Disrupted Small-World Networks in Children with Drug-Naïve Attention-Deficit/Hyperactivity Disorder: A DTI-Based Network Analysis

Abstract Attention-deficit/hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders, while the potential neurological mechanisms are poorly understood. To explore the alterations in the white matter (WM) structural connectome in children with drug-naïve ADHD, forty-nine ADHD and 51 age- and gender-matched typically developing (TD) children aged 6–14 years were enrolled. WM structural connectivity based on deterministic diffusion tensor imaging (DTI) was constructed in 90 cortical and subcortical regions, and topological parameters of the resulting graphs were calculated. Network metrics were compared between two groups. The concentration index and the total cancellation test scores of digit cancellation test were used to evaluate clinical symptom severity in ADHD. Then, a partial correlation analysis was performed to explore the relationship between significant topologic metrics and clinical symptom severity. Compared to TD group, ADHD showed an increase in the characteristic path length (Lp), normalized clustering coefficient (γ), small worldness (σ), and a decrease in the global efficiency (Eglob) (all p < 0.05). Furthermore, ADHD showed reduced nodal centralities mainly in the regions of default mode network (DMN), central executive network (CEN), basal ganglia, and bilateral thalamus (all p < 0.05). After performing Benjamini-Hochberg’s procedure, only the left orbital part of superior frontal gyrus and the left caudate were statistically significant (p < 0.05, FDR-corrected). In addition, the concentration index of ADHD was negatively correlated with the nodal betweenness of the left orbital part of the middle frontal gyrus (r = −0.302, p = 0.042). Our findings revealed an ADHD-related shift of WM network topology toward “regularization” pattern, characterized by decreased global network integration, which is also reflected by changed nodal centralities involving DMN, CEN, basal ganglia, and bilateral thalamus. ADHD could be understood by examining the dysfunction of large-scale spatially distributed neural networks.


Abstract
Attention-deficit/hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders, while the potential neurological mechanisms are poorly understood.To explore the alterations in the white matter (WM) structural connectome in children with drug-naïve ADHD, fortynine ADHD and 51 age-and gender-matched typically developing (TD) children aged 6-14 years were enrolled.WM structural connectivity based on deterministic diffusion tensor imaging (DTI) was constructed in 90 cortical and subcortical regions, and topological parameters of the resulting graphs were calculated.Network metrics were compared between two groups.The concentration index and the total cancellation test scores of digit cancellation test were used to evaluate clinical symptom severity in ADHD.Then, a partial correlation analysis was performed to explore the relationship between significant topologic metrics and clinical symptom severity.Compared to TD group, ADHD showed an increase in the characteristic path length (L p ), normalized clustering coefficient (γ), small worldness (σ), and a decrease in the global efficiency (E glob ) (all p < 0.05).Furthermore, ADHD showed reduced nodal centralities mainly in the regions of default mode network (DMN), central executive network (CEN), basal ganglia, and bilateral thalamus (all p < 0.05).After performing Benjamini-Hochberg's procedure, only the left orbital part of superior frontal gyrus and the left caudate were statistically significant (p < 0.05, FDR-corrected).In addition, the concentration index of ADHD was negatively correlated with the nodal betweenness of the left orbital part of the middle frontal gyrus (r = −0.302,p = 0.042).Our findings revealed an ADHDrelated shift of WM network topology toward "regularization" pattern, characterized by decreased global network integration, which is also reflected by changed nodal centralities involving DMN, CEN, basal ganglia, and bilateral thalamus.ADHD could be understood by examining the dysfunction of large-scale spatially distributed neural networks.

Introduction
Attention-deficit/hyperactivity disorder (ADHD) is one of the most common neurodevelopmental disorders characterized by age-inappropriate inattention and hyperactivity/ impulsivity.The estimated prevalence of ADHD is up to 9% in school-age children and may continue into adulthood [1].Increasing evidence suggested that brain structural and functional abnormalities may be associated with the development and progression of ADHD [2].However, the potential neurological mechanisms underlying ADHD are poorly understood.
Currently, whole-brain voxel-based analysis and tractbased spatial statistics analysis are used to examine white matter (WM) structural integrity in ADHD and widespread altered WM microstructural are mainly located in the corpus callosum, frontostriatal circuits, cingulum, and corticospinal tract [3][4][5].Nevertheless, the brain is a complex and integrative network [6], using an integrated approach to analyze the organizational framework of the brain, which provides a more comprehensive understanding of the whole-brain network abnormalities than traditional regional or voxel-based analysis.Graph theory analysis, in which the brain network is represented in the mathematical "graph" comprising nodes linked by edges, enables the study of a large-scale brain structural network.WM connectome obtained from diffusion tensor imaging (DTI) maps the structural connectivity between gray matter regions using WM tractography, is an essential tool to understand structural brain networks.Structural brain connectome based on DTI approach shows that human WM networks exhibit a "small-world" network, which has dense local connections and few long connections, making a balance between local and global structural characteristics [7,8].Thus far, disrupted WM networks were found in neuropsychiatric [9] or neurodevelopmental disorders, including pediatric ADHD [10,11]; however, those inclusion criteria of patients, inappropriate age distribution, as well as the reported results were heterogeneous.For example, Beare et al. [12] and Hong et al. [13] recruited ADHD participants treated with psychostimulants or other medications, but the stimulant medication may lead to changes in brain structure [14].Moreover, in the study of Beare et al. [12], the participants were 9-17 years, which is a period of dynamic brain change affected by pubertal hormones.Emerging evidence also indicates that the clinical symptoms of ADHD (mainly hyperactivity/impulsivity) tend to decrease with age [15,16], and some structural alterations observed in childhood ADHD may normalize with age [17].Considering these inconsistencies, we re-cruited drug-naïve pediatric ADHD aged 6-14 years, which is an age group that reflects the disorder-related abnormalities to the greatest extent.
Using graph theory analysis, this study aimed to evaluate brain WM structural connectome in drug-naïve pediatric ADHD from a macroscopic network perspective.Furthermore, the specific morphological network patterns of brain WM networks were discussed, and the associations between altered network metrics and the core behavioral symptom severity in ADHD were further explored.

Participants
This study was approved by the Institutional Review Board of the First Affiliated Hospital of Sun Yat-sen University (No. [2019] 328) and has been registered at https://clinicaltrials.gov/ (Identifier: ChiCTR2100048109).Written informed consent was obtained from the patients' parents or guardians.
Drug-naïve ADHD participants without psychiatric comorbidities confirmed by two experienced pediatricians were recruited from April 2019 to March 2020.The patients were diagnosed with ADHD according to the Diagnostic and Statistical Manual of Mental Disorders' criteria (4th edition [DSM-4]), as well as the parent and teacher reports on Conners Symptom Questionnaire (Conners CK.Conners 3rd Edition; Toronto: Multi-Health Systems; 2008).The parent and teacher ADHD indices ≥75th percentile was considered ADHD-positive.The typically developing (TD) volunteer children were established to be without the diagnosis of ADHD based on the same reports, with both parent and teacher ADHD indices <75th percentile.The TD volunteer children had no first-degree relatives with a known history of psychiatric illness.Meanwhile, recruited TD children were matched for age, gender, and right-handedness.
The exclusion criteria were (1) those with a history of any psychotic and/or mood disorder and/or current medication with psychoactive drugs; (2) with a history of seizure and/or loss of consciousness for >5 min; (3) with a history of brain injury or surgery of head; (4) contraindication to MRI; (5) MRI data with scanning artifacts and motion artifacts manually excluded by two experienced radiologists.
The digit cancellation test is taken to assess individual attention concentration levels based on the time and accuracy of completing the task referring to previous studies [18][19][20].Then, total cancellation test scores and the concentration index were obtained from digit cancellation test in ADHD group.

Image Acquisition
MRI examinations of all participants were performed on a 3.0T scanner (SIGNA Pioneer GE Healthcare, WI, USA) using 32channel head coils.A coronal T2-weighted sequence was collected to rule out any cranial organic lesion.The sagittal threedimensional T1-weighted fast spoiled gradient echo-based sequence (T1-FSPGR) with 1.00 mm isotropic resolution and DTI scan were performed for each subject.DTI scans were obtained using a single-shot echo-planar imaging sequence.Images were acquired axially parallel to the anterior-posterior commissure (AC-PC) to cover the entire brain with the following parameters: TR/TE = 10,000/88.6ms; FOV = 25.6 cm; matrix size = 128 × 128; voxel size = 2 mm 3 ; NEX = 1; number of directions = 32; b value = 1,000 s/mm 2 .
Data Processing and WM Network Construction PANDA software (https://www.nitrc.org/projects/panda/)was used to conduct data preprocessing and brain WM network construction.Preprocessing steps of each fractional anisotropic (FA) map included (1) skull removal with brain extraction tool; (2) eddy correct; (3) DT fitting or building DT models and obtaining the fractional anisotropic maps (DTIFIT); (4) nonlinear registration to Montreal Neurological Institute (MNI) space with a voxel size of 2 × 2 × 2 mm 3 .In the current study, whole-brain WM tracts were reconstructed by deterministic fiber tractography, which was continued until either an angle >45°was reached or the FA was <0.2 using the fiber assignment by continuous tracking algorithm [21].
The automated anatomical labeling (AAL) atlas divided the whole brain into 90 cortical and subcortical regions [22], and nodes were defined as brain regions using AAL atlas.First, each T1-weighted image was co-registered to individual b0 images in native diffusion space using a linear transformation, and then nonlinearly mapped to MNI space.Next, the derived transformation parameters were inverted and used to warp the AAL atlas from MNI space to the native diffusion space, in which the discrete labeling values were preserved by using a nearest neighbor interpolation method.Next, the averaged FA of linking fibers was calculated to define the network's edges.Finally, weighted and undirected symmetrical anatomical 90 × 90 matrices were obtained for each participant.
The area under the curve was calculated over the sparsity range from S 1 = 0.01 to Sn = 0.40 with an interval of ΔS = 0.01 for each network metric.This approach provides a summarized scalar for the topologic characterization of brain networks independent of a single threshold selection, which can minimize the effects of discrepancies in the overall correlation strength between groups and enables exploration of between-group differences in relative network organization [21].The between-group differences of nodal profiles were presented using the BrainNet Viewer toolbox (https://www.nitrc.org/projects/bnv/).Statistical Analysis SPSS v25.0 (IBM Corp., Armonk, NY, USA) was used to compare clinical characteristics.Independent-sample t tests were used to compare quantitative variables, and χ 2 tests were used to compare qualitative variables.A p value < 0.05 was considered to be statistically significant.
Nonparametric permutation testing was performed on the area under the curve of each network metric to assess between-group differences [23].Partial correlation analyses were used to examine relationships between significant topologic metrics and clinical symptom severity, with age and gender used as the controlled covariance.A p value <0.05 was considered to be statistically significant.The Benjamini-Hochberg false-discovery rate (BHFDR) method was used to control for the error of multiple comparisons.

Demographic and Clinical Comparisons
Demographic and clinical features are shown in Table 1.There were no significant differences among the two groups in age, sex, or years of education (all p > 0.05).

Global Properties of the WM Networks
Both ADHD group and TD group have the smallworld topology of the WM brain networks (σ > 1) in the defined threshold range.However, compared to TD group, ADHD showed increased L p (p = 0.019), γ (p = 0.024), and σ (p = 0.029), while there were no significant differences in the C p (p = 0.565) and λ (p = 0.414).For network efficiency, the ADHD group showed decreased E glob (p = 0.019) compared with the TD group, with no significant differences in the E loc (p = 0.657) (shown in Fig. 1).

Nodal Properties of the WM Networks
According to the AAL-90 atlas, compared to the TD group, the ADHD group showed altered nodal profiles mainly in 4 categories: (1) the default mode network (DMN) that included bilateral dorsolateral of superior frontal gyrus (SFGdor), bilateral parahippocampal gyrus, bilateral precuneus, bilateral gyrus rectus, and right posterior cingulate gyrus (PCG.R); (2) the central executive network (CEN) that included the bilateral orbital part of superior frontal gyrus (ORBsup), the left orbital part of middle frontal gyrus (ORBmid.L) and the left triangular part of inferior frontal gyrus (IFGtriang.L); (3) the basal ganglia that included bilateral caudate nucleus, right putamen (PUT.R), and right pallidum (PAL.R); and (4) bilateral thalamus (THA).Altered nodal profiles were also seen in the visual network-related right superior occipital gyrus (SOG.R) and salience network (SN)-related left amygdala (AMYG.L) (shown in Table 2; Fig. 2).There were no significant differences in other nodal properties.However, after performing Benjamini-Hochberg's procedure, only left ORBsup and left caudate were statistically significant (p < 0.05, FDRcorrected).
Disrupted Small-World Networks in ADHD Data are represented as the mean ± SD.For comparisons of demographics, p values are obtained using two-sample t test or χ 2 test; p < 0.05 was considered significant.ADHD, attention-deficit/hyperactivity disorder; TD, typically developing; SD, standard deviation.

Correlations between Network Metrics and Clinical Variables
The correlation analysis was used to explore the inner relationship between the significant network metrics of WM structural brain networks and symptom severity in ADHD.Nodal betweenness of the ORBmid.L was negatively associated with the concentration index (r = −0.302,p = 0.042, shown in Fig. 3) in ADHD.However, no significant correlations were found between other topological global/nodal properties and clinical variables.

Main Findings
The present study used graph-theoretical analyses to examine the topological organization of the WM brain networks in drug-naïve ADHD patients without any psychiatric comorbidity.The main findings of this study are (i) at the global level, ADHD showed decreased global network integration, i.e., a shift to a "regularization" pattern; (ii) at the nodal level, altered nodal profiles were mainly located in DMN, CEN, basal ganglia, and bilateral THA, which are known to be involved in ADHD; (iii) the nodal betweenness of ORBmid.L was negatively correlated with the symptom severity in ADHD.These findings suggest that ADHD could be understood by examining the dysfunction of large-scale spatially distributed neural networks.

Disrupted Small-World Networks in ADHD
The human brain network is typically organized as a small-world network that features two main organizational principles: segregation, reflected by clustering coefficient (C p ), normalized clustering coefficient (γ) or local efficiency (E loc ), and integration, reflected by a shorter path length (L p ), normalized characteristic path length (λ), and global efficiency (E glob ).A normal neocortex and different areas in the brain have the small- Disrupted Small-World Networks in ADHD world structure [24].However, several studies suggested that the changes of the small-world can be explored in various neurological and psychiatric disorders, including schizophrenia [25], bipolar disorder [9], major depression disorder [26], and pediatric ADHD [10,11].
Our study demonstrated a similar small-world structure in both ADHD and TD groups (σ > 1).However, compared to the TD group, the ADHD group showed lower E glob and higher L p, γ, and σ.Lower E glob and higher L p means that the ability of integration and the efficiency of information transfer in parallel over the whole brain network was decreased.This means that the WM network of ADHD presented a less optimized topological organization and a tendency to shift toward a closer-to-regular network, characterized as a high-cost, low-profit manner [8,27].The current morphological WM network findings are also consistent with previous DTI-based structural brain network studies [5,28], which showed a shift toward a regularization pattern ("weaker small worldness") in ADHD [29].Previous functional connectome studies of ADHD children reported alterations of small-world topological properties [30].Based on functional and structural network studies, ADHD could be understood by the dysfunction of large-scale spatially distributed neural networks, especially in wholebrain insufficient integration.

Alterations in Regional Topology Metrics
In the present study, patients with ADHD showed altered nodal profiles, mainly in the DMN, CEN, basal ganglia, and bilateral THA, which are known to be involved in ADHD [31].The well-known triple network model hypothesis of psychopathology suggests that the CEN, DMN, and SN are the core neurocognitive networks [32] which have been reported to be greatly involved in ADHD [33].Resting-state functional connectivity analysis [31,34] revealed altered connectivity in the DMN and frontoparietal network (also known as CEN) but also in reward-related (mainly corpus striatum) [35] and affective circuits [34] in ADHD, which is consistent with our findings.
Interestingly, we found that ADHD showed hyperand hypo-connectivity of the DMN, which appeared in different models in the previous studies.Although DMN shows higher activity and stronger functional connectivity during the resting state, it is also used as a buffer against the disruption of attentional processes from external stimuli [36].Decreased nodal efficiency of the DMN likely contributes to inadequate suppression for disruption and leads to inattention symptoms.However, increased nodal efficiency of the DMN may be due to inadequate suppression as well as hyper-activation [37].These might suggest that the abnormal transition of DMN leads to momentary attention lapses in ADHD.
Altered nodal metrics in ADHD were also found in basal ganglia and bilateral THA.Structural neuroimaging results of ADHD have demonstrated that the most prominent deficit regions were subcortical regions, including the basal ganglia and insula [3][4][5].Furthermore, the basal ganglia and bilateral THA are the core components of the cortico-striatal-thalamo-cortical (CSTC) loops, which involve impulsivity, cognitive, and limbic (or affective) [38].In principle, decreased nodal profiles in basal ganglia could cause CSTC loop-modulated dysfunction, leading to the behavioral symptoms of ADHD.Overall, our results reemphasized these subcortical regions as the key pathophysiology of ADHD.

Significant Relation between Nodal Network Profiles and Clinical Variables
In this study, decreased nodal betweenness in the ORBmid.L was observed in ADHD and was negatively associated with the clinical symptom severity.ORBmid is the key structure of the orbitofrontal cortex (OFC), involved in the regulation of cognition, emotion, and attention.OFC was reported to mediate attentional shifting by reducing interference from salient stimuli [39].Recent studies showed that the OFC is related to various brain regions of attention modulation according to functional connectivity [40], and the orbitofrontal abnormalities resulted in significant attentional deficits [41].Therefore, we inferred that altered nodal properties of ORBmid.L lead to the abnormal dysregulated top-down control of the prefrontal cortex, which in turn leads to the failure of top-down control mechanisms associated with abnormal attention in patients with ADHD.
In conclusion, our findings complement and extend the existing research and provide a structural network mechanism for pediatric ADHD.Furthermore, the observed abnormalities suggest that the pathogenesis of ADHD can be explained by the dysfunction of large-scale spatially distributed neural networks.

Conclusions
Our findings complement and extend the existing research and provide a structural network basis for functional network alterations of pediatric ADHD.Furthermore, these abnormalities suggest that the pathogenesis of ADHD can be explained by the dysfunction of large-scale spatially distributed neural networks.Thus, the disrupted structural network identified in this study, without the potential confound of medications and psychiatric comorbidities, provides novel insights into the pathophysiology of ADHD.

Study Limitation
This study has a few limitations.First, this was a single-center study with small sample size.Second, although the two groups did not significantly differ in sex distribution, we could not achieve a gendermatching population for each group (the number of girls in the overall sample was very limited).Thus, a larger sample size is needed to examine whether there are differences in WM topological properties between males and females.Third, the study is a cross-sectional analysis; thus, we did not discuss the developmental changes in brain network metrics.Fourth, the global properties and correlation analysis did not survive after FDR correction, perhaps because of our relatively small sample size.In the future, multi-center studies with a large sample size combined with structural and functional networks may facilitate more comprehension of the neurobiological in ADHD.

Statement of Ethics
This study was approved by the Institutional Review Board of the First Affiliated Hospital of Sun Yat-sen University (No. [2019] 328) and has been registered at https://clinicaltrials.gov/ (Identifier: ChiCTR2100048109).Written informed consent was obtained from patients' parents or guardians.

Fig. 3 .
Fig. 3. Relationship between the nodal betweenness of the left orbital part of the middle frontal gyrus (ORBmin.L) and the concentration index in the ADHD group.The results of correlation analysis between the concentration index (x-axis) and nodal betweenness of the left orbital part of the middle frontal gyrus (r = −0.302;p = 0.042).

Table 1 .
Demographic and clinical characteristics of participants

Table 2 .
Regions showing altered nodal metrics in the ADHD group and the TD group All p values were obtained using a permutation test.The brain regions were defined by AAL.L, left; R, right; CEN, central executive network; DMN, default mode network; SN, salience network; VN, visual network; ADHD, attention-deficit/hyperactivity disorder; TD, typically developing.*Uncorrected p < 0.05; **p < 0.05 after performing the Benjamini-Hochberg false discovery rate correction.