Abstract
Attention-deficit/hyperactivity disorder (ADHD) is characterized by involuntary fluctuations of attention in continuous performance tasks (CPTs) wherein attention must be sustained over long periods of time. The neuronal basis underlying aberrant attentional fluctuations in time scales from seconds to minutes have remained poorly understood. Neuronal alpha- and gammaband oscillations are thought to implement attentional and top-down control of sensorimotor processing. We hypothesized that aberrant behavioral fluctuations in ADHD would be caused by aberrant endogenous brain dynamics in alpha and gamma-band oscillations and specifically by their aberrant long-range temporal correlations (LRTCs). We measured brain activity with magnetoencephalography (MEG) from adult participants diagnosed with ADHD (N = 19) and from healthy control subjects (N = 20) during resting state and two CPTs; a threshold stimulus detection task and a Go/NoGo task. We then estimated LRTCs of neuronal oscillations and behavioral fluctuations with detrended fluctuation analysis (DFA). ADHD was associated with aberrant LRTCs in both behavioral performance and of neuronal oscillations. LRTCs were correlated with symptom severity with a U-shaped correlations indicating that the LRTCs were largest with moderate symptom scores. These data demonstrate the presence of aberrant temporal dynamics of neuronal oscillations in adult ADHD patients, which may underlie involuntary attentional fluctuations in ADHD. Taken that LRTCs are a hallmark of brain critical dynamics, these data show that moderate ADHD symptoms scores maximize brain criticality which is thought to be beneficial for performance.
Introduction
Attention-deficit/hyperactivity disorder (ADHD) is a heritable neurodevelopmental condition that is characterized by detrimental levels of impulsive, inattentive and/or hyperactive traits.1,2 ADHD affects 2.5–3.4 % of the adult population worldwide and its incidence has increased markedly in the last decade.3 Current diagnostic tools for ADHD are largely based on subjective appraisals such as clinical interviews, self-reports, and observations by caregivers. However there are no reliable biomarkers that would enhance either the precision or accuracy of the diagnosis.4 The development of reliable biomarkers and diagnostic criteria for ADHD are critically dependent on understanding the neuronal substrates of the disease symptoms, which may reveal ADHD-specific neuronal biotypes, or neurotypes, with abnormalities in neuronal dynamics despite no clear differences in behavioral testing.5
The cognitive functioning of individuals with ADHD is characterized by altered reaction times (RTs) and increased reaction time variability (RTV), which reflect the drifting of attention, loss of focusing, and failures in executive functions.6–8 Such cognitive deficits are especially observable in excessive performance fluctuations in continuous performance tasks (CPTs) where attention must be sustained for a longer period of time in a monotonous or “boring” environment.9,10 Elevated RTV and decreased response inhibition reflect individual differences in executive functioning and constitute an endophenotype in ADHD.8 Attentional and executive functioning drifts involve time scales from seconds to hundreds of seconds in both healthy subjects and clinical populations. Behavioral performance in CPTs is characterized by slow oscillations or “streaks” in RT time-series,8,11 and power-law long-range temporal correlations (LRTCs),12,13 which are likely driven by similarly individually variable LRTCs in endogenous brain dynamics.14 However, how behavioral LRTCs are associated with ADHD symptomatology or even how deficits in RTV and other psychophysical measures in ADHD patients are generally related to LRTCs have remained unclear.
Magnetic resonance imaging (MRI) of the brain has established that ADHD is associated with structural changes in surface-based morphometry (SBM) measures such as gyrification,15 as well as with structural and functional connectivity16 measures; albeit findings are somewhat inconsistent.17 At the functional level, evidence for abnormal neuronal dynamics in ADHD patients have been found in electro- (EEG) and magnetoencephalography (MEG).18–21 Alpha (8-12 Hz) and beta (13-30 Hz) band oscillations, which are fundamental for attention and topdown control,21–26 show reduced task-dependent suppression and lateralization of local oscillation amplitudes in adults21,27–29 and children30 with ADHD. Such aberrant modulation of oscillation amplitudes are thought to arise from decreased structural connectivity (SC) between fronto-parietal (FP) regions.28 However, even though these prior studies convergently indicate abnormal event-related and sub-second oscillation dynamics in ADHD, such shortrange dynamics do not directly explain the temporally long-range and long time-scale deficits in sustained attention.
Oscillation amplitudes have been found to exhibit not only short-range correlations but also slow fluctuations over several seconds up to minutes. Similar to fluctuations in behavioral performance, these neuronal activity fluctuations are slowly decaying, scale-free (i.e. without characteristic temporal scale) and can be described with power-law LRTCs.14,31–35 LRTCs are thought to be the hallmarks of brain criticality36 and indicate that healthy brain activity is in the critical state between order and disorder that is thought to maximize the information processing capacity. Importantly, there are also large individual variability in neuronal LRTCs within healthy brain activity which predict variability of LRTCs in behavioral performance in CPT tasks and therefore predict the length of behavioral streaks in a trait-like manner.11,14,33 Indicative of their critical role in healthy brain dynamics, LRTCs are attenuated in patients with epilepsy,37 Alzheimer’s disease,38 schizophrenia or schizoaffective disorders,39–41 autism spectrum disorder42,43 and major depression,44,45 compared to neurotypical control subjects. However, it remains unknown whether neuronal LRTCs are abnormal also in ADHD patients, even though specifically this patient group struggles to sustain attention over long temporal durations and their behavioral performance is characterized by aberrant dynamics (e.g. increased RTV) especially in CPTs.
We tested here whether ADHD would be characterized by aberrant neuronal LRTCs by recording MEG data during two different CPTs from adult ADHD patients. We estimated behavioral LRTCs from the CPT RT time-series, and neuronal LRTCs from MEG oscillation amplitude fluctuations, using detrended fluctuation analysis (DFA). We found altered neural LRTCs in oscillation amplitude fluctuations in ADHD patients and demonstrated that they were correlated with both the variability in behavioral LRTCs and the severity of ADHD symptoms.
Methods and Materials
Participants
The sample comprised of N = 23 healthy neurotypical control participants (mean: 33 years old, range: 18-57; 15 females, 1 left-handed), and N = 23 patients with prior diagnosis of ADHD (mean: 36 years old, range: 26-59; 13 females, 3 left-handed) recruited from a previous study46 and from the University of Helsinki students. The sample size was known to induce reliable LRTCs based on a previous study.14 Inclusion criteria for all participants were: aged 18-60 years, male or female, normal or corrected to normal vision, and compatibility with MEG and MRI. Exclusion criteria included any substance abuse or medications active in the central nervous system and ADHD patients were asked to refrain from medications 24 hours before the MEG recordings. The control group exclusion criteria included any neurological or neuropsychiatric disorder and for the ADHD group any comorbid neurological or neuropsychiatric disorder or IQ < 80. This research was conducted in compliance with the Declaration of Helsinki and approved by the Research Ethics Board at Helsinki University Central Hospital. All participants gave written informed consent prior to participation.
Tasks and Experiments
MEG data was collected during resting state (RS) activity and during two visual CPTs collected on separate days. Both tasks were presented in a projector screen inside a magnetically shielded room during MEG recordings.
Resting state
Eyes-open 10 min resting state MEG data, during which participants fixated at the fixation cross in the center of the screen, were collected before task data collection. Data for RS were collected from N = 23 control participant and N = 23 ADHD participants.
Visual Threshold of Stimulus Detection Task (TSDT)
Participants were instructed to maintain fixation and detect the presence of stimuli presented at the visual threshold14 and individually pre-calibrated to achieve a 50% detection rate using the adaptive QUEST algorithm (Supplementary Table 1).50 Here misses reflect lapses in sustained attention. TSDT task was generated using Psychtoolbox-3. Visual stimuli of two shapes (50 % probability) were presented in any orientation with duration of 0.1 seconds in slowly-varying gradient noise with a diameter of 10° 51 and with a random stimulus onset asynchrony (SOA) of 1.5 – 4.5 s (Fig. 1A). During MEG, 500 stimulus trials were collected for each subject during a run of ~25 min.
Go/NoGo task
Go/NoGo task is as in,12 where commission errors (i.e., erroneous responses to NoGo stimuli) constitute a measure of lack of sufficient cognitive control. The task was coded using Presentation™ software (Neurobehavioral Systems, Inc., Albany, CA, USA). The stimulus was presented with a size of 1° for 0.1 s on a grey background with mean luminance of 107 cd/m2 (Fig. 1B). The Go (75% of trials) and NoGo (25% of trials) stimuli were distributed randomly in the stimulus stream with SOA of 1 s.12 Participants were instructed to respond as quickly as possible for the Go and withhold the responses for NoGo stimuli (Fig. 1B). A total of 1000 trials were collected for each participant.
Behavioral Data analysis
Detection Hit Rates (HRs) and Reaction times (RTs) for TDST were computed as the proportion of correct detection, with a response provided between 200–1500 ms from stimulus onset. Despite the calibration before experiments, HRs for participants varied and participants with HR < 25% were excluded from the analysis leaving N=19 ADHD and N=19 control subjects. HRs for the Go/NoGo task, were computed for the correctly responded Go-trials (‘Hits’) and incorrectly responded NoGo-trials (‘Commission errors’ i.e false alarms) for responses provided between 150–800 ms from the stimulus onset. Here, high HRs are indicative of successful task performance and HR < 70% for the Go stimuli were excluded leaving N=19 ADHD and N=20 control participants. Reaction time variability (RTV) was derived by computing standard deviations over the individual RTs in Go-trials. In the TSDT task, behavioral LRTCs were computed over the continuous sequence of Hit and Miss trials as in14. In the Go/NoGo task, RTs for Hits, and Commission Errors i.e. False Alarms (FA) were all included in the RT timeseries, as in12. LRTCs were then estimated from the RT time-series using Detrended Fluctuation Analysis (DFA) as in52. The scaling exponent β is a measure of temporal clustering such that higher β values indicate stronger temporal dependencies while values closer to 0.5 are associated with uncorrelated noise.
MEG and MRI Recordings
MEG data were recorded in the BioMag laboratory in HUS Medical Imaging Centre using 306-channel MEG (Elekta Neuromag TRIUX) with a sampling rate of 1000 Hz and online passband of 0.03-330 Hz. Electro-oculography (EOG) data were recorded to detect and then remove ocular artifacts. Behavioral responses were recorded with electromyography of abductor/flexor of pollicis brevis muscle and detected with an automatic algorithm. T1-weighted anatomical MRI scans for cortical surface reconstruction models were obtained for each subject at a resolution of 1 x 1 x 1 mm (MP-RAGE) with a 1.5-T MRI scanner (Siemens, Germany).
MEG preprocessing, source analysis, and surface parcellations
Source reconstruction and data analysis followed previously published procedures.53–55 Maxfilter software (Elekta Neuromag) was used to suppress extra-cranial noise and interpolate bad channels. Independent component analysis (ICA) (MATLAB) was used for identifying and extracting ocular and heartbeat artefacts. The MEG data were Morlet-wavelet filtered between, fmin = 3 Hz… fmax = 148 Hz with m = 5. Freesurfer software was used for volumetric segmentation of the MRI data, surface reconstruction, flattening, and cortical parcellation/labeling with the Destrieux atlas. MNE software56 was used for MEG-MRI colocalization, and preparation of the forward and inverse operators. Noise-covariance matrices (NCMs) were computed with a regularization constant of 0.05 using 200–250 Hz filtered baseline data (−0.75 - 0.25 s) for both TSDT and RS-data for the Go/NoGo task (4×1 min for NCM and 2×1 min for ADHD). Source-model dipoles had fixed orientation with 5 mm resolution. Source source-space vertices were collapsed into 400 cortical parcels using fidelity optimized parcellations57 before further data-analysis.
Analysis of LRTCs from Neuronal MEG Data
To estimate the LRTCs of oscillation amplitude fluctuations, we applied DFA on the amplitude envelope of the Morlet-wavelet filtered parcel time-series signals52 by segmenting data into time windows Δt from 1 to 225 seconds. Each segment of integrated data was then locally fitted similarly as for behavioral DFA.
Statistical analysis of behavioral data
Non-parametric Mann-Whitney U-test was used to test differences between the control and ADHD groups. The correlation between performance and the clinical total scores were assessed with Spearman’s ranked correlation test. The correlation between behavioral DFAs and the clinical scores, and RTV in the Go/NoGo -task, were assessed with Spearman’s ranked correlation test.
Statistical analysis of MEG data
Before statistical testing, the MEG data were collapsed into 200 parcels to decrease the impact of individual variability in brain functional anatomy. To identify the most significant and robust effects, statistical analyses were performed separately for each frequency and parcel. Within-group differences between task-state and resting-state activity were obtained with Wilcoxon signed-ranked test while group differences were obtained using Welch’s t-test. The correlations between clinical scores and neuronal and behavioral scaling laws were estimated with Spearman rank correlation and quadratic correlations tests. Multiple comparisons were controlled with false discovery rate (FDR) correction at given the alpha-level (i.e., 5% at α = 0.05). In addition, we applied a Q threshold to indicate the proportion of significant observations that could thereafter arise by chance in any of the wavelet frequencies. Only events that were above both corrected thresholds were considered significant.
Data availability statement
Anonymized data underlying figures will be made available upon publication.
Results
Clinical evaluations and Symptom data
We first validated that the ADHD participants had clinical symptoms as defined by ASRS, BIS, and BADDS scores, and that subjects in the control group had scores defined for neurotypical groups (Supplementary Table 1). For the ADHD group, the mean and standard deviation (SD) were 78.22 ± 12.88 for BIS, 67 ± 19.88 for BADDS, and 16.5 ± 3.68 ASRS (Fig. 1E). For the control participants, total scores were for ASRS were 8.79 ± 2.74 and for BIS were 59.58 ± 10.00, both of which were significantly lower than in the ADHD group (ASRS, Mann-Whitney U-test, p = 2.1E-7, BIS, Mann-Whitney U-test, p = 3.8E-5, Fig 1E). Although these symptom scores are not officially part of any diagnosis criterium for ADHD, be it ICD-10 or DSM-5, and have been used mostly for screening purposes or evaluating the response for treatments, it has been suggested that a total BIS sum of 72 or above should be used to classify an individual as highly impulsive.58 In the case of ASRS, the maximum total score is 30, but for screening purposes certain questions weigh more than others. In this study, there were only two out of 23 control subjects who fulfilled these screening criteria. Taken together, these data show that ADHD symptoms in the control participants were mostly below or in the subclinical regime.
Behavioral task performance
We first assessed the classical behavioral measures of ADHD using TSDT (Fig. 1A) and Go/NoGo (Fig. 1B) tasks by estimating measures from the behavioral time-series data (Fig. 1C-D). Taken that the TSDT task performance was pre-calibrated to a 50% stimulus-detection rate (Hit Rate, HR), as expected, the between-group differences in HRs were not significant for TSDT (Fig. 1F) (Control group (Ctrl) 40.35 ± 11.76 % (mean ± SD), ADHD patients 49.45 ± 16.24%, Mann-Whitney U-test, p = 0.087) nor were there differences in reaction times (RTs) (Ctrl 0.56 ± 0.0455 s, ADHD 0.54 ± 0.058 s, Mann-Whitney U-test, p = 0.26). In the Go/NoGo task, HRs (Ctrl 91.5 ± 8.31 %, ADHD 91.27 ± 6.86 %, Mann-Whitney U-test, p = 0.77) nor RTs (Control 0.27 ± 0.039 s, ADHD 0.28 ± 0.035 ms, Mann-Whitney U-test, p = 0.14) were significantly different between the groups neither were the False alarms (FA) (Fig 1 E) (Ctrl 0.12 ± 0.08 %, ADHD patients 0.18 ± 0.14 %, Mann-Whitney U-test, p = 0.14, Fig. 1G). However, as found previously, the reaction time variability (RTV) was larger for the ADHD group compared to the control group (Ctrl 0.076 s ± 0.022, ADHD patients 0.089 s ± 0.022, Mann-Whitney U-test, p = 0.038, Fig. 1G).
We then identified long-range temporal correlations (LRTCs), or “streaks”, in behavioral Hit Miss times-series (Fig. 1C) and RT time-series (Fig. 1D) data for the TSDT and Go/NoGo tasks, respectively, by using DFA (Fig 1H). DFA exponents did not differ between control and ADHD groups for TSDT (0.74 ± 0.095 and 0.78 ± 0.086 respectively, Mann-Whitney U-test, p = 0.34, Fig 1H), but in the Go/NoGo task, DFA exponents were larger in the control than in the ADHD group (0.74 ± 0.066 in the control group and 0.69 ± 0.06 in the ADHD group, Mann-Whitney U-test, p = 0.018).
Neuronal LRTCs differ between ADHD and control groups
Oscillation amplitude envelopes were extracted from Morlet-wavelet filtered time-series of source-reconstructed MEG data (Fig. 2A) and used detrended fluctuation analysis (DFA) to identify DFA exponents (Fig. 2B). We first conducted a whole-brain-level analysis of neuronal LRTCs from oscillation amplitude envelopes such that the neuronal DFA scaling exponents were computed across all parcels for the task. The whole-brain mean DFAs averaged across all parcels (brain areas) peaked in the alpha-band (α, 8-13 Hz) in both tasks (Fig. 2C).
To assess the group differences at parcel-level anatomical resolution, we computed parcel-wise differences in LRTCs between the groups. In line with the whole-brain-average analysis, the DFA exponents were smaller for control than for the ADHD group in the theta-alpha (θ–α, 610 Hz) band in the TSDT task and in the low theta (θ, 3–4 Hz) band in the Go/NoGo task (Fig. 2D) (Welch’s t-test,p < 0.05, FDR corrected). Stronger LRTCs of theta-alpha oscillations in ADHD patients than in control participants were found in the occipital cortex in both tasks and during the TSDT task additionally in posterior parietal cortex (PPC) and temporal cortex (Fig. 2E).
Different task-dependent modulation of LRTCs in ADHD patients
We then asked in parcel-level resolution whether these differential task effects between groups on LRTCs would be caused by different task dependent modulation compared to rest within the groups. In the control group, the theta-alpha (θ–α, 6-13 Hz) band DFA scaling exponents were suppressed in the task conditions compared to the in resting-state (RS) and increased from the RS in the wide gamma-band (γ, 30–60 Hz) (Fig. 3A) (Wilcoxon signed rank test, p < 0.05, FDR corrected). In the ADHD group there was only a slight suppression in the theta-alpha band DFAs and a robust increase in the wide gamma-band DFAs) during task compared to RS in both tasks (Fig. 3A). These differences in the task dependent modulations of DFA exponents were significantly different between the groups (Welch’s t-test, p < 0.05, FDR corrected).
The suppression of theta-alpha band DFAs in TSDT task was widespread in the control group but restricted to somatomotor (SM) system in ADHD patients (Fig. 3B). This suggests that, during the performance of sensory detection task, ADHD patients successfully inhibit LRTCs only in the motor system, but not in other brain regions like the control group. In the Go/NoGo task, in which successful performance is dependent on inhibition and on the activity of PFC, the suppression of theta-alpha band DFAs was restricted to SM in control group but was also found in the PFC in the ADHD group, pointing to abnormal regulations of LRTCs in the key region underlying task performance (Fig. 3B). In contrast to suppressed theta-alpha DFAs, gamma band DFAs in both tasks were slightly increased during task performance in the control group (Fig. 3A) (Wilcoxon signed rank test, p < 0.05, FDR corrected) in visual and PFC regions (Fig. 3C), but robustly increased in ADHD group in nearly all brain regions (Fig. 3C).
Neuronal LRTCs predict LRTCs in behavioral performance
We next tested whether neuronal LRTCs could predict LRTCs in behavioral task performance. Neuronal and behavioral DFAs in the TSDT task showed significant robust positive correlations for ADHD group in alpha and beta bands, which was not seen for the control group (Fig. 4A, Spearman rank correlation, p < 0.05, FDR corrected). In the Go/NoGo task behavioral and neuronal DFAs were correlated positively with performance in the control group in the alpha band (Fig. 4A) while in ADHD group there was a negative correlation between thetaalpha band DFAs and task performance. In ADHD patients, these correlations arised from the activity in visual systems (Fig. 4B-C) while in control group positive correlations arised from the PFC.
Neuronal LRTCs correlate with the ADHD symptom scores
Finally, to test whether aberrant LRTCs were clinically relevant, we investigated if LRTCs were correlated with ADHD symptoms as quantified with ASRS and BIS scores. In both tasks, neuronal DFAs in the theta-beta band were negatively correlated with ASRS and BIS scores in ADHD patients (Fig. 5A). This indicated that smaller DFA exponents in patients were found to accompany worse ADHD symptoms in several parcels. In contrast, in healthy participants larger DFA exponents accompanied non-clinical increase in the ASSR and BIS scores (Fig 5B). To further understand this bimodal distribution, we next averaged DFAs over all cortical parcels at a peak correlation frequency of 17 Hz and plotted the correlation of DFA exponents with ASRS and BIS scores (Fig. 5C). Taken the opposite correlations between LRTCs and symptom scores in the healthy controls and ADHD groups, we tested whether the correlations would be explained by an inverted U-shaped curve (i.e parabolic) where largest LRTCs are obtained with intermediate symptom score values. To this end, we estimated the partial quadratic correlations by regressing out the linear components between DFA exponents with ASRS and BIS scores. As predicted, quadratic correlations were significant both in TSDT and Go/NoGo tasks, demonstrating that maximal LRTCs values are obtained in the border between neurotypical behavior and ADHD symptoms (Fig. 5C).
Discussion
LRTCs are a characteristic feature of behavioral performance and healthy brain activity59 while their aberrancies are a characteristic feature of many brain diseases.37–45 We investigated whether aberrant behavioral fluctuations in sustained tasks in ADHD could arise from aberrant long-range temporal neuronal dynamics. We show here that behavioral LRTCs are decreased in ADHD in Go/NoGo task measuring response inhibition and cognitive control, this being paralleled by aberrant LRTCs of oscillation amplitude fluctuations. The deviant neuronal LRTCs were correlated with inter-individual variability in behavioral LRTCs both in TSDT and Go/NoGo CPTs, suggesting that aberrant neuronal fluctuations can be the cause of aberrant behavioral fluctuations in adult ADHD patients. Overall, these data demonstrate that altered LRTCs in oscillation amplitudes might constitute to the aberrant involuntary fluctuations of attention of adults with ADHD.
ADHD patients had aberrant LRTCs in alpha and gamma-band oscillation amplitude fluctuations compared to neurotypical controls, both during resting-state and CPT execution. This shows that in ADHD, oscillation dynamics is altered across time scales from seconds to minutes, in addition to the previously found sub-second time scale abnormalities in alpha band for adult,21,27,29 adolescent60,61 and child ADHD patients,30 and in theta-62 and gamma-bands for adult ADHD.20 These results are also in line with a recent study reporting intrinsic temporal variability in functional connectivity in ADHD patients63 and show that aberrant temporal dynamics is also a characteristic of local neuronal processing. ADHD patients had larger alpha band LRTCs in RS compared control group suggestive of longer periods of mind wandering64,65 In contrast, while in the control group alpha-band LRTCs were suppressed in task compared to RS-activity, this suppression was not evident in ADHD group. Instead, increased gamma-band LRTCs in the Go/NoGo task compared to RS was only observed in the ADHD group. This differential pattern of alpha and gamma frequency band LRTCs changes in response to task performance suggests that ADHD patients might have developed a compensatory mechanism to overcome the deficits in sustaining attention in experimental conditions.66 The localization of LRTCs was widespread across cortical regions, in line with prior studies of LRTCs location14 and results showing that altered neuronal processing in ADHD characterizes multiple brain systems in terms of fMRI BOLD signal strength67–75 and functional connectivity16,70,73,76 as well as MEG oscillation power and connectivity28,77 analyses. DAN (Dorsal Attention Network) hyperactivation and stronger within-network connectivity of both VAN (Ventral Attention Network) and DAN78 have previously been interpreted as either a compensatory mechanism79 or deficient segregation of these networks. This implies that ADHD patients are challenged in engaging whichever network is more relevant for current behavioral goals. This supports the idea that increased task-dependent gamma-band LRTCs found in Go/NoGo task for ADHD patients reflect a compensatory mechanism to overcome difficulties in task performance demanding cognitive control. Opposite changes of alpha- and gamma-band LRTCs could lead to the behavioral deficits observed in ADHD, such as the increased attention to distracting information reflected in stronger ERPs80 and a tendency for mind wandering81. In line with this idea, alpha and beta band LRTCs were correlated with behavioral DFA during task performance in TSDT task more strongly in ADHD patients. This suggest that sensory-driven task differences in LRTCs in oscillation amplitude fluctuations could lead to aberrant behavioral fluctuations observed previously to characterize ADHD.82 These findings are in line with hypoarousal and unstable vigilance regulation as a possible factor contributing to ADHD symptoms.83,84
Importantly, neuronal LRTCs predicted the ASRS and BIS symptom scores with quadratic correlations indicating inverted U-shaped correlations where LRTCs are maximal for intermediate symptom scores in the regime between neurotypical features and mild ADHD symptoms. In contrast, both low symptom scores in control subjects and high symptom scores in ADHD patients were associated with smaller LRTCs. As high LRTCs are the hallmarks of brain criticality,36 this demonstrates that moderate symptom scores and mild ADHD are associated with the brain operating closer to the critical point. Since, brain criticality is thought to maximize flexible routing of information, storage capacity,36 and optimal performance in behavioral data,12 this suggest that moderate ADHD symptoms scores are associated with these optimal brain states and are obtained in neurotypical controls with strong ADHD features and in ADHD patients with mild symptoms. In contrast, neurotypical controls with low ADHD symptom scores and ADHD patients with high symptoms scores were associated with smaller DFA exponents are thought to reflect suboptimal brain dynamics that might be detrimental for behavioral performance, yielding to less flexible information processing and storage capacity. These data constitute evidence that altered LRTCs are not only a core disease mechanism causing ADHD symptoms and behavioral deficits, but also leads to behavioral variability and putatively to personality characteristics in the healthy neurotypical population.
Importantly excitation-inhibition (E/I) (E/I ratio) is the main control parameter of critical dynamics such that it emerges at E/I balance, while excess inhibition or excitation leads to sub- or super-critical dynamics, respectively85–91. In line with this framework, LRTCs are influenced by pharmacological modulations of neuronal excitability92–95 as well as by polymorphism in neuromodulatory genes96. Aberrant LRTCs in ADHD could hence arise via mechanisms that influence the E/I ratio at the systems level or in specific neuronal circuits. Specifically, the norepinephrine (NE) system modulates attentional allocation97 and NE reuptake inhibitor are used for treatment of ADHD.98 Modulation of NE levels would therefore be a plausible candidate mechanisms influencing LRTCs. However, taken the inverted U-shaped correlations between the LRTCs and ADHD symptoms, the optimal dose giving rise to maximal LRTCs are likely to grossly vary across individuals.
Funding
This study was supported by funding from the Sigrid Juselius foundation to SP and JMP, Ella and Georg Ehrnrooth Foundation to IMM, and the Academy of Finland (SA 1294761) to JS.
Competing interests
All authors report no biomedical financial interests or potential conflicts of interest.
Acknowledgements
We thank Anna Lampinen for collecting the neuropsychological data.
Abbreviations
- ADHD
- Attention Deficit and Hyperactivity Disorder
- ASRS
- Adult ADHD Self-Report Scale
- BADDS
- Brown Attention Deficit Disorder Symptom Assessment Scale
- BIS
- Barrett Impulsiveness Scale
- CPT
- Continuous Performance Task
- DAN
- Dorsal Attention Network
- DFA
- Detrended Fluctuation Analysis
- EOG
- Electro-oculography
- FA
- False Alarm
- FDR
- False Discovery Rate
- HR
- Hit Rate
- ICA
- Independent Component Analysis
- IQ
- Intelligence quotient
- lPFC
- lateral Prefrontal Cortex
- LRTC
- Long-Range Temporal Correlation
- MEG
- Magnetoencephalography
- mPFC
- media Prefrontal Cortex
- mPPC
- medial Parietal Cortex
- NE
- Norepinephrine
- PPC
- Posterior Parietal Cortex
- RS
- resting state
- RT
- Reaction Time
- RTV
- Reaction Time Variability
- SBM
- Surface-based Morphometry
- SOA
- Stimulus Onset Asynchrony
- TSDT
- Threshold of Stimulus Detection Task
- VAN
- Ventral Attention Network
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.
- 20.↵
- 21.↵
- 22.
- 23.
- 24.
- 25.
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.
- 33.↵
- 34.
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.
- 69.
- 70.↵
- 71.
- 72.
- 73.↵
- 74.
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.
- 87.
- 88.
- 89.
- 90.
- 91.↵
- 92.↵
- 93.
- 94.
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.