Maturation in ADHD

Lag in maturation of the brain’s intrinsic functional architecture in ADHD 1 Abstract Attention deficit/hyperactivity disorder (ADHD) is the most common psychiatric disorder of childhood, and there is great interest in understanding its neurobiological basis. A highly influential neurodevelopmental hypothesis proposes ADHD is associated with a lag in maturation of critical brain variables. In the last decade, resting state functional imaging has enabled detailed investigation of connectivity patterns of the brain, and has revealed that the brain is functionally organized into a number of large-scale networks. In this study, we extend the maturational lag hypothesis in ADHD to the brain’s developing functional architecture. We demonstrate that the developing relationships between default network and task positive networks—which are the focus of current models of attention dysfunction in ADHD—exhibit significant and specific maturational lag in the disorder. 2 Main Text Attention deficit/hyperactivity disorder (ADHD) is a serious neuropsychiatric disorder characterized by inattention, hyperactivity, and impulsivity. One influential neurodevelopmental model of the disorder posits a lag in the maturational trajectories of key brain variables. This model has mostly been tested by examining developmental pathways of structural features of the brain (1, 2). In recent years, however, theorists have increasingly used resting state fMRI—scanning subjects in a task-free resting state—to explore the functional architecture of the brain. This work has led to the recognition that the human brain is organized into several largescale intrinsic connectivity networks (ICNs), each associated with specific neurocognitive functions (3, 4). ICNs have been shown to undergo significant maturation from childhood to early adulthood, exhibiting characteristic patterns of integration (increased connectivity with age) and segregation (decreased connectivity with age) with other ICNs (5–7). These advances raise new possibilities for investigating maturational lag in ADHD in the developing ICN architecture of the brain (8). Independent lines of research suggest that attention dysfunction in ADHD is linked to altered ICN interrelationships. According to current theoretical models (9, 10), inattention in ADHD involves altered competitive balance between default network, an ICN implicated in internally-directed mentation (11, 12), and several ‘taskpositive’ ICNs (TPNs)—including dorsal attention network (DAN), ventral attention network (VAN), and frontoparietal network (FPN)—that are involved in cognitively demanding externally focused processing. Consistent with these models, previous resting state fMRI studies in ADHD have reliably found abnormalities in functional connections within DMN and in its interconnections with TPNs (13, 14). Importantly, however, it is not currently known whether these abnormalities reliably observed in ADHD are linked to maturational lag. The current study sought to investigate this question. Based on current network models of ADHD, we hypothesized that maturational lag in ADHD in the brain’s intrinsic functional architecture would be focused within DMN and in its interconnections with three TPNs: DAN, VAN, and FPN. In order to test this hypothesis, we used connectomic methods. Connectivity alterations in DMN and TPNs have been shown to be highly variable across individual connections (15). Standard strategies that examine connectivity at a single or a handful of ‘seed’ regions, or that average connectivity values across entire ICNs, are likely to produce summary measures that mislead about underlying fine-grained patterns of variation, or can miss trends that are visible only when looking across large populations of connections. In order to produce comprehensive connectomic maps of the cortex, we placed 907 regions of interest (ROIs) at regular intervals throughout the cortex, and calculated connectivity between each pair of ROIs. Then using a multiple regression approach, we calculated the effect of age and the effect of ADHD at each connection of the connectome. This yielded two comprehensive maps of the effect of age and the effect of ADHD, respectively. 3 Examining patterns of correspondence across connections in these whole-cortex connectomic maps provides a powerful way to investigate the maturational lag hypothesis. Of note, we performed the connectomic analyses described below twice. The analysis described in the main text utilizes effect of age and effect ADHD estimates derived from participants in the ADHD-200 sample (16) , which after demographic and quality control exclusions encompassed 421 participants [288 typically developing controls (TDC), 135 ADHD]. We also performed the analysis a second time using effect of age estimates derived from a completely independent sample: 185 TDC participants from the multisite ABIDE sample (17). Results were remarkably similar in this second analysis (all significant statistical tests reported below from the first analysis were replicated in the second analysis, in some cases with higher levels of statistical significance; see Supplement). By repeating our analysis with a second large, multisite dataset, we provide additional support for the reliability of our findings. The maturational lag hypothesis predicts that in ICNs exhibiting maturational lag, there will be large and statistically significant negative correlations across connections between the effect of age and effect of ADHD. The rationale for this prediction is that, according to the hypothesis, ADHD selectively affects features of the brain that undergo maturation and imposes a lag on their maturational trajectory. Thus, to the extent that a connection tends to more strongly integrate with age, then it should be correspondingly weaker in ADHD relative to TDCs. Conversely, to the extent that a connection tends to more strongly segregate with age, it should be correspondingly stronger in ADHD relative to TDCs. Figure 1 shows the Pearson’s correlation coefficients across connections of the connectome between the effect of age and the effect of ADHD. These were calculated separately for each of seven major ICNs and their interconnections, with ICN boundaries determined by a widely used seven network parcellation (18). Three network pairs showed highly statistically significant negative correlations evidencing maturational lag: DMN-DMN, DMN-VAN, and DMN-FPN. Moreover, these negative correlations were relatively specific to these DMN-TPN interconnections (only two other cells were significant at p<0.05). This result is consistent with our a priori predictions that maturational lag in ADHD would be focused in DMN-TPN interconnections. Figure 2 shows the scatter plots of the relationship between the effect of age and effect of ADHD in DMN-DMN, DMN-VAN, and DMN-FPN. We were interested in the question of whether maturational lag due to ADHD could be detected in the subpopulation of connections that exhibit relatively strong maturation with age (based on a p<0.01 threshold). If ADHD does produce a lag in the maturation of these connections, then these strongly maturing connections should be more likely to lie in the upper left and lower right quadrants; in these ‘lagged quadrants’, the directional effect of maturation, i.e., integration versus segregation, is opposed by 4 the effect of ADHD. We found that for these strongly maturing connections, the odds of lying in a lagged quadrant were indeed substantially elevated (DMN-DMN 6.68:1; DMN-VAN 2.21:1; DMN-FPN 3.26:1), and permutation tests revealed that odds as high as these were unlikely to have arisen by chance [DMN-DMN (p<0.0001), DMNVAN (p=0.03), DMN-FPN (p<0.0001)]. We next visualized the inter-region distribution of these strongly maturing connections that are lagged in ADHD (Figure 3). Many of these inter-region links are noteworthy because of their hypothesized roles in normal attention functioning and their contribution to attention dysfunction in ADHD. According to current network models of ADHD, DMN and TPNs are, respectively, specialized for introspective versus extrospective orientations of attention (9). Individuals with the disorder exhibit insufficient regulatory control over DMN. This leads to inappropriate intrusion of DMN during externally demanding tasks, producing lapses in attention (9, 19), distractibility (20), and increased variability in task performance (21). Consistent with this ‘network regulation’ model of ADHD, our results show that critical regions involved in network switching and adaptive cognitive control display maturational lag in ADHD. Figure 3B shows a prominent lag in ADHD in anterior insula, a key hub in VAN. Anterior insula is implicated in detecting salient changes in the external environment (22, 23), and regulating shifts between introspective versus extrospective modes of attention (22, 24). There was also a lag in ADHD in integration of interconnections linking DMN and FPN (Figure 3C). FPN is a core network underlying adaptive cognitive control (25). This network flexibly binds to other ICNs in accordance with task demands, and rapidly modulates widespread ICN interconnectivity patterns (25, 26). In addition, FPN has been proposed to specifically modulate DMN activity in task contexts where introspective modes of attention are relevant (e.g., planning for one’s personal future) (26, 27). Our finding of developmental lag in regions such as anterior insula and FPN thus potentially suggests a developmental mechanism for the network regulation deficiencies proposed in current models of ADHD. We also observed maturational lag in integration of the dmPFC and PCC (Figure 3A)—the two regions that are widely thought to make up the functional core of the DMN (28). Functional abnormalities in DMN are among the most reliably seen in ADHD (8, 14, 29). Interestingly, a growing body of evidence links these functional abnormalities with disturbances in structural development (15). DMPFC and PCC are connected by the cingulum fiber bundle, and maturation of this structural connection has been shown to predict maturation of functional connectivity between these two regions (30). In a seminal investigation (2), Shaw and colleagues found diffuse maturational delay in cortical thickness in ADHD, with peak delays in medial prefrontal regions in DMN. These findings across imaging modalities underscore a broader point: The functional abnormalities we observed within and between DMN and TPNs are likely to be part of a larger picture of development delays of structural characteristics such as cortical thickness and white matter tracts (see 15 for a review). Joint, multimodal investigation of this ensemble of 5 structural and functional characteristics in ADHD promises to provide new insights into the mechanisms that produce and sustain the disorder. In summary, this study extends the maturational lag hypothesis in ADHD to the brain’s developing functional architecture, and in particular to connections within and between DMN and TPNs. Our results add a new developmental perspective to current network models of ADHD, and provide strong initial evidence that altered DMN-TPN relationships that are reliably seen in ADHD are related to delays in typical maturational patterns. References 1. L. P. Castellanos F, Developmental trajectories of brain volume abnormalities in children and adolescents with attention-deficit/hyperactivity disorder, JAMA 288, 1740–1748 (2002). 2. P. Shaw et al., Attention-deficit/hyperactivity disorder is characterized by a delay in cortical maturation, Proc. Natl. Acad. Sci. U. S. A. 104, 19649–19654 (2007). 3. M. D. Fox et al., The human brain is intrinsically organized into dynamic, anticorrelated functional networks, Proc Natl Acad Sci U A 102, 9673–8 (2005). 4. J. D. Power et al., Functional Network Organization of the Human Brain, Neuron 72, 665–678 (2011). 5. J. S. Anderson, M. A. Ferguson, M. Lopez-Larson, D. Yurgelun-Todd, Connectivity gradients between the default mode and attention control networks, Brain Connect. 1, 147–157 (2011). 6. D. A. Fair et al., Development of distinct control networks through segregation and integration, Proc. Natl. Acad. Sci. U. S. A. 104, 13507–13512 (2007). 7. D. A. Fair et al., The maturing architecture of the brain’s default network, Proc. Natl. Acad. Sci. 105, 4028–4032 (2008). 8. D. A. Fair et al., Atypical Default Network Connectivity in Youth with AttentionDeficit/Hyperactivity Disorder, Biol. Psychiatry 68, 1084–1091 (2010). 9. E. J. S. Sonuga-Barke, F. X. Castellanos, Spontaneous attentional fluctuations in impaired states and pathological conditions: A neurobiological hypothesis, Neurosci. Biobehav. Rev. 31, 977–986 (2007). 10. F. X. Castellanos, E. Proal, Large-scale brain systems in ADHD: beyond the prefrontal–striatal model, Trends Cogn. Sci. 16, 17–26 (2012). 11. M. E. Raichle et al., A default mode of brain function, Proc. Natl. Acad. Sci. 98, 676–682 (2001). 6 12. R. L. Buckner, J. R. Andrews-Hanna, D. L. Schacter, The Brain’s Default Network, Ann. N. Y. Acad. Sci. 1124, 1–38 (2008). 13. L. Tian et al., Altered resting-state functional connectivity patterns of anterior cingulate cortex in adolescents with attention deficit hyperactivity disorder, Neurosci. Lett. 400, 39–43 (2006). 14. F. X. Castellanos et al., Cingulate-Precuneus Interactions: A New Locus of Dysfunction in Adult Attention-Deficit/Hyperactivity Disorder, Biol. Psychiatry 63, 332–337 (2008). 15. K. Konrad, S. B. Eickhoff, Is the ADHD brain wired differently? A review on structural and functional connectivity in attention deficit hyperactivity disorder, Hum. Brain Mapp. 31, 904–916 (2010). 16. The ADHD-200 Consortium: A Model to Advance the Translational Potential of Neuroimaging in Clinical Neuroscience, Front. Syst. Neurosci. 6 (2012), doi:10.3389/fnsys.2012.00062. 17. A. Di Martino et al., The autism brain imaging data exchange: towards a largescale evaluation of the intrinsic brain architecture in autism, Mol. Psychiatry (2013), doi:10.1038/mp.2013.78. 18. B. T. T. Yeo et al., The organization of the human cerebral cortex estimated by intrinsic functional connectivity, J. Neurophysiol. 106, 1125–1165 (2011). 19. D. H. Weissman, K. C. Roberts, K. M. Visscher, M. G. Woldorff, The neural bases of momentary lapses in attention, Nat Neurosci 9, 971–8 (2006). 20. C. Fassbender et al., A lack of default network suppression is linked to increased distractibility in ADHD, Brain Res. 1273, 114–128 (2009). 21. F. X. Castellanos et al., Varieties of attention-deficit/hyperactivity disorderrelated intra-individual variability, Biol. Psychiatry 57, 1416–1423 (2005). 22. V. Menon, L. Q. Uddin, Saliency, switching, attention and control: a network model of insula function, Brain Struct. Funct. 214, 655–667 (2010). 23. M. A. Eckert et al., At the heart of the ventral attention system: the right anterior insula, Hum. Brain Mapp. 30, 2530–2541 (2009). 24. D. Sridharan, D. J. Levitin, V. Menon, A critical role for the right fronto-insular cortex in switching between central-executive and default-mode networks, Proc Natl Acad Sci U A 105, 12569–74 (2008). 25. M. W. Cole et al., Multi-task connectivity reveals flexible hubs for adaptive task control, Nat. Neurosci. 16, 1348–1355 (2013). 7 26. R. N. Spreng, J. Sepulcre, G. R. Turner, W. D. Stevens, D. L. Schacter, Intrinsic Architecture Underlying the Relations among the Default, Dorsal Attention, and Frontoparietal Control Networks of the Human Brain, J. Cogn. Neurosci. 25, 74–86 (2012). 27. R. N. Spreng, W. D. Stevens, J. P. Chamberlain, A. W. Gilmore, D. L. Schacter, Default network activity, coupled with the frontoparietal control network, supports goal-directed cognition, NeuroImage 53, 303–317 (2010). 28. J. R. Andrews-Hanna, J. S. Reidler, J. Sepulcre, R. Poulin, R. L. Buckner, Functionalanatomic fractionation of the brain’s default network, Neuron 65, 550–562 (2010). 29. L. Q. Uddin et al., Network homogeneity reveals decreased integrity of defaultmode network in ADHD, J. Neurosci. Methods 169, 249–254 (2008). 30. K. Supekar et al., Development of functional and structural connectivity within the default mode network in young children, NeuroImage 52, 290–301 (2010). 8 Figures Figure 1: Correlations across connections between effect of age and effect of ADHD. Correlations are shown separately for interconnections between seven major intrinsic connectivity networks. We had a priori hypotheses that we would observe maturational lag in the cells shaded red. Consistent with our hypotheses, highly statistically significant negative correlations indicating maturational lag were observed in DMN-DMN, DMN-VAN, and DMN-FPN. VN=Visual Network; SMN=Somatomotor Network; LN=Limbic Network; DAN=Dorsal Attention Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network; DMN=Default Mode Network. Figure 2: Scatter plots of the relationship between the effect of age and effect of ADHD. Weakly maturing connections are shown in light grey. Strongly maturing connections that are lagged in ADHD are shown in blue and red. Percentages refer to the percent of strongly maturing connections that lie within the respective quadrant. DMN=Default Mode Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network. 9 Figure 3: We identified populations of strongly maturing connections that exhibit maturational lag in ADHD. These circle graphs depict the inter-region distribution of these lagged connections. There is prominent lag in: A) Links between dorsomedial prefrontal cortex (dmPFC) and posterior cingulate cortex (PCC), B) Anterior insula (aINS) connections with PCC, and C) FPN-DMN interconnections. The functional significance of these region-to-region interconnections is discussed in the text. DMN=Default Mode Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network. Abbreviations for names of all subregions are listed in the Supplement. 10 Supplement (includes Supplemental Methods, 3 Supplemental Figures, 2 Supplemental Tables, and Supplemental References) Supplemental Methods Subjects and Scanning Subjects and scans for this study derive from two large, multi-site datasets ADHD-200 sample –A total of 757 participants underwent resting state scanning at seven contributing sites. The dataset comprised 481 typically developing control participants and 276 participants with a DSM-IV-TR diagnosis of ADHD. Demographic information is shown in Table S1. Autism Brain Image Date Exchange (ABIDE) –This sample comprised 573 typically developing control (TDC) participants from 21 contributing sites. While participants with autism were also available from this dataset, importantly, only healthy controls were used from this sample. Demographic information is shown in Table S2. In previous reports using the ADHD-200 dataset (Fair et al., 2013) and the ABIDE dataset (Di Martino et al., 2013), comprehensive descriptions of data provenance, additional demographic/phenotypic information, and scanner protocols are available. Imaging Sample Selection Analyses were limited to participants with: 1) MPRAGE anatomical images, with consistent near-full brain coverage (i.e., superior extent included the majority of frontal and parietal cortex and inferior extent included the temporal lobes) with successful registration; 2) Complete phenotypic information for main phenotypic variables (diagnosis, age, handedness); 2) Full IQ (FIQ) within two SD of the overall sample mean, 3) mean framewise displacement (FD) within two SD of the sample mean; 4) at least 40% of frames retained after application of framewise censoring for motion (‘motion scrubbing’; see below). Of note, for participants lacking a F4 or F2 IQ score, FIQ was estimated by averaging the participant’s performance and verbal IQ scores. The age range of the ADHD sample spanned 7.2 years minimum and 21.8 years maximum. The TDC participants from the ABIDE dataset had a wider age range, and since we were primarily interested in age effects, the age range from the ABIDE sample was trimmed to fall within that of the ADHD-200 sample. After applying these sample selection criteria, we analyzed resting state scans from 421 individuals (TDC=288; ADHD=133) from seven sites in the ADHD-200 sample and 185 TDC participants from 9 sites from the ABIDE sample. Preprocessing 11 Preprocessing steps were performed using statistical parametric mapping (SPM8; www.fil.ion.ucl.ac.uk/spm). Scans were slice-time corrected, realigned to the tenth scan in the experiment for correction of head motion, and co-registered with the high-resolution T1-weighted image. Normalization was performed using the voxelbased morphometry toolbox implemented in SPM8. The high-resolution T1weighted image was segmented into tissue types, bias-corrected, registered to MNI space, and then normalized using Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) (Ashburner, 2007). The resulting deformation fields were then applied to the functional images. Smoothing of functional data was performed with an 8 mm3 kernel. Connectome Generation In order to produce a whole-brain resting state functional connectome, we placed 4.35 mm radius ROIs encompassing 19 3x3x3 mm voxels in a regular grid spaced at 12 mm intervals throughout the brain. Given our 8 mm smoothing kernel, this density of sampling also ensured that information from the entirety of brain was comprehensively sampled. Since our main interest was cortical ICNs, we next removed all ROIs that fell more than 5 mm (Euclidean distance) from the Yeo and colleagues ICN parcellation of the brain (Yeo et al., 2011), yielding 907 ROIs in total. We chose the Yeo and colleagues network map because their study was based on a large number of subjects (1000 participants), they included multiple convergent methods to assess reliability, and their parcellation was derived using grid-based connectomic methods similar to the current study. Spatially averaged time series were extracted from each of the ROIs. Linear detrending was next performed, followed by nuisance regression. Nuisance regression: Regression was performed to remove nuisance effects. Regressors included six motion regressors generated from the realignment step, as well as their first derivatives. White matter and cerebrospinal fluid masks were generated from the VBM-based tissue segmentation step noted above, and eroded using the fslmaths program from FSL to eliminate border regions of potentially ambiguous tissue type. The top five principal components of the BOLD time series were extracted from each of the masks and included as regressors in the model—a method that has been demonstrated to effectively remove signals arising from the cardiac and respiratory cycle (Behzadi et al., 2007). The time-series for each ROI was then band-passed filtered in the .01 – .10 Hz range. Next, motion scrubbing was performed. Motion Scrubbing: Individual frames with excessive head motion were censored from the time series. The framewise displacement threshold for excessive motion was set at 0.2 mm (Fair et al., 2013). One frame before and two frames after the target frame was also removed to account for temporal blurring (Power et al., 2012). Subjects with more than 60% of their frames removed by scrubbing were 12 excluded from further analysis, a threshold justified by simulations conducted by other groups (Fair et al., 2013), as well as by our group. Next Pearson product-moment correlation coefficients were calculated pairwise between time courses for each of the 907 ROIs, producing a cross-correlation map with 410,871 non-redundant entries. Fisher’s r-z transformation was then applied. Calculating effect of ADHD and effect of age in the ADHD-200 sample Using the ADHD-200 dataset, we calculated the effect of ADHD at each connection of the connectome. Separately for each connection, we fit a multiple regression model that modeled the effect of disease (ADHD versus TDC) and the effect of age, while controlling for the effects of site, gender, full scale IQ, and handedness. We also included terms modeling the linear and quadratic effects of mean motion (framewise displacement averaged across the entire scan) to absorb any residual variance unaccounted for by the previously mentioned motion correction steps (i.e., removal of high motion subjects, regression of motion from the resting state time series, and scrubbing of high motion frames). We retained the betas representing the effect of ADHD and the effect of age from this analysis. These betas constitute, respectively, connectomic maps that represents the effect of ADHD and effect of age at each connection of the whole-cortex connectome (positive ADHD beta = greater connectivity in ADHD; positive age beta = greater connectivity with age). Calculating the effect of age in the ABIDE sample Using the TDC participants from the ABIDE dataset, we calculated the effect of age on each connection of the connectome. For each connection, we fit an identical regression model to that used in the ADHD-200 dataset above, except that since only TDC participants were included in the analysis, there was no term modeling the effect of disease. We retained the betas representing the effect of age from this analysis, yielding a connectomic map that represents the effect of age at each connection of the connectome (positive age beta = greater connectivity with age). Permutation Tests ICNs exhibit substantial intra- and inter-network dependency across connections. Thus we assessed the significance of all statistical tests with non-parametric permutation tests, which are robust to deviations from independence and normality assumptions of conventional tests (Good, 2000). Correlation tests: The distribution under chance of correlations between the effect of age and effect ADHD was generated by randomly permuting the disease labels (TDC versus ADHD) of the 421 ADHD-200 participants 10,000 times. At each iteration, we refit the multiple regression models described above and recalculated betas for the effect of ADHD. The correlation across connections between effect age and effect of ADHD was then recalculated. The correlation between age betas and ADHD betas that was actually observed was located in this null distribution in terms of rank, and statistical significance was set as this rank value divided by 10,000. 13 Since the multiple regression models fit at each iteration of the permutation test included covariates (gender, full scale IQ, and handedness, etc.), the procedure of Freedman and Lane (1983) was followed. In brief, a multiple regression model was first estimated with nuisance covariates alone, residuals were formed and were permuted. The covariate effect of interest was then included in the subsequent model, creating an approximate realization of data under the null hypothesis, and the statistical test of interest was calculated on this data (see FSL Randomise http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Randomise/Theory for a neuroimaging implementation). Odds of lying in ADHD-lagged quadrants: Our permutation approach to assessing statistical significance of observed odds of lying in ADHD-lagged quadrants followed a similar approach. We randomly permuted the disease labels (TDC versus ADHD) of the 421 ADHD-200 participants 10,000 times. At each iteration, we refit the multiple regression models described above and recalculated betas for the effect of ADHD (following the procedure of Freedman and Lane described above to deal with covariates). Next, we recalculated the odds of lying in a lagged quadrant (upper left and lower right) versus an unlagged quadrant (upper right and lower left). The odds of being in a lagged quadrant that was actually observed was located in this null distribution in terms of rank, and statistical significance was set as this rank value divided by 10,000. Circular Connection Graphs Generation In Figure 3 of the main manuscript and Figure S3 of the Supplement, we display visualizations that highlight regional distribution of ADHD effects within the implicated ICNs (i.e., DMN, VAN, and FPN). To do this, we leveraged the fact that these ICNs are comprised of a number of spatially distinct subregions. More specifically, DMN, VAN, and FPN networks were split into a number of distinct contiguous subregions using the “pick cluster” function in XJView. This yielded 21 large contiguous regions such as insula, posterior cingulate cortex, and supplementary motor area, many of which have right and left components (see Figure 3). Of note, since anterior medial frontal cortex comprises a massive contiguous region that is commonly split into ventral and dorsal components (e.g., Amodio and Frith, 2006), we split this region at the MNI z = 0 line. We then created circular graphs in which these subregions are connected by arcs. In computing the width of an arc connecting two subregions, we did the following separately for each of the three sets of network-to-network connections: Let P be the size of the population of strongly maturing connections (after p<0.01 thresholding of connectomic maps) that are lagged in ADHD (i.e., the connections colored blue and red in Figure 2 of the main manuscript). Let L be the number of connections within this population that links a pair of regions. The width the arc connecting this pair of regions was then set to be proportional to L/P. These circular graphs allow appreciation of the spatial distribution across the subregions of DMN, VAN, and FPN of this population of highly maturing connections that are lagged in ADHD. 14 Of note, to enhance the readability of the circular graphs, arcs that represent less than 1% of the internetwork connections were omitted. In addition, if subregions participated in less than 1% of total connections, they were omitted from the graph (for regions with right and left components, we required that both sides participate in less 1% of connections). The abbreviations used for the subregions in figures 2,3&4 are as follows: DMN: dmPFC=dorsomedial PFC, vmPFC=ventromedial PFC, ANG=angular gyrus, PCC=posterior cingulate cingulate, LTL=lateral temporal lobe, oIFG=orbital inferior frontal gyrus. Not shown: medial temporal lobe. VAN: aPCN=anterior precuneus, SMG=supramarginal gyrus, aINS=anterior insula, SMA=supplementary motor area. Not shown: superior lateral PFC, precentral cortex. FPN: pPCN=posterior precuneus, LPL=lateral parietal lobe, MCC=middle cingulate cortex, SFG=superior frontal gyrus, oMFG=orbital middle frontal gyrus, lPFC=lateral PFC. Not shown: middle insula, inferior temporal lobe. Supplemental Figures Figure S1: Using the effect of age estimates from a second independent sample (the ABIDE sample), we calculated correlations across connections between effect of age and effect of ADHD. Correlations are shown separately for interconnections between seven major intrinsic connectivity networks listed in bold across the diagonal. We had a priori hypotheses that we would observe maturational lag in the cells shaded red. Consistent with the analysis presented in the main manuscript, highly statically significant negative correlations indicating maturational lag were observed in DMNDMN, DMN-VAN, and DMN-FPN. VN=Visual Network; SMN=Somatomotor Network; 15 LN=Limbic Network; DAN=Dorsal Attention Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network; DMN=Default Mode Network. Figure S2: Using the effect of age estimates from a second independent sample (the ABIDE sample), we plotted the relationship between the effect of age and effect of ADHD. Weakly maturing connections are shown in light grey. Strongly maturing connections that are lagged in ADHD lie in the upper left and lower right quadrants ( ‘lagged quadrants’) and are shown in blue and red. Percentages refer to the percent of strongly maturing connections that lie within the respective quadrant. Further analysis of these scatter plots replicated the findings reported in the main manuscript. We found that for strongly maturing connections, the odds of lying in a lagged quadrant were substantially elevated (DMN-DMN 9.07:1; DMN-VAN 2.61:1; DMN-FPN 4.27:1), and permutation tests revealed that odds as high as these were unlikely to have arisen by chance [DMN-DMN (p<0.0002), DMN-VAN (p=0.0003), DMN-FPN (p<0.0001)]. DMN=Default Mode Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network. 16 Figure S3: Using the effect of age estimates from the ABIDE sample, we identified populations of strongly maturing connections that exhibit maturational lag in ADHD. These circle graphs depict the inter-region distribution of these lagged connections. Replicating the findings from Figure 3 in the main manuscript, there is prominent lag in: A) Links between dorsomedial prefrontal cortex (dmPFC) and posterior cingulate cortex (PCC), B) Anterior insula (aINS) connections with PCC, and C) FPN-DMN interconnections. DMN=Default Mode Network; VAN=Ventral Attention Network; FPN=Frontoparietal Network. Abbreviations for names of all subregions are listed in the Supplement. 17 Supplemental Tables Post-Exclusions Pre-Exclusions ADHD-200 Typically Developing Controls ADHD-200 ADHD Participants Site n Age % Male IQ n Age % Male IQ NYU 93 12.1±3.1 45.2% 110.7±13.9 116 11.3±2.7 77.6% 106.4±14.0 Beijing Pittsburgh OHSU NeuroImage Washington KKI Total 116 89 41 22 59 61 481 11.7±1.7 15.1±2.9 8.9±1.2 17.3±2.6 11.5±3.9 10.3±1.3 12.2±3.3 61.2% 51.7% 43.9% 50.0% 52.5% 55.7% 52.9% 118.1±13.3 109.8±11.5 118.7±12.6 111.2 116±14.1 111.5±10.3 113.7±12.9 78 12.4±2.0 105.4±13.2 37 23 8.8±1.0 16.7±3.0 22 276 10.2±1.6 11.6±3.0 91.0% NA 70.3% 82.6% NA 54.6% 79.1% NYU 49 12.7±2.9 44.9% 113.6±11.8 52 11.7±3.1 71.2% 107.5±12.9 Beijing Pittsburgh OHSU NeuroImage Washington KKI Total 89 54 19 15 24 38 288 118.1±11.9 113.1±9.9 116.6±12.5 111.2 114.9±11.2 111.2±11.5 114.8±11.3 47 12.5±2.1 91.5% NA 105.8±12.5 15 9 9.2±1.3 16.1±2.4 112.3±12.5 111.2* 10 133 11.1±1.7 12.0±2.9 73.3% 88.9% NA 60% 79.0% 11.8±1.8 15.7±2.8 9.2±1.5 16.0±2.4 13.8±4.1 10.5±1.3 12.8±3.2 58.4% 48.2% 47.4% 46.7% 45.8% 55.3% 51.4% 108.5±13.9 111.2* 106.0±15.2 106.8±13.3 100.8±14.4 107.2±12.5 Table S1: Sample characteristics of the ADHD-200 dataset. Sample characteristics are shown both prior to and after application of exclusion and quality control criteria. NYU=New York University; OHSU=Oregon Health and Science University; KKI= Kennedy Krieger Institute. 18 Post-Exclusions Pre-Exclusions ABIDE Typically Developing Controls Site n Age % Male IQ Caltech 19 28.9±11.2 78.9% 114.2±9.7 CMU KKI Leuven 1 Leuven 2 MAX MUN NYU OHSU OLIN Pittsburgh SBL SDSU Stanford Trinity UCLA 1 UCLA 2 UM 1 UM 2 USM YALE Total 13 33 15 20 33 105 15 16 27 15 22 20 25 33 14 55 22 43 28 573 26.8±5.7 10.2±1.3 23.3±2.9 14.3±1.5 26.2±9.8 15.8±6.3 10.1±1.1 16.9±3.7 18.9±6.6 33.7±6.6 14.2±1.9 10.0±1.6 17.1±3.8 13.2±2.1 12.2±1.1 14.1±3.2 16.6±3.9 21.4±7.6 12.7±2.8 17.1±7.7 76.9% 72.7% 100% 75% 87.9% 75.2% 100% 87.5% 85.2% 100% 72.7% 80% 100% 87.9% 85.7% 69.1% 95.5% 100% 71.4% 82.7% 114.6±9.7 113.4±9.3 114.8±12.9 110.7±8.6 111.5±8.8 113.2±13.1 115.7±11.1 114.9±16.5 110.1±9.4 111.2* 108.1±10.5 112.1±15.4 110.9±12.2 105.2±10.6 109.5±12.8 107.1±9.6 111.1±9.3 115.1±13.8 105.0±17.4 111.2±12.0 KKI 13 10.5±1.4 61.5% 112.9±11.1 NYU SDSU Trinity UCLA 1 UM 1 UM 2 USM YALE Total 62 14 12 17 31 14 11 11 185 13.3±3.9 14.3±1.3 16.6±3.0 13.7±2.3 14.6±3.1 15.8±1.5 15.9±2.7 13.4±2.7 14.0±3.3 72.6% 78.6% 100% 76.5% 67.7% 92.9% 100% 54.5% 75.7% 113.9±10.9 105.9±9.9 107.6±11.5 104.9±9.7 109.1±9.9 109.3±11.0 110.1±10.7 103.5±11.1 110.0±11.0 Table S2: Sample characteristics of the ABIDE dataset. Sample characteristics are shown both prior to and after application of exclusion and quality control criteria. CMU=Carnegie Mellon University; KKI= Kennedy Krieger Institute; Leuven=; Max Mun= Ludwig Maximilians University Munich; NYU=New York University; OHSU=Oregon Health and Science University; Olin=Olin, Institute of Living at Hartford Hospital; SBL= Social Brain Lab BCN NIC UMC Groningen and Netherlands Institute for Neurosciences; SDSU=San Diego State University; USM=University of Utah School of Medicine; YALE=Yale Child Study Center. 19 Supplemental References 1. D. A. Fair et al., Distinct neural signatures detected for ADHD subtypes after controlling for micro-movements in resting state functional connectivity MRI data, Front. Syst. Neurosci. 6, 80 (2013). 2. A. Di Martino et al., The autism brain imaging data exchange: towards a largescale evaluation of the intrinsic brain architecture in autism, Mol. Psychiatry (2013), doi:10.1038/mp.2013.78. 3. J. Ashburner, A fast diffeomorphic image registration algorithm, NeuroImage 38, 95–113 (2007). 4. B. T. T. Yeo et al., The organization of the human cerebral cortex estimated by intrinsic functional connectivity, J. Neurophysiol. 106, 1125–1165 (2011). 5. Y. Behzadi, K. Restom, J. Liau, T. T. Liu, A component based noise correction method (CompCor) for BOLD and perfusion based fMRI, NeuroImage 37, 90–101 (2007). 6. J. D. Power, K. A. Barnes, A. Z. Snyder, B. L. Schlaggar, S. E. Petersen, Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion, NeuroImage 59, 2142–2154 (2012). 7. P. Good, Permutation Tests: A Practical Guide to Resampling Methods for Testing Hypotheses (Springer, ed. 2nd, 2000). 8. D. M. Amodio, C. D. Frith, Meeting of minds: the medial frontal cortex and social cognition, Nat Rev Neurosci 7, 268–77 (2006). 20