimulant drugs were withheld at least 24 h before scanning. Estimates of FSIQ above 80, right-handedness, and absence of other chronic medical conditions were required for all participants”. For the adult group, participants were men or women ages 21–50 years; NIH racial/ethnic category either White, not Hispanic or Latino; or Hispanic or Latino, of any racial group; primary language either English or Spanish; completed at least 8 years of formal education; no significant medical illness; adequate cooperation to complete assessments; visual acuity 20/60 or better; and urinalysis negative for drugs of abuse (Cocaine; Methamphetamine; Morphine; THC; and Benzodiazepines). ADHD criteria were assessed using Adult ADHD Interview and excluded anyone with other disorders; stable medications were permitted for patients. For MRI studies we excluded participants who were left handed, who believed they might be pregnant or had other contraindications to scanning (e.g., claustrophobia, metal in the body, body too large to fit in the scanner). Diagnoses followed the DSM-IV—Text Revision and were based on the Structured Clinical Interview for DSM-IV (SCID-I) supplemented by the Adult ADHD Interview (a structured interview form derived from the KSADS-PL, in order to enable a more detailed characterization of lifetime history of ADHD in adults. Interviewers/raters were trained according to the criteria; in brief minimum standards of acceptable symptom agreement were overall kappa of 0.75, a kappa specificity of 0.75, and sensitivity of 0.75, and 0.85 kappa for diagnostic accuracy. Diagnostic and Symptom elicitation skill was also assessed with the SCID Checklist of Interviewer Behaviors and the Symptom Checklist of Interviewer Behaviors. Ongoing quality assurance checks documented kappa above 0.75 for each rater annually during the course of the study. All participants used in the current study were diagnosed with ADHD, and their scores have been used for the correlation analysis in the present study. Detailed information about the subjects can be found in Table 1.

For the adult dataset, MRI data were acquired on one of two 3T Siemens Trio scanners, located at the Ahmanson-Lovelace Brain Mapping Center (Siemens version syngo MR B15) and the Staglin Center for Cognitive Neuroscience (Siemens version syngo MR B17) at UCLA. Functional MRI data were collected using a T2*-weighted echo-planar imaging (EPI) sequence with the following parameters: slice thickness = 4 mm, slices = 34, TR = 2 s, TE = 30 ms, flip angle = 90°, matrix 64 × 64, FOV = 192 mm. A T2-weighted matched-bandwidth high-resolution anatomical scan (with the same slice prescription as the fMRI scan) and MPRAGE were collected. The parameters for the high-resolution scan were: 4 mm slices, TR/TE = 5,000/34 ms, 4 averages, matrix = 128 × 128, 90-degree flip angle. The parameters for MPRAGE were the following: TR = 1.9 s, TE = 2.26 ms, FOV = 250 mm, matrix = 256 × 256, sagittal plane, slice thickness = 1 mm, 176 slices. For the child and adolescent datasets, MRI data were obtained using Siemens Magnetom Allegra Syngo Mr 2004a. FMRI data were collected using an echo-planar imaging sequence with the following parameters: slice thickness: 4 mm, Slices: 33, TR: 2,000 ms, TE: 15 ms, Rotation = 90°, FoV phase: 80.0%, FoV read = 240 mm. In addition, T1-weighted images were acquired using the following parameters: Slice thickness = 1.33 ms, TR = 2,530 ms, TE = 3.25 ms, rotation = 0 degrees, FoV phase = 100.0%, FoV read = 256 mm.

To determine the gray matter alterations in ADHD, One-way Anova as implemented in SPM12 was performed on the smoothed GM images of the three ADHD age groups. Age, gender, and total intracranial volume (TIV) were used as covariates for statistical analyses. Furthermore, given that our data were obtained from different sites, and several studies have shown the effect of multi-site in different ADHD age groups (Hong et al., 2017; Zhou et al., 2019), we regressed the effects of the site in our analyses.

As mentioned in section 2.3 above, One-way Anova computed as implemented in SPM12; in brief, the smoothed gray matter images of all three ADHD age groups were loaded onto the SPM12 toolbox using the One-way Anova option. Gender, age, TIV and sites were used as covariates, and Family-wise error rate (FWE) correction was employed for multiple comparison corrections. In addition, we examined the relationship between altered regions and symptomatic values; we performed correlation analysis between regions that show significant gray matter differences and disease severity. To this end, we defined a sphere ROI (radius = 7 mm) using MarsBaR toolbox⁴. We then applied the general linear model (GLM) across all subjects, and using the extracted ROIs; parametric values were obtained. These values were subsequently used as independent variables against the disease severity using MATLAB partial correlation option. Age, gender, and site were regressed as covariates. Two subjects did not have disease severity scores and were excluded from the correlations analysis. FWE was used for multiple comparison corrections for all statistical analyses.

We use the signed path coefficient-based GCA implemented in the REST toolkit⁵. GCA was performed on a voxel-wise basis for all the voxels. The right pallidum region with Montreal Neurologic Institute coordinates (18, 2, -2 with radius = 7 mm) was selected from a one-way Anova test of a whole-brain voxel-based morphometric comparison among all three (3) ADHD patient groups and correlated with disease severity in the current study. We, therefore, selected this region for the GCA analysis. The order was set to one (1), and GCA was Z-transformed. GCA from the seed region (right pallidum) to the whole-brain (Outflow) and from the whole-brain to the seed region (Inflow) were measured. We further investigated the causal effect among significant whole-brain areas identified in GCA by calculating an ROI-ROI causality for outflow and inflow. The signed-path coefficient GCA was performed to build an ROI causal network that characterized the causal relationships among the ROIs. Age, gender, site, and mean framewise displacement (mean FD) were regressed out as covariates during the analyses. FWE (P < 0.05) was used for multiple comparison corrections.

One-way Anova was used to determine the significant gray matter volume differences among the three ADHD age groups. Regions showing significant differences in gray matter volume include: the bilateral pallidum, bilateral thalamus, bilateral insula, Temporal_Sup_L, and the right cerebellum (P < 0.05, FWE corrected), as demonstrated in Figure 1 and Supplementary Table S1. For the correlation analysis between significant gray matter volume regions and disease severity, only the right pallidum was found to show a significant relationship (positive) with disease severity (Figure 2).

Using the right pallidum as seed, significant causal effect from the seed region to the occipital_Mid_R, bilateral fusiform, postcentral_L, Paracentral_Lobule_Left, Amygdala_L, and the right cerebellum was observed (P < 0.05, FWE corrected). The results are shown in Figure 3A and Supplementary Table S2. Also, from Figure 3B and Supplementary Table S3, significant causal effects from the anterior cingulate cortex, cerebellum_Crus2_L, putamen_L, caudate_L, bilateral superior temporal pole, middle cingulate cortex, prefrontal cortex, precentral_R, and the Supp_Motor_Area_L to the seed region (right pallidum) was observed (P < 0.05, FWE corrected).

Regions that showed significant causality from the seed to significant whole-brain regions and from whole-brain regions to the seed were extracted and used for the ROI-ROI GCA. Figure 4 represents the results from the ROI-ROI GC analysis. The results showed directional networks revealing interregional causal relationships among the ROIs. For outflow ROI-ROI GC analysis (Figure 4A), the amygdala, postcentral gyrus, and paracentral lobule mainly projected causal effects to the fusiform gyri and cerebellar regions. The middle occipital gyrus and right cerebellum_4_5 projected causal effects towards each other. For the inflow ROI-ROI GCA (Figure 4B), the cingulate cortex projected causal effects towards the other ROIs and received causal effects from the superior temporal gyri. Also, the Cingulate cortex, middle frontal gyri, basal ganglia, precentral gyrus, SMA, and the right superior temporal gyrus projected causal effects towards the left cerebellum crus2, which served as a causal target.

The significant gray matter volume differences observed in ADHD patients in the current study were found in the bilateral pallidum, bilateral thalamus, bilateral insula, superior temporal cortex, and the right cerebellum. Reduced globus pallidus volume in ADHD patients has been reported in several previous ADHD studies (Frodl and Skokauskas, 2012; Hoogman et al., 2017). The pallidum regulates voluntary movement (Gillies et al., 2019); specifically, hyperkinesia is among the pronounced characteristic symptoms in ADHD patients (Sempere-Tortosa et al., 2021); that is, damage to the pallidum, as shown in this study, implicates movement disorder. Also, the correlation analysis indicating a positive relationship between the right pallidum and disease severity; implied that the higher the reduction of gray matter volume in the right pallidum, the higher the disease severity. Consistent with previous studies (Mills et al., 2012), the thalamus has been found to be implicated in ADHD patients. The thalamus plays several crucial roles, such as regulating alertness and consciousness and relaying sensory and motor signals between the brain and the body (Torrico and Munakomi, 2022). Defects in this region may complicate motor, sensory and cognitive functioning in ADHD patients. The insula cortex, which is part of the salience network, has been linked with symptoms of ADHD and plays an essential role in affective and cognitive processes (Menon and Uddin, 2010; Janes et al., 2018). Consistent with our findings, previous studies have found structural deficits and functional abnormalities in the insula of ADHD patients. Lopez-Larson et al. (2012) reported decreased gray matter volume in the anterior bilateral insula, associated with attention and inhibition problems. Some recent works also showed reduced functional connectivity between the amygdala and insula interaction in regulating emotion in ADHD children (Hulvershorn et al., 2014; Yu et al., 2020); these findings further show the implication of the insula in ADHD. The superior temporal gyrus, one of the gyri of the temporal lobe, has been reported to exhibit reduced activation and overlap with decreased gray matter volumes in the putamen and anterior insula in ADHD patients (Norman et al., 2016). The superior temporal gyrus is involved in language and mediates spatial awareness and exploration (Karnath, 2001); hence alterations in the gray matter volume of this brain region in the current study might imply language and spatial awareness impairments in ADHD patients. Consistent with previous studies, the cerebellum also exhibited defects in gray matter volume in the present study. Pathogenesis of ADHD is associated with abnormalities in the function and morphology of fronto-striatal brain regions (Durston et al., 2011). However, recent studies suggest that structural and functional abnormalities in ADHD are not limited to the fronto-striatal areas but to other regions, including the thalamus (Ivanov et al., 2010; Mills et al., 2012), the cerebellum (Valera et al., 2007), among others. The cerebellum is involved in motor control and higher cognitive processes, including learning (Steinlin, 2008), attention shifting (Golla et al., 2005), visual-spatial processing (Ivry et al., 2002), working memory (Cooper et al., 2012; Stoodley et al., 2012), and emotion (Ferrucci et al., 2012). Accordingly, the deficit of the cerebellum may suggest impairments in cognitive control and attentional mechanisms in ADHD patients.

For effective connectivity, the right pallidum as a seed demonstrated causal effects (outflow) on the Occipital_Mid_R, bilateral fusiform, Postcentral_L, Paracentral_Lobule_L, Amygdala_L, and the right cerebellum indicating that the right pallidum precedes and allows prediction of these brain regions. Also, the anterior cingulate cortex, prefrontal cortex, cerebellum_Crus2_L, Putamen_L, Caudate_L, bilateral superior temporal pole, middle cingulate cortex, Precentral_R, and the Supp_Motor_Area_L demonstrated causal effects on the seed region (inflow), indicating that these brain regions precedes and allow prediction of the right pallidum. The ROI-ROI GCA showed a directional network for both outflow and inflow. The outflow ROI directional network demonstrated that the right pallidum was its pivot. Disruptions in the right pallidum might be related to disruptions in other brain areas in the outflow connectivity graph (Figure 4A). The amygdala was identified as the transition point and received causal effects from the right pallidum and projected to other regions. The right fusiform gyri and the cerebellar regions received more causal effects from the other regions and were identified as the causal targets. The ACC was identified as the core of the inflow ROI directional causal relationships (Figure 4B). Changes in the ACC potentially conferred causal effects to all other regions. Consistent with previous ADHD studies (Frodl and Skokauskas, 2012; Bledsoe et al., 2013), the ACC has been reported to be implicated in ADHD. More projections were made from the frontal cortex and basal ganglia regions to the left cerebellum, suggesting frontal-striatal-cerebellar connectivity, reported in ADHD (Giedd et al., 2001). The cerebellum has been identified as a causal target in both inflow and outflow ROI causal relationships; this further provides evidence of the involvement of the cerebellum in ADHD pathophysiology. The GCA captured the effect of the abnormality of the right pallidum on other regions, the effects of the abnormalities in other regions on the right pallidum, and the directional causal relationships among these brain regions.

The occipital cortex is responsible for visual processing and has been reported in previous ADHD studies (Wolf et al., 2009; Sokunbi et al., 2013); therefore, the seed’s prediction of this brain region indicates the implication of the occipital cortex in ADHD. Consistent with previous studies (Hoogman et al., 2019), the fusiform also showed activation in the current study. This activation of the bilateral fusiform was caused by the neural activity in the right pallidum. The occipital cortex and the fusiform gyri are part of the visual network, suggested to be responsible for processing information about static and moving objects, spatial awareness and guidance of action, and object recognition (James et al., 2003; Nassi and Callaway, 2009) and have been reported to show abnormal connectivity in several ADHD studies (Benli et al., 2018; Agoalikum et al., 2021). Our results also showed that the neural activity in the seed region could predict the neural activity in the left postcentral gyrus and the left paracentral lobule, which are part of the somatosensory system reported to be implicated in ADHD (Parush et al., 1997). The somatosensory system is associated with touch, nerve pathways, and parallel receptors for the sensation of pain, temperature, body position, and movement (Sherman, 2019). The somatosensory cortex is also said to be involved in emotional regulation (Kropf et al., 2019), which is suggested to be associated with ADHD (Retz et al., 2012; Corbisiero et al., 2013). The amygdala is said to be involved in emotional regulation especially negative emotions (Bonnet et al., 2015); hence, the activation of the left amygdala caused by the right pallidum might suggest emotional impairment in ADHD patients. Previous structural and functional studies have reported abnormal connectivity in the cerebellum of ADHD patients (Valera et al., 2007; Agoalikum et al., 2021). Deficits in gray matter volume and cerebellum activation further provide evidence of the cerebellum’s implication in ADHD. The cingulate cortex, especially the anterior cingulate cortex (ACC), has been reported to show symptom severity and dysfunction in attention, motor control, and volumetric reductions in ADHD patients (Makris et al., 2007; Bush et al., 2008; Frodl and Skokauskas, 2012; Bledsoe et al., 2013; Sanfratello et al., 2019), hence the ACC predicting the right pallidum in the current study provides further evidence of the implication of this region in ADHD. The prefrontal cortex is essential in cognitive control functioning, thereby regulating attention and behavior via its extensive connectivity to the motor and sensory cortices, as well as subcortical regions like the cerebellum and basal ganglia (Arnsten, 2009). The prefrontal cortex, cerebellum, and the basal ganglia regions predicted the right pallidum (seed) in the current study; this is consistent with the frontal-striatal-cerebellar circuits reported in ADHD (Giedd et al., 2001), suggesting that the prefrontal cortex and its extensive connections are associated with ADHD symptoms. The Right precentral and left supplementary motor areas which are part of the sensorimotor network (SMN), also predicted the seed region in the current study. The SMN is involved in the coordination of sensory and motor actions and has been associated with hyperactivity in ADHD patients (Sörös et al., 2019). Our results also showed that neural activity in the bilateral superior temporal pole could predict the neural activity in the right pallidum, which shows defects in gray matter volume in the current study. The temporal pole is involved in functions such as semantic processing, memory, and emotional and social behavior, among others (Córcoles-Parada et al., 2019), which may be associated with ADHD.

Although we showed the structural differences in ADHD and the causal relationship between the right pallidum and whole brain, as well as among outflow and inflow ROIs in the three ADHD patient groups, the current study has some limitations. Albeit the control of sex variable in the statistical analysis, the male to female ratio in the current study was not the same, which may affect the results. Also, the sample size was a limitation in this study, as connectivity tends to be more stable with an increasing number of participants. We were not able to investigate whether our relatively small sample influenced the result presented, we suggest that future studies recruit larger sample sizes to include a balanced sex ratio.

In conclusion, the current study investigated the structural differences among ADHD patients and the causal relationship between the right pallidum and the rest of the brain and among significant outflow and inflow regions, using VBM and coefficient-based GCA. The significant gray matter volume defects in the right pallidum were positively correlated with disease severity. Also, significant outflow and inflow regions obtained from the GC analysis revealed directional causal relationships among ROIs. Generally, the current study demonstrated structural differences among the three ADHD patient groups and the ability of GCA to capture the causal connectivity between brain regions. Furthermore, this work highlights the evidence of the frontal-striatal-cerebellar circuits in ADHD and provides new insights into the effective connectivity of the right pallidum and the pathophysiology of ADHD.