Disrupted signal variability of spontaneous neural activity in children with attention-deficit/hyperactivity disorder

: Brain signal variability (BSV) has shown to be powerful in characterizing human brain development and neuropsychiatric disorders. Multiscale entropy (MSE) is a novel method for quantifying the variability of brain signal, and helps elucidate complex dynamic pathological mechanisms in children with attention-deficit/hyperactivity disorder (ADHD). Here, multiple-channel resting-state functional near-infrared spectroscopy (fNIRS) imaging data were acquired from 42 children with ADHD and 41 healthy controls (HCs) and then BSV was calculated for each participant based on the MSE analysis. Compared with HCs, ADHD group exhibited reduced BSV in both high-order and primary brain functional networks, e.g., the default mode, frontoparietal, attention and visual networks. Intriguingly, the BSV aberrations negatively correlated with ADHD symptoms in the frontoparietal network and negatively correlated with reaction time variability in the frontoparietal, default mode, somatomotor and attention networks. This study demonstrates a wide alternation in the moment-to-moment variability of spontaneous brain signal in children with ADHD, and highlights the potential for using MSE metric as a disease biomarker. ADHD core symptoms and cognitive function. The results revealed decreased BSV in children with ADHD compared to typically developing controls and decreased brain variability in ADHD was accompanied by increased ADHD core symptoms and disrupted cognitive function. Our results provide novel insights and potential biomarkers for the diagnosis of ADHD. Replications of these experiments in larger, independent population samples with wider age ranges are needed to confirm our preliminary findings.


Introduction
Attention-deficit/hyperactivity disorder (ADHD) is the most common neurodevelopmental disorder during childhood and adolescence, with approximately 7.2% of school-aged children meeting diagnostic criteria [1]. The children with ADHD commonly have some typical symptoms of inattention, hyperactivity and impulsivity [2], and the symptoms continue into adulthood in approximately 50% to 60% [3]. The exploration of neuropathological mechanism and development of ADHD is of significant importance.
Neuroimaging technique provides a potential tool for exploring brain function disorder in children with ADHD. Accumulating evidence has suggested dysfunction in some important brain functional connectivity and network topology in the children with ADHD. For example, Zhan et al. found the disconnection between visual and the other brain regions in children with ADHD [4]. Furthermore, it has also been found that the ADHD patients exhibited decreased global efficiency [5] and increased local efficiency [6] compared to healthy controls (HCs). These studies, from the views of large-scale network, demonstrate that brain functional organization is Scale for Children [33] ≥ 80; 3) medication-naïve (stimulant and non-stimulant medications) and free of other medical interventions. Patients with a diagnosis of schizophrenia, pervasive developmental disorders, bipolar disorder, epilepsy, mental retardation, or other brain disorders were excluded. It should be noted that this study only focused on boys with ADHD considering that ADHD is more frequently prominent in males with a male-to-female ratio of 4:1 in clinical samples [34].
Forty-one age-matched healthy control boys (aged 8 to 11 years old, Mean ± SD = 9.5 ± 0.8) were enrolled from a primary school in the local community. Individuals with ADHD, mania, bipolar disorder, other major psychiatric disorders, family history of psychosis, and severe physical diseases were excluded.
This study was approved by the Medical Research Ethics Committee of Peking University Sixth Hospital. Written informed consent was obtained from the parents of all participants (and participants themselves if they were at least 10 years old) before the experiment.

Diagnoses and assessment
In addition to the clarification of an ADHD diagnosis, the CDIS was also used to assess ADHD subtypes and evaluate comorbidities. Among the recruited boys with ADHD, 22 (52.4%) met the criteria for the predominantly inattentive subtype (ADHD-I), and 20 (47.6%) met the criteria for the combined subtype (ADHD-C) ( Abbreviations: ADHD, attention-deficit/hyperactivity disorder; HCs, healthy controls; SD, standard deviation; IQ, intelligence quotient; ADHD-I, ADHD inattentive subtype; ADHD-C, ADHD combined subtype. a The P value was obtained using two-sample t-test. The cognitive function was assessed using the reaction time variability (RTV) index adopted from a Stop Signal Task in the Cambridge Neuropsychological Test Automated Battery (CANTAB). Each participant was asked to complete a 20-min task, in which the participant must respond to an arrow stimulus by selecting one of two options, depending on the direction at which the arrow points. The RTV was calculated as the ratio of standard deviation to mean value of the reaction time for correct responses for analyses [35]. The RTV data were only available for 39 ADHD patients (21 ADHD-I and 18 ADHD-C) and 24 HCs.

Data acquisition
A multichannel near-infrared optical imaging system (Hui Chuang, China) with continuous waves and a 17 Hz sampling rate was used to collect rs-fNIRS data. Similar to our previous study [5], twenty-four light sources (each with two wavelengths: 670 and 830 nm) and twenty-eight detectors were arranged on the participant's head, with adjacent source and detector pairs being 3 cm apart and composed of 80 different measurement channels ( Fig. 1(A)). Using the external auditory canals and vertex of the participants as landmarks, the position of these probes was established according to the international 10-20 system. Specifically, measurement channels 35 and 36 were placed around T3 and T4, and the midpoint of the two channels was localized in Cz ( Fig. 1(B)). Considering changes of the averaged head circumference from 51.5 cm in 8-years-old children to 53.2 cm in 12-years-old children [36], we estimated the measurement inconsistency across participants was in the range of 0.09 cm (i.e., [(53.2-51.5)/53.2] × 3 cm; 3cm is the S-D separation). The positions of the measurement channels were validated by the spatial coordinates, which were acquired from the structural MRI image of an arbitrary participant with a Siemens 3.0 Tesla scanner. The channels were labeled with vitamin E capsules during structural MRI scanning and were projected on the network templates from Yeo et al. [37] in further analyses. According to the templates, six functional networks, i.e., default mode (DMN), frontoparietal (FPN), ventral attention (VAN), somatomotor (SMN), dorsal attention (DAN) and visual (VN) networks, were involved and displayed with different colors in our study ( Fig. 1(C)).

Data preprocessing
The in-house FC-NIRS package (http://www.nitrc.org/projects/fcnirs [19]) was used to preprocess the current resting-state fNIRS data in this study. First, motion artifacts were examined and removed from the optical signals using a spline interpolation method [38]. This method detected the motion-induced artifacts by calculating moving standard deviation (MSD) within sliding time windows in a window length of 2 seconds. The MSD values larger than a predefined threshold (e.g., 5 MSD [19]) were regarded as artifacts. The time series that represented the motion artifacts was further modeled via a cubic spline interpolation, which was then subtracted from the original signal of the time series. The resulting signal was considered to be free of motion artifacts. Next, the signals were bandpass filtered with a frequency of 0.01 Hz-0.1 Hz [39] to diminish the interference of low-frequency drift and high-frequency neurophysiological noise [40]. Subsequently, they were transformed into the concentrations of oxyhemoglobin (HbO) and deoxyhemoglobin (HbR) using the modified Beer-Lambert law [41] with a differential path-length factor of 6 for two wavelengths [41][42][43][44][45]. Finally, an eight-minute stable hemoglobin time series was extracted from each participant. Note that the HbO data were used for subsequent analysis considering its better signal-to-noise ratio [46].

Multiscale entropy and brain signal variability
The multiscale entropy (MSE) algorithm, which has been widely used for characterizing neural signal complexity, was adopted to estimate brain signal variability (BSV) in this study. MSE was calculated according to the following steps similar to our previous research [27]. First, the HbO concentration time series {x 1 , . . . x i , . . . x N } was downsampled to obtain a coarse-grained time series for different timescales t. The coarse-grained time series y t was generated by averaging the data points within nonoverlapping windows of length t. Each value of the coarse-grained time series, y t j , was computed using the following equation: (1) x i denotes the value of time point i in the original time series; N defines the number of time points; and j is the index of each coarse-grained time series. Second, the sample entropy of each downsampled time series, S E , was calculated by: where In Eqs. (2) and (3), m defines the pattern length indicating that m consecutive data points are used for pattern matching, and r specifies the similarity criterion indicating the threshold portion of the time series standard. In this study, pattern length was set to m=2 and the similarity criterion was chosen to r = 0.2, which were judged to be optimal and statistically valid following the method used in [27,47,48]. The data points are considered to have indistinguishable amplitude values when the absolute amplitude difference among them is lower than r. v m i is the vector defined as {v m i = y i , y i+1 , . . . y i+m−1 }; the definition of v m j is similar to that of v m i , and |v m i − v m j | means the Chebychev distance between v m i and v m j . MSE, the sample entropy across different temporal scales, quantifies the signal variability by estimating the predictability of amplitude patterns across a time series. While lower MSE values indicate a low complexity and a high degree of determinacy for the signal, higher MSE values reflect a high complexity and a low degree of predictability or rich information for the signal.
2.6. Statistical analysis 2.6.1. Between-group differences in brain signal variability To evaluate group differences in BSV, the MSE values of each channel between the ADHD and HCs were compared using a multiple linear regression model, in which MSE was considered as dependent variable, group as independent variable, and age, IQ, age × group (i.e., interaction effect of age and group) and IQ × group as covariates. In the model, significant group differences were determined by P values lower than 0.05. False discovery rate (FDR) correction [49] was used to control the multiple testing error by correcting P values.

Correlation between brain signal variability and ADHD core symptoms/reaction time variability
Pearson correlation analyses were performed between MSE and ADHD symptom scores (inattentive, hyperactive/impulsive and total scores) as well as between MSE and RTV in the ADHD and HCs, respectively. Before the correlation analyses, the effects of age and IQ were removed by multiple linear regression. FDR correction was conducted to correct the multiple comparisons.

Demographic and core symptom
The demographic and core symptom data are presented in Table 1. There were no significant differences in age (P = 0.819) or IQ (P = 0.076) between the ADHD and HCs. The ADHD group had significantly higher inattentive scores (P < 0.001), hyperactive/impulsive scores (P < 0.001) and total scores (P < 0.001) than the HCs group.

Spatial distribution of brain signal variability in the ADHD and HCs
Similar to previous studies [27,47,50], the area under the curve (AUC) according to the sample entropy across different temporal scales (i.e., MSE) was adopted to evaluate brain signal variability (BSV). To obtain an intuitive visual presentation, we adopted bilinear interpolation algorithm to smooth MSE values across the whole brain. The bilinear interpolation algorithm performs interpolation calculations in both directions and can be extended from the linear interpolation between variables. Figure 2(A) shows the group-averaged MSE in ADHD and HC groups, respectively. We found that the ADHD and HC groups showed similar spatial distributions in MSE, with large values located in regions of frontal, parietal and occipital cortices (Fig. 2(A)). However, quantitatively, the MSE values in ADHD were much smaller than those in HCs. The mean MSE values were 0.62 ± 0.07 (mean and standard deviation) for ADHD patients and 0.65 ± 0.08 for HCs (Fig. 2(B)).

Statistical difference in brain signal variability between ADHD and HCs
Group statistical differences in MSE between ADHD and HC were shown in Fig. 3. Specifically, MSE values on 23 measurement channels, mostly involving the default mode (DMN), frontopatietal (FPN), visual (VN), ventral (VAN) and dorsal (DAN) networks, were significantly reduced in the ADHD group (Fig. 3(A)), while two channels in default mode-network still showed significantly increased MSE values in ADHD patients (P values < 0.05, FDR correction) (Fig. 3(B)). Significant MSE differences in measurement channels between ADHD and HCs were included in Table S1 in the supplementary material. 3.4. Correlation between brain signal variability and ADHD core symptoms/reaction time variability correction) were found in FPN, which revealed that lower brain signal variability in the FPN network was associated with more severe symptoms in the ADHD group. The above correlations were retained only in the ADHD-C subtype when the analyses were repeated in the two subtypes separately (r = -0.54; P = 0.014; P = 0.085, FDR correction for hyperactive/impulsive scores; r = -0.47; P = 0.038; P = 0.230, FDR correction for total scores, respectively).  (Fig. 4(B)).

Discussion
In our present study, we used multiscale entropy (MSE) analysis to evaluate fNIRS signal variability in children with ADHD and typically developed controls. As we expected, categorical analyses indicated decreased brain signal variability (BSV) measured by MSE in boys with ADHD compared with age-matched healthy controls. Further quantitative analyses showed a negative association between altered MSE values and ADHD core symptoms, indicating lower BSV in the frontoparietal (FPN) network associated with more severe hyperactive/impulsive symptoms. Another important and interesting finding was that decreased BSV was negatively associated with increased reaction time variability (RTV) in children with ADHD.
Despite conflicts, strong evidence from recent studies has indicated that signal variability might increase with maturation. In the study by [16], the authors used two measures to evaluate BSV: principal component analysis (PCA) and MSE; both indicated increased brain variability with maturation from childhood (8-9 years old) to young adulthood (20-33 years old). When extending the analysis to include healthy aging, the authors found an inverted U-shaped curve with the development of BSV from childhood to old age [51]. For our present study, we focused on children with ADHD and healthy controls aged 8-12 years old. As supported by a previous study [16,52], BSV should increase with maturation. ADHD has been considered a neurodevelopmental disorder with delayed maturation in brain development. Briefly, a marked delay of approximately 3 years of cortical maturation was found in ADHD, with the most prominent delay in the prefrontal regions [29]. Consistent with this, Qian et al. [28] found that children with ADHD had development-related delays in inhibition and shifting functions. Considering the delayed development of ADHD and the 'increasing BSV' theory for typical developing individuals [30], our present study indicated that the MSE values in children with ADHD were indeed lower compared with HCs.
Another interesting finding is the negative correlation between decreased BSV and the increased cognitive variability (reaction time variability, RTV) in children with ADHD. The greater BSV representing well-functioning neural systems could enable the brain to respond to a task or the environment more accurately and smoothly, leading to more stable behavior or cognitive function [53]. That is, if the BSV decreases, the brain cannot process the information efficiently and quickly, leading to increased cognitive variability. This negative correlation has been confirmed in children with normal development, in which increased BSV accompanies lower behavioral variability [16]. Numerous studies have indicated increased RTV in patients with ADHD [54][55][56]. In fact, in a previous study, frontocentral theta-band phase variability was indicated to be closely linked with RTV in children with ADHD both phenotypically and genetically [57]. However, frontocentral theta-band phase variability does not account for the temporal disorganization of neural dynamics, while our findings indicate a correlation between moment-to-moment BSV and cognitive variability. RTV is an important index of sustained attention. Impaired sustained attention always occurred in children with ADHD [58], which was even state-independent and should be an endophenotype of ADHD [59]. Our present study provides the potential neural evidence underlying increased RTV in children with ADHD. This 'brain-behavior' relationship will also promote our understanding of the pathogenesis of ADHD. Significantly, the deficiency of sustained attention occurs not only in ADHD, but also in other psychiatric disorders, such as anxiety disorders [60] and mood disorders [61]. As such, our present finding will also provide meaningful reference for the research of other psychiatric disorders.
As for the functional brain networks, we found that the children with ADHD showed reduced BSV in both high-order (DMN, FPN, VAN and DAN) and primary brain functional networks (VN). These findings are interesting because they demonstrate a wide alternation in the momentto-moment variability of spontaneous brain signal in children with ADHD. These results did not change when considering ADHD subtypes (Table S2). However, the quantitative analyses of ADHD symptoms only indicated the association between altered BSV in FPN and hyperactiveimpulsive symptoms. The FPN has been suggested to be a flexible hub for cognitive control [62]. Dysfunction of FPN is a well-confirmed finding in the MRI study of ADHD and ADHD-related cognitive dysfunction, including impulsive symptoms [8,63] and inhibition [64]. Our current findings support the critical role of FPN in ADHD, especially the hyperactive/impulsive symptoms, on a temporal scale. The other networks revealed from the categorical analyses, including DMN, VAN, SMN, are also worthy of further exploration with respect to other behavioral domains and cognitive functions. In addition, the brain networks are not isolated but have complex and multiple connections. Therefore, the investigation of the inter-correlation of BSV among these brain networks may also be helpful to understand the neural dynamic complexity in participants with ADHD more comprehensively.
Several limitations should be considered for our current study. First, the sample size is relatively small, and we could not investigate the potentially explicit confounding effects of comorbidities on our findings. In addition, most results were marginal after corrections for multiple comparisons, which may also be due to the limited sample size. Second, we only included males for analyses, which may limit the generalization of our findings only to males. Third, we only recruited children as participants. Adults recruited in the future could enable us to better understand the age-related changes in BSV in participants with ADHD. Finally, although we adopted the spline interpolation method to remove head motion artifacts, the potential influence of head motion could still exist. Future studies are expected to combine motion-removing approaches and motion record system to eliminate the influence of head motions.

Conclusion
In summary, we utilized resting-state fNIRS imaging to explore the brain signal variability (BSV) in children with ADHD and estimated its association with ADHD core symptoms and cognitive function. The results revealed decreased BSV in children with ADHD compared to typically developing controls and decreased brain variability in ADHD was accompanied by increased ADHD core symptoms and disrupted cognitive function. Our results provide novel insights and potential biomarkers for the diagnosis of ADHD. Replications of these experiments in larger, independent population samples with wider age ranges are needed to confirm our preliminary findings.