The Developmental Chronnecto-Genomics (Dev-CoG) Study: A Multimodal Study on the Developing Brain

Abstract

Brain development has largely been studied through unimodal analysis of neuroimaging data, providing independent results for structural and functional data. However, structure clearly impacts function and vice versa, pointing to the need for performing multimodal data collection and analysis to improve our understanding of brain development, and to further inform models of typical and atypical brain development across the lifespan. Ultimately, such models should also incorporate genetic and epigenetic mechanisms underlying brain structure and function, although currently this area is poorly specified. To this end, we are reporting here a multi-site, multi-modal dataset that captures cognitive function, brain structure and function, and genetic and epigenetic measures to better quantify the factors that influence brain development in children originally aged 9–14 years. Data collection for the Developmental Chronnecto-Genomics (Dev-CoG) study (http://devcog.mrn.org/) includes cognitive, emotional, and social performance scales, structural and functional MRI, diffusion MRI, magnetoencephalography (MEG), and saliva collection for DNA analysis of single nucleotide polymorphisms (SNPs) and DNA methylation patterns. Across two sites (The Mind Research Network and the University of Nebraska Medical Center), data from over 200 participants were collected and these children were re-tested annually for at least 3 years. The data collection protocol, sample demographics, and data quality measures for the dataset are presented here. The sample will be made freely available through the collaborative informatics and neuroimaging suite (COINS) database at the conclusion of the study.

1. Introduction

As neuroimaging methodologies mature, investigators have increasingly focused on understanding the dynamic changes in both brain structure and function that occur throughout childhood. These studies are in part motivated by the need to understand how the brain differs in children with developmental disorders to improve diagnosis and treatment. Examination of the developing brain is further motivated by recent studies indicating that multiple adult psychiatric disorders originate in childhood. The growing field of the Developmental Origins of Health and Disease (DOHAD) (Wadhwa et al. 2009), which emphasizes the impact of fetal environment on health, underlines the importance of developmental factors to improve health across the lifespan. Despite clear changes in both brain structure and function from infancy to adulthood (de Graaf-Peters and Hadders-Algra 2006), most developmental studies have remained as unimodal reports such that we have gained knowledge of structure and function independent of each other. Due to the complex relationship between structure and function and the large variation in environmental variables in human studies, a more complete understanding of brain development requires a multimodal approach that captures both hemodynamic and neurophysiological measures across a longitudinal cohort (Calhoun and Sui 2016).

A number of large pediatric cohorts have provided important information with regards to the development of brain structure and function during adolescence using MRI. For example, the Philadelphia Neurodevelopmental Cohort (PNC) included 1000 children aged 8–21 years and collected structural (T1, T2 and diffusion scans) and functional MRI sequences during rest, a fractal N-back task and an emotion identification task, in combination with genetic information that was gathered previously from these children (Satterthwaite et al. 2016; Satterthwaite et al. 2014). The Pediatric Imaging, Neurocognition and Genetics (PING) study similarly collected 735 structural scans (T1, T2, and diffusion scans) and resting functional MRI (Fjell et al. 2012; Ostby et al. 2012) in children aged 4–21 years. Additionally, the National Consortium on Alcohol and NeuroDevelopment in Adolescence (NCANDA) study is examining risk factors for drinking within an adolescent population (N=808; 12–21 years) and includes psychosocial measures in addition to structural and functional MRI measures (Pfefferbaum et al. 2016). The IMAGEN study, a European based consortium, collected structural and functional MRI data, genetics and behavioral assessments from 2000 adolescents at ages 14, 16, 19 and 22 years (Maricic et al. 2020, Schumann et al. 2010). Interestingly the Healthy Brain Network study took a different approach of scanning a small number of adults (N=13) over many scan sessions (N=12) to provide reliability data for healthy brain networks (O’Connor et al. 2017). The human connectome project – development (HCP-D) has extended the initial HCP study to examine the brain connectome in 1300 children 5–21 years of age with a subset of individuals 9–17 years of age receiving repeat scans to capture changes across puberty (Sommerville et al. 2018). Finally, the ongoing Adolescent Brain Cognitive Development (ABCD) study will collect longitudinal data in 10,000 children starting at 9–10 years with longitudinal scans every two years into early adulthood, while collecting structural (T1, T2, and diffusion scans) and functional (rest, monetary incentive delay task, stop signal task and the emotional N-back task) MRI, with an emphasis on understanding the precursors to addictive behavior (Casey et al. 2018). However, these large cohort studies mostly do not include collection of neurophysiological data captured by either EEG or MEG, although a subset of the N-CANDA cohort (N=40) were examined using EEG during a standard polysomnography protocol (de Zambotti et al. 2016). There remains a notable lack of multimodal cohort studies that include MRI, genetic, epigenetic, and neurophysiological measures to understand the transition into adolescence in a comprehensive way.

One of the most consistent developmental changes is the myelination of white matter tracts that primarily occurs after birth. The peak rate of myelination occurs in the first two years of life (de Graaf-Peters and Hadders-Algra 2006), but white matter tracts continue to develop and myelinate well into adulthood with cerebral white matter volume peaking in the 4^th decade of life (Jernigan et al. 2011; Lebel et al. 2017). Myelination has a direct impact on the rate of information transfer between different brain regions by increasing neuronal conduction times. The effect of myelination is demonstrated in part by the ubiquitous findings that evoked response latencies decrease with increasing age across childhood (Allison et al. 1984). Decreases in latency occur until young adulthood and persistent decreases in latency are observed across broad age ranges as well as with studies examining a narrower age range (Gage et al. 2003; Roberts and Hall 2008; Stephen et al. 2017). For example, latencies of the auditory M100 peak localized to auditory cortex are inversely correlated with age (Roberts et al. 2008) in children 7–13 years of age. While most of the large developmental studies have utilized MRI to examine brain development, MRI does not provide the high temporal resolution needed to track changes in conduction time throughout the developing brain. Both MEG and EEG provide direct measures of neuronal responses and reliably report decreasing peak latencies with increasing age until approximately the 4^th decade of life at which point demyelination begins and latencies again lengthen with a slower time constant across adulthood (Albrecht et al. 2000; Allison et al. 1984; Lippe et al. 2009; Paetau et al. 1995; Pihko et al. 2009; Stephen et al. 2017; Wakai et al. 2007). Therefore, MEG/EEG provides complementary temporal information to MRI.

Neural oscillations are ubiquitous in mammalian brains and are recognized as resulting from activation of populations of neurons within a complex cortical network. Small changes in the timing parameters of either local or distant connectivity parameters lead to changes in these neural oscillations either by desynchronizing the activity – resulting in decreased oscillatory power, or alterations in oscillation frequency. Therefore, developmental changes in cortical timing parameters are also expected to alter neural oscillations across development. For example, Cho and colleagues (2015) demonstrated a U-shaped change in gamma power in response to auditory steady-state responses in children aged 8–22 years – demonstrating a rapidly changing profile of gamma oscillations across this age range. They proposed that this may be related to the slow shift from GABA-A α1 to α2 receptors, which changes the time constant of inhibitory neurons leading to faster kinetics (Hashimoto et al. 2009). This change in neuronal kinetics allows for faster recovery time of the individual neurons leading to the capacity to generate higher frequency oscillations. This is consistent with the EEG and MEG literature across the broad developmental span which indicates that low frequency power decreases with increasing age, whereas high frequency power increases with increasing age (Clarke et al. 2001; Somsen et al. 1997). Furthermore, specific physiological frequency bands shift in frequency with increasing age. For example, sensorimotor mu rhythm is functionally defined by the change in mu rhythm power between active versus rest conditions of the sensorimotor system. The frequency of maximal responsiveness to this manipulation increases with increasing age such that the maximal peak frequency is 6–9 Hz in infants and evolves to 8–13 Hz in adults (Berchicci et al. 2011; Saby and Marshall 2012). Therefore, neural oscillations change in multiple ways across development revealing both local and global changes in timing parameters. However, beyond these broad changes across the frequency spectrum, our understanding of the role of neural oscillations in cognitive development in children remains poor.

Both genetics and environment interactively shape the developmental trajectories of brain and cognition. Epigenetic factors, without changing DNA sequences or regulating gene expression, influence both genetics and environmental insults (Jaffe et al. 2016; Kundakovic and Champagne 2014). Using DNA methylation as an example, the process by which methyl groups are attached to mostly Cytosine-Guanine dinucleotides (CpG) in DNA sequence, is known to vary over time (Horvath 2013), and is affected by genetic variation (Zhang et al. 2010), life style (Lim and Song 2012; Ronn et al. 2013), environmental stress (Seifuddin et al. 2017; Zannas et al. 2015), and exposure to substance use (Kyzar et al. 2016; Zhang et al. 2013). Also, a longitudinal study on DNA methylation, prenatal stress and brain structure (Walton et al. 2017) has shown that methylation in the gene SP6 was associated longitudinally and positively with amygdala:hippocampal volume ratio at birth and age 7, and was increased by stressful life events during pregnancy at both time points. Furthermore, the amygdala:hippocampal volume ratio was correlated positively with psychosis‐like symptoms at age 18 years, and was larger in patients with schizophrenia compared to controls. Another recent study examined the relations among cortisol levels, DNA methylation age and hippocampal volume in 46 adolescent girls, and found cortisol production was associated with accelerated DNA methylation age. This in turn was associated with reduced left hippocampal volume, suggesting accelerated DNA methylation in the aging process may be an epigenetic marker linking hypothalamic–pituitary–adrenal axis dysregulation with neural structure (Davis et al. 2017). These exciting findings on DNA methylation provide both challenges and promising directions to understand the factors that influence the developmental trajectory of the brain, but epigenetics remains a rather new biological measure yet to be explored.

Rationale

The goal of the current study was to obtain a comprehensive dataset spanning multiple time scales to gain a broader understanding of the environmental or epigenetic factors that impact brain development and neural oscillations across the 9–17 year age range. The study collected both structural and functional brain measures using MRI, consistent with other neurodevelopmental cohorts (e.g., PING, PNC, ABCD, and NCANDA), while also adding the assessment of neural oscillations and evoked activity with MEG. We chose MEG based on its good spatial and high temporal resolution (Baillet 2017). Through the use of multimodal imaging and a cohort-sequential longitudinal design, this study examined brain development across multiple time scales from milliseconds (MEG), to seconds (MEG & fMRI), to minutes (MEG & fMRI), to years (longitudinal MEG, fMRI, DWI & sMRI) as depicted in Figure 1.

Figure 1.

A depiction of the Chronnectome.

To understand the role of environment and genetic profile, the participants and their parents completed a series of cognitive testing protocols, demographic and behavioral questionnaires, and provided saliva samples for the assessment of genetic and DNA methylation factors.

2. Methods/Design

2.1. Cohort description

The cohort was recruited from two sites: The Mind Research Network (MRN) located in Albuquerque, New Mexico and the University of Nebraska Medical Center (UNMC) located in Omaha, Nebraska. The study is funded through the National Science Foundation’s Established Program to Stimulate Competitive Research (NSF EPSCoR), which is designed to increase research capacity within a subset of states in the USA. The studies were approved by the relevant institutional review board at each data collection site (Advarrra IRB – MRN and UNMC IRB – Nebraska) and data sharing across study sites was written into the consent forms and the study protocols. All study protocols were approved by the IRB prior to study initiation and the research was carried out in compliance with the Declaration of Helsinki. The de-identified dataset will be made available to the broader research community at the completion of the study.

The recruitment goal was set to recruit 100 participants from each site in year one with additional participants recruited in subsequent years to account for attrition across longitudinal study visits. After a 5-month site integration, calibration, and study preparation period, each site met the goal of recruiting 100 participants over a 15-month time span. Individuals were recruited from the local communities through advertisements, posted flyers and community events. The demographics from each site matched the local demographics in racial and ethnic categories.

2.2. Participant recruitment and study enrollment

Parents were engaged for the phone screening portion of the study on behalf of their child. First, the study was described with an emphasis on the longitudinal visits. If the parent/child remained interested in the study, the participant was recruited to participate based on the inclusion and exclusion criteria. The inclusion criteria for the study were: English speaking, age 9–14 years at enrollment and both child and parent were able and willing to assent/consent to the study. The exclusion criteria for the study were: current pregnancy, unable to consent/assent, history of developmental delays, history of developmental disorders, history of epilepsy or other neurological disorders, parental history of major psychiatric or neurological disorders, self-reported prenatal exposure to alcohol or drugs, medication use, an individual education plan indicative of a developmental delay/disorder, contraindication to MRI (MRI screening form was reviewed), and orthodontia (e.g. braces or spacers). Regular medications that were not excluded were: asthma and allergy medications, laxatives, and vitamins. We encouraged parents not to participate if orthodontia were planned within the study time frame. If their child qualified, the participant was scheduled for their year 1 visit. The inclusion criteria were broad with the intention of capturing variance within the typically developing population of children defined based on whether the child was diagnosed with known developmental or psychiatric disorders. This approach likely does not exclude children with psychopathology, but typically excludes children who are taking psychotropic medication.

Once parents and their child agreed to be a part of the study, they were scheduled for their first study visit. At the first study visit, the entire study was described in detail including the benefits and risks associated with study participation. After all questions about the study were addressed, one parent was asked to complete the signed informed consent and the child was asked to complete the child assent. Written informed consent/assent was obtained prior to initiation of study procedures. The term Session refers to the annual visit. Annual sessions were divided into two equally-timed visits of ~2.5 hours.

2.3. Neuropsychological Testing

Dev-CoG participants were asked to complete a wide range of assessments that examined cognitive and mental processes (e.g. attention, memory, motor skills). Session 1 included the following measures:

• The Wechsler Abbreviated Scale of Intelligence (Second Edition; WASI-II; (Wechsler 2011)) was used to assess intelligence. Participants were administered all four subtests: Vocabulary, Similarities, Block Design and Matrix Reasoning; the composite scores for verbal comprehension and perceptual reasoning abilities, and Full-Scale IQ were estimated.

• The NIH Toolbox Cognition Battery (NIHTB-CB; (Weintraub et al. 2013)) assesses six cognitive domains: Attention, Episodic Memory, Executive Function, Language, Processing Speed and Working Memory. The battery yields a composite score for Cognitive Function, Fluid Cognition and Crystallized Cognition. All assessments were administered using the NIH Toolbox iPad application during all three Sessions, and at each administration the full NIHTB-CB age-appropriate assessments were completed.

• The participants also completed the NIH Toolbox 9-Hole Pegboard Dexterity Test (Reuben et al. 2013), which is a part of the NIH Toolbox Motor Battery and assesses motor dexterity.

During Session 2, participants were asked to complete the following tests in lieu of WASI-II. The remaining assessments listed above were completed for each Session:

• The Rey-Ostereith Complex Figure Test (CFT, (Lezak 2004)) assessed visual-spatial perception, memory and organization.

• The California Verbal Learning Test-Children’s version (CVLT–C; (Delis et al. 1994)) recall trials measured a child’s verbal learning curve and memory abilities.

• The Children and Adolescent Memory Profile (ChAMP; (Sherman and Brooks 2015)) Objects and Places Memory Subtest measured a child’s short and long-delay visual domain memory.

• The Delis–Kaplan Executive Function System (D–KEFS; (Delis et al. 2001)) measured a child’s executive function abilities. Dev-CoG participants were asked to complete two of the D-KEFS Verbal fluency subtests (i.e. F-A-S and Animals/Boy’s Names).

During Session 3, participants were asked to complete the following:

• The Attention and Executive Function tests of the Developmental NEuroPSYchological Assessment (NEPSY; (Korkman et al. 2007)). This includes the animal sorting, auditory attention response set, clocks, design fluency (9–12 year olds only) and Inhibition tests.

2.4. Demographic Questionnaires

2.4.1. Child Questionnaires

• Physical Activity Questionnaire for Older Children (PAQ-C; (Kowalski et al. 2004)). The PAQ-C assessed physical activity in children over the past seven days. It accounts for any unusual activity during the previous week which is not part of the total summary activity score.

• The Edinburgh Handedness Inventory (EHI; (Oldfield 1971)) was used to assess handedness.

2.4.2. Parent Questionnaires

• Demographics Questionnaires: During Session 1, parents were asked about the participant’s age, gender, ethnic/racial background, medical history, language(s) spoken, and education level. Parents were asked additional questions about the participant’s birth information (e.g. birth weight and preterm birth), mental health history and the family’s mental health history for Session 2 at MRN. MEG/MRI contraindications were also assessed. During remaining sessions, the questionnaire inquired about any recent changes (e.g. medical history) in the participant’s life since their prior visit.

• Guide to Measuring Household Food Security (Bickel et al. 2000) assessed the food security status within the home over the past 12 months.

• Barratt Simplified Measure of Social Status (BSMSS; (Barratt 2006)) assessed the participant’s social status and provided estimate scores for the parent’s education, occupation and overall SES.

2.5. Social & Emotional Questionnaires

2.5.1. Child Questionnaires

Participants were administered questionnaires that asked about their behavioral and emotional strengths and difficulties. Younger children were assisted while the older children completed the questionnaires on their own. Where necessary, the scales were reproduced via special permission from the publisher and adapted for digital administration. These questionnaires were completed at all Sessions with the exception of the Trauma questionnaires, where the children only repeated these if they had experienced any additional traumas since the prior Session. The measures included:

• Mood and Feelings Questionnaire: Short Version, Child Self-Report (MFQ-C; (Angold and Costello 1987)) that assessed how the participant has been feeling or acting in the past two weeks.

• Trauma Symptom Checklist for Children (TSCC; (Briere 1996)) measured internalized distress, including symptoms associated with Post-Traumatic Stress Disorder (PTSD).

• Trauma History Profile (THP; (Pynoos and Steinberg 2002)) was derived from the trauma history portion of the UCLA PTSD Reaction Index for DSM IV (Steinberg et al. 2004) and assessed any form of maltreatment and other forms of trauma.

• Strengths and Difficulties Questionnaire-Child (SDQ-C; (Goodman et al. 1998)) assessed a child’s behavioral and emotional problems.

• Participants also completed the age-appropriate sections of the NIH Toolbox Emotion Battery (NIHTB-EB; (Salsman et al. 2013)). The NIHTB-EB measured emotional health across four domains: social relationships, psychological well-being, negative affect and stress and self-efficacy. All assessments are administered using the NIH Toolbox iPad application.

• The Perceived Stress Survey assessed the participant’s perception of their coping resources in distressful situations. This measure was only given to participants aged 13–15 years.

2.5.2. Parent Questionnaires

Parents were asked to complete self-administered questionnaires behavioral and emotional strengths and difficulties, and home environment. These questionnaires were administered during all Sessions, unless stated otherwise. The questionnaires include:

• Strengths and Difficulties Questionnaire-Parent (SDQ-P; (Goodman 1997)) measured a child’s behavioral and emotional problems similar to the SDQ-C. The two age-appropriate versions were used for participants 9–10 years and 11+ years. A total difficulty score was derived by summing all problem scales.

• Child Behavior Checklist - Parent Report/Adaptive Functioning (CBCL-PR; (Achenbach 1991)) is a standardized tool that measures a child’s behavior and emotional difficulties. The questionnaire yielded scores for internalizing and externalizing behaviors and the child’s adaptive functioning.

• Conners ADHD Rating Scale 3- Parent Short Form (C3-PSF; (Conners 2008)) was used to assess Attention Deficit/Hyperactivity Disorder (ADHD) behaviors although children with diagnosed ADHD at recruitment were excluded from the study.

• Parenting Stress Index-Short Form (PSI-SF; (Abidin 2012)) provided scores for parental distress, parent-child dysfunctional interaction and difficult child subscales.

• NIH Toolbox Emotion Parent Proxy Battery (Salsman et al. 2013) was completed by parents of participants aged 9–12 years.

2.6. Neuroimaging

The following tasks were performed by each participant during MEG and/or MRI data acquisition. For all behavioral tasks that required a response, practice was provided before scanning. Additional tasks were added/removed in subsequent sessions and at the two sites.

Resting State -- Eyes Open and Eyes Closed (MRN & UNMC; All Sessions; MEG and fMRI):

Participants completed eyes open (EO) and eyes closed (EC) resting state scans in both MEG and fMRI. The order of EO/EC scans was counterbalanced between participants, with EO/EC order maintained for each participant across MEG/fMRI sessions for all follow-up visits. The order was randomized based on whether the randomly assigned subject number was even or odd. In MEG, the EO scan was preceded by 30 seconds of prompted eye blinks. For both MEG and fMRI, each EO and EC scan was recorded for 5 minutes while participants rested quietly without thinking of anything in particular. For the EO scans, participants were instructed to keep their eyes open and focused on the central fixation cross. For the EC scans, participants were instructed to close their eyes and to not fall asleep.

Multisensory (MRN & UNMC; All Sessions; MEG and fMRI):

The multisensory task was performed across all sessions in both MEG and fMRI. Prior research shows that multisensory facilitation does not reach mature levels until about 14 years of age (Brandwein et al. 2011). The visual stimulus was a full-screen, black and white vertical grating (0.25 cycles/degree). The auditory stimulus was a 40 Hz modulated 1000 Hz tone. In the multisensory condition, the auditory and visual stimuli were presented simultaneously. Baseline fixation was a gray screen with a red fixation box (1°) in the center. Participants were instructed to press the index finger as soon as they saw anything, heard anything, or both.

In MEG, each trial began with fixation for an inter-trial interval (ITI) pseudo-randomly jittered between 2400–2600 milliseconds (ms) in 10 ms increments. Following fixation, a sensory stimulus (auditory, visual, or audio-visual) was presented for 800 ms. Participants were instructed to press the index finger as soon as any stimulus appeared (and/or was heard). Children performed a single block of 300 trials, with trials in pseudo-randomized order. Task duration was approximately 18 minutes. The task was coded in E-Prime software (https://pstnet.com/products/e-prime/).

In fMRI, each trial began with fixation with an ITI of (MRN: 3680, 4600, or 5520 ms or UNMC: 3340, 4260, 5180); ITIs were equally distributed between conditions. Following fixation, a sensory stimulus (auditory, visual, or audio-visual) was presented for 800 ms. Participants were instructed to press the index finger as soon as any stimulus appeared (and/or was heard). Children performed two runs of 42 trials (84 total trials), with trials in pseudo-randomized order. Two blocks were separated by a self-timed break. Task duration was approximately 7 minutes. The task was coded in Presentation software (https://www.neurobs.com/) at MRN and E-Prime at UNMC.

Motor Complexity (MRN & UNMC; All Sessions; MEG):

This task (Heinrichs-Graham and Wilson 2015a) was utilized to assess movement planning and enabled comparison of motor preparation for simple versus complex motor sequences. Simple sequences (50% of trials) were number sets in sequential order (i.e.: 1-2-3 or 2-3-4), whereas complex sequences (50% of trials) were non-sequential (i.e.: 4-1-3 or 2-4-1).

A five-button response glove was worn on the right hand, with the following number-finger pairings: 1=index, 2=middle, 3=ring, 4=pinky; the thumb was not used. Children learned and became familiar with the number-finger pairings and task procedures during a practice session. For each trial, participants viewed a baseline fixation cross (3700 ms), followed by a set of three numbers in black type (500 ms), describing the sequence. Children were instructed to press the corresponding buttons in sequence (as presented from left to right), as soon as the numbers changed color from black to blue. The “cue to move” number set was displayed in blue (2200 ms), during which time button presses were recorded. Participants completed two blocks of 80 trials each (160 total trials), with trials in pseudo-randomized order and a self-timed intermission between blocks. Total task duration was approximately 16 minutes.

Verbal Working Memory (MRN & UNMC; Session 1; MEG):

This Sternberg-type Verbal Working Memory (VWM) task (Sternberg 1966) has been used to understand the oscillatory activities underlying verbal memory storage, maintenance, and retrieval (Heinrichs-Graham and Wilson 2015b; Proskovec et al. 2016). In this task, each trial began with the presentation of a baseline fixation grid consisting of 6 empty boxes for 1300 ms. During the encoding phase, letters were presented in each of the 6 boxes for 2000 ms. Subsequently, the letters disappeared and were replaced by an empty grid during the maintenance phase (3000 ms). In the retrieval phase (900 ms), a single letter appeared in the upper-middle grid box. Participants were instructed to indicate by button press whether the single letter matched any of the letters displayed in that trial’s encoding phase (“index finger”=match, “middle finger”=non-match). The trials were presented in pseudo-random order, with half containing matching letters. Children first practiced the task and then performed two blocks of 64 trials (128 trials total), with a self-timed break between blocks, for a total duration of ~16 minutes.

SART - Random (MRN; Session 2; MEG):

The Sustained Attention to Response Task (Robertson et al. 1997) is commonly used to measure the ability to withhold a response to rare, unexpected stimuli amidst a stream of rapid responses to frequent stimuli. In this version, numbers (1–9) were presented one at a time, in random order, with instructions to press the index finger upon appearance of any number except the number “3”, during which they should withhold their response. On each trial, participants viewed a blank fixation screen for 280 ms, followed by the rapid presentation of a number (1–9) (148 ms), and then a blank screen which appeared for a short interval (670 ms). Responses were registered during the number presentation as well as during the subsequent interval. 945 trials were performed in succession, resulting in total task duration of ~15 minutes.

Modified Cambridge Gambling Task (MRN; Session 3; MEG).

The modified Cambridge Gambling task (mCGT) was designed to eliminate the need for a touch screen response as is required for the Cambridge Gambling Task (CGT) in the CANTAB battery, while maintaining the different aspects of the CGT that includes a reduction on the working memory load in comparison to the Iowa Gambling Task. The task is similar to the Cake task described by (Van Leijenhorst et al. 2008) and captures the relation between betting risk relative to probability and impulsivity. One advantage of this version of the task is that central fixation is maintained throughout the task, minimizing the confounding artifacts related to eye movement. Also, unlike the CGT, the mCGT response was made with a button press minimizing movement and motor-related artifact during MEG data collection. The participant was presented with a pie chart with different proportions of red/blue wedges on each trial. The task was to first predict whether a yellow token was hidden under a red or blue wedge. Participants were instructed to press a button with their left index finger to choose red and right index finger to choose blue. Once the color prediction was chosen, the subject then decided what percentage of their current points to wager that their prediction (red/blue) was true. Wager choices were presented in either ascending or descending order and the participant was instructed to press the button with their right thumb when the wager they chose appeared. Total points were presented below the wager amount. Once the wager was placed, the final screen appeared in which the location of the yellow token was revealed and the resulting “win” or “loss” was presented along with the updated total points. This task took approximately 20 minutes to complete.

Vis-Attend (UNMC; Session 2; MEG):

This visuospatial attention task (Wiesman et al. 2017; Wiesman et al. 2018) required participants to make judgements based on the spatial positon of stimuli. During the task, participants fixated on a crosshair presented centrally. After a variable ISI (range: 1900–2100 ms), an 8×8 grid was presented for 800 ms at one of four positions relative to the fixation: above right, below right, above left, or below and to the left. The left/right orientations were defined as a lateral offset of 75% of the grid from the center of fixation. Before the task began, participants were instructed to respond via button press with their right hand whether the grid was positioned to the left (index finger) or right (middle finger) of the fixation point upon presentation of the grid. Each participant performed 240 repetitions of the task concurrent with MEG recording.

Abstract Reasoning (UNMC; Sessions 2 and 3; MEG).

Abstract reasoning is a type of fluid intelligence, or a person’s ability to apply logical rules and patterns to presented information. It is one of the last developmental milestones to be reached. In this study, we evaluated abstract reasoning ability using a non-progressive, modified version of Raven’s matrices (Raven 1936; Raven et al. 2003). Participants were initially presented with four empty boxes (2 × 2 matrix) for 2750 +/− 250 ms, which served as the inter-stimulus baseline period. One of the two bottom boxes was outlined in white. These boxes were then filled with various shapes for 4000 ms. The participant responded as to whether the pattern of shapes in the highlighted box correctly completed the pattern, such that the relationship between the boxes in the bottom row matched the relationship found between the boxes in the top row (“index finger” = yes, “middle finger” = no). Relationships could be matched in color, position, order, or shape. Participants first practiced the tasks, then completed two blocks of 60 trials (120 trials total), with a self-timed break between blocks, for task duration of about 14 minutes.

Flanker Task (UNMC; Session 3; MEG).

This selective attention and response competition task (McDermott et al. 2017) required participants to indicate the direction (left or right) of a central arrow stimulus. Each trial began with a fixation that was presented for an interval of 1450–1550 ms. A row of 5 arrows was then presented for 2500 ms and participants were instructed to indicate with their right hand whether the middle arrow was pointing to the left (index finger) or right (middle finger). The 200 total trials were pseudo-randomized and equally split between congruent and incongruent conditions. In the congruent condition, the four flanking arrows (two per side) pointed in the same direction as the middle arrow, whereas the four flanking arrows pointed in the opposite direction as the middle arrow in the incongruent condition. Of note, the left and right pointing arrows were equally represented in the congruent and incongruent conditions, and the overall MEG recording time was about 14 minutes for the task.

2.6.1. MRI parameters

MRI data at the Mind Research Network were collected on a Siemens 3T TrioTim, with a 32-channel radio frequency coil. MRI data at UNMC were collected on a Siemens 3T Magnetom Skyra, with a 32-channel radio frequency coil. In addition to the fMRI sessions described above, a T1 weighted structural scan, a diffusion tensor imaging sequence, and distortion correction sequences were obtained (see Table 1 for MRI parameters).

Table 1.

MRI parameters

	FOV (mm)	Resolution (mm)	Flip Angle	TE (ms)	TR (ms)	Slices orientation	AF/MB	Time (min:s)
MRN
Localizer	280×280	1.9×1.5.8.0		5	20	1/Sag		0:10
Distortion_corr_PA	248×248	3.0×3.0×3.0	90/180	73	7220	56/Trans	1	0:22
Distortion_corr_AP	248×248	3.0×3.0×3.0	90/180	73	7220	56/Trans	1	0:22
Rest fMRI (open/closed)	248×248	3.0×3.0×3.0	44	29	460	56/Trans	8	5:06, 5:06
Task fMRI (Multisensory*2)	248×248	3.0×3.0×3.0	44	29	460	56/Trans	8	3:57, 3:57
T1w	256×256	1.0×1.0×1.0	8	1.9	2400	192/Sag		5:45
DTI (AP/PA) b-value 2400 s/mm2 (44, 47, 42, 40)	224×224	2.0×2.0×2.0	84/157	108	4000	72/Trans	3	3:12, 3:24, 3:04, 2:56
UNMC
Localizer		1.5×1.5×8	40	5	20	1/Sag		0:10
Distortion_corr_PA	268×268	3.3×3.3×3.0	90/180	73	7220	48/Trans	1	0:22
Distortion_corr_AP	268×268	3.3×3.3×3.0	90/180	73	7220	48/Trans	1	0:22
Rest fMRI (open/closed)	268×268	3.3×3.3×3.0	44	29	460	48/Trans	8	5:06, 5:06
Task fMRI (Multisensory *2)	268×268	3.3×3.3×3.0	44	29	460	48/Trans	8	3:57, 3:57
T1w	256×256	1.0×1.0×1.0	8	1.94	2400	192/Sag		5:45
DTI AP/PA (b-value 2400 s/mm2) diffusion directions (44, 47, 42, 40)	224×224	2.0×2.0×2.0	90/180	108	4000	72/Trans	3	3:12, 3:24, 3:04, 2:56
Hippocampal Volume	170×170	0.4×0.4×2.0	150	70	7790	30/Trans		6:23

Abbreviations: AP- anterior/posterior; PA – posterior/anterior; Sag – sagittal; AF/MB – Acceleration factor/Multi-band factor

2.6.2. MEG parameters

MEG data at both sites were collected using a whole-cortex 306-channel MEG system (Elekta Neuromag/York MEGIN). Prior to scanning, 3D digitization was performed (Polhemus Fastrak), to collect positioning data for fiducials, four head-position indicator (HPI) coils, and the scalp surface. The scalp surface was used to optimize the co-registration of MEG and structural MRI data. The HPI coils were continuously monitored throughout the recording, which enabled offline head movement correction. Electro-oculogram (two channels) and electrocardiogram (two channels) were also recorded along with the MEG at the MRN site; frontal MEG channels were used to identify these artifacts for the UNMC site. EOG and ECG artifacts were identified and removed during pre-processing using SSP at both sites (Uusitalo and Ilmoniemi 1997). MEG data were sampled at 1 kHz, with an acquisition passband of 0.1 to 330 Hz.

2.7. Genetics

Saliva samples were collected at each annual visit for the purpose of performing genome wide analysis of single nucleotide polymorphisms (SNPs) and DNA methylation. The SNP and methylation analysis of the first two sessions were performed in one batch to eliminate batch effects. The saliva was collected using the DNA Genotek, Inc. Oragene saliva collection kits. A minimum of 2 mL of saliva was collected immediately following the MRI session. Participants were not provided with food, water or gum prior to saliva collection. Participants were encouraged to take their time in providing the required amount of saliva. Once sufficient saliva was collected, the Oragene preservative was released into the saliva and the sample was marked with the subject number and date of collection.

DNA SNPs were measured by Illumina Omni express exome v1.6 array. DNA methylation was measured by Illumina Infinium Methylation EPIC array, which covers over 850,000 methylation sites at single nucleotide resolution. It includes 99% of RefSeq genes, 95% of CpG islands, high coverage of enhancer regions, and other content categories.

2.8. Adjustments to Task design

The initial design of the study was intended to collect repeat assessments on all measures at the annual visits (Sessions 1–3). However, a number of assessments used are relatively stable across childhood including IQ measures and may lead to artificially increased scores with multiple repeat tests due to practice effects. Therefore, the WASI was only administered in Session 1 and additional cognitive measures were added in subsequent sessions to gain a broader understanding of the cognitive profile of the cohort (see above).

Furthermore, in Session 2 a finger prick procedure was performed using commercially available lancers and collection of dry blood spots on filter paper. The blood was then assessed for lead levels within a 6-month window through the local laboratories. At MRN, blood collection occurred during Session 2 MEG visit. At UNMC, blood collection occurred during the Session 3 MEG visit. Standard protocol was to clean the finger, lance the finger, discard the first drop of blood and collect at least two blood spots on filter paper, clean the finger again and apply a bandage. This procedure was optional and not all children chose to participate in the finger prick procedure. Approximately 93% of children at MRN agreed to participate and were successful in providing the required number of blood spots.

Finally, pubertal status was assessed by measuring testosterone, estradiol, and DHEA based on a saliva sample obtained during Session 2. The child was requested to provide an additional 0.5 mL of saliva which was collected in the same saliva collection tube and pipetted to a separate storage tube after saliva collection – before the Oragene preservative was mixed with the saliva for DNA preservation. The saliva for pubertal status was then stored in a −20 (UNMC) or −80 (MRN) degree freezer until processing.

2.9. Data Capture

Identifiable subject information was stored in the online HIPAA-compliant MICIS database. Upon enrollment, the subject was assigned an 8-digit randomized subject number. All assessments were entered into the Collaborative Informatics and Neuroimaging Suite (COINS; https://coins.trendscenter.org; (Scott et al. 2011)) database, with direct data entry occurring for all questionnaires for both children and parents and electronic transfer of data from the NIH Toolbox to COINS. Neuropsychological assessments that required offline data collection were double entered after scoring. MRI and MEG data were also captured via COINS. Upon data transfer, each imaging session and task was registered as a separate scan session in COINS to log data acquisition and to capture scan notes. Automated analysis pipelines for each MRI modality and for MEG were launched once the data were archived. The data will be openly available to the broader research community at the completion of the study through the COINS (https://coins.trendscenter.org) data exchange (Wood et al. 2014). Upon contact through COINS, data use agreements will be required according to institutional guidelines.

2.9. Preliminary analysis of MEG and DTI data

We have implemented a preliminary multi-modal analysis of MEG and DTI data to demonstrate one method for integrating across the multiple modalities. The jICA approach has been demonstrated for MEG and DTI data previously (Stephen et al. 2013) and is described in detail in this prior publication. Briefly, the MEG data were converted to a single timeseries (vector) to capture the temporal variation across time, condition and cortical region. The DTI FA maps were also converted to a single vector capturing the variation in FA values across the brain for each participant. Each participant represented a row vector in the MEG and DTI matrix. The MEG and DTI matrices were concatenated after the data were normalized to match variance across modalities. This combined MEG/DTI matrix was submitted to an ICA deconvolution to capture covariation of MEG/DTI. In this example we used the source timecourses described in our previous publication from the Verbal Working Memory task described above (Embury et al. 2019). The DTI data were processed using the FSL processing pipeline. We examined whether joint components differed in individuals with high or low D’ on the task, where D’ is the signal detection index of the likelihood that a participant can detect a signal of interest and summarizes false alarm rate and hit rate into one measure.

3. Cohort Description

The age distribution for the Dev-CoG cohort is shown in Table 2. The cohort included 203 participants with 101 recruited at MRN and 102 recruited at UNMC. The goal was to recruit 100 participants at each site and follow the children with up to 3 consecutive annual visits across the 4-year project period. Additional participants were recruited to account for attrition across the longitudinal study. Demographic information and cognitive performance are included in Table 3. The groups across the two sites were well matched with regards to age and gender distribution. The racial and ethnic diversity of both the MRN and UNMC cohorts matched the local demographics. The socioeconomic scale used ranges from 1–66 demonstrating that both sites represented a wide range along the socio-economic scale. There was also considerable range in the IQ measures in both cohorts again demonstrating a representative sample of children that span the full range of intelligence of a typically developing child. Finally, the sample included children who had variable physical activity levels (5-point scale).

Table 2.

Number of participants at each age at enrollment

	MRN		UNMC		Combined		Total
Age	M	F	M	F	M	F
9	14	10	8	8	22	18	40
10	6	9	10	12	16	21	37
11	6	9	12	12	18	21	39
12	9	8	5	10	14	18	32
13	7	10	9	7	16	17	33
14	9	4	7	2	16	6	22
Total	51	50	51	51	102	101	203

Table 3.

Demographics of the study population

Demographics	MRN (101)	UNMC (102)
Age at enrollment Mean±STD (Range)	11.3±1.75 (9–14)	11.2 ± 1.55 (9–14)
Gender (M/F)	51M/50F	51M/51F
Race (N)
Caucasian	86	87
African American	3	5
Asian	2	1
American Indian/Alaska Native	6	2
Native Hawaiian/Pacific Islander	1	2
More than 1 race	1	5
Unknown	2	0
Ethnicity (% Hispanic)	41.6%	7.8%
Handedness (R/L/A)	96/3/1	96/6
Bilingual (%)	23.7%	6.3%
WASI-II IQ Mean±STD (Range)	108.6±15.2 (72–139)	112.06 ±16.12 (68–148)
Overall SES Mean±STD (Range)	42.6±12.3 (17–66)	48.15 ± 11.01 (15–64.67)
Education	16.4±3.4 (5–21)	17.17 ± 2.84 (7–21)
Occupation	26.4±10.6 (1–45)	31.18 ± 8.40 (8–44)
Physical Activity Mean±STD (Range)	2.73±0.77 (1.22–4.69)	2.65 ± .657 (1.45–3.37)

3.1. Neuroimaging Data quality Assessment

We assessed data quality across the sample through examining motion during the resting state scans. We identified 8% of the sessions during eyes open and 8% during eyes closed with sufficient motion to be considered outliers during fMRI and 3% of the sessions during eyes open and 4% during eyes closed as outliers due to motion in MEG. The vast majority of sessions were retained for further analysis (see Figure 2). We also examined the correlation between age and motion to determine if motion may contaminate our interpretation of age-related changes. In most cases, there was not a significant correlation between motion vs. age with fMRI eyes open (r = 0.021, p = 0.76), fMRI eyes closed (r = −.113, p = .12) and MEG eyes open (r = −.118, p = 0.098) not showing significant correlation with age and only MEG eyes closed showing a weak correlation with age (r = −0.15, p = 0.027), which did not remain significant when accounting for multiple comparisons. However, the pattern of correlation between motion and age grew stronger in fMRI across sessions, with a significant correlation with age in session 3 for fMRI eyes open (r = −0.34, p = 0.002) whereas fMRI eyes closed remained nonsignificant (r = −0.19, p = 0.1).

Figure 2.

Motion in fMRI and MEG across sessions and sites.

Finally, we examined head size to determine if this variable should be considered as a covariate in future analyses. We estimated head size based on the total intracrancial volume estimate (eTIV) provided by Freesurfer 5.3, through which all structural MRIs were processed. There was a nonsignificant trend for an increase in eTIV with age (r = 0.122, p = 0.09). Despite having no outliers in the dataset, the 9 year old children had greater eTIV than 10 year old children. This can be understood through the large cluster of children above the regression line in the correlation plot shown in Figure 3.

Figure 3.

There was a nonsignificant trend for an increase in estimated total intracranial volume (eTIV) with age.

Participant Success/Retention Rate

Overall, the retention rate of participants across the three sessions was high. The initial visit for session 1 was always the MEG visit with the MRI visit following. Therefore, Session 1 success rate for MEG was set at 100% (Table 5: individuals who screened in and did not complete visit 1 were not counted). Retention rates are reported as year to year retention (of those who completed Session 2 how many successfully completed Session 3). Session 1 MRI success rate was 96% with some parents children either declining to participate in the MRI or the child being unwilling or unable to complete the MRI. When accounting for new participants who were recruited to account for attrition there were 501 MEG sessions completed and 486 MRI sessions completed. Success rates were not significantly different across sites.

Table 5.

Participant success/retention rates.

	Session 1	Session 2	Session 3
MEG (%/# sessions)	100% (215)	77% (166)	79% (120)
MRI (%/# sessions)	96% (208)	80% (166)	75% (112)

3.2. Cognitive and Behavioral Results

We also examined cognitive and behavioral measures across sites. Overall, the sample was well-matched as shown by the demographic and other questionnaire results in Figure 4. As expected, by the differing demographics of the state of NM vs. NE the children in NM came from lower socioeconomic status (SES) households although this remained at the trend level (t = −1.86, p = 0.064). The number of traumas experienced (χ²= 1.1, p = 0.57), parental stress (t = 0.57, p = 0.57), and physical activity (t = 1.36, p = 0.17) were similar across sites. Despite the similarity in some of these measures, there were significant differences in the moods and feelings questionnaire (t = −2.1, p = 0.033) and the WASI Full scale IQ (t = −2.22, p = 0.027), although it is important to note that both sites had mean IQs above 108 where the population mean is normed to 100 (Table 3).

Figure 4.

Demographic measures compared between NM and NE. SES – socioeconomic status, PAQ – physical activity questionnaire, MFQ – moods and feelings questionnaire.

Other questionnaires also demonstrated many similarities with some differences, Figure 5. For example, attention deficits (greater inattention based on parental report) were higher in children from NM relative to children from NE (t = 2.46, p = 0.014) but internalizing (t = −.57, p = .56), externalizing (t = 0.94, p = 0.34), hyperactivity (t = 0.07, p = 0.94) and Child behavior checklist total score (t = 0.56, p = 0.57) were statistically equivalent between sites.

Figure 5.

Parental questionnaires relating to child behavior. Connors Inattention and Hyperactivity – two summary measures from the Connors parent questionnaire to assess ADHD. CBCL – child behavior checklist

Finally, we examined whether children differed in cognitive performance as assessed by the NIH Toolbox, Figure 6. There were no significant site differences in the dimensional card sort (t = −1.4, p = 0.16) or spatial working memory (t = −1.4, p = 0.15) tasks. However, there were significant differences in the Flanker Inhibitory control task (t = −2.3, p = 0.022), the Oral Reading (t = −4.0, p <0.001), PSM (t = −4.74, p < 0.001) and the picture vocabulary (t = −2.1, p = 0.039) tasks.

Figure 6.

NIH Toolbox measures obtained from the NM and NE sites. DCCS – Dimensional change card sort, SWM – spatial working memory, FICA – Flanker inhibitory control and attention, Oral Reading – Oral reading and recognition, PSM – picture sequence memory, Pic Vocab – picture vocabulary.

Due to the demonstrated effects of SES on cognitive performance, we examined whether site differences remained when including SES as a covariate. With this model, IQ and Connors inattention no longer demonstrated site differences, however the NIH TB measures of oral reading (F(1,171) = 8.1, p = 0.005) and picture sequence memory (F(1,171) = 5.9, p = 0.017) remained significantly different by site.

3.3. Analytic approach and Emerging Results

The data obtained in this project will allow us to expand our assessment of the chronnectome (Calhoun et al. 2014) across development. Initially defined by our group, the chronnectome describes the brain through dynamic nodal activity and varying connectivity patterns that change in predictable and meaningful ways. Prior studies have examined this through assessing first static functional connectivity across minutes using fMRI (Allen et al. 2011) and extending this analysis to dynamic functional connectivity of fMRI resting data (Allen et al. 2014; Sakoglu et al. 2010) to examine changes in these connectivity patterns over the time span of seconds. With the addition of MEG, we will be able to further extend our assessment of the chronnectome to the order of tens or 100s of milliseconds, in addition to examining the chronnectome of neural oscillations across seconds and minutes. At the other extreme, we will also examine changes in the chronnectome across years and developmental stages within the DevCog cohort.

Our preliminary results applying the chronnectomic approach have been published based on the PNC and PING datasets. For example, using the PING dataset Faghiri et al. (2018) identified five connectivity states that were well represented through assessment of the dynamic connectivity patterns across a 6.5 minute resting state fMRI task. The dwell time of certain states was correlated with age indicating either a decrease or increase in dwell time with increasing age. Furthermore, the pattern of transitions between states differed for a subset of the states. These results demonstrate that assessment of dynamic FNC provided unique insights into which cortical networks continue to evolve across middle childhood. Instead, Zille et al. (2018) applied a sparse connectivity pattern approach to examine dynamic connectivity. This particular approach allows for the networks to be different across different age groups as may occur throughout development. Application of this approach to the PNC rs-fMRI dataset demonstrated that the connectivity patterns were weighted differently between children vs. young adults.

Similar developmental changes in the chronnectome have been demonstrated through the analysis of the neural oscillations measured with MEG. For example, Heinrichs-Graham et al. (2018) combined the DevCog data with a previously acquired adult dataset to map the lifespan trajectory of neural oscillations associated with motor control. The results indicate that baseline beta oscillations differ across age with older adults generating greater beta power during rest relative to young adults and children. In contrast, event-related desynchronization decreases linearly with increasing age. However, both baseline beta and absolute beta power during movement demonstrates a nonlinear U-shaped pattern across the age spectrum. This study represented the largest study describing changes in event related beta oscillations across the lifespan to date. Two other recent MEG studies using this dataset focused on the developmental trajectory of working memory (Embury et al. 2018) and motor-related gamma activity (Trevarrow et al. 2018). Both studies found robust developmental changes in neural oscillatory activity during the transition from childhood to adolescence in task-related brain regions.

We also implemented a preliminary multi-modal analysis of the DevCog dataset using the verbal working memory task and DTI data. The results indicate that children with higher D’ show a trend (p = 0.064) for an increased loading factor in one component (see Figure 7). This same component was significantly correlated with age (r = .25, p = 0.027). The MEG showed group differences (high/low D’) in this component in left anterior cingulate, but this component did not describe any significant signal in left postcentral gyrus. This demonstrates how the jICA approach can separate function and structure. The white matter regions associated with this component are shown and correspond with posterior white matter tracts. While these results are preliminary, it provides evidence that joint ICA may be a useful tool to identify joint components that systematically change with age.

Figure 7.

Joint component that was significantly positively correlated with age (r = 0.248). DTI component clusters are shown above and example regions showing regional specificity – significant signal shown in left cingulate cortex (left panel) and no signal represented by this component in left postcentral gyrus (right panel). MEG timecourses were derived from beamformer regions that showed significant activation during a working memory task (Embury et al. 2019).

4. Discussion

The Dev-CoG study provides a multi-site dataset comprised of neuropsychological testing, parent/child questionnaires, a broad set of neuroimaging assessments, and genetic and epigenetic results in typically-developing children aged 9–17 years. To the best of our knowledge, this is the first comprehensive pediatric neuroimaging dataset that includes MEG data in combination with structural and functional MRI, genetics and epigenetics. This is also the first large neuroimaging dataset that includes both genetics and epigenetics with a target number of 483 sessions across sites and years, which was exceeded for both MEG (501) and MRI (486) scan sessions. The cohort includes children with a broad spectrum of cognitive abilities within the normal range as well as stratification across the socio-economic spectrum. Recent multi-modal studies are providing growing evidence of the complementary nature of hemodynamic and electrophysiological approaches in understanding typical and atypical brain function (Cetin et al. 2016; Houck et al. 2017).

Neural oscillations are now recognized to be fundamental components of brain function (Basar et al. 2001; Singer 2011), yet the mechanisms that lead to alterations in these neural oscillations across disorders are still poorly understood. As the brain grows, timing parameters of the neural networks change (Uhlhaas et al. 2009) leading to changes in frequency and spectral power with age. Understanding how neural oscillations assessed with MEG relate to connectivity assessed with fMRI measured at a different timescale, will likely provide us with valuable insight into important neurophysiological changes that unfold during adolescence. These data will also provide key normative data against which developmental disorders can be compared and contrasted in the search for biomarkers and treatment targets, and thereby improve overall outcomes as children and adolescents grow into adulthood.

The DevCog dataset provides within-subject multimodal measures during an important developmental milestone: the transition through puberty. This time window is pivotal for a number of pediatric and adult disorders. For example, a new study tracked development in 845 children 6–10 years of age to identify early markers of psychiatric illness (Muetzel et al. 2018) based on the research indicating that problems in childhood predict psychiatric disorders in adulthood (Hofstra et al. 2002). They found that baseline measures of both internalizing and externalizing behaviors predicted smaller gray matter volume and fractional anisotropy at later ages. However, baseline gray matter and fractional anisotropy measures did not predict behavioral measures at older ages. Therefore, early effects of psychopathology are evident in middle childhood, but these structural measures alone could not predict emergence of psychiatric illness in children. Furthermore, while the ABCD study is designed to track brain development in typically-developing children, the goal of the study is to look for early predictors of addiction as well as to examine the effects of substance use on brain development, based on the robust finding that addiction largely originates in childhood (Burge et al. 2004). However, the ABCD study is not capturing neurophysiological measures (EEG or MEG) of brain development – limiting the brain measures to multimodal MRI. We are motivated to examine neurophysiological measures in children based on the association between neural oscillations and behavior (Cavanagh et al. 2010; Mazaheri et al. 2014; Rivolta et al. 2015; Uhlhaas et al. 2006b; Uhlhaas et al. 2009; Van Der Werf et al. 2013). For example, frontal theta power measured with MEG differed in individuals with a gambling addiction relative to controls and theta power was positively associated with gambling severity (Dymond et al. 2014). At the same time impulsivity has been identified as a common predictor across different types of addiction (Choi et al. 2014). Therefore, it has been proposed that theta oscillations are a key component in mediating response inhibition (Cavanagh and Frank 2014), a well-described deficit in individuals with addiction risk (Norman et al. 2011). We expect the dataset will provide important insights into the links between neural oscillations and how different environmental insults impact both emotional and cognitive development during this time.

Recent fMRI studies have underlined the importance of motion in influencing brain connectivity estimates (Power et al. 2012; Satterthwaite et al. 2012). Therefore, minimizing motion during scanning is important and accounting for age-related changes in amount of motion is also paramount. Interestingly in this study we did not find a correlation between age and motion in either fMRI or MEG resting scans across the full cohort. There was a difference in the amount of motion between EO vs. EC with more motion in EC vs. EO in fMRI and the opposite pattern in MEG. This may be related to the difference in positioning between the two modalities where children were laying down during fMRI but seated during the MEG. We also found that over study visits, the amount of motion increased from Session 1 to Session 3, with a significant association with age present by Session 3. We attribute this pattern to decreased emphasis on reducing movement for return visits, perhaps assuming that the children remained aware of the expectations to minimize motion for the return visits based on their prior experience. This result underlines the importance of continuing to emphasize the need to remain still during scanning for children in longitudinal studies.

The difference in demographic and performance variables across sites described above is partly accounted for by differences in socioeconomic status. Prior experimental and epidemiological studies have demonstrated a relationship between parental SES and a child’s performance in school and on intelligence tests (Hackman and Farah 2009; Johnson et al. 2016; Raizada and Kishiyama 2010). The influence of parental SES on child performance is largely attributed to environmental factors including increased family stress and reduced access to child enrichment activities. Interestingly, low parental SES has also been associated with increased attention problems in children (Hampton Wray et al. 2017; Rowland et al. 2018). Our site differences in IQ and inattention are consistent with these prior results where site differences in IQ and inattention were no longer significant when SES was included as a covariate. The results indicate the importance of accounting for family circumstances in child developmental studies. Yet at the same time, our resting state fMRI results halve not shown a relationship to SES in the current cohort (Faghiri et al. 2019), indicating that the effects of parental SES only selectively impact development. Notably, the number of traumas and reported parental stress are similar across sites. At the same time, there was considerable variability in the number of traumas experienced by children within this cohort and it is important to note that exposure to trauma is still implicated in altering brain development as demonstrated through our results derived from this cohort (Badura-Brack et al. 2020).

Both SNPs and epigenetics add another dimension to multi-modal brain imaging, and their integration can help identify intrinsic correlations between genetics and brain structures and functions, and further the link with behavioral and cognitive changes. To perform this multimodal integration, we have proposed sparse multiple canonical correlation analysis (SMCCA) (Hu, et al. 2018) to extract contributing features from each dataset while maximizing the cross-modality correlation, and found successful applications in detecting three-way correlations between SNPs, fMRI and methylation in MCIC data. In addition, we applied a graph-based semi-supervised learning (GSSL) approach (Bai, et al. 2019) for further integrative analysis, which accounts for genetics, brain imaging (endophenotypes), and environmental factors (epigenomics) respectively. Due to the high dimensionality, group structure, and mixed type of brain imaging and omics data, detection of their interactions is especially challenging. We have proposed a novel distance correlation based approach (Fang, et al. 2017) and applied this to 866 PNC samples with fMRI images and SNP profiles. We uncovered several statistically significant and biologically interesting interactions. These approaches are generalizable for three or more data modalities, and we plan to combine them with MEG data incorporate fine temporal information. There are some recognized limitations to the currently designed study. First, while we were able to match MEG systems across sites, the MRI scanners were not identical contributing some site-specific variability. Second, while many multi-site MRI studies have been conducted, fewer multi-site MEG studies have been conducted to date. Therefore, there is still less known about how to best harmonize across sites. A strength of this current study is that the MEG machine is the same across site and to the best of our ability (inclusion of stimulus timing testing), we harmonized the stimulus delivery variables, thereby limiting the site-specific variability which will be examined more closely in further analysis of the current dataset. A third limitation of the current dataset is the sample size. To compete with large normative datasets, the sample will need to be expanded to be able to understand each of the factors that influence neurodevelopment (SES, trauma, rearing environment, etc.). One challenge that remains is that collection and analysis of multi-modal datasets are costly. Yet our prior studies indicate that MRI and MEG contribute unique information about brain function and are therefore key to understanding brain function more broadly, while incorporating both modifiable and unmodifiable genetic information will better inform us of factors that lead to familial risk of diseases.

5. Conclusions

The DevCog study (http://devcog.mrn.org/) provides an important contribution to normative developmental neuroimaging datasets through the incorporation of a broad multimodal assessment of development across middle childhood. Prior studies provide strong evidence on concurrent changes in structure and function at both the neuronal and behavioral timescale. Our prior studies confirm that multimodal measures offer complementary views of brain function, supporting the need to assess brain development across the temporal and spatial scales available from each of the imaging modalities. Increasing evidence supports the need to understand typical brain development to improve the health of individuals across the age spectrum. The DevCog study will allow us to more fully capture the influence of developmental biological processes (functional and structural neuronal changes and genetic and epigenetic factors) at different timescales (milliseconds to years) on human behavior, contributing a rich resource for the study of brain development.

Table 4.

Estimated total intracranial volume (eTIV) by age

Age (years)	9	10	11	12	13	14
N	34	37	35	30	31	27
eTIV (SD)	1,513,724 (18,783)	1,473,973 (26,850)	1,508,910 (20,210)	1,532,481 (23,831)	1,542,192 (26,134)	1,552,594 (33,399)